1 //
2 // Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 //
5 // This code is free software; you can redistribute it and/or modify it
6 // under the terms of the GNU General Public License version 2 only, as
7 // published by the Free Software Foundation.
8 //
9 // This code is distributed in the hope that it will be useful, but WITHOUT
10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 // version 2 for more details (a copy is included in the LICENSE file that
13 // accompanied this code).
14 //
15 // You should have received a copy of the GNU General Public License version
16 // 2 along with this work; if not, write to the Free Software Foundation,
17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 //
19 // Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 // or visit www.oracle.com if you need additional information or have any
21 // questions.
22 //
23 //
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 // it does if the polling page is more than disp32 away.
579 bool SafePointNode::needs_polling_address_input()
580 {
581 return Assembler::is_polling_page_far();
582 }
583
584 //
585 // Compute padding required for nodes which need alignment
586 //
587
588 // The address of the call instruction needs to be 4-byte aligned to
589 // ensure that it does not span a cache line so that it can be patched.
590 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
591 {
592 current_offset += 1; // skip call opcode byte
593 return round_to(current_offset, alignment_required()) - current_offset;
594 }
595
596 // The address of the call instruction needs to be 4-byte aligned to
597 // ensure that it does not span a cache line so that it can be patched.
598 int CallStaticJavaHandleNode::compute_padding(int current_offset) const
599 {
600 current_offset += preserve_SP_size(); // skip mov rbp, rsp
601 current_offset += 1; // skip call opcode byte
602 return round_to(current_offset, alignment_required()) - current_offset;
603 }
604
605 // The address of the call instruction needs to be 4-byte aligned to
606 // ensure that it does not span a cache line so that it can be patched.
607 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
608 {
609 current_offset += 11; // skip movq instruction + call opcode byte
610 return round_to(current_offset, alignment_required()) - current_offset;
611 }
612
613 #ifndef PRODUCT
614 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
615 {
616 st->print("INT3");
617 }
618 #endif
619
620 // EMIT_RM()
621 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) {
622 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
623 cbuf.insts()->emit_int8(c);
624 }
625
626 // EMIT_CC()
627 void emit_cc(CodeBuffer &cbuf, int f1, int f2) {
628 unsigned char c = (unsigned char) (f1 | f2);
629 cbuf.insts()->emit_int8(c);
630 }
631
632 // EMIT_OPCODE()
633 void emit_opcode(CodeBuffer &cbuf, int code) {
634 cbuf.insts()->emit_int8((unsigned char) code);
635 }
636
637 // EMIT_OPCODE() w/ relocation information
638 void emit_opcode(CodeBuffer &cbuf,
639 int code, relocInfo::relocType reloc, int offset, int format)
640 {
641 cbuf.relocate(cbuf.insts_mark() + offset, reloc, format);
642 emit_opcode(cbuf, code);
643 }
644
645 // EMIT_D8()
646 void emit_d8(CodeBuffer &cbuf, int d8) {
647 cbuf.insts()->emit_int8((unsigned char) d8);
648 }
649
650 // EMIT_D16()
651 void emit_d16(CodeBuffer &cbuf, int d16) {
652 cbuf.insts()->emit_int16(d16);
653 }
654
655 // EMIT_D32()
656 void emit_d32(CodeBuffer &cbuf, int d32) {
657 cbuf.insts()->emit_int32(d32);
658 }
659
660 // EMIT_D64()
661 void emit_d64(CodeBuffer &cbuf, int64_t d64) {
662 cbuf.insts()->emit_int64(d64);
663 }
664
665 // emit 32 bit value and construct relocation entry from relocInfo::relocType
666 void emit_d32_reloc(CodeBuffer& cbuf,
667 int d32,
668 relocInfo::relocType reloc,
669 int format)
670 {
671 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
672 cbuf.relocate(cbuf.insts_mark(), reloc, format);
673 cbuf.insts()->emit_int32(d32);
674 }
675
676 // emit 32 bit value and construct relocation entry from RelocationHolder
677 void emit_d32_reloc(CodeBuffer& cbuf, int d32, RelocationHolder const& rspec, int format) {
678 #ifdef ASSERT
679 if (rspec.reloc()->type() == relocInfo::oop_type &&
680 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
681 assert(oop((intptr_t)d32)->is_oop() && (ScavengeRootsInCode || !oop((intptr_t)d32)->is_scavengable()), "cannot embed scavengable oops in code");
682 }
683 #endif
684 cbuf.relocate(cbuf.insts_mark(), rspec, format);
685 cbuf.insts()->emit_int32(d32);
686 }
687
688 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
689 address next_ip = cbuf.insts_end() + 4;
690 emit_d32_reloc(cbuf, (int) (addr - next_ip),
691 external_word_Relocation::spec(addr),
692 RELOC_DISP32);
693 }
694
695
696 // emit 64 bit value and construct relocation entry from relocInfo::relocType
697 void emit_d64_reloc(CodeBuffer& cbuf, int64_t d64, relocInfo::relocType reloc, int format) {
698 cbuf.relocate(cbuf.insts_mark(), reloc, format);
699 cbuf.insts()->emit_int64(d64);
700 }
701
702 // emit 64 bit value and construct relocation entry from RelocationHolder
703 void emit_d64_reloc(CodeBuffer& cbuf, int64_t d64, RelocationHolder const& rspec, int format) {
704 #ifdef ASSERT
705 if (rspec.reloc()->type() == relocInfo::oop_type &&
706 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
707 assert(oop(d64)->is_oop() && (ScavengeRootsInCode || !oop(d64)->is_scavengable()),
708 "cannot embed scavengable oops in code");
709 }
710 #endif
711 cbuf.relocate(cbuf.insts_mark(), rspec, format);
712 cbuf.insts()->emit_int64(d64);
713 }
714
715 // Access stack slot for load or store
716 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
717 {
718 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src])
719 if (-0x80 <= disp && disp < 0x80) {
720 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte
721 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
722 emit_d8(cbuf, disp); // Displacement // R/M byte
723 } else {
724 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte
725 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
726 emit_d32(cbuf, disp); // Displacement // R/M byte
727 }
728 }
729
730 // rRegI ereg, memory mem) %{ // emit_reg_mem
731 void encode_RegMem(CodeBuffer &cbuf,
732 int reg,
733 int base, int index, int scale, int disp, bool disp_is_oop)
734 {
735 assert(!disp_is_oop, "cannot have disp");
736 int regenc = reg & 7;
737 int baseenc = base & 7;
738 int indexenc = index & 7;
739
740 // There is no index & no scale, use form without SIB byte
741 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
742 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
743 if (disp == 0 && base != RBP_enc && base != R13_enc) {
744 emit_rm(cbuf, 0x0, regenc, baseenc); // *
745 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
746 // If 8-bit displacement, mode 0x1
747 emit_rm(cbuf, 0x1, regenc, baseenc); // *
748 emit_d8(cbuf, disp);
749 } else {
750 // If 32-bit displacement
751 if (base == -1) { // Special flag for absolute address
752 emit_rm(cbuf, 0x0, regenc, 0x5); // *
753 if (disp_is_oop) {
754 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
755 } else {
756 emit_d32(cbuf, disp);
757 }
758 } else {
759 // Normal base + offset
760 emit_rm(cbuf, 0x2, regenc, baseenc); // *
761 if (disp_is_oop) {
762 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
763 } else {
764 emit_d32(cbuf, disp);
765 }
766 }
767 }
768 } else {
769 // Else, encode with the SIB byte
770 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
771 if (disp == 0 && base != RBP_enc && base != R13_enc) {
772 // If no displacement
773 emit_rm(cbuf, 0x0, regenc, 0x4); // *
774 emit_rm(cbuf, scale, indexenc, baseenc);
775 } else {
776 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
777 // If 8-bit displacement, mode 0x1
778 emit_rm(cbuf, 0x1, regenc, 0x4); // *
779 emit_rm(cbuf, scale, indexenc, baseenc);
780 emit_d8(cbuf, disp);
781 } else {
782 // If 32-bit displacement
783 if (base == 0x04 ) {
784 emit_rm(cbuf, 0x2, regenc, 0x4);
785 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
786 } else {
787 emit_rm(cbuf, 0x2, regenc, 0x4);
788 emit_rm(cbuf, scale, indexenc, baseenc); // *
789 }
790 if (disp_is_oop) {
791 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
792 } else {
793 emit_d32(cbuf, disp);
794 }
795 }
796 }
797 }
798 }
799
800 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
801 {
802 if (dstenc != srcenc) {
803 if (dstenc < 8) {
804 if (srcenc >= 8) {
805 emit_opcode(cbuf, Assembler::REX_B);
806 srcenc -= 8;
807 }
808 } else {
809 if (srcenc < 8) {
810 emit_opcode(cbuf, Assembler::REX_R);
811 } else {
812 emit_opcode(cbuf, Assembler::REX_RB);
813 srcenc -= 8;
814 }
815 dstenc -= 8;
816 }
817
818 emit_opcode(cbuf, 0x8B);
819 emit_rm(cbuf, 0x3, dstenc, srcenc);
820 }
821 }
822
823 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
824 if( dst_encoding == src_encoding ) {
825 // reg-reg copy, use an empty encoding
826 } else {
827 MacroAssembler _masm(&cbuf);
828
829 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
830 }
831 }
832
833 // This could be in MacroAssembler but it's fairly C2 specific
834 void emit_cmpfp_fixup(MacroAssembler& _masm) {
835 Label exit;
836 __ jccb(Assembler::noParity, exit);
837 __ pushf();
838 __ andq(Address(rsp, 0), 0xffffff2b);
839 __ popf();
840 __ bind(exit);
841 __ nop(); // (target for branch to avoid branch to branch)
842 }
843
844
845 //=============================================================================
846 const bool Matcher::constant_table_absolute_addressing = true;
847 const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
848
849 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
850 // Empty encoding
851 }
852
853 uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
854 return 0;
855 }
856
857 #ifndef PRODUCT
858 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
859 st->print("# MachConstantBaseNode (empty encoding)");
860 }
861 #endif
862
863
864 //=============================================================================
865 #ifndef PRODUCT
866 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
867 {
868 Compile* C = ra_->C;
869
870 int framesize = C->frame_slots() << LogBytesPerInt;
871 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
872 // Remove wordSize for return adr already pushed
873 // and another for the RBP we are going to save
874 framesize -= 2*wordSize;
875 bool need_nop = true;
876
877 // Calls to C2R adapters often do not accept exceptional returns.
878 // We require that their callers must bang for them. But be
879 // careful, because some VM calls (such as call site linkage) can
880 // use several kilobytes of stack. But the stack safety zone should
881 // account for that. See bugs 4446381, 4468289, 4497237.
882 if (C->need_stack_bang(framesize)) {
883 st->print_cr("# stack bang"); st->print("\t");
884 need_nop = false;
885 }
886 st->print_cr("pushq rbp"); st->print("\t");
887
888 if (VerifyStackAtCalls) {
889 // Majik cookie to verify stack depth
890 st->print_cr("pushq 0xffffffffbadb100d"
891 "\t# Majik cookie for stack depth check");
892 st->print("\t");
893 framesize -= wordSize; // Remove 2 for cookie
894 need_nop = false;
895 }
896
897 if (framesize) {
898 st->print("subq rsp, #%d\t# Create frame", framesize);
899 if (framesize < 0x80 && need_nop) {
900 st->print("\n\tnop\t# nop for patch_verified_entry");
901 }
902 }
903 }
904 #endif
905
906 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
907 {
908 Compile* C = ra_->C;
909
910 // WARNING: Initial instruction MUST be 5 bytes or longer so that
911 // NativeJump::patch_verified_entry will be able to patch out the entry
912 // code safely. The fldcw is ok at 6 bytes, the push to verify stack
913 // depth is ok at 5 bytes, the frame allocation can be either 3 or
914 // 6 bytes. So if we don't do the fldcw or the push then we must
915 // use the 6 byte frame allocation even if we have no frame. :-(
916 // If method sets FPU control word do it now
917
918 int framesize = C->frame_slots() << LogBytesPerInt;
919 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
920 // Remove wordSize for return adr already pushed
921 // and another for the RBP we are going to save
922 framesize -= 2*wordSize;
923 bool need_nop = true;
924
925 // Calls to C2R adapters often do not accept exceptional returns.
926 // We require that their callers must bang for them. But be
927 // careful, because some VM calls (such as call site linkage) can
928 // use several kilobytes of stack. But the stack safety zone should
929 // account for that. See bugs 4446381, 4468289, 4497237.
930 if (C->need_stack_bang(framesize)) {
931 MacroAssembler masm(&cbuf);
932 masm.generate_stack_overflow_check(framesize);
933 need_nop = false;
934 }
935
936 // We always push rbp so that on return to interpreter rbp will be
937 // restored correctly and we can correct the stack.
938 emit_opcode(cbuf, 0x50 | RBP_enc);
939
940 if (VerifyStackAtCalls) {
941 // Majik cookie to verify stack depth
942 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
943 emit_d32(cbuf, 0xbadb100d);
944 framesize -= wordSize; // Remove 2 for cookie
945 need_nop = false;
946 }
947
948 if (framesize) {
949 emit_opcode(cbuf, Assembler::REX_W);
950 if (framesize < 0x80) {
951 emit_opcode(cbuf, 0x83); // sub SP,#framesize
952 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
953 emit_d8(cbuf, framesize);
954 if (need_nop) {
955 emit_opcode(cbuf, 0x90); // nop
956 }
957 } else {
958 emit_opcode(cbuf, 0x81); // sub SP,#framesize
959 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
960 emit_d32(cbuf, framesize);
961 }
962 }
963
964 C->set_frame_complete(cbuf.insts_size());
965
966 #ifdef ASSERT
967 if (VerifyStackAtCalls) {
968 Label L;
969 MacroAssembler masm(&cbuf);
970 masm.push(rax);
971 masm.mov(rax, rsp);
972 masm.andptr(rax, StackAlignmentInBytes-1);
973 masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
974 masm.pop(rax);
975 masm.jcc(Assembler::equal, L);
976 masm.stop("Stack is not properly aligned!");
977 masm.bind(L);
978 }
979 #endif
980 }
981
982 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
983 {
984 return MachNode::size(ra_); // too many variables; just compute it
985 // the hard way
986 }
987
988 int MachPrologNode::reloc() const
989 {
990 return 0; // a large enough number
991 }
992
993 //=============================================================================
994 #ifndef PRODUCT
995 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
996 {
997 Compile* C = ra_->C;
998 int framesize = C->frame_slots() << LogBytesPerInt;
999 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1000 // Remove word for return adr already pushed
1001 // and RBP
1002 framesize -= 2*wordSize;
1003
1004 if (framesize) {
1005 st->print_cr("addq rsp, %d\t# Destroy frame", framesize);
1006 st->print("\t");
1007 }
1008
1009 st->print_cr("popq rbp");
1010 if (do_polling() && C->is_method_compilation()) {
1011 st->print("\t");
1012 if (Assembler::is_polling_page_far()) {
1013 st->print_cr("movq rscratch1, #polling_page_address\n\t"
1014 "testl rax, [rscratch1]\t"
1015 "# Safepoint: poll for GC");
1016 } else {
1017 st->print_cr("testl rax, [rip + #offset_to_poll_page]\t"
1018 "# Safepoint: poll for GC");
1019 }
1020 }
1021 }
1022 #endif
1023
1024 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1025 {
1026 Compile* C = ra_->C;
1027 int framesize = C->frame_slots() << LogBytesPerInt;
1028 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1029 // Remove word for return adr already pushed
1030 // and RBP
1031 framesize -= 2*wordSize;
1032
1033 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1034
1035 if (framesize) {
1036 emit_opcode(cbuf, Assembler::REX_W);
1037 if (framesize < 0x80) {
1038 emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1039 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1040 emit_d8(cbuf, framesize);
1041 } else {
1042 emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1043 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1044 emit_d32(cbuf, framesize);
1045 }
1046 }
1047
1048 // popq rbp
1049 emit_opcode(cbuf, 0x58 | RBP_enc);
1050
1051 if (do_polling() && C->is_method_compilation()) {
1052 MacroAssembler _masm(&cbuf);
1053 AddressLiteral polling_page(os::get_polling_page(), relocInfo::poll_return_type);
1054 if (Assembler::is_polling_page_far()) {
1055 __ lea(rscratch1, polling_page);
1056 __ relocate(relocInfo::poll_return_type);
1057 __ testl(rax, Address(rscratch1, 0));
1058 } else {
1059 __ testl(rax, polling_page);
1060 }
1061 }
1062 }
1063
1064 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1065 {
1066 return MachNode::size(ra_); // too many variables; just compute it
1067 // the hard way
1068 }
1069
1070 int MachEpilogNode::reloc() const
1071 {
1072 return 2; // a large enough number
1073 }
1074
1075 const Pipeline* MachEpilogNode::pipeline() const
1076 {
1077 return MachNode::pipeline_class();
1078 }
1079
1080 int MachEpilogNode::safepoint_offset() const
1081 {
1082 return 0;
1083 }
1084
1085 //=============================================================================
1086
1087 enum RC {
1088 rc_bad,
1089 rc_int,
1090 rc_float,
1091 rc_stack
1092 };
1093
1094 static enum RC rc_class(OptoReg::Name reg)
1095 {
1096 if( !OptoReg::is_valid(reg) ) return rc_bad;
1097
1098 if (OptoReg::is_stack(reg)) return rc_stack;
1099
1100 VMReg r = OptoReg::as_VMReg(reg);
1101
1102 if (r->is_Register()) return rc_int;
1103
1104 assert(r->is_XMMRegister(), "must be");
1105 return rc_float;
1106 }
1107
1108 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1109 PhaseRegAlloc* ra_,
1110 bool do_size,
1111 outputStream* st) const
1112 {
1113
1114 // Get registers to move
1115 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1116 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1117 OptoReg::Name dst_second = ra_->get_reg_second(this);
1118 OptoReg::Name dst_first = ra_->get_reg_first(this);
1119
1120 enum RC src_second_rc = rc_class(src_second);
1121 enum RC src_first_rc = rc_class(src_first);
1122 enum RC dst_second_rc = rc_class(dst_second);
1123 enum RC dst_first_rc = rc_class(dst_first);
1124
1125 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1126 "must move at least 1 register" );
1127
1128 if (src_first == dst_first && src_second == dst_second) {
1129 // Self copy, no move
1130 return 0;
1131 } else if (src_first_rc == rc_stack) {
1132 // mem ->
1133 if (dst_first_rc == rc_stack) {
1134 // mem -> mem
1135 assert(src_second != dst_first, "overlap");
1136 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1137 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1138 // 64-bit
1139 int src_offset = ra_->reg2offset(src_first);
1140 int dst_offset = ra_->reg2offset(dst_first);
1141 if (cbuf) {
1142 emit_opcode(*cbuf, 0xFF);
1143 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1144
1145 emit_opcode(*cbuf, 0x8F);
1146 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1147
1148 #ifndef PRODUCT
1149 } else if (!do_size) {
1150 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1151 "popq [rsp + #%d]",
1152 src_offset,
1153 dst_offset);
1154 #endif
1155 }
1156 return
1157 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1158 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1159 } else {
1160 // 32-bit
1161 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1162 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1163 // No pushl/popl, so:
1164 int src_offset = ra_->reg2offset(src_first);
1165 int dst_offset = ra_->reg2offset(dst_first);
1166 if (cbuf) {
1167 emit_opcode(*cbuf, Assembler::REX_W);
1168 emit_opcode(*cbuf, 0x89);
1169 emit_opcode(*cbuf, 0x44);
1170 emit_opcode(*cbuf, 0x24);
1171 emit_opcode(*cbuf, 0xF8);
1172
1173 emit_opcode(*cbuf, 0x8B);
1174 encode_RegMem(*cbuf,
1175 RAX_enc,
1176 RSP_enc, 0x4, 0, src_offset,
1177 false);
1178
1179 emit_opcode(*cbuf, 0x89);
1180 encode_RegMem(*cbuf,
1181 RAX_enc,
1182 RSP_enc, 0x4, 0, dst_offset,
1183 false);
1184
1185 emit_opcode(*cbuf, Assembler::REX_W);
1186 emit_opcode(*cbuf, 0x8B);
1187 emit_opcode(*cbuf, 0x44);
1188 emit_opcode(*cbuf, 0x24);
1189 emit_opcode(*cbuf, 0xF8);
1190
1191 #ifndef PRODUCT
1192 } else if (!do_size) {
1193 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1194 "movl rax, [rsp + #%d]\n\t"
1195 "movl [rsp + #%d], rax\n\t"
1196 "movq rax, [rsp - #8]",
1197 src_offset,
1198 dst_offset);
1199 #endif
1200 }
1201 return
1202 5 + // movq
1203 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1204 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1205 5; // movq
1206 }
1207 } else if (dst_first_rc == rc_int) {
1208 // mem -> gpr
1209 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1210 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1211 // 64-bit
1212 int offset = ra_->reg2offset(src_first);
1213 if (cbuf) {
1214 if (Matcher::_regEncode[dst_first] < 8) {
1215 emit_opcode(*cbuf, Assembler::REX_W);
1216 } else {
1217 emit_opcode(*cbuf, Assembler::REX_WR);
1218 }
1219 emit_opcode(*cbuf, 0x8B);
1220 encode_RegMem(*cbuf,
1221 Matcher::_regEncode[dst_first],
1222 RSP_enc, 0x4, 0, offset,
1223 false);
1224 #ifndef PRODUCT
1225 } else if (!do_size) {
1226 st->print("movq %s, [rsp + #%d]\t# spill",
1227 Matcher::regName[dst_first],
1228 offset);
1229 #endif
1230 }
1231 return
1232 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1233 } else {
1234 // 32-bit
1235 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1236 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1237 int offset = ra_->reg2offset(src_first);
1238 if (cbuf) {
1239 if (Matcher::_regEncode[dst_first] >= 8) {
1240 emit_opcode(*cbuf, Assembler::REX_R);
1241 }
1242 emit_opcode(*cbuf, 0x8B);
1243 encode_RegMem(*cbuf,
1244 Matcher::_regEncode[dst_first],
1245 RSP_enc, 0x4, 0, offset,
1246 false);
1247 #ifndef PRODUCT
1248 } else if (!do_size) {
1249 st->print("movl %s, [rsp + #%d]\t# spill",
1250 Matcher::regName[dst_first],
1251 offset);
1252 #endif
1253 }
1254 return
1255 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1256 ((Matcher::_regEncode[dst_first] < 8)
1257 ? 3
1258 : 4); // REX
1259 }
1260 } else if (dst_first_rc == rc_float) {
1261 // mem-> xmm
1262 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1263 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1264 // 64-bit
1265 int offset = ra_->reg2offset(src_first);
1266 if (cbuf) {
1267 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1268 if (Matcher::_regEncode[dst_first] >= 8) {
1269 emit_opcode(*cbuf, Assembler::REX_R);
1270 }
1271 emit_opcode(*cbuf, 0x0F);
1272 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1273 encode_RegMem(*cbuf,
1274 Matcher::_regEncode[dst_first],
1275 RSP_enc, 0x4, 0, offset,
1276 false);
1277 #ifndef PRODUCT
1278 } else if (!do_size) {
1279 st->print("%s %s, [rsp + #%d]\t# spill",
1280 UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1281 Matcher::regName[dst_first],
1282 offset);
1283 #endif
1284 }
1285 return
1286 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1287 ((Matcher::_regEncode[dst_first] < 8)
1288 ? 5
1289 : 6); // REX
1290 } else {
1291 // 32-bit
1292 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1293 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1294 int offset = ra_->reg2offset(src_first);
1295 if (cbuf) {
1296 emit_opcode(*cbuf, 0xF3);
1297 if (Matcher::_regEncode[dst_first] >= 8) {
1298 emit_opcode(*cbuf, Assembler::REX_R);
1299 }
1300 emit_opcode(*cbuf, 0x0F);
1301 emit_opcode(*cbuf, 0x10);
1302 encode_RegMem(*cbuf,
1303 Matcher::_regEncode[dst_first],
1304 RSP_enc, 0x4, 0, offset,
1305 false);
1306 #ifndef PRODUCT
1307 } else if (!do_size) {
1308 st->print("movss %s, [rsp + #%d]\t# spill",
1309 Matcher::regName[dst_first],
1310 offset);
1311 #endif
1312 }
1313 return
1314 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1315 ((Matcher::_regEncode[dst_first] < 8)
1316 ? 5
1317 : 6); // REX
1318 }
1319 }
1320 } else if (src_first_rc == rc_int) {
1321 // gpr ->
1322 if (dst_first_rc == rc_stack) {
1323 // gpr -> mem
1324 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1325 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1326 // 64-bit
1327 int offset = ra_->reg2offset(dst_first);
1328 if (cbuf) {
1329 if (Matcher::_regEncode[src_first] < 8) {
1330 emit_opcode(*cbuf, Assembler::REX_W);
1331 } else {
1332 emit_opcode(*cbuf, Assembler::REX_WR);
1333 }
1334 emit_opcode(*cbuf, 0x89);
1335 encode_RegMem(*cbuf,
1336 Matcher::_regEncode[src_first],
1337 RSP_enc, 0x4, 0, offset,
1338 false);
1339 #ifndef PRODUCT
1340 } else if (!do_size) {
1341 st->print("movq [rsp + #%d], %s\t# spill",
1342 offset,
1343 Matcher::regName[src_first]);
1344 #endif
1345 }
1346 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1347 } else {
1348 // 32-bit
1349 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1350 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1351 int offset = ra_->reg2offset(dst_first);
1352 if (cbuf) {
1353 if (Matcher::_regEncode[src_first] >= 8) {
1354 emit_opcode(*cbuf, Assembler::REX_R);
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("movl [rsp + #%d], %s\t# spill",
1364 offset,
1365 Matcher::regName[src_first]);
1366 #endif
1367 }
1368 return
1369 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1370 ((Matcher::_regEncode[src_first] < 8)
1371 ? 3
1372 : 4); // REX
1373 }
1374 } else if (dst_first_rc == rc_int) {
1375 // gpr -> gpr
1376 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1377 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1378 // 64-bit
1379 if (cbuf) {
1380 if (Matcher::_regEncode[dst_first] < 8) {
1381 if (Matcher::_regEncode[src_first] < 8) {
1382 emit_opcode(*cbuf, Assembler::REX_W);
1383 } else {
1384 emit_opcode(*cbuf, Assembler::REX_WB);
1385 }
1386 } else {
1387 if (Matcher::_regEncode[src_first] < 8) {
1388 emit_opcode(*cbuf, Assembler::REX_WR);
1389 } else {
1390 emit_opcode(*cbuf, Assembler::REX_WRB);
1391 }
1392 }
1393 emit_opcode(*cbuf, 0x8B);
1394 emit_rm(*cbuf, 0x3,
1395 Matcher::_regEncode[dst_first] & 7,
1396 Matcher::_regEncode[src_first] & 7);
1397 #ifndef PRODUCT
1398 } else if (!do_size) {
1399 st->print("movq %s, %s\t# spill",
1400 Matcher::regName[dst_first],
1401 Matcher::regName[src_first]);
1402 #endif
1403 }
1404 return 3; // REX
1405 } else {
1406 // 32-bit
1407 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1408 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1409 if (cbuf) {
1410 if (Matcher::_regEncode[dst_first] < 8) {
1411 if (Matcher::_regEncode[src_first] >= 8) {
1412 emit_opcode(*cbuf, Assembler::REX_B);
1413 }
1414 } else {
1415 if (Matcher::_regEncode[src_first] < 8) {
1416 emit_opcode(*cbuf, Assembler::REX_R);
1417 } else {
1418 emit_opcode(*cbuf, Assembler::REX_RB);
1419 }
1420 }
1421 emit_opcode(*cbuf, 0x8B);
1422 emit_rm(*cbuf, 0x3,
1423 Matcher::_regEncode[dst_first] & 7,
1424 Matcher::_regEncode[src_first] & 7);
1425 #ifndef PRODUCT
1426 } else if (!do_size) {
1427 st->print("movl %s, %s\t# spill",
1428 Matcher::regName[dst_first],
1429 Matcher::regName[src_first]);
1430 #endif
1431 }
1432 return
1433 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1434 ? 2
1435 : 3; // REX
1436 }
1437 } else if (dst_first_rc == rc_float) {
1438 // gpr -> xmm
1439 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1440 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1441 // 64-bit
1442 if (cbuf) {
1443 emit_opcode(*cbuf, 0x66);
1444 if (Matcher::_regEncode[dst_first] < 8) {
1445 if (Matcher::_regEncode[src_first] < 8) {
1446 emit_opcode(*cbuf, Assembler::REX_W);
1447 } else {
1448 emit_opcode(*cbuf, Assembler::REX_WB);
1449 }
1450 } else {
1451 if (Matcher::_regEncode[src_first] < 8) {
1452 emit_opcode(*cbuf, Assembler::REX_WR);
1453 } else {
1454 emit_opcode(*cbuf, Assembler::REX_WRB);
1455 }
1456 }
1457 emit_opcode(*cbuf, 0x0F);
1458 emit_opcode(*cbuf, 0x6E);
1459 emit_rm(*cbuf, 0x3,
1460 Matcher::_regEncode[dst_first] & 7,
1461 Matcher::_regEncode[src_first] & 7);
1462 #ifndef PRODUCT
1463 } else if (!do_size) {
1464 st->print("movdq %s, %s\t# spill",
1465 Matcher::regName[dst_first],
1466 Matcher::regName[src_first]);
1467 #endif
1468 }
1469 return 5; // REX
1470 } else {
1471 // 32-bit
1472 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1473 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1474 if (cbuf) {
1475 emit_opcode(*cbuf, 0x66);
1476 if (Matcher::_regEncode[dst_first] < 8) {
1477 if (Matcher::_regEncode[src_first] >= 8) {
1478 emit_opcode(*cbuf, Assembler::REX_B);
1479 }
1480 } else {
1481 if (Matcher::_regEncode[src_first] < 8) {
1482 emit_opcode(*cbuf, Assembler::REX_R);
1483 } else {
1484 emit_opcode(*cbuf, Assembler::REX_RB);
1485 }
1486 }
1487 emit_opcode(*cbuf, 0x0F);
1488 emit_opcode(*cbuf, 0x6E);
1489 emit_rm(*cbuf, 0x3,
1490 Matcher::_regEncode[dst_first] & 7,
1491 Matcher::_regEncode[src_first] & 7);
1492 #ifndef PRODUCT
1493 } else if (!do_size) {
1494 st->print("movdl %s, %s\t# spill",
1495 Matcher::regName[dst_first],
1496 Matcher::regName[src_first]);
1497 #endif
1498 }
1499 return
1500 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1501 ? 4
1502 : 5; // REX
1503 }
1504 }
1505 } else if (src_first_rc == rc_float) {
1506 // xmm ->
1507 if (dst_first_rc == rc_stack) {
1508 // xmm -> mem
1509 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1510 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1511 // 64-bit
1512 int offset = ra_->reg2offset(dst_first);
1513 if (cbuf) {
1514 emit_opcode(*cbuf, 0xF2);
1515 if (Matcher::_regEncode[src_first] >= 8) {
1516 emit_opcode(*cbuf, Assembler::REX_R);
1517 }
1518 emit_opcode(*cbuf, 0x0F);
1519 emit_opcode(*cbuf, 0x11);
1520 encode_RegMem(*cbuf,
1521 Matcher::_regEncode[src_first],
1522 RSP_enc, 0x4, 0, offset,
1523 false);
1524 #ifndef PRODUCT
1525 } else if (!do_size) {
1526 st->print("movsd [rsp + #%d], %s\t# spill",
1527 offset,
1528 Matcher::regName[src_first]);
1529 #endif
1530 }
1531 return
1532 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1533 ((Matcher::_regEncode[src_first] < 8)
1534 ? 5
1535 : 6); // REX
1536 } else {
1537 // 32-bit
1538 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1539 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1540 int offset = ra_->reg2offset(dst_first);
1541 if (cbuf) {
1542 emit_opcode(*cbuf, 0xF3);
1543 if (Matcher::_regEncode[src_first] >= 8) {
1544 emit_opcode(*cbuf, Assembler::REX_R);
1545 }
1546 emit_opcode(*cbuf, 0x0F);
1547 emit_opcode(*cbuf, 0x11);
1548 encode_RegMem(*cbuf,
1549 Matcher::_regEncode[src_first],
1550 RSP_enc, 0x4, 0, offset,
1551 false);
1552 #ifndef PRODUCT
1553 } else if (!do_size) {
1554 st->print("movss [rsp + #%d], %s\t# spill",
1555 offset,
1556 Matcher::regName[src_first]);
1557 #endif
1558 }
1559 return
1560 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1561 ((Matcher::_regEncode[src_first] < 8)
1562 ? 5
1563 : 6); // REX
1564 }
1565 } else if (dst_first_rc == rc_int) {
1566 // xmm -> gpr
1567 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1568 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1569 // 64-bit
1570 if (cbuf) {
1571 emit_opcode(*cbuf, 0x66);
1572 if (Matcher::_regEncode[dst_first] < 8) {
1573 if (Matcher::_regEncode[src_first] < 8) {
1574 emit_opcode(*cbuf, Assembler::REX_W);
1575 } else {
1576 emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1577 }
1578 } else {
1579 if (Matcher::_regEncode[src_first] < 8) {
1580 emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1581 } else {
1582 emit_opcode(*cbuf, Assembler::REX_WRB);
1583 }
1584 }
1585 emit_opcode(*cbuf, 0x0F);
1586 emit_opcode(*cbuf, 0x7E);
1587 emit_rm(*cbuf, 0x3,
1588 Matcher::_regEncode[src_first] & 7,
1589 Matcher::_regEncode[dst_first] & 7);
1590 #ifndef PRODUCT
1591 } else if (!do_size) {
1592 st->print("movdq %s, %s\t# spill",
1593 Matcher::regName[dst_first],
1594 Matcher::regName[src_first]);
1595 #endif
1596 }
1597 return 5; // REX
1598 } else {
1599 // 32-bit
1600 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1601 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1602 if (cbuf) {
1603 emit_opcode(*cbuf, 0x66);
1604 if (Matcher::_regEncode[dst_first] < 8) {
1605 if (Matcher::_regEncode[src_first] >= 8) {
1606 emit_opcode(*cbuf, Assembler::REX_R); // attention!
1607 }
1608 } else {
1609 if (Matcher::_regEncode[src_first] < 8) {
1610 emit_opcode(*cbuf, Assembler::REX_B); // attention!
1611 } else {
1612 emit_opcode(*cbuf, Assembler::REX_RB);
1613 }
1614 }
1615 emit_opcode(*cbuf, 0x0F);
1616 emit_opcode(*cbuf, 0x7E);
1617 emit_rm(*cbuf, 0x3,
1618 Matcher::_regEncode[src_first] & 7,
1619 Matcher::_regEncode[dst_first] & 7);
1620 #ifndef PRODUCT
1621 } else if (!do_size) {
1622 st->print("movdl %s, %s\t# spill",
1623 Matcher::regName[dst_first],
1624 Matcher::regName[src_first]);
1625 #endif
1626 }
1627 return
1628 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1629 ? 4
1630 : 5; // REX
1631 }
1632 } else if (dst_first_rc == rc_float) {
1633 // xmm -> xmm
1634 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1635 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1636 // 64-bit
1637 if (cbuf) {
1638 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1639 if (Matcher::_regEncode[dst_first] < 8) {
1640 if (Matcher::_regEncode[src_first] >= 8) {
1641 emit_opcode(*cbuf, Assembler::REX_B);
1642 }
1643 } else {
1644 if (Matcher::_regEncode[src_first] < 8) {
1645 emit_opcode(*cbuf, Assembler::REX_R);
1646 } else {
1647 emit_opcode(*cbuf, Assembler::REX_RB);
1648 }
1649 }
1650 emit_opcode(*cbuf, 0x0F);
1651 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1652 emit_rm(*cbuf, 0x3,
1653 Matcher::_regEncode[dst_first] & 7,
1654 Matcher::_regEncode[src_first] & 7);
1655 #ifndef PRODUCT
1656 } else if (!do_size) {
1657 st->print("%s %s, %s\t# spill",
1658 UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1659 Matcher::regName[dst_first],
1660 Matcher::regName[src_first]);
1661 #endif
1662 }
1663 return
1664 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1665 ? 4
1666 : 5; // REX
1667 } else {
1668 // 32-bit
1669 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1670 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1671 if (cbuf) {
1672 if (!UseXmmRegToRegMoveAll)
1673 emit_opcode(*cbuf, 0xF3);
1674 if (Matcher::_regEncode[dst_first] < 8) {
1675 if (Matcher::_regEncode[src_first] >= 8) {
1676 emit_opcode(*cbuf, Assembler::REX_B);
1677 }
1678 } else {
1679 if (Matcher::_regEncode[src_first] < 8) {
1680 emit_opcode(*cbuf, Assembler::REX_R);
1681 } else {
1682 emit_opcode(*cbuf, Assembler::REX_RB);
1683 }
1684 }
1685 emit_opcode(*cbuf, 0x0F);
1686 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1687 emit_rm(*cbuf, 0x3,
1688 Matcher::_regEncode[dst_first] & 7,
1689 Matcher::_regEncode[src_first] & 7);
1690 #ifndef PRODUCT
1691 } else if (!do_size) {
1692 st->print("%s %s, %s\t# spill",
1693 UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1694 Matcher::regName[dst_first],
1695 Matcher::regName[src_first]);
1696 #endif
1697 }
1698 return
1699 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1700 ? (UseXmmRegToRegMoveAll ? 3 : 4)
1701 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1702 }
1703 }
1704 }
1705
1706 assert(0," foo ");
1707 Unimplemented();
1708
1709 return 0;
1710 }
1711
1712 #ifndef PRODUCT
1713 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1714 {
1715 implementation(NULL, ra_, false, st);
1716 }
1717 #endif
1718
1719 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1720 {
1721 implementation(&cbuf, ra_, false, NULL);
1722 }
1723
1724 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1725 {
1726 return implementation(NULL, ra_, true, NULL);
1727 }
1728
1729 //=============================================================================
1730 #ifndef PRODUCT
1731 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1732 {
1733 st->print("nop \t# %d bytes pad for loops and calls", _count);
1734 }
1735 #endif
1736
1737 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1738 {
1739 MacroAssembler _masm(&cbuf);
1740 __ nop(_count);
1741 }
1742
1743 uint MachNopNode::size(PhaseRegAlloc*) const
1744 {
1745 return _count;
1746 }
1747
1748
1749 //=============================================================================
1750 #ifndef PRODUCT
1751 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1752 {
1753 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1754 int reg = ra_->get_reg_first(this);
1755 st->print("leaq %s, [rsp + #%d]\t# box lock",
1756 Matcher::regName[reg], offset);
1757 }
1758 #endif
1759
1760 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1761 {
1762 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1763 int reg = ra_->get_encode(this);
1764 if (offset >= 0x80) {
1765 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1766 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1767 emit_rm(cbuf, 0x2, reg & 7, 0x04);
1768 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1769 emit_d32(cbuf, offset);
1770 } else {
1771 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1772 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1773 emit_rm(cbuf, 0x1, reg & 7, 0x04);
1774 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1775 emit_d8(cbuf, offset);
1776 }
1777 }
1778
1779 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1780 {
1781 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1782 return (offset < 0x80) ? 5 : 8; // REX
1783 }
1784
1785 //=============================================================================
1786
1787 // emit call stub, compiled java to interpreter
1788 void emit_java_to_interp(CodeBuffer& cbuf)
1789 {
1790 // Stub is fixed up when the corresponding call is converted from
1791 // calling compiled code to calling interpreted code.
1792 // movq rbx, 0
1793 // jmp -5 # to self
1794
1795 address mark = cbuf.insts_mark(); // get mark within main instrs section
1796
1797 // Note that the code buffer's insts_mark is always relative to insts.
1798 // That's why we must use the macroassembler to generate a stub.
1799 MacroAssembler _masm(&cbuf);
1800
1801 address base =
1802 __ start_a_stub(Compile::MAX_stubs_size);
1803 if (base == NULL) return; // CodeBuffer::expand failed
1804 // static stub relocation stores the instruction address of the call
1805 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1806 // static stub relocation also tags the methodOop in the code-stream.
1807 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time
1808 // This is recognized as unresolved by relocs/nativeinst/ic code
1809 __ jump(RuntimeAddress(__ pc()));
1810
1811 // Update current stubs pointer and restore insts_end.
1812 __ end_a_stub();
1813 }
1814
1815 // size of call stub, compiled java to interpretor
1816 uint size_java_to_interp()
1817 {
1818 return 15; // movq (1+1+8); jmp (1+4)
1819 }
1820
1821 // relocation entries for call stub, compiled java to interpretor
1822 uint reloc_java_to_interp()
1823 {
1824 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1825 }
1826
1827 //=============================================================================
1828 #ifndef PRODUCT
1829 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1830 {
1831 if (UseCompressedOops) {
1832 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t# compressed klass");
1833 if (Universe::narrow_oop_shift() != 0) {
1834 st->print_cr("\tdecode_heap_oop_not_null rscratch1, rscratch1");
1835 }
1836 st->print_cr("\tcmpq rax, rscratch1\t # Inline cache check");
1837 } else {
1838 st->print_cr("\tcmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t"
1839 "# Inline cache check");
1840 }
1841 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1842 st->print_cr("\tnop\t# nops to align entry point");
1843 }
1844 #endif
1845
1846 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1847 {
1848 MacroAssembler masm(&cbuf);
1849 uint insts_size = cbuf.insts_size();
1850 if (UseCompressedOops) {
1851 masm.load_klass(rscratch1, j_rarg0);
1852 masm.cmpptr(rax, rscratch1);
1853 } else {
1854 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1855 }
1856
1857 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1858
1859 /* WARNING these NOPs are critical so that verified entry point is properly
1860 4 bytes aligned for patching by NativeJump::patch_verified_entry() */
1861 int nops_cnt = 4 - ((cbuf.insts_size() - insts_size) & 0x3);
1862 if (OptoBreakpoint) {
1863 // Leave space for int3
1864 nops_cnt -= 1;
1865 }
1866 nops_cnt &= 0x3; // Do not add nops if code is aligned.
1867 if (nops_cnt > 0)
1868 masm.nop(nops_cnt);
1869 }
1870
1871 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1872 {
1873 return MachNode::size(ra_); // too many variables; just compute it
1874 // the hard way
1875 }
1876
1877
1878 //=============================================================================
1879 uint size_exception_handler()
1880 {
1881 // NativeCall instruction size is the same as NativeJump.
1882 // Note that this value is also credited (in output.cpp) to
1883 // the size of the code section.
1884 return NativeJump::instruction_size;
1885 }
1886
1887 // Emit exception handler code.
1888 int emit_exception_handler(CodeBuffer& cbuf)
1889 {
1890
1891 // Note that the code buffer's insts_mark is always relative to insts.
1892 // That's why we must use the macroassembler to generate a handler.
1893 MacroAssembler _masm(&cbuf);
1894 address base =
1895 __ start_a_stub(size_exception_handler());
1896 if (base == NULL) return 0; // CodeBuffer::expand failed
1897 int offset = __ offset();
1898 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
1899 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1900 __ end_a_stub();
1901 return offset;
1902 }
1903
1904 uint size_deopt_handler()
1905 {
1906 // three 5 byte instructions
1907 return 15;
1908 }
1909
1910 // Emit deopt handler code.
1911 int emit_deopt_handler(CodeBuffer& cbuf)
1912 {
1913
1914 // Note that the code buffer's insts_mark is always relative to insts.
1915 // That's why we must use the macroassembler to generate a handler.
1916 MacroAssembler _masm(&cbuf);
1917 address base =
1918 __ start_a_stub(size_deopt_handler());
1919 if (base == NULL) return 0; // CodeBuffer::expand failed
1920 int offset = __ offset();
1921 address the_pc = (address) __ pc();
1922 Label next;
1923 // push a "the_pc" on the stack without destroying any registers
1924 // as they all may be live.
1925
1926 // push address of "next"
1927 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1928 __ bind(next);
1929 // adjust it so it matches "the_pc"
1930 __ subptr(Address(rsp, 0), __ offset() - offset);
1931 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1932 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1933 __ end_a_stub();
1934 return offset;
1935 }
1936
1937
1938 const bool Matcher::match_rule_supported(int opcode) {
1939 if (!has_match_rule(opcode))
1940 return false;
1941
1942 return true; // Per default match rules are supported.
1943 }
1944
1945 int Matcher::regnum_to_fpu_offset(int regnum)
1946 {
1947 return regnum - 32; // The FP registers are in the second chunk
1948 }
1949
1950 // This is UltraSparc specific, true just means we have fast l2f conversion
1951 const bool Matcher::convL2FSupported(void) {
1952 return true;
1953 }
1954
1955 // Vector width in bytes
1956 const uint Matcher::vector_width_in_bytes(void) {
1957 return 8;
1958 }
1959
1960 // Vector ideal reg
1961 const uint Matcher::vector_ideal_reg(void) {
1962 return Op_RegD;
1963 }
1964
1965 // Is this branch offset short enough that a short branch can be used?
1966 //
1967 // NOTE: If the platform does not provide any short branch variants, then
1968 // this method should return false for offset 0.
1969 bool Matcher::is_short_branch_offset(int rule, int offset) {
1970 // the short version of jmpConUCF2 contains multiple branches,
1971 // making the reach slightly less
1972 if (rule == jmpConUCF2_rule)
1973 return (-126 <= offset && offset <= 125);
1974 return (-128 <= offset && offset <= 127);
1975 }
1976
1977 const bool Matcher::isSimpleConstant64(jlong value) {
1978 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1979 //return value == (int) value; // Cf. storeImmL and immL32.
1980
1981 // Probably always true, even if a temp register is required.
1982 return true;
1983 }
1984
1985 // The ecx parameter to rep stosq for the ClearArray node is in words.
1986 const bool Matcher::init_array_count_is_in_bytes = false;
1987
1988 // Threshold size for cleararray.
1989 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1990
1991 // Should the Matcher clone shifts on addressing modes, expecting them
1992 // to be subsumed into complex addressing expressions or compute them
1993 // into registers? True for Intel but false for most RISCs
1994 const bool Matcher::clone_shift_expressions = true;
1995
1996 // Do we need to mask the count passed to shift instructions or does
1997 // the cpu only look at the lower 5/6 bits anyway?
1998 const bool Matcher::need_masked_shift_count = false;
1999
2000 bool Matcher::narrow_oop_use_complex_address() {
2001 assert(UseCompressedOops, "only for compressed oops code");
2002 return (LogMinObjAlignmentInBytes <= 3);
2003 }
2004
2005 // Is it better to copy float constants, or load them directly from
2006 // memory? Intel can load a float constant from a direct address,
2007 // requiring no extra registers. Most RISCs will have to materialize
2008 // an address into a register first, so they would do better to copy
2009 // the constant from stack.
2010 const bool Matcher::rematerialize_float_constants = true; // XXX
2011
2012 // If CPU can load and store mis-aligned doubles directly then no
2013 // fixup is needed. Else we split the double into 2 integer pieces
2014 // and move it piece-by-piece. Only happens when passing doubles into
2015 // C code as the Java calling convention forces doubles to be aligned.
2016 const bool Matcher::misaligned_doubles_ok = true;
2017
2018 // No-op on amd64
2019 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2020
2021 // Advertise here if the CPU requires explicit rounding operations to
2022 // implement the UseStrictFP mode.
2023 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2024
2025 // Are floats conerted to double when stored to stack during deoptimization?
2026 // On x64 it is stored without convertion so we can use normal access.
2027 bool Matcher::float_in_double() { return false; }
2028
2029 // Do ints take an entire long register or just half?
2030 const bool Matcher::int_in_long = true;
2031
2032 // Return whether or not this register is ever used as an argument.
2033 // This function is used on startup to build the trampoline stubs in
2034 // generateOptoStub. Registers not mentioned will be killed by the VM
2035 // call in the trampoline, and arguments in those registers not be
2036 // available to the callee.
2037 bool Matcher::can_be_java_arg(int reg)
2038 {
2039 return
2040 reg == RDI_num || reg == RDI_H_num ||
2041 reg == RSI_num || reg == RSI_H_num ||
2042 reg == RDX_num || reg == RDX_H_num ||
2043 reg == RCX_num || reg == RCX_H_num ||
2044 reg == R8_num || reg == R8_H_num ||
2045 reg == R9_num || reg == R9_H_num ||
2046 reg == R12_num || reg == R12_H_num ||
2047 reg == XMM0_num || reg == XMM0_H_num ||
2048 reg == XMM1_num || reg == XMM1_H_num ||
2049 reg == XMM2_num || reg == XMM2_H_num ||
2050 reg == XMM3_num || reg == XMM3_H_num ||
2051 reg == XMM4_num || reg == XMM4_H_num ||
2052 reg == XMM5_num || reg == XMM5_H_num ||
2053 reg == XMM6_num || reg == XMM6_H_num ||
2054 reg == XMM7_num || reg == XMM7_H_num;
2055 }
2056
2057 bool Matcher::is_spillable_arg(int reg)
2058 {
2059 return can_be_java_arg(reg);
2060 }
2061
2062 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
2063 // In 64 bit mode a code which use multiply when
2064 // devisor is constant is faster than hardware
2065 // DIV instruction (it uses MulHiL).
2066 return false;
2067 }
2068
2069 // Register for DIVI projection of divmodI
2070 RegMask Matcher::divI_proj_mask() {
2071 return INT_RAX_REG_mask;
2072 }
2073
2074 // Register for MODI projection of divmodI
2075 RegMask Matcher::modI_proj_mask() {
2076 return INT_RDX_REG_mask;
2077 }
2078
2079 // Register for DIVL projection of divmodL
2080 RegMask Matcher::divL_proj_mask() {
2081 return LONG_RAX_REG_mask;
2082 }
2083
2084 // Register for MODL projection of divmodL
2085 RegMask Matcher::modL_proj_mask() {
2086 return LONG_RDX_REG_mask;
2087 }
2088
2089 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2090 return PTR_RBP_REG_mask;
2091 }
2092
2093 static Address build_address(int b, int i, int s, int d) {
2094 Register index = as_Register(i);
2095 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2096 if (index == rsp) {
2097 index = noreg;
2098 scale = Address::no_scale;
2099 }
2100 Address addr(as_Register(b), index, scale, d);
2101 return addr;
2102 }
2103
2104 %}
2105
2106 //----------ENCODING BLOCK-----------------------------------------------------
2107 // This block specifies the encoding classes used by the compiler to
2108 // output byte streams. Encoding classes are parameterized macros
2109 // used by Machine Instruction Nodes in order to generate the bit
2110 // encoding of the instruction. Operands specify their base encoding
2111 // interface with the interface keyword. There are currently
2112 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2113 // COND_INTER. REG_INTER causes an operand to generate a function
2114 // which returns its register number when queried. CONST_INTER causes
2115 // an operand to generate a function which returns the value of the
2116 // constant when queried. MEMORY_INTER causes an operand to generate
2117 // four functions which return the Base Register, the Index Register,
2118 // the Scale Value, and the Offset Value of the operand when queried.
2119 // COND_INTER causes an operand to generate six functions which return
2120 // the encoding code (ie - encoding bits for the instruction)
2121 // associated with each basic boolean condition for a conditional
2122 // instruction.
2123 //
2124 // Instructions specify two basic values for encoding. Again, a
2125 // function is available to check if the constant displacement is an
2126 // oop. They use the ins_encode keyword to specify their encoding
2127 // classes (which must be a sequence of enc_class names, and their
2128 // parameters, specified in the encoding block), and they use the
2129 // opcode keyword to specify, in order, their primary, secondary, and
2130 // tertiary opcode. Only the opcode sections which a particular
2131 // instruction needs for encoding need to be specified.
2132 encode %{
2133 // Build emit functions for each basic byte or larger field in the
2134 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2135 // from C++ code in the enc_class source block. Emit functions will
2136 // live in the main source block for now. In future, we can
2137 // generalize this by adding a syntax that specifies the sizes of
2138 // fields in an order, so that the adlc can build the emit functions
2139 // automagically
2140
2141 // Emit primary opcode
2142 enc_class OpcP
2143 %{
2144 emit_opcode(cbuf, $primary);
2145 %}
2146
2147 // Emit secondary opcode
2148 enc_class OpcS
2149 %{
2150 emit_opcode(cbuf, $secondary);
2151 %}
2152
2153 // Emit tertiary opcode
2154 enc_class OpcT
2155 %{
2156 emit_opcode(cbuf, $tertiary);
2157 %}
2158
2159 // Emit opcode directly
2160 enc_class Opcode(immI d8)
2161 %{
2162 emit_opcode(cbuf, $d8$$constant);
2163 %}
2164
2165 // Emit size prefix
2166 enc_class SizePrefix
2167 %{
2168 emit_opcode(cbuf, 0x66);
2169 %}
2170
2171 enc_class reg(rRegI reg)
2172 %{
2173 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2174 %}
2175
2176 enc_class reg_reg(rRegI dst, rRegI src)
2177 %{
2178 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2179 %}
2180
2181 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2182 %{
2183 emit_opcode(cbuf, $opcode$$constant);
2184 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2185 %}
2186
2187 enc_class cmpfp_fixup() %{
2188 MacroAssembler _masm(&cbuf);
2189 emit_cmpfp_fixup(_masm);
2190 %}
2191
2192 enc_class cmpfp3(rRegI dst)
2193 %{
2194 int dstenc = $dst$$reg;
2195
2196 // movl $dst, -1
2197 if (dstenc >= 8) {
2198 emit_opcode(cbuf, Assembler::REX_B);
2199 }
2200 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2201 emit_d32(cbuf, -1);
2202
2203 // jp,s done
2204 emit_opcode(cbuf, 0x7A);
2205 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2206
2207 // jb,s done
2208 emit_opcode(cbuf, 0x72);
2209 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2210
2211 // setne $dst
2212 if (dstenc >= 4) {
2213 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2214 }
2215 emit_opcode(cbuf, 0x0F);
2216 emit_opcode(cbuf, 0x95);
2217 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2218
2219 // movzbl $dst, $dst
2220 if (dstenc >= 4) {
2221 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2222 }
2223 emit_opcode(cbuf, 0x0F);
2224 emit_opcode(cbuf, 0xB6);
2225 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2226 %}
2227
2228 enc_class cdql_enc(no_rax_rdx_RegI div)
2229 %{
2230 // Full implementation of Java idiv and irem; checks for
2231 // special case as described in JVM spec., p.243 & p.271.
2232 //
2233 // normal case special case
2234 //
2235 // input : rax: dividend min_int
2236 // reg: divisor -1
2237 //
2238 // output: rax: quotient (= rax idiv reg) min_int
2239 // rdx: remainder (= rax irem reg) 0
2240 //
2241 // Code sequnce:
2242 //
2243 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2244 // 5: 75 07/08 jne e <normal>
2245 // 7: 33 d2 xor %edx,%edx
2246 // [div >= 8 -> offset + 1]
2247 // [REX_B]
2248 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2249 // c: 74 03/04 je 11 <done>
2250 // 000000000000000e <normal>:
2251 // e: 99 cltd
2252 // [div >= 8 -> offset + 1]
2253 // [REX_B]
2254 // f: f7 f9 idiv $div
2255 // 0000000000000011 <done>:
2256
2257 // cmp $0x80000000,%eax
2258 emit_opcode(cbuf, 0x3d);
2259 emit_d8(cbuf, 0x00);
2260 emit_d8(cbuf, 0x00);
2261 emit_d8(cbuf, 0x00);
2262 emit_d8(cbuf, 0x80);
2263
2264 // jne e <normal>
2265 emit_opcode(cbuf, 0x75);
2266 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2267
2268 // xor %edx,%edx
2269 emit_opcode(cbuf, 0x33);
2270 emit_d8(cbuf, 0xD2);
2271
2272 // cmp $0xffffffffffffffff,%ecx
2273 if ($div$$reg >= 8) {
2274 emit_opcode(cbuf, Assembler::REX_B);
2275 }
2276 emit_opcode(cbuf, 0x83);
2277 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2278 emit_d8(cbuf, 0xFF);
2279
2280 // je 11 <done>
2281 emit_opcode(cbuf, 0x74);
2282 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2283
2284 // <normal>
2285 // cltd
2286 emit_opcode(cbuf, 0x99);
2287
2288 // idivl (note: must be emitted by the user of this rule)
2289 // <done>
2290 %}
2291
2292 enc_class cdqq_enc(no_rax_rdx_RegL div)
2293 %{
2294 // Full implementation of Java ldiv and lrem; checks for
2295 // special case as described in JVM spec., p.243 & p.271.
2296 //
2297 // normal case special case
2298 //
2299 // input : rax: dividend min_long
2300 // reg: divisor -1
2301 //
2302 // output: rax: quotient (= rax idiv reg) min_long
2303 // rdx: remainder (= rax irem reg) 0
2304 //
2305 // Code sequnce:
2306 //
2307 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2308 // 7: 00 00 80
2309 // a: 48 39 d0 cmp %rdx,%rax
2310 // d: 75 08 jne 17 <normal>
2311 // f: 33 d2 xor %edx,%edx
2312 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2313 // 15: 74 05 je 1c <done>
2314 // 0000000000000017 <normal>:
2315 // 17: 48 99 cqto
2316 // 19: 48 f7 f9 idiv $div
2317 // 000000000000001c <done>:
2318
2319 // mov $0x8000000000000000,%rdx
2320 emit_opcode(cbuf, Assembler::REX_W);
2321 emit_opcode(cbuf, 0xBA);
2322 emit_d8(cbuf, 0x00);
2323 emit_d8(cbuf, 0x00);
2324 emit_d8(cbuf, 0x00);
2325 emit_d8(cbuf, 0x00);
2326 emit_d8(cbuf, 0x00);
2327 emit_d8(cbuf, 0x00);
2328 emit_d8(cbuf, 0x00);
2329 emit_d8(cbuf, 0x80);
2330
2331 // cmp %rdx,%rax
2332 emit_opcode(cbuf, Assembler::REX_W);
2333 emit_opcode(cbuf, 0x39);
2334 emit_d8(cbuf, 0xD0);
2335
2336 // jne 17 <normal>
2337 emit_opcode(cbuf, 0x75);
2338 emit_d8(cbuf, 0x08);
2339
2340 // xor %edx,%edx
2341 emit_opcode(cbuf, 0x33);
2342 emit_d8(cbuf, 0xD2);
2343
2344 // cmp $0xffffffffffffffff,$div
2345 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2346 emit_opcode(cbuf, 0x83);
2347 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2348 emit_d8(cbuf, 0xFF);
2349
2350 // je 1e <done>
2351 emit_opcode(cbuf, 0x74);
2352 emit_d8(cbuf, 0x05);
2353
2354 // <normal>
2355 // cqto
2356 emit_opcode(cbuf, Assembler::REX_W);
2357 emit_opcode(cbuf, 0x99);
2358
2359 // idivq (note: must be emitted by the user of this rule)
2360 // <done>
2361 %}
2362
2363 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2364 enc_class OpcSE(immI imm)
2365 %{
2366 // Emit primary opcode and set sign-extend bit
2367 // Check for 8-bit immediate, and set sign extend bit in opcode
2368 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2369 emit_opcode(cbuf, $primary | 0x02);
2370 } else {
2371 // 32-bit immediate
2372 emit_opcode(cbuf, $primary);
2373 }
2374 %}
2375
2376 enc_class OpcSErm(rRegI dst, immI imm)
2377 %{
2378 // OpcSEr/m
2379 int dstenc = $dst$$reg;
2380 if (dstenc >= 8) {
2381 emit_opcode(cbuf, Assembler::REX_B);
2382 dstenc -= 8;
2383 }
2384 // Emit primary opcode and set sign-extend bit
2385 // Check for 8-bit immediate, and set sign extend bit in opcode
2386 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2387 emit_opcode(cbuf, $primary | 0x02);
2388 } else {
2389 // 32-bit immediate
2390 emit_opcode(cbuf, $primary);
2391 }
2392 // Emit r/m byte with secondary opcode, after primary opcode.
2393 emit_rm(cbuf, 0x3, $secondary, dstenc);
2394 %}
2395
2396 enc_class OpcSErm_wide(rRegL dst, immI imm)
2397 %{
2398 // OpcSEr/m
2399 int dstenc = $dst$$reg;
2400 if (dstenc < 8) {
2401 emit_opcode(cbuf, Assembler::REX_W);
2402 } else {
2403 emit_opcode(cbuf, Assembler::REX_WB);
2404 dstenc -= 8;
2405 }
2406 // Emit primary opcode and set sign-extend bit
2407 // Check for 8-bit immediate, and set sign extend bit in opcode
2408 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2409 emit_opcode(cbuf, $primary | 0x02);
2410 } else {
2411 // 32-bit immediate
2412 emit_opcode(cbuf, $primary);
2413 }
2414 // Emit r/m byte with secondary opcode, after primary opcode.
2415 emit_rm(cbuf, 0x3, $secondary, dstenc);
2416 %}
2417
2418 enc_class Con8or32(immI imm)
2419 %{
2420 // Check for 8-bit immediate, and set sign extend bit in opcode
2421 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2422 $$$emit8$imm$$constant;
2423 } else {
2424 // 32-bit immediate
2425 $$$emit32$imm$$constant;
2426 }
2427 %}
2428
2429 enc_class Lbl(label labl)
2430 %{
2431 // GOTO
2432 Label* l = $labl$$label;
2433 emit_d32(cbuf, (l->loc_pos() - (cbuf.insts_size() + 4)));
2434 %}
2435
2436 enc_class LblShort(label labl)
2437 %{
2438 // GOTO
2439 Label* l = $labl$$label;
2440 int disp = l->loc_pos() - (cbuf.insts_size() + 1);
2441 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2442 emit_d8(cbuf, disp);
2443 %}
2444
2445 enc_class opc2_reg(rRegI dst)
2446 %{
2447 // BSWAP
2448 emit_cc(cbuf, $secondary, $dst$$reg);
2449 %}
2450
2451 enc_class opc3_reg(rRegI dst)
2452 %{
2453 // BSWAP
2454 emit_cc(cbuf, $tertiary, $dst$$reg);
2455 %}
2456
2457 enc_class reg_opc(rRegI div)
2458 %{
2459 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2460 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2461 %}
2462
2463 enc_class Jcc(cmpOp cop, label labl)
2464 %{
2465 // JCC
2466 Label* l = $labl$$label;
2467 $$$emit8$primary;
2468 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2469 emit_d32(cbuf, (l->loc_pos() - (cbuf.insts_size() + 4)));
2470 %}
2471
2472 enc_class JccShort (cmpOp cop, label labl)
2473 %{
2474 // JCC
2475 Label *l = $labl$$label;
2476 emit_cc(cbuf, $primary, $cop$$cmpcode);
2477 int disp = l->loc_pos() - (cbuf.insts_size() + 1);
2478 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2479 emit_d8(cbuf, disp);
2480 %}
2481
2482 enc_class enc_cmov(cmpOp cop)
2483 %{
2484 // CMOV
2485 $$$emit8$primary;
2486 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2487 %}
2488
2489 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2490 %{
2491 // Invert sense of branch from sense of cmov
2492 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2493 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2494 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2495 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2496 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2497 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2498 if ($dst$$reg < 8) {
2499 if ($src$$reg >= 8) {
2500 emit_opcode(cbuf, Assembler::REX_B);
2501 }
2502 } else {
2503 if ($src$$reg < 8) {
2504 emit_opcode(cbuf, Assembler::REX_R);
2505 } else {
2506 emit_opcode(cbuf, Assembler::REX_RB);
2507 }
2508 }
2509 emit_opcode(cbuf, 0x0F);
2510 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2511 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2512 %}
2513
2514 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2515 %{
2516 // Invert sense of branch from sense of cmov
2517 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2518 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2519
2520 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2521 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2522 if ($dst$$reg < 8) {
2523 if ($src$$reg >= 8) {
2524 emit_opcode(cbuf, Assembler::REX_B);
2525 }
2526 } else {
2527 if ($src$$reg < 8) {
2528 emit_opcode(cbuf, Assembler::REX_R);
2529 } else {
2530 emit_opcode(cbuf, Assembler::REX_RB);
2531 }
2532 }
2533 emit_opcode(cbuf, 0x0F);
2534 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2535 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2536 %}
2537
2538 enc_class enc_PartialSubtypeCheck()
2539 %{
2540 Register Rrdi = as_Register(RDI_enc); // result register
2541 Register Rrax = as_Register(RAX_enc); // super class
2542 Register Rrcx = as_Register(RCX_enc); // killed
2543 Register Rrsi = as_Register(RSI_enc); // sub class
2544 Label miss;
2545 const bool set_cond_codes = true;
2546
2547 MacroAssembler _masm(&cbuf);
2548 __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi,
2549 NULL, &miss,
2550 /*set_cond_codes:*/ true);
2551 if ($primary) {
2552 __ xorptr(Rrdi, Rrdi);
2553 }
2554 __ bind(miss);
2555 %}
2556
2557 enc_class Java_To_Interpreter(method meth)
2558 %{
2559 // CALL Java_To_Interpreter
2560 // This is the instruction starting address for relocation info.
2561 cbuf.set_insts_mark();
2562 $$$emit8$primary;
2563 // CALL directly to the runtime
2564 emit_d32_reloc(cbuf,
2565 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2566 runtime_call_Relocation::spec(),
2567 RELOC_DISP32);
2568 %}
2569
2570 enc_class preserve_SP %{
2571 debug_only(int off0 = cbuf.insts_size());
2572 MacroAssembler _masm(&cbuf);
2573 // RBP is preserved across all calls, even compiled calls.
2574 // Use it to preserve RSP in places where the callee might change the SP.
2575 __ movptr(rbp_mh_SP_save, rsp);
2576 debug_only(int off1 = cbuf.insts_size());
2577 assert(off1 - off0 == preserve_SP_size(), "correct size prediction");
2578 %}
2579
2580 enc_class restore_SP %{
2581 MacroAssembler _masm(&cbuf);
2582 __ movptr(rsp, rbp_mh_SP_save);
2583 %}
2584
2585 enc_class Java_Static_Call(method meth)
2586 %{
2587 // JAVA STATIC CALL
2588 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2589 // determine who we intended to call.
2590 cbuf.set_insts_mark();
2591 $$$emit8$primary;
2592
2593 if (!_method) {
2594 emit_d32_reloc(cbuf,
2595 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2596 runtime_call_Relocation::spec(),
2597 RELOC_DISP32);
2598 } else if (_optimized_virtual) {
2599 emit_d32_reloc(cbuf,
2600 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2601 opt_virtual_call_Relocation::spec(),
2602 RELOC_DISP32);
2603 } else {
2604 emit_d32_reloc(cbuf,
2605 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2606 static_call_Relocation::spec(),
2607 RELOC_DISP32);
2608 }
2609 if (_method) {
2610 // Emit stub for static call
2611 emit_java_to_interp(cbuf);
2612 }
2613 %}
2614
2615 enc_class Java_Dynamic_Call(method meth)
2616 %{
2617 // JAVA DYNAMIC CALL
2618 // !!!!!
2619 // Generate "movq rax, -1", placeholder instruction to load oop-info
2620 // emit_call_dynamic_prologue( cbuf );
2621 cbuf.set_insts_mark();
2622
2623 // movq rax, -1
2624 emit_opcode(cbuf, Assembler::REX_W);
2625 emit_opcode(cbuf, 0xB8 | RAX_enc);
2626 emit_d64_reloc(cbuf,
2627 (int64_t) Universe::non_oop_word(),
2628 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2629 address virtual_call_oop_addr = cbuf.insts_mark();
2630 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2631 // who we intended to call.
2632 cbuf.set_insts_mark();
2633 $$$emit8$primary;
2634 emit_d32_reloc(cbuf,
2635 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2636 virtual_call_Relocation::spec(virtual_call_oop_addr),
2637 RELOC_DISP32);
2638 %}
2639
2640 enc_class Java_Compiled_Call(method meth)
2641 %{
2642 // JAVA COMPILED CALL
2643 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2644
2645 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2646 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2647
2648 // callq *disp(%rax)
2649 cbuf.set_insts_mark();
2650 $$$emit8$primary;
2651 if (disp < 0x80) {
2652 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2653 emit_d8(cbuf, disp); // Displacement
2654 } else {
2655 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2656 emit_d32(cbuf, disp); // Displacement
2657 }
2658 %}
2659
2660 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2661 %{
2662 // SAL, SAR, SHR
2663 int dstenc = $dst$$reg;
2664 if (dstenc >= 8) {
2665 emit_opcode(cbuf, Assembler::REX_B);
2666 dstenc -= 8;
2667 }
2668 $$$emit8$primary;
2669 emit_rm(cbuf, 0x3, $secondary, dstenc);
2670 $$$emit8$shift$$constant;
2671 %}
2672
2673 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2674 %{
2675 // SAL, SAR, SHR
2676 int dstenc = $dst$$reg;
2677 if (dstenc < 8) {
2678 emit_opcode(cbuf, Assembler::REX_W);
2679 } else {
2680 emit_opcode(cbuf, Assembler::REX_WB);
2681 dstenc -= 8;
2682 }
2683 $$$emit8$primary;
2684 emit_rm(cbuf, 0x3, $secondary, dstenc);
2685 $$$emit8$shift$$constant;
2686 %}
2687
2688 enc_class load_immI(rRegI dst, immI src)
2689 %{
2690 int dstenc = $dst$$reg;
2691 if (dstenc >= 8) {
2692 emit_opcode(cbuf, Assembler::REX_B);
2693 dstenc -= 8;
2694 }
2695 emit_opcode(cbuf, 0xB8 | dstenc);
2696 $$$emit32$src$$constant;
2697 %}
2698
2699 enc_class load_immL(rRegL dst, immL src)
2700 %{
2701 int dstenc = $dst$$reg;
2702 if (dstenc < 8) {
2703 emit_opcode(cbuf, Assembler::REX_W);
2704 } else {
2705 emit_opcode(cbuf, Assembler::REX_WB);
2706 dstenc -= 8;
2707 }
2708 emit_opcode(cbuf, 0xB8 | dstenc);
2709 emit_d64(cbuf, $src$$constant);
2710 %}
2711
2712 enc_class load_immUL32(rRegL dst, immUL32 src)
2713 %{
2714 // same as load_immI, but this time we care about zeroes in the high word
2715 int dstenc = $dst$$reg;
2716 if (dstenc >= 8) {
2717 emit_opcode(cbuf, Assembler::REX_B);
2718 dstenc -= 8;
2719 }
2720 emit_opcode(cbuf, 0xB8 | dstenc);
2721 $$$emit32$src$$constant;
2722 %}
2723
2724 enc_class load_immL32(rRegL dst, immL32 src)
2725 %{
2726 int dstenc = $dst$$reg;
2727 if (dstenc < 8) {
2728 emit_opcode(cbuf, Assembler::REX_W);
2729 } else {
2730 emit_opcode(cbuf, Assembler::REX_WB);
2731 dstenc -= 8;
2732 }
2733 emit_opcode(cbuf, 0xC7);
2734 emit_rm(cbuf, 0x03, 0x00, dstenc);
2735 $$$emit32$src$$constant;
2736 %}
2737
2738 enc_class load_immP31(rRegP dst, immP32 src)
2739 %{
2740 // same as load_immI, but this time we care about zeroes in the high word
2741 int dstenc = $dst$$reg;
2742 if (dstenc >= 8) {
2743 emit_opcode(cbuf, Assembler::REX_B);
2744 dstenc -= 8;
2745 }
2746 emit_opcode(cbuf, 0xB8 | dstenc);
2747 $$$emit32$src$$constant;
2748 %}
2749
2750 enc_class load_immP(rRegP dst, immP src)
2751 %{
2752 int dstenc = $dst$$reg;
2753 if (dstenc < 8) {
2754 emit_opcode(cbuf, Assembler::REX_W);
2755 } else {
2756 emit_opcode(cbuf, Assembler::REX_WB);
2757 dstenc -= 8;
2758 }
2759 emit_opcode(cbuf, 0xB8 | dstenc);
2760 // This next line should be generated from ADLC
2761 if ($src->constant_is_oop()) {
2762 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2763 } else {
2764 emit_d64(cbuf, $src$$constant);
2765 }
2766 %}
2767
2768 // Encode a reg-reg copy. If it is useless, then empty encoding.
2769 enc_class enc_copy(rRegI dst, rRegI src)
2770 %{
2771 encode_copy(cbuf, $dst$$reg, $src$$reg);
2772 %}
2773
2774 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2775 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2776 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2777 %}
2778
2779 enc_class enc_copy_always(rRegI dst, rRegI src)
2780 %{
2781 int srcenc = $src$$reg;
2782 int dstenc = $dst$$reg;
2783
2784 if (dstenc < 8) {
2785 if (srcenc >= 8) {
2786 emit_opcode(cbuf, Assembler::REX_B);
2787 srcenc -= 8;
2788 }
2789 } else {
2790 if (srcenc < 8) {
2791 emit_opcode(cbuf, Assembler::REX_R);
2792 } else {
2793 emit_opcode(cbuf, Assembler::REX_RB);
2794 srcenc -= 8;
2795 }
2796 dstenc -= 8;
2797 }
2798
2799 emit_opcode(cbuf, 0x8B);
2800 emit_rm(cbuf, 0x3, dstenc, srcenc);
2801 %}
2802
2803 enc_class enc_copy_wide(rRegL dst, rRegL src)
2804 %{
2805 int srcenc = $src$$reg;
2806 int dstenc = $dst$$reg;
2807
2808 if (dstenc != srcenc) {
2809 if (dstenc < 8) {
2810 if (srcenc < 8) {
2811 emit_opcode(cbuf, Assembler::REX_W);
2812 } else {
2813 emit_opcode(cbuf, Assembler::REX_WB);
2814 srcenc -= 8;
2815 }
2816 } else {
2817 if (srcenc < 8) {
2818 emit_opcode(cbuf, Assembler::REX_WR);
2819 } else {
2820 emit_opcode(cbuf, Assembler::REX_WRB);
2821 srcenc -= 8;
2822 }
2823 dstenc -= 8;
2824 }
2825 emit_opcode(cbuf, 0x8B);
2826 emit_rm(cbuf, 0x3, dstenc, srcenc);
2827 }
2828 %}
2829
2830 enc_class Con32(immI src)
2831 %{
2832 // Output immediate
2833 $$$emit32$src$$constant;
2834 %}
2835
2836 enc_class Con64(immL src)
2837 %{
2838 // Output immediate
2839 emit_d64($src$$constant);
2840 %}
2841
2842 enc_class Con32F_as_bits(immF src)
2843 %{
2844 // Output Float immediate bits
2845 jfloat jf = $src$$constant;
2846 jint jf_as_bits = jint_cast(jf);
2847 emit_d32(cbuf, jf_as_bits);
2848 %}
2849
2850 enc_class Con16(immI src)
2851 %{
2852 // Output immediate
2853 $$$emit16$src$$constant;
2854 %}
2855
2856 // How is this different from Con32??? XXX
2857 enc_class Con_d32(immI src)
2858 %{
2859 emit_d32(cbuf,$src$$constant);
2860 %}
2861
2862 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2863 // Output immediate memory reference
2864 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2865 emit_d32(cbuf, 0x00);
2866 %}
2867
2868 enc_class lock_prefix()
2869 %{
2870 if (os::is_MP()) {
2871 emit_opcode(cbuf, 0xF0); // lock
2872 }
2873 %}
2874
2875 enc_class REX_mem(memory mem)
2876 %{
2877 if ($mem$$base >= 8) {
2878 if ($mem$$index < 8) {
2879 emit_opcode(cbuf, Assembler::REX_B);
2880 } else {
2881 emit_opcode(cbuf, Assembler::REX_XB);
2882 }
2883 } else {
2884 if ($mem$$index >= 8) {
2885 emit_opcode(cbuf, Assembler::REX_X);
2886 }
2887 }
2888 %}
2889
2890 enc_class REX_mem_wide(memory mem)
2891 %{
2892 if ($mem$$base >= 8) {
2893 if ($mem$$index < 8) {
2894 emit_opcode(cbuf, Assembler::REX_WB);
2895 } else {
2896 emit_opcode(cbuf, Assembler::REX_WXB);
2897 }
2898 } else {
2899 if ($mem$$index < 8) {
2900 emit_opcode(cbuf, Assembler::REX_W);
2901 } else {
2902 emit_opcode(cbuf, Assembler::REX_WX);
2903 }
2904 }
2905 %}
2906
2907 // for byte regs
2908 enc_class REX_breg(rRegI reg)
2909 %{
2910 if ($reg$$reg >= 4) {
2911 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
2912 }
2913 %}
2914
2915 // for byte regs
2916 enc_class REX_reg_breg(rRegI dst, rRegI src)
2917 %{
2918 if ($dst$$reg < 8) {
2919 if ($src$$reg >= 4) {
2920 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
2921 }
2922 } else {
2923 if ($src$$reg < 8) {
2924 emit_opcode(cbuf, Assembler::REX_R);
2925 } else {
2926 emit_opcode(cbuf, Assembler::REX_RB);
2927 }
2928 }
2929 %}
2930
2931 // for byte regs
2932 enc_class REX_breg_mem(rRegI reg, memory mem)
2933 %{
2934 if ($reg$$reg < 8) {
2935 if ($mem$$base < 8) {
2936 if ($mem$$index >= 8) {
2937 emit_opcode(cbuf, Assembler::REX_X);
2938 } else if ($reg$$reg >= 4) {
2939 emit_opcode(cbuf, Assembler::REX);
2940 }
2941 } else {
2942 if ($mem$$index < 8) {
2943 emit_opcode(cbuf, Assembler::REX_B);
2944 } else {
2945 emit_opcode(cbuf, Assembler::REX_XB);
2946 }
2947 }
2948 } else {
2949 if ($mem$$base < 8) {
2950 if ($mem$$index < 8) {
2951 emit_opcode(cbuf, Assembler::REX_R);
2952 } else {
2953 emit_opcode(cbuf, Assembler::REX_RX);
2954 }
2955 } else {
2956 if ($mem$$index < 8) {
2957 emit_opcode(cbuf, Assembler::REX_RB);
2958 } else {
2959 emit_opcode(cbuf, Assembler::REX_RXB);
2960 }
2961 }
2962 }
2963 %}
2964
2965 enc_class REX_reg(rRegI reg)
2966 %{
2967 if ($reg$$reg >= 8) {
2968 emit_opcode(cbuf, Assembler::REX_B);
2969 }
2970 %}
2971
2972 enc_class REX_reg_wide(rRegI reg)
2973 %{
2974 if ($reg$$reg < 8) {
2975 emit_opcode(cbuf, Assembler::REX_W);
2976 } else {
2977 emit_opcode(cbuf, Assembler::REX_WB);
2978 }
2979 %}
2980
2981 enc_class REX_reg_reg(rRegI dst, rRegI src)
2982 %{
2983 if ($dst$$reg < 8) {
2984 if ($src$$reg >= 8) {
2985 emit_opcode(cbuf, Assembler::REX_B);
2986 }
2987 } else {
2988 if ($src$$reg < 8) {
2989 emit_opcode(cbuf, Assembler::REX_R);
2990 } else {
2991 emit_opcode(cbuf, Assembler::REX_RB);
2992 }
2993 }
2994 %}
2995
2996 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
2997 %{
2998 if ($dst$$reg < 8) {
2999 if ($src$$reg < 8) {
3000 emit_opcode(cbuf, Assembler::REX_W);
3001 } else {
3002 emit_opcode(cbuf, Assembler::REX_WB);
3003 }
3004 } else {
3005 if ($src$$reg < 8) {
3006 emit_opcode(cbuf, Assembler::REX_WR);
3007 } else {
3008 emit_opcode(cbuf, Assembler::REX_WRB);
3009 }
3010 }
3011 %}
3012
3013 enc_class REX_reg_mem(rRegI reg, memory mem)
3014 %{
3015 if ($reg$$reg < 8) {
3016 if ($mem$$base < 8) {
3017 if ($mem$$index >= 8) {
3018 emit_opcode(cbuf, Assembler::REX_X);
3019 }
3020 } else {
3021 if ($mem$$index < 8) {
3022 emit_opcode(cbuf, Assembler::REX_B);
3023 } else {
3024 emit_opcode(cbuf, Assembler::REX_XB);
3025 }
3026 }
3027 } else {
3028 if ($mem$$base < 8) {
3029 if ($mem$$index < 8) {
3030 emit_opcode(cbuf, Assembler::REX_R);
3031 } else {
3032 emit_opcode(cbuf, Assembler::REX_RX);
3033 }
3034 } else {
3035 if ($mem$$index < 8) {
3036 emit_opcode(cbuf, Assembler::REX_RB);
3037 } else {
3038 emit_opcode(cbuf, Assembler::REX_RXB);
3039 }
3040 }
3041 }
3042 %}
3043
3044 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3045 %{
3046 if ($reg$$reg < 8) {
3047 if ($mem$$base < 8) {
3048 if ($mem$$index < 8) {
3049 emit_opcode(cbuf, Assembler::REX_W);
3050 } else {
3051 emit_opcode(cbuf, Assembler::REX_WX);
3052 }
3053 } else {
3054 if ($mem$$index < 8) {
3055 emit_opcode(cbuf, Assembler::REX_WB);
3056 } else {
3057 emit_opcode(cbuf, Assembler::REX_WXB);
3058 }
3059 }
3060 } else {
3061 if ($mem$$base < 8) {
3062 if ($mem$$index < 8) {
3063 emit_opcode(cbuf, Assembler::REX_WR);
3064 } else {
3065 emit_opcode(cbuf, Assembler::REX_WRX);
3066 }
3067 } else {
3068 if ($mem$$index < 8) {
3069 emit_opcode(cbuf, Assembler::REX_WRB);
3070 } else {
3071 emit_opcode(cbuf, Assembler::REX_WRXB);
3072 }
3073 }
3074 }
3075 %}
3076
3077 enc_class reg_mem(rRegI ereg, memory mem)
3078 %{
3079 // High registers handle in encode_RegMem
3080 int reg = $ereg$$reg;
3081 int base = $mem$$base;
3082 int index = $mem$$index;
3083 int scale = $mem$$scale;
3084 int disp = $mem$$disp;
3085 bool disp_is_oop = $mem->disp_is_oop();
3086
3087 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3088 %}
3089
3090 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3091 %{
3092 int rm_byte_opcode = $rm_opcode$$constant;
3093
3094 // High registers handle in encode_RegMem
3095 int base = $mem$$base;
3096 int index = $mem$$index;
3097 int scale = $mem$$scale;
3098 int displace = $mem$$disp;
3099
3100 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3101 // working with static
3102 // globals
3103 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3104 disp_is_oop);
3105 %}
3106
3107 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3108 %{
3109 int reg_encoding = $dst$$reg;
3110 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3111 int index = 0x04; // 0x04 indicates no index
3112 int scale = 0x00; // 0x00 indicates no scale
3113 int displace = $src1$$constant; // 0x00 indicates no displacement
3114 bool disp_is_oop = false;
3115 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3116 disp_is_oop);
3117 %}
3118
3119 enc_class neg_reg(rRegI dst)
3120 %{
3121 int dstenc = $dst$$reg;
3122 if (dstenc >= 8) {
3123 emit_opcode(cbuf, Assembler::REX_B);
3124 dstenc -= 8;
3125 }
3126 // NEG $dst
3127 emit_opcode(cbuf, 0xF7);
3128 emit_rm(cbuf, 0x3, 0x03, dstenc);
3129 %}
3130
3131 enc_class neg_reg_wide(rRegI dst)
3132 %{
3133 int dstenc = $dst$$reg;
3134 if (dstenc < 8) {
3135 emit_opcode(cbuf, Assembler::REX_W);
3136 } else {
3137 emit_opcode(cbuf, Assembler::REX_WB);
3138 dstenc -= 8;
3139 }
3140 // NEG $dst
3141 emit_opcode(cbuf, 0xF7);
3142 emit_rm(cbuf, 0x3, 0x03, dstenc);
3143 %}
3144
3145 enc_class setLT_reg(rRegI dst)
3146 %{
3147 int dstenc = $dst$$reg;
3148 if (dstenc >= 8) {
3149 emit_opcode(cbuf, Assembler::REX_B);
3150 dstenc -= 8;
3151 } else if (dstenc >= 4) {
3152 emit_opcode(cbuf, Assembler::REX);
3153 }
3154 // SETLT $dst
3155 emit_opcode(cbuf, 0x0F);
3156 emit_opcode(cbuf, 0x9C);
3157 emit_rm(cbuf, 0x3, 0x0, dstenc);
3158 %}
3159
3160 enc_class setNZ_reg(rRegI dst)
3161 %{
3162 int dstenc = $dst$$reg;
3163 if (dstenc >= 8) {
3164 emit_opcode(cbuf, Assembler::REX_B);
3165 dstenc -= 8;
3166 } else if (dstenc >= 4) {
3167 emit_opcode(cbuf, Assembler::REX);
3168 }
3169 // SETNZ $dst
3170 emit_opcode(cbuf, 0x0F);
3171 emit_opcode(cbuf, 0x95);
3172 emit_rm(cbuf, 0x3, 0x0, dstenc);
3173 %}
3174
3175
3176 // Compare the lonogs and set -1, 0, or 1 into dst
3177 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3178 %{
3179 int src1enc = $src1$$reg;
3180 int src2enc = $src2$$reg;
3181 int dstenc = $dst$$reg;
3182
3183 // cmpq $src1, $src2
3184 if (src1enc < 8) {
3185 if (src2enc < 8) {
3186 emit_opcode(cbuf, Assembler::REX_W);
3187 } else {
3188 emit_opcode(cbuf, Assembler::REX_WB);
3189 }
3190 } else {
3191 if (src2enc < 8) {
3192 emit_opcode(cbuf, Assembler::REX_WR);
3193 } else {
3194 emit_opcode(cbuf, Assembler::REX_WRB);
3195 }
3196 }
3197 emit_opcode(cbuf, 0x3B);
3198 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3199
3200 // movl $dst, -1
3201 if (dstenc >= 8) {
3202 emit_opcode(cbuf, Assembler::REX_B);
3203 }
3204 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3205 emit_d32(cbuf, -1);
3206
3207 // jl,s done
3208 emit_opcode(cbuf, 0x7C);
3209 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3210
3211 // setne $dst
3212 if (dstenc >= 4) {
3213 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3214 }
3215 emit_opcode(cbuf, 0x0F);
3216 emit_opcode(cbuf, 0x95);
3217 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3218
3219 // movzbl $dst, $dst
3220 if (dstenc >= 4) {
3221 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3222 }
3223 emit_opcode(cbuf, 0x0F);
3224 emit_opcode(cbuf, 0xB6);
3225 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3226 %}
3227
3228 enc_class Push_ResultXD(regD dst) %{
3229 int dstenc = $dst$$reg;
3230
3231 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3232
3233 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3234 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3235 if (dstenc >= 8) {
3236 emit_opcode(cbuf, Assembler::REX_R);
3237 }
3238 emit_opcode (cbuf, 0x0F );
3239 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3240 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3241
3242 // add rsp,8
3243 emit_opcode(cbuf, Assembler::REX_W);
3244 emit_opcode(cbuf,0x83);
3245 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3246 emit_d8(cbuf,0x08);
3247 %}
3248
3249 enc_class Push_SrcXD(regD src) %{
3250 int srcenc = $src$$reg;
3251
3252 // subq rsp,#8
3253 emit_opcode(cbuf, Assembler::REX_W);
3254 emit_opcode(cbuf, 0x83);
3255 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3256 emit_d8(cbuf, 0x8);
3257
3258 // movsd [rsp],src
3259 emit_opcode(cbuf, 0xF2);
3260 if (srcenc >= 8) {
3261 emit_opcode(cbuf, Assembler::REX_R);
3262 }
3263 emit_opcode(cbuf, 0x0F);
3264 emit_opcode(cbuf, 0x11);
3265 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3266
3267 // fldd [rsp]
3268 emit_opcode(cbuf, 0x66);
3269 emit_opcode(cbuf, 0xDD);
3270 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3271 %}
3272
3273
3274 enc_class movq_ld(regD dst, memory mem) %{
3275 MacroAssembler _masm(&cbuf);
3276 __ movq($dst$$XMMRegister, $mem$$Address);
3277 %}
3278
3279 enc_class movq_st(memory mem, regD src) %{
3280 MacroAssembler _masm(&cbuf);
3281 __ movq($mem$$Address, $src$$XMMRegister);
3282 %}
3283
3284 enc_class pshufd_8x8(regF dst, regF src) %{
3285 MacroAssembler _masm(&cbuf);
3286
3287 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3288 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3289 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3290 %}
3291
3292 enc_class pshufd_4x16(regF dst, regF src) %{
3293 MacroAssembler _masm(&cbuf);
3294
3295 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3296 %}
3297
3298 enc_class pshufd(regD dst, regD src, int mode) %{
3299 MacroAssembler _masm(&cbuf);
3300
3301 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3302 %}
3303
3304 enc_class pxor(regD dst, regD src) %{
3305 MacroAssembler _masm(&cbuf);
3306
3307 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3308 %}
3309
3310 enc_class mov_i2x(regD dst, rRegI src) %{
3311 MacroAssembler _masm(&cbuf);
3312
3313 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3314 %}
3315
3316 // obj: object to lock
3317 // box: box address (header location) -- killed
3318 // tmp: rax -- killed
3319 // scr: rbx -- killed
3320 //
3321 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3322 // from i486.ad. See that file for comments.
3323 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3324 // use the shorter encoding. (Movl clears the high-order 32-bits).
3325
3326
3327 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3328 %{
3329 Register objReg = as_Register((int)$obj$$reg);
3330 Register boxReg = as_Register((int)$box$$reg);
3331 Register tmpReg = as_Register($tmp$$reg);
3332 Register scrReg = as_Register($scr$$reg);
3333 MacroAssembler masm(&cbuf);
3334
3335 // Verify uniqueness of register assignments -- necessary but not sufficient
3336 assert (objReg != boxReg && objReg != tmpReg &&
3337 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3338
3339 if (_counters != NULL) {
3340 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3341 }
3342 if (EmitSync & 1) {
3343 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3344 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3345 masm.cmpptr(rsp, (int32_t)NULL_WORD) ;
3346 } else
3347 if (EmitSync & 2) {
3348 Label DONE_LABEL;
3349 if (UseBiasedLocking) {
3350 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3351 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3352 }
3353 // QQQ was movl...
3354 masm.movptr(tmpReg, 0x1);
3355 masm.orptr(tmpReg, Address(objReg, 0));
3356 masm.movptr(Address(boxReg, 0), tmpReg);
3357 if (os::is_MP()) {
3358 masm.lock();
3359 }
3360 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3361 masm.jcc(Assembler::equal, DONE_LABEL);
3362
3363 // Recursive locking
3364 masm.subptr(tmpReg, rsp);
3365 masm.andptr(tmpReg, 7 - os::vm_page_size());
3366 masm.movptr(Address(boxReg, 0), tmpReg);
3367
3368 masm.bind(DONE_LABEL);
3369 masm.nop(); // avoid branch to branch
3370 } else {
3371 Label DONE_LABEL, IsInflated, Egress;
3372
3373 masm.movptr(tmpReg, Address(objReg, 0)) ;
3374 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3375 masm.jcc (Assembler::notZero, IsInflated) ;
3376
3377 // it's stack-locked, biased or neutral
3378 // TODO: optimize markword triage order to reduce the number of
3379 // conditional branches in the most common cases.
3380 // Beware -- there's a subtle invariant that fetch of the markword
3381 // at [FETCH], below, will never observe a biased encoding (*101b).
3382 // If this invariant is not held we'll suffer exclusion (safety) failure.
3383
3384 if (UseBiasedLocking && !UseOptoBiasInlining) {
3385 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3386 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH]
3387 }
3388
3389 // was q will it destroy high?
3390 masm.orl (tmpReg, 1) ;
3391 masm.movptr(Address(boxReg, 0), tmpReg) ;
3392 if (os::is_MP()) { masm.lock(); }
3393 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3394 if (_counters != NULL) {
3395 masm.cond_inc32(Assembler::equal,
3396 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3397 }
3398 masm.jcc (Assembler::equal, DONE_LABEL);
3399
3400 // Recursive locking
3401 masm.subptr(tmpReg, rsp);
3402 masm.andptr(tmpReg, 7 - os::vm_page_size());
3403 masm.movptr(Address(boxReg, 0), tmpReg);
3404 if (_counters != NULL) {
3405 masm.cond_inc32(Assembler::equal,
3406 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3407 }
3408 masm.jmp (DONE_LABEL) ;
3409
3410 masm.bind (IsInflated) ;
3411 // It's inflated
3412
3413 // TODO: someday avoid the ST-before-CAS penalty by
3414 // relocating (deferring) the following ST.
3415 // We should also think about trying a CAS without having
3416 // fetched _owner. If the CAS is successful we may
3417 // avoid an RTO->RTS upgrade on the $line.
3418 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3419 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3420
3421 masm.mov (boxReg, tmpReg) ;
3422 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3423 masm.testptr(tmpReg, tmpReg) ;
3424 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3425
3426 // It's inflated and appears unlocked
3427 if (os::is_MP()) { masm.lock(); }
3428 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3429 // Intentional fall-through into DONE_LABEL ...
3430
3431 masm.bind (DONE_LABEL) ;
3432 masm.nop () ; // avoid jmp to jmp
3433 }
3434 %}
3435
3436 // obj: object to unlock
3437 // box: box address (displaced header location), killed
3438 // RBX: killed tmp; cannot be obj nor box
3439 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3440 %{
3441
3442 Register objReg = as_Register($obj$$reg);
3443 Register boxReg = as_Register($box$$reg);
3444 Register tmpReg = as_Register($tmp$$reg);
3445 MacroAssembler masm(&cbuf);
3446
3447 if (EmitSync & 4) {
3448 masm.cmpptr(rsp, 0) ;
3449 } else
3450 if (EmitSync & 8) {
3451 Label DONE_LABEL;
3452 if (UseBiasedLocking) {
3453 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3454 }
3455
3456 // Check whether the displaced header is 0
3457 //(=> recursive unlock)
3458 masm.movptr(tmpReg, Address(boxReg, 0));
3459 masm.testptr(tmpReg, tmpReg);
3460 masm.jcc(Assembler::zero, DONE_LABEL);
3461
3462 // If not recursive lock, reset the header to displaced header
3463 if (os::is_MP()) {
3464 masm.lock();
3465 }
3466 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3467 masm.bind(DONE_LABEL);
3468 masm.nop(); // avoid branch to branch
3469 } else {
3470 Label DONE_LABEL, Stacked, CheckSucc ;
3471
3472 if (UseBiasedLocking && !UseOptoBiasInlining) {
3473 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3474 }
3475
3476 masm.movptr(tmpReg, Address(objReg, 0)) ;
3477 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ;
3478 masm.jcc (Assembler::zero, DONE_LABEL) ;
3479 masm.testl (tmpReg, 0x02) ;
3480 masm.jcc (Assembler::zero, Stacked) ;
3481
3482 // It's inflated
3483 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3484 masm.xorptr(boxReg, r15_thread) ;
3485 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3486 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3487 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3488 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3489 masm.jcc (Assembler::notZero, CheckSucc) ;
3490 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3491 masm.jmp (DONE_LABEL) ;
3492
3493 if ((EmitSync & 65536) == 0) {
3494 Label LSuccess, LGoSlowPath ;
3495 masm.bind (CheckSucc) ;
3496 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3497 masm.jcc (Assembler::zero, LGoSlowPath) ;
3498
3499 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3500 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3501 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3502 // are all faster when the write buffer is populated.
3503 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3504 if (os::is_MP()) {
3505 masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3506 }
3507 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3508 masm.jcc (Assembler::notZero, LSuccess) ;
3509
3510 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX
3511 if (os::is_MP()) { masm.lock(); }
3512 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3513 masm.jcc (Assembler::notEqual, LSuccess) ;
3514 // Intentional fall-through into slow-path
3515
3516 masm.bind (LGoSlowPath) ;
3517 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3518 masm.jmp (DONE_LABEL) ;
3519
3520 masm.bind (LSuccess) ;
3521 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3522 masm.jmp (DONE_LABEL) ;
3523 }
3524
3525 masm.bind (Stacked) ;
3526 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch
3527 if (os::is_MP()) { masm.lock(); }
3528 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3529
3530 if (EmitSync & 65536) {
3531 masm.bind (CheckSucc) ;
3532 }
3533 masm.bind(DONE_LABEL);
3534 if (EmitSync & 32768) {
3535 masm.nop(); // avoid branch to branch
3536 }
3537 }
3538 %}
3539
3540
3541 enc_class enc_rethrow()
3542 %{
3543 cbuf.set_insts_mark();
3544 emit_opcode(cbuf, 0xE9); // jmp entry
3545 emit_d32_reloc(cbuf,
3546 (int) (OptoRuntime::rethrow_stub() - cbuf.insts_end() - 4),
3547 runtime_call_Relocation::spec(),
3548 RELOC_DISP32);
3549 %}
3550
3551 enc_class absF_encoding(regF dst)
3552 %{
3553 int dstenc = $dst$$reg;
3554 address signmask_address = (address) StubRoutines::x86::float_sign_mask();
3555
3556 cbuf.set_insts_mark();
3557 if (dstenc >= 8) {
3558 emit_opcode(cbuf, Assembler::REX_R);
3559 dstenc -= 8;
3560 }
3561 // XXX reg_mem doesn't support RIP-relative addressing yet
3562 emit_opcode(cbuf, 0x0F);
3563 emit_opcode(cbuf, 0x54);
3564 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3565 emit_d32_reloc(cbuf, signmask_address);
3566 %}
3567
3568 enc_class absD_encoding(regD dst)
3569 %{
3570 int dstenc = $dst$$reg;
3571 address signmask_address = (address) StubRoutines::x86::double_sign_mask();
3572
3573 cbuf.set_insts_mark();
3574 emit_opcode(cbuf, 0x66);
3575 if (dstenc >= 8) {
3576 emit_opcode(cbuf, Assembler::REX_R);
3577 dstenc -= 8;
3578 }
3579 // XXX reg_mem doesn't support RIP-relative addressing yet
3580 emit_opcode(cbuf, 0x0F);
3581 emit_opcode(cbuf, 0x54);
3582 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3583 emit_d32_reloc(cbuf, signmask_address);
3584 %}
3585
3586 enc_class negF_encoding(regF dst)
3587 %{
3588 int dstenc = $dst$$reg;
3589 address signflip_address = (address) StubRoutines::x86::float_sign_flip();
3590
3591 cbuf.set_insts_mark();
3592 if (dstenc >= 8) {
3593 emit_opcode(cbuf, Assembler::REX_R);
3594 dstenc -= 8;
3595 }
3596 // XXX reg_mem doesn't support RIP-relative addressing yet
3597 emit_opcode(cbuf, 0x0F);
3598 emit_opcode(cbuf, 0x57);
3599 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3600 emit_d32_reloc(cbuf, signflip_address);
3601 %}
3602
3603 enc_class negD_encoding(regD dst)
3604 %{
3605 int dstenc = $dst$$reg;
3606 address signflip_address = (address) StubRoutines::x86::double_sign_flip();
3607
3608 cbuf.set_insts_mark();
3609 emit_opcode(cbuf, 0x66);
3610 if (dstenc >= 8) {
3611 emit_opcode(cbuf, Assembler::REX_R);
3612 dstenc -= 8;
3613 }
3614 // XXX reg_mem doesn't support RIP-relative addressing yet
3615 emit_opcode(cbuf, 0x0F);
3616 emit_opcode(cbuf, 0x57);
3617 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3618 emit_d32_reloc(cbuf, signflip_address);
3619 %}
3620
3621 enc_class f2i_fixup(rRegI dst, regF src)
3622 %{
3623 int dstenc = $dst$$reg;
3624 int srcenc = $src$$reg;
3625
3626 // cmpl $dst, #0x80000000
3627 if (dstenc >= 8) {
3628 emit_opcode(cbuf, Assembler::REX_B);
3629 }
3630 emit_opcode(cbuf, 0x81);
3631 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3632 emit_d32(cbuf, 0x80000000);
3633
3634 // jne,s done
3635 emit_opcode(cbuf, 0x75);
3636 if (srcenc < 8 && dstenc < 8) {
3637 emit_d8(cbuf, 0xF);
3638 } else if (srcenc >= 8 && dstenc >= 8) {
3639 emit_d8(cbuf, 0x11);
3640 } else {
3641 emit_d8(cbuf, 0x10);
3642 }
3643
3644 // subq rsp, #8
3645 emit_opcode(cbuf, Assembler::REX_W);
3646 emit_opcode(cbuf, 0x83);
3647 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3648 emit_d8(cbuf, 8);
3649
3650 // movss [rsp], $src
3651 emit_opcode(cbuf, 0xF3);
3652 if (srcenc >= 8) {
3653 emit_opcode(cbuf, Assembler::REX_R);
3654 }
3655 emit_opcode(cbuf, 0x0F);
3656 emit_opcode(cbuf, 0x11);
3657 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3658
3659 // call f2i_fixup
3660 cbuf.set_insts_mark();
3661 emit_opcode(cbuf, 0xE8);
3662 emit_d32_reloc(cbuf,
3663 (int)
3664 (StubRoutines::x86::f2i_fixup() - cbuf.insts_end() - 4),
3665 runtime_call_Relocation::spec(),
3666 RELOC_DISP32);
3667
3668 // popq $dst
3669 if (dstenc >= 8) {
3670 emit_opcode(cbuf, Assembler::REX_B);
3671 }
3672 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3673
3674 // done:
3675 %}
3676
3677 enc_class f2l_fixup(rRegL dst, regF src)
3678 %{
3679 int dstenc = $dst$$reg;
3680 int srcenc = $src$$reg;
3681 address const_address = (address) StubRoutines::x86::double_sign_flip();
3682
3683 // cmpq $dst, [0x8000000000000000]
3684 cbuf.set_insts_mark();
3685 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3686 emit_opcode(cbuf, 0x39);
3687 // XXX reg_mem doesn't support RIP-relative addressing yet
3688 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3689 emit_d32_reloc(cbuf, const_address);
3690
3691
3692 // jne,s done
3693 emit_opcode(cbuf, 0x75);
3694 if (srcenc < 8 && dstenc < 8) {
3695 emit_d8(cbuf, 0xF);
3696 } else if (srcenc >= 8 && dstenc >= 8) {
3697 emit_d8(cbuf, 0x11);
3698 } else {
3699 emit_d8(cbuf, 0x10);
3700 }
3701
3702 // subq rsp, #8
3703 emit_opcode(cbuf, Assembler::REX_W);
3704 emit_opcode(cbuf, 0x83);
3705 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3706 emit_d8(cbuf, 8);
3707
3708 // movss [rsp], $src
3709 emit_opcode(cbuf, 0xF3);
3710 if (srcenc >= 8) {
3711 emit_opcode(cbuf, Assembler::REX_R);
3712 }
3713 emit_opcode(cbuf, 0x0F);
3714 emit_opcode(cbuf, 0x11);
3715 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3716
3717 // call f2l_fixup
3718 cbuf.set_insts_mark();
3719 emit_opcode(cbuf, 0xE8);
3720 emit_d32_reloc(cbuf,
3721 (int)
3722 (StubRoutines::x86::f2l_fixup() - cbuf.insts_end() - 4),
3723 runtime_call_Relocation::spec(),
3724 RELOC_DISP32);
3725
3726 // popq $dst
3727 if (dstenc >= 8) {
3728 emit_opcode(cbuf, Assembler::REX_B);
3729 }
3730 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3731
3732 // done:
3733 %}
3734
3735 enc_class d2i_fixup(rRegI dst, regD src)
3736 %{
3737 int dstenc = $dst$$reg;
3738 int srcenc = $src$$reg;
3739
3740 // cmpl $dst, #0x80000000
3741 if (dstenc >= 8) {
3742 emit_opcode(cbuf, Assembler::REX_B);
3743 }
3744 emit_opcode(cbuf, 0x81);
3745 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3746 emit_d32(cbuf, 0x80000000);
3747
3748 // jne,s done
3749 emit_opcode(cbuf, 0x75);
3750 if (srcenc < 8 && dstenc < 8) {
3751 emit_d8(cbuf, 0xF);
3752 } else if (srcenc >= 8 && dstenc >= 8) {
3753 emit_d8(cbuf, 0x11);
3754 } else {
3755 emit_d8(cbuf, 0x10);
3756 }
3757
3758 // subq rsp, #8
3759 emit_opcode(cbuf, Assembler::REX_W);
3760 emit_opcode(cbuf, 0x83);
3761 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3762 emit_d8(cbuf, 8);
3763
3764 // movsd [rsp], $src
3765 emit_opcode(cbuf, 0xF2);
3766 if (srcenc >= 8) {
3767 emit_opcode(cbuf, Assembler::REX_R);
3768 }
3769 emit_opcode(cbuf, 0x0F);
3770 emit_opcode(cbuf, 0x11);
3771 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3772
3773 // call d2i_fixup
3774 cbuf.set_insts_mark();
3775 emit_opcode(cbuf, 0xE8);
3776 emit_d32_reloc(cbuf,
3777 (int)
3778 (StubRoutines::x86::d2i_fixup() - cbuf.insts_end() - 4),
3779 runtime_call_Relocation::spec(),
3780 RELOC_DISP32);
3781
3782 // popq $dst
3783 if (dstenc >= 8) {
3784 emit_opcode(cbuf, Assembler::REX_B);
3785 }
3786 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3787
3788 // done:
3789 %}
3790
3791 enc_class d2l_fixup(rRegL dst, regD src)
3792 %{
3793 int dstenc = $dst$$reg;
3794 int srcenc = $src$$reg;
3795 address const_address = (address) StubRoutines::x86::double_sign_flip();
3796
3797 // cmpq $dst, [0x8000000000000000]
3798 cbuf.set_insts_mark();
3799 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3800 emit_opcode(cbuf, 0x39);
3801 // XXX reg_mem doesn't support RIP-relative addressing yet
3802 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3803 emit_d32_reloc(cbuf, const_address);
3804
3805
3806 // jne,s done
3807 emit_opcode(cbuf, 0x75);
3808 if (srcenc < 8 && dstenc < 8) {
3809 emit_d8(cbuf, 0xF);
3810 } else if (srcenc >= 8 && dstenc >= 8) {
3811 emit_d8(cbuf, 0x11);
3812 } else {
3813 emit_d8(cbuf, 0x10);
3814 }
3815
3816 // subq rsp, #8
3817 emit_opcode(cbuf, Assembler::REX_W);
3818 emit_opcode(cbuf, 0x83);
3819 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3820 emit_d8(cbuf, 8);
3821
3822 // movsd [rsp], $src
3823 emit_opcode(cbuf, 0xF2);
3824 if (srcenc >= 8) {
3825 emit_opcode(cbuf, Assembler::REX_R);
3826 }
3827 emit_opcode(cbuf, 0x0F);
3828 emit_opcode(cbuf, 0x11);
3829 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3830
3831 // call d2l_fixup
3832 cbuf.set_insts_mark();
3833 emit_opcode(cbuf, 0xE8);
3834 emit_d32_reloc(cbuf,
3835 (int)
3836 (StubRoutines::x86::d2l_fixup() - cbuf.insts_end() - 4),
3837 runtime_call_Relocation::spec(),
3838 RELOC_DISP32);
3839
3840 // popq $dst
3841 if (dstenc >= 8) {
3842 emit_opcode(cbuf, Assembler::REX_B);
3843 }
3844 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3845
3846 // done:
3847 %}
3848 %}
3849
3850
3851
3852 //----------FRAME--------------------------------------------------------------
3853 // Definition of frame structure and management information.
3854 //
3855 // S T A C K L A Y O U T Allocators stack-slot number
3856 // | (to get allocators register number
3857 // G Owned by | | v add OptoReg::stack0())
3858 // r CALLER | |
3859 // o | +--------+ pad to even-align allocators stack-slot
3860 // w V | pad0 | numbers; owned by CALLER
3861 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
3862 // h ^ | in | 5
3863 // | | args | 4 Holes in incoming args owned by SELF
3864 // | | | | 3
3865 // | | +--------+
3866 // V | | old out| Empty on Intel, window on Sparc
3867 // | old |preserve| Must be even aligned.
3868 // | SP-+--------+----> Matcher::_old_SP, even aligned
3869 // | | in | 3 area for Intel ret address
3870 // Owned by |preserve| Empty on Sparc.
3871 // SELF +--------+
3872 // | | pad2 | 2 pad to align old SP
3873 // | +--------+ 1
3874 // | | locks | 0
3875 // | +--------+----> OptoReg::stack0(), even aligned
3876 // | | pad1 | 11 pad to align new SP
3877 // | +--------+
3878 // | | | 10
3879 // | | spills | 9 spills
3880 // V | | 8 (pad0 slot for callee)
3881 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
3882 // ^ | out | 7
3883 // | | args | 6 Holes in outgoing args owned by CALLEE
3884 // Owned by +--------+
3885 // CALLEE | new out| 6 Empty on Intel, window on Sparc
3886 // | new |preserve| Must be even-aligned.
3887 // | SP-+--------+----> Matcher::_new_SP, even aligned
3888 // | | |
3889 //
3890 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
3891 // known from SELF's arguments and the Java calling convention.
3892 // Region 6-7 is determined per call site.
3893 // Note 2: If the calling convention leaves holes in the incoming argument
3894 // area, those holes are owned by SELF. Holes in the outgoing area
3895 // are owned by the CALLEE. Holes should not be nessecary in the
3896 // incoming area, as the Java calling convention is completely under
3897 // the control of the AD file. Doubles can be sorted and packed to
3898 // avoid holes. Holes in the outgoing arguments may be nessecary for
3899 // varargs C calling conventions.
3900 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
3901 // even aligned with pad0 as needed.
3902 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
3903 // region 6-11 is even aligned; it may be padded out more so that
3904 // the region from SP to FP meets the minimum stack alignment.
3905 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
3906 // alignment. Region 11, pad1, may be dynamically extended so that
3907 // SP meets the minimum alignment.
3908
3909 frame
3910 %{
3911 // What direction does stack grow in (assumed to be same for C & Java)
3912 stack_direction(TOWARDS_LOW);
3913
3914 // These three registers define part of the calling convention
3915 // between compiled code and the interpreter.
3916 inline_cache_reg(RAX); // Inline Cache Register
3917 interpreter_method_oop_reg(RBX); // Method Oop Register when
3918 // calling interpreter
3919
3920 // Optional: name the operand used by cisc-spilling to access
3921 // [stack_pointer + offset]
3922 cisc_spilling_operand_name(indOffset32);
3923
3924 // Number of stack slots consumed by locking an object
3925 sync_stack_slots(2);
3926
3927 // Compiled code's Frame Pointer
3928 frame_pointer(RSP);
3929
3930 // Interpreter stores its frame pointer in a register which is
3931 // stored to the stack by I2CAdaptors.
3932 // I2CAdaptors convert from interpreted java to compiled java.
3933 interpreter_frame_pointer(RBP);
3934
3935 // Stack alignment requirement
3936 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
3937
3938 // Number of stack slots between incoming argument block and the start of
3939 // a new frame. The PROLOG must add this many slots to the stack. The
3940 // EPILOG must remove this many slots. amd64 needs two slots for
3941 // return address.
3942 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
3943
3944 // Number of outgoing stack slots killed above the out_preserve_stack_slots
3945 // for calls to C. Supports the var-args backing area for register parms.
3946 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
3947
3948 // The after-PROLOG location of the return address. Location of
3949 // return address specifies a type (REG or STACK) and a number
3950 // representing the register number (i.e. - use a register name) or
3951 // stack slot.
3952 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
3953 // Otherwise, it is above the locks and verification slot and alignment word
3954 return_addr(STACK - 2 +
3955 round_to(2 + 2 * VerifyStackAtCalls +
3956 Compile::current()->fixed_slots(),
3957 WordsPerLong * 2));
3958
3959 // Body of function which returns an integer array locating
3960 // arguments either in registers or in stack slots. Passed an array
3961 // of ideal registers called "sig" and a "length" count. Stack-slot
3962 // offsets are based on outgoing arguments, i.e. a CALLER setting up
3963 // arguments for a CALLEE. Incoming stack arguments are
3964 // automatically biased by the preserve_stack_slots field above.
3965
3966 calling_convention
3967 %{
3968 // No difference between ingoing/outgoing just pass false
3969 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
3970 %}
3971
3972 c_calling_convention
3973 %{
3974 // This is obviously always outgoing
3975 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
3976 %}
3977
3978 // Location of compiled Java return values. Same as C for now.
3979 return_value
3980 %{
3981 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
3982 "only return normal values");
3983
3984 static const int lo[Op_RegL + 1] = {
3985 0,
3986 0,
3987 RAX_num, // Op_RegN
3988 RAX_num, // Op_RegI
3989 RAX_num, // Op_RegP
3990 XMM0_num, // Op_RegF
3991 XMM0_num, // Op_RegD
3992 RAX_num // Op_RegL
3993 };
3994 static const int hi[Op_RegL + 1] = {
3995 0,
3996 0,
3997 OptoReg::Bad, // Op_RegN
3998 OptoReg::Bad, // Op_RegI
3999 RAX_H_num, // Op_RegP
4000 OptoReg::Bad, // Op_RegF
4001 XMM0_H_num, // Op_RegD
4002 RAX_H_num // Op_RegL
4003 };
4004 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4005 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4006 %}
4007 %}
4008
4009 //----------ATTRIBUTES---------------------------------------------------------
4010 //----------Operand Attributes-------------------------------------------------
4011 op_attrib op_cost(0); // Required cost attribute
4012
4013 //----------Instruction Attributes---------------------------------------------
4014 ins_attrib ins_cost(100); // Required cost attribute
4015 ins_attrib ins_size(8); // Required size attribute (in bits)
4016 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4017 // a non-matching short branch variant
4018 // of some long branch?
4019 ins_attrib ins_alignment(1); // Required alignment attribute (must
4020 // be a power of 2) specifies the
4021 // alignment that some part of the
4022 // instruction (not necessarily the
4023 // start) requires. If > 1, a
4024 // compute_padding() function must be
4025 // provided for the instruction
4026
4027 //----------OPERANDS-----------------------------------------------------------
4028 // Operand definitions must precede instruction definitions for correct parsing
4029 // in the ADLC because operands constitute user defined types which are used in
4030 // instruction definitions.
4031
4032 //----------Simple Operands----------------------------------------------------
4033 // Immediate Operands
4034 // Integer Immediate
4035 operand immI()
4036 %{
4037 match(ConI);
4038
4039 op_cost(10);
4040 format %{ %}
4041 interface(CONST_INTER);
4042 %}
4043
4044 // Constant for test vs zero
4045 operand immI0()
4046 %{
4047 predicate(n->get_int() == 0);
4048 match(ConI);
4049
4050 op_cost(0);
4051 format %{ %}
4052 interface(CONST_INTER);
4053 %}
4054
4055 // Constant for increment
4056 operand immI1()
4057 %{
4058 predicate(n->get_int() == 1);
4059 match(ConI);
4060
4061 op_cost(0);
4062 format %{ %}
4063 interface(CONST_INTER);
4064 %}
4065
4066 // Constant for decrement
4067 operand immI_M1()
4068 %{
4069 predicate(n->get_int() == -1);
4070 match(ConI);
4071
4072 op_cost(0);
4073 format %{ %}
4074 interface(CONST_INTER);
4075 %}
4076
4077 // Valid scale values for addressing modes
4078 operand immI2()
4079 %{
4080 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4081 match(ConI);
4082
4083 format %{ %}
4084 interface(CONST_INTER);
4085 %}
4086
4087 operand immI8()
4088 %{
4089 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4090 match(ConI);
4091
4092 op_cost(5);
4093 format %{ %}
4094 interface(CONST_INTER);
4095 %}
4096
4097 operand immI16()
4098 %{
4099 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4100 match(ConI);
4101
4102 op_cost(10);
4103 format %{ %}
4104 interface(CONST_INTER);
4105 %}
4106
4107 // Constant for long shifts
4108 operand immI_32()
4109 %{
4110 predicate( n->get_int() == 32 );
4111 match(ConI);
4112
4113 op_cost(0);
4114 format %{ %}
4115 interface(CONST_INTER);
4116 %}
4117
4118 // Constant for long shifts
4119 operand immI_64()
4120 %{
4121 predicate( n->get_int() == 64 );
4122 match(ConI);
4123
4124 op_cost(0);
4125 format %{ %}
4126 interface(CONST_INTER);
4127 %}
4128
4129 // Pointer Immediate
4130 operand immP()
4131 %{
4132 match(ConP);
4133
4134 op_cost(10);
4135 format %{ %}
4136 interface(CONST_INTER);
4137 %}
4138
4139 // NULL Pointer Immediate
4140 operand immP0()
4141 %{
4142 predicate(n->get_ptr() == 0);
4143 match(ConP);
4144
4145 op_cost(5);
4146 format %{ %}
4147 interface(CONST_INTER);
4148 %}
4149
4150 operand immP_poll() %{
4151 predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
4152 match(ConP);
4153
4154 // formats are generated automatically for constants and base registers
4155 format %{ %}
4156 interface(CONST_INTER);
4157 %}
4158
4159 // Pointer Immediate
4160 operand immN() %{
4161 match(ConN);
4162
4163 op_cost(10);
4164 format %{ %}
4165 interface(CONST_INTER);
4166 %}
4167
4168 // NULL Pointer Immediate
4169 operand immN0() %{
4170 predicate(n->get_narrowcon() == 0);
4171 match(ConN);
4172
4173 op_cost(5);
4174 format %{ %}
4175 interface(CONST_INTER);
4176 %}
4177
4178 operand immP31()
4179 %{
4180 predicate(!n->as_Type()->type()->isa_oopptr()
4181 && (n->get_ptr() >> 31) == 0);
4182 match(ConP);
4183
4184 op_cost(5);
4185 format %{ %}
4186 interface(CONST_INTER);
4187 %}
4188
4189
4190 // Long Immediate
4191 operand immL()
4192 %{
4193 match(ConL);
4194
4195 op_cost(20);
4196 format %{ %}
4197 interface(CONST_INTER);
4198 %}
4199
4200 // Long Immediate 8-bit
4201 operand immL8()
4202 %{
4203 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4204 match(ConL);
4205
4206 op_cost(5);
4207 format %{ %}
4208 interface(CONST_INTER);
4209 %}
4210
4211 // Long Immediate 32-bit unsigned
4212 operand immUL32()
4213 %{
4214 predicate(n->get_long() == (unsigned int) (n->get_long()));
4215 match(ConL);
4216
4217 op_cost(10);
4218 format %{ %}
4219 interface(CONST_INTER);
4220 %}
4221
4222 // Long Immediate 32-bit signed
4223 operand immL32()
4224 %{
4225 predicate(n->get_long() == (int) (n->get_long()));
4226 match(ConL);
4227
4228 op_cost(15);
4229 format %{ %}
4230 interface(CONST_INTER);
4231 %}
4232
4233 // Long Immediate zero
4234 operand immL0()
4235 %{
4236 predicate(n->get_long() == 0L);
4237 match(ConL);
4238
4239 op_cost(10);
4240 format %{ %}
4241 interface(CONST_INTER);
4242 %}
4243
4244 // Constant for increment
4245 operand immL1()
4246 %{
4247 predicate(n->get_long() == 1);
4248 match(ConL);
4249
4250 format %{ %}
4251 interface(CONST_INTER);
4252 %}
4253
4254 // Constant for decrement
4255 operand immL_M1()
4256 %{
4257 predicate(n->get_long() == -1);
4258 match(ConL);
4259
4260 format %{ %}
4261 interface(CONST_INTER);
4262 %}
4263
4264 // Long Immediate: the value 10
4265 operand immL10()
4266 %{
4267 predicate(n->get_long() == 10);
4268 match(ConL);
4269
4270 format %{ %}
4271 interface(CONST_INTER);
4272 %}
4273
4274 // Long immediate from 0 to 127.
4275 // Used for a shorter form of long mul by 10.
4276 operand immL_127()
4277 %{
4278 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4279 match(ConL);
4280
4281 op_cost(10);
4282 format %{ %}
4283 interface(CONST_INTER);
4284 %}
4285
4286 // Long Immediate: low 32-bit mask
4287 operand immL_32bits()
4288 %{
4289 predicate(n->get_long() == 0xFFFFFFFFL);
4290 match(ConL);
4291 op_cost(20);
4292
4293 format %{ %}
4294 interface(CONST_INTER);
4295 %}
4296
4297 // Float Immediate zero
4298 operand immF0()
4299 %{
4300 predicate(jint_cast(n->getf()) == 0);
4301 match(ConF);
4302
4303 op_cost(5);
4304 format %{ %}
4305 interface(CONST_INTER);
4306 %}
4307
4308 // Float Immediate
4309 operand immF()
4310 %{
4311 match(ConF);
4312
4313 op_cost(15);
4314 format %{ %}
4315 interface(CONST_INTER);
4316 %}
4317
4318 // Double Immediate zero
4319 operand immD0()
4320 %{
4321 predicate(jlong_cast(n->getd()) == 0);
4322 match(ConD);
4323
4324 op_cost(5);
4325 format %{ %}
4326 interface(CONST_INTER);
4327 %}
4328
4329 // Double Immediate
4330 operand immD()
4331 %{
4332 match(ConD);
4333
4334 op_cost(15);
4335 format %{ %}
4336 interface(CONST_INTER);
4337 %}
4338
4339 // Immediates for special shifts (sign extend)
4340
4341 // Constants for increment
4342 operand immI_16()
4343 %{
4344 predicate(n->get_int() == 16);
4345 match(ConI);
4346
4347 format %{ %}
4348 interface(CONST_INTER);
4349 %}
4350
4351 operand immI_24()
4352 %{
4353 predicate(n->get_int() == 24);
4354 match(ConI);
4355
4356 format %{ %}
4357 interface(CONST_INTER);
4358 %}
4359
4360 // Constant for byte-wide masking
4361 operand immI_255()
4362 %{
4363 predicate(n->get_int() == 255);
4364 match(ConI);
4365
4366 format %{ %}
4367 interface(CONST_INTER);
4368 %}
4369
4370 // Constant for short-wide masking
4371 operand immI_65535()
4372 %{
4373 predicate(n->get_int() == 65535);
4374 match(ConI);
4375
4376 format %{ %}
4377 interface(CONST_INTER);
4378 %}
4379
4380 // Constant for byte-wide masking
4381 operand immL_255()
4382 %{
4383 predicate(n->get_long() == 255);
4384 match(ConL);
4385
4386 format %{ %}
4387 interface(CONST_INTER);
4388 %}
4389
4390 // Constant for short-wide masking
4391 operand immL_65535()
4392 %{
4393 predicate(n->get_long() == 65535);
4394 match(ConL);
4395
4396 format %{ %}
4397 interface(CONST_INTER);
4398 %}
4399
4400 // Register Operands
4401 // Integer Register
4402 operand rRegI()
4403 %{
4404 constraint(ALLOC_IN_RC(int_reg));
4405 match(RegI);
4406
4407 match(rax_RegI);
4408 match(rbx_RegI);
4409 match(rcx_RegI);
4410 match(rdx_RegI);
4411 match(rdi_RegI);
4412
4413 format %{ %}
4414 interface(REG_INTER);
4415 %}
4416
4417 // Special Registers
4418 operand rax_RegI()
4419 %{
4420 constraint(ALLOC_IN_RC(int_rax_reg));
4421 match(RegI);
4422 match(rRegI);
4423
4424 format %{ "RAX" %}
4425 interface(REG_INTER);
4426 %}
4427
4428 // Special Registers
4429 operand rbx_RegI()
4430 %{
4431 constraint(ALLOC_IN_RC(int_rbx_reg));
4432 match(RegI);
4433 match(rRegI);
4434
4435 format %{ "RBX" %}
4436 interface(REG_INTER);
4437 %}
4438
4439 operand rcx_RegI()
4440 %{
4441 constraint(ALLOC_IN_RC(int_rcx_reg));
4442 match(RegI);
4443 match(rRegI);
4444
4445 format %{ "RCX" %}
4446 interface(REG_INTER);
4447 %}
4448
4449 operand rdx_RegI()
4450 %{
4451 constraint(ALLOC_IN_RC(int_rdx_reg));
4452 match(RegI);
4453 match(rRegI);
4454
4455 format %{ "RDX" %}
4456 interface(REG_INTER);
4457 %}
4458
4459 operand rdi_RegI()
4460 %{
4461 constraint(ALLOC_IN_RC(int_rdi_reg));
4462 match(RegI);
4463 match(rRegI);
4464
4465 format %{ "RDI" %}
4466 interface(REG_INTER);
4467 %}
4468
4469 operand no_rcx_RegI()
4470 %{
4471 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4472 match(RegI);
4473 match(rax_RegI);
4474 match(rbx_RegI);
4475 match(rdx_RegI);
4476 match(rdi_RegI);
4477
4478 format %{ %}
4479 interface(REG_INTER);
4480 %}
4481
4482 operand no_rax_rdx_RegI()
4483 %{
4484 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4485 match(RegI);
4486 match(rbx_RegI);
4487 match(rcx_RegI);
4488 match(rdi_RegI);
4489
4490 format %{ %}
4491 interface(REG_INTER);
4492 %}
4493
4494 // Pointer Register
4495 operand any_RegP()
4496 %{
4497 constraint(ALLOC_IN_RC(any_reg));
4498 match(RegP);
4499 match(rax_RegP);
4500 match(rbx_RegP);
4501 match(rdi_RegP);
4502 match(rsi_RegP);
4503 match(rbp_RegP);
4504 match(r15_RegP);
4505 match(rRegP);
4506
4507 format %{ %}
4508 interface(REG_INTER);
4509 %}
4510
4511 operand rRegP()
4512 %{
4513 constraint(ALLOC_IN_RC(ptr_reg));
4514 match(RegP);
4515 match(rax_RegP);
4516 match(rbx_RegP);
4517 match(rdi_RegP);
4518 match(rsi_RegP);
4519 match(rbp_RegP);
4520 match(r15_RegP); // See Q&A below about r15_RegP.
4521
4522 format %{ %}
4523 interface(REG_INTER);
4524 %}
4525
4526 operand rRegN() %{
4527 constraint(ALLOC_IN_RC(int_reg));
4528 match(RegN);
4529
4530 format %{ %}
4531 interface(REG_INTER);
4532 %}
4533
4534 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4535 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4536 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4537 // The output of an instruction is controlled by the allocator, which respects
4538 // register class masks, not match rules. Unless an instruction mentions
4539 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4540 // by the allocator as an input.
4541
4542 operand no_rax_RegP()
4543 %{
4544 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4545 match(RegP);
4546 match(rbx_RegP);
4547 match(rsi_RegP);
4548 match(rdi_RegP);
4549
4550 format %{ %}
4551 interface(REG_INTER);
4552 %}
4553
4554 operand no_rbp_RegP()
4555 %{
4556 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4557 match(RegP);
4558 match(rbx_RegP);
4559 match(rsi_RegP);
4560 match(rdi_RegP);
4561
4562 format %{ %}
4563 interface(REG_INTER);
4564 %}
4565
4566 operand no_rax_rbx_RegP()
4567 %{
4568 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4569 match(RegP);
4570 match(rsi_RegP);
4571 match(rdi_RegP);
4572
4573 format %{ %}
4574 interface(REG_INTER);
4575 %}
4576
4577 // Special Registers
4578 // Return a pointer value
4579 operand rax_RegP()
4580 %{
4581 constraint(ALLOC_IN_RC(ptr_rax_reg));
4582 match(RegP);
4583 match(rRegP);
4584
4585 format %{ %}
4586 interface(REG_INTER);
4587 %}
4588
4589 // Special Registers
4590 // Return a compressed pointer value
4591 operand rax_RegN()
4592 %{
4593 constraint(ALLOC_IN_RC(int_rax_reg));
4594 match(RegN);
4595 match(rRegN);
4596
4597 format %{ %}
4598 interface(REG_INTER);
4599 %}
4600
4601 // Used in AtomicAdd
4602 operand rbx_RegP()
4603 %{
4604 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4605 match(RegP);
4606 match(rRegP);
4607
4608 format %{ %}
4609 interface(REG_INTER);
4610 %}
4611
4612 operand rsi_RegP()
4613 %{
4614 constraint(ALLOC_IN_RC(ptr_rsi_reg));
4615 match(RegP);
4616 match(rRegP);
4617
4618 format %{ %}
4619 interface(REG_INTER);
4620 %}
4621
4622 // Used in rep stosq
4623 operand rdi_RegP()
4624 %{
4625 constraint(ALLOC_IN_RC(ptr_rdi_reg));
4626 match(RegP);
4627 match(rRegP);
4628
4629 format %{ %}
4630 interface(REG_INTER);
4631 %}
4632
4633 operand rbp_RegP()
4634 %{
4635 constraint(ALLOC_IN_RC(ptr_rbp_reg));
4636 match(RegP);
4637 match(rRegP);
4638
4639 format %{ %}
4640 interface(REG_INTER);
4641 %}
4642
4643 operand r15_RegP()
4644 %{
4645 constraint(ALLOC_IN_RC(ptr_r15_reg));
4646 match(RegP);
4647 match(rRegP);
4648
4649 format %{ %}
4650 interface(REG_INTER);
4651 %}
4652
4653 operand rRegL()
4654 %{
4655 constraint(ALLOC_IN_RC(long_reg));
4656 match(RegL);
4657 match(rax_RegL);
4658 match(rdx_RegL);
4659
4660 format %{ %}
4661 interface(REG_INTER);
4662 %}
4663
4664 // Special Registers
4665 operand no_rax_rdx_RegL()
4666 %{
4667 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4668 match(RegL);
4669 match(rRegL);
4670
4671 format %{ %}
4672 interface(REG_INTER);
4673 %}
4674
4675 operand no_rax_RegL()
4676 %{
4677 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4678 match(RegL);
4679 match(rRegL);
4680 match(rdx_RegL);
4681
4682 format %{ %}
4683 interface(REG_INTER);
4684 %}
4685
4686 operand no_rcx_RegL()
4687 %{
4688 constraint(ALLOC_IN_RC(long_no_rcx_reg));
4689 match(RegL);
4690 match(rRegL);
4691
4692 format %{ %}
4693 interface(REG_INTER);
4694 %}
4695
4696 operand rax_RegL()
4697 %{
4698 constraint(ALLOC_IN_RC(long_rax_reg));
4699 match(RegL);
4700 match(rRegL);
4701
4702 format %{ "RAX" %}
4703 interface(REG_INTER);
4704 %}
4705
4706 operand rcx_RegL()
4707 %{
4708 constraint(ALLOC_IN_RC(long_rcx_reg));
4709 match(RegL);
4710 match(rRegL);
4711
4712 format %{ %}
4713 interface(REG_INTER);
4714 %}
4715
4716 operand rdx_RegL()
4717 %{
4718 constraint(ALLOC_IN_RC(long_rdx_reg));
4719 match(RegL);
4720 match(rRegL);
4721
4722 format %{ %}
4723 interface(REG_INTER);
4724 %}
4725
4726 // Flags register, used as output of compare instructions
4727 operand rFlagsReg()
4728 %{
4729 constraint(ALLOC_IN_RC(int_flags));
4730 match(RegFlags);
4731
4732 format %{ "RFLAGS" %}
4733 interface(REG_INTER);
4734 %}
4735
4736 // Flags register, used as output of FLOATING POINT compare instructions
4737 operand rFlagsRegU()
4738 %{
4739 constraint(ALLOC_IN_RC(int_flags));
4740 match(RegFlags);
4741
4742 format %{ "RFLAGS_U" %}
4743 interface(REG_INTER);
4744 %}
4745
4746 operand rFlagsRegUCF() %{
4747 constraint(ALLOC_IN_RC(int_flags));
4748 match(RegFlags);
4749 predicate(false);
4750
4751 format %{ "RFLAGS_U_CF" %}
4752 interface(REG_INTER);
4753 %}
4754
4755 // Float register operands
4756 operand regF()
4757 %{
4758 constraint(ALLOC_IN_RC(float_reg));
4759 match(RegF);
4760
4761 format %{ %}
4762 interface(REG_INTER);
4763 %}
4764
4765 // Double register operands
4766 operand regD()
4767 %{
4768 constraint(ALLOC_IN_RC(double_reg));
4769 match(RegD);
4770
4771 format %{ %}
4772 interface(REG_INTER);
4773 %}
4774
4775
4776 //----------Memory Operands----------------------------------------------------
4777 // Direct Memory Operand
4778 // operand direct(immP addr)
4779 // %{
4780 // match(addr);
4781
4782 // format %{ "[$addr]" %}
4783 // interface(MEMORY_INTER) %{
4784 // base(0xFFFFFFFF);
4785 // index(0x4);
4786 // scale(0x0);
4787 // disp($addr);
4788 // %}
4789 // %}
4790
4791 // Indirect Memory Operand
4792 operand indirect(any_RegP reg)
4793 %{
4794 constraint(ALLOC_IN_RC(ptr_reg));
4795 match(reg);
4796
4797 format %{ "[$reg]" %}
4798 interface(MEMORY_INTER) %{
4799 base($reg);
4800 index(0x4);
4801 scale(0x0);
4802 disp(0x0);
4803 %}
4804 %}
4805
4806 // Indirect Memory Plus Short Offset Operand
4807 operand indOffset8(any_RegP reg, immL8 off)
4808 %{
4809 constraint(ALLOC_IN_RC(ptr_reg));
4810 match(AddP reg off);
4811
4812 format %{ "[$reg + $off (8-bit)]" %}
4813 interface(MEMORY_INTER) %{
4814 base($reg);
4815 index(0x4);
4816 scale(0x0);
4817 disp($off);
4818 %}
4819 %}
4820
4821 // Indirect Memory Plus Long Offset Operand
4822 operand indOffset32(any_RegP reg, immL32 off)
4823 %{
4824 constraint(ALLOC_IN_RC(ptr_reg));
4825 match(AddP reg off);
4826
4827 format %{ "[$reg + $off (32-bit)]" %}
4828 interface(MEMORY_INTER) %{
4829 base($reg);
4830 index(0x4);
4831 scale(0x0);
4832 disp($off);
4833 %}
4834 %}
4835
4836 // Indirect Memory Plus Index Register Plus Offset Operand
4837 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
4838 %{
4839 constraint(ALLOC_IN_RC(ptr_reg));
4840 match(AddP (AddP reg lreg) off);
4841
4842 op_cost(10);
4843 format %{"[$reg + $off + $lreg]" %}
4844 interface(MEMORY_INTER) %{
4845 base($reg);
4846 index($lreg);
4847 scale(0x0);
4848 disp($off);
4849 %}
4850 %}
4851
4852 // Indirect Memory Plus Index Register Plus Offset Operand
4853 operand indIndex(any_RegP reg, rRegL lreg)
4854 %{
4855 constraint(ALLOC_IN_RC(ptr_reg));
4856 match(AddP reg lreg);
4857
4858 op_cost(10);
4859 format %{"[$reg + $lreg]" %}
4860 interface(MEMORY_INTER) %{
4861 base($reg);
4862 index($lreg);
4863 scale(0x0);
4864 disp(0x0);
4865 %}
4866 %}
4867
4868 // Indirect Memory Times Scale Plus Index Register
4869 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
4870 %{
4871 constraint(ALLOC_IN_RC(ptr_reg));
4872 match(AddP reg (LShiftL lreg scale));
4873
4874 op_cost(10);
4875 format %{"[$reg + $lreg << $scale]" %}
4876 interface(MEMORY_INTER) %{
4877 base($reg);
4878 index($lreg);
4879 scale($scale);
4880 disp(0x0);
4881 %}
4882 %}
4883
4884 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4885 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
4886 %{
4887 constraint(ALLOC_IN_RC(ptr_reg));
4888 match(AddP (AddP reg (LShiftL lreg scale)) off);
4889
4890 op_cost(10);
4891 format %{"[$reg + $off + $lreg << $scale]" %}
4892 interface(MEMORY_INTER) %{
4893 base($reg);
4894 index($lreg);
4895 scale($scale);
4896 disp($off);
4897 %}
4898 %}
4899
4900 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
4901 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
4902 %{
4903 constraint(ALLOC_IN_RC(ptr_reg));
4904 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
4905 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
4906
4907 op_cost(10);
4908 format %{"[$reg + $off + $idx << $scale]" %}
4909 interface(MEMORY_INTER) %{
4910 base($reg);
4911 index($idx);
4912 scale($scale);
4913 disp($off);
4914 %}
4915 %}
4916
4917 // Indirect Narrow Oop Plus Offset Operand
4918 // Note: x86 architecture doesn't support "scale * index + offset" without a base
4919 // we can't free r12 even with Universe::narrow_oop_base() == NULL.
4920 operand indCompressedOopOffset(rRegN reg, immL32 off) %{
4921 predicate(UseCompressedOops && (Universe::narrow_oop_shift() == Address::times_8));
4922 constraint(ALLOC_IN_RC(ptr_reg));
4923 match(AddP (DecodeN reg) off);
4924
4925 op_cost(10);
4926 format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %}
4927 interface(MEMORY_INTER) %{
4928 base(0xc); // R12
4929 index($reg);
4930 scale(0x3);
4931 disp($off);
4932 %}
4933 %}
4934
4935 // Indirect Memory Operand
4936 operand indirectNarrow(rRegN reg)
4937 %{
4938 predicate(Universe::narrow_oop_shift() == 0);
4939 constraint(ALLOC_IN_RC(ptr_reg));
4940 match(DecodeN reg);
4941
4942 format %{ "[$reg]" %}
4943 interface(MEMORY_INTER) %{
4944 base($reg);
4945 index(0x4);
4946 scale(0x0);
4947 disp(0x0);
4948 %}
4949 %}
4950
4951 // Indirect Memory Plus Short Offset Operand
4952 operand indOffset8Narrow(rRegN reg, immL8 off)
4953 %{
4954 predicate(Universe::narrow_oop_shift() == 0);
4955 constraint(ALLOC_IN_RC(ptr_reg));
4956 match(AddP (DecodeN reg) off);
4957
4958 format %{ "[$reg + $off (8-bit)]" %}
4959 interface(MEMORY_INTER) %{
4960 base($reg);
4961 index(0x4);
4962 scale(0x0);
4963 disp($off);
4964 %}
4965 %}
4966
4967 // Indirect Memory Plus Long Offset Operand
4968 operand indOffset32Narrow(rRegN reg, immL32 off)
4969 %{
4970 predicate(Universe::narrow_oop_shift() == 0);
4971 constraint(ALLOC_IN_RC(ptr_reg));
4972 match(AddP (DecodeN reg) off);
4973
4974 format %{ "[$reg + $off (32-bit)]" %}
4975 interface(MEMORY_INTER) %{
4976 base($reg);
4977 index(0x4);
4978 scale(0x0);
4979 disp($off);
4980 %}
4981 %}
4982
4983 // Indirect Memory Plus Index Register Plus Offset Operand
4984 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off)
4985 %{
4986 predicate(Universe::narrow_oop_shift() == 0);
4987 constraint(ALLOC_IN_RC(ptr_reg));
4988 match(AddP (AddP (DecodeN reg) lreg) off);
4989
4990 op_cost(10);
4991 format %{"[$reg + $off + $lreg]" %}
4992 interface(MEMORY_INTER) %{
4993 base($reg);
4994 index($lreg);
4995 scale(0x0);
4996 disp($off);
4997 %}
4998 %}
4999
5000 // Indirect Memory Plus Index Register Plus Offset Operand
5001 operand indIndexNarrow(rRegN reg, rRegL lreg)
5002 %{
5003 predicate(Universe::narrow_oop_shift() == 0);
5004 constraint(ALLOC_IN_RC(ptr_reg));
5005 match(AddP (DecodeN reg) lreg);
5006
5007 op_cost(10);
5008 format %{"[$reg + $lreg]" %}
5009 interface(MEMORY_INTER) %{
5010 base($reg);
5011 index($lreg);
5012 scale(0x0);
5013 disp(0x0);
5014 %}
5015 %}
5016
5017 // Indirect Memory Times Scale Plus Index Register
5018 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale)
5019 %{
5020 predicate(Universe::narrow_oop_shift() == 0);
5021 constraint(ALLOC_IN_RC(ptr_reg));
5022 match(AddP (DecodeN reg) (LShiftL lreg scale));
5023
5024 op_cost(10);
5025 format %{"[$reg + $lreg << $scale]" %}
5026 interface(MEMORY_INTER) %{
5027 base($reg);
5028 index($lreg);
5029 scale($scale);
5030 disp(0x0);
5031 %}
5032 %}
5033
5034 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5035 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
5036 %{
5037 predicate(Universe::narrow_oop_shift() == 0);
5038 constraint(ALLOC_IN_RC(ptr_reg));
5039 match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off);
5040
5041 op_cost(10);
5042 format %{"[$reg + $off + $lreg << $scale]" %}
5043 interface(MEMORY_INTER) %{
5044 base($reg);
5045 index($lreg);
5046 scale($scale);
5047 disp($off);
5048 %}
5049 %}
5050
5051 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5052 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale)
5053 %{
5054 constraint(ALLOC_IN_RC(ptr_reg));
5055 predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5056 match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off);
5057
5058 op_cost(10);
5059 format %{"[$reg + $off + $idx << $scale]" %}
5060 interface(MEMORY_INTER) %{
5061 base($reg);
5062 index($idx);
5063 scale($scale);
5064 disp($off);
5065 %}
5066 %}
5067
5068
5069 //----------Special Memory Operands--------------------------------------------
5070 // Stack Slot Operand - This operand is used for loading and storing temporary
5071 // values on the stack where a match requires a value to
5072 // flow through memory.
5073 operand stackSlotP(sRegP reg)
5074 %{
5075 constraint(ALLOC_IN_RC(stack_slots));
5076 // No match rule because this operand is only generated in matching
5077
5078 format %{ "[$reg]" %}
5079 interface(MEMORY_INTER) %{
5080 base(0x4); // RSP
5081 index(0x4); // No Index
5082 scale(0x0); // No Scale
5083 disp($reg); // Stack Offset
5084 %}
5085 %}
5086
5087 operand stackSlotI(sRegI reg)
5088 %{
5089 constraint(ALLOC_IN_RC(stack_slots));
5090 // No match rule because this operand is only generated in matching
5091
5092 format %{ "[$reg]" %}
5093 interface(MEMORY_INTER) %{
5094 base(0x4); // RSP
5095 index(0x4); // No Index
5096 scale(0x0); // No Scale
5097 disp($reg); // Stack Offset
5098 %}
5099 %}
5100
5101 operand stackSlotF(sRegF reg)
5102 %{
5103 constraint(ALLOC_IN_RC(stack_slots));
5104 // No match rule because this operand is only generated in matching
5105
5106 format %{ "[$reg]" %}
5107 interface(MEMORY_INTER) %{
5108 base(0x4); // RSP
5109 index(0x4); // No Index
5110 scale(0x0); // No Scale
5111 disp($reg); // Stack Offset
5112 %}
5113 %}
5114
5115 operand stackSlotD(sRegD reg)
5116 %{
5117 constraint(ALLOC_IN_RC(stack_slots));
5118 // No match rule because this operand is only generated in matching
5119
5120 format %{ "[$reg]" %}
5121 interface(MEMORY_INTER) %{
5122 base(0x4); // RSP
5123 index(0x4); // No Index
5124 scale(0x0); // No Scale
5125 disp($reg); // Stack Offset
5126 %}
5127 %}
5128 operand stackSlotL(sRegL reg)
5129 %{
5130 constraint(ALLOC_IN_RC(stack_slots));
5131 // No match rule because this operand is only generated in matching
5132
5133 format %{ "[$reg]" %}
5134 interface(MEMORY_INTER) %{
5135 base(0x4); // RSP
5136 index(0x4); // No Index
5137 scale(0x0); // No Scale
5138 disp($reg); // Stack Offset
5139 %}
5140 %}
5141
5142 //----------Conditional Branch Operands----------------------------------------
5143 // Comparison Op - This is the operation of the comparison, and is limited to
5144 // the following set of codes:
5145 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5146 //
5147 // Other attributes of the comparison, such as unsignedness, are specified
5148 // by the comparison instruction that sets a condition code flags register.
5149 // That result is represented by a flags operand whose subtype is appropriate
5150 // to the unsignedness (etc.) of the comparison.
5151 //
5152 // Later, the instruction which matches both the Comparison Op (a Bool) and
5153 // the flags (produced by the Cmp) specifies the coding of the comparison op
5154 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5155
5156 // Comparision Code
5157 operand cmpOp()
5158 %{
5159 match(Bool);
5160
5161 format %{ "" %}
5162 interface(COND_INTER) %{
5163 equal(0x4, "e");
5164 not_equal(0x5, "ne");
5165 less(0xC, "l");
5166 greater_equal(0xD, "ge");
5167 less_equal(0xE, "le");
5168 greater(0xF, "g");
5169 %}
5170 %}
5171
5172 // Comparison Code, unsigned compare. Used by FP also, with
5173 // C2 (unordered) turned into GT or LT already. The other bits
5174 // C0 and C3 are turned into Carry & Zero flags.
5175 operand cmpOpU()
5176 %{
5177 match(Bool);
5178
5179 format %{ "" %}
5180 interface(COND_INTER) %{
5181 equal(0x4, "e");
5182 not_equal(0x5, "ne");
5183 less(0x2, "b");
5184 greater_equal(0x3, "nb");
5185 less_equal(0x6, "be");
5186 greater(0x7, "nbe");
5187 %}
5188 %}
5189
5190
5191 // Floating comparisons that don't require any fixup for the unordered case
5192 operand cmpOpUCF() %{
5193 match(Bool);
5194 predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5195 n->as_Bool()->_test._test == BoolTest::ge ||
5196 n->as_Bool()->_test._test == BoolTest::le ||
5197 n->as_Bool()->_test._test == BoolTest::gt);
5198 format %{ "" %}
5199 interface(COND_INTER) %{
5200 equal(0x4, "e");
5201 not_equal(0x5, "ne");
5202 less(0x2, "b");
5203 greater_equal(0x3, "nb");
5204 less_equal(0x6, "be");
5205 greater(0x7, "nbe");
5206 %}
5207 %}
5208
5209
5210 // Floating comparisons that can be fixed up with extra conditional jumps
5211 operand cmpOpUCF2() %{
5212 match(Bool);
5213 predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5214 n->as_Bool()->_test._test == BoolTest::eq);
5215 format %{ "" %}
5216 interface(COND_INTER) %{
5217 equal(0x4, "e");
5218 not_equal(0x5, "ne");
5219 less(0x2, "b");
5220 greater_equal(0x3, "nb");
5221 less_equal(0x6, "be");
5222 greater(0x7, "nbe");
5223 %}
5224 %}
5225
5226
5227 //----------OPERAND CLASSES----------------------------------------------------
5228 // Operand Classes are groups of operands that are used as to simplify
5229 // instruction definitions by not requiring the AD writer to specify separate
5230 // instructions for every form of operand when the instruction accepts
5231 // multiple operand types with the same basic encoding and format. The classic
5232 // case of this is memory operands.
5233
5234 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5235 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5236 indCompressedOopOffset,
5237 indirectNarrow, indOffset8Narrow, indOffset32Narrow,
5238 indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
5239 indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
5240
5241 //----------PIPELINE-----------------------------------------------------------
5242 // Rules which define the behavior of the target architectures pipeline.
5243 pipeline %{
5244
5245 //----------ATTRIBUTES---------------------------------------------------------
5246 attributes %{
5247 variable_size_instructions; // Fixed size instructions
5248 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5249 instruction_unit_size = 1; // An instruction is 1 bytes long
5250 instruction_fetch_unit_size = 16; // The processor fetches one line
5251 instruction_fetch_units = 1; // of 16 bytes
5252
5253 // List of nop instructions
5254 nops( MachNop );
5255 %}
5256
5257 //----------RESOURCES----------------------------------------------------------
5258 // Resources are the functional units available to the machine
5259
5260 // Generic P2/P3 pipeline
5261 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5262 // 3 instructions decoded per cycle.
5263 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5264 // 3 ALU op, only ALU0 handles mul instructions.
5265 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5266 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5267 BR, FPU,
5268 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5269
5270 //----------PIPELINE DESCRIPTION-----------------------------------------------
5271 // Pipeline Description specifies the stages in the machine's pipeline
5272
5273 // Generic P2/P3 pipeline
5274 pipe_desc(S0, S1, S2, S3, S4, S5);
5275
5276 //----------PIPELINE CLASSES---------------------------------------------------
5277 // Pipeline Classes describe the stages in which input and output are
5278 // referenced by the hardware pipeline.
5279
5280 // Naming convention: ialu or fpu
5281 // Then: _reg
5282 // Then: _reg if there is a 2nd register
5283 // Then: _long if it's a pair of instructions implementing a long
5284 // Then: _fat if it requires the big decoder
5285 // Or: _mem if it requires the big decoder and a memory unit.
5286
5287 // Integer ALU reg operation
5288 pipe_class ialu_reg(rRegI dst)
5289 %{
5290 single_instruction;
5291 dst : S4(write);
5292 dst : S3(read);
5293 DECODE : S0; // any decoder
5294 ALU : S3; // any alu
5295 %}
5296
5297 // Long ALU reg operation
5298 pipe_class ialu_reg_long(rRegL dst)
5299 %{
5300 instruction_count(2);
5301 dst : S4(write);
5302 dst : S3(read);
5303 DECODE : S0(2); // any 2 decoders
5304 ALU : S3(2); // both alus
5305 %}
5306
5307 // Integer ALU reg operation using big decoder
5308 pipe_class ialu_reg_fat(rRegI dst)
5309 %{
5310 single_instruction;
5311 dst : S4(write);
5312 dst : S3(read);
5313 D0 : S0; // big decoder only
5314 ALU : S3; // any alu
5315 %}
5316
5317 // Long ALU reg operation using big decoder
5318 pipe_class ialu_reg_long_fat(rRegL dst)
5319 %{
5320 instruction_count(2);
5321 dst : S4(write);
5322 dst : S3(read);
5323 D0 : S0(2); // big decoder only; twice
5324 ALU : S3(2); // any 2 alus
5325 %}
5326
5327 // Integer ALU reg-reg operation
5328 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5329 %{
5330 single_instruction;
5331 dst : S4(write);
5332 src : S3(read);
5333 DECODE : S0; // any decoder
5334 ALU : S3; // any alu
5335 %}
5336
5337 // Long ALU reg-reg operation
5338 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5339 %{
5340 instruction_count(2);
5341 dst : S4(write);
5342 src : S3(read);
5343 DECODE : S0(2); // any 2 decoders
5344 ALU : S3(2); // both alus
5345 %}
5346
5347 // Integer ALU reg-reg operation
5348 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5349 %{
5350 single_instruction;
5351 dst : S4(write);
5352 src : S3(read);
5353 D0 : S0; // big decoder only
5354 ALU : S3; // any alu
5355 %}
5356
5357 // Long ALU reg-reg operation
5358 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5359 %{
5360 instruction_count(2);
5361 dst : S4(write);
5362 src : S3(read);
5363 D0 : S0(2); // big decoder only; twice
5364 ALU : S3(2); // both alus
5365 %}
5366
5367 // Integer ALU reg-mem operation
5368 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5369 %{
5370 single_instruction;
5371 dst : S5(write);
5372 mem : S3(read);
5373 D0 : S0; // big decoder only
5374 ALU : S4; // any alu
5375 MEM : S3; // any mem
5376 %}
5377
5378 // Integer mem operation (prefetch)
5379 pipe_class ialu_mem(memory mem)
5380 %{
5381 single_instruction;
5382 mem : S3(read);
5383 D0 : S0; // big decoder only
5384 MEM : S3; // any mem
5385 %}
5386
5387 // Integer Store to Memory
5388 pipe_class ialu_mem_reg(memory mem, rRegI src)
5389 %{
5390 single_instruction;
5391 mem : S3(read);
5392 src : S5(read);
5393 D0 : S0; // big decoder only
5394 ALU : S4; // any alu
5395 MEM : S3;
5396 %}
5397
5398 // // Long Store to Memory
5399 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5400 // %{
5401 // instruction_count(2);
5402 // mem : S3(read);
5403 // src : S5(read);
5404 // D0 : S0(2); // big decoder only; twice
5405 // ALU : S4(2); // any 2 alus
5406 // MEM : S3(2); // Both mems
5407 // %}
5408
5409 // Integer Store to Memory
5410 pipe_class ialu_mem_imm(memory mem)
5411 %{
5412 single_instruction;
5413 mem : S3(read);
5414 D0 : S0; // big decoder only
5415 ALU : S4; // any alu
5416 MEM : S3;
5417 %}
5418
5419 // Integer ALU0 reg-reg operation
5420 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5421 %{
5422 single_instruction;
5423 dst : S4(write);
5424 src : S3(read);
5425 D0 : S0; // Big decoder only
5426 ALU0 : S3; // only alu0
5427 %}
5428
5429 // Integer ALU0 reg-mem operation
5430 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5431 %{
5432 single_instruction;
5433 dst : S5(write);
5434 mem : S3(read);
5435 D0 : S0; // big decoder only
5436 ALU0 : S4; // ALU0 only
5437 MEM : S3; // any mem
5438 %}
5439
5440 // Integer ALU reg-reg operation
5441 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5442 %{
5443 single_instruction;
5444 cr : S4(write);
5445 src1 : S3(read);
5446 src2 : S3(read);
5447 DECODE : S0; // any decoder
5448 ALU : S3; // any alu
5449 %}
5450
5451 // Integer ALU reg-imm operation
5452 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5453 %{
5454 single_instruction;
5455 cr : S4(write);
5456 src1 : S3(read);
5457 DECODE : S0; // any decoder
5458 ALU : S3; // any alu
5459 %}
5460
5461 // Integer ALU reg-mem operation
5462 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5463 %{
5464 single_instruction;
5465 cr : S4(write);
5466 src1 : S3(read);
5467 src2 : S3(read);
5468 D0 : S0; // big decoder only
5469 ALU : S4; // any alu
5470 MEM : S3;
5471 %}
5472
5473 // Conditional move reg-reg
5474 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5475 %{
5476 instruction_count(4);
5477 y : S4(read);
5478 q : S3(read);
5479 p : S3(read);
5480 DECODE : S0(4); // any decoder
5481 %}
5482
5483 // Conditional move reg-reg
5484 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5485 %{
5486 single_instruction;
5487 dst : S4(write);
5488 src : S3(read);
5489 cr : S3(read);
5490 DECODE : S0; // any decoder
5491 %}
5492
5493 // Conditional move reg-mem
5494 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5495 %{
5496 single_instruction;
5497 dst : S4(write);
5498 src : S3(read);
5499 cr : S3(read);
5500 DECODE : S0; // any decoder
5501 MEM : S3;
5502 %}
5503
5504 // Conditional move reg-reg long
5505 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5506 %{
5507 single_instruction;
5508 dst : S4(write);
5509 src : S3(read);
5510 cr : S3(read);
5511 DECODE : S0(2); // any 2 decoders
5512 %}
5513
5514 // XXX
5515 // // Conditional move double reg-reg
5516 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5517 // %{
5518 // single_instruction;
5519 // dst : S4(write);
5520 // src : S3(read);
5521 // cr : S3(read);
5522 // DECODE : S0; // any decoder
5523 // %}
5524
5525 // Float reg-reg operation
5526 pipe_class fpu_reg(regD dst)
5527 %{
5528 instruction_count(2);
5529 dst : S3(read);
5530 DECODE : S0(2); // any 2 decoders
5531 FPU : S3;
5532 %}
5533
5534 // Float reg-reg operation
5535 pipe_class fpu_reg_reg(regD dst, regD src)
5536 %{
5537 instruction_count(2);
5538 dst : S4(write);
5539 src : S3(read);
5540 DECODE : S0(2); // any 2 decoders
5541 FPU : S3;
5542 %}
5543
5544 // Float reg-reg operation
5545 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5546 %{
5547 instruction_count(3);
5548 dst : S4(write);
5549 src1 : S3(read);
5550 src2 : S3(read);
5551 DECODE : S0(3); // any 3 decoders
5552 FPU : S3(2);
5553 %}
5554
5555 // Float reg-reg operation
5556 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5557 %{
5558 instruction_count(4);
5559 dst : S4(write);
5560 src1 : S3(read);
5561 src2 : S3(read);
5562 src3 : S3(read);
5563 DECODE : S0(4); // any 3 decoders
5564 FPU : S3(2);
5565 %}
5566
5567 // Float reg-reg operation
5568 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5569 %{
5570 instruction_count(4);
5571 dst : S4(write);
5572 src1 : S3(read);
5573 src2 : S3(read);
5574 src3 : S3(read);
5575 DECODE : S1(3); // any 3 decoders
5576 D0 : S0; // Big decoder only
5577 FPU : S3(2);
5578 MEM : S3;
5579 %}
5580
5581 // Float reg-mem operation
5582 pipe_class fpu_reg_mem(regD dst, memory mem)
5583 %{
5584 instruction_count(2);
5585 dst : S5(write);
5586 mem : S3(read);
5587 D0 : S0; // big decoder only
5588 DECODE : S1; // any decoder for FPU POP
5589 FPU : S4;
5590 MEM : S3; // any mem
5591 %}
5592
5593 // Float reg-mem operation
5594 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5595 %{
5596 instruction_count(3);
5597 dst : S5(write);
5598 src1 : S3(read);
5599 mem : S3(read);
5600 D0 : S0; // big decoder only
5601 DECODE : S1(2); // any decoder for FPU POP
5602 FPU : S4;
5603 MEM : S3; // any mem
5604 %}
5605
5606 // Float mem-reg operation
5607 pipe_class fpu_mem_reg(memory mem, regD src)
5608 %{
5609 instruction_count(2);
5610 src : S5(read);
5611 mem : S3(read);
5612 DECODE : S0; // any decoder for FPU PUSH
5613 D0 : S1; // big decoder only
5614 FPU : S4;
5615 MEM : S3; // any mem
5616 %}
5617
5618 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5619 %{
5620 instruction_count(3);
5621 src1 : S3(read);
5622 src2 : S3(read);
5623 mem : S3(read);
5624 DECODE : S0(2); // any decoder for FPU PUSH
5625 D0 : S1; // big decoder only
5626 FPU : S4;
5627 MEM : S3; // any mem
5628 %}
5629
5630 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5631 %{
5632 instruction_count(3);
5633 src1 : S3(read);
5634 src2 : S3(read);
5635 mem : S4(read);
5636 DECODE : S0; // any decoder for FPU PUSH
5637 D0 : S0(2); // big decoder only
5638 FPU : S4;
5639 MEM : S3(2); // any mem
5640 %}
5641
5642 pipe_class fpu_mem_mem(memory dst, memory src1)
5643 %{
5644 instruction_count(2);
5645 src1 : S3(read);
5646 dst : S4(read);
5647 D0 : S0(2); // big decoder only
5648 MEM : S3(2); // any mem
5649 %}
5650
5651 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
5652 %{
5653 instruction_count(3);
5654 src1 : S3(read);
5655 src2 : S3(read);
5656 dst : S4(read);
5657 D0 : S0(3); // big decoder only
5658 FPU : S4;
5659 MEM : S3(3); // any mem
5660 %}
5661
5662 pipe_class fpu_mem_reg_con(memory mem, regD src1)
5663 %{
5664 instruction_count(3);
5665 src1 : S4(read);
5666 mem : S4(read);
5667 DECODE : S0; // any decoder for FPU PUSH
5668 D0 : S0(2); // big decoder only
5669 FPU : S4;
5670 MEM : S3(2); // any mem
5671 %}
5672
5673 // Float load constant
5674 pipe_class fpu_reg_con(regD dst)
5675 %{
5676 instruction_count(2);
5677 dst : S5(write);
5678 D0 : S0; // big decoder only for the load
5679 DECODE : S1; // any decoder for FPU POP
5680 FPU : S4;
5681 MEM : S3; // any mem
5682 %}
5683
5684 // Float load constant
5685 pipe_class fpu_reg_reg_con(regD dst, regD src)
5686 %{
5687 instruction_count(3);
5688 dst : S5(write);
5689 src : S3(read);
5690 D0 : S0; // big decoder only for the load
5691 DECODE : S1(2); // any decoder for FPU POP
5692 FPU : S4;
5693 MEM : S3; // any mem
5694 %}
5695
5696 // UnConditional branch
5697 pipe_class pipe_jmp(label labl)
5698 %{
5699 single_instruction;
5700 BR : S3;
5701 %}
5702
5703 // Conditional branch
5704 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
5705 %{
5706 single_instruction;
5707 cr : S1(read);
5708 BR : S3;
5709 %}
5710
5711 // Allocation idiom
5712 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
5713 %{
5714 instruction_count(1); force_serialization;
5715 fixed_latency(6);
5716 heap_ptr : S3(read);
5717 DECODE : S0(3);
5718 D0 : S2;
5719 MEM : S3;
5720 ALU : S3(2);
5721 dst : S5(write);
5722 BR : S5;
5723 %}
5724
5725 // Generic big/slow expanded idiom
5726 pipe_class pipe_slow()
5727 %{
5728 instruction_count(10); multiple_bundles; force_serialization;
5729 fixed_latency(100);
5730 D0 : S0(2);
5731 MEM : S3(2);
5732 %}
5733
5734 // The real do-nothing guy
5735 pipe_class empty()
5736 %{
5737 instruction_count(0);
5738 %}
5739
5740 // Define the class for the Nop node
5741 define
5742 %{
5743 MachNop = empty;
5744 %}
5745
5746 %}
5747
5748 //----------INSTRUCTIONS-------------------------------------------------------
5749 //
5750 // match -- States which machine-independent subtree may be replaced
5751 // by this instruction.
5752 // ins_cost -- The estimated cost of this instruction is used by instruction
5753 // selection to identify a minimum cost tree of machine
5754 // instructions that matches a tree of machine-independent
5755 // instructions.
5756 // format -- A string providing the disassembly for this instruction.
5757 // The value of an instruction's operand may be inserted
5758 // by referring to it with a '$' prefix.
5759 // opcode -- Three instruction opcodes may be provided. These are referred
5760 // to within an encode class as $primary, $secondary, and $tertiary
5761 // rrspectively. The primary opcode is commonly used to
5762 // indicate the type of machine instruction, while secondary
5763 // and tertiary are often used for prefix options or addressing
5764 // modes.
5765 // ins_encode -- A list of encode classes with parameters. The encode class
5766 // name must have been defined in an 'enc_class' specification
5767 // in the encode section of the architecture description.
5768
5769
5770 //----------Load/Store/Move Instructions---------------------------------------
5771 //----------Load Instructions--------------------------------------------------
5772
5773 // Load Byte (8 bit signed)
5774 instruct loadB(rRegI dst, memory mem)
5775 %{
5776 match(Set dst (LoadB mem));
5777
5778 ins_cost(125);
5779 format %{ "movsbl $dst, $mem\t# byte" %}
5780
5781 ins_encode %{
5782 __ movsbl($dst$$Register, $mem$$Address);
5783 %}
5784
5785 ins_pipe(ialu_reg_mem);
5786 %}
5787
5788 // Load Byte (8 bit signed) into Long Register
5789 instruct loadB2L(rRegL dst, memory mem)
5790 %{
5791 match(Set dst (ConvI2L (LoadB mem)));
5792
5793 ins_cost(125);
5794 format %{ "movsbq $dst, $mem\t# byte -> long" %}
5795
5796 ins_encode %{
5797 __ movsbq($dst$$Register, $mem$$Address);
5798 %}
5799
5800 ins_pipe(ialu_reg_mem);
5801 %}
5802
5803 // Load Unsigned Byte (8 bit UNsigned)
5804 instruct loadUB(rRegI dst, memory mem)
5805 %{
5806 match(Set dst (LoadUB mem));
5807
5808 ins_cost(125);
5809 format %{ "movzbl $dst, $mem\t# ubyte" %}
5810
5811 ins_encode %{
5812 __ movzbl($dst$$Register, $mem$$Address);
5813 %}
5814
5815 ins_pipe(ialu_reg_mem);
5816 %}
5817
5818 // Load Unsigned Byte (8 bit UNsigned) into Long Register
5819 instruct loadUB2L(rRegL dst, memory mem)
5820 %{
5821 match(Set dst (ConvI2L (LoadUB mem)));
5822
5823 ins_cost(125);
5824 format %{ "movzbq $dst, $mem\t# ubyte -> long" %}
5825
5826 ins_encode %{
5827 __ movzbq($dst$$Register, $mem$$Address);
5828 %}
5829
5830 ins_pipe(ialu_reg_mem);
5831 %}
5832
5833 // Load Unsigned Byte (8 bit UNsigned) with a 8-bit mask into Long Register
5834 instruct loadUB2L_immI8(rRegL dst, memory mem, immI8 mask, rFlagsReg cr) %{
5835 match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5836 effect(KILL cr);
5837
5838 format %{ "movzbq $dst, $mem\t# ubyte & 8-bit mask -> long\n\t"
5839 "andl $dst, $mask" %}
5840 ins_encode %{
5841 Register Rdst = $dst$$Register;
5842 __ movzbq(Rdst, $mem$$Address);
5843 __ andl(Rdst, $mask$$constant);
5844 %}
5845 ins_pipe(ialu_reg_mem);
5846 %}
5847
5848 // Load Short (16 bit signed)
5849 instruct loadS(rRegI dst, memory mem)
5850 %{
5851 match(Set dst (LoadS mem));
5852
5853 ins_cost(125);
5854 format %{ "movswl $dst, $mem\t# short" %}
5855
5856 ins_encode %{
5857 __ movswl($dst$$Register, $mem$$Address);
5858 %}
5859
5860 ins_pipe(ialu_reg_mem);
5861 %}
5862
5863 // Load Short (16 bit signed) to Byte (8 bit signed)
5864 instruct loadS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5865 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5866
5867 ins_cost(125);
5868 format %{ "movsbl $dst, $mem\t# short -> byte" %}
5869 ins_encode %{
5870 __ movsbl($dst$$Register, $mem$$Address);
5871 %}
5872 ins_pipe(ialu_reg_mem);
5873 %}
5874
5875 // Load Short (16 bit signed) into Long Register
5876 instruct loadS2L(rRegL dst, memory mem)
5877 %{
5878 match(Set dst (ConvI2L (LoadS mem)));
5879
5880 ins_cost(125);
5881 format %{ "movswq $dst, $mem\t# short -> long" %}
5882
5883 ins_encode %{
5884 __ movswq($dst$$Register, $mem$$Address);
5885 %}
5886
5887 ins_pipe(ialu_reg_mem);
5888 %}
5889
5890 // Load Unsigned Short/Char (16 bit UNsigned)
5891 instruct loadUS(rRegI dst, memory mem)
5892 %{
5893 match(Set dst (LoadUS mem));
5894
5895 ins_cost(125);
5896 format %{ "movzwl $dst, $mem\t# ushort/char" %}
5897
5898 ins_encode %{
5899 __ movzwl($dst$$Register, $mem$$Address);
5900 %}
5901
5902 ins_pipe(ialu_reg_mem);
5903 %}
5904
5905 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5906 instruct loadUS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5907 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5908
5909 ins_cost(125);
5910 format %{ "movsbl $dst, $mem\t# ushort -> byte" %}
5911 ins_encode %{
5912 __ movsbl($dst$$Register, $mem$$Address);
5913 %}
5914 ins_pipe(ialu_reg_mem);
5915 %}
5916
5917 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
5918 instruct loadUS2L(rRegL dst, memory mem)
5919 %{
5920 match(Set dst (ConvI2L (LoadUS mem)));
5921
5922 ins_cost(125);
5923 format %{ "movzwq $dst, $mem\t# ushort/char -> long" %}
5924
5925 ins_encode %{
5926 __ movzwq($dst$$Register, $mem$$Address);
5927 %}
5928
5929 ins_pipe(ialu_reg_mem);
5930 %}
5931
5932 // Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
5933 instruct loadUS2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
5934 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5935
5936 format %{ "movzbq $dst, $mem\t# ushort/char & 0xFF -> long" %}
5937 ins_encode %{
5938 __ movzbq($dst$$Register, $mem$$Address);
5939 %}
5940 ins_pipe(ialu_reg_mem);
5941 %}
5942
5943 // Load Unsigned Short/Char (16 bit UNsigned) with mask into Long Register
5944 instruct loadUS2L_immI16(rRegL dst, memory mem, immI16 mask, rFlagsReg cr) %{
5945 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5946 effect(KILL cr);
5947
5948 format %{ "movzwq $dst, $mem\t# ushort/char & 16-bit mask -> long\n\t"
5949 "andl $dst, $mask" %}
5950 ins_encode %{
5951 Register Rdst = $dst$$Register;
5952 __ movzwq(Rdst, $mem$$Address);
5953 __ andl(Rdst, $mask$$constant);
5954 %}
5955 ins_pipe(ialu_reg_mem);
5956 %}
5957
5958 // Load Integer
5959 instruct loadI(rRegI dst, memory mem)
5960 %{
5961 match(Set dst (LoadI mem));
5962
5963 ins_cost(125);
5964 format %{ "movl $dst, $mem\t# int" %}
5965
5966 ins_encode %{
5967 __ movl($dst$$Register, $mem$$Address);
5968 %}
5969
5970 ins_pipe(ialu_reg_mem);
5971 %}
5972
5973 // Load Integer (32 bit signed) to Byte (8 bit signed)
5974 instruct loadI2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5975 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
5976
5977 ins_cost(125);
5978 format %{ "movsbl $dst, $mem\t# int -> byte" %}
5979 ins_encode %{
5980 __ movsbl($dst$$Register, $mem$$Address);
5981 %}
5982 ins_pipe(ialu_reg_mem);
5983 %}
5984
5985 // Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
5986 instruct loadI2UB(rRegI dst, memory mem, immI_255 mask) %{
5987 match(Set dst (AndI (LoadI mem) mask));
5988
5989 ins_cost(125);
5990 format %{ "movzbl $dst, $mem\t# int -> ubyte" %}
5991 ins_encode %{
5992 __ movzbl($dst$$Register, $mem$$Address);
5993 %}
5994 ins_pipe(ialu_reg_mem);
5995 %}
5996
5997 // Load Integer (32 bit signed) to Short (16 bit signed)
5998 instruct loadI2S(rRegI dst, memory mem, immI_16 sixteen) %{
5999 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
6000
6001 ins_cost(125);
6002 format %{ "movswl $dst, $mem\t# int -> short" %}
6003 ins_encode %{
6004 __ movswl($dst$$Register, $mem$$Address);
6005 %}
6006 ins_pipe(ialu_reg_mem);
6007 %}
6008
6009 // Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
6010 instruct loadI2US(rRegI dst, memory mem, immI_65535 mask) %{
6011 match(Set dst (AndI (LoadI mem) mask));
6012
6013 ins_cost(125);
6014 format %{ "movzwl $dst, $mem\t# int -> ushort/char" %}
6015 ins_encode %{
6016 __ movzwl($dst$$Register, $mem$$Address);
6017 %}
6018 ins_pipe(ialu_reg_mem);
6019 %}
6020
6021 // Load Integer into Long Register
6022 instruct loadI2L(rRegL dst, memory mem)
6023 %{
6024 match(Set dst (ConvI2L (LoadI mem)));
6025
6026 ins_cost(125);
6027 format %{ "movslq $dst, $mem\t# int -> long" %}
6028
6029 ins_encode %{
6030 __ movslq($dst$$Register, $mem$$Address);
6031 %}
6032
6033 ins_pipe(ialu_reg_mem);
6034 %}
6035
6036 // Load Integer with mask 0xFF into Long Register
6037 instruct loadI2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
6038 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6039
6040 format %{ "movzbq $dst, $mem\t# int & 0xFF -> long" %}
6041 ins_encode %{
6042 __ movzbq($dst$$Register, $mem$$Address);
6043 %}
6044 ins_pipe(ialu_reg_mem);
6045 %}
6046
6047 // Load Integer with mask 0xFFFF into Long Register
6048 instruct loadI2L_immI_65535(rRegL dst, memory mem, immI_65535 mask) %{
6049 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6050
6051 format %{ "movzwq $dst, $mem\t# int & 0xFFFF -> long" %}
6052 ins_encode %{
6053 __ movzwq($dst$$Register, $mem$$Address);
6054 %}
6055 ins_pipe(ialu_reg_mem);
6056 %}
6057
6058 // Load Integer with a 32-bit mask into Long Register
6059 instruct loadI2L_immI(rRegL dst, memory mem, immI mask, rFlagsReg cr) %{
6060 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6061 effect(KILL cr);
6062
6063 format %{ "movl $dst, $mem\t# int & 32-bit mask -> long\n\t"
6064 "andl $dst, $mask" %}
6065 ins_encode %{
6066 Register Rdst = $dst$$Register;
6067 __ movl(Rdst, $mem$$Address);
6068 __ andl(Rdst, $mask$$constant);
6069 %}
6070 ins_pipe(ialu_reg_mem);
6071 %}
6072
6073 // Load Unsigned Integer into Long Register
6074 instruct loadUI2L(rRegL dst, memory mem)
6075 %{
6076 match(Set dst (LoadUI2L mem));
6077
6078 ins_cost(125);
6079 format %{ "movl $dst, $mem\t# uint -> long" %}
6080
6081 ins_encode %{
6082 __ movl($dst$$Register, $mem$$Address);
6083 %}
6084
6085 ins_pipe(ialu_reg_mem);
6086 %}
6087
6088 // Load Long
6089 instruct loadL(rRegL dst, memory mem)
6090 %{
6091 match(Set dst (LoadL mem));
6092
6093 ins_cost(125);
6094 format %{ "movq $dst, $mem\t# long" %}
6095
6096 ins_encode %{
6097 __ movq($dst$$Register, $mem$$Address);
6098 %}
6099
6100 ins_pipe(ialu_reg_mem); // XXX
6101 %}
6102
6103 // Load Range
6104 instruct loadRange(rRegI dst, memory mem)
6105 %{
6106 match(Set dst (LoadRange mem));
6107
6108 ins_cost(125); // XXX
6109 format %{ "movl $dst, $mem\t# range" %}
6110 opcode(0x8B);
6111 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6112 ins_pipe(ialu_reg_mem);
6113 %}
6114
6115 // Load Pointer
6116 instruct loadP(rRegP dst, memory mem)
6117 %{
6118 match(Set dst (LoadP mem));
6119
6120 ins_cost(125); // XXX
6121 format %{ "movq $dst, $mem\t# ptr" %}
6122 opcode(0x8B);
6123 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6124 ins_pipe(ialu_reg_mem); // XXX
6125 %}
6126
6127 // Load Compressed Pointer
6128 instruct loadN(rRegN dst, memory mem)
6129 %{
6130 match(Set dst (LoadN mem));
6131
6132 ins_cost(125); // XXX
6133 format %{ "movl $dst, $mem\t# compressed ptr" %}
6134 ins_encode %{
6135 __ movl($dst$$Register, $mem$$Address);
6136 %}
6137 ins_pipe(ialu_reg_mem); // XXX
6138 %}
6139
6140
6141 // Load Klass Pointer
6142 instruct loadKlass(rRegP dst, memory mem)
6143 %{
6144 match(Set dst (LoadKlass mem));
6145
6146 ins_cost(125); // XXX
6147 format %{ "movq $dst, $mem\t# class" %}
6148 opcode(0x8B);
6149 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6150 ins_pipe(ialu_reg_mem); // XXX
6151 %}
6152
6153 // Load narrow Klass Pointer
6154 instruct loadNKlass(rRegN dst, memory mem)
6155 %{
6156 match(Set dst (LoadNKlass mem));
6157
6158 ins_cost(125); // XXX
6159 format %{ "movl $dst, $mem\t# compressed klass ptr" %}
6160 ins_encode %{
6161 __ movl($dst$$Register, $mem$$Address);
6162 %}
6163 ins_pipe(ialu_reg_mem); // XXX
6164 %}
6165
6166 // Load Float
6167 instruct loadF(regF dst, memory mem)
6168 %{
6169 match(Set dst (LoadF mem));
6170
6171 ins_cost(145); // XXX
6172 format %{ "movss $dst, $mem\t# float" %}
6173 opcode(0xF3, 0x0F, 0x10);
6174 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6175 ins_pipe(pipe_slow); // XXX
6176 %}
6177
6178 // Load Double
6179 instruct loadD_partial(regD dst, memory mem)
6180 %{
6181 predicate(!UseXmmLoadAndClearUpper);
6182 match(Set dst (LoadD mem));
6183
6184 ins_cost(145); // XXX
6185 format %{ "movlpd $dst, $mem\t# double" %}
6186 opcode(0x66, 0x0F, 0x12);
6187 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6188 ins_pipe(pipe_slow); // XXX
6189 %}
6190
6191 instruct loadD(regD dst, memory mem)
6192 %{
6193 predicate(UseXmmLoadAndClearUpper);
6194 match(Set dst (LoadD mem));
6195
6196 ins_cost(145); // XXX
6197 format %{ "movsd $dst, $mem\t# double" %}
6198 opcode(0xF2, 0x0F, 0x10);
6199 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6200 ins_pipe(pipe_slow); // XXX
6201 %}
6202
6203 // Load Aligned Packed Byte to XMM register
6204 instruct loadA8B(regD dst, memory mem) %{
6205 match(Set dst (Load8B mem));
6206 ins_cost(125);
6207 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6208 ins_encode( movq_ld(dst, mem));
6209 ins_pipe( pipe_slow );
6210 %}
6211
6212 // Load Aligned Packed Short to XMM register
6213 instruct loadA4S(regD dst, memory mem) %{
6214 match(Set dst (Load4S mem));
6215 ins_cost(125);
6216 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6217 ins_encode( movq_ld(dst, mem));
6218 ins_pipe( pipe_slow );
6219 %}
6220
6221 // Load Aligned Packed Char to XMM register
6222 instruct loadA4C(regD dst, memory mem) %{
6223 match(Set dst (Load4C mem));
6224 ins_cost(125);
6225 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6226 ins_encode( movq_ld(dst, mem));
6227 ins_pipe( pipe_slow );
6228 %}
6229
6230 // Load Aligned Packed Integer to XMM register
6231 instruct load2IU(regD dst, memory mem) %{
6232 match(Set dst (Load2I mem));
6233 ins_cost(125);
6234 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6235 ins_encode( movq_ld(dst, mem));
6236 ins_pipe( pipe_slow );
6237 %}
6238
6239 // Load Aligned Packed Single to XMM
6240 instruct loadA2F(regD dst, memory mem) %{
6241 match(Set dst (Load2F mem));
6242 ins_cost(145);
6243 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6244 ins_encode( movq_ld(dst, mem));
6245 ins_pipe( pipe_slow );
6246 %}
6247
6248 // Load Effective Address
6249 instruct leaP8(rRegP dst, indOffset8 mem)
6250 %{
6251 match(Set dst mem);
6252
6253 ins_cost(110); // XXX
6254 format %{ "leaq $dst, $mem\t# ptr 8" %}
6255 opcode(0x8D);
6256 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6257 ins_pipe(ialu_reg_reg_fat);
6258 %}
6259
6260 instruct leaP32(rRegP dst, indOffset32 mem)
6261 %{
6262 match(Set dst mem);
6263
6264 ins_cost(110);
6265 format %{ "leaq $dst, $mem\t# ptr 32" %}
6266 opcode(0x8D);
6267 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6268 ins_pipe(ialu_reg_reg_fat);
6269 %}
6270
6271 // instruct leaPIdx(rRegP dst, indIndex mem)
6272 // %{
6273 // match(Set dst mem);
6274
6275 // ins_cost(110);
6276 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6277 // opcode(0x8D);
6278 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6279 // ins_pipe(ialu_reg_reg_fat);
6280 // %}
6281
6282 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6283 %{
6284 match(Set dst mem);
6285
6286 ins_cost(110);
6287 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6288 opcode(0x8D);
6289 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6290 ins_pipe(ialu_reg_reg_fat);
6291 %}
6292
6293 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6294 %{
6295 match(Set dst mem);
6296
6297 ins_cost(110);
6298 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6299 opcode(0x8D);
6300 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6301 ins_pipe(ialu_reg_reg_fat);
6302 %}
6303
6304 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6305 %{
6306 match(Set dst mem);
6307
6308 ins_cost(110);
6309 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6310 opcode(0x8D);
6311 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6312 ins_pipe(ialu_reg_reg_fat);
6313 %}
6314
6315 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem)
6316 %{
6317 match(Set dst mem);
6318
6319 ins_cost(110);
6320 format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %}
6321 opcode(0x8D);
6322 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6323 ins_pipe(ialu_reg_reg_fat);
6324 %}
6325
6326 // Load Effective Address which uses Narrow (32-bits) oop
6327 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem)
6328 %{
6329 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
6330 match(Set dst mem);
6331
6332 ins_cost(110);
6333 format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %}
6334 opcode(0x8D);
6335 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6336 ins_pipe(ialu_reg_reg_fat);
6337 %}
6338
6339 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem)
6340 %{
6341 predicate(Universe::narrow_oop_shift() == 0);
6342 match(Set dst mem);
6343
6344 ins_cost(110); // XXX
6345 format %{ "leaq $dst, $mem\t# ptr off8narrow" %}
6346 opcode(0x8D);
6347 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6348 ins_pipe(ialu_reg_reg_fat);
6349 %}
6350
6351 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem)
6352 %{
6353 predicate(Universe::narrow_oop_shift() == 0);
6354 match(Set dst mem);
6355
6356 ins_cost(110);
6357 format %{ "leaq $dst, $mem\t# ptr off32narrow" %}
6358 opcode(0x8D);
6359 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6360 ins_pipe(ialu_reg_reg_fat);
6361 %}
6362
6363 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem)
6364 %{
6365 predicate(Universe::narrow_oop_shift() == 0);
6366 match(Set dst mem);
6367
6368 ins_cost(110);
6369 format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %}
6370 opcode(0x8D);
6371 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6372 ins_pipe(ialu_reg_reg_fat);
6373 %}
6374
6375 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem)
6376 %{
6377 predicate(Universe::narrow_oop_shift() == 0);
6378 match(Set dst mem);
6379
6380 ins_cost(110);
6381 format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %}
6382 opcode(0x8D);
6383 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6384 ins_pipe(ialu_reg_reg_fat);
6385 %}
6386
6387 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem)
6388 %{
6389 predicate(Universe::narrow_oop_shift() == 0);
6390 match(Set dst mem);
6391
6392 ins_cost(110);
6393 format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %}
6394 opcode(0x8D);
6395 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6396 ins_pipe(ialu_reg_reg_fat);
6397 %}
6398
6399 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem)
6400 %{
6401 predicate(Universe::narrow_oop_shift() == 0);
6402 match(Set dst mem);
6403
6404 ins_cost(110);
6405 format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %}
6406 opcode(0x8D);
6407 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6408 ins_pipe(ialu_reg_reg_fat);
6409 %}
6410
6411 instruct loadConI(rRegI dst, immI src)
6412 %{
6413 match(Set dst src);
6414
6415 format %{ "movl $dst, $src\t# int" %}
6416 ins_encode(load_immI(dst, src));
6417 ins_pipe(ialu_reg_fat); // XXX
6418 %}
6419
6420 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6421 %{
6422 match(Set dst src);
6423 effect(KILL cr);
6424
6425 ins_cost(50);
6426 format %{ "xorl $dst, $dst\t# int" %}
6427 opcode(0x33); /* + rd */
6428 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6429 ins_pipe(ialu_reg);
6430 %}
6431
6432 instruct loadConL(rRegL dst, immL src)
6433 %{
6434 match(Set dst src);
6435
6436 ins_cost(150);
6437 format %{ "movq $dst, $src\t# long" %}
6438 ins_encode(load_immL(dst, src));
6439 ins_pipe(ialu_reg);
6440 %}
6441
6442 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6443 %{
6444 match(Set dst src);
6445 effect(KILL cr);
6446
6447 ins_cost(50);
6448 format %{ "xorl $dst, $dst\t# long" %}
6449 opcode(0x33); /* + rd */
6450 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6451 ins_pipe(ialu_reg); // XXX
6452 %}
6453
6454 instruct loadConUL32(rRegL dst, immUL32 src)
6455 %{
6456 match(Set dst src);
6457
6458 ins_cost(60);
6459 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6460 ins_encode(load_immUL32(dst, src));
6461 ins_pipe(ialu_reg);
6462 %}
6463
6464 instruct loadConL32(rRegL dst, immL32 src)
6465 %{
6466 match(Set dst src);
6467
6468 ins_cost(70);
6469 format %{ "movq $dst, $src\t# long (32-bit)" %}
6470 ins_encode(load_immL32(dst, src));
6471 ins_pipe(ialu_reg);
6472 %}
6473
6474 instruct loadConP(rRegP dst, immP con) %{
6475 match(Set dst con);
6476
6477 format %{ "movq $dst, $con\t# ptr" %}
6478 ins_encode(load_immP(dst, con));
6479 ins_pipe(ialu_reg_fat); // XXX
6480 %}
6481
6482 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6483 %{
6484 match(Set dst src);
6485 effect(KILL cr);
6486
6487 ins_cost(50);
6488 format %{ "xorl $dst, $dst\t# ptr" %}
6489 opcode(0x33); /* + rd */
6490 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6491 ins_pipe(ialu_reg);
6492 %}
6493
6494 instruct loadConP_poll(rRegP dst, immP_poll src) %{
6495 match(Set dst src);
6496 format %{ "movq $dst, $src\t!ptr" %}
6497 ins_encode %{
6498 AddressLiteral polling_page(os::get_polling_page(), relocInfo::poll_type);
6499 __ lea($dst$$Register, polling_page);
6500 %}
6501 ins_pipe(ialu_reg_fat);
6502 %}
6503
6504 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6505 %{
6506 match(Set dst src);
6507 effect(KILL cr);
6508
6509 ins_cost(60);
6510 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6511 ins_encode(load_immP31(dst, src));
6512 ins_pipe(ialu_reg);
6513 %}
6514
6515 instruct loadConF(regF dst, immF con) %{
6516 match(Set dst con);
6517 ins_cost(125);
6518 format %{ "movss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
6519 ins_encode %{
6520 __ movflt($dst$$XMMRegister, $constantaddress($con));
6521 %}
6522 ins_pipe(pipe_slow);
6523 %}
6524
6525 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6526 match(Set dst src);
6527 effect(KILL cr);
6528 format %{ "xorq $dst, $src\t# compressed NULL ptr" %}
6529 ins_encode %{
6530 __ xorq($dst$$Register, $dst$$Register);
6531 %}
6532 ins_pipe(ialu_reg);
6533 %}
6534
6535 instruct loadConN(rRegN dst, immN src) %{
6536 match(Set dst src);
6537
6538 ins_cost(125);
6539 format %{ "movl $dst, $src\t# compressed ptr" %}
6540 ins_encode %{
6541 address con = (address)$src$$constant;
6542 if (con == NULL) {
6543 ShouldNotReachHere();
6544 } else {
6545 __ set_narrow_oop($dst$$Register, (jobject)$src$$constant);
6546 }
6547 %}
6548 ins_pipe(ialu_reg_fat); // XXX
6549 %}
6550
6551 instruct loadConF0(regF dst, immF0 src)
6552 %{
6553 match(Set dst src);
6554 ins_cost(100);
6555
6556 format %{ "xorps $dst, $dst\t# float 0.0" %}
6557 opcode(0x0F, 0x57);
6558 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6559 ins_pipe(pipe_slow);
6560 %}
6561
6562 // Use the same format since predicate() can not be used here.
6563 instruct loadConD(regD dst, immD con) %{
6564 match(Set dst con);
6565 ins_cost(125);
6566 format %{ "movsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
6567 ins_encode %{
6568 __ movdbl($dst$$XMMRegister, $constantaddress($con));
6569 %}
6570 ins_pipe(pipe_slow);
6571 %}
6572
6573 instruct loadConD0(regD dst, immD0 src)
6574 %{
6575 match(Set dst src);
6576 ins_cost(100);
6577
6578 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6579 opcode(0x66, 0x0F, 0x57);
6580 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6581 ins_pipe(pipe_slow);
6582 %}
6583
6584 instruct loadSSI(rRegI dst, stackSlotI src)
6585 %{
6586 match(Set dst src);
6587
6588 ins_cost(125);
6589 format %{ "movl $dst, $src\t# int stk" %}
6590 opcode(0x8B);
6591 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6592 ins_pipe(ialu_reg_mem);
6593 %}
6594
6595 instruct loadSSL(rRegL dst, stackSlotL src)
6596 %{
6597 match(Set dst src);
6598
6599 ins_cost(125);
6600 format %{ "movq $dst, $src\t# long stk" %}
6601 opcode(0x8B);
6602 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6603 ins_pipe(ialu_reg_mem);
6604 %}
6605
6606 instruct loadSSP(rRegP dst, stackSlotP src)
6607 %{
6608 match(Set dst src);
6609
6610 ins_cost(125);
6611 format %{ "movq $dst, $src\t# ptr stk" %}
6612 opcode(0x8B);
6613 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6614 ins_pipe(ialu_reg_mem);
6615 %}
6616
6617 instruct loadSSF(regF dst, stackSlotF src)
6618 %{
6619 match(Set dst src);
6620
6621 ins_cost(125);
6622 format %{ "movss $dst, $src\t# float stk" %}
6623 opcode(0xF3, 0x0F, 0x10);
6624 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6625 ins_pipe(pipe_slow); // XXX
6626 %}
6627
6628 // Use the same format since predicate() can not be used here.
6629 instruct loadSSD(regD dst, stackSlotD src)
6630 %{
6631 match(Set dst src);
6632
6633 ins_cost(125);
6634 format %{ "movsd $dst, $src\t# double stk" %}
6635 ins_encode %{
6636 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6637 %}
6638 ins_pipe(pipe_slow); // XXX
6639 %}
6640
6641 // Prefetch instructions.
6642 // Must be safe to execute with invalid address (cannot fault).
6643
6644 instruct prefetchr( memory mem ) %{
6645 predicate(ReadPrefetchInstr==3);
6646 match(PrefetchRead mem);
6647 ins_cost(125);
6648
6649 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6650 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6651 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6652 ins_pipe(ialu_mem);
6653 %}
6654
6655 instruct prefetchrNTA( memory mem ) %{
6656 predicate(ReadPrefetchInstr==0);
6657 match(PrefetchRead mem);
6658 ins_cost(125);
6659
6660 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6661 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6662 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6663 ins_pipe(ialu_mem);
6664 %}
6665
6666 instruct prefetchrT0( memory mem ) %{
6667 predicate(ReadPrefetchInstr==1);
6668 match(PrefetchRead mem);
6669 ins_cost(125);
6670
6671 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6672 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6673 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6674 ins_pipe(ialu_mem);
6675 %}
6676
6677 instruct prefetchrT2( memory mem ) %{
6678 predicate(ReadPrefetchInstr==2);
6679 match(PrefetchRead mem);
6680 ins_cost(125);
6681
6682 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6683 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6684 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6685 ins_pipe(ialu_mem);
6686 %}
6687
6688 instruct prefetchw( memory mem ) %{
6689 predicate(AllocatePrefetchInstr==3);
6690 match(PrefetchWrite mem);
6691 ins_cost(125);
6692
6693 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6694 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6695 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6696 ins_pipe(ialu_mem);
6697 %}
6698
6699 instruct prefetchwNTA( memory mem ) %{
6700 predicate(AllocatePrefetchInstr==0);
6701 match(PrefetchWrite mem);
6702 ins_cost(125);
6703
6704 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6705 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6706 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6707 ins_pipe(ialu_mem);
6708 %}
6709
6710 instruct prefetchwT0( memory mem ) %{
6711 predicate(AllocatePrefetchInstr==1);
6712 match(PrefetchWrite mem);
6713 ins_cost(125);
6714
6715 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6716 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6717 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6718 ins_pipe(ialu_mem);
6719 %}
6720
6721 instruct prefetchwT2( memory mem ) %{
6722 predicate(AllocatePrefetchInstr==2);
6723 match(PrefetchWrite mem);
6724 ins_cost(125);
6725
6726 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6727 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6728 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6729 ins_pipe(ialu_mem);
6730 %}
6731
6732 //----------Store Instructions-------------------------------------------------
6733
6734 // Store Byte
6735 instruct storeB(memory mem, rRegI src)
6736 %{
6737 match(Set mem (StoreB mem src));
6738
6739 ins_cost(125); // XXX
6740 format %{ "movb $mem, $src\t# byte" %}
6741 opcode(0x88);
6742 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6743 ins_pipe(ialu_mem_reg);
6744 %}
6745
6746 // Store Char/Short
6747 instruct storeC(memory mem, rRegI src)
6748 %{
6749 match(Set mem (StoreC mem src));
6750
6751 ins_cost(125); // XXX
6752 format %{ "movw $mem, $src\t# char/short" %}
6753 opcode(0x89);
6754 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6755 ins_pipe(ialu_mem_reg);
6756 %}
6757
6758 // Store Integer
6759 instruct storeI(memory mem, rRegI src)
6760 %{
6761 match(Set mem (StoreI mem src));
6762
6763 ins_cost(125); // XXX
6764 format %{ "movl $mem, $src\t# int" %}
6765 opcode(0x89);
6766 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6767 ins_pipe(ialu_mem_reg);
6768 %}
6769
6770 // Store Long
6771 instruct storeL(memory mem, rRegL src)
6772 %{
6773 match(Set mem (StoreL mem src));
6774
6775 ins_cost(125); // XXX
6776 format %{ "movq $mem, $src\t# long" %}
6777 opcode(0x89);
6778 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6779 ins_pipe(ialu_mem_reg); // XXX
6780 %}
6781
6782 // Store Pointer
6783 instruct storeP(memory mem, any_RegP src)
6784 %{
6785 match(Set mem (StoreP mem src));
6786
6787 ins_cost(125); // XXX
6788 format %{ "movq $mem, $src\t# ptr" %}
6789 opcode(0x89);
6790 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6791 ins_pipe(ialu_mem_reg);
6792 %}
6793
6794 instruct storeImmP0(memory mem, immP0 zero)
6795 %{
6796 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6797 match(Set mem (StoreP mem zero));
6798
6799 ins_cost(125); // XXX
6800 format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %}
6801 ins_encode %{
6802 __ movq($mem$$Address, r12);
6803 %}
6804 ins_pipe(ialu_mem_reg);
6805 %}
6806
6807 // Store NULL Pointer, mark word, or other simple pointer constant.
6808 instruct storeImmP(memory mem, immP31 src)
6809 %{
6810 match(Set mem (StoreP mem src));
6811
6812 ins_cost(150); // XXX
6813 format %{ "movq $mem, $src\t# ptr" %}
6814 opcode(0xC7); /* C7 /0 */
6815 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6816 ins_pipe(ialu_mem_imm);
6817 %}
6818
6819 // Store Compressed Pointer
6820 instruct storeN(memory mem, rRegN src)
6821 %{
6822 match(Set mem (StoreN mem src));
6823
6824 ins_cost(125); // XXX
6825 format %{ "movl $mem, $src\t# compressed ptr" %}
6826 ins_encode %{
6827 __ movl($mem$$Address, $src$$Register);
6828 %}
6829 ins_pipe(ialu_mem_reg);
6830 %}
6831
6832 instruct storeImmN0(memory mem, immN0 zero)
6833 %{
6834 predicate(Universe::narrow_oop_base() == NULL);
6835 match(Set mem (StoreN mem zero));
6836
6837 ins_cost(125); // XXX
6838 format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %}
6839 ins_encode %{
6840 __ movl($mem$$Address, r12);
6841 %}
6842 ins_pipe(ialu_mem_reg);
6843 %}
6844
6845 instruct storeImmN(memory mem, immN src)
6846 %{
6847 match(Set mem (StoreN mem src));
6848
6849 ins_cost(150); // XXX
6850 format %{ "movl $mem, $src\t# compressed ptr" %}
6851 ins_encode %{
6852 address con = (address)$src$$constant;
6853 if (con == NULL) {
6854 __ movl($mem$$Address, (int32_t)0);
6855 } else {
6856 __ set_narrow_oop($mem$$Address, (jobject)$src$$constant);
6857 }
6858 %}
6859 ins_pipe(ialu_mem_imm);
6860 %}
6861
6862 // Store Integer Immediate
6863 instruct storeImmI0(memory mem, immI0 zero)
6864 %{
6865 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6866 match(Set mem (StoreI mem zero));
6867
6868 ins_cost(125); // XXX
6869 format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %}
6870 ins_encode %{
6871 __ movl($mem$$Address, r12);
6872 %}
6873 ins_pipe(ialu_mem_reg);
6874 %}
6875
6876 instruct storeImmI(memory mem, immI src)
6877 %{
6878 match(Set mem (StoreI mem src));
6879
6880 ins_cost(150);
6881 format %{ "movl $mem, $src\t# int" %}
6882 opcode(0xC7); /* C7 /0 */
6883 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6884 ins_pipe(ialu_mem_imm);
6885 %}
6886
6887 // Store Long Immediate
6888 instruct storeImmL0(memory mem, immL0 zero)
6889 %{
6890 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6891 match(Set mem (StoreL mem zero));
6892
6893 ins_cost(125); // XXX
6894 format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %}
6895 ins_encode %{
6896 __ movq($mem$$Address, r12);
6897 %}
6898 ins_pipe(ialu_mem_reg);
6899 %}
6900
6901 instruct storeImmL(memory mem, immL32 src)
6902 %{
6903 match(Set mem (StoreL mem src));
6904
6905 ins_cost(150);
6906 format %{ "movq $mem, $src\t# long" %}
6907 opcode(0xC7); /* C7 /0 */
6908 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6909 ins_pipe(ialu_mem_imm);
6910 %}
6911
6912 // Store Short/Char Immediate
6913 instruct storeImmC0(memory mem, immI0 zero)
6914 %{
6915 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6916 match(Set mem (StoreC mem zero));
6917
6918 ins_cost(125); // XXX
6919 format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %}
6920 ins_encode %{
6921 __ movw($mem$$Address, r12);
6922 %}
6923 ins_pipe(ialu_mem_reg);
6924 %}
6925
6926 instruct storeImmI16(memory mem, immI16 src)
6927 %{
6928 predicate(UseStoreImmI16);
6929 match(Set mem (StoreC mem src));
6930
6931 ins_cost(150);
6932 format %{ "movw $mem, $src\t# short/char" %}
6933 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
6934 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
6935 ins_pipe(ialu_mem_imm);
6936 %}
6937
6938 // Store Byte Immediate
6939 instruct storeImmB0(memory mem, immI0 zero)
6940 %{
6941 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6942 match(Set mem (StoreB mem zero));
6943
6944 ins_cost(125); // XXX
6945 format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %}
6946 ins_encode %{
6947 __ movb($mem$$Address, r12);
6948 %}
6949 ins_pipe(ialu_mem_reg);
6950 %}
6951
6952 instruct storeImmB(memory mem, immI8 src)
6953 %{
6954 match(Set mem (StoreB mem src));
6955
6956 ins_cost(150); // XXX
6957 format %{ "movb $mem, $src\t# byte" %}
6958 opcode(0xC6); /* C6 /0 */
6959 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
6960 ins_pipe(ialu_mem_imm);
6961 %}
6962
6963 // Store Aligned Packed Byte XMM register to memory
6964 instruct storeA8B(memory mem, regD src) %{
6965 match(Set mem (Store8B mem src));
6966 ins_cost(145);
6967 format %{ "MOVQ $mem,$src\t! packed8B" %}
6968 ins_encode( movq_st(mem, src));
6969 ins_pipe( pipe_slow );
6970 %}
6971
6972 // Store Aligned Packed Char/Short XMM register to memory
6973 instruct storeA4C(memory mem, regD src) %{
6974 match(Set mem (Store4C mem src));
6975 ins_cost(145);
6976 format %{ "MOVQ $mem,$src\t! packed4C" %}
6977 ins_encode( movq_st(mem, src));
6978 ins_pipe( pipe_slow );
6979 %}
6980
6981 // Store Aligned Packed Integer XMM register to memory
6982 instruct storeA2I(memory mem, regD src) %{
6983 match(Set mem (Store2I mem src));
6984 ins_cost(145);
6985 format %{ "MOVQ $mem,$src\t! packed2I" %}
6986 ins_encode( movq_st(mem, src));
6987 ins_pipe( pipe_slow );
6988 %}
6989
6990 // Store CMS card-mark Immediate
6991 instruct storeImmCM0_reg(memory mem, immI0 zero)
6992 %{
6993 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6994 match(Set mem (StoreCM mem zero));
6995
6996 ins_cost(125); // XXX
6997 format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %}
6998 ins_encode %{
6999 __ movb($mem$$Address, r12);
7000 %}
7001 ins_pipe(ialu_mem_reg);
7002 %}
7003
7004 instruct storeImmCM0(memory mem, immI0 src)
7005 %{
7006 match(Set mem (StoreCM mem src));
7007
7008 ins_cost(150); // XXX
7009 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
7010 opcode(0xC6); /* C6 /0 */
7011 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7012 ins_pipe(ialu_mem_imm);
7013 %}
7014
7015 // Store Aligned Packed Single Float XMM register to memory
7016 instruct storeA2F(memory mem, regD src) %{
7017 match(Set mem (Store2F mem src));
7018 ins_cost(145);
7019 format %{ "MOVQ $mem,$src\t! packed2F" %}
7020 ins_encode( movq_st(mem, src));
7021 ins_pipe( pipe_slow );
7022 %}
7023
7024 // Store Float
7025 instruct storeF(memory mem, regF src)
7026 %{
7027 match(Set mem (StoreF mem src));
7028
7029 ins_cost(95); // XXX
7030 format %{ "movss $mem, $src\t# float" %}
7031 opcode(0xF3, 0x0F, 0x11);
7032 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7033 ins_pipe(pipe_slow); // XXX
7034 %}
7035
7036 // Store immediate Float value (it is faster than store from XMM register)
7037 instruct storeF0(memory mem, immF0 zero)
7038 %{
7039 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7040 match(Set mem (StoreF mem zero));
7041
7042 ins_cost(25); // XXX
7043 format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %}
7044 ins_encode %{
7045 __ movl($mem$$Address, r12);
7046 %}
7047 ins_pipe(ialu_mem_reg);
7048 %}
7049
7050 instruct storeF_imm(memory mem, immF src)
7051 %{
7052 match(Set mem (StoreF mem src));
7053
7054 ins_cost(50);
7055 format %{ "movl $mem, $src\t# float" %}
7056 opcode(0xC7); /* C7 /0 */
7057 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7058 ins_pipe(ialu_mem_imm);
7059 %}
7060
7061 // Store Double
7062 instruct storeD(memory mem, regD src)
7063 %{
7064 match(Set mem (StoreD mem src));
7065
7066 ins_cost(95); // XXX
7067 format %{ "movsd $mem, $src\t# double" %}
7068 opcode(0xF2, 0x0F, 0x11);
7069 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7070 ins_pipe(pipe_slow); // XXX
7071 %}
7072
7073 // Store immediate double 0.0 (it is faster than store from XMM register)
7074 instruct storeD0_imm(memory mem, immD0 src)
7075 %{
7076 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
7077 match(Set mem (StoreD mem src));
7078
7079 ins_cost(50);
7080 format %{ "movq $mem, $src\t# double 0." %}
7081 opcode(0xC7); /* C7 /0 */
7082 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7083 ins_pipe(ialu_mem_imm);
7084 %}
7085
7086 instruct storeD0(memory mem, immD0 zero)
7087 %{
7088 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7089 match(Set mem (StoreD mem zero));
7090
7091 ins_cost(25); // XXX
7092 format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %}
7093 ins_encode %{
7094 __ movq($mem$$Address, r12);
7095 %}
7096 ins_pipe(ialu_mem_reg);
7097 %}
7098
7099 instruct storeSSI(stackSlotI dst, rRegI src)
7100 %{
7101 match(Set dst src);
7102
7103 ins_cost(100);
7104 format %{ "movl $dst, $src\t# int stk" %}
7105 opcode(0x89);
7106 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7107 ins_pipe( ialu_mem_reg );
7108 %}
7109
7110 instruct storeSSL(stackSlotL dst, rRegL src)
7111 %{
7112 match(Set dst src);
7113
7114 ins_cost(100);
7115 format %{ "movq $dst, $src\t# long stk" %}
7116 opcode(0x89);
7117 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7118 ins_pipe(ialu_mem_reg);
7119 %}
7120
7121 instruct storeSSP(stackSlotP dst, rRegP src)
7122 %{
7123 match(Set dst src);
7124
7125 ins_cost(100);
7126 format %{ "movq $dst, $src\t# ptr stk" %}
7127 opcode(0x89);
7128 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7129 ins_pipe(ialu_mem_reg);
7130 %}
7131
7132 instruct storeSSF(stackSlotF dst, regF src)
7133 %{
7134 match(Set dst src);
7135
7136 ins_cost(95); // XXX
7137 format %{ "movss $dst, $src\t# float stk" %}
7138 opcode(0xF3, 0x0F, 0x11);
7139 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7140 ins_pipe(pipe_slow); // XXX
7141 %}
7142
7143 instruct storeSSD(stackSlotD dst, regD src)
7144 %{
7145 match(Set dst src);
7146
7147 ins_cost(95); // XXX
7148 format %{ "movsd $dst, $src\t# double stk" %}
7149 opcode(0xF2, 0x0F, 0x11);
7150 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7151 ins_pipe(pipe_slow); // XXX
7152 %}
7153
7154 //----------BSWAP Instructions-------------------------------------------------
7155 instruct bytes_reverse_int(rRegI dst) %{
7156 match(Set dst (ReverseBytesI dst));
7157
7158 format %{ "bswapl $dst" %}
7159 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
7160 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
7161 ins_pipe( ialu_reg );
7162 %}
7163
7164 instruct bytes_reverse_long(rRegL dst) %{
7165 match(Set dst (ReverseBytesL dst));
7166
7167 format %{ "bswapq $dst" %}
7168
7169 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
7170 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
7171 ins_pipe( ialu_reg);
7172 %}
7173
7174 instruct bytes_reverse_unsigned_short(rRegI dst) %{
7175 match(Set dst (ReverseBytesUS dst));
7176
7177 format %{ "bswapl $dst\n\t"
7178 "shrl $dst,16\n\t" %}
7179 ins_encode %{
7180 __ bswapl($dst$$Register);
7181 __ shrl($dst$$Register, 16);
7182 %}
7183 ins_pipe( ialu_reg );
7184 %}
7185
7186 instruct bytes_reverse_short(rRegI dst) %{
7187 match(Set dst (ReverseBytesS dst));
7188
7189 format %{ "bswapl $dst\n\t"
7190 "sar $dst,16\n\t" %}
7191 ins_encode %{
7192 __ bswapl($dst$$Register);
7193 __ sarl($dst$$Register, 16);
7194 %}
7195 ins_pipe( ialu_reg );
7196 %}
7197
7198 //---------- Zeros Count Instructions ------------------------------------------
7199
7200 instruct countLeadingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7201 predicate(UseCountLeadingZerosInstruction);
7202 match(Set dst (CountLeadingZerosI src));
7203 effect(KILL cr);
7204
7205 format %{ "lzcntl $dst, $src\t# count leading zeros (int)" %}
7206 ins_encode %{
7207 __ lzcntl($dst$$Register, $src$$Register);
7208 %}
7209 ins_pipe(ialu_reg);
7210 %}
7211
7212 instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, rFlagsReg cr) %{
7213 predicate(!UseCountLeadingZerosInstruction);
7214 match(Set dst (CountLeadingZerosI src));
7215 effect(KILL cr);
7216
7217 format %{ "bsrl $dst, $src\t# count leading zeros (int)\n\t"
7218 "jnz skip\n\t"
7219 "movl $dst, -1\n"
7220 "skip:\n\t"
7221 "negl $dst\n\t"
7222 "addl $dst, 31" %}
7223 ins_encode %{
7224 Register Rdst = $dst$$Register;
7225 Register Rsrc = $src$$Register;
7226 Label skip;
7227 __ bsrl(Rdst, Rsrc);
7228 __ jccb(Assembler::notZero, skip);
7229 __ movl(Rdst, -1);
7230 __ bind(skip);
7231 __ negl(Rdst);
7232 __ addl(Rdst, BitsPerInt - 1);
7233 %}
7234 ins_pipe(ialu_reg);
7235 %}
7236
7237 instruct countLeadingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7238 predicate(UseCountLeadingZerosInstruction);
7239 match(Set dst (CountLeadingZerosL src));
7240 effect(KILL cr);
7241
7242 format %{ "lzcntq $dst, $src\t# count leading zeros (long)" %}
7243 ins_encode %{
7244 __ lzcntq($dst$$Register, $src$$Register);
7245 %}
7246 ins_pipe(ialu_reg);
7247 %}
7248
7249 instruct countLeadingZerosL_bsr(rRegI dst, rRegL src, rFlagsReg cr) %{
7250 predicate(!UseCountLeadingZerosInstruction);
7251 match(Set dst (CountLeadingZerosL src));
7252 effect(KILL cr);
7253
7254 format %{ "bsrq $dst, $src\t# count leading zeros (long)\n\t"
7255 "jnz skip\n\t"
7256 "movl $dst, -1\n"
7257 "skip:\n\t"
7258 "negl $dst\n\t"
7259 "addl $dst, 63" %}
7260 ins_encode %{
7261 Register Rdst = $dst$$Register;
7262 Register Rsrc = $src$$Register;
7263 Label skip;
7264 __ bsrq(Rdst, Rsrc);
7265 __ jccb(Assembler::notZero, skip);
7266 __ movl(Rdst, -1);
7267 __ bind(skip);
7268 __ negl(Rdst);
7269 __ addl(Rdst, BitsPerLong - 1);
7270 %}
7271 ins_pipe(ialu_reg);
7272 %}
7273
7274 instruct countTrailingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7275 match(Set dst (CountTrailingZerosI src));
7276 effect(KILL cr);
7277
7278 format %{ "bsfl $dst, $src\t# count trailing zeros (int)\n\t"
7279 "jnz done\n\t"
7280 "movl $dst, 32\n"
7281 "done:" %}
7282 ins_encode %{
7283 Register Rdst = $dst$$Register;
7284 Label done;
7285 __ bsfl(Rdst, $src$$Register);
7286 __ jccb(Assembler::notZero, done);
7287 __ movl(Rdst, BitsPerInt);
7288 __ bind(done);
7289 %}
7290 ins_pipe(ialu_reg);
7291 %}
7292
7293 instruct countTrailingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7294 match(Set dst (CountTrailingZerosL src));
7295 effect(KILL cr);
7296
7297 format %{ "bsfq $dst, $src\t# count trailing zeros (long)\n\t"
7298 "jnz done\n\t"
7299 "movl $dst, 64\n"
7300 "done:" %}
7301 ins_encode %{
7302 Register Rdst = $dst$$Register;
7303 Label done;
7304 __ bsfq(Rdst, $src$$Register);
7305 __ jccb(Assembler::notZero, done);
7306 __ movl(Rdst, BitsPerLong);
7307 __ bind(done);
7308 %}
7309 ins_pipe(ialu_reg);
7310 %}
7311
7312
7313 //---------- Population Count Instructions -------------------------------------
7314
7315 instruct popCountI(rRegI dst, rRegI src) %{
7316 predicate(UsePopCountInstruction);
7317 match(Set dst (PopCountI src));
7318
7319 format %{ "popcnt $dst, $src" %}
7320 ins_encode %{
7321 __ popcntl($dst$$Register, $src$$Register);
7322 %}
7323 ins_pipe(ialu_reg);
7324 %}
7325
7326 instruct popCountI_mem(rRegI dst, memory mem) %{
7327 predicate(UsePopCountInstruction);
7328 match(Set dst (PopCountI (LoadI mem)));
7329
7330 format %{ "popcnt $dst, $mem" %}
7331 ins_encode %{
7332 __ popcntl($dst$$Register, $mem$$Address);
7333 %}
7334 ins_pipe(ialu_reg);
7335 %}
7336
7337 // Note: Long.bitCount(long) returns an int.
7338 instruct popCountL(rRegI dst, rRegL src) %{
7339 predicate(UsePopCountInstruction);
7340 match(Set dst (PopCountL src));
7341
7342 format %{ "popcnt $dst, $src" %}
7343 ins_encode %{
7344 __ popcntq($dst$$Register, $src$$Register);
7345 %}
7346 ins_pipe(ialu_reg);
7347 %}
7348
7349 // Note: Long.bitCount(long) returns an int.
7350 instruct popCountL_mem(rRegI dst, memory mem) %{
7351 predicate(UsePopCountInstruction);
7352 match(Set dst (PopCountL (LoadL mem)));
7353
7354 format %{ "popcnt $dst, $mem" %}
7355 ins_encode %{
7356 __ popcntq($dst$$Register, $mem$$Address);
7357 %}
7358 ins_pipe(ialu_reg);
7359 %}
7360
7361
7362 //----------MemBar Instructions-----------------------------------------------
7363 // Memory barrier flavors
7364
7365 instruct membar_acquire()
7366 %{
7367 match(MemBarAcquire);
7368 ins_cost(0);
7369
7370 size(0);
7371 format %{ "MEMBAR-acquire ! (empty encoding)" %}
7372 ins_encode();
7373 ins_pipe(empty);
7374 %}
7375
7376 instruct membar_acquire_lock()
7377 %{
7378 match(MemBarAcquireLock);
7379 ins_cost(0);
7380
7381 size(0);
7382 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7383 ins_encode();
7384 ins_pipe(empty);
7385 %}
7386
7387 instruct membar_release()
7388 %{
7389 match(MemBarRelease);
7390 ins_cost(0);
7391
7392 size(0);
7393 format %{ "MEMBAR-release ! (empty encoding)" %}
7394 ins_encode();
7395 ins_pipe(empty);
7396 %}
7397
7398 instruct membar_release_lock()
7399 %{
7400 match(MemBarReleaseLock);
7401 ins_cost(0);
7402
7403 size(0);
7404 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7405 ins_encode();
7406 ins_pipe(empty);
7407 %}
7408
7409 instruct membar_volatile(rFlagsReg cr) %{
7410 match(MemBarVolatile);
7411 effect(KILL cr);
7412 ins_cost(400);
7413
7414 format %{
7415 $$template
7416 if (os::is_MP()) {
7417 $$emit$$"lock addl [rsp + #0], 0\t! membar_volatile"
7418 } else {
7419 $$emit$$"MEMBAR-volatile ! (empty encoding)"
7420 }
7421 %}
7422 ins_encode %{
7423 __ membar(Assembler::StoreLoad);
7424 %}
7425 ins_pipe(pipe_slow);
7426 %}
7427
7428 instruct unnecessary_membar_volatile()
7429 %{
7430 match(MemBarVolatile);
7431 predicate(Matcher::post_store_load_barrier(n));
7432 ins_cost(0);
7433
7434 size(0);
7435 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7436 ins_encode();
7437 ins_pipe(empty);
7438 %}
7439
7440 //----------Move Instructions--------------------------------------------------
7441
7442 instruct castX2P(rRegP dst, rRegL src)
7443 %{
7444 match(Set dst (CastX2P src));
7445
7446 format %{ "movq $dst, $src\t# long->ptr" %}
7447 ins_encode(enc_copy_wide(dst, src));
7448 ins_pipe(ialu_reg_reg); // XXX
7449 %}
7450
7451 instruct castP2X(rRegL dst, rRegP src)
7452 %{
7453 match(Set dst (CastP2X src));
7454
7455 format %{ "movq $dst, $src\t# ptr -> long" %}
7456 ins_encode(enc_copy_wide(dst, src));
7457 ins_pipe(ialu_reg_reg); // XXX
7458 %}
7459
7460
7461 // Convert oop pointer into compressed form
7462 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7463 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7464 match(Set dst (EncodeP src));
7465 effect(KILL cr);
7466 format %{ "encode_heap_oop $dst,$src" %}
7467 ins_encode %{
7468 Register s = $src$$Register;
7469 Register d = $dst$$Register;
7470 if (s != d) {
7471 __ movq(d, s);
7472 }
7473 __ encode_heap_oop(d);
7474 %}
7475 ins_pipe(ialu_reg_long);
7476 %}
7477
7478 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7479 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7480 match(Set dst (EncodeP src));
7481 effect(KILL cr);
7482 format %{ "encode_heap_oop_not_null $dst,$src" %}
7483 ins_encode %{
7484 __ encode_heap_oop_not_null($dst$$Register, $src$$Register);
7485 %}
7486 ins_pipe(ialu_reg_long);
7487 %}
7488
7489 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7490 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7491 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7492 match(Set dst (DecodeN src));
7493 effect(KILL cr);
7494 format %{ "decode_heap_oop $dst,$src" %}
7495 ins_encode %{
7496 Register s = $src$$Register;
7497 Register d = $dst$$Register;
7498 if (s != d) {
7499 __ movq(d, s);
7500 }
7501 __ decode_heap_oop(d);
7502 %}
7503 ins_pipe(ialu_reg_long);
7504 %}
7505
7506 instruct decodeHeapOop_not_null(rRegP dst, rRegN src, rFlagsReg cr) %{
7507 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7508 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7509 match(Set dst (DecodeN src));
7510 effect(KILL cr);
7511 format %{ "decode_heap_oop_not_null $dst,$src" %}
7512 ins_encode %{
7513 Register s = $src$$Register;
7514 Register d = $dst$$Register;
7515 if (s != d) {
7516 __ decode_heap_oop_not_null(d, s);
7517 } else {
7518 __ decode_heap_oop_not_null(d);
7519 }
7520 %}
7521 ins_pipe(ialu_reg_long);
7522 %}
7523
7524
7525 //----------Conditional Move---------------------------------------------------
7526 // Jump
7527 // dummy instruction for generating temp registers
7528 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7529 match(Jump (LShiftL switch_val shift));
7530 ins_cost(350);
7531 predicate(false);
7532 effect(TEMP dest);
7533
7534 format %{ "leaq $dest, [$constantaddress]\n\t"
7535 "jmp [$dest + $switch_val << $shift]\n\t" %}
7536 ins_encode %{
7537 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7538 // to do that and the compiler is using that register as one it can allocate.
7539 // So we build it all by hand.
7540 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
7541 // ArrayAddress dispatch(table, index);
7542 Address dispatch($dest$$Register, $switch_val$$Register, (Address::ScaleFactor) $shift$$constant);
7543 __ lea($dest$$Register, $constantaddress);
7544 __ jmp(dispatch);
7545 %}
7546 ins_pipe(pipe_jmp);
7547 %}
7548
7549 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7550 match(Jump (AddL (LShiftL switch_val shift) offset));
7551 ins_cost(350);
7552 effect(TEMP dest);
7553
7554 format %{ "leaq $dest, [$constantaddress]\n\t"
7555 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7556 ins_encode %{
7557 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7558 // to do that and the compiler is using that register as one it can allocate.
7559 // So we build it all by hand.
7560 // Address index(noreg, switch_reg, (Address::ScaleFactor) $shift$$constant, (int) $offset$$constant);
7561 // ArrayAddress dispatch(table, index);
7562 Address dispatch($dest$$Register, $switch_val$$Register, (Address::ScaleFactor) $shift$$constant, (int) $offset$$constant);
7563 __ lea($dest$$Register, $constantaddress);
7564 __ jmp(dispatch);
7565 %}
7566 ins_pipe(pipe_jmp);
7567 %}
7568
7569 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7570 match(Jump switch_val);
7571 ins_cost(350);
7572 effect(TEMP dest);
7573
7574 format %{ "leaq $dest, [$constantaddress]\n\t"
7575 "jmp [$dest + $switch_val]\n\t" %}
7576 ins_encode %{
7577 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7578 // to do that and the compiler is using that register as one it can allocate.
7579 // So we build it all by hand.
7580 // Address index(noreg, switch_reg, Address::times_1);
7581 // ArrayAddress dispatch(table, index);
7582 Address dispatch($dest$$Register, $switch_val$$Register, Address::times_1);
7583 __ lea($dest$$Register, $constantaddress);
7584 __ jmp(dispatch);
7585 %}
7586 ins_pipe(pipe_jmp);
7587 %}
7588
7589 // Conditional move
7590 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7591 %{
7592 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7593
7594 ins_cost(200); // XXX
7595 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7596 opcode(0x0F, 0x40);
7597 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7598 ins_pipe(pipe_cmov_reg);
7599 %}
7600
7601 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7602 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7603
7604 ins_cost(200); // XXX
7605 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7606 opcode(0x0F, 0x40);
7607 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7608 ins_pipe(pipe_cmov_reg);
7609 %}
7610
7611 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7612 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7613 ins_cost(200);
7614 expand %{
7615 cmovI_regU(cop, cr, dst, src);
7616 %}
7617 %}
7618
7619 // Conditional move
7620 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7621 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7622
7623 ins_cost(250); // XXX
7624 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7625 opcode(0x0F, 0x40);
7626 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7627 ins_pipe(pipe_cmov_mem);
7628 %}
7629
7630 // Conditional move
7631 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7632 %{
7633 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7634
7635 ins_cost(250); // XXX
7636 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7637 opcode(0x0F, 0x40);
7638 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7639 ins_pipe(pipe_cmov_mem);
7640 %}
7641
7642 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7643 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7644 ins_cost(250);
7645 expand %{
7646 cmovI_memU(cop, cr, dst, src);
7647 %}
7648 %}
7649
7650 // Conditional move
7651 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7652 %{
7653 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7654
7655 ins_cost(200); // XXX
7656 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7657 opcode(0x0F, 0x40);
7658 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7659 ins_pipe(pipe_cmov_reg);
7660 %}
7661
7662 // Conditional move
7663 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7664 %{
7665 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7666
7667 ins_cost(200); // XXX
7668 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7669 opcode(0x0F, 0x40);
7670 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7671 ins_pipe(pipe_cmov_reg);
7672 %}
7673
7674 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7675 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7676 ins_cost(200);
7677 expand %{
7678 cmovN_regU(cop, cr, dst, src);
7679 %}
7680 %}
7681
7682 // Conditional move
7683 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7684 %{
7685 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7686
7687 ins_cost(200); // XXX
7688 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7689 opcode(0x0F, 0x40);
7690 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7691 ins_pipe(pipe_cmov_reg); // XXX
7692 %}
7693
7694 // Conditional move
7695 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7696 %{
7697 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7698
7699 ins_cost(200); // XXX
7700 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7701 opcode(0x0F, 0x40);
7702 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7703 ins_pipe(pipe_cmov_reg); // XXX
7704 %}
7705
7706 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7707 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7708 ins_cost(200);
7709 expand %{
7710 cmovP_regU(cop, cr, dst, src);
7711 %}
7712 %}
7713
7714 // DISABLED: Requires the ADLC to emit a bottom_type call that
7715 // correctly meets the two pointer arguments; one is an incoming
7716 // register but the other is a memory operand. ALSO appears to
7717 // be buggy with implicit null checks.
7718 //
7719 //// Conditional move
7720 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7721 //%{
7722 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7723 // ins_cost(250);
7724 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7725 // opcode(0x0F,0x40);
7726 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7727 // ins_pipe( pipe_cmov_mem );
7728 //%}
7729 //
7730 //// Conditional move
7731 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7732 //%{
7733 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7734 // ins_cost(250);
7735 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7736 // opcode(0x0F,0x40);
7737 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7738 // ins_pipe( pipe_cmov_mem );
7739 //%}
7740
7741 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7742 %{
7743 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7744
7745 ins_cost(200); // XXX
7746 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7747 opcode(0x0F, 0x40);
7748 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7749 ins_pipe(pipe_cmov_reg); // XXX
7750 %}
7751
7752 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7753 %{
7754 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7755
7756 ins_cost(200); // XXX
7757 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7758 opcode(0x0F, 0x40);
7759 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7760 ins_pipe(pipe_cmov_mem); // XXX
7761 %}
7762
7763 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7764 %{
7765 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7766
7767 ins_cost(200); // XXX
7768 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7769 opcode(0x0F, 0x40);
7770 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7771 ins_pipe(pipe_cmov_reg); // XXX
7772 %}
7773
7774 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7775 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7776 ins_cost(200);
7777 expand %{
7778 cmovL_regU(cop, cr, dst, src);
7779 %}
7780 %}
7781
7782 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7783 %{
7784 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7785
7786 ins_cost(200); // XXX
7787 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7788 opcode(0x0F, 0x40);
7789 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7790 ins_pipe(pipe_cmov_mem); // XXX
7791 %}
7792
7793 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
7794 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7795 ins_cost(200);
7796 expand %{
7797 cmovL_memU(cop, cr, dst, src);
7798 %}
7799 %}
7800
7801 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7802 %{
7803 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7804
7805 ins_cost(200); // XXX
7806 format %{ "jn$cop skip\t# signed cmove float\n\t"
7807 "movss $dst, $src\n"
7808 "skip:" %}
7809 ins_encode(enc_cmovf_branch(cop, dst, src));
7810 ins_pipe(pipe_slow);
7811 %}
7812
7813 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7814 // %{
7815 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7816
7817 // ins_cost(200); // XXX
7818 // format %{ "jn$cop skip\t# signed cmove float\n\t"
7819 // "movss $dst, $src\n"
7820 // "skip:" %}
7821 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7822 // ins_pipe(pipe_slow);
7823 // %}
7824
7825 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7826 %{
7827 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7828
7829 ins_cost(200); // XXX
7830 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
7831 "movss $dst, $src\n"
7832 "skip:" %}
7833 ins_encode(enc_cmovf_branch(cop, dst, src));
7834 ins_pipe(pipe_slow);
7835 %}
7836
7837 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
7838 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7839 ins_cost(200);
7840 expand %{
7841 cmovF_regU(cop, cr, dst, src);
7842 %}
7843 %}
7844
7845 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7846 %{
7847 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7848
7849 ins_cost(200); // XXX
7850 format %{ "jn$cop skip\t# signed cmove double\n\t"
7851 "movsd $dst, $src\n"
7852 "skip:" %}
7853 ins_encode(enc_cmovd_branch(cop, dst, src));
7854 ins_pipe(pipe_slow);
7855 %}
7856
7857 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7858 %{
7859 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7860
7861 ins_cost(200); // XXX
7862 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
7863 "movsd $dst, $src\n"
7864 "skip:" %}
7865 ins_encode(enc_cmovd_branch(cop, dst, src));
7866 ins_pipe(pipe_slow);
7867 %}
7868
7869 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
7870 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7871 ins_cost(200);
7872 expand %{
7873 cmovD_regU(cop, cr, dst, src);
7874 %}
7875 %}
7876
7877 //----------Arithmetic Instructions--------------------------------------------
7878 //----------Addition Instructions----------------------------------------------
7879
7880 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7881 %{
7882 match(Set dst (AddI dst src));
7883 effect(KILL cr);
7884
7885 format %{ "addl $dst, $src\t# int" %}
7886 opcode(0x03);
7887 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7888 ins_pipe(ialu_reg_reg);
7889 %}
7890
7891 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7892 %{
7893 match(Set dst (AddI dst src));
7894 effect(KILL cr);
7895
7896 format %{ "addl $dst, $src\t# int" %}
7897 opcode(0x81, 0x00); /* /0 id */
7898 ins_encode(OpcSErm(dst, src), Con8or32(src));
7899 ins_pipe( ialu_reg );
7900 %}
7901
7902 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7903 %{
7904 match(Set dst (AddI dst (LoadI src)));
7905 effect(KILL cr);
7906
7907 ins_cost(125); // XXX
7908 format %{ "addl $dst, $src\t# int" %}
7909 opcode(0x03);
7910 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7911 ins_pipe(ialu_reg_mem);
7912 %}
7913
7914 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7915 %{
7916 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7917 effect(KILL cr);
7918
7919 ins_cost(150); // XXX
7920 format %{ "addl $dst, $src\t# int" %}
7921 opcode(0x01); /* Opcode 01 /r */
7922 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7923 ins_pipe(ialu_mem_reg);
7924 %}
7925
7926 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7927 %{
7928 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7929 effect(KILL cr);
7930
7931 ins_cost(125); // XXX
7932 format %{ "addl $dst, $src\t# int" %}
7933 opcode(0x81); /* Opcode 81 /0 id */
7934 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7935 ins_pipe(ialu_mem_imm);
7936 %}
7937
7938 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7939 %{
7940 predicate(UseIncDec);
7941 match(Set dst (AddI dst src));
7942 effect(KILL cr);
7943
7944 format %{ "incl $dst\t# int" %}
7945 opcode(0xFF, 0x00); // FF /0
7946 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7947 ins_pipe(ialu_reg);
7948 %}
7949
7950 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
7951 %{
7952 predicate(UseIncDec);
7953 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7954 effect(KILL cr);
7955
7956 ins_cost(125); // XXX
7957 format %{ "incl $dst\t# int" %}
7958 opcode(0xFF); /* Opcode FF /0 */
7959 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
7960 ins_pipe(ialu_mem_imm);
7961 %}
7962
7963 // XXX why does that use AddI
7964 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
7965 %{
7966 predicate(UseIncDec);
7967 match(Set dst (AddI dst src));
7968 effect(KILL cr);
7969
7970 format %{ "decl $dst\t# int" %}
7971 opcode(0xFF, 0x01); // FF /1
7972 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7973 ins_pipe(ialu_reg);
7974 %}
7975
7976 // XXX why does that use AddI
7977 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
7978 %{
7979 predicate(UseIncDec);
7980 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7981 effect(KILL cr);
7982
7983 ins_cost(125); // XXX
7984 format %{ "decl $dst\t# int" %}
7985 opcode(0xFF); /* Opcode FF /1 */
7986 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
7987 ins_pipe(ialu_mem_imm);
7988 %}
7989
7990 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
7991 %{
7992 match(Set dst (AddI src0 src1));
7993
7994 ins_cost(110);
7995 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
7996 opcode(0x8D); /* 0x8D /r */
7997 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7998 ins_pipe(ialu_reg_reg);
7999 %}
8000
8001 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8002 %{
8003 match(Set dst (AddL dst src));
8004 effect(KILL cr);
8005
8006 format %{ "addq $dst, $src\t# long" %}
8007 opcode(0x03);
8008 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8009 ins_pipe(ialu_reg_reg);
8010 %}
8011
8012 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
8013 %{
8014 match(Set dst (AddL dst src));
8015 effect(KILL cr);
8016
8017 format %{ "addq $dst, $src\t# long" %}
8018 opcode(0x81, 0x00); /* /0 id */
8019 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8020 ins_pipe( ialu_reg );
8021 %}
8022
8023 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8024 %{
8025 match(Set dst (AddL dst (LoadL src)));
8026 effect(KILL cr);
8027
8028 ins_cost(125); // XXX
8029 format %{ "addq $dst, $src\t# long" %}
8030 opcode(0x03);
8031 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8032 ins_pipe(ialu_reg_mem);
8033 %}
8034
8035 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8036 %{
8037 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8038 effect(KILL cr);
8039
8040 ins_cost(150); // XXX
8041 format %{ "addq $dst, $src\t# long" %}
8042 opcode(0x01); /* Opcode 01 /r */
8043 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8044 ins_pipe(ialu_mem_reg);
8045 %}
8046
8047 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8048 %{
8049 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8050 effect(KILL cr);
8051
8052 ins_cost(125); // XXX
8053 format %{ "addq $dst, $src\t# long" %}
8054 opcode(0x81); /* Opcode 81 /0 id */
8055 ins_encode(REX_mem_wide(dst),
8056 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
8057 ins_pipe(ialu_mem_imm);
8058 %}
8059
8060 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
8061 %{
8062 predicate(UseIncDec);
8063 match(Set dst (AddL dst src));
8064 effect(KILL cr);
8065
8066 format %{ "incq $dst\t# long" %}
8067 opcode(0xFF, 0x00); // FF /0
8068 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8069 ins_pipe(ialu_reg);
8070 %}
8071
8072 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
8073 %{
8074 predicate(UseIncDec);
8075 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8076 effect(KILL cr);
8077
8078 ins_cost(125); // XXX
8079 format %{ "incq $dst\t# long" %}
8080 opcode(0xFF); /* Opcode FF /0 */
8081 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
8082 ins_pipe(ialu_mem_imm);
8083 %}
8084
8085 // XXX why does that use AddL
8086 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
8087 %{
8088 predicate(UseIncDec);
8089 match(Set dst (AddL dst src));
8090 effect(KILL cr);
8091
8092 format %{ "decq $dst\t# long" %}
8093 opcode(0xFF, 0x01); // FF /1
8094 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8095 ins_pipe(ialu_reg);
8096 %}
8097
8098 // XXX why does that use AddL
8099 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
8100 %{
8101 predicate(UseIncDec);
8102 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8103 effect(KILL cr);
8104
8105 ins_cost(125); // XXX
8106 format %{ "decq $dst\t# long" %}
8107 opcode(0xFF); /* Opcode FF /1 */
8108 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
8109 ins_pipe(ialu_mem_imm);
8110 %}
8111
8112 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
8113 %{
8114 match(Set dst (AddL src0 src1));
8115
8116 ins_cost(110);
8117 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
8118 opcode(0x8D); /* 0x8D /r */
8119 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8120 ins_pipe(ialu_reg_reg);
8121 %}
8122
8123 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
8124 %{
8125 match(Set dst (AddP dst src));
8126 effect(KILL cr);
8127
8128 format %{ "addq $dst, $src\t# ptr" %}
8129 opcode(0x03);
8130 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8131 ins_pipe(ialu_reg_reg);
8132 %}
8133
8134 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
8135 %{
8136 match(Set dst (AddP dst src));
8137 effect(KILL cr);
8138
8139 format %{ "addq $dst, $src\t# ptr" %}
8140 opcode(0x81, 0x00); /* /0 id */
8141 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8142 ins_pipe( ialu_reg );
8143 %}
8144
8145 // XXX addP mem ops ????
8146
8147 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
8148 %{
8149 match(Set dst (AddP src0 src1));
8150
8151 ins_cost(110);
8152 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
8153 opcode(0x8D); /* 0x8D /r */
8154 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
8155 ins_pipe(ialu_reg_reg);
8156 %}
8157
8158 instruct checkCastPP(rRegP dst)
8159 %{
8160 match(Set dst (CheckCastPP dst));
8161
8162 size(0);
8163 format %{ "# checkcastPP of $dst" %}
8164 ins_encode(/* empty encoding */);
8165 ins_pipe(empty);
8166 %}
8167
8168 instruct castPP(rRegP dst)
8169 %{
8170 match(Set dst (CastPP dst));
8171
8172 size(0);
8173 format %{ "# castPP of $dst" %}
8174 ins_encode(/* empty encoding */);
8175 ins_pipe(empty);
8176 %}
8177
8178 instruct castII(rRegI dst)
8179 %{
8180 match(Set dst (CastII dst));
8181
8182 size(0);
8183 format %{ "# castII of $dst" %}
8184 ins_encode(/* empty encoding */);
8185 ins_cost(0);
8186 ins_pipe(empty);
8187 %}
8188
8189 // LoadP-locked same as a regular LoadP when used with compare-swap
8190 instruct loadPLocked(rRegP dst, memory mem)
8191 %{
8192 match(Set dst (LoadPLocked mem));
8193
8194 ins_cost(125); // XXX
8195 format %{ "movq $dst, $mem\t# ptr locked" %}
8196 opcode(0x8B);
8197 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8198 ins_pipe(ialu_reg_mem); // XXX
8199 %}
8200
8201 // LoadL-locked - same as a regular LoadL when used with compare-swap
8202 instruct loadLLocked(rRegL dst, memory mem)
8203 %{
8204 match(Set dst (LoadLLocked mem));
8205
8206 ins_cost(125); // XXX
8207 format %{ "movq $dst, $mem\t# long locked" %}
8208 opcode(0x8B);
8209 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8210 ins_pipe(ialu_reg_mem); // XXX
8211 %}
8212
8213 // Conditional-store of the updated heap-top.
8214 // Used during allocation of the shared heap.
8215 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
8216
8217 instruct storePConditional(memory heap_top_ptr,
8218 rax_RegP oldval, rRegP newval,
8219 rFlagsReg cr)
8220 %{
8221 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
8222
8223 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
8224 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
8225 opcode(0x0F, 0xB1);
8226 ins_encode(lock_prefix,
8227 REX_reg_mem_wide(newval, heap_top_ptr),
8228 OpcP, OpcS,
8229 reg_mem(newval, heap_top_ptr));
8230 ins_pipe(pipe_cmpxchg);
8231 %}
8232
8233 // Conditional-store of an int value.
8234 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8235 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
8236 %{
8237 match(Set cr (StoreIConditional mem (Binary oldval newval)));
8238 effect(KILL oldval);
8239
8240 format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8241 opcode(0x0F, 0xB1);
8242 ins_encode(lock_prefix,
8243 REX_reg_mem(newval, mem),
8244 OpcP, OpcS,
8245 reg_mem(newval, mem));
8246 ins_pipe(pipe_cmpxchg);
8247 %}
8248
8249 // Conditional-store of a long value.
8250 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8251 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
8252 %{
8253 match(Set cr (StoreLConditional mem (Binary oldval newval)));
8254 effect(KILL oldval);
8255
8256 format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8257 opcode(0x0F, 0xB1);
8258 ins_encode(lock_prefix,
8259 REX_reg_mem_wide(newval, mem),
8260 OpcP, OpcS,
8261 reg_mem(newval, mem));
8262 ins_pipe(pipe_cmpxchg);
8263 %}
8264
8265
8266 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
8267 instruct compareAndSwapP(rRegI res,
8268 memory mem_ptr,
8269 rax_RegP oldval, rRegP newval,
8270 rFlagsReg cr)
8271 %{
8272 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
8273 effect(KILL cr, KILL oldval);
8274
8275 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8276 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8277 "sete $res\n\t"
8278 "movzbl $res, $res" %}
8279 opcode(0x0F, 0xB1);
8280 ins_encode(lock_prefix,
8281 REX_reg_mem_wide(newval, mem_ptr),
8282 OpcP, OpcS,
8283 reg_mem(newval, mem_ptr),
8284 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8285 REX_reg_breg(res, res), // movzbl
8286 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8287 ins_pipe( pipe_cmpxchg );
8288 %}
8289
8290 instruct compareAndSwapL(rRegI res,
8291 memory mem_ptr,
8292 rax_RegL oldval, rRegL newval,
8293 rFlagsReg cr)
8294 %{
8295 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
8296 effect(KILL cr, KILL oldval);
8297
8298 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8299 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8300 "sete $res\n\t"
8301 "movzbl $res, $res" %}
8302 opcode(0x0F, 0xB1);
8303 ins_encode(lock_prefix,
8304 REX_reg_mem_wide(newval, mem_ptr),
8305 OpcP, OpcS,
8306 reg_mem(newval, mem_ptr),
8307 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8308 REX_reg_breg(res, res), // movzbl
8309 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8310 ins_pipe( pipe_cmpxchg );
8311 %}
8312
8313 instruct compareAndSwapI(rRegI res,
8314 memory mem_ptr,
8315 rax_RegI oldval, rRegI newval,
8316 rFlagsReg cr)
8317 %{
8318 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
8319 effect(KILL cr, KILL oldval);
8320
8321 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8322 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8323 "sete $res\n\t"
8324 "movzbl $res, $res" %}
8325 opcode(0x0F, 0xB1);
8326 ins_encode(lock_prefix,
8327 REX_reg_mem(newval, mem_ptr),
8328 OpcP, OpcS,
8329 reg_mem(newval, mem_ptr),
8330 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8331 REX_reg_breg(res, res), // movzbl
8332 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8333 ins_pipe( pipe_cmpxchg );
8334 %}
8335
8336
8337 instruct compareAndSwapN(rRegI res,
8338 memory mem_ptr,
8339 rax_RegN oldval, rRegN newval,
8340 rFlagsReg cr) %{
8341 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
8342 effect(KILL cr, KILL oldval);
8343
8344 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8345 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8346 "sete $res\n\t"
8347 "movzbl $res, $res" %}
8348 opcode(0x0F, 0xB1);
8349 ins_encode(lock_prefix,
8350 REX_reg_mem(newval, mem_ptr),
8351 OpcP, OpcS,
8352 reg_mem(newval, mem_ptr),
8353 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8354 REX_reg_breg(res, res), // movzbl
8355 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8356 ins_pipe( pipe_cmpxchg );
8357 %}
8358
8359 //----------Subtraction Instructions-------------------------------------------
8360
8361 // Integer Subtraction Instructions
8362 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8363 %{
8364 match(Set dst (SubI dst src));
8365 effect(KILL cr);
8366
8367 format %{ "subl $dst, $src\t# int" %}
8368 opcode(0x2B);
8369 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8370 ins_pipe(ialu_reg_reg);
8371 %}
8372
8373 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8374 %{
8375 match(Set dst (SubI dst src));
8376 effect(KILL cr);
8377
8378 format %{ "subl $dst, $src\t# int" %}
8379 opcode(0x81, 0x05); /* Opcode 81 /5 */
8380 ins_encode(OpcSErm(dst, src), Con8or32(src));
8381 ins_pipe(ialu_reg);
8382 %}
8383
8384 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8385 %{
8386 match(Set dst (SubI dst (LoadI src)));
8387 effect(KILL cr);
8388
8389 ins_cost(125);
8390 format %{ "subl $dst, $src\t# int" %}
8391 opcode(0x2B);
8392 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8393 ins_pipe(ialu_reg_mem);
8394 %}
8395
8396 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8397 %{
8398 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8399 effect(KILL cr);
8400
8401 ins_cost(150);
8402 format %{ "subl $dst, $src\t# int" %}
8403 opcode(0x29); /* Opcode 29 /r */
8404 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8405 ins_pipe(ialu_mem_reg);
8406 %}
8407
8408 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8409 %{
8410 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8411 effect(KILL cr);
8412
8413 ins_cost(125); // XXX
8414 format %{ "subl $dst, $src\t# int" %}
8415 opcode(0x81); /* Opcode 81 /5 id */
8416 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8417 ins_pipe(ialu_mem_imm);
8418 %}
8419
8420 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8421 %{
8422 match(Set dst (SubL dst src));
8423 effect(KILL cr);
8424
8425 format %{ "subq $dst, $src\t# long" %}
8426 opcode(0x2B);
8427 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8428 ins_pipe(ialu_reg_reg);
8429 %}
8430
8431 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8432 %{
8433 match(Set dst (SubL dst src));
8434 effect(KILL cr);
8435
8436 format %{ "subq $dst, $src\t# long" %}
8437 opcode(0x81, 0x05); /* Opcode 81 /5 */
8438 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8439 ins_pipe(ialu_reg);
8440 %}
8441
8442 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8443 %{
8444 match(Set dst (SubL dst (LoadL src)));
8445 effect(KILL cr);
8446
8447 ins_cost(125);
8448 format %{ "subq $dst, $src\t# long" %}
8449 opcode(0x2B);
8450 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8451 ins_pipe(ialu_reg_mem);
8452 %}
8453
8454 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8455 %{
8456 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8457 effect(KILL cr);
8458
8459 ins_cost(150);
8460 format %{ "subq $dst, $src\t# long" %}
8461 opcode(0x29); /* Opcode 29 /r */
8462 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8463 ins_pipe(ialu_mem_reg);
8464 %}
8465
8466 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8467 %{
8468 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8469 effect(KILL cr);
8470
8471 ins_cost(125); // XXX
8472 format %{ "subq $dst, $src\t# long" %}
8473 opcode(0x81); /* Opcode 81 /5 id */
8474 ins_encode(REX_mem_wide(dst),
8475 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8476 ins_pipe(ialu_mem_imm);
8477 %}
8478
8479 // Subtract from a pointer
8480 // XXX hmpf???
8481 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8482 %{
8483 match(Set dst (AddP dst (SubI zero src)));
8484 effect(KILL cr);
8485
8486 format %{ "subq $dst, $src\t# ptr - int" %}
8487 opcode(0x2B);
8488 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8489 ins_pipe(ialu_reg_reg);
8490 %}
8491
8492 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8493 %{
8494 match(Set dst (SubI zero dst));
8495 effect(KILL cr);
8496
8497 format %{ "negl $dst\t# int" %}
8498 opcode(0xF7, 0x03); // Opcode F7 /3
8499 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8500 ins_pipe(ialu_reg);
8501 %}
8502
8503 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8504 %{
8505 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8506 effect(KILL cr);
8507
8508 format %{ "negl $dst\t# int" %}
8509 opcode(0xF7, 0x03); // Opcode F7 /3
8510 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8511 ins_pipe(ialu_reg);
8512 %}
8513
8514 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8515 %{
8516 match(Set dst (SubL zero dst));
8517 effect(KILL cr);
8518
8519 format %{ "negq $dst\t# long" %}
8520 opcode(0xF7, 0x03); // Opcode F7 /3
8521 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8522 ins_pipe(ialu_reg);
8523 %}
8524
8525 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8526 %{
8527 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8528 effect(KILL cr);
8529
8530 format %{ "negq $dst\t# long" %}
8531 opcode(0xF7, 0x03); // Opcode F7 /3
8532 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8533 ins_pipe(ialu_reg);
8534 %}
8535
8536
8537 //----------Multiplication/Division Instructions-------------------------------
8538 // Integer Multiplication Instructions
8539 // Multiply Register
8540
8541 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8542 %{
8543 match(Set dst (MulI dst src));
8544 effect(KILL cr);
8545
8546 ins_cost(300);
8547 format %{ "imull $dst, $src\t# int" %}
8548 opcode(0x0F, 0xAF);
8549 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8550 ins_pipe(ialu_reg_reg_alu0);
8551 %}
8552
8553 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8554 %{
8555 match(Set dst (MulI src imm));
8556 effect(KILL cr);
8557
8558 ins_cost(300);
8559 format %{ "imull $dst, $src, $imm\t# int" %}
8560 opcode(0x69); /* 69 /r id */
8561 ins_encode(REX_reg_reg(dst, src),
8562 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8563 ins_pipe(ialu_reg_reg_alu0);
8564 %}
8565
8566 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8567 %{
8568 match(Set dst (MulI dst (LoadI src)));
8569 effect(KILL cr);
8570
8571 ins_cost(350);
8572 format %{ "imull $dst, $src\t# int" %}
8573 opcode(0x0F, 0xAF);
8574 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8575 ins_pipe(ialu_reg_mem_alu0);
8576 %}
8577
8578 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8579 %{
8580 match(Set dst (MulI (LoadI src) imm));
8581 effect(KILL cr);
8582
8583 ins_cost(300);
8584 format %{ "imull $dst, $src, $imm\t# int" %}
8585 opcode(0x69); /* 69 /r id */
8586 ins_encode(REX_reg_mem(dst, src),
8587 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8588 ins_pipe(ialu_reg_mem_alu0);
8589 %}
8590
8591 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8592 %{
8593 match(Set dst (MulL dst src));
8594 effect(KILL cr);
8595
8596 ins_cost(300);
8597 format %{ "imulq $dst, $src\t# long" %}
8598 opcode(0x0F, 0xAF);
8599 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8600 ins_pipe(ialu_reg_reg_alu0);
8601 %}
8602
8603 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8604 %{
8605 match(Set dst (MulL src imm));
8606 effect(KILL cr);
8607
8608 ins_cost(300);
8609 format %{ "imulq $dst, $src, $imm\t# long" %}
8610 opcode(0x69); /* 69 /r id */
8611 ins_encode(REX_reg_reg_wide(dst, src),
8612 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8613 ins_pipe(ialu_reg_reg_alu0);
8614 %}
8615
8616 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8617 %{
8618 match(Set dst (MulL dst (LoadL src)));
8619 effect(KILL cr);
8620
8621 ins_cost(350);
8622 format %{ "imulq $dst, $src\t# long" %}
8623 opcode(0x0F, 0xAF);
8624 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8625 ins_pipe(ialu_reg_mem_alu0);
8626 %}
8627
8628 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8629 %{
8630 match(Set dst (MulL (LoadL src) imm));
8631 effect(KILL cr);
8632
8633 ins_cost(300);
8634 format %{ "imulq $dst, $src, $imm\t# long" %}
8635 opcode(0x69); /* 69 /r id */
8636 ins_encode(REX_reg_mem_wide(dst, src),
8637 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8638 ins_pipe(ialu_reg_mem_alu0);
8639 %}
8640
8641 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8642 %{
8643 match(Set dst (MulHiL src rax));
8644 effect(USE_KILL rax, KILL cr);
8645
8646 ins_cost(300);
8647 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %}
8648 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8649 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8650 ins_pipe(ialu_reg_reg_alu0);
8651 %}
8652
8653 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8654 rFlagsReg cr)
8655 %{
8656 match(Set rax (DivI rax div));
8657 effect(KILL rdx, KILL cr);
8658
8659 ins_cost(30*100+10*100); // XXX
8660 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8661 "jne,s normal\n\t"
8662 "xorl rdx, rdx\n\t"
8663 "cmpl $div, -1\n\t"
8664 "je,s done\n"
8665 "normal: cdql\n\t"
8666 "idivl $div\n"
8667 "done:" %}
8668 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8669 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8670 ins_pipe(ialu_reg_reg_alu0);
8671 %}
8672
8673 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8674 rFlagsReg cr)
8675 %{
8676 match(Set rax (DivL rax div));
8677 effect(KILL rdx, KILL cr);
8678
8679 ins_cost(30*100+10*100); // XXX
8680 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8681 "cmpq rax, rdx\n\t"
8682 "jne,s normal\n\t"
8683 "xorl rdx, rdx\n\t"
8684 "cmpq $div, -1\n\t"
8685 "je,s done\n"
8686 "normal: cdqq\n\t"
8687 "idivq $div\n"
8688 "done:" %}
8689 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8690 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8691 ins_pipe(ialu_reg_reg_alu0);
8692 %}
8693
8694 // Integer DIVMOD with Register, both quotient and mod results
8695 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8696 rFlagsReg cr)
8697 %{
8698 match(DivModI rax div);
8699 effect(KILL cr);
8700
8701 ins_cost(30*100+10*100); // XXX
8702 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8703 "jne,s normal\n\t"
8704 "xorl rdx, rdx\n\t"
8705 "cmpl $div, -1\n\t"
8706 "je,s done\n"
8707 "normal: cdql\n\t"
8708 "idivl $div\n"
8709 "done:" %}
8710 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8711 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8712 ins_pipe(pipe_slow);
8713 %}
8714
8715 // Long DIVMOD with Register, both quotient and mod results
8716 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8717 rFlagsReg cr)
8718 %{
8719 match(DivModL rax div);
8720 effect(KILL cr);
8721
8722 ins_cost(30*100+10*100); // XXX
8723 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8724 "cmpq rax, rdx\n\t"
8725 "jne,s normal\n\t"
8726 "xorl rdx, rdx\n\t"
8727 "cmpq $div, -1\n\t"
8728 "je,s done\n"
8729 "normal: cdqq\n\t"
8730 "idivq $div\n"
8731 "done:" %}
8732 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8733 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8734 ins_pipe(pipe_slow);
8735 %}
8736
8737 //----------- DivL-By-Constant-Expansions--------------------------------------
8738 // DivI cases are handled by the compiler
8739
8740 // Magic constant, reciprocal of 10
8741 instruct loadConL_0x6666666666666667(rRegL dst)
8742 %{
8743 effect(DEF dst);
8744
8745 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8746 ins_encode(load_immL(dst, 0x6666666666666667));
8747 ins_pipe(ialu_reg);
8748 %}
8749
8750 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8751 %{
8752 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8753
8754 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8755 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8756 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8757 ins_pipe(ialu_reg_reg_alu0);
8758 %}
8759
8760 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8761 %{
8762 effect(USE_DEF dst, KILL cr);
8763
8764 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8765 opcode(0xC1, 0x7); /* C1 /7 ib */
8766 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8767 ins_pipe(ialu_reg);
8768 %}
8769
8770 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8771 %{
8772 effect(USE_DEF dst, KILL cr);
8773
8774 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8775 opcode(0xC1, 0x7); /* C1 /7 ib */
8776 ins_encode(reg_opc_imm_wide(dst, 0x2));
8777 ins_pipe(ialu_reg);
8778 %}
8779
8780 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8781 %{
8782 match(Set dst (DivL src div));
8783
8784 ins_cost((5+8)*100);
8785 expand %{
8786 rax_RegL rax; // Killed temp
8787 rFlagsReg cr; // Killed
8788 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
8789 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
8790 sarL_rReg_63(src, cr); // sarq src, 63
8791 sarL_rReg_2(dst, cr); // sarq rdx, 2
8792 subL_rReg(dst, src, cr); // subl rdx, src
8793 %}
8794 %}
8795
8796 //-----------------------------------------------------------------------------
8797
8798 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8799 rFlagsReg cr)
8800 %{
8801 match(Set rdx (ModI rax div));
8802 effect(KILL rax, KILL cr);
8803
8804 ins_cost(300); // XXX
8805 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
8806 "jne,s normal\n\t"
8807 "xorl rdx, rdx\n\t"
8808 "cmpl $div, -1\n\t"
8809 "je,s done\n"
8810 "normal: cdql\n\t"
8811 "idivl $div\n"
8812 "done:" %}
8813 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8814 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8815 ins_pipe(ialu_reg_reg_alu0);
8816 %}
8817
8818 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8819 rFlagsReg cr)
8820 %{
8821 match(Set rdx (ModL rax div));
8822 effect(KILL rax, KILL cr);
8823
8824 ins_cost(300); // XXX
8825 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
8826 "cmpq rax, rdx\n\t"
8827 "jne,s normal\n\t"
8828 "xorl rdx, rdx\n\t"
8829 "cmpq $div, -1\n\t"
8830 "je,s done\n"
8831 "normal: cdqq\n\t"
8832 "idivq $div\n"
8833 "done:" %}
8834 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8835 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8836 ins_pipe(ialu_reg_reg_alu0);
8837 %}
8838
8839 // Integer Shift Instructions
8840 // Shift Left by one
8841 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8842 %{
8843 match(Set dst (LShiftI dst shift));
8844 effect(KILL cr);
8845
8846 format %{ "sall $dst, $shift" %}
8847 opcode(0xD1, 0x4); /* D1 /4 */
8848 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8849 ins_pipe(ialu_reg);
8850 %}
8851
8852 // Shift Left by one
8853 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8854 %{
8855 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8856 effect(KILL cr);
8857
8858 format %{ "sall $dst, $shift\t" %}
8859 opcode(0xD1, 0x4); /* D1 /4 */
8860 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8861 ins_pipe(ialu_mem_imm);
8862 %}
8863
8864 // Shift Left by 8-bit immediate
8865 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8866 %{
8867 match(Set dst (LShiftI dst shift));
8868 effect(KILL cr);
8869
8870 format %{ "sall $dst, $shift" %}
8871 opcode(0xC1, 0x4); /* C1 /4 ib */
8872 ins_encode(reg_opc_imm(dst, shift));
8873 ins_pipe(ialu_reg);
8874 %}
8875
8876 // Shift Left by 8-bit immediate
8877 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8878 %{
8879 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8880 effect(KILL cr);
8881
8882 format %{ "sall $dst, $shift" %}
8883 opcode(0xC1, 0x4); /* C1 /4 ib */
8884 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8885 ins_pipe(ialu_mem_imm);
8886 %}
8887
8888 // Shift Left by variable
8889 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8890 %{
8891 match(Set dst (LShiftI dst shift));
8892 effect(KILL cr);
8893
8894 format %{ "sall $dst, $shift" %}
8895 opcode(0xD3, 0x4); /* D3 /4 */
8896 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8897 ins_pipe(ialu_reg_reg);
8898 %}
8899
8900 // Shift Left by variable
8901 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8902 %{
8903 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8904 effect(KILL cr);
8905
8906 format %{ "sall $dst, $shift" %}
8907 opcode(0xD3, 0x4); /* D3 /4 */
8908 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8909 ins_pipe(ialu_mem_reg);
8910 %}
8911
8912 // Arithmetic shift right by one
8913 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8914 %{
8915 match(Set dst (RShiftI dst shift));
8916 effect(KILL cr);
8917
8918 format %{ "sarl $dst, $shift" %}
8919 opcode(0xD1, 0x7); /* D1 /7 */
8920 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8921 ins_pipe(ialu_reg);
8922 %}
8923
8924 // Arithmetic shift right by one
8925 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8926 %{
8927 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8928 effect(KILL cr);
8929
8930 format %{ "sarl $dst, $shift" %}
8931 opcode(0xD1, 0x7); /* D1 /7 */
8932 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8933 ins_pipe(ialu_mem_imm);
8934 %}
8935
8936 // Arithmetic Shift Right by 8-bit immediate
8937 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8938 %{
8939 match(Set dst (RShiftI dst shift));
8940 effect(KILL cr);
8941
8942 format %{ "sarl $dst, $shift" %}
8943 opcode(0xC1, 0x7); /* C1 /7 ib */
8944 ins_encode(reg_opc_imm(dst, shift));
8945 ins_pipe(ialu_mem_imm);
8946 %}
8947
8948 // Arithmetic Shift Right by 8-bit immediate
8949 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8950 %{
8951 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8952 effect(KILL cr);
8953
8954 format %{ "sarl $dst, $shift" %}
8955 opcode(0xC1, 0x7); /* C1 /7 ib */
8956 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8957 ins_pipe(ialu_mem_imm);
8958 %}
8959
8960 // Arithmetic Shift Right by variable
8961 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8962 %{
8963 match(Set dst (RShiftI dst shift));
8964 effect(KILL cr);
8965
8966 format %{ "sarl $dst, $shift" %}
8967 opcode(0xD3, 0x7); /* D3 /7 */
8968 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8969 ins_pipe(ialu_reg_reg);
8970 %}
8971
8972 // Arithmetic Shift Right by variable
8973 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8974 %{
8975 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8976 effect(KILL cr);
8977
8978 format %{ "sarl $dst, $shift" %}
8979 opcode(0xD3, 0x7); /* D3 /7 */
8980 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8981 ins_pipe(ialu_mem_reg);
8982 %}
8983
8984 // Logical shift right by one
8985 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8986 %{
8987 match(Set dst (URShiftI dst shift));
8988 effect(KILL cr);
8989
8990 format %{ "shrl $dst, $shift" %}
8991 opcode(0xD1, 0x5); /* D1 /5 */
8992 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8993 ins_pipe(ialu_reg);
8994 %}
8995
8996 // Logical shift right by one
8997 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8998 %{
8999 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9000 effect(KILL cr);
9001
9002 format %{ "shrl $dst, $shift" %}
9003 opcode(0xD1, 0x5); /* D1 /5 */
9004 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9005 ins_pipe(ialu_mem_imm);
9006 %}
9007
9008 // Logical Shift Right by 8-bit immediate
9009 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9010 %{
9011 match(Set dst (URShiftI dst shift));
9012 effect(KILL cr);
9013
9014 format %{ "shrl $dst, $shift" %}
9015 opcode(0xC1, 0x5); /* C1 /5 ib */
9016 ins_encode(reg_opc_imm(dst, shift));
9017 ins_pipe(ialu_reg);
9018 %}
9019
9020 // Logical Shift Right by 8-bit immediate
9021 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9022 %{
9023 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9024 effect(KILL cr);
9025
9026 format %{ "shrl $dst, $shift" %}
9027 opcode(0xC1, 0x5); /* C1 /5 ib */
9028 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9029 ins_pipe(ialu_mem_imm);
9030 %}
9031
9032 // Logical Shift Right by variable
9033 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9034 %{
9035 match(Set dst (URShiftI dst shift));
9036 effect(KILL cr);
9037
9038 format %{ "shrl $dst, $shift" %}
9039 opcode(0xD3, 0x5); /* D3 /5 */
9040 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9041 ins_pipe(ialu_reg_reg);
9042 %}
9043
9044 // Logical Shift Right by variable
9045 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9046 %{
9047 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9048 effect(KILL cr);
9049
9050 format %{ "shrl $dst, $shift" %}
9051 opcode(0xD3, 0x5); /* D3 /5 */
9052 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9053 ins_pipe(ialu_mem_reg);
9054 %}
9055
9056 // Long Shift Instructions
9057 // Shift Left by one
9058 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9059 %{
9060 match(Set dst (LShiftL dst shift));
9061 effect(KILL cr);
9062
9063 format %{ "salq $dst, $shift" %}
9064 opcode(0xD1, 0x4); /* D1 /4 */
9065 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9066 ins_pipe(ialu_reg);
9067 %}
9068
9069 // Shift Left by one
9070 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9071 %{
9072 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9073 effect(KILL cr);
9074
9075 format %{ "salq $dst, $shift" %}
9076 opcode(0xD1, 0x4); /* D1 /4 */
9077 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9078 ins_pipe(ialu_mem_imm);
9079 %}
9080
9081 // Shift Left by 8-bit immediate
9082 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9083 %{
9084 match(Set dst (LShiftL dst shift));
9085 effect(KILL cr);
9086
9087 format %{ "salq $dst, $shift" %}
9088 opcode(0xC1, 0x4); /* C1 /4 ib */
9089 ins_encode(reg_opc_imm_wide(dst, shift));
9090 ins_pipe(ialu_reg);
9091 %}
9092
9093 // Shift Left by 8-bit immediate
9094 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9095 %{
9096 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9097 effect(KILL cr);
9098
9099 format %{ "salq $dst, $shift" %}
9100 opcode(0xC1, 0x4); /* C1 /4 ib */
9101 ins_encode(REX_mem_wide(dst), OpcP,
9102 RM_opc_mem(secondary, dst), Con8or32(shift));
9103 ins_pipe(ialu_mem_imm);
9104 %}
9105
9106 // Shift Left by variable
9107 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9108 %{
9109 match(Set dst (LShiftL dst shift));
9110 effect(KILL cr);
9111
9112 format %{ "salq $dst, $shift" %}
9113 opcode(0xD3, 0x4); /* D3 /4 */
9114 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9115 ins_pipe(ialu_reg_reg);
9116 %}
9117
9118 // Shift Left by variable
9119 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9120 %{
9121 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9122 effect(KILL cr);
9123
9124 format %{ "salq $dst, $shift" %}
9125 opcode(0xD3, 0x4); /* D3 /4 */
9126 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9127 ins_pipe(ialu_mem_reg);
9128 %}
9129
9130 // Arithmetic shift right by one
9131 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9132 %{
9133 match(Set dst (RShiftL dst shift));
9134 effect(KILL cr);
9135
9136 format %{ "sarq $dst, $shift" %}
9137 opcode(0xD1, 0x7); /* D1 /7 */
9138 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9139 ins_pipe(ialu_reg);
9140 %}
9141
9142 // Arithmetic shift right by one
9143 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9144 %{
9145 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9146 effect(KILL cr);
9147
9148 format %{ "sarq $dst, $shift" %}
9149 opcode(0xD1, 0x7); /* D1 /7 */
9150 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9151 ins_pipe(ialu_mem_imm);
9152 %}
9153
9154 // Arithmetic Shift Right by 8-bit immediate
9155 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9156 %{
9157 match(Set dst (RShiftL dst shift));
9158 effect(KILL cr);
9159
9160 format %{ "sarq $dst, $shift" %}
9161 opcode(0xC1, 0x7); /* C1 /7 ib */
9162 ins_encode(reg_opc_imm_wide(dst, shift));
9163 ins_pipe(ialu_mem_imm);
9164 %}
9165
9166 // Arithmetic Shift Right by 8-bit immediate
9167 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9168 %{
9169 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9170 effect(KILL cr);
9171
9172 format %{ "sarq $dst, $shift" %}
9173 opcode(0xC1, 0x7); /* C1 /7 ib */
9174 ins_encode(REX_mem_wide(dst), OpcP,
9175 RM_opc_mem(secondary, dst), Con8or32(shift));
9176 ins_pipe(ialu_mem_imm);
9177 %}
9178
9179 // Arithmetic Shift Right by variable
9180 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9181 %{
9182 match(Set dst (RShiftL dst shift));
9183 effect(KILL cr);
9184
9185 format %{ "sarq $dst, $shift" %}
9186 opcode(0xD3, 0x7); /* D3 /7 */
9187 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9188 ins_pipe(ialu_reg_reg);
9189 %}
9190
9191 // Arithmetic Shift Right by variable
9192 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9193 %{
9194 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9195 effect(KILL cr);
9196
9197 format %{ "sarq $dst, $shift" %}
9198 opcode(0xD3, 0x7); /* D3 /7 */
9199 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9200 ins_pipe(ialu_mem_reg);
9201 %}
9202
9203 // Logical shift right by one
9204 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9205 %{
9206 match(Set dst (URShiftL dst shift));
9207 effect(KILL cr);
9208
9209 format %{ "shrq $dst, $shift" %}
9210 opcode(0xD1, 0x5); /* D1 /5 */
9211 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
9212 ins_pipe(ialu_reg);
9213 %}
9214
9215 // Logical shift right by one
9216 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9217 %{
9218 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9219 effect(KILL cr);
9220
9221 format %{ "shrq $dst, $shift" %}
9222 opcode(0xD1, 0x5); /* D1 /5 */
9223 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9224 ins_pipe(ialu_mem_imm);
9225 %}
9226
9227 // Logical Shift Right by 8-bit immediate
9228 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9229 %{
9230 match(Set dst (URShiftL dst shift));
9231 effect(KILL cr);
9232
9233 format %{ "shrq $dst, $shift" %}
9234 opcode(0xC1, 0x5); /* C1 /5 ib */
9235 ins_encode(reg_opc_imm_wide(dst, shift));
9236 ins_pipe(ialu_reg);
9237 %}
9238
9239
9240 // Logical Shift Right by 8-bit immediate
9241 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9242 %{
9243 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9244 effect(KILL cr);
9245
9246 format %{ "shrq $dst, $shift" %}
9247 opcode(0xC1, 0x5); /* C1 /5 ib */
9248 ins_encode(REX_mem_wide(dst), OpcP,
9249 RM_opc_mem(secondary, dst), Con8or32(shift));
9250 ins_pipe(ialu_mem_imm);
9251 %}
9252
9253 // Logical Shift Right by variable
9254 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9255 %{
9256 match(Set dst (URShiftL dst shift));
9257 effect(KILL cr);
9258
9259 format %{ "shrq $dst, $shift" %}
9260 opcode(0xD3, 0x5); /* D3 /5 */
9261 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9262 ins_pipe(ialu_reg_reg);
9263 %}
9264
9265 // Logical Shift Right by variable
9266 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9267 %{
9268 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9269 effect(KILL cr);
9270
9271 format %{ "shrq $dst, $shift" %}
9272 opcode(0xD3, 0x5); /* D3 /5 */
9273 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9274 ins_pipe(ialu_mem_reg);
9275 %}
9276
9277 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
9278 // This idiom is used by the compiler for the i2b bytecode.
9279 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
9280 %{
9281 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
9282
9283 format %{ "movsbl $dst, $src\t# i2b" %}
9284 opcode(0x0F, 0xBE);
9285 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9286 ins_pipe(ialu_reg_reg);
9287 %}
9288
9289 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
9290 // This idiom is used by the compiler the i2s bytecode.
9291 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
9292 %{
9293 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
9294
9295 format %{ "movswl $dst, $src\t# i2s" %}
9296 opcode(0x0F, 0xBF);
9297 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9298 ins_pipe(ialu_reg_reg);
9299 %}
9300
9301 // ROL/ROR instructions
9302
9303 // ROL expand
9304 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
9305 effect(KILL cr, USE_DEF dst);
9306
9307 format %{ "roll $dst" %}
9308 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9309 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9310 ins_pipe(ialu_reg);
9311 %}
9312
9313 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
9314 effect(USE_DEF dst, USE shift, KILL cr);
9315
9316 format %{ "roll $dst, $shift" %}
9317 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9318 ins_encode( reg_opc_imm(dst, shift) );
9319 ins_pipe(ialu_reg);
9320 %}
9321
9322 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9323 %{
9324 effect(USE_DEF dst, USE shift, KILL cr);
9325
9326 format %{ "roll $dst, $shift" %}
9327 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9328 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9329 ins_pipe(ialu_reg_reg);
9330 %}
9331 // end of ROL expand
9332
9333 // Rotate Left by one
9334 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9335 %{
9336 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9337
9338 expand %{
9339 rolI_rReg_imm1(dst, cr);
9340 %}
9341 %}
9342
9343 // Rotate Left by 8-bit immediate
9344 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9345 %{
9346 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9347 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9348
9349 expand %{
9350 rolI_rReg_imm8(dst, lshift, cr);
9351 %}
9352 %}
9353
9354 // Rotate Left by variable
9355 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9356 %{
9357 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
9358
9359 expand %{
9360 rolI_rReg_CL(dst, shift, cr);
9361 %}
9362 %}
9363
9364 // Rotate Left by variable
9365 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9366 %{
9367 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9368
9369 expand %{
9370 rolI_rReg_CL(dst, shift, cr);
9371 %}
9372 %}
9373
9374 // ROR expand
9375 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9376 %{
9377 effect(USE_DEF dst, KILL cr);
9378
9379 format %{ "rorl $dst" %}
9380 opcode(0xD1, 0x1); /* D1 /1 */
9381 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9382 ins_pipe(ialu_reg);
9383 %}
9384
9385 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9386 %{
9387 effect(USE_DEF dst, USE shift, KILL cr);
9388
9389 format %{ "rorl $dst, $shift" %}
9390 opcode(0xC1, 0x1); /* C1 /1 ib */
9391 ins_encode(reg_opc_imm(dst, shift));
9392 ins_pipe(ialu_reg);
9393 %}
9394
9395 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9396 %{
9397 effect(USE_DEF dst, USE shift, KILL cr);
9398
9399 format %{ "rorl $dst, $shift" %}
9400 opcode(0xD3, 0x1); /* D3 /1 */
9401 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9402 ins_pipe(ialu_reg_reg);
9403 %}
9404 // end of ROR expand
9405
9406 // Rotate Right by one
9407 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9408 %{
9409 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9410
9411 expand %{
9412 rorI_rReg_imm1(dst, cr);
9413 %}
9414 %}
9415
9416 // Rotate Right by 8-bit immediate
9417 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9418 %{
9419 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9420 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9421
9422 expand %{
9423 rorI_rReg_imm8(dst, rshift, cr);
9424 %}
9425 %}
9426
9427 // Rotate Right by variable
9428 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9429 %{
9430 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9431
9432 expand %{
9433 rorI_rReg_CL(dst, shift, cr);
9434 %}
9435 %}
9436
9437 // Rotate Right by variable
9438 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9439 %{
9440 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9441
9442 expand %{
9443 rorI_rReg_CL(dst, shift, cr);
9444 %}
9445 %}
9446
9447 // for long rotate
9448 // ROL expand
9449 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9450 effect(USE_DEF dst, KILL cr);
9451
9452 format %{ "rolq $dst" %}
9453 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9454 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9455 ins_pipe(ialu_reg);
9456 %}
9457
9458 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9459 effect(USE_DEF dst, USE shift, KILL cr);
9460
9461 format %{ "rolq $dst, $shift" %}
9462 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9463 ins_encode( reg_opc_imm_wide(dst, shift) );
9464 ins_pipe(ialu_reg);
9465 %}
9466
9467 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9468 %{
9469 effect(USE_DEF dst, USE shift, KILL cr);
9470
9471 format %{ "rolq $dst, $shift" %}
9472 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9473 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9474 ins_pipe(ialu_reg_reg);
9475 %}
9476 // end of ROL expand
9477
9478 // Rotate Left by one
9479 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9480 %{
9481 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9482
9483 expand %{
9484 rolL_rReg_imm1(dst, cr);
9485 %}
9486 %}
9487
9488 // Rotate Left by 8-bit immediate
9489 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9490 %{
9491 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9492 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9493
9494 expand %{
9495 rolL_rReg_imm8(dst, lshift, cr);
9496 %}
9497 %}
9498
9499 // Rotate Left by variable
9500 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9501 %{
9502 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9503
9504 expand %{
9505 rolL_rReg_CL(dst, shift, cr);
9506 %}
9507 %}
9508
9509 // Rotate Left by variable
9510 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9511 %{
9512 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9513
9514 expand %{
9515 rolL_rReg_CL(dst, shift, cr);
9516 %}
9517 %}
9518
9519 // ROR expand
9520 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9521 %{
9522 effect(USE_DEF dst, KILL cr);
9523
9524 format %{ "rorq $dst" %}
9525 opcode(0xD1, 0x1); /* D1 /1 */
9526 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9527 ins_pipe(ialu_reg);
9528 %}
9529
9530 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9531 %{
9532 effect(USE_DEF dst, USE shift, KILL cr);
9533
9534 format %{ "rorq $dst, $shift" %}
9535 opcode(0xC1, 0x1); /* C1 /1 ib */
9536 ins_encode(reg_opc_imm_wide(dst, shift));
9537 ins_pipe(ialu_reg);
9538 %}
9539
9540 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9541 %{
9542 effect(USE_DEF dst, USE shift, KILL cr);
9543
9544 format %{ "rorq $dst, $shift" %}
9545 opcode(0xD3, 0x1); /* D3 /1 */
9546 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9547 ins_pipe(ialu_reg_reg);
9548 %}
9549 // end of ROR expand
9550
9551 // Rotate Right by one
9552 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9553 %{
9554 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9555
9556 expand %{
9557 rorL_rReg_imm1(dst, cr);
9558 %}
9559 %}
9560
9561 // Rotate Right by 8-bit immediate
9562 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9563 %{
9564 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9565 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9566
9567 expand %{
9568 rorL_rReg_imm8(dst, rshift, cr);
9569 %}
9570 %}
9571
9572 // Rotate Right by variable
9573 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9574 %{
9575 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9576
9577 expand %{
9578 rorL_rReg_CL(dst, shift, cr);
9579 %}
9580 %}
9581
9582 // Rotate Right by variable
9583 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9584 %{
9585 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9586
9587 expand %{
9588 rorL_rReg_CL(dst, shift, cr);
9589 %}
9590 %}
9591
9592 // Logical Instructions
9593
9594 // Integer Logical Instructions
9595
9596 // And Instructions
9597 // And Register with Register
9598 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9599 %{
9600 match(Set dst (AndI dst src));
9601 effect(KILL cr);
9602
9603 format %{ "andl $dst, $src\t# int" %}
9604 opcode(0x23);
9605 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9606 ins_pipe(ialu_reg_reg);
9607 %}
9608
9609 // And Register with Immediate 255
9610 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9611 %{
9612 match(Set dst (AndI dst src));
9613
9614 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9615 opcode(0x0F, 0xB6);
9616 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9617 ins_pipe(ialu_reg);
9618 %}
9619
9620 // And Register with Immediate 255 and promote to long
9621 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9622 %{
9623 match(Set dst (ConvI2L (AndI src mask)));
9624
9625 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9626 opcode(0x0F, 0xB6);
9627 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9628 ins_pipe(ialu_reg);
9629 %}
9630
9631 // And Register with Immediate 65535
9632 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9633 %{
9634 match(Set dst (AndI dst src));
9635
9636 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9637 opcode(0x0F, 0xB7);
9638 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9639 ins_pipe(ialu_reg);
9640 %}
9641
9642 // And Register with Immediate 65535 and promote to long
9643 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9644 %{
9645 match(Set dst (ConvI2L (AndI src mask)));
9646
9647 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9648 opcode(0x0F, 0xB7);
9649 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9650 ins_pipe(ialu_reg);
9651 %}
9652
9653 // And Register with Immediate
9654 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9655 %{
9656 match(Set dst (AndI dst src));
9657 effect(KILL cr);
9658
9659 format %{ "andl $dst, $src\t# int" %}
9660 opcode(0x81, 0x04); /* Opcode 81 /4 */
9661 ins_encode(OpcSErm(dst, src), Con8or32(src));
9662 ins_pipe(ialu_reg);
9663 %}
9664
9665 // And Register with Memory
9666 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9667 %{
9668 match(Set dst (AndI dst (LoadI src)));
9669 effect(KILL cr);
9670
9671 ins_cost(125);
9672 format %{ "andl $dst, $src\t# int" %}
9673 opcode(0x23);
9674 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9675 ins_pipe(ialu_reg_mem);
9676 %}
9677
9678 // And Memory with Register
9679 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9680 %{
9681 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9682 effect(KILL cr);
9683
9684 ins_cost(150);
9685 format %{ "andl $dst, $src\t# int" %}
9686 opcode(0x21); /* Opcode 21 /r */
9687 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9688 ins_pipe(ialu_mem_reg);
9689 %}
9690
9691 // And Memory with Immediate
9692 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9693 %{
9694 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9695 effect(KILL cr);
9696
9697 ins_cost(125);
9698 format %{ "andl $dst, $src\t# int" %}
9699 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9700 ins_encode(REX_mem(dst), OpcSE(src),
9701 RM_opc_mem(secondary, dst), Con8or32(src));
9702 ins_pipe(ialu_mem_imm);
9703 %}
9704
9705 // Or Instructions
9706 // Or Register with Register
9707 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9708 %{
9709 match(Set dst (OrI dst src));
9710 effect(KILL cr);
9711
9712 format %{ "orl $dst, $src\t# int" %}
9713 opcode(0x0B);
9714 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9715 ins_pipe(ialu_reg_reg);
9716 %}
9717
9718 // Or Register with Immediate
9719 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9720 %{
9721 match(Set dst (OrI dst src));
9722 effect(KILL cr);
9723
9724 format %{ "orl $dst, $src\t# int" %}
9725 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9726 ins_encode(OpcSErm(dst, src), Con8or32(src));
9727 ins_pipe(ialu_reg);
9728 %}
9729
9730 // Or Register with Memory
9731 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9732 %{
9733 match(Set dst (OrI dst (LoadI src)));
9734 effect(KILL cr);
9735
9736 ins_cost(125);
9737 format %{ "orl $dst, $src\t# int" %}
9738 opcode(0x0B);
9739 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9740 ins_pipe(ialu_reg_mem);
9741 %}
9742
9743 // Or Memory with Register
9744 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9745 %{
9746 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9747 effect(KILL cr);
9748
9749 ins_cost(150);
9750 format %{ "orl $dst, $src\t# int" %}
9751 opcode(0x09); /* Opcode 09 /r */
9752 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9753 ins_pipe(ialu_mem_reg);
9754 %}
9755
9756 // Or Memory with Immediate
9757 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9758 %{
9759 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9760 effect(KILL cr);
9761
9762 ins_cost(125);
9763 format %{ "orl $dst, $src\t# int" %}
9764 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9765 ins_encode(REX_mem(dst), OpcSE(src),
9766 RM_opc_mem(secondary, dst), Con8or32(src));
9767 ins_pipe(ialu_mem_imm);
9768 %}
9769
9770 // Xor Instructions
9771 // Xor Register with Register
9772 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9773 %{
9774 match(Set dst (XorI dst src));
9775 effect(KILL cr);
9776
9777 format %{ "xorl $dst, $src\t# int" %}
9778 opcode(0x33);
9779 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9780 ins_pipe(ialu_reg_reg);
9781 %}
9782
9783 // Xor Register with Immediate -1
9784 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9785 match(Set dst (XorI dst imm));
9786
9787 format %{ "not $dst" %}
9788 ins_encode %{
9789 __ notl($dst$$Register);
9790 %}
9791 ins_pipe(ialu_reg);
9792 %}
9793
9794 // Xor Register with Immediate
9795 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9796 %{
9797 match(Set dst (XorI dst src));
9798 effect(KILL cr);
9799
9800 format %{ "xorl $dst, $src\t# int" %}
9801 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9802 ins_encode(OpcSErm(dst, src), Con8or32(src));
9803 ins_pipe(ialu_reg);
9804 %}
9805
9806 // Xor Register with Memory
9807 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9808 %{
9809 match(Set dst (XorI dst (LoadI src)));
9810 effect(KILL cr);
9811
9812 ins_cost(125);
9813 format %{ "xorl $dst, $src\t# int" %}
9814 opcode(0x33);
9815 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9816 ins_pipe(ialu_reg_mem);
9817 %}
9818
9819 // Xor Memory with Register
9820 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9821 %{
9822 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9823 effect(KILL cr);
9824
9825 ins_cost(150);
9826 format %{ "xorl $dst, $src\t# int" %}
9827 opcode(0x31); /* Opcode 31 /r */
9828 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9829 ins_pipe(ialu_mem_reg);
9830 %}
9831
9832 // Xor Memory with Immediate
9833 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9834 %{
9835 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9836 effect(KILL cr);
9837
9838 ins_cost(125);
9839 format %{ "xorl $dst, $src\t# int" %}
9840 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9841 ins_encode(REX_mem(dst), OpcSE(src),
9842 RM_opc_mem(secondary, dst), Con8or32(src));
9843 ins_pipe(ialu_mem_imm);
9844 %}
9845
9846
9847 // Long Logical Instructions
9848
9849 // And Instructions
9850 // And Register with Register
9851 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9852 %{
9853 match(Set dst (AndL dst src));
9854 effect(KILL cr);
9855
9856 format %{ "andq $dst, $src\t# long" %}
9857 opcode(0x23);
9858 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9859 ins_pipe(ialu_reg_reg);
9860 %}
9861
9862 // And Register with Immediate 255
9863 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9864 %{
9865 match(Set dst (AndL dst src));
9866
9867 format %{ "movzbq $dst, $dst\t# long & 0xFF" %}
9868 opcode(0x0F, 0xB6);
9869 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9870 ins_pipe(ialu_reg);
9871 %}
9872
9873 // And Register with Immediate 65535
9874 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src)
9875 %{
9876 match(Set dst (AndL dst src));
9877
9878 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
9879 opcode(0x0F, 0xB7);
9880 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9881 ins_pipe(ialu_reg);
9882 %}
9883
9884 // And Register with Immediate
9885 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9886 %{
9887 match(Set dst (AndL dst src));
9888 effect(KILL cr);
9889
9890 format %{ "andq $dst, $src\t# long" %}
9891 opcode(0x81, 0x04); /* Opcode 81 /4 */
9892 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9893 ins_pipe(ialu_reg);
9894 %}
9895
9896 // And Register with Memory
9897 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9898 %{
9899 match(Set dst (AndL dst (LoadL src)));
9900 effect(KILL cr);
9901
9902 ins_cost(125);
9903 format %{ "andq $dst, $src\t# long" %}
9904 opcode(0x23);
9905 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9906 ins_pipe(ialu_reg_mem);
9907 %}
9908
9909 // And Memory with Register
9910 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9911 %{
9912 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9913 effect(KILL cr);
9914
9915 ins_cost(150);
9916 format %{ "andq $dst, $src\t# long" %}
9917 opcode(0x21); /* Opcode 21 /r */
9918 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9919 ins_pipe(ialu_mem_reg);
9920 %}
9921
9922 // And Memory with Immediate
9923 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9924 %{
9925 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9926 effect(KILL cr);
9927
9928 ins_cost(125);
9929 format %{ "andq $dst, $src\t# long" %}
9930 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9931 ins_encode(REX_mem_wide(dst), OpcSE(src),
9932 RM_opc_mem(secondary, dst), Con8or32(src));
9933 ins_pipe(ialu_mem_imm);
9934 %}
9935
9936 // Or Instructions
9937 // Or Register with Register
9938 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9939 %{
9940 match(Set dst (OrL dst src));
9941 effect(KILL cr);
9942
9943 format %{ "orq $dst, $src\t# long" %}
9944 opcode(0x0B);
9945 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9946 ins_pipe(ialu_reg_reg);
9947 %}
9948
9949 // Use any_RegP to match R15 (TLS register) without spilling.
9950 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
9951 match(Set dst (OrL dst (CastP2X src)));
9952 effect(KILL cr);
9953
9954 format %{ "orq $dst, $src\t# long" %}
9955 opcode(0x0B);
9956 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9957 ins_pipe(ialu_reg_reg);
9958 %}
9959
9960
9961 // Or Register with Immediate
9962 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9963 %{
9964 match(Set dst (OrL dst src));
9965 effect(KILL cr);
9966
9967 format %{ "orq $dst, $src\t# long" %}
9968 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9969 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9970 ins_pipe(ialu_reg);
9971 %}
9972
9973 // Or Register with Memory
9974 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9975 %{
9976 match(Set dst (OrL dst (LoadL src)));
9977 effect(KILL cr);
9978
9979 ins_cost(125);
9980 format %{ "orq $dst, $src\t# long" %}
9981 opcode(0x0B);
9982 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9983 ins_pipe(ialu_reg_mem);
9984 %}
9985
9986 // Or Memory with Register
9987 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9988 %{
9989 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9990 effect(KILL cr);
9991
9992 ins_cost(150);
9993 format %{ "orq $dst, $src\t# long" %}
9994 opcode(0x09); /* Opcode 09 /r */
9995 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9996 ins_pipe(ialu_mem_reg);
9997 %}
9998
9999 // Or Memory with Immediate
10000 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10001 %{
10002 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10003 effect(KILL cr);
10004
10005 ins_cost(125);
10006 format %{ "orq $dst, $src\t# long" %}
10007 opcode(0x81, 0x1); /* Opcode 81 /1 id */
10008 ins_encode(REX_mem_wide(dst), OpcSE(src),
10009 RM_opc_mem(secondary, dst), Con8or32(src));
10010 ins_pipe(ialu_mem_imm);
10011 %}
10012
10013 // Xor Instructions
10014 // Xor Register with Register
10015 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10016 %{
10017 match(Set dst (XorL dst src));
10018 effect(KILL cr);
10019
10020 format %{ "xorq $dst, $src\t# long" %}
10021 opcode(0x33);
10022 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10023 ins_pipe(ialu_reg_reg);
10024 %}
10025
10026 // Xor Register with Immediate -1
10027 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
10028 match(Set dst (XorL dst imm));
10029
10030 format %{ "notq $dst" %}
10031 ins_encode %{
10032 __ notq($dst$$Register);
10033 %}
10034 ins_pipe(ialu_reg);
10035 %}
10036
10037 // Xor Register with Immediate
10038 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10039 %{
10040 match(Set dst (XorL dst src));
10041 effect(KILL cr);
10042
10043 format %{ "xorq $dst, $src\t# long" %}
10044 opcode(0x81, 0x06); /* Opcode 81 /6 id */
10045 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10046 ins_pipe(ialu_reg);
10047 %}
10048
10049 // Xor Register with Memory
10050 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10051 %{
10052 match(Set dst (XorL dst (LoadL src)));
10053 effect(KILL cr);
10054
10055 ins_cost(125);
10056 format %{ "xorq $dst, $src\t# long" %}
10057 opcode(0x33);
10058 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10059 ins_pipe(ialu_reg_mem);
10060 %}
10061
10062 // Xor Memory with Register
10063 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10064 %{
10065 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10066 effect(KILL cr);
10067
10068 ins_cost(150);
10069 format %{ "xorq $dst, $src\t# long" %}
10070 opcode(0x31); /* Opcode 31 /r */
10071 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10072 ins_pipe(ialu_mem_reg);
10073 %}
10074
10075 // Xor Memory with Immediate
10076 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10077 %{
10078 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10079 effect(KILL cr);
10080
10081 ins_cost(125);
10082 format %{ "xorq $dst, $src\t# long" %}
10083 opcode(0x81, 0x6); /* Opcode 81 /6 id */
10084 ins_encode(REX_mem_wide(dst), OpcSE(src),
10085 RM_opc_mem(secondary, dst), Con8or32(src));
10086 ins_pipe(ialu_mem_imm);
10087 %}
10088
10089 // Convert Int to Boolean
10090 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
10091 %{
10092 match(Set dst (Conv2B src));
10093 effect(KILL cr);
10094
10095 format %{ "testl $src, $src\t# ci2b\n\t"
10096 "setnz $dst\n\t"
10097 "movzbl $dst, $dst" %}
10098 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
10099 setNZ_reg(dst),
10100 REX_reg_breg(dst, dst), // movzbl
10101 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10102 ins_pipe(pipe_slow); // XXX
10103 %}
10104
10105 // Convert Pointer to Boolean
10106 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
10107 %{
10108 match(Set dst (Conv2B src));
10109 effect(KILL cr);
10110
10111 format %{ "testq $src, $src\t# cp2b\n\t"
10112 "setnz $dst\n\t"
10113 "movzbl $dst, $dst" %}
10114 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
10115 setNZ_reg(dst),
10116 REX_reg_breg(dst, dst), // movzbl
10117 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10118 ins_pipe(pipe_slow); // XXX
10119 %}
10120
10121 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
10122 %{
10123 match(Set dst (CmpLTMask p q));
10124 effect(KILL cr);
10125
10126 ins_cost(400); // XXX
10127 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
10128 "setlt $dst\n\t"
10129 "movzbl $dst, $dst\n\t"
10130 "negl $dst" %}
10131 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
10132 setLT_reg(dst),
10133 REX_reg_breg(dst, dst), // movzbl
10134 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
10135 neg_reg(dst));
10136 ins_pipe(pipe_slow);
10137 %}
10138
10139 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
10140 %{
10141 match(Set dst (CmpLTMask dst zero));
10142 effect(KILL cr);
10143
10144 ins_cost(100); // XXX
10145 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
10146 opcode(0xC1, 0x7); /* C1 /7 ib */
10147 ins_encode(reg_opc_imm(dst, 0x1F));
10148 ins_pipe(ialu_reg);
10149 %}
10150
10151
10152 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y, rRegI tmp, rFlagsReg cr)
10153 %{
10154 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
10155 effect(TEMP tmp, KILL cr);
10156
10157 ins_cost(400); // XXX
10158 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
10159 "sbbl $tmp, $tmp\n\t"
10160 "andl $tmp, $y\n\t"
10161 "addl $p, $tmp" %}
10162 ins_encode %{
10163 Register Rp = $p$$Register;
10164 Register Rq = $q$$Register;
10165 Register Ry = $y$$Register;
10166 Register Rt = $tmp$$Register;
10167 __ subl(Rp, Rq);
10168 __ sbbl(Rt, Rt);
10169 __ andl(Rt, Ry);
10170 __ addl(Rp, Rt);
10171 %}
10172 ins_pipe(pipe_cmplt);
10173 %}
10174
10175 //---------- FP Instructions------------------------------------------------
10176
10177 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
10178 %{
10179 match(Set cr (CmpF src1 src2));
10180
10181 ins_cost(145);
10182 format %{ "ucomiss $src1, $src2\n\t"
10183 "jnp,s exit\n\t"
10184 "pushfq\t# saw NaN, set CF\n\t"
10185 "andq [rsp], #0xffffff2b\n\t"
10186 "popfq\n"
10187 "exit: nop\t# avoid branch to branch" %}
10188 opcode(0x0F, 0x2E);
10189 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10190 cmpfp_fixup);
10191 ins_pipe(pipe_slow);
10192 %}
10193
10194 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
10195 match(Set cr (CmpF src1 src2));
10196
10197 ins_cost(145);
10198 format %{ "ucomiss $src1, $src2" %}
10199 ins_encode %{
10200 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10201 %}
10202 ins_pipe(pipe_slow);
10203 %}
10204
10205 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
10206 %{
10207 match(Set cr (CmpF src1 (LoadF src2)));
10208
10209 ins_cost(145);
10210 format %{ "ucomiss $src1, $src2\n\t"
10211 "jnp,s exit\n\t"
10212 "pushfq\t# saw NaN, set CF\n\t"
10213 "andq [rsp], #0xffffff2b\n\t"
10214 "popfq\n"
10215 "exit: nop\t# avoid branch to branch" %}
10216 opcode(0x0F, 0x2E);
10217 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10218 cmpfp_fixup);
10219 ins_pipe(pipe_slow);
10220 %}
10221
10222 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
10223 match(Set cr (CmpF src1 (LoadF src2)));
10224
10225 ins_cost(100);
10226 format %{ "ucomiss $src1, $src2" %}
10227 opcode(0x0F, 0x2E);
10228 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
10229 ins_pipe(pipe_slow);
10230 %}
10231
10232 instruct cmpF_cc_imm(rFlagsRegU cr, regF src, immF con) %{
10233 match(Set cr (CmpF src con));
10234
10235 ins_cost(145);
10236 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con\n\t"
10237 "jnp,s exit\n\t"
10238 "pushfq\t# saw NaN, set CF\n\t"
10239 "andq [rsp], #0xffffff2b\n\t"
10240 "popfq\n"
10241 "exit: nop\t# avoid branch to branch" %}
10242 ins_encode %{
10243 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10244 emit_cmpfp_fixup(_masm);
10245 %}
10246 ins_pipe(pipe_slow);
10247 %}
10248
10249 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src, immF con) %{
10250 match(Set cr (CmpF src con));
10251 ins_cost(100);
10252 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con" %}
10253 ins_encode %{
10254 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10255 %}
10256 ins_pipe(pipe_slow);
10257 %}
10258
10259 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
10260 %{
10261 match(Set cr (CmpD src1 src2));
10262
10263 ins_cost(145);
10264 format %{ "ucomisd $src1, $src2\n\t"
10265 "jnp,s exit\n\t"
10266 "pushfq\t# saw NaN, set CF\n\t"
10267 "andq [rsp], #0xffffff2b\n\t"
10268 "popfq\n"
10269 "exit: nop\t# avoid branch to branch" %}
10270 opcode(0x66, 0x0F, 0x2E);
10271 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10272 cmpfp_fixup);
10273 ins_pipe(pipe_slow);
10274 %}
10275
10276 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
10277 match(Set cr (CmpD src1 src2));
10278
10279 ins_cost(100);
10280 format %{ "ucomisd $src1, $src2 test" %}
10281 ins_encode %{
10282 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
10283 %}
10284 ins_pipe(pipe_slow);
10285 %}
10286
10287 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
10288 %{
10289 match(Set cr (CmpD src1 (LoadD src2)));
10290
10291 ins_cost(145);
10292 format %{ "ucomisd $src1, $src2\n\t"
10293 "jnp,s exit\n\t"
10294 "pushfq\t# saw NaN, set CF\n\t"
10295 "andq [rsp], #0xffffff2b\n\t"
10296 "popfq\n"
10297 "exit: nop\t# avoid branch to branch" %}
10298 opcode(0x66, 0x0F, 0x2E);
10299 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10300 cmpfp_fixup);
10301 ins_pipe(pipe_slow);
10302 %}
10303
10304 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
10305 match(Set cr (CmpD src1 (LoadD src2)));
10306
10307 ins_cost(100);
10308 format %{ "ucomisd $src1, $src2" %}
10309 opcode(0x66, 0x0F, 0x2E);
10310 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
10311 ins_pipe(pipe_slow);
10312 %}
10313
10314 instruct cmpD_cc_imm(rFlagsRegU cr, regD src, immD con) %{
10315 match(Set cr (CmpD src con));
10316
10317 ins_cost(145);
10318 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con\n\t"
10319 "jnp,s exit\n\t"
10320 "pushfq\t# saw NaN, set CF\n\t"
10321 "andq [rsp], #0xffffff2b\n\t"
10322 "popfq\n"
10323 "exit: nop\t# avoid branch to branch" %}
10324 ins_encode %{
10325 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10326 emit_cmpfp_fixup(_masm);
10327 %}
10328 ins_pipe(pipe_slow);
10329 %}
10330
10331 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src, immD con) %{
10332 match(Set cr (CmpD src con));
10333 ins_cost(100);
10334 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con" %}
10335 ins_encode %{
10336 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10337 %}
10338 ins_pipe(pipe_slow);
10339 %}
10340
10341 // Compare into -1,0,1
10342 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
10343 %{
10344 match(Set dst (CmpF3 src1 src2));
10345 effect(KILL cr);
10346
10347 ins_cost(275);
10348 format %{ "ucomiss $src1, $src2\n\t"
10349 "movl $dst, #-1\n\t"
10350 "jp,s done\n\t"
10351 "jb,s done\n\t"
10352 "setne $dst\n\t"
10353 "movzbl $dst, $dst\n"
10354 "done:" %}
10355
10356 opcode(0x0F, 0x2E);
10357 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10358 cmpfp3(dst));
10359 ins_pipe(pipe_slow);
10360 %}
10361
10362 // Compare into -1,0,1
10363 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
10364 %{
10365 match(Set dst (CmpF3 src1 (LoadF src2)));
10366 effect(KILL cr);
10367
10368 ins_cost(275);
10369 format %{ "ucomiss $src1, $src2\n\t"
10370 "movl $dst, #-1\n\t"
10371 "jp,s done\n\t"
10372 "jb,s done\n\t"
10373 "setne $dst\n\t"
10374 "movzbl $dst, $dst\n"
10375 "done:" %}
10376
10377 opcode(0x0F, 0x2E);
10378 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10379 cmpfp3(dst));
10380 ins_pipe(pipe_slow);
10381 %}
10382
10383 // Compare into -1,0,1
10384 instruct cmpF_imm(rRegI dst, regF src, immF con, rFlagsReg cr) %{
10385 match(Set dst (CmpF3 src con));
10386 effect(KILL cr);
10387
10388 ins_cost(275);
10389 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con\n\t"
10390 "movl $dst, #-1\n\t"
10391 "jp,s done\n\t"
10392 "jb,s done\n\t"
10393 "setne $dst\n\t"
10394 "movzbl $dst, $dst\n"
10395 "done:" %}
10396 ins_encode %{
10397 Label L_done;
10398 Register Rdst = $dst$$Register;
10399 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10400 __ movl(Rdst, -1);
10401 __ jcc(Assembler::parity, L_done);
10402 __ jcc(Assembler::below, L_done);
10403 __ setb(Assembler::notEqual, Rdst);
10404 __ movzbl(Rdst, Rdst);
10405 __ bind(L_done);
10406 %}
10407 ins_pipe(pipe_slow);
10408 %}
10409
10410 // Compare into -1,0,1
10411 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10412 %{
10413 match(Set dst (CmpD3 src1 src2));
10414 effect(KILL cr);
10415
10416 ins_cost(275);
10417 format %{ "ucomisd $src1, $src2\n\t"
10418 "movl $dst, #-1\n\t"
10419 "jp,s done\n\t"
10420 "jb,s done\n\t"
10421 "setne $dst\n\t"
10422 "movzbl $dst, $dst\n"
10423 "done:" %}
10424
10425 opcode(0x66, 0x0F, 0x2E);
10426 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10427 cmpfp3(dst));
10428 ins_pipe(pipe_slow);
10429 %}
10430
10431 // Compare into -1,0,1
10432 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10433 %{
10434 match(Set dst (CmpD3 src1 (LoadD src2)));
10435 effect(KILL cr);
10436
10437 ins_cost(275);
10438 format %{ "ucomisd $src1, $src2\n\t"
10439 "movl $dst, #-1\n\t"
10440 "jp,s done\n\t"
10441 "jb,s done\n\t"
10442 "setne $dst\n\t"
10443 "movzbl $dst, $dst\n"
10444 "done:" %}
10445
10446 opcode(0x66, 0x0F, 0x2E);
10447 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10448 cmpfp3(dst));
10449 ins_pipe(pipe_slow);
10450 %}
10451
10452 // Compare into -1,0,1
10453 instruct cmpD_imm(rRegI dst, regD src, immD con, rFlagsReg cr) %{
10454 match(Set dst (CmpD3 src con));
10455 effect(KILL cr);
10456
10457 ins_cost(275);
10458 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con\n\t"
10459 "movl $dst, #-1\n\t"
10460 "jp,s done\n\t"
10461 "jb,s done\n\t"
10462 "setne $dst\n\t"
10463 "movzbl $dst, $dst\n"
10464 "done:" %}
10465 ins_encode %{
10466 Register Rdst = $dst$$Register;
10467 Label L_done;
10468 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10469 __ movl(Rdst, -1);
10470 __ jcc(Assembler::parity, L_done);
10471 __ jcc(Assembler::below, L_done);
10472 __ setb(Assembler::notEqual, Rdst);
10473 __ movzbl(Rdst, Rdst);
10474 __ bind(L_done);
10475 %}
10476 ins_pipe(pipe_slow);
10477 %}
10478
10479 instruct addF_reg(regF dst, regF src)
10480 %{
10481 match(Set dst (AddF dst src));
10482
10483 format %{ "addss $dst, $src" %}
10484 ins_cost(150); // XXX
10485 opcode(0xF3, 0x0F, 0x58);
10486 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10487 ins_pipe(pipe_slow);
10488 %}
10489
10490 instruct addF_mem(regF dst, memory src)
10491 %{
10492 match(Set dst (AddF dst (LoadF src)));
10493
10494 format %{ "addss $dst, $src" %}
10495 ins_cost(150); // XXX
10496 opcode(0xF3, 0x0F, 0x58);
10497 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10498 ins_pipe(pipe_slow);
10499 %}
10500
10501 instruct addF_imm(regF dst, immF con) %{
10502 match(Set dst (AddF dst con));
10503 format %{ "addss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10504 ins_cost(150); // XXX
10505 ins_encode %{
10506 __ addss($dst$$XMMRegister, $constantaddress($con));
10507 %}
10508 ins_pipe(pipe_slow);
10509 %}
10510
10511 instruct addD_reg(regD dst, regD src)
10512 %{
10513 match(Set dst (AddD dst src));
10514
10515 format %{ "addsd $dst, $src" %}
10516 ins_cost(150); // XXX
10517 opcode(0xF2, 0x0F, 0x58);
10518 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10519 ins_pipe(pipe_slow);
10520 %}
10521
10522 instruct addD_mem(regD dst, memory src)
10523 %{
10524 match(Set dst (AddD dst (LoadD src)));
10525
10526 format %{ "addsd $dst, $src" %}
10527 ins_cost(150); // XXX
10528 opcode(0xF2, 0x0F, 0x58);
10529 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10530 ins_pipe(pipe_slow);
10531 %}
10532
10533 instruct addD_imm(regD dst, immD con) %{
10534 match(Set dst (AddD dst con));
10535 format %{ "addsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10536 ins_cost(150); // XXX
10537 ins_encode %{
10538 __ addsd($dst$$XMMRegister, $constantaddress($con));
10539 %}
10540 ins_pipe(pipe_slow);
10541 %}
10542
10543 instruct subF_reg(regF dst, regF src)
10544 %{
10545 match(Set dst (SubF dst src));
10546
10547 format %{ "subss $dst, $src" %}
10548 ins_cost(150); // XXX
10549 opcode(0xF3, 0x0F, 0x5C);
10550 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10551 ins_pipe(pipe_slow);
10552 %}
10553
10554 instruct subF_mem(regF dst, memory src)
10555 %{
10556 match(Set dst (SubF dst (LoadF src)));
10557
10558 format %{ "subss $dst, $src" %}
10559 ins_cost(150); // XXX
10560 opcode(0xF3, 0x0F, 0x5C);
10561 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10562 ins_pipe(pipe_slow);
10563 %}
10564
10565 instruct subF_imm(regF dst, immF con) %{
10566 match(Set dst (SubF dst con));
10567 format %{ "subss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10568 ins_cost(150); // XXX
10569 ins_encode %{
10570 __ subss($dst$$XMMRegister, $constantaddress($con));
10571 %}
10572 ins_pipe(pipe_slow);
10573 %}
10574
10575 instruct subD_reg(regD dst, regD src)
10576 %{
10577 match(Set dst (SubD dst src));
10578
10579 format %{ "subsd $dst, $src" %}
10580 ins_cost(150); // XXX
10581 opcode(0xF2, 0x0F, 0x5C);
10582 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10583 ins_pipe(pipe_slow);
10584 %}
10585
10586 instruct subD_mem(regD dst, memory src)
10587 %{
10588 match(Set dst (SubD dst (LoadD src)));
10589
10590 format %{ "subsd $dst, $src" %}
10591 ins_cost(150); // XXX
10592 opcode(0xF2, 0x0F, 0x5C);
10593 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10594 ins_pipe(pipe_slow);
10595 %}
10596
10597 instruct subD_imm(regD dst, immD con) %{
10598 match(Set dst (SubD dst con));
10599 format %{ "subsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10600 ins_cost(150); // XXX
10601 ins_encode %{
10602 __ subsd($dst$$XMMRegister, $constantaddress($con));
10603 %}
10604 ins_pipe(pipe_slow);
10605 %}
10606
10607 instruct mulF_reg(regF dst, regF src)
10608 %{
10609 match(Set dst (MulF dst src));
10610
10611 format %{ "mulss $dst, $src" %}
10612 ins_cost(150); // XXX
10613 opcode(0xF3, 0x0F, 0x59);
10614 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10615 ins_pipe(pipe_slow);
10616 %}
10617
10618 instruct mulF_mem(regF dst, memory src)
10619 %{
10620 match(Set dst (MulF dst (LoadF src)));
10621
10622 format %{ "mulss $dst, $src" %}
10623 ins_cost(150); // XXX
10624 opcode(0xF3, 0x0F, 0x59);
10625 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10626 ins_pipe(pipe_slow);
10627 %}
10628
10629 instruct mulF_imm(regF dst, immF con) %{
10630 match(Set dst (MulF dst con));
10631 format %{ "mulss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10632 ins_cost(150); // XXX
10633 ins_encode %{
10634 __ mulss($dst$$XMMRegister, $constantaddress($con));
10635 %}
10636 ins_pipe(pipe_slow);
10637 %}
10638
10639 instruct mulD_reg(regD dst, regD src)
10640 %{
10641 match(Set dst (MulD dst src));
10642
10643 format %{ "mulsd $dst, $src" %}
10644 ins_cost(150); // XXX
10645 opcode(0xF2, 0x0F, 0x59);
10646 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10647 ins_pipe(pipe_slow);
10648 %}
10649
10650 instruct mulD_mem(regD dst, memory src)
10651 %{
10652 match(Set dst (MulD dst (LoadD src)));
10653
10654 format %{ "mulsd $dst, $src" %}
10655 ins_cost(150); // XXX
10656 opcode(0xF2, 0x0F, 0x59);
10657 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10658 ins_pipe(pipe_slow);
10659 %}
10660
10661 instruct mulD_imm(regD dst, immD con) %{
10662 match(Set dst (MulD dst con));
10663 format %{ "mulsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10664 ins_cost(150); // XXX
10665 ins_encode %{
10666 __ mulsd($dst$$XMMRegister, $constantaddress($con));
10667 %}
10668 ins_pipe(pipe_slow);
10669 %}
10670
10671 instruct divF_reg(regF dst, regF src)
10672 %{
10673 match(Set dst (DivF dst src));
10674
10675 format %{ "divss $dst, $src" %}
10676 ins_cost(150); // XXX
10677 opcode(0xF3, 0x0F, 0x5E);
10678 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10679 ins_pipe(pipe_slow);
10680 %}
10681
10682 instruct divF_mem(regF dst, memory src)
10683 %{
10684 match(Set dst (DivF dst (LoadF src)));
10685
10686 format %{ "divss $dst, $src" %}
10687 ins_cost(150); // XXX
10688 opcode(0xF3, 0x0F, 0x5E);
10689 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10690 ins_pipe(pipe_slow);
10691 %}
10692
10693 instruct divF_imm(regF dst, immF con) %{
10694 match(Set dst (DivF dst con));
10695 format %{ "divss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10696 ins_cost(150); // XXX
10697 ins_encode %{
10698 __ divss($dst$$XMMRegister, $constantaddress($con));
10699 %}
10700 ins_pipe(pipe_slow);
10701 %}
10702
10703 instruct divD_reg(regD dst, regD src)
10704 %{
10705 match(Set dst (DivD dst src));
10706
10707 format %{ "divsd $dst, $src" %}
10708 ins_cost(150); // XXX
10709 opcode(0xF2, 0x0F, 0x5E);
10710 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10711 ins_pipe(pipe_slow);
10712 %}
10713
10714 instruct divD_mem(regD dst, memory src)
10715 %{
10716 match(Set dst (DivD dst (LoadD src)));
10717
10718 format %{ "divsd $dst, $src" %}
10719 ins_cost(150); // XXX
10720 opcode(0xF2, 0x0F, 0x5E);
10721 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10722 ins_pipe(pipe_slow);
10723 %}
10724
10725 instruct divD_imm(regD dst, immD con) %{
10726 match(Set dst (DivD dst con));
10727 format %{ "divsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10728 ins_cost(150); // XXX
10729 ins_encode %{
10730 __ divsd($dst$$XMMRegister, $constantaddress($con));
10731 %}
10732 ins_pipe(pipe_slow);
10733 %}
10734
10735 instruct sqrtF_reg(regF dst, regF src)
10736 %{
10737 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10738
10739 format %{ "sqrtss $dst, $src" %}
10740 ins_cost(150); // XXX
10741 opcode(0xF3, 0x0F, 0x51);
10742 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10743 ins_pipe(pipe_slow);
10744 %}
10745
10746 instruct sqrtF_mem(regF dst, memory src)
10747 %{
10748 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10749
10750 format %{ "sqrtss $dst, $src" %}
10751 ins_cost(150); // XXX
10752 opcode(0xF3, 0x0F, 0x51);
10753 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10754 ins_pipe(pipe_slow);
10755 %}
10756
10757 instruct sqrtF_imm(regF dst, immF con) %{
10758 match(Set dst (ConvD2F (SqrtD (ConvF2D con))));
10759 format %{ "sqrtss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10760 ins_cost(150); // XXX
10761 ins_encode %{
10762 __ sqrtss($dst$$XMMRegister, $constantaddress($con));
10763 %}
10764 ins_pipe(pipe_slow);
10765 %}
10766
10767 instruct sqrtD_reg(regD dst, regD src)
10768 %{
10769 match(Set dst (SqrtD src));
10770
10771 format %{ "sqrtsd $dst, $src" %}
10772 ins_cost(150); // XXX
10773 opcode(0xF2, 0x0F, 0x51);
10774 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10775 ins_pipe(pipe_slow);
10776 %}
10777
10778 instruct sqrtD_mem(regD dst, memory src)
10779 %{
10780 match(Set dst (SqrtD (LoadD src)));
10781
10782 format %{ "sqrtsd $dst, $src" %}
10783 ins_cost(150); // XXX
10784 opcode(0xF2, 0x0F, 0x51);
10785 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10786 ins_pipe(pipe_slow);
10787 %}
10788
10789 instruct sqrtD_imm(regD dst, immD con) %{
10790 match(Set dst (SqrtD con));
10791 format %{ "sqrtsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10792 ins_cost(150); // XXX
10793 ins_encode %{
10794 __ sqrtsd($dst$$XMMRegister, $constantaddress($con));
10795 %}
10796 ins_pipe(pipe_slow);
10797 %}
10798
10799 instruct absF_reg(regF dst)
10800 %{
10801 match(Set dst (AbsF dst));
10802
10803 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
10804 ins_encode(absF_encoding(dst));
10805 ins_pipe(pipe_slow);
10806 %}
10807
10808 instruct absD_reg(regD dst)
10809 %{
10810 match(Set dst (AbsD dst));
10811
10812 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
10813 "# abs double by sign masking" %}
10814 ins_encode(absD_encoding(dst));
10815 ins_pipe(pipe_slow);
10816 %}
10817
10818 instruct negF_reg(regF dst)
10819 %{
10820 match(Set dst (NegF dst));
10821
10822 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
10823 ins_encode(negF_encoding(dst));
10824 ins_pipe(pipe_slow);
10825 %}
10826
10827 instruct negD_reg(regD dst)
10828 %{
10829 match(Set dst (NegD dst));
10830
10831 format %{ "xorpd $dst, [0x8000000000000000]\t"
10832 "# neg double by sign flipping" %}
10833 ins_encode(negD_encoding(dst));
10834 ins_pipe(pipe_slow);
10835 %}
10836
10837 // -----------Trig and Trancendental Instructions------------------------------
10838 instruct cosD_reg(regD dst) %{
10839 match(Set dst (CosD dst));
10840
10841 format %{ "dcos $dst\n\t" %}
10842 opcode(0xD9, 0xFF);
10843 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10844 ins_pipe( pipe_slow );
10845 %}
10846
10847 instruct sinD_reg(regD dst) %{
10848 match(Set dst (SinD dst));
10849
10850 format %{ "dsin $dst\n\t" %}
10851 opcode(0xD9, 0xFE);
10852 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10853 ins_pipe( pipe_slow );
10854 %}
10855
10856 instruct tanD_reg(regD dst) %{
10857 match(Set dst (TanD dst));
10858
10859 format %{ "dtan $dst\n\t" %}
10860 ins_encode( Push_SrcXD(dst),
10861 Opcode(0xD9), Opcode(0xF2), //fptan
10862 Opcode(0xDD), Opcode(0xD8), //fstp st
10863 Push_ResultXD(dst) );
10864 ins_pipe( pipe_slow );
10865 %}
10866
10867 instruct log10D_reg(regD dst) %{
10868 // The source and result Double operands in XMM registers
10869 match(Set dst (Log10D dst));
10870 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
10871 // fyl2x ; compute log_10(2) * log_2(x)
10872 format %{ "fldlg2\t\t\t#Log10\n\t"
10873 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10874 %}
10875 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
10876 Push_SrcXD(dst),
10877 Opcode(0xD9), Opcode(0xF1), // fyl2x
10878 Push_ResultXD(dst));
10879
10880 ins_pipe( pipe_slow );
10881 %}
10882
10883 instruct logD_reg(regD dst) %{
10884 // The source and result Double operands in XMM registers
10885 match(Set dst (LogD dst));
10886 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
10887 // fyl2x ; compute log_e(2) * log_2(x)
10888 format %{ "fldln2\t\t\t#Log_e\n\t"
10889 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10890 %}
10891 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
10892 Push_SrcXD(dst),
10893 Opcode(0xD9), Opcode(0xF1), // fyl2x
10894 Push_ResultXD(dst));
10895 ins_pipe( pipe_slow );
10896 %}
10897
10898
10899
10900 //----------Arithmetic Conversion Instructions---------------------------------
10901
10902 instruct roundFloat_nop(regF dst)
10903 %{
10904 match(Set dst (RoundFloat dst));
10905
10906 ins_cost(0);
10907 ins_encode();
10908 ins_pipe(empty);
10909 %}
10910
10911 instruct roundDouble_nop(regD dst)
10912 %{
10913 match(Set dst (RoundDouble dst));
10914
10915 ins_cost(0);
10916 ins_encode();
10917 ins_pipe(empty);
10918 %}
10919
10920 instruct convF2D_reg_reg(regD dst, regF src)
10921 %{
10922 match(Set dst (ConvF2D src));
10923
10924 format %{ "cvtss2sd $dst, $src" %}
10925 opcode(0xF3, 0x0F, 0x5A);
10926 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10927 ins_pipe(pipe_slow); // XXX
10928 %}
10929
10930 instruct convF2D_reg_mem(regD dst, memory src)
10931 %{
10932 match(Set dst (ConvF2D (LoadF src)));
10933
10934 format %{ "cvtss2sd $dst, $src" %}
10935 opcode(0xF3, 0x0F, 0x5A);
10936 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10937 ins_pipe(pipe_slow); // XXX
10938 %}
10939
10940 instruct convD2F_reg_reg(regF dst, regD src)
10941 %{
10942 match(Set dst (ConvD2F src));
10943
10944 format %{ "cvtsd2ss $dst, $src" %}
10945 opcode(0xF2, 0x0F, 0x5A);
10946 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10947 ins_pipe(pipe_slow); // XXX
10948 %}
10949
10950 instruct convD2F_reg_mem(regF dst, memory src)
10951 %{
10952 match(Set dst (ConvD2F (LoadD src)));
10953
10954 format %{ "cvtsd2ss $dst, $src" %}
10955 opcode(0xF2, 0x0F, 0x5A);
10956 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10957 ins_pipe(pipe_slow); // XXX
10958 %}
10959
10960 // XXX do mem variants
10961 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
10962 %{
10963 match(Set dst (ConvF2I src));
10964 effect(KILL cr);
10965
10966 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
10967 "cmpl $dst, #0x80000000\n\t"
10968 "jne,s done\n\t"
10969 "subq rsp, #8\n\t"
10970 "movss [rsp], $src\n\t"
10971 "call f2i_fixup\n\t"
10972 "popq $dst\n"
10973 "done: "%}
10974 opcode(0xF3, 0x0F, 0x2C);
10975 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10976 f2i_fixup(dst, src));
10977 ins_pipe(pipe_slow);
10978 %}
10979
10980 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
10981 %{
10982 match(Set dst (ConvF2L src));
10983 effect(KILL cr);
10984
10985 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
10986 "cmpq $dst, [0x8000000000000000]\n\t"
10987 "jne,s done\n\t"
10988 "subq rsp, #8\n\t"
10989 "movss [rsp], $src\n\t"
10990 "call f2l_fixup\n\t"
10991 "popq $dst\n"
10992 "done: "%}
10993 opcode(0xF3, 0x0F, 0x2C);
10994 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10995 f2l_fixup(dst, src));
10996 ins_pipe(pipe_slow);
10997 %}
10998
10999 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
11000 %{
11001 match(Set dst (ConvD2I src));
11002 effect(KILL cr);
11003
11004 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
11005 "cmpl $dst, #0x80000000\n\t"
11006 "jne,s done\n\t"
11007 "subq rsp, #8\n\t"
11008 "movsd [rsp], $src\n\t"
11009 "call d2i_fixup\n\t"
11010 "popq $dst\n"
11011 "done: "%}
11012 opcode(0xF2, 0x0F, 0x2C);
11013 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11014 d2i_fixup(dst, src));
11015 ins_pipe(pipe_slow);
11016 %}
11017
11018 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
11019 %{
11020 match(Set dst (ConvD2L src));
11021 effect(KILL cr);
11022
11023 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
11024 "cmpq $dst, [0x8000000000000000]\n\t"
11025 "jne,s done\n\t"
11026 "subq rsp, #8\n\t"
11027 "movsd [rsp], $src\n\t"
11028 "call d2l_fixup\n\t"
11029 "popq $dst\n"
11030 "done: "%}
11031 opcode(0xF2, 0x0F, 0x2C);
11032 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11033 d2l_fixup(dst, src));
11034 ins_pipe(pipe_slow);
11035 %}
11036
11037 instruct convI2F_reg_reg(regF dst, rRegI src)
11038 %{
11039 predicate(!UseXmmI2F);
11040 match(Set dst (ConvI2F src));
11041
11042 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11043 opcode(0xF3, 0x0F, 0x2A);
11044 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11045 ins_pipe(pipe_slow); // XXX
11046 %}
11047
11048 instruct convI2F_reg_mem(regF dst, memory src)
11049 %{
11050 match(Set dst (ConvI2F (LoadI src)));
11051
11052 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11053 opcode(0xF3, 0x0F, 0x2A);
11054 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11055 ins_pipe(pipe_slow); // XXX
11056 %}
11057
11058 instruct convI2D_reg_reg(regD dst, rRegI src)
11059 %{
11060 predicate(!UseXmmI2D);
11061 match(Set dst (ConvI2D src));
11062
11063 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11064 opcode(0xF2, 0x0F, 0x2A);
11065 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11066 ins_pipe(pipe_slow); // XXX
11067 %}
11068
11069 instruct convI2D_reg_mem(regD dst, memory src)
11070 %{
11071 match(Set dst (ConvI2D (LoadI src)));
11072
11073 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11074 opcode(0xF2, 0x0F, 0x2A);
11075 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11076 ins_pipe(pipe_slow); // XXX
11077 %}
11078
11079 instruct convXI2F_reg(regF dst, rRegI src)
11080 %{
11081 predicate(UseXmmI2F);
11082 match(Set dst (ConvI2F src));
11083
11084 format %{ "movdl $dst, $src\n\t"
11085 "cvtdq2psl $dst, $dst\t# i2f" %}
11086 ins_encode %{
11087 __ movdl($dst$$XMMRegister, $src$$Register);
11088 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11089 %}
11090 ins_pipe(pipe_slow); // XXX
11091 %}
11092
11093 instruct convXI2D_reg(regD dst, rRegI src)
11094 %{
11095 predicate(UseXmmI2D);
11096 match(Set dst (ConvI2D src));
11097
11098 format %{ "movdl $dst, $src\n\t"
11099 "cvtdq2pdl $dst, $dst\t# i2d" %}
11100 ins_encode %{
11101 __ movdl($dst$$XMMRegister, $src$$Register);
11102 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11103 %}
11104 ins_pipe(pipe_slow); // XXX
11105 %}
11106
11107 instruct convL2F_reg_reg(regF dst, rRegL src)
11108 %{
11109 match(Set dst (ConvL2F src));
11110
11111 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11112 opcode(0xF3, 0x0F, 0x2A);
11113 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11114 ins_pipe(pipe_slow); // XXX
11115 %}
11116
11117 instruct convL2F_reg_mem(regF dst, memory src)
11118 %{
11119 match(Set dst (ConvL2F (LoadL src)));
11120
11121 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11122 opcode(0xF3, 0x0F, 0x2A);
11123 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11124 ins_pipe(pipe_slow); // XXX
11125 %}
11126
11127 instruct convL2D_reg_reg(regD dst, rRegL src)
11128 %{
11129 match(Set dst (ConvL2D src));
11130
11131 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11132 opcode(0xF2, 0x0F, 0x2A);
11133 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11134 ins_pipe(pipe_slow); // XXX
11135 %}
11136
11137 instruct convL2D_reg_mem(regD dst, memory src)
11138 %{
11139 match(Set dst (ConvL2D (LoadL src)));
11140
11141 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11142 opcode(0xF2, 0x0F, 0x2A);
11143 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11144 ins_pipe(pipe_slow); // XXX
11145 %}
11146
11147 instruct convI2L_reg_reg(rRegL dst, rRegI src)
11148 %{
11149 match(Set dst (ConvI2L src));
11150
11151 ins_cost(125);
11152 format %{ "movslq $dst, $src\t# i2l" %}
11153 ins_encode %{
11154 __ movslq($dst$$Register, $src$$Register);
11155 %}
11156 ins_pipe(ialu_reg_reg);
11157 %}
11158
11159 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
11160 // %{
11161 // match(Set dst (ConvI2L src));
11162 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
11163 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
11164 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
11165 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
11166 // ((const TypeNode*) n)->type()->is_long()->_lo ==
11167 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
11168
11169 // format %{ "movl $dst, $src\t# unsigned i2l" %}
11170 // ins_encode(enc_copy(dst, src));
11171 // // opcode(0x63); // needs REX.W
11172 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11173 // ins_pipe(ialu_reg_reg);
11174 // %}
11175
11176 // Zero-extend convert int to long
11177 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
11178 %{
11179 match(Set dst (AndL (ConvI2L src) mask));
11180
11181 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11182 ins_encode(enc_copy(dst, src));
11183 ins_pipe(ialu_reg_reg);
11184 %}
11185
11186 // Zero-extend convert int to long
11187 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
11188 %{
11189 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
11190
11191 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11192 opcode(0x8B);
11193 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11194 ins_pipe(ialu_reg_mem);
11195 %}
11196
11197 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
11198 %{
11199 match(Set dst (AndL src mask));
11200
11201 format %{ "movl $dst, $src\t# zero-extend long" %}
11202 ins_encode(enc_copy_always(dst, src));
11203 ins_pipe(ialu_reg_reg);
11204 %}
11205
11206 instruct convL2I_reg_reg(rRegI dst, rRegL src)
11207 %{
11208 match(Set dst (ConvL2I src));
11209
11210 format %{ "movl $dst, $src\t# l2i" %}
11211 ins_encode(enc_copy_always(dst, src));
11212 ins_pipe(ialu_reg_reg);
11213 %}
11214
11215
11216 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11217 match(Set dst (MoveF2I src));
11218 effect(DEF dst, USE src);
11219
11220 ins_cost(125);
11221 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
11222 opcode(0x8B);
11223 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11224 ins_pipe(ialu_reg_mem);
11225 %}
11226
11227 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
11228 match(Set dst (MoveI2F src));
11229 effect(DEF dst, USE src);
11230
11231 ins_cost(125);
11232 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
11233 opcode(0xF3, 0x0F, 0x10);
11234 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11235 ins_pipe(pipe_slow);
11236 %}
11237
11238 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
11239 match(Set dst (MoveD2L src));
11240 effect(DEF dst, USE src);
11241
11242 ins_cost(125);
11243 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
11244 opcode(0x8B);
11245 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
11246 ins_pipe(ialu_reg_mem);
11247 %}
11248
11249 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
11250 predicate(!UseXmmLoadAndClearUpper);
11251 match(Set dst (MoveL2D src));
11252 effect(DEF dst, USE src);
11253
11254 ins_cost(125);
11255 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
11256 opcode(0x66, 0x0F, 0x12);
11257 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11258 ins_pipe(pipe_slow);
11259 %}
11260
11261 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
11262 predicate(UseXmmLoadAndClearUpper);
11263 match(Set dst (MoveL2D src));
11264 effect(DEF dst, USE src);
11265
11266 ins_cost(125);
11267 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
11268 opcode(0xF2, 0x0F, 0x10);
11269 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11270 ins_pipe(pipe_slow);
11271 %}
11272
11273
11274 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
11275 match(Set dst (MoveF2I src));
11276 effect(DEF dst, USE src);
11277
11278 ins_cost(95); // XXX
11279 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
11280 opcode(0xF3, 0x0F, 0x11);
11281 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11282 ins_pipe(pipe_slow);
11283 %}
11284
11285 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11286 match(Set dst (MoveI2F src));
11287 effect(DEF dst, USE src);
11288
11289 ins_cost(100);
11290 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
11291 opcode(0x89);
11292 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
11293 ins_pipe( ialu_mem_reg );
11294 %}
11295
11296 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
11297 match(Set dst (MoveD2L src));
11298 effect(DEF dst, USE src);
11299
11300 ins_cost(95); // XXX
11301 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
11302 opcode(0xF2, 0x0F, 0x11);
11303 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11304 ins_pipe(pipe_slow);
11305 %}
11306
11307 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
11308 match(Set dst (MoveL2D src));
11309 effect(DEF dst, USE src);
11310
11311 ins_cost(100);
11312 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
11313 opcode(0x89);
11314 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
11315 ins_pipe(ialu_mem_reg);
11316 %}
11317
11318 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
11319 match(Set dst (MoveF2I src));
11320 effect(DEF dst, USE src);
11321 ins_cost(85);
11322 format %{ "movd $dst,$src\t# MoveF2I" %}
11323 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
11324 ins_pipe( pipe_slow );
11325 %}
11326
11327 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
11328 match(Set dst (MoveD2L src));
11329 effect(DEF dst, USE src);
11330 ins_cost(85);
11331 format %{ "movd $dst,$src\t# MoveD2L" %}
11332 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
11333 ins_pipe( pipe_slow );
11334 %}
11335
11336 // The next instructions have long latency and use Int unit. Set high cost.
11337 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
11338 match(Set dst (MoveI2F src));
11339 effect(DEF dst, USE src);
11340 ins_cost(300);
11341 format %{ "movd $dst,$src\t# MoveI2F" %}
11342 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
11343 ins_pipe( pipe_slow );
11344 %}
11345
11346 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
11347 match(Set dst (MoveL2D src));
11348 effect(DEF dst, USE src);
11349 ins_cost(300);
11350 format %{ "movd $dst,$src\t# MoveL2D" %}
11351 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
11352 ins_pipe( pipe_slow );
11353 %}
11354
11355 // Replicate scalar to packed byte (1 byte) values in xmm
11356 instruct Repl8B_reg(regD dst, regD src) %{
11357 match(Set dst (Replicate8B src));
11358 format %{ "MOVDQA $dst,$src\n\t"
11359 "PUNPCKLBW $dst,$dst\n\t"
11360 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11361 ins_encode( pshufd_8x8(dst, src));
11362 ins_pipe( pipe_slow );
11363 %}
11364
11365 // Replicate scalar to packed byte (1 byte) values in xmm
11366 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11367 match(Set dst (Replicate8B src));
11368 format %{ "MOVD $dst,$src\n\t"
11369 "PUNPCKLBW $dst,$dst\n\t"
11370 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11371 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11372 ins_pipe( pipe_slow );
11373 %}
11374
11375 // Replicate scalar zero to packed byte (1 byte) values in xmm
11376 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11377 match(Set dst (Replicate8B zero));
11378 format %{ "PXOR $dst,$dst\t! replicate8B" %}
11379 ins_encode( pxor(dst, dst));
11380 ins_pipe( fpu_reg_reg );
11381 %}
11382
11383 // Replicate scalar to packed shore (2 byte) values in xmm
11384 instruct Repl4S_reg(regD dst, regD src) %{
11385 match(Set dst (Replicate4S src));
11386 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11387 ins_encode( pshufd_4x16(dst, src));
11388 ins_pipe( fpu_reg_reg );
11389 %}
11390
11391 // Replicate scalar to packed shore (2 byte) values in xmm
11392 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11393 match(Set dst (Replicate4S src));
11394 format %{ "MOVD $dst,$src\n\t"
11395 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11396 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11397 ins_pipe( fpu_reg_reg );
11398 %}
11399
11400 // Replicate scalar zero to packed short (2 byte) values in xmm
11401 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11402 match(Set dst (Replicate4S zero));
11403 format %{ "PXOR $dst,$dst\t! replicate4S" %}
11404 ins_encode( pxor(dst, dst));
11405 ins_pipe( fpu_reg_reg );
11406 %}
11407
11408 // Replicate scalar to packed char (2 byte) values in xmm
11409 instruct Repl4C_reg(regD dst, regD src) %{
11410 match(Set dst (Replicate4C src));
11411 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11412 ins_encode( pshufd_4x16(dst, src));
11413 ins_pipe( fpu_reg_reg );
11414 %}
11415
11416 // Replicate scalar to packed char (2 byte) values in xmm
11417 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11418 match(Set dst (Replicate4C src));
11419 format %{ "MOVD $dst,$src\n\t"
11420 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11421 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11422 ins_pipe( fpu_reg_reg );
11423 %}
11424
11425 // Replicate scalar zero to packed char (2 byte) values in xmm
11426 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11427 match(Set dst (Replicate4C zero));
11428 format %{ "PXOR $dst,$dst\t! replicate4C" %}
11429 ins_encode( pxor(dst, dst));
11430 ins_pipe( fpu_reg_reg );
11431 %}
11432
11433 // Replicate scalar to packed integer (4 byte) values in xmm
11434 instruct Repl2I_reg(regD dst, regD src) %{
11435 match(Set dst (Replicate2I src));
11436 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11437 ins_encode( pshufd(dst, src, 0x00));
11438 ins_pipe( fpu_reg_reg );
11439 %}
11440
11441 // Replicate scalar to packed integer (4 byte) values in xmm
11442 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11443 match(Set dst (Replicate2I src));
11444 format %{ "MOVD $dst,$src\n\t"
11445 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11446 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11447 ins_pipe( fpu_reg_reg );
11448 %}
11449
11450 // Replicate scalar zero to packed integer (2 byte) values in xmm
11451 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11452 match(Set dst (Replicate2I zero));
11453 format %{ "PXOR $dst,$dst\t! replicate2I" %}
11454 ins_encode( pxor(dst, dst));
11455 ins_pipe( fpu_reg_reg );
11456 %}
11457
11458 // Replicate scalar to packed single precision floating point values in xmm
11459 instruct Repl2F_reg(regD dst, regD src) %{
11460 match(Set dst (Replicate2F src));
11461 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11462 ins_encode( pshufd(dst, src, 0xe0));
11463 ins_pipe( fpu_reg_reg );
11464 %}
11465
11466 // Replicate scalar to packed single precision floating point values in xmm
11467 instruct Repl2F_regF(regD dst, regF src) %{
11468 match(Set dst (Replicate2F src));
11469 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11470 ins_encode( pshufd(dst, src, 0xe0));
11471 ins_pipe( fpu_reg_reg );
11472 %}
11473
11474 // Replicate scalar to packed single precision floating point values in xmm
11475 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11476 match(Set dst (Replicate2F zero));
11477 format %{ "PXOR $dst,$dst\t! replicate2F" %}
11478 ins_encode( pxor(dst, dst));
11479 ins_pipe( fpu_reg_reg );
11480 %}
11481
11482
11483 // =======================================================================
11484 // fast clearing of an array
11485 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11486 rFlagsReg cr)
11487 %{
11488 match(Set dummy (ClearArray cnt base));
11489 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11490
11491 format %{ "xorl rax, rax\t# ClearArray:\n\t"
11492 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11493 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11494 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11495 ins_pipe(pipe_slow);
11496 %}
11497
11498 instruct string_compare(rdi_RegP str1, rcx_RegI cnt1, rsi_RegP str2, rdx_RegI cnt2,
11499 rax_RegI result, regD tmp1, rFlagsReg cr)
11500 %{
11501 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11502 effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
11503
11504 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp1" %}
11505 ins_encode %{
11506 __ string_compare($str1$$Register, $str2$$Register,
11507 $cnt1$$Register, $cnt2$$Register, $result$$Register,
11508 $tmp1$$XMMRegister);
11509 %}
11510 ins_pipe( pipe_slow );
11511 %}
11512
11513 // fast search of substring with known size.
11514 instruct string_indexof_con(rdi_RegP str1, rdx_RegI cnt1, rsi_RegP str2, immI int_cnt2,
11515 rbx_RegI result, regD vec, rax_RegI cnt2, rcx_RegI tmp, rFlagsReg cr)
11516 %{
11517 predicate(UseSSE42Intrinsics);
11518 match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
11519 effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, KILL cnt2, KILL tmp, KILL cr);
11520
11521 format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result // KILL $vec, $cnt1, $cnt2, $tmp" %}
11522 ins_encode %{
11523 int icnt2 = (int)$int_cnt2$$constant;
11524 if (icnt2 >= 8) {
11525 // IndexOf for constant substrings with size >= 8 elements
11526 // which don't need to be loaded through stack.
11527 __ string_indexofC8($str1$$Register, $str2$$Register,
11528 $cnt1$$Register, $cnt2$$Register,
11529 icnt2, $result$$Register,
11530 $vec$$XMMRegister, $tmp$$Register);
11531 } else {
11532 // Small strings are loaded through stack if they cross page boundary.
11533 __ string_indexof($str1$$Register, $str2$$Register,
11534 $cnt1$$Register, $cnt2$$Register,
11535 icnt2, $result$$Register,
11536 $vec$$XMMRegister, $tmp$$Register);
11537 }
11538 %}
11539 ins_pipe( pipe_slow );
11540 %}
11541
11542 instruct string_indexof(rdi_RegP str1, rdx_RegI cnt1, rsi_RegP str2, rax_RegI cnt2,
11543 rbx_RegI result, regD vec, rcx_RegI tmp, rFlagsReg cr)
11544 %{
11545 predicate(UseSSE42Intrinsics);
11546 match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
11547 effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp, KILL cr);
11548
11549 format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result // KILL all" %}
11550 ins_encode %{
11551 __ string_indexof($str1$$Register, $str2$$Register,
11552 $cnt1$$Register, $cnt2$$Register,
11553 (-1), $result$$Register,
11554 $vec$$XMMRegister, $tmp$$Register);
11555 %}
11556 ins_pipe( pipe_slow );
11557 %}
11558
11559 // fast string equals
11560 instruct string_equals(rdi_RegP str1, rsi_RegP str2, rcx_RegI cnt, rax_RegI result,
11561 regD tmp1, regD tmp2, rbx_RegI tmp3, rFlagsReg cr)
11562 %{
11563 match(Set result (StrEquals (Binary str1 str2) cnt));
11564 effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);
11565
11566 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp1, $tmp2, $tmp3" %}
11567 ins_encode %{
11568 __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
11569 $cnt$$Register, $result$$Register, $tmp3$$Register,
11570 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11571 %}
11572 ins_pipe( pipe_slow );
11573 %}
11574
11575 // fast array equals
11576 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI result,
11577 regD tmp1, regD tmp2, rcx_RegI tmp3, rbx_RegI tmp4, rFlagsReg cr)
11578 %{
11579 match(Set result (AryEq ary1 ary2));
11580 effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
11581 //ins_cost(300);
11582
11583 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
11584 ins_encode %{
11585 __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
11586 $tmp3$$Register, $result$$Register, $tmp4$$Register,
11587 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11588 %}
11589 ins_pipe( pipe_slow );
11590 %}
11591
11592 //----------Control Flow Instructions------------------------------------------
11593 // Signed compare Instructions
11594
11595 // XXX more variants!!
11596 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11597 %{
11598 match(Set cr (CmpI op1 op2));
11599 effect(DEF cr, USE op1, USE op2);
11600
11601 format %{ "cmpl $op1, $op2" %}
11602 opcode(0x3B); /* Opcode 3B /r */
11603 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11604 ins_pipe(ialu_cr_reg_reg);
11605 %}
11606
11607 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11608 %{
11609 match(Set cr (CmpI op1 op2));
11610
11611 format %{ "cmpl $op1, $op2" %}
11612 opcode(0x81, 0x07); /* Opcode 81 /7 */
11613 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11614 ins_pipe(ialu_cr_reg_imm);
11615 %}
11616
11617 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11618 %{
11619 match(Set cr (CmpI op1 (LoadI op2)));
11620
11621 ins_cost(500); // XXX
11622 format %{ "cmpl $op1, $op2" %}
11623 opcode(0x3B); /* Opcode 3B /r */
11624 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11625 ins_pipe(ialu_cr_reg_mem);
11626 %}
11627
11628 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11629 %{
11630 match(Set cr (CmpI src zero));
11631
11632 format %{ "testl $src, $src" %}
11633 opcode(0x85);
11634 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11635 ins_pipe(ialu_cr_reg_imm);
11636 %}
11637
11638 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11639 %{
11640 match(Set cr (CmpI (AndI src con) zero));
11641
11642 format %{ "testl $src, $con" %}
11643 opcode(0xF7, 0x00);
11644 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11645 ins_pipe(ialu_cr_reg_imm);
11646 %}
11647
11648 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11649 %{
11650 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11651
11652 format %{ "testl $src, $mem" %}
11653 opcode(0x85);
11654 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11655 ins_pipe(ialu_cr_reg_mem);
11656 %}
11657
11658 // Unsigned compare Instructions; really, same as signed except they
11659 // produce an rFlagsRegU instead of rFlagsReg.
11660 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11661 %{
11662 match(Set cr (CmpU op1 op2));
11663
11664 format %{ "cmpl $op1, $op2\t# unsigned" %}
11665 opcode(0x3B); /* Opcode 3B /r */
11666 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11667 ins_pipe(ialu_cr_reg_reg);
11668 %}
11669
11670 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11671 %{
11672 match(Set cr (CmpU op1 op2));
11673
11674 format %{ "cmpl $op1, $op2\t# unsigned" %}
11675 opcode(0x81,0x07); /* Opcode 81 /7 */
11676 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11677 ins_pipe(ialu_cr_reg_imm);
11678 %}
11679
11680 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11681 %{
11682 match(Set cr (CmpU op1 (LoadI op2)));
11683
11684 ins_cost(500); // XXX
11685 format %{ "cmpl $op1, $op2\t# unsigned" %}
11686 opcode(0x3B); /* Opcode 3B /r */
11687 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11688 ins_pipe(ialu_cr_reg_mem);
11689 %}
11690
11691 // // // Cisc-spilled version of cmpU_rReg
11692 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11693 // //%{
11694 // // match(Set cr (CmpU (LoadI op1) op2));
11695 // //
11696 // // format %{ "CMPu $op1,$op2" %}
11697 // // ins_cost(500);
11698 // // opcode(0x39); /* Opcode 39 /r */
11699 // // ins_encode( OpcP, reg_mem( op1, op2) );
11700 // //%}
11701
11702 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11703 %{
11704 match(Set cr (CmpU src zero));
11705
11706 format %{ "testl $src, $src\t# unsigned" %}
11707 opcode(0x85);
11708 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11709 ins_pipe(ialu_cr_reg_imm);
11710 %}
11711
11712 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11713 %{
11714 match(Set cr (CmpP op1 op2));
11715
11716 format %{ "cmpq $op1, $op2\t# ptr" %}
11717 opcode(0x3B); /* Opcode 3B /r */
11718 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11719 ins_pipe(ialu_cr_reg_reg);
11720 %}
11721
11722 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11723 %{
11724 match(Set cr (CmpP op1 (LoadP op2)));
11725
11726 ins_cost(500); // XXX
11727 format %{ "cmpq $op1, $op2\t# ptr" %}
11728 opcode(0x3B); /* Opcode 3B /r */
11729 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11730 ins_pipe(ialu_cr_reg_mem);
11731 %}
11732
11733 // // // Cisc-spilled version of cmpP_rReg
11734 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11735 // //%{
11736 // // match(Set cr (CmpP (LoadP op1) op2));
11737 // //
11738 // // format %{ "CMPu $op1,$op2" %}
11739 // // ins_cost(500);
11740 // // opcode(0x39); /* Opcode 39 /r */
11741 // // ins_encode( OpcP, reg_mem( op1, op2) );
11742 // //%}
11743
11744 // XXX this is generalized by compP_rReg_mem???
11745 // Compare raw pointer (used in out-of-heap check).
11746 // Only works because non-oop pointers must be raw pointers
11747 // and raw pointers have no anti-dependencies.
11748 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11749 %{
11750 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11751 match(Set cr (CmpP op1 (LoadP op2)));
11752
11753 format %{ "cmpq $op1, $op2\t# raw ptr" %}
11754 opcode(0x3B); /* Opcode 3B /r */
11755 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11756 ins_pipe(ialu_cr_reg_mem);
11757 %}
11758
11759 // This will generate a signed flags result. This should be OK since
11760 // any compare to a zero should be eq/neq.
11761 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11762 %{
11763 match(Set cr (CmpP src zero));
11764
11765 format %{ "testq $src, $src\t# ptr" %}
11766 opcode(0x85);
11767 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11768 ins_pipe(ialu_cr_reg_imm);
11769 %}
11770
11771 // This will generate a signed flags result. This should be OK since
11772 // any compare to a zero should be eq/neq.
11773 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero)
11774 %{
11775 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
11776 match(Set cr (CmpP (LoadP op) zero));
11777
11778 ins_cost(500); // XXX
11779 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11780 opcode(0xF7); /* Opcode F7 /0 */
11781 ins_encode(REX_mem_wide(op),
11782 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11783 ins_pipe(ialu_cr_reg_imm);
11784 %}
11785
11786 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero)
11787 %{
11788 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
11789 match(Set cr (CmpP (LoadP mem) zero));
11790
11791 format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %}
11792 ins_encode %{
11793 __ cmpq(r12, $mem$$Address);
11794 %}
11795 ins_pipe(ialu_cr_reg_mem);
11796 %}
11797
11798 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11799 %{
11800 match(Set cr (CmpN op1 op2));
11801
11802 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11803 ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %}
11804 ins_pipe(ialu_cr_reg_reg);
11805 %}
11806
11807 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11808 %{
11809 match(Set cr (CmpN src (LoadN mem)));
11810
11811 format %{ "cmpl $src, $mem\t# compressed ptr" %}
11812 ins_encode %{
11813 __ cmpl($src$$Register, $mem$$Address);
11814 %}
11815 ins_pipe(ialu_cr_reg_mem);
11816 %}
11817
11818 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{
11819 match(Set cr (CmpN op1 op2));
11820
11821 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11822 ins_encode %{
11823 __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant);
11824 %}
11825 ins_pipe(ialu_cr_reg_imm);
11826 %}
11827
11828 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src)
11829 %{
11830 match(Set cr (CmpN src (LoadN mem)));
11831
11832 format %{ "cmpl $mem, $src\t# compressed ptr" %}
11833 ins_encode %{
11834 __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant);
11835 %}
11836 ins_pipe(ialu_cr_reg_mem);
11837 %}
11838
11839 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11840 match(Set cr (CmpN src zero));
11841
11842 format %{ "testl $src, $src\t# compressed ptr" %}
11843 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11844 ins_pipe(ialu_cr_reg_imm);
11845 %}
11846
11847 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero)
11848 %{
11849 predicate(Universe::narrow_oop_base() != NULL);
11850 match(Set cr (CmpN (LoadN mem) zero));
11851
11852 ins_cost(500); // XXX
11853 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %}
11854 ins_encode %{
11855 __ cmpl($mem$$Address, (int)0xFFFFFFFF);
11856 %}
11857 ins_pipe(ialu_cr_reg_mem);
11858 %}
11859
11860 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero)
11861 %{
11862 predicate(Universe::narrow_oop_base() == NULL);
11863 match(Set cr (CmpN (LoadN mem) zero));
11864
11865 format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %}
11866 ins_encode %{
11867 __ cmpl(r12, $mem$$Address);
11868 %}
11869 ins_pipe(ialu_cr_reg_mem);
11870 %}
11871
11872 // Yanked all unsigned pointer compare operations.
11873 // Pointer compares are done with CmpP which is already unsigned.
11874
11875 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11876 %{
11877 match(Set cr (CmpL op1 op2));
11878
11879 format %{ "cmpq $op1, $op2" %}
11880 opcode(0x3B); /* Opcode 3B /r */
11881 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11882 ins_pipe(ialu_cr_reg_reg);
11883 %}
11884
11885 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11886 %{
11887 match(Set cr (CmpL op1 op2));
11888
11889 format %{ "cmpq $op1, $op2" %}
11890 opcode(0x81, 0x07); /* Opcode 81 /7 */
11891 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11892 ins_pipe(ialu_cr_reg_imm);
11893 %}
11894
11895 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11896 %{
11897 match(Set cr (CmpL op1 (LoadL op2)));
11898
11899 format %{ "cmpq $op1, $op2" %}
11900 opcode(0x3B); /* Opcode 3B /r */
11901 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11902 ins_pipe(ialu_cr_reg_mem);
11903 %}
11904
11905 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11906 %{
11907 match(Set cr (CmpL src zero));
11908
11909 format %{ "testq $src, $src" %}
11910 opcode(0x85);
11911 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11912 ins_pipe(ialu_cr_reg_imm);
11913 %}
11914
11915 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11916 %{
11917 match(Set cr (CmpL (AndL src con) zero));
11918
11919 format %{ "testq $src, $con\t# long" %}
11920 opcode(0xF7, 0x00);
11921 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11922 ins_pipe(ialu_cr_reg_imm);
11923 %}
11924
11925 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
11926 %{
11927 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
11928
11929 format %{ "testq $src, $mem" %}
11930 opcode(0x85);
11931 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
11932 ins_pipe(ialu_cr_reg_mem);
11933 %}
11934
11935 // Manifest a CmpL result in an integer register. Very painful.
11936 // This is the test to avoid.
11937 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
11938 %{
11939 match(Set dst (CmpL3 src1 src2));
11940 effect(KILL flags);
11941
11942 ins_cost(275); // XXX
11943 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
11944 "movl $dst, -1\n\t"
11945 "jl,s done\n\t"
11946 "setne $dst\n\t"
11947 "movzbl $dst, $dst\n\t"
11948 "done:" %}
11949 ins_encode(cmpl3_flag(src1, src2, dst));
11950 ins_pipe(pipe_slow);
11951 %}
11952
11953 //----------Max and Min--------------------------------------------------------
11954 // Min Instructions
11955
11956 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
11957 %{
11958 effect(USE_DEF dst, USE src, USE cr);
11959
11960 format %{ "cmovlgt $dst, $src\t# min" %}
11961 opcode(0x0F, 0x4F);
11962 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11963 ins_pipe(pipe_cmov_reg);
11964 %}
11965
11966
11967 instruct minI_rReg(rRegI dst, rRegI src)
11968 %{
11969 match(Set dst (MinI dst src));
11970
11971 ins_cost(200);
11972 expand %{
11973 rFlagsReg cr;
11974 compI_rReg(cr, dst, src);
11975 cmovI_reg_g(dst, src, cr);
11976 %}
11977 %}
11978
11979 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
11980 %{
11981 effect(USE_DEF dst, USE src, USE cr);
11982
11983 format %{ "cmovllt $dst, $src\t# max" %}
11984 opcode(0x0F, 0x4C);
11985 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11986 ins_pipe(pipe_cmov_reg);
11987 %}
11988
11989
11990 instruct maxI_rReg(rRegI dst, rRegI src)
11991 %{
11992 match(Set dst (MaxI dst src));
11993
11994 ins_cost(200);
11995 expand %{
11996 rFlagsReg cr;
11997 compI_rReg(cr, dst, src);
11998 cmovI_reg_l(dst, src, cr);
11999 %}
12000 %}
12001
12002 // ============================================================================
12003 // Branch Instructions
12004
12005 // Jump Direct - Label defines a relative address from JMP+1
12006 instruct jmpDir(label labl)
12007 %{
12008 match(Goto);
12009 effect(USE labl);
12010
12011 ins_cost(300);
12012 format %{ "jmp $labl" %}
12013 size(5);
12014 opcode(0xE9);
12015 ins_encode(OpcP, Lbl(labl));
12016 ins_pipe(pipe_jmp);
12017 %}
12018
12019 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12020 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
12021 %{
12022 match(If cop cr);
12023 effect(USE labl);
12024
12025 ins_cost(300);
12026 format %{ "j$cop $labl" %}
12027 size(6);
12028 opcode(0x0F, 0x80);
12029 ins_encode(Jcc(cop, labl));
12030 ins_pipe(pipe_jcc);
12031 %}
12032
12033 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12034 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
12035 %{
12036 match(CountedLoopEnd cop cr);
12037 effect(USE labl);
12038
12039 ins_cost(300);
12040 format %{ "j$cop $labl\t# loop end" %}
12041 size(6);
12042 opcode(0x0F, 0x80);
12043 ins_encode(Jcc(cop, labl));
12044 ins_pipe(pipe_jcc);
12045 %}
12046
12047 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12048 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12049 match(CountedLoopEnd cop cmp);
12050 effect(USE labl);
12051
12052 ins_cost(300);
12053 format %{ "j$cop,u $labl\t# loop end" %}
12054 size(6);
12055 opcode(0x0F, 0x80);
12056 ins_encode(Jcc(cop, labl));
12057 ins_pipe(pipe_jcc);
12058 %}
12059
12060 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12061 match(CountedLoopEnd cop cmp);
12062 effect(USE labl);
12063
12064 ins_cost(200);
12065 format %{ "j$cop,u $labl\t# loop end" %}
12066 size(6);
12067 opcode(0x0F, 0x80);
12068 ins_encode(Jcc(cop, labl));
12069 ins_pipe(pipe_jcc);
12070 %}
12071
12072 // Jump Direct Conditional - using unsigned comparison
12073 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12074 match(If cop cmp);
12075 effect(USE labl);
12076
12077 ins_cost(300);
12078 format %{ "j$cop,u $labl" %}
12079 size(6);
12080 opcode(0x0F, 0x80);
12081 ins_encode(Jcc(cop, labl));
12082 ins_pipe(pipe_jcc);
12083 %}
12084
12085 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12086 match(If cop cmp);
12087 effect(USE labl);
12088
12089 ins_cost(200);
12090 format %{ "j$cop,u $labl" %}
12091 size(6);
12092 opcode(0x0F, 0x80);
12093 ins_encode(Jcc(cop, labl));
12094 ins_pipe(pipe_jcc);
12095 %}
12096
12097 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12098 match(If cop cmp);
12099 effect(USE labl);
12100
12101 ins_cost(200);
12102 format %{ $$template
12103 if ($cop$$cmpcode == Assembler::notEqual) {
12104 $$emit$$"jp,u $labl\n\t"
12105 $$emit$$"j$cop,u $labl"
12106 } else {
12107 $$emit$$"jp,u done\n\t"
12108 $$emit$$"j$cop,u $labl\n\t"
12109 $$emit$$"done:"
12110 }
12111 %}
12112 size(12);
12113 opcode(0x0F, 0x80);
12114 ins_encode %{
12115 Label* l = $labl$$label;
12116 $$$emit8$primary;
12117 emit_cc(cbuf, $secondary, Assembler::parity);
12118 int parity_disp = -1;
12119 if ($cop$$cmpcode == Assembler::notEqual) {
12120 // the two jumps 6 bytes apart so the jump distances are too
12121 parity_disp = l->loc_pos() - (cbuf.insts_size() + 4);
12122 } else if ($cop$$cmpcode == Assembler::equal) {
12123 parity_disp = 6;
12124 } else {
12125 ShouldNotReachHere();
12126 }
12127 emit_d32(cbuf, parity_disp);
12128 $$$emit8$primary;
12129 emit_cc(cbuf, $secondary, $cop$$cmpcode);
12130 int disp = l->loc_pos() - (cbuf.insts_size() + 4);
12131 emit_d32(cbuf, disp);
12132 %}
12133 ins_pipe(pipe_jcc);
12134 %}
12135
12136 // ============================================================================
12137 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
12138 // superklass array for an instance of the superklass. Set a hidden
12139 // internal cache on a hit (cache is checked with exposed code in
12140 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
12141 // encoding ALSO sets flags.
12142
12143 instruct partialSubtypeCheck(rdi_RegP result,
12144 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12145 rFlagsReg cr)
12146 %{
12147 match(Set result (PartialSubtypeCheck sub super));
12148 effect(KILL rcx, KILL cr);
12149
12150 ins_cost(1100); // slightly larger than the next version
12151 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12152 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12153 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12154 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
12155 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
12156 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12157 "xorq $result, $result\t\t Hit: rdi zero\n\t"
12158 "miss:\t" %}
12159
12160 opcode(0x1); // Force a XOR of RDI
12161 ins_encode(enc_PartialSubtypeCheck());
12162 ins_pipe(pipe_slow);
12163 %}
12164
12165 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
12166 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12167 immP0 zero,
12168 rdi_RegP result)
12169 %{
12170 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12171 effect(KILL rcx, KILL result);
12172
12173 ins_cost(1000);
12174 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12175 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12176 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12177 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
12178 "jne,s miss\t\t# Missed: flags nz\n\t"
12179 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12180 "miss:\t" %}
12181
12182 opcode(0x0); // No need to XOR RDI
12183 ins_encode(enc_PartialSubtypeCheck());
12184 ins_pipe(pipe_slow);
12185 %}
12186
12187 // ============================================================================
12188 // Branch Instructions -- short offset versions
12189 //
12190 // These instructions are used to replace jumps of a long offset (the default
12191 // match) with jumps of a shorter offset. These instructions are all tagged
12192 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12193 // match rules in general matching. Instead, the ADLC generates a conversion
12194 // method in the MachNode which can be used to do in-place replacement of the
12195 // long variant with the shorter variant. The compiler will determine if a
12196 // branch can be taken by the is_short_branch_offset() predicate in the machine
12197 // specific code section of the file.
12198
12199 // Jump Direct - Label defines a relative address from JMP+1
12200 instruct jmpDir_short(label labl) %{
12201 match(Goto);
12202 effect(USE labl);
12203
12204 ins_cost(300);
12205 format %{ "jmp,s $labl" %}
12206 size(2);
12207 opcode(0xEB);
12208 ins_encode(OpcP, LblShort(labl));
12209 ins_pipe(pipe_jmp);
12210 ins_short_branch(1);
12211 %}
12212
12213 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12214 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
12215 match(If cop cr);
12216 effect(USE labl);
12217
12218 ins_cost(300);
12219 format %{ "j$cop,s $labl" %}
12220 size(2);
12221 opcode(0x70);
12222 ins_encode(JccShort(cop, labl));
12223 ins_pipe(pipe_jcc);
12224 ins_short_branch(1);
12225 %}
12226
12227 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12228 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
12229 match(CountedLoopEnd cop cr);
12230 effect(USE labl);
12231
12232 ins_cost(300);
12233 format %{ "j$cop,s $labl\t# loop end" %}
12234 size(2);
12235 opcode(0x70);
12236 ins_encode(JccShort(cop, labl));
12237 ins_pipe(pipe_jcc);
12238 ins_short_branch(1);
12239 %}
12240
12241 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12242 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12243 match(CountedLoopEnd cop cmp);
12244 effect(USE labl);
12245
12246 ins_cost(300);
12247 format %{ "j$cop,us $labl\t# loop end" %}
12248 size(2);
12249 opcode(0x70);
12250 ins_encode(JccShort(cop, labl));
12251 ins_pipe(pipe_jcc);
12252 ins_short_branch(1);
12253 %}
12254
12255 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12256 match(CountedLoopEnd cop cmp);
12257 effect(USE labl);
12258
12259 ins_cost(300);
12260 format %{ "j$cop,us $labl\t# loop end" %}
12261 size(2);
12262 opcode(0x70);
12263 ins_encode(JccShort(cop, labl));
12264 ins_pipe(pipe_jcc);
12265 ins_short_branch(1);
12266 %}
12267
12268 // Jump Direct Conditional - using unsigned comparison
12269 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12270 match(If cop cmp);
12271 effect(USE labl);
12272
12273 ins_cost(300);
12274 format %{ "j$cop,us $labl" %}
12275 size(2);
12276 opcode(0x70);
12277 ins_encode(JccShort(cop, labl));
12278 ins_pipe(pipe_jcc);
12279 ins_short_branch(1);
12280 %}
12281
12282 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12283 match(If cop cmp);
12284 effect(USE labl);
12285
12286 ins_cost(300);
12287 format %{ "j$cop,us $labl" %}
12288 size(2);
12289 opcode(0x70);
12290 ins_encode(JccShort(cop, labl));
12291 ins_pipe(pipe_jcc);
12292 ins_short_branch(1);
12293 %}
12294
12295 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12296 match(If cop cmp);
12297 effect(USE labl);
12298
12299 ins_cost(300);
12300 format %{ $$template
12301 if ($cop$$cmpcode == Assembler::notEqual) {
12302 $$emit$$"jp,u,s $labl\n\t"
12303 $$emit$$"j$cop,u,s $labl"
12304 } else {
12305 $$emit$$"jp,u,s done\n\t"
12306 $$emit$$"j$cop,u,s $labl\n\t"
12307 $$emit$$"done:"
12308 }
12309 %}
12310 size(4);
12311 opcode(0x70);
12312 ins_encode %{
12313 Label* l = $labl$$label;
12314 emit_cc(cbuf, $primary, Assembler::parity);
12315 int parity_disp = -1;
12316 if ($cop$$cmpcode == Assembler::notEqual) {
12317 parity_disp = l->loc_pos() - (cbuf.insts_size() + 1);
12318 } else if ($cop$$cmpcode == Assembler::equal) {
12319 parity_disp = 2;
12320 } else {
12321 ShouldNotReachHere();
12322 }
12323 emit_d8(cbuf, parity_disp);
12324 emit_cc(cbuf, $primary, $cop$$cmpcode);
12325 int disp = l->loc_pos() - (cbuf.insts_size() + 1);
12326 emit_d8(cbuf, disp);
12327 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
12328 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
12329 %}
12330 ins_pipe(pipe_jcc);
12331 ins_short_branch(1);
12332 %}
12333
12334 // ============================================================================
12335 // inlined locking and unlocking
12336
12337 instruct cmpFastLock(rFlagsReg cr,
12338 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
12339 %{
12340 match(Set cr (FastLock object box));
12341 effect(TEMP tmp, TEMP scr);
12342
12343 ins_cost(300);
12344 format %{ "fastlock $object,$box,$tmp,$scr" %}
12345 ins_encode(Fast_Lock(object, box, tmp, scr));
12346 ins_pipe(pipe_slow);
12347 %}
12348
12349 instruct cmpFastUnlock(rFlagsReg cr,
12350 rRegP object, rax_RegP box, rRegP tmp)
12351 %{
12352 match(Set cr (FastUnlock object box));
12353 effect(TEMP tmp);
12354
12355 ins_cost(300);
12356 format %{ "fastunlock $object, $box, $tmp" %}
12357 ins_encode(Fast_Unlock(object, box, tmp));
12358 ins_pipe(pipe_slow);
12359 %}
12360
12361
12362 // ============================================================================
12363 // Safepoint Instructions
12364 instruct safePoint_poll(rFlagsReg cr)
12365 %{
12366 predicate(!Assembler::is_polling_page_far());
12367 match(SafePoint);
12368 effect(KILL cr);
12369
12370 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
12371 "# Safepoint: poll for GC" %}
12372 ins_cost(125);
12373 ins_encode %{
12374 AddressLiteral addr(os::get_polling_page(), relocInfo::poll_type);
12375 __ testl(rax, addr);
12376 %}
12377 ins_pipe(ialu_reg_mem);
12378 %}
12379
12380 instruct safePoint_poll_far(rFlagsReg cr, rRegP poll)
12381 %{
12382 predicate(Assembler::is_polling_page_far());
12383 match(SafePoint poll);
12384 effect(KILL cr, USE poll);
12385
12386 format %{ "testl rax, [$poll]\t"
12387 "# Safepoint: poll for GC" %}
12388 ins_cost(125);
12389 ins_encode %{
12390 __ relocate(relocInfo::poll_type);
12391 __ testl(rax, Address($poll$$Register, 0));
12392 %}
12393 ins_pipe(ialu_reg_mem);
12394 %}
12395
12396 // ============================================================================
12397 // Procedure Call/Return Instructions
12398 // Call Java Static Instruction
12399 // Note: If this code changes, the corresponding ret_addr_offset() and
12400 // compute_padding() functions will have to be adjusted.
12401 instruct CallStaticJavaDirect(method meth) %{
12402 match(CallStaticJava);
12403 predicate(!((CallStaticJavaNode*) n)->is_method_handle_invoke());
12404 effect(USE meth);
12405
12406 ins_cost(300);
12407 format %{ "call,static " %}
12408 opcode(0xE8); /* E8 cd */
12409 ins_encode(Java_Static_Call(meth), call_epilog);
12410 ins_pipe(pipe_slow);
12411 ins_alignment(4);
12412 %}
12413
12414 // Call Java Static Instruction (method handle version)
12415 // Note: If this code changes, the corresponding ret_addr_offset() and
12416 // compute_padding() functions will have to be adjusted.
12417 instruct CallStaticJavaHandle(method meth, rbp_RegP rbp_mh_SP_save) %{
12418 match(CallStaticJava);
12419 predicate(((CallStaticJavaNode*) n)->is_method_handle_invoke());
12420 effect(USE meth);
12421 // RBP is saved by all callees (for interpreter stack correction).
12422 // We use it here for a similar purpose, in {preserve,restore}_SP.
12423
12424 ins_cost(300);
12425 format %{ "call,static/MethodHandle " %}
12426 opcode(0xE8); /* E8 cd */
12427 ins_encode(preserve_SP,
12428 Java_Static_Call(meth),
12429 restore_SP,
12430 call_epilog);
12431 ins_pipe(pipe_slow);
12432 ins_alignment(4);
12433 %}
12434
12435 // Call Java Dynamic Instruction
12436 // Note: If this code changes, the corresponding ret_addr_offset() and
12437 // compute_padding() functions will have to be adjusted.
12438 instruct CallDynamicJavaDirect(method meth)
12439 %{
12440 match(CallDynamicJava);
12441 effect(USE meth);
12442
12443 ins_cost(300);
12444 format %{ "movq rax, #Universe::non_oop_word()\n\t"
12445 "call,dynamic " %}
12446 opcode(0xE8); /* E8 cd */
12447 ins_encode(Java_Dynamic_Call(meth), call_epilog);
12448 ins_pipe(pipe_slow);
12449 ins_alignment(4);
12450 %}
12451
12452 // Call Runtime Instruction
12453 instruct CallRuntimeDirect(method meth)
12454 %{
12455 match(CallRuntime);
12456 effect(USE meth);
12457
12458 ins_cost(300);
12459 format %{ "call,runtime " %}
12460 opcode(0xE8); /* E8 cd */
12461 ins_encode(Java_To_Runtime(meth));
12462 ins_pipe(pipe_slow);
12463 %}
12464
12465 // Call runtime without safepoint
12466 instruct CallLeafDirect(method meth)
12467 %{
12468 match(CallLeaf);
12469 effect(USE meth);
12470
12471 ins_cost(300);
12472 format %{ "call_leaf,runtime " %}
12473 opcode(0xE8); /* E8 cd */
12474 ins_encode(Java_To_Runtime(meth));
12475 ins_pipe(pipe_slow);
12476 %}
12477
12478 // Call runtime without safepoint
12479 instruct CallLeafNoFPDirect(method meth)
12480 %{
12481 match(CallLeafNoFP);
12482 effect(USE meth);
12483
12484 ins_cost(300);
12485 format %{ "call_leaf_nofp,runtime " %}
12486 opcode(0xE8); /* E8 cd */
12487 ins_encode(Java_To_Runtime(meth));
12488 ins_pipe(pipe_slow);
12489 %}
12490
12491 // Return Instruction
12492 // Remove the return address & jump to it.
12493 // Notice: We always emit a nop after a ret to make sure there is room
12494 // for safepoint patching
12495 instruct Ret()
12496 %{
12497 match(Return);
12498
12499 format %{ "ret" %}
12500 opcode(0xC3);
12501 ins_encode(OpcP);
12502 ins_pipe(pipe_jmp);
12503 %}
12504
12505 // Tail Call; Jump from runtime stub to Java code.
12506 // Also known as an 'interprocedural jump'.
12507 // Target of jump will eventually return to caller.
12508 // TailJump below removes the return address.
12509 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12510 %{
12511 match(TailCall jump_target method_oop);
12512
12513 ins_cost(300);
12514 format %{ "jmp $jump_target\t# rbx holds method oop" %}
12515 opcode(0xFF, 0x4); /* Opcode FF /4 */
12516 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12517 ins_pipe(pipe_jmp);
12518 %}
12519
12520 // Tail Jump; remove the return address; jump to target.
12521 // TailCall above leaves the return address around.
12522 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12523 %{
12524 match(TailJump jump_target ex_oop);
12525
12526 ins_cost(300);
12527 format %{ "popq rdx\t# pop return address\n\t"
12528 "jmp $jump_target" %}
12529 opcode(0xFF, 0x4); /* Opcode FF /4 */
12530 ins_encode(Opcode(0x5a), // popq rdx
12531 REX_reg(jump_target), OpcP, reg_opc(jump_target));
12532 ins_pipe(pipe_jmp);
12533 %}
12534
12535 // Create exception oop: created by stack-crawling runtime code.
12536 // Created exception is now available to this handler, and is setup
12537 // just prior to jumping to this handler. No code emitted.
12538 instruct CreateException(rax_RegP ex_oop)
12539 %{
12540 match(Set ex_oop (CreateEx));
12541
12542 size(0);
12543 // use the following format syntax
12544 format %{ "# exception oop is in rax; no code emitted" %}
12545 ins_encode();
12546 ins_pipe(empty);
12547 %}
12548
12549 // Rethrow exception:
12550 // The exception oop will come in the first argument position.
12551 // Then JUMP (not call) to the rethrow stub code.
12552 instruct RethrowException()
12553 %{
12554 match(Rethrow);
12555
12556 // use the following format syntax
12557 format %{ "jmp rethrow_stub" %}
12558 ins_encode(enc_rethrow);
12559 ins_pipe(pipe_jmp);
12560 %}
12561
12562
12563 //----------PEEPHOLE RULES-----------------------------------------------------
12564 // These must follow all instruction definitions as they use the names
12565 // defined in the instructions definitions.
12566 //
12567 // peepmatch ( root_instr_name [preceding_instruction]* );
12568 //
12569 // peepconstraint %{
12570 // (instruction_number.operand_name relational_op instruction_number.operand_name
12571 // [, ...] );
12572 // // instruction numbers are zero-based using left to right order in peepmatch
12573 //
12574 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12575 // // provide an instruction_number.operand_name for each operand that appears
12576 // // in the replacement instruction's match rule
12577 //
12578 // ---------VM FLAGS---------------------------------------------------------
12579 //
12580 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12581 //
12582 // Each peephole rule is given an identifying number starting with zero and
12583 // increasing by one in the order seen by the parser. An individual peephole
12584 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12585 // on the command-line.
12586 //
12587 // ---------CURRENT LIMITATIONS----------------------------------------------
12588 //
12589 // Only match adjacent instructions in same basic block
12590 // Only equality constraints
12591 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12592 // Only one replacement instruction
12593 //
12594 // ---------EXAMPLE----------------------------------------------------------
12595 //
12596 // // pertinent parts of existing instructions in architecture description
12597 // instruct movI(rRegI dst, rRegI src)
12598 // %{
12599 // match(Set dst (CopyI src));
12600 // %}
12601 //
12602 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12603 // %{
12604 // match(Set dst (AddI dst src));
12605 // effect(KILL cr);
12606 // %}
12607 //
12608 // // Change (inc mov) to lea
12609 // peephole %{
12610 // // increment preceeded by register-register move
12611 // peepmatch ( incI_rReg movI );
12612 // // require that the destination register of the increment
12613 // // match the destination register of the move
12614 // peepconstraint ( 0.dst == 1.dst );
12615 // // construct a replacement instruction that sets
12616 // // the destination to ( move's source register + one )
12617 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12618 // %}
12619 //
12620
12621 // Implementation no longer uses movX instructions since
12622 // machine-independent system no longer uses CopyX nodes.
12623 //
12624 // peephole
12625 // %{
12626 // peepmatch (incI_rReg movI);
12627 // peepconstraint (0.dst == 1.dst);
12628 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12629 // %}
12630
12631 // peephole
12632 // %{
12633 // peepmatch (decI_rReg movI);
12634 // peepconstraint (0.dst == 1.dst);
12635 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12636 // %}
12637
12638 // peephole
12639 // %{
12640 // peepmatch (addI_rReg_imm movI);
12641 // peepconstraint (0.dst == 1.dst);
12642 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12643 // %}
12644
12645 // peephole
12646 // %{
12647 // peepmatch (incL_rReg movL);
12648 // peepconstraint (0.dst == 1.dst);
12649 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12650 // %}
12651
12652 // peephole
12653 // %{
12654 // peepmatch (decL_rReg movL);
12655 // peepconstraint (0.dst == 1.dst);
12656 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12657 // %}
12658
12659 // peephole
12660 // %{
12661 // peepmatch (addL_rReg_imm movL);
12662 // peepconstraint (0.dst == 1.dst);
12663 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12664 // %}
12665
12666 // peephole
12667 // %{
12668 // peepmatch (addP_rReg_imm movP);
12669 // peepconstraint (0.dst == 1.dst);
12670 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12671 // %}
12672
12673 // // Change load of spilled value to only a spill
12674 // instruct storeI(memory mem, rRegI src)
12675 // %{
12676 // match(Set mem (StoreI mem src));
12677 // %}
12678 //
12679 // instruct loadI(rRegI dst, memory mem)
12680 // %{
12681 // match(Set dst (LoadI mem));
12682 // %}
12683 //
12684
12685 peephole
12686 %{
12687 peepmatch (loadI storeI);
12688 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12689 peepreplace (storeI(1.mem 1.mem 1.src));
12690 %}
12691
12692 peephole
12693 %{
12694 peepmatch (loadL storeL);
12695 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12696 peepreplace (storeL(1.mem 1.mem 1.src));
12697 %}
12698
12699 //----------SMARTSPILL RULES---------------------------------------------------
12700 // These must follow all instruction definitions as they use the names
12701 // defined in the instructions definitions.