Thu, 19 Mar 2009 09:13:24 -0700
Merge
1 //
2 // Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved.
3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 //
5 // This code is free software; you can redistribute it and/or modify it
6 // under the terms of the GNU General Public License version 2 only, as
7 // published by the Free Software Foundation.
8 //
9 // This code is distributed in the hope that it will be useful, but WITHOUT
10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 // version 2 for more details (a copy is included in the LICENSE file that
13 // accompanied this code).
14 //
15 // You should have received a copy of the GNU General Public License version
16 // 2 along with this work; if not, write to the Free Software Foundation,
17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 //
19 // Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 // CA 95054 USA or visit www.sun.com if you need additional information or
21 // have any questions.
22 //
23 //
25 // AMD64 Architecture Description File
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.
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.
62 // General Registers
63 // R8-R15 must be encoded with REX. (RSP, RBP, RSI, RDI need REX when
64 // used as byte registers)
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.
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());
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());
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());
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());
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());
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());
89 #ifdef _WIN64
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());
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());
97 #else
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());
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());
105 #endif
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());
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());
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());
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());
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());
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());
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());
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());
132 // Floating Point Registers
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
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());
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());
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());
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());
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());
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());
162 #ifdef _WIN64
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());
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());
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());
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());
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());
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());
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());
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());
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());
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());
194 #else
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());
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());
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());
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());
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());
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());
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());
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());
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());
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());
226 #endif // _WIN64
228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
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.
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);
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);
273 alloc_class chunk2(RFLAGS);
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 //
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);
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);
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);
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);
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);
358 // Singleton class for RAX pointer register
359 reg_class ptr_rax_reg(RAX, RAX_H);
361 // Singleton class for RBX pointer register
362 reg_class ptr_rbx_reg(RBX, RBX_H);
364 // Singleton class for RSI pointer register
365 reg_class ptr_rsi_reg(RSI, RSI_H);
367 // Singleton class for RDI pointer register
368 reg_class ptr_rdi_reg(RDI, RDI_H);
370 // Singleton class for RBP pointer register
371 reg_class ptr_rbp_reg(RBP, RBP_H);
373 // Singleton class for stack pointer
374 reg_class ptr_rsp_reg(RSP, RSP_H);
376 // Singleton class for TLS pointer
377 reg_class ptr_r15_reg(R15, R15_H);
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);
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);
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);
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);
435 // Singleton class for RAX long register
436 reg_class long_rax_reg(RAX, RAX_H);
438 // Singleton class for RCX long register
439 reg_class long_rcx_reg(RCX, RCX_H);
441 // Singleton class for RDX long register
442 reg_class long_rdx_reg(RDX, RDX_H);
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);
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);
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);
486 // Singleton class for RAX int register
487 reg_class int_rax_reg(RAX);
489 // Singleton class for RBX int register
490 reg_class int_rbx_reg(RBX);
492 // Singleton class for RCX int register
493 reg_class int_rcx_reg(RCX);
495 // Singleton class for RCX int register
496 reg_class int_rdx_reg(RDX);
498 // Singleton class for RCX int register
499 reg_class int_rdi_reg(RDI);
501 // Singleton class for instruction pointer
502 // reg_class ip_reg(RIP);
504 // Singleton class for condition codes
505 reg_class int_flags(RFLAGS);
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);
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 %}
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
552 #define __ _masm.
554 // !!!!! Special hack to get all types of calls to specify the byte offset
555 // from the start of the call to the point where the return address
556 // will point.
557 int MachCallStaticJavaNode::ret_addr_offset()
558 {
559 return 5; // 5 bytes from start of call to where return address points
560 }
562 int MachCallDynamicJavaNode::ret_addr_offset()
563 {
564 return 15; // 15 bytes from start of call to where return address points
565 }
567 // In os_cpu .ad file
568 // int MachCallRuntimeNode::ret_addr_offset()
570 // Indicate if the safepoint node needs the polling page as an input.
571 // Since amd64 does not have absolute addressing but RIP-relative
572 // addressing and the polling page is within 2G, it doesn't.
573 bool SafePointNode::needs_polling_address_input()
574 {
575 return false;
576 }
578 //
579 // Compute padding required for nodes which need alignment
580 //
582 // The address of the call instruction needs to be 4-byte aligned to
583 // ensure that it does not span a cache line so that it can be patched.
584 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
585 {
586 current_offset += 1; // skip call opcode byte
587 return round_to(current_offset, alignment_required()) - current_offset;
588 }
590 // The address of the call instruction needs to be 4-byte aligned to
591 // ensure that it does not span a cache line so that it can be patched.
592 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
593 {
594 current_offset += 11; // skip movq instruction + call opcode byte
595 return round_to(current_offset, alignment_required()) - current_offset;
596 }
598 #ifndef PRODUCT
599 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
600 {
601 st->print("INT3");
602 }
603 #endif
605 // EMIT_RM()
606 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3)
607 {
608 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
609 *(cbuf.code_end()) = c;
610 cbuf.set_code_end(cbuf.code_end() + 1);
611 }
613 // EMIT_CC()
614 void emit_cc(CodeBuffer &cbuf, int f1, int f2)
615 {
616 unsigned char c = (unsigned char) (f1 | f2);
617 *(cbuf.code_end()) = c;
618 cbuf.set_code_end(cbuf.code_end() + 1);
619 }
621 // EMIT_OPCODE()
622 void emit_opcode(CodeBuffer &cbuf, int code)
623 {
624 *(cbuf.code_end()) = (unsigned char) code;
625 cbuf.set_code_end(cbuf.code_end() + 1);
626 }
628 // EMIT_OPCODE() w/ relocation information
629 void emit_opcode(CodeBuffer &cbuf,
630 int code, relocInfo::relocType reloc, int offset, int format)
631 {
632 cbuf.relocate(cbuf.inst_mark() + offset, reloc, format);
633 emit_opcode(cbuf, code);
634 }
636 // EMIT_D8()
637 void emit_d8(CodeBuffer &cbuf, int d8)
638 {
639 *(cbuf.code_end()) = (unsigned char) d8;
640 cbuf.set_code_end(cbuf.code_end() + 1);
641 }
643 // EMIT_D16()
644 void emit_d16(CodeBuffer &cbuf, int d16)
645 {
646 *((short *)(cbuf.code_end())) = d16;
647 cbuf.set_code_end(cbuf.code_end() + 2);
648 }
650 // EMIT_D32()
651 void emit_d32(CodeBuffer &cbuf, int d32)
652 {
653 *((int *)(cbuf.code_end())) = d32;
654 cbuf.set_code_end(cbuf.code_end() + 4);
655 }
657 // EMIT_D64()
658 void emit_d64(CodeBuffer &cbuf, int64_t d64)
659 {
660 *((int64_t*) (cbuf.code_end())) = d64;
661 cbuf.set_code_end(cbuf.code_end() + 8);
662 }
664 // emit 32 bit value and construct relocation entry from relocInfo::relocType
665 void emit_d32_reloc(CodeBuffer& cbuf,
666 int d32,
667 relocInfo::relocType reloc,
668 int format)
669 {
670 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
671 cbuf.relocate(cbuf.inst_mark(), reloc, format);
673 *((int*) (cbuf.code_end())) = d32;
674 cbuf.set_code_end(cbuf.code_end() + 4);
675 }
677 // emit 32 bit value and construct relocation entry from RelocationHolder
678 void emit_d32_reloc(CodeBuffer& cbuf,
679 int d32,
680 RelocationHolder const& rspec,
681 int format)
682 {
683 #ifdef ASSERT
684 if (rspec.reloc()->type() == relocInfo::oop_type &&
685 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
686 assert(oop((intptr_t)d32)->is_oop() && oop((intptr_t)d32)->is_perm(), "cannot embed non-perm oops in code");
687 }
688 #endif
689 cbuf.relocate(cbuf.inst_mark(), rspec, format);
691 *((int* )(cbuf.code_end())) = d32;
692 cbuf.set_code_end(cbuf.code_end() + 4);
693 }
695 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
696 address next_ip = cbuf.code_end() + 4;
697 emit_d32_reloc(cbuf, (int) (addr - next_ip),
698 external_word_Relocation::spec(addr),
699 RELOC_DISP32);
700 }
703 // emit 64 bit value and construct relocation entry from relocInfo::relocType
704 void emit_d64_reloc(CodeBuffer& cbuf,
705 int64_t d64,
706 relocInfo::relocType reloc,
707 int format)
708 {
709 cbuf.relocate(cbuf.inst_mark(), reloc, format);
711 *((int64_t*) (cbuf.code_end())) = d64;
712 cbuf.set_code_end(cbuf.code_end() + 8);
713 }
715 // emit 64 bit value and construct relocation entry from RelocationHolder
716 void emit_d64_reloc(CodeBuffer& cbuf,
717 int64_t d64,
718 RelocationHolder const& rspec,
719 int format)
720 {
721 #ifdef ASSERT
722 if (rspec.reloc()->type() == relocInfo::oop_type &&
723 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
724 assert(oop(d64)->is_oop() && oop(d64)->is_perm(),
725 "cannot embed non-perm oops in code");
726 }
727 #endif
728 cbuf.relocate(cbuf.inst_mark(), rspec, format);
730 *((int64_t*) (cbuf.code_end())) = d64;
731 cbuf.set_code_end(cbuf.code_end() + 8);
732 }
734 // Access stack slot for load or store
735 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
736 {
737 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src])
738 if (-0x80 <= disp && disp < 0x80) {
739 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte
740 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
741 emit_d8(cbuf, disp); // Displacement // R/M byte
742 } else {
743 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte
744 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
745 emit_d32(cbuf, disp); // Displacement // R/M byte
746 }
747 }
749 // rRegI ereg, memory mem) %{ // emit_reg_mem
750 void encode_RegMem(CodeBuffer &cbuf,
751 int reg,
752 int base, int index, int scale, int disp, bool disp_is_oop)
753 {
754 assert(!disp_is_oop, "cannot have disp");
755 int regenc = reg & 7;
756 int baseenc = base & 7;
757 int indexenc = index & 7;
759 // There is no index & no scale, use form without SIB byte
760 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
761 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
762 if (disp == 0 && base != RBP_enc && base != R13_enc) {
763 emit_rm(cbuf, 0x0, regenc, baseenc); // *
764 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
765 // If 8-bit displacement, mode 0x1
766 emit_rm(cbuf, 0x1, regenc, baseenc); // *
767 emit_d8(cbuf, disp);
768 } else {
769 // If 32-bit displacement
770 if (base == -1) { // Special flag for absolute address
771 emit_rm(cbuf, 0x0, regenc, 0x5); // *
772 if (disp_is_oop) {
773 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
774 } else {
775 emit_d32(cbuf, disp);
776 }
777 } else {
778 // Normal base + offset
779 emit_rm(cbuf, 0x2, regenc, baseenc); // *
780 if (disp_is_oop) {
781 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
782 } else {
783 emit_d32(cbuf, disp);
784 }
785 }
786 }
787 } else {
788 // Else, encode with the SIB byte
789 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
790 if (disp == 0 && base != RBP_enc && base != R13_enc) {
791 // If no displacement
792 emit_rm(cbuf, 0x0, regenc, 0x4); // *
793 emit_rm(cbuf, scale, indexenc, baseenc);
794 } else {
795 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
796 // If 8-bit displacement, mode 0x1
797 emit_rm(cbuf, 0x1, regenc, 0x4); // *
798 emit_rm(cbuf, scale, indexenc, baseenc);
799 emit_d8(cbuf, disp);
800 } else {
801 // If 32-bit displacement
802 if (base == 0x04 ) {
803 emit_rm(cbuf, 0x2, regenc, 0x4);
804 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
805 } else {
806 emit_rm(cbuf, 0x2, regenc, 0x4);
807 emit_rm(cbuf, scale, indexenc, baseenc); // *
808 }
809 if (disp_is_oop) {
810 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
811 } else {
812 emit_d32(cbuf, disp);
813 }
814 }
815 }
816 }
817 }
819 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
820 {
821 if (dstenc != srcenc) {
822 if (dstenc < 8) {
823 if (srcenc >= 8) {
824 emit_opcode(cbuf, Assembler::REX_B);
825 srcenc -= 8;
826 }
827 } else {
828 if (srcenc < 8) {
829 emit_opcode(cbuf, Assembler::REX_R);
830 } else {
831 emit_opcode(cbuf, Assembler::REX_RB);
832 srcenc -= 8;
833 }
834 dstenc -= 8;
835 }
837 emit_opcode(cbuf, 0x8B);
838 emit_rm(cbuf, 0x3, dstenc, srcenc);
839 }
840 }
842 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
843 if( dst_encoding == src_encoding ) {
844 // reg-reg copy, use an empty encoding
845 } else {
846 MacroAssembler _masm(&cbuf);
848 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
849 }
850 }
853 //=============================================================================
854 #ifndef PRODUCT
855 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
856 {
857 Compile* C = ra_->C;
859 int framesize = C->frame_slots() << LogBytesPerInt;
860 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
861 // Remove wordSize for return adr already pushed
862 // and another for the RBP we are going to save
863 framesize -= 2*wordSize;
864 bool need_nop = true;
866 // Calls to C2R adapters often do not accept exceptional returns.
867 // We require that their callers must bang for them. But be
868 // careful, because some VM calls (such as call site linkage) can
869 // use several kilobytes of stack. But the stack safety zone should
870 // account for that. See bugs 4446381, 4468289, 4497237.
871 if (C->need_stack_bang(framesize)) {
872 st->print_cr("# stack bang"); st->print("\t");
873 need_nop = false;
874 }
875 st->print_cr("pushq rbp"); st->print("\t");
877 if (VerifyStackAtCalls) {
878 // Majik cookie to verify stack depth
879 st->print_cr("pushq 0xffffffffbadb100d"
880 "\t# Majik cookie for stack depth check");
881 st->print("\t");
882 framesize -= wordSize; // Remove 2 for cookie
883 need_nop = false;
884 }
886 if (framesize) {
887 st->print("subq rsp, #%d\t# Create frame", framesize);
888 if (framesize < 0x80 && need_nop) {
889 st->print("\n\tnop\t# nop for patch_verified_entry");
890 }
891 }
892 }
893 #endif
895 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
896 {
897 Compile* C = ra_->C;
899 // WARNING: Initial instruction MUST be 5 bytes or longer so that
900 // NativeJump::patch_verified_entry will be able to patch out the entry
901 // code safely. The fldcw is ok at 6 bytes, the push to verify stack
902 // depth is ok at 5 bytes, the frame allocation can be either 3 or
903 // 6 bytes. So if we don't do the fldcw or the push then we must
904 // use the 6 byte frame allocation even if we have no frame. :-(
905 // If method sets FPU control word do it now
907 int framesize = C->frame_slots() << LogBytesPerInt;
908 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
909 // Remove wordSize for return adr already pushed
910 // and another for the RBP we are going to save
911 framesize -= 2*wordSize;
912 bool need_nop = true;
914 // Calls to C2R adapters often do not accept exceptional returns.
915 // We require that their callers must bang for them. But be
916 // careful, because some VM calls (such as call site linkage) can
917 // use several kilobytes of stack. But the stack safety zone should
918 // account for that. See bugs 4446381, 4468289, 4497237.
919 if (C->need_stack_bang(framesize)) {
920 MacroAssembler masm(&cbuf);
921 masm.generate_stack_overflow_check(framesize);
922 need_nop = false;
923 }
925 // We always push rbp so that on return to interpreter rbp will be
926 // restored correctly and we can correct the stack.
927 emit_opcode(cbuf, 0x50 | RBP_enc);
929 if (VerifyStackAtCalls) {
930 // Majik cookie to verify stack depth
931 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
932 emit_d32(cbuf, 0xbadb100d);
933 framesize -= wordSize; // Remove 2 for cookie
934 need_nop = false;
935 }
937 if (framesize) {
938 emit_opcode(cbuf, Assembler::REX_W);
939 if (framesize < 0x80) {
940 emit_opcode(cbuf, 0x83); // sub SP,#framesize
941 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
942 emit_d8(cbuf, framesize);
943 if (need_nop) {
944 emit_opcode(cbuf, 0x90); // nop
945 }
946 } else {
947 emit_opcode(cbuf, 0x81); // sub SP,#framesize
948 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
949 emit_d32(cbuf, framesize);
950 }
951 }
953 C->set_frame_complete(cbuf.code_end() - cbuf.code_begin());
955 #ifdef ASSERT
956 if (VerifyStackAtCalls) {
957 Label L;
958 MacroAssembler masm(&cbuf);
959 masm.push(rax);
960 masm.mov(rax, rsp);
961 masm.andptr(rax, StackAlignmentInBytes-1);
962 masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
963 masm.pop(rax);
964 masm.jcc(Assembler::equal, L);
965 masm.stop("Stack is not properly aligned!");
966 masm.bind(L);
967 }
968 #endif
969 }
971 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
972 {
973 return MachNode::size(ra_); // too many variables; just compute it
974 // the hard way
975 }
977 int MachPrologNode::reloc() const
978 {
979 return 0; // a large enough number
980 }
982 //=============================================================================
983 #ifndef PRODUCT
984 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
985 {
986 Compile* C = ra_->C;
987 int framesize = C->frame_slots() << LogBytesPerInt;
988 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
989 // Remove word for return adr already pushed
990 // and RBP
991 framesize -= 2*wordSize;
993 if (framesize) {
994 st->print_cr("addq\trsp, %d\t# Destroy frame", framesize);
995 st->print("\t");
996 }
998 st->print_cr("popq\trbp");
999 if (do_polling() && C->is_method_compilation()) {
1000 st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t"
1001 "# Safepoint: poll for GC");
1002 st->print("\t");
1003 }
1004 }
1005 #endif
1007 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1008 {
1009 Compile* C = ra_->C;
1010 int framesize = C->frame_slots() << LogBytesPerInt;
1011 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1012 // Remove word for return adr already pushed
1013 // and RBP
1014 framesize -= 2*wordSize;
1016 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1018 if (framesize) {
1019 emit_opcode(cbuf, Assembler::REX_W);
1020 if (framesize < 0x80) {
1021 emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1022 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1023 emit_d8(cbuf, framesize);
1024 } else {
1025 emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1026 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1027 emit_d32(cbuf, framesize);
1028 }
1029 }
1031 // popq rbp
1032 emit_opcode(cbuf, 0x58 | RBP_enc);
1034 if (do_polling() && C->is_method_compilation()) {
1035 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
1036 // XXX reg_mem doesn't support RIP-relative addressing yet
1037 cbuf.set_inst_mark();
1038 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX
1039 emit_opcode(cbuf, 0x85); // testl
1040 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
1041 // cbuf.inst_mark() is beginning of instruction
1042 emit_d32_reloc(cbuf, os::get_polling_page());
1043 // relocInfo::poll_return_type,
1044 }
1045 }
1047 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1048 {
1049 Compile* C = ra_->C;
1050 int framesize = C->frame_slots() << LogBytesPerInt;
1051 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1052 // Remove word for return adr already pushed
1053 // and RBP
1054 framesize -= 2*wordSize;
1056 uint size = 0;
1058 if (do_polling() && C->is_method_compilation()) {
1059 size += 6;
1060 }
1062 // count popq rbp
1063 size++;
1065 if (framesize) {
1066 if (framesize < 0x80) {
1067 size += 4;
1068 } else if (framesize) {
1069 size += 7;
1070 }
1071 }
1073 return size;
1074 }
1076 int MachEpilogNode::reloc() const
1077 {
1078 return 2; // a large enough number
1079 }
1081 const Pipeline* MachEpilogNode::pipeline() const
1082 {
1083 return MachNode::pipeline_class();
1084 }
1086 int MachEpilogNode::safepoint_offset() const
1087 {
1088 return 0;
1089 }
1091 //=============================================================================
1093 enum RC {
1094 rc_bad,
1095 rc_int,
1096 rc_float,
1097 rc_stack
1098 };
1100 static enum RC rc_class(OptoReg::Name reg)
1101 {
1102 if( !OptoReg::is_valid(reg) ) return rc_bad;
1104 if (OptoReg::is_stack(reg)) return rc_stack;
1106 VMReg r = OptoReg::as_VMReg(reg);
1108 if (r->is_Register()) return rc_int;
1110 assert(r->is_XMMRegister(), "must be");
1111 return rc_float;
1112 }
1114 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1115 PhaseRegAlloc* ra_,
1116 bool do_size,
1117 outputStream* st) const
1118 {
1120 // Get registers to move
1121 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1122 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1123 OptoReg::Name dst_second = ra_->get_reg_second(this);
1124 OptoReg::Name dst_first = ra_->get_reg_first(this);
1126 enum RC src_second_rc = rc_class(src_second);
1127 enum RC src_first_rc = rc_class(src_first);
1128 enum RC dst_second_rc = rc_class(dst_second);
1129 enum RC dst_first_rc = rc_class(dst_first);
1131 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1132 "must move at least 1 register" );
1134 if (src_first == dst_first && src_second == dst_second) {
1135 // Self copy, no move
1136 return 0;
1137 } else if (src_first_rc == rc_stack) {
1138 // mem ->
1139 if (dst_first_rc == rc_stack) {
1140 // mem -> mem
1141 assert(src_second != dst_first, "overlap");
1142 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1143 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1144 // 64-bit
1145 int src_offset = ra_->reg2offset(src_first);
1146 int dst_offset = ra_->reg2offset(dst_first);
1147 if (cbuf) {
1148 emit_opcode(*cbuf, 0xFF);
1149 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1151 emit_opcode(*cbuf, 0x8F);
1152 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1154 #ifndef PRODUCT
1155 } else if (!do_size) {
1156 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1157 "popq [rsp + #%d]",
1158 src_offset,
1159 dst_offset);
1160 #endif
1161 }
1162 return
1163 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1164 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1165 } else {
1166 // 32-bit
1167 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1168 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1169 // No pushl/popl, so:
1170 int src_offset = ra_->reg2offset(src_first);
1171 int dst_offset = ra_->reg2offset(dst_first);
1172 if (cbuf) {
1173 emit_opcode(*cbuf, Assembler::REX_W);
1174 emit_opcode(*cbuf, 0x89);
1175 emit_opcode(*cbuf, 0x44);
1176 emit_opcode(*cbuf, 0x24);
1177 emit_opcode(*cbuf, 0xF8);
1179 emit_opcode(*cbuf, 0x8B);
1180 encode_RegMem(*cbuf,
1181 RAX_enc,
1182 RSP_enc, 0x4, 0, src_offset,
1183 false);
1185 emit_opcode(*cbuf, 0x89);
1186 encode_RegMem(*cbuf,
1187 RAX_enc,
1188 RSP_enc, 0x4, 0, dst_offset,
1189 false);
1191 emit_opcode(*cbuf, Assembler::REX_W);
1192 emit_opcode(*cbuf, 0x8B);
1193 emit_opcode(*cbuf, 0x44);
1194 emit_opcode(*cbuf, 0x24);
1195 emit_opcode(*cbuf, 0xF8);
1197 #ifndef PRODUCT
1198 } else if (!do_size) {
1199 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1200 "movl rax, [rsp + #%d]\n\t"
1201 "movl [rsp + #%d], rax\n\t"
1202 "movq rax, [rsp - #8]",
1203 src_offset,
1204 dst_offset);
1205 #endif
1206 }
1207 return
1208 5 + // movq
1209 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1210 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1211 5; // movq
1212 }
1213 } else if (dst_first_rc == rc_int) {
1214 // mem -> gpr
1215 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1216 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1217 // 64-bit
1218 int offset = ra_->reg2offset(src_first);
1219 if (cbuf) {
1220 if (Matcher::_regEncode[dst_first] < 8) {
1221 emit_opcode(*cbuf, Assembler::REX_W);
1222 } else {
1223 emit_opcode(*cbuf, Assembler::REX_WR);
1224 }
1225 emit_opcode(*cbuf, 0x8B);
1226 encode_RegMem(*cbuf,
1227 Matcher::_regEncode[dst_first],
1228 RSP_enc, 0x4, 0, offset,
1229 false);
1230 #ifndef PRODUCT
1231 } else if (!do_size) {
1232 st->print("movq %s, [rsp + #%d]\t# spill",
1233 Matcher::regName[dst_first],
1234 offset);
1235 #endif
1236 }
1237 return
1238 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1239 } else {
1240 // 32-bit
1241 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1242 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1243 int offset = ra_->reg2offset(src_first);
1244 if (cbuf) {
1245 if (Matcher::_regEncode[dst_first] >= 8) {
1246 emit_opcode(*cbuf, Assembler::REX_R);
1247 }
1248 emit_opcode(*cbuf, 0x8B);
1249 encode_RegMem(*cbuf,
1250 Matcher::_regEncode[dst_first],
1251 RSP_enc, 0x4, 0, offset,
1252 false);
1253 #ifndef PRODUCT
1254 } else if (!do_size) {
1255 st->print("movl %s, [rsp + #%d]\t# spill",
1256 Matcher::regName[dst_first],
1257 offset);
1258 #endif
1259 }
1260 return
1261 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1262 ((Matcher::_regEncode[dst_first] < 8)
1263 ? 3
1264 : 4); // REX
1265 }
1266 } else if (dst_first_rc == rc_float) {
1267 // mem-> xmm
1268 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1269 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1270 // 64-bit
1271 int offset = ra_->reg2offset(src_first);
1272 if (cbuf) {
1273 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1274 if (Matcher::_regEncode[dst_first] >= 8) {
1275 emit_opcode(*cbuf, Assembler::REX_R);
1276 }
1277 emit_opcode(*cbuf, 0x0F);
1278 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1279 encode_RegMem(*cbuf,
1280 Matcher::_regEncode[dst_first],
1281 RSP_enc, 0x4, 0, offset,
1282 false);
1283 #ifndef PRODUCT
1284 } else if (!do_size) {
1285 st->print("%s %s, [rsp + #%d]\t# spill",
1286 UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1287 Matcher::regName[dst_first],
1288 offset);
1289 #endif
1290 }
1291 return
1292 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1293 ((Matcher::_regEncode[dst_first] < 8)
1294 ? 5
1295 : 6); // REX
1296 } else {
1297 // 32-bit
1298 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1299 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1300 int offset = ra_->reg2offset(src_first);
1301 if (cbuf) {
1302 emit_opcode(*cbuf, 0xF3);
1303 if (Matcher::_regEncode[dst_first] >= 8) {
1304 emit_opcode(*cbuf, Assembler::REX_R);
1305 }
1306 emit_opcode(*cbuf, 0x0F);
1307 emit_opcode(*cbuf, 0x10);
1308 encode_RegMem(*cbuf,
1309 Matcher::_regEncode[dst_first],
1310 RSP_enc, 0x4, 0, offset,
1311 false);
1312 #ifndef PRODUCT
1313 } else if (!do_size) {
1314 st->print("movss %s, [rsp + #%d]\t# spill",
1315 Matcher::regName[dst_first],
1316 offset);
1317 #endif
1318 }
1319 return
1320 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1321 ((Matcher::_regEncode[dst_first] < 8)
1322 ? 5
1323 : 6); // REX
1324 }
1325 }
1326 } else if (src_first_rc == rc_int) {
1327 // gpr ->
1328 if (dst_first_rc == rc_stack) {
1329 // gpr -> mem
1330 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1331 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1332 // 64-bit
1333 int offset = ra_->reg2offset(dst_first);
1334 if (cbuf) {
1335 if (Matcher::_regEncode[src_first] < 8) {
1336 emit_opcode(*cbuf, Assembler::REX_W);
1337 } else {
1338 emit_opcode(*cbuf, Assembler::REX_WR);
1339 }
1340 emit_opcode(*cbuf, 0x89);
1341 encode_RegMem(*cbuf,
1342 Matcher::_regEncode[src_first],
1343 RSP_enc, 0x4, 0, offset,
1344 false);
1345 #ifndef PRODUCT
1346 } else if (!do_size) {
1347 st->print("movq [rsp + #%d], %s\t# spill",
1348 offset,
1349 Matcher::regName[src_first]);
1350 #endif
1351 }
1352 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1353 } else {
1354 // 32-bit
1355 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1356 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1357 int offset = ra_->reg2offset(dst_first);
1358 if (cbuf) {
1359 if (Matcher::_regEncode[src_first] >= 8) {
1360 emit_opcode(*cbuf, Assembler::REX_R);
1361 }
1362 emit_opcode(*cbuf, 0x89);
1363 encode_RegMem(*cbuf,
1364 Matcher::_regEncode[src_first],
1365 RSP_enc, 0x4, 0, offset,
1366 false);
1367 #ifndef PRODUCT
1368 } else if (!do_size) {
1369 st->print("movl [rsp + #%d], %s\t# spill",
1370 offset,
1371 Matcher::regName[src_first]);
1372 #endif
1373 }
1374 return
1375 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1376 ((Matcher::_regEncode[src_first] < 8)
1377 ? 3
1378 : 4); // REX
1379 }
1380 } else if (dst_first_rc == rc_int) {
1381 // gpr -> gpr
1382 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1383 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1384 // 64-bit
1385 if (cbuf) {
1386 if (Matcher::_regEncode[dst_first] < 8) {
1387 if (Matcher::_regEncode[src_first] < 8) {
1388 emit_opcode(*cbuf, Assembler::REX_W);
1389 } else {
1390 emit_opcode(*cbuf, Assembler::REX_WB);
1391 }
1392 } else {
1393 if (Matcher::_regEncode[src_first] < 8) {
1394 emit_opcode(*cbuf, Assembler::REX_WR);
1395 } else {
1396 emit_opcode(*cbuf, Assembler::REX_WRB);
1397 }
1398 }
1399 emit_opcode(*cbuf, 0x8B);
1400 emit_rm(*cbuf, 0x3,
1401 Matcher::_regEncode[dst_first] & 7,
1402 Matcher::_regEncode[src_first] & 7);
1403 #ifndef PRODUCT
1404 } else if (!do_size) {
1405 st->print("movq %s, %s\t# spill",
1406 Matcher::regName[dst_first],
1407 Matcher::regName[src_first]);
1408 #endif
1409 }
1410 return 3; // REX
1411 } else {
1412 // 32-bit
1413 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1414 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1415 if (cbuf) {
1416 if (Matcher::_regEncode[dst_first] < 8) {
1417 if (Matcher::_regEncode[src_first] >= 8) {
1418 emit_opcode(*cbuf, Assembler::REX_B);
1419 }
1420 } else {
1421 if (Matcher::_regEncode[src_first] < 8) {
1422 emit_opcode(*cbuf, Assembler::REX_R);
1423 } else {
1424 emit_opcode(*cbuf, Assembler::REX_RB);
1425 }
1426 }
1427 emit_opcode(*cbuf, 0x8B);
1428 emit_rm(*cbuf, 0x3,
1429 Matcher::_regEncode[dst_first] & 7,
1430 Matcher::_regEncode[src_first] & 7);
1431 #ifndef PRODUCT
1432 } else if (!do_size) {
1433 st->print("movl %s, %s\t# spill",
1434 Matcher::regName[dst_first],
1435 Matcher::regName[src_first]);
1436 #endif
1437 }
1438 return
1439 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1440 ? 2
1441 : 3; // REX
1442 }
1443 } else if (dst_first_rc == rc_float) {
1444 // gpr -> xmm
1445 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1446 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1447 // 64-bit
1448 if (cbuf) {
1449 emit_opcode(*cbuf, 0x66);
1450 if (Matcher::_regEncode[dst_first] < 8) {
1451 if (Matcher::_regEncode[src_first] < 8) {
1452 emit_opcode(*cbuf, Assembler::REX_W);
1453 } else {
1454 emit_opcode(*cbuf, Assembler::REX_WB);
1455 }
1456 } else {
1457 if (Matcher::_regEncode[src_first] < 8) {
1458 emit_opcode(*cbuf, Assembler::REX_WR);
1459 } else {
1460 emit_opcode(*cbuf, Assembler::REX_WRB);
1461 }
1462 }
1463 emit_opcode(*cbuf, 0x0F);
1464 emit_opcode(*cbuf, 0x6E);
1465 emit_rm(*cbuf, 0x3,
1466 Matcher::_regEncode[dst_first] & 7,
1467 Matcher::_regEncode[src_first] & 7);
1468 #ifndef PRODUCT
1469 } else if (!do_size) {
1470 st->print("movdq %s, %s\t# spill",
1471 Matcher::regName[dst_first],
1472 Matcher::regName[src_first]);
1473 #endif
1474 }
1475 return 5; // REX
1476 } else {
1477 // 32-bit
1478 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1479 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1480 if (cbuf) {
1481 emit_opcode(*cbuf, 0x66);
1482 if (Matcher::_regEncode[dst_first] < 8) {
1483 if (Matcher::_regEncode[src_first] >= 8) {
1484 emit_opcode(*cbuf, Assembler::REX_B);
1485 }
1486 } else {
1487 if (Matcher::_regEncode[src_first] < 8) {
1488 emit_opcode(*cbuf, Assembler::REX_R);
1489 } else {
1490 emit_opcode(*cbuf, Assembler::REX_RB);
1491 }
1492 }
1493 emit_opcode(*cbuf, 0x0F);
1494 emit_opcode(*cbuf, 0x6E);
1495 emit_rm(*cbuf, 0x3,
1496 Matcher::_regEncode[dst_first] & 7,
1497 Matcher::_regEncode[src_first] & 7);
1498 #ifndef PRODUCT
1499 } else if (!do_size) {
1500 st->print("movdl %s, %s\t# spill",
1501 Matcher::regName[dst_first],
1502 Matcher::regName[src_first]);
1503 #endif
1504 }
1505 return
1506 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1507 ? 4
1508 : 5; // REX
1509 }
1510 }
1511 } else if (src_first_rc == rc_float) {
1512 // xmm ->
1513 if (dst_first_rc == rc_stack) {
1514 // xmm -> mem
1515 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1516 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1517 // 64-bit
1518 int offset = ra_->reg2offset(dst_first);
1519 if (cbuf) {
1520 emit_opcode(*cbuf, 0xF2);
1521 if (Matcher::_regEncode[src_first] >= 8) {
1522 emit_opcode(*cbuf, Assembler::REX_R);
1523 }
1524 emit_opcode(*cbuf, 0x0F);
1525 emit_opcode(*cbuf, 0x11);
1526 encode_RegMem(*cbuf,
1527 Matcher::_regEncode[src_first],
1528 RSP_enc, 0x4, 0, offset,
1529 false);
1530 #ifndef PRODUCT
1531 } else if (!do_size) {
1532 st->print("movsd [rsp + #%d], %s\t# spill",
1533 offset,
1534 Matcher::regName[src_first]);
1535 #endif
1536 }
1537 return
1538 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1539 ((Matcher::_regEncode[src_first] < 8)
1540 ? 5
1541 : 6); // REX
1542 } else {
1543 // 32-bit
1544 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1545 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1546 int offset = ra_->reg2offset(dst_first);
1547 if (cbuf) {
1548 emit_opcode(*cbuf, 0xF3);
1549 if (Matcher::_regEncode[src_first] >= 8) {
1550 emit_opcode(*cbuf, Assembler::REX_R);
1551 }
1552 emit_opcode(*cbuf, 0x0F);
1553 emit_opcode(*cbuf, 0x11);
1554 encode_RegMem(*cbuf,
1555 Matcher::_regEncode[src_first],
1556 RSP_enc, 0x4, 0, offset,
1557 false);
1558 #ifndef PRODUCT
1559 } else if (!do_size) {
1560 st->print("movss [rsp + #%d], %s\t# spill",
1561 offset,
1562 Matcher::regName[src_first]);
1563 #endif
1564 }
1565 return
1566 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1567 ((Matcher::_regEncode[src_first] < 8)
1568 ? 5
1569 : 6); // REX
1570 }
1571 } else if (dst_first_rc == rc_int) {
1572 // xmm -> gpr
1573 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1574 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1575 // 64-bit
1576 if (cbuf) {
1577 emit_opcode(*cbuf, 0x66);
1578 if (Matcher::_regEncode[dst_first] < 8) {
1579 if (Matcher::_regEncode[src_first] < 8) {
1580 emit_opcode(*cbuf, Assembler::REX_W);
1581 } else {
1582 emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1583 }
1584 } else {
1585 if (Matcher::_regEncode[src_first] < 8) {
1586 emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1587 } else {
1588 emit_opcode(*cbuf, Assembler::REX_WRB);
1589 }
1590 }
1591 emit_opcode(*cbuf, 0x0F);
1592 emit_opcode(*cbuf, 0x7E);
1593 emit_rm(*cbuf, 0x3,
1594 Matcher::_regEncode[dst_first] & 7,
1595 Matcher::_regEncode[src_first] & 7);
1596 #ifndef PRODUCT
1597 } else if (!do_size) {
1598 st->print("movdq %s, %s\t# spill",
1599 Matcher::regName[dst_first],
1600 Matcher::regName[src_first]);
1601 #endif
1602 }
1603 return 5; // REX
1604 } else {
1605 // 32-bit
1606 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1607 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1608 if (cbuf) {
1609 emit_opcode(*cbuf, 0x66);
1610 if (Matcher::_regEncode[dst_first] < 8) {
1611 if (Matcher::_regEncode[src_first] >= 8) {
1612 emit_opcode(*cbuf, Assembler::REX_R); // attention!
1613 }
1614 } else {
1615 if (Matcher::_regEncode[src_first] < 8) {
1616 emit_opcode(*cbuf, Assembler::REX_B); // attention!
1617 } else {
1618 emit_opcode(*cbuf, Assembler::REX_RB);
1619 }
1620 }
1621 emit_opcode(*cbuf, 0x0F);
1622 emit_opcode(*cbuf, 0x7E);
1623 emit_rm(*cbuf, 0x3,
1624 Matcher::_regEncode[dst_first] & 7,
1625 Matcher::_regEncode[src_first] & 7);
1626 #ifndef PRODUCT
1627 } else if (!do_size) {
1628 st->print("movdl %s, %s\t# spill",
1629 Matcher::regName[dst_first],
1630 Matcher::regName[src_first]);
1631 #endif
1632 }
1633 return
1634 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1635 ? 4
1636 : 5; // REX
1637 }
1638 } else if (dst_first_rc == rc_float) {
1639 // xmm -> xmm
1640 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1641 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1642 // 64-bit
1643 if (cbuf) {
1644 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1645 if (Matcher::_regEncode[dst_first] < 8) {
1646 if (Matcher::_regEncode[src_first] >= 8) {
1647 emit_opcode(*cbuf, Assembler::REX_B);
1648 }
1649 } else {
1650 if (Matcher::_regEncode[src_first] < 8) {
1651 emit_opcode(*cbuf, Assembler::REX_R);
1652 } else {
1653 emit_opcode(*cbuf, Assembler::REX_RB);
1654 }
1655 }
1656 emit_opcode(*cbuf, 0x0F);
1657 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1658 emit_rm(*cbuf, 0x3,
1659 Matcher::_regEncode[dst_first] & 7,
1660 Matcher::_regEncode[src_first] & 7);
1661 #ifndef PRODUCT
1662 } else if (!do_size) {
1663 st->print("%s %s, %s\t# spill",
1664 UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1665 Matcher::regName[dst_first],
1666 Matcher::regName[src_first]);
1667 #endif
1668 }
1669 return
1670 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1671 ? 4
1672 : 5; // REX
1673 } else {
1674 // 32-bit
1675 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1676 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1677 if (cbuf) {
1678 if (!UseXmmRegToRegMoveAll)
1679 emit_opcode(*cbuf, 0xF3);
1680 if (Matcher::_regEncode[dst_first] < 8) {
1681 if (Matcher::_regEncode[src_first] >= 8) {
1682 emit_opcode(*cbuf, Assembler::REX_B);
1683 }
1684 } else {
1685 if (Matcher::_regEncode[src_first] < 8) {
1686 emit_opcode(*cbuf, Assembler::REX_R);
1687 } else {
1688 emit_opcode(*cbuf, Assembler::REX_RB);
1689 }
1690 }
1691 emit_opcode(*cbuf, 0x0F);
1692 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1693 emit_rm(*cbuf, 0x3,
1694 Matcher::_regEncode[dst_first] & 7,
1695 Matcher::_regEncode[src_first] & 7);
1696 #ifndef PRODUCT
1697 } else if (!do_size) {
1698 st->print("%s %s, %s\t# spill",
1699 UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1700 Matcher::regName[dst_first],
1701 Matcher::regName[src_first]);
1702 #endif
1703 }
1704 return
1705 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1706 ? (UseXmmRegToRegMoveAll ? 3 : 4)
1707 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1708 }
1709 }
1710 }
1712 assert(0," foo ");
1713 Unimplemented();
1715 return 0;
1716 }
1718 #ifndef PRODUCT
1719 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1720 {
1721 implementation(NULL, ra_, false, st);
1722 }
1723 #endif
1725 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1726 {
1727 implementation(&cbuf, ra_, false, NULL);
1728 }
1730 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1731 {
1732 return implementation(NULL, ra_, true, NULL);
1733 }
1735 //=============================================================================
1736 #ifndef PRODUCT
1737 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1738 {
1739 st->print("nop \t# %d bytes pad for loops and calls", _count);
1740 }
1741 #endif
1743 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1744 {
1745 MacroAssembler _masm(&cbuf);
1746 __ nop(_count);
1747 }
1749 uint MachNopNode::size(PhaseRegAlloc*) const
1750 {
1751 return _count;
1752 }
1755 //=============================================================================
1756 #ifndef PRODUCT
1757 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1758 {
1759 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1760 int reg = ra_->get_reg_first(this);
1761 st->print("leaq %s, [rsp + #%d]\t# box lock",
1762 Matcher::regName[reg], offset);
1763 }
1764 #endif
1766 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1767 {
1768 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1769 int reg = ra_->get_encode(this);
1770 if (offset >= 0x80) {
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, 0x2, reg & 7, 0x04);
1774 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1775 emit_d32(cbuf, offset);
1776 } else {
1777 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1778 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1779 emit_rm(cbuf, 0x1, reg & 7, 0x04);
1780 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1781 emit_d8(cbuf, offset);
1782 }
1783 }
1785 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1786 {
1787 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1788 return (offset < 0x80) ? 5 : 8; // REX
1789 }
1791 //=============================================================================
1793 // emit call stub, compiled java to interpreter
1794 void emit_java_to_interp(CodeBuffer& cbuf)
1795 {
1796 // Stub is fixed up when the corresponding call is converted from
1797 // calling compiled code to calling interpreted code.
1798 // movq rbx, 0
1799 // jmp -5 # to self
1801 address mark = cbuf.inst_mark(); // get mark within main instrs section
1803 // Note that the code buffer's inst_mark is always relative to insts.
1804 // That's why we must use the macroassembler to generate a stub.
1805 MacroAssembler _masm(&cbuf);
1807 address base =
1808 __ start_a_stub(Compile::MAX_stubs_size);
1809 if (base == NULL) return; // CodeBuffer::expand failed
1810 // static stub relocation stores the instruction address of the call
1811 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1812 // static stub relocation also tags the methodOop in the code-stream.
1813 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time
1814 // This is recognized as unresolved by relocs/nativeinst/ic code
1815 __ jump(RuntimeAddress(__ pc()));
1817 // Update current stubs pointer and restore code_end.
1818 __ end_a_stub();
1819 }
1821 // size of call stub, compiled java to interpretor
1822 uint size_java_to_interp()
1823 {
1824 return 15; // movq (1+1+8); jmp (1+4)
1825 }
1827 // relocation entries for call stub, compiled java to interpretor
1828 uint reloc_java_to_interp()
1829 {
1830 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1831 }
1833 //=============================================================================
1834 #ifndef PRODUCT
1835 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1836 {
1837 if (UseCompressedOops) {
1838 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
1839 if (Universe::narrow_oop_shift() != 0) {
1840 st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]");
1841 }
1842 st->print_cr("cmpq rax, rscratch1\t # Inline cache check");
1843 } else {
1844 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
1845 "# Inline cache check", oopDesc::klass_offset_in_bytes());
1846 }
1847 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1848 st->print_cr("\tnop");
1849 if (!OptoBreakpoint) {
1850 st->print_cr("\tnop");
1851 }
1852 }
1853 #endif
1855 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1856 {
1857 MacroAssembler masm(&cbuf);
1858 #ifdef ASSERT
1859 uint code_size = cbuf.code_size();
1860 #endif
1861 if (UseCompressedOops) {
1862 masm.load_klass(rscratch1, j_rarg0);
1863 masm.cmpptr(rax, rscratch1);
1864 } else {
1865 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1866 }
1868 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1870 /* WARNING these NOPs are critical so that verified entry point is properly
1871 aligned for patching by NativeJump::patch_verified_entry() */
1872 int nops_cnt = 1;
1873 if (!OptoBreakpoint) {
1874 // Leave space for int3
1875 nops_cnt += 1;
1876 }
1877 if (UseCompressedOops) {
1878 // ??? divisible by 4 is aligned?
1879 nops_cnt += 1;
1880 }
1881 masm.nop(nops_cnt);
1883 assert(cbuf.code_size() - code_size == size(ra_),
1884 "checking code size of inline cache node");
1885 }
1887 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1888 {
1889 if (UseCompressedOops) {
1890 if (Universe::narrow_oop_shift() == 0) {
1891 return OptoBreakpoint ? 15 : 16;
1892 } else {
1893 return OptoBreakpoint ? 19 : 20;
1894 }
1895 } else {
1896 return OptoBreakpoint ? 11 : 12;
1897 }
1898 }
1901 //=============================================================================
1902 uint size_exception_handler()
1903 {
1904 // NativeCall instruction size is the same as NativeJump.
1905 // Note that this value is also credited (in output.cpp) to
1906 // the size of the code section.
1907 return NativeJump::instruction_size;
1908 }
1910 // Emit exception handler code.
1911 int emit_exception_handler(CodeBuffer& cbuf)
1912 {
1914 // Note that the code buffer's inst_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_exception_handler());
1919 if (base == NULL) return 0; // CodeBuffer::expand failed
1920 int offset = __ offset();
1921 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
1922 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1923 __ end_a_stub();
1924 return offset;
1925 }
1927 uint size_deopt_handler()
1928 {
1929 // three 5 byte instructions
1930 return 15;
1931 }
1933 // Emit deopt handler code.
1934 int emit_deopt_handler(CodeBuffer& cbuf)
1935 {
1937 // Note that the code buffer's inst_mark is always relative to insts.
1938 // That's why we must use the macroassembler to generate a handler.
1939 MacroAssembler _masm(&cbuf);
1940 address base =
1941 __ start_a_stub(size_deopt_handler());
1942 if (base == NULL) return 0; // CodeBuffer::expand failed
1943 int offset = __ offset();
1944 address the_pc = (address) __ pc();
1945 Label next;
1946 // push a "the_pc" on the stack without destroying any registers
1947 // as they all may be live.
1949 // push address of "next"
1950 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1951 __ bind(next);
1952 // adjust it so it matches "the_pc"
1953 __ subptr(Address(rsp, 0), __ offset() - offset);
1954 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1955 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1956 __ end_a_stub();
1957 return offset;
1958 }
1960 static void emit_double_constant(CodeBuffer& cbuf, double x) {
1961 int mark = cbuf.insts()->mark_off();
1962 MacroAssembler _masm(&cbuf);
1963 address double_address = __ double_constant(x);
1964 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1965 emit_d32_reloc(cbuf,
1966 (int) (double_address - cbuf.code_end() - 4),
1967 internal_word_Relocation::spec(double_address),
1968 RELOC_DISP32);
1969 }
1971 static void emit_float_constant(CodeBuffer& cbuf, float x) {
1972 int mark = cbuf.insts()->mark_off();
1973 MacroAssembler _masm(&cbuf);
1974 address float_address = __ float_constant(x);
1975 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1976 emit_d32_reloc(cbuf,
1977 (int) (float_address - cbuf.code_end() - 4),
1978 internal_word_Relocation::spec(float_address),
1979 RELOC_DISP32);
1980 }
1983 int Matcher::regnum_to_fpu_offset(int regnum)
1984 {
1985 return regnum - 32; // The FP registers are in the second chunk
1986 }
1988 // This is UltraSparc specific, true just means we have fast l2f conversion
1989 const bool Matcher::convL2FSupported(void) {
1990 return true;
1991 }
1993 // Vector width in bytes
1994 const uint Matcher::vector_width_in_bytes(void) {
1995 return 8;
1996 }
1998 // Vector ideal reg
1999 const uint Matcher::vector_ideal_reg(void) {
2000 return Op_RegD;
2001 }
2003 // Is this branch offset short enough that a short branch can be used?
2004 //
2005 // NOTE: If the platform does not provide any short branch variants, then
2006 // this method should return false for offset 0.
2007 bool Matcher::is_short_branch_offset(int rule, int offset) {
2008 // the short version of jmpConUCF2 contains multiple branches,
2009 // making the reach slightly less
2010 if (rule == jmpConUCF2_rule)
2011 return (-126 <= offset && offset <= 125);
2012 return (-128 <= offset && offset <= 127);
2013 }
2015 const bool Matcher::isSimpleConstant64(jlong value) {
2016 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
2017 //return value == (int) value; // Cf. storeImmL and immL32.
2019 // Probably always true, even if a temp register is required.
2020 return true;
2021 }
2023 // The ecx parameter to rep stosq for the ClearArray node is in words.
2024 const bool Matcher::init_array_count_is_in_bytes = false;
2026 // Threshold size for cleararray.
2027 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2029 // Should the Matcher clone shifts on addressing modes, expecting them
2030 // to be subsumed into complex addressing expressions or compute them
2031 // into registers? True for Intel but false for most RISCs
2032 const bool Matcher::clone_shift_expressions = true;
2034 // Is it better to copy float constants, or load them directly from
2035 // memory? Intel can load a float constant from a direct address,
2036 // requiring no extra registers. Most RISCs will have to materialize
2037 // an address into a register first, so they would do better to copy
2038 // the constant from stack.
2039 const bool Matcher::rematerialize_float_constants = true; // XXX
2041 // If CPU can load and store mis-aligned doubles directly then no
2042 // fixup is needed. Else we split the double into 2 integer pieces
2043 // and move it piece-by-piece. Only happens when passing doubles into
2044 // C code as the Java calling convention forces doubles to be aligned.
2045 const bool Matcher::misaligned_doubles_ok = true;
2047 // No-op on amd64
2048 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2050 // Advertise here if the CPU requires explicit rounding operations to
2051 // implement the UseStrictFP mode.
2052 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2054 // Do floats take an entire double register or just half?
2055 const bool Matcher::float_in_double = true;
2056 // Do ints take an entire long register or just half?
2057 const bool Matcher::int_in_long = true;
2059 // Return whether or not this register is ever used as an argument.
2060 // This function is used on startup to build the trampoline stubs in
2061 // generateOptoStub. Registers not mentioned will be killed by the VM
2062 // call in the trampoline, and arguments in those registers not be
2063 // available to the callee.
2064 bool Matcher::can_be_java_arg(int reg)
2065 {
2066 return
2067 reg == RDI_num || reg == RDI_H_num ||
2068 reg == RSI_num || reg == RSI_H_num ||
2069 reg == RDX_num || reg == RDX_H_num ||
2070 reg == RCX_num || reg == RCX_H_num ||
2071 reg == R8_num || reg == R8_H_num ||
2072 reg == R9_num || reg == R9_H_num ||
2073 reg == R12_num || reg == R12_H_num ||
2074 reg == XMM0_num || reg == XMM0_H_num ||
2075 reg == XMM1_num || reg == XMM1_H_num ||
2076 reg == XMM2_num || reg == XMM2_H_num ||
2077 reg == XMM3_num || reg == XMM3_H_num ||
2078 reg == XMM4_num || reg == XMM4_H_num ||
2079 reg == XMM5_num || reg == XMM5_H_num ||
2080 reg == XMM6_num || reg == XMM6_H_num ||
2081 reg == XMM7_num || reg == XMM7_H_num;
2082 }
2084 bool Matcher::is_spillable_arg(int reg)
2085 {
2086 return can_be_java_arg(reg);
2087 }
2089 // Register for DIVI projection of divmodI
2090 RegMask Matcher::divI_proj_mask() {
2091 return INT_RAX_REG_mask;
2092 }
2094 // Register for MODI projection of divmodI
2095 RegMask Matcher::modI_proj_mask() {
2096 return INT_RDX_REG_mask;
2097 }
2099 // Register for DIVL projection of divmodL
2100 RegMask Matcher::divL_proj_mask() {
2101 return LONG_RAX_REG_mask;
2102 }
2104 // Register for MODL projection of divmodL
2105 RegMask Matcher::modL_proj_mask() {
2106 return LONG_RDX_REG_mask;
2107 }
2109 static Address build_address(int b, int i, int s, int d) {
2110 Register index = as_Register(i);
2111 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2112 if (index == rsp) {
2113 index = noreg;
2114 scale = Address::no_scale;
2115 }
2116 Address addr(as_Register(b), index, scale, d);
2117 return addr;
2118 }
2120 %}
2122 //----------ENCODING BLOCK-----------------------------------------------------
2123 // This block specifies the encoding classes used by the compiler to
2124 // output byte streams. Encoding classes are parameterized macros
2125 // used by Machine Instruction Nodes in order to generate the bit
2126 // encoding of the instruction. Operands specify their base encoding
2127 // interface with the interface keyword. There are currently
2128 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2129 // COND_INTER. REG_INTER causes an operand to generate a function
2130 // which returns its register number when queried. CONST_INTER causes
2131 // an operand to generate a function which returns the value of the
2132 // constant when queried. MEMORY_INTER causes an operand to generate
2133 // four functions which return the Base Register, the Index Register,
2134 // the Scale Value, and the Offset Value of the operand when queried.
2135 // COND_INTER causes an operand to generate six functions which return
2136 // the encoding code (ie - encoding bits for the instruction)
2137 // associated with each basic boolean condition for a conditional
2138 // instruction.
2139 //
2140 // Instructions specify two basic values for encoding. Again, a
2141 // function is available to check if the constant displacement is an
2142 // oop. They use the ins_encode keyword to specify their encoding
2143 // classes (which must be a sequence of enc_class names, and their
2144 // parameters, specified in the encoding block), and they use the
2145 // opcode keyword to specify, in order, their primary, secondary, and
2146 // tertiary opcode. Only the opcode sections which a particular
2147 // instruction needs for encoding need to be specified.
2148 encode %{
2149 // Build emit functions for each basic byte or larger field in the
2150 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2151 // from C++ code in the enc_class source block. Emit functions will
2152 // live in the main source block for now. In future, we can
2153 // generalize this by adding a syntax that specifies the sizes of
2154 // fields in an order, so that the adlc can build the emit functions
2155 // automagically
2157 // Emit primary opcode
2158 enc_class OpcP
2159 %{
2160 emit_opcode(cbuf, $primary);
2161 %}
2163 // Emit secondary opcode
2164 enc_class OpcS
2165 %{
2166 emit_opcode(cbuf, $secondary);
2167 %}
2169 // Emit tertiary opcode
2170 enc_class OpcT
2171 %{
2172 emit_opcode(cbuf, $tertiary);
2173 %}
2175 // Emit opcode directly
2176 enc_class Opcode(immI d8)
2177 %{
2178 emit_opcode(cbuf, $d8$$constant);
2179 %}
2181 // Emit size prefix
2182 enc_class SizePrefix
2183 %{
2184 emit_opcode(cbuf, 0x66);
2185 %}
2187 enc_class reg(rRegI reg)
2188 %{
2189 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2190 %}
2192 enc_class reg_reg(rRegI dst, rRegI src)
2193 %{
2194 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2195 %}
2197 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2198 %{
2199 emit_opcode(cbuf, $opcode$$constant);
2200 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2201 %}
2203 enc_class cmpfp_fixup()
2204 %{
2205 // jnp,s exit
2206 emit_opcode(cbuf, 0x7B);
2207 emit_d8(cbuf, 0x0A);
2209 // pushfq
2210 emit_opcode(cbuf, 0x9C);
2212 // andq $0xffffff2b, (%rsp)
2213 emit_opcode(cbuf, Assembler::REX_W);
2214 emit_opcode(cbuf, 0x81);
2215 emit_opcode(cbuf, 0x24);
2216 emit_opcode(cbuf, 0x24);
2217 emit_d32(cbuf, 0xffffff2b);
2219 // popfq
2220 emit_opcode(cbuf, 0x9D);
2222 // nop (target for branch to avoid branch to branch)
2223 emit_opcode(cbuf, 0x90);
2224 %}
2226 enc_class cmpfp3(rRegI dst)
2227 %{
2228 int dstenc = $dst$$reg;
2230 // movl $dst, -1
2231 if (dstenc >= 8) {
2232 emit_opcode(cbuf, Assembler::REX_B);
2233 }
2234 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2235 emit_d32(cbuf, -1);
2237 // jp,s done
2238 emit_opcode(cbuf, 0x7A);
2239 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2241 // jb,s done
2242 emit_opcode(cbuf, 0x72);
2243 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2245 // setne $dst
2246 if (dstenc >= 4) {
2247 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2248 }
2249 emit_opcode(cbuf, 0x0F);
2250 emit_opcode(cbuf, 0x95);
2251 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2253 // movzbl $dst, $dst
2254 if (dstenc >= 4) {
2255 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2256 }
2257 emit_opcode(cbuf, 0x0F);
2258 emit_opcode(cbuf, 0xB6);
2259 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2260 %}
2262 enc_class cdql_enc(no_rax_rdx_RegI div)
2263 %{
2264 // Full implementation of Java idiv and irem; checks for
2265 // special case as described in JVM spec., p.243 & p.271.
2266 //
2267 // normal case special case
2268 //
2269 // input : rax: dividend min_int
2270 // reg: divisor -1
2271 //
2272 // output: rax: quotient (= rax idiv reg) min_int
2273 // rdx: remainder (= rax irem reg) 0
2274 //
2275 // Code sequnce:
2276 //
2277 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2278 // 5: 75 07/08 jne e <normal>
2279 // 7: 33 d2 xor %edx,%edx
2280 // [div >= 8 -> offset + 1]
2281 // [REX_B]
2282 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2283 // c: 74 03/04 je 11 <done>
2284 // 000000000000000e <normal>:
2285 // e: 99 cltd
2286 // [div >= 8 -> offset + 1]
2287 // [REX_B]
2288 // f: f7 f9 idiv $div
2289 // 0000000000000011 <done>:
2291 // cmp $0x80000000,%eax
2292 emit_opcode(cbuf, 0x3d);
2293 emit_d8(cbuf, 0x00);
2294 emit_d8(cbuf, 0x00);
2295 emit_d8(cbuf, 0x00);
2296 emit_d8(cbuf, 0x80);
2298 // jne e <normal>
2299 emit_opcode(cbuf, 0x75);
2300 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2302 // xor %edx,%edx
2303 emit_opcode(cbuf, 0x33);
2304 emit_d8(cbuf, 0xD2);
2306 // cmp $0xffffffffffffffff,%ecx
2307 if ($div$$reg >= 8) {
2308 emit_opcode(cbuf, Assembler::REX_B);
2309 }
2310 emit_opcode(cbuf, 0x83);
2311 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2312 emit_d8(cbuf, 0xFF);
2314 // je 11 <done>
2315 emit_opcode(cbuf, 0x74);
2316 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2318 // <normal>
2319 // cltd
2320 emit_opcode(cbuf, 0x99);
2322 // idivl (note: must be emitted by the user of this rule)
2323 // <done>
2324 %}
2326 enc_class cdqq_enc(no_rax_rdx_RegL div)
2327 %{
2328 // Full implementation of Java ldiv and lrem; checks for
2329 // special case as described in JVM spec., p.243 & p.271.
2330 //
2331 // normal case special case
2332 //
2333 // input : rax: dividend min_long
2334 // reg: divisor -1
2335 //
2336 // output: rax: quotient (= rax idiv reg) min_long
2337 // rdx: remainder (= rax irem reg) 0
2338 //
2339 // Code sequnce:
2340 //
2341 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2342 // 7: 00 00 80
2343 // a: 48 39 d0 cmp %rdx,%rax
2344 // d: 75 08 jne 17 <normal>
2345 // f: 33 d2 xor %edx,%edx
2346 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2347 // 15: 74 05 je 1c <done>
2348 // 0000000000000017 <normal>:
2349 // 17: 48 99 cqto
2350 // 19: 48 f7 f9 idiv $div
2351 // 000000000000001c <done>:
2353 // mov $0x8000000000000000,%rdx
2354 emit_opcode(cbuf, Assembler::REX_W);
2355 emit_opcode(cbuf, 0xBA);
2356 emit_d8(cbuf, 0x00);
2357 emit_d8(cbuf, 0x00);
2358 emit_d8(cbuf, 0x00);
2359 emit_d8(cbuf, 0x00);
2360 emit_d8(cbuf, 0x00);
2361 emit_d8(cbuf, 0x00);
2362 emit_d8(cbuf, 0x00);
2363 emit_d8(cbuf, 0x80);
2365 // cmp %rdx,%rax
2366 emit_opcode(cbuf, Assembler::REX_W);
2367 emit_opcode(cbuf, 0x39);
2368 emit_d8(cbuf, 0xD0);
2370 // jne 17 <normal>
2371 emit_opcode(cbuf, 0x75);
2372 emit_d8(cbuf, 0x08);
2374 // xor %edx,%edx
2375 emit_opcode(cbuf, 0x33);
2376 emit_d8(cbuf, 0xD2);
2378 // cmp $0xffffffffffffffff,$div
2379 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2380 emit_opcode(cbuf, 0x83);
2381 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2382 emit_d8(cbuf, 0xFF);
2384 // je 1e <done>
2385 emit_opcode(cbuf, 0x74);
2386 emit_d8(cbuf, 0x05);
2388 // <normal>
2389 // cqto
2390 emit_opcode(cbuf, Assembler::REX_W);
2391 emit_opcode(cbuf, 0x99);
2393 // idivq (note: must be emitted by the user of this rule)
2394 // <done>
2395 %}
2397 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2398 enc_class OpcSE(immI imm)
2399 %{
2400 // Emit primary opcode and set sign-extend bit
2401 // Check for 8-bit immediate, and set sign extend bit in opcode
2402 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2403 emit_opcode(cbuf, $primary | 0x02);
2404 } else {
2405 // 32-bit immediate
2406 emit_opcode(cbuf, $primary);
2407 }
2408 %}
2410 enc_class OpcSErm(rRegI dst, immI imm)
2411 %{
2412 // OpcSEr/m
2413 int dstenc = $dst$$reg;
2414 if (dstenc >= 8) {
2415 emit_opcode(cbuf, Assembler::REX_B);
2416 dstenc -= 8;
2417 }
2418 // Emit primary opcode and set sign-extend bit
2419 // Check for 8-bit immediate, and set sign extend bit in opcode
2420 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2421 emit_opcode(cbuf, $primary | 0x02);
2422 } else {
2423 // 32-bit immediate
2424 emit_opcode(cbuf, $primary);
2425 }
2426 // Emit r/m byte with secondary opcode, after primary opcode.
2427 emit_rm(cbuf, 0x3, $secondary, dstenc);
2428 %}
2430 enc_class OpcSErm_wide(rRegL dst, immI imm)
2431 %{
2432 // OpcSEr/m
2433 int dstenc = $dst$$reg;
2434 if (dstenc < 8) {
2435 emit_opcode(cbuf, Assembler::REX_W);
2436 } else {
2437 emit_opcode(cbuf, Assembler::REX_WB);
2438 dstenc -= 8;
2439 }
2440 // Emit primary opcode and set sign-extend bit
2441 // Check for 8-bit immediate, and set sign extend bit in opcode
2442 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2443 emit_opcode(cbuf, $primary | 0x02);
2444 } else {
2445 // 32-bit immediate
2446 emit_opcode(cbuf, $primary);
2447 }
2448 // Emit r/m byte with secondary opcode, after primary opcode.
2449 emit_rm(cbuf, 0x3, $secondary, dstenc);
2450 %}
2452 enc_class Con8or32(immI imm)
2453 %{
2454 // Check for 8-bit immediate, and set sign extend bit in opcode
2455 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2456 $$$emit8$imm$$constant;
2457 } else {
2458 // 32-bit immediate
2459 $$$emit32$imm$$constant;
2460 }
2461 %}
2463 enc_class Lbl(label labl)
2464 %{
2465 // JMP, CALL
2466 Label* l = $labl$$label;
2467 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2468 %}
2470 enc_class LblShort(label labl)
2471 %{
2472 // JMP, CALL
2473 Label* l = $labl$$label;
2474 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2475 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2476 emit_d8(cbuf, disp);
2477 %}
2479 enc_class opc2_reg(rRegI dst)
2480 %{
2481 // BSWAP
2482 emit_cc(cbuf, $secondary, $dst$$reg);
2483 %}
2485 enc_class opc3_reg(rRegI dst)
2486 %{
2487 // BSWAP
2488 emit_cc(cbuf, $tertiary, $dst$$reg);
2489 %}
2491 enc_class reg_opc(rRegI div)
2492 %{
2493 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2494 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2495 %}
2497 enc_class Jcc(cmpOp cop, label labl)
2498 %{
2499 // JCC
2500 Label* l = $labl$$label;
2501 $$$emit8$primary;
2502 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2503 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2504 %}
2506 enc_class JccShort (cmpOp cop, label labl)
2507 %{
2508 // JCC
2509 Label *l = $labl$$label;
2510 emit_cc(cbuf, $primary, $cop$$cmpcode);
2511 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2512 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2513 emit_d8(cbuf, disp);
2514 %}
2516 enc_class enc_cmov(cmpOp cop)
2517 %{
2518 // CMOV
2519 $$$emit8$primary;
2520 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2521 %}
2523 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2524 %{
2525 // Invert sense of branch from sense of cmov
2526 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2527 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2528 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2529 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2530 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2531 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2532 if ($dst$$reg < 8) {
2533 if ($src$$reg >= 8) {
2534 emit_opcode(cbuf, Assembler::REX_B);
2535 }
2536 } else {
2537 if ($src$$reg < 8) {
2538 emit_opcode(cbuf, Assembler::REX_R);
2539 } else {
2540 emit_opcode(cbuf, Assembler::REX_RB);
2541 }
2542 }
2543 emit_opcode(cbuf, 0x0F);
2544 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2545 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2546 %}
2548 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2549 %{
2550 // Invert sense of branch from sense of cmov
2551 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2552 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2554 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2555 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2556 if ($dst$$reg < 8) {
2557 if ($src$$reg >= 8) {
2558 emit_opcode(cbuf, Assembler::REX_B);
2559 }
2560 } else {
2561 if ($src$$reg < 8) {
2562 emit_opcode(cbuf, Assembler::REX_R);
2563 } else {
2564 emit_opcode(cbuf, Assembler::REX_RB);
2565 }
2566 }
2567 emit_opcode(cbuf, 0x0F);
2568 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2569 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2570 %}
2572 enc_class enc_PartialSubtypeCheck()
2573 %{
2574 Register Rrdi = as_Register(RDI_enc); // result register
2575 Register Rrax = as_Register(RAX_enc); // super class
2576 Register Rrcx = as_Register(RCX_enc); // killed
2577 Register Rrsi = as_Register(RSI_enc); // sub class
2578 Label miss;
2579 const bool set_cond_codes = true;
2581 MacroAssembler _masm(&cbuf);
2582 __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi,
2583 NULL, &miss,
2584 /*set_cond_codes:*/ true);
2585 if ($primary) {
2586 __ xorptr(Rrdi, Rrdi);
2587 }
2588 __ bind(miss);
2589 %}
2591 enc_class Java_To_Interpreter(method meth)
2592 %{
2593 // CALL Java_To_Interpreter
2594 // This is the instruction starting address for relocation info.
2595 cbuf.set_inst_mark();
2596 $$$emit8$primary;
2597 // CALL directly to the runtime
2598 emit_d32_reloc(cbuf,
2599 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2600 runtime_call_Relocation::spec(),
2601 RELOC_DISP32);
2602 %}
2604 enc_class Java_Static_Call(method meth)
2605 %{
2606 // JAVA STATIC CALL
2607 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2608 // determine who we intended to call.
2609 cbuf.set_inst_mark();
2610 $$$emit8$primary;
2612 if (!_method) {
2613 emit_d32_reloc(cbuf,
2614 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2615 runtime_call_Relocation::spec(),
2616 RELOC_DISP32);
2617 } else if (_optimized_virtual) {
2618 emit_d32_reloc(cbuf,
2619 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2620 opt_virtual_call_Relocation::spec(),
2621 RELOC_DISP32);
2622 } else {
2623 emit_d32_reloc(cbuf,
2624 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2625 static_call_Relocation::spec(),
2626 RELOC_DISP32);
2627 }
2628 if (_method) {
2629 // Emit stub for static call
2630 emit_java_to_interp(cbuf);
2631 }
2632 %}
2634 enc_class Java_Dynamic_Call(method meth)
2635 %{
2636 // JAVA DYNAMIC CALL
2637 // !!!!!
2638 // Generate "movq rax, -1", placeholder instruction to load oop-info
2639 // emit_call_dynamic_prologue( cbuf );
2640 cbuf.set_inst_mark();
2642 // movq rax, -1
2643 emit_opcode(cbuf, Assembler::REX_W);
2644 emit_opcode(cbuf, 0xB8 | RAX_enc);
2645 emit_d64_reloc(cbuf,
2646 (int64_t) Universe::non_oop_word(),
2647 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2648 address virtual_call_oop_addr = cbuf.inst_mark();
2649 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2650 // who we intended to call.
2651 cbuf.set_inst_mark();
2652 $$$emit8$primary;
2653 emit_d32_reloc(cbuf,
2654 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2655 virtual_call_Relocation::spec(virtual_call_oop_addr),
2656 RELOC_DISP32);
2657 %}
2659 enc_class Java_Compiled_Call(method meth)
2660 %{
2661 // JAVA COMPILED CALL
2662 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2664 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2665 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2667 // callq *disp(%rax)
2668 cbuf.set_inst_mark();
2669 $$$emit8$primary;
2670 if (disp < 0x80) {
2671 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2672 emit_d8(cbuf, disp); // Displacement
2673 } else {
2674 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2675 emit_d32(cbuf, disp); // Displacement
2676 }
2677 %}
2679 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2680 %{
2681 // SAL, SAR, SHR
2682 int dstenc = $dst$$reg;
2683 if (dstenc >= 8) {
2684 emit_opcode(cbuf, Assembler::REX_B);
2685 dstenc -= 8;
2686 }
2687 $$$emit8$primary;
2688 emit_rm(cbuf, 0x3, $secondary, dstenc);
2689 $$$emit8$shift$$constant;
2690 %}
2692 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2693 %{
2694 // SAL, SAR, SHR
2695 int dstenc = $dst$$reg;
2696 if (dstenc < 8) {
2697 emit_opcode(cbuf, Assembler::REX_W);
2698 } else {
2699 emit_opcode(cbuf, Assembler::REX_WB);
2700 dstenc -= 8;
2701 }
2702 $$$emit8$primary;
2703 emit_rm(cbuf, 0x3, $secondary, dstenc);
2704 $$$emit8$shift$$constant;
2705 %}
2707 enc_class load_immI(rRegI dst, immI src)
2708 %{
2709 int dstenc = $dst$$reg;
2710 if (dstenc >= 8) {
2711 emit_opcode(cbuf, Assembler::REX_B);
2712 dstenc -= 8;
2713 }
2714 emit_opcode(cbuf, 0xB8 | dstenc);
2715 $$$emit32$src$$constant;
2716 %}
2718 enc_class load_immL(rRegL dst, immL src)
2719 %{
2720 int dstenc = $dst$$reg;
2721 if (dstenc < 8) {
2722 emit_opcode(cbuf, Assembler::REX_W);
2723 } else {
2724 emit_opcode(cbuf, Assembler::REX_WB);
2725 dstenc -= 8;
2726 }
2727 emit_opcode(cbuf, 0xB8 | dstenc);
2728 emit_d64(cbuf, $src$$constant);
2729 %}
2731 enc_class load_immUL32(rRegL dst, immUL32 src)
2732 %{
2733 // same as load_immI, but this time we care about zeroes in the high word
2734 int dstenc = $dst$$reg;
2735 if (dstenc >= 8) {
2736 emit_opcode(cbuf, Assembler::REX_B);
2737 dstenc -= 8;
2738 }
2739 emit_opcode(cbuf, 0xB8 | dstenc);
2740 $$$emit32$src$$constant;
2741 %}
2743 enc_class load_immL32(rRegL dst, immL32 src)
2744 %{
2745 int dstenc = $dst$$reg;
2746 if (dstenc < 8) {
2747 emit_opcode(cbuf, Assembler::REX_W);
2748 } else {
2749 emit_opcode(cbuf, Assembler::REX_WB);
2750 dstenc -= 8;
2751 }
2752 emit_opcode(cbuf, 0xC7);
2753 emit_rm(cbuf, 0x03, 0x00, dstenc);
2754 $$$emit32$src$$constant;
2755 %}
2757 enc_class load_immP31(rRegP dst, immP32 src)
2758 %{
2759 // same as load_immI, but this time we care about zeroes in the high word
2760 int dstenc = $dst$$reg;
2761 if (dstenc >= 8) {
2762 emit_opcode(cbuf, Assembler::REX_B);
2763 dstenc -= 8;
2764 }
2765 emit_opcode(cbuf, 0xB8 | dstenc);
2766 $$$emit32$src$$constant;
2767 %}
2769 enc_class load_immP(rRegP dst, immP src)
2770 %{
2771 int dstenc = $dst$$reg;
2772 if (dstenc < 8) {
2773 emit_opcode(cbuf, Assembler::REX_W);
2774 } else {
2775 emit_opcode(cbuf, Assembler::REX_WB);
2776 dstenc -= 8;
2777 }
2778 emit_opcode(cbuf, 0xB8 | dstenc);
2779 // This next line should be generated from ADLC
2780 if ($src->constant_is_oop()) {
2781 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2782 } else {
2783 emit_d64(cbuf, $src$$constant);
2784 }
2785 %}
2787 enc_class load_immF(regF dst, immF con)
2788 %{
2789 // XXX reg_mem doesn't support RIP-relative addressing yet
2790 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2791 emit_float_constant(cbuf, $con$$constant);
2792 %}
2794 enc_class load_immD(regD dst, immD con)
2795 %{
2796 // XXX reg_mem doesn't support RIP-relative addressing yet
2797 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2798 emit_double_constant(cbuf, $con$$constant);
2799 %}
2801 enc_class load_conF (regF dst, immF con) %{ // Load float constant
2802 emit_opcode(cbuf, 0xF3);
2803 if ($dst$$reg >= 8) {
2804 emit_opcode(cbuf, Assembler::REX_R);
2805 }
2806 emit_opcode(cbuf, 0x0F);
2807 emit_opcode(cbuf, 0x10);
2808 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2809 emit_float_constant(cbuf, $con$$constant);
2810 %}
2812 enc_class load_conD (regD dst, immD con) %{ // Load double constant
2813 // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
2814 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
2815 if ($dst$$reg >= 8) {
2816 emit_opcode(cbuf, Assembler::REX_R);
2817 }
2818 emit_opcode(cbuf, 0x0F);
2819 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
2820 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2821 emit_double_constant(cbuf, $con$$constant);
2822 %}
2824 // Encode a reg-reg copy. If it is useless, then empty encoding.
2825 enc_class enc_copy(rRegI dst, rRegI src)
2826 %{
2827 encode_copy(cbuf, $dst$$reg, $src$$reg);
2828 %}
2830 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2831 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2832 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2833 %}
2835 enc_class enc_copy_always(rRegI dst, rRegI src)
2836 %{
2837 int srcenc = $src$$reg;
2838 int dstenc = $dst$$reg;
2840 if (dstenc < 8) {
2841 if (srcenc >= 8) {
2842 emit_opcode(cbuf, Assembler::REX_B);
2843 srcenc -= 8;
2844 }
2845 } else {
2846 if (srcenc < 8) {
2847 emit_opcode(cbuf, Assembler::REX_R);
2848 } else {
2849 emit_opcode(cbuf, Assembler::REX_RB);
2850 srcenc -= 8;
2851 }
2852 dstenc -= 8;
2853 }
2855 emit_opcode(cbuf, 0x8B);
2856 emit_rm(cbuf, 0x3, dstenc, srcenc);
2857 %}
2859 enc_class enc_copy_wide(rRegL dst, rRegL src)
2860 %{
2861 int srcenc = $src$$reg;
2862 int dstenc = $dst$$reg;
2864 if (dstenc != srcenc) {
2865 if (dstenc < 8) {
2866 if (srcenc < 8) {
2867 emit_opcode(cbuf, Assembler::REX_W);
2868 } else {
2869 emit_opcode(cbuf, Assembler::REX_WB);
2870 srcenc -= 8;
2871 }
2872 } else {
2873 if (srcenc < 8) {
2874 emit_opcode(cbuf, Assembler::REX_WR);
2875 } else {
2876 emit_opcode(cbuf, Assembler::REX_WRB);
2877 srcenc -= 8;
2878 }
2879 dstenc -= 8;
2880 }
2881 emit_opcode(cbuf, 0x8B);
2882 emit_rm(cbuf, 0x3, dstenc, srcenc);
2883 }
2884 %}
2886 enc_class Con32(immI src)
2887 %{
2888 // Output immediate
2889 $$$emit32$src$$constant;
2890 %}
2892 enc_class Con64(immL src)
2893 %{
2894 // Output immediate
2895 emit_d64($src$$constant);
2896 %}
2898 enc_class Con32F_as_bits(immF src)
2899 %{
2900 // Output Float immediate bits
2901 jfloat jf = $src$$constant;
2902 jint jf_as_bits = jint_cast(jf);
2903 emit_d32(cbuf, jf_as_bits);
2904 %}
2906 enc_class Con16(immI src)
2907 %{
2908 // Output immediate
2909 $$$emit16$src$$constant;
2910 %}
2912 // How is this different from Con32??? XXX
2913 enc_class Con_d32(immI src)
2914 %{
2915 emit_d32(cbuf,$src$$constant);
2916 %}
2918 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2919 // Output immediate memory reference
2920 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2921 emit_d32(cbuf, 0x00);
2922 %}
2924 enc_class jump_enc(rRegL switch_val, rRegI dest) %{
2925 MacroAssembler masm(&cbuf);
2927 Register switch_reg = as_Register($switch_val$$reg);
2928 Register dest_reg = as_Register($dest$$reg);
2929 address table_base = masm.address_table_constant(_index2label);
2931 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2932 // to do that and the compiler is using that register as one it can allocate.
2933 // So we build it all by hand.
2934 // Address index(noreg, switch_reg, Address::times_1);
2935 // ArrayAddress dispatch(table, index);
2937 Address dispatch(dest_reg, switch_reg, Address::times_1);
2939 masm.lea(dest_reg, InternalAddress(table_base));
2940 masm.jmp(dispatch);
2941 %}
2943 enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
2944 MacroAssembler masm(&cbuf);
2946 Register switch_reg = as_Register($switch_val$$reg);
2947 Register dest_reg = as_Register($dest$$reg);
2948 address table_base = masm.address_table_constant(_index2label);
2950 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2951 // to do that and the compiler is using that register as one it can allocate.
2952 // So we build it all by hand.
2953 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2954 // ArrayAddress dispatch(table, index);
2956 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2958 masm.lea(dest_reg, InternalAddress(table_base));
2959 masm.jmp(dispatch);
2960 %}
2962 enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
2963 MacroAssembler masm(&cbuf);
2965 Register switch_reg = as_Register($switch_val$$reg);
2966 Register dest_reg = as_Register($dest$$reg);
2967 address table_base = masm.address_table_constant(_index2label);
2969 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2970 // to do that and the compiler is using that register as one it can allocate.
2971 // So we build it all by hand.
2972 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
2973 // ArrayAddress dispatch(table, index);
2975 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
2976 masm.lea(dest_reg, InternalAddress(table_base));
2977 masm.jmp(dispatch);
2979 %}
2981 enc_class lock_prefix()
2982 %{
2983 if (os::is_MP()) {
2984 emit_opcode(cbuf, 0xF0); // lock
2985 }
2986 %}
2988 enc_class REX_mem(memory mem)
2989 %{
2990 if ($mem$$base >= 8) {
2991 if ($mem$$index < 8) {
2992 emit_opcode(cbuf, Assembler::REX_B);
2993 } else {
2994 emit_opcode(cbuf, Assembler::REX_XB);
2995 }
2996 } else {
2997 if ($mem$$index >= 8) {
2998 emit_opcode(cbuf, Assembler::REX_X);
2999 }
3000 }
3001 %}
3003 enc_class REX_mem_wide(memory mem)
3004 %{
3005 if ($mem$$base >= 8) {
3006 if ($mem$$index < 8) {
3007 emit_opcode(cbuf, Assembler::REX_WB);
3008 } else {
3009 emit_opcode(cbuf, Assembler::REX_WXB);
3010 }
3011 } else {
3012 if ($mem$$index < 8) {
3013 emit_opcode(cbuf, Assembler::REX_W);
3014 } else {
3015 emit_opcode(cbuf, Assembler::REX_WX);
3016 }
3017 }
3018 %}
3020 // for byte regs
3021 enc_class REX_breg(rRegI reg)
3022 %{
3023 if ($reg$$reg >= 4) {
3024 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3025 }
3026 %}
3028 // for byte regs
3029 enc_class REX_reg_breg(rRegI dst, rRegI src)
3030 %{
3031 if ($dst$$reg < 8) {
3032 if ($src$$reg >= 4) {
3033 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3034 }
3035 } else {
3036 if ($src$$reg < 8) {
3037 emit_opcode(cbuf, Assembler::REX_R);
3038 } else {
3039 emit_opcode(cbuf, Assembler::REX_RB);
3040 }
3041 }
3042 %}
3044 // for byte regs
3045 enc_class REX_breg_mem(rRegI reg, memory mem)
3046 %{
3047 if ($reg$$reg < 8) {
3048 if ($mem$$base < 8) {
3049 if ($mem$$index >= 8) {
3050 emit_opcode(cbuf, Assembler::REX_X);
3051 } else if ($reg$$reg >= 4) {
3052 emit_opcode(cbuf, Assembler::REX);
3053 }
3054 } else {
3055 if ($mem$$index < 8) {
3056 emit_opcode(cbuf, Assembler::REX_B);
3057 } else {
3058 emit_opcode(cbuf, Assembler::REX_XB);
3059 }
3060 }
3061 } else {
3062 if ($mem$$base < 8) {
3063 if ($mem$$index < 8) {
3064 emit_opcode(cbuf, Assembler::REX_R);
3065 } else {
3066 emit_opcode(cbuf, Assembler::REX_RX);
3067 }
3068 } else {
3069 if ($mem$$index < 8) {
3070 emit_opcode(cbuf, Assembler::REX_RB);
3071 } else {
3072 emit_opcode(cbuf, Assembler::REX_RXB);
3073 }
3074 }
3075 }
3076 %}
3078 enc_class REX_reg(rRegI reg)
3079 %{
3080 if ($reg$$reg >= 8) {
3081 emit_opcode(cbuf, Assembler::REX_B);
3082 }
3083 %}
3085 enc_class REX_reg_wide(rRegI reg)
3086 %{
3087 if ($reg$$reg < 8) {
3088 emit_opcode(cbuf, Assembler::REX_W);
3089 } else {
3090 emit_opcode(cbuf, Assembler::REX_WB);
3091 }
3092 %}
3094 enc_class REX_reg_reg(rRegI dst, rRegI src)
3095 %{
3096 if ($dst$$reg < 8) {
3097 if ($src$$reg >= 8) {
3098 emit_opcode(cbuf, Assembler::REX_B);
3099 }
3100 } else {
3101 if ($src$$reg < 8) {
3102 emit_opcode(cbuf, Assembler::REX_R);
3103 } else {
3104 emit_opcode(cbuf, Assembler::REX_RB);
3105 }
3106 }
3107 %}
3109 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3110 %{
3111 if ($dst$$reg < 8) {
3112 if ($src$$reg < 8) {
3113 emit_opcode(cbuf, Assembler::REX_W);
3114 } else {
3115 emit_opcode(cbuf, Assembler::REX_WB);
3116 }
3117 } else {
3118 if ($src$$reg < 8) {
3119 emit_opcode(cbuf, Assembler::REX_WR);
3120 } else {
3121 emit_opcode(cbuf, Assembler::REX_WRB);
3122 }
3123 }
3124 %}
3126 enc_class REX_reg_mem(rRegI reg, memory mem)
3127 %{
3128 if ($reg$$reg < 8) {
3129 if ($mem$$base < 8) {
3130 if ($mem$$index >= 8) {
3131 emit_opcode(cbuf, Assembler::REX_X);
3132 }
3133 } else {
3134 if ($mem$$index < 8) {
3135 emit_opcode(cbuf, Assembler::REX_B);
3136 } else {
3137 emit_opcode(cbuf, Assembler::REX_XB);
3138 }
3139 }
3140 } else {
3141 if ($mem$$base < 8) {
3142 if ($mem$$index < 8) {
3143 emit_opcode(cbuf, Assembler::REX_R);
3144 } else {
3145 emit_opcode(cbuf, Assembler::REX_RX);
3146 }
3147 } else {
3148 if ($mem$$index < 8) {
3149 emit_opcode(cbuf, Assembler::REX_RB);
3150 } else {
3151 emit_opcode(cbuf, Assembler::REX_RXB);
3152 }
3153 }
3154 }
3155 %}
3157 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3158 %{
3159 if ($reg$$reg < 8) {
3160 if ($mem$$base < 8) {
3161 if ($mem$$index < 8) {
3162 emit_opcode(cbuf, Assembler::REX_W);
3163 } else {
3164 emit_opcode(cbuf, Assembler::REX_WX);
3165 }
3166 } else {
3167 if ($mem$$index < 8) {
3168 emit_opcode(cbuf, Assembler::REX_WB);
3169 } else {
3170 emit_opcode(cbuf, Assembler::REX_WXB);
3171 }
3172 }
3173 } else {
3174 if ($mem$$base < 8) {
3175 if ($mem$$index < 8) {
3176 emit_opcode(cbuf, Assembler::REX_WR);
3177 } else {
3178 emit_opcode(cbuf, Assembler::REX_WRX);
3179 }
3180 } else {
3181 if ($mem$$index < 8) {
3182 emit_opcode(cbuf, Assembler::REX_WRB);
3183 } else {
3184 emit_opcode(cbuf, Assembler::REX_WRXB);
3185 }
3186 }
3187 }
3188 %}
3190 enc_class reg_mem(rRegI ereg, memory mem)
3191 %{
3192 // High registers handle in encode_RegMem
3193 int reg = $ereg$$reg;
3194 int base = $mem$$base;
3195 int index = $mem$$index;
3196 int scale = $mem$$scale;
3197 int disp = $mem$$disp;
3198 bool disp_is_oop = $mem->disp_is_oop();
3200 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3201 %}
3203 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3204 %{
3205 int rm_byte_opcode = $rm_opcode$$constant;
3207 // High registers handle in encode_RegMem
3208 int base = $mem$$base;
3209 int index = $mem$$index;
3210 int scale = $mem$$scale;
3211 int displace = $mem$$disp;
3213 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3214 // working with static
3215 // globals
3216 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3217 disp_is_oop);
3218 %}
3220 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3221 %{
3222 int reg_encoding = $dst$$reg;
3223 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3224 int index = 0x04; // 0x04 indicates no index
3225 int scale = 0x00; // 0x00 indicates no scale
3226 int displace = $src1$$constant; // 0x00 indicates no displacement
3227 bool disp_is_oop = false;
3228 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3229 disp_is_oop);
3230 %}
3232 enc_class neg_reg(rRegI dst)
3233 %{
3234 int dstenc = $dst$$reg;
3235 if (dstenc >= 8) {
3236 emit_opcode(cbuf, Assembler::REX_B);
3237 dstenc -= 8;
3238 }
3239 // NEG $dst
3240 emit_opcode(cbuf, 0xF7);
3241 emit_rm(cbuf, 0x3, 0x03, dstenc);
3242 %}
3244 enc_class neg_reg_wide(rRegI dst)
3245 %{
3246 int dstenc = $dst$$reg;
3247 if (dstenc < 8) {
3248 emit_opcode(cbuf, Assembler::REX_W);
3249 } else {
3250 emit_opcode(cbuf, Assembler::REX_WB);
3251 dstenc -= 8;
3252 }
3253 // NEG $dst
3254 emit_opcode(cbuf, 0xF7);
3255 emit_rm(cbuf, 0x3, 0x03, dstenc);
3256 %}
3258 enc_class setLT_reg(rRegI dst)
3259 %{
3260 int dstenc = $dst$$reg;
3261 if (dstenc >= 8) {
3262 emit_opcode(cbuf, Assembler::REX_B);
3263 dstenc -= 8;
3264 } else if (dstenc >= 4) {
3265 emit_opcode(cbuf, Assembler::REX);
3266 }
3267 // SETLT $dst
3268 emit_opcode(cbuf, 0x0F);
3269 emit_opcode(cbuf, 0x9C);
3270 emit_rm(cbuf, 0x3, 0x0, dstenc);
3271 %}
3273 enc_class setNZ_reg(rRegI dst)
3274 %{
3275 int dstenc = $dst$$reg;
3276 if (dstenc >= 8) {
3277 emit_opcode(cbuf, Assembler::REX_B);
3278 dstenc -= 8;
3279 } else if (dstenc >= 4) {
3280 emit_opcode(cbuf, Assembler::REX);
3281 }
3282 // SETNZ $dst
3283 emit_opcode(cbuf, 0x0F);
3284 emit_opcode(cbuf, 0x95);
3285 emit_rm(cbuf, 0x3, 0x0, dstenc);
3286 %}
3288 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3289 rcx_RegI tmp)
3290 %{
3291 // cadd_cmpLT
3293 int tmpReg = $tmp$$reg;
3295 int penc = $p$$reg;
3296 int qenc = $q$$reg;
3297 int yenc = $y$$reg;
3299 // subl $p,$q
3300 if (penc < 8) {
3301 if (qenc >= 8) {
3302 emit_opcode(cbuf, Assembler::REX_B);
3303 }
3304 } else {
3305 if (qenc < 8) {
3306 emit_opcode(cbuf, Assembler::REX_R);
3307 } else {
3308 emit_opcode(cbuf, Assembler::REX_RB);
3309 }
3310 }
3311 emit_opcode(cbuf, 0x2B);
3312 emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3314 // sbbl $tmp, $tmp
3315 emit_opcode(cbuf, 0x1B);
3316 emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3318 // andl $tmp, $y
3319 if (yenc >= 8) {
3320 emit_opcode(cbuf, Assembler::REX_B);
3321 }
3322 emit_opcode(cbuf, 0x23);
3323 emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3325 // addl $p,$tmp
3326 if (penc >= 8) {
3327 emit_opcode(cbuf, Assembler::REX_R);
3328 }
3329 emit_opcode(cbuf, 0x03);
3330 emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3331 %}
3333 // Compare the lonogs and set -1, 0, or 1 into dst
3334 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3335 %{
3336 int src1enc = $src1$$reg;
3337 int src2enc = $src2$$reg;
3338 int dstenc = $dst$$reg;
3340 // cmpq $src1, $src2
3341 if (src1enc < 8) {
3342 if (src2enc < 8) {
3343 emit_opcode(cbuf, Assembler::REX_W);
3344 } else {
3345 emit_opcode(cbuf, Assembler::REX_WB);
3346 }
3347 } else {
3348 if (src2enc < 8) {
3349 emit_opcode(cbuf, Assembler::REX_WR);
3350 } else {
3351 emit_opcode(cbuf, Assembler::REX_WRB);
3352 }
3353 }
3354 emit_opcode(cbuf, 0x3B);
3355 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3357 // movl $dst, -1
3358 if (dstenc >= 8) {
3359 emit_opcode(cbuf, Assembler::REX_B);
3360 }
3361 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3362 emit_d32(cbuf, -1);
3364 // jl,s done
3365 emit_opcode(cbuf, 0x7C);
3366 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3368 // setne $dst
3369 if (dstenc >= 4) {
3370 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3371 }
3372 emit_opcode(cbuf, 0x0F);
3373 emit_opcode(cbuf, 0x95);
3374 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3376 // movzbl $dst, $dst
3377 if (dstenc >= 4) {
3378 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3379 }
3380 emit_opcode(cbuf, 0x0F);
3381 emit_opcode(cbuf, 0xB6);
3382 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3383 %}
3385 enc_class Push_ResultXD(regD dst) %{
3386 int dstenc = $dst$$reg;
3388 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3390 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3391 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3392 if (dstenc >= 8) {
3393 emit_opcode(cbuf, Assembler::REX_R);
3394 }
3395 emit_opcode (cbuf, 0x0F );
3396 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3397 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3399 // add rsp,8
3400 emit_opcode(cbuf, Assembler::REX_W);
3401 emit_opcode(cbuf,0x83);
3402 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3403 emit_d8(cbuf,0x08);
3404 %}
3406 enc_class Push_SrcXD(regD src) %{
3407 int srcenc = $src$$reg;
3409 // subq rsp,#8
3410 emit_opcode(cbuf, Assembler::REX_W);
3411 emit_opcode(cbuf, 0x83);
3412 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3413 emit_d8(cbuf, 0x8);
3415 // movsd [rsp],src
3416 emit_opcode(cbuf, 0xF2);
3417 if (srcenc >= 8) {
3418 emit_opcode(cbuf, Assembler::REX_R);
3419 }
3420 emit_opcode(cbuf, 0x0F);
3421 emit_opcode(cbuf, 0x11);
3422 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3424 // fldd [rsp]
3425 emit_opcode(cbuf, 0x66);
3426 emit_opcode(cbuf, 0xDD);
3427 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3428 %}
3431 enc_class movq_ld(regD dst, memory mem) %{
3432 MacroAssembler _masm(&cbuf);
3433 __ movq($dst$$XMMRegister, $mem$$Address);
3434 %}
3436 enc_class movq_st(memory mem, regD src) %{
3437 MacroAssembler _masm(&cbuf);
3438 __ movq($mem$$Address, $src$$XMMRegister);
3439 %}
3441 enc_class pshufd_8x8(regF dst, regF src) %{
3442 MacroAssembler _masm(&cbuf);
3444 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3445 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3446 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3447 %}
3449 enc_class pshufd_4x16(regF dst, regF src) %{
3450 MacroAssembler _masm(&cbuf);
3452 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3453 %}
3455 enc_class pshufd(regD dst, regD src, int mode) %{
3456 MacroAssembler _masm(&cbuf);
3458 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3459 %}
3461 enc_class pxor(regD dst, regD src) %{
3462 MacroAssembler _masm(&cbuf);
3464 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3465 %}
3467 enc_class mov_i2x(regD dst, rRegI src) %{
3468 MacroAssembler _masm(&cbuf);
3470 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3471 %}
3473 // obj: object to lock
3474 // box: box address (header location) -- killed
3475 // tmp: rax -- killed
3476 // scr: rbx -- killed
3477 //
3478 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3479 // from i486.ad. See that file for comments.
3480 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3481 // use the shorter encoding. (Movl clears the high-order 32-bits).
3484 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3485 %{
3486 Register objReg = as_Register((int)$obj$$reg);
3487 Register boxReg = as_Register((int)$box$$reg);
3488 Register tmpReg = as_Register($tmp$$reg);
3489 Register scrReg = as_Register($scr$$reg);
3490 MacroAssembler masm(&cbuf);
3492 // Verify uniqueness of register assignments -- necessary but not sufficient
3493 assert (objReg != boxReg && objReg != tmpReg &&
3494 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3496 if (_counters != NULL) {
3497 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3498 }
3499 if (EmitSync & 1) {
3500 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3501 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3502 masm.cmpptr(rsp, (int32_t)NULL_WORD) ;
3503 } else
3504 if (EmitSync & 2) {
3505 Label DONE_LABEL;
3506 if (UseBiasedLocking) {
3507 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3508 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3509 }
3510 // QQQ was movl...
3511 masm.movptr(tmpReg, 0x1);
3512 masm.orptr(tmpReg, Address(objReg, 0));
3513 masm.movptr(Address(boxReg, 0), tmpReg);
3514 if (os::is_MP()) {
3515 masm.lock();
3516 }
3517 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3518 masm.jcc(Assembler::equal, DONE_LABEL);
3520 // Recursive locking
3521 masm.subptr(tmpReg, rsp);
3522 masm.andptr(tmpReg, 7 - os::vm_page_size());
3523 masm.movptr(Address(boxReg, 0), tmpReg);
3525 masm.bind(DONE_LABEL);
3526 masm.nop(); // avoid branch to branch
3527 } else {
3528 Label DONE_LABEL, IsInflated, Egress;
3530 masm.movptr(tmpReg, Address(objReg, 0)) ;
3531 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3532 masm.jcc (Assembler::notZero, IsInflated) ;
3534 // it's stack-locked, biased or neutral
3535 // TODO: optimize markword triage order to reduce the number of
3536 // conditional branches in the most common cases.
3537 // Beware -- there's a subtle invariant that fetch of the markword
3538 // at [FETCH], below, will never observe a biased encoding (*101b).
3539 // If this invariant is not held we'll suffer exclusion (safety) failure.
3541 if (UseBiasedLocking && !UseOptoBiasInlining) {
3542 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3543 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH]
3544 }
3546 // was q will it destroy high?
3547 masm.orl (tmpReg, 1) ;
3548 masm.movptr(Address(boxReg, 0), tmpReg) ;
3549 if (os::is_MP()) { masm.lock(); }
3550 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3551 if (_counters != NULL) {
3552 masm.cond_inc32(Assembler::equal,
3553 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3554 }
3555 masm.jcc (Assembler::equal, DONE_LABEL);
3557 // Recursive locking
3558 masm.subptr(tmpReg, rsp);
3559 masm.andptr(tmpReg, 7 - os::vm_page_size());
3560 masm.movptr(Address(boxReg, 0), tmpReg);
3561 if (_counters != NULL) {
3562 masm.cond_inc32(Assembler::equal,
3563 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3564 }
3565 masm.jmp (DONE_LABEL) ;
3567 masm.bind (IsInflated) ;
3568 // It's inflated
3570 // TODO: someday avoid the ST-before-CAS penalty by
3571 // relocating (deferring) the following ST.
3572 // We should also think about trying a CAS without having
3573 // fetched _owner. If the CAS is successful we may
3574 // avoid an RTO->RTS upgrade on the $line.
3575 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3576 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3578 masm.mov (boxReg, tmpReg) ;
3579 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3580 masm.testptr(tmpReg, tmpReg) ;
3581 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3583 // It's inflated and appears unlocked
3584 if (os::is_MP()) { masm.lock(); }
3585 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3586 // Intentional fall-through into DONE_LABEL ...
3588 masm.bind (DONE_LABEL) ;
3589 masm.nop () ; // avoid jmp to jmp
3590 }
3591 %}
3593 // obj: object to unlock
3594 // box: box address (displaced header location), killed
3595 // RBX: killed tmp; cannot be obj nor box
3596 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3597 %{
3599 Register objReg = as_Register($obj$$reg);
3600 Register boxReg = as_Register($box$$reg);
3601 Register tmpReg = as_Register($tmp$$reg);
3602 MacroAssembler masm(&cbuf);
3604 if (EmitSync & 4) {
3605 masm.cmpptr(rsp, 0) ;
3606 } else
3607 if (EmitSync & 8) {
3608 Label DONE_LABEL;
3609 if (UseBiasedLocking) {
3610 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3611 }
3613 // Check whether the displaced header is 0
3614 //(=> recursive unlock)
3615 masm.movptr(tmpReg, Address(boxReg, 0));
3616 masm.testptr(tmpReg, tmpReg);
3617 masm.jcc(Assembler::zero, DONE_LABEL);
3619 // If not recursive lock, reset the header to displaced header
3620 if (os::is_MP()) {
3621 masm.lock();
3622 }
3623 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3624 masm.bind(DONE_LABEL);
3625 masm.nop(); // avoid branch to branch
3626 } else {
3627 Label DONE_LABEL, Stacked, CheckSucc ;
3629 if (UseBiasedLocking && !UseOptoBiasInlining) {
3630 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3631 }
3633 masm.movptr(tmpReg, Address(objReg, 0)) ;
3634 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ;
3635 masm.jcc (Assembler::zero, DONE_LABEL) ;
3636 masm.testl (tmpReg, 0x02) ;
3637 masm.jcc (Assembler::zero, Stacked) ;
3639 // It's inflated
3640 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3641 masm.xorptr(boxReg, r15_thread) ;
3642 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3643 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3644 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3645 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3646 masm.jcc (Assembler::notZero, CheckSucc) ;
3647 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3648 masm.jmp (DONE_LABEL) ;
3650 if ((EmitSync & 65536) == 0) {
3651 Label LSuccess, LGoSlowPath ;
3652 masm.bind (CheckSucc) ;
3653 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3654 masm.jcc (Assembler::zero, LGoSlowPath) ;
3656 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3657 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3658 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3659 // are all faster when the write buffer is populated.
3660 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3661 if (os::is_MP()) {
3662 masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3663 }
3664 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3665 masm.jcc (Assembler::notZero, LSuccess) ;
3667 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX
3668 if (os::is_MP()) { masm.lock(); }
3669 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3670 masm.jcc (Assembler::notEqual, LSuccess) ;
3671 // Intentional fall-through into slow-path
3673 masm.bind (LGoSlowPath) ;
3674 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3675 masm.jmp (DONE_LABEL) ;
3677 masm.bind (LSuccess) ;
3678 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3679 masm.jmp (DONE_LABEL) ;
3680 }
3682 masm.bind (Stacked) ;
3683 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch
3684 if (os::is_MP()) { masm.lock(); }
3685 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3687 if (EmitSync & 65536) {
3688 masm.bind (CheckSucc) ;
3689 }
3690 masm.bind(DONE_LABEL);
3691 if (EmitSync & 32768) {
3692 masm.nop(); // avoid branch to branch
3693 }
3694 }
3695 %}
3697 enc_class enc_String_Compare()
3698 %{
3699 Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
3700 POP_LABEL, DONE_LABEL, CONT_LABEL,
3701 WHILE_HEAD_LABEL;
3702 MacroAssembler masm(&cbuf);
3704 // Get the first character position in both strings
3705 // [8] char array, [12] offset, [16] count
3706 int value_offset = java_lang_String::value_offset_in_bytes();
3707 int offset_offset = java_lang_String::offset_offset_in_bytes();
3708 int count_offset = java_lang_String::count_offset_in_bytes();
3709 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3711 masm.load_heap_oop(rax, Address(rsi, value_offset));
3712 masm.movl(rcx, Address(rsi, offset_offset));
3713 masm.lea(rax, Address(rax, rcx, Address::times_2, base_offset));
3714 masm.load_heap_oop(rbx, Address(rdi, value_offset));
3715 masm.movl(rcx, Address(rdi, offset_offset));
3716 masm.lea(rbx, Address(rbx, rcx, Address::times_2, base_offset));
3718 // Compute the minimum of the string lengths(rsi) and the
3719 // difference of the string lengths (stack)
3721 masm.movl(rdi, Address(rdi, count_offset));
3722 masm.movl(rsi, Address(rsi, count_offset));
3723 masm.movl(rcx, rdi);
3724 masm.subl(rdi, rsi);
3725 masm.push(rdi);
3726 masm.cmov(Assembler::lessEqual, rsi, rcx);
3728 // Is the minimum length zero?
3729 masm.bind(RCX_GOOD_LABEL);
3730 masm.testl(rsi, rsi);
3731 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3733 // Load first characters
3734 masm.load_unsigned_short(rcx, Address(rbx, 0));
3735 masm.load_unsigned_short(rdi, Address(rax, 0));
3737 // Compare first characters
3738 masm.subl(rcx, rdi);
3739 masm.jcc(Assembler::notZero, POP_LABEL);
3740 masm.decrementl(rsi);
3741 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3743 {
3744 // Check after comparing first character to see if strings are equivalent
3745 Label LSkip2;
3746 // Check if the strings start at same location
3747 masm.cmpptr(rbx, rax);
3748 masm.jcc(Assembler::notEqual, LSkip2);
3750 // Check if the length difference is zero (from stack)
3751 masm.cmpl(Address(rsp, 0), 0x0);
3752 masm.jcc(Assembler::equal, LENGTH_DIFF_LABEL);
3754 // Strings might not be equivalent
3755 masm.bind(LSkip2);
3756 }
3758 // Shift RAX and RBX to the end of the arrays, negate min
3759 masm.lea(rax, Address(rax, rsi, Address::times_2, 2));
3760 masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2));
3761 masm.negptr(rsi);
3763 // Compare the rest of the characters
3764 masm.bind(WHILE_HEAD_LABEL);
3765 masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0));
3766 masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0));
3767 masm.subl(rcx, rdi);
3768 masm.jcc(Assembler::notZero, POP_LABEL);
3769 masm.increment(rsi);
3770 masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
3772 // Strings are equal up to min length. Return the length difference.
3773 masm.bind(LENGTH_DIFF_LABEL);
3774 masm.pop(rcx);
3775 masm.jmp(DONE_LABEL);
3777 // Discard the stored length difference
3778 masm.bind(POP_LABEL);
3779 masm.addptr(rsp, 8);
3781 // That's it
3782 masm.bind(DONE_LABEL);
3783 %}
3785 enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, rbx_RegI tmp2, rcx_RegI result) %{
3786 Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP;
3787 MacroAssembler masm(&cbuf);
3789 Register ary1Reg = as_Register($ary1$$reg);
3790 Register ary2Reg = as_Register($ary2$$reg);
3791 Register tmp1Reg = as_Register($tmp1$$reg);
3792 Register tmp2Reg = as_Register($tmp2$$reg);
3793 Register resultReg = as_Register($result$$reg);
3795 int length_offset = arrayOopDesc::length_offset_in_bytes();
3796 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3798 // Check the input args
3799 masm.cmpq(ary1Reg, ary2Reg);
3800 masm.jcc(Assembler::equal, TRUE_LABEL);
3801 masm.testq(ary1Reg, ary1Reg);
3802 masm.jcc(Assembler::zero, FALSE_LABEL);
3803 masm.testq(ary2Reg, ary2Reg);
3804 masm.jcc(Assembler::zero, FALSE_LABEL);
3806 // Check the lengths
3807 masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
3808 masm.movl(resultReg, Address(ary2Reg, length_offset));
3809 masm.cmpl(tmp2Reg, resultReg);
3810 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3811 masm.testl(resultReg, resultReg);
3812 masm.jcc(Assembler::zero, TRUE_LABEL);
3814 // Get the number of 4 byte vectors to compare
3815 masm.shrl(resultReg, 1);
3817 // Check for odd-length arrays
3818 masm.andl(tmp2Reg, 1);
3819 masm.testl(tmp2Reg, tmp2Reg);
3820 masm.jcc(Assembler::zero, COMPARE_LOOP_HDR);
3822 // Compare 2-byte "tail" at end of arrays
3823 masm.load_unsigned_short(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
3824 masm.load_unsigned_short(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
3825 masm.cmpl(tmp1Reg, tmp2Reg);
3826 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3827 masm.testl(resultReg, resultReg);
3828 masm.jcc(Assembler::zero, TRUE_LABEL);
3830 // Setup compare loop
3831 masm.bind(COMPARE_LOOP_HDR);
3832 // Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays
3833 masm.leaq(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
3834 masm.leaq(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
3835 masm.negq(resultReg);
3837 // 4-byte-wide compare loop
3838 masm.bind(COMPARE_LOOP);
3839 masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0));
3840 masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0));
3841 masm.cmpl(ary1Reg, ary2Reg);
3842 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3843 masm.incrementq(resultReg);
3844 masm.jcc(Assembler::notZero, COMPARE_LOOP);
3846 masm.bind(TRUE_LABEL);
3847 masm.movl(resultReg, 1); // return true
3848 masm.jmp(DONE_LABEL);
3850 masm.bind(FALSE_LABEL);
3851 masm.xorl(resultReg, resultReg); // return false
3853 // That's it
3854 masm.bind(DONE_LABEL);
3855 %}
3857 enc_class enc_rethrow()
3858 %{
3859 cbuf.set_inst_mark();
3860 emit_opcode(cbuf, 0xE9); // jmp entry
3861 emit_d32_reloc(cbuf,
3862 (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
3863 runtime_call_Relocation::spec(),
3864 RELOC_DISP32);
3865 %}
3867 enc_class absF_encoding(regF dst)
3868 %{
3869 int dstenc = $dst$$reg;
3870 address signmask_address = (address) StubRoutines::x86::float_sign_mask();
3872 cbuf.set_inst_mark();
3873 if (dstenc >= 8) {
3874 emit_opcode(cbuf, Assembler::REX_R);
3875 dstenc -= 8;
3876 }
3877 // XXX reg_mem doesn't support RIP-relative addressing yet
3878 emit_opcode(cbuf, 0x0F);
3879 emit_opcode(cbuf, 0x54);
3880 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3881 emit_d32_reloc(cbuf, signmask_address);
3882 %}
3884 enc_class absD_encoding(regD dst)
3885 %{
3886 int dstenc = $dst$$reg;
3887 address signmask_address = (address) StubRoutines::x86::double_sign_mask();
3889 cbuf.set_inst_mark();
3890 emit_opcode(cbuf, 0x66);
3891 if (dstenc >= 8) {
3892 emit_opcode(cbuf, Assembler::REX_R);
3893 dstenc -= 8;
3894 }
3895 // XXX reg_mem doesn't support RIP-relative addressing yet
3896 emit_opcode(cbuf, 0x0F);
3897 emit_opcode(cbuf, 0x54);
3898 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3899 emit_d32_reloc(cbuf, signmask_address);
3900 %}
3902 enc_class negF_encoding(regF dst)
3903 %{
3904 int dstenc = $dst$$reg;
3905 address signflip_address = (address) StubRoutines::x86::float_sign_flip();
3907 cbuf.set_inst_mark();
3908 if (dstenc >= 8) {
3909 emit_opcode(cbuf, Assembler::REX_R);
3910 dstenc -= 8;
3911 }
3912 // XXX reg_mem doesn't support RIP-relative addressing yet
3913 emit_opcode(cbuf, 0x0F);
3914 emit_opcode(cbuf, 0x57);
3915 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3916 emit_d32_reloc(cbuf, signflip_address);
3917 %}
3919 enc_class negD_encoding(regD dst)
3920 %{
3921 int dstenc = $dst$$reg;
3922 address signflip_address = (address) StubRoutines::x86::double_sign_flip();
3924 cbuf.set_inst_mark();
3925 emit_opcode(cbuf, 0x66);
3926 if (dstenc >= 8) {
3927 emit_opcode(cbuf, Assembler::REX_R);
3928 dstenc -= 8;
3929 }
3930 // XXX reg_mem doesn't support RIP-relative addressing yet
3931 emit_opcode(cbuf, 0x0F);
3932 emit_opcode(cbuf, 0x57);
3933 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3934 emit_d32_reloc(cbuf, signflip_address);
3935 %}
3937 enc_class f2i_fixup(rRegI dst, regF src)
3938 %{
3939 int dstenc = $dst$$reg;
3940 int srcenc = $src$$reg;
3942 // cmpl $dst, #0x80000000
3943 if (dstenc >= 8) {
3944 emit_opcode(cbuf, Assembler::REX_B);
3945 }
3946 emit_opcode(cbuf, 0x81);
3947 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3948 emit_d32(cbuf, 0x80000000);
3950 // jne,s done
3951 emit_opcode(cbuf, 0x75);
3952 if (srcenc < 8 && dstenc < 8) {
3953 emit_d8(cbuf, 0xF);
3954 } else if (srcenc >= 8 && dstenc >= 8) {
3955 emit_d8(cbuf, 0x11);
3956 } else {
3957 emit_d8(cbuf, 0x10);
3958 }
3960 // subq rsp, #8
3961 emit_opcode(cbuf, Assembler::REX_W);
3962 emit_opcode(cbuf, 0x83);
3963 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3964 emit_d8(cbuf, 8);
3966 // movss [rsp], $src
3967 emit_opcode(cbuf, 0xF3);
3968 if (srcenc >= 8) {
3969 emit_opcode(cbuf, Assembler::REX_R);
3970 }
3971 emit_opcode(cbuf, 0x0F);
3972 emit_opcode(cbuf, 0x11);
3973 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3975 // call f2i_fixup
3976 cbuf.set_inst_mark();
3977 emit_opcode(cbuf, 0xE8);
3978 emit_d32_reloc(cbuf,
3979 (int)
3980 (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4),
3981 runtime_call_Relocation::spec(),
3982 RELOC_DISP32);
3984 // popq $dst
3985 if (dstenc >= 8) {
3986 emit_opcode(cbuf, Assembler::REX_B);
3987 }
3988 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3990 // done:
3991 %}
3993 enc_class f2l_fixup(rRegL dst, regF src)
3994 %{
3995 int dstenc = $dst$$reg;
3996 int srcenc = $src$$reg;
3997 address const_address = (address) StubRoutines::x86::double_sign_flip();
3999 // cmpq $dst, [0x8000000000000000]
4000 cbuf.set_inst_mark();
4001 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4002 emit_opcode(cbuf, 0x39);
4003 // XXX reg_mem doesn't support RIP-relative addressing yet
4004 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4005 emit_d32_reloc(cbuf, const_address);
4008 // jne,s done
4009 emit_opcode(cbuf, 0x75);
4010 if (srcenc < 8 && dstenc < 8) {
4011 emit_d8(cbuf, 0xF);
4012 } else if (srcenc >= 8 && dstenc >= 8) {
4013 emit_d8(cbuf, 0x11);
4014 } else {
4015 emit_d8(cbuf, 0x10);
4016 }
4018 // subq rsp, #8
4019 emit_opcode(cbuf, Assembler::REX_W);
4020 emit_opcode(cbuf, 0x83);
4021 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4022 emit_d8(cbuf, 8);
4024 // movss [rsp], $src
4025 emit_opcode(cbuf, 0xF3);
4026 if (srcenc >= 8) {
4027 emit_opcode(cbuf, Assembler::REX_R);
4028 }
4029 emit_opcode(cbuf, 0x0F);
4030 emit_opcode(cbuf, 0x11);
4031 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4033 // call f2l_fixup
4034 cbuf.set_inst_mark();
4035 emit_opcode(cbuf, 0xE8);
4036 emit_d32_reloc(cbuf,
4037 (int)
4038 (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4),
4039 runtime_call_Relocation::spec(),
4040 RELOC_DISP32);
4042 // popq $dst
4043 if (dstenc >= 8) {
4044 emit_opcode(cbuf, Assembler::REX_B);
4045 }
4046 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4048 // done:
4049 %}
4051 enc_class d2i_fixup(rRegI dst, regD src)
4052 %{
4053 int dstenc = $dst$$reg;
4054 int srcenc = $src$$reg;
4056 // cmpl $dst, #0x80000000
4057 if (dstenc >= 8) {
4058 emit_opcode(cbuf, Assembler::REX_B);
4059 }
4060 emit_opcode(cbuf, 0x81);
4061 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
4062 emit_d32(cbuf, 0x80000000);
4064 // jne,s done
4065 emit_opcode(cbuf, 0x75);
4066 if (srcenc < 8 && dstenc < 8) {
4067 emit_d8(cbuf, 0xF);
4068 } else if (srcenc >= 8 && dstenc >= 8) {
4069 emit_d8(cbuf, 0x11);
4070 } else {
4071 emit_d8(cbuf, 0x10);
4072 }
4074 // subq rsp, #8
4075 emit_opcode(cbuf, Assembler::REX_W);
4076 emit_opcode(cbuf, 0x83);
4077 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4078 emit_d8(cbuf, 8);
4080 // movsd [rsp], $src
4081 emit_opcode(cbuf, 0xF2);
4082 if (srcenc >= 8) {
4083 emit_opcode(cbuf, Assembler::REX_R);
4084 }
4085 emit_opcode(cbuf, 0x0F);
4086 emit_opcode(cbuf, 0x11);
4087 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4089 // call d2i_fixup
4090 cbuf.set_inst_mark();
4091 emit_opcode(cbuf, 0xE8);
4092 emit_d32_reloc(cbuf,
4093 (int)
4094 (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4),
4095 runtime_call_Relocation::spec(),
4096 RELOC_DISP32);
4098 // popq $dst
4099 if (dstenc >= 8) {
4100 emit_opcode(cbuf, Assembler::REX_B);
4101 }
4102 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4104 // done:
4105 %}
4107 enc_class d2l_fixup(rRegL dst, regD src)
4108 %{
4109 int dstenc = $dst$$reg;
4110 int srcenc = $src$$reg;
4111 address const_address = (address) StubRoutines::x86::double_sign_flip();
4113 // cmpq $dst, [0x8000000000000000]
4114 cbuf.set_inst_mark();
4115 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4116 emit_opcode(cbuf, 0x39);
4117 // XXX reg_mem doesn't support RIP-relative addressing yet
4118 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4119 emit_d32_reloc(cbuf, const_address);
4122 // jne,s done
4123 emit_opcode(cbuf, 0x75);
4124 if (srcenc < 8 && dstenc < 8) {
4125 emit_d8(cbuf, 0xF);
4126 } else if (srcenc >= 8 && dstenc >= 8) {
4127 emit_d8(cbuf, 0x11);
4128 } else {
4129 emit_d8(cbuf, 0x10);
4130 }
4132 // subq rsp, #8
4133 emit_opcode(cbuf, Assembler::REX_W);
4134 emit_opcode(cbuf, 0x83);
4135 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4136 emit_d8(cbuf, 8);
4138 // movsd [rsp], $src
4139 emit_opcode(cbuf, 0xF2);
4140 if (srcenc >= 8) {
4141 emit_opcode(cbuf, Assembler::REX_R);
4142 }
4143 emit_opcode(cbuf, 0x0F);
4144 emit_opcode(cbuf, 0x11);
4145 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4147 // call d2l_fixup
4148 cbuf.set_inst_mark();
4149 emit_opcode(cbuf, 0xE8);
4150 emit_d32_reloc(cbuf,
4151 (int)
4152 (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4),
4153 runtime_call_Relocation::spec(),
4154 RELOC_DISP32);
4156 // popq $dst
4157 if (dstenc >= 8) {
4158 emit_opcode(cbuf, Assembler::REX_B);
4159 }
4160 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4162 // done:
4163 %}
4165 enc_class enc_membar_acquire
4166 %{
4167 // [jk] not needed currently, if you enable this and it really
4168 // emits code don't forget to the remove the "size(0)" line in
4169 // membar_acquire()
4170 // MacroAssembler masm(&cbuf);
4171 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4172 // Assembler::LoadLoad));
4173 %}
4175 enc_class enc_membar_release
4176 %{
4177 // [jk] not needed currently, if you enable this and it really
4178 // emits code don't forget to the remove the "size(0)" line in
4179 // membar_release()
4180 // MacroAssembler masm(&cbuf);
4181 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4182 // Assembler::StoreStore));
4183 %}
4185 enc_class enc_membar_volatile
4186 %{
4187 MacroAssembler masm(&cbuf);
4188 masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
4189 Assembler::StoreStore));
4190 %}
4192 // Safepoint Poll. This polls the safepoint page, and causes an
4193 // exception if it is not readable. Unfortunately, it kills
4194 // RFLAGS in the process.
4195 enc_class enc_safepoint_poll
4196 %{
4197 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
4198 // XXX reg_mem doesn't support RIP-relative addressing yet
4199 cbuf.set_inst_mark();
4200 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
4201 emit_opcode(cbuf, 0x85); // testl
4202 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
4203 // cbuf.inst_mark() is beginning of instruction
4204 emit_d32_reloc(cbuf, os::get_polling_page());
4205 // relocInfo::poll_type,
4206 %}
4207 %}
4211 //----------FRAME--------------------------------------------------------------
4212 // Definition of frame structure and management information.
4213 //
4214 // S T A C K L A Y O U T Allocators stack-slot number
4215 // | (to get allocators register number
4216 // G Owned by | | v add OptoReg::stack0())
4217 // r CALLER | |
4218 // o | +--------+ pad to even-align allocators stack-slot
4219 // w V | pad0 | numbers; owned by CALLER
4220 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
4221 // h ^ | in | 5
4222 // | | args | 4 Holes in incoming args owned by SELF
4223 // | | | | 3
4224 // | | +--------+
4225 // V | | old out| Empty on Intel, window on Sparc
4226 // | old |preserve| Must be even aligned.
4227 // | SP-+--------+----> Matcher::_old_SP, even aligned
4228 // | | in | 3 area for Intel ret address
4229 // Owned by |preserve| Empty on Sparc.
4230 // SELF +--------+
4231 // | | pad2 | 2 pad to align old SP
4232 // | +--------+ 1
4233 // | | locks | 0
4234 // | +--------+----> OptoReg::stack0(), even aligned
4235 // | | pad1 | 11 pad to align new SP
4236 // | +--------+
4237 // | | | 10
4238 // | | spills | 9 spills
4239 // V | | 8 (pad0 slot for callee)
4240 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
4241 // ^ | out | 7
4242 // | | args | 6 Holes in outgoing args owned by CALLEE
4243 // Owned by +--------+
4244 // CALLEE | new out| 6 Empty on Intel, window on Sparc
4245 // | new |preserve| Must be even-aligned.
4246 // | SP-+--------+----> Matcher::_new_SP, even aligned
4247 // | | |
4248 //
4249 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
4250 // known from SELF's arguments and the Java calling convention.
4251 // Region 6-7 is determined per call site.
4252 // Note 2: If the calling convention leaves holes in the incoming argument
4253 // area, those holes are owned by SELF. Holes in the outgoing area
4254 // are owned by the CALLEE. Holes should not be nessecary in the
4255 // incoming area, as the Java calling convention is completely under
4256 // the control of the AD file. Doubles can be sorted and packed to
4257 // avoid holes. Holes in the outgoing arguments may be nessecary for
4258 // varargs C calling conventions.
4259 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
4260 // even aligned with pad0 as needed.
4261 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
4262 // region 6-11 is even aligned; it may be padded out more so that
4263 // the region from SP to FP meets the minimum stack alignment.
4264 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
4265 // alignment. Region 11, pad1, may be dynamically extended so that
4266 // SP meets the minimum alignment.
4268 frame
4269 %{
4270 // What direction does stack grow in (assumed to be same for C & Java)
4271 stack_direction(TOWARDS_LOW);
4273 // These three registers define part of the calling convention
4274 // between compiled code and the interpreter.
4275 inline_cache_reg(RAX); // Inline Cache Register
4276 interpreter_method_oop_reg(RBX); // Method Oop Register when
4277 // calling interpreter
4279 // Optional: name the operand used by cisc-spilling to access
4280 // [stack_pointer + offset]
4281 cisc_spilling_operand_name(indOffset32);
4283 // Number of stack slots consumed by locking an object
4284 sync_stack_slots(2);
4286 // Compiled code's Frame Pointer
4287 frame_pointer(RSP);
4289 // Interpreter stores its frame pointer in a register which is
4290 // stored to the stack by I2CAdaptors.
4291 // I2CAdaptors convert from interpreted java to compiled java.
4292 interpreter_frame_pointer(RBP);
4294 // Stack alignment requirement
4295 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
4297 // Number of stack slots between incoming argument block and the start of
4298 // a new frame. The PROLOG must add this many slots to the stack. The
4299 // EPILOG must remove this many slots. amd64 needs two slots for
4300 // return address.
4301 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
4303 // Number of outgoing stack slots killed above the out_preserve_stack_slots
4304 // for calls to C. Supports the var-args backing area for register parms.
4305 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
4307 // The after-PROLOG location of the return address. Location of
4308 // return address specifies a type (REG or STACK) and a number
4309 // representing the register number (i.e. - use a register name) or
4310 // stack slot.
4311 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4312 // Otherwise, it is above the locks and verification slot and alignment word
4313 return_addr(STACK - 2 +
4314 round_to(2 + 2 * VerifyStackAtCalls +
4315 Compile::current()->fixed_slots(),
4316 WordsPerLong * 2));
4318 // Body of function which returns an integer array locating
4319 // arguments either in registers or in stack slots. Passed an array
4320 // of ideal registers called "sig" and a "length" count. Stack-slot
4321 // offsets are based on outgoing arguments, i.e. a CALLER setting up
4322 // arguments for a CALLEE. Incoming stack arguments are
4323 // automatically biased by the preserve_stack_slots field above.
4325 calling_convention
4326 %{
4327 // No difference between ingoing/outgoing just pass false
4328 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4329 %}
4331 c_calling_convention
4332 %{
4333 // This is obviously always outgoing
4334 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4335 %}
4337 // Location of compiled Java return values. Same as C for now.
4338 return_value
4339 %{
4340 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4341 "only return normal values");
4343 static const int lo[Op_RegL + 1] = {
4344 0,
4345 0,
4346 RAX_num, // Op_RegN
4347 RAX_num, // Op_RegI
4348 RAX_num, // Op_RegP
4349 XMM0_num, // Op_RegF
4350 XMM0_num, // Op_RegD
4351 RAX_num // Op_RegL
4352 };
4353 static const int hi[Op_RegL + 1] = {
4354 0,
4355 0,
4356 OptoReg::Bad, // Op_RegN
4357 OptoReg::Bad, // Op_RegI
4358 RAX_H_num, // Op_RegP
4359 OptoReg::Bad, // Op_RegF
4360 XMM0_H_num, // Op_RegD
4361 RAX_H_num // Op_RegL
4362 };
4363 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4364 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4365 %}
4366 %}
4368 //----------ATTRIBUTES---------------------------------------------------------
4369 //----------Operand Attributes-------------------------------------------------
4370 op_attrib op_cost(0); // Required cost attribute
4372 //----------Instruction Attributes---------------------------------------------
4373 ins_attrib ins_cost(100); // Required cost attribute
4374 ins_attrib ins_size(8); // Required size attribute (in bits)
4375 ins_attrib ins_pc_relative(0); // Required PC Relative flag
4376 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4377 // a non-matching short branch variant
4378 // of some long branch?
4379 ins_attrib ins_alignment(1); // Required alignment attribute (must
4380 // be a power of 2) specifies the
4381 // alignment that some part of the
4382 // instruction (not necessarily the
4383 // start) requires. If > 1, a
4384 // compute_padding() function must be
4385 // provided for the instruction
4387 //----------OPERANDS-----------------------------------------------------------
4388 // Operand definitions must precede instruction definitions for correct parsing
4389 // in the ADLC because operands constitute user defined types which are used in
4390 // instruction definitions.
4392 //----------Simple Operands----------------------------------------------------
4393 // Immediate Operands
4394 // Integer Immediate
4395 operand immI()
4396 %{
4397 match(ConI);
4399 op_cost(10);
4400 format %{ %}
4401 interface(CONST_INTER);
4402 %}
4404 // Constant for test vs zero
4405 operand immI0()
4406 %{
4407 predicate(n->get_int() == 0);
4408 match(ConI);
4410 op_cost(0);
4411 format %{ %}
4412 interface(CONST_INTER);
4413 %}
4415 // Constant for increment
4416 operand immI1()
4417 %{
4418 predicate(n->get_int() == 1);
4419 match(ConI);
4421 op_cost(0);
4422 format %{ %}
4423 interface(CONST_INTER);
4424 %}
4426 // Constant for decrement
4427 operand immI_M1()
4428 %{
4429 predicate(n->get_int() == -1);
4430 match(ConI);
4432 op_cost(0);
4433 format %{ %}
4434 interface(CONST_INTER);
4435 %}
4437 // Valid scale values for addressing modes
4438 operand immI2()
4439 %{
4440 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4441 match(ConI);
4443 format %{ %}
4444 interface(CONST_INTER);
4445 %}
4447 operand immI8()
4448 %{
4449 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4450 match(ConI);
4452 op_cost(5);
4453 format %{ %}
4454 interface(CONST_INTER);
4455 %}
4457 operand immI16()
4458 %{
4459 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4460 match(ConI);
4462 op_cost(10);
4463 format %{ %}
4464 interface(CONST_INTER);
4465 %}
4467 // Constant for long shifts
4468 operand immI_32()
4469 %{
4470 predicate( n->get_int() == 32 );
4471 match(ConI);
4473 op_cost(0);
4474 format %{ %}
4475 interface(CONST_INTER);
4476 %}
4478 // Constant for long shifts
4479 operand immI_64()
4480 %{
4481 predicate( n->get_int() == 64 );
4482 match(ConI);
4484 op_cost(0);
4485 format %{ %}
4486 interface(CONST_INTER);
4487 %}
4489 // Pointer Immediate
4490 operand immP()
4491 %{
4492 match(ConP);
4494 op_cost(10);
4495 format %{ %}
4496 interface(CONST_INTER);
4497 %}
4499 // NULL Pointer Immediate
4500 operand immP0()
4501 %{
4502 predicate(n->get_ptr() == 0);
4503 match(ConP);
4505 op_cost(5);
4506 format %{ %}
4507 interface(CONST_INTER);
4508 %}
4510 // Pointer Immediate
4511 operand immN() %{
4512 match(ConN);
4514 op_cost(10);
4515 format %{ %}
4516 interface(CONST_INTER);
4517 %}
4519 // NULL Pointer Immediate
4520 operand immN0() %{
4521 predicate(n->get_narrowcon() == 0);
4522 match(ConN);
4524 op_cost(5);
4525 format %{ %}
4526 interface(CONST_INTER);
4527 %}
4529 operand immP31()
4530 %{
4531 predicate(!n->as_Type()->type()->isa_oopptr()
4532 && (n->get_ptr() >> 31) == 0);
4533 match(ConP);
4535 op_cost(5);
4536 format %{ %}
4537 interface(CONST_INTER);
4538 %}
4541 // Long Immediate
4542 operand immL()
4543 %{
4544 match(ConL);
4546 op_cost(20);
4547 format %{ %}
4548 interface(CONST_INTER);
4549 %}
4551 // Long Immediate 8-bit
4552 operand immL8()
4553 %{
4554 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4555 match(ConL);
4557 op_cost(5);
4558 format %{ %}
4559 interface(CONST_INTER);
4560 %}
4562 // Long Immediate 32-bit unsigned
4563 operand immUL32()
4564 %{
4565 predicate(n->get_long() == (unsigned int) (n->get_long()));
4566 match(ConL);
4568 op_cost(10);
4569 format %{ %}
4570 interface(CONST_INTER);
4571 %}
4573 // Long Immediate 32-bit signed
4574 operand immL32()
4575 %{
4576 predicate(n->get_long() == (int) (n->get_long()));
4577 match(ConL);
4579 op_cost(15);
4580 format %{ %}
4581 interface(CONST_INTER);
4582 %}
4584 // Long Immediate zero
4585 operand immL0()
4586 %{
4587 predicate(n->get_long() == 0L);
4588 match(ConL);
4590 op_cost(10);
4591 format %{ %}
4592 interface(CONST_INTER);
4593 %}
4595 // Constant for increment
4596 operand immL1()
4597 %{
4598 predicate(n->get_long() == 1);
4599 match(ConL);
4601 format %{ %}
4602 interface(CONST_INTER);
4603 %}
4605 // Constant for decrement
4606 operand immL_M1()
4607 %{
4608 predicate(n->get_long() == -1);
4609 match(ConL);
4611 format %{ %}
4612 interface(CONST_INTER);
4613 %}
4615 // Long Immediate: the value 10
4616 operand immL10()
4617 %{
4618 predicate(n->get_long() == 10);
4619 match(ConL);
4621 format %{ %}
4622 interface(CONST_INTER);
4623 %}
4625 // Long immediate from 0 to 127.
4626 // Used for a shorter form of long mul by 10.
4627 operand immL_127()
4628 %{
4629 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4630 match(ConL);
4632 op_cost(10);
4633 format %{ %}
4634 interface(CONST_INTER);
4635 %}
4637 // Long Immediate: low 32-bit mask
4638 operand immL_32bits()
4639 %{
4640 predicate(n->get_long() == 0xFFFFFFFFL);
4641 match(ConL);
4642 op_cost(20);
4644 format %{ %}
4645 interface(CONST_INTER);
4646 %}
4648 // Float Immediate zero
4649 operand immF0()
4650 %{
4651 predicate(jint_cast(n->getf()) == 0);
4652 match(ConF);
4654 op_cost(5);
4655 format %{ %}
4656 interface(CONST_INTER);
4657 %}
4659 // Float Immediate
4660 operand immF()
4661 %{
4662 match(ConF);
4664 op_cost(15);
4665 format %{ %}
4666 interface(CONST_INTER);
4667 %}
4669 // Double Immediate zero
4670 operand immD0()
4671 %{
4672 predicate(jlong_cast(n->getd()) == 0);
4673 match(ConD);
4675 op_cost(5);
4676 format %{ %}
4677 interface(CONST_INTER);
4678 %}
4680 // Double Immediate
4681 operand immD()
4682 %{
4683 match(ConD);
4685 op_cost(15);
4686 format %{ %}
4687 interface(CONST_INTER);
4688 %}
4690 // Immediates for special shifts (sign extend)
4692 // Constants for increment
4693 operand immI_16()
4694 %{
4695 predicate(n->get_int() == 16);
4696 match(ConI);
4698 format %{ %}
4699 interface(CONST_INTER);
4700 %}
4702 operand immI_24()
4703 %{
4704 predicate(n->get_int() == 24);
4705 match(ConI);
4707 format %{ %}
4708 interface(CONST_INTER);
4709 %}
4711 // Constant for byte-wide masking
4712 operand immI_255()
4713 %{
4714 predicate(n->get_int() == 255);
4715 match(ConI);
4717 format %{ %}
4718 interface(CONST_INTER);
4719 %}
4721 // Constant for short-wide masking
4722 operand immI_65535()
4723 %{
4724 predicate(n->get_int() == 65535);
4725 match(ConI);
4727 format %{ %}
4728 interface(CONST_INTER);
4729 %}
4731 // Constant for byte-wide masking
4732 operand immL_255()
4733 %{
4734 predicate(n->get_long() == 255);
4735 match(ConL);
4737 format %{ %}
4738 interface(CONST_INTER);
4739 %}
4741 // Constant for short-wide masking
4742 operand immL_65535()
4743 %{
4744 predicate(n->get_long() == 65535);
4745 match(ConL);
4747 format %{ %}
4748 interface(CONST_INTER);
4749 %}
4751 // Register Operands
4752 // Integer Register
4753 operand rRegI()
4754 %{
4755 constraint(ALLOC_IN_RC(int_reg));
4756 match(RegI);
4758 match(rax_RegI);
4759 match(rbx_RegI);
4760 match(rcx_RegI);
4761 match(rdx_RegI);
4762 match(rdi_RegI);
4764 format %{ %}
4765 interface(REG_INTER);
4766 %}
4768 // Special Registers
4769 operand rax_RegI()
4770 %{
4771 constraint(ALLOC_IN_RC(int_rax_reg));
4772 match(RegI);
4773 match(rRegI);
4775 format %{ "RAX" %}
4776 interface(REG_INTER);
4777 %}
4779 // Special Registers
4780 operand rbx_RegI()
4781 %{
4782 constraint(ALLOC_IN_RC(int_rbx_reg));
4783 match(RegI);
4784 match(rRegI);
4786 format %{ "RBX" %}
4787 interface(REG_INTER);
4788 %}
4790 operand rcx_RegI()
4791 %{
4792 constraint(ALLOC_IN_RC(int_rcx_reg));
4793 match(RegI);
4794 match(rRegI);
4796 format %{ "RCX" %}
4797 interface(REG_INTER);
4798 %}
4800 operand rdx_RegI()
4801 %{
4802 constraint(ALLOC_IN_RC(int_rdx_reg));
4803 match(RegI);
4804 match(rRegI);
4806 format %{ "RDX" %}
4807 interface(REG_INTER);
4808 %}
4810 operand rdi_RegI()
4811 %{
4812 constraint(ALLOC_IN_RC(int_rdi_reg));
4813 match(RegI);
4814 match(rRegI);
4816 format %{ "RDI" %}
4817 interface(REG_INTER);
4818 %}
4820 operand no_rcx_RegI()
4821 %{
4822 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4823 match(RegI);
4824 match(rax_RegI);
4825 match(rbx_RegI);
4826 match(rdx_RegI);
4827 match(rdi_RegI);
4829 format %{ %}
4830 interface(REG_INTER);
4831 %}
4833 operand no_rax_rdx_RegI()
4834 %{
4835 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4836 match(RegI);
4837 match(rbx_RegI);
4838 match(rcx_RegI);
4839 match(rdi_RegI);
4841 format %{ %}
4842 interface(REG_INTER);
4843 %}
4845 // Pointer Register
4846 operand any_RegP()
4847 %{
4848 constraint(ALLOC_IN_RC(any_reg));
4849 match(RegP);
4850 match(rax_RegP);
4851 match(rbx_RegP);
4852 match(rdi_RegP);
4853 match(rsi_RegP);
4854 match(rbp_RegP);
4855 match(r15_RegP);
4856 match(rRegP);
4858 format %{ %}
4859 interface(REG_INTER);
4860 %}
4862 operand rRegP()
4863 %{
4864 constraint(ALLOC_IN_RC(ptr_reg));
4865 match(RegP);
4866 match(rax_RegP);
4867 match(rbx_RegP);
4868 match(rdi_RegP);
4869 match(rsi_RegP);
4870 match(rbp_RegP);
4871 match(r15_RegP); // See Q&A below about r15_RegP.
4873 format %{ %}
4874 interface(REG_INTER);
4875 %}
4877 operand rRegN() %{
4878 constraint(ALLOC_IN_RC(int_reg));
4879 match(RegN);
4881 format %{ %}
4882 interface(REG_INTER);
4883 %}
4885 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4886 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4887 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4888 // The output of an instruction is controlled by the allocator, which respects
4889 // register class masks, not match rules. Unless an instruction mentions
4890 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4891 // by the allocator as an input.
4893 operand no_rax_RegP()
4894 %{
4895 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4896 match(RegP);
4897 match(rbx_RegP);
4898 match(rsi_RegP);
4899 match(rdi_RegP);
4901 format %{ %}
4902 interface(REG_INTER);
4903 %}
4905 operand no_rbp_RegP()
4906 %{
4907 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4908 match(RegP);
4909 match(rbx_RegP);
4910 match(rsi_RegP);
4911 match(rdi_RegP);
4913 format %{ %}
4914 interface(REG_INTER);
4915 %}
4917 operand no_rax_rbx_RegP()
4918 %{
4919 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4920 match(RegP);
4921 match(rsi_RegP);
4922 match(rdi_RegP);
4924 format %{ %}
4925 interface(REG_INTER);
4926 %}
4928 // Special Registers
4929 // Return a pointer value
4930 operand rax_RegP()
4931 %{
4932 constraint(ALLOC_IN_RC(ptr_rax_reg));
4933 match(RegP);
4934 match(rRegP);
4936 format %{ %}
4937 interface(REG_INTER);
4938 %}
4940 // Special Registers
4941 // Return a compressed pointer value
4942 operand rax_RegN()
4943 %{
4944 constraint(ALLOC_IN_RC(int_rax_reg));
4945 match(RegN);
4946 match(rRegN);
4948 format %{ %}
4949 interface(REG_INTER);
4950 %}
4952 // Used in AtomicAdd
4953 operand rbx_RegP()
4954 %{
4955 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4956 match(RegP);
4957 match(rRegP);
4959 format %{ %}
4960 interface(REG_INTER);
4961 %}
4963 operand rsi_RegP()
4964 %{
4965 constraint(ALLOC_IN_RC(ptr_rsi_reg));
4966 match(RegP);
4967 match(rRegP);
4969 format %{ %}
4970 interface(REG_INTER);
4971 %}
4973 // Used in rep stosq
4974 operand rdi_RegP()
4975 %{
4976 constraint(ALLOC_IN_RC(ptr_rdi_reg));
4977 match(RegP);
4978 match(rRegP);
4980 format %{ %}
4981 interface(REG_INTER);
4982 %}
4984 operand rbp_RegP()
4985 %{
4986 constraint(ALLOC_IN_RC(ptr_rbp_reg));
4987 match(RegP);
4988 match(rRegP);
4990 format %{ %}
4991 interface(REG_INTER);
4992 %}
4994 operand r15_RegP()
4995 %{
4996 constraint(ALLOC_IN_RC(ptr_r15_reg));
4997 match(RegP);
4998 match(rRegP);
5000 format %{ %}
5001 interface(REG_INTER);
5002 %}
5004 operand rRegL()
5005 %{
5006 constraint(ALLOC_IN_RC(long_reg));
5007 match(RegL);
5008 match(rax_RegL);
5009 match(rdx_RegL);
5011 format %{ %}
5012 interface(REG_INTER);
5013 %}
5015 // Special Registers
5016 operand no_rax_rdx_RegL()
5017 %{
5018 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5019 match(RegL);
5020 match(rRegL);
5022 format %{ %}
5023 interface(REG_INTER);
5024 %}
5026 operand no_rax_RegL()
5027 %{
5028 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5029 match(RegL);
5030 match(rRegL);
5031 match(rdx_RegL);
5033 format %{ %}
5034 interface(REG_INTER);
5035 %}
5037 operand no_rcx_RegL()
5038 %{
5039 constraint(ALLOC_IN_RC(long_no_rcx_reg));
5040 match(RegL);
5041 match(rRegL);
5043 format %{ %}
5044 interface(REG_INTER);
5045 %}
5047 operand rax_RegL()
5048 %{
5049 constraint(ALLOC_IN_RC(long_rax_reg));
5050 match(RegL);
5051 match(rRegL);
5053 format %{ "RAX" %}
5054 interface(REG_INTER);
5055 %}
5057 operand rcx_RegL()
5058 %{
5059 constraint(ALLOC_IN_RC(long_rcx_reg));
5060 match(RegL);
5061 match(rRegL);
5063 format %{ %}
5064 interface(REG_INTER);
5065 %}
5067 operand rdx_RegL()
5068 %{
5069 constraint(ALLOC_IN_RC(long_rdx_reg));
5070 match(RegL);
5071 match(rRegL);
5073 format %{ %}
5074 interface(REG_INTER);
5075 %}
5077 // Flags register, used as output of compare instructions
5078 operand rFlagsReg()
5079 %{
5080 constraint(ALLOC_IN_RC(int_flags));
5081 match(RegFlags);
5083 format %{ "RFLAGS" %}
5084 interface(REG_INTER);
5085 %}
5087 // Flags register, used as output of FLOATING POINT compare instructions
5088 operand rFlagsRegU()
5089 %{
5090 constraint(ALLOC_IN_RC(int_flags));
5091 match(RegFlags);
5093 format %{ "RFLAGS_U" %}
5094 interface(REG_INTER);
5095 %}
5097 operand rFlagsRegUCF() %{
5098 constraint(ALLOC_IN_RC(int_flags));
5099 match(RegFlags);
5100 predicate(false);
5102 format %{ "RFLAGS_U_CF" %}
5103 interface(REG_INTER);
5104 %}
5106 // Float register operands
5107 operand regF()
5108 %{
5109 constraint(ALLOC_IN_RC(float_reg));
5110 match(RegF);
5112 format %{ %}
5113 interface(REG_INTER);
5114 %}
5116 // Double register operands
5117 operand regD()
5118 %{
5119 constraint(ALLOC_IN_RC(double_reg));
5120 match(RegD);
5122 format %{ %}
5123 interface(REG_INTER);
5124 %}
5127 //----------Memory Operands----------------------------------------------------
5128 // Direct Memory Operand
5129 // operand direct(immP addr)
5130 // %{
5131 // match(addr);
5133 // format %{ "[$addr]" %}
5134 // interface(MEMORY_INTER) %{
5135 // base(0xFFFFFFFF);
5136 // index(0x4);
5137 // scale(0x0);
5138 // disp($addr);
5139 // %}
5140 // %}
5142 // Indirect Memory Operand
5143 operand indirect(any_RegP reg)
5144 %{
5145 constraint(ALLOC_IN_RC(ptr_reg));
5146 match(reg);
5148 format %{ "[$reg]" %}
5149 interface(MEMORY_INTER) %{
5150 base($reg);
5151 index(0x4);
5152 scale(0x0);
5153 disp(0x0);
5154 %}
5155 %}
5157 // Indirect Memory Plus Short Offset Operand
5158 operand indOffset8(any_RegP reg, immL8 off)
5159 %{
5160 constraint(ALLOC_IN_RC(ptr_reg));
5161 match(AddP reg off);
5163 format %{ "[$reg + $off (8-bit)]" %}
5164 interface(MEMORY_INTER) %{
5165 base($reg);
5166 index(0x4);
5167 scale(0x0);
5168 disp($off);
5169 %}
5170 %}
5172 // Indirect Memory Plus Long Offset Operand
5173 operand indOffset32(any_RegP reg, immL32 off)
5174 %{
5175 constraint(ALLOC_IN_RC(ptr_reg));
5176 match(AddP reg off);
5178 format %{ "[$reg + $off (32-bit)]" %}
5179 interface(MEMORY_INTER) %{
5180 base($reg);
5181 index(0x4);
5182 scale(0x0);
5183 disp($off);
5184 %}
5185 %}
5187 // Indirect Memory Plus Index Register Plus Offset Operand
5188 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
5189 %{
5190 constraint(ALLOC_IN_RC(ptr_reg));
5191 match(AddP (AddP reg lreg) off);
5193 op_cost(10);
5194 format %{"[$reg + $off + $lreg]" %}
5195 interface(MEMORY_INTER) %{
5196 base($reg);
5197 index($lreg);
5198 scale(0x0);
5199 disp($off);
5200 %}
5201 %}
5203 // Indirect Memory Plus Index Register Plus Offset Operand
5204 operand indIndex(any_RegP reg, rRegL lreg)
5205 %{
5206 constraint(ALLOC_IN_RC(ptr_reg));
5207 match(AddP reg lreg);
5209 op_cost(10);
5210 format %{"[$reg + $lreg]" %}
5211 interface(MEMORY_INTER) %{
5212 base($reg);
5213 index($lreg);
5214 scale(0x0);
5215 disp(0x0);
5216 %}
5217 %}
5219 // Indirect Memory Times Scale Plus Index Register
5220 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
5221 %{
5222 constraint(ALLOC_IN_RC(ptr_reg));
5223 match(AddP reg (LShiftL lreg scale));
5225 op_cost(10);
5226 format %{"[$reg + $lreg << $scale]" %}
5227 interface(MEMORY_INTER) %{
5228 base($reg);
5229 index($lreg);
5230 scale($scale);
5231 disp(0x0);
5232 %}
5233 %}
5235 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5236 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
5237 %{
5238 constraint(ALLOC_IN_RC(ptr_reg));
5239 match(AddP (AddP reg (LShiftL lreg scale)) off);
5241 op_cost(10);
5242 format %{"[$reg + $off + $lreg << $scale]" %}
5243 interface(MEMORY_INTER) %{
5244 base($reg);
5245 index($lreg);
5246 scale($scale);
5247 disp($off);
5248 %}
5249 %}
5251 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5252 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
5253 %{
5254 constraint(ALLOC_IN_RC(ptr_reg));
5255 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5256 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
5258 op_cost(10);
5259 format %{"[$reg + $off + $idx << $scale]" %}
5260 interface(MEMORY_INTER) %{
5261 base($reg);
5262 index($idx);
5263 scale($scale);
5264 disp($off);
5265 %}
5266 %}
5268 // Indirect Narrow Oop Plus Offset Operand
5269 // Note: x86 architecture doesn't support "scale * index + offset" without a base
5270 // we can't free r12 even with Universe::narrow_oop_base() == NULL.
5271 operand indCompressedOopOffset(rRegN reg, immL32 off) %{
5272 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
5273 constraint(ALLOC_IN_RC(ptr_reg));
5274 match(AddP (DecodeN reg) off);
5276 op_cost(10);
5277 format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %}
5278 interface(MEMORY_INTER) %{
5279 base(0xc); // R12
5280 index($reg);
5281 scale(0x3);
5282 disp($off);
5283 %}
5284 %}
5286 // Indirect Memory Operand
5287 operand indirectNarrow(rRegN reg)
5288 %{
5289 predicate(Universe::narrow_oop_shift() == 0);
5290 constraint(ALLOC_IN_RC(ptr_reg));
5291 match(DecodeN reg);
5293 format %{ "[$reg]" %}
5294 interface(MEMORY_INTER) %{
5295 base($reg);
5296 index(0x4);
5297 scale(0x0);
5298 disp(0x0);
5299 %}
5300 %}
5302 // Indirect Memory Plus Short Offset Operand
5303 operand indOffset8Narrow(rRegN reg, immL8 off)
5304 %{
5305 predicate(Universe::narrow_oop_shift() == 0);
5306 constraint(ALLOC_IN_RC(ptr_reg));
5307 match(AddP (DecodeN reg) off);
5309 format %{ "[$reg + $off (8-bit)]" %}
5310 interface(MEMORY_INTER) %{
5311 base($reg);
5312 index(0x4);
5313 scale(0x0);
5314 disp($off);
5315 %}
5316 %}
5318 // Indirect Memory Plus Long Offset Operand
5319 operand indOffset32Narrow(rRegN reg, immL32 off)
5320 %{
5321 predicate(Universe::narrow_oop_shift() == 0);
5322 constraint(ALLOC_IN_RC(ptr_reg));
5323 match(AddP (DecodeN reg) off);
5325 format %{ "[$reg + $off (32-bit)]" %}
5326 interface(MEMORY_INTER) %{
5327 base($reg);
5328 index(0x4);
5329 scale(0x0);
5330 disp($off);
5331 %}
5332 %}
5334 // Indirect Memory Plus Index Register Plus Offset Operand
5335 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off)
5336 %{
5337 predicate(Universe::narrow_oop_shift() == 0);
5338 constraint(ALLOC_IN_RC(ptr_reg));
5339 match(AddP (AddP (DecodeN reg) lreg) off);
5341 op_cost(10);
5342 format %{"[$reg + $off + $lreg]" %}
5343 interface(MEMORY_INTER) %{
5344 base($reg);
5345 index($lreg);
5346 scale(0x0);
5347 disp($off);
5348 %}
5349 %}
5351 // Indirect Memory Plus Index Register Plus Offset Operand
5352 operand indIndexNarrow(rRegN reg, rRegL lreg)
5353 %{
5354 predicate(Universe::narrow_oop_shift() == 0);
5355 constraint(ALLOC_IN_RC(ptr_reg));
5356 match(AddP (DecodeN reg) lreg);
5358 op_cost(10);
5359 format %{"[$reg + $lreg]" %}
5360 interface(MEMORY_INTER) %{
5361 base($reg);
5362 index($lreg);
5363 scale(0x0);
5364 disp(0x0);
5365 %}
5366 %}
5368 // Indirect Memory Times Scale Plus Index Register
5369 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale)
5370 %{
5371 predicate(Universe::narrow_oop_shift() == 0);
5372 constraint(ALLOC_IN_RC(ptr_reg));
5373 match(AddP (DecodeN reg) (LShiftL lreg scale));
5375 op_cost(10);
5376 format %{"[$reg + $lreg << $scale]" %}
5377 interface(MEMORY_INTER) %{
5378 base($reg);
5379 index($lreg);
5380 scale($scale);
5381 disp(0x0);
5382 %}
5383 %}
5385 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5386 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
5387 %{
5388 predicate(Universe::narrow_oop_shift() == 0);
5389 constraint(ALLOC_IN_RC(ptr_reg));
5390 match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off);
5392 op_cost(10);
5393 format %{"[$reg + $off + $lreg << $scale]" %}
5394 interface(MEMORY_INTER) %{
5395 base($reg);
5396 index($lreg);
5397 scale($scale);
5398 disp($off);
5399 %}
5400 %}
5402 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5403 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale)
5404 %{
5405 constraint(ALLOC_IN_RC(ptr_reg));
5406 predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5407 match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off);
5409 op_cost(10);
5410 format %{"[$reg + $off + $idx << $scale]" %}
5411 interface(MEMORY_INTER) %{
5412 base($reg);
5413 index($idx);
5414 scale($scale);
5415 disp($off);
5416 %}
5417 %}
5420 //----------Special Memory Operands--------------------------------------------
5421 // Stack Slot Operand - This operand is used for loading and storing temporary
5422 // values on the stack where a match requires a value to
5423 // flow through memory.
5424 operand stackSlotP(sRegP reg)
5425 %{
5426 constraint(ALLOC_IN_RC(stack_slots));
5427 // No match rule because this operand is only generated in matching
5429 format %{ "[$reg]" %}
5430 interface(MEMORY_INTER) %{
5431 base(0x4); // RSP
5432 index(0x4); // No Index
5433 scale(0x0); // No Scale
5434 disp($reg); // Stack Offset
5435 %}
5436 %}
5438 operand stackSlotI(sRegI reg)
5439 %{
5440 constraint(ALLOC_IN_RC(stack_slots));
5441 // No match rule because this operand is only generated in matching
5443 format %{ "[$reg]" %}
5444 interface(MEMORY_INTER) %{
5445 base(0x4); // RSP
5446 index(0x4); // No Index
5447 scale(0x0); // No Scale
5448 disp($reg); // Stack Offset
5449 %}
5450 %}
5452 operand stackSlotF(sRegF reg)
5453 %{
5454 constraint(ALLOC_IN_RC(stack_slots));
5455 // No match rule because this operand is only generated in matching
5457 format %{ "[$reg]" %}
5458 interface(MEMORY_INTER) %{
5459 base(0x4); // RSP
5460 index(0x4); // No Index
5461 scale(0x0); // No Scale
5462 disp($reg); // Stack Offset
5463 %}
5464 %}
5466 operand stackSlotD(sRegD reg)
5467 %{
5468 constraint(ALLOC_IN_RC(stack_slots));
5469 // No match rule because this operand is only generated in matching
5471 format %{ "[$reg]" %}
5472 interface(MEMORY_INTER) %{
5473 base(0x4); // RSP
5474 index(0x4); // No Index
5475 scale(0x0); // No Scale
5476 disp($reg); // Stack Offset
5477 %}
5478 %}
5479 operand stackSlotL(sRegL reg)
5480 %{
5481 constraint(ALLOC_IN_RC(stack_slots));
5482 // No match rule because this operand is only generated in matching
5484 format %{ "[$reg]" %}
5485 interface(MEMORY_INTER) %{
5486 base(0x4); // RSP
5487 index(0x4); // No Index
5488 scale(0x0); // No Scale
5489 disp($reg); // Stack Offset
5490 %}
5491 %}
5493 //----------Conditional Branch Operands----------------------------------------
5494 // Comparison Op - This is the operation of the comparison, and is limited to
5495 // the following set of codes:
5496 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5497 //
5498 // Other attributes of the comparison, such as unsignedness, are specified
5499 // by the comparison instruction that sets a condition code flags register.
5500 // That result is represented by a flags operand whose subtype is appropriate
5501 // to the unsignedness (etc.) of the comparison.
5502 //
5503 // Later, the instruction which matches both the Comparison Op (a Bool) and
5504 // the flags (produced by the Cmp) specifies the coding of the comparison op
5505 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5507 // Comparision Code
5508 operand cmpOp()
5509 %{
5510 match(Bool);
5512 format %{ "" %}
5513 interface(COND_INTER) %{
5514 equal(0x4, "e");
5515 not_equal(0x5, "ne");
5516 less(0xC, "l");
5517 greater_equal(0xD, "ge");
5518 less_equal(0xE, "le");
5519 greater(0xF, "g");
5520 %}
5521 %}
5523 // Comparison Code, unsigned compare. Used by FP also, with
5524 // C2 (unordered) turned into GT or LT already. The other bits
5525 // C0 and C3 are turned into Carry & Zero flags.
5526 operand cmpOpU()
5527 %{
5528 match(Bool);
5530 format %{ "" %}
5531 interface(COND_INTER) %{
5532 equal(0x4, "e");
5533 not_equal(0x5, "ne");
5534 less(0x2, "b");
5535 greater_equal(0x3, "nb");
5536 less_equal(0x6, "be");
5537 greater(0x7, "nbe");
5538 %}
5539 %}
5542 // Floating comparisons that don't require any fixup for the unordered case
5543 operand cmpOpUCF() %{
5544 match(Bool);
5545 predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5546 n->as_Bool()->_test._test == BoolTest::ge ||
5547 n->as_Bool()->_test._test == BoolTest::le ||
5548 n->as_Bool()->_test._test == BoolTest::gt);
5549 format %{ "" %}
5550 interface(COND_INTER) %{
5551 equal(0x4, "e");
5552 not_equal(0x5, "ne");
5553 less(0x2, "b");
5554 greater_equal(0x3, "nb");
5555 less_equal(0x6, "be");
5556 greater(0x7, "nbe");
5557 %}
5558 %}
5561 // Floating comparisons that can be fixed up with extra conditional jumps
5562 operand cmpOpUCF2() %{
5563 match(Bool);
5564 predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5565 n->as_Bool()->_test._test == BoolTest::eq);
5566 format %{ "" %}
5567 interface(COND_INTER) %{
5568 equal(0x4, "e");
5569 not_equal(0x5, "ne");
5570 less(0x2, "b");
5571 greater_equal(0x3, "nb");
5572 less_equal(0x6, "be");
5573 greater(0x7, "nbe");
5574 %}
5575 %}
5578 //----------OPERAND CLASSES----------------------------------------------------
5579 // Operand Classes are groups of operands that are used as to simplify
5580 // instruction definitions by not requiring the AD writer to specify separate
5581 // instructions for every form of operand when the instruction accepts
5582 // multiple operand types with the same basic encoding and format. The classic
5583 // case of this is memory operands.
5585 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5586 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5587 indCompressedOopOffset,
5588 indirectNarrow, indOffset8Narrow, indOffset32Narrow,
5589 indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
5590 indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
5592 //----------PIPELINE-----------------------------------------------------------
5593 // Rules which define the behavior of the target architectures pipeline.
5594 pipeline %{
5596 //----------ATTRIBUTES---------------------------------------------------------
5597 attributes %{
5598 variable_size_instructions; // Fixed size instructions
5599 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5600 instruction_unit_size = 1; // An instruction is 1 bytes long
5601 instruction_fetch_unit_size = 16; // The processor fetches one line
5602 instruction_fetch_units = 1; // of 16 bytes
5604 // List of nop instructions
5605 nops( MachNop );
5606 %}
5608 //----------RESOURCES----------------------------------------------------------
5609 // Resources are the functional units available to the machine
5611 // Generic P2/P3 pipeline
5612 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5613 // 3 instructions decoded per cycle.
5614 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5615 // 3 ALU op, only ALU0 handles mul instructions.
5616 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5617 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5618 BR, FPU,
5619 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5621 //----------PIPELINE DESCRIPTION-----------------------------------------------
5622 // Pipeline Description specifies the stages in the machine's pipeline
5624 // Generic P2/P3 pipeline
5625 pipe_desc(S0, S1, S2, S3, S4, S5);
5627 //----------PIPELINE CLASSES---------------------------------------------------
5628 // Pipeline Classes describe the stages in which input and output are
5629 // referenced by the hardware pipeline.
5631 // Naming convention: ialu or fpu
5632 // Then: _reg
5633 // Then: _reg if there is a 2nd register
5634 // Then: _long if it's a pair of instructions implementing a long
5635 // Then: _fat if it requires the big decoder
5636 // Or: _mem if it requires the big decoder and a memory unit.
5638 // Integer ALU reg operation
5639 pipe_class ialu_reg(rRegI dst)
5640 %{
5641 single_instruction;
5642 dst : S4(write);
5643 dst : S3(read);
5644 DECODE : S0; // any decoder
5645 ALU : S3; // any alu
5646 %}
5648 // Long ALU reg operation
5649 pipe_class ialu_reg_long(rRegL dst)
5650 %{
5651 instruction_count(2);
5652 dst : S4(write);
5653 dst : S3(read);
5654 DECODE : S0(2); // any 2 decoders
5655 ALU : S3(2); // both alus
5656 %}
5658 // Integer ALU reg operation using big decoder
5659 pipe_class ialu_reg_fat(rRegI dst)
5660 %{
5661 single_instruction;
5662 dst : S4(write);
5663 dst : S3(read);
5664 D0 : S0; // big decoder only
5665 ALU : S3; // any alu
5666 %}
5668 // Long ALU reg operation using big decoder
5669 pipe_class ialu_reg_long_fat(rRegL dst)
5670 %{
5671 instruction_count(2);
5672 dst : S4(write);
5673 dst : S3(read);
5674 D0 : S0(2); // big decoder only; twice
5675 ALU : S3(2); // any 2 alus
5676 %}
5678 // Integer ALU reg-reg operation
5679 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5680 %{
5681 single_instruction;
5682 dst : S4(write);
5683 src : S3(read);
5684 DECODE : S0; // any decoder
5685 ALU : S3; // any alu
5686 %}
5688 // Long ALU reg-reg operation
5689 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5690 %{
5691 instruction_count(2);
5692 dst : S4(write);
5693 src : S3(read);
5694 DECODE : S0(2); // any 2 decoders
5695 ALU : S3(2); // both alus
5696 %}
5698 // Integer ALU reg-reg operation
5699 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5700 %{
5701 single_instruction;
5702 dst : S4(write);
5703 src : S3(read);
5704 D0 : S0; // big decoder only
5705 ALU : S3; // any alu
5706 %}
5708 // Long ALU reg-reg operation
5709 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5710 %{
5711 instruction_count(2);
5712 dst : S4(write);
5713 src : S3(read);
5714 D0 : S0(2); // big decoder only; twice
5715 ALU : S3(2); // both alus
5716 %}
5718 // Integer ALU reg-mem operation
5719 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5720 %{
5721 single_instruction;
5722 dst : S5(write);
5723 mem : S3(read);
5724 D0 : S0; // big decoder only
5725 ALU : S4; // any alu
5726 MEM : S3; // any mem
5727 %}
5729 // Integer mem operation (prefetch)
5730 pipe_class ialu_mem(memory mem)
5731 %{
5732 single_instruction;
5733 mem : S3(read);
5734 D0 : S0; // big decoder only
5735 MEM : S3; // any mem
5736 %}
5738 // Integer Store to Memory
5739 pipe_class ialu_mem_reg(memory mem, rRegI src)
5740 %{
5741 single_instruction;
5742 mem : S3(read);
5743 src : S5(read);
5744 D0 : S0; // big decoder only
5745 ALU : S4; // any alu
5746 MEM : S3;
5747 %}
5749 // // Long Store to Memory
5750 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5751 // %{
5752 // instruction_count(2);
5753 // mem : S3(read);
5754 // src : S5(read);
5755 // D0 : S0(2); // big decoder only; twice
5756 // ALU : S4(2); // any 2 alus
5757 // MEM : S3(2); // Both mems
5758 // %}
5760 // Integer Store to Memory
5761 pipe_class ialu_mem_imm(memory mem)
5762 %{
5763 single_instruction;
5764 mem : S3(read);
5765 D0 : S0; // big decoder only
5766 ALU : S4; // any alu
5767 MEM : S3;
5768 %}
5770 // Integer ALU0 reg-reg operation
5771 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5772 %{
5773 single_instruction;
5774 dst : S4(write);
5775 src : S3(read);
5776 D0 : S0; // Big decoder only
5777 ALU0 : S3; // only alu0
5778 %}
5780 // Integer ALU0 reg-mem operation
5781 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5782 %{
5783 single_instruction;
5784 dst : S5(write);
5785 mem : S3(read);
5786 D0 : S0; // big decoder only
5787 ALU0 : S4; // ALU0 only
5788 MEM : S3; // any mem
5789 %}
5791 // Integer ALU reg-reg operation
5792 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5793 %{
5794 single_instruction;
5795 cr : S4(write);
5796 src1 : S3(read);
5797 src2 : S3(read);
5798 DECODE : S0; // any decoder
5799 ALU : S3; // any alu
5800 %}
5802 // Integer ALU reg-imm operation
5803 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5804 %{
5805 single_instruction;
5806 cr : S4(write);
5807 src1 : S3(read);
5808 DECODE : S0; // any decoder
5809 ALU : S3; // any alu
5810 %}
5812 // Integer ALU reg-mem operation
5813 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5814 %{
5815 single_instruction;
5816 cr : S4(write);
5817 src1 : S3(read);
5818 src2 : S3(read);
5819 D0 : S0; // big decoder only
5820 ALU : S4; // any alu
5821 MEM : S3;
5822 %}
5824 // Conditional move reg-reg
5825 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5826 %{
5827 instruction_count(4);
5828 y : S4(read);
5829 q : S3(read);
5830 p : S3(read);
5831 DECODE : S0(4); // any decoder
5832 %}
5834 // Conditional move reg-reg
5835 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5836 %{
5837 single_instruction;
5838 dst : S4(write);
5839 src : S3(read);
5840 cr : S3(read);
5841 DECODE : S0; // any decoder
5842 %}
5844 // Conditional move reg-mem
5845 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5846 %{
5847 single_instruction;
5848 dst : S4(write);
5849 src : S3(read);
5850 cr : S3(read);
5851 DECODE : S0; // any decoder
5852 MEM : S3;
5853 %}
5855 // Conditional move reg-reg long
5856 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5857 %{
5858 single_instruction;
5859 dst : S4(write);
5860 src : S3(read);
5861 cr : S3(read);
5862 DECODE : S0(2); // any 2 decoders
5863 %}
5865 // XXX
5866 // // Conditional move double reg-reg
5867 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5868 // %{
5869 // single_instruction;
5870 // dst : S4(write);
5871 // src : S3(read);
5872 // cr : S3(read);
5873 // DECODE : S0; // any decoder
5874 // %}
5876 // Float reg-reg operation
5877 pipe_class fpu_reg(regD dst)
5878 %{
5879 instruction_count(2);
5880 dst : S3(read);
5881 DECODE : S0(2); // any 2 decoders
5882 FPU : S3;
5883 %}
5885 // Float reg-reg operation
5886 pipe_class fpu_reg_reg(regD dst, regD src)
5887 %{
5888 instruction_count(2);
5889 dst : S4(write);
5890 src : S3(read);
5891 DECODE : S0(2); // any 2 decoders
5892 FPU : S3;
5893 %}
5895 // Float reg-reg operation
5896 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5897 %{
5898 instruction_count(3);
5899 dst : S4(write);
5900 src1 : S3(read);
5901 src2 : S3(read);
5902 DECODE : S0(3); // any 3 decoders
5903 FPU : S3(2);
5904 %}
5906 // Float reg-reg operation
5907 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5908 %{
5909 instruction_count(4);
5910 dst : S4(write);
5911 src1 : S3(read);
5912 src2 : S3(read);
5913 src3 : S3(read);
5914 DECODE : S0(4); // any 3 decoders
5915 FPU : S3(2);
5916 %}
5918 // Float reg-reg operation
5919 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5920 %{
5921 instruction_count(4);
5922 dst : S4(write);
5923 src1 : S3(read);
5924 src2 : S3(read);
5925 src3 : S3(read);
5926 DECODE : S1(3); // any 3 decoders
5927 D0 : S0; // Big decoder only
5928 FPU : S3(2);
5929 MEM : S3;
5930 %}
5932 // Float reg-mem operation
5933 pipe_class fpu_reg_mem(regD dst, memory mem)
5934 %{
5935 instruction_count(2);
5936 dst : S5(write);
5937 mem : S3(read);
5938 D0 : S0; // big decoder only
5939 DECODE : S1; // any decoder for FPU POP
5940 FPU : S4;
5941 MEM : S3; // any mem
5942 %}
5944 // Float reg-mem operation
5945 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5946 %{
5947 instruction_count(3);
5948 dst : S5(write);
5949 src1 : S3(read);
5950 mem : S3(read);
5951 D0 : S0; // big decoder only
5952 DECODE : S1(2); // any decoder for FPU POP
5953 FPU : S4;
5954 MEM : S3; // any mem
5955 %}
5957 // Float mem-reg operation
5958 pipe_class fpu_mem_reg(memory mem, regD src)
5959 %{
5960 instruction_count(2);
5961 src : S5(read);
5962 mem : S3(read);
5963 DECODE : S0; // any decoder for FPU PUSH
5964 D0 : S1; // big decoder only
5965 FPU : S4;
5966 MEM : S3; // any mem
5967 %}
5969 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5970 %{
5971 instruction_count(3);
5972 src1 : S3(read);
5973 src2 : S3(read);
5974 mem : S3(read);
5975 DECODE : S0(2); // any decoder for FPU PUSH
5976 D0 : S1; // big decoder only
5977 FPU : S4;
5978 MEM : S3; // any mem
5979 %}
5981 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5982 %{
5983 instruction_count(3);
5984 src1 : S3(read);
5985 src2 : S3(read);
5986 mem : S4(read);
5987 DECODE : S0; // any decoder for FPU PUSH
5988 D0 : S0(2); // big decoder only
5989 FPU : S4;
5990 MEM : S3(2); // any mem
5991 %}
5993 pipe_class fpu_mem_mem(memory dst, memory src1)
5994 %{
5995 instruction_count(2);
5996 src1 : S3(read);
5997 dst : S4(read);
5998 D0 : S0(2); // big decoder only
5999 MEM : S3(2); // any mem
6000 %}
6002 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
6003 %{
6004 instruction_count(3);
6005 src1 : S3(read);
6006 src2 : S3(read);
6007 dst : S4(read);
6008 D0 : S0(3); // big decoder only
6009 FPU : S4;
6010 MEM : S3(3); // any mem
6011 %}
6013 pipe_class fpu_mem_reg_con(memory mem, regD src1)
6014 %{
6015 instruction_count(3);
6016 src1 : S4(read);
6017 mem : S4(read);
6018 DECODE : S0; // any decoder for FPU PUSH
6019 D0 : S0(2); // big decoder only
6020 FPU : S4;
6021 MEM : S3(2); // any mem
6022 %}
6024 // Float load constant
6025 pipe_class fpu_reg_con(regD dst)
6026 %{
6027 instruction_count(2);
6028 dst : S5(write);
6029 D0 : S0; // big decoder only for the load
6030 DECODE : S1; // any decoder for FPU POP
6031 FPU : S4;
6032 MEM : S3; // any mem
6033 %}
6035 // Float load constant
6036 pipe_class fpu_reg_reg_con(regD dst, regD src)
6037 %{
6038 instruction_count(3);
6039 dst : S5(write);
6040 src : S3(read);
6041 D0 : S0; // big decoder only for the load
6042 DECODE : S1(2); // any decoder for FPU POP
6043 FPU : S4;
6044 MEM : S3; // any mem
6045 %}
6047 // UnConditional branch
6048 pipe_class pipe_jmp(label labl)
6049 %{
6050 single_instruction;
6051 BR : S3;
6052 %}
6054 // Conditional branch
6055 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
6056 %{
6057 single_instruction;
6058 cr : S1(read);
6059 BR : S3;
6060 %}
6062 // Allocation idiom
6063 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
6064 %{
6065 instruction_count(1); force_serialization;
6066 fixed_latency(6);
6067 heap_ptr : S3(read);
6068 DECODE : S0(3);
6069 D0 : S2;
6070 MEM : S3;
6071 ALU : S3(2);
6072 dst : S5(write);
6073 BR : S5;
6074 %}
6076 // Generic big/slow expanded idiom
6077 pipe_class pipe_slow()
6078 %{
6079 instruction_count(10); multiple_bundles; force_serialization;
6080 fixed_latency(100);
6081 D0 : S0(2);
6082 MEM : S3(2);
6083 %}
6085 // The real do-nothing guy
6086 pipe_class empty()
6087 %{
6088 instruction_count(0);
6089 %}
6091 // Define the class for the Nop node
6092 define
6093 %{
6094 MachNop = empty;
6095 %}
6097 %}
6099 //----------INSTRUCTIONS-------------------------------------------------------
6100 //
6101 // match -- States which machine-independent subtree may be replaced
6102 // by this instruction.
6103 // ins_cost -- The estimated cost of this instruction is used by instruction
6104 // selection to identify a minimum cost tree of machine
6105 // instructions that matches a tree of machine-independent
6106 // instructions.
6107 // format -- A string providing the disassembly for this instruction.
6108 // The value of an instruction's operand may be inserted
6109 // by referring to it with a '$' prefix.
6110 // opcode -- Three instruction opcodes may be provided. These are referred
6111 // to within an encode class as $primary, $secondary, and $tertiary
6112 // rrspectively. The primary opcode is commonly used to
6113 // indicate the type of machine instruction, while secondary
6114 // and tertiary are often used for prefix options or addressing
6115 // modes.
6116 // ins_encode -- A list of encode classes with parameters. The encode class
6117 // name must have been defined in an 'enc_class' specification
6118 // in the encode section of the architecture description.
6121 //----------Load/Store/Move Instructions---------------------------------------
6122 //----------Load Instructions--------------------------------------------------
6124 // Load Byte (8 bit signed)
6125 instruct loadB(rRegI dst, memory mem)
6126 %{
6127 match(Set dst (LoadB mem));
6129 ins_cost(125);
6130 format %{ "movsbl $dst, $mem\t# byte" %}
6132 ins_encode %{
6133 __ movsbl($dst$$Register, $mem$$Address);
6134 %}
6136 ins_pipe(ialu_reg_mem);
6137 %}
6139 // Load Byte (8 bit signed) into Long Register
6140 instruct loadB2L(rRegL dst, memory mem)
6141 %{
6142 match(Set dst (ConvI2L (LoadB mem)));
6144 ins_cost(125);
6145 format %{ "movsbq $dst, $mem\t# byte -> long" %}
6147 ins_encode %{
6148 __ movsbq($dst$$Register, $mem$$Address);
6149 %}
6151 ins_pipe(ialu_reg_mem);
6152 %}
6154 // Load Unsigned Byte (8 bit UNsigned)
6155 instruct loadUB(rRegI dst, memory mem)
6156 %{
6157 match(Set dst (LoadUB mem));
6159 ins_cost(125);
6160 format %{ "movzbl $dst, $mem\t# ubyte" %}
6162 ins_encode %{
6163 __ movzbl($dst$$Register, $mem$$Address);
6164 %}
6166 ins_pipe(ialu_reg_mem);
6167 %}
6169 // Load Unsigned Byte (8 bit UNsigned) into Long Register
6170 instruct loadUB2L(rRegL dst, memory mem)
6171 %{
6172 match(Set dst (ConvI2L (LoadUB mem)));
6174 ins_cost(125);
6175 format %{ "movzbq $dst, $mem\t# ubyte -> long" %}
6177 ins_encode %{
6178 __ movzbq($dst$$Register, $mem$$Address);
6179 %}
6181 ins_pipe(ialu_reg_mem);
6182 %}
6184 // Load Short (16 bit signed)
6185 instruct loadS(rRegI dst, memory mem)
6186 %{
6187 match(Set dst (LoadS mem));
6189 ins_cost(125);
6190 format %{ "movswl $dst, $mem\t# short" %}
6192 ins_encode %{
6193 __ movswl($dst$$Register, $mem$$Address);
6194 %}
6196 ins_pipe(ialu_reg_mem);
6197 %}
6199 // Load Short (16 bit signed) into Long Register
6200 instruct loadS2L(rRegL dst, memory mem)
6201 %{
6202 match(Set dst (ConvI2L (LoadS mem)));
6204 ins_cost(125);
6205 format %{ "movswq $dst, $mem\t# short -> long" %}
6207 ins_encode %{
6208 __ movswq($dst$$Register, $mem$$Address);
6209 %}
6211 ins_pipe(ialu_reg_mem);
6212 %}
6214 // Load Unsigned Short/Char (16 bit UNsigned)
6215 instruct loadUS(rRegI dst, memory mem)
6216 %{
6217 match(Set dst (LoadUS mem));
6219 ins_cost(125);
6220 format %{ "movzwl $dst, $mem\t# ushort/char" %}
6222 ins_encode %{
6223 __ movzwl($dst$$Register, $mem$$Address);
6224 %}
6226 ins_pipe(ialu_reg_mem);
6227 %}
6229 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
6230 instruct loadUS2L(rRegL dst, memory mem)
6231 %{
6232 match(Set dst (ConvI2L (LoadUS mem)));
6234 ins_cost(125);
6235 format %{ "movzwq $dst, $mem\t# ushort/char -> long" %}
6237 ins_encode %{
6238 __ movzwq($dst$$Register, $mem$$Address);
6239 %}
6241 ins_pipe(ialu_reg_mem);
6242 %}
6244 // Load Integer
6245 instruct loadI(rRegI dst, memory mem)
6246 %{
6247 match(Set dst (LoadI mem));
6249 ins_cost(125);
6250 format %{ "movl $dst, $mem\t# int" %}
6252 ins_encode %{
6253 __ movl($dst$$Register, $mem$$Address);
6254 %}
6256 ins_pipe(ialu_reg_mem);
6257 %}
6259 // Load Integer into Long Register
6260 instruct loadI2L(rRegL dst, memory mem)
6261 %{
6262 match(Set dst (ConvI2L (LoadI mem)));
6264 ins_cost(125);
6265 format %{ "movslq $dst, $mem\t# int -> long" %}
6267 ins_encode %{
6268 __ movslq($dst$$Register, $mem$$Address);
6269 %}
6271 ins_pipe(ialu_reg_mem);
6272 %}
6274 // Load Unsigned Integer into Long Register
6275 instruct loadUI2L(rRegL dst, memory mem)
6276 %{
6277 match(Set dst (LoadUI2L mem));
6279 ins_cost(125);
6280 format %{ "movl $dst, $mem\t# uint -> long" %}
6282 ins_encode %{
6283 __ movl($dst$$Register, $mem$$Address);
6284 %}
6286 ins_pipe(ialu_reg_mem);
6287 %}
6289 // Load Long
6290 instruct loadL(rRegL dst, memory mem)
6291 %{
6292 match(Set dst (LoadL mem));
6294 ins_cost(125);
6295 format %{ "movq $dst, $mem\t# long" %}
6297 ins_encode %{
6298 __ movq($dst$$Register, $mem$$Address);
6299 %}
6301 ins_pipe(ialu_reg_mem); // XXX
6302 %}
6304 // Load Range
6305 instruct loadRange(rRegI dst, memory mem)
6306 %{
6307 match(Set dst (LoadRange mem));
6309 ins_cost(125); // XXX
6310 format %{ "movl $dst, $mem\t# range" %}
6311 opcode(0x8B);
6312 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6313 ins_pipe(ialu_reg_mem);
6314 %}
6316 // Load Pointer
6317 instruct loadP(rRegP dst, memory mem)
6318 %{
6319 match(Set dst (LoadP mem));
6321 ins_cost(125); // XXX
6322 format %{ "movq $dst, $mem\t# ptr" %}
6323 opcode(0x8B);
6324 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6325 ins_pipe(ialu_reg_mem); // XXX
6326 %}
6328 // Load Compressed Pointer
6329 instruct loadN(rRegN dst, memory mem)
6330 %{
6331 match(Set dst (LoadN mem));
6333 ins_cost(125); // XXX
6334 format %{ "movl $dst, $mem\t# compressed ptr" %}
6335 ins_encode %{
6336 __ movl($dst$$Register, $mem$$Address);
6337 %}
6338 ins_pipe(ialu_reg_mem); // XXX
6339 %}
6342 // Load Klass Pointer
6343 instruct loadKlass(rRegP dst, memory mem)
6344 %{
6345 match(Set dst (LoadKlass mem));
6347 ins_cost(125); // XXX
6348 format %{ "movq $dst, $mem\t# class" %}
6349 opcode(0x8B);
6350 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6351 ins_pipe(ialu_reg_mem); // XXX
6352 %}
6354 // Load narrow Klass Pointer
6355 instruct loadNKlass(rRegN dst, memory mem)
6356 %{
6357 match(Set dst (LoadNKlass mem));
6359 ins_cost(125); // XXX
6360 format %{ "movl $dst, $mem\t# compressed klass ptr" %}
6361 ins_encode %{
6362 __ movl($dst$$Register, $mem$$Address);
6363 %}
6364 ins_pipe(ialu_reg_mem); // XXX
6365 %}
6367 // Load Float
6368 instruct loadF(regF dst, memory mem)
6369 %{
6370 match(Set dst (LoadF mem));
6372 ins_cost(145); // XXX
6373 format %{ "movss $dst, $mem\t# float" %}
6374 opcode(0xF3, 0x0F, 0x10);
6375 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6376 ins_pipe(pipe_slow); // XXX
6377 %}
6379 // Load Double
6380 instruct loadD_partial(regD dst, memory mem)
6381 %{
6382 predicate(!UseXmmLoadAndClearUpper);
6383 match(Set dst (LoadD mem));
6385 ins_cost(145); // XXX
6386 format %{ "movlpd $dst, $mem\t# double" %}
6387 opcode(0x66, 0x0F, 0x12);
6388 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6389 ins_pipe(pipe_slow); // XXX
6390 %}
6392 instruct loadD(regD dst, memory mem)
6393 %{
6394 predicate(UseXmmLoadAndClearUpper);
6395 match(Set dst (LoadD mem));
6397 ins_cost(145); // XXX
6398 format %{ "movsd $dst, $mem\t# double" %}
6399 opcode(0xF2, 0x0F, 0x10);
6400 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6401 ins_pipe(pipe_slow); // XXX
6402 %}
6404 // Load Aligned Packed Byte to XMM register
6405 instruct loadA8B(regD dst, memory mem) %{
6406 match(Set dst (Load8B mem));
6407 ins_cost(125);
6408 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6409 ins_encode( movq_ld(dst, mem));
6410 ins_pipe( pipe_slow );
6411 %}
6413 // Load Aligned Packed Short to XMM register
6414 instruct loadA4S(regD dst, memory mem) %{
6415 match(Set dst (Load4S mem));
6416 ins_cost(125);
6417 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6418 ins_encode( movq_ld(dst, mem));
6419 ins_pipe( pipe_slow );
6420 %}
6422 // Load Aligned Packed Char to XMM register
6423 instruct loadA4C(regD dst, memory mem) %{
6424 match(Set dst (Load4C mem));
6425 ins_cost(125);
6426 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6427 ins_encode( movq_ld(dst, mem));
6428 ins_pipe( pipe_slow );
6429 %}
6431 // Load Aligned Packed Integer to XMM register
6432 instruct load2IU(regD dst, memory mem) %{
6433 match(Set dst (Load2I mem));
6434 ins_cost(125);
6435 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6436 ins_encode( movq_ld(dst, mem));
6437 ins_pipe( pipe_slow );
6438 %}
6440 // Load Aligned Packed Single to XMM
6441 instruct loadA2F(regD dst, memory mem) %{
6442 match(Set dst (Load2F mem));
6443 ins_cost(145);
6444 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6445 ins_encode( movq_ld(dst, mem));
6446 ins_pipe( pipe_slow );
6447 %}
6449 // Load Effective Address
6450 instruct leaP8(rRegP dst, indOffset8 mem)
6451 %{
6452 match(Set dst mem);
6454 ins_cost(110); // XXX
6455 format %{ "leaq $dst, $mem\t# ptr 8" %}
6456 opcode(0x8D);
6457 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6458 ins_pipe(ialu_reg_reg_fat);
6459 %}
6461 instruct leaP32(rRegP dst, indOffset32 mem)
6462 %{
6463 match(Set dst mem);
6465 ins_cost(110);
6466 format %{ "leaq $dst, $mem\t# ptr 32" %}
6467 opcode(0x8D);
6468 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6469 ins_pipe(ialu_reg_reg_fat);
6470 %}
6472 // instruct leaPIdx(rRegP dst, indIndex mem)
6473 // %{
6474 // match(Set dst mem);
6476 // ins_cost(110);
6477 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6478 // opcode(0x8D);
6479 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6480 // ins_pipe(ialu_reg_reg_fat);
6481 // %}
6483 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6484 %{
6485 match(Set dst mem);
6487 ins_cost(110);
6488 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6489 opcode(0x8D);
6490 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6491 ins_pipe(ialu_reg_reg_fat);
6492 %}
6494 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6495 %{
6496 match(Set dst mem);
6498 ins_cost(110);
6499 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6500 opcode(0x8D);
6501 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6502 ins_pipe(ialu_reg_reg_fat);
6503 %}
6505 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6506 %{
6507 match(Set dst mem);
6509 ins_cost(110);
6510 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6511 opcode(0x8D);
6512 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6513 ins_pipe(ialu_reg_reg_fat);
6514 %}
6516 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem)
6517 %{
6518 match(Set dst mem);
6520 ins_cost(110);
6521 format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %}
6522 opcode(0x8D);
6523 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6524 ins_pipe(ialu_reg_reg_fat);
6525 %}
6527 // Load Effective Address which uses Narrow (32-bits) oop
6528 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem)
6529 %{
6530 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
6531 match(Set dst mem);
6533 ins_cost(110);
6534 format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %}
6535 opcode(0x8D);
6536 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6537 ins_pipe(ialu_reg_reg_fat);
6538 %}
6540 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem)
6541 %{
6542 predicate(Universe::narrow_oop_shift() == 0);
6543 match(Set dst mem);
6545 ins_cost(110); // XXX
6546 format %{ "leaq $dst, $mem\t# ptr off8narrow" %}
6547 opcode(0x8D);
6548 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6549 ins_pipe(ialu_reg_reg_fat);
6550 %}
6552 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem)
6553 %{
6554 predicate(Universe::narrow_oop_shift() == 0);
6555 match(Set dst mem);
6557 ins_cost(110);
6558 format %{ "leaq $dst, $mem\t# ptr off32narrow" %}
6559 opcode(0x8D);
6560 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6561 ins_pipe(ialu_reg_reg_fat);
6562 %}
6564 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem)
6565 %{
6566 predicate(Universe::narrow_oop_shift() == 0);
6567 match(Set dst mem);
6569 ins_cost(110);
6570 format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %}
6571 opcode(0x8D);
6572 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6573 ins_pipe(ialu_reg_reg_fat);
6574 %}
6576 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem)
6577 %{
6578 predicate(Universe::narrow_oop_shift() == 0);
6579 match(Set dst mem);
6581 ins_cost(110);
6582 format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %}
6583 opcode(0x8D);
6584 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6585 ins_pipe(ialu_reg_reg_fat);
6586 %}
6588 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem)
6589 %{
6590 predicate(Universe::narrow_oop_shift() == 0);
6591 match(Set dst mem);
6593 ins_cost(110);
6594 format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %}
6595 opcode(0x8D);
6596 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6597 ins_pipe(ialu_reg_reg_fat);
6598 %}
6600 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem)
6601 %{
6602 predicate(Universe::narrow_oop_shift() == 0);
6603 match(Set dst mem);
6605 ins_cost(110);
6606 format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %}
6607 opcode(0x8D);
6608 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6609 ins_pipe(ialu_reg_reg_fat);
6610 %}
6612 instruct loadConI(rRegI dst, immI src)
6613 %{
6614 match(Set dst src);
6616 format %{ "movl $dst, $src\t# int" %}
6617 ins_encode(load_immI(dst, src));
6618 ins_pipe(ialu_reg_fat); // XXX
6619 %}
6621 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6622 %{
6623 match(Set dst src);
6624 effect(KILL cr);
6626 ins_cost(50);
6627 format %{ "xorl $dst, $dst\t# int" %}
6628 opcode(0x33); /* + rd */
6629 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6630 ins_pipe(ialu_reg);
6631 %}
6633 instruct loadConL(rRegL dst, immL src)
6634 %{
6635 match(Set dst src);
6637 ins_cost(150);
6638 format %{ "movq $dst, $src\t# long" %}
6639 ins_encode(load_immL(dst, src));
6640 ins_pipe(ialu_reg);
6641 %}
6643 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6644 %{
6645 match(Set dst src);
6646 effect(KILL cr);
6648 ins_cost(50);
6649 format %{ "xorl $dst, $dst\t# long" %}
6650 opcode(0x33); /* + rd */
6651 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6652 ins_pipe(ialu_reg); // XXX
6653 %}
6655 instruct loadConUL32(rRegL dst, immUL32 src)
6656 %{
6657 match(Set dst src);
6659 ins_cost(60);
6660 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6661 ins_encode(load_immUL32(dst, src));
6662 ins_pipe(ialu_reg);
6663 %}
6665 instruct loadConL32(rRegL dst, immL32 src)
6666 %{
6667 match(Set dst src);
6669 ins_cost(70);
6670 format %{ "movq $dst, $src\t# long (32-bit)" %}
6671 ins_encode(load_immL32(dst, src));
6672 ins_pipe(ialu_reg);
6673 %}
6675 instruct loadConP(rRegP dst, immP src)
6676 %{
6677 match(Set dst src);
6679 format %{ "movq $dst, $src\t# ptr" %}
6680 ins_encode(load_immP(dst, src));
6681 ins_pipe(ialu_reg_fat); // XXX
6682 %}
6684 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6685 %{
6686 match(Set dst src);
6687 effect(KILL cr);
6689 ins_cost(50);
6690 format %{ "xorl $dst, $dst\t# ptr" %}
6691 opcode(0x33); /* + rd */
6692 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6693 ins_pipe(ialu_reg);
6694 %}
6696 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6697 %{
6698 match(Set dst src);
6699 effect(KILL cr);
6701 ins_cost(60);
6702 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6703 ins_encode(load_immP31(dst, src));
6704 ins_pipe(ialu_reg);
6705 %}
6707 instruct loadConF(regF dst, immF src)
6708 %{
6709 match(Set dst src);
6710 ins_cost(125);
6712 format %{ "movss $dst, [$src]" %}
6713 ins_encode(load_conF(dst, src));
6714 ins_pipe(pipe_slow);
6715 %}
6717 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6718 match(Set dst src);
6719 effect(KILL cr);
6720 format %{ "xorq $dst, $src\t# compressed NULL ptr" %}
6721 ins_encode %{
6722 __ xorq($dst$$Register, $dst$$Register);
6723 %}
6724 ins_pipe(ialu_reg);
6725 %}
6727 instruct loadConN(rRegN dst, immN src) %{
6728 match(Set dst src);
6730 ins_cost(125);
6731 format %{ "movl $dst, $src\t# compressed ptr" %}
6732 ins_encode %{
6733 address con = (address)$src$$constant;
6734 if (con == NULL) {
6735 ShouldNotReachHere();
6736 } else {
6737 __ set_narrow_oop($dst$$Register, (jobject)$src$$constant);
6738 }
6739 %}
6740 ins_pipe(ialu_reg_fat); // XXX
6741 %}
6743 instruct loadConF0(regF dst, immF0 src)
6744 %{
6745 match(Set dst src);
6746 ins_cost(100);
6748 format %{ "xorps $dst, $dst\t# float 0.0" %}
6749 opcode(0x0F, 0x57);
6750 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6751 ins_pipe(pipe_slow);
6752 %}
6754 // Use the same format since predicate() can not be used here.
6755 instruct loadConD(regD dst, immD src)
6756 %{
6757 match(Set dst src);
6758 ins_cost(125);
6760 format %{ "movsd $dst, [$src]" %}
6761 ins_encode(load_conD(dst, src));
6762 ins_pipe(pipe_slow);
6763 %}
6765 instruct loadConD0(regD dst, immD0 src)
6766 %{
6767 match(Set dst src);
6768 ins_cost(100);
6770 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6771 opcode(0x66, 0x0F, 0x57);
6772 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6773 ins_pipe(pipe_slow);
6774 %}
6776 instruct loadSSI(rRegI dst, stackSlotI src)
6777 %{
6778 match(Set dst src);
6780 ins_cost(125);
6781 format %{ "movl $dst, $src\t# int stk" %}
6782 opcode(0x8B);
6783 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6784 ins_pipe(ialu_reg_mem);
6785 %}
6787 instruct loadSSL(rRegL dst, stackSlotL src)
6788 %{
6789 match(Set dst src);
6791 ins_cost(125);
6792 format %{ "movq $dst, $src\t# long stk" %}
6793 opcode(0x8B);
6794 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6795 ins_pipe(ialu_reg_mem);
6796 %}
6798 instruct loadSSP(rRegP dst, stackSlotP src)
6799 %{
6800 match(Set dst src);
6802 ins_cost(125);
6803 format %{ "movq $dst, $src\t# ptr stk" %}
6804 opcode(0x8B);
6805 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6806 ins_pipe(ialu_reg_mem);
6807 %}
6809 instruct loadSSF(regF dst, stackSlotF src)
6810 %{
6811 match(Set dst src);
6813 ins_cost(125);
6814 format %{ "movss $dst, $src\t# float stk" %}
6815 opcode(0xF3, 0x0F, 0x10);
6816 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6817 ins_pipe(pipe_slow); // XXX
6818 %}
6820 // Use the same format since predicate() can not be used here.
6821 instruct loadSSD(regD dst, stackSlotD src)
6822 %{
6823 match(Set dst src);
6825 ins_cost(125);
6826 format %{ "movsd $dst, $src\t# double stk" %}
6827 ins_encode %{
6828 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6829 %}
6830 ins_pipe(pipe_slow); // XXX
6831 %}
6833 // Prefetch instructions.
6834 // Must be safe to execute with invalid address (cannot fault).
6836 instruct prefetchr( memory mem ) %{
6837 predicate(ReadPrefetchInstr==3);
6838 match(PrefetchRead mem);
6839 ins_cost(125);
6841 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6842 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6843 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6844 ins_pipe(ialu_mem);
6845 %}
6847 instruct prefetchrNTA( memory mem ) %{
6848 predicate(ReadPrefetchInstr==0);
6849 match(PrefetchRead mem);
6850 ins_cost(125);
6852 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6853 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6854 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6855 ins_pipe(ialu_mem);
6856 %}
6858 instruct prefetchrT0( memory mem ) %{
6859 predicate(ReadPrefetchInstr==1);
6860 match(PrefetchRead mem);
6861 ins_cost(125);
6863 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6864 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6865 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6866 ins_pipe(ialu_mem);
6867 %}
6869 instruct prefetchrT2( memory mem ) %{
6870 predicate(ReadPrefetchInstr==2);
6871 match(PrefetchRead mem);
6872 ins_cost(125);
6874 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6875 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6876 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6877 ins_pipe(ialu_mem);
6878 %}
6880 instruct prefetchw( memory mem ) %{
6881 predicate(AllocatePrefetchInstr==3);
6882 match(PrefetchWrite mem);
6883 ins_cost(125);
6885 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6886 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6887 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6888 ins_pipe(ialu_mem);
6889 %}
6891 instruct prefetchwNTA( memory mem ) %{
6892 predicate(AllocatePrefetchInstr==0);
6893 match(PrefetchWrite mem);
6894 ins_cost(125);
6896 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6897 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6898 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6899 ins_pipe(ialu_mem);
6900 %}
6902 instruct prefetchwT0( memory mem ) %{
6903 predicate(AllocatePrefetchInstr==1);
6904 match(PrefetchWrite mem);
6905 ins_cost(125);
6907 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6908 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6909 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6910 ins_pipe(ialu_mem);
6911 %}
6913 instruct prefetchwT2( memory mem ) %{
6914 predicate(AllocatePrefetchInstr==2);
6915 match(PrefetchWrite mem);
6916 ins_cost(125);
6918 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6919 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6920 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6921 ins_pipe(ialu_mem);
6922 %}
6924 //----------Store Instructions-------------------------------------------------
6926 // Store Byte
6927 instruct storeB(memory mem, rRegI src)
6928 %{
6929 match(Set mem (StoreB mem src));
6931 ins_cost(125); // XXX
6932 format %{ "movb $mem, $src\t# byte" %}
6933 opcode(0x88);
6934 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6935 ins_pipe(ialu_mem_reg);
6936 %}
6938 // Store Char/Short
6939 instruct storeC(memory mem, rRegI src)
6940 %{
6941 match(Set mem (StoreC mem src));
6943 ins_cost(125); // XXX
6944 format %{ "movw $mem, $src\t# char/short" %}
6945 opcode(0x89);
6946 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6947 ins_pipe(ialu_mem_reg);
6948 %}
6950 // Store Integer
6951 instruct storeI(memory mem, rRegI src)
6952 %{
6953 match(Set mem (StoreI mem src));
6955 ins_cost(125); // XXX
6956 format %{ "movl $mem, $src\t# int" %}
6957 opcode(0x89);
6958 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6959 ins_pipe(ialu_mem_reg);
6960 %}
6962 // Store Long
6963 instruct storeL(memory mem, rRegL src)
6964 %{
6965 match(Set mem (StoreL mem src));
6967 ins_cost(125); // XXX
6968 format %{ "movq $mem, $src\t# long" %}
6969 opcode(0x89);
6970 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6971 ins_pipe(ialu_mem_reg); // XXX
6972 %}
6974 // Store Pointer
6975 instruct storeP(memory mem, any_RegP src)
6976 %{
6977 match(Set mem (StoreP mem src));
6979 ins_cost(125); // XXX
6980 format %{ "movq $mem, $src\t# ptr" %}
6981 opcode(0x89);
6982 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6983 ins_pipe(ialu_mem_reg);
6984 %}
6986 instruct storeImmP0(memory mem, immP0 zero)
6987 %{
6988 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6989 match(Set mem (StoreP mem zero));
6991 ins_cost(125); // XXX
6992 format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %}
6993 ins_encode %{
6994 __ movq($mem$$Address, r12);
6995 %}
6996 ins_pipe(ialu_mem_reg);
6997 %}
6999 // Store NULL Pointer, mark word, or other simple pointer constant.
7000 instruct storeImmP(memory mem, immP31 src)
7001 %{
7002 match(Set mem (StoreP mem src));
7004 ins_cost(150); // XXX
7005 format %{ "movq $mem, $src\t# ptr" %}
7006 opcode(0xC7); /* C7 /0 */
7007 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7008 ins_pipe(ialu_mem_imm);
7009 %}
7011 // Store Compressed Pointer
7012 instruct storeN(memory mem, rRegN src)
7013 %{
7014 match(Set mem (StoreN mem src));
7016 ins_cost(125); // XXX
7017 format %{ "movl $mem, $src\t# compressed ptr" %}
7018 ins_encode %{
7019 __ movl($mem$$Address, $src$$Register);
7020 %}
7021 ins_pipe(ialu_mem_reg);
7022 %}
7024 instruct storeImmN0(memory mem, immN0 zero)
7025 %{
7026 predicate(Universe::narrow_oop_base() == NULL);
7027 match(Set mem (StoreN mem zero));
7029 ins_cost(125); // XXX
7030 format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %}
7031 ins_encode %{
7032 __ movl($mem$$Address, r12);
7033 %}
7034 ins_pipe(ialu_mem_reg);
7035 %}
7037 instruct storeImmN(memory mem, immN src)
7038 %{
7039 match(Set mem (StoreN mem src));
7041 ins_cost(150); // XXX
7042 format %{ "movl $mem, $src\t# compressed ptr" %}
7043 ins_encode %{
7044 address con = (address)$src$$constant;
7045 if (con == NULL) {
7046 __ movl($mem$$Address, (int32_t)0);
7047 } else {
7048 __ set_narrow_oop($mem$$Address, (jobject)$src$$constant);
7049 }
7050 %}
7051 ins_pipe(ialu_mem_imm);
7052 %}
7054 // Store Integer Immediate
7055 instruct storeImmI0(memory mem, immI0 zero)
7056 %{
7057 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7058 match(Set mem (StoreI mem zero));
7060 ins_cost(125); // XXX
7061 format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %}
7062 ins_encode %{
7063 __ movl($mem$$Address, r12);
7064 %}
7065 ins_pipe(ialu_mem_reg);
7066 %}
7068 instruct storeImmI(memory mem, immI src)
7069 %{
7070 match(Set mem (StoreI mem src));
7072 ins_cost(150);
7073 format %{ "movl $mem, $src\t# int" %}
7074 opcode(0xC7); /* C7 /0 */
7075 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7076 ins_pipe(ialu_mem_imm);
7077 %}
7079 // Store Long Immediate
7080 instruct storeImmL0(memory mem, immL0 zero)
7081 %{
7082 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7083 match(Set mem (StoreL mem zero));
7085 ins_cost(125); // XXX
7086 format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %}
7087 ins_encode %{
7088 __ movq($mem$$Address, r12);
7089 %}
7090 ins_pipe(ialu_mem_reg);
7091 %}
7093 instruct storeImmL(memory mem, immL32 src)
7094 %{
7095 match(Set mem (StoreL mem src));
7097 ins_cost(150);
7098 format %{ "movq $mem, $src\t# long" %}
7099 opcode(0xC7); /* C7 /0 */
7100 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7101 ins_pipe(ialu_mem_imm);
7102 %}
7104 // Store Short/Char Immediate
7105 instruct storeImmC0(memory mem, immI0 zero)
7106 %{
7107 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7108 match(Set mem (StoreC mem zero));
7110 ins_cost(125); // XXX
7111 format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %}
7112 ins_encode %{
7113 __ movw($mem$$Address, r12);
7114 %}
7115 ins_pipe(ialu_mem_reg);
7116 %}
7118 instruct storeImmI16(memory mem, immI16 src)
7119 %{
7120 predicate(UseStoreImmI16);
7121 match(Set mem (StoreC mem src));
7123 ins_cost(150);
7124 format %{ "movw $mem, $src\t# short/char" %}
7125 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
7126 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
7127 ins_pipe(ialu_mem_imm);
7128 %}
7130 // Store Byte Immediate
7131 instruct storeImmB0(memory mem, immI0 zero)
7132 %{
7133 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7134 match(Set mem (StoreB mem zero));
7136 ins_cost(125); // XXX
7137 format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %}
7138 ins_encode %{
7139 __ movb($mem$$Address, r12);
7140 %}
7141 ins_pipe(ialu_mem_reg);
7142 %}
7144 instruct storeImmB(memory mem, immI8 src)
7145 %{
7146 match(Set mem (StoreB mem src));
7148 ins_cost(150); // XXX
7149 format %{ "movb $mem, $src\t# byte" %}
7150 opcode(0xC6); /* C6 /0 */
7151 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7152 ins_pipe(ialu_mem_imm);
7153 %}
7155 // Store Aligned Packed Byte XMM register to memory
7156 instruct storeA8B(memory mem, regD src) %{
7157 match(Set mem (Store8B mem src));
7158 ins_cost(145);
7159 format %{ "MOVQ $mem,$src\t! packed8B" %}
7160 ins_encode( movq_st(mem, src));
7161 ins_pipe( pipe_slow );
7162 %}
7164 // Store Aligned Packed Char/Short XMM register to memory
7165 instruct storeA4C(memory mem, regD src) %{
7166 match(Set mem (Store4C mem src));
7167 ins_cost(145);
7168 format %{ "MOVQ $mem,$src\t! packed4C" %}
7169 ins_encode( movq_st(mem, src));
7170 ins_pipe( pipe_slow );
7171 %}
7173 // Store Aligned Packed Integer XMM register to memory
7174 instruct storeA2I(memory mem, regD src) %{
7175 match(Set mem (Store2I mem src));
7176 ins_cost(145);
7177 format %{ "MOVQ $mem,$src\t! packed2I" %}
7178 ins_encode( movq_st(mem, src));
7179 ins_pipe( pipe_slow );
7180 %}
7182 // Store CMS card-mark Immediate
7183 instruct storeImmCM0_reg(memory mem, immI0 zero)
7184 %{
7185 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7186 match(Set mem (StoreCM mem zero));
7188 ins_cost(125); // XXX
7189 format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %}
7190 ins_encode %{
7191 __ movb($mem$$Address, r12);
7192 %}
7193 ins_pipe(ialu_mem_reg);
7194 %}
7196 instruct storeImmCM0(memory mem, immI0 src)
7197 %{
7198 match(Set mem (StoreCM mem src));
7200 ins_cost(150); // XXX
7201 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
7202 opcode(0xC6); /* C6 /0 */
7203 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7204 ins_pipe(ialu_mem_imm);
7205 %}
7207 // Store Aligned Packed Single Float XMM register to memory
7208 instruct storeA2F(memory mem, regD src) %{
7209 match(Set mem (Store2F mem src));
7210 ins_cost(145);
7211 format %{ "MOVQ $mem,$src\t! packed2F" %}
7212 ins_encode( movq_st(mem, src));
7213 ins_pipe( pipe_slow );
7214 %}
7216 // Store Float
7217 instruct storeF(memory mem, regF src)
7218 %{
7219 match(Set mem (StoreF mem src));
7221 ins_cost(95); // XXX
7222 format %{ "movss $mem, $src\t# float" %}
7223 opcode(0xF3, 0x0F, 0x11);
7224 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7225 ins_pipe(pipe_slow); // XXX
7226 %}
7228 // Store immediate Float value (it is faster than store from XMM register)
7229 instruct storeF0(memory mem, immF0 zero)
7230 %{
7231 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7232 match(Set mem (StoreF mem zero));
7234 ins_cost(25); // XXX
7235 format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %}
7236 ins_encode %{
7237 __ movl($mem$$Address, r12);
7238 %}
7239 ins_pipe(ialu_mem_reg);
7240 %}
7242 instruct storeF_imm(memory mem, immF src)
7243 %{
7244 match(Set mem (StoreF mem src));
7246 ins_cost(50);
7247 format %{ "movl $mem, $src\t# float" %}
7248 opcode(0xC7); /* C7 /0 */
7249 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7250 ins_pipe(ialu_mem_imm);
7251 %}
7253 // Store Double
7254 instruct storeD(memory mem, regD src)
7255 %{
7256 match(Set mem (StoreD mem src));
7258 ins_cost(95); // XXX
7259 format %{ "movsd $mem, $src\t# double" %}
7260 opcode(0xF2, 0x0F, 0x11);
7261 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7262 ins_pipe(pipe_slow); // XXX
7263 %}
7265 // Store immediate double 0.0 (it is faster than store from XMM register)
7266 instruct storeD0_imm(memory mem, immD0 src)
7267 %{
7268 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
7269 match(Set mem (StoreD mem src));
7271 ins_cost(50);
7272 format %{ "movq $mem, $src\t# double 0." %}
7273 opcode(0xC7); /* C7 /0 */
7274 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7275 ins_pipe(ialu_mem_imm);
7276 %}
7278 instruct storeD0(memory mem, immD0 zero)
7279 %{
7280 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7281 match(Set mem (StoreD mem zero));
7283 ins_cost(25); // XXX
7284 format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %}
7285 ins_encode %{
7286 __ movq($mem$$Address, r12);
7287 %}
7288 ins_pipe(ialu_mem_reg);
7289 %}
7291 instruct storeSSI(stackSlotI dst, rRegI src)
7292 %{
7293 match(Set dst src);
7295 ins_cost(100);
7296 format %{ "movl $dst, $src\t# int stk" %}
7297 opcode(0x89);
7298 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7299 ins_pipe( ialu_mem_reg );
7300 %}
7302 instruct storeSSL(stackSlotL dst, rRegL src)
7303 %{
7304 match(Set dst src);
7306 ins_cost(100);
7307 format %{ "movq $dst, $src\t# long stk" %}
7308 opcode(0x89);
7309 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7310 ins_pipe(ialu_mem_reg);
7311 %}
7313 instruct storeSSP(stackSlotP dst, rRegP src)
7314 %{
7315 match(Set dst src);
7317 ins_cost(100);
7318 format %{ "movq $dst, $src\t# ptr stk" %}
7319 opcode(0x89);
7320 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7321 ins_pipe(ialu_mem_reg);
7322 %}
7324 instruct storeSSF(stackSlotF dst, regF src)
7325 %{
7326 match(Set dst src);
7328 ins_cost(95); // XXX
7329 format %{ "movss $dst, $src\t# float stk" %}
7330 opcode(0xF3, 0x0F, 0x11);
7331 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7332 ins_pipe(pipe_slow); // XXX
7333 %}
7335 instruct storeSSD(stackSlotD dst, regD src)
7336 %{
7337 match(Set dst src);
7339 ins_cost(95); // XXX
7340 format %{ "movsd $dst, $src\t# double stk" %}
7341 opcode(0xF2, 0x0F, 0x11);
7342 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7343 ins_pipe(pipe_slow); // XXX
7344 %}
7346 //----------BSWAP Instructions-------------------------------------------------
7347 instruct bytes_reverse_int(rRegI dst) %{
7348 match(Set dst (ReverseBytesI dst));
7350 format %{ "bswapl $dst" %}
7351 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
7352 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
7353 ins_pipe( ialu_reg );
7354 %}
7356 instruct bytes_reverse_long(rRegL dst) %{
7357 match(Set dst (ReverseBytesL dst));
7359 format %{ "bswapq $dst" %}
7361 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
7362 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
7363 ins_pipe( ialu_reg);
7364 %}
7366 instruct loadI_reversed(rRegI dst, memory src) %{
7367 match(Set dst (ReverseBytesI (LoadI src)));
7369 format %{ "bswap_movl $dst, $src" %}
7370 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7371 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
7372 ins_pipe( ialu_reg_mem );
7373 %}
7375 instruct loadL_reversed(rRegL dst, memory src) %{
7376 match(Set dst (ReverseBytesL (LoadL src)));
7378 format %{ "bswap_movq $dst, $src" %}
7379 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7380 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
7381 ins_pipe( ialu_reg_mem );
7382 %}
7384 instruct storeI_reversed(memory dst, rRegI src) %{
7385 match(Set dst (StoreI dst (ReverseBytesI src)));
7387 format %{ "movl_bswap $dst, $src" %}
7388 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7389 ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
7390 ins_pipe( ialu_mem_reg );
7391 %}
7393 instruct storeL_reversed(memory dst, rRegL src) %{
7394 match(Set dst (StoreL dst (ReverseBytesL src)));
7396 format %{ "movq_bswap $dst, $src" %}
7397 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7398 ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
7399 ins_pipe( ialu_mem_reg );
7400 %}
7403 //---------- Population Count Instructions -------------------------------------
7405 instruct popCountI(rRegI dst, rRegI src) %{
7406 predicate(UsePopCountInstruction);
7407 match(Set dst (PopCountI src));
7409 format %{ "popcnt $dst, $src" %}
7410 ins_encode %{
7411 __ popcntl($dst$$Register, $src$$Register);
7412 %}
7413 ins_pipe(ialu_reg);
7414 %}
7416 instruct popCountI_mem(rRegI dst, memory mem) %{
7417 predicate(UsePopCountInstruction);
7418 match(Set dst (PopCountI (LoadI mem)));
7420 format %{ "popcnt $dst, $mem" %}
7421 ins_encode %{
7422 __ popcntl($dst$$Register, $mem$$Address);
7423 %}
7424 ins_pipe(ialu_reg);
7425 %}
7427 // Note: Long.bitCount(long) returns an int.
7428 instruct popCountL(rRegI dst, rRegL src) %{
7429 predicate(UsePopCountInstruction);
7430 match(Set dst (PopCountL src));
7432 format %{ "popcnt $dst, $src" %}
7433 ins_encode %{
7434 __ popcntq($dst$$Register, $src$$Register);
7435 %}
7436 ins_pipe(ialu_reg);
7437 %}
7439 // Note: Long.bitCount(long) returns an int.
7440 instruct popCountL_mem(rRegI dst, memory mem) %{
7441 predicate(UsePopCountInstruction);
7442 match(Set dst (PopCountL (LoadL mem)));
7444 format %{ "popcnt $dst, $mem" %}
7445 ins_encode %{
7446 __ popcntq($dst$$Register, $mem$$Address);
7447 %}
7448 ins_pipe(ialu_reg);
7449 %}
7452 //----------MemBar Instructions-----------------------------------------------
7453 // Memory barrier flavors
7455 instruct membar_acquire()
7456 %{
7457 match(MemBarAcquire);
7458 ins_cost(0);
7460 size(0);
7461 format %{ "MEMBAR-acquire" %}
7462 ins_encode();
7463 ins_pipe(empty);
7464 %}
7466 instruct membar_acquire_lock()
7467 %{
7468 match(MemBarAcquire);
7469 predicate(Matcher::prior_fast_lock(n));
7470 ins_cost(0);
7472 size(0);
7473 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7474 ins_encode();
7475 ins_pipe(empty);
7476 %}
7478 instruct membar_release()
7479 %{
7480 match(MemBarRelease);
7481 ins_cost(0);
7483 size(0);
7484 format %{ "MEMBAR-release" %}
7485 ins_encode();
7486 ins_pipe(empty);
7487 %}
7489 instruct membar_release_lock()
7490 %{
7491 match(MemBarRelease);
7492 predicate(Matcher::post_fast_unlock(n));
7493 ins_cost(0);
7495 size(0);
7496 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7497 ins_encode();
7498 ins_pipe(empty);
7499 %}
7501 instruct membar_volatile()
7502 %{
7503 match(MemBarVolatile);
7504 ins_cost(400);
7506 format %{ "MEMBAR-volatile" %}
7507 ins_encode(enc_membar_volatile);
7508 ins_pipe(pipe_slow);
7509 %}
7511 instruct unnecessary_membar_volatile()
7512 %{
7513 match(MemBarVolatile);
7514 predicate(Matcher::post_store_load_barrier(n));
7515 ins_cost(0);
7517 size(0);
7518 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7519 ins_encode();
7520 ins_pipe(empty);
7521 %}
7523 //----------Move Instructions--------------------------------------------------
7525 instruct castX2P(rRegP dst, rRegL src)
7526 %{
7527 match(Set dst (CastX2P src));
7529 format %{ "movq $dst, $src\t# long->ptr" %}
7530 ins_encode(enc_copy_wide(dst, src));
7531 ins_pipe(ialu_reg_reg); // XXX
7532 %}
7534 instruct castP2X(rRegL dst, rRegP src)
7535 %{
7536 match(Set dst (CastP2X src));
7538 format %{ "movq $dst, $src\t# ptr -> long" %}
7539 ins_encode(enc_copy_wide(dst, src));
7540 ins_pipe(ialu_reg_reg); // XXX
7541 %}
7544 // Convert oop pointer into compressed form
7545 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7546 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7547 match(Set dst (EncodeP src));
7548 effect(KILL cr);
7549 format %{ "encode_heap_oop $dst,$src" %}
7550 ins_encode %{
7551 Register s = $src$$Register;
7552 Register d = $dst$$Register;
7553 if (s != d) {
7554 __ movq(d, s);
7555 }
7556 __ encode_heap_oop(d);
7557 %}
7558 ins_pipe(ialu_reg_long);
7559 %}
7561 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7562 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7563 match(Set dst (EncodeP src));
7564 effect(KILL cr);
7565 format %{ "encode_heap_oop_not_null $dst,$src" %}
7566 ins_encode %{
7567 __ encode_heap_oop_not_null($dst$$Register, $src$$Register);
7568 %}
7569 ins_pipe(ialu_reg_long);
7570 %}
7572 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7573 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7574 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7575 match(Set dst (DecodeN src));
7576 effect(KILL cr);
7577 format %{ "decode_heap_oop $dst,$src" %}
7578 ins_encode %{
7579 Register s = $src$$Register;
7580 Register d = $dst$$Register;
7581 if (s != d) {
7582 __ movq(d, s);
7583 }
7584 __ decode_heap_oop(d);
7585 %}
7586 ins_pipe(ialu_reg_long);
7587 %}
7589 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{
7590 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7591 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7592 match(Set dst (DecodeN src));
7593 format %{ "decode_heap_oop_not_null $dst,$src" %}
7594 ins_encode %{
7595 Register s = $src$$Register;
7596 Register d = $dst$$Register;
7597 if (s != d) {
7598 __ decode_heap_oop_not_null(d, s);
7599 } else {
7600 __ decode_heap_oop_not_null(d);
7601 }
7602 %}
7603 ins_pipe(ialu_reg_long);
7604 %}
7607 //----------Conditional Move---------------------------------------------------
7608 // Jump
7609 // dummy instruction for generating temp registers
7610 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7611 match(Jump (LShiftL switch_val shift));
7612 ins_cost(350);
7613 predicate(false);
7614 effect(TEMP dest);
7616 format %{ "leaq $dest, table_base\n\t"
7617 "jmp [$dest + $switch_val << $shift]\n\t" %}
7618 ins_encode(jump_enc_offset(switch_val, shift, dest));
7619 ins_pipe(pipe_jmp);
7620 ins_pc_relative(1);
7621 %}
7623 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7624 match(Jump (AddL (LShiftL switch_val shift) offset));
7625 ins_cost(350);
7626 effect(TEMP dest);
7628 format %{ "leaq $dest, table_base\n\t"
7629 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7630 ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
7631 ins_pipe(pipe_jmp);
7632 ins_pc_relative(1);
7633 %}
7635 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7636 match(Jump switch_val);
7637 ins_cost(350);
7638 effect(TEMP dest);
7640 format %{ "leaq $dest, table_base\n\t"
7641 "jmp [$dest + $switch_val]\n\t" %}
7642 ins_encode(jump_enc(switch_val, dest));
7643 ins_pipe(pipe_jmp);
7644 ins_pc_relative(1);
7645 %}
7647 // Conditional move
7648 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7649 %{
7650 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7652 ins_cost(200); // XXX
7653 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7654 opcode(0x0F, 0x40);
7655 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7656 ins_pipe(pipe_cmov_reg);
7657 %}
7659 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7660 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7662 ins_cost(200); // XXX
7663 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7664 opcode(0x0F, 0x40);
7665 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7666 ins_pipe(pipe_cmov_reg);
7667 %}
7669 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7670 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7671 ins_cost(200);
7672 expand %{
7673 cmovI_regU(cop, cr, dst, src);
7674 %}
7675 %}
7677 // Conditional move
7678 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7679 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7681 ins_cost(250); // XXX
7682 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7683 opcode(0x0F, 0x40);
7684 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7685 ins_pipe(pipe_cmov_mem);
7686 %}
7688 // Conditional move
7689 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7690 %{
7691 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7693 ins_cost(250); // XXX
7694 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7695 opcode(0x0F, 0x40);
7696 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7697 ins_pipe(pipe_cmov_mem);
7698 %}
7700 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7701 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7702 ins_cost(250);
7703 expand %{
7704 cmovI_memU(cop, cr, dst, src);
7705 %}
7706 %}
7708 // Conditional move
7709 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7710 %{
7711 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7713 ins_cost(200); // XXX
7714 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7715 opcode(0x0F, 0x40);
7716 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7717 ins_pipe(pipe_cmov_reg);
7718 %}
7720 // Conditional move
7721 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7722 %{
7723 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7725 ins_cost(200); // XXX
7726 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7727 opcode(0x0F, 0x40);
7728 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7729 ins_pipe(pipe_cmov_reg);
7730 %}
7732 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7733 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7734 ins_cost(200);
7735 expand %{
7736 cmovN_regU(cop, cr, dst, src);
7737 %}
7738 %}
7740 // Conditional move
7741 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7742 %{
7743 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7745 ins_cost(200); // XXX
7746 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7747 opcode(0x0F, 0x40);
7748 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7749 ins_pipe(pipe_cmov_reg); // XXX
7750 %}
7752 // Conditional move
7753 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7754 %{
7755 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7757 ins_cost(200); // XXX
7758 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7759 opcode(0x0F, 0x40);
7760 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7761 ins_pipe(pipe_cmov_reg); // XXX
7762 %}
7764 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7765 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7766 ins_cost(200);
7767 expand %{
7768 cmovP_regU(cop, cr, dst, src);
7769 %}
7770 %}
7772 // DISABLED: Requires the ADLC to emit a bottom_type call that
7773 // correctly meets the two pointer arguments; one is an incoming
7774 // register but the other is a memory operand. ALSO appears to
7775 // be buggy with implicit null checks.
7776 //
7777 //// Conditional move
7778 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7779 //%{
7780 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7781 // ins_cost(250);
7782 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7783 // opcode(0x0F,0x40);
7784 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7785 // ins_pipe( pipe_cmov_mem );
7786 //%}
7787 //
7788 //// Conditional move
7789 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7790 //%{
7791 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7792 // ins_cost(250);
7793 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7794 // opcode(0x0F,0x40);
7795 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7796 // ins_pipe( pipe_cmov_mem );
7797 //%}
7799 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7800 %{
7801 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7803 ins_cost(200); // XXX
7804 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7805 opcode(0x0F, 0x40);
7806 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7807 ins_pipe(pipe_cmov_reg); // XXX
7808 %}
7810 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7811 %{
7812 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7814 ins_cost(200); // XXX
7815 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7816 opcode(0x0F, 0x40);
7817 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7818 ins_pipe(pipe_cmov_mem); // XXX
7819 %}
7821 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7822 %{
7823 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7825 ins_cost(200); // XXX
7826 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7827 opcode(0x0F, 0x40);
7828 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7829 ins_pipe(pipe_cmov_reg); // XXX
7830 %}
7832 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7833 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7834 ins_cost(200);
7835 expand %{
7836 cmovL_regU(cop, cr, dst, src);
7837 %}
7838 %}
7840 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7841 %{
7842 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7844 ins_cost(200); // XXX
7845 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7846 opcode(0x0F, 0x40);
7847 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7848 ins_pipe(pipe_cmov_mem); // XXX
7849 %}
7851 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
7852 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7853 ins_cost(200);
7854 expand %{
7855 cmovL_memU(cop, cr, dst, src);
7856 %}
7857 %}
7859 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7860 %{
7861 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7863 ins_cost(200); // XXX
7864 format %{ "jn$cop skip\t# signed cmove float\n\t"
7865 "movss $dst, $src\n"
7866 "skip:" %}
7867 ins_encode(enc_cmovf_branch(cop, dst, src));
7868 ins_pipe(pipe_slow);
7869 %}
7871 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7872 // %{
7873 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7875 // ins_cost(200); // XXX
7876 // format %{ "jn$cop skip\t# signed cmove float\n\t"
7877 // "movss $dst, $src\n"
7878 // "skip:" %}
7879 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7880 // ins_pipe(pipe_slow);
7881 // %}
7883 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7884 %{
7885 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7887 ins_cost(200); // XXX
7888 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
7889 "movss $dst, $src\n"
7890 "skip:" %}
7891 ins_encode(enc_cmovf_branch(cop, dst, src));
7892 ins_pipe(pipe_slow);
7893 %}
7895 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
7896 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7897 ins_cost(200);
7898 expand %{
7899 cmovF_regU(cop, cr, dst, src);
7900 %}
7901 %}
7903 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7904 %{
7905 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7907 ins_cost(200); // XXX
7908 format %{ "jn$cop skip\t# signed cmove double\n\t"
7909 "movsd $dst, $src\n"
7910 "skip:" %}
7911 ins_encode(enc_cmovd_branch(cop, dst, src));
7912 ins_pipe(pipe_slow);
7913 %}
7915 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7916 %{
7917 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7919 ins_cost(200); // XXX
7920 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
7921 "movsd $dst, $src\n"
7922 "skip:" %}
7923 ins_encode(enc_cmovd_branch(cop, dst, src));
7924 ins_pipe(pipe_slow);
7925 %}
7927 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
7928 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7929 ins_cost(200);
7930 expand %{
7931 cmovD_regU(cop, cr, dst, src);
7932 %}
7933 %}
7935 //----------Arithmetic Instructions--------------------------------------------
7936 //----------Addition Instructions----------------------------------------------
7938 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7939 %{
7940 match(Set dst (AddI dst src));
7941 effect(KILL cr);
7943 format %{ "addl $dst, $src\t# int" %}
7944 opcode(0x03);
7945 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7946 ins_pipe(ialu_reg_reg);
7947 %}
7949 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7950 %{
7951 match(Set dst (AddI dst src));
7952 effect(KILL cr);
7954 format %{ "addl $dst, $src\t# int" %}
7955 opcode(0x81, 0x00); /* /0 id */
7956 ins_encode(OpcSErm(dst, src), Con8or32(src));
7957 ins_pipe( ialu_reg );
7958 %}
7960 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7961 %{
7962 match(Set dst (AddI dst (LoadI src)));
7963 effect(KILL cr);
7965 ins_cost(125); // XXX
7966 format %{ "addl $dst, $src\t# int" %}
7967 opcode(0x03);
7968 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7969 ins_pipe(ialu_reg_mem);
7970 %}
7972 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7973 %{
7974 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7975 effect(KILL cr);
7977 ins_cost(150); // XXX
7978 format %{ "addl $dst, $src\t# int" %}
7979 opcode(0x01); /* Opcode 01 /r */
7980 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7981 ins_pipe(ialu_mem_reg);
7982 %}
7984 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7985 %{
7986 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7987 effect(KILL cr);
7989 ins_cost(125); // XXX
7990 format %{ "addl $dst, $src\t# int" %}
7991 opcode(0x81); /* Opcode 81 /0 id */
7992 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7993 ins_pipe(ialu_mem_imm);
7994 %}
7996 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7997 %{
7998 predicate(UseIncDec);
7999 match(Set dst (AddI dst src));
8000 effect(KILL cr);
8002 format %{ "incl $dst\t# int" %}
8003 opcode(0xFF, 0x00); // FF /0
8004 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8005 ins_pipe(ialu_reg);
8006 %}
8008 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
8009 %{
8010 predicate(UseIncDec);
8011 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8012 effect(KILL cr);
8014 ins_cost(125); // XXX
8015 format %{ "incl $dst\t# int" %}
8016 opcode(0xFF); /* Opcode FF /0 */
8017 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
8018 ins_pipe(ialu_mem_imm);
8019 %}
8021 // XXX why does that use AddI
8022 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
8023 %{
8024 predicate(UseIncDec);
8025 match(Set dst (AddI dst src));
8026 effect(KILL cr);
8028 format %{ "decl $dst\t# int" %}
8029 opcode(0xFF, 0x01); // FF /1
8030 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8031 ins_pipe(ialu_reg);
8032 %}
8034 // XXX why does that use AddI
8035 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
8036 %{
8037 predicate(UseIncDec);
8038 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8039 effect(KILL cr);
8041 ins_cost(125); // XXX
8042 format %{ "decl $dst\t# int" %}
8043 opcode(0xFF); /* Opcode FF /1 */
8044 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
8045 ins_pipe(ialu_mem_imm);
8046 %}
8048 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
8049 %{
8050 match(Set dst (AddI src0 src1));
8052 ins_cost(110);
8053 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
8054 opcode(0x8D); /* 0x8D /r */
8055 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8056 ins_pipe(ialu_reg_reg);
8057 %}
8059 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8060 %{
8061 match(Set dst (AddL dst src));
8062 effect(KILL cr);
8064 format %{ "addq $dst, $src\t# long" %}
8065 opcode(0x03);
8066 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8067 ins_pipe(ialu_reg_reg);
8068 %}
8070 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
8071 %{
8072 match(Set dst (AddL dst src));
8073 effect(KILL cr);
8075 format %{ "addq $dst, $src\t# long" %}
8076 opcode(0x81, 0x00); /* /0 id */
8077 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8078 ins_pipe( ialu_reg );
8079 %}
8081 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8082 %{
8083 match(Set dst (AddL dst (LoadL src)));
8084 effect(KILL cr);
8086 ins_cost(125); // XXX
8087 format %{ "addq $dst, $src\t# long" %}
8088 opcode(0x03);
8089 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8090 ins_pipe(ialu_reg_mem);
8091 %}
8093 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8094 %{
8095 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8096 effect(KILL cr);
8098 ins_cost(150); // XXX
8099 format %{ "addq $dst, $src\t# long" %}
8100 opcode(0x01); /* Opcode 01 /r */
8101 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8102 ins_pipe(ialu_mem_reg);
8103 %}
8105 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8106 %{
8107 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8108 effect(KILL cr);
8110 ins_cost(125); // XXX
8111 format %{ "addq $dst, $src\t# long" %}
8112 opcode(0x81); /* Opcode 81 /0 id */
8113 ins_encode(REX_mem_wide(dst),
8114 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
8115 ins_pipe(ialu_mem_imm);
8116 %}
8118 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
8119 %{
8120 predicate(UseIncDec);
8121 match(Set dst (AddL dst src));
8122 effect(KILL cr);
8124 format %{ "incq $dst\t# long" %}
8125 opcode(0xFF, 0x00); // FF /0
8126 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8127 ins_pipe(ialu_reg);
8128 %}
8130 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
8131 %{
8132 predicate(UseIncDec);
8133 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8134 effect(KILL cr);
8136 ins_cost(125); // XXX
8137 format %{ "incq $dst\t# long" %}
8138 opcode(0xFF); /* Opcode FF /0 */
8139 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
8140 ins_pipe(ialu_mem_imm);
8141 %}
8143 // XXX why does that use AddL
8144 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
8145 %{
8146 predicate(UseIncDec);
8147 match(Set dst (AddL dst src));
8148 effect(KILL cr);
8150 format %{ "decq $dst\t# long" %}
8151 opcode(0xFF, 0x01); // FF /1
8152 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8153 ins_pipe(ialu_reg);
8154 %}
8156 // XXX why does that use AddL
8157 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
8158 %{
8159 predicate(UseIncDec);
8160 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8161 effect(KILL cr);
8163 ins_cost(125); // XXX
8164 format %{ "decq $dst\t# long" %}
8165 opcode(0xFF); /* Opcode FF /1 */
8166 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
8167 ins_pipe(ialu_mem_imm);
8168 %}
8170 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
8171 %{
8172 match(Set dst (AddL src0 src1));
8174 ins_cost(110);
8175 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
8176 opcode(0x8D); /* 0x8D /r */
8177 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8178 ins_pipe(ialu_reg_reg);
8179 %}
8181 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
8182 %{
8183 match(Set dst (AddP dst src));
8184 effect(KILL cr);
8186 format %{ "addq $dst, $src\t# ptr" %}
8187 opcode(0x03);
8188 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8189 ins_pipe(ialu_reg_reg);
8190 %}
8192 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
8193 %{
8194 match(Set dst (AddP dst src));
8195 effect(KILL cr);
8197 format %{ "addq $dst, $src\t# ptr" %}
8198 opcode(0x81, 0x00); /* /0 id */
8199 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8200 ins_pipe( ialu_reg );
8201 %}
8203 // XXX addP mem ops ????
8205 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
8206 %{
8207 match(Set dst (AddP src0 src1));
8209 ins_cost(110);
8210 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
8211 opcode(0x8D); /* 0x8D /r */
8212 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
8213 ins_pipe(ialu_reg_reg);
8214 %}
8216 instruct checkCastPP(rRegP dst)
8217 %{
8218 match(Set dst (CheckCastPP dst));
8220 size(0);
8221 format %{ "# checkcastPP of $dst" %}
8222 ins_encode(/* empty encoding */);
8223 ins_pipe(empty);
8224 %}
8226 instruct castPP(rRegP dst)
8227 %{
8228 match(Set dst (CastPP dst));
8230 size(0);
8231 format %{ "# castPP of $dst" %}
8232 ins_encode(/* empty encoding */);
8233 ins_pipe(empty);
8234 %}
8236 instruct castII(rRegI dst)
8237 %{
8238 match(Set dst (CastII dst));
8240 size(0);
8241 format %{ "# castII of $dst" %}
8242 ins_encode(/* empty encoding */);
8243 ins_cost(0);
8244 ins_pipe(empty);
8245 %}
8247 // LoadP-locked same as a regular LoadP when used with compare-swap
8248 instruct loadPLocked(rRegP dst, memory mem)
8249 %{
8250 match(Set dst (LoadPLocked mem));
8252 ins_cost(125); // XXX
8253 format %{ "movq $dst, $mem\t# ptr locked" %}
8254 opcode(0x8B);
8255 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8256 ins_pipe(ialu_reg_mem); // XXX
8257 %}
8259 // LoadL-locked - same as a regular LoadL when used with compare-swap
8260 instruct loadLLocked(rRegL dst, memory mem)
8261 %{
8262 match(Set dst (LoadLLocked mem));
8264 ins_cost(125); // XXX
8265 format %{ "movq $dst, $mem\t# long locked" %}
8266 opcode(0x8B);
8267 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8268 ins_pipe(ialu_reg_mem); // XXX
8269 %}
8271 // Conditional-store of the updated heap-top.
8272 // Used during allocation of the shared heap.
8273 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
8275 instruct storePConditional(memory heap_top_ptr,
8276 rax_RegP oldval, rRegP newval,
8277 rFlagsReg cr)
8278 %{
8279 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
8281 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
8282 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
8283 opcode(0x0F, 0xB1);
8284 ins_encode(lock_prefix,
8285 REX_reg_mem_wide(newval, heap_top_ptr),
8286 OpcP, OpcS,
8287 reg_mem(newval, heap_top_ptr));
8288 ins_pipe(pipe_cmpxchg);
8289 %}
8291 // Conditional-store of an int value.
8292 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8293 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
8294 %{
8295 match(Set cr (StoreIConditional mem (Binary oldval newval)));
8296 effect(KILL oldval);
8298 format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8299 opcode(0x0F, 0xB1);
8300 ins_encode(lock_prefix,
8301 REX_reg_mem(newval, mem),
8302 OpcP, OpcS,
8303 reg_mem(newval, mem));
8304 ins_pipe(pipe_cmpxchg);
8305 %}
8307 // Conditional-store of a long value.
8308 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8309 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
8310 %{
8311 match(Set cr (StoreLConditional mem (Binary oldval newval)));
8312 effect(KILL oldval);
8314 format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8315 opcode(0x0F, 0xB1);
8316 ins_encode(lock_prefix,
8317 REX_reg_mem_wide(newval, mem),
8318 OpcP, OpcS,
8319 reg_mem(newval, mem));
8320 ins_pipe(pipe_cmpxchg);
8321 %}
8324 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
8325 instruct compareAndSwapP(rRegI res,
8326 memory mem_ptr,
8327 rax_RegP oldval, rRegP newval,
8328 rFlagsReg cr)
8329 %{
8330 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
8331 effect(KILL cr, KILL oldval);
8333 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8334 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8335 "sete $res\n\t"
8336 "movzbl $res, $res" %}
8337 opcode(0x0F, 0xB1);
8338 ins_encode(lock_prefix,
8339 REX_reg_mem_wide(newval, mem_ptr),
8340 OpcP, OpcS,
8341 reg_mem(newval, mem_ptr),
8342 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8343 REX_reg_breg(res, res), // movzbl
8344 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8345 ins_pipe( pipe_cmpxchg );
8346 %}
8348 instruct compareAndSwapL(rRegI res,
8349 memory mem_ptr,
8350 rax_RegL oldval, rRegL newval,
8351 rFlagsReg cr)
8352 %{
8353 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
8354 effect(KILL cr, KILL oldval);
8356 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8357 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8358 "sete $res\n\t"
8359 "movzbl $res, $res" %}
8360 opcode(0x0F, 0xB1);
8361 ins_encode(lock_prefix,
8362 REX_reg_mem_wide(newval, mem_ptr),
8363 OpcP, OpcS,
8364 reg_mem(newval, mem_ptr),
8365 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8366 REX_reg_breg(res, res), // movzbl
8367 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8368 ins_pipe( pipe_cmpxchg );
8369 %}
8371 instruct compareAndSwapI(rRegI res,
8372 memory mem_ptr,
8373 rax_RegI oldval, rRegI newval,
8374 rFlagsReg cr)
8375 %{
8376 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
8377 effect(KILL cr, KILL oldval);
8379 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8380 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8381 "sete $res\n\t"
8382 "movzbl $res, $res" %}
8383 opcode(0x0F, 0xB1);
8384 ins_encode(lock_prefix,
8385 REX_reg_mem(newval, mem_ptr),
8386 OpcP, OpcS,
8387 reg_mem(newval, mem_ptr),
8388 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8389 REX_reg_breg(res, res), // movzbl
8390 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8391 ins_pipe( pipe_cmpxchg );
8392 %}
8395 instruct compareAndSwapN(rRegI res,
8396 memory mem_ptr,
8397 rax_RegN oldval, rRegN newval,
8398 rFlagsReg cr) %{
8399 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
8400 effect(KILL cr, KILL oldval);
8402 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8403 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8404 "sete $res\n\t"
8405 "movzbl $res, $res" %}
8406 opcode(0x0F, 0xB1);
8407 ins_encode(lock_prefix,
8408 REX_reg_mem(newval, mem_ptr),
8409 OpcP, OpcS,
8410 reg_mem(newval, mem_ptr),
8411 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8412 REX_reg_breg(res, res), // movzbl
8413 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8414 ins_pipe( pipe_cmpxchg );
8415 %}
8417 //----------Subtraction Instructions-------------------------------------------
8419 // Integer Subtraction Instructions
8420 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8421 %{
8422 match(Set dst (SubI dst src));
8423 effect(KILL cr);
8425 format %{ "subl $dst, $src\t# int" %}
8426 opcode(0x2B);
8427 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8428 ins_pipe(ialu_reg_reg);
8429 %}
8431 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8432 %{
8433 match(Set dst (SubI dst src));
8434 effect(KILL cr);
8436 format %{ "subl $dst, $src\t# int" %}
8437 opcode(0x81, 0x05); /* Opcode 81 /5 */
8438 ins_encode(OpcSErm(dst, src), Con8or32(src));
8439 ins_pipe(ialu_reg);
8440 %}
8442 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8443 %{
8444 match(Set dst (SubI dst (LoadI src)));
8445 effect(KILL cr);
8447 ins_cost(125);
8448 format %{ "subl $dst, $src\t# int" %}
8449 opcode(0x2B);
8450 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8451 ins_pipe(ialu_reg_mem);
8452 %}
8454 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8455 %{
8456 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8457 effect(KILL cr);
8459 ins_cost(150);
8460 format %{ "subl $dst, $src\t# int" %}
8461 opcode(0x29); /* Opcode 29 /r */
8462 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8463 ins_pipe(ialu_mem_reg);
8464 %}
8466 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8467 %{
8468 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8469 effect(KILL cr);
8471 ins_cost(125); // XXX
8472 format %{ "subl $dst, $src\t# int" %}
8473 opcode(0x81); /* Opcode 81 /5 id */
8474 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8475 ins_pipe(ialu_mem_imm);
8476 %}
8478 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8479 %{
8480 match(Set dst (SubL dst src));
8481 effect(KILL cr);
8483 format %{ "subq $dst, $src\t# long" %}
8484 opcode(0x2B);
8485 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8486 ins_pipe(ialu_reg_reg);
8487 %}
8489 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8490 %{
8491 match(Set dst (SubL dst src));
8492 effect(KILL cr);
8494 format %{ "subq $dst, $src\t# long" %}
8495 opcode(0x81, 0x05); /* Opcode 81 /5 */
8496 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8497 ins_pipe(ialu_reg);
8498 %}
8500 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8501 %{
8502 match(Set dst (SubL dst (LoadL src)));
8503 effect(KILL cr);
8505 ins_cost(125);
8506 format %{ "subq $dst, $src\t# long" %}
8507 opcode(0x2B);
8508 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8509 ins_pipe(ialu_reg_mem);
8510 %}
8512 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8513 %{
8514 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8515 effect(KILL cr);
8517 ins_cost(150);
8518 format %{ "subq $dst, $src\t# long" %}
8519 opcode(0x29); /* Opcode 29 /r */
8520 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8521 ins_pipe(ialu_mem_reg);
8522 %}
8524 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8525 %{
8526 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8527 effect(KILL cr);
8529 ins_cost(125); // XXX
8530 format %{ "subq $dst, $src\t# long" %}
8531 opcode(0x81); /* Opcode 81 /5 id */
8532 ins_encode(REX_mem_wide(dst),
8533 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8534 ins_pipe(ialu_mem_imm);
8535 %}
8537 // Subtract from a pointer
8538 // XXX hmpf???
8539 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8540 %{
8541 match(Set dst (AddP dst (SubI zero src)));
8542 effect(KILL cr);
8544 format %{ "subq $dst, $src\t# ptr - int" %}
8545 opcode(0x2B);
8546 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8547 ins_pipe(ialu_reg_reg);
8548 %}
8550 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8551 %{
8552 match(Set dst (SubI zero dst));
8553 effect(KILL cr);
8555 format %{ "negl $dst\t# int" %}
8556 opcode(0xF7, 0x03); // Opcode F7 /3
8557 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8558 ins_pipe(ialu_reg);
8559 %}
8561 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8562 %{
8563 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8564 effect(KILL cr);
8566 format %{ "negl $dst\t# int" %}
8567 opcode(0xF7, 0x03); // Opcode F7 /3
8568 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8569 ins_pipe(ialu_reg);
8570 %}
8572 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8573 %{
8574 match(Set dst (SubL zero dst));
8575 effect(KILL cr);
8577 format %{ "negq $dst\t# long" %}
8578 opcode(0xF7, 0x03); // Opcode F7 /3
8579 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8580 ins_pipe(ialu_reg);
8581 %}
8583 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8584 %{
8585 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8586 effect(KILL cr);
8588 format %{ "negq $dst\t# long" %}
8589 opcode(0xF7, 0x03); // Opcode F7 /3
8590 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8591 ins_pipe(ialu_reg);
8592 %}
8595 //----------Multiplication/Division Instructions-------------------------------
8596 // Integer Multiplication Instructions
8597 // Multiply Register
8599 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8600 %{
8601 match(Set dst (MulI dst src));
8602 effect(KILL cr);
8604 ins_cost(300);
8605 format %{ "imull $dst, $src\t# int" %}
8606 opcode(0x0F, 0xAF);
8607 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8608 ins_pipe(ialu_reg_reg_alu0);
8609 %}
8611 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8612 %{
8613 match(Set dst (MulI src imm));
8614 effect(KILL cr);
8616 ins_cost(300);
8617 format %{ "imull $dst, $src, $imm\t# int" %}
8618 opcode(0x69); /* 69 /r id */
8619 ins_encode(REX_reg_reg(dst, src),
8620 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8621 ins_pipe(ialu_reg_reg_alu0);
8622 %}
8624 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8625 %{
8626 match(Set dst (MulI dst (LoadI src)));
8627 effect(KILL cr);
8629 ins_cost(350);
8630 format %{ "imull $dst, $src\t# int" %}
8631 opcode(0x0F, 0xAF);
8632 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8633 ins_pipe(ialu_reg_mem_alu0);
8634 %}
8636 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8637 %{
8638 match(Set dst (MulI (LoadI src) imm));
8639 effect(KILL cr);
8641 ins_cost(300);
8642 format %{ "imull $dst, $src, $imm\t# int" %}
8643 opcode(0x69); /* 69 /r id */
8644 ins_encode(REX_reg_mem(dst, src),
8645 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8646 ins_pipe(ialu_reg_mem_alu0);
8647 %}
8649 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8650 %{
8651 match(Set dst (MulL dst src));
8652 effect(KILL cr);
8654 ins_cost(300);
8655 format %{ "imulq $dst, $src\t# long" %}
8656 opcode(0x0F, 0xAF);
8657 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8658 ins_pipe(ialu_reg_reg_alu0);
8659 %}
8661 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8662 %{
8663 match(Set dst (MulL src imm));
8664 effect(KILL cr);
8666 ins_cost(300);
8667 format %{ "imulq $dst, $src, $imm\t# long" %}
8668 opcode(0x69); /* 69 /r id */
8669 ins_encode(REX_reg_reg_wide(dst, src),
8670 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8671 ins_pipe(ialu_reg_reg_alu0);
8672 %}
8674 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8675 %{
8676 match(Set dst (MulL dst (LoadL src)));
8677 effect(KILL cr);
8679 ins_cost(350);
8680 format %{ "imulq $dst, $src\t# long" %}
8681 opcode(0x0F, 0xAF);
8682 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8683 ins_pipe(ialu_reg_mem_alu0);
8684 %}
8686 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8687 %{
8688 match(Set dst (MulL (LoadL src) imm));
8689 effect(KILL cr);
8691 ins_cost(300);
8692 format %{ "imulq $dst, $src, $imm\t# long" %}
8693 opcode(0x69); /* 69 /r id */
8694 ins_encode(REX_reg_mem_wide(dst, src),
8695 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8696 ins_pipe(ialu_reg_mem_alu0);
8697 %}
8699 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8700 %{
8701 match(Set dst (MulHiL src rax));
8702 effect(USE_KILL rax, KILL cr);
8704 ins_cost(300);
8705 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %}
8706 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8707 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8708 ins_pipe(ialu_reg_reg_alu0);
8709 %}
8711 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8712 rFlagsReg cr)
8713 %{
8714 match(Set rax (DivI rax div));
8715 effect(KILL rdx, KILL cr);
8717 ins_cost(30*100+10*100); // XXX
8718 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8719 "jne,s normal\n\t"
8720 "xorl rdx, rdx\n\t"
8721 "cmpl $div, -1\n\t"
8722 "je,s done\n"
8723 "normal: cdql\n\t"
8724 "idivl $div\n"
8725 "done:" %}
8726 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8727 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8728 ins_pipe(ialu_reg_reg_alu0);
8729 %}
8731 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8732 rFlagsReg cr)
8733 %{
8734 match(Set rax (DivL rax div));
8735 effect(KILL rdx, KILL cr);
8737 ins_cost(30*100+10*100); // XXX
8738 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8739 "cmpq rax, rdx\n\t"
8740 "jne,s normal\n\t"
8741 "xorl rdx, rdx\n\t"
8742 "cmpq $div, -1\n\t"
8743 "je,s done\n"
8744 "normal: cdqq\n\t"
8745 "idivq $div\n"
8746 "done:" %}
8747 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8748 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8749 ins_pipe(ialu_reg_reg_alu0);
8750 %}
8752 // Integer DIVMOD with Register, both quotient and mod results
8753 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8754 rFlagsReg cr)
8755 %{
8756 match(DivModI rax div);
8757 effect(KILL cr);
8759 ins_cost(30*100+10*100); // XXX
8760 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8761 "jne,s normal\n\t"
8762 "xorl rdx, rdx\n\t"
8763 "cmpl $div, -1\n\t"
8764 "je,s done\n"
8765 "normal: cdql\n\t"
8766 "idivl $div\n"
8767 "done:" %}
8768 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8769 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8770 ins_pipe(pipe_slow);
8771 %}
8773 // Long DIVMOD with Register, both quotient and mod results
8774 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8775 rFlagsReg cr)
8776 %{
8777 match(DivModL rax div);
8778 effect(KILL cr);
8780 ins_cost(30*100+10*100); // XXX
8781 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8782 "cmpq rax, rdx\n\t"
8783 "jne,s normal\n\t"
8784 "xorl rdx, rdx\n\t"
8785 "cmpq $div, -1\n\t"
8786 "je,s done\n"
8787 "normal: cdqq\n\t"
8788 "idivq $div\n"
8789 "done:" %}
8790 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8791 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8792 ins_pipe(pipe_slow);
8793 %}
8795 //----------- DivL-By-Constant-Expansions--------------------------------------
8796 // DivI cases are handled by the compiler
8798 // Magic constant, reciprocal of 10
8799 instruct loadConL_0x6666666666666667(rRegL dst)
8800 %{
8801 effect(DEF dst);
8803 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8804 ins_encode(load_immL(dst, 0x6666666666666667));
8805 ins_pipe(ialu_reg);
8806 %}
8808 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8809 %{
8810 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8812 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8813 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8814 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8815 ins_pipe(ialu_reg_reg_alu0);
8816 %}
8818 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8819 %{
8820 effect(USE_DEF dst, KILL cr);
8822 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8823 opcode(0xC1, 0x7); /* C1 /7 ib */
8824 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8825 ins_pipe(ialu_reg);
8826 %}
8828 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8829 %{
8830 effect(USE_DEF dst, KILL cr);
8832 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8833 opcode(0xC1, 0x7); /* C1 /7 ib */
8834 ins_encode(reg_opc_imm_wide(dst, 0x2));
8835 ins_pipe(ialu_reg);
8836 %}
8838 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8839 %{
8840 match(Set dst (DivL src div));
8842 ins_cost((5+8)*100);
8843 expand %{
8844 rax_RegL rax; // Killed temp
8845 rFlagsReg cr; // Killed
8846 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
8847 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
8848 sarL_rReg_63(src, cr); // sarq src, 63
8849 sarL_rReg_2(dst, cr); // sarq rdx, 2
8850 subL_rReg(dst, src, cr); // subl rdx, src
8851 %}
8852 %}
8854 //-----------------------------------------------------------------------------
8856 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8857 rFlagsReg cr)
8858 %{
8859 match(Set rdx (ModI rax div));
8860 effect(KILL rax, KILL cr);
8862 ins_cost(300); // XXX
8863 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
8864 "jne,s normal\n\t"
8865 "xorl rdx, rdx\n\t"
8866 "cmpl $div, -1\n\t"
8867 "je,s done\n"
8868 "normal: cdql\n\t"
8869 "idivl $div\n"
8870 "done:" %}
8871 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8872 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8873 ins_pipe(ialu_reg_reg_alu0);
8874 %}
8876 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8877 rFlagsReg cr)
8878 %{
8879 match(Set rdx (ModL rax div));
8880 effect(KILL rax, KILL cr);
8882 ins_cost(300); // XXX
8883 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
8884 "cmpq rax, rdx\n\t"
8885 "jne,s normal\n\t"
8886 "xorl rdx, rdx\n\t"
8887 "cmpq $div, -1\n\t"
8888 "je,s done\n"
8889 "normal: cdqq\n\t"
8890 "idivq $div\n"
8891 "done:" %}
8892 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8893 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8894 ins_pipe(ialu_reg_reg_alu0);
8895 %}
8897 // Integer Shift Instructions
8898 // Shift Left by one
8899 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8900 %{
8901 match(Set dst (LShiftI dst shift));
8902 effect(KILL cr);
8904 format %{ "sall $dst, $shift" %}
8905 opcode(0xD1, 0x4); /* D1 /4 */
8906 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8907 ins_pipe(ialu_reg);
8908 %}
8910 // Shift Left by one
8911 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8912 %{
8913 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8914 effect(KILL cr);
8916 format %{ "sall $dst, $shift\t" %}
8917 opcode(0xD1, 0x4); /* D1 /4 */
8918 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8919 ins_pipe(ialu_mem_imm);
8920 %}
8922 // Shift Left by 8-bit immediate
8923 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8924 %{
8925 match(Set dst (LShiftI dst shift));
8926 effect(KILL cr);
8928 format %{ "sall $dst, $shift" %}
8929 opcode(0xC1, 0x4); /* C1 /4 ib */
8930 ins_encode(reg_opc_imm(dst, shift));
8931 ins_pipe(ialu_reg);
8932 %}
8934 // Shift Left by 8-bit immediate
8935 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8936 %{
8937 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8938 effect(KILL cr);
8940 format %{ "sall $dst, $shift" %}
8941 opcode(0xC1, 0x4); /* C1 /4 ib */
8942 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8943 ins_pipe(ialu_mem_imm);
8944 %}
8946 // Shift Left by variable
8947 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8948 %{
8949 match(Set dst (LShiftI dst shift));
8950 effect(KILL cr);
8952 format %{ "sall $dst, $shift" %}
8953 opcode(0xD3, 0x4); /* D3 /4 */
8954 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8955 ins_pipe(ialu_reg_reg);
8956 %}
8958 // Shift Left by variable
8959 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8960 %{
8961 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8962 effect(KILL cr);
8964 format %{ "sall $dst, $shift" %}
8965 opcode(0xD3, 0x4); /* D3 /4 */
8966 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8967 ins_pipe(ialu_mem_reg);
8968 %}
8970 // Arithmetic shift right by one
8971 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8972 %{
8973 match(Set dst (RShiftI dst shift));
8974 effect(KILL cr);
8976 format %{ "sarl $dst, $shift" %}
8977 opcode(0xD1, 0x7); /* D1 /7 */
8978 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8979 ins_pipe(ialu_reg);
8980 %}
8982 // Arithmetic shift right by one
8983 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8984 %{
8985 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8986 effect(KILL cr);
8988 format %{ "sarl $dst, $shift" %}
8989 opcode(0xD1, 0x7); /* D1 /7 */
8990 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8991 ins_pipe(ialu_mem_imm);
8992 %}
8994 // Arithmetic Shift Right by 8-bit immediate
8995 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8996 %{
8997 match(Set dst (RShiftI dst shift));
8998 effect(KILL cr);
9000 format %{ "sarl $dst, $shift" %}
9001 opcode(0xC1, 0x7); /* C1 /7 ib */
9002 ins_encode(reg_opc_imm(dst, shift));
9003 ins_pipe(ialu_mem_imm);
9004 %}
9006 // Arithmetic Shift Right by 8-bit immediate
9007 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9008 %{
9009 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9010 effect(KILL cr);
9012 format %{ "sarl $dst, $shift" %}
9013 opcode(0xC1, 0x7); /* C1 /7 ib */
9014 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9015 ins_pipe(ialu_mem_imm);
9016 %}
9018 // Arithmetic Shift Right by variable
9019 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9020 %{
9021 match(Set dst (RShiftI dst shift));
9022 effect(KILL cr);
9024 format %{ "sarl $dst, $shift" %}
9025 opcode(0xD3, 0x7); /* D3 /7 */
9026 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9027 ins_pipe(ialu_reg_reg);
9028 %}
9030 // Arithmetic Shift Right by variable
9031 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9032 %{
9033 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9034 effect(KILL cr);
9036 format %{ "sarl $dst, $shift" %}
9037 opcode(0xD3, 0x7); /* D3 /7 */
9038 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9039 ins_pipe(ialu_mem_reg);
9040 %}
9042 // Logical shift right by one
9043 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
9044 %{
9045 match(Set dst (URShiftI dst shift));
9046 effect(KILL cr);
9048 format %{ "shrl $dst, $shift" %}
9049 opcode(0xD1, 0x5); /* D1 /5 */
9050 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9051 ins_pipe(ialu_reg);
9052 %}
9054 // Logical shift right by one
9055 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9056 %{
9057 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9058 effect(KILL cr);
9060 format %{ "shrl $dst, $shift" %}
9061 opcode(0xD1, 0x5); /* D1 /5 */
9062 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9063 ins_pipe(ialu_mem_imm);
9064 %}
9066 // Logical Shift Right by 8-bit immediate
9067 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9068 %{
9069 match(Set dst (URShiftI dst shift));
9070 effect(KILL cr);
9072 format %{ "shrl $dst, $shift" %}
9073 opcode(0xC1, 0x5); /* C1 /5 ib */
9074 ins_encode(reg_opc_imm(dst, shift));
9075 ins_pipe(ialu_reg);
9076 %}
9078 // Logical Shift Right by 8-bit immediate
9079 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9080 %{
9081 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9082 effect(KILL cr);
9084 format %{ "shrl $dst, $shift" %}
9085 opcode(0xC1, 0x5); /* C1 /5 ib */
9086 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9087 ins_pipe(ialu_mem_imm);
9088 %}
9090 // Logical Shift Right by variable
9091 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9092 %{
9093 match(Set dst (URShiftI dst shift));
9094 effect(KILL cr);
9096 format %{ "shrl $dst, $shift" %}
9097 opcode(0xD3, 0x5); /* D3 /5 */
9098 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9099 ins_pipe(ialu_reg_reg);
9100 %}
9102 // Logical Shift Right by variable
9103 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9104 %{
9105 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9106 effect(KILL cr);
9108 format %{ "shrl $dst, $shift" %}
9109 opcode(0xD3, 0x5); /* D3 /5 */
9110 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9111 ins_pipe(ialu_mem_reg);
9112 %}
9114 // Long Shift Instructions
9115 // Shift Left by one
9116 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9117 %{
9118 match(Set dst (LShiftL dst shift));
9119 effect(KILL cr);
9121 format %{ "salq $dst, $shift" %}
9122 opcode(0xD1, 0x4); /* D1 /4 */
9123 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9124 ins_pipe(ialu_reg);
9125 %}
9127 // Shift Left by one
9128 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9129 %{
9130 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9131 effect(KILL cr);
9133 format %{ "salq $dst, $shift" %}
9134 opcode(0xD1, 0x4); /* D1 /4 */
9135 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9136 ins_pipe(ialu_mem_imm);
9137 %}
9139 // Shift Left by 8-bit immediate
9140 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9141 %{
9142 match(Set dst (LShiftL dst shift));
9143 effect(KILL cr);
9145 format %{ "salq $dst, $shift" %}
9146 opcode(0xC1, 0x4); /* C1 /4 ib */
9147 ins_encode(reg_opc_imm_wide(dst, shift));
9148 ins_pipe(ialu_reg);
9149 %}
9151 // Shift Left by 8-bit immediate
9152 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9153 %{
9154 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9155 effect(KILL cr);
9157 format %{ "salq $dst, $shift" %}
9158 opcode(0xC1, 0x4); /* C1 /4 ib */
9159 ins_encode(REX_mem_wide(dst), OpcP,
9160 RM_opc_mem(secondary, dst), Con8or32(shift));
9161 ins_pipe(ialu_mem_imm);
9162 %}
9164 // Shift Left by variable
9165 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9166 %{
9167 match(Set dst (LShiftL dst shift));
9168 effect(KILL cr);
9170 format %{ "salq $dst, $shift" %}
9171 opcode(0xD3, 0x4); /* D3 /4 */
9172 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9173 ins_pipe(ialu_reg_reg);
9174 %}
9176 // Shift Left by variable
9177 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9178 %{
9179 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9180 effect(KILL cr);
9182 format %{ "salq $dst, $shift" %}
9183 opcode(0xD3, 0x4); /* D3 /4 */
9184 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9185 ins_pipe(ialu_mem_reg);
9186 %}
9188 // Arithmetic shift right by one
9189 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9190 %{
9191 match(Set dst (RShiftL dst shift));
9192 effect(KILL cr);
9194 format %{ "sarq $dst, $shift" %}
9195 opcode(0xD1, 0x7); /* D1 /7 */
9196 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9197 ins_pipe(ialu_reg);
9198 %}
9200 // Arithmetic shift right by one
9201 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9202 %{
9203 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9204 effect(KILL cr);
9206 format %{ "sarq $dst, $shift" %}
9207 opcode(0xD1, 0x7); /* D1 /7 */
9208 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9209 ins_pipe(ialu_mem_imm);
9210 %}
9212 // Arithmetic Shift Right by 8-bit immediate
9213 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9214 %{
9215 match(Set dst (RShiftL dst shift));
9216 effect(KILL cr);
9218 format %{ "sarq $dst, $shift" %}
9219 opcode(0xC1, 0x7); /* C1 /7 ib */
9220 ins_encode(reg_opc_imm_wide(dst, shift));
9221 ins_pipe(ialu_mem_imm);
9222 %}
9224 // Arithmetic Shift Right by 8-bit immediate
9225 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9226 %{
9227 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9228 effect(KILL cr);
9230 format %{ "sarq $dst, $shift" %}
9231 opcode(0xC1, 0x7); /* C1 /7 ib */
9232 ins_encode(REX_mem_wide(dst), OpcP,
9233 RM_opc_mem(secondary, dst), Con8or32(shift));
9234 ins_pipe(ialu_mem_imm);
9235 %}
9237 // Arithmetic Shift Right by variable
9238 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9239 %{
9240 match(Set dst (RShiftL dst shift));
9241 effect(KILL cr);
9243 format %{ "sarq $dst, $shift" %}
9244 opcode(0xD3, 0x7); /* D3 /7 */
9245 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9246 ins_pipe(ialu_reg_reg);
9247 %}
9249 // Arithmetic Shift Right by variable
9250 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9251 %{
9252 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9253 effect(KILL cr);
9255 format %{ "sarq $dst, $shift" %}
9256 opcode(0xD3, 0x7); /* D3 /7 */
9257 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9258 ins_pipe(ialu_mem_reg);
9259 %}
9261 // Logical shift right by one
9262 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9263 %{
9264 match(Set dst (URShiftL dst shift));
9265 effect(KILL cr);
9267 format %{ "shrq $dst, $shift" %}
9268 opcode(0xD1, 0x5); /* D1 /5 */
9269 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
9270 ins_pipe(ialu_reg);
9271 %}
9273 // Logical shift right by one
9274 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9275 %{
9276 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9277 effect(KILL cr);
9279 format %{ "shrq $dst, $shift" %}
9280 opcode(0xD1, 0x5); /* D1 /5 */
9281 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9282 ins_pipe(ialu_mem_imm);
9283 %}
9285 // Logical Shift Right by 8-bit immediate
9286 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9287 %{
9288 match(Set dst (URShiftL dst shift));
9289 effect(KILL cr);
9291 format %{ "shrq $dst, $shift" %}
9292 opcode(0xC1, 0x5); /* C1 /5 ib */
9293 ins_encode(reg_opc_imm_wide(dst, shift));
9294 ins_pipe(ialu_reg);
9295 %}
9298 // Logical Shift Right by 8-bit immediate
9299 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9300 %{
9301 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9302 effect(KILL cr);
9304 format %{ "shrq $dst, $shift" %}
9305 opcode(0xC1, 0x5); /* C1 /5 ib */
9306 ins_encode(REX_mem_wide(dst), OpcP,
9307 RM_opc_mem(secondary, dst), Con8or32(shift));
9308 ins_pipe(ialu_mem_imm);
9309 %}
9311 // Logical Shift Right by variable
9312 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9313 %{
9314 match(Set dst (URShiftL dst shift));
9315 effect(KILL cr);
9317 format %{ "shrq $dst, $shift" %}
9318 opcode(0xD3, 0x5); /* D3 /5 */
9319 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9320 ins_pipe(ialu_reg_reg);
9321 %}
9323 // Logical Shift Right by variable
9324 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9325 %{
9326 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9327 effect(KILL cr);
9329 format %{ "shrq $dst, $shift" %}
9330 opcode(0xD3, 0x5); /* D3 /5 */
9331 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9332 ins_pipe(ialu_mem_reg);
9333 %}
9335 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
9336 // This idiom is used by the compiler for the i2b bytecode.
9337 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
9338 %{
9339 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
9341 format %{ "movsbl $dst, $src\t# i2b" %}
9342 opcode(0x0F, 0xBE);
9343 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9344 ins_pipe(ialu_reg_reg);
9345 %}
9347 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
9348 // This idiom is used by the compiler the i2s bytecode.
9349 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
9350 %{
9351 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
9353 format %{ "movswl $dst, $src\t# i2s" %}
9354 opcode(0x0F, 0xBF);
9355 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9356 ins_pipe(ialu_reg_reg);
9357 %}
9359 // ROL/ROR instructions
9361 // ROL expand
9362 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
9363 effect(KILL cr, USE_DEF dst);
9365 format %{ "roll $dst" %}
9366 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9367 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9368 ins_pipe(ialu_reg);
9369 %}
9371 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
9372 effect(USE_DEF dst, USE shift, KILL cr);
9374 format %{ "roll $dst, $shift" %}
9375 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9376 ins_encode( reg_opc_imm(dst, shift) );
9377 ins_pipe(ialu_reg);
9378 %}
9380 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9381 %{
9382 effect(USE_DEF dst, USE shift, KILL cr);
9384 format %{ "roll $dst, $shift" %}
9385 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9386 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9387 ins_pipe(ialu_reg_reg);
9388 %}
9389 // end of ROL expand
9391 // Rotate Left by one
9392 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9393 %{
9394 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9396 expand %{
9397 rolI_rReg_imm1(dst, cr);
9398 %}
9399 %}
9401 // Rotate Left by 8-bit immediate
9402 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9403 %{
9404 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9405 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9407 expand %{
9408 rolI_rReg_imm8(dst, lshift, cr);
9409 %}
9410 %}
9412 // Rotate Left by variable
9413 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9414 %{
9415 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
9417 expand %{
9418 rolI_rReg_CL(dst, shift, cr);
9419 %}
9420 %}
9422 // Rotate Left by variable
9423 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9424 %{
9425 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9427 expand %{
9428 rolI_rReg_CL(dst, shift, cr);
9429 %}
9430 %}
9432 // ROR expand
9433 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9434 %{
9435 effect(USE_DEF dst, KILL cr);
9437 format %{ "rorl $dst" %}
9438 opcode(0xD1, 0x1); /* D1 /1 */
9439 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9440 ins_pipe(ialu_reg);
9441 %}
9443 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9444 %{
9445 effect(USE_DEF dst, USE shift, KILL cr);
9447 format %{ "rorl $dst, $shift" %}
9448 opcode(0xC1, 0x1); /* C1 /1 ib */
9449 ins_encode(reg_opc_imm(dst, shift));
9450 ins_pipe(ialu_reg);
9451 %}
9453 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9454 %{
9455 effect(USE_DEF dst, USE shift, KILL cr);
9457 format %{ "rorl $dst, $shift" %}
9458 opcode(0xD3, 0x1); /* D3 /1 */
9459 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9460 ins_pipe(ialu_reg_reg);
9461 %}
9462 // end of ROR expand
9464 // Rotate Right by one
9465 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9466 %{
9467 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9469 expand %{
9470 rorI_rReg_imm1(dst, cr);
9471 %}
9472 %}
9474 // Rotate Right by 8-bit immediate
9475 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9476 %{
9477 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9478 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9480 expand %{
9481 rorI_rReg_imm8(dst, rshift, cr);
9482 %}
9483 %}
9485 // Rotate Right by variable
9486 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9487 %{
9488 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9490 expand %{
9491 rorI_rReg_CL(dst, shift, cr);
9492 %}
9493 %}
9495 // Rotate Right by variable
9496 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9497 %{
9498 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9500 expand %{
9501 rorI_rReg_CL(dst, shift, cr);
9502 %}
9503 %}
9505 // for long rotate
9506 // ROL expand
9507 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9508 effect(USE_DEF dst, KILL cr);
9510 format %{ "rolq $dst" %}
9511 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9512 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9513 ins_pipe(ialu_reg);
9514 %}
9516 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9517 effect(USE_DEF dst, USE shift, KILL cr);
9519 format %{ "rolq $dst, $shift" %}
9520 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9521 ins_encode( reg_opc_imm_wide(dst, shift) );
9522 ins_pipe(ialu_reg);
9523 %}
9525 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9526 %{
9527 effect(USE_DEF dst, USE shift, KILL cr);
9529 format %{ "rolq $dst, $shift" %}
9530 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9531 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9532 ins_pipe(ialu_reg_reg);
9533 %}
9534 // end of ROL expand
9536 // Rotate Left by one
9537 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9538 %{
9539 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9541 expand %{
9542 rolL_rReg_imm1(dst, cr);
9543 %}
9544 %}
9546 // Rotate Left by 8-bit immediate
9547 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9548 %{
9549 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9550 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9552 expand %{
9553 rolL_rReg_imm8(dst, lshift, cr);
9554 %}
9555 %}
9557 // Rotate Left by variable
9558 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9559 %{
9560 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9562 expand %{
9563 rolL_rReg_CL(dst, shift, cr);
9564 %}
9565 %}
9567 // Rotate Left by variable
9568 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9569 %{
9570 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9572 expand %{
9573 rolL_rReg_CL(dst, shift, cr);
9574 %}
9575 %}
9577 // ROR expand
9578 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9579 %{
9580 effect(USE_DEF dst, KILL cr);
9582 format %{ "rorq $dst" %}
9583 opcode(0xD1, 0x1); /* D1 /1 */
9584 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9585 ins_pipe(ialu_reg);
9586 %}
9588 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9589 %{
9590 effect(USE_DEF dst, USE shift, KILL cr);
9592 format %{ "rorq $dst, $shift" %}
9593 opcode(0xC1, 0x1); /* C1 /1 ib */
9594 ins_encode(reg_opc_imm_wide(dst, shift));
9595 ins_pipe(ialu_reg);
9596 %}
9598 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9599 %{
9600 effect(USE_DEF dst, USE shift, KILL cr);
9602 format %{ "rorq $dst, $shift" %}
9603 opcode(0xD3, 0x1); /* D3 /1 */
9604 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9605 ins_pipe(ialu_reg_reg);
9606 %}
9607 // end of ROR expand
9609 // Rotate Right by one
9610 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9611 %{
9612 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9614 expand %{
9615 rorL_rReg_imm1(dst, cr);
9616 %}
9617 %}
9619 // Rotate Right by 8-bit immediate
9620 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9621 %{
9622 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9623 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9625 expand %{
9626 rorL_rReg_imm8(dst, rshift, cr);
9627 %}
9628 %}
9630 // Rotate Right by variable
9631 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9632 %{
9633 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9635 expand %{
9636 rorL_rReg_CL(dst, shift, cr);
9637 %}
9638 %}
9640 // Rotate Right by variable
9641 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9642 %{
9643 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9645 expand %{
9646 rorL_rReg_CL(dst, shift, cr);
9647 %}
9648 %}
9650 // Logical Instructions
9652 // Integer Logical Instructions
9654 // And Instructions
9655 // And Register with Register
9656 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9657 %{
9658 match(Set dst (AndI dst src));
9659 effect(KILL cr);
9661 format %{ "andl $dst, $src\t# int" %}
9662 opcode(0x23);
9663 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9664 ins_pipe(ialu_reg_reg);
9665 %}
9667 // And Register with Immediate 255
9668 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9669 %{
9670 match(Set dst (AndI dst src));
9672 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9673 opcode(0x0F, 0xB6);
9674 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9675 ins_pipe(ialu_reg);
9676 %}
9678 // And Register with Immediate 255 and promote to long
9679 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9680 %{
9681 match(Set dst (ConvI2L (AndI src mask)));
9683 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9684 opcode(0x0F, 0xB6);
9685 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9686 ins_pipe(ialu_reg);
9687 %}
9689 // And Register with Immediate 65535
9690 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9691 %{
9692 match(Set dst (AndI dst src));
9694 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9695 opcode(0x0F, 0xB7);
9696 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9697 ins_pipe(ialu_reg);
9698 %}
9700 // And Register with Immediate 65535 and promote to long
9701 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9702 %{
9703 match(Set dst (ConvI2L (AndI src mask)));
9705 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9706 opcode(0x0F, 0xB7);
9707 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9708 ins_pipe(ialu_reg);
9709 %}
9711 // And Register with Immediate
9712 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9713 %{
9714 match(Set dst (AndI dst src));
9715 effect(KILL cr);
9717 format %{ "andl $dst, $src\t# int" %}
9718 opcode(0x81, 0x04); /* Opcode 81 /4 */
9719 ins_encode(OpcSErm(dst, src), Con8or32(src));
9720 ins_pipe(ialu_reg);
9721 %}
9723 // And Register with Memory
9724 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9725 %{
9726 match(Set dst (AndI dst (LoadI src)));
9727 effect(KILL cr);
9729 ins_cost(125);
9730 format %{ "andl $dst, $src\t# int" %}
9731 opcode(0x23);
9732 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9733 ins_pipe(ialu_reg_mem);
9734 %}
9736 // And Memory with Register
9737 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9738 %{
9739 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9740 effect(KILL cr);
9742 ins_cost(150);
9743 format %{ "andl $dst, $src\t# int" %}
9744 opcode(0x21); /* Opcode 21 /r */
9745 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9746 ins_pipe(ialu_mem_reg);
9747 %}
9749 // And Memory with Immediate
9750 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9751 %{
9752 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9753 effect(KILL cr);
9755 ins_cost(125);
9756 format %{ "andl $dst, $src\t# int" %}
9757 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9758 ins_encode(REX_mem(dst), OpcSE(src),
9759 RM_opc_mem(secondary, dst), Con8or32(src));
9760 ins_pipe(ialu_mem_imm);
9761 %}
9763 // Or Instructions
9764 // Or Register with Register
9765 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9766 %{
9767 match(Set dst (OrI dst src));
9768 effect(KILL cr);
9770 format %{ "orl $dst, $src\t# int" %}
9771 opcode(0x0B);
9772 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9773 ins_pipe(ialu_reg_reg);
9774 %}
9776 // Or Register with Immediate
9777 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9778 %{
9779 match(Set dst (OrI dst src));
9780 effect(KILL cr);
9782 format %{ "orl $dst, $src\t# int" %}
9783 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9784 ins_encode(OpcSErm(dst, src), Con8or32(src));
9785 ins_pipe(ialu_reg);
9786 %}
9788 // Or Register with Memory
9789 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9790 %{
9791 match(Set dst (OrI dst (LoadI src)));
9792 effect(KILL cr);
9794 ins_cost(125);
9795 format %{ "orl $dst, $src\t# int" %}
9796 opcode(0x0B);
9797 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9798 ins_pipe(ialu_reg_mem);
9799 %}
9801 // Or Memory with Register
9802 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9803 %{
9804 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9805 effect(KILL cr);
9807 ins_cost(150);
9808 format %{ "orl $dst, $src\t# int" %}
9809 opcode(0x09); /* Opcode 09 /r */
9810 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9811 ins_pipe(ialu_mem_reg);
9812 %}
9814 // Or Memory with Immediate
9815 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9816 %{
9817 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9818 effect(KILL cr);
9820 ins_cost(125);
9821 format %{ "orl $dst, $src\t# int" %}
9822 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9823 ins_encode(REX_mem(dst), OpcSE(src),
9824 RM_opc_mem(secondary, dst), Con8or32(src));
9825 ins_pipe(ialu_mem_imm);
9826 %}
9828 // Xor Instructions
9829 // Xor Register with Register
9830 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9831 %{
9832 match(Set dst (XorI dst src));
9833 effect(KILL cr);
9835 format %{ "xorl $dst, $src\t# int" %}
9836 opcode(0x33);
9837 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9838 ins_pipe(ialu_reg_reg);
9839 %}
9841 // Xor Register with Immediate -1
9842 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9843 match(Set dst (XorI dst imm));
9845 format %{ "not $dst" %}
9846 ins_encode %{
9847 __ notl($dst$$Register);
9848 %}
9849 ins_pipe(ialu_reg);
9850 %}
9852 // Xor Register with Immediate
9853 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9854 %{
9855 match(Set dst (XorI dst src));
9856 effect(KILL cr);
9858 format %{ "xorl $dst, $src\t# int" %}
9859 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9860 ins_encode(OpcSErm(dst, src), Con8or32(src));
9861 ins_pipe(ialu_reg);
9862 %}
9864 // Xor Register with Memory
9865 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9866 %{
9867 match(Set dst (XorI dst (LoadI src)));
9868 effect(KILL cr);
9870 ins_cost(125);
9871 format %{ "xorl $dst, $src\t# int" %}
9872 opcode(0x33);
9873 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9874 ins_pipe(ialu_reg_mem);
9875 %}
9877 // Xor Memory with Register
9878 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9879 %{
9880 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9881 effect(KILL cr);
9883 ins_cost(150);
9884 format %{ "xorl $dst, $src\t# int" %}
9885 opcode(0x31); /* Opcode 31 /r */
9886 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9887 ins_pipe(ialu_mem_reg);
9888 %}
9890 // Xor Memory with Immediate
9891 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9892 %{
9893 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9894 effect(KILL cr);
9896 ins_cost(125);
9897 format %{ "xorl $dst, $src\t# int" %}
9898 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9899 ins_encode(REX_mem(dst), OpcSE(src),
9900 RM_opc_mem(secondary, dst), Con8or32(src));
9901 ins_pipe(ialu_mem_imm);
9902 %}
9905 // Long Logical Instructions
9907 // And Instructions
9908 // And Register with Register
9909 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9910 %{
9911 match(Set dst (AndL dst src));
9912 effect(KILL cr);
9914 format %{ "andq $dst, $src\t# long" %}
9915 opcode(0x23);
9916 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9917 ins_pipe(ialu_reg_reg);
9918 %}
9920 // And Register with Immediate 255
9921 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9922 %{
9923 match(Set dst (AndL dst src));
9925 format %{ "movzbq $dst, $dst\t# long & 0xFF" %}
9926 opcode(0x0F, 0xB6);
9927 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9928 ins_pipe(ialu_reg);
9929 %}
9931 // And Register with Immediate 65535
9932 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src)
9933 %{
9934 match(Set dst (AndL dst src));
9936 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
9937 opcode(0x0F, 0xB7);
9938 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9939 ins_pipe(ialu_reg);
9940 %}
9942 // And Register with Immediate
9943 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9944 %{
9945 match(Set dst (AndL dst src));
9946 effect(KILL cr);
9948 format %{ "andq $dst, $src\t# long" %}
9949 opcode(0x81, 0x04); /* Opcode 81 /4 */
9950 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9951 ins_pipe(ialu_reg);
9952 %}
9954 // And Register with Memory
9955 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9956 %{
9957 match(Set dst (AndL dst (LoadL src)));
9958 effect(KILL cr);
9960 ins_cost(125);
9961 format %{ "andq $dst, $src\t# long" %}
9962 opcode(0x23);
9963 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9964 ins_pipe(ialu_reg_mem);
9965 %}
9967 // And Memory with Register
9968 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9969 %{
9970 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9971 effect(KILL cr);
9973 ins_cost(150);
9974 format %{ "andq $dst, $src\t# long" %}
9975 opcode(0x21); /* Opcode 21 /r */
9976 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9977 ins_pipe(ialu_mem_reg);
9978 %}
9980 // And Memory with Immediate
9981 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9982 %{
9983 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9984 effect(KILL cr);
9986 ins_cost(125);
9987 format %{ "andq $dst, $src\t# long" %}
9988 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9989 ins_encode(REX_mem_wide(dst), OpcSE(src),
9990 RM_opc_mem(secondary, dst), Con8or32(src));
9991 ins_pipe(ialu_mem_imm);
9992 %}
9994 // Or Instructions
9995 // Or Register with Register
9996 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9997 %{
9998 match(Set dst (OrL dst src));
9999 effect(KILL cr);
10001 format %{ "orq $dst, $src\t# long" %}
10002 opcode(0x0B);
10003 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10004 ins_pipe(ialu_reg_reg);
10005 %}
10007 // Use any_RegP to match R15 (TLS register) without spilling.
10008 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
10009 match(Set dst (OrL dst (CastP2X src)));
10010 effect(KILL cr);
10012 format %{ "orq $dst, $src\t# long" %}
10013 opcode(0x0B);
10014 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10015 ins_pipe(ialu_reg_reg);
10016 %}
10019 // Or Register with Immediate
10020 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10021 %{
10022 match(Set dst (OrL dst src));
10023 effect(KILL cr);
10025 format %{ "orq $dst, $src\t# long" %}
10026 opcode(0x81, 0x01); /* Opcode 81 /1 id */
10027 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10028 ins_pipe(ialu_reg);
10029 %}
10031 // Or Register with Memory
10032 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10033 %{
10034 match(Set dst (OrL dst (LoadL src)));
10035 effect(KILL cr);
10037 ins_cost(125);
10038 format %{ "orq $dst, $src\t# long" %}
10039 opcode(0x0B);
10040 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10041 ins_pipe(ialu_reg_mem);
10042 %}
10044 // Or Memory with Register
10045 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10046 %{
10047 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10048 effect(KILL cr);
10050 ins_cost(150);
10051 format %{ "orq $dst, $src\t# long" %}
10052 opcode(0x09); /* Opcode 09 /r */
10053 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10054 ins_pipe(ialu_mem_reg);
10055 %}
10057 // Or Memory with Immediate
10058 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10059 %{
10060 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10061 effect(KILL cr);
10063 ins_cost(125);
10064 format %{ "orq $dst, $src\t# long" %}
10065 opcode(0x81, 0x1); /* Opcode 81 /1 id */
10066 ins_encode(REX_mem_wide(dst), OpcSE(src),
10067 RM_opc_mem(secondary, dst), Con8or32(src));
10068 ins_pipe(ialu_mem_imm);
10069 %}
10071 // Xor Instructions
10072 // Xor Register with Register
10073 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10074 %{
10075 match(Set dst (XorL dst src));
10076 effect(KILL cr);
10078 format %{ "xorq $dst, $src\t# long" %}
10079 opcode(0x33);
10080 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10081 ins_pipe(ialu_reg_reg);
10082 %}
10084 // Xor Register with Immediate -1
10085 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
10086 match(Set dst (XorL dst imm));
10088 format %{ "notq $dst" %}
10089 ins_encode %{
10090 __ notq($dst$$Register);
10091 %}
10092 ins_pipe(ialu_reg);
10093 %}
10095 // Xor Register with Immediate
10096 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10097 %{
10098 match(Set dst (XorL dst src));
10099 effect(KILL cr);
10101 format %{ "xorq $dst, $src\t# long" %}
10102 opcode(0x81, 0x06); /* Opcode 81 /6 id */
10103 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10104 ins_pipe(ialu_reg);
10105 %}
10107 // Xor Register with Memory
10108 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10109 %{
10110 match(Set dst (XorL dst (LoadL src)));
10111 effect(KILL cr);
10113 ins_cost(125);
10114 format %{ "xorq $dst, $src\t# long" %}
10115 opcode(0x33);
10116 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10117 ins_pipe(ialu_reg_mem);
10118 %}
10120 // Xor Memory with Register
10121 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10122 %{
10123 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10124 effect(KILL cr);
10126 ins_cost(150);
10127 format %{ "xorq $dst, $src\t# long" %}
10128 opcode(0x31); /* Opcode 31 /r */
10129 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10130 ins_pipe(ialu_mem_reg);
10131 %}
10133 // Xor Memory with Immediate
10134 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10135 %{
10136 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10137 effect(KILL cr);
10139 ins_cost(125);
10140 format %{ "xorq $dst, $src\t# long" %}
10141 opcode(0x81, 0x6); /* Opcode 81 /6 id */
10142 ins_encode(REX_mem_wide(dst), OpcSE(src),
10143 RM_opc_mem(secondary, dst), Con8or32(src));
10144 ins_pipe(ialu_mem_imm);
10145 %}
10147 // Convert Int to Boolean
10148 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
10149 %{
10150 match(Set dst (Conv2B src));
10151 effect(KILL cr);
10153 format %{ "testl $src, $src\t# ci2b\n\t"
10154 "setnz $dst\n\t"
10155 "movzbl $dst, $dst" %}
10156 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
10157 setNZ_reg(dst),
10158 REX_reg_breg(dst, dst), // movzbl
10159 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10160 ins_pipe(pipe_slow); // XXX
10161 %}
10163 // Convert Pointer to Boolean
10164 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
10165 %{
10166 match(Set dst (Conv2B src));
10167 effect(KILL cr);
10169 format %{ "testq $src, $src\t# cp2b\n\t"
10170 "setnz $dst\n\t"
10171 "movzbl $dst, $dst" %}
10172 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
10173 setNZ_reg(dst),
10174 REX_reg_breg(dst, dst), // movzbl
10175 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10176 ins_pipe(pipe_slow); // XXX
10177 %}
10179 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
10180 %{
10181 match(Set dst (CmpLTMask p q));
10182 effect(KILL cr);
10184 ins_cost(400); // XXX
10185 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
10186 "setlt $dst\n\t"
10187 "movzbl $dst, $dst\n\t"
10188 "negl $dst" %}
10189 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
10190 setLT_reg(dst),
10191 REX_reg_breg(dst, dst), // movzbl
10192 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
10193 neg_reg(dst));
10194 ins_pipe(pipe_slow);
10195 %}
10197 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
10198 %{
10199 match(Set dst (CmpLTMask dst zero));
10200 effect(KILL cr);
10202 ins_cost(100); // XXX
10203 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
10204 opcode(0xC1, 0x7); /* C1 /7 ib */
10205 ins_encode(reg_opc_imm(dst, 0x1F));
10206 ins_pipe(ialu_reg);
10207 %}
10210 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
10211 rRegI tmp,
10212 rFlagsReg cr)
10213 %{
10214 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
10215 effect(TEMP tmp, KILL cr);
10217 ins_cost(400); // XXX
10218 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
10219 "sbbl $tmp, $tmp\n\t"
10220 "andl $tmp, $y\n\t"
10221 "addl $p, $tmp" %}
10222 ins_encode(enc_cmpLTP(p, q, y, tmp));
10223 ins_pipe(pipe_cmplt);
10224 %}
10226 /* If I enable this, I encourage spilling in the inner loop of compress.
10227 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
10228 %{
10229 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
10230 effect( TEMP tmp, KILL cr );
10231 ins_cost(400);
10233 format %{ "SUB $p,$q\n\t"
10234 "SBB RCX,RCX\n\t"
10235 "AND RCX,$y\n\t"
10236 "ADD $p,RCX" %}
10237 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
10238 %}
10239 */
10241 //---------- FP Instructions------------------------------------------------
10243 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
10244 %{
10245 match(Set cr (CmpF src1 src2));
10247 ins_cost(145);
10248 format %{ "ucomiss $src1, $src2\n\t"
10249 "jnp,s exit\n\t"
10250 "pushfq\t# saw NaN, set CF\n\t"
10251 "andq [rsp], #0xffffff2b\n\t"
10252 "popfq\n"
10253 "exit: nop\t# avoid branch to branch" %}
10254 opcode(0x0F, 0x2E);
10255 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10256 cmpfp_fixup);
10257 ins_pipe(pipe_slow);
10258 %}
10260 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
10261 match(Set cr (CmpF src1 src2));
10263 ins_cost(145);
10264 format %{ "ucomiss $src1, $src2" %}
10265 ins_encode %{
10266 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10267 %}
10268 ins_pipe(pipe_slow);
10269 %}
10271 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
10272 %{
10273 match(Set cr (CmpF src1 (LoadF src2)));
10275 ins_cost(145);
10276 format %{ "ucomiss $src1, $src2\n\t"
10277 "jnp,s exit\n\t"
10278 "pushfq\t# saw NaN, set CF\n\t"
10279 "andq [rsp], #0xffffff2b\n\t"
10280 "popfq\n"
10281 "exit: nop\t# avoid branch to branch" %}
10282 opcode(0x0F, 0x2E);
10283 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10284 cmpfp_fixup);
10285 ins_pipe(pipe_slow);
10286 %}
10288 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
10289 match(Set cr (CmpF src1 (LoadF src2)));
10291 ins_cost(100);
10292 format %{ "ucomiss $src1, $src2" %}
10293 opcode(0x0F, 0x2E);
10294 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
10295 ins_pipe(pipe_slow);
10296 %}
10298 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
10299 %{
10300 match(Set cr (CmpF src1 src2));
10302 ins_cost(145);
10303 format %{ "ucomiss $src1, $src2\n\t"
10304 "jnp,s exit\n\t"
10305 "pushfq\t# saw NaN, set CF\n\t"
10306 "andq [rsp], #0xffffff2b\n\t"
10307 "popfq\n"
10308 "exit: nop\t# avoid branch to branch" %}
10309 opcode(0x0F, 0x2E);
10310 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10311 cmpfp_fixup);
10312 ins_pipe(pipe_slow);
10313 %}
10315 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{
10316 match(Set cr (CmpF src1 src2));
10318 ins_cost(100);
10319 format %{ "ucomiss $src1, $src2" %}
10320 opcode(0x0F, 0x2E);
10321 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2));
10322 ins_pipe(pipe_slow);
10323 %}
10325 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
10326 %{
10327 match(Set cr (CmpD src1 src2));
10329 ins_cost(145);
10330 format %{ "ucomisd $src1, $src2\n\t"
10331 "jnp,s exit\n\t"
10332 "pushfq\t# saw NaN, set CF\n\t"
10333 "andq [rsp], #0xffffff2b\n\t"
10334 "popfq\n"
10335 "exit: nop\t# avoid branch to branch" %}
10336 opcode(0x66, 0x0F, 0x2E);
10337 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10338 cmpfp_fixup);
10339 ins_pipe(pipe_slow);
10340 %}
10342 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
10343 match(Set cr (CmpD src1 src2));
10345 ins_cost(100);
10346 format %{ "ucomisd $src1, $src2 test" %}
10347 ins_encode %{
10348 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
10349 %}
10350 ins_pipe(pipe_slow);
10351 %}
10353 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
10354 %{
10355 match(Set cr (CmpD src1 (LoadD src2)));
10357 ins_cost(145);
10358 format %{ "ucomisd $src1, $src2\n\t"
10359 "jnp,s exit\n\t"
10360 "pushfq\t# saw NaN, set CF\n\t"
10361 "andq [rsp], #0xffffff2b\n\t"
10362 "popfq\n"
10363 "exit: nop\t# avoid branch to branch" %}
10364 opcode(0x66, 0x0F, 0x2E);
10365 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10366 cmpfp_fixup);
10367 ins_pipe(pipe_slow);
10368 %}
10370 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
10371 match(Set cr (CmpD src1 (LoadD src2)));
10373 ins_cost(100);
10374 format %{ "ucomisd $src1, $src2" %}
10375 opcode(0x66, 0x0F, 0x2E);
10376 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
10377 ins_pipe(pipe_slow);
10378 %}
10380 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
10381 %{
10382 match(Set cr (CmpD src1 src2));
10384 ins_cost(145);
10385 format %{ "ucomisd $src1, [$src2]\n\t"
10386 "jnp,s exit\n\t"
10387 "pushfq\t# saw NaN, set CF\n\t"
10388 "andq [rsp], #0xffffff2b\n\t"
10389 "popfq\n"
10390 "exit: nop\t# avoid branch to branch" %}
10391 opcode(0x66, 0x0F, 0x2E);
10392 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10393 cmpfp_fixup);
10394 ins_pipe(pipe_slow);
10395 %}
10397 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{
10398 match(Set cr (CmpD src1 src2));
10400 ins_cost(100);
10401 format %{ "ucomisd $src1, [$src2]" %}
10402 opcode(0x66, 0x0F, 0x2E);
10403 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2));
10404 ins_pipe(pipe_slow);
10405 %}
10407 // Compare into -1,0,1
10408 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
10409 %{
10410 match(Set dst (CmpF3 src1 src2));
10411 effect(KILL cr);
10413 ins_cost(275);
10414 format %{ "ucomiss $src1, $src2\n\t"
10415 "movl $dst, #-1\n\t"
10416 "jp,s done\n\t"
10417 "jb,s done\n\t"
10418 "setne $dst\n\t"
10419 "movzbl $dst, $dst\n"
10420 "done:" %}
10422 opcode(0x0F, 0x2E);
10423 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10424 cmpfp3(dst));
10425 ins_pipe(pipe_slow);
10426 %}
10428 // Compare into -1,0,1
10429 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
10430 %{
10431 match(Set dst (CmpF3 src1 (LoadF src2)));
10432 effect(KILL cr);
10434 ins_cost(275);
10435 format %{ "ucomiss $src1, $src2\n\t"
10436 "movl $dst, #-1\n\t"
10437 "jp,s done\n\t"
10438 "jb,s done\n\t"
10439 "setne $dst\n\t"
10440 "movzbl $dst, $dst\n"
10441 "done:" %}
10443 opcode(0x0F, 0x2E);
10444 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10445 cmpfp3(dst));
10446 ins_pipe(pipe_slow);
10447 %}
10449 // Compare into -1,0,1
10450 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
10451 %{
10452 match(Set dst (CmpF3 src1 src2));
10453 effect(KILL cr);
10455 ins_cost(275);
10456 format %{ "ucomiss $src1, [$src2]\n\t"
10457 "movl $dst, #-1\n\t"
10458 "jp,s done\n\t"
10459 "jb,s done\n\t"
10460 "setne $dst\n\t"
10461 "movzbl $dst, $dst\n"
10462 "done:" %}
10464 opcode(0x0F, 0x2E);
10465 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10466 cmpfp3(dst));
10467 ins_pipe(pipe_slow);
10468 %}
10470 // Compare into -1,0,1
10471 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10472 %{
10473 match(Set dst (CmpD3 src1 src2));
10474 effect(KILL cr);
10476 ins_cost(275);
10477 format %{ "ucomisd $src1, $src2\n\t"
10478 "movl $dst, #-1\n\t"
10479 "jp,s done\n\t"
10480 "jb,s done\n\t"
10481 "setne $dst\n\t"
10482 "movzbl $dst, $dst\n"
10483 "done:" %}
10485 opcode(0x66, 0x0F, 0x2E);
10486 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10487 cmpfp3(dst));
10488 ins_pipe(pipe_slow);
10489 %}
10491 // Compare into -1,0,1
10492 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10493 %{
10494 match(Set dst (CmpD3 src1 (LoadD src2)));
10495 effect(KILL cr);
10497 ins_cost(275);
10498 format %{ "ucomisd $src1, $src2\n\t"
10499 "movl $dst, #-1\n\t"
10500 "jp,s done\n\t"
10501 "jb,s done\n\t"
10502 "setne $dst\n\t"
10503 "movzbl $dst, $dst\n"
10504 "done:" %}
10506 opcode(0x66, 0x0F, 0x2E);
10507 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10508 cmpfp3(dst));
10509 ins_pipe(pipe_slow);
10510 %}
10512 // Compare into -1,0,1
10513 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
10514 %{
10515 match(Set dst (CmpD3 src1 src2));
10516 effect(KILL cr);
10518 ins_cost(275);
10519 format %{ "ucomisd $src1, [$src2]\n\t"
10520 "movl $dst, #-1\n\t"
10521 "jp,s done\n\t"
10522 "jb,s done\n\t"
10523 "setne $dst\n\t"
10524 "movzbl $dst, $dst\n"
10525 "done:" %}
10527 opcode(0x66, 0x0F, 0x2E);
10528 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10529 cmpfp3(dst));
10530 ins_pipe(pipe_slow);
10531 %}
10533 instruct addF_reg(regF dst, regF src)
10534 %{
10535 match(Set dst (AddF dst src));
10537 format %{ "addss $dst, $src" %}
10538 ins_cost(150); // XXX
10539 opcode(0xF3, 0x0F, 0x58);
10540 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10541 ins_pipe(pipe_slow);
10542 %}
10544 instruct addF_mem(regF dst, memory src)
10545 %{
10546 match(Set dst (AddF dst (LoadF src)));
10548 format %{ "addss $dst, $src" %}
10549 ins_cost(150); // XXX
10550 opcode(0xF3, 0x0F, 0x58);
10551 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10552 ins_pipe(pipe_slow);
10553 %}
10555 instruct addF_imm(regF dst, immF src)
10556 %{
10557 match(Set dst (AddF dst src));
10559 format %{ "addss $dst, [$src]" %}
10560 ins_cost(150); // XXX
10561 opcode(0xF3, 0x0F, 0x58);
10562 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10563 ins_pipe(pipe_slow);
10564 %}
10566 instruct addD_reg(regD dst, regD src)
10567 %{
10568 match(Set dst (AddD dst src));
10570 format %{ "addsd $dst, $src" %}
10571 ins_cost(150); // XXX
10572 opcode(0xF2, 0x0F, 0x58);
10573 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10574 ins_pipe(pipe_slow);
10575 %}
10577 instruct addD_mem(regD dst, memory src)
10578 %{
10579 match(Set dst (AddD dst (LoadD src)));
10581 format %{ "addsd $dst, $src" %}
10582 ins_cost(150); // XXX
10583 opcode(0xF2, 0x0F, 0x58);
10584 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10585 ins_pipe(pipe_slow);
10586 %}
10588 instruct addD_imm(regD dst, immD src)
10589 %{
10590 match(Set dst (AddD dst src));
10592 format %{ "addsd $dst, [$src]" %}
10593 ins_cost(150); // XXX
10594 opcode(0xF2, 0x0F, 0x58);
10595 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10596 ins_pipe(pipe_slow);
10597 %}
10599 instruct subF_reg(regF dst, regF src)
10600 %{
10601 match(Set dst (SubF dst src));
10603 format %{ "subss $dst, $src" %}
10604 ins_cost(150); // XXX
10605 opcode(0xF3, 0x0F, 0x5C);
10606 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10607 ins_pipe(pipe_slow);
10608 %}
10610 instruct subF_mem(regF dst, memory src)
10611 %{
10612 match(Set dst (SubF dst (LoadF src)));
10614 format %{ "subss $dst, $src" %}
10615 ins_cost(150); // XXX
10616 opcode(0xF3, 0x0F, 0x5C);
10617 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10618 ins_pipe(pipe_slow);
10619 %}
10621 instruct subF_imm(regF dst, immF src)
10622 %{
10623 match(Set dst (SubF dst src));
10625 format %{ "subss $dst, [$src]" %}
10626 ins_cost(150); // XXX
10627 opcode(0xF3, 0x0F, 0x5C);
10628 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10629 ins_pipe(pipe_slow);
10630 %}
10632 instruct subD_reg(regD dst, regD src)
10633 %{
10634 match(Set dst (SubD dst src));
10636 format %{ "subsd $dst, $src" %}
10637 ins_cost(150); // XXX
10638 opcode(0xF2, 0x0F, 0x5C);
10639 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10640 ins_pipe(pipe_slow);
10641 %}
10643 instruct subD_mem(regD dst, memory src)
10644 %{
10645 match(Set dst (SubD dst (LoadD src)));
10647 format %{ "subsd $dst, $src" %}
10648 ins_cost(150); // XXX
10649 opcode(0xF2, 0x0F, 0x5C);
10650 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10651 ins_pipe(pipe_slow);
10652 %}
10654 instruct subD_imm(regD dst, immD src)
10655 %{
10656 match(Set dst (SubD dst src));
10658 format %{ "subsd $dst, [$src]" %}
10659 ins_cost(150); // XXX
10660 opcode(0xF2, 0x0F, 0x5C);
10661 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10662 ins_pipe(pipe_slow);
10663 %}
10665 instruct mulF_reg(regF dst, regF src)
10666 %{
10667 match(Set dst (MulF dst src));
10669 format %{ "mulss $dst, $src" %}
10670 ins_cost(150); // XXX
10671 opcode(0xF3, 0x0F, 0x59);
10672 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10673 ins_pipe(pipe_slow);
10674 %}
10676 instruct mulF_mem(regF dst, memory src)
10677 %{
10678 match(Set dst (MulF dst (LoadF src)));
10680 format %{ "mulss $dst, $src" %}
10681 ins_cost(150); // XXX
10682 opcode(0xF3, 0x0F, 0x59);
10683 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10684 ins_pipe(pipe_slow);
10685 %}
10687 instruct mulF_imm(regF dst, immF src)
10688 %{
10689 match(Set dst (MulF dst src));
10691 format %{ "mulss $dst, [$src]" %}
10692 ins_cost(150); // XXX
10693 opcode(0xF3, 0x0F, 0x59);
10694 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10695 ins_pipe(pipe_slow);
10696 %}
10698 instruct mulD_reg(regD dst, regD src)
10699 %{
10700 match(Set dst (MulD dst src));
10702 format %{ "mulsd $dst, $src" %}
10703 ins_cost(150); // XXX
10704 opcode(0xF2, 0x0F, 0x59);
10705 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10706 ins_pipe(pipe_slow);
10707 %}
10709 instruct mulD_mem(regD dst, memory src)
10710 %{
10711 match(Set dst (MulD dst (LoadD src)));
10713 format %{ "mulsd $dst, $src" %}
10714 ins_cost(150); // XXX
10715 opcode(0xF2, 0x0F, 0x59);
10716 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10717 ins_pipe(pipe_slow);
10718 %}
10720 instruct mulD_imm(regD dst, immD src)
10721 %{
10722 match(Set dst (MulD dst src));
10724 format %{ "mulsd $dst, [$src]" %}
10725 ins_cost(150); // XXX
10726 opcode(0xF2, 0x0F, 0x59);
10727 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10728 ins_pipe(pipe_slow);
10729 %}
10731 instruct divF_reg(regF dst, regF src)
10732 %{
10733 match(Set dst (DivF dst src));
10735 format %{ "divss $dst, $src" %}
10736 ins_cost(150); // XXX
10737 opcode(0xF3, 0x0F, 0x5E);
10738 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10739 ins_pipe(pipe_slow);
10740 %}
10742 instruct divF_mem(regF dst, memory src)
10743 %{
10744 match(Set dst (DivF dst (LoadF src)));
10746 format %{ "divss $dst, $src" %}
10747 ins_cost(150); // XXX
10748 opcode(0xF3, 0x0F, 0x5E);
10749 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10750 ins_pipe(pipe_slow);
10751 %}
10753 instruct divF_imm(regF dst, immF src)
10754 %{
10755 match(Set dst (DivF dst src));
10757 format %{ "divss $dst, [$src]" %}
10758 ins_cost(150); // XXX
10759 opcode(0xF3, 0x0F, 0x5E);
10760 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10761 ins_pipe(pipe_slow);
10762 %}
10764 instruct divD_reg(regD dst, regD src)
10765 %{
10766 match(Set dst (DivD dst src));
10768 format %{ "divsd $dst, $src" %}
10769 ins_cost(150); // XXX
10770 opcode(0xF2, 0x0F, 0x5E);
10771 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10772 ins_pipe(pipe_slow);
10773 %}
10775 instruct divD_mem(regD dst, memory src)
10776 %{
10777 match(Set dst (DivD dst (LoadD src)));
10779 format %{ "divsd $dst, $src" %}
10780 ins_cost(150); // XXX
10781 opcode(0xF2, 0x0F, 0x5E);
10782 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10783 ins_pipe(pipe_slow);
10784 %}
10786 instruct divD_imm(regD dst, immD src)
10787 %{
10788 match(Set dst (DivD dst src));
10790 format %{ "divsd $dst, [$src]" %}
10791 ins_cost(150); // XXX
10792 opcode(0xF2, 0x0F, 0x5E);
10793 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10794 ins_pipe(pipe_slow);
10795 %}
10797 instruct sqrtF_reg(regF dst, regF src)
10798 %{
10799 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10801 format %{ "sqrtss $dst, $src" %}
10802 ins_cost(150); // XXX
10803 opcode(0xF3, 0x0F, 0x51);
10804 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10805 ins_pipe(pipe_slow);
10806 %}
10808 instruct sqrtF_mem(regF dst, memory src)
10809 %{
10810 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10812 format %{ "sqrtss $dst, $src" %}
10813 ins_cost(150); // XXX
10814 opcode(0xF3, 0x0F, 0x51);
10815 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10816 ins_pipe(pipe_slow);
10817 %}
10819 instruct sqrtF_imm(regF dst, immF src)
10820 %{
10821 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10823 format %{ "sqrtss $dst, [$src]" %}
10824 ins_cost(150); // XXX
10825 opcode(0xF3, 0x0F, 0x51);
10826 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10827 ins_pipe(pipe_slow);
10828 %}
10830 instruct sqrtD_reg(regD dst, regD src)
10831 %{
10832 match(Set dst (SqrtD src));
10834 format %{ "sqrtsd $dst, $src" %}
10835 ins_cost(150); // XXX
10836 opcode(0xF2, 0x0F, 0x51);
10837 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10838 ins_pipe(pipe_slow);
10839 %}
10841 instruct sqrtD_mem(regD dst, memory src)
10842 %{
10843 match(Set dst (SqrtD (LoadD src)));
10845 format %{ "sqrtsd $dst, $src" %}
10846 ins_cost(150); // XXX
10847 opcode(0xF2, 0x0F, 0x51);
10848 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10849 ins_pipe(pipe_slow);
10850 %}
10852 instruct sqrtD_imm(regD dst, immD src)
10853 %{
10854 match(Set dst (SqrtD src));
10856 format %{ "sqrtsd $dst, [$src]" %}
10857 ins_cost(150); // XXX
10858 opcode(0xF2, 0x0F, 0x51);
10859 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10860 ins_pipe(pipe_slow);
10861 %}
10863 instruct absF_reg(regF dst)
10864 %{
10865 match(Set dst (AbsF dst));
10867 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
10868 ins_encode(absF_encoding(dst));
10869 ins_pipe(pipe_slow);
10870 %}
10872 instruct absD_reg(regD dst)
10873 %{
10874 match(Set dst (AbsD dst));
10876 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
10877 "# abs double by sign masking" %}
10878 ins_encode(absD_encoding(dst));
10879 ins_pipe(pipe_slow);
10880 %}
10882 instruct negF_reg(regF dst)
10883 %{
10884 match(Set dst (NegF dst));
10886 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
10887 ins_encode(negF_encoding(dst));
10888 ins_pipe(pipe_slow);
10889 %}
10891 instruct negD_reg(regD dst)
10892 %{
10893 match(Set dst (NegD dst));
10895 format %{ "xorpd $dst, [0x8000000000000000]\t"
10896 "# neg double by sign flipping" %}
10897 ins_encode(negD_encoding(dst));
10898 ins_pipe(pipe_slow);
10899 %}
10901 // -----------Trig and Trancendental Instructions------------------------------
10902 instruct cosD_reg(regD dst) %{
10903 match(Set dst (CosD dst));
10905 format %{ "dcos $dst\n\t" %}
10906 opcode(0xD9, 0xFF);
10907 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10908 ins_pipe( pipe_slow );
10909 %}
10911 instruct sinD_reg(regD dst) %{
10912 match(Set dst (SinD dst));
10914 format %{ "dsin $dst\n\t" %}
10915 opcode(0xD9, 0xFE);
10916 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10917 ins_pipe( pipe_slow );
10918 %}
10920 instruct tanD_reg(regD dst) %{
10921 match(Set dst (TanD dst));
10923 format %{ "dtan $dst\n\t" %}
10924 ins_encode( Push_SrcXD(dst),
10925 Opcode(0xD9), Opcode(0xF2), //fptan
10926 Opcode(0xDD), Opcode(0xD8), //fstp st
10927 Push_ResultXD(dst) );
10928 ins_pipe( pipe_slow );
10929 %}
10931 instruct log10D_reg(regD dst) %{
10932 // The source and result Double operands in XMM registers
10933 match(Set dst (Log10D dst));
10934 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
10935 // fyl2x ; compute log_10(2) * log_2(x)
10936 format %{ "fldlg2\t\t\t#Log10\n\t"
10937 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10938 %}
10939 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
10940 Push_SrcXD(dst),
10941 Opcode(0xD9), Opcode(0xF1), // fyl2x
10942 Push_ResultXD(dst));
10944 ins_pipe( pipe_slow );
10945 %}
10947 instruct logD_reg(regD dst) %{
10948 // The source and result Double operands in XMM registers
10949 match(Set dst (LogD dst));
10950 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
10951 // fyl2x ; compute log_e(2) * log_2(x)
10952 format %{ "fldln2\t\t\t#Log_e\n\t"
10953 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10954 %}
10955 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
10956 Push_SrcXD(dst),
10957 Opcode(0xD9), Opcode(0xF1), // fyl2x
10958 Push_ResultXD(dst));
10959 ins_pipe( pipe_slow );
10960 %}
10964 //----------Arithmetic Conversion Instructions---------------------------------
10966 instruct roundFloat_nop(regF dst)
10967 %{
10968 match(Set dst (RoundFloat dst));
10970 ins_cost(0);
10971 ins_encode();
10972 ins_pipe(empty);
10973 %}
10975 instruct roundDouble_nop(regD dst)
10976 %{
10977 match(Set dst (RoundDouble dst));
10979 ins_cost(0);
10980 ins_encode();
10981 ins_pipe(empty);
10982 %}
10984 instruct convF2D_reg_reg(regD dst, regF src)
10985 %{
10986 match(Set dst (ConvF2D src));
10988 format %{ "cvtss2sd $dst, $src" %}
10989 opcode(0xF3, 0x0F, 0x5A);
10990 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10991 ins_pipe(pipe_slow); // XXX
10992 %}
10994 instruct convF2D_reg_mem(regD dst, memory src)
10995 %{
10996 match(Set dst (ConvF2D (LoadF src)));
10998 format %{ "cvtss2sd $dst, $src" %}
10999 opcode(0xF3, 0x0F, 0x5A);
11000 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11001 ins_pipe(pipe_slow); // XXX
11002 %}
11004 instruct convD2F_reg_reg(regF dst, regD src)
11005 %{
11006 match(Set dst (ConvD2F src));
11008 format %{ "cvtsd2ss $dst, $src" %}
11009 opcode(0xF2, 0x0F, 0x5A);
11010 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11011 ins_pipe(pipe_slow); // XXX
11012 %}
11014 instruct convD2F_reg_mem(regF dst, memory src)
11015 %{
11016 match(Set dst (ConvD2F (LoadD src)));
11018 format %{ "cvtsd2ss $dst, $src" %}
11019 opcode(0xF2, 0x0F, 0x5A);
11020 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11021 ins_pipe(pipe_slow); // XXX
11022 %}
11024 // XXX do mem variants
11025 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
11026 %{
11027 match(Set dst (ConvF2I src));
11028 effect(KILL cr);
11030 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
11031 "cmpl $dst, #0x80000000\n\t"
11032 "jne,s done\n\t"
11033 "subq rsp, #8\n\t"
11034 "movss [rsp], $src\n\t"
11035 "call f2i_fixup\n\t"
11036 "popq $dst\n"
11037 "done: "%}
11038 opcode(0xF3, 0x0F, 0x2C);
11039 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11040 f2i_fixup(dst, src));
11041 ins_pipe(pipe_slow);
11042 %}
11044 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
11045 %{
11046 match(Set dst (ConvF2L src));
11047 effect(KILL cr);
11049 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
11050 "cmpq $dst, [0x8000000000000000]\n\t"
11051 "jne,s done\n\t"
11052 "subq rsp, #8\n\t"
11053 "movss [rsp], $src\n\t"
11054 "call f2l_fixup\n\t"
11055 "popq $dst\n"
11056 "done: "%}
11057 opcode(0xF3, 0x0F, 0x2C);
11058 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11059 f2l_fixup(dst, src));
11060 ins_pipe(pipe_slow);
11061 %}
11063 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
11064 %{
11065 match(Set dst (ConvD2I src));
11066 effect(KILL cr);
11068 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
11069 "cmpl $dst, #0x80000000\n\t"
11070 "jne,s done\n\t"
11071 "subq rsp, #8\n\t"
11072 "movsd [rsp], $src\n\t"
11073 "call d2i_fixup\n\t"
11074 "popq $dst\n"
11075 "done: "%}
11076 opcode(0xF2, 0x0F, 0x2C);
11077 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11078 d2i_fixup(dst, src));
11079 ins_pipe(pipe_slow);
11080 %}
11082 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
11083 %{
11084 match(Set dst (ConvD2L src));
11085 effect(KILL cr);
11087 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
11088 "cmpq $dst, [0x8000000000000000]\n\t"
11089 "jne,s done\n\t"
11090 "subq rsp, #8\n\t"
11091 "movsd [rsp], $src\n\t"
11092 "call d2l_fixup\n\t"
11093 "popq $dst\n"
11094 "done: "%}
11095 opcode(0xF2, 0x0F, 0x2C);
11096 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11097 d2l_fixup(dst, src));
11098 ins_pipe(pipe_slow);
11099 %}
11101 instruct convI2F_reg_reg(regF dst, rRegI src)
11102 %{
11103 predicate(!UseXmmI2F);
11104 match(Set dst (ConvI2F src));
11106 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11107 opcode(0xF3, 0x0F, 0x2A);
11108 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11109 ins_pipe(pipe_slow); // XXX
11110 %}
11112 instruct convI2F_reg_mem(regF dst, memory src)
11113 %{
11114 match(Set dst (ConvI2F (LoadI src)));
11116 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11117 opcode(0xF3, 0x0F, 0x2A);
11118 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11119 ins_pipe(pipe_slow); // XXX
11120 %}
11122 instruct convI2D_reg_reg(regD dst, rRegI src)
11123 %{
11124 predicate(!UseXmmI2D);
11125 match(Set dst (ConvI2D src));
11127 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11128 opcode(0xF2, 0x0F, 0x2A);
11129 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11130 ins_pipe(pipe_slow); // XXX
11131 %}
11133 instruct convI2D_reg_mem(regD dst, memory src)
11134 %{
11135 match(Set dst (ConvI2D (LoadI src)));
11137 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11138 opcode(0xF2, 0x0F, 0x2A);
11139 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11140 ins_pipe(pipe_slow); // XXX
11141 %}
11143 instruct convXI2F_reg(regF dst, rRegI src)
11144 %{
11145 predicate(UseXmmI2F);
11146 match(Set dst (ConvI2F src));
11148 format %{ "movdl $dst, $src\n\t"
11149 "cvtdq2psl $dst, $dst\t# i2f" %}
11150 ins_encode %{
11151 __ movdl($dst$$XMMRegister, $src$$Register);
11152 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11153 %}
11154 ins_pipe(pipe_slow); // XXX
11155 %}
11157 instruct convXI2D_reg(regD dst, rRegI src)
11158 %{
11159 predicate(UseXmmI2D);
11160 match(Set dst (ConvI2D src));
11162 format %{ "movdl $dst, $src\n\t"
11163 "cvtdq2pdl $dst, $dst\t# i2d" %}
11164 ins_encode %{
11165 __ movdl($dst$$XMMRegister, $src$$Register);
11166 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11167 %}
11168 ins_pipe(pipe_slow); // XXX
11169 %}
11171 instruct convL2F_reg_reg(regF dst, rRegL src)
11172 %{
11173 match(Set dst (ConvL2F src));
11175 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11176 opcode(0xF3, 0x0F, 0x2A);
11177 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11178 ins_pipe(pipe_slow); // XXX
11179 %}
11181 instruct convL2F_reg_mem(regF dst, memory src)
11182 %{
11183 match(Set dst (ConvL2F (LoadL src)));
11185 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11186 opcode(0xF3, 0x0F, 0x2A);
11187 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11188 ins_pipe(pipe_slow); // XXX
11189 %}
11191 instruct convL2D_reg_reg(regD dst, rRegL src)
11192 %{
11193 match(Set dst (ConvL2D src));
11195 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11196 opcode(0xF2, 0x0F, 0x2A);
11197 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11198 ins_pipe(pipe_slow); // XXX
11199 %}
11201 instruct convL2D_reg_mem(regD dst, memory src)
11202 %{
11203 match(Set dst (ConvL2D (LoadL src)));
11205 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11206 opcode(0xF2, 0x0F, 0x2A);
11207 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11208 ins_pipe(pipe_slow); // XXX
11209 %}
11211 instruct convI2L_reg_reg(rRegL dst, rRegI src)
11212 %{
11213 match(Set dst (ConvI2L src));
11215 ins_cost(125);
11216 format %{ "movslq $dst, $src\t# i2l" %}
11217 opcode(0x63); // needs REX.W
11218 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11219 ins_pipe(ialu_reg_reg);
11220 %}
11222 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
11223 // %{
11224 // match(Set dst (ConvI2L src));
11225 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
11226 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
11227 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
11228 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
11229 // ((const TypeNode*) n)->type()->is_long()->_lo ==
11230 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
11232 // format %{ "movl $dst, $src\t# unsigned i2l" %}
11233 // ins_encode(enc_copy(dst, src));
11234 // // opcode(0x63); // needs REX.W
11235 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11236 // ins_pipe(ialu_reg_reg);
11237 // %}
11239 // Zero-extend convert int to long
11240 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
11241 %{
11242 match(Set dst (AndL (ConvI2L src) mask));
11244 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11245 ins_encode(enc_copy(dst, src));
11246 ins_pipe(ialu_reg_reg);
11247 %}
11249 // Zero-extend convert int to long
11250 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
11251 %{
11252 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
11254 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11255 opcode(0x8B);
11256 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11257 ins_pipe(ialu_reg_mem);
11258 %}
11260 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
11261 %{
11262 match(Set dst (AndL src mask));
11264 format %{ "movl $dst, $src\t# zero-extend long" %}
11265 ins_encode(enc_copy_always(dst, src));
11266 ins_pipe(ialu_reg_reg);
11267 %}
11269 instruct convL2I_reg_reg(rRegI dst, rRegL src)
11270 %{
11271 match(Set dst (ConvL2I src));
11273 format %{ "movl $dst, $src\t# l2i" %}
11274 ins_encode(enc_copy_always(dst, src));
11275 ins_pipe(ialu_reg_reg);
11276 %}
11279 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11280 match(Set dst (MoveF2I src));
11281 effect(DEF dst, USE src);
11283 ins_cost(125);
11284 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
11285 opcode(0x8B);
11286 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11287 ins_pipe(ialu_reg_mem);
11288 %}
11290 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
11291 match(Set dst (MoveI2F src));
11292 effect(DEF dst, USE src);
11294 ins_cost(125);
11295 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
11296 opcode(0xF3, 0x0F, 0x10);
11297 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11298 ins_pipe(pipe_slow);
11299 %}
11301 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
11302 match(Set dst (MoveD2L src));
11303 effect(DEF dst, USE src);
11305 ins_cost(125);
11306 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
11307 opcode(0x8B);
11308 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
11309 ins_pipe(ialu_reg_mem);
11310 %}
11312 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
11313 predicate(!UseXmmLoadAndClearUpper);
11314 match(Set dst (MoveL2D src));
11315 effect(DEF dst, USE src);
11317 ins_cost(125);
11318 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
11319 opcode(0x66, 0x0F, 0x12);
11320 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11321 ins_pipe(pipe_slow);
11322 %}
11324 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
11325 predicate(UseXmmLoadAndClearUpper);
11326 match(Set dst (MoveL2D src));
11327 effect(DEF dst, USE src);
11329 ins_cost(125);
11330 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
11331 opcode(0xF2, 0x0F, 0x10);
11332 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11333 ins_pipe(pipe_slow);
11334 %}
11337 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
11338 match(Set dst (MoveF2I src));
11339 effect(DEF dst, USE src);
11341 ins_cost(95); // XXX
11342 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
11343 opcode(0xF3, 0x0F, 0x11);
11344 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11345 ins_pipe(pipe_slow);
11346 %}
11348 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11349 match(Set dst (MoveI2F src));
11350 effect(DEF dst, USE src);
11352 ins_cost(100);
11353 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
11354 opcode(0x89);
11355 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
11356 ins_pipe( ialu_mem_reg );
11357 %}
11359 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
11360 match(Set dst (MoveD2L src));
11361 effect(DEF dst, USE src);
11363 ins_cost(95); // XXX
11364 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
11365 opcode(0xF2, 0x0F, 0x11);
11366 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11367 ins_pipe(pipe_slow);
11368 %}
11370 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
11371 match(Set dst (MoveL2D src));
11372 effect(DEF dst, USE src);
11374 ins_cost(100);
11375 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
11376 opcode(0x89);
11377 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
11378 ins_pipe(ialu_mem_reg);
11379 %}
11381 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
11382 match(Set dst (MoveF2I src));
11383 effect(DEF dst, USE src);
11384 ins_cost(85);
11385 format %{ "movd $dst,$src\t# MoveF2I" %}
11386 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
11387 ins_pipe( pipe_slow );
11388 %}
11390 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
11391 match(Set dst (MoveD2L src));
11392 effect(DEF dst, USE src);
11393 ins_cost(85);
11394 format %{ "movd $dst,$src\t# MoveD2L" %}
11395 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
11396 ins_pipe( pipe_slow );
11397 %}
11399 // The next instructions have long latency and use Int unit. Set high cost.
11400 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
11401 match(Set dst (MoveI2F src));
11402 effect(DEF dst, USE src);
11403 ins_cost(300);
11404 format %{ "movd $dst,$src\t# MoveI2F" %}
11405 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
11406 ins_pipe( pipe_slow );
11407 %}
11409 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
11410 match(Set dst (MoveL2D src));
11411 effect(DEF dst, USE src);
11412 ins_cost(300);
11413 format %{ "movd $dst,$src\t# MoveL2D" %}
11414 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
11415 ins_pipe( pipe_slow );
11416 %}
11418 // Replicate scalar to packed byte (1 byte) values in xmm
11419 instruct Repl8B_reg(regD dst, regD src) %{
11420 match(Set dst (Replicate8B src));
11421 format %{ "MOVDQA $dst,$src\n\t"
11422 "PUNPCKLBW $dst,$dst\n\t"
11423 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11424 ins_encode( pshufd_8x8(dst, src));
11425 ins_pipe( pipe_slow );
11426 %}
11428 // Replicate scalar to packed byte (1 byte) values in xmm
11429 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11430 match(Set dst (Replicate8B src));
11431 format %{ "MOVD $dst,$src\n\t"
11432 "PUNPCKLBW $dst,$dst\n\t"
11433 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11434 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11435 ins_pipe( pipe_slow );
11436 %}
11438 // Replicate scalar zero to packed byte (1 byte) values in xmm
11439 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11440 match(Set dst (Replicate8B zero));
11441 format %{ "PXOR $dst,$dst\t! replicate8B" %}
11442 ins_encode( pxor(dst, dst));
11443 ins_pipe( fpu_reg_reg );
11444 %}
11446 // Replicate scalar to packed shore (2 byte) values in xmm
11447 instruct Repl4S_reg(regD dst, regD src) %{
11448 match(Set dst (Replicate4S src));
11449 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11450 ins_encode( pshufd_4x16(dst, src));
11451 ins_pipe( fpu_reg_reg );
11452 %}
11454 // Replicate scalar to packed shore (2 byte) values in xmm
11455 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11456 match(Set dst (Replicate4S src));
11457 format %{ "MOVD $dst,$src\n\t"
11458 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11459 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11460 ins_pipe( fpu_reg_reg );
11461 %}
11463 // Replicate scalar zero to packed short (2 byte) values in xmm
11464 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11465 match(Set dst (Replicate4S zero));
11466 format %{ "PXOR $dst,$dst\t! replicate4S" %}
11467 ins_encode( pxor(dst, dst));
11468 ins_pipe( fpu_reg_reg );
11469 %}
11471 // Replicate scalar to packed char (2 byte) values in xmm
11472 instruct Repl4C_reg(regD dst, regD src) %{
11473 match(Set dst (Replicate4C src));
11474 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11475 ins_encode( pshufd_4x16(dst, src));
11476 ins_pipe( fpu_reg_reg );
11477 %}
11479 // Replicate scalar to packed char (2 byte) values in xmm
11480 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11481 match(Set dst (Replicate4C src));
11482 format %{ "MOVD $dst,$src\n\t"
11483 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11484 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11485 ins_pipe( fpu_reg_reg );
11486 %}
11488 // Replicate scalar zero to packed char (2 byte) values in xmm
11489 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11490 match(Set dst (Replicate4C zero));
11491 format %{ "PXOR $dst,$dst\t! replicate4C" %}
11492 ins_encode( pxor(dst, dst));
11493 ins_pipe( fpu_reg_reg );
11494 %}
11496 // Replicate scalar to packed integer (4 byte) values in xmm
11497 instruct Repl2I_reg(regD dst, regD src) %{
11498 match(Set dst (Replicate2I src));
11499 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11500 ins_encode( pshufd(dst, src, 0x00));
11501 ins_pipe( fpu_reg_reg );
11502 %}
11504 // Replicate scalar to packed integer (4 byte) values in xmm
11505 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11506 match(Set dst (Replicate2I src));
11507 format %{ "MOVD $dst,$src\n\t"
11508 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11509 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11510 ins_pipe( fpu_reg_reg );
11511 %}
11513 // Replicate scalar zero to packed integer (2 byte) values in xmm
11514 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11515 match(Set dst (Replicate2I zero));
11516 format %{ "PXOR $dst,$dst\t! replicate2I" %}
11517 ins_encode( pxor(dst, dst));
11518 ins_pipe( fpu_reg_reg );
11519 %}
11521 // Replicate scalar to packed single precision floating point values in xmm
11522 instruct Repl2F_reg(regD dst, regD src) %{
11523 match(Set dst (Replicate2F src));
11524 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11525 ins_encode( pshufd(dst, src, 0xe0));
11526 ins_pipe( fpu_reg_reg );
11527 %}
11529 // Replicate scalar to packed single precision floating point values in xmm
11530 instruct Repl2F_regF(regD dst, regF src) %{
11531 match(Set dst (Replicate2F src));
11532 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11533 ins_encode( pshufd(dst, src, 0xe0));
11534 ins_pipe( fpu_reg_reg );
11535 %}
11537 // Replicate scalar to packed single precision floating point values in xmm
11538 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11539 match(Set dst (Replicate2F zero));
11540 format %{ "PXOR $dst,$dst\t! replicate2F" %}
11541 ins_encode( pxor(dst, dst));
11542 ins_pipe( fpu_reg_reg );
11543 %}
11546 // =======================================================================
11547 // fast clearing of an array
11548 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11549 rFlagsReg cr)
11550 %{
11551 match(Set dummy (ClearArray cnt base));
11552 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11554 format %{ "xorl rax, rax\t# ClearArray:\n\t"
11555 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11556 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11557 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11558 ins_pipe(pipe_slow);
11559 %}
11561 instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1,
11562 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr)
11563 %{
11564 match(Set result (StrComp str1 str2));
11565 effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
11566 //ins_cost(300);
11568 format %{ "String Compare $str1, $str2 -> $result // XXX KILL RAX, RBX" %}
11569 ins_encode( enc_String_Compare() );
11570 ins_pipe( pipe_slow );
11571 %}
11573 // fast array equals
11574 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1,
11575 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr) %{
11576 match(Set result (AryEq ary1 ary2));
11577 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr);
11578 //ins_cost(300);
11580 format %{ "Array Equals $ary1,$ary2 -> $result // KILL RAX, RBX" %}
11581 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) );
11582 ins_pipe( pipe_slow );
11583 %}
11585 //----------Control Flow Instructions------------------------------------------
11586 // Signed compare Instructions
11588 // XXX more variants!!
11589 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11590 %{
11591 match(Set cr (CmpI op1 op2));
11592 effect(DEF cr, USE op1, USE op2);
11594 format %{ "cmpl $op1, $op2" %}
11595 opcode(0x3B); /* Opcode 3B /r */
11596 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11597 ins_pipe(ialu_cr_reg_reg);
11598 %}
11600 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11601 %{
11602 match(Set cr (CmpI op1 op2));
11604 format %{ "cmpl $op1, $op2" %}
11605 opcode(0x81, 0x07); /* Opcode 81 /7 */
11606 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11607 ins_pipe(ialu_cr_reg_imm);
11608 %}
11610 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11611 %{
11612 match(Set cr (CmpI op1 (LoadI op2)));
11614 ins_cost(500); // XXX
11615 format %{ "cmpl $op1, $op2" %}
11616 opcode(0x3B); /* Opcode 3B /r */
11617 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11618 ins_pipe(ialu_cr_reg_mem);
11619 %}
11621 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11622 %{
11623 match(Set cr (CmpI src zero));
11625 format %{ "testl $src, $src" %}
11626 opcode(0x85);
11627 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11628 ins_pipe(ialu_cr_reg_imm);
11629 %}
11631 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11632 %{
11633 match(Set cr (CmpI (AndI src con) zero));
11635 format %{ "testl $src, $con" %}
11636 opcode(0xF7, 0x00);
11637 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11638 ins_pipe(ialu_cr_reg_imm);
11639 %}
11641 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11642 %{
11643 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11645 format %{ "testl $src, $mem" %}
11646 opcode(0x85);
11647 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11648 ins_pipe(ialu_cr_reg_mem);
11649 %}
11651 // Unsigned compare Instructions; really, same as signed except they
11652 // produce an rFlagsRegU instead of rFlagsReg.
11653 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11654 %{
11655 match(Set cr (CmpU op1 op2));
11657 format %{ "cmpl $op1, $op2\t# unsigned" %}
11658 opcode(0x3B); /* Opcode 3B /r */
11659 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11660 ins_pipe(ialu_cr_reg_reg);
11661 %}
11663 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11664 %{
11665 match(Set cr (CmpU op1 op2));
11667 format %{ "cmpl $op1, $op2\t# unsigned" %}
11668 opcode(0x81,0x07); /* Opcode 81 /7 */
11669 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11670 ins_pipe(ialu_cr_reg_imm);
11671 %}
11673 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11674 %{
11675 match(Set cr (CmpU op1 (LoadI op2)));
11677 ins_cost(500); // XXX
11678 format %{ "cmpl $op1, $op2\t# unsigned" %}
11679 opcode(0x3B); /* Opcode 3B /r */
11680 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11681 ins_pipe(ialu_cr_reg_mem);
11682 %}
11684 // // // Cisc-spilled version of cmpU_rReg
11685 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11686 // //%{
11687 // // match(Set cr (CmpU (LoadI op1) op2));
11688 // //
11689 // // format %{ "CMPu $op1,$op2" %}
11690 // // ins_cost(500);
11691 // // opcode(0x39); /* Opcode 39 /r */
11692 // // ins_encode( OpcP, reg_mem( op1, op2) );
11693 // //%}
11695 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11696 %{
11697 match(Set cr (CmpU src zero));
11699 format %{ "testl $src, $src\t# unsigned" %}
11700 opcode(0x85);
11701 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11702 ins_pipe(ialu_cr_reg_imm);
11703 %}
11705 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11706 %{
11707 match(Set cr (CmpP op1 op2));
11709 format %{ "cmpq $op1, $op2\t# ptr" %}
11710 opcode(0x3B); /* Opcode 3B /r */
11711 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11712 ins_pipe(ialu_cr_reg_reg);
11713 %}
11715 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11716 %{
11717 match(Set cr (CmpP op1 (LoadP op2)));
11719 ins_cost(500); // XXX
11720 format %{ "cmpq $op1, $op2\t# ptr" %}
11721 opcode(0x3B); /* Opcode 3B /r */
11722 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11723 ins_pipe(ialu_cr_reg_mem);
11724 %}
11726 // // // Cisc-spilled version of cmpP_rReg
11727 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11728 // //%{
11729 // // match(Set cr (CmpP (LoadP op1) op2));
11730 // //
11731 // // format %{ "CMPu $op1,$op2" %}
11732 // // ins_cost(500);
11733 // // opcode(0x39); /* Opcode 39 /r */
11734 // // ins_encode( OpcP, reg_mem( op1, op2) );
11735 // //%}
11737 // XXX this is generalized by compP_rReg_mem???
11738 // Compare raw pointer (used in out-of-heap check).
11739 // Only works because non-oop pointers must be raw pointers
11740 // and raw pointers have no anti-dependencies.
11741 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11742 %{
11743 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11744 match(Set cr (CmpP op1 (LoadP op2)));
11746 format %{ "cmpq $op1, $op2\t# raw ptr" %}
11747 opcode(0x3B); /* Opcode 3B /r */
11748 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11749 ins_pipe(ialu_cr_reg_mem);
11750 %}
11752 // This will generate a signed flags result. This should be OK since
11753 // any compare to a zero should be eq/neq.
11754 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11755 %{
11756 match(Set cr (CmpP src zero));
11758 format %{ "testq $src, $src\t# ptr" %}
11759 opcode(0x85);
11760 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11761 ins_pipe(ialu_cr_reg_imm);
11762 %}
11764 // This will generate a signed flags result. This should be OK since
11765 // any compare to a zero should be eq/neq.
11766 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero)
11767 %{
11768 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
11769 match(Set cr (CmpP (LoadP op) zero));
11771 ins_cost(500); // XXX
11772 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11773 opcode(0xF7); /* Opcode F7 /0 */
11774 ins_encode(REX_mem_wide(op),
11775 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11776 ins_pipe(ialu_cr_reg_imm);
11777 %}
11779 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero)
11780 %{
11781 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
11782 match(Set cr (CmpP (LoadP mem) zero));
11784 format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %}
11785 ins_encode %{
11786 __ cmpq(r12, $mem$$Address);
11787 %}
11788 ins_pipe(ialu_cr_reg_mem);
11789 %}
11791 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11792 %{
11793 match(Set cr (CmpN op1 op2));
11795 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11796 ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %}
11797 ins_pipe(ialu_cr_reg_reg);
11798 %}
11800 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11801 %{
11802 match(Set cr (CmpN src (LoadN mem)));
11804 format %{ "cmpl $src, $mem\t# compressed ptr" %}
11805 ins_encode %{
11806 __ cmpl($src$$Register, $mem$$Address);
11807 %}
11808 ins_pipe(ialu_cr_reg_mem);
11809 %}
11811 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{
11812 match(Set cr (CmpN op1 op2));
11814 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11815 ins_encode %{
11816 __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant);
11817 %}
11818 ins_pipe(ialu_cr_reg_imm);
11819 %}
11821 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src)
11822 %{
11823 match(Set cr (CmpN src (LoadN mem)));
11825 format %{ "cmpl $mem, $src\t# compressed ptr" %}
11826 ins_encode %{
11827 __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant);
11828 %}
11829 ins_pipe(ialu_cr_reg_mem);
11830 %}
11832 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11833 match(Set cr (CmpN src zero));
11835 format %{ "testl $src, $src\t# compressed ptr" %}
11836 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11837 ins_pipe(ialu_cr_reg_imm);
11838 %}
11840 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero)
11841 %{
11842 predicate(Universe::narrow_oop_base() != NULL);
11843 match(Set cr (CmpN (LoadN mem) zero));
11845 ins_cost(500); // XXX
11846 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %}
11847 ins_encode %{
11848 __ cmpl($mem$$Address, (int)0xFFFFFFFF);
11849 %}
11850 ins_pipe(ialu_cr_reg_mem);
11851 %}
11853 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero)
11854 %{
11855 predicate(Universe::narrow_oop_base() == NULL);
11856 match(Set cr (CmpN (LoadN mem) zero));
11858 format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %}
11859 ins_encode %{
11860 __ cmpl(r12, $mem$$Address);
11861 %}
11862 ins_pipe(ialu_cr_reg_mem);
11863 %}
11865 // Yanked all unsigned pointer compare operations.
11866 // Pointer compares are done with CmpP which is already unsigned.
11868 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11869 %{
11870 match(Set cr (CmpL op1 op2));
11872 format %{ "cmpq $op1, $op2" %}
11873 opcode(0x3B); /* Opcode 3B /r */
11874 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11875 ins_pipe(ialu_cr_reg_reg);
11876 %}
11878 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11879 %{
11880 match(Set cr (CmpL op1 op2));
11882 format %{ "cmpq $op1, $op2" %}
11883 opcode(0x81, 0x07); /* Opcode 81 /7 */
11884 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11885 ins_pipe(ialu_cr_reg_imm);
11886 %}
11888 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11889 %{
11890 match(Set cr (CmpL op1 (LoadL op2)));
11892 format %{ "cmpq $op1, $op2" %}
11893 opcode(0x3B); /* Opcode 3B /r */
11894 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11895 ins_pipe(ialu_cr_reg_mem);
11896 %}
11898 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11899 %{
11900 match(Set cr (CmpL src zero));
11902 format %{ "testq $src, $src" %}
11903 opcode(0x85);
11904 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11905 ins_pipe(ialu_cr_reg_imm);
11906 %}
11908 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11909 %{
11910 match(Set cr (CmpL (AndL src con) zero));
11912 format %{ "testq $src, $con\t# long" %}
11913 opcode(0xF7, 0x00);
11914 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11915 ins_pipe(ialu_cr_reg_imm);
11916 %}
11918 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
11919 %{
11920 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
11922 format %{ "testq $src, $mem" %}
11923 opcode(0x85);
11924 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
11925 ins_pipe(ialu_cr_reg_mem);
11926 %}
11928 // Manifest a CmpL result in an integer register. Very painful.
11929 // This is the test to avoid.
11930 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
11931 %{
11932 match(Set dst (CmpL3 src1 src2));
11933 effect(KILL flags);
11935 ins_cost(275); // XXX
11936 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
11937 "movl $dst, -1\n\t"
11938 "jl,s done\n\t"
11939 "setne $dst\n\t"
11940 "movzbl $dst, $dst\n\t"
11941 "done:" %}
11942 ins_encode(cmpl3_flag(src1, src2, dst));
11943 ins_pipe(pipe_slow);
11944 %}
11946 //----------Max and Min--------------------------------------------------------
11947 // Min Instructions
11949 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
11950 %{
11951 effect(USE_DEF dst, USE src, USE cr);
11953 format %{ "cmovlgt $dst, $src\t# min" %}
11954 opcode(0x0F, 0x4F);
11955 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11956 ins_pipe(pipe_cmov_reg);
11957 %}
11960 instruct minI_rReg(rRegI dst, rRegI src)
11961 %{
11962 match(Set dst (MinI dst src));
11964 ins_cost(200);
11965 expand %{
11966 rFlagsReg cr;
11967 compI_rReg(cr, dst, src);
11968 cmovI_reg_g(dst, src, cr);
11969 %}
11970 %}
11972 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
11973 %{
11974 effect(USE_DEF dst, USE src, USE cr);
11976 format %{ "cmovllt $dst, $src\t# max" %}
11977 opcode(0x0F, 0x4C);
11978 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11979 ins_pipe(pipe_cmov_reg);
11980 %}
11983 instruct maxI_rReg(rRegI dst, rRegI src)
11984 %{
11985 match(Set dst (MaxI dst src));
11987 ins_cost(200);
11988 expand %{
11989 rFlagsReg cr;
11990 compI_rReg(cr, dst, src);
11991 cmovI_reg_l(dst, src, cr);
11992 %}
11993 %}
11995 // ============================================================================
11996 // Branch Instructions
11998 // Jump Direct - Label defines a relative address from JMP+1
11999 instruct jmpDir(label labl)
12000 %{
12001 match(Goto);
12002 effect(USE labl);
12004 ins_cost(300);
12005 format %{ "jmp $labl" %}
12006 size(5);
12007 opcode(0xE9);
12008 ins_encode(OpcP, Lbl(labl));
12009 ins_pipe(pipe_jmp);
12010 ins_pc_relative(1);
12011 %}
12013 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12014 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
12015 %{
12016 match(If cop cr);
12017 effect(USE labl);
12019 ins_cost(300);
12020 format %{ "j$cop $labl" %}
12021 size(6);
12022 opcode(0x0F, 0x80);
12023 ins_encode(Jcc(cop, labl));
12024 ins_pipe(pipe_jcc);
12025 ins_pc_relative(1);
12026 %}
12028 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12029 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
12030 %{
12031 match(CountedLoopEnd cop cr);
12032 effect(USE labl);
12034 ins_cost(300);
12035 format %{ "j$cop $labl\t# loop end" %}
12036 size(6);
12037 opcode(0x0F, 0x80);
12038 ins_encode(Jcc(cop, labl));
12039 ins_pipe(pipe_jcc);
12040 ins_pc_relative(1);
12041 %}
12043 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12044 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12045 match(CountedLoopEnd cop cmp);
12046 effect(USE labl);
12048 ins_cost(300);
12049 format %{ "j$cop,u $labl\t# loop end" %}
12050 size(6);
12051 opcode(0x0F, 0x80);
12052 ins_encode(Jcc(cop, labl));
12053 ins_pipe(pipe_jcc);
12054 ins_pc_relative(1);
12055 %}
12057 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12058 match(CountedLoopEnd cop cmp);
12059 effect(USE labl);
12061 ins_cost(200);
12062 format %{ "j$cop,u $labl\t# loop end" %}
12063 size(6);
12064 opcode(0x0F, 0x80);
12065 ins_encode(Jcc(cop, labl));
12066 ins_pipe(pipe_jcc);
12067 ins_pc_relative(1);
12068 %}
12070 // Jump Direct Conditional - using unsigned comparison
12071 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12072 match(If cop cmp);
12073 effect(USE labl);
12075 ins_cost(300);
12076 format %{ "j$cop,u $labl" %}
12077 size(6);
12078 opcode(0x0F, 0x80);
12079 ins_encode(Jcc(cop, labl));
12080 ins_pipe(pipe_jcc);
12081 ins_pc_relative(1);
12082 %}
12084 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12085 match(If cop cmp);
12086 effect(USE labl);
12088 ins_cost(200);
12089 format %{ "j$cop,u $labl" %}
12090 size(6);
12091 opcode(0x0F, 0x80);
12092 ins_encode(Jcc(cop, labl));
12093 ins_pipe(pipe_jcc);
12094 ins_pc_relative(1);
12095 %}
12097 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12098 match(If cop cmp);
12099 effect(USE labl);
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 ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
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 ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
12131 emit_d32(cbuf, disp);
12132 %}
12133 ins_pipe(pipe_jcc);
12134 ins_pc_relative(1);
12135 %}
12137 // ============================================================================
12138 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
12139 // superklass array for an instance of the superklass. Set a hidden
12140 // internal cache on a hit (cache is checked with exposed code in
12141 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
12142 // encoding ALSO sets flags.
12144 instruct partialSubtypeCheck(rdi_RegP result,
12145 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12146 rFlagsReg cr)
12147 %{
12148 match(Set result (PartialSubtypeCheck sub super));
12149 effect(KILL rcx, KILL cr);
12151 ins_cost(1100); // slightly larger than the next version
12152 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12153 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12154 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12155 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
12156 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
12157 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12158 "xorq $result, $result\t\t Hit: rdi zero\n\t"
12159 "miss:\t" %}
12161 opcode(0x1); // Force a XOR of RDI
12162 ins_encode(enc_PartialSubtypeCheck());
12163 ins_pipe(pipe_slow);
12164 %}
12166 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
12167 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12168 immP0 zero,
12169 rdi_RegP result)
12170 %{
12171 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12172 effect(KILL rcx, KILL result);
12174 ins_cost(1000);
12175 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12176 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12177 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12178 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
12179 "jne,s miss\t\t# Missed: flags nz\n\t"
12180 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12181 "miss:\t" %}
12183 opcode(0x0); // No need to XOR RDI
12184 ins_encode(enc_PartialSubtypeCheck());
12185 ins_pipe(pipe_slow);
12186 %}
12188 // ============================================================================
12189 // Branch Instructions -- short offset versions
12190 //
12191 // These instructions are used to replace jumps of a long offset (the default
12192 // match) with jumps of a shorter offset. These instructions are all tagged
12193 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12194 // match rules in general matching. Instead, the ADLC generates a conversion
12195 // method in the MachNode which can be used to do in-place replacement of the
12196 // long variant with the shorter variant. The compiler will determine if a
12197 // branch can be taken by the is_short_branch_offset() predicate in the machine
12198 // specific code section of the file.
12200 // Jump Direct - Label defines a relative address from JMP+1
12201 instruct jmpDir_short(label labl) %{
12202 match(Goto);
12203 effect(USE labl);
12205 ins_cost(300);
12206 format %{ "jmp,s $labl" %}
12207 size(2);
12208 opcode(0xEB);
12209 ins_encode(OpcP, LblShort(labl));
12210 ins_pipe(pipe_jmp);
12211 ins_pc_relative(1);
12212 ins_short_branch(1);
12213 %}
12215 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12216 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
12217 match(If cop cr);
12218 effect(USE labl);
12220 ins_cost(300);
12221 format %{ "j$cop,s $labl" %}
12222 size(2);
12223 opcode(0x70);
12224 ins_encode(JccShort(cop, labl));
12225 ins_pipe(pipe_jcc);
12226 ins_pc_relative(1);
12227 ins_short_branch(1);
12228 %}
12230 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12231 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
12232 match(CountedLoopEnd cop cr);
12233 effect(USE labl);
12235 ins_cost(300);
12236 format %{ "j$cop,s $labl\t# loop end" %}
12237 size(2);
12238 opcode(0x70);
12239 ins_encode(JccShort(cop, labl));
12240 ins_pipe(pipe_jcc);
12241 ins_pc_relative(1);
12242 ins_short_branch(1);
12243 %}
12245 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12246 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12247 match(CountedLoopEnd cop cmp);
12248 effect(USE labl);
12250 ins_cost(300);
12251 format %{ "j$cop,us $labl\t# loop end" %}
12252 size(2);
12253 opcode(0x70);
12254 ins_encode(JccShort(cop, labl));
12255 ins_pipe(pipe_jcc);
12256 ins_pc_relative(1);
12257 ins_short_branch(1);
12258 %}
12260 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12261 match(CountedLoopEnd cop cmp);
12262 effect(USE labl);
12264 ins_cost(300);
12265 format %{ "j$cop,us $labl\t# loop end" %}
12266 size(2);
12267 opcode(0x70);
12268 ins_encode(JccShort(cop, labl));
12269 ins_pipe(pipe_jcc);
12270 ins_pc_relative(1);
12271 ins_short_branch(1);
12272 %}
12274 // Jump Direct Conditional - using unsigned comparison
12275 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12276 match(If cop cmp);
12277 effect(USE labl);
12279 ins_cost(300);
12280 format %{ "j$cop,us $labl" %}
12281 size(2);
12282 opcode(0x70);
12283 ins_encode(JccShort(cop, labl));
12284 ins_pipe(pipe_jcc);
12285 ins_pc_relative(1);
12286 ins_short_branch(1);
12287 %}
12289 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12290 match(If cop cmp);
12291 effect(USE labl);
12293 ins_cost(300);
12294 format %{ "j$cop,us $labl" %}
12295 size(2);
12296 opcode(0x70);
12297 ins_encode(JccShort(cop, labl));
12298 ins_pipe(pipe_jcc);
12299 ins_pc_relative(1);
12300 ins_short_branch(1);
12301 %}
12303 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12304 match(If cop cmp);
12305 effect(USE labl);
12307 ins_cost(300);
12308 format %{ $$template
12309 if ($cop$$cmpcode == Assembler::notEqual) {
12310 $$emit$$"jp,u,s $labl\n\t"
12311 $$emit$$"j$cop,u,s $labl"
12312 } else {
12313 $$emit$$"jp,u,s done\n\t"
12314 $$emit$$"j$cop,u,s $labl\n\t"
12315 $$emit$$"done:"
12316 }
12317 %}
12318 size(4);
12319 opcode(0x70);
12320 ins_encode %{
12321 Label* l = $labl$$label;
12322 emit_cc(cbuf, $primary, Assembler::parity);
12323 int parity_disp = -1;
12324 if ($cop$$cmpcode == Assembler::notEqual) {
12325 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12326 } else if ($cop$$cmpcode == Assembler::equal) {
12327 parity_disp = 2;
12328 } else {
12329 ShouldNotReachHere();
12330 }
12331 emit_d8(cbuf, parity_disp);
12332 emit_cc(cbuf, $primary, $cop$$cmpcode);
12333 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12334 emit_d8(cbuf, disp);
12335 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
12336 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
12337 %}
12338 ins_pipe(pipe_jcc);
12339 ins_pc_relative(1);
12340 ins_short_branch(1);
12341 %}
12343 // ============================================================================
12344 // inlined locking and unlocking
12346 instruct cmpFastLock(rFlagsReg cr,
12347 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
12348 %{
12349 match(Set cr (FastLock object box));
12350 effect(TEMP tmp, TEMP scr);
12352 ins_cost(300);
12353 format %{ "fastlock $object,$box,$tmp,$scr" %}
12354 ins_encode(Fast_Lock(object, box, tmp, scr));
12355 ins_pipe(pipe_slow);
12356 ins_pc_relative(1);
12357 %}
12359 instruct cmpFastUnlock(rFlagsReg cr,
12360 rRegP object, rax_RegP box, rRegP tmp)
12361 %{
12362 match(Set cr (FastUnlock object box));
12363 effect(TEMP tmp);
12365 ins_cost(300);
12366 format %{ "fastunlock $object, $box, $tmp" %}
12367 ins_encode(Fast_Unlock(object, box, tmp));
12368 ins_pipe(pipe_slow);
12369 ins_pc_relative(1);
12370 %}
12373 // ============================================================================
12374 // Safepoint Instructions
12375 instruct safePoint_poll(rFlagsReg cr)
12376 %{
12377 match(SafePoint);
12378 effect(KILL cr);
12380 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
12381 "# Safepoint: poll for GC" %}
12382 size(6); // Opcode + ModRM + Disp32 == 6 bytes
12383 ins_cost(125);
12384 ins_encode(enc_safepoint_poll);
12385 ins_pipe(ialu_reg_mem);
12386 %}
12388 // ============================================================================
12389 // Procedure Call/Return Instructions
12390 // Call Java Static Instruction
12391 // Note: If this code changes, the corresponding ret_addr_offset() and
12392 // compute_padding() functions will have to be adjusted.
12393 instruct CallStaticJavaDirect(method meth)
12394 %{
12395 match(CallStaticJava);
12396 effect(USE meth);
12398 ins_cost(300);
12399 format %{ "call,static " %}
12400 opcode(0xE8); /* E8 cd */
12401 ins_encode(Java_Static_Call(meth), call_epilog);
12402 ins_pipe(pipe_slow);
12403 ins_pc_relative(1);
12404 ins_alignment(4);
12405 %}
12407 // Call Java Dynamic Instruction
12408 // Note: If this code changes, the corresponding ret_addr_offset() and
12409 // compute_padding() functions will have to be adjusted.
12410 instruct CallDynamicJavaDirect(method meth)
12411 %{
12412 match(CallDynamicJava);
12413 effect(USE meth);
12415 ins_cost(300);
12416 format %{ "movq rax, #Universe::non_oop_word()\n\t"
12417 "call,dynamic " %}
12418 opcode(0xE8); /* E8 cd */
12419 ins_encode(Java_Dynamic_Call(meth), call_epilog);
12420 ins_pipe(pipe_slow);
12421 ins_pc_relative(1);
12422 ins_alignment(4);
12423 %}
12425 // Call Runtime Instruction
12426 instruct CallRuntimeDirect(method meth)
12427 %{
12428 match(CallRuntime);
12429 effect(USE meth);
12431 ins_cost(300);
12432 format %{ "call,runtime " %}
12433 opcode(0xE8); /* E8 cd */
12434 ins_encode(Java_To_Runtime(meth));
12435 ins_pipe(pipe_slow);
12436 ins_pc_relative(1);
12437 %}
12439 // Call runtime without safepoint
12440 instruct CallLeafDirect(method meth)
12441 %{
12442 match(CallLeaf);
12443 effect(USE meth);
12445 ins_cost(300);
12446 format %{ "call_leaf,runtime " %}
12447 opcode(0xE8); /* E8 cd */
12448 ins_encode(Java_To_Runtime(meth));
12449 ins_pipe(pipe_slow);
12450 ins_pc_relative(1);
12451 %}
12453 // Call runtime without safepoint
12454 instruct CallLeafNoFPDirect(method meth)
12455 %{
12456 match(CallLeafNoFP);
12457 effect(USE meth);
12459 ins_cost(300);
12460 format %{ "call_leaf_nofp,runtime " %}
12461 opcode(0xE8); /* E8 cd */
12462 ins_encode(Java_To_Runtime(meth));
12463 ins_pipe(pipe_slow);
12464 ins_pc_relative(1);
12465 %}
12467 // Return Instruction
12468 // Remove the return address & jump to it.
12469 // Notice: We always emit a nop after a ret to make sure there is room
12470 // for safepoint patching
12471 instruct Ret()
12472 %{
12473 match(Return);
12475 format %{ "ret" %}
12476 opcode(0xC3);
12477 ins_encode(OpcP);
12478 ins_pipe(pipe_jmp);
12479 %}
12481 // Tail Call; Jump from runtime stub to Java code.
12482 // Also known as an 'interprocedural jump'.
12483 // Target of jump will eventually return to caller.
12484 // TailJump below removes the return address.
12485 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12486 %{
12487 match(TailCall jump_target method_oop);
12489 ins_cost(300);
12490 format %{ "jmp $jump_target\t# rbx holds method oop" %}
12491 opcode(0xFF, 0x4); /* Opcode FF /4 */
12492 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12493 ins_pipe(pipe_jmp);
12494 %}
12496 // Tail Jump; remove the return address; jump to target.
12497 // TailCall above leaves the return address around.
12498 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12499 %{
12500 match(TailJump jump_target ex_oop);
12502 ins_cost(300);
12503 format %{ "popq rdx\t# pop return address\n\t"
12504 "jmp $jump_target" %}
12505 opcode(0xFF, 0x4); /* Opcode FF /4 */
12506 ins_encode(Opcode(0x5a), // popq rdx
12507 REX_reg(jump_target), OpcP, reg_opc(jump_target));
12508 ins_pipe(pipe_jmp);
12509 %}
12511 // Create exception oop: created by stack-crawling runtime code.
12512 // Created exception is now available to this handler, and is setup
12513 // just prior to jumping to this handler. No code emitted.
12514 instruct CreateException(rax_RegP ex_oop)
12515 %{
12516 match(Set ex_oop (CreateEx));
12518 size(0);
12519 // use the following format syntax
12520 format %{ "# exception oop is in rax; no code emitted" %}
12521 ins_encode();
12522 ins_pipe(empty);
12523 %}
12525 // Rethrow exception:
12526 // The exception oop will come in the first argument position.
12527 // Then JUMP (not call) to the rethrow stub code.
12528 instruct RethrowException()
12529 %{
12530 match(Rethrow);
12532 // use the following format syntax
12533 format %{ "jmp rethrow_stub" %}
12534 ins_encode(enc_rethrow);
12535 ins_pipe(pipe_jmp);
12536 %}
12539 //----------PEEPHOLE RULES-----------------------------------------------------
12540 // These must follow all instruction definitions as they use the names
12541 // defined in the instructions definitions.
12542 //
12543 // peepmatch ( root_instr_name [preceding_instruction]* );
12544 //
12545 // peepconstraint %{
12546 // (instruction_number.operand_name relational_op instruction_number.operand_name
12547 // [, ...] );
12548 // // instruction numbers are zero-based using left to right order in peepmatch
12549 //
12550 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12551 // // provide an instruction_number.operand_name for each operand that appears
12552 // // in the replacement instruction's match rule
12553 //
12554 // ---------VM FLAGS---------------------------------------------------------
12555 //
12556 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12557 //
12558 // Each peephole rule is given an identifying number starting with zero and
12559 // increasing by one in the order seen by the parser. An individual peephole
12560 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12561 // on the command-line.
12562 //
12563 // ---------CURRENT LIMITATIONS----------------------------------------------
12564 //
12565 // Only match adjacent instructions in same basic block
12566 // Only equality constraints
12567 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12568 // Only one replacement instruction
12569 //
12570 // ---------EXAMPLE----------------------------------------------------------
12571 //
12572 // // pertinent parts of existing instructions in architecture description
12573 // instruct movI(rRegI dst, rRegI src)
12574 // %{
12575 // match(Set dst (CopyI src));
12576 // %}
12577 //
12578 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12579 // %{
12580 // match(Set dst (AddI dst src));
12581 // effect(KILL cr);
12582 // %}
12583 //
12584 // // Change (inc mov) to lea
12585 // peephole %{
12586 // // increment preceeded by register-register move
12587 // peepmatch ( incI_rReg movI );
12588 // // require that the destination register of the increment
12589 // // match the destination register of the move
12590 // peepconstraint ( 0.dst == 1.dst );
12591 // // construct a replacement instruction that sets
12592 // // the destination to ( move's source register + one )
12593 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12594 // %}
12595 //
12597 // Implementation no longer uses movX instructions since
12598 // machine-independent system no longer uses CopyX nodes.
12599 //
12600 // peephole
12601 // %{
12602 // peepmatch (incI_rReg movI);
12603 // peepconstraint (0.dst == 1.dst);
12604 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12605 // %}
12607 // peephole
12608 // %{
12609 // peepmatch (decI_rReg movI);
12610 // peepconstraint (0.dst == 1.dst);
12611 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12612 // %}
12614 // peephole
12615 // %{
12616 // peepmatch (addI_rReg_imm movI);
12617 // peepconstraint (0.dst == 1.dst);
12618 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12619 // %}
12621 // peephole
12622 // %{
12623 // peepmatch (incL_rReg movL);
12624 // peepconstraint (0.dst == 1.dst);
12625 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12626 // %}
12628 // peephole
12629 // %{
12630 // peepmatch (decL_rReg movL);
12631 // peepconstraint (0.dst == 1.dst);
12632 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12633 // %}
12635 // peephole
12636 // %{
12637 // peepmatch (addL_rReg_imm movL);
12638 // peepconstraint (0.dst == 1.dst);
12639 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12640 // %}
12642 // peephole
12643 // %{
12644 // peepmatch (addP_rReg_imm movP);
12645 // peepconstraint (0.dst == 1.dst);
12646 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12647 // %}
12649 // // Change load of spilled value to only a spill
12650 // instruct storeI(memory mem, rRegI src)
12651 // %{
12652 // match(Set mem (StoreI mem src));
12653 // %}
12654 //
12655 // instruct loadI(rRegI dst, memory mem)
12656 // %{
12657 // match(Set dst (LoadI mem));
12658 // %}
12659 //
12661 peephole
12662 %{
12663 peepmatch (loadI storeI);
12664 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12665 peepreplace (storeI(1.mem 1.mem 1.src));
12666 %}
12668 peephole
12669 %{
12670 peepmatch (loadL storeL);
12671 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12672 peepreplace (storeL(1.mem 1.mem 1.src));
12673 %}
12675 //----------SMARTSPILL RULES---------------------------------------------------
12676 // These must follow all instruction definitions as they use the names
12677 // defined in the instructions definitions.