Sun, 13 Apr 2008 17:43:42 -0400
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold
1 //
2 // Copyright 2003-2007 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 R12, R12_H,
330 R13, R13_H,
331 R14, R14_H);
333 reg_class ptr_no_rbp_reg(RDX, RDX_H,
334 RAX, RAX_H,
335 RDI, RDI_H,
336 RSI, RSI_H,
337 RCX, RCX_H,
338 RBX, RBX_H,
339 R8, R8_H,
340 R9, R9_H,
341 R10, R10_H,
342 R11, R11_H,
343 R12, R12_H,
344 R13, R13_H,
345 R14, R14_H);
347 // Class for all pointer registers except RAX, RBX and RSP
348 reg_class ptr_no_rax_rbx_reg(RDX, RDX_H,
349 RBP, RBP_H,
350 RDI, RDI_H,
351 RSI, RSI_H,
352 RCX, RCX_H,
353 R8, R8_H,
354 R9, R9_H,
355 R10, R10_H,
356 R11, R11_H,
357 R12, R12_H,
358 R13, R13_H,
359 R14, R14_H);
361 // Singleton class for RAX pointer register
362 reg_class ptr_rax_reg(RAX, RAX_H);
364 // Singleton class for RBX pointer register
365 reg_class ptr_rbx_reg(RBX, RBX_H);
367 // Singleton class for RSI pointer register
368 reg_class ptr_rsi_reg(RSI, RSI_H);
370 // Singleton class for RDI pointer register
371 reg_class ptr_rdi_reg(RDI, RDI_H);
373 // Singleton class for RBP pointer register
374 reg_class ptr_rbp_reg(RBP, RBP_H);
376 // Singleton class for stack pointer
377 reg_class ptr_rsp_reg(RSP, RSP_H);
379 // Singleton class for TLS pointer
380 reg_class ptr_r15_reg(R15, R15_H);
382 // Class for all long registers (except RSP)
383 reg_class long_reg(RAX, RAX_H,
384 RDX, RDX_H,
385 RBP, RBP_H,
386 RDI, RDI_H,
387 RSI, RSI_H,
388 RCX, RCX_H,
389 RBX, RBX_H,
390 R8, R8_H,
391 R9, R9_H,
392 R10, R10_H,
393 R11, R11_H,
394 R13, R13_H,
395 R14, R14_H);
397 // Class for all long registers except RAX, RDX (and RSP)
398 reg_class long_no_rax_rdx_reg(RBP, RBP_H,
399 RDI, RDI_H,
400 RSI, RSI_H,
401 RCX, RCX_H,
402 RBX, RBX_H,
403 R8, R8_H,
404 R9, R9_H,
405 R10, R10_H,
406 R11, R11_H,
407 R13, R13_H,
408 R14, R14_H);
410 // Class for all long registers except RCX (and RSP)
411 reg_class long_no_rcx_reg(RBP, RBP_H,
412 RDI, RDI_H,
413 RSI, RSI_H,
414 RAX, RAX_H,
415 RDX, RDX_H,
416 RBX, RBX_H,
417 R8, R8_H,
418 R9, R9_H,
419 R10, R10_H,
420 R11, R11_H,
421 R13, R13_H,
422 R14, R14_H);
424 // Class for all long registers except RAX (and RSP)
425 reg_class long_no_rax_reg(RBP, RBP_H,
426 RDX, RDX_H,
427 RDI, RDI_H,
428 RSI, RSI_H,
429 RCX, RCX_H,
430 RBX, RBX_H,
431 R8, R8_H,
432 R9, R9_H,
433 R10, R10_H,
434 R11, R11_H,
435 R13, R13_H,
436 R14, R14_H);
438 // Singleton class for RAX long register
439 reg_class long_rax_reg(RAX, RAX_H);
441 // Singleton class for RCX long register
442 reg_class long_rcx_reg(RCX, RCX_H);
444 // Singleton class for RDX long register
445 reg_class long_rdx_reg(RDX, RDX_H);
447 // Singleton class for R12 long register
448 reg_class long_r12_reg(R12, R12_H);
450 // Class for all int registers (except RSP)
451 reg_class int_reg(RAX,
452 RDX,
453 RBP,
454 RDI,
455 RSI,
456 RCX,
457 RBX,
458 R8,
459 R9,
460 R10,
461 R11,
462 R13,
463 R14);
465 // Class for all int registers except RCX (and RSP)
466 reg_class int_no_rcx_reg(RAX,
467 RDX,
468 RBP,
469 RDI,
470 RSI,
471 RBX,
472 R8,
473 R9,
474 R10,
475 R11,
476 R13,
477 R14);
479 // Class for all int registers except RAX, RDX (and RSP)
480 reg_class int_no_rax_rdx_reg(RBP,
481 RDI
482 RSI,
483 RCX,
484 RBX,
485 R8,
486 R9,
487 R10,
488 R11,
489 R13,
490 R14);
492 // Singleton class for RAX int register
493 reg_class int_rax_reg(RAX);
495 // Singleton class for RBX int register
496 reg_class int_rbx_reg(RBX);
498 // Singleton class for RCX int register
499 reg_class int_rcx_reg(RCX);
501 // Singleton class for RCX int register
502 reg_class int_rdx_reg(RDX);
504 // Singleton class for RCX int register
505 reg_class int_rdi_reg(RDI);
507 // Singleton class for instruction pointer
508 // reg_class ip_reg(RIP);
510 // Singleton class for condition codes
511 reg_class int_flags(RFLAGS);
513 // Class for all float registers
514 reg_class float_reg(XMM0,
515 XMM1,
516 XMM2,
517 XMM3,
518 XMM4,
519 XMM5,
520 XMM6,
521 XMM7,
522 XMM8,
523 XMM9,
524 XMM10,
525 XMM11,
526 XMM12,
527 XMM13,
528 XMM14,
529 XMM15);
531 // Class for all double registers
532 reg_class double_reg(XMM0, XMM0_H,
533 XMM1, XMM1_H,
534 XMM2, XMM2_H,
535 XMM3, XMM3_H,
536 XMM4, XMM4_H,
537 XMM5, XMM5_H,
538 XMM6, XMM6_H,
539 XMM7, XMM7_H,
540 XMM8, XMM8_H,
541 XMM9, XMM9_H,
542 XMM10, XMM10_H,
543 XMM11, XMM11_H,
544 XMM12, XMM12_H,
545 XMM13, XMM13_H,
546 XMM14, XMM14_H,
547 XMM15, XMM15_H);
548 %}
551 //----------SOURCE BLOCK-------------------------------------------------------
552 // This is a block of C++ code which provides values, functions, and
553 // definitions necessary in the rest of the architecture description
554 source %{
555 #define RELOC_IMM64 Assembler::imm64_operand
556 #define RELOC_DISP32 Assembler::disp32_operand
558 #define __ _masm.
560 // !!!!! Special hack to get all types of calls to specify the byte offset
561 // from the start of the call to the point where the return address
562 // will point.
563 int MachCallStaticJavaNode::ret_addr_offset()
564 {
565 return 5; // 5 bytes from start of call to where return address points
566 }
568 int MachCallDynamicJavaNode::ret_addr_offset()
569 {
570 return 15; // 15 bytes from start of call to where return address points
571 }
573 // In os_cpu .ad file
574 // int MachCallRuntimeNode::ret_addr_offset()
576 // Indicate if the safepoint node needs the polling page as an input.
577 // Since amd64 does not have absolute addressing but RIP-relative
578 // addressing and the polling page is within 2G, it doesn't.
579 bool SafePointNode::needs_polling_address_input()
580 {
581 return false;
582 }
584 //
585 // Compute padding required for nodes which need alignment
586 //
588 // The address of the call instruction needs to be 4-byte aligned to
589 // ensure that it does not span a cache line so that it can be patched.
590 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
591 {
592 current_offset += 1; // skip call opcode byte
593 return round_to(current_offset, alignment_required()) - current_offset;
594 }
596 // The address of the call instruction needs to be 4-byte aligned to
597 // ensure that it does not span a cache line so that it can be patched.
598 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
599 {
600 current_offset += 11; // skip movq instruction + call opcode byte
601 return round_to(current_offset, alignment_required()) - current_offset;
602 }
604 #ifndef PRODUCT
605 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
606 {
607 st->print("INT3");
608 }
609 #endif
611 // EMIT_RM()
612 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3)
613 {
614 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
615 *(cbuf.code_end()) = c;
616 cbuf.set_code_end(cbuf.code_end() + 1);
617 }
619 // EMIT_CC()
620 void emit_cc(CodeBuffer &cbuf, int f1, int f2)
621 {
622 unsigned char c = (unsigned char) (f1 | f2);
623 *(cbuf.code_end()) = c;
624 cbuf.set_code_end(cbuf.code_end() + 1);
625 }
627 // EMIT_OPCODE()
628 void emit_opcode(CodeBuffer &cbuf, int code)
629 {
630 *(cbuf.code_end()) = (unsigned char) code;
631 cbuf.set_code_end(cbuf.code_end() + 1);
632 }
634 // EMIT_OPCODE() w/ relocation information
635 void emit_opcode(CodeBuffer &cbuf,
636 int code, relocInfo::relocType reloc, int offset, int format)
637 {
638 cbuf.relocate(cbuf.inst_mark() + offset, reloc, format);
639 emit_opcode(cbuf, code);
640 }
642 // EMIT_D8()
643 void emit_d8(CodeBuffer &cbuf, int d8)
644 {
645 *(cbuf.code_end()) = (unsigned char) d8;
646 cbuf.set_code_end(cbuf.code_end() + 1);
647 }
649 // EMIT_D16()
650 void emit_d16(CodeBuffer &cbuf, int d16)
651 {
652 *((short *)(cbuf.code_end())) = d16;
653 cbuf.set_code_end(cbuf.code_end() + 2);
654 }
656 // EMIT_D32()
657 void emit_d32(CodeBuffer &cbuf, int d32)
658 {
659 *((int *)(cbuf.code_end())) = d32;
660 cbuf.set_code_end(cbuf.code_end() + 4);
661 }
663 // EMIT_D64()
664 void emit_d64(CodeBuffer &cbuf, int64_t d64)
665 {
666 *((int64_t*) (cbuf.code_end())) = d64;
667 cbuf.set_code_end(cbuf.code_end() + 8);
668 }
670 // emit 32 bit value and construct relocation entry from relocInfo::relocType
671 void emit_d32_reloc(CodeBuffer& cbuf,
672 int d32,
673 relocInfo::relocType reloc,
674 int format)
675 {
676 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
677 cbuf.relocate(cbuf.inst_mark(), reloc, format);
679 *((int*) (cbuf.code_end())) = d32;
680 cbuf.set_code_end(cbuf.code_end() + 4);
681 }
683 // emit 32 bit value and construct relocation entry from RelocationHolder
684 void emit_d32_reloc(CodeBuffer& cbuf,
685 int d32,
686 RelocationHolder const& rspec,
687 int format)
688 {
689 #ifdef ASSERT
690 if (rspec.reloc()->type() == relocInfo::oop_type &&
691 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
692 assert(oop((intptr_t)d32)->is_oop() && oop((intptr_t)d32)->is_perm(), "cannot embed non-perm oops in code");
693 }
694 #endif
695 cbuf.relocate(cbuf.inst_mark(), rspec, format);
697 *((int* )(cbuf.code_end())) = d32;
698 cbuf.set_code_end(cbuf.code_end() + 4);
699 }
701 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
702 address next_ip = cbuf.code_end() + 4;
703 emit_d32_reloc(cbuf, (int) (addr - next_ip),
704 external_word_Relocation::spec(addr),
705 RELOC_DISP32);
706 }
709 // emit 64 bit value and construct relocation entry from relocInfo::relocType
710 void emit_d64_reloc(CodeBuffer& cbuf,
711 int64_t d64,
712 relocInfo::relocType reloc,
713 int format)
714 {
715 cbuf.relocate(cbuf.inst_mark(), reloc, format);
717 *((int64_t*) (cbuf.code_end())) = d64;
718 cbuf.set_code_end(cbuf.code_end() + 8);
719 }
721 // emit 64 bit value and construct relocation entry from RelocationHolder
722 void emit_d64_reloc(CodeBuffer& cbuf,
723 int64_t d64,
724 RelocationHolder const& rspec,
725 int format)
726 {
727 #ifdef ASSERT
728 if (rspec.reloc()->type() == relocInfo::oop_type &&
729 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
730 assert(oop(d64)->is_oop() && oop(d64)->is_perm(),
731 "cannot embed non-perm oops in code");
732 }
733 #endif
734 cbuf.relocate(cbuf.inst_mark(), rspec, format);
736 *((int64_t*) (cbuf.code_end())) = d64;
737 cbuf.set_code_end(cbuf.code_end() + 8);
738 }
740 // Access stack slot for load or store
741 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
742 {
743 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src])
744 if (-0x80 <= disp && disp < 0x80) {
745 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte
746 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
747 emit_d8(cbuf, disp); // Displacement // R/M byte
748 } else {
749 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte
750 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
751 emit_d32(cbuf, disp); // Displacement // R/M byte
752 }
753 }
755 // rRegI ereg, memory mem) %{ // emit_reg_mem
756 void encode_RegMem(CodeBuffer &cbuf,
757 int reg,
758 int base, int index, int scale, int disp, bool disp_is_oop)
759 {
760 assert(!disp_is_oop, "cannot have disp");
761 int regenc = reg & 7;
762 int baseenc = base & 7;
763 int indexenc = index & 7;
765 // There is no index & no scale, use form without SIB byte
766 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
767 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
768 if (disp == 0 && base != RBP_enc && base != R13_enc) {
769 emit_rm(cbuf, 0x0, regenc, baseenc); // *
770 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
771 // If 8-bit displacement, mode 0x1
772 emit_rm(cbuf, 0x1, regenc, baseenc); // *
773 emit_d8(cbuf, disp);
774 } else {
775 // If 32-bit displacement
776 if (base == -1) { // Special flag for absolute address
777 emit_rm(cbuf, 0x0, regenc, 0x5); // *
778 if (disp_is_oop) {
779 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
780 } else {
781 emit_d32(cbuf, disp);
782 }
783 } else {
784 // Normal base + offset
785 emit_rm(cbuf, 0x2, regenc, baseenc); // *
786 if (disp_is_oop) {
787 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
788 } else {
789 emit_d32(cbuf, disp);
790 }
791 }
792 }
793 } else {
794 // Else, encode with the SIB byte
795 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
796 if (disp == 0 && base != RBP_enc && base != R13_enc) {
797 // If no displacement
798 emit_rm(cbuf, 0x0, regenc, 0x4); // *
799 emit_rm(cbuf, scale, indexenc, baseenc);
800 } else {
801 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
802 // If 8-bit displacement, mode 0x1
803 emit_rm(cbuf, 0x1, regenc, 0x4); // *
804 emit_rm(cbuf, scale, indexenc, baseenc);
805 emit_d8(cbuf, disp);
806 } else {
807 // If 32-bit displacement
808 if (base == 0x04 ) {
809 emit_rm(cbuf, 0x2, regenc, 0x4);
810 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
811 } else {
812 emit_rm(cbuf, 0x2, regenc, 0x4);
813 emit_rm(cbuf, scale, indexenc, baseenc); // *
814 }
815 if (disp_is_oop) {
816 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
817 } else {
818 emit_d32(cbuf, disp);
819 }
820 }
821 }
822 }
823 }
825 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
826 {
827 if (dstenc != srcenc) {
828 if (dstenc < 8) {
829 if (srcenc >= 8) {
830 emit_opcode(cbuf, Assembler::REX_B);
831 srcenc -= 8;
832 }
833 } else {
834 if (srcenc < 8) {
835 emit_opcode(cbuf, Assembler::REX_R);
836 } else {
837 emit_opcode(cbuf, Assembler::REX_RB);
838 srcenc -= 8;
839 }
840 dstenc -= 8;
841 }
843 emit_opcode(cbuf, 0x8B);
844 emit_rm(cbuf, 0x3, dstenc, srcenc);
845 }
846 }
848 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
849 if( dst_encoding == src_encoding ) {
850 // reg-reg copy, use an empty encoding
851 } else {
852 MacroAssembler _masm(&cbuf);
854 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
855 }
856 }
859 //=============================================================================
860 #ifndef PRODUCT
861 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
862 {
863 Compile* C = ra_->C;
865 int framesize = C->frame_slots() << LogBytesPerInt;
866 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
867 // Remove wordSize for return adr already pushed
868 // and another for the RBP we are going to save
869 framesize -= 2*wordSize;
870 bool need_nop = true;
872 // Calls to C2R adapters often do not accept exceptional returns.
873 // We require that their callers must bang for them. But be
874 // careful, because some VM calls (such as call site linkage) can
875 // use several kilobytes of stack. But the stack safety zone should
876 // account for that. See bugs 4446381, 4468289, 4497237.
877 if (C->need_stack_bang(framesize)) {
878 st->print_cr("# stack bang"); st->print("\t");
879 need_nop = false;
880 }
881 st->print_cr("pushq rbp"); st->print("\t");
883 if (VerifyStackAtCalls) {
884 // Majik cookie to verify stack depth
885 st->print_cr("pushq 0xffffffffbadb100d"
886 "\t# Majik cookie for stack depth check");
887 st->print("\t");
888 framesize -= wordSize; // Remove 2 for cookie
889 need_nop = false;
890 }
892 if (framesize) {
893 st->print("subq rsp, #%d\t# Create frame", framesize);
894 if (framesize < 0x80 && need_nop) {
895 st->print("\n\tnop\t# nop for patch_verified_entry");
896 }
897 }
898 }
899 #endif
901 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
902 {
903 Compile* C = ra_->C;
905 // WARNING: Initial instruction MUST be 5 bytes or longer so that
906 // NativeJump::patch_verified_entry will be able to patch out the entry
907 // code safely. The fldcw is ok at 6 bytes, the push to verify stack
908 // depth is ok at 5 bytes, the frame allocation can be either 3 or
909 // 6 bytes. So if we don't do the fldcw or the push then we must
910 // use the 6 byte frame allocation even if we have no frame. :-(
911 // If method sets FPU control word do it now
913 int framesize = C->frame_slots() << LogBytesPerInt;
914 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
915 // Remove wordSize for return adr already pushed
916 // and another for the RBP we are going to save
917 framesize -= 2*wordSize;
918 bool need_nop = true;
920 // Calls to C2R adapters often do not accept exceptional returns.
921 // We require that their callers must bang for them. But be
922 // careful, because some VM calls (such as call site linkage) can
923 // use several kilobytes of stack. But the stack safety zone should
924 // account for that. See bugs 4446381, 4468289, 4497237.
925 if (C->need_stack_bang(framesize)) {
926 MacroAssembler masm(&cbuf);
927 masm.generate_stack_overflow_check(framesize);
928 need_nop = false;
929 }
931 // We always push rbp so that on return to interpreter rbp will be
932 // restored correctly and we can correct the stack.
933 emit_opcode(cbuf, 0x50 | RBP_enc);
935 if (VerifyStackAtCalls) {
936 // Majik cookie to verify stack depth
937 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
938 emit_d32(cbuf, 0xbadb100d);
939 framesize -= wordSize; // Remove 2 for cookie
940 need_nop = false;
941 }
943 if (framesize) {
944 emit_opcode(cbuf, Assembler::REX_W);
945 if (framesize < 0x80) {
946 emit_opcode(cbuf, 0x83); // sub SP,#framesize
947 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
948 emit_d8(cbuf, framesize);
949 if (need_nop) {
950 emit_opcode(cbuf, 0x90); // nop
951 }
952 } else {
953 emit_opcode(cbuf, 0x81); // sub SP,#framesize
954 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
955 emit_d32(cbuf, framesize);
956 }
957 }
959 C->set_frame_complete(cbuf.code_end() - cbuf.code_begin());
961 #ifdef ASSERT
962 if (VerifyStackAtCalls) {
963 Label L;
964 MacroAssembler masm(&cbuf);
965 masm.pushq(rax);
966 masm.movq(rax, rsp);
967 masm.andq(rax, StackAlignmentInBytes-1);
968 masm.cmpq(rax, StackAlignmentInBytes-wordSize);
969 masm.popq(rax);
970 masm.jcc(Assembler::equal, L);
971 masm.stop("Stack is not properly aligned!");
972 masm.bind(L);
973 }
974 #endif
975 }
977 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
978 {
979 return MachNode::size(ra_); // too many variables; just compute it
980 // the hard way
981 }
983 int MachPrologNode::reloc() const
984 {
985 return 0; // a large enough number
986 }
988 //=============================================================================
989 #ifndef PRODUCT
990 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
991 {
992 Compile* C = ra_->C;
993 int framesize = C->frame_slots() << LogBytesPerInt;
994 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
995 // Remove word for return adr already pushed
996 // and RBP
997 framesize -= 2*wordSize;
999 if (framesize) {
1000 st->print_cr("addq\trsp, %d\t# Destroy frame", framesize);
1001 st->print("\t");
1002 }
1004 st->print_cr("popq\trbp");
1005 if (do_polling() && C->is_method_compilation()) {
1006 st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t"
1007 "# Safepoint: poll for GC");
1008 st->print("\t");
1009 }
1010 }
1011 #endif
1013 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1014 {
1015 Compile* C = ra_->C;
1016 int framesize = C->frame_slots() << LogBytesPerInt;
1017 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1018 // Remove word for return adr already pushed
1019 // and RBP
1020 framesize -= 2*wordSize;
1022 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1024 if (framesize) {
1025 emit_opcode(cbuf, Assembler::REX_W);
1026 if (framesize < 0x80) {
1027 emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1028 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1029 emit_d8(cbuf, framesize);
1030 } else {
1031 emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1032 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1033 emit_d32(cbuf, framesize);
1034 }
1035 }
1037 // popq rbp
1038 emit_opcode(cbuf, 0x58 | RBP_enc);
1040 if (do_polling() && C->is_method_compilation()) {
1041 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
1042 // XXX reg_mem doesn't support RIP-relative addressing yet
1043 cbuf.set_inst_mark();
1044 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX
1045 emit_opcode(cbuf, 0x85); // testl
1046 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
1047 // cbuf.inst_mark() is beginning of instruction
1048 emit_d32_reloc(cbuf, os::get_polling_page());
1049 // relocInfo::poll_return_type,
1050 }
1051 }
1053 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1054 {
1055 Compile* C = ra_->C;
1056 int framesize = C->frame_slots() << LogBytesPerInt;
1057 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1058 // Remove word for return adr already pushed
1059 // and RBP
1060 framesize -= 2*wordSize;
1062 uint size = 0;
1064 if (do_polling() && C->is_method_compilation()) {
1065 size += 6;
1066 }
1068 // count popq rbp
1069 size++;
1071 if (framesize) {
1072 if (framesize < 0x80) {
1073 size += 4;
1074 } else if (framesize) {
1075 size += 7;
1076 }
1077 }
1079 return size;
1080 }
1082 int MachEpilogNode::reloc() const
1083 {
1084 return 2; // a large enough number
1085 }
1087 const Pipeline* MachEpilogNode::pipeline() const
1088 {
1089 return MachNode::pipeline_class();
1090 }
1092 int MachEpilogNode::safepoint_offset() const
1093 {
1094 return 0;
1095 }
1097 //=============================================================================
1099 enum RC {
1100 rc_bad,
1101 rc_int,
1102 rc_float,
1103 rc_stack
1104 };
1106 static enum RC rc_class(OptoReg::Name reg)
1107 {
1108 if( !OptoReg::is_valid(reg) ) return rc_bad;
1110 if (OptoReg::is_stack(reg)) return rc_stack;
1112 VMReg r = OptoReg::as_VMReg(reg);
1114 if (r->is_Register()) return rc_int;
1116 assert(r->is_XMMRegister(), "must be");
1117 return rc_float;
1118 }
1120 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1121 PhaseRegAlloc* ra_,
1122 bool do_size,
1123 outputStream* st) const
1124 {
1126 // Get registers to move
1127 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1128 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1129 OptoReg::Name dst_second = ra_->get_reg_second(this);
1130 OptoReg::Name dst_first = ra_->get_reg_first(this);
1132 enum RC src_second_rc = rc_class(src_second);
1133 enum RC src_first_rc = rc_class(src_first);
1134 enum RC dst_second_rc = rc_class(dst_second);
1135 enum RC dst_first_rc = rc_class(dst_first);
1137 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1138 "must move at least 1 register" );
1140 if (src_first == dst_first && src_second == dst_second) {
1141 // Self copy, no move
1142 return 0;
1143 } else if (src_first_rc == rc_stack) {
1144 // mem ->
1145 if (dst_first_rc == rc_stack) {
1146 // mem -> mem
1147 assert(src_second != dst_first, "overlap");
1148 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1149 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1150 // 64-bit
1151 int src_offset = ra_->reg2offset(src_first);
1152 int dst_offset = ra_->reg2offset(dst_first);
1153 if (cbuf) {
1154 emit_opcode(*cbuf, 0xFF);
1155 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1157 emit_opcode(*cbuf, 0x8F);
1158 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1160 #ifndef PRODUCT
1161 } else if (!do_size) {
1162 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1163 "popq [rsp + #%d]",
1164 src_offset,
1165 dst_offset);
1166 #endif
1167 }
1168 return
1169 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1170 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1171 } else {
1172 // 32-bit
1173 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1174 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1175 // No pushl/popl, so:
1176 int src_offset = ra_->reg2offset(src_first);
1177 int dst_offset = ra_->reg2offset(dst_first);
1178 if (cbuf) {
1179 emit_opcode(*cbuf, Assembler::REX_W);
1180 emit_opcode(*cbuf, 0x89);
1181 emit_opcode(*cbuf, 0x44);
1182 emit_opcode(*cbuf, 0x24);
1183 emit_opcode(*cbuf, 0xF8);
1185 emit_opcode(*cbuf, 0x8B);
1186 encode_RegMem(*cbuf,
1187 RAX_enc,
1188 RSP_enc, 0x4, 0, src_offset,
1189 false);
1191 emit_opcode(*cbuf, 0x89);
1192 encode_RegMem(*cbuf,
1193 RAX_enc,
1194 RSP_enc, 0x4, 0, dst_offset,
1195 false);
1197 emit_opcode(*cbuf, Assembler::REX_W);
1198 emit_opcode(*cbuf, 0x8B);
1199 emit_opcode(*cbuf, 0x44);
1200 emit_opcode(*cbuf, 0x24);
1201 emit_opcode(*cbuf, 0xF8);
1203 #ifndef PRODUCT
1204 } else if (!do_size) {
1205 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1206 "movl rax, [rsp + #%d]\n\t"
1207 "movl [rsp + #%d], rax\n\t"
1208 "movq rax, [rsp - #8]",
1209 src_offset,
1210 dst_offset);
1211 #endif
1212 }
1213 return
1214 5 + // movq
1215 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1216 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1217 5; // movq
1218 }
1219 } else if (dst_first_rc == rc_int) {
1220 // mem -> gpr
1221 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1222 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1223 // 64-bit
1224 int offset = ra_->reg2offset(src_first);
1225 if (cbuf) {
1226 if (Matcher::_regEncode[dst_first] < 8) {
1227 emit_opcode(*cbuf, Assembler::REX_W);
1228 } else {
1229 emit_opcode(*cbuf, Assembler::REX_WR);
1230 }
1231 emit_opcode(*cbuf, 0x8B);
1232 encode_RegMem(*cbuf,
1233 Matcher::_regEncode[dst_first],
1234 RSP_enc, 0x4, 0, offset,
1235 false);
1236 #ifndef PRODUCT
1237 } else if (!do_size) {
1238 st->print("movq %s, [rsp + #%d]\t# spill",
1239 Matcher::regName[dst_first],
1240 offset);
1241 #endif
1242 }
1243 return
1244 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1245 } else {
1246 // 32-bit
1247 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1248 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1249 int offset = ra_->reg2offset(src_first);
1250 if (cbuf) {
1251 if (Matcher::_regEncode[dst_first] >= 8) {
1252 emit_opcode(*cbuf, Assembler::REX_R);
1253 }
1254 emit_opcode(*cbuf, 0x8B);
1255 encode_RegMem(*cbuf,
1256 Matcher::_regEncode[dst_first],
1257 RSP_enc, 0x4, 0, offset,
1258 false);
1259 #ifndef PRODUCT
1260 } else if (!do_size) {
1261 st->print("movl %s, [rsp + #%d]\t# spill",
1262 Matcher::regName[dst_first],
1263 offset);
1264 #endif
1265 }
1266 return
1267 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1268 ((Matcher::_regEncode[dst_first] < 8)
1269 ? 3
1270 : 4); // REX
1271 }
1272 } else if (dst_first_rc == rc_float) {
1273 // mem-> xmm
1274 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1275 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1276 // 64-bit
1277 int offset = ra_->reg2offset(src_first);
1278 if (cbuf) {
1279 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1280 if (Matcher::_regEncode[dst_first] >= 8) {
1281 emit_opcode(*cbuf, Assembler::REX_R);
1282 }
1283 emit_opcode(*cbuf, 0x0F);
1284 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1285 encode_RegMem(*cbuf,
1286 Matcher::_regEncode[dst_first],
1287 RSP_enc, 0x4, 0, offset,
1288 false);
1289 #ifndef PRODUCT
1290 } else if (!do_size) {
1291 st->print("%s %s, [rsp + #%d]\t# spill",
1292 UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1293 Matcher::regName[dst_first],
1294 offset);
1295 #endif
1296 }
1297 return
1298 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1299 ((Matcher::_regEncode[dst_first] < 8)
1300 ? 5
1301 : 6); // REX
1302 } else {
1303 // 32-bit
1304 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1305 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1306 int offset = ra_->reg2offset(src_first);
1307 if (cbuf) {
1308 emit_opcode(*cbuf, 0xF3);
1309 if (Matcher::_regEncode[dst_first] >= 8) {
1310 emit_opcode(*cbuf, Assembler::REX_R);
1311 }
1312 emit_opcode(*cbuf, 0x0F);
1313 emit_opcode(*cbuf, 0x10);
1314 encode_RegMem(*cbuf,
1315 Matcher::_regEncode[dst_first],
1316 RSP_enc, 0x4, 0, offset,
1317 false);
1318 #ifndef PRODUCT
1319 } else if (!do_size) {
1320 st->print("movss %s, [rsp + #%d]\t# spill",
1321 Matcher::regName[dst_first],
1322 offset);
1323 #endif
1324 }
1325 return
1326 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1327 ((Matcher::_regEncode[dst_first] < 8)
1328 ? 5
1329 : 6); // REX
1330 }
1331 }
1332 } else if (src_first_rc == rc_int) {
1333 // gpr ->
1334 if (dst_first_rc == rc_stack) {
1335 // gpr -> mem
1336 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1337 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1338 // 64-bit
1339 int offset = ra_->reg2offset(dst_first);
1340 if (cbuf) {
1341 if (Matcher::_regEncode[src_first] < 8) {
1342 emit_opcode(*cbuf, Assembler::REX_W);
1343 } else {
1344 emit_opcode(*cbuf, Assembler::REX_WR);
1345 }
1346 emit_opcode(*cbuf, 0x89);
1347 encode_RegMem(*cbuf,
1348 Matcher::_regEncode[src_first],
1349 RSP_enc, 0x4, 0, offset,
1350 false);
1351 #ifndef PRODUCT
1352 } else if (!do_size) {
1353 st->print("movq [rsp + #%d], %s\t# spill",
1354 offset,
1355 Matcher::regName[src_first]);
1356 #endif
1357 }
1358 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1359 } else {
1360 // 32-bit
1361 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1362 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1363 int offset = ra_->reg2offset(dst_first);
1364 if (cbuf) {
1365 if (Matcher::_regEncode[src_first] >= 8) {
1366 emit_opcode(*cbuf, Assembler::REX_R);
1367 }
1368 emit_opcode(*cbuf, 0x89);
1369 encode_RegMem(*cbuf,
1370 Matcher::_regEncode[src_first],
1371 RSP_enc, 0x4, 0, offset,
1372 false);
1373 #ifndef PRODUCT
1374 } else if (!do_size) {
1375 st->print("movl [rsp + #%d], %s\t# spill",
1376 offset,
1377 Matcher::regName[src_first]);
1378 #endif
1379 }
1380 return
1381 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1382 ((Matcher::_regEncode[src_first] < 8)
1383 ? 3
1384 : 4); // REX
1385 }
1386 } else if (dst_first_rc == rc_int) {
1387 // gpr -> gpr
1388 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1389 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1390 // 64-bit
1391 if (cbuf) {
1392 if (Matcher::_regEncode[dst_first] < 8) {
1393 if (Matcher::_regEncode[src_first] < 8) {
1394 emit_opcode(*cbuf, Assembler::REX_W);
1395 } else {
1396 emit_opcode(*cbuf, Assembler::REX_WB);
1397 }
1398 } else {
1399 if (Matcher::_regEncode[src_first] < 8) {
1400 emit_opcode(*cbuf, Assembler::REX_WR);
1401 } else {
1402 emit_opcode(*cbuf, Assembler::REX_WRB);
1403 }
1404 }
1405 emit_opcode(*cbuf, 0x8B);
1406 emit_rm(*cbuf, 0x3,
1407 Matcher::_regEncode[dst_first] & 7,
1408 Matcher::_regEncode[src_first] & 7);
1409 #ifndef PRODUCT
1410 } else if (!do_size) {
1411 st->print("movq %s, %s\t# spill",
1412 Matcher::regName[dst_first],
1413 Matcher::regName[src_first]);
1414 #endif
1415 }
1416 return 3; // REX
1417 } else {
1418 // 32-bit
1419 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1420 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1421 if (cbuf) {
1422 if (Matcher::_regEncode[dst_first] < 8) {
1423 if (Matcher::_regEncode[src_first] >= 8) {
1424 emit_opcode(*cbuf, Assembler::REX_B);
1425 }
1426 } else {
1427 if (Matcher::_regEncode[src_first] < 8) {
1428 emit_opcode(*cbuf, Assembler::REX_R);
1429 } else {
1430 emit_opcode(*cbuf, Assembler::REX_RB);
1431 }
1432 }
1433 emit_opcode(*cbuf, 0x8B);
1434 emit_rm(*cbuf, 0x3,
1435 Matcher::_regEncode[dst_first] & 7,
1436 Matcher::_regEncode[src_first] & 7);
1437 #ifndef PRODUCT
1438 } else if (!do_size) {
1439 st->print("movl %s, %s\t# spill",
1440 Matcher::regName[dst_first],
1441 Matcher::regName[src_first]);
1442 #endif
1443 }
1444 return
1445 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1446 ? 2
1447 : 3; // REX
1448 }
1449 } else if (dst_first_rc == rc_float) {
1450 // gpr -> xmm
1451 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1452 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1453 // 64-bit
1454 if (cbuf) {
1455 emit_opcode(*cbuf, 0x66);
1456 if (Matcher::_regEncode[dst_first] < 8) {
1457 if (Matcher::_regEncode[src_first] < 8) {
1458 emit_opcode(*cbuf, Assembler::REX_W);
1459 } else {
1460 emit_opcode(*cbuf, Assembler::REX_WB);
1461 }
1462 } else {
1463 if (Matcher::_regEncode[src_first] < 8) {
1464 emit_opcode(*cbuf, Assembler::REX_WR);
1465 } else {
1466 emit_opcode(*cbuf, Assembler::REX_WRB);
1467 }
1468 }
1469 emit_opcode(*cbuf, 0x0F);
1470 emit_opcode(*cbuf, 0x6E);
1471 emit_rm(*cbuf, 0x3,
1472 Matcher::_regEncode[dst_first] & 7,
1473 Matcher::_regEncode[src_first] & 7);
1474 #ifndef PRODUCT
1475 } else if (!do_size) {
1476 st->print("movdq %s, %s\t# spill",
1477 Matcher::regName[dst_first],
1478 Matcher::regName[src_first]);
1479 #endif
1480 }
1481 return 5; // REX
1482 } else {
1483 // 32-bit
1484 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1485 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1486 if (cbuf) {
1487 emit_opcode(*cbuf, 0x66);
1488 if (Matcher::_regEncode[dst_first] < 8) {
1489 if (Matcher::_regEncode[src_first] >= 8) {
1490 emit_opcode(*cbuf, Assembler::REX_B);
1491 }
1492 } else {
1493 if (Matcher::_regEncode[src_first] < 8) {
1494 emit_opcode(*cbuf, Assembler::REX_R);
1495 } else {
1496 emit_opcode(*cbuf, Assembler::REX_RB);
1497 }
1498 }
1499 emit_opcode(*cbuf, 0x0F);
1500 emit_opcode(*cbuf, 0x6E);
1501 emit_rm(*cbuf, 0x3,
1502 Matcher::_regEncode[dst_first] & 7,
1503 Matcher::_regEncode[src_first] & 7);
1504 #ifndef PRODUCT
1505 } else if (!do_size) {
1506 st->print("movdl %s, %s\t# spill",
1507 Matcher::regName[dst_first],
1508 Matcher::regName[src_first]);
1509 #endif
1510 }
1511 return
1512 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1513 ? 4
1514 : 5; // REX
1515 }
1516 }
1517 } else if (src_first_rc == rc_float) {
1518 // xmm ->
1519 if (dst_first_rc == rc_stack) {
1520 // xmm -> mem
1521 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1522 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1523 // 64-bit
1524 int offset = ra_->reg2offset(dst_first);
1525 if (cbuf) {
1526 emit_opcode(*cbuf, 0xF2);
1527 if (Matcher::_regEncode[src_first] >= 8) {
1528 emit_opcode(*cbuf, Assembler::REX_R);
1529 }
1530 emit_opcode(*cbuf, 0x0F);
1531 emit_opcode(*cbuf, 0x11);
1532 encode_RegMem(*cbuf,
1533 Matcher::_regEncode[src_first],
1534 RSP_enc, 0x4, 0, offset,
1535 false);
1536 #ifndef PRODUCT
1537 } else if (!do_size) {
1538 st->print("movsd [rsp + #%d], %s\t# spill",
1539 offset,
1540 Matcher::regName[src_first]);
1541 #endif
1542 }
1543 return
1544 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1545 ((Matcher::_regEncode[src_first] < 8)
1546 ? 5
1547 : 6); // REX
1548 } else {
1549 // 32-bit
1550 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1551 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1552 int offset = ra_->reg2offset(dst_first);
1553 if (cbuf) {
1554 emit_opcode(*cbuf, 0xF3);
1555 if (Matcher::_regEncode[src_first] >= 8) {
1556 emit_opcode(*cbuf, Assembler::REX_R);
1557 }
1558 emit_opcode(*cbuf, 0x0F);
1559 emit_opcode(*cbuf, 0x11);
1560 encode_RegMem(*cbuf,
1561 Matcher::_regEncode[src_first],
1562 RSP_enc, 0x4, 0, offset,
1563 false);
1564 #ifndef PRODUCT
1565 } else if (!do_size) {
1566 st->print("movss [rsp + #%d], %s\t# spill",
1567 offset,
1568 Matcher::regName[src_first]);
1569 #endif
1570 }
1571 return
1572 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1573 ((Matcher::_regEncode[src_first] < 8)
1574 ? 5
1575 : 6); // REX
1576 }
1577 } else if (dst_first_rc == rc_int) {
1578 // xmm -> gpr
1579 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1580 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1581 // 64-bit
1582 if (cbuf) {
1583 emit_opcode(*cbuf, 0x66);
1584 if (Matcher::_regEncode[dst_first] < 8) {
1585 if (Matcher::_regEncode[src_first] < 8) {
1586 emit_opcode(*cbuf, Assembler::REX_W);
1587 } else {
1588 emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1589 }
1590 } else {
1591 if (Matcher::_regEncode[src_first] < 8) {
1592 emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1593 } else {
1594 emit_opcode(*cbuf, Assembler::REX_WRB);
1595 }
1596 }
1597 emit_opcode(*cbuf, 0x0F);
1598 emit_opcode(*cbuf, 0x7E);
1599 emit_rm(*cbuf, 0x3,
1600 Matcher::_regEncode[dst_first] & 7,
1601 Matcher::_regEncode[src_first] & 7);
1602 #ifndef PRODUCT
1603 } else if (!do_size) {
1604 st->print("movdq %s, %s\t# spill",
1605 Matcher::regName[dst_first],
1606 Matcher::regName[src_first]);
1607 #endif
1608 }
1609 return 5; // REX
1610 } else {
1611 // 32-bit
1612 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1613 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1614 if (cbuf) {
1615 emit_opcode(*cbuf, 0x66);
1616 if (Matcher::_regEncode[dst_first] < 8) {
1617 if (Matcher::_regEncode[src_first] >= 8) {
1618 emit_opcode(*cbuf, Assembler::REX_R); // attention!
1619 }
1620 } else {
1621 if (Matcher::_regEncode[src_first] < 8) {
1622 emit_opcode(*cbuf, Assembler::REX_B); // attention!
1623 } else {
1624 emit_opcode(*cbuf, Assembler::REX_RB);
1625 }
1626 }
1627 emit_opcode(*cbuf, 0x0F);
1628 emit_opcode(*cbuf, 0x7E);
1629 emit_rm(*cbuf, 0x3,
1630 Matcher::_regEncode[dst_first] & 7,
1631 Matcher::_regEncode[src_first] & 7);
1632 #ifndef PRODUCT
1633 } else if (!do_size) {
1634 st->print("movdl %s, %s\t# spill",
1635 Matcher::regName[dst_first],
1636 Matcher::regName[src_first]);
1637 #endif
1638 }
1639 return
1640 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1641 ? 4
1642 : 5; // REX
1643 }
1644 } else if (dst_first_rc == rc_float) {
1645 // xmm -> xmm
1646 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1647 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1648 // 64-bit
1649 if (cbuf) {
1650 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1651 if (Matcher::_regEncode[dst_first] < 8) {
1652 if (Matcher::_regEncode[src_first] >= 8) {
1653 emit_opcode(*cbuf, Assembler::REX_B);
1654 }
1655 } else {
1656 if (Matcher::_regEncode[src_first] < 8) {
1657 emit_opcode(*cbuf, Assembler::REX_R);
1658 } else {
1659 emit_opcode(*cbuf, Assembler::REX_RB);
1660 }
1661 }
1662 emit_opcode(*cbuf, 0x0F);
1663 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1664 emit_rm(*cbuf, 0x3,
1665 Matcher::_regEncode[dst_first] & 7,
1666 Matcher::_regEncode[src_first] & 7);
1667 #ifndef PRODUCT
1668 } else if (!do_size) {
1669 st->print("%s %s, %s\t# spill",
1670 UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1671 Matcher::regName[dst_first],
1672 Matcher::regName[src_first]);
1673 #endif
1674 }
1675 return
1676 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1677 ? 4
1678 : 5; // REX
1679 } else {
1680 // 32-bit
1681 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1682 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1683 if (cbuf) {
1684 if (!UseXmmRegToRegMoveAll)
1685 emit_opcode(*cbuf, 0xF3);
1686 if (Matcher::_regEncode[dst_first] < 8) {
1687 if (Matcher::_regEncode[src_first] >= 8) {
1688 emit_opcode(*cbuf, Assembler::REX_B);
1689 }
1690 } else {
1691 if (Matcher::_regEncode[src_first] < 8) {
1692 emit_opcode(*cbuf, Assembler::REX_R);
1693 } else {
1694 emit_opcode(*cbuf, Assembler::REX_RB);
1695 }
1696 }
1697 emit_opcode(*cbuf, 0x0F);
1698 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1699 emit_rm(*cbuf, 0x3,
1700 Matcher::_regEncode[dst_first] & 7,
1701 Matcher::_regEncode[src_first] & 7);
1702 #ifndef PRODUCT
1703 } else if (!do_size) {
1704 st->print("%s %s, %s\t# spill",
1705 UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1706 Matcher::regName[dst_first],
1707 Matcher::regName[src_first]);
1708 #endif
1709 }
1710 return
1711 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1712 ? (UseXmmRegToRegMoveAll ? 3 : 4)
1713 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1714 }
1715 }
1716 }
1718 assert(0," foo ");
1719 Unimplemented();
1721 return 0;
1722 }
1724 #ifndef PRODUCT
1725 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1726 {
1727 implementation(NULL, ra_, false, st);
1728 }
1729 #endif
1731 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1732 {
1733 implementation(&cbuf, ra_, false, NULL);
1734 }
1736 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1737 {
1738 return implementation(NULL, ra_, true, NULL);
1739 }
1741 //=============================================================================
1742 #ifndef PRODUCT
1743 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1744 {
1745 st->print("nop \t# %d bytes pad for loops and calls", _count);
1746 }
1747 #endif
1749 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1750 {
1751 MacroAssembler _masm(&cbuf);
1752 __ nop(_count);
1753 }
1755 uint MachNopNode::size(PhaseRegAlloc*) const
1756 {
1757 return _count;
1758 }
1761 //=============================================================================
1762 #ifndef PRODUCT
1763 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1764 {
1765 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1766 int reg = ra_->get_reg_first(this);
1767 st->print("leaq %s, [rsp + #%d]\t# box lock",
1768 Matcher::regName[reg], offset);
1769 }
1770 #endif
1772 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1773 {
1774 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1775 int reg = ra_->get_encode(this);
1776 if (offset >= 0x80) {
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, 0x2, reg & 7, 0x04);
1780 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1781 emit_d32(cbuf, offset);
1782 } else {
1783 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1784 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1785 emit_rm(cbuf, 0x1, reg & 7, 0x04);
1786 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1787 emit_d8(cbuf, offset);
1788 }
1789 }
1791 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1792 {
1793 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1794 return (offset < 0x80) ? 5 : 8; // REX
1795 }
1797 //=============================================================================
1799 // emit call stub, compiled java to interpreter
1800 void emit_java_to_interp(CodeBuffer& cbuf)
1801 {
1802 // Stub is fixed up when the corresponding call is converted from
1803 // calling compiled code to calling interpreted code.
1804 // movq rbx, 0
1805 // jmp -5 # to self
1807 address mark = cbuf.inst_mark(); // get mark within main instrs section
1809 // Note that the code buffer's inst_mark is always relative to insts.
1810 // That's why we must use the macroassembler to generate a stub.
1811 MacroAssembler _masm(&cbuf);
1813 address base =
1814 __ start_a_stub(Compile::MAX_stubs_size);
1815 if (base == NULL) return; // CodeBuffer::expand failed
1816 // static stub relocation stores the instruction address of the call
1817 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1818 // static stub relocation also tags the methodOop in the code-stream.
1819 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time
1820 __ jump(RuntimeAddress(__ pc()));
1822 // Update current stubs pointer and restore code_end.
1823 __ end_a_stub();
1824 }
1826 // size of call stub, compiled java to interpretor
1827 uint size_java_to_interp()
1828 {
1829 return 15; // movq (1+1+8); jmp (1+4)
1830 }
1832 // relocation entries for call stub, compiled java to interpretor
1833 uint reloc_java_to_interp()
1834 {
1835 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1836 }
1838 //=============================================================================
1839 #ifndef PRODUCT
1840 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1841 {
1842 if (UseCompressedOops) {
1843 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
1844 st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]");
1845 st->print_cr("cmpq rax, rscratch1\t # Inline cache check");
1846 } else {
1847 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
1848 "# Inline cache check", oopDesc::klass_offset_in_bytes());
1849 }
1850 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1851 st->print_cr("\tnop");
1852 if (!OptoBreakpoint) {
1853 st->print_cr("\tnop");
1854 }
1855 }
1856 #endif
1858 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1859 {
1860 MacroAssembler masm(&cbuf);
1861 #ifdef ASSERT
1862 uint code_size = cbuf.code_size();
1863 #endif
1864 if (UseCompressedOops) {
1865 masm.load_klass(rscratch1, j_rarg0);
1866 masm.cmpq(rax, rscratch1);
1867 } else {
1868 masm.cmpq(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1869 }
1871 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1873 /* WARNING these NOPs are critical so that verified entry point is properly
1874 aligned for patching by NativeJump::patch_verified_entry() */
1875 int nops_cnt = 1;
1876 if (!OptoBreakpoint) {
1877 // Leave space for int3
1878 nops_cnt += 1;
1879 }
1880 if (UseCompressedOops) {
1881 // ??? divisible by 4 is aligned?
1882 nops_cnt += 1;
1883 }
1884 masm.nop(nops_cnt);
1886 assert(cbuf.code_size() - code_size == size(ra_),
1887 "checking code size of inline cache node");
1888 }
1890 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1891 {
1892 if (UseCompressedOops) {
1893 return OptoBreakpoint ? 19 : 20;
1894 } else {
1895 return OptoBreakpoint ? 11 : 12;
1896 }
1897 }
1900 //=============================================================================
1901 uint size_exception_handler()
1902 {
1903 // NativeCall instruction size is the same as NativeJump.
1904 // Note that this value is also credited (in output.cpp) to
1905 // the size of the code section.
1906 return NativeJump::instruction_size;
1907 }
1909 // Emit exception handler code.
1910 int emit_exception_handler(CodeBuffer& cbuf)
1911 {
1913 // Note that the code buffer's inst_mark is always relative to insts.
1914 // That's why we must use the macroassembler to generate a handler.
1915 MacroAssembler _masm(&cbuf);
1916 address base =
1917 __ start_a_stub(size_exception_handler());
1918 if (base == NULL) return 0; // CodeBuffer::expand failed
1919 int offset = __ offset();
1920 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
1921 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1922 __ end_a_stub();
1923 return offset;
1924 }
1926 uint size_deopt_handler()
1927 {
1928 // three 5 byte instructions
1929 return 15;
1930 }
1932 // Emit deopt handler code.
1933 int emit_deopt_handler(CodeBuffer& cbuf)
1934 {
1936 // Note that the code buffer's inst_mark is always relative to insts.
1937 // That's why we must use the macroassembler to generate a handler.
1938 MacroAssembler _masm(&cbuf);
1939 address base =
1940 __ start_a_stub(size_deopt_handler());
1941 if (base == NULL) return 0; // CodeBuffer::expand failed
1942 int offset = __ offset();
1943 address the_pc = (address) __ pc();
1944 Label next;
1945 // push a "the_pc" on the stack without destroying any registers
1946 // as they all may be live.
1948 // push address of "next"
1949 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1950 __ bind(next);
1951 // adjust it so it matches "the_pc"
1952 __ subq(Address(rsp, 0), __ offset() - offset);
1953 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1954 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1955 __ end_a_stub();
1956 return offset;
1957 }
1959 static void emit_double_constant(CodeBuffer& cbuf, double x) {
1960 int mark = cbuf.insts()->mark_off();
1961 MacroAssembler _masm(&cbuf);
1962 address double_address = __ double_constant(x);
1963 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1964 emit_d32_reloc(cbuf,
1965 (int) (double_address - cbuf.code_end() - 4),
1966 internal_word_Relocation::spec(double_address),
1967 RELOC_DISP32);
1968 }
1970 static void emit_float_constant(CodeBuffer& cbuf, float x) {
1971 int mark = cbuf.insts()->mark_off();
1972 MacroAssembler _masm(&cbuf);
1973 address float_address = __ float_constant(x);
1974 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1975 emit_d32_reloc(cbuf,
1976 (int) (float_address - cbuf.code_end() - 4),
1977 internal_word_Relocation::spec(float_address),
1978 RELOC_DISP32);
1979 }
1982 int Matcher::regnum_to_fpu_offset(int regnum)
1983 {
1984 return regnum - 32; // The FP registers are in the second chunk
1985 }
1987 // This is UltraSparc specific, true just means we have fast l2f conversion
1988 const bool Matcher::convL2FSupported(void) {
1989 return true;
1990 }
1992 // Vector width in bytes
1993 const uint Matcher::vector_width_in_bytes(void) {
1994 return 8;
1995 }
1997 // Vector ideal reg
1998 const uint Matcher::vector_ideal_reg(void) {
1999 return Op_RegD;
2000 }
2002 // Is this branch offset short enough that a short branch can be used?
2003 //
2004 // NOTE: If the platform does not provide any short branch variants, then
2005 // this method should return false for offset 0.
2006 bool Matcher::is_short_branch_offset(int offset)
2007 {
2008 return -0x80 <= offset && offset < 0x80;
2009 }
2011 const bool Matcher::isSimpleConstant64(jlong value) {
2012 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
2013 //return value == (int) value; // Cf. storeImmL and immL32.
2015 // Probably always true, even if a temp register is required.
2016 return true;
2017 }
2019 // The ecx parameter to rep stosq for the ClearArray node is in words.
2020 const bool Matcher::init_array_count_is_in_bytes = false;
2022 // Threshold size for cleararray.
2023 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2025 // Should the Matcher clone shifts on addressing modes, expecting them
2026 // to be subsumed into complex addressing expressions or compute them
2027 // into registers? True for Intel but false for most RISCs
2028 const bool Matcher::clone_shift_expressions = true;
2030 // Is it better to copy float constants, or load them directly from
2031 // memory? Intel can load a float constant from a direct address,
2032 // requiring no extra registers. Most RISCs will have to materialize
2033 // an address into a register first, so they would do better to copy
2034 // the constant from stack.
2035 const bool Matcher::rematerialize_float_constants = true; // XXX
2037 // If CPU can load and store mis-aligned doubles directly then no
2038 // fixup is needed. Else we split the double into 2 integer pieces
2039 // and move it piece-by-piece. Only happens when passing doubles into
2040 // C code as the Java calling convention forces doubles to be aligned.
2041 const bool Matcher::misaligned_doubles_ok = true;
2043 // No-op on amd64
2044 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2046 // Advertise here if the CPU requires explicit rounding operations to
2047 // implement the UseStrictFP mode.
2048 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2050 // Do floats take an entire double register or just half?
2051 const bool Matcher::float_in_double = true;
2052 // Do ints take an entire long register or just half?
2053 const bool Matcher::int_in_long = true;
2055 // Return whether or not this register is ever used as an argument.
2056 // This function is used on startup to build the trampoline stubs in
2057 // generateOptoStub. Registers not mentioned will be killed by the VM
2058 // call in the trampoline, and arguments in those registers not be
2059 // available to the callee.
2060 bool Matcher::can_be_java_arg(int reg)
2061 {
2062 return
2063 reg == RDI_num || reg == RDI_H_num ||
2064 reg == RSI_num || reg == RSI_H_num ||
2065 reg == RDX_num || reg == RDX_H_num ||
2066 reg == RCX_num || reg == RCX_H_num ||
2067 reg == R8_num || reg == R8_H_num ||
2068 reg == R9_num || reg == R9_H_num ||
2069 reg == R12_num || reg == R12_H_num ||
2070 reg == XMM0_num || reg == XMM0_H_num ||
2071 reg == XMM1_num || reg == XMM1_H_num ||
2072 reg == XMM2_num || reg == XMM2_H_num ||
2073 reg == XMM3_num || reg == XMM3_H_num ||
2074 reg == XMM4_num || reg == XMM4_H_num ||
2075 reg == XMM5_num || reg == XMM5_H_num ||
2076 reg == XMM6_num || reg == XMM6_H_num ||
2077 reg == XMM7_num || reg == XMM7_H_num;
2078 }
2080 bool Matcher::is_spillable_arg(int reg)
2081 {
2082 return can_be_java_arg(reg);
2083 }
2085 // Register for DIVI projection of divmodI
2086 RegMask Matcher::divI_proj_mask() {
2087 return INT_RAX_REG_mask;
2088 }
2090 // Register for MODI projection of divmodI
2091 RegMask Matcher::modI_proj_mask() {
2092 return INT_RDX_REG_mask;
2093 }
2095 // Register for DIVL projection of divmodL
2096 RegMask Matcher::divL_proj_mask() {
2097 return LONG_RAX_REG_mask;
2098 }
2100 // Register for MODL projection of divmodL
2101 RegMask Matcher::modL_proj_mask() {
2102 return LONG_RDX_REG_mask;
2103 }
2105 static Address build_address(int b, int i, int s, int d) {
2106 Register index = as_Register(i);
2107 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2108 if (index == rsp) {
2109 index = noreg;
2110 scale = Address::no_scale;
2111 }
2112 Address addr(as_Register(b), index, scale, d);
2113 return addr;
2114 }
2116 %}
2118 //----------ENCODING BLOCK-----------------------------------------------------
2119 // This block specifies the encoding classes used by the compiler to
2120 // output byte streams. Encoding classes are parameterized macros
2121 // used by Machine Instruction Nodes in order to generate the bit
2122 // encoding of the instruction. Operands specify their base encoding
2123 // interface with the interface keyword. There are currently
2124 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2125 // COND_INTER. REG_INTER causes an operand to generate a function
2126 // which returns its register number when queried. CONST_INTER causes
2127 // an operand to generate a function which returns the value of the
2128 // constant when queried. MEMORY_INTER causes an operand to generate
2129 // four functions which return the Base Register, the Index Register,
2130 // the Scale Value, and the Offset Value of the operand when queried.
2131 // COND_INTER causes an operand to generate six functions which return
2132 // the encoding code (ie - encoding bits for the instruction)
2133 // associated with each basic boolean condition for a conditional
2134 // instruction.
2135 //
2136 // Instructions specify two basic values for encoding. Again, a
2137 // function is available to check if the constant displacement is an
2138 // oop. They use the ins_encode keyword to specify their encoding
2139 // classes (which must be a sequence of enc_class names, and their
2140 // parameters, specified in the encoding block), and they use the
2141 // opcode keyword to specify, in order, their primary, secondary, and
2142 // tertiary opcode. Only the opcode sections which a particular
2143 // instruction needs for encoding need to be specified.
2144 encode %{
2145 // Build emit functions for each basic byte or larger field in the
2146 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2147 // from C++ code in the enc_class source block. Emit functions will
2148 // live in the main source block for now. In future, we can
2149 // generalize this by adding a syntax that specifies the sizes of
2150 // fields in an order, so that the adlc can build the emit functions
2151 // automagically
2153 // Emit primary opcode
2154 enc_class OpcP
2155 %{
2156 emit_opcode(cbuf, $primary);
2157 %}
2159 // Emit secondary opcode
2160 enc_class OpcS
2161 %{
2162 emit_opcode(cbuf, $secondary);
2163 %}
2165 // Emit tertiary opcode
2166 enc_class OpcT
2167 %{
2168 emit_opcode(cbuf, $tertiary);
2169 %}
2171 // Emit opcode directly
2172 enc_class Opcode(immI d8)
2173 %{
2174 emit_opcode(cbuf, $d8$$constant);
2175 %}
2177 // Emit size prefix
2178 enc_class SizePrefix
2179 %{
2180 emit_opcode(cbuf, 0x66);
2181 %}
2183 enc_class reg(rRegI reg)
2184 %{
2185 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2186 %}
2188 enc_class reg_reg(rRegI dst, rRegI src)
2189 %{
2190 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2191 %}
2193 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2194 %{
2195 emit_opcode(cbuf, $opcode$$constant);
2196 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2197 %}
2199 enc_class cmpfp_fixup()
2200 %{
2201 // jnp,s exit
2202 emit_opcode(cbuf, 0x7B);
2203 emit_d8(cbuf, 0x0A);
2205 // pushfq
2206 emit_opcode(cbuf, 0x9C);
2208 // andq $0xffffff2b, (%rsp)
2209 emit_opcode(cbuf, Assembler::REX_W);
2210 emit_opcode(cbuf, 0x81);
2211 emit_opcode(cbuf, 0x24);
2212 emit_opcode(cbuf, 0x24);
2213 emit_d32(cbuf, 0xffffff2b);
2215 // popfq
2216 emit_opcode(cbuf, 0x9D);
2218 // nop (target for branch to avoid branch to branch)
2219 emit_opcode(cbuf, 0x90);
2220 %}
2222 enc_class cmpfp3(rRegI dst)
2223 %{
2224 int dstenc = $dst$$reg;
2226 // movl $dst, -1
2227 if (dstenc >= 8) {
2228 emit_opcode(cbuf, Assembler::REX_B);
2229 }
2230 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2231 emit_d32(cbuf, -1);
2233 // jp,s done
2234 emit_opcode(cbuf, 0x7A);
2235 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2237 // jb,s done
2238 emit_opcode(cbuf, 0x72);
2239 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2241 // setne $dst
2242 if (dstenc >= 4) {
2243 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2244 }
2245 emit_opcode(cbuf, 0x0F);
2246 emit_opcode(cbuf, 0x95);
2247 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2249 // movzbl $dst, $dst
2250 if (dstenc >= 4) {
2251 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2252 }
2253 emit_opcode(cbuf, 0x0F);
2254 emit_opcode(cbuf, 0xB6);
2255 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2256 %}
2258 enc_class cdql_enc(no_rax_rdx_RegI div)
2259 %{
2260 // Full implementation of Java idiv and irem; checks for
2261 // special case as described in JVM spec., p.243 & p.271.
2262 //
2263 // normal case special case
2264 //
2265 // input : rax: dividend min_int
2266 // reg: divisor -1
2267 //
2268 // output: rax: quotient (= rax idiv reg) min_int
2269 // rdx: remainder (= rax irem reg) 0
2270 //
2271 // Code sequnce:
2272 //
2273 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2274 // 5: 75 07/08 jne e <normal>
2275 // 7: 33 d2 xor %edx,%edx
2276 // [div >= 8 -> offset + 1]
2277 // [REX_B]
2278 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2279 // c: 74 03/04 je 11 <done>
2280 // 000000000000000e <normal>:
2281 // e: 99 cltd
2282 // [div >= 8 -> offset + 1]
2283 // [REX_B]
2284 // f: f7 f9 idiv $div
2285 // 0000000000000011 <done>:
2287 // cmp $0x80000000,%eax
2288 emit_opcode(cbuf, 0x3d);
2289 emit_d8(cbuf, 0x00);
2290 emit_d8(cbuf, 0x00);
2291 emit_d8(cbuf, 0x00);
2292 emit_d8(cbuf, 0x80);
2294 // jne e <normal>
2295 emit_opcode(cbuf, 0x75);
2296 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2298 // xor %edx,%edx
2299 emit_opcode(cbuf, 0x33);
2300 emit_d8(cbuf, 0xD2);
2302 // cmp $0xffffffffffffffff,%ecx
2303 if ($div$$reg >= 8) {
2304 emit_opcode(cbuf, Assembler::REX_B);
2305 }
2306 emit_opcode(cbuf, 0x83);
2307 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2308 emit_d8(cbuf, 0xFF);
2310 // je 11 <done>
2311 emit_opcode(cbuf, 0x74);
2312 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2314 // <normal>
2315 // cltd
2316 emit_opcode(cbuf, 0x99);
2318 // idivl (note: must be emitted by the user of this rule)
2319 // <done>
2320 %}
2322 enc_class cdqq_enc(no_rax_rdx_RegL div)
2323 %{
2324 // Full implementation of Java ldiv and lrem; checks for
2325 // special case as described in JVM spec., p.243 & p.271.
2326 //
2327 // normal case special case
2328 //
2329 // input : rax: dividend min_long
2330 // reg: divisor -1
2331 //
2332 // output: rax: quotient (= rax idiv reg) min_long
2333 // rdx: remainder (= rax irem reg) 0
2334 //
2335 // Code sequnce:
2336 //
2337 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2338 // 7: 00 00 80
2339 // a: 48 39 d0 cmp %rdx,%rax
2340 // d: 75 08 jne 17 <normal>
2341 // f: 33 d2 xor %edx,%edx
2342 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2343 // 15: 74 05 je 1c <done>
2344 // 0000000000000017 <normal>:
2345 // 17: 48 99 cqto
2346 // 19: 48 f7 f9 idiv $div
2347 // 000000000000001c <done>:
2349 // mov $0x8000000000000000,%rdx
2350 emit_opcode(cbuf, Assembler::REX_W);
2351 emit_opcode(cbuf, 0xBA);
2352 emit_d8(cbuf, 0x00);
2353 emit_d8(cbuf, 0x00);
2354 emit_d8(cbuf, 0x00);
2355 emit_d8(cbuf, 0x00);
2356 emit_d8(cbuf, 0x00);
2357 emit_d8(cbuf, 0x00);
2358 emit_d8(cbuf, 0x00);
2359 emit_d8(cbuf, 0x80);
2361 // cmp %rdx,%rax
2362 emit_opcode(cbuf, Assembler::REX_W);
2363 emit_opcode(cbuf, 0x39);
2364 emit_d8(cbuf, 0xD0);
2366 // jne 17 <normal>
2367 emit_opcode(cbuf, 0x75);
2368 emit_d8(cbuf, 0x08);
2370 // xor %edx,%edx
2371 emit_opcode(cbuf, 0x33);
2372 emit_d8(cbuf, 0xD2);
2374 // cmp $0xffffffffffffffff,$div
2375 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2376 emit_opcode(cbuf, 0x83);
2377 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2378 emit_d8(cbuf, 0xFF);
2380 // je 1e <done>
2381 emit_opcode(cbuf, 0x74);
2382 emit_d8(cbuf, 0x05);
2384 // <normal>
2385 // cqto
2386 emit_opcode(cbuf, Assembler::REX_W);
2387 emit_opcode(cbuf, 0x99);
2389 // idivq (note: must be emitted by the user of this rule)
2390 // <done>
2391 %}
2393 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2394 enc_class OpcSE(immI imm)
2395 %{
2396 // Emit primary opcode and set sign-extend bit
2397 // Check for 8-bit immediate, and set sign extend bit in opcode
2398 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2399 emit_opcode(cbuf, $primary | 0x02);
2400 } else {
2401 // 32-bit immediate
2402 emit_opcode(cbuf, $primary);
2403 }
2404 %}
2406 enc_class OpcSErm(rRegI dst, immI imm)
2407 %{
2408 // OpcSEr/m
2409 int dstenc = $dst$$reg;
2410 if (dstenc >= 8) {
2411 emit_opcode(cbuf, Assembler::REX_B);
2412 dstenc -= 8;
2413 }
2414 // Emit primary opcode and set sign-extend bit
2415 // Check for 8-bit immediate, and set sign extend bit in opcode
2416 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2417 emit_opcode(cbuf, $primary | 0x02);
2418 } else {
2419 // 32-bit immediate
2420 emit_opcode(cbuf, $primary);
2421 }
2422 // Emit r/m byte with secondary opcode, after primary opcode.
2423 emit_rm(cbuf, 0x3, $secondary, dstenc);
2424 %}
2426 enc_class OpcSErm_wide(rRegL dst, immI imm)
2427 %{
2428 // OpcSEr/m
2429 int dstenc = $dst$$reg;
2430 if (dstenc < 8) {
2431 emit_opcode(cbuf, Assembler::REX_W);
2432 } else {
2433 emit_opcode(cbuf, Assembler::REX_WB);
2434 dstenc -= 8;
2435 }
2436 // Emit primary opcode and set sign-extend bit
2437 // Check for 8-bit immediate, and set sign extend bit in opcode
2438 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2439 emit_opcode(cbuf, $primary | 0x02);
2440 } else {
2441 // 32-bit immediate
2442 emit_opcode(cbuf, $primary);
2443 }
2444 // Emit r/m byte with secondary opcode, after primary opcode.
2445 emit_rm(cbuf, 0x3, $secondary, dstenc);
2446 %}
2448 enc_class Con8or32(immI imm)
2449 %{
2450 // Check for 8-bit immediate, and set sign extend bit in opcode
2451 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2452 $$$emit8$imm$$constant;
2453 } else {
2454 // 32-bit immediate
2455 $$$emit32$imm$$constant;
2456 }
2457 %}
2459 enc_class Lbl(label labl)
2460 %{
2461 // JMP, CALL
2462 Label* l = $labl$$label;
2463 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2464 %}
2466 enc_class LblShort(label labl)
2467 %{
2468 // JMP, CALL
2469 Label* l = $labl$$label;
2470 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2471 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2472 emit_d8(cbuf, disp);
2473 %}
2475 enc_class opc2_reg(rRegI dst)
2476 %{
2477 // BSWAP
2478 emit_cc(cbuf, $secondary, $dst$$reg);
2479 %}
2481 enc_class opc3_reg(rRegI dst)
2482 %{
2483 // BSWAP
2484 emit_cc(cbuf, $tertiary, $dst$$reg);
2485 %}
2487 enc_class reg_opc(rRegI div)
2488 %{
2489 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2490 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2491 %}
2493 enc_class Jcc(cmpOp cop, label labl)
2494 %{
2495 // JCC
2496 Label* l = $labl$$label;
2497 $$$emit8$primary;
2498 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2499 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2500 %}
2502 enc_class JccShort (cmpOp cop, label labl)
2503 %{
2504 // JCC
2505 Label *l = $labl$$label;
2506 emit_cc(cbuf, $primary, $cop$$cmpcode);
2507 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2508 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2509 emit_d8(cbuf, disp);
2510 %}
2512 enc_class enc_cmov(cmpOp cop)
2513 %{
2514 // CMOV
2515 $$$emit8$primary;
2516 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2517 %}
2519 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2520 %{
2521 // Invert sense of branch from sense of cmov
2522 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2523 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2524 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2525 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2526 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2527 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2528 if ($dst$$reg < 8) {
2529 if ($src$$reg >= 8) {
2530 emit_opcode(cbuf, Assembler::REX_B);
2531 }
2532 } else {
2533 if ($src$$reg < 8) {
2534 emit_opcode(cbuf, Assembler::REX_R);
2535 } else {
2536 emit_opcode(cbuf, Assembler::REX_RB);
2537 }
2538 }
2539 emit_opcode(cbuf, 0x0F);
2540 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2541 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2542 %}
2544 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2545 %{
2546 // Invert sense of branch from sense of cmov
2547 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2548 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2550 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2551 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2552 if ($dst$$reg < 8) {
2553 if ($src$$reg >= 8) {
2554 emit_opcode(cbuf, Assembler::REX_B);
2555 }
2556 } else {
2557 if ($src$$reg < 8) {
2558 emit_opcode(cbuf, Assembler::REX_R);
2559 } else {
2560 emit_opcode(cbuf, Assembler::REX_RB);
2561 }
2562 }
2563 emit_opcode(cbuf, 0x0F);
2564 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2565 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2566 %}
2568 enc_class enc_PartialSubtypeCheck()
2569 %{
2570 Register Rrdi = as_Register(RDI_enc); // result register
2571 Register Rrax = as_Register(RAX_enc); // super class
2572 Register Rrcx = as_Register(RCX_enc); // killed
2573 Register Rrsi = as_Register(RSI_enc); // sub class
2574 Label hit, miss, cmiss;
2576 MacroAssembler _masm(&cbuf);
2577 // Compare super with sub directly, since super is not in its own SSA.
2578 // The compiler used to emit this test, but we fold it in here,
2579 // to allow platform-specific tweaking on sparc.
2580 __ cmpq(Rrax, Rrsi);
2581 __ jcc(Assembler::equal, hit);
2582 #ifndef PRODUCT
2583 __ lea(Rrcx, ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
2584 __ incrementl(Address(Rrcx, 0));
2585 #endif //PRODUCT
2586 __ movq(Rrdi, Address(Rrsi,
2587 sizeof(oopDesc) +
2588 Klass::secondary_supers_offset_in_bytes()));
2589 __ movl(Rrcx, Address(Rrdi, arrayOopDesc::length_offset_in_bytes()));
2590 __ addq(Rrdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
2591 if (UseCompressedOops) {
2592 __ encode_heap_oop(Rrax);
2593 __ repne_scanl();
2594 __ jcc(Assembler::notEqual, cmiss);
2595 __ decode_heap_oop(Rrax);
2596 __ movq(Address(Rrsi,
2597 sizeof(oopDesc) +
2598 Klass::secondary_super_cache_offset_in_bytes()),
2599 Rrax);
2600 __ jmp(hit);
2601 __ bind(cmiss);
2602 __ decode_heap_oop(Rrax);
2603 __ jmp(miss);
2604 } else {
2605 __ repne_scanq();
2606 __ jcc(Assembler::notEqual, miss);
2607 __ movq(Address(Rrsi,
2608 sizeof(oopDesc) +
2609 Klass::secondary_super_cache_offset_in_bytes()),
2610 Rrax);
2611 }
2612 __ bind(hit);
2613 if ($primary) {
2614 __ xorq(Rrdi, Rrdi);
2615 }
2616 __ bind(miss);
2617 %}
2619 enc_class Java_To_Interpreter(method meth)
2620 %{
2621 // CALL Java_To_Interpreter
2622 // This is the instruction starting address for relocation info.
2623 cbuf.set_inst_mark();
2624 $$$emit8$primary;
2625 // CALL directly to the runtime
2626 emit_d32_reloc(cbuf,
2627 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2628 runtime_call_Relocation::spec(),
2629 RELOC_DISP32);
2630 %}
2632 enc_class Java_Static_Call(method meth)
2633 %{
2634 // JAVA STATIC CALL
2635 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2636 // determine who we intended to call.
2637 cbuf.set_inst_mark();
2638 $$$emit8$primary;
2640 if (!_method) {
2641 emit_d32_reloc(cbuf,
2642 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2643 runtime_call_Relocation::spec(),
2644 RELOC_DISP32);
2645 } else if (_optimized_virtual) {
2646 emit_d32_reloc(cbuf,
2647 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2648 opt_virtual_call_Relocation::spec(),
2649 RELOC_DISP32);
2650 } else {
2651 emit_d32_reloc(cbuf,
2652 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2653 static_call_Relocation::spec(),
2654 RELOC_DISP32);
2655 }
2656 if (_method) {
2657 // Emit stub for static call
2658 emit_java_to_interp(cbuf);
2659 }
2660 %}
2662 enc_class Java_Dynamic_Call(method meth)
2663 %{
2664 // JAVA DYNAMIC CALL
2665 // !!!!!
2666 // Generate "movq rax, -1", placeholder instruction to load oop-info
2667 // emit_call_dynamic_prologue( cbuf );
2668 cbuf.set_inst_mark();
2670 // movq rax, -1
2671 emit_opcode(cbuf, Assembler::REX_W);
2672 emit_opcode(cbuf, 0xB8 | RAX_enc);
2673 emit_d64_reloc(cbuf,
2674 (int64_t) Universe::non_oop_word(),
2675 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2676 address virtual_call_oop_addr = cbuf.inst_mark();
2677 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2678 // who we intended to call.
2679 cbuf.set_inst_mark();
2680 $$$emit8$primary;
2681 emit_d32_reloc(cbuf,
2682 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2683 virtual_call_Relocation::spec(virtual_call_oop_addr),
2684 RELOC_DISP32);
2685 %}
2687 enc_class Java_Compiled_Call(method meth)
2688 %{
2689 // JAVA COMPILED CALL
2690 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2692 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2693 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2695 // callq *disp(%rax)
2696 cbuf.set_inst_mark();
2697 $$$emit8$primary;
2698 if (disp < 0x80) {
2699 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2700 emit_d8(cbuf, disp); // Displacement
2701 } else {
2702 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2703 emit_d32(cbuf, disp); // Displacement
2704 }
2705 %}
2707 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2708 %{
2709 // SAL, SAR, SHR
2710 int dstenc = $dst$$reg;
2711 if (dstenc >= 8) {
2712 emit_opcode(cbuf, Assembler::REX_B);
2713 dstenc -= 8;
2714 }
2715 $$$emit8$primary;
2716 emit_rm(cbuf, 0x3, $secondary, dstenc);
2717 $$$emit8$shift$$constant;
2718 %}
2720 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2721 %{
2722 // SAL, SAR, SHR
2723 int dstenc = $dst$$reg;
2724 if (dstenc < 8) {
2725 emit_opcode(cbuf, Assembler::REX_W);
2726 } else {
2727 emit_opcode(cbuf, Assembler::REX_WB);
2728 dstenc -= 8;
2729 }
2730 $$$emit8$primary;
2731 emit_rm(cbuf, 0x3, $secondary, dstenc);
2732 $$$emit8$shift$$constant;
2733 %}
2735 enc_class load_immI(rRegI dst, immI src)
2736 %{
2737 int dstenc = $dst$$reg;
2738 if (dstenc >= 8) {
2739 emit_opcode(cbuf, Assembler::REX_B);
2740 dstenc -= 8;
2741 }
2742 emit_opcode(cbuf, 0xB8 | dstenc);
2743 $$$emit32$src$$constant;
2744 %}
2746 enc_class load_immL(rRegL dst, immL src)
2747 %{
2748 int dstenc = $dst$$reg;
2749 if (dstenc < 8) {
2750 emit_opcode(cbuf, Assembler::REX_W);
2751 } else {
2752 emit_opcode(cbuf, Assembler::REX_WB);
2753 dstenc -= 8;
2754 }
2755 emit_opcode(cbuf, 0xB8 | dstenc);
2756 emit_d64(cbuf, $src$$constant);
2757 %}
2759 enc_class load_immUL32(rRegL dst, immUL32 src)
2760 %{
2761 // same as load_immI, but this time we care about zeroes in the high word
2762 int dstenc = $dst$$reg;
2763 if (dstenc >= 8) {
2764 emit_opcode(cbuf, Assembler::REX_B);
2765 dstenc -= 8;
2766 }
2767 emit_opcode(cbuf, 0xB8 | dstenc);
2768 $$$emit32$src$$constant;
2769 %}
2771 enc_class load_immL32(rRegL dst, immL32 src)
2772 %{
2773 int dstenc = $dst$$reg;
2774 if (dstenc < 8) {
2775 emit_opcode(cbuf, Assembler::REX_W);
2776 } else {
2777 emit_opcode(cbuf, Assembler::REX_WB);
2778 dstenc -= 8;
2779 }
2780 emit_opcode(cbuf, 0xC7);
2781 emit_rm(cbuf, 0x03, 0x00, dstenc);
2782 $$$emit32$src$$constant;
2783 %}
2785 enc_class load_immP31(rRegP dst, immP32 src)
2786 %{
2787 // same as load_immI, but this time we care about zeroes in the high word
2788 int dstenc = $dst$$reg;
2789 if (dstenc >= 8) {
2790 emit_opcode(cbuf, Assembler::REX_B);
2791 dstenc -= 8;
2792 }
2793 emit_opcode(cbuf, 0xB8 | dstenc);
2794 $$$emit32$src$$constant;
2795 %}
2797 enc_class load_immP(rRegP dst, immP src)
2798 %{
2799 int dstenc = $dst$$reg;
2800 if (dstenc < 8) {
2801 emit_opcode(cbuf, Assembler::REX_W);
2802 } else {
2803 emit_opcode(cbuf, Assembler::REX_WB);
2804 dstenc -= 8;
2805 }
2806 emit_opcode(cbuf, 0xB8 | dstenc);
2807 // This next line should be generated from ADLC
2808 if ($src->constant_is_oop()) {
2809 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2810 } else {
2811 emit_d64(cbuf, $src$$constant);
2812 }
2813 %}
2815 enc_class load_immF(regF dst, immF con)
2816 %{
2817 // XXX reg_mem doesn't support RIP-relative addressing yet
2818 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2819 emit_float_constant(cbuf, $con$$constant);
2820 %}
2822 enc_class load_immD(regD dst, immD con)
2823 %{
2824 // XXX reg_mem doesn't support RIP-relative addressing yet
2825 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2826 emit_double_constant(cbuf, $con$$constant);
2827 %}
2829 enc_class load_conF (regF dst, immF con) %{ // Load float constant
2830 emit_opcode(cbuf, 0xF3);
2831 if ($dst$$reg >= 8) {
2832 emit_opcode(cbuf, Assembler::REX_R);
2833 }
2834 emit_opcode(cbuf, 0x0F);
2835 emit_opcode(cbuf, 0x10);
2836 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2837 emit_float_constant(cbuf, $con$$constant);
2838 %}
2840 enc_class load_conD (regD dst, immD con) %{ // Load double constant
2841 // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
2842 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
2843 if ($dst$$reg >= 8) {
2844 emit_opcode(cbuf, Assembler::REX_R);
2845 }
2846 emit_opcode(cbuf, 0x0F);
2847 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
2848 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2849 emit_double_constant(cbuf, $con$$constant);
2850 %}
2852 // Encode a reg-reg copy. If it is useless, then empty encoding.
2853 enc_class enc_copy(rRegI dst, rRegI src)
2854 %{
2855 encode_copy(cbuf, $dst$$reg, $src$$reg);
2856 %}
2858 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2859 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2860 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2861 %}
2863 enc_class enc_copy_always(rRegI dst, rRegI src)
2864 %{
2865 int srcenc = $src$$reg;
2866 int dstenc = $dst$$reg;
2868 if (dstenc < 8) {
2869 if (srcenc >= 8) {
2870 emit_opcode(cbuf, Assembler::REX_B);
2871 srcenc -= 8;
2872 }
2873 } else {
2874 if (srcenc < 8) {
2875 emit_opcode(cbuf, Assembler::REX_R);
2876 } else {
2877 emit_opcode(cbuf, Assembler::REX_RB);
2878 srcenc -= 8;
2879 }
2880 dstenc -= 8;
2881 }
2883 emit_opcode(cbuf, 0x8B);
2884 emit_rm(cbuf, 0x3, dstenc, srcenc);
2885 %}
2887 enc_class enc_copy_wide(rRegL dst, rRegL src)
2888 %{
2889 int srcenc = $src$$reg;
2890 int dstenc = $dst$$reg;
2892 if (dstenc != srcenc) {
2893 if (dstenc < 8) {
2894 if (srcenc < 8) {
2895 emit_opcode(cbuf, Assembler::REX_W);
2896 } else {
2897 emit_opcode(cbuf, Assembler::REX_WB);
2898 srcenc -= 8;
2899 }
2900 } else {
2901 if (srcenc < 8) {
2902 emit_opcode(cbuf, Assembler::REX_WR);
2903 } else {
2904 emit_opcode(cbuf, Assembler::REX_WRB);
2905 srcenc -= 8;
2906 }
2907 dstenc -= 8;
2908 }
2909 emit_opcode(cbuf, 0x8B);
2910 emit_rm(cbuf, 0x3, dstenc, srcenc);
2911 }
2912 %}
2914 enc_class Con32(immI src)
2915 %{
2916 // Output immediate
2917 $$$emit32$src$$constant;
2918 %}
2920 enc_class Con64(immL src)
2921 %{
2922 // Output immediate
2923 emit_d64($src$$constant);
2924 %}
2926 enc_class Con32F_as_bits(immF src)
2927 %{
2928 // Output Float immediate bits
2929 jfloat jf = $src$$constant;
2930 jint jf_as_bits = jint_cast(jf);
2931 emit_d32(cbuf, jf_as_bits);
2932 %}
2934 enc_class Con16(immI src)
2935 %{
2936 // Output immediate
2937 $$$emit16$src$$constant;
2938 %}
2940 // How is this different from Con32??? XXX
2941 enc_class Con_d32(immI src)
2942 %{
2943 emit_d32(cbuf,$src$$constant);
2944 %}
2946 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2947 // Output immediate memory reference
2948 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2949 emit_d32(cbuf, 0x00);
2950 %}
2952 enc_class jump_enc(rRegL switch_val, rRegI dest) %{
2953 MacroAssembler masm(&cbuf);
2955 Register switch_reg = as_Register($switch_val$$reg);
2956 Register dest_reg = as_Register($dest$$reg);
2957 address table_base = masm.address_table_constant(_index2label);
2959 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2960 // to do that and the compiler is using that register as one it can allocate.
2961 // So we build it all by hand.
2962 // Address index(noreg, switch_reg, Address::times_1);
2963 // ArrayAddress dispatch(table, index);
2965 Address dispatch(dest_reg, switch_reg, Address::times_1);
2967 masm.lea(dest_reg, InternalAddress(table_base));
2968 masm.jmp(dispatch);
2969 %}
2971 enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
2972 MacroAssembler masm(&cbuf);
2974 Register switch_reg = as_Register($switch_val$$reg);
2975 Register dest_reg = as_Register($dest$$reg);
2976 address table_base = masm.address_table_constant(_index2label);
2978 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2979 // to do that and the compiler is using that register as one it can allocate.
2980 // So we build it all by hand.
2981 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2982 // ArrayAddress dispatch(table, index);
2984 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2986 masm.lea(dest_reg, InternalAddress(table_base));
2987 masm.jmp(dispatch);
2988 %}
2990 enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
2991 MacroAssembler masm(&cbuf);
2993 Register switch_reg = as_Register($switch_val$$reg);
2994 Register dest_reg = as_Register($dest$$reg);
2995 address table_base = masm.address_table_constant(_index2label);
2997 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2998 // to do that and the compiler is using that register as one it can allocate.
2999 // So we build it all by hand.
3000 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3001 // ArrayAddress dispatch(table, index);
3003 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3004 masm.lea(dest_reg, InternalAddress(table_base));
3005 masm.jmp(dispatch);
3007 %}
3009 enc_class lock_prefix()
3010 %{
3011 if (os::is_MP()) {
3012 emit_opcode(cbuf, 0xF0); // lock
3013 }
3014 %}
3016 enc_class REX_mem(memory mem)
3017 %{
3018 if ($mem$$base >= 8) {
3019 if ($mem$$index < 8) {
3020 emit_opcode(cbuf, Assembler::REX_B);
3021 } else {
3022 emit_opcode(cbuf, Assembler::REX_XB);
3023 }
3024 } else {
3025 if ($mem$$index >= 8) {
3026 emit_opcode(cbuf, Assembler::REX_X);
3027 }
3028 }
3029 %}
3031 enc_class REX_mem_wide(memory mem)
3032 %{
3033 if ($mem$$base >= 8) {
3034 if ($mem$$index < 8) {
3035 emit_opcode(cbuf, Assembler::REX_WB);
3036 } else {
3037 emit_opcode(cbuf, Assembler::REX_WXB);
3038 }
3039 } else {
3040 if ($mem$$index < 8) {
3041 emit_opcode(cbuf, Assembler::REX_W);
3042 } else {
3043 emit_opcode(cbuf, Assembler::REX_WX);
3044 }
3045 }
3046 %}
3048 // for byte regs
3049 enc_class REX_breg(rRegI reg)
3050 %{
3051 if ($reg$$reg >= 4) {
3052 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3053 }
3054 %}
3056 // for byte regs
3057 enc_class REX_reg_breg(rRegI dst, rRegI src)
3058 %{
3059 if ($dst$$reg < 8) {
3060 if ($src$$reg >= 4) {
3061 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3062 }
3063 } else {
3064 if ($src$$reg < 8) {
3065 emit_opcode(cbuf, Assembler::REX_R);
3066 } else {
3067 emit_opcode(cbuf, Assembler::REX_RB);
3068 }
3069 }
3070 %}
3072 // for byte regs
3073 enc_class REX_breg_mem(rRegI reg, memory mem)
3074 %{
3075 if ($reg$$reg < 8) {
3076 if ($mem$$base < 8) {
3077 if ($mem$$index >= 8) {
3078 emit_opcode(cbuf, Assembler::REX_X);
3079 } else if ($reg$$reg >= 4) {
3080 emit_opcode(cbuf, Assembler::REX);
3081 }
3082 } else {
3083 if ($mem$$index < 8) {
3084 emit_opcode(cbuf, Assembler::REX_B);
3085 } else {
3086 emit_opcode(cbuf, Assembler::REX_XB);
3087 }
3088 }
3089 } else {
3090 if ($mem$$base < 8) {
3091 if ($mem$$index < 8) {
3092 emit_opcode(cbuf, Assembler::REX_R);
3093 } else {
3094 emit_opcode(cbuf, Assembler::REX_RX);
3095 }
3096 } else {
3097 if ($mem$$index < 8) {
3098 emit_opcode(cbuf, Assembler::REX_RB);
3099 } else {
3100 emit_opcode(cbuf, Assembler::REX_RXB);
3101 }
3102 }
3103 }
3104 %}
3106 enc_class REX_reg(rRegI reg)
3107 %{
3108 if ($reg$$reg >= 8) {
3109 emit_opcode(cbuf, Assembler::REX_B);
3110 }
3111 %}
3113 enc_class REX_reg_wide(rRegI reg)
3114 %{
3115 if ($reg$$reg < 8) {
3116 emit_opcode(cbuf, Assembler::REX_W);
3117 } else {
3118 emit_opcode(cbuf, Assembler::REX_WB);
3119 }
3120 %}
3122 enc_class REX_reg_reg(rRegI dst, rRegI src)
3123 %{
3124 if ($dst$$reg < 8) {
3125 if ($src$$reg >= 8) {
3126 emit_opcode(cbuf, Assembler::REX_B);
3127 }
3128 } else {
3129 if ($src$$reg < 8) {
3130 emit_opcode(cbuf, Assembler::REX_R);
3131 } else {
3132 emit_opcode(cbuf, Assembler::REX_RB);
3133 }
3134 }
3135 %}
3137 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3138 %{
3139 if ($dst$$reg < 8) {
3140 if ($src$$reg < 8) {
3141 emit_opcode(cbuf, Assembler::REX_W);
3142 } else {
3143 emit_opcode(cbuf, Assembler::REX_WB);
3144 }
3145 } else {
3146 if ($src$$reg < 8) {
3147 emit_opcode(cbuf, Assembler::REX_WR);
3148 } else {
3149 emit_opcode(cbuf, Assembler::REX_WRB);
3150 }
3151 }
3152 %}
3154 enc_class REX_reg_mem(rRegI reg, memory mem)
3155 %{
3156 if ($reg$$reg < 8) {
3157 if ($mem$$base < 8) {
3158 if ($mem$$index >= 8) {
3159 emit_opcode(cbuf, Assembler::REX_X);
3160 }
3161 } else {
3162 if ($mem$$index < 8) {
3163 emit_opcode(cbuf, Assembler::REX_B);
3164 } else {
3165 emit_opcode(cbuf, Assembler::REX_XB);
3166 }
3167 }
3168 } else {
3169 if ($mem$$base < 8) {
3170 if ($mem$$index < 8) {
3171 emit_opcode(cbuf, Assembler::REX_R);
3172 } else {
3173 emit_opcode(cbuf, Assembler::REX_RX);
3174 }
3175 } else {
3176 if ($mem$$index < 8) {
3177 emit_opcode(cbuf, Assembler::REX_RB);
3178 } else {
3179 emit_opcode(cbuf, Assembler::REX_RXB);
3180 }
3181 }
3182 }
3183 %}
3185 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3186 %{
3187 if ($reg$$reg < 8) {
3188 if ($mem$$base < 8) {
3189 if ($mem$$index < 8) {
3190 emit_opcode(cbuf, Assembler::REX_W);
3191 } else {
3192 emit_opcode(cbuf, Assembler::REX_WX);
3193 }
3194 } else {
3195 if ($mem$$index < 8) {
3196 emit_opcode(cbuf, Assembler::REX_WB);
3197 } else {
3198 emit_opcode(cbuf, Assembler::REX_WXB);
3199 }
3200 }
3201 } else {
3202 if ($mem$$base < 8) {
3203 if ($mem$$index < 8) {
3204 emit_opcode(cbuf, Assembler::REX_WR);
3205 } else {
3206 emit_opcode(cbuf, Assembler::REX_WRX);
3207 }
3208 } else {
3209 if ($mem$$index < 8) {
3210 emit_opcode(cbuf, Assembler::REX_WRB);
3211 } else {
3212 emit_opcode(cbuf, Assembler::REX_WRXB);
3213 }
3214 }
3215 }
3216 %}
3218 enc_class reg_mem(rRegI ereg, memory mem)
3219 %{
3220 // High registers handle in encode_RegMem
3221 int reg = $ereg$$reg;
3222 int base = $mem$$base;
3223 int index = $mem$$index;
3224 int scale = $mem$$scale;
3225 int disp = $mem$$disp;
3226 bool disp_is_oop = $mem->disp_is_oop();
3228 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3229 %}
3231 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3232 %{
3233 int rm_byte_opcode = $rm_opcode$$constant;
3235 // High registers handle in encode_RegMem
3236 int base = $mem$$base;
3237 int index = $mem$$index;
3238 int scale = $mem$$scale;
3239 int displace = $mem$$disp;
3241 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3242 // working with static
3243 // globals
3244 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3245 disp_is_oop);
3246 %}
3248 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3249 %{
3250 int reg_encoding = $dst$$reg;
3251 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3252 int index = 0x04; // 0x04 indicates no index
3253 int scale = 0x00; // 0x00 indicates no scale
3254 int displace = $src1$$constant; // 0x00 indicates no displacement
3255 bool disp_is_oop = false;
3256 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3257 disp_is_oop);
3258 %}
3260 enc_class neg_reg(rRegI dst)
3261 %{
3262 int dstenc = $dst$$reg;
3263 if (dstenc >= 8) {
3264 emit_opcode(cbuf, Assembler::REX_B);
3265 dstenc -= 8;
3266 }
3267 // NEG $dst
3268 emit_opcode(cbuf, 0xF7);
3269 emit_rm(cbuf, 0x3, 0x03, dstenc);
3270 %}
3272 enc_class neg_reg_wide(rRegI dst)
3273 %{
3274 int dstenc = $dst$$reg;
3275 if (dstenc < 8) {
3276 emit_opcode(cbuf, Assembler::REX_W);
3277 } else {
3278 emit_opcode(cbuf, Assembler::REX_WB);
3279 dstenc -= 8;
3280 }
3281 // NEG $dst
3282 emit_opcode(cbuf, 0xF7);
3283 emit_rm(cbuf, 0x3, 0x03, dstenc);
3284 %}
3286 enc_class setLT_reg(rRegI dst)
3287 %{
3288 int dstenc = $dst$$reg;
3289 if (dstenc >= 8) {
3290 emit_opcode(cbuf, Assembler::REX_B);
3291 dstenc -= 8;
3292 } else if (dstenc >= 4) {
3293 emit_opcode(cbuf, Assembler::REX);
3294 }
3295 // SETLT $dst
3296 emit_opcode(cbuf, 0x0F);
3297 emit_opcode(cbuf, 0x9C);
3298 emit_rm(cbuf, 0x3, 0x0, dstenc);
3299 %}
3301 enc_class setNZ_reg(rRegI dst)
3302 %{
3303 int dstenc = $dst$$reg;
3304 if (dstenc >= 8) {
3305 emit_opcode(cbuf, Assembler::REX_B);
3306 dstenc -= 8;
3307 } else if (dstenc >= 4) {
3308 emit_opcode(cbuf, Assembler::REX);
3309 }
3310 // SETNZ $dst
3311 emit_opcode(cbuf, 0x0F);
3312 emit_opcode(cbuf, 0x95);
3313 emit_rm(cbuf, 0x3, 0x0, dstenc);
3314 %}
3316 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3317 rcx_RegI tmp)
3318 %{
3319 // cadd_cmpLT
3321 int tmpReg = $tmp$$reg;
3323 int penc = $p$$reg;
3324 int qenc = $q$$reg;
3325 int yenc = $y$$reg;
3327 // subl $p,$q
3328 if (penc < 8) {
3329 if (qenc >= 8) {
3330 emit_opcode(cbuf, Assembler::REX_B);
3331 }
3332 } else {
3333 if (qenc < 8) {
3334 emit_opcode(cbuf, Assembler::REX_R);
3335 } else {
3336 emit_opcode(cbuf, Assembler::REX_RB);
3337 }
3338 }
3339 emit_opcode(cbuf, 0x2B);
3340 emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3342 // sbbl $tmp, $tmp
3343 emit_opcode(cbuf, 0x1B);
3344 emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3346 // andl $tmp, $y
3347 if (yenc >= 8) {
3348 emit_opcode(cbuf, Assembler::REX_B);
3349 }
3350 emit_opcode(cbuf, 0x23);
3351 emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3353 // addl $p,$tmp
3354 if (penc >= 8) {
3355 emit_opcode(cbuf, Assembler::REX_R);
3356 }
3357 emit_opcode(cbuf, 0x03);
3358 emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3359 %}
3361 // Compare the lonogs and set -1, 0, or 1 into dst
3362 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3363 %{
3364 int src1enc = $src1$$reg;
3365 int src2enc = $src2$$reg;
3366 int dstenc = $dst$$reg;
3368 // cmpq $src1, $src2
3369 if (src1enc < 8) {
3370 if (src2enc < 8) {
3371 emit_opcode(cbuf, Assembler::REX_W);
3372 } else {
3373 emit_opcode(cbuf, Assembler::REX_WB);
3374 }
3375 } else {
3376 if (src2enc < 8) {
3377 emit_opcode(cbuf, Assembler::REX_WR);
3378 } else {
3379 emit_opcode(cbuf, Assembler::REX_WRB);
3380 }
3381 }
3382 emit_opcode(cbuf, 0x3B);
3383 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3385 // movl $dst, -1
3386 if (dstenc >= 8) {
3387 emit_opcode(cbuf, Assembler::REX_B);
3388 }
3389 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3390 emit_d32(cbuf, -1);
3392 // jl,s done
3393 emit_opcode(cbuf, 0x7C);
3394 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3396 // setne $dst
3397 if (dstenc >= 4) {
3398 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3399 }
3400 emit_opcode(cbuf, 0x0F);
3401 emit_opcode(cbuf, 0x95);
3402 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3404 // movzbl $dst, $dst
3405 if (dstenc >= 4) {
3406 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3407 }
3408 emit_opcode(cbuf, 0x0F);
3409 emit_opcode(cbuf, 0xB6);
3410 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3411 %}
3413 enc_class Push_ResultXD(regD dst) %{
3414 int dstenc = $dst$$reg;
3416 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3418 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3419 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3420 if (dstenc >= 8) {
3421 emit_opcode(cbuf, Assembler::REX_R);
3422 }
3423 emit_opcode (cbuf, 0x0F );
3424 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3425 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3427 // add rsp,8
3428 emit_opcode(cbuf, Assembler::REX_W);
3429 emit_opcode(cbuf,0x83);
3430 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3431 emit_d8(cbuf,0x08);
3432 %}
3434 enc_class Push_SrcXD(regD src) %{
3435 int srcenc = $src$$reg;
3437 // subq rsp,#8
3438 emit_opcode(cbuf, Assembler::REX_W);
3439 emit_opcode(cbuf, 0x83);
3440 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3441 emit_d8(cbuf, 0x8);
3443 // movsd [rsp],src
3444 emit_opcode(cbuf, 0xF2);
3445 if (srcenc >= 8) {
3446 emit_opcode(cbuf, Assembler::REX_R);
3447 }
3448 emit_opcode(cbuf, 0x0F);
3449 emit_opcode(cbuf, 0x11);
3450 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3452 // fldd [rsp]
3453 emit_opcode(cbuf, 0x66);
3454 emit_opcode(cbuf, 0xDD);
3455 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3456 %}
3459 enc_class movq_ld(regD dst, memory mem) %{
3460 MacroAssembler _masm(&cbuf);
3461 Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3462 __ movq(as_XMMRegister($dst$$reg), madr);
3463 %}
3465 enc_class movq_st(memory mem, regD src) %{
3466 MacroAssembler _masm(&cbuf);
3467 Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3468 __ movq(madr, as_XMMRegister($src$$reg));
3469 %}
3471 enc_class pshufd_8x8(regF dst, regF src) %{
3472 MacroAssembler _masm(&cbuf);
3474 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3475 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3476 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3477 %}
3479 enc_class pshufd_4x16(regF dst, regF src) %{
3480 MacroAssembler _masm(&cbuf);
3482 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3483 %}
3485 enc_class pshufd(regD dst, regD src, int mode) %{
3486 MacroAssembler _masm(&cbuf);
3488 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3489 %}
3491 enc_class pxor(regD dst, regD src) %{
3492 MacroAssembler _masm(&cbuf);
3494 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3495 %}
3497 enc_class mov_i2x(regD dst, rRegI src) %{
3498 MacroAssembler _masm(&cbuf);
3500 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3501 %}
3503 // obj: object to lock
3504 // box: box address (header location) -- killed
3505 // tmp: rax -- killed
3506 // scr: rbx -- killed
3507 //
3508 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3509 // from i486.ad. See that file for comments.
3510 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3511 // use the shorter encoding. (Movl clears the high-order 32-bits).
3514 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3515 %{
3516 Register objReg = as_Register((int)$obj$$reg);
3517 Register boxReg = as_Register((int)$box$$reg);
3518 Register tmpReg = as_Register($tmp$$reg);
3519 Register scrReg = as_Register($scr$$reg);
3520 MacroAssembler masm(&cbuf);
3522 // Verify uniqueness of register assignments -- necessary but not sufficient
3523 assert (objReg != boxReg && objReg != tmpReg &&
3524 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3526 if (_counters != NULL) {
3527 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3528 }
3529 if (EmitSync & 1) {
3530 masm.movptr (Address(boxReg, 0), intptr_t(markOopDesc::unused_mark())) ;
3531 masm.cmpq (rsp, 0) ;
3532 } else
3533 if (EmitSync & 2) {
3534 Label DONE_LABEL;
3535 if (UseBiasedLocking) {
3536 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3537 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3538 }
3539 masm.movl(tmpReg, 0x1);
3540 masm.orq(tmpReg, Address(objReg, 0));
3541 masm.movq(Address(boxReg, 0), tmpReg);
3542 if (os::is_MP()) {
3543 masm.lock();
3544 }
3545 masm.cmpxchgq(boxReg, Address(objReg, 0)); // Updates tmpReg
3546 masm.jcc(Assembler::equal, DONE_LABEL);
3548 // Recursive locking
3549 masm.subq(tmpReg, rsp);
3550 masm.andq(tmpReg, 7 - os::vm_page_size());
3551 masm.movq(Address(boxReg, 0), tmpReg);
3553 masm.bind(DONE_LABEL);
3554 masm.nop(); // avoid branch to branch
3555 } else {
3556 Label DONE_LABEL, IsInflated, Egress;
3558 masm.movq (tmpReg, Address(objReg, 0)) ;
3559 masm.testq (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3560 masm.jcc (Assembler::notZero, IsInflated) ;
3562 // it's stack-locked, biased or neutral
3563 // TODO: optimize markword triage order to reduce the number of
3564 // conditional branches in the most common cases.
3565 // Beware -- there's a subtle invariant that fetch of the markword
3566 // at [FETCH], below, will never observe a biased encoding (*101b).
3567 // If this invariant is not held we'll suffer exclusion (safety) failure.
3569 if (UseBiasedLocking) {
3570 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3571 masm.movq (tmpReg, Address(objReg, 0)) ; // [FETCH]
3572 }
3574 masm.orq (tmpReg, 1) ;
3575 masm.movq (Address(boxReg, 0), tmpReg) ;
3576 if (os::is_MP()) { masm.lock(); }
3577 masm.cmpxchgq(boxReg, Address(objReg, 0)); // Updates tmpReg
3578 if (_counters != NULL) {
3579 masm.cond_inc32(Assembler::equal,
3580 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3581 }
3582 masm.jcc (Assembler::equal, DONE_LABEL);
3584 // Recursive locking
3585 masm.subq (tmpReg, rsp);
3586 masm.andq (tmpReg, 7 - os::vm_page_size());
3587 masm.movq (Address(boxReg, 0), tmpReg);
3588 if (_counters != NULL) {
3589 masm.cond_inc32(Assembler::equal,
3590 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3591 }
3592 masm.jmp (DONE_LABEL) ;
3594 masm.bind (IsInflated) ;
3595 // It's inflated
3597 // TODO: someday avoid the ST-before-CAS penalty by
3598 // relocating (deferring) the following ST.
3599 // We should also think about trying a CAS without having
3600 // fetched _owner. If the CAS is successful we may
3601 // avoid an RTO->RTS upgrade on the $line.
3602 masm.movptr(Address(boxReg, 0), intptr_t(markOopDesc::unused_mark())) ;
3604 masm.movq (boxReg, tmpReg) ;
3605 masm.movq (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3606 masm.testq (tmpReg, tmpReg) ;
3607 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3609 // It's inflated and appears unlocked
3610 if (os::is_MP()) { masm.lock(); }
3611 masm.cmpxchgq(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3612 // Intentional fall-through into DONE_LABEL ...
3614 masm.bind (DONE_LABEL) ;
3615 masm.nop () ; // avoid jmp to jmp
3616 }
3617 %}
3619 // obj: object to unlock
3620 // box: box address (displaced header location), killed
3621 // RBX: killed tmp; cannot be obj nor box
3622 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3623 %{
3625 Register objReg = as_Register($obj$$reg);
3626 Register boxReg = as_Register($box$$reg);
3627 Register tmpReg = as_Register($tmp$$reg);
3628 MacroAssembler masm(&cbuf);
3630 if (EmitSync & 4) {
3631 masm.cmpq (rsp, 0) ;
3632 } else
3633 if (EmitSync & 8) {
3634 Label DONE_LABEL;
3635 if (UseBiasedLocking) {
3636 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3637 }
3639 // Check whether the displaced header is 0
3640 //(=> recursive unlock)
3641 masm.movq(tmpReg, Address(boxReg, 0));
3642 masm.testq(tmpReg, tmpReg);
3643 masm.jcc(Assembler::zero, DONE_LABEL);
3645 // If not recursive lock, reset the header to displaced header
3646 if (os::is_MP()) {
3647 masm.lock();
3648 }
3649 masm.cmpxchgq(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3650 masm.bind(DONE_LABEL);
3651 masm.nop(); // avoid branch to branch
3652 } else {
3653 Label DONE_LABEL, Stacked, CheckSucc ;
3655 if (UseBiasedLocking) {
3656 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3657 }
3659 masm.movq (tmpReg, Address(objReg, 0)) ;
3660 masm.cmpq (Address(boxReg, 0), (int)NULL_WORD) ;
3661 masm.jcc (Assembler::zero, DONE_LABEL) ;
3662 masm.testq (tmpReg, 0x02) ;
3663 masm.jcc (Assembler::zero, Stacked) ;
3665 // It's inflated
3666 masm.movq (boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3667 masm.xorq (boxReg, r15_thread) ;
3668 masm.orq (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3669 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3670 masm.movq (boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3671 masm.orq (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3672 masm.jcc (Assembler::notZero, CheckSucc) ;
3673 masm.mov64 (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int)NULL_WORD) ;
3674 masm.jmp (DONE_LABEL) ;
3676 if ((EmitSync & 65536) == 0) {
3677 Label LSuccess, LGoSlowPath ;
3678 masm.bind (CheckSucc) ;
3679 masm.cmpq (Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int)NULL_WORD) ;
3680 masm.jcc (Assembler::zero, LGoSlowPath) ;
3682 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3683 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3684 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3685 // are all faster when the write buffer is populated.
3686 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int)NULL_WORD) ;
3687 if (os::is_MP()) {
3688 masm.lock () ; masm.addq (Address(rsp, 0), 0) ;
3689 }
3690 masm.cmpq (Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int)NULL_WORD) ;
3691 masm.jcc (Assembler::notZero, LSuccess) ;
3693 masm.movptr (boxReg, (int)NULL_WORD) ; // box is really EAX
3694 if (os::is_MP()) { masm.lock(); }
3695 masm.cmpxchgq (r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3696 masm.jcc (Assembler::notEqual, LSuccess) ;
3697 // Intentional fall-through into slow-path
3699 masm.bind (LGoSlowPath) ;
3700 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3701 masm.jmp (DONE_LABEL) ;
3703 masm.bind (LSuccess) ;
3704 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3705 masm.jmp (DONE_LABEL) ;
3706 }
3708 masm.bind (Stacked) ;
3709 masm.movq (tmpReg, Address (boxReg, 0)) ; // re-fetch
3710 if (os::is_MP()) { masm.lock(); }
3711 masm.cmpxchgq(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3713 if (EmitSync & 65536) {
3714 masm.bind (CheckSucc) ;
3715 }
3716 masm.bind(DONE_LABEL);
3717 if (EmitSync & 32768) {
3718 masm.nop(); // avoid branch to branch
3719 }
3720 }
3721 %}
3723 enc_class enc_String_Compare()
3724 %{
3725 Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
3726 POP_LABEL, DONE_LABEL, CONT_LABEL,
3727 WHILE_HEAD_LABEL;
3728 MacroAssembler masm(&cbuf);
3730 // Get the first character position in both strings
3731 // [8] char array, [12] offset, [16] count
3732 int value_offset = java_lang_String::value_offset_in_bytes();
3733 int offset_offset = java_lang_String::offset_offset_in_bytes();
3734 int count_offset = java_lang_String::count_offset_in_bytes();
3735 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3737 masm.load_heap_oop(rax, Address(rsi, value_offset));
3738 masm.movl(rcx, Address(rsi, offset_offset));
3739 masm.leaq(rax, Address(rax, rcx, Address::times_2, base_offset));
3740 masm.load_heap_oop(rbx, Address(rdi, value_offset));
3741 masm.movl(rcx, Address(rdi, offset_offset));
3742 masm.leaq(rbx, Address(rbx, rcx, Address::times_2, base_offset));
3744 // Compute the minimum of the string lengths(rsi) and the
3745 // difference of the string lengths (stack)
3747 masm.movl(rdi, Address(rdi, count_offset));
3748 masm.movl(rsi, Address(rsi, count_offset));
3749 masm.movl(rcx, rdi);
3750 masm.subl(rdi, rsi);
3751 masm.pushq(rdi);
3752 masm.cmovl(Assembler::lessEqual, rsi, rcx);
3754 // Is the minimum length zero?
3755 masm.bind(RCX_GOOD_LABEL);
3756 masm.testl(rsi, rsi);
3757 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3759 // Load first characters
3760 masm.load_unsigned_word(rcx, Address(rbx, 0));
3761 masm.load_unsigned_word(rdi, Address(rax, 0));
3763 // Compare first characters
3764 masm.subl(rcx, rdi);
3765 masm.jcc(Assembler::notZero, POP_LABEL);
3766 masm.decrementl(rsi);
3767 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3769 {
3770 // Check after comparing first character to see if strings are equivalent
3771 Label LSkip2;
3772 // Check if the strings start at same location
3773 masm.cmpq(rbx, rax);
3774 masm.jcc(Assembler::notEqual, LSkip2);
3776 // Check if the length difference is zero (from stack)
3777 masm.cmpl(Address(rsp, 0), 0x0);
3778 masm.jcc(Assembler::equal, LENGTH_DIFF_LABEL);
3780 // Strings might not be equivalent
3781 masm.bind(LSkip2);
3782 }
3784 // Shift RAX and RBX to the end of the arrays, negate min
3785 masm.leaq(rax, Address(rax, rsi, Address::times_2, 2));
3786 masm.leaq(rbx, Address(rbx, rsi, Address::times_2, 2));
3787 masm.negq(rsi);
3789 // Compare the rest of the characters
3790 masm.bind(WHILE_HEAD_LABEL);
3791 masm.load_unsigned_word(rcx, Address(rbx, rsi, Address::times_2, 0));
3792 masm.load_unsigned_word(rdi, Address(rax, rsi, Address::times_2, 0));
3793 masm.subl(rcx, rdi);
3794 masm.jcc(Assembler::notZero, POP_LABEL);
3795 masm.incrementq(rsi);
3796 masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
3798 // Strings are equal up to min length. Return the length difference.
3799 masm.bind(LENGTH_DIFF_LABEL);
3800 masm.popq(rcx);
3801 masm.jmp(DONE_LABEL);
3803 // Discard the stored length difference
3804 masm.bind(POP_LABEL);
3805 masm.addq(rsp, 8);
3807 // That's it
3808 masm.bind(DONE_LABEL);
3809 %}
3811 enc_class enc_rethrow()
3812 %{
3813 cbuf.set_inst_mark();
3814 emit_opcode(cbuf, 0xE9); // jmp entry
3815 emit_d32_reloc(cbuf,
3816 (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
3817 runtime_call_Relocation::spec(),
3818 RELOC_DISP32);
3819 %}
3821 enc_class absF_encoding(regF dst)
3822 %{
3823 int dstenc = $dst$$reg;
3824 address signmask_address = (address) StubRoutines::amd64::float_sign_mask();
3826 cbuf.set_inst_mark();
3827 if (dstenc >= 8) {
3828 emit_opcode(cbuf, Assembler::REX_R);
3829 dstenc -= 8;
3830 }
3831 // XXX reg_mem doesn't support RIP-relative addressing yet
3832 emit_opcode(cbuf, 0x0F);
3833 emit_opcode(cbuf, 0x54);
3834 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3835 emit_d32_reloc(cbuf, signmask_address);
3836 %}
3838 enc_class absD_encoding(regD dst)
3839 %{
3840 int dstenc = $dst$$reg;
3841 address signmask_address = (address) StubRoutines::amd64::double_sign_mask();
3843 cbuf.set_inst_mark();
3844 emit_opcode(cbuf, 0x66);
3845 if (dstenc >= 8) {
3846 emit_opcode(cbuf, Assembler::REX_R);
3847 dstenc -= 8;
3848 }
3849 // XXX reg_mem doesn't support RIP-relative addressing yet
3850 emit_opcode(cbuf, 0x0F);
3851 emit_opcode(cbuf, 0x54);
3852 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3853 emit_d32_reloc(cbuf, signmask_address);
3854 %}
3856 enc_class negF_encoding(regF dst)
3857 %{
3858 int dstenc = $dst$$reg;
3859 address signflip_address = (address) StubRoutines::amd64::float_sign_flip();
3861 cbuf.set_inst_mark();
3862 if (dstenc >= 8) {
3863 emit_opcode(cbuf, Assembler::REX_R);
3864 dstenc -= 8;
3865 }
3866 // XXX reg_mem doesn't support RIP-relative addressing yet
3867 emit_opcode(cbuf, 0x0F);
3868 emit_opcode(cbuf, 0x57);
3869 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3870 emit_d32_reloc(cbuf, signflip_address);
3871 %}
3873 enc_class negD_encoding(regD dst)
3874 %{
3875 int dstenc = $dst$$reg;
3876 address signflip_address = (address) StubRoutines::amd64::double_sign_flip();
3878 cbuf.set_inst_mark();
3879 emit_opcode(cbuf, 0x66);
3880 if (dstenc >= 8) {
3881 emit_opcode(cbuf, Assembler::REX_R);
3882 dstenc -= 8;
3883 }
3884 // XXX reg_mem doesn't support RIP-relative addressing yet
3885 emit_opcode(cbuf, 0x0F);
3886 emit_opcode(cbuf, 0x57);
3887 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3888 emit_d32_reloc(cbuf, signflip_address);
3889 %}
3891 enc_class f2i_fixup(rRegI dst, regF src)
3892 %{
3893 int dstenc = $dst$$reg;
3894 int srcenc = $src$$reg;
3896 // cmpl $dst, #0x80000000
3897 if (dstenc >= 8) {
3898 emit_opcode(cbuf, Assembler::REX_B);
3899 }
3900 emit_opcode(cbuf, 0x81);
3901 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3902 emit_d32(cbuf, 0x80000000);
3904 // jne,s done
3905 emit_opcode(cbuf, 0x75);
3906 if (srcenc < 8 && dstenc < 8) {
3907 emit_d8(cbuf, 0xF);
3908 } else if (srcenc >= 8 && dstenc >= 8) {
3909 emit_d8(cbuf, 0x11);
3910 } else {
3911 emit_d8(cbuf, 0x10);
3912 }
3914 // subq rsp, #8
3915 emit_opcode(cbuf, Assembler::REX_W);
3916 emit_opcode(cbuf, 0x83);
3917 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3918 emit_d8(cbuf, 8);
3920 // movss [rsp], $src
3921 emit_opcode(cbuf, 0xF3);
3922 if (srcenc >= 8) {
3923 emit_opcode(cbuf, Assembler::REX_R);
3924 }
3925 emit_opcode(cbuf, 0x0F);
3926 emit_opcode(cbuf, 0x11);
3927 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3929 // call f2i_fixup
3930 cbuf.set_inst_mark();
3931 emit_opcode(cbuf, 0xE8);
3932 emit_d32_reloc(cbuf,
3933 (int)
3934 (StubRoutines::amd64::f2i_fixup() - cbuf.code_end() - 4),
3935 runtime_call_Relocation::spec(),
3936 RELOC_DISP32);
3938 // popq $dst
3939 if (dstenc >= 8) {
3940 emit_opcode(cbuf, Assembler::REX_B);
3941 }
3942 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3944 // done:
3945 %}
3947 enc_class f2l_fixup(rRegL dst, regF src)
3948 %{
3949 int dstenc = $dst$$reg;
3950 int srcenc = $src$$reg;
3951 address const_address = (address) StubRoutines::amd64::double_sign_flip();
3953 // cmpq $dst, [0x8000000000000000]
3954 cbuf.set_inst_mark();
3955 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3956 emit_opcode(cbuf, 0x39);
3957 // XXX reg_mem doesn't support RIP-relative addressing yet
3958 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3959 emit_d32_reloc(cbuf, const_address);
3962 // jne,s done
3963 emit_opcode(cbuf, 0x75);
3964 if (srcenc < 8 && dstenc < 8) {
3965 emit_d8(cbuf, 0xF);
3966 } else if (srcenc >= 8 && dstenc >= 8) {
3967 emit_d8(cbuf, 0x11);
3968 } else {
3969 emit_d8(cbuf, 0x10);
3970 }
3972 // subq rsp, #8
3973 emit_opcode(cbuf, Assembler::REX_W);
3974 emit_opcode(cbuf, 0x83);
3975 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3976 emit_d8(cbuf, 8);
3978 // movss [rsp], $src
3979 emit_opcode(cbuf, 0xF3);
3980 if (srcenc >= 8) {
3981 emit_opcode(cbuf, Assembler::REX_R);
3982 }
3983 emit_opcode(cbuf, 0x0F);
3984 emit_opcode(cbuf, 0x11);
3985 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3987 // call f2l_fixup
3988 cbuf.set_inst_mark();
3989 emit_opcode(cbuf, 0xE8);
3990 emit_d32_reloc(cbuf,
3991 (int)
3992 (StubRoutines::amd64::f2l_fixup() - cbuf.code_end() - 4),
3993 runtime_call_Relocation::spec(),
3994 RELOC_DISP32);
3996 // popq $dst
3997 if (dstenc >= 8) {
3998 emit_opcode(cbuf, Assembler::REX_B);
3999 }
4000 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4002 // done:
4003 %}
4005 enc_class d2i_fixup(rRegI dst, regD src)
4006 %{
4007 int dstenc = $dst$$reg;
4008 int srcenc = $src$$reg;
4010 // cmpl $dst, #0x80000000
4011 if (dstenc >= 8) {
4012 emit_opcode(cbuf, Assembler::REX_B);
4013 }
4014 emit_opcode(cbuf, 0x81);
4015 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
4016 emit_d32(cbuf, 0x80000000);
4018 // jne,s done
4019 emit_opcode(cbuf, 0x75);
4020 if (srcenc < 8 && dstenc < 8) {
4021 emit_d8(cbuf, 0xF);
4022 } else if (srcenc >= 8 && dstenc >= 8) {
4023 emit_d8(cbuf, 0x11);
4024 } else {
4025 emit_d8(cbuf, 0x10);
4026 }
4028 // subq rsp, #8
4029 emit_opcode(cbuf, Assembler::REX_W);
4030 emit_opcode(cbuf, 0x83);
4031 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4032 emit_d8(cbuf, 8);
4034 // movsd [rsp], $src
4035 emit_opcode(cbuf, 0xF2);
4036 if (srcenc >= 8) {
4037 emit_opcode(cbuf, Assembler::REX_R);
4038 }
4039 emit_opcode(cbuf, 0x0F);
4040 emit_opcode(cbuf, 0x11);
4041 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4043 // call d2i_fixup
4044 cbuf.set_inst_mark();
4045 emit_opcode(cbuf, 0xE8);
4046 emit_d32_reloc(cbuf,
4047 (int)
4048 (StubRoutines::amd64::d2i_fixup() - cbuf.code_end() - 4),
4049 runtime_call_Relocation::spec(),
4050 RELOC_DISP32);
4052 // popq $dst
4053 if (dstenc >= 8) {
4054 emit_opcode(cbuf, Assembler::REX_B);
4055 }
4056 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4058 // done:
4059 %}
4061 enc_class d2l_fixup(rRegL dst, regD src)
4062 %{
4063 int dstenc = $dst$$reg;
4064 int srcenc = $src$$reg;
4065 address const_address = (address) StubRoutines::amd64::double_sign_flip();
4067 // cmpq $dst, [0x8000000000000000]
4068 cbuf.set_inst_mark();
4069 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4070 emit_opcode(cbuf, 0x39);
4071 // XXX reg_mem doesn't support RIP-relative addressing yet
4072 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4073 emit_d32_reloc(cbuf, const_address);
4076 // jne,s done
4077 emit_opcode(cbuf, 0x75);
4078 if (srcenc < 8 && dstenc < 8) {
4079 emit_d8(cbuf, 0xF);
4080 } else if (srcenc >= 8 && dstenc >= 8) {
4081 emit_d8(cbuf, 0x11);
4082 } else {
4083 emit_d8(cbuf, 0x10);
4084 }
4086 // subq rsp, #8
4087 emit_opcode(cbuf, Assembler::REX_W);
4088 emit_opcode(cbuf, 0x83);
4089 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4090 emit_d8(cbuf, 8);
4092 // movsd [rsp], $src
4093 emit_opcode(cbuf, 0xF2);
4094 if (srcenc >= 8) {
4095 emit_opcode(cbuf, Assembler::REX_R);
4096 }
4097 emit_opcode(cbuf, 0x0F);
4098 emit_opcode(cbuf, 0x11);
4099 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4101 // call d2l_fixup
4102 cbuf.set_inst_mark();
4103 emit_opcode(cbuf, 0xE8);
4104 emit_d32_reloc(cbuf,
4105 (int)
4106 (StubRoutines::amd64::d2l_fixup() - cbuf.code_end() - 4),
4107 runtime_call_Relocation::spec(),
4108 RELOC_DISP32);
4110 // popq $dst
4111 if (dstenc >= 8) {
4112 emit_opcode(cbuf, Assembler::REX_B);
4113 }
4114 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4116 // done:
4117 %}
4119 enc_class enc_membar_acquire
4120 %{
4121 // [jk] not needed currently, if you enable this and it really
4122 // emits code don't forget to the remove the "size(0)" line in
4123 // membar_acquire()
4124 // MacroAssembler masm(&cbuf);
4125 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4126 // Assembler::LoadLoad));
4127 %}
4129 enc_class enc_membar_release
4130 %{
4131 // [jk] not needed currently, if you enable this and it really
4132 // emits code don't forget to the remove the "size(0)" line in
4133 // membar_release()
4134 // MacroAssembler masm(&cbuf);
4135 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4136 // Assembler::StoreStore));
4137 %}
4139 enc_class enc_membar_volatile
4140 %{
4141 MacroAssembler masm(&cbuf);
4142 masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
4143 Assembler::StoreStore));
4144 %}
4146 // Safepoint Poll. This polls the safepoint page, and causes an
4147 // exception if it is not readable. Unfortunately, it kills
4148 // RFLAGS in the process.
4149 enc_class enc_safepoint_poll
4150 %{
4151 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
4152 // XXX reg_mem doesn't support RIP-relative addressing yet
4153 cbuf.set_inst_mark();
4154 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
4155 emit_opcode(cbuf, 0x85); // testl
4156 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
4157 // cbuf.inst_mark() is beginning of instruction
4158 emit_d32_reloc(cbuf, os::get_polling_page());
4159 // relocInfo::poll_type,
4160 %}
4161 %}
4165 //----------FRAME--------------------------------------------------------------
4166 // Definition of frame structure and management information.
4167 //
4168 // S T A C K L A Y O U T Allocators stack-slot number
4169 // | (to get allocators register number
4170 // G Owned by | | v add OptoReg::stack0())
4171 // r CALLER | |
4172 // o | +--------+ pad to even-align allocators stack-slot
4173 // w V | pad0 | numbers; owned by CALLER
4174 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
4175 // h ^ | in | 5
4176 // | | args | 4 Holes in incoming args owned by SELF
4177 // | | | | 3
4178 // | | +--------+
4179 // V | | old out| Empty on Intel, window on Sparc
4180 // | old |preserve| Must be even aligned.
4181 // | SP-+--------+----> Matcher::_old_SP, even aligned
4182 // | | in | 3 area for Intel ret address
4183 // Owned by |preserve| Empty on Sparc.
4184 // SELF +--------+
4185 // | | pad2 | 2 pad to align old SP
4186 // | +--------+ 1
4187 // | | locks | 0
4188 // | +--------+----> OptoReg::stack0(), even aligned
4189 // | | pad1 | 11 pad to align new SP
4190 // | +--------+
4191 // | | | 10
4192 // | | spills | 9 spills
4193 // V | | 8 (pad0 slot for callee)
4194 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
4195 // ^ | out | 7
4196 // | | args | 6 Holes in outgoing args owned by CALLEE
4197 // Owned by +--------+
4198 // CALLEE | new out| 6 Empty on Intel, window on Sparc
4199 // | new |preserve| Must be even-aligned.
4200 // | SP-+--------+----> Matcher::_new_SP, even aligned
4201 // | | |
4202 //
4203 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
4204 // known from SELF's arguments and the Java calling convention.
4205 // Region 6-7 is determined per call site.
4206 // Note 2: If the calling convention leaves holes in the incoming argument
4207 // area, those holes are owned by SELF. Holes in the outgoing area
4208 // are owned by the CALLEE. Holes should not be nessecary in the
4209 // incoming area, as the Java calling convention is completely under
4210 // the control of the AD file. Doubles can be sorted and packed to
4211 // avoid holes. Holes in the outgoing arguments may be nessecary for
4212 // varargs C calling conventions.
4213 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
4214 // even aligned with pad0 as needed.
4215 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
4216 // region 6-11 is even aligned; it may be padded out more so that
4217 // the region from SP to FP meets the minimum stack alignment.
4218 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
4219 // alignment. Region 11, pad1, may be dynamically extended so that
4220 // SP meets the minimum alignment.
4222 frame
4223 %{
4224 // What direction does stack grow in (assumed to be same for C & Java)
4225 stack_direction(TOWARDS_LOW);
4227 // These three registers define part of the calling convention
4228 // between compiled code and the interpreter.
4229 inline_cache_reg(RAX); // Inline Cache Register
4230 interpreter_method_oop_reg(RBX); // Method Oop Register when
4231 // calling interpreter
4233 // Optional: name the operand used by cisc-spilling to access
4234 // [stack_pointer + offset]
4235 cisc_spilling_operand_name(indOffset32);
4237 // Number of stack slots consumed by locking an object
4238 sync_stack_slots(2);
4240 // Compiled code's Frame Pointer
4241 frame_pointer(RSP);
4243 // Interpreter stores its frame pointer in a register which is
4244 // stored to the stack by I2CAdaptors.
4245 // I2CAdaptors convert from interpreted java to compiled java.
4246 interpreter_frame_pointer(RBP);
4248 // Stack alignment requirement
4249 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
4251 // Number of stack slots between incoming argument block and the start of
4252 // a new frame. The PROLOG must add this many slots to the stack. The
4253 // EPILOG must remove this many slots. amd64 needs two slots for
4254 // return address.
4255 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
4257 // Number of outgoing stack slots killed above the out_preserve_stack_slots
4258 // for calls to C. Supports the var-args backing area for register parms.
4259 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
4261 // The after-PROLOG location of the return address. Location of
4262 // return address specifies a type (REG or STACK) and a number
4263 // representing the register number (i.e. - use a register name) or
4264 // stack slot.
4265 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4266 // Otherwise, it is above the locks and verification slot and alignment word
4267 return_addr(STACK - 2 +
4268 round_to(2 + 2 * VerifyStackAtCalls +
4269 Compile::current()->fixed_slots(),
4270 WordsPerLong * 2));
4272 // Body of function which returns an integer array locating
4273 // arguments either in registers or in stack slots. Passed an array
4274 // of ideal registers called "sig" and a "length" count. Stack-slot
4275 // offsets are based on outgoing arguments, i.e. a CALLER setting up
4276 // arguments for a CALLEE. Incoming stack arguments are
4277 // automatically biased by the preserve_stack_slots field above.
4279 calling_convention
4280 %{
4281 // No difference between ingoing/outgoing just pass false
4282 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4283 %}
4285 c_calling_convention
4286 %{
4287 // This is obviously always outgoing
4288 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4289 %}
4291 // Location of compiled Java return values. Same as C for now.
4292 return_value
4293 %{
4294 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4295 "only return normal values");
4297 static const int lo[Op_RegL + 1] = {
4298 0,
4299 0,
4300 RAX_num, // Op_RegN
4301 RAX_num, // Op_RegI
4302 RAX_num, // Op_RegP
4303 XMM0_num, // Op_RegF
4304 XMM0_num, // Op_RegD
4305 RAX_num // Op_RegL
4306 };
4307 static const int hi[Op_RegL + 1] = {
4308 0,
4309 0,
4310 OptoReg::Bad, // Op_RegN
4311 OptoReg::Bad, // Op_RegI
4312 RAX_H_num, // Op_RegP
4313 OptoReg::Bad, // Op_RegF
4314 XMM0_H_num, // Op_RegD
4315 RAX_H_num // Op_RegL
4316 };
4317 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4318 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4319 %}
4320 %}
4322 //----------ATTRIBUTES---------------------------------------------------------
4323 //----------Operand Attributes-------------------------------------------------
4324 op_attrib op_cost(0); // Required cost attribute
4326 //----------Instruction Attributes---------------------------------------------
4327 ins_attrib ins_cost(100); // Required cost attribute
4328 ins_attrib ins_size(8); // Required size attribute (in bits)
4329 ins_attrib ins_pc_relative(0); // Required PC Relative flag
4330 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4331 // a non-matching short branch variant
4332 // of some long branch?
4333 ins_attrib ins_alignment(1); // Required alignment attribute (must
4334 // be a power of 2) specifies the
4335 // alignment that some part of the
4336 // instruction (not necessarily the
4337 // start) requires. If > 1, a
4338 // compute_padding() function must be
4339 // provided for the instruction
4341 //----------OPERANDS-----------------------------------------------------------
4342 // Operand definitions must precede instruction definitions for correct parsing
4343 // in the ADLC because operands constitute user defined types which are used in
4344 // instruction definitions.
4346 //----------Simple Operands----------------------------------------------------
4347 // Immediate Operands
4348 // Integer Immediate
4349 operand immI()
4350 %{
4351 match(ConI);
4353 op_cost(10);
4354 format %{ %}
4355 interface(CONST_INTER);
4356 %}
4358 // Constant for test vs zero
4359 operand immI0()
4360 %{
4361 predicate(n->get_int() == 0);
4362 match(ConI);
4364 op_cost(0);
4365 format %{ %}
4366 interface(CONST_INTER);
4367 %}
4369 // Constant for increment
4370 operand immI1()
4371 %{
4372 predicate(n->get_int() == 1);
4373 match(ConI);
4375 op_cost(0);
4376 format %{ %}
4377 interface(CONST_INTER);
4378 %}
4380 // Constant for decrement
4381 operand immI_M1()
4382 %{
4383 predicate(n->get_int() == -1);
4384 match(ConI);
4386 op_cost(0);
4387 format %{ %}
4388 interface(CONST_INTER);
4389 %}
4391 // Valid scale values for addressing modes
4392 operand immI2()
4393 %{
4394 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4395 match(ConI);
4397 format %{ %}
4398 interface(CONST_INTER);
4399 %}
4401 operand immI8()
4402 %{
4403 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4404 match(ConI);
4406 op_cost(5);
4407 format %{ %}
4408 interface(CONST_INTER);
4409 %}
4411 operand immI16()
4412 %{
4413 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4414 match(ConI);
4416 op_cost(10);
4417 format %{ %}
4418 interface(CONST_INTER);
4419 %}
4421 // Constant for long shifts
4422 operand immI_32()
4423 %{
4424 predicate( n->get_int() == 32 );
4425 match(ConI);
4427 op_cost(0);
4428 format %{ %}
4429 interface(CONST_INTER);
4430 %}
4432 // Constant for long shifts
4433 operand immI_64()
4434 %{
4435 predicate( n->get_int() == 64 );
4436 match(ConI);
4438 op_cost(0);
4439 format %{ %}
4440 interface(CONST_INTER);
4441 %}
4443 // Pointer Immediate
4444 operand immP()
4445 %{
4446 match(ConP);
4448 op_cost(10);
4449 format %{ %}
4450 interface(CONST_INTER);
4451 %}
4453 // NULL Pointer Immediate
4454 operand immP0()
4455 %{
4456 predicate(n->get_ptr() == 0);
4457 match(ConP);
4459 op_cost(5);
4460 format %{ %}
4461 interface(CONST_INTER);
4462 %}
4464 // Pointer Immediate
4465 operand immN() %{
4466 match(ConN);
4468 op_cost(10);
4469 format %{ %}
4470 interface(CONST_INTER);
4471 %}
4473 // NULL Pointer Immediate
4474 operand immN0() %{
4475 predicate(n->get_narrowcon() == 0);
4476 match(ConN);
4478 op_cost(5);
4479 format %{ %}
4480 interface(CONST_INTER);
4481 %}
4483 operand immP31()
4484 %{
4485 predicate(!n->as_Type()->type()->isa_oopptr()
4486 && (n->get_ptr() >> 31) == 0);
4487 match(ConP);
4489 op_cost(5);
4490 format %{ %}
4491 interface(CONST_INTER);
4492 %}
4495 // Long Immediate
4496 operand immL()
4497 %{
4498 match(ConL);
4500 op_cost(20);
4501 format %{ %}
4502 interface(CONST_INTER);
4503 %}
4505 // Long Immediate 8-bit
4506 operand immL8()
4507 %{
4508 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4509 match(ConL);
4511 op_cost(5);
4512 format %{ %}
4513 interface(CONST_INTER);
4514 %}
4516 // Long Immediate 32-bit unsigned
4517 operand immUL32()
4518 %{
4519 predicate(n->get_long() == (unsigned int) (n->get_long()));
4520 match(ConL);
4522 op_cost(10);
4523 format %{ %}
4524 interface(CONST_INTER);
4525 %}
4527 // Long Immediate 32-bit signed
4528 operand immL32()
4529 %{
4530 predicate(n->get_long() == (int) (n->get_long()));
4531 match(ConL);
4533 op_cost(15);
4534 format %{ %}
4535 interface(CONST_INTER);
4536 %}
4538 // Long Immediate zero
4539 operand immL0()
4540 %{
4541 predicate(n->get_long() == 0L);
4542 match(ConL);
4544 op_cost(10);
4545 format %{ %}
4546 interface(CONST_INTER);
4547 %}
4549 // Constant for increment
4550 operand immL1()
4551 %{
4552 predicate(n->get_long() == 1);
4553 match(ConL);
4555 format %{ %}
4556 interface(CONST_INTER);
4557 %}
4559 // Constant for decrement
4560 operand immL_M1()
4561 %{
4562 predicate(n->get_long() == -1);
4563 match(ConL);
4565 format %{ %}
4566 interface(CONST_INTER);
4567 %}
4569 // Long Immediate: the value 10
4570 operand immL10()
4571 %{
4572 predicate(n->get_long() == 10);
4573 match(ConL);
4575 format %{ %}
4576 interface(CONST_INTER);
4577 %}
4579 // Long immediate from 0 to 127.
4580 // Used for a shorter form of long mul by 10.
4581 operand immL_127()
4582 %{
4583 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4584 match(ConL);
4586 op_cost(10);
4587 format %{ %}
4588 interface(CONST_INTER);
4589 %}
4591 // Long Immediate: low 32-bit mask
4592 operand immL_32bits()
4593 %{
4594 predicate(n->get_long() == 0xFFFFFFFFL);
4595 match(ConL);
4596 op_cost(20);
4598 format %{ %}
4599 interface(CONST_INTER);
4600 %}
4602 // Float Immediate zero
4603 operand immF0()
4604 %{
4605 predicate(jint_cast(n->getf()) == 0);
4606 match(ConF);
4608 op_cost(5);
4609 format %{ %}
4610 interface(CONST_INTER);
4611 %}
4613 // Float Immediate
4614 operand immF()
4615 %{
4616 match(ConF);
4618 op_cost(15);
4619 format %{ %}
4620 interface(CONST_INTER);
4621 %}
4623 // Double Immediate zero
4624 operand immD0()
4625 %{
4626 predicate(jlong_cast(n->getd()) == 0);
4627 match(ConD);
4629 op_cost(5);
4630 format %{ %}
4631 interface(CONST_INTER);
4632 %}
4634 // Double Immediate
4635 operand immD()
4636 %{
4637 match(ConD);
4639 op_cost(15);
4640 format %{ %}
4641 interface(CONST_INTER);
4642 %}
4644 // Immediates for special shifts (sign extend)
4646 // Constants for increment
4647 operand immI_16()
4648 %{
4649 predicate(n->get_int() == 16);
4650 match(ConI);
4652 format %{ %}
4653 interface(CONST_INTER);
4654 %}
4656 operand immI_24()
4657 %{
4658 predicate(n->get_int() == 24);
4659 match(ConI);
4661 format %{ %}
4662 interface(CONST_INTER);
4663 %}
4665 // Constant for byte-wide masking
4666 operand immI_255()
4667 %{
4668 predicate(n->get_int() == 255);
4669 match(ConI);
4671 format %{ %}
4672 interface(CONST_INTER);
4673 %}
4675 // Constant for short-wide masking
4676 operand immI_65535()
4677 %{
4678 predicate(n->get_int() == 65535);
4679 match(ConI);
4681 format %{ %}
4682 interface(CONST_INTER);
4683 %}
4685 // Constant for byte-wide masking
4686 operand immL_255()
4687 %{
4688 predicate(n->get_long() == 255);
4689 match(ConL);
4691 format %{ %}
4692 interface(CONST_INTER);
4693 %}
4695 // Constant for short-wide masking
4696 operand immL_65535()
4697 %{
4698 predicate(n->get_long() == 65535);
4699 match(ConL);
4701 format %{ %}
4702 interface(CONST_INTER);
4703 %}
4705 // Register Operands
4706 // Integer Register
4707 operand rRegI()
4708 %{
4709 constraint(ALLOC_IN_RC(int_reg));
4710 match(RegI);
4712 match(rax_RegI);
4713 match(rbx_RegI);
4714 match(rcx_RegI);
4715 match(rdx_RegI);
4716 match(rdi_RegI);
4718 format %{ %}
4719 interface(REG_INTER);
4720 %}
4722 // Special Registers
4723 operand rax_RegI()
4724 %{
4725 constraint(ALLOC_IN_RC(int_rax_reg));
4726 match(RegI);
4727 match(rRegI);
4729 format %{ "RAX" %}
4730 interface(REG_INTER);
4731 %}
4733 // Special Registers
4734 operand rbx_RegI()
4735 %{
4736 constraint(ALLOC_IN_RC(int_rbx_reg));
4737 match(RegI);
4738 match(rRegI);
4740 format %{ "RBX" %}
4741 interface(REG_INTER);
4742 %}
4744 operand rcx_RegI()
4745 %{
4746 constraint(ALLOC_IN_RC(int_rcx_reg));
4747 match(RegI);
4748 match(rRegI);
4750 format %{ "RCX" %}
4751 interface(REG_INTER);
4752 %}
4754 operand rdx_RegI()
4755 %{
4756 constraint(ALLOC_IN_RC(int_rdx_reg));
4757 match(RegI);
4758 match(rRegI);
4760 format %{ "RDX" %}
4761 interface(REG_INTER);
4762 %}
4764 operand rdi_RegI()
4765 %{
4766 constraint(ALLOC_IN_RC(int_rdi_reg));
4767 match(RegI);
4768 match(rRegI);
4770 format %{ "RDI" %}
4771 interface(REG_INTER);
4772 %}
4774 operand no_rcx_RegI()
4775 %{
4776 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4777 match(RegI);
4778 match(rax_RegI);
4779 match(rbx_RegI);
4780 match(rdx_RegI);
4781 match(rdi_RegI);
4783 format %{ %}
4784 interface(REG_INTER);
4785 %}
4787 operand no_rax_rdx_RegI()
4788 %{
4789 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4790 match(RegI);
4791 match(rbx_RegI);
4792 match(rcx_RegI);
4793 match(rdi_RegI);
4795 format %{ %}
4796 interface(REG_INTER);
4797 %}
4799 // Pointer Register
4800 operand any_RegP()
4801 %{
4802 constraint(ALLOC_IN_RC(any_reg));
4803 match(RegP);
4804 match(rax_RegP);
4805 match(rbx_RegP);
4806 match(rdi_RegP);
4807 match(rsi_RegP);
4808 match(rbp_RegP);
4809 match(r15_RegP);
4810 match(rRegP);
4812 format %{ %}
4813 interface(REG_INTER);
4814 %}
4816 operand rRegP()
4817 %{
4818 constraint(ALLOC_IN_RC(ptr_reg));
4819 match(RegP);
4820 match(rax_RegP);
4821 match(rbx_RegP);
4822 match(rdi_RegP);
4823 match(rsi_RegP);
4824 match(rbp_RegP);
4825 match(r15_RegP); // See Q&A below about r15_RegP.
4827 format %{ %}
4828 interface(REG_INTER);
4829 %}
4832 operand r12RegL() %{
4833 constraint(ALLOC_IN_RC(long_r12_reg));
4834 match(RegL);
4836 format %{ %}
4837 interface(REG_INTER);
4838 %}
4840 operand rRegN() %{
4841 constraint(ALLOC_IN_RC(int_reg));
4842 match(RegN);
4844 format %{ %}
4845 interface(REG_INTER);
4846 %}
4848 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4849 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4850 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4851 // The output of an instruction is controlled by the allocator, which respects
4852 // register class masks, not match rules. Unless an instruction mentions
4853 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4854 // by the allocator as an input.
4856 operand no_rax_RegP()
4857 %{
4858 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4859 match(RegP);
4860 match(rbx_RegP);
4861 match(rsi_RegP);
4862 match(rdi_RegP);
4864 format %{ %}
4865 interface(REG_INTER);
4866 %}
4868 operand no_rbp_RegP()
4869 %{
4870 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4871 match(RegP);
4872 match(rbx_RegP);
4873 match(rsi_RegP);
4874 match(rdi_RegP);
4876 format %{ %}
4877 interface(REG_INTER);
4878 %}
4880 operand no_rax_rbx_RegP()
4881 %{
4882 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4883 match(RegP);
4884 match(rsi_RegP);
4885 match(rdi_RegP);
4887 format %{ %}
4888 interface(REG_INTER);
4889 %}
4891 // Special Registers
4892 // Return a pointer value
4893 operand rax_RegP()
4894 %{
4895 constraint(ALLOC_IN_RC(ptr_rax_reg));
4896 match(RegP);
4897 match(rRegP);
4899 format %{ %}
4900 interface(REG_INTER);
4901 %}
4903 // Special Registers
4904 // Return a compressed pointer value
4905 operand rax_RegN()
4906 %{
4907 constraint(ALLOC_IN_RC(int_rax_reg));
4908 match(RegN);
4909 match(rRegN);
4911 format %{ %}
4912 interface(REG_INTER);
4913 %}
4915 // Used in AtomicAdd
4916 operand rbx_RegP()
4917 %{
4918 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4919 match(RegP);
4920 match(rRegP);
4922 format %{ %}
4923 interface(REG_INTER);
4924 %}
4926 operand rsi_RegP()
4927 %{
4928 constraint(ALLOC_IN_RC(ptr_rsi_reg));
4929 match(RegP);
4930 match(rRegP);
4932 format %{ %}
4933 interface(REG_INTER);
4934 %}
4936 // Used in rep stosq
4937 operand rdi_RegP()
4938 %{
4939 constraint(ALLOC_IN_RC(ptr_rdi_reg));
4940 match(RegP);
4941 match(rRegP);
4943 format %{ %}
4944 interface(REG_INTER);
4945 %}
4947 operand rbp_RegP()
4948 %{
4949 constraint(ALLOC_IN_RC(ptr_rbp_reg));
4950 match(RegP);
4951 match(rRegP);
4953 format %{ %}
4954 interface(REG_INTER);
4955 %}
4957 operand r15_RegP()
4958 %{
4959 constraint(ALLOC_IN_RC(ptr_r15_reg));
4960 match(RegP);
4961 match(rRegP);
4963 format %{ %}
4964 interface(REG_INTER);
4965 %}
4967 operand rRegL()
4968 %{
4969 constraint(ALLOC_IN_RC(long_reg));
4970 match(RegL);
4971 match(rax_RegL);
4972 match(rdx_RegL);
4974 format %{ %}
4975 interface(REG_INTER);
4976 %}
4978 // Special Registers
4979 operand no_rax_rdx_RegL()
4980 %{
4981 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4982 match(RegL);
4983 match(rRegL);
4985 format %{ %}
4986 interface(REG_INTER);
4987 %}
4989 operand no_rax_RegL()
4990 %{
4991 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4992 match(RegL);
4993 match(rRegL);
4994 match(rdx_RegL);
4996 format %{ %}
4997 interface(REG_INTER);
4998 %}
5000 operand no_rcx_RegL()
5001 %{
5002 constraint(ALLOC_IN_RC(long_no_rcx_reg));
5003 match(RegL);
5004 match(rRegL);
5006 format %{ %}
5007 interface(REG_INTER);
5008 %}
5010 operand rax_RegL()
5011 %{
5012 constraint(ALLOC_IN_RC(long_rax_reg));
5013 match(RegL);
5014 match(rRegL);
5016 format %{ "RAX" %}
5017 interface(REG_INTER);
5018 %}
5020 operand rcx_RegL()
5021 %{
5022 constraint(ALLOC_IN_RC(long_rcx_reg));
5023 match(RegL);
5024 match(rRegL);
5026 format %{ %}
5027 interface(REG_INTER);
5028 %}
5030 operand rdx_RegL()
5031 %{
5032 constraint(ALLOC_IN_RC(long_rdx_reg));
5033 match(RegL);
5034 match(rRegL);
5036 format %{ %}
5037 interface(REG_INTER);
5038 %}
5040 // Flags register, used as output of compare instructions
5041 operand rFlagsReg()
5042 %{
5043 constraint(ALLOC_IN_RC(int_flags));
5044 match(RegFlags);
5046 format %{ "RFLAGS" %}
5047 interface(REG_INTER);
5048 %}
5050 // Flags register, used as output of FLOATING POINT compare instructions
5051 operand rFlagsRegU()
5052 %{
5053 constraint(ALLOC_IN_RC(int_flags));
5054 match(RegFlags);
5056 format %{ "RFLAGS_U" %}
5057 interface(REG_INTER);
5058 %}
5060 // Float register operands
5061 operand regF()
5062 %{
5063 constraint(ALLOC_IN_RC(float_reg));
5064 match(RegF);
5066 format %{ %}
5067 interface(REG_INTER);
5068 %}
5070 // Double register operands
5071 operand regD()
5072 %{
5073 constraint(ALLOC_IN_RC(double_reg));
5074 match(RegD);
5076 format %{ %}
5077 interface(REG_INTER);
5078 %}
5081 //----------Memory Operands----------------------------------------------------
5082 // Direct Memory Operand
5083 // operand direct(immP addr)
5084 // %{
5085 // match(addr);
5087 // format %{ "[$addr]" %}
5088 // interface(MEMORY_INTER) %{
5089 // base(0xFFFFFFFF);
5090 // index(0x4);
5091 // scale(0x0);
5092 // disp($addr);
5093 // %}
5094 // %}
5096 // Indirect Memory Operand
5097 operand indirect(any_RegP reg)
5098 %{
5099 constraint(ALLOC_IN_RC(ptr_reg));
5100 match(reg);
5102 format %{ "[$reg]" %}
5103 interface(MEMORY_INTER) %{
5104 base($reg);
5105 index(0x4);
5106 scale(0x0);
5107 disp(0x0);
5108 %}
5109 %}
5111 // Indirect Memory Plus Short Offset Operand
5112 operand indOffset8(any_RegP reg, immL8 off)
5113 %{
5114 constraint(ALLOC_IN_RC(ptr_reg));
5115 match(AddP reg off);
5117 format %{ "[$reg + $off (8-bit)]" %}
5118 interface(MEMORY_INTER) %{
5119 base($reg);
5120 index(0x4);
5121 scale(0x0);
5122 disp($off);
5123 %}
5124 %}
5126 // Indirect Memory Plus Long Offset Operand
5127 operand indOffset32(any_RegP reg, immL32 off)
5128 %{
5129 constraint(ALLOC_IN_RC(ptr_reg));
5130 match(AddP reg off);
5132 format %{ "[$reg + $off (32-bit)]" %}
5133 interface(MEMORY_INTER) %{
5134 base($reg);
5135 index(0x4);
5136 scale(0x0);
5137 disp($off);
5138 %}
5139 %}
5141 // Indirect Memory Plus Index Register Plus Offset Operand
5142 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
5143 %{
5144 constraint(ALLOC_IN_RC(ptr_reg));
5145 match(AddP (AddP reg lreg) off);
5147 op_cost(10);
5148 format %{"[$reg + $off + $lreg]" %}
5149 interface(MEMORY_INTER) %{
5150 base($reg);
5151 index($lreg);
5152 scale(0x0);
5153 disp($off);
5154 %}
5155 %}
5157 // Indirect Memory Plus Index Register Plus Offset Operand
5158 operand indIndex(any_RegP reg, rRegL lreg)
5159 %{
5160 constraint(ALLOC_IN_RC(ptr_reg));
5161 match(AddP reg lreg);
5163 op_cost(10);
5164 format %{"[$reg + $lreg]" %}
5165 interface(MEMORY_INTER) %{
5166 base($reg);
5167 index($lreg);
5168 scale(0x0);
5169 disp(0x0);
5170 %}
5171 %}
5173 // Indirect Memory Times Scale Plus Index Register
5174 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
5175 %{
5176 constraint(ALLOC_IN_RC(ptr_reg));
5177 match(AddP reg (LShiftL lreg scale));
5179 op_cost(10);
5180 format %{"[$reg + $lreg << $scale]" %}
5181 interface(MEMORY_INTER) %{
5182 base($reg);
5183 index($lreg);
5184 scale($scale);
5185 disp(0x0);
5186 %}
5187 %}
5189 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5190 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
5191 %{
5192 constraint(ALLOC_IN_RC(ptr_reg));
5193 match(AddP (AddP reg (LShiftL lreg scale)) off);
5195 op_cost(10);
5196 format %{"[$reg + $off + $lreg << $scale]" %}
5197 interface(MEMORY_INTER) %{
5198 base($reg);
5199 index($lreg);
5200 scale($scale);
5201 disp($off);
5202 %}
5203 %}
5205 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5206 operand indIndexScaleOffsetComp(rRegN src, immL32 off, r12RegL base) %{
5207 constraint(ALLOC_IN_RC(ptr_reg));
5208 match(AddP (DecodeN src base) off);
5210 op_cost(10);
5211 format %{"[$base + $src << 3 + $off] (compressed)" %}
5212 interface(MEMORY_INTER) %{
5213 base($base);
5214 index($src);
5215 scale(0x3);
5216 disp($off);
5217 %}
5218 %}
5220 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5221 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
5222 %{
5223 constraint(ALLOC_IN_RC(ptr_reg));
5224 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5225 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
5227 op_cost(10);
5228 format %{"[$reg + $off + $idx << $scale]" %}
5229 interface(MEMORY_INTER) %{
5230 base($reg);
5231 index($idx);
5232 scale($scale);
5233 disp($off);
5234 %}
5235 %}
5237 //----------Special Memory Operands--------------------------------------------
5238 // Stack Slot Operand - This operand is used for loading and storing temporary
5239 // values on the stack where a match requires a value to
5240 // flow through memory.
5241 operand stackSlotP(sRegP reg)
5242 %{
5243 constraint(ALLOC_IN_RC(stack_slots));
5244 // No match rule because this operand is only generated in matching
5246 format %{ "[$reg]" %}
5247 interface(MEMORY_INTER) %{
5248 base(0x4); // RSP
5249 index(0x4); // No Index
5250 scale(0x0); // No Scale
5251 disp($reg); // Stack Offset
5252 %}
5253 %}
5255 operand stackSlotI(sRegI reg)
5256 %{
5257 constraint(ALLOC_IN_RC(stack_slots));
5258 // No match rule because this operand is only generated in matching
5260 format %{ "[$reg]" %}
5261 interface(MEMORY_INTER) %{
5262 base(0x4); // RSP
5263 index(0x4); // No Index
5264 scale(0x0); // No Scale
5265 disp($reg); // Stack Offset
5266 %}
5267 %}
5269 operand stackSlotF(sRegF reg)
5270 %{
5271 constraint(ALLOC_IN_RC(stack_slots));
5272 // No match rule because this operand is only generated in matching
5274 format %{ "[$reg]" %}
5275 interface(MEMORY_INTER) %{
5276 base(0x4); // RSP
5277 index(0x4); // No Index
5278 scale(0x0); // No Scale
5279 disp($reg); // Stack Offset
5280 %}
5281 %}
5283 operand stackSlotD(sRegD reg)
5284 %{
5285 constraint(ALLOC_IN_RC(stack_slots));
5286 // No match rule because this operand is only generated in matching
5288 format %{ "[$reg]" %}
5289 interface(MEMORY_INTER) %{
5290 base(0x4); // RSP
5291 index(0x4); // No Index
5292 scale(0x0); // No Scale
5293 disp($reg); // Stack Offset
5294 %}
5295 %}
5296 operand stackSlotL(sRegL reg)
5297 %{
5298 constraint(ALLOC_IN_RC(stack_slots));
5299 // No match rule because this operand is only generated in matching
5301 format %{ "[$reg]" %}
5302 interface(MEMORY_INTER) %{
5303 base(0x4); // RSP
5304 index(0x4); // No Index
5305 scale(0x0); // No Scale
5306 disp($reg); // Stack Offset
5307 %}
5308 %}
5310 //----------Conditional Branch Operands----------------------------------------
5311 // Comparison Op - This is the operation of the comparison, and is limited to
5312 // the following set of codes:
5313 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5314 //
5315 // Other attributes of the comparison, such as unsignedness, are specified
5316 // by the comparison instruction that sets a condition code flags register.
5317 // That result is represented by a flags operand whose subtype is appropriate
5318 // to the unsignedness (etc.) of the comparison.
5319 //
5320 // Later, the instruction which matches both the Comparison Op (a Bool) and
5321 // the flags (produced by the Cmp) specifies the coding of the comparison op
5322 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5324 // Comparision Code
5325 operand cmpOp()
5326 %{
5327 match(Bool);
5329 format %{ "" %}
5330 interface(COND_INTER) %{
5331 equal(0x4);
5332 not_equal(0x5);
5333 less(0xC);
5334 greater_equal(0xD);
5335 less_equal(0xE);
5336 greater(0xF);
5337 %}
5338 %}
5340 // Comparison Code, unsigned compare. Used by FP also, with
5341 // C2 (unordered) turned into GT or LT already. The other bits
5342 // C0 and C3 are turned into Carry & Zero flags.
5343 operand cmpOpU()
5344 %{
5345 match(Bool);
5347 format %{ "" %}
5348 interface(COND_INTER) %{
5349 equal(0x4);
5350 not_equal(0x5);
5351 less(0x2);
5352 greater_equal(0x3);
5353 less_equal(0x6);
5354 greater(0x7);
5355 %}
5356 %}
5359 //----------OPERAND CLASSES----------------------------------------------------
5360 // Operand Classes are groups of operands that are used as to simplify
5361 // instruction definitions by not requiring the AD writer to specify seperate
5362 // instructions for every form of operand when the instruction accepts
5363 // multiple operand types with the same basic encoding and format. The classic
5364 // case of this is memory operands.
5366 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5367 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5368 indIndexScaleOffsetComp);
5370 //----------PIPELINE-----------------------------------------------------------
5371 // Rules which define the behavior of the target architectures pipeline.
5372 pipeline %{
5374 //----------ATTRIBUTES---------------------------------------------------------
5375 attributes %{
5376 variable_size_instructions; // Fixed size instructions
5377 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5378 instruction_unit_size = 1; // An instruction is 1 bytes long
5379 instruction_fetch_unit_size = 16; // The processor fetches one line
5380 instruction_fetch_units = 1; // of 16 bytes
5382 // List of nop instructions
5383 nops( MachNop );
5384 %}
5386 //----------RESOURCES----------------------------------------------------------
5387 // Resources are the functional units available to the machine
5389 // Generic P2/P3 pipeline
5390 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5391 // 3 instructions decoded per cycle.
5392 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5393 // 3 ALU op, only ALU0 handles mul instructions.
5394 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5395 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5396 BR, FPU,
5397 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5399 //----------PIPELINE DESCRIPTION-----------------------------------------------
5400 // Pipeline Description specifies the stages in the machine's pipeline
5402 // Generic P2/P3 pipeline
5403 pipe_desc(S0, S1, S2, S3, S4, S5);
5405 //----------PIPELINE CLASSES---------------------------------------------------
5406 // Pipeline Classes describe the stages in which input and output are
5407 // referenced by the hardware pipeline.
5409 // Naming convention: ialu or fpu
5410 // Then: _reg
5411 // Then: _reg if there is a 2nd register
5412 // Then: _long if it's a pair of instructions implementing a long
5413 // Then: _fat if it requires the big decoder
5414 // Or: _mem if it requires the big decoder and a memory unit.
5416 // Integer ALU reg operation
5417 pipe_class ialu_reg(rRegI dst)
5418 %{
5419 single_instruction;
5420 dst : S4(write);
5421 dst : S3(read);
5422 DECODE : S0; // any decoder
5423 ALU : S3; // any alu
5424 %}
5426 // Long ALU reg operation
5427 pipe_class ialu_reg_long(rRegL dst)
5428 %{
5429 instruction_count(2);
5430 dst : S4(write);
5431 dst : S3(read);
5432 DECODE : S0(2); // any 2 decoders
5433 ALU : S3(2); // both alus
5434 %}
5436 // Integer ALU reg operation using big decoder
5437 pipe_class ialu_reg_fat(rRegI dst)
5438 %{
5439 single_instruction;
5440 dst : S4(write);
5441 dst : S3(read);
5442 D0 : S0; // big decoder only
5443 ALU : S3; // any alu
5444 %}
5446 // Long ALU reg operation using big decoder
5447 pipe_class ialu_reg_long_fat(rRegL dst)
5448 %{
5449 instruction_count(2);
5450 dst : S4(write);
5451 dst : S3(read);
5452 D0 : S0(2); // big decoder only; twice
5453 ALU : S3(2); // any 2 alus
5454 %}
5456 // Integer ALU reg-reg operation
5457 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5458 %{
5459 single_instruction;
5460 dst : S4(write);
5461 src : S3(read);
5462 DECODE : S0; // any decoder
5463 ALU : S3; // any alu
5464 %}
5466 // Long ALU reg-reg operation
5467 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5468 %{
5469 instruction_count(2);
5470 dst : S4(write);
5471 src : S3(read);
5472 DECODE : S0(2); // any 2 decoders
5473 ALU : S3(2); // both alus
5474 %}
5476 // Integer ALU reg-reg operation
5477 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5478 %{
5479 single_instruction;
5480 dst : S4(write);
5481 src : S3(read);
5482 D0 : S0; // big decoder only
5483 ALU : S3; // any alu
5484 %}
5486 // Long ALU reg-reg operation
5487 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5488 %{
5489 instruction_count(2);
5490 dst : S4(write);
5491 src : S3(read);
5492 D0 : S0(2); // big decoder only; twice
5493 ALU : S3(2); // both alus
5494 %}
5496 // Integer ALU reg-mem operation
5497 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5498 %{
5499 single_instruction;
5500 dst : S5(write);
5501 mem : S3(read);
5502 D0 : S0; // big decoder only
5503 ALU : S4; // any alu
5504 MEM : S3; // any mem
5505 %}
5507 // Integer mem operation (prefetch)
5508 pipe_class ialu_mem(memory mem)
5509 %{
5510 single_instruction;
5511 mem : S3(read);
5512 D0 : S0; // big decoder only
5513 MEM : S3; // any mem
5514 %}
5516 // Integer Store to Memory
5517 pipe_class ialu_mem_reg(memory mem, rRegI src)
5518 %{
5519 single_instruction;
5520 mem : S3(read);
5521 src : S5(read);
5522 D0 : S0; // big decoder only
5523 ALU : S4; // any alu
5524 MEM : S3;
5525 %}
5527 // // Long Store to Memory
5528 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5529 // %{
5530 // instruction_count(2);
5531 // mem : S3(read);
5532 // src : S5(read);
5533 // D0 : S0(2); // big decoder only; twice
5534 // ALU : S4(2); // any 2 alus
5535 // MEM : S3(2); // Both mems
5536 // %}
5538 // Integer Store to Memory
5539 pipe_class ialu_mem_imm(memory mem)
5540 %{
5541 single_instruction;
5542 mem : S3(read);
5543 D0 : S0; // big decoder only
5544 ALU : S4; // any alu
5545 MEM : S3;
5546 %}
5548 // Integer ALU0 reg-reg operation
5549 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5550 %{
5551 single_instruction;
5552 dst : S4(write);
5553 src : S3(read);
5554 D0 : S0; // Big decoder only
5555 ALU0 : S3; // only alu0
5556 %}
5558 // Integer ALU0 reg-mem operation
5559 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5560 %{
5561 single_instruction;
5562 dst : S5(write);
5563 mem : S3(read);
5564 D0 : S0; // big decoder only
5565 ALU0 : S4; // ALU0 only
5566 MEM : S3; // any mem
5567 %}
5569 // Integer ALU reg-reg operation
5570 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5571 %{
5572 single_instruction;
5573 cr : S4(write);
5574 src1 : S3(read);
5575 src2 : S3(read);
5576 DECODE : S0; // any decoder
5577 ALU : S3; // any alu
5578 %}
5580 // Integer ALU reg-imm operation
5581 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5582 %{
5583 single_instruction;
5584 cr : S4(write);
5585 src1 : S3(read);
5586 DECODE : S0; // any decoder
5587 ALU : S3; // any alu
5588 %}
5590 // Integer ALU reg-mem operation
5591 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5592 %{
5593 single_instruction;
5594 cr : S4(write);
5595 src1 : S3(read);
5596 src2 : S3(read);
5597 D0 : S0; // big decoder only
5598 ALU : S4; // any alu
5599 MEM : S3;
5600 %}
5602 // Conditional move reg-reg
5603 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5604 %{
5605 instruction_count(4);
5606 y : S4(read);
5607 q : S3(read);
5608 p : S3(read);
5609 DECODE : S0(4); // any decoder
5610 %}
5612 // Conditional move reg-reg
5613 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5614 %{
5615 single_instruction;
5616 dst : S4(write);
5617 src : S3(read);
5618 cr : S3(read);
5619 DECODE : S0; // any decoder
5620 %}
5622 // Conditional move reg-mem
5623 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5624 %{
5625 single_instruction;
5626 dst : S4(write);
5627 src : S3(read);
5628 cr : S3(read);
5629 DECODE : S0; // any decoder
5630 MEM : S3;
5631 %}
5633 // Conditional move reg-reg long
5634 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5635 %{
5636 single_instruction;
5637 dst : S4(write);
5638 src : S3(read);
5639 cr : S3(read);
5640 DECODE : S0(2); // any 2 decoders
5641 %}
5643 // XXX
5644 // // Conditional move double reg-reg
5645 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5646 // %{
5647 // single_instruction;
5648 // dst : S4(write);
5649 // src : S3(read);
5650 // cr : S3(read);
5651 // DECODE : S0; // any decoder
5652 // %}
5654 // Float reg-reg operation
5655 pipe_class fpu_reg(regD dst)
5656 %{
5657 instruction_count(2);
5658 dst : S3(read);
5659 DECODE : S0(2); // any 2 decoders
5660 FPU : S3;
5661 %}
5663 // Float reg-reg operation
5664 pipe_class fpu_reg_reg(regD dst, regD src)
5665 %{
5666 instruction_count(2);
5667 dst : S4(write);
5668 src : S3(read);
5669 DECODE : S0(2); // any 2 decoders
5670 FPU : S3;
5671 %}
5673 // Float reg-reg operation
5674 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5675 %{
5676 instruction_count(3);
5677 dst : S4(write);
5678 src1 : S3(read);
5679 src2 : S3(read);
5680 DECODE : S0(3); // any 3 decoders
5681 FPU : S3(2);
5682 %}
5684 // Float reg-reg operation
5685 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5686 %{
5687 instruction_count(4);
5688 dst : S4(write);
5689 src1 : S3(read);
5690 src2 : S3(read);
5691 src3 : S3(read);
5692 DECODE : S0(4); // any 3 decoders
5693 FPU : S3(2);
5694 %}
5696 // Float reg-reg operation
5697 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5698 %{
5699 instruction_count(4);
5700 dst : S4(write);
5701 src1 : S3(read);
5702 src2 : S3(read);
5703 src3 : S3(read);
5704 DECODE : S1(3); // any 3 decoders
5705 D0 : S0; // Big decoder only
5706 FPU : S3(2);
5707 MEM : S3;
5708 %}
5710 // Float reg-mem operation
5711 pipe_class fpu_reg_mem(regD dst, memory mem)
5712 %{
5713 instruction_count(2);
5714 dst : S5(write);
5715 mem : S3(read);
5716 D0 : S0; // big decoder only
5717 DECODE : S1; // any decoder for FPU POP
5718 FPU : S4;
5719 MEM : S3; // any mem
5720 %}
5722 // Float reg-mem operation
5723 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5724 %{
5725 instruction_count(3);
5726 dst : S5(write);
5727 src1 : S3(read);
5728 mem : S3(read);
5729 D0 : S0; // big decoder only
5730 DECODE : S1(2); // any decoder for FPU POP
5731 FPU : S4;
5732 MEM : S3; // any mem
5733 %}
5735 // Float mem-reg operation
5736 pipe_class fpu_mem_reg(memory mem, regD src)
5737 %{
5738 instruction_count(2);
5739 src : S5(read);
5740 mem : S3(read);
5741 DECODE : S0; // any decoder for FPU PUSH
5742 D0 : S1; // big decoder only
5743 FPU : S4;
5744 MEM : S3; // any mem
5745 %}
5747 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5748 %{
5749 instruction_count(3);
5750 src1 : S3(read);
5751 src2 : S3(read);
5752 mem : S3(read);
5753 DECODE : S0(2); // any decoder for FPU PUSH
5754 D0 : S1; // big decoder only
5755 FPU : S4;
5756 MEM : S3; // any mem
5757 %}
5759 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5760 %{
5761 instruction_count(3);
5762 src1 : S3(read);
5763 src2 : S3(read);
5764 mem : S4(read);
5765 DECODE : S0; // any decoder for FPU PUSH
5766 D0 : S0(2); // big decoder only
5767 FPU : S4;
5768 MEM : S3(2); // any mem
5769 %}
5771 pipe_class fpu_mem_mem(memory dst, memory src1)
5772 %{
5773 instruction_count(2);
5774 src1 : S3(read);
5775 dst : S4(read);
5776 D0 : S0(2); // big decoder only
5777 MEM : S3(2); // any mem
5778 %}
5780 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
5781 %{
5782 instruction_count(3);
5783 src1 : S3(read);
5784 src2 : S3(read);
5785 dst : S4(read);
5786 D0 : S0(3); // big decoder only
5787 FPU : S4;
5788 MEM : S3(3); // any mem
5789 %}
5791 pipe_class fpu_mem_reg_con(memory mem, regD src1)
5792 %{
5793 instruction_count(3);
5794 src1 : S4(read);
5795 mem : S4(read);
5796 DECODE : S0; // any decoder for FPU PUSH
5797 D0 : S0(2); // big decoder only
5798 FPU : S4;
5799 MEM : S3(2); // any mem
5800 %}
5802 // Float load constant
5803 pipe_class fpu_reg_con(regD dst)
5804 %{
5805 instruction_count(2);
5806 dst : S5(write);
5807 D0 : S0; // big decoder only for the load
5808 DECODE : S1; // any decoder for FPU POP
5809 FPU : S4;
5810 MEM : S3; // any mem
5811 %}
5813 // Float load constant
5814 pipe_class fpu_reg_reg_con(regD dst, regD src)
5815 %{
5816 instruction_count(3);
5817 dst : S5(write);
5818 src : S3(read);
5819 D0 : S0; // big decoder only for the load
5820 DECODE : S1(2); // any decoder for FPU POP
5821 FPU : S4;
5822 MEM : S3; // any mem
5823 %}
5825 // UnConditional branch
5826 pipe_class pipe_jmp(label labl)
5827 %{
5828 single_instruction;
5829 BR : S3;
5830 %}
5832 // Conditional branch
5833 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
5834 %{
5835 single_instruction;
5836 cr : S1(read);
5837 BR : S3;
5838 %}
5840 // Allocation idiom
5841 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
5842 %{
5843 instruction_count(1); force_serialization;
5844 fixed_latency(6);
5845 heap_ptr : S3(read);
5846 DECODE : S0(3);
5847 D0 : S2;
5848 MEM : S3;
5849 ALU : S3(2);
5850 dst : S5(write);
5851 BR : S5;
5852 %}
5854 // Generic big/slow expanded idiom
5855 pipe_class pipe_slow()
5856 %{
5857 instruction_count(10); multiple_bundles; force_serialization;
5858 fixed_latency(100);
5859 D0 : S0(2);
5860 MEM : S3(2);
5861 %}
5863 // The real do-nothing guy
5864 pipe_class empty()
5865 %{
5866 instruction_count(0);
5867 %}
5869 // Define the class for the Nop node
5870 define
5871 %{
5872 MachNop = empty;
5873 %}
5875 %}
5877 //----------INSTRUCTIONS-------------------------------------------------------
5878 //
5879 // match -- States which machine-independent subtree may be replaced
5880 // by this instruction.
5881 // ins_cost -- The estimated cost of this instruction is used by instruction
5882 // selection to identify a minimum cost tree of machine
5883 // instructions that matches a tree of machine-independent
5884 // instructions.
5885 // format -- A string providing the disassembly for this instruction.
5886 // The value of an instruction's operand may be inserted
5887 // by referring to it with a '$' prefix.
5888 // opcode -- Three instruction opcodes may be provided. These are referred
5889 // to within an encode class as $primary, $secondary, and $tertiary
5890 // rrspectively. The primary opcode is commonly used to
5891 // indicate the type of machine instruction, while secondary
5892 // and tertiary are often used for prefix options or addressing
5893 // modes.
5894 // ins_encode -- A list of encode classes with parameters. The encode class
5895 // name must have been defined in an 'enc_class' specification
5896 // in the encode section of the architecture description.
5899 //----------Load/Store/Move Instructions---------------------------------------
5900 //----------Load Instructions--------------------------------------------------
5902 // Load Byte (8 bit signed)
5903 instruct loadB(rRegI dst, memory mem)
5904 %{
5905 match(Set dst (LoadB mem));
5907 ins_cost(125);
5908 format %{ "movsbl $dst, $mem\t# byte" %}
5909 opcode(0x0F, 0xBE);
5910 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5911 ins_pipe(ialu_reg_mem);
5912 %}
5914 // Load Byte (8 bit signed) into long
5915 // instruct loadB2L(rRegL dst, memory mem)
5916 // %{
5917 // match(Set dst (ConvI2L (LoadB mem)));
5919 // ins_cost(125);
5920 // format %{ "movsbq $dst, $mem\t# byte -> long" %}
5921 // opcode(0x0F, 0xBE);
5922 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5923 // ins_pipe(ialu_reg_mem);
5924 // %}
5926 // Load Byte (8 bit UNsigned)
5927 instruct loadUB(rRegI dst, memory mem, immI_255 bytemask)
5928 %{
5929 match(Set dst (AndI (LoadB mem) bytemask));
5931 ins_cost(125);
5932 format %{ "movzbl $dst, $mem\t# ubyte" %}
5933 opcode(0x0F, 0xB6);
5934 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5935 ins_pipe(ialu_reg_mem);
5936 %}
5938 // Load Byte (8 bit UNsigned) into long
5939 // instruct loadUB2L(rRegL dst, memory mem, immI_255 bytemask)
5940 // %{
5941 // match(Set dst (ConvI2L (AndI (LoadB mem) bytemask)));
5943 // ins_cost(125);
5944 // format %{ "movzbl $dst, $mem\t# ubyte -> long" %}
5945 // opcode(0x0F, 0xB6);
5946 // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5947 // ins_pipe(ialu_reg_mem);
5948 // %}
5950 // Load Short (16 bit signed)
5951 instruct loadS(rRegI dst, memory mem)
5952 %{
5953 match(Set dst (LoadS mem));
5955 ins_cost(125); // XXX
5956 format %{ "movswl $dst, $mem\t# short" %}
5957 opcode(0x0F, 0xBF);
5958 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5959 ins_pipe(ialu_reg_mem);
5960 %}
5962 // Load Short (16 bit signed) into long
5963 // instruct loadS2L(rRegL dst, memory mem)
5964 // %{
5965 // match(Set dst (ConvI2L (LoadS mem)));
5967 // ins_cost(125); // XXX
5968 // format %{ "movswq $dst, $mem\t# short -> long" %}
5969 // opcode(0x0F, 0xBF);
5970 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5971 // ins_pipe(ialu_reg_mem);
5972 // %}
5974 // Load Char (16 bit UNsigned)
5975 instruct loadC(rRegI dst, memory mem)
5976 %{
5977 match(Set dst (LoadC mem));
5979 ins_cost(125);
5980 format %{ "movzwl $dst, $mem\t# char" %}
5981 opcode(0x0F, 0xB7);
5982 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5983 ins_pipe(ialu_reg_mem);
5984 %}
5986 // Load Char (16 bit UNsigned) into long
5987 // instruct loadC2L(rRegL dst, memory mem)
5988 // %{
5989 // match(Set dst (ConvI2L (LoadC mem)));
5991 // ins_cost(125);
5992 // format %{ "movzwl $dst, $mem\t# char -> long" %}
5993 // opcode(0x0F, 0xB7);
5994 // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
5995 // ins_pipe(ialu_reg_mem);
5996 // %}
5998 // Load Integer
5999 instruct loadI(rRegI dst, memory mem)
6000 %{
6001 match(Set dst (LoadI mem));
6003 ins_cost(125); // XXX
6004 format %{ "movl $dst, $mem\t# int" %}
6005 opcode(0x8B);
6006 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6007 ins_pipe(ialu_reg_mem);
6008 %}
6010 // Load Long
6011 instruct loadL(rRegL dst, memory mem)
6012 %{
6013 match(Set dst (LoadL mem));
6015 ins_cost(125); // XXX
6016 format %{ "movq $dst, $mem\t# long" %}
6017 opcode(0x8B);
6018 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6019 ins_pipe(ialu_reg_mem); // XXX
6020 %}
6022 // Load Range
6023 instruct loadRange(rRegI dst, memory mem)
6024 %{
6025 match(Set dst (LoadRange mem));
6027 ins_cost(125); // XXX
6028 format %{ "movl $dst, $mem\t# range" %}
6029 opcode(0x8B);
6030 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6031 ins_pipe(ialu_reg_mem);
6032 %}
6034 // Load Pointer
6035 instruct loadP(rRegP dst, memory mem)
6036 %{
6037 match(Set dst (LoadP mem));
6039 ins_cost(125); // XXX
6040 format %{ "movq $dst, $mem\t# ptr" %}
6041 opcode(0x8B);
6042 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6043 ins_pipe(ialu_reg_mem); // XXX
6044 %}
6046 // Load Compressed Pointer
6047 instruct loadN(rRegN dst, memory mem, rFlagsReg cr)
6048 %{
6049 match(Set dst (LoadN mem));
6050 effect(KILL cr);
6052 ins_cost(125); // XXX
6053 format %{ "movl $dst, $mem\t# compressed ptr" %}
6054 ins_encode %{
6055 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6056 Register dst = as_Register($dst$$reg);
6057 __ movl(dst, addr);
6058 %}
6059 ins_pipe(ialu_reg_mem); // XXX
6060 %}
6063 // Load Klass Pointer
6064 instruct loadKlass(rRegP dst, memory mem)
6065 %{
6066 match(Set dst (LoadKlass mem));
6067 predicate(!n->in(MemNode::Address)->bottom_type()->is_narrow());
6069 ins_cost(125); // XXX
6070 format %{ "movq $dst, $mem\t# class" %}
6071 opcode(0x8B);
6072 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6073 ins_pipe(ialu_reg_mem); // XXX
6074 %}
6076 // Load Klass Pointer
6077 instruct loadKlassComp(rRegP dst, memory mem)
6078 %{
6079 match(Set dst (LoadKlass mem));
6080 predicate(n->in(MemNode::Address)->bottom_type()->is_narrow());
6082 ins_cost(125); // XXX
6083 format %{ "movl $dst, $mem\t# compressed class" %}
6084 ins_encode %{
6085 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6086 Register dst = as_Register($dst$$reg);
6087 __ movl(dst, addr);
6088 // klass is never null in the header but this is generated for all
6089 // klass loads not just the _klass field in the header.
6090 __ decode_heap_oop(dst);
6091 %}
6092 ins_pipe(ialu_reg_mem); // XXX
6093 %}
6095 // Load Float
6096 instruct loadF(regF dst, memory mem)
6097 %{
6098 match(Set dst (LoadF mem));
6100 ins_cost(145); // XXX
6101 format %{ "movss $dst, $mem\t# float" %}
6102 opcode(0xF3, 0x0F, 0x10);
6103 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6104 ins_pipe(pipe_slow); // XXX
6105 %}
6107 // Load Double
6108 instruct loadD_partial(regD dst, memory mem)
6109 %{
6110 predicate(!UseXmmLoadAndClearUpper);
6111 match(Set dst (LoadD mem));
6113 ins_cost(145); // XXX
6114 format %{ "movlpd $dst, $mem\t# double" %}
6115 opcode(0x66, 0x0F, 0x12);
6116 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6117 ins_pipe(pipe_slow); // XXX
6118 %}
6120 instruct loadD(regD dst, memory mem)
6121 %{
6122 predicate(UseXmmLoadAndClearUpper);
6123 match(Set dst (LoadD mem));
6125 ins_cost(145); // XXX
6126 format %{ "movsd $dst, $mem\t# double" %}
6127 opcode(0xF2, 0x0F, 0x10);
6128 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6129 ins_pipe(pipe_slow); // XXX
6130 %}
6132 // Load Aligned Packed Byte to XMM register
6133 instruct loadA8B(regD dst, memory mem) %{
6134 match(Set dst (Load8B mem));
6135 ins_cost(125);
6136 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6137 ins_encode( movq_ld(dst, mem));
6138 ins_pipe( pipe_slow );
6139 %}
6141 // Load Aligned Packed Short to XMM register
6142 instruct loadA4S(regD dst, memory mem) %{
6143 match(Set dst (Load4S mem));
6144 ins_cost(125);
6145 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6146 ins_encode( movq_ld(dst, mem));
6147 ins_pipe( pipe_slow );
6148 %}
6150 // Load Aligned Packed Char to XMM register
6151 instruct loadA4C(regD dst, memory mem) %{
6152 match(Set dst (Load4C mem));
6153 ins_cost(125);
6154 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6155 ins_encode( movq_ld(dst, mem));
6156 ins_pipe( pipe_slow );
6157 %}
6159 // Load Aligned Packed Integer to XMM register
6160 instruct load2IU(regD dst, memory mem) %{
6161 match(Set dst (Load2I mem));
6162 ins_cost(125);
6163 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6164 ins_encode( movq_ld(dst, mem));
6165 ins_pipe( pipe_slow );
6166 %}
6168 // Load Aligned Packed Single to XMM
6169 instruct loadA2F(regD dst, memory mem) %{
6170 match(Set dst (Load2F mem));
6171 ins_cost(145);
6172 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6173 ins_encode( movq_ld(dst, mem));
6174 ins_pipe( pipe_slow );
6175 %}
6177 // Load Effective Address
6178 instruct leaP8(rRegP dst, indOffset8 mem)
6179 %{
6180 match(Set dst mem);
6182 ins_cost(110); // XXX
6183 format %{ "leaq $dst, $mem\t# ptr 8" %}
6184 opcode(0x8D);
6185 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6186 ins_pipe(ialu_reg_reg_fat);
6187 %}
6189 instruct leaP32(rRegP dst, indOffset32 mem)
6190 %{
6191 match(Set dst mem);
6193 ins_cost(110);
6194 format %{ "leaq $dst, $mem\t# ptr 32" %}
6195 opcode(0x8D);
6196 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6197 ins_pipe(ialu_reg_reg_fat);
6198 %}
6200 // instruct leaPIdx(rRegP dst, indIndex mem)
6201 // %{
6202 // match(Set dst mem);
6204 // ins_cost(110);
6205 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6206 // opcode(0x8D);
6207 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6208 // ins_pipe(ialu_reg_reg_fat);
6209 // %}
6211 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6212 %{
6213 match(Set dst mem);
6215 ins_cost(110);
6216 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6217 opcode(0x8D);
6218 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6219 ins_pipe(ialu_reg_reg_fat);
6220 %}
6222 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6223 %{
6224 match(Set dst mem);
6226 ins_cost(110);
6227 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6228 opcode(0x8D);
6229 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6230 ins_pipe(ialu_reg_reg_fat);
6231 %}
6233 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6234 %{
6235 match(Set dst mem);
6237 ins_cost(110);
6238 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6239 opcode(0x8D);
6240 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6241 ins_pipe(ialu_reg_reg_fat);
6242 %}
6244 instruct loadConI(rRegI dst, immI src)
6245 %{
6246 match(Set dst src);
6248 format %{ "movl $dst, $src\t# int" %}
6249 ins_encode(load_immI(dst, src));
6250 ins_pipe(ialu_reg_fat); // XXX
6251 %}
6253 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6254 %{
6255 match(Set dst src);
6256 effect(KILL cr);
6258 ins_cost(50);
6259 format %{ "xorl $dst, $dst\t# int" %}
6260 opcode(0x33); /* + rd */
6261 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6262 ins_pipe(ialu_reg);
6263 %}
6265 instruct loadConL(rRegL dst, immL src)
6266 %{
6267 match(Set dst src);
6269 ins_cost(150);
6270 format %{ "movq $dst, $src\t# long" %}
6271 ins_encode(load_immL(dst, src));
6272 ins_pipe(ialu_reg);
6273 %}
6275 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6276 %{
6277 match(Set dst src);
6278 effect(KILL cr);
6280 ins_cost(50);
6281 format %{ "xorl $dst, $dst\t# long" %}
6282 opcode(0x33); /* + rd */
6283 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6284 ins_pipe(ialu_reg); // XXX
6285 %}
6287 instruct loadConUL32(rRegL dst, immUL32 src)
6288 %{
6289 match(Set dst src);
6291 ins_cost(60);
6292 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6293 ins_encode(load_immUL32(dst, src));
6294 ins_pipe(ialu_reg);
6295 %}
6297 instruct loadConL32(rRegL dst, immL32 src)
6298 %{
6299 match(Set dst src);
6301 ins_cost(70);
6302 format %{ "movq $dst, $src\t# long (32-bit)" %}
6303 ins_encode(load_immL32(dst, src));
6304 ins_pipe(ialu_reg);
6305 %}
6307 instruct loadConP(rRegP dst, immP src)
6308 %{
6309 match(Set dst src);
6311 format %{ "movq $dst, $src\t# ptr" %}
6312 ins_encode(load_immP(dst, src));
6313 ins_pipe(ialu_reg_fat); // XXX
6314 %}
6316 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6317 %{
6318 match(Set dst src);
6319 effect(KILL cr);
6321 ins_cost(50);
6322 format %{ "xorl $dst, $dst\t# ptr" %}
6323 opcode(0x33); /* + rd */
6324 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6325 ins_pipe(ialu_reg);
6326 %}
6328 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6329 %{
6330 match(Set dst src);
6331 effect(KILL cr);
6333 ins_cost(60);
6334 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6335 ins_encode(load_immP31(dst, src));
6336 ins_pipe(ialu_reg);
6337 %}
6339 instruct loadConF(regF dst, immF src)
6340 %{
6341 match(Set dst src);
6342 ins_cost(125);
6344 format %{ "movss $dst, [$src]" %}
6345 ins_encode(load_conF(dst, src));
6346 ins_pipe(pipe_slow);
6347 %}
6349 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6350 match(Set dst src);
6351 effect(KILL cr);
6352 format %{ "xorq $dst, $src\t# compressed ptr" %}
6353 ins_encode %{
6354 Register dst = $dst$$Register;
6355 __ xorq(dst, dst);
6356 %}
6357 ins_pipe(ialu_reg);
6358 %}
6360 instruct loadConN(rRegN dst, immN src) %{
6361 match(Set dst src);
6363 ins_cost(125);
6364 format %{ "movl $dst, $src\t# compressed ptr" %}
6365 ins_encode %{
6366 address con = (address)$src$$constant;
6367 Register dst = $dst$$Register;
6368 if (con == NULL) {
6369 ShouldNotReachHere();
6370 } else {
6371 __ movoop(dst, (jobject)$src$$constant);
6372 __ encode_heap_oop_not_null(dst);
6373 }
6374 %}
6375 ins_pipe(ialu_reg_fat); // XXX
6376 %}
6378 instruct loadConF0(regF dst, immF0 src)
6379 %{
6380 match(Set dst src);
6381 ins_cost(100);
6383 format %{ "xorps $dst, $dst\t# float 0.0" %}
6384 opcode(0x0F, 0x57);
6385 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6386 ins_pipe(pipe_slow);
6387 %}
6389 // Use the same format since predicate() can not be used here.
6390 instruct loadConD(regD dst, immD src)
6391 %{
6392 match(Set dst src);
6393 ins_cost(125);
6395 format %{ "movsd $dst, [$src]" %}
6396 ins_encode(load_conD(dst, src));
6397 ins_pipe(pipe_slow);
6398 %}
6400 instruct loadConD0(regD dst, immD0 src)
6401 %{
6402 match(Set dst src);
6403 ins_cost(100);
6405 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6406 opcode(0x66, 0x0F, 0x57);
6407 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6408 ins_pipe(pipe_slow);
6409 %}
6411 instruct loadSSI(rRegI dst, stackSlotI src)
6412 %{
6413 match(Set dst src);
6415 ins_cost(125);
6416 format %{ "movl $dst, $src\t# int stk" %}
6417 opcode(0x8B);
6418 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6419 ins_pipe(ialu_reg_mem);
6420 %}
6422 instruct loadSSL(rRegL dst, stackSlotL src)
6423 %{
6424 match(Set dst src);
6426 ins_cost(125);
6427 format %{ "movq $dst, $src\t# long stk" %}
6428 opcode(0x8B);
6429 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6430 ins_pipe(ialu_reg_mem);
6431 %}
6433 instruct loadSSP(rRegP dst, stackSlotP src)
6434 %{
6435 match(Set dst src);
6437 ins_cost(125);
6438 format %{ "movq $dst, $src\t# ptr stk" %}
6439 opcode(0x8B);
6440 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6441 ins_pipe(ialu_reg_mem);
6442 %}
6444 instruct loadSSF(regF dst, stackSlotF src)
6445 %{
6446 match(Set dst src);
6448 ins_cost(125);
6449 format %{ "movss $dst, $src\t# float stk" %}
6450 opcode(0xF3, 0x0F, 0x10);
6451 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6452 ins_pipe(pipe_slow); // XXX
6453 %}
6455 // Use the same format since predicate() can not be used here.
6456 instruct loadSSD(regD dst, stackSlotD src)
6457 %{
6458 match(Set dst src);
6460 ins_cost(125);
6461 format %{ "movsd $dst, $src\t# double stk" %}
6462 ins_encode %{
6463 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6464 %}
6465 ins_pipe(pipe_slow); // XXX
6466 %}
6468 // Prefetch instructions.
6469 // Must be safe to execute with invalid address (cannot fault).
6471 instruct prefetchr( memory mem ) %{
6472 predicate(ReadPrefetchInstr==3);
6473 match(PrefetchRead mem);
6474 ins_cost(125);
6476 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6477 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6478 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6479 ins_pipe(ialu_mem);
6480 %}
6482 instruct prefetchrNTA( memory mem ) %{
6483 predicate(ReadPrefetchInstr==0);
6484 match(PrefetchRead mem);
6485 ins_cost(125);
6487 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6488 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6489 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6490 ins_pipe(ialu_mem);
6491 %}
6493 instruct prefetchrT0( memory mem ) %{
6494 predicate(ReadPrefetchInstr==1);
6495 match(PrefetchRead mem);
6496 ins_cost(125);
6498 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6499 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6500 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6501 ins_pipe(ialu_mem);
6502 %}
6504 instruct prefetchrT2( memory mem ) %{
6505 predicate(ReadPrefetchInstr==2);
6506 match(PrefetchRead mem);
6507 ins_cost(125);
6509 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6510 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6511 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6512 ins_pipe(ialu_mem);
6513 %}
6515 instruct prefetchw( memory mem ) %{
6516 predicate(AllocatePrefetchInstr==3);
6517 match(PrefetchWrite mem);
6518 ins_cost(125);
6520 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6521 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6522 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6523 ins_pipe(ialu_mem);
6524 %}
6526 instruct prefetchwNTA( memory mem ) %{
6527 predicate(AllocatePrefetchInstr==0);
6528 match(PrefetchWrite mem);
6529 ins_cost(125);
6531 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6532 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6533 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6534 ins_pipe(ialu_mem);
6535 %}
6537 instruct prefetchwT0( memory mem ) %{
6538 predicate(AllocatePrefetchInstr==1);
6539 match(PrefetchWrite mem);
6540 ins_cost(125);
6542 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6543 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6544 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6545 ins_pipe(ialu_mem);
6546 %}
6548 instruct prefetchwT2( memory mem ) %{
6549 predicate(AllocatePrefetchInstr==2);
6550 match(PrefetchWrite mem);
6551 ins_cost(125);
6553 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6554 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6555 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6556 ins_pipe(ialu_mem);
6557 %}
6559 //----------Store Instructions-------------------------------------------------
6561 // Store Byte
6562 instruct storeB(memory mem, rRegI src)
6563 %{
6564 match(Set mem (StoreB mem src));
6566 ins_cost(125); // XXX
6567 format %{ "movb $mem, $src\t# byte" %}
6568 opcode(0x88);
6569 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6570 ins_pipe(ialu_mem_reg);
6571 %}
6573 // Store Char/Short
6574 instruct storeC(memory mem, rRegI src)
6575 %{
6576 match(Set mem (StoreC mem src));
6578 ins_cost(125); // XXX
6579 format %{ "movw $mem, $src\t# char/short" %}
6580 opcode(0x89);
6581 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6582 ins_pipe(ialu_mem_reg);
6583 %}
6585 // Store Integer
6586 instruct storeI(memory mem, rRegI src)
6587 %{
6588 match(Set mem (StoreI mem src));
6590 ins_cost(125); // XXX
6591 format %{ "movl $mem, $src\t# int" %}
6592 opcode(0x89);
6593 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6594 ins_pipe(ialu_mem_reg);
6595 %}
6597 // Store Long
6598 instruct storeL(memory mem, rRegL src)
6599 %{
6600 match(Set mem (StoreL mem src));
6602 ins_cost(125); // XXX
6603 format %{ "movq $mem, $src\t# long" %}
6604 opcode(0x89);
6605 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6606 ins_pipe(ialu_mem_reg); // XXX
6607 %}
6609 // Store Pointer
6610 instruct storeP(memory mem, any_RegP src)
6611 %{
6612 match(Set mem (StoreP mem src));
6614 ins_cost(125); // XXX
6615 format %{ "movq $mem, $src\t# ptr" %}
6616 opcode(0x89);
6617 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6618 ins_pipe(ialu_mem_reg);
6619 %}
6621 // Store NULL Pointer, mark word, or other simple pointer constant.
6622 instruct storeImmP(memory mem, immP31 src)
6623 %{
6624 match(Set mem (StoreP mem src));
6626 ins_cost(125); // XXX
6627 format %{ "movq $mem, $src\t# ptr" %}
6628 opcode(0xC7); /* C7 /0 */
6629 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6630 ins_pipe(ialu_mem_imm);
6631 %}
6633 // Store Compressed Pointer
6634 instruct storeN(memory mem, rRegN src, rFlagsReg cr)
6635 %{
6636 match(Set mem (StoreN mem src));
6637 effect(KILL cr);
6639 ins_cost(125); // XXX
6640 format %{ "movl $mem, $src\t# ptr" %}
6641 ins_encode %{
6642 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6643 Register src = as_Register($src$$reg);
6644 __ movl(addr, src);
6645 %}
6646 ins_pipe(ialu_mem_reg);
6647 %}
6649 // Store Integer Immediate
6650 instruct storeImmI(memory mem, immI src)
6651 %{
6652 match(Set mem (StoreI mem src));
6654 ins_cost(150);
6655 format %{ "movl $mem, $src\t# int" %}
6656 opcode(0xC7); /* C7 /0 */
6657 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6658 ins_pipe(ialu_mem_imm);
6659 %}
6661 // Store Long Immediate
6662 instruct storeImmL(memory mem, immL32 src)
6663 %{
6664 match(Set mem (StoreL mem src));
6666 ins_cost(150);
6667 format %{ "movq $mem, $src\t# long" %}
6668 opcode(0xC7); /* C7 /0 */
6669 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6670 ins_pipe(ialu_mem_imm);
6671 %}
6673 // Store Short/Char Immediate
6674 instruct storeImmI16(memory mem, immI16 src)
6675 %{
6676 predicate(UseStoreImmI16);
6677 match(Set mem (StoreC mem src));
6679 ins_cost(150);
6680 format %{ "movw $mem, $src\t# short/char" %}
6681 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
6682 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
6683 ins_pipe(ialu_mem_imm);
6684 %}
6686 // Store Byte Immediate
6687 instruct storeImmB(memory mem, immI8 src)
6688 %{
6689 match(Set mem (StoreB mem src));
6691 ins_cost(150); // XXX
6692 format %{ "movb $mem, $src\t# byte" %}
6693 opcode(0xC6); /* C6 /0 */
6694 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
6695 ins_pipe(ialu_mem_imm);
6696 %}
6698 // Store Aligned Packed Byte XMM register to memory
6699 instruct storeA8B(memory mem, regD src) %{
6700 match(Set mem (Store8B mem src));
6701 ins_cost(145);
6702 format %{ "MOVQ $mem,$src\t! packed8B" %}
6703 ins_encode( movq_st(mem, src));
6704 ins_pipe( pipe_slow );
6705 %}
6707 // Store Aligned Packed Char/Short XMM register to memory
6708 instruct storeA4C(memory mem, regD src) %{
6709 match(Set mem (Store4C mem src));
6710 ins_cost(145);
6711 format %{ "MOVQ $mem,$src\t! packed4C" %}
6712 ins_encode( movq_st(mem, src));
6713 ins_pipe( pipe_slow );
6714 %}
6716 // Store Aligned Packed Integer XMM register to memory
6717 instruct storeA2I(memory mem, regD src) %{
6718 match(Set mem (Store2I mem src));
6719 ins_cost(145);
6720 format %{ "MOVQ $mem,$src\t! packed2I" %}
6721 ins_encode( movq_st(mem, src));
6722 ins_pipe( pipe_slow );
6723 %}
6725 // Store CMS card-mark Immediate
6726 instruct storeImmCM0(memory mem, immI0 src)
6727 %{
6728 match(Set mem (StoreCM mem src));
6730 ins_cost(150); // XXX
6731 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
6732 opcode(0xC6); /* C6 /0 */
6733 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
6734 ins_pipe(ialu_mem_imm);
6735 %}
6737 // Store Aligned Packed Single Float XMM register to memory
6738 instruct storeA2F(memory mem, regD src) %{
6739 match(Set mem (Store2F mem src));
6740 ins_cost(145);
6741 format %{ "MOVQ $mem,$src\t! packed2F" %}
6742 ins_encode( movq_st(mem, src));
6743 ins_pipe( pipe_slow );
6744 %}
6746 // Store Float
6747 instruct storeF(memory mem, regF src)
6748 %{
6749 match(Set mem (StoreF mem src));
6751 ins_cost(95); // XXX
6752 format %{ "movss $mem, $src\t# float" %}
6753 opcode(0xF3, 0x0F, 0x11);
6754 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
6755 ins_pipe(pipe_slow); // XXX
6756 %}
6758 // Store immediate Float value (it is faster than store from XMM register)
6759 instruct storeF_imm(memory mem, immF src)
6760 %{
6761 match(Set mem (StoreF mem src));
6763 ins_cost(50);
6764 format %{ "movl $mem, $src\t# float" %}
6765 opcode(0xC7); /* C7 /0 */
6766 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
6767 ins_pipe(ialu_mem_imm);
6768 %}
6770 // Store Double
6771 instruct storeD(memory mem, regD src)
6772 %{
6773 match(Set mem (StoreD mem src));
6775 ins_cost(95); // XXX
6776 format %{ "movsd $mem, $src\t# double" %}
6777 opcode(0xF2, 0x0F, 0x11);
6778 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
6779 ins_pipe(pipe_slow); // XXX
6780 %}
6782 // Store immediate double 0.0 (it is faster than store from XMM register)
6783 instruct storeD0_imm(memory mem, immD0 src)
6784 %{
6785 match(Set mem (StoreD mem src));
6787 ins_cost(50);
6788 format %{ "movq $mem, $src\t# double 0." %}
6789 opcode(0xC7); /* C7 /0 */
6790 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
6791 ins_pipe(ialu_mem_imm);
6792 %}
6794 instruct storeSSI(stackSlotI dst, rRegI src)
6795 %{
6796 match(Set dst src);
6798 ins_cost(100);
6799 format %{ "movl $dst, $src\t# int stk" %}
6800 opcode(0x89);
6801 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
6802 ins_pipe( ialu_mem_reg );
6803 %}
6805 instruct storeSSL(stackSlotL dst, rRegL src)
6806 %{
6807 match(Set dst src);
6809 ins_cost(100);
6810 format %{ "movq $dst, $src\t# long stk" %}
6811 opcode(0x89);
6812 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
6813 ins_pipe(ialu_mem_reg);
6814 %}
6816 instruct storeSSP(stackSlotP dst, rRegP src)
6817 %{
6818 match(Set dst src);
6820 ins_cost(100);
6821 format %{ "movq $dst, $src\t# ptr stk" %}
6822 opcode(0x89);
6823 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
6824 ins_pipe(ialu_mem_reg);
6825 %}
6827 instruct storeSSF(stackSlotF dst, regF src)
6828 %{
6829 match(Set dst src);
6831 ins_cost(95); // XXX
6832 format %{ "movss $dst, $src\t# float stk" %}
6833 opcode(0xF3, 0x0F, 0x11);
6834 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
6835 ins_pipe(pipe_slow); // XXX
6836 %}
6838 instruct storeSSD(stackSlotD dst, regD src)
6839 %{
6840 match(Set dst src);
6842 ins_cost(95); // XXX
6843 format %{ "movsd $dst, $src\t# double stk" %}
6844 opcode(0xF2, 0x0F, 0x11);
6845 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
6846 ins_pipe(pipe_slow); // XXX
6847 %}
6849 //----------BSWAP Instructions-------------------------------------------------
6850 instruct bytes_reverse_int(rRegI dst) %{
6851 match(Set dst (ReverseBytesI dst));
6853 format %{ "bswapl $dst" %}
6854 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
6855 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
6856 ins_pipe( ialu_reg );
6857 %}
6859 instruct bytes_reverse_long(rRegL dst) %{
6860 match(Set dst (ReverseBytesL dst));
6862 format %{ "bswapq $dst" %}
6864 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
6865 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
6866 ins_pipe( ialu_reg);
6867 %}
6869 instruct loadI_reversed(rRegI dst, memory src) %{
6870 match(Set dst (ReverseBytesI (LoadI src)));
6872 format %{ "bswap_movl $dst, $src" %}
6873 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
6874 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
6875 ins_pipe( ialu_reg_mem );
6876 %}
6878 instruct loadL_reversed(rRegL dst, memory src) %{
6879 match(Set dst (ReverseBytesL (LoadL src)));
6881 format %{ "bswap_movq $dst, $src" %}
6882 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
6883 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
6884 ins_pipe( ialu_reg_mem );
6885 %}
6887 instruct storeI_reversed(memory dst, rRegI src) %{
6888 match(Set dst (StoreI dst (ReverseBytesI src)));
6890 format %{ "movl_bswap $dst, $src" %}
6891 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
6892 ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
6893 ins_pipe( ialu_mem_reg );
6894 %}
6896 instruct storeL_reversed(memory dst, rRegL src) %{
6897 match(Set dst (StoreL dst (ReverseBytesL src)));
6899 format %{ "movq_bswap $dst, $src" %}
6900 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
6901 ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
6902 ins_pipe( ialu_mem_reg );
6903 %}
6905 //----------MemBar Instructions-----------------------------------------------
6906 // Memory barrier flavors
6908 instruct membar_acquire()
6909 %{
6910 match(MemBarAcquire);
6911 ins_cost(0);
6913 size(0);
6914 format %{ "MEMBAR-acquire" %}
6915 ins_encode();
6916 ins_pipe(empty);
6917 %}
6919 instruct membar_acquire_lock()
6920 %{
6921 match(MemBarAcquire);
6922 predicate(Matcher::prior_fast_lock(n));
6923 ins_cost(0);
6925 size(0);
6926 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
6927 ins_encode();
6928 ins_pipe(empty);
6929 %}
6931 instruct membar_release()
6932 %{
6933 match(MemBarRelease);
6934 ins_cost(0);
6936 size(0);
6937 format %{ "MEMBAR-release" %}
6938 ins_encode();
6939 ins_pipe(empty);
6940 %}
6942 instruct membar_release_lock()
6943 %{
6944 match(MemBarRelease);
6945 predicate(Matcher::post_fast_unlock(n));
6946 ins_cost(0);
6948 size(0);
6949 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
6950 ins_encode();
6951 ins_pipe(empty);
6952 %}
6954 instruct membar_volatile()
6955 %{
6956 match(MemBarVolatile);
6957 ins_cost(400);
6959 format %{ "MEMBAR-volatile" %}
6960 ins_encode(enc_membar_volatile);
6961 ins_pipe(pipe_slow);
6962 %}
6964 instruct unnecessary_membar_volatile()
6965 %{
6966 match(MemBarVolatile);
6967 predicate(Matcher::post_store_load_barrier(n));
6968 ins_cost(0);
6970 size(0);
6971 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
6972 ins_encode();
6973 ins_pipe(empty);
6974 %}
6976 //----------Move Instructions--------------------------------------------------
6978 instruct castX2P(rRegP dst, rRegL src)
6979 %{
6980 match(Set dst (CastX2P src));
6982 format %{ "movq $dst, $src\t# long->ptr" %}
6983 ins_encode(enc_copy_wide(dst, src));
6984 ins_pipe(ialu_reg_reg); // XXX
6985 %}
6987 instruct castP2X(rRegL dst, rRegP src)
6988 %{
6989 match(Set dst (CastP2X src));
6991 format %{ "movq $dst, $src\t# ptr -> long" %}
6992 ins_encode(enc_copy_wide(dst, src));
6993 ins_pipe(ialu_reg_reg); // XXX
6994 %}
6997 // Convert oop pointer into compressed form
6998 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
6999 match(Set dst (EncodeP src));
7000 effect(KILL cr);
7001 format %{ "encode_heap_oop $dst,$src" %}
7002 ins_encode %{
7003 Register s = $src$$Register;
7004 Register d = $dst$$Register;
7005 if (s != d) {
7006 __ movq(d, s);
7007 }
7008 __ encode_heap_oop(d);
7009 %}
7010 ins_pipe(ialu_reg_long);
7011 %}
7013 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7014 match(Set dst (DecodeN src));
7015 effect(KILL cr);
7016 format %{ "decode_heap_oop $dst,$src" %}
7017 ins_encode %{
7018 Register s = $src$$Register;
7019 Register d = $dst$$Register;
7020 if (s != d) {
7021 __ movq(d, s);
7022 }
7023 __ decode_heap_oop(d);
7024 %}
7025 ins_pipe(ialu_reg_long);
7026 %}
7029 //----------Conditional Move---------------------------------------------------
7030 // Jump
7031 // dummy instruction for generating temp registers
7032 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7033 match(Jump (LShiftL switch_val shift));
7034 ins_cost(350);
7035 predicate(false);
7036 effect(TEMP dest);
7038 format %{ "leaq $dest, table_base\n\t"
7039 "jmp [$dest + $switch_val << $shift]\n\t" %}
7040 ins_encode(jump_enc_offset(switch_val, shift, dest));
7041 ins_pipe(pipe_jmp);
7042 ins_pc_relative(1);
7043 %}
7045 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7046 match(Jump (AddL (LShiftL switch_val shift) offset));
7047 ins_cost(350);
7048 effect(TEMP dest);
7050 format %{ "leaq $dest, table_base\n\t"
7051 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7052 ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
7053 ins_pipe(pipe_jmp);
7054 ins_pc_relative(1);
7055 %}
7057 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7058 match(Jump switch_val);
7059 ins_cost(350);
7060 effect(TEMP dest);
7062 format %{ "leaq $dest, table_base\n\t"
7063 "jmp [$dest + $switch_val]\n\t" %}
7064 ins_encode(jump_enc(switch_val, dest));
7065 ins_pipe(pipe_jmp);
7066 ins_pc_relative(1);
7067 %}
7069 // Conditional move
7070 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7071 %{
7072 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7074 ins_cost(200); // XXX
7075 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7076 opcode(0x0F, 0x40);
7077 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7078 ins_pipe(pipe_cmov_reg);
7079 %}
7081 instruct cmovI_regU(rRegI dst, rRegI src, rFlagsRegU cr, cmpOpU cop)
7082 %{
7083 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7085 ins_cost(200); // XXX
7086 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7087 opcode(0x0F, 0x40);
7088 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7089 ins_pipe(pipe_cmov_reg);
7090 %}
7092 // Conditional move
7093 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src)
7094 %{
7095 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7097 ins_cost(250); // XXX
7098 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7099 opcode(0x0F, 0x40);
7100 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7101 ins_pipe(pipe_cmov_mem);
7102 %}
7104 // Conditional move
7105 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7106 %{
7107 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7109 ins_cost(250); // XXX
7110 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7111 opcode(0x0F, 0x40);
7112 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7113 ins_pipe(pipe_cmov_mem);
7114 %}
7116 // Conditional move
7117 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7118 %{
7119 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7121 ins_cost(200); // XXX
7122 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7123 opcode(0x0F, 0x40);
7124 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7125 ins_pipe(pipe_cmov_reg); // XXX
7126 %}
7128 // Conditional move
7129 instruct cmovP_regU(rRegP dst, rRegP src, rFlagsRegU cr, cmpOpU cop)
7130 %{
7131 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7133 ins_cost(200); // XXX
7134 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7135 opcode(0x0F, 0x40);
7136 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7137 ins_pipe(pipe_cmov_reg); // XXX
7138 %}
7140 // DISABLED: Requires the ADLC to emit a bottom_type call that
7141 // correctly meets the two pointer arguments; one is an incoming
7142 // register but the other is a memory operand. ALSO appears to
7143 // be buggy with implicit null checks.
7144 //
7145 //// Conditional move
7146 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7147 //%{
7148 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7149 // ins_cost(250);
7150 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7151 // opcode(0x0F,0x40);
7152 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7153 // ins_pipe( pipe_cmov_mem );
7154 //%}
7155 //
7156 //// Conditional move
7157 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7158 //%{
7159 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7160 // ins_cost(250);
7161 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7162 // opcode(0x0F,0x40);
7163 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7164 // ins_pipe( pipe_cmov_mem );
7165 //%}
7167 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7168 %{
7169 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7171 ins_cost(200); // XXX
7172 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7173 opcode(0x0F, 0x40);
7174 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7175 ins_pipe(pipe_cmov_reg); // XXX
7176 %}
7178 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7179 %{
7180 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7182 ins_cost(200); // XXX
7183 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7184 opcode(0x0F, 0x40);
7185 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7186 ins_pipe(pipe_cmov_mem); // XXX
7187 %}
7189 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7190 %{
7191 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7193 ins_cost(200); // XXX
7194 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7195 opcode(0x0F, 0x40);
7196 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7197 ins_pipe(pipe_cmov_reg); // XXX
7198 %}
7200 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7201 %{
7202 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7204 ins_cost(200); // XXX
7205 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7206 opcode(0x0F, 0x40);
7207 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7208 ins_pipe(pipe_cmov_mem); // XXX
7209 %}
7211 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7212 %{
7213 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7215 ins_cost(200); // XXX
7216 format %{ "jn$cop skip\t# signed cmove float\n\t"
7217 "movss $dst, $src\n"
7218 "skip:" %}
7219 ins_encode(enc_cmovf_branch(cop, dst, src));
7220 ins_pipe(pipe_slow);
7221 %}
7223 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7224 // %{
7225 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7227 // ins_cost(200); // XXX
7228 // format %{ "jn$cop skip\t# signed cmove float\n\t"
7229 // "movss $dst, $src\n"
7230 // "skip:" %}
7231 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7232 // ins_pipe(pipe_slow);
7233 // %}
7235 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7236 %{
7237 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7239 ins_cost(200); // XXX
7240 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
7241 "movss $dst, $src\n"
7242 "skip:" %}
7243 ins_encode(enc_cmovf_branch(cop, dst, src));
7244 ins_pipe(pipe_slow);
7245 %}
7247 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7248 %{
7249 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7251 ins_cost(200); // XXX
7252 format %{ "jn$cop skip\t# signed cmove double\n\t"
7253 "movsd $dst, $src\n"
7254 "skip:" %}
7255 ins_encode(enc_cmovd_branch(cop, dst, src));
7256 ins_pipe(pipe_slow);
7257 %}
7259 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7260 %{
7261 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7263 ins_cost(200); // XXX
7264 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
7265 "movsd $dst, $src\n"
7266 "skip:" %}
7267 ins_encode(enc_cmovd_branch(cop, dst, src));
7268 ins_pipe(pipe_slow);
7269 %}
7271 //----------Arithmetic Instructions--------------------------------------------
7272 //----------Addition Instructions----------------------------------------------
7274 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7275 %{
7276 match(Set dst (AddI dst src));
7277 effect(KILL cr);
7279 format %{ "addl $dst, $src\t# int" %}
7280 opcode(0x03);
7281 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7282 ins_pipe(ialu_reg_reg);
7283 %}
7285 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7286 %{
7287 match(Set dst (AddI dst src));
7288 effect(KILL cr);
7290 format %{ "addl $dst, $src\t# int" %}
7291 opcode(0x81, 0x00); /* /0 id */
7292 ins_encode(OpcSErm(dst, src), Con8or32(src));
7293 ins_pipe( ialu_reg );
7294 %}
7296 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7297 %{
7298 match(Set dst (AddI dst (LoadI src)));
7299 effect(KILL cr);
7301 ins_cost(125); // XXX
7302 format %{ "addl $dst, $src\t# int" %}
7303 opcode(0x03);
7304 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7305 ins_pipe(ialu_reg_mem);
7306 %}
7308 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7309 %{
7310 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7311 effect(KILL cr);
7313 ins_cost(150); // XXX
7314 format %{ "addl $dst, $src\t# int" %}
7315 opcode(0x01); /* Opcode 01 /r */
7316 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7317 ins_pipe(ialu_mem_reg);
7318 %}
7320 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7321 %{
7322 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7323 effect(KILL cr);
7325 ins_cost(125); // XXX
7326 format %{ "addl $dst, $src\t# int" %}
7327 opcode(0x81); /* Opcode 81 /0 id */
7328 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7329 ins_pipe(ialu_mem_imm);
7330 %}
7332 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7333 %{
7334 predicate(UseIncDec);
7335 match(Set dst (AddI dst src));
7336 effect(KILL cr);
7338 format %{ "incl $dst\t# int" %}
7339 opcode(0xFF, 0x00); // FF /0
7340 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7341 ins_pipe(ialu_reg);
7342 %}
7344 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
7345 %{
7346 predicate(UseIncDec);
7347 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7348 effect(KILL cr);
7350 ins_cost(125); // XXX
7351 format %{ "incl $dst\t# int" %}
7352 opcode(0xFF); /* Opcode FF /0 */
7353 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
7354 ins_pipe(ialu_mem_imm);
7355 %}
7357 // XXX why does that use AddI
7358 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
7359 %{
7360 predicate(UseIncDec);
7361 match(Set dst (AddI dst src));
7362 effect(KILL cr);
7364 format %{ "decl $dst\t# int" %}
7365 opcode(0xFF, 0x01); // FF /1
7366 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7367 ins_pipe(ialu_reg);
7368 %}
7370 // XXX why does that use AddI
7371 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
7372 %{
7373 predicate(UseIncDec);
7374 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7375 effect(KILL cr);
7377 ins_cost(125); // XXX
7378 format %{ "decl $dst\t# int" %}
7379 opcode(0xFF); /* Opcode FF /1 */
7380 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
7381 ins_pipe(ialu_mem_imm);
7382 %}
7384 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
7385 %{
7386 match(Set dst (AddI src0 src1));
7388 ins_cost(110);
7389 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
7390 opcode(0x8D); /* 0x8D /r */
7391 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7392 ins_pipe(ialu_reg_reg);
7393 %}
7395 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
7396 %{
7397 match(Set dst (AddL dst src));
7398 effect(KILL cr);
7400 format %{ "addq $dst, $src\t# long" %}
7401 opcode(0x03);
7402 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7403 ins_pipe(ialu_reg_reg);
7404 %}
7406 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
7407 %{
7408 match(Set dst (AddL dst src));
7409 effect(KILL cr);
7411 format %{ "addq $dst, $src\t# long" %}
7412 opcode(0x81, 0x00); /* /0 id */
7413 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7414 ins_pipe( ialu_reg );
7415 %}
7417 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
7418 %{
7419 match(Set dst (AddL dst (LoadL src)));
7420 effect(KILL cr);
7422 ins_cost(125); // XXX
7423 format %{ "addq $dst, $src\t# long" %}
7424 opcode(0x03);
7425 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
7426 ins_pipe(ialu_reg_mem);
7427 %}
7429 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
7430 %{
7431 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7432 effect(KILL cr);
7434 ins_cost(150); // XXX
7435 format %{ "addq $dst, $src\t# long" %}
7436 opcode(0x01); /* Opcode 01 /r */
7437 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7438 ins_pipe(ialu_mem_reg);
7439 %}
7441 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
7442 %{
7443 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7444 effect(KILL cr);
7446 ins_cost(125); // XXX
7447 format %{ "addq $dst, $src\t# long" %}
7448 opcode(0x81); /* Opcode 81 /0 id */
7449 ins_encode(REX_mem_wide(dst),
7450 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7451 ins_pipe(ialu_mem_imm);
7452 %}
7454 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
7455 %{
7456 predicate(UseIncDec);
7457 match(Set dst (AddL dst src));
7458 effect(KILL cr);
7460 format %{ "incq $dst\t# long" %}
7461 opcode(0xFF, 0x00); // FF /0
7462 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7463 ins_pipe(ialu_reg);
7464 %}
7466 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
7467 %{
7468 predicate(UseIncDec);
7469 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7470 effect(KILL cr);
7472 ins_cost(125); // XXX
7473 format %{ "incq $dst\t# long" %}
7474 opcode(0xFF); /* Opcode FF /0 */
7475 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
7476 ins_pipe(ialu_mem_imm);
7477 %}
7479 // XXX why does that use AddL
7480 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
7481 %{
7482 predicate(UseIncDec);
7483 match(Set dst (AddL dst src));
7484 effect(KILL cr);
7486 format %{ "decq $dst\t# long" %}
7487 opcode(0xFF, 0x01); // FF /1
7488 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7489 ins_pipe(ialu_reg);
7490 %}
7492 // XXX why does that use AddL
7493 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
7494 %{
7495 predicate(UseIncDec);
7496 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7497 effect(KILL cr);
7499 ins_cost(125); // XXX
7500 format %{ "decq $dst\t# long" %}
7501 opcode(0xFF); /* Opcode FF /1 */
7502 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
7503 ins_pipe(ialu_mem_imm);
7504 %}
7506 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
7507 %{
7508 match(Set dst (AddL src0 src1));
7510 ins_cost(110);
7511 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
7512 opcode(0x8D); /* 0x8D /r */
7513 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7514 ins_pipe(ialu_reg_reg);
7515 %}
7517 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
7518 %{
7519 match(Set dst (AddP dst src));
7520 effect(KILL cr);
7522 format %{ "addq $dst, $src\t# ptr" %}
7523 opcode(0x03);
7524 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7525 ins_pipe(ialu_reg_reg);
7526 %}
7528 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
7529 %{
7530 match(Set dst (AddP dst src));
7531 effect(KILL cr);
7533 format %{ "addq $dst, $src\t# ptr" %}
7534 opcode(0x81, 0x00); /* /0 id */
7535 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7536 ins_pipe( ialu_reg );
7537 %}
7539 // XXX addP mem ops ????
7541 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
7542 %{
7543 match(Set dst (AddP src0 src1));
7545 ins_cost(110);
7546 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
7547 opcode(0x8D); /* 0x8D /r */
7548 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
7549 ins_pipe(ialu_reg_reg);
7550 %}
7552 instruct checkCastPP(rRegP dst)
7553 %{
7554 match(Set dst (CheckCastPP dst));
7556 size(0);
7557 format %{ "# checkcastPP of $dst" %}
7558 ins_encode(/* empty encoding */);
7559 ins_pipe(empty);
7560 %}
7562 instruct castPP(rRegP dst)
7563 %{
7564 match(Set dst (CastPP dst));
7566 size(0);
7567 format %{ "# castPP of $dst" %}
7568 ins_encode(/* empty encoding */);
7569 ins_pipe(empty);
7570 %}
7572 instruct castII(rRegI dst)
7573 %{
7574 match(Set dst (CastII dst));
7576 size(0);
7577 format %{ "# castII of $dst" %}
7578 ins_encode(/* empty encoding */);
7579 ins_cost(0);
7580 ins_pipe(empty);
7581 %}
7583 // LoadP-locked same as a regular LoadP when used with compare-swap
7584 instruct loadPLocked(rRegP dst, memory mem)
7585 %{
7586 match(Set dst (LoadPLocked mem));
7588 ins_cost(125); // XXX
7589 format %{ "movq $dst, $mem\t# ptr locked" %}
7590 opcode(0x8B);
7591 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
7592 ins_pipe(ialu_reg_mem); // XXX
7593 %}
7595 // LoadL-locked - same as a regular LoadL when used with compare-swap
7596 instruct loadLLocked(rRegL dst, memory mem)
7597 %{
7598 match(Set dst (LoadLLocked mem));
7600 ins_cost(125); // XXX
7601 format %{ "movq $dst, $mem\t# long locked" %}
7602 opcode(0x8B);
7603 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
7604 ins_pipe(ialu_reg_mem); // XXX
7605 %}
7607 // Conditional-store of the updated heap-top.
7608 // Used during allocation of the shared heap.
7609 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
7611 instruct storePConditional(memory heap_top_ptr,
7612 rax_RegP oldval, rRegP newval,
7613 rFlagsReg cr)
7614 %{
7615 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
7617 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
7618 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
7619 opcode(0x0F, 0xB1);
7620 ins_encode(lock_prefix,
7621 REX_reg_mem_wide(newval, heap_top_ptr),
7622 OpcP, OpcS,
7623 reg_mem(newval, heap_top_ptr));
7624 ins_pipe(pipe_cmpxchg);
7625 %}
7627 // Conditional-store of a long value
7628 // Returns a boolean value (0/1) on success. Implemented with a
7629 // CMPXCHG8 on Intel. mem_ptr can actually be in either RSI or RDI
7631 instruct storeLConditional(rRegI res,
7632 memory mem_ptr,
7633 rax_RegL oldval, rRegL newval,
7634 rFlagsReg cr)
7635 %{
7636 match(Set res (StoreLConditional mem_ptr (Binary oldval newval)));
7637 effect(KILL cr);
7639 format %{ "cmpxchgq $mem_ptr, $newval\t# (long) "
7640 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7641 "sete $res\n\t"
7642 "movzbl $res, $res" %}
7643 opcode(0x0F, 0xB1);
7644 ins_encode(lock_prefix,
7645 REX_reg_mem_wide(newval, mem_ptr),
7646 OpcP, OpcS,
7647 reg_mem(newval, mem_ptr),
7648 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7649 REX_reg_breg(res, res), // movzbl
7650 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7651 ins_pipe(pipe_cmpxchg);
7652 %}
7654 // Conditional-store of a long value
7655 // ZF flag is set on success, reset otherwise. Implemented with a
7656 // CMPXCHG8 on Intel. mem_ptr can actually be in either RSI or RDI
7657 instruct storeLConditional_flags(memory mem_ptr,
7658 rax_RegL oldval, rRegL newval,
7659 rFlagsReg cr,
7660 immI0 zero)
7661 %{
7662 match(Set cr (CmpI (StoreLConditional mem_ptr (Binary oldval newval)) zero));
7664 format %{ "cmpxchgq $mem_ptr, $newval\t# (long) "
7665 "If rax == $mem_ptr then store $newval into $mem_ptr" %}
7666 opcode(0x0F, 0xB1);
7667 ins_encode(lock_prefix,
7668 REX_reg_mem_wide(newval, mem_ptr),
7669 OpcP, OpcS,
7670 reg_mem(newval, mem_ptr));
7671 ins_pipe(pipe_cmpxchg);
7672 %}
7674 instruct compareAndSwapP(rRegI res,
7675 memory mem_ptr,
7676 rax_RegP oldval, rRegP newval,
7677 rFlagsReg cr)
7678 %{
7679 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7680 effect(KILL cr, KILL oldval);
7682 format %{ "cmpxchgq $mem_ptr,$newval\t# "
7683 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7684 "sete $res\n\t"
7685 "movzbl $res, $res" %}
7686 opcode(0x0F, 0xB1);
7687 ins_encode(lock_prefix,
7688 REX_reg_mem_wide(newval, mem_ptr),
7689 OpcP, OpcS,
7690 reg_mem(newval, mem_ptr),
7691 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7692 REX_reg_breg(res, res), // movzbl
7693 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7694 ins_pipe( pipe_cmpxchg );
7695 %}
7697 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7698 instruct compareAndSwapL(rRegI res,
7699 memory mem_ptr,
7700 rax_RegL oldval, rRegL newval,
7701 rFlagsReg cr)
7702 %{
7703 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7704 effect(KILL cr, KILL oldval);
7706 format %{ "cmpxchgq $mem_ptr,$newval\t# "
7707 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7708 "sete $res\n\t"
7709 "movzbl $res, $res" %}
7710 opcode(0x0F, 0xB1);
7711 ins_encode(lock_prefix,
7712 REX_reg_mem_wide(newval, mem_ptr),
7713 OpcP, OpcS,
7714 reg_mem(newval, mem_ptr),
7715 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7716 REX_reg_breg(res, res), // movzbl
7717 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7718 ins_pipe( pipe_cmpxchg );
7719 %}
7721 instruct compareAndSwapI(rRegI res,
7722 memory mem_ptr,
7723 rax_RegI oldval, rRegI newval,
7724 rFlagsReg cr)
7725 %{
7726 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7727 effect(KILL cr, KILL oldval);
7729 format %{ "cmpxchgl $mem_ptr,$newval\t# "
7730 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7731 "sete $res\n\t"
7732 "movzbl $res, $res" %}
7733 opcode(0x0F, 0xB1);
7734 ins_encode(lock_prefix,
7735 REX_reg_mem(newval, mem_ptr),
7736 OpcP, OpcS,
7737 reg_mem(newval, mem_ptr),
7738 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7739 REX_reg_breg(res, res), // movzbl
7740 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7741 ins_pipe( pipe_cmpxchg );
7742 %}
7745 instruct compareAndSwapN(rRegI res,
7746 memory mem_ptr,
7747 rax_RegN oldval, rRegN newval,
7748 rFlagsReg cr) %{
7749 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
7750 effect(KILL cr, KILL oldval);
7752 format %{ "cmpxchgl $mem_ptr,$newval\t# "
7753 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7754 "sete $res\n\t"
7755 "movzbl $res, $res" %}
7756 opcode(0x0F, 0xB1);
7757 ins_encode(lock_prefix,
7758 REX_reg_mem(newval, mem_ptr),
7759 OpcP, OpcS,
7760 reg_mem(newval, mem_ptr),
7761 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7762 REX_reg_breg(res, res), // movzbl
7763 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7764 ins_pipe( pipe_cmpxchg );
7765 %}
7767 //----------Subtraction Instructions-------------------------------------------
7769 // Integer Subtraction Instructions
7770 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7771 %{
7772 match(Set dst (SubI dst src));
7773 effect(KILL cr);
7775 format %{ "subl $dst, $src\t# int" %}
7776 opcode(0x2B);
7777 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7778 ins_pipe(ialu_reg_reg);
7779 %}
7781 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7782 %{
7783 match(Set dst (SubI dst src));
7784 effect(KILL cr);
7786 format %{ "subl $dst, $src\t# int" %}
7787 opcode(0x81, 0x05); /* Opcode 81 /5 */
7788 ins_encode(OpcSErm(dst, src), Con8or32(src));
7789 ins_pipe(ialu_reg);
7790 %}
7792 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7793 %{
7794 match(Set dst (SubI dst (LoadI src)));
7795 effect(KILL cr);
7797 ins_cost(125);
7798 format %{ "subl $dst, $src\t# int" %}
7799 opcode(0x2B);
7800 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7801 ins_pipe(ialu_reg_mem);
7802 %}
7804 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7805 %{
7806 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
7807 effect(KILL cr);
7809 ins_cost(150);
7810 format %{ "subl $dst, $src\t# int" %}
7811 opcode(0x29); /* Opcode 29 /r */
7812 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7813 ins_pipe(ialu_mem_reg);
7814 %}
7816 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
7817 %{
7818 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
7819 effect(KILL cr);
7821 ins_cost(125); // XXX
7822 format %{ "subl $dst, $src\t# int" %}
7823 opcode(0x81); /* Opcode 81 /5 id */
7824 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
7825 ins_pipe(ialu_mem_imm);
7826 %}
7828 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
7829 %{
7830 match(Set dst (SubL dst src));
7831 effect(KILL cr);
7833 format %{ "subq $dst, $src\t# long" %}
7834 opcode(0x2B);
7835 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7836 ins_pipe(ialu_reg_reg);
7837 %}
7839 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
7840 %{
7841 match(Set dst (SubL dst src));
7842 effect(KILL cr);
7844 format %{ "subq $dst, $src\t# long" %}
7845 opcode(0x81, 0x05); /* Opcode 81 /5 */
7846 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7847 ins_pipe(ialu_reg);
7848 %}
7850 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
7851 %{
7852 match(Set dst (SubL dst (LoadL src)));
7853 effect(KILL cr);
7855 ins_cost(125);
7856 format %{ "subq $dst, $src\t# long" %}
7857 opcode(0x2B);
7858 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
7859 ins_pipe(ialu_reg_mem);
7860 %}
7862 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
7863 %{
7864 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
7865 effect(KILL cr);
7867 ins_cost(150);
7868 format %{ "subq $dst, $src\t# long" %}
7869 opcode(0x29); /* Opcode 29 /r */
7870 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7871 ins_pipe(ialu_mem_reg);
7872 %}
7874 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
7875 %{
7876 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
7877 effect(KILL cr);
7879 ins_cost(125); // XXX
7880 format %{ "subq $dst, $src\t# long" %}
7881 opcode(0x81); /* Opcode 81 /5 id */
7882 ins_encode(REX_mem_wide(dst),
7883 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
7884 ins_pipe(ialu_mem_imm);
7885 %}
7887 // Subtract from a pointer
7888 // XXX hmpf???
7889 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
7890 %{
7891 match(Set dst (AddP dst (SubI zero src)));
7892 effect(KILL cr);
7894 format %{ "subq $dst, $src\t# ptr - int" %}
7895 opcode(0x2B);
7896 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7897 ins_pipe(ialu_reg_reg);
7898 %}
7900 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
7901 %{
7902 match(Set dst (SubI zero dst));
7903 effect(KILL cr);
7905 format %{ "negl $dst\t# int" %}
7906 opcode(0xF7, 0x03); // Opcode F7 /3
7907 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7908 ins_pipe(ialu_reg);
7909 %}
7911 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
7912 %{
7913 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
7914 effect(KILL cr);
7916 format %{ "negl $dst\t# int" %}
7917 opcode(0xF7, 0x03); // Opcode F7 /3
7918 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
7919 ins_pipe(ialu_reg);
7920 %}
7922 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
7923 %{
7924 match(Set dst (SubL zero dst));
7925 effect(KILL cr);
7927 format %{ "negq $dst\t# long" %}
7928 opcode(0xF7, 0x03); // Opcode F7 /3
7929 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7930 ins_pipe(ialu_reg);
7931 %}
7933 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
7934 %{
7935 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
7936 effect(KILL cr);
7938 format %{ "negq $dst\t# long" %}
7939 opcode(0xF7, 0x03); // Opcode F7 /3
7940 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
7941 ins_pipe(ialu_reg);
7942 %}
7945 //----------Multiplication/Division Instructions-------------------------------
7946 // Integer Multiplication Instructions
7947 // Multiply Register
7949 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7950 %{
7951 match(Set dst (MulI dst src));
7952 effect(KILL cr);
7954 ins_cost(300);
7955 format %{ "imull $dst, $src\t# int" %}
7956 opcode(0x0F, 0xAF);
7957 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
7958 ins_pipe(ialu_reg_reg_alu0);
7959 %}
7961 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
7962 %{
7963 match(Set dst (MulI src imm));
7964 effect(KILL cr);
7966 ins_cost(300);
7967 format %{ "imull $dst, $src, $imm\t# int" %}
7968 opcode(0x69); /* 69 /r id */
7969 ins_encode(REX_reg_reg(dst, src),
7970 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
7971 ins_pipe(ialu_reg_reg_alu0);
7972 %}
7974 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
7975 %{
7976 match(Set dst (MulI dst (LoadI src)));
7977 effect(KILL cr);
7979 ins_cost(350);
7980 format %{ "imull $dst, $src\t# int" %}
7981 opcode(0x0F, 0xAF);
7982 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
7983 ins_pipe(ialu_reg_mem_alu0);
7984 %}
7986 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
7987 %{
7988 match(Set dst (MulI (LoadI src) imm));
7989 effect(KILL cr);
7991 ins_cost(300);
7992 format %{ "imull $dst, $src, $imm\t# int" %}
7993 opcode(0x69); /* 69 /r id */
7994 ins_encode(REX_reg_mem(dst, src),
7995 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
7996 ins_pipe(ialu_reg_mem_alu0);
7997 %}
7999 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8000 %{
8001 match(Set dst (MulL dst src));
8002 effect(KILL cr);
8004 ins_cost(300);
8005 format %{ "imulq $dst, $src\t# long" %}
8006 opcode(0x0F, 0xAF);
8007 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8008 ins_pipe(ialu_reg_reg_alu0);
8009 %}
8011 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8012 %{
8013 match(Set dst (MulL src imm));
8014 effect(KILL cr);
8016 ins_cost(300);
8017 format %{ "imulq $dst, $src, $imm\t# long" %}
8018 opcode(0x69); /* 69 /r id */
8019 ins_encode(REX_reg_reg_wide(dst, src),
8020 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8021 ins_pipe(ialu_reg_reg_alu0);
8022 %}
8024 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8025 %{
8026 match(Set dst (MulL dst (LoadL src)));
8027 effect(KILL cr);
8029 ins_cost(350);
8030 format %{ "imulq $dst, $src\t# long" %}
8031 opcode(0x0F, 0xAF);
8032 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8033 ins_pipe(ialu_reg_mem_alu0);
8034 %}
8036 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8037 %{
8038 match(Set dst (MulL (LoadL src) imm));
8039 effect(KILL cr);
8041 ins_cost(300);
8042 format %{ "imulq $dst, $src, $imm\t# long" %}
8043 opcode(0x69); /* 69 /r id */
8044 ins_encode(REX_reg_mem_wide(dst, src),
8045 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8046 ins_pipe(ialu_reg_mem_alu0);
8047 %}
8049 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8050 rFlagsReg cr)
8051 %{
8052 match(Set rax (DivI rax div));
8053 effect(KILL rdx, KILL cr);
8055 ins_cost(30*100+10*100); // XXX
8056 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8057 "jne,s normal\n\t"
8058 "xorl rdx, rdx\n\t"
8059 "cmpl $div, -1\n\t"
8060 "je,s done\n"
8061 "normal: cdql\n\t"
8062 "idivl $div\n"
8063 "done:" %}
8064 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8065 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8066 ins_pipe(ialu_reg_reg_alu0);
8067 %}
8069 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8070 rFlagsReg cr)
8071 %{
8072 match(Set rax (DivL rax div));
8073 effect(KILL rdx, KILL cr);
8075 ins_cost(30*100+10*100); // XXX
8076 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8077 "cmpq rax, rdx\n\t"
8078 "jne,s normal\n\t"
8079 "xorl rdx, rdx\n\t"
8080 "cmpq $div, -1\n\t"
8081 "je,s done\n"
8082 "normal: cdqq\n\t"
8083 "idivq $div\n"
8084 "done:" %}
8085 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8086 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8087 ins_pipe(ialu_reg_reg_alu0);
8088 %}
8090 // Integer DIVMOD with Register, both quotient and mod results
8091 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8092 rFlagsReg cr)
8093 %{
8094 match(DivModI rax div);
8095 effect(KILL cr);
8097 ins_cost(30*100+10*100); // XXX
8098 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8099 "jne,s normal\n\t"
8100 "xorl rdx, rdx\n\t"
8101 "cmpl $div, -1\n\t"
8102 "je,s done\n"
8103 "normal: cdql\n\t"
8104 "idivl $div\n"
8105 "done:" %}
8106 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8107 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8108 ins_pipe(pipe_slow);
8109 %}
8111 // Long DIVMOD with Register, both quotient and mod results
8112 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8113 rFlagsReg cr)
8114 %{
8115 match(DivModL rax div);
8116 effect(KILL cr);
8118 ins_cost(30*100+10*100); // XXX
8119 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8120 "cmpq rax, rdx\n\t"
8121 "jne,s normal\n\t"
8122 "xorl rdx, rdx\n\t"
8123 "cmpq $div, -1\n\t"
8124 "je,s done\n"
8125 "normal: cdqq\n\t"
8126 "idivq $div\n"
8127 "done:" %}
8128 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8129 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8130 ins_pipe(pipe_slow);
8131 %}
8133 //----------- DivL-By-Constant-Expansions--------------------------------------
8134 // DivI cases are handled by the compiler
8136 // Magic constant, reciprical of 10
8137 instruct loadConL_0x6666666666666667(rRegL dst)
8138 %{
8139 effect(DEF dst);
8141 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8142 ins_encode(load_immL(dst, 0x6666666666666667));
8143 ins_pipe(ialu_reg);
8144 %}
8146 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8147 %{
8148 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8150 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8151 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8152 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8153 ins_pipe(ialu_reg_reg_alu0);
8154 %}
8156 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8157 %{
8158 effect(USE_DEF dst, KILL cr);
8160 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8161 opcode(0xC1, 0x7); /* C1 /7 ib */
8162 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8163 ins_pipe(ialu_reg);
8164 %}
8166 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8167 %{
8168 effect(USE_DEF dst, KILL cr);
8170 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8171 opcode(0xC1, 0x7); /* C1 /7 ib */
8172 ins_encode(reg_opc_imm_wide(dst, 0x2));
8173 ins_pipe(ialu_reg);
8174 %}
8176 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8177 %{
8178 match(Set dst (DivL src div));
8180 ins_cost((5+8)*100);
8181 expand %{
8182 rax_RegL rax; // Killed temp
8183 rFlagsReg cr; // Killed
8184 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
8185 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
8186 sarL_rReg_63(src, cr); // sarq src, 63
8187 sarL_rReg_2(dst, cr); // sarq rdx, 2
8188 subL_rReg(dst, src, cr); // subl rdx, src
8189 %}
8190 %}
8192 //-----------------------------------------------------------------------------
8194 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8195 rFlagsReg cr)
8196 %{
8197 match(Set rdx (ModI rax div));
8198 effect(KILL rax, KILL cr);
8200 ins_cost(300); // XXX
8201 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
8202 "jne,s normal\n\t"
8203 "xorl rdx, rdx\n\t"
8204 "cmpl $div, -1\n\t"
8205 "je,s done\n"
8206 "normal: cdql\n\t"
8207 "idivl $div\n"
8208 "done:" %}
8209 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8210 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8211 ins_pipe(ialu_reg_reg_alu0);
8212 %}
8214 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8215 rFlagsReg cr)
8216 %{
8217 match(Set rdx (ModL rax div));
8218 effect(KILL rax, KILL cr);
8220 ins_cost(300); // XXX
8221 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
8222 "cmpq rax, rdx\n\t"
8223 "jne,s normal\n\t"
8224 "xorl rdx, rdx\n\t"
8225 "cmpq $div, -1\n\t"
8226 "je,s done\n"
8227 "normal: cdqq\n\t"
8228 "idivq $div\n"
8229 "done:" %}
8230 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8231 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8232 ins_pipe(ialu_reg_reg_alu0);
8233 %}
8235 // Integer Shift Instructions
8236 // Shift Left by one
8237 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8238 %{
8239 match(Set dst (LShiftI dst shift));
8240 effect(KILL cr);
8242 format %{ "sall $dst, $shift" %}
8243 opcode(0xD1, 0x4); /* D1 /4 */
8244 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8245 ins_pipe(ialu_reg);
8246 %}
8248 // Shift Left by one
8249 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8250 %{
8251 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8252 effect(KILL cr);
8254 format %{ "sall $dst, $shift\t" %}
8255 opcode(0xD1, 0x4); /* D1 /4 */
8256 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8257 ins_pipe(ialu_mem_imm);
8258 %}
8260 // Shift Left by 8-bit immediate
8261 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8262 %{
8263 match(Set dst (LShiftI dst shift));
8264 effect(KILL cr);
8266 format %{ "sall $dst, $shift" %}
8267 opcode(0xC1, 0x4); /* C1 /4 ib */
8268 ins_encode(reg_opc_imm(dst, shift));
8269 ins_pipe(ialu_reg);
8270 %}
8272 // Shift Left by 8-bit immediate
8273 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8274 %{
8275 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8276 effect(KILL cr);
8278 format %{ "sall $dst, $shift" %}
8279 opcode(0xC1, 0x4); /* C1 /4 ib */
8280 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8281 ins_pipe(ialu_mem_imm);
8282 %}
8284 // Shift Left by variable
8285 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8286 %{
8287 match(Set dst (LShiftI dst shift));
8288 effect(KILL cr);
8290 format %{ "sall $dst, $shift" %}
8291 opcode(0xD3, 0x4); /* D3 /4 */
8292 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8293 ins_pipe(ialu_reg_reg);
8294 %}
8296 // Shift Left by variable
8297 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8298 %{
8299 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8300 effect(KILL cr);
8302 format %{ "sall $dst, $shift" %}
8303 opcode(0xD3, 0x4); /* D3 /4 */
8304 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8305 ins_pipe(ialu_mem_reg);
8306 %}
8308 // Arithmetic shift right by one
8309 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8310 %{
8311 match(Set dst (RShiftI dst shift));
8312 effect(KILL cr);
8314 format %{ "sarl $dst, $shift" %}
8315 opcode(0xD1, 0x7); /* D1 /7 */
8316 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8317 ins_pipe(ialu_reg);
8318 %}
8320 // Arithmetic shift right by one
8321 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8322 %{
8323 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8324 effect(KILL cr);
8326 format %{ "sarl $dst, $shift" %}
8327 opcode(0xD1, 0x7); /* D1 /7 */
8328 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8329 ins_pipe(ialu_mem_imm);
8330 %}
8332 // Arithmetic Shift Right by 8-bit immediate
8333 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8334 %{
8335 match(Set dst (RShiftI dst shift));
8336 effect(KILL cr);
8338 format %{ "sarl $dst, $shift" %}
8339 opcode(0xC1, 0x7); /* C1 /7 ib */
8340 ins_encode(reg_opc_imm(dst, shift));
8341 ins_pipe(ialu_mem_imm);
8342 %}
8344 // Arithmetic Shift Right by 8-bit immediate
8345 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8346 %{
8347 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8348 effect(KILL cr);
8350 format %{ "sarl $dst, $shift" %}
8351 opcode(0xC1, 0x7); /* C1 /7 ib */
8352 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8353 ins_pipe(ialu_mem_imm);
8354 %}
8356 // Arithmetic Shift Right by variable
8357 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8358 %{
8359 match(Set dst (RShiftI dst shift));
8360 effect(KILL cr);
8362 format %{ "sarl $dst, $shift" %}
8363 opcode(0xD3, 0x7); /* D3 /7 */
8364 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8365 ins_pipe(ialu_reg_reg);
8366 %}
8368 // Arithmetic Shift Right by variable
8369 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8370 %{
8371 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8372 effect(KILL cr);
8374 format %{ "sarl $dst, $shift" %}
8375 opcode(0xD3, 0x7); /* D3 /7 */
8376 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8377 ins_pipe(ialu_mem_reg);
8378 %}
8380 // Logical shift right by one
8381 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8382 %{
8383 match(Set dst (URShiftI dst shift));
8384 effect(KILL cr);
8386 format %{ "shrl $dst, $shift" %}
8387 opcode(0xD1, 0x5); /* D1 /5 */
8388 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8389 ins_pipe(ialu_reg);
8390 %}
8392 // Logical shift right by one
8393 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8394 %{
8395 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8396 effect(KILL cr);
8398 format %{ "shrl $dst, $shift" %}
8399 opcode(0xD1, 0x5); /* D1 /5 */
8400 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8401 ins_pipe(ialu_mem_imm);
8402 %}
8404 // Logical Shift Right by 8-bit immediate
8405 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8406 %{
8407 match(Set dst (URShiftI dst shift));
8408 effect(KILL cr);
8410 format %{ "shrl $dst, $shift" %}
8411 opcode(0xC1, 0x5); /* C1 /5 ib */
8412 ins_encode(reg_opc_imm(dst, shift));
8413 ins_pipe(ialu_reg);
8414 %}
8416 // Logical Shift Right by 8-bit immediate
8417 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8418 %{
8419 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8420 effect(KILL cr);
8422 format %{ "shrl $dst, $shift" %}
8423 opcode(0xC1, 0x5); /* C1 /5 ib */
8424 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8425 ins_pipe(ialu_mem_imm);
8426 %}
8428 // Logical Shift Right by variable
8429 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8430 %{
8431 match(Set dst (URShiftI dst shift));
8432 effect(KILL cr);
8434 format %{ "shrl $dst, $shift" %}
8435 opcode(0xD3, 0x5); /* D3 /5 */
8436 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8437 ins_pipe(ialu_reg_reg);
8438 %}
8440 // Logical Shift Right by variable
8441 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8442 %{
8443 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8444 effect(KILL cr);
8446 format %{ "shrl $dst, $shift" %}
8447 opcode(0xD3, 0x5); /* D3 /5 */
8448 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8449 ins_pipe(ialu_mem_reg);
8450 %}
8452 // Long Shift Instructions
8453 // Shift Left by one
8454 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8455 %{
8456 match(Set dst (LShiftL dst shift));
8457 effect(KILL cr);
8459 format %{ "salq $dst, $shift" %}
8460 opcode(0xD1, 0x4); /* D1 /4 */
8461 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8462 ins_pipe(ialu_reg);
8463 %}
8465 // Shift Left by one
8466 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8467 %{
8468 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8469 effect(KILL cr);
8471 format %{ "salq $dst, $shift" %}
8472 opcode(0xD1, 0x4); /* D1 /4 */
8473 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8474 ins_pipe(ialu_mem_imm);
8475 %}
8477 // Shift Left by 8-bit immediate
8478 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8479 %{
8480 match(Set dst (LShiftL dst shift));
8481 effect(KILL cr);
8483 format %{ "salq $dst, $shift" %}
8484 opcode(0xC1, 0x4); /* C1 /4 ib */
8485 ins_encode(reg_opc_imm_wide(dst, shift));
8486 ins_pipe(ialu_reg);
8487 %}
8489 // Shift Left by 8-bit immediate
8490 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8491 %{
8492 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8493 effect(KILL cr);
8495 format %{ "salq $dst, $shift" %}
8496 opcode(0xC1, 0x4); /* C1 /4 ib */
8497 ins_encode(REX_mem_wide(dst), OpcP,
8498 RM_opc_mem(secondary, dst), Con8or32(shift));
8499 ins_pipe(ialu_mem_imm);
8500 %}
8502 // Shift Left by variable
8503 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8504 %{
8505 match(Set dst (LShiftL dst shift));
8506 effect(KILL cr);
8508 format %{ "salq $dst, $shift" %}
8509 opcode(0xD3, 0x4); /* D3 /4 */
8510 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8511 ins_pipe(ialu_reg_reg);
8512 %}
8514 // Shift Left by variable
8515 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8516 %{
8517 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8518 effect(KILL cr);
8520 format %{ "salq $dst, $shift" %}
8521 opcode(0xD3, 0x4); /* D3 /4 */
8522 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8523 ins_pipe(ialu_mem_reg);
8524 %}
8526 // Arithmetic shift right by one
8527 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8528 %{
8529 match(Set dst (RShiftL dst shift));
8530 effect(KILL cr);
8532 format %{ "sarq $dst, $shift" %}
8533 opcode(0xD1, 0x7); /* D1 /7 */
8534 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8535 ins_pipe(ialu_reg);
8536 %}
8538 // Arithmetic shift right by one
8539 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8540 %{
8541 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8542 effect(KILL cr);
8544 format %{ "sarq $dst, $shift" %}
8545 opcode(0xD1, 0x7); /* D1 /7 */
8546 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8547 ins_pipe(ialu_mem_imm);
8548 %}
8550 // Arithmetic Shift Right by 8-bit immediate
8551 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8552 %{
8553 match(Set dst (RShiftL dst shift));
8554 effect(KILL cr);
8556 format %{ "sarq $dst, $shift" %}
8557 opcode(0xC1, 0x7); /* C1 /7 ib */
8558 ins_encode(reg_opc_imm_wide(dst, shift));
8559 ins_pipe(ialu_mem_imm);
8560 %}
8562 // Arithmetic Shift Right by 8-bit immediate
8563 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8564 %{
8565 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8566 effect(KILL cr);
8568 format %{ "sarq $dst, $shift" %}
8569 opcode(0xC1, 0x7); /* C1 /7 ib */
8570 ins_encode(REX_mem_wide(dst), OpcP,
8571 RM_opc_mem(secondary, dst), Con8or32(shift));
8572 ins_pipe(ialu_mem_imm);
8573 %}
8575 // Arithmetic Shift Right by variable
8576 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8577 %{
8578 match(Set dst (RShiftL dst shift));
8579 effect(KILL cr);
8581 format %{ "sarq $dst, $shift" %}
8582 opcode(0xD3, 0x7); /* D3 /7 */
8583 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8584 ins_pipe(ialu_reg_reg);
8585 %}
8587 // Arithmetic Shift Right by variable
8588 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8589 %{
8590 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8591 effect(KILL cr);
8593 format %{ "sarq $dst, $shift" %}
8594 opcode(0xD3, 0x7); /* D3 /7 */
8595 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8596 ins_pipe(ialu_mem_reg);
8597 %}
8599 // Logical shift right by one
8600 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8601 %{
8602 match(Set dst (URShiftL dst shift));
8603 effect(KILL cr);
8605 format %{ "shrq $dst, $shift" %}
8606 opcode(0xD1, 0x5); /* D1 /5 */
8607 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
8608 ins_pipe(ialu_reg);
8609 %}
8611 // Logical shift right by one
8612 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8613 %{
8614 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8615 effect(KILL cr);
8617 format %{ "shrq $dst, $shift" %}
8618 opcode(0xD1, 0x5); /* D1 /5 */
8619 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8620 ins_pipe(ialu_mem_imm);
8621 %}
8623 // Logical Shift Right by 8-bit immediate
8624 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8625 %{
8626 match(Set dst (URShiftL dst shift));
8627 effect(KILL cr);
8629 format %{ "shrq $dst, $shift" %}
8630 opcode(0xC1, 0x5); /* C1 /5 ib */
8631 ins_encode(reg_opc_imm_wide(dst, shift));
8632 ins_pipe(ialu_reg);
8633 %}
8635 // Logical Shift Right by 8-bit immediate
8636 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8637 %{
8638 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8639 effect(KILL cr);
8641 format %{ "shrq $dst, $shift" %}
8642 opcode(0xC1, 0x5); /* C1 /5 ib */
8643 ins_encode(REX_mem_wide(dst), OpcP,
8644 RM_opc_mem(secondary, dst), Con8or32(shift));
8645 ins_pipe(ialu_mem_imm);
8646 %}
8648 // Logical Shift Right by variable
8649 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8650 %{
8651 match(Set dst (URShiftL dst shift));
8652 effect(KILL cr);
8654 format %{ "shrq $dst, $shift" %}
8655 opcode(0xD3, 0x5); /* D3 /5 */
8656 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8657 ins_pipe(ialu_reg_reg);
8658 %}
8660 // Logical Shift Right by variable
8661 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8662 %{
8663 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8664 effect(KILL cr);
8666 format %{ "shrq $dst, $shift" %}
8667 opcode(0xD3, 0x5); /* D3 /5 */
8668 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8669 ins_pipe(ialu_mem_reg);
8670 %}
8672 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
8673 // This idiom is used by the compiler for the i2b bytecode.
8674 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
8675 %{
8676 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
8678 format %{ "movsbl $dst, $src\t# i2b" %}
8679 opcode(0x0F, 0xBE);
8680 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8681 ins_pipe(ialu_reg_reg);
8682 %}
8684 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
8685 // This idiom is used by the compiler the i2s bytecode.
8686 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
8687 %{
8688 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
8690 format %{ "movswl $dst, $src\t# i2s" %}
8691 opcode(0x0F, 0xBF);
8692 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8693 ins_pipe(ialu_reg_reg);
8694 %}
8696 // ROL/ROR instructions
8698 // ROL expand
8699 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
8700 effect(KILL cr, USE_DEF dst);
8702 format %{ "roll $dst" %}
8703 opcode(0xD1, 0x0); /* Opcode D1 /0 */
8704 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8705 ins_pipe(ialu_reg);
8706 %}
8708 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
8709 effect(USE_DEF dst, USE shift, KILL cr);
8711 format %{ "roll $dst, $shift" %}
8712 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
8713 ins_encode( reg_opc_imm(dst, shift) );
8714 ins_pipe(ialu_reg);
8715 %}
8717 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
8718 %{
8719 effect(USE_DEF dst, USE shift, KILL cr);
8721 format %{ "roll $dst, $shift" %}
8722 opcode(0xD3, 0x0); /* Opcode D3 /0 */
8723 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8724 ins_pipe(ialu_reg_reg);
8725 %}
8726 // end of ROL expand
8728 // Rotate Left by one
8729 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
8730 %{
8731 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8733 expand %{
8734 rolI_rReg_imm1(dst, cr);
8735 %}
8736 %}
8738 // Rotate Left by 8-bit immediate
8739 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
8740 %{
8741 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8742 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8744 expand %{
8745 rolI_rReg_imm8(dst, lshift, cr);
8746 %}
8747 %}
8749 // Rotate Left by variable
8750 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
8751 %{
8752 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
8754 expand %{
8755 rolI_rReg_CL(dst, shift, cr);
8756 %}
8757 %}
8759 // Rotate Left by variable
8760 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
8761 %{
8762 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
8764 expand %{
8765 rolI_rReg_CL(dst, shift, cr);
8766 %}
8767 %}
8769 // ROR expand
8770 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
8771 %{
8772 effect(USE_DEF dst, KILL cr);
8774 format %{ "rorl $dst" %}
8775 opcode(0xD1, 0x1); /* D1 /1 */
8776 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8777 ins_pipe(ialu_reg);
8778 %}
8780 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
8781 %{
8782 effect(USE_DEF dst, USE shift, KILL cr);
8784 format %{ "rorl $dst, $shift" %}
8785 opcode(0xC1, 0x1); /* C1 /1 ib */
8786 ins_encode(reg_opc_imm(dst, shift));
8787 ins_pipe(ialu_reg);
8788 %}
8790 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
8791 %{
8792 effect(USE_DEF dst, USE shift, KILL cr);
8794 format %{ "rorl $dst, $shift" %}
8795 opcode(0xD3, 0x1); /* D3 /1 */
8796 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8797 ins_pipe(ialu_reg_reg);
8798 %}
8799 // end of ROR expand
8801 // Rotate Right by one
8802 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
8803 %{
8804 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8806 expand %{
8807 rorI_rReg_imm1(dst, cr);
8808 %}
8809 %}
8811 // Rotate Right by 8-bit immediate
8812 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
8813 %{
8814 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8815 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8817 expand %{
8818 rorI_rReg_imm8(dst, rshift, cr);
8819 %}
8820 %}
8822 // Rotate Right by variable
8823 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
8824 %{
8825 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
8827 expand %{
8828 rorI_rReg_CL(dst, shift, cr);
8829 %}
8830 %}
8832 // Rotate Right by variable
8833 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
8834 %{
8835 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
8837 expand %{
8838 rorI_rReg_CL(dst, shift, cr);
8839 %}
8840 %}
8842 // for long rotate
8843 // ROL expand
8844 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
8845 effect(USE_DEF dst, KILL cr);
8847 format %{ "rolq $dst" %}
8848 opcode(0xD1, 0x0); /* Opcode D1 /0 */
8849 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8850 ins_pipe(ialu_reg);
8851 %}
8853 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
8854 effect(USE_DEF dst, USE shift, KILL cr);
8856 format %{ "rolq $dst, $shift" %}
8857 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
8858 ins_encode( reg_opc_imm_wide(dst, shift) );
8859 ins_pipe(ialu_reg);
8860 %}
8862 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
8863 %{
8864 effect(USE_DEF dst, USE shift, KILL cr);
8866 format %{ "rolq $dst, $shift" %}
8867 opcode(0xD3, 0x0); /* Opcode D3 /0 */
8868 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8869 ins_pipe(ialu_reg_reg);
8870 %}
8871 // end of ROL expand
8873 // Rotate Left by one
8874 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
8875 %{
8876 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
8878 expand %{
8879 rolL_rReg_imm1(dst, cr);
8880 %}
8881 %}
8883 // Rotate Left by 8-bit immediate
8884 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
8885 %{
8886 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
8887 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
8889 expand %{
8890 rolL_rReg_imm8(dst, lshift, cr);
8891 %}
8892 %}
8894 // Rotate Left by variable
8895 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
8896 %{
8897 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
8899 expand %{
8900 rolL_rReg_CL(dst, shift, cr);
8901 %}
8902 %}
8904 // Rotate Left by variable
8905 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
8906 %{
8907 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
8909 expand %{
8910 rolL_rReg_CL(dst, shift, cr);
8911 %}
8912 %}
8914 // ROR expand
8915 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
8916 %{
8917 effect(USE_DEF dst, KILL cr);
8919 format %{ "rorq $dst" %}
8920 opcode(0xD1, 0x1); /* D1 /1 */
8921 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8922 ins_pipe(ialu_reg);
8923 %}
8925 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
8926 %{
8927 effect(USE_DEF dst, USE shift, KILL cr);
8929 format %{ "rorq $dst, $shift" %}
8930 opcode(0xC1, 0x1); /* C1 /1 ib */
8931 ins_encode(reg_opc_imm_wide(dst, shift));
8932 ins_pipe(ialu_reg);
8933 %}
8935 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
8936 %{
8937 effect(USE_DEF dst, USE shift, KILL cr);
8939 format %{ "rorq $dst, $shift" %}
8940 opcode(0xD3, 0x1); /* D3 /1 */
8941 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8942 ins_pipe(ialu_reg_reg);
8943 %}
8944 // end of ROR expand
8946 // Rotate Right by one
8947 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
8948 %{
8949 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
8951 expand %{
8952 rorL_rReg_imm1(dst, cr);
8953 %}
8954 %}
8956 // Rotate Right by 8-bit immediate
8957 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
8958 %{
8959 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
8960 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
8962 expand %{
8963 rorL_rReg_imm8(dst, rshift, cr);
8964 %}
8965 %}
8967 // Rotate Right by variable
8968 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
8969 %{
8970 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
8972 expand %{
8973 rorL_rReg_CL(dst, shift, cr);
8974 %}
8975 %}
8977 // Rotate Right by variable
8978 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
8979 %{
8980 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
8982 expand %{
8983 rorL_rReg_CL(dst, shift, cr);
8984 %}
8985 %}
8987 // Logical Instructions
8989 // Integer Logical Instructions
8991 // And Instructions
8992 // And Register with Register
8993 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8994 %{
8995 match(Set dst (AndI dst src));
8996 effect(KILL cr);
8998 format %{ "andl $dst, $src\t# int" %}
8999 opcode(0x23);
9000 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9001 ins_pipe(ialu_reg_reg);
9002 %}
9004 // And Register with Immediate 255
9005 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9006 %{
9007 match(Set dst (AndI dst src));
9009 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9010 opcode(0x0F, 0xB6);
9011 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9012 ins_pipe(ialu_reg);
9013 %}
9015 // And Register with Immediate 255 and promote to long
9016 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9017 %{
9018 match(Set dst (ConvI2L (AndI src mask)));
9020 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9021 opcode(0x0F, 0xB6);
9022 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9023 ins_pipe(ialu_reg);
9024 %}
9026 // And Register with Immediate 65535
9027 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9028 %{
9029 match(Set dst (AndI dst src));
9031 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9032 opcode(0x0F, 0xB7);
9033 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9034 ins_pipe(ialu_reg);
9035 %}
9037 // And Register with Immediate 65535 and promote to long
9038 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9039 %{
9040 match(Set dst (ConvI2L (AndI src mask)));
9042 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9043 opcode(0x0F, 0xB7);
9044 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9045 ins_pipe(ialu_reg);
9046 %}
9048 // And Register with Immediate
9049 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9050 %{
9051 match(Set dst (AndI dst src));
9052 effect(KILL cr);
9054 format %{ "andl $dst, $src\t# int" %}
9055 opcode(0x81, 0x04); /* Opcode 81 /4 */
9056 ins_encode(OpcSErm(dst, src), Con8or32(src));
9057 ins_pipe(ialu_reg);
9058 %}
9060 // And Register with Memory
9061 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9062 %{
9063 match(Set dst (AndI dst (LoadI src)));
9064 effect(KILL cr);
9066 ins_cost(125);
9067 format %{ "andl $dst, $src\t# int" %}
9068 opcode(0x23);
9069 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9070 ins_pipe(ialu_reg_mem);
9071 %}
9073 // And Memory with Register
9074 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9075 %{
9076 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9077 effect(KILL cr);
9079 ins_cost(150);
9080 format %{ "andl $dst, $src\t# int" %}
9081 opcode(0x21); /* Opcode 21 /r */
9082 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9083 ins_pipe(ialu_mem_reg);
9084 %}
9086 // And Memory with Immediate
9087 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9088 %{
9089 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9090 effect(KILL cr);
9092 ins_cost(125);
9093 format %{ "andl $dst, $src\t# int" %}
9094 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9095 ins_encode(REX_mem(dst), OpcSE(src),
9096 RM_opc_mem(secondary, dst), Con8or32(src));
9097 ins_pipe(ialu_mem_imm);
9098 %}
9100 // Or Instructions
9101 // Or Register with Register
9102 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9103 %{
9104 match(Set dst (OrI dst src));
9105 effect(KILL cr);
9107 format %{ "orl $dst, $src\t# int" %}
9108 opcode(0x0B);
9109 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9110 ins_pipe(ialu_reg_reg);
9111 %}
9113 // Or Register with Immediate
9114 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9115 %{
9116 match(Set dst (OrI dst src));
9117 effect(KILL cr);
9119 format %{ "orl $dst, $src\t# int" %}
9120 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9121 ins_encode(OpcSErm(dst, src), Con8or32(src));
9122 ins_pipe(ialu_reg);
9123 %}
9125 // Or Register with Memory
9126 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9127 %{
9128 match(Set dst (OrI dst (LoadI src)));
9129 effect(KILL cr);
9131 ins_cost(125);
9132 format %{ "orl $dst, $src\t# int" %}
9133 opcode(0x0B);
9134 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9135 ins_pipe(ialu_reg_mem);
9136 %}
9138 // Or Memory with Register
9139 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9140 %{
9141 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9142 effect(KILL cr);
9144 ins_cost(150);
9145 format %{ "orl $dst, $src\t# int" %}
9146 opcode(0x09); /* Opcode 09 /r */
9147 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9148 ins_pipe(ialu_mem_reg);
9149 %}
9151 // Or Memory with Immediate
9152 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9153 %{
9154 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9155 effect(KILL cr);
9157 ins_cost(125);
9158 format %{ "orl $dst, $src\t# int" %}
9159 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9160 ins_encode(REX_mem(dst), OpcSE(src),
9161 RM_opc_mem(secondary, dst), Con8or32(src));
9162 ins_pipe(ialu_mem_imm);
9163 %}
9165 // Xor Instructions
9166 // Xor Register with Register
9167 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9168 %{
9169 match(Set dst (XorI dst src));
9170 effect(KILL cr);
9172 format %{ "xorl $dst, $src\t# int" %}
9173 opcode(0x33);
9174 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9175 ins_pipe(ialu_reg_reg);
9176 %}
9178 // Xor Register with Immediate
9179 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9180 %{
9181 match(Set dst (XorI dst src));
9182 effect(KILL cr);
9184 format %{ "xorl $dst, $src\t# int" %}
9185 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9186 ins_encode(OpcSErm(dst, src), Con8or32(src));
9187 ins_pipe(ialu_reg);
9188 %}
9190 // Xor Register with Memory
9191 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9192 %{
9193 match(Set dst (XorI dst (LoadI src)));
9194 effect(KILL cr);
9196 ins_cost(125);
9197 format %{ "xorl $dst, $src\t# int" %}
9198 opcode(0x33);
9199 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9200 ins_pipe(ialu_reg_mem);
9201 %}
9203 // Xor Memory with Register
9204 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9205 %{
9206 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9207 effect(KILL cr);
9209 ins_cost(150);
9210 format %{ "xorl $dst, $src\t# int" %}
9211 opcode(0x31); /* Opcode 31 /r */
9212 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9213 ins_pipe(ialu_mem_reg);
9214 %}
9216 // Xor Memory with Immediate
9217 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9218 %{
9219 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9220 effect(KILL cr);
9222 ins_cost(125);
9223 format %{ "xorl $dst, $src\t# int" %}
9224 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9225 ins_encode(REX_mem(dst), OpcSE(src),
9226 RM_opc_mem(secondary, dst), Con8or32(src));
9227 ins_pipe(ialu_mem_imm);
9228 %}
9231 // Long Logical Instructions
9233 // And Instructions
9234 // And Register with Register
9235 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9236 %{
9237 match(Set dst (AndL dst src));
9238 effect(KILL cr);
9240 format %{ "andq $dst, $src\t# long" %}
9241 opcode(0x23);
9242 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9243 ins_pipe(ialu_reg_reg);
9244 %}
9246 // And Register with Immediate 255
9247 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9248 %{
9249 match(Set dst (AndL dst src));
9251 format %{ "movzbq $dst, $src\t# long & 0xFF" %}
9252 opcode(0x0F, 0xB6);
9253 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9254 ins_pipe(ialu_reg);
9255 %}
9257 // And Register with Immediate 65535
9258 instruct andL_rReg_imm65535(rRegI dst, immL_65535 src)
9259 %{
9260 match(Set dst (AndL dst src));
9262 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
9263 opcode(0x0F, 0xB7);
9264 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9265 ins_pipe(ialu_reg);
9266 %}
9268 // And Register with Immediate
9269 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9270 %{
9271 match(Set dst (AndL dst src));
9272 effect(KILL cr);
9274 format %{ "andq $dst, $src\t# long" %}
9275 opcode(0x81, 0x04); /* Opcode 81 /4 */
9276 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9277 ins_pipe(ialu_reg);
9278 %}
9280 // And Register with Memory
9281 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9282 %{
9283 match(Set dst (AndL dst (LoadL src)));
9284 effect(KILL cr);
9286 ins_cost(125);
9287 format %{ "andq $dst, $src\t# long" %}
9288 opcode(0x23);
9289 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9290 ins_pipe(ialu_reg_mem);
9291 %}
9293 // And Memory with Register
9294 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9295 %{
9296 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9297 effect(KILL cr);
9299 ins_cost(150);
9300 format %{ "andq $dst, $src\t# long" %}
9301 opcode(0x21); /* Opcode 21 /r */
9302 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9303 ins_pipe(ialu_mem_reg);
9304 %}
9306 // And Memory with Immediate
9307 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9308 %{
9309 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9310 effect(KILL cr);
9312 ins_cost(125);
9313 format %{ "andq $dst, $src\t# long" %}
9314 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9315 ins_encode(REX_mem_wide(dst), OpcSE(src),
9316 RM_opc_mem(secondary, dst), Con8or32(src));
9317 ins_pipe(ialu_mem_imm);
9318 %}
9320 // Or Instructions
9321 // Or Register with Register
9322 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9323 %{
9324 match(Set dst (OrL dst src));
9325 effect(KILL cr);
9327 format %{ "orq $dst, $src\t# long" %}
9328 opcode(0x0B);
9329 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9330 ins_pipe(ialu_reg_reg);
9331 %}
9333 // Or Register with Immediate
9334 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9335 %{
9336 match(Set dst (OrL dst src));
9337 effect(KILL cr);
9339 format %{ "orq $dst, $src\t# long" %}
9340 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9341 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9342 ins_pipe(ialu_reg);
9343 %}
9345 // Or Register with Memory
9346 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9347 %{
9348 match(Set dst (OrL dst (LoadL src)));
9349 effect(KILL cr);
9351 ins_cost(125);
9352 format %{ "orq $dst, $src\t# long" %}
9353 opcode(0x0B);
9354 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9355 ins_pipe(ialu_reg_mem);
9356 %}
9358 // Or Memory with Register
9359 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9360 %{
9361 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9362 effect(KILL cr);
9364 ins_cost(150);
9365 format %{ "orq $dst, $src\t# long" %}
9366 opcode(0x09); /* Opcode 09 /r */
9367 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9368 ins_pipe(ialu_mem_reg);
9369 %}
9371 // Or Memory with Immediate
9372 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9373 %{
9374 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9375 effect(KILL cr);
9377 ins_cost(125);
9378 format %{ "orq $dst, $src\t# long" %}
9379 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9380 ins_encode(REX_mem_wide(dst), OpcSE(src),
9381 RM_opc_mem(secondary, dst), Con8or32(src));
9382 ins_pipe(ialu_mem_imm);
9383 %}
9385 // Xor Instructions
9386 // Xor Register with Register
9387 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9388 %{
9389 match(Set dst (XorL dst src));
9390 effect(KILL cr);
9392 format %{ "xorq $dst, $src\t# long" %}
9393 opcode(0x33);
9394 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9395 ins_pipe(ialu_reg_reg);
9396 %}
9398 // Xor Register with Immediate
9399 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9400 %{
9401 match(Set dst (XorL dst src));
9402 effect(KILL cr);
9404 format %{ "xorq $dst, $src\t# long" %}
9405 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9406 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9407 ins_pipe(ialu_reg);
9408 %}
9410 // Xor Register with Memory
9411 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9412 %{
9413 match(Set dst (XorL dst (LoadL src)));
9414 effect(KILL cr);
9416 ins_cost(125);
9417 format %{ "xorq $dst, $src\t# long" %}
9418 opcode(0x33);
9419 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9420 ins_pipe(ialu_reg_mem);
9421 %}
9423 // Xor Memory with Register
9424 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9425 %{
9426 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9427 effect(KILL cr);
9429 ins_cost(150);
9430 format %{ "xorq $dst, $src\t# long" %}
9431 opcode(0x31); /* Opcode 31 /r */
9432 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9433 ins_pipe(ialu_mem_reg);
9434 %}
9436 // Xor Memory with Immediate
9437 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9438 %{
9439 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9440 effect(KILL cr);
9442 ins_cost(125);
9443 format %{ "xorq $dst, $src\t# long" %}
9444 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9445 ins_encode(REX_mem_wide(dst), OpcSE(src),
9446 RM_opc_mem(secondary, dst), Con8or32(src));
9447 ins_pipe(ialu_mem_imm);
9448 %}
9450 // Convert Int to Boolean
9451 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
9452 %{
9453 match(Set dst (Conv2B src));
9454 effect(KILL cr);
9456 format %{ "testl $src, $src\t# ci2b\n\t"
9457 "setnz $dst\n\t"
9458 "movzbl $dst, $dst" %}
9459 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
9460 setNZ_reg(dst),
9461 REX_reg_breg(dst, dst), // movzbl
9462 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
9463 ins_pipe(pipe_slow); // XXX
9464 %}
9466 // Convert Pointer to Boolean
9467 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
9468 %{
9469 match(Set dst (Conv2B src));
9470 effect(KILL cr);
9472 format %{ "testq $src, $src\t# cp2b\n\t"
9473 "setnz $dst\n\t"
9474 "movzbl $dst, $dst" %}
9475 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
9476 setNZ_reg(dst),
9477 REX_reg_breg(dst, dst), // movzbl
9478 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
9479 ins_pipe(pipe_slow); // XXX
9480 %}
9482 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
9483 %{
9484 match(Set dst (CmpLTMask p q));
9485 effect(KILL cr);
9487 ins_cost(400); // XXX
9488 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
9489 "setlt $dst\n\t"
9490 "movzbl $dst, $dst\n\t"
9491 "negl $dst" %}
9492 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
9493 setLT_reg(dst),
9494 REX_reg_breg(dst, dst), // movzbl
9495 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
9496 neg_reg(dst));
9497 ins_pipe(pipe_slow);
9498 %}
9500 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
9501 %{
9502 match(Set dst (CmpLTMask dst zero));
9503 effect(KILL cr);
9505 ins_cost(100); // XXX
9506 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
9507 opcode(0xC1, 0x7); /* C1 /7 ib */
9508 ins_encode(reg_opc_imm(dst, 0x1F));
9509 ins_pipe(ialu_reg);
9510 %}
9513 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
9514 rRegI tmp,
9515 rFlagsReg cr)
9516 %{
9517 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
9518 effect(TEMP tmp, KILL cr);
9520 ins_cost(400); // XXX
9521 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
9522 "sbbl $tmp, $tmp\n\t"
9523 "andl $tmp, $y\n\t"
9524 "addl $p, $tmp" %}
9525 ins_encode(enc_cmpLTP(p, q, y, tmp));
9526 ins_pipe(pipe_cmplt);
9527 %}
9529 /* If I enable this, I encourage spilling in the inner loop of compress.
9530 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
9531 %{
9532 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
9533 effect( TEMP tmp, KILL cr );
9534 ins_cost(400);
9536 format %{ "SUB $p,$q\n\t"
9537 "SBB RCX,RCX\n\t"
9538 "AND RCX,$y\n\t"
9539 "ADD $p,RCX" %}
9540 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
9541 %}
9542 */
9544 //---------- FP Instructions------------------------------------------------
9546 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
9547 %{
9548 match(Set cr (CmpF src1 src2));
9550 ins_cost(145);
9551 format %{ "ucomiss $src1, $src2\n\t"
9552 "jnp,s exit\n\t"
9553 "pushfq\t# saw NaN, set CF\n\t"
9554 "andq [rsp], #0xffffff2b\n\t"
9555 "popfq\n"
9556 "exit: nop\t# avoid branch to branch" %}
9557 opcode(0x0F, 0x2E);
9558 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
9559 cmpfp_fixup);
9560 ins_pipe(pipe_slow);
9561 %}
9563 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
9564 %{
9565 match(Set cr (CmpF src1 (LoadF src2)));
9567 ins_cost(145);
9568 format %{ "ucomiss $src1, $src2\n\t"
9569 "jnp,s exit\n\t"
9570 "pushfq\t# saw NaN, set CF\n\t"
9571 "andq [rsp], #0xffffff2b\n\t"
9572 "popfq\n"
9573 "exit: nop\t# avoid branch to branch" %}
9574 opcode(0x0F, 0x2E);
9575 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
9576 cmpfp_fixup);
9577 ins_pipe(pipe_slow);
9578 %}
9580 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
9581 %{
9582 match(Set cr (CmpF src1 src2));
9584 ins_cost(145);
9585 format %{ "ucomiss $src1, $src2\n\t"
9586 "jnp,s exit\n\t"
9587 "pushfq\t# saw NaN, set CF\n\t"
9588 "andq [rsp], #0xffffff2b\n\t"
9589 "popfq\n"
9590 "exit: nop\t# avoid branch to branch" %}
9591 opcode(0x0F, 0x2E);
9592 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
9593 cmpfp_fixup);
9594 ins_pipe(pipe_slow);
9595 %}
9597 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
9598 %{
9599 match(Set cr (CmpD src1 src2));
9601 ins_cost(145);
9602 format %{ "ucomisd $src1, $src2\n\t"
9603 "jnp,s exit\n\t"
9604 "pushfq\t# saw NaN, set CF\n\t"
9605 "andq [rsp], #0xffffff2b\n\t"
9606 "popfq\n"
9607 "exit: nop\t# avoid branch to branch" %}
9608 opcode(0x66, 0x0F, 0x2E);
9609 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
9610 cmpfp_fixup);
9611 ins_pipe(pipe_slow);
9612 %}
9614 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
9615 %{
9616 match(Set cr (CmpD src1 (LoadD src2)));
9618 ins_cost(145);
9619 format %{ "ucomisd $src1, $src2\n\t"
9620 "jnp,s exit\n\t"
9621 "pushfq\t# saw NaN, set CF\n\t"
9622 "andq [rsp], #0xffffff2b\n\t"
9623 "popfq\n"
9624 "exit: nop\t# avoid branch to branch" %}
9625 opcode(0x66, 0x0F, 0x2E);
9626 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
9627 cmpfp_fixup);
9628 ins_pipe(pipe_slow);
9629 %}
9631 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
9632 %{
9633 match(Set cr (CmpD src1 src2));
9635 ins_cost(145);
9636 format %{ "ucomisd $src1, [$src2]\n\t"
9637 "jnp,s exit\n\t"
9638 "pushfq\t# saw NaN, set CF\n\t"
9639 "andq [rsp], #0xffffff2b\n\t"
9640 "popfq\n"
9641 "exit: nop\t# avoid branch to branch" %}
9642 opcode(0x66, 0x0F, 0x2E);
9643 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
9644 cmpfp_fixup);
9645 ins_pipe(pipe_slow);
9646 %}
9648 // Compare into -1,0,1
9649 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
9650 %{
9651 match(Set dst (CmpF3 src1 src2));
9652 effect(KILL cr);
9654 ins_cost(275);
9655 format %{ "ucomiss $src1, $src2\n\t"
9656 "movl $dst, #-1\n\t"
9657 "jp,s done\n\t"
9658 "jb,s done\n\t"
9659 "setne $dst\n\t"
9660 "movzbl $dst, $dst\n"
9661 "done:" %}
9663 opcode(0x0F, 0x2E);
9664 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
9665 cmpfp3(dst));
9666 ins_pipe(pipe_slow);
9667 %}
9669 // Compare into -1,0,1
9670 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
9671 %{
9672 match(Set dst (CmpF3 src1 (LoadF src2)));
9673 effect(KILL cr);
9675 ins_cost(275);
9676 format %{ "ucomiss $src1, $src2\n\t"
9677 "movl $dst, #-1\n\t"
9678 "jp,s done\n\t"
9679 "jb,s done\n\t"
9680 "setne $dst\n\t"
9681 "movzbl $dst, $dst\n"
9682 "done:" %}
9684 opcode(0x0F, 0x2E);
9685 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
9686 cmpfp3(dst));
9687 ins_pipe(pipe_slow);
9688 %}
9690 // Compare into -1,0,1
9691 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
9692 %{
9693 match(Set dst (CmpF3 src1 src2));
9694 effect(KILL cr);
9696 ins_cost(275);
9697 format %{ "ucomiss $src1, [$src2]\n\t"
9698 "movl $dst, #-1\n\t"
9699 "jp,s done\n\t"
9700 "jb,s done\n\t"
9701 "setne $dst\n\t"
9702 "movzbl $dst, $dst\n"
9703 "done:" %}
9705 opcode(0x0F, 0x2E);
9706 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
9707 cmpfp3(dst));
9708 ins_pipe(pipe_slow);
9709 %}
9711 // Compare into -1,0,1
9712 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
9713 %{
9714 match(Set dst (CmpD3 src1 src2));
9715 effect(KILL cr);
9717 ins_cost(275);
9718 format %{ "ucomisd $src1, $src2\n\t"
9719 "movl $dst, #-1\n\t"
9720 "jp,s done\n\t"
9721 "jb,s done\n\t"
9722 "setne $dst\n\t"
9723 "movzbl $dst, $dst\n"
9724 "done:" %}
9726 opcode(0x66, 0x0F, 0x2E);
9727 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
9728 cmpfp3(dst));
9729 ins_pipe(pipe_slow);
9730 %}
9732 // Compare into -1,0,1
9733 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
9734 %{
9735 match(Set dst (CmpD3 src1 (LoadD src2)));
9736 effect(KILL cr);
9738 ins_cost(275);
9739 format %{ "ucomisd $src1, $src2\n\t"
9740 "movl $dst, #-1\n\t"
9741 "jp,s done\n\t"
9742 "jb,s done\n\t"
9743 "setne $dst\n\t"
9744 "movzbl $dst, $dst\n"
9745 "done:" %}
9747 opcode(0x66, 0x0F, 0x2E);
9748 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
9749 cmpfp3(dst));
9750 ins_pipe(pipe_slow);
9751 %}
9753 // Compare into -1,0,1
9754 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
9755 %{
9756 match(Set dst (CmpD3 src1 src2));
9757 effect(KILL cr);
9759 ins_cost(275);
9760 format %{ "ucomisd $src1, [$src2]\n\t"
9761 "movl $dst, #-1\n\t"
9762 "jp,s done\n\t"
9763 "jb,s done\n\t"
9764 "setne $dst\n\t"
9765 "movzbl $dst, $dst\n"
9766 "done:" %}
9768 opcode(0x66, 0x0F, 0x2E);
9769 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
9770 cmpfp3(dst));
9771 ins_pipe(pipe_slow);
9772 %}
9774 instruct addF_reg(regF dst, regF src)
9775 %{
9776 match(Set dst (AddF dst src));
9778 format %{ "addss $dst, $src" %}
9779 ins_cost(150); // XXX
9780 opcode(0xF3, 0x0F, 0x58);
9781 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9782 ins_pipe(pipe_slow);
9783 %}
9785 instruct addF_mem(regF dst, memory src)
9786 %{
9787 match(Set dst (AddF dst (LoadF src)));
9789 format %{ "addss $dst, $src" %}
9790 ins_cost(150); // XXX
9791 opcode(0xF3, 0x0F, 0x58);
9792 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9793 ins_pipe(pipe_slow);
9794 %}
9796 instruct addF_imm(regF dst, immF src)
9797 %{
9798 match(Set dst (AddF dst src));
9800 format %{ "addss $dst, [$src]" %}
9801 ins_cost(150); // XXX
9802 opcode(0xF3, 0x0F, 0x58);
9803 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
9804 ins_pipe(pipe_slow);
9805 %}
9807 instruct addD_reg(regD dst, regD src)
9808 %{
9809 match(Set dst (AddD dst src));
9811 format %{ "addsd $dst, $src" %}
9812 ins_cost(150); // XXX
9813 opcode(0xF2, 0x0F, 0x58);
9814 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9815 ins_pipe(pipe_slow);
9816 %}
9818 instruct addD_mem(regD dst, memory src)
9819 %{
9820 match(Set dst (AddD dst (LoadD src)));
9822 format %{ "addsd $dst, $src" %}
9823 ins_cost(150); // XXX
9824 opcode(0xF2, 0x0F, 0x58);
9825 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9826 ins_pipe(pipe_slow);
9827 %}
9829 instruct addD_imm(regD dst, immD src)
9830 %{
9831 match(Set dst (AddD dst src));
9833 format %{ "addsd $dst, [$src]" %}
9834 ins_cost(150); // XXX
9835 opcode(0xF2, 0x0F, 0x58);
9836 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
9837 ins_pipe(pipe_slow);
9838 %}
9840 instruct subF_reg(regF dst, regF src)
9841 %{
9842 match(Set dst (SubF dst src));
9844 format %{ "subss $dst, $src" %}
9845 ins_cost(150); // XXX
9846 opcode(0xF3, 0x0F, 0x5C);
9847 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9848 ins_pipe(pipe_slow);
9849 %}
9851 instruct subF_mem(regF dst, memory src)
9852 %{
9853 match(Set dst (SubF dst (LoadF src)));
9855 format %{ "subss $dst, $src" %}
9856 ins_cost(150); // XXX
9857 opcode(0xF3, 0x0F, 0x5C);
9858 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9859 ins_pipe(pipe_slow);
9860 %}
9862 instruct subF_imm(regF dst, immF src)
9863 %{
9864 match(Set dst (SubF dst src));
9866 format %{ "subss $dst, [$src]" %}
9867 ins_cost(150); // XXX
9868 opcode(0xF3, 0x0F, 0x5C);
9869 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
9870 ins_pipe(pipe_slow);
9871 %}
9873 instruct subD_reg(regD dst, regD src)
9874 %{
9875 match(Set dst (SubD dst src));
9877 format %{ "subsd $dst, $src" %}
9878 ins_cost(150); // XXX
9879 opcode(0xF2, 0x0F, 0x5C);
9880 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9881 ins_pipe(pipe_slow);
9882 %}
9884 instruct subD_mem(regD dst, memory src)
9885 %{
9886 match(Set dst (SubD dst (LoadD src)));
9888 format %{ "subsd $dst, $src" %}
9889 ins_cost(150); // XXX
9890 opcode(0xF2, 0x0F, 0x5C);
9891 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9892 ins_pipe(pipe_slow);
9893 %}
9895 instruct subD_imm(regD dst, immD src)
9896 %{
9897 match(Set dst (SubD dst src));
9899 format %{ "subsd $dst, [$src]" %}
9900 ins_cost(150); // XXX
9901 opcode(0xF2, 0x0F, 0x5C);
9902 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
9903 ins_pipe(pipe_slow);
9904 %}
9906 instruct mulF_reg(regF dst, regF src)
9907 %{
9908 match(Set dst (MulF dst src));
9910 format %{ "mulss $dst, $src" %}
9911 ins_cost(150); // XXX
9912 opcode(0xF3, 0x0F, 0x59);
9913 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9914 ins_pipe(pipe_slow);
9915 %}
9917 instruct mulF_mem(regF dst, memory src)
9918 %{
9919 match(Set dst (MulF dst (LoadF src)));
9921 format %{ "mulss $dst, $src" %}
9922 ins_cost(150); // XXX
9923 opcode(0xF3, 0x0F, 0x59);
9924 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9925 ins_pipe(pipe_slow);
9926 %}
9928 instruct mulF_imm(regF dst, immF src)
9929 %{
9930 match(Set dst (MulF dst src));
9932 format %{ "mulss $dst, [$src]" %}
9933 ins_cost(150); // XXX
9934 opcode(0xF3, 0x0F, 0x59);
9935 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
9936 ins_pipe(pipe_slow);
9937 %}
9939 instruct mulD_reg(regD dst, regD src)
9940 %{
9941 match(Set dst (MulD dst src));
9943 format %{ "mulsd $dst, $src" %}
9944 ins_cost(150); // XXX
9945 opcode(0xF2, 0x0F, 0x59);
9946 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9947 ins_pipe(pipe_slow);
9948 %}
9950 instruct mulD_mem(regD dst, memory src)
9951 %{
9952 match(Set dst (MulD dst (LoadD src)));
9954 format %{ "mulsd $dst, $src" %}
9955 ins_cost(150); // XXX
9956 opcode(0xF2, 0x0F, 0x59);
9957 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9958 ins_pipe(pipe_slow);
9959 %}
9961 instruct mulD_imm(regD dst, immD src)
9962 %{
9963 match(Set dst (MulD dst src));
9965 format %{ "mulsd $dst, [$src]" %}
9966 ins_cost(150); // XXX
9967 opcode(0xF2, 0x0F, 0x59);
9968 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
9969 ins_pipe(pipe_slow);
9970 %}
9972 instruct divF_reg(regF dst, regF src)
9973 %{
9974 match(Set dst (DivF dst src));
9976 format %{ "divss $dst, $src" %}
9977 ins_cost(150); // XXX
9978 opcode(0xF3, 0x0F, 0x5E);
9979 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
9980 ins_pipe(pipe_slow);
9981 %}
9983 instruct divF_mem(regF dst, memory src)
9984 %{
9985 match(Set dst (DivF dst (LoadF src)));
9987 format %{ "divss $dst, $src" %}
9988 ins_cost(150); // XXX
9989 opcode(0xF3, 0x0F, 0x5E);
9990 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
9991 ins_pipe(pipe_slow);
9992 %}
9994 instruct divF_imm(regF dst, immF src)
9995 %{
9996 match(Set dst (DivF dst src));
9998 format %{ "divss $dst, [$src]" %}
9999 ins_cost(150); // XXX
10000 opcode(0xF3, 0x0F, 0x5E);
10001 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10002 ins_pipe(pipe_slow);
10003 %}
10005 instruct divD_reg(regD dst, regD src)
10006 %{
10007 match(Set dst (DivD dst src));
10009 format %{ "divsd $dst, $src" %}
10010 ins_cost(150); // XXX
10011 opcode(0xF2, 0x0F, 0x5E);
10012 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10013 ins_pipe(pipe_slow);
10014 %}
10016 instruct divD_mem(regD dst, memory src)
10017 %{
10018 match(Set dst (DivD dst (LoadD src)));
10020 format %{ "divsd $dst, $src" %}
10021 ins_cost(150); // XXX
10022 opcode(0xF2, 0x0F, 0x5E);
10023 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10024 ins_pipe(pipe_slow);
10025 %}
10027 instruct divD_imm(regD dst, immD src)
10028 %{
10029 match(Set dst (DivD dst src));
10031 format %{ "divsd $dst, [$src]" %}
10032 ins_cost(150); // XXX
10033 opcode(0xF2, 0x0F, 0x5E);
10034 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10035 ins_pipe(pipe_slow);
10036 %}
10038 instruct sqrtF_reg(regF dst, regF src)
10039 %{
10040 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10042 format %{ "sqrtss $dst, $src" %}
10043 ins_cost(150); // XXX
10044 opcode(0xF3, 0x0F, 0x51);
10045 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10046 ins_pipe(pipe_slow);
10047 %}
10049 instruct sqrtF_mem(regF dst, memory src)
10050 %{
10051 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10053 format %{ "sqrtss $dst, $src" %}
10054 ins_cost(150); // XXX
10055 opcode(0xF3, 0x0F, 0x51);
10056 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10057 ins_pipe(pipe_slow);
10058 %}
10060 instruct sqrtF_imm(regF dst, immF src)
10061 %{
10062 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10064 format %{ "sqrtss $dst, [$src]" %}
10065 ins_cost(150); // XXX
10066 opcode(0xF3, 0x0F, 0x51);
10067 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10068 ins_pipe(pipe_slow);
10069 %}
10071 instruct sqrtD_reg(regD dst, regD src)
10072 %{
10073 match(Set dst (SqrtD src));
10075 format %{ "sqrtsd $dst, $src" %}
10076 ins_cost(150); // XXX
10077 opcode(0xF2, 0x0F, 0x51);
10078 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10079 ins_pipe(pipe_slow);
10080 %}
10082 instruct sqrtD_mem(regD dst, memory src)
10083 %{
10084 match(Set dst (SqrtD (LoadD src)));
10086 format %{ "sqrtsd $dst, $src" %}
10087 ins_cost(150); // XXX
10088 opcode(0xF2, 0x0F, 0x51);
10089 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10090 ins_pipe(pipe_slow);
10091 %}
10093 instruct sqrtD_imm(regD dst, immD src)
10094 %{
10095 match(Set dst (SqrtD src));
10097 format %{ "sqrtsd $dst, [$src]" %}
10098 ins_cost(150); // XXX
10099 opcode(0xF2, 0x0F, 0x51);
10100 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10101 ins_pipe(pipe_slow);
10102 %}
10104 instruct absF_reg(regF dst)
10105 %{
10106 match(Set dst (AbsF dst));
10108 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
10109 ins_encode(absF_encoding(dst));
10110 ins_pipe(pipe_slow);
10111 %}
10113 instruct absD_reg(regD dst)
10114 %{
10115 match(Set dst (AbsD dst));
10117 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
10118 "# abs double by sign masking" %}
10119 ins_encode(absD_encoding(dst));
10120 ins_pipe(pipe_slow);
10121 %}
10123 instruct negF_reg(regF dst)
10124 %{
10125 match(Set dst (NegF dst));
10127 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
10128 ins_encode(negF_encoding(dst));
10129 ins_pipe(pipe_slow);
10130 %}
10132 instruct negD_reg(regD dst)
10133 %{
10134 match(Set dst (NegD dst));
10136 format %{ "xorpd $dst, [0x8000000000000000]\t"
10137 "# neg double by sign flipping" %}
10138 ins_encode(negD_encoding(dst));
10139 ins_pipe(pipe_slow);
10140 %}
10142 // -----------Trig and Trancendental Instructions------------------------------
10143 instruct cosD_reg(regD dst) %{
10144 match(Set dst (CosD dst));
10146 format %{ "dcos $dst\n\t" %}
10147 opcode(0xD9, 0xFF);
10148 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10149 ins_pipe( pipe_slow );
10150 %}
10152 instruct sinD_reg(regD dst) %{
10153 match(Set dst (SinD dst));
10155 format %{ "dsin $dst\n\t" %}
10156 opcode(0xD9, 0xFE);
10157 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10158 ins_pipe( pipe_slow );
10159 %}
10161 instruct tanD_reg(regD dst) %{
10162 match(Set dst (TanD dst));
10164 format %{ "dtan $dst\n\t" %}
10165 ins_encode( Push_SrcXD(dst),
10166 Opcode(0xD9), Opcode(0xF2), //fptan
10167 Opcode(0xDD), Opcode(0xD8), //fstp st
10168 Push_ResultXD(dst) );
10169 ins_pipe( pipe_slow );
10170 %}
10172 instruct log10D_reg(regD dst) %{
10173 // The source and result Double operands in XMM registers
10174 match(Set dst (Log10D dst));
10175 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
10176 // fyl2x ; compute log_10(2) * log_2(x)
10177 format %{ "fldlg2\t\t\t#Log10\n\t"
10178 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10179 %}
10180 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
10181 Push_SrcXD(dst),
10182 Opcode(0xD9), Opcode(0xF1), // fyl2x
10183 Push_ResultXD(dst));
10185 ins_pipe( pipe_slow );
10186 %}
10188 instruct logD_reg(regD dst) %{
10189 // The source and result Double operands in XMM registers
10190 match(Set dst (LogD dst));
10191 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
10192 // fyl2x ; compute log_e(2) * log_2(x)
10193 format %{ "fldln2\t\t\t#Log_e\n\t"
10194 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10195 %}
10196 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
10197 Push_SrcXD(dst),
10198 Opcode(0xD9), Opcode(0xF1), // fyl2x
10199 Push_ResultXD(dst));
10200 ins_pipe( pipe_slow );
10201 %}
10205 //----------Arithmetic Conversion Instructions---------------------------------
10207 instruct roundFloat_nop(regF dst)
10208 %{
10209 match(Set dst (RoundFloat dst));
10211 ins_cost(0);
10212 ins_encode();
10213 ins_pipe(empty);
10214 %}
10216 instruct roundDouble_nop(regD dst)
10217 %{
10218 match(Set dst (RoundDouble dst));
10220 ins_cost(0);
10221 ins_encode();
10222 ins_pipe(empty);
10223 %}
10225 instruct convF2D_reg_reg(regD dst, regF src)
10226 %{
10227 match(Set dst (ConvF2D src));
10229 format %{ "cvtss2sd $dst, $src" %}
10230 opcode(0xF3, 0x0F, 0x5A);
10231 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10232 ins_pipe(pipe_slow); // XXX
10233 %}
10235 instruct convF2D_reg_mem(regD dst, memory src)
10236 %{
10237 match(Set dst (ConvF2D (LoadF src)));
10239 format %{ "cvtss2sd $dst, $src" %}
10240 opcode(0xF3, 0x0F, 0x5A);
10241 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10242 ins_pipe(pipe_slow); // XXX
10243 %}
10245 instruct convD2F_reg_reg(regF dst, regD src)
10246 %{
10247 match(Set dst (ConvD2F src));
10249 format %{ "cvtsd2ss $dst, $src" %}
10250 opcode(0xF2, 0x0F, 0x5A);
10251 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10252 ins_pipe(pipe_slow); // XXX
10253 %}
10255 instruct convD2F_reg_mem(regF dst, memory src)
10256 %{
10257 match(Set dst (ConvD2F (LoadD src)));
10259 format %{ "cvtsd2ss $dst, $src" %}
10260 opcode(0xF2, 0x0F, 0x5A);
10261 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10262 ins_pipe(pipe_slow); // XXX
10263 %}
10265 // XXX do mem variants
10266 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
10267 %{
10268 match(Set dst (ConvF2I src));
10269 effect(KILL cr);
10271 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
10272 "cmpl $dst, #0x80000000\n\t"
10273 "jne,s done\n\t"
10274 "subq rsp, #8\n\t"
10275 "movss [rsp], $src\n\t"
10276 "call f2i_fixup\n\t"
10277 "popq $dst\n"
10278 "done: "%}
10279 opcode(0xF3, 0x0F, 0x2C);
10280 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10281 f2i_fixup(dst, src));
10282 ins_pipe(pipe_slow);
10283 %}
10285 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
10286 %{
10287 match(Set dst (ConvF2L src));
10288 effect(KILL cr);
10290 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
10291 "cmpq $dst, [0x8000000000000000]\n\t"
10292 "jne,s done\n\t"
10293 "subq rsp, #8\n\t"
10294 "movss [rsp], $src\n\t"
10295 "call f2l_fixup\n\t"
10296 "popq $dst\n"
10297 "done: "%}
10298 opcode(0xF3, 0x0F, 0x2C);
10299 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10300 f2l_fixup(dst, src));
10301 ins_pipe(pipe_slow);
10302 %}
10304 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
10305 %{
10306 match(Set dst (ConvD2I src));
10307 effect(KILL cr);
10309 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
10310 "cmpl $dst, #0x80000000\n\t"
10311 "jne,s done\n\t"
10312 "subq rsp, #8\n\t"
10313 "movsd [rsp], $src\n\t"
10314 "call d2i_fixup\n\t"
10315 "popq $dst\n"
10316 "done: "%}
10317 opcode(0xF2, 0x0F, 0x2C);
10318 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10319 d2i_fixup(dst, src));
10320 ins_pipe(pipe_slow);
10321 %}
10323 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
10324 %{
10325 match(Set dst (ConvD2L src));
10326 effect(KILL cr);
10328 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
10329 "cmpq $dst, [0x8000000000000000]\n\t"
10330 "jne,s done\n\t"
10331 "subq rsp, #8\n\t"
10332 "movsd [rsp], $src\n\t"
10333 "call d2l_fixup\n\t"
10334 "popq $dst\n"
10335 "done: "%}
10336 opcode(0xF2, 0x0F, 0x2C);
10337 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10338 d2l_fixup(dst, src));
10339 ins_pipe(pipe_slow);
10340 %}
10342 instruct convI2F_reg_reg(regF dst, rRegI src)
10343 %{
10344 predicate(!UseXmmI2F);
10345 match(Set dst (ConvI2F src));
10347 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10348 opcode(0xF3, 0x0F, 0x2A);
10349 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10350 ins_pipe(pipe_slow); // XXX
10351 %}
10353 instruct convI2F_reg_mem(regF dst, memory src)
10354 %{
10355 match(Set dst (ConvI2F (LoadI src)));
10357 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10358 opcode(0xF3, 0x0F, 0x2A);
10359 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10360 ins_pipe(pipe_slow); // XXX
10361 %}
10363 instruct convI2D_reg_reg(regD dst, rRegI src)
10364 %{
10365 predicate(!UseXmmI2D);
10366 match(Set dst (ConvI2D src));
10368 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10369 opcode(0xF2, 0x0F, 0x2A);
10370 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10371 ins_pipe(pipe_slow); // XXX
10372 %}
10374 instruct convI2D_reg_mem(regD dst, memory src)
10375 %{
10376 match(Set dst (ConvI2D (LoadI src)));
10378 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10379 opcode(0xF2, 0x0F, 0x2A);
10380 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10381 ins_pipe(pipe_slow); // XXX
10382 %}
10384 instruct convXI2F_reg(regF dst, rRegI src)
10385 %{
10386 predicate(UseXmmI2F);
10387 match(Set dst (ConvI2F src));
10389 format %{ "movdl $dst, $src\n\t"
10390 "cvtdq2psl $dst, $dst\t# i2f" %}
10391 ins_encode %{
10392 __ movdl($dst$$XMMRegister, $src$$Register);
10393 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
10394 %}
10395 ins_pipe(pipe_slow); // XXX
10396 %}
10398 instruct convXI2D_reg(regD dst, rRegI src)
10399 %{
10400 predicate(UseXmmI2D);
10401 match(Set dst (ConvI2D src));
10403 format %{ "movdl $dst, $src\n\t"
10404 "cvtdq2pdl $dst, $dst\t# i2d" %}
10405 ins_encode %{
10406 __ movdl($dst$$XMMRegister, $src$$Register);
10407 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
10408 %}
10409 ins_pipe(pipe_slow); // XXX
10410 %}
10412 instruct convL2F_reg_reg(regF dst, rRegL src)
10413 %{
10414 match(Set dst (ConvL2F src));
10416 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
10417 opcode(0xF3, 0x0F, 0x2A);
10418 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
10419 ins_pipe(pipe_slow); // XXX
10420 %}
10422 instruct convL2F_reg_mem(regF dst, memory src)
10423 %{
10424 match(Set dst (ConvL2F (LoadL src)));
10426 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
10427 opcode(0xF3, 0x0F, 0x2A);
10428 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
10429 ins_pipe(pipe_slow); // XXX
10430 %}
10432 instruct convL2D_reg_reg(regD dst, rRegL src)
10433 %{
10434 match(Set dst (ConvL2D src));
10436 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
10437 opcode(0xF2, 0x0F, 0x2A);
10438 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
10439 ins_pipe(pipe_slow); // XXX
10440 %}
10442 instruct convL2D_reg_mem(regD dst, memory src)
10443 %{
10444 match(Set dst (ConvL2D (LoadL src)));
10446 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
10447 opcode(0xF2, 0x0F, 0x2A);
10448 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
10449 ins_pipe(pipe_slow); // XXX
10450 %}
10452 instruct convI2L_reg_reg(rRegL dst, rRegI src)
10453 %{
10454 match(Set dst (ConvI2L src));
10456 ins_cost(125);
10457 format %{ "movslq $dst, $src\t# i2l" %}
10458 opcode(0x63); // needs REX.W
10459 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
10460 ins_pipe(ialu_reg_reg);
10461 %}
10463 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
10464 // %{
10465 // match(Set dst (ConvI2L src));
10466 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
10467 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
10468 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
10469 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
10470 // ((const TypeNode*) n)->type()->is_long()->_lo ==
10471 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
10473 // format %{ "movl $dst, $src\t# unsigned i2l" %}
10474 // ins_encode(enc_copy(dst, src));
10475 // // opcode(0x63); // needs REX.W
10476 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
10477 // ins_pipe(ialu_reg_reg);
10478 // %}
10480 instruct convI2L_reg_mem(rRegL dst, memory src)
10481 %{
10482 match(Set dst (ConvI2L (LoadI src)));
10484 format %{ "movslq $dst, $src\t# i2l" %}
10485 opcode(0x63); // needs REX.W
10486 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst,src));
10487 ins_pipe(ialu_reg_mem);
10488 %}
10490 // Zero-extend convert int to long
10491 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
10492 %{
10493 match(Set dst (AndL (ConvI2L src) mask));
10495 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
10496 ins_encode(enc_copy(dst, src));
10497 ins_pipe(ialu_reg_reg);
10498 %}
10500 // Zero-extend convert int to long
10501 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
10502 %{
10503 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
10505 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
10506 opcode(0x8B);
10507 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
10508 ins_pipe(ialu_reg_mem);
10509 %}
10511 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
10512 %{
10513 match(Set dst (AndL src mask));
10515 format %{ "movl $dst, $src\t# zero-extend long" %}
10516 ins_encode(enc_copy_always(dst, src));
10517 ins_pipe(ialu_reg_reg);
10518 %}
10520 instruct convL2I_reg_reg(rRegI dst, rRegL src)
10521 %{
10522 match(Set dst (ConvL2I src));
10524 format %{ "movl $dst, $src\t# l2i" %}
10525 ins_encode(enc_copy_always(dst, src));
10526 ins_pipe(ialu_reg_reg);
10527 %}
10530 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
10531 match(Set dst (MoveF2I src));
10532 effect(DEF dst, USE src);
10534 ins_cost(125);
10535 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
10536 opcode(0x8B);
10537 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
10538 ins_pipe(ialu_reg_mem);
10539 %}
10541 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
10542 match(Set dst (MoveI2F src));
10543 effect(DEF dst, USE src);
10545 ins_cost(125);
10546 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
10547 opcode(0xF3, 0x0F, 0x10);
10548 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10549 ins_pipe(pipe_slow);
10550 %}
10552 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
10553 match(Set dst (MoveD2L src));
10554 effect(DEF dst, USE src);
10556 ins_cost(125);
10557 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
10558 opcode(0x8B);
10559 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10560 ins_pipe(ialu_reg_mem);
10561 %}
10563 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
10564 predicate(!UseXmmLoadAndClearUpper);
10565 match(Set dst (MoveL2D src));
10566 effect(DEF dst, USE src);
10568 ins_cost(125);
10569 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
10570 opcode(0x66, 0x0F, 0x12);
10571 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10572 ins_pipe(pipe_slow);
10573 %}
10575 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
10576 predicate(UseXmmLoadAndClearUpper);
10577 match(Set dst (MoveL2D src));
10578 effect(DEF dst, USE src);
10580 ins_cost(125);
10581 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
10582 opcode(0xF2, 0x0F, 0x10);
10583 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10584 ins_pipe(pipe_slow);
10585 %}
10588 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
10589 match(Set dst (MoveF2I src));
10590 effect(DEF dst, USE src);
10592 ins_cost(95); // XXX
10593 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
10594 opcode(0xF3, 0x0F, 0x11);
10595 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
10596 ins_pipe(pipe_slow);
10597 %}
10599 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
10600 match(Set dst (MoveI2F src));
10601 effect(DEF dst, USE src);
10603 ins_cost(100);
10604 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
10605 opcode(0x89);
10606 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
10607 ins_pipe( ialu_mem_reg );
10608 %}
10610 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
10611 match(Set dst (MoveD2L src));
10612 effect(DEF dst, USE src);
10614 ins_cost(95); // XXX
10615 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
10616 opcode(0xF2, 0x0F, 0x11);
10617 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
10618 ins_pipe(pipe_slow);
10619 %}
10621 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
10622 match(Set dst (MoveL2D src));
10623 effect(DEF dst, USE src);
10625 ins_cost(100);
10626 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
10627 opcode(0x89);
10628 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10629 ins_pipe(ialu_mem_reg);
10630 %}
10632 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
10633 match(Set dst (MoveF2I src));
10634 effect(DEF dst, USE src);
10635 ins_cost(85);
10636 format %{ "movd $dst,$src\t# MoveF2I" %}
10637 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
10638 ins_pipe( pipe_slow );
10639 %}
10641 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
10642 match(Set dst (MoveD2L src));
10643 effect(DEF dst, USE src);
10644 ins_cost(85);
10645 format %{ "movd $dst,$src\t# MoveD2L" %}
10646 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
10647 ins_pipe( pipe_slow );
10648 %}
10650 // The next instructions have long latency and use Int unit. Set high cost.
10651 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
10652 match(Set dst (MoveI2F src));
10653 effect(DEF dst, USE src);
10654 ins_cost(300);
10655 format %{ "movd $dst,$src\t# MoveI2F" %}
10656 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
10657 ins_pipe( pipe_slow );
10658 %}
10660 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
10661 match(Set dst (MoveL2D src));
10662 effect(DEF dst, USE src);
10663 ins_cost(300);
10664 format %{ "movd $dst,$src\t# MoveL2D" %}
10665 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
10666 ins_pipe( pipe_slow );
10667 %}
10669 // Replicate scalar to packed byte (1 byte) values in xmm
10670 instruct Repl8B_reg(regD dst, regD src) %{
10671 match(Set dst (Replicate8B src));
10672 format %{ "MOVDQA $dst,$src\n\t"
10673 "PUNPCKLBW $dst,$dst\n\t"
10674 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
10675 ins_encode( pshufd_8x8(dst, src));
10676 ins_pipe( pipe_slow );
10677 %}
10679 // Replicate scalar to packed byte (1 byte) values in xmm
10680 instruct Repl8B_rRegI(regD dst, rRegI src) %{
10681 match(Set dst (Replicate8B src));
10682 format %{ "MOVD $dst,$src\n\t"
10683 "PUNPCKLBW $dst,$dst\n\t"
10684 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
10685 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
10686 ins_pipe( pipe_slow );
10687 %}
10689 // Replicate scalar zero to packed byte (1 byte) values in xmm
10690 instruct Repl8B_immI0(regD dst, immI0 zero) %{
10691 match(Set dst (Replicate8B zero));
10692 format %{ "PXOR $dst,$dst\t! replicate8B" %}
10693 ins_encode( pxor(dst, dst));
10694 ins_pipe( fpu_reg_reg );
10695 %}
10697 // Replicate scalar to packed shore (2 byte) values in xmm
10698 instruct Repl4S_reg(regD dst, regD src) %{
10699 match(Set dst (Replicate4S src));
10700 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
10701 ins_encode( pshufd_4x16(dst, src));
10702 ins_pipe( fpu_reg_reg );
10703 %}
10705 // Replicate scalar to packed shore (2 byte) values in xmm
10706 instruct Repl4S_rRegI(regD dst, rRegI src) %{
10707 match(Set dst (Replicate4S src));
10708 format %{ "MOVD $dst,$src\n\t"
10709 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
10710 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
10711 ins_pipe( fpu_reg_reg );
10712 %}
10714 // Replicate scalar zero to packed short (2 byte) values in xmm
10715 instruct Repl4S_immI0(regD dst, immI0 zero) %{
10716 match(Set dst (Replicate4S zero));
10717 format %{ "PXOR $dst,$dst\t! replicate4S" %}
10718 ins_encode( pxor(dst, dst));
10719 ins_pipe( fpu_reg_reg );
10720 %}
10722 // Replicate scalar to packed char (2 byte) values in xmm
10723 instruct Repl4C_reg(regD dst, regD src) %{
10724 match(Set dst (Replicate4C src));
10725 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
10726 ins_encode( pshufd_4x16(dst, src));
10727 ins_pipe( fpu_reg_reg );
10728 %}
10730 // Replicate scalar to packed char (2 byte) values in xmm
10731 instruct Repl4C_rRegI(regD dst, rRegI src) %{
10732 match(Set dst (Replicate4C src));
10733 format %{ "MOVD $dst,$src\n\t"
10734 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
10735 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
10736 ins_pipe( fpu_reg_reg );
10737 %}
10739 // Replicate scalar zero to packed char (2 byte) values in xmm
10740 instruct Repl4C_immI0(regD dst, immI0 zero) %{
10741 match(Set dst (Replicate4C zero));
10742 format %{ "PXOR $dst,$dst\t! replicate4C" %}
10743 ins_encode( pxor(dst, dst));
10744 ins_pipe( fpu_reg_reg );
10745 %}
10747 // Replicate scalar to packed integer (4 byte) values in xmm
10748 instruct Repl2I_reg(regD dst, regD src) %{
10749 match(Set dst (Replicate2I src));
10750 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
10751 ins_encode( pshufd(dst, src, 0x00));
10752 ins_pipe( fpu_reg_reg );
10753 %}
10755 // Replicate scalar to packed integer (4 byte) values in xmm
10756 instruct Repl2I_rRegI(regD dst, rRegI src) %{
10757 match(Set dst (Replicate2I src));
10758 format %{ "MOVD $dst,$src\n\t"
10759 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
10760 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
10761 ins_pipe( fpu_reg_reg );
10762 %}
10764 // Replicate scalar zero to packed integer (2 byte) values in xmm
10765 instruct Repl2I_immI0(regD dst, immI0 zero) %{
10766 match(Set dst (Replicate2I zero));
10767 format %{ "PXOR $dst,$dst\t! replicate2I" %}
10768 ins_encode( pxor(dst, dst));
10769 ins_pipe( fpu_reg_reg );
10770 %}
10772 // Replicate scalar to packed single precision floating point values in xmm
10773 instruct Repl2F_reg(regD dst, regD src) %{
10774 match(Set dst (Replicate2F src));
10775 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
10776 ins_encode( pshufd(dst, src, 0xe0));
10777 ins_pipe( fpu_reg_reg );
10778 %}
10780 // Replicate scalar to packed single precision floating point values in xmm
10781 instruct Repl2F_regF(regD dst, regF src) %{
10782 match(Set dst (Replicate2F src));
10783 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
10784 ins_encode( pshufd(dst, src, 0xe0));
10785 ins_pipe( fpu_reg_reg );
10786 %}
10788 // Replicate scalar to packed single precision floating point values in xmm
10789 instruct Repl2F_immF0(regD dst, immF0 zero) %{
10790 match(Set dst (Replicate2F zero));
10791 format %{ "PXOR $dst,$dst\t! replicate2F" %}
10792 ins_encode( pxor(dst, dst));
10793 ins_pipe( fpu_reg_reg );
10794 %}
10797 // =======================================================================
10798 // fast clearing of an array
10799 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
10800 rFlagsReg cr)
10801 %{
10802 match(Set dummy (ClearArray cnt base));
10803 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
10805 format %{ "xorl rax, rax\t# ClearArray:\n\t"
10806 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
10807 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
10808 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
10809 ins_pipe(pipe_slow);
10810 %}
10812 instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1,
10813 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr)
10814 %{
10815 match(Set result (StrComp str1 str2));
10816 effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
10817 //ins_cost(300);
10819 format %{ "String Compare $str1, $str2 -> $result // XXX KILL RAX, RBX" %}
10820 ins_encode( enc_String_Compare() );
10821 ins_pipe( pipe_slow );
10822 %}
10824 //----------Control Flow Instructions------------------------------------------
10825 // Signed compare Instructions
10827 // XXX more variants!!
10828 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
10829 %{
10830 match(Set cr (CmpI op1 op2));
10831 effect(DEF cr, USE op1, USE op2);
10833 format %{ "cmpl $op1, $op2" %}
10834 opcode(0x3B); /* Opcode 3B /r */
10835 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
10836 ins_pipe(ialu_cr_reg_reg);
10837 %}
10839 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
10840 %{
10841 match(Set cr (CmpI op1 op2));
10843 format %{ "cmpl $op1, $op2" %}
10844 opcode(0x81, 0x07); /* Opcode 81 /7 */
10845 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
10846 ins_pipe(ialu_cr_reg_imm);
10847 %}
10849 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
10850 %{
10851 match(Set cr (CmpI op1 (LoadI op2)));
10853 ins_cost(500); // XXX
10854 format %{ "cmpl $op1, $op2" %}
10855 opcode(0x3B); /* Opcode 3B /r */
10856 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
10857 ins_pipe(ialu_cr_reg_mem);
10858 %}
10860 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
10861 %{
10862 match(Set cr (CmpI src zero));
10864 format %{ "testl $src, $src" %}
10865 opcode(0x85);
10866 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
10867 ins_pipe(ialu_cr_reg_imm);
10868 %}
10870 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
10871 %{
10872 match(Set cr (CmpI (AndI src con) zero));
10874 format %{ "testl $src, $con" %}
10875 opcode(0xF7, 0x00);
10876 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
10877 ins_pipe(ialu_cr_reg_imm);
10878 %}
10880 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
10881 %{
10882 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
10884 format %{ "testl $src, $mem" %}
10885 opcode(0x85);
10886 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
10887 ins_pipe(ialu_cr_reg_mem);
10888 %}
10890 // Unsigned compare Instructions; really, same as signed except they
10891 // produce an rFlagsRegU instead of rFlagsReg.
10892 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
10893 %{
10894 match(Set cr (CmpU op1 op2));
10896 format %{ "cmpl $op1, $op2\t# unsigned" %}
10897 opcode(0x3B); /* Opcode 3B /r */
10898 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
10899 ins_pipe(ialu_cr_reg_reg);
10900 %}
10902 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
10903 %{
10904 match(Set cr (CmpU op1 op2));
10906 format %{ "cmpl $op1, $op2\t# unsigned" %}
10907 opcode(0x81,0x07); /* Opcode 81 /7 */
10908 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
10909 ins_pipe(ialu_cr_reg_imm);
10910 %}
10912 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
10913 %{
10914 match(Set cr (CmpU op1 (LoadI op2)));
10916 ins_cost(500); // XXX
10917 format %{ "cmpl $op1, $op2\t# unsigned" %}
10918 opcode(0x3B); /* Opcode 3B /r */
10919 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
10920 ins_pipe(ialu_cr_reg_mem);
10921 %}
10923 // // // Cisc-spilled version of cmpU_rReg
10924 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
10925 // //%{
10926 // // match(Set cr (CmpU (LoadI op1) op2));
10927 // //
10928 // // format %{ "CMPu $op1,$op2" %}
10929 // // ins_cost(500);
10930 // // opcode(0x39); /* Opcode 39 /r */
10931 // // ins_encode( OpcP, reg_mem( op1, op2) );
10932 // //%}
10934 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
10935 %{
10936 match(Set cr (CmpU src zero));
10938 format %{ "testl $src, $src\t# unsigned" %}
10939 opcode(0x85);
10940 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
10941 ins_pipe(ialu_cr_reg_imm);
10942 %}
10944 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
10945 %{
10946 match(Set cr (CmpP op1 op2));
10948 format %{ "cmpq $op1, $op2\t# ptr" %}
10949 opcode(0x3B); /* Opcode 3B /r */
10950 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
10951 ins_pipe(ialu_cr_reg_reg);
10952 %}
10954 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
10955 %{
10956 match(Set cr (CmpP op1 (LoadP op2)));
10958 ins_cost(500); // XXX
10959 format %{ "cmpq $op1, $op2\t# ptr" %}
10960 opcode(0x3B); /* Opcode 3B /r */
10961 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
10962 ins_pipe(ialu_cr_reg_mem);
10963 %}
10965 // // // Cisc-spilled version of cmpP_rReg
10966 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
10967 // //%{
10968 // // match(Set cr (CmpP (LoadP op1) op2));
10969 // //
10970 // // format %{ "CMPu $op1,$op2" %}
10971 // // ins_cost(500);
10972 // // opcode(0x39); /* Opcode 39 /r */
10973 // // ins_encode( OpcP, reg_mem( op1, op2) );
10974 // //%}
10976 // XXX this is generalized by compP_rReg_mem???
10977 // Compare raw pointer (used in out-of-heap check).
10978 // Only works because non-oop pointers must be raw pointers
10979 // and raw pointers have no anti-dependencies.
10980 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
10981 %{
10982 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
10983 match(Set cr (CmpP op1 (LoadP op2)));
10985 format %{ "cmpq $op1, $op2\t# raw ptr" %}
10986 opcode(0x3B); /* Opcode 3B /r */
10987 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
10988 ins_pipe(ialu_cr_reg_mem);
10989 %}
10991 // This will generate a signed flags result. This should be OK since
10992 // any compare to a zero should be eq/neq.
10993 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
10994 %{
10995 match(Set cr (CmpP src zero));
10997 format %{ "testq $src, $src\t# ptr" %}
10998 opcode(0x85);
10999 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11000 ins_pipe(ialu_cr_reg_imm);
11001 %}
11003 // This will generate a signed flags result. This should be OK since
11004 // any compare to a zero should be eq/neq.
11005 instruct testP_reg_mem(rFlagsReg cr, memory op, immP0 zero)
11006 %{
11007 match(Set cr (CmpP (LoadP op) zero));
11009 ins_cost(500); // XXX
11010 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11011 opcode(0xF7); /* Opcode F7 /0 */
11012 ins_encode(REX_mem_wide(op),
11013 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11014 ins_pipe(ialu_cr_reg_imm);
11015 %}
11017 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11018 match(Set cr (CmpN src zero));
11020 format %{ "testl $src, $src" %}
11021 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11022 ins_pipe(ialu_cr_reg_imm);
11023 %}
11025 // Yanked all unsigned pointer compare operations.
11026 // Pointer compares are done with CmpP which is already unsigned.
11028 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11029 %{
11030 match(Set cr (CmpL op1 op2));
11032 format %{ "cmpq $op1, $op2" %}
11033 opcode(0x3B); /* Opcode 3B /r */
11034 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11035 ins_pipe(ialu_cr_reg_reg);
11036 %}
11038 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11039 %{
11040 match(Set cr (CmpL op1 op2));
11042 format %{ "cmpq $op1, $op2" %}
11043 opcode(0x81, 0x07); /* Opcode 81 /7 */
11044 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11045 ins_pipe(ialu_cr_reg_imm);
11046 %}
11048 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11049 %{
11050 match(Set cr (CmpL op1 (LoadL op2)));
11052 ins_cost(500); // XXX
11053 format %{ "cmpq $op1, $op2" %}
11054 opcode(0x3B); /* Opcode 3B /r */
11055 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11056 ins_pipe(ialu_cr_reg_mem);
11057 %}
11059 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11060 %{
11061 match(Set cr (CmpL src zero));
11063 format %{ "testq $src, $src" %}
11064 opcode(0x85);
11065 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11066 ins_pipe(ialu_cr_reg_imm);
11067 %}
11069 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11070 %{
11071 match(Set cr (CmpL (AndL src con) zero));
11073 format %{ "testq $src, $con\t# long" %}
11074 opcode(0xF7, 0x00);
11075 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11076 ins_pipe(ialu_cr_reg_imm);
11077 %}
11079 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
11080 %{
11081 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
11083 format %{ "testq $src, $mem" %}
11084 opcode(0x85);
11085 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
11086 ins_pipe(ialu_cr_reg_mem);
11087 %}
11089 // Manifest a CmpL result in an integer register. Very painful.
11090 // This is the test to avoid.
11091 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
11092 %{
11093 match(Set dst (CmpL3 src1 src2));
11094 effect(KILL flags);
11096 ins_cost(275); // XXX
11097 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
11098 "movl $dst, -1\n\t"
11099 "jl,s done\n\t"
11100 "setne $dst\n\t"
11101 "movzbl $dst, $dst\n\t"
11102 "done:" %}
11103 ins_encode(cmpl3_flag(src1, src2, dst));
11104 ins_pipe(pipe_slow);
11105 %}
11107 //----------Max and Min--------------------------------------------------------
11108 // Min Instructions
11110 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
11111 %{
11112 effect(USE_DEF dst, USE src, USE cr);
11114 format %{ "cmovlgt $dst, $src\t# min" %}
11115 opcode(0x0F, 0x4F);
11116 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11117 ins_pipe(pipe_cmov_reg);
11118 %}
11121 instruct minI_rReg(rRegI dst, rRegI src)
11122 %{
11123 match(Set dst (MinI dst src));
11125 ins_cost(200);
11126 expand %{
11127 rFlagsReg cr;
11128 compI_rReg(cr, dst, src);
11129 cmovI_reg_g(dst, src, cr);
11130 %}
11131 %}
11133 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
11134 %{
11135 effect(USE_DEF dst, USE src, USE cr);
11137 format %{ "cmovllt $dst, $src\t# max" %}
11138 opcode(0x0F, 0x4C);
11139 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11140 ins_pipe(pipe_cmov_reg);
11141 %}
11144 instruct maxI_rReg(rRegI dst, rRegI src)
11145 %{
11146 match(Set dst (MaxI dst src));
11148 ins_cost(200);
11149 expand %{
11150 rFlagsReg cr;
11151 compI_rReg(cr, dst, src);
11152 cmovI_reg_l(dst, src, cr);
11153 %}
11154 %}
11156 // ============================================================================
11157 // Branch Instructions
11159 // Jump Direct - Label defines a relative address from JMP+1
11160 instruct jmpDir(label labl)
11161 %{
11162 match(Goto);
11163 effect(USE labl);
11165 ins_cost(300);
11166 format %{ "jmp $labl" %}
11167 size(5);
11168 opcode(0xE9);
11169 ins_encode(OpcP, Lbl(labl));
11170 ins_pipe(pipe_jmp);
11171 ins_pc_relative(1);
11172 %}
11174 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11175 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
11176 %{
11177 match(If cop cr);
11178 effect(USE labl);
11180 ins_cost(300);
11181 format %{ "j$cop $labl" %}
11182 size(6);
11183 opcode(0x0F, 0x80);
11184 ins_encode(Jcc(cop, labl));
11185 ins_pipe(pipe_jcc);
11186 ins_pc_relative(1);
11187 %}
11189 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11190 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
11191 %{
11192 match(CountedLoopEnd cop cr);
11193 effect(USE labl);
11195 ins_cost(300);
11196 format %{ "j$cop $labl\t# loop end" %}
11197 size(6);
11198 opcode(0x0F, 0x80);
11199 ins_encode(Jcc(cop, labl));
11200 ins_pipe(pipe_jcc);
11201 ins_pc_relative(1);
11202 %}
11204 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11205 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl)
11206 %{
11207 match(CountedLoopEnd cop cmp);
11208 effect(USE labl);
11210 ins_cost(300);
11211 format %{ "j$cop,u $labl\t# loop end" %}
11212 size(6);
11213 opcode(0x0F, 0x80);
11214 ins_encode(Jcc(cop, labl));
11215 ins_pipe(pipe_jcc);
11216 ins_pc_relative(1);
11217 %}
11219 // Jump Direct Conditional - using unsigned comparison
11220 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl)
11221 %{
11222 match(If cop cmp);
11223 effect(USE labl);
11225 ins_cost(300);
11226 format %{ "j$cop,u $labl" %}
11227 size(6);
11228 opcode(0x0F, 0x80);
11229 ins_encode(Jcc(cop, labl));
11230 ins_pipe(pipe_jcc);
11231 ins_pc_relative(1);
11232 %}
11234 // ============================================================================
11235 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
11236 // superklass array for an instance of the superklass. Set a hidden
11237 // internal cache on a hit (cache is checked with exposed code in
11238 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
11239 // encoding ALSO sets flags.
11241 instruct partialSubtypeCheck(rdi_RegP result,
11242 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
11243 rFlagsReg cr)
11244 %{
11245 match(Set result (PartialSubtypeCheck sub super));
11246 effect(KILL rcx, KILL cr);
11248 ins_cost(1100); // slightly larger than the next version
11249 format %{ "cmpq rax, rsi\n\t"
11250 "jeq,s hit\n\t"
11251 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
11252 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
11253 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
11254 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
11255 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
11256 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
11257 "hit:\n\t"
11258 "xorq $result, $result\t\t Hit: rdi zero\n\t"
11259 "miss:\t" %}
11261 opcode(0x1); // Force a XOR of RDI
11262 ins_encode(enc_PartialSubtypeCheck());
11263 ins_pipe(pipe_slow);
11264 %}
11266 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
11267 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
11268 immP0 zero,
11269 rdi_RegP result)
11270 %{
11271 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
11272 predicate(!UseCompressedOops); // decoding oop kills condition codes
11273 effect(KILL rcx, KILL result);
11275 ins_cost(1000);
11276 format %{ "cmpq rax, rsi\n\t"
11277 "jeq,s miss\t# Actually a hit; we are done.\n\t"
11278 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
11279 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
11280 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
11281 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
11282 "jne,s miss\t\t# Missed: flags nz\n\t"
11283 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
11284 "miss:\t" %}
11286 opcode(0x0); // No need to XOR RDI
11287 ins_encode(enc_PartialSubtypeCheck());
11288 ins_pipe(pipe_slow);
11289 %}
11291 // ============================================================================
11292 // Branch Instructions -- short offset versions
11293 //
11294 // These instructions are used to replace jumps of a long offset (the default
11295 // match) with jumps of a shorter offset. These instructions are all tagged
11296 // with the ins_short_branch attribute, which causes the ADLC to suppress the
11297 // match rules in general matching. Instead, the ADLC generates a conversion
11298 // method in the MachNode which can be used to do in-place replacement of the
11299 // long variant with the shorter variant. The compiler will determine if a
11300 // branch can be taken by the is_short_branch_offset() predicate in the machine
11301 // specific code section of the file.
11303 // Jump Direct - Label defines a relative address from JMP+1
11304 instruct jmpDir_short(label labl)
11305 %{
11306 match(Goto);
11307 effect(USE labl);
11309 ins_cost(300);
11310 format %{ "jmp,s $labl" %}
11311 size(2);
11312 opcode(0xEB);
11313 ins_encode(OpcP, LblShort(labl));
11314 ins_pipe(pipe_jmp);
11315 ins_pc_relative(1);
11316 ins_short_branch(1);
11317 %}
11319 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11320 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl)
11321 %{
11322 match(If cop cr);
11323 effect(USE labl);
11325 ins_cost(300);
11326 format %{ "j$cop,s $labl" %}
11327 size(2);
11328 opcode(0x70);
11329 ins_encode(JccShort(cop, labl));
11330 ins_pipe(pipe_jcc);
11331 ins_pc_relative(1);
11332 ins_short_branch(1);
11333 %}
11335 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11336 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl)
11337 %{
11338 match(CountedLoopEnd cop cr);
11339 effect(USE labl);
11341 ins_cost(300);
11342 format %{ "j$cop,s $labl" %}
11343 size(2);
11344 opcode(0x70);
11345 ins_encode(JccShort(cop, labl));
11346 ins_pipe(pipe_jcc);
11347 ins_pc_relative(1);
11348 ins_short_branch(1);
11349 %}
11351 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11352 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl)
11353 %{
11354 match(CountedLoopEnd cop cmp);
11355 effect(USE labl);
11357 ins_cost(300);
11358 format %{ "j$cop,us $labl" %}
11359 size(2);
11360 opcode(0x70);
11361 ins_encode(JccShort(cop, labl));
11362 ins_pipe(pipe_jcc);
11363 ins_pc_relative(1);
11364 ins_short_branch(1);
11365 %}
11367 // Jump Direct Conditional - using unsigned comparison
11368 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl)
11369 %{
11370 match(If cop cmp);
11371 effect(USE labl);
11373 ins_cost(300);
11374 format %{ "j$cop,us $labl" %}
11375 size(2);
11376 opcode(0x70);
11377 ins_encode(JccShort(cop, labl));
11378 ins_pipe(pipe_jcc);
11379 ins_pc_relative(1);
11380 ins_short_branch(1);
11381 %}
11383 // ============================================================================
11384 // inlined locking and unlocking
11386 instruct cmpFastLock(rFlagsReg cr,
11387 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
11388 %{
11389 match(Set cr (FastLock object box));
11390 effect(TEMP tmp, TEMP scr);
11392 ins_cost(300);
11393 format %{ "fastlock $object,$box,$tmp,$scr" %}
11394 ins_encode(Fast_Lock(object, box, tmp, scr));
11395 ins_pipe(pipe_slow);
11396 ins_pc_relative(1);
11397 %}
11399 instruct cmpFastUnlock(rFlagsReg cr,
11400 rRegP object, rax_RegP box, rRegP tmp)
11401 %{
11402 match(Set cr (FastUnlock object box));
11403 effect(TEMP tmp);
11405 ins_cost(300);
11406 format %{ "fastunlock $object, $box, $tmp" %}
11407 ins_encode(Fast_Unlock(object, box, tmp));
11408 ins_pipe(pipe_slow);
11409 ins_pc_relative(1);
11410 %}
11413 // ============================================================================
11414 // Safepoint Instructions
11415 instruct safePoint_poll(rFlagsReg cr)
11416 %{
11417 match(SafePoint);
11418 effect(KILL cr);
11420 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
11421 "# Safepoint: poll for GC" %}
11422 size(6); // Opcode + ModRM + Disp32 == 6 bytes
11423 ins_cost(125);
11424 ins_encode(enc_safepoint_poll);
11425 ins_pipe(ialu_reg_mem);
11426 %}
11428 // ============================================================================
11429 // Procedure Call/Return Instructions
11430 // Call Java Static Instruction
11431 // Note: If this code changes, the corresponding ret_addr_offset() and
11432 // compute_padding() functions will have to be adjusted.
11433 instruct CallStaticJavaDirect(method meth)
11434 %{
11435 match(CallStaticJava);
11436 effect(USE meth);
11438 ins_cost(300);
11439 format %{ "call,static " %}
11440 opcode(0xE8); /* E8 cd */
11441 ins_encode(Java_Static_Call(meth), call_epilog);
11442 ins_pipe(pipe_slow);
11443 ins_pc_relative(1);
11444 ins_alignment(4);
11445 %}
11447 // Call Java Dynamic Instruction
11448 // Note: If this code changes, the corresponding ret_addr_offset() and
11449 // compute_padding() functions will have to be adjusted.
11450 instruct CallDynamicJavaDirect(method meth)
11451 %{
11452 match(CallDynamicJava);
11453 effect(USE meth);
11455 ins_cost(300);
11456 format %{ "movq rax, #Universe::non_oop_word()\n\t"
11457 "call,dynamic " %}
11458 opcode(0xE8); /* E8 cd */
11459 ins_encode(Java_Dynamic_Call(meth), call_epilog);
11460 ins_pipe(pipe_slow);
11461 ins_pc_relative(1);
11462 ins_alignment(4);
11463 %}
11465 // Call Runtime Instruction
11466 instruct CallRuntimeDirect(method meth)
11467 %{
11468 match(CallRuntime);
11469 effect(USE meth);
11471 ins_cost(300);
11472 format %{ "call,runtime " %}
11473 opcode(0xE8); /* E8 cd */
11474 ins_encode(Java_To_Runtime(meth));
11475 ins_pipe(pipe_slow);
11476 ins_pc_relative(1);
11477 %}
11479 // Call runtime without safepoint
11480 instruct CallLeafDirect(method meth)
11481 %{
11482 match(CallLeaf);
11483 effect(USE meth);
11485 ins_cost(300);
11486 format %{ "call_leaf,runtime " %}
11487 opcode(0xE8); /* E8 cd */
11488 ins_encode(Java_To_Runtime(meth));
11489 ins_pipe(pipe_slow);
11490 ins_pc_relative(1);
11491 %}
11493 // Call runtime without safepoint
11494 instruct CallLeafNoFPDirect(method meth)
11495 %{
11496 match(CallLeafNoFP);
11497 effect(USE meth);
11499 ins_cost(300);
11500 format %{ "call_leaf_nofp,runtime " %}
11501 opcode(0xE8); /* E8 cd */
11502 ins_encode(Java_To_Runtime(meth));
11503 ins_pipe(pipe_slow);
11504 ins_pc_relative(1);
11505 %}
11507 // Return Instruction
11508 // Remove the return address & jump to it.
11509 // Notice: We always emit a nop after a ret to make sure there is room
11510 // for safepoint patching
11511 instruct Ret()
11512 %{
11513 match(Return);
11515 format %{ "ret" %}
11516 opcode(0xC3);
11517 ins_encode(OpcP);
11518 ins_pipe(pipe_jmp);
11519 %}
11521 // Tail Call; Jump from runtime stub to Java code.
11522 // Also known as an 'interprocedural jump'.
11523 // Target of jump will eventually return to caller.
11524 // TailJump below removes the return address.
11525 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
11526 %{
11527 match(TailCall jump_target method_oop);
11529 ins_cost(300);
11530 format %{ "jmp $jump_target\t# rbx holds method oop" %}
11531 opcode(0xFF, 0x4); /* Opcode FF /4 */
11532 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
11533 ins_pipe(pipe_jmp);
11534 %}
11536 // Tail Jump; remove the return address; jump to target.
11537 // TailCall above leaves the return address around.
11538 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
11539 %{
11540 match(TailJump jump_target ex_oop);
11542 ins_cost(300);
11543 format %{ "popq rdx\t# pop return address\n\t"
11544 "jmp $jump_target" %}
11545 opcode(0xFF, 0x4); /* Opcode FF /4 */
11546 ins_encode(Opcode(0x5a), // popq rdx
11547 REX_reg(jump_target), OpcP, reg_opc(jump_target));
11548 ins_pipe(pipe_jmp);
11549 %}
11551 // Create exception oop: created by stack-crawling runtime code.
11552 // Created exception is now available to this handler, and is setup
11553 // just prior to jumping to this handler. No code emitted.
11554 instruct CreateException(rax_RegP ex_oop)
11555 %{
11556 match(Set ex_oop (CreateEx));
11558 size(0);
11559 // use the following format syntax
11560 format %{ "# exception oop is in rax; no code emitted" %}
11561 ins_encode();
11562 ins_pipe(empty);
11563 %}
11565 // Rethrow exception:
11566 // The exception oop will come in the first argument position.
11567 // Then JUMP (not call) to the rethrow stub code.
11568 instruct RethrowException()
11569 %{
11570 match(Rethrow);
11572 // use the following format syntax
11573 format %{ "jmp rethrow_stub" %}
11574 ins_encode(enc_rethrow);
11575 ins_pipe(pipe_jmp);
11576 %}
11579 //----------PEEPHOLE RULES-----------------------------------------------------
11580 // These must follow all instruction definitions as they use the names
11581 // defined in the instructions definitions.
11582 //
11583 // peepmatch ( root_instr_name [precerding_instruction]* );
11584 //
11585 // peepconstraint %{
11586 // (instruction_number.operand_name relational_op instruction_number.operand_name
11587 // [, ...] );
11588 // // instruction numbers are zero-based using left to right order in peepmatch
11589 //
11590 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
11591 // // provide an instruction_number.operand_name for each operand that appears
11592 // // in the replacement instruction's match rule
11593 //
11594 // ---------VM FLAGS---------------------------------------------------------
11595 //
11596 // All peephole optimizations can be turned off using -XX:-OptoPeephole
11597 //
11598 // Each peephole rule is given an identifying number starting with zero and
11599 // increasing by one in the order seen by the parser. An individual peephole
11600 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
11601 // on the command-line.
11602 //
11603 // ---------CURRENT LIMITATIONS----------------------------------------------
11604 //
11605 // Only match adjacent instructions in same basic block
11606 // Only equality constraints
11607 // Only constraints between operands, not (0.dest_reg == RAX_enc)
11608 // Only one replacement instruction
11609 //
11610 // ---------EXAMPLE----------------------------------------------------------
11611 //
11612 // // pertinent parts of existing instructions in architecture description
11613 // instruct movI(rRegI dst, rRegI src)
11614 // %{
11615 // match(Set dst (CopyI src));
11616 // %}
11617 //
11618 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
11619 // %{
11620 // match(Set dst (AddI dst src));
11621 // effect(KILL cr);
11622 // %}
11623 //
11624 // // Change (inc mov) to lea
11625 // peephole %{
11626 // // increment preceeded by register-register move
11627 // peepmatch ( incI_rReg movI );
11628 // // require that the destination register of the increment
11629 // // match the destination register of the move
11630 // peepconstraint ( 0.dst == 1.dst );
11631 // // construct a replacement instruction that sets
11632 // // the destination to ( move's source register + one )
11633 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
11634 // %}
11635 //
11637 // Implementation no longer uses movX instructions since
11638 // machine-independent system no longer uses CopyX nodes.
11639 //
11640 // peephole
11641 // %{
11642 // peepmatch (incI_rReg movI);
11643 // peepconstraint (0.dst == 1.dst);
11644 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
11645 // %}
11647 // peephole
11648 // %{
11649 // peepmatch (decI_rReg movI);
11650 // peepconstraint (0.dst == 1.dst);
11651 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
11652 // %}
11654 // peephole
11655 // %{
11656 // peepmatch (addI_rReg_imm movI);
11657 // peepconstraint (0.dst == 1.dst);
11658 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
11659 // %}
11661 // peephole
11662 // %{
11663 // peepmatch (incL_rReg movL);
11664 // peepconstraint (0.dst == 1.dst);
11665 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
11666 // %}
11668 // peephole
11669 // %{
11670 // peepmatch (decL_rReg movL);
11671 // peepconstraint (0.dst == 1.dst);
11672 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
11673 // %}
11675 // peephole
11676 // %{
11677 // peepmatch (addL_rReg_imm movL);
11678 // peepconstraint (0.dst == 1.dst);
11679 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
11680 // %}
11682 // peephole
11683 // %{
11684 // peepmatch (addP_rReg_imm movP);
11685 // peepconstraint (0.dst == 1.dst);
11686 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
11687 // %}
11689 // // Change load of spilled value to only a spill
11690 // instruct storeI(memory mem, rRegI src)
11691 // %{
11692 // match(Set mem (StoreI mem src));
11693 // %}
11694 //
11695 // instruct loadI(rRegI dst, memory mem)
11696 // %{
11697 // match(Set dst (LoadI mem));
11698 // %}
11699 //
11701 peephole
11702 %{
11703 peepmatch (loadI storeI);
11704 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
11705 peepreplace (storeI(1.mem 1.mem 1.src));
11706 %}
11708 peephole
11709 %{
11710 peepmatch (loadL storeL);
11711 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
11712 peepreplace (storeL(1.mem 1.mem 1.src));
11713 %}
11715 //----------SMARTSPILL RULES---------------------------------------------------
11716 // These must follow all instruction definitions as they use the names
11717 // defined in the instructions definitions.