Tue, 28 Oct 2008 09:31:30 -0700
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
Reviewed-by: kvn, rasbold
1 //
2 // Copyright 2003-2008 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::imm_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.push(rax);
966 masm.mov(rax, rsp);
967 masm.andptr(rax, StackAlignmentInBytes-1);
968 masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
969 masm.pop(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 // This is recognized as unresolved by relocs/nativeinst/ic code
1821 __ jump(RuntimeAddress(__ pc()));
1823 // Update current stubs pointer and restore code_end.
1824 __ end_a_stub();
1825 }
1827 // size of call stub, compiled java to interpretor
1828 uint size_java_to_interp()
1829 {
1830 return 15; // movq (1+1+8); jmp (1+4)
1831 }
1833 // relocation entries for call stub, compiled java to interpretor
1834 uint reloc_java_to_interp()
1835 {
1836 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1837 }
1839 //=============================================================================
1840 #ifndef PRODUCT
1841 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1842 {
1843 if (UseCompressedOops) {
1844 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
1845 st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]");
1846 st->print_cr("cmpq rax, rscratch1\t # Inline cache check");
1847 } else {
1848 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
1849 "# Inline cache check", oopDesc::klass_offset_in_bytes());
1850 }
1851 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1852 st->print_cr("\tnop");
1853 if (!OptoBreakpoint) {
1854 st->print_cr("\tnop");
1855 }
1856 }
1857 #endif
1859 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1860 {
1861 MacroAssembler masm(&cbuf);
1862 #ifdef ASSERT
1863 uint code_size = cbuf.code_size();
1864 #endif
1865 if (UseCompressedOops) {
1866 masm.load_klass(rscratch1, j_rarg0);
1867 masm.cmpptr(rax, rscratch1);
1868 } else {
1869 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1870 }
1872 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1874 /* WARNING these NOPs are critical so that verified entry point is properly
1875 aligned for patching by NativeJump::patch_verified_entry() */
1876 int nops_cnt = 1;
1877 if (!OptoBreakpoint) {
1878 // Leave space for int3
1879 nops_cnt += 1;
1880 }
1881 if (UseCompressedOops) {
1882 // ??? divisible by 4 is aligned?
1883 nops_cnt += 1;
1884 }
1885 masm.nop(nops_cnt);
1887 assert(cbuf.code_size() - code_size == size(ra_),
1888 "checking code size of inline cache node");
1889 }
1891 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1892 {
1893 if (UseCompressedOops) {
1894 return OptoBreakpoint ? 19 : 20;
1895 } else {
1896 return OptoBreakpoint ? 11 : 12;
1897 }
1898 }
1901 //=============================================================================
1902 uint size_exception_handler()
1903 {
1904 // NativeCall instruction size is the same as NativeJump.
1905 // Note that this value is also credited (in output.cpp) to
1906 // the size of the code section.
1907 return NativeJump::instruction_size;
1908 }
1910 // Emit exception handler code.
1911 int emit_exception_handler(CodeBuffer& cbuf)
1912 {
1914 // Note that the code buffer's inst_mark is always relative to insts.
1915 // That's why we must use the macroassembler to generate a handler.
1916 MacroAssembler _masm(&cbuf);
1917 address base =
1918 __ start_a_stub(size_exception_handler());
1919 if (base == NULL) return 0; // CodeBuffer::expand failed
1920 int offset = __ offset();
1921 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
1922 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1923 __ end_a_stub();
1924 return offset;
1925 }
1927 uint size_deopt_handler()
1928 {
1929 // three 5 byte instructions
1930 return 15;
1931 }
1933 // Emit deopt handler code.
1934 int emit_deopt_handler(CodeBuffer& cbuf)
1935 {
1937 // Note that the code buffer's inst_mark is always relative to insts.
1938 // That's why we must use the macroassembler to generate a handler.
1939 MacroAssembler _masm(&cbuf);
1940 address base =
1941 __ start_a_stub(size_deopt_handler());
1942 if (base == NULL) return 0; // CodeBuffer::expand failed
1943 int offset = __ offset();
1944 address the_pc = (address) __ pc();
1945 Label next;
1946 // push a "the_pc" on the stack without destroying any registers
1947 // as they all may be live.
1949 // push address of "next"
1950 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1951 __ bind(next);
1952 // adjust it so it matches "the_pc"
1953 __ subptr(Address(rsp, 0), __ offset() - offset);
1954 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1955 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1956 __ end_a_stub();
1957 return offset;
1958 }
1960 static void emit_double_constant(CodeBuffer& cbuf, double x) {
1961 int mark = cbuf.insts()->mark_off();
1962 MacroAssembler _masm(&cbuf);
1963 address double_address = __ double_constant(x);
1964 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1965 emit_d32_reloc(cbuf,
1966 (int) (double_address - cbuf.code_end() - 4),
1967 internal_word_Relocation::spec(double_address),
1968 RELOC_DISP32);
1969 }
1971 static void emit_float_constant(CodeBuffer& cbuf, float x) {
1972 int mark = cbuf.insts()->mark_off();
1973 MacroAssembler _masm(&cbuf);
1974 address float_address = __ float_constant(x);
1975 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1976 emit_d32_reloc(cbuf,
1977 (int) (float_address - cbuf.code_end() - 4),
1978 internal_word_Relocation::spec(float_address),
1979 RELOC_DISP32);
1980 }
1983 int Matcher::regnum_to_fpu_offset(int regnum)
1984 {
1985 return regnum - 32; // The FP registers are in the second chunk
1986 }
1988 // This is UltraSparc specific, true just means we have fast l2f conversion
1989 const bool Matcher::convL2FSupported(void) {
1990 return true;
1991 }
1993 // Vector width in bytes
1994 const uint Matcher::vector_width_in_bytes(void) {
1995 return 8;
1996 }
1998 // Vector ideal reg
1999 const uint Matcher::vector_ideal_reg(void) {
2000 return Op_RegD;
2001 }
2003 // Is this branch offset short enough that a short branch can be used?
2004 //
2005 // NOTE: If the platform does not provide any short branch variants, then
2006 // this method should return false for offset 0.
2007 bool Matcher::is_short_branch_offset(int rule, int offset) {
2008 // the short version of jmpConUCF2 contains multiple branches,
2009 // making the reach slightly less
2010 if (rule == jmpConUCF2_rule)
2011 return (-126 <= offset && offset <= 125);
2012 return (-128 <= offset && offset <= 127);
2013 }
2015 const bool Matcher::isSimpleConstant64(jlong value) {
2016 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
2017 //return value == (int) value; // Cf. storeImmL and immL32.
2019 // Probably always true, even if a temp register is required.
2020 return true;
2021 }
2023 // The ecx parameter to rep stosq for the ClearArray node is in words.
2024 const bool Matcher::init_array_count_is_in_bytes = false;
2026 // Threshold size for cleararray.
2027 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2029 // Should the Matcher clone shifts on addressing modes, expecting them
2030 // to be subsumed into complex addressing expressions or compute them
2031 // into registers? True for Intel but false for most RISCs
2032 const bool Matcher::clone_shift_expressions = true;
2034 // Is it better to copy float constants, or load them directly from
2035 // memory? Intel can load a float constant from a direct address,
2036 // requiring no extra registers. Most RISCs will have to materialize
2037 // an address into a register first, so they would do better to copy
2038 // the constant from stack.
2039 const bool Matcher::rematerialize_float_constants = true; // XXX
2041 // If CPU can load and store mis-aligned doubles directly then no
2042 // fixup is needed. Else we split the double into 2 integer pieces
2043 // and move it piece-by-piece. Only happens when passing doubles into
2044 // C code as the Java calling convention forces doubles to be aligned.
2045 const bool Matcher::misaligned_doubles_ok = true;
2047 // No-op on amd64
2048 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2050 // Advertise here if the CPU requires explicit rounding operations to
2051 // implement the UseStrictFP mode.
2052 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2054 // Do floats take an entire double register or just half?
2055 const bool Matcher::float_in_double = true;
2056 // Do ints take an entire long register or just half?
2057 const bool Matcher::int_in_long = true;
2059 // Return whether or not this register is ever used as an argument.
2060 // This function is used on startup to build the trampoline stubs in
2061 // generateOptoStub. Registers not mentioned will be killed by the VM
2062 // call in the trampoline, and arguments in those registers not be
2063 // available to the callee.
2064 bool Matcher::can_be_java_arg(int reg)
2065 {
2066 return
2067 reg == RDI_num || reg == RDI_H_num ||
2068 reg == RSI_num || reg == RSI_H_num ||
2069 reg == RDX_num || reg == RDX_H_num ||
2070 reg == RCX_num || reg == RCX_H_num ||
2071 reg == R8_num || reg == R8_H_num ||
2072 reg == R9_num || reg == R9_H_num ||
2073 reg == R12_num || reg == R12_H_num ||
2074 reg == XMM0_num || reg == XMM0_H_num ||
2075 reg == XMM1_num || reg == XMM1_H_num ||
2076 reg == XMM2_num || reg == XMM2_H_num ||
2077 reg == XMM3_num || reg == XMM3_H_num ||
2078 reg == XMM4_num || reg == XMM4_H_num ||
2079 reg == XMM5_num || reg == XMM5_H_num ||
2080 reg == XMM6_num || reg == XMM6_H_num ||
2081 reg == XMM7_num || reg == XMM7_H_num;
2082 }
2084 bool Matcher::is_spillable_arg(int reg)
2085 {
2086 return can_be_java_arg(reg);
2087 }
2089 // Register for DIVI projection of divmodI
2090 RegMask Matcher::divI_proj_mask() {
2091 return INT_RAX_REG_mask;
2092 }
2094 // Register for MODI projection of divmodI
2095 RegMask Matcher::modI_proj_mask() {
2096 return INT_RDX_REG_mask;
2097 }
2099 // Register for DIVL projection of divmodL
2100 RegMask Matcher::divL_proj_mask() {
2101 return LONG_RAX_REG_mask;
2102 }
2104 // Register for MODL projection of divmodL
2105 RegMask Matcher::modL_proj_mask() {
2106 return LONG_RDX_REG_mask;
2107 }
2109 static Address build_address(int b, int i, int s, int d) {
2110 Register index = as_Register(i);
2111 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2112 if (index == rsp) {
2113 index = noreg;
2114 scale = Address::no_scale;
2115 }
2116 Address addr(as_Register(b), index, scale, d);
2117 return addr;
2118 }
2120 %}
2122 //----------ENCODING BLOCK-----------------------------------------------------
2123 // This block specifies the encoding classes used by the compiler to
2124 // output byte streams. Encoding classes are parameterized macros
2125 // used by Machine Instruction Nodes in order to generate the bit
2126 // encoding of the instruction. Operands specify their base encoding
2127 // interface with the interface keyword. There are currently
2128 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2129 // COND_INTER. REG_INTER causes an operand to generate a function
2130 // which returns its register number when queried. CONST_INTER causes
2131 // an operand to generate a function which returns the value of the
2132 // constant when queried. MEMORY_INTER causes an operand to generate
2133 // four functions which return the Base Register, the Index Register,
2134 // the Scale Value, and the Offset Value of the operand when queried.
2135 // COND_INTER causes an operand to generate six functions which return
2136 // the encoding code (ie - encoding bits for the instruction)
2137 // associated with each basic boolean condition for a conditional
2138 // instruction.
2139 //
2140 // Instructions specify two basic values for encoding. Again, a
2141 // function is available to check if the constant displacement is an
2142 // oop. They use the ins_encode keyword to specify their encoding
2143 // classes (which must be a sequence of enc_class names, and their
2144 // parameters, specified in the encoding block), and they use the
2145 // opcode keyword to specify, in order, their primary, secondary, and
2146 // tertiary opcode. Only the opcode sections which a particular
2147 // instruction needs for encoding need to be specified.
2148 encode %{
2149 // Build emit functions for each basic byte or larger field in the
2150 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2151 // from C++ code in the enc_class source block. Emit functions will
2152 // live in the main source block for now. In future, we can
2153 // generalize this by adding a syntax that specifies the sizes of
2154 // fields in an order, so that the adlc can build the emit functions
2155 // automagically
2157 // Emit primary opcode
2158 enc_class OpcP
2159 %{
2160 emit_opcode(cbuf, $primary);
2161 %}
2163 // Emit secondary opcode
2164 enc_class OpcS
2165 %{
2166 emit_opcode(cbuf, $secondary);
2167 %}
2169 // Emit tertiary opcode
2170 enc_class OpcT
2171 %{
2172 emit_opcode(cbuf, $tertiary);
2173 %}
2175 // Emit opcode directly
2176 enc_class Opcode(immI d8)
2177 %{
2178 emit_opcode(cbuf, $d8$$constant);
2179 %}
2181 // Emit size prefix
2182 enc_class SizePrefix
2183 %{
2184 emit_opcode(cbuf, 0x66);
2185 %}
2187 enc_class reg(rRegI reg)
2188 %{
2189 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2190 %}
2192 enc_class reg_reg(rRegI dst, rRegI src)
2193 %{
2194 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2195 %}
2197 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2198 %{
2199 emit_opcode(cbuf, $opcode$$constant);
2200 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2201 %}
2203 enc_class cmpfp_fixup()
2204 %{
2205 // jnp,s exit
2206 emit_opcode(cbuf, 0x7B);
2207 emit_d8(cbuf, 0x0A);
2209 // pushfq
2210 emit_opcode(cbuf, 0x9C);
2212 // andq $0xffffff2b, (%rsp)
2213 emit_opcode(cbuf, Assembler::REX_W);
2214 emit_opcode(cbuf, 0x81);
2215 emit_opcode(cbuf, 0x24);
2216 emit_opcode(cbuf, 0x24);
2217 emit_d32(cbuf, 0xffffff2b);
2219 // popfq
2220 emit_opcode(cbuf, 0x9D);
2222 // nop (target for branch to avoid branch to branch)
2223 emit_opcode(cbuf, 0x90);
2224 %}
2226 enc_class cmpfp3(rRegI dst)
2227 %{
2228 int dstenc = $dst$$reg;
2230 // movl $dst, -1
2231 if (dstenc >= 8) {
2232 emit_opcode(cbuf, Assembler::REX_B);
2233 }
2234 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2235 emit_d32(cbuf, -1);
2237 // jp,s done
2238 emit_opcode(cbuf, 0x7A);
2239 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2241 // jb,s done
2242 emit_opcode(cbuf, 0x72);
2243 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2245 // setne $dst
2246 if (dstenc >= 4) {
2247 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2248 }
2249 emit_opcode(cbuf, 0x0F);
2250 emit_opcode(cbuf, 0x95);
2251 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2253 // movzbl $dst, $dst
2254 if (dstenc >= 4) {
2255 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2256 }
2257 emit_opcode(cbuf, 0x0F);
2258 emit_opcode(cbuf, 0xB6);
2259 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2260 %}
2262 enc_class cdql_enc(no_rax_rdx_RegI div)
2263 %{
2264 // Full implementation of Java idiv and irem; checks for
2265 // special case as described in JVM spec., p.243 & p.271.
2266 //
2267 // normal case special case
2268 //
2269 // input : rax: dividend min_int
2270 // reg: divisor -1
2271 //
2272 // output: rax: quotient (= rax idiv reg) min_int
2273 // rdx: remainder (= rax irem reg) 0
2274 //
2275 // Code sequnce:
2276 //
2277 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2278 // 5: 75 07/08 jne e <normal>
2279 // 7: 33 d2 xor %edx,%edx
2280 // [div >= 8 -> offset + 1]
2281 // [REX_B]
2282 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2283 // c: 74 03/04 je 11 <done>
2284 // 000000000000000e <normal>:
2285 // e: 99 cltd
2286 // [div >= 8 -> offset + 1]
2287 // [REX_B]
2288 // f: f7 f9 idiv $div
2289 // 0000000000000011 <done>:
2291 // cmp $0x80000000,%eax
2292 emit_opcode(cbuf, 0x3d);
2293 emit_d8(cbuf, 0x00);
2294 emit_d8(cbuf, 0x00);
2295 emit_d8(cbuf, 0x00);
2296 emit_d8(cbuf, 0x80);
2298 // jne e <normal>
2299 emit_opcode(cbuf, 0x75);
2300 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2302 // xor %edx,%edx
2303 emit_opcode(cbuf, 0x33);
2304 emit_d8(cbuf, 0xD2);
2306 // cmp $0xffffffffffffffff,%ecx
2307 if ($div$$reg >= 8) {
2308 emit_opcode(cbuf, Assembler::REX_B);
2309 }
2310 emit_opcode(cbuf, 0x83);
2311 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2312 emit_d8(cbuf, 0xFF);
2314 // je 11 <done>
2315 emit_opcode(cbuf, 0x74);
2316 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2318 // <normal>
2319 // cltd
2320 emit_opcode(cbuf, 0x99);
2322 // idivl (note: must be emitted by the user of this rule)
2323 // <done>
2324 %}
2326 enc_class cdqq_enc(no_rax_rdx_RegL div)
2327 %{
2328 // Full implementation of Java ldiv and lrem; checks for
2329 // special case as described in JVM spec., p.243 & p.271.
2330 //
2331 // normal case special case
2332 //
2333 // input : rax: dividend min_long
2334 // reg: divisor -1
2335 //
2336 // output: rax: quotient (= rax idiv reg) min_long
2337 // rdx: remainder (= rax irem reg) 0
2338 //
2339 // Code sequnce:
2340 //
2341 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2342 // 7: 00 00 80
2343 // a: 48 39 d0 cmp %rdx,%rax
2344 // d: 75 08 jne 17 <normal>
2345 // f: 33 d2 xor %edx,%edx
2346 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2347 // 15: 74 05 je 1c <done>
2348 // 0000000000000017 <normal>:
2349 // 17: 48 99 cqto
2350 // 19: 48 f7 f9 idiv $div
2351 // 000000000000001c <done>:
2353 // mov $0x8000000000000000,%rdx
2354 emit_opcode(cbuf, Assembler::REX_W);
2355 emit_opcode(cbuf, 0xBA);
2356 emit_d8(cbuf, 0x00);
2357 emit_d8(cbuf, 0x00);
2358 emit_d8(cbuf, 0x00);
2359 emit_d8(cbuf, 0x00);
2360 emit_d8(cbuf, 0x00);
2361 emit_d8(cbuf, 0x00);
2362 emit_d8(cbuf, 0x00);
2363 emit_d8(cbuf, 0x80);
2365 // cmp %rdx,%rax
2366 emit_opcode(cbuf, Assembler::REX_W);
2367 emit_opcode(cbuf, 0x39);
2368 emit_d8(cbuf, 0xD0);
2370 // jne 17 <normal>
2371 emit_opcode(cbuf, 0x75);
2372 emit_d8(cbuf, 0x08);
2374 // xor %edx,%edx
2375 emit_opcode(cbuf, 0x33);
2376 emit_d8(cbuf, 0xD2);
2378 // cmp $0xffffffffffffffff,$div
2379 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2380 emit_opcode(cbuf, 0x83);
2381 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2382 emit_d8(cbuf, 0xFF);
2384 // je 1e <done>
2385 emit_opcode(cbuf, 0x74);
2386 emit_d8(cbuf, 0x05);
2388 // <normal>
2389 // cqto
2390 emit_opcode(cbuf, Assembler::REX_W);
2391 emit_opcode(cbuf, 0x99);
2393 // idivq (note: must be emitted by the user of this rule)
2394 // <done>
2395 %}
2397 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2398 enc_class OpcSE(immI imm)
2399 %{
2400 // Emit primary opcode and set sign-extend bit
2401 // Check for 8-bit immediate, and set sign extend bit in opcode
2402 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2403 emit_opcode(cbuf, $primary | 0x02);
2404 } else {
2405 // 32-bit immediate
2406 emit_opcode(cbuf, $primary);
2407 }
2408 %}
2410 enc_class OpcSErm(rRegI dst, immI imm)
2411 %{
2412 // OpcSEr/m
2413 int dstenc = $dst$$reg;
2414 if (dstenc >= 8) {
2415 emit_opcode(cbuf, Assembler::REX_B);
2416 dstenc -= 8;
2417 }
2418 // Emit primary opcode and set sign-extend bit
2419 // Check for 8-bit immediate, and set sign extend bit in opcode
2420 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2421 emit_opcode(cbuf, $primary | 0x02);
2422 } else {
2423 // 32-bit immediate
2424 emit_opcode(cbuf, $primary);
2425 }
2426 // Emit r/m byte with secondary opcode, after primary opcode.
2427 emit_rm(cbuf, 0x3, $secondary, dstenc);
2428 %}
2430 enc_class OpcSErm_wide(rRegL dst, immI imm)
2431 %{
2432 // OpcSEr/m
2433 int dstenc = $dst$$reg;
2434 if (dstenc < 8) {
2435 emit_opcode(cbuf, Assembler::REX_W);
2436 } else {
2437 emit_opcode(cbuf, Assembler::REX_WB);
2438 dstenc -= 8;
2439 }
2440 // Emit primary opcode and set sign-extend bit
2441 // Check for 8-bit immediate, and set sign extend bit in opcode
2442 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2443 emit_opcode(cbuf, $primary | 0x02);
2444 } else {
2445 // 32-bit immediate
2446 emit_opcode(cbuf, $primary);
2447 }
2448 // Emit r/m byte with secondary opcode, after primary opcode.
2449 emit_rm(cbuf, 0x3, $secondary, dstenc);
2450 %}
2452 enc_class Con8or32(immI imm)
2453 %{
2454 // Check for 8-bit immediate, and set sign extend bit in opcode
2455 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2456 $$$emit8$imm$$constant;
2457 } else {
2458 // 32-bit immediate
2459 $$$emit32$imm$$constant;
2460 }
2461 %}
2463 enc_class Lbl(label labl)
2464 %{
2465 // JMP, CALL
2466 Label* l = $labl$$label;
2467 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2468 %}
2470 enc_class LblShort(label labl)
2471 %{
2472 // JMP, CALL
2473 Label* l = $labl$$label;
2474 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2475 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2476 emit_d8(cbuf, disp);
2477 %}
2479 enc_class opc2_reg(rRegI dst)
2480 %{
2481 // BSWAP
2482 emit_cc(cbuf, $secondary, $dst$$reg);
2483 %}
2485 enc_class opc3_reg(rRegI dst)
2486 %{
2487 // BSWAP
2488 emit_cc(cbuf, $tertiary, $dst$$reg);
2489 %}
2491 enc_class reg_opc(rRegI div)
2492 %{
2493 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2494 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2495 %}
2497 enc_class Jcc(cmpOp cop, label labl)
2498 %{
2499 // JCC
2500 Label* l = $labl$$label;
2501 $$$emit8$primary;
2502 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2503 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2504 %}
2506 enc_class JccShort (cmpOp cop, label labl)
2507 %{
2508 // JCC
2509 Label *l = $labl$$label;
2510 emit_cc(cbuf, $primary, $cop$$cmpcode);
2511 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2512 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2513 emit_d8(cbuf, disp);
2514 %}
2516 enc_class enc_cmov(cmpOp cop)
2517 %{
2518 // CMOV
2519 $$$emit8$primary;
2520 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2521 %}
2523 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2524 %{
2525 // Invert sense of branch from sense of cmov
2526 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2527 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2528 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2529 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2530 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2531 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2532 if ($dst$$reg < 8) {
2533 if ($src$$reg >= 8) {
2534 emit_opcode(cbuf, Assembler::REX_B);
2535 }
2536 } else {
2537 if ($src$$reg < 8) {
2538 emit_opcode(cbuf, Assembler::REX_R);
2539 } else {
2540 emit_opcode(cbuf, Assembler::REX_RB);
2541 }
2542 }
2543 emit_opcode(cbuf, 0x0F);
2544 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2545 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2546 %}
2548 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2549 %{
2550 // Invert sense of branch from sense of cmov
2551 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2552 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2554 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2555 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2556 if ($dst$$reg < 8) {
2557 if ($src$$reg >= 8) {
2558 emit_opcode(cbuf, Assembler::REX_B);
2559 }
2560 } else {
2561 if ($src$$reg < 8) {
2562 emit_opcode(cbuf, Assembler::REX_R);
2563 } else {
2564 emit_opcode(cbuf, Assembler::REX_RB);
2565 }
2566 }
2567 emit_opcode(cbuf, 0x0F);
2568 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2569 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2570 %}
2572 enc_class enc_PartialSubtypeCheck()
2573 %{
2574 Register Rrdi = as_Register(RDI_enc); // result register
2575 Register Rrax = as_Register(RAX_enc); // super class
2576 Register Rrcx = as_Register(RCX_enc); // killed
2577 Register Rrsi = as_Register(RSI_enc); // sub class
2578 Label hit, miss, cmiss;
2580 MacroAssembler _masm(&cbuf);
2581 // Compare super with sub directly, since super is not in its own SSA.
2582 // The compiler used to emit this test, but we fold it in here,
2583 // to allow platform-specific tweaking on sparc.
2584 __ cmpptr(Rrax, Rrsi);
2585 __ jcc(Assembler::equal, hit);
2586 #ifndef PRODUCT
2587 __ lea(Rrcx, ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
2588 __ incrementl(Address(Rrcx, 0));
2589 #endif //PRODUCT
2590 __ movptr(Rrdi, Address(Rrsi,
2591 sizeof(oopDesc) +
2592 Klass::secondary_supers_offset_in_bytes()));
2593 __ movl(Rrcx, Address(Rrdi, arrayOopDesc::length_offset_in_bytes()));
2594 __ addptr(Rrdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
2595 if (UseCompressedOops) {
2596 __ encode_heap_oop(Rrax);
2597 __ repne_scanl();
2598 __ jcc(Assembler::notEqual, cmiss);
2599 __ decode_heap_oop(Rrax);
2600 __ movptr(Address(Rrsi,
2601 sizeof(oopDesc) +
2602 Klass::secondary_super_cache_offset_in_bytes()),
2603 Rrax);
2604 __ jmp(hit);
2605 __ bind(cmiss);
2606 __ decode_heap_oop(Rrax);
2607 __ jmp(miss);
2608 } else {
2609 __ repne_scan();
2610 __ jcc(Assembler::notEqual, miss);
2611 __ movptr(Address(Rrsi,
2612 sizeof(oopDesc) +
2613 Klass::secondary_super_cache_offset_in_bytes()),
2614 Rrax);
2615 }
2616 __ bind(hit);
2617 if ($primary) {
2618 __ xorptr(Rrdi, Rrdi);
2619 }
2620 __ bind(miss);
2621 %}
2623 enc_class Java_To_Interpreter(method meth)
2624 %{
2625 // CALL Java_To_Interpreter
2626 // This is the instruction starting address for relocation info.
2627 cbuf.set_inst_mark();
2628 $$$emit8$primary;
2629 // CALL directly to the runtime
2630 emit_d32_reloc(cbuf,
2631 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2632 runtime_call_Relocation::spec(),
2633 RELOC_DISP32);
2634 %}
2636 enc_class Java_Static_Call(method meth)
2637 %{
2638 // JAVA STATIC CALL
2639 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2640 // determine who we intended to call.
2641 cbuf.set_inst_mark();
2642 $$$emit8$primary;
2644 if (!_method) {
2645 emit_d32_reloc(cbuf,
2646 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2647 runtime_call_Relocation::spec(),
2648 RELOC_DISP32);
2649 } else if (_optimized_virtual) {
2650 emit_d32_reloc(cbuf,
2651 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2652 opt_virtual_call_Relocation::spec(),
2653 RELOC_DISP32);
2654 } else {
2655 emit_d32_reloc(cbuf,
2656 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2657 static_call_Relocation::spec(),
2658 RELOC_DISP32);
2659 }
2660 if (_method) {
2661 // Emit stub for static call
2662 emit_java_to_interp(cbuf);
2663 }
2664 %}
2666 enc_class Java_Dynamic_Call(method meth)
2667 %{
2668 // JAVA DYNAMIC CALL
2669 // !!!!!
2670 // Generate "movq rax, -1", placeholder instruction to load oop-info
2671 // emit_call_dynamic_prologue( cbuf );
2672 cbuf.set_inst_mark();
2674 // movq rax, -1
2675 emit_opcode(cbuf, Assembler::REX_W);
2676 emit_opcode(cbuf, 0xB8 | RAX_enc);
2677 emit_d64_reloc(cbuf,
2678 (int64_t) Universe::non_oop_word(),
2679 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2680 address virtual_call_oop_addr = cbuf.inst_mark();
2681 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2682 // who we intended to call.
2683 cbuf.set_inst_mark();
2684 $$$emit8$primary;
2685 emit_d32_reloc(cbuf,
2686 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2687 virtual_call_Relocation::spec(virtual_call_oop_addr),
2688 RELOC_DISP32);
2689 %}
2691 enc_class Java_Compiled_Call(method meth)
2692 %{
2693 // JAVA COMPILED CALL
2694 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2696 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2697 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2699 // callq *disp(%rax)
2700 cbuf.set_inst_mark();
2701 $$$emit8$primary;
2702 if (disp < 0x80) {
2703 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2704 emit_d8(cbuf, disp); // Displacement
2705 } else {
2706 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2707 emit_d32(cbuf, disp); // Displacement
2708 }
2709 %}
2711 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2712 %{
2713 // SAL, SAR, SHR
2714 int dstenc = $dst$$reg;
2715 if (dstenc >= 8) {
2716 emit_opcode(cbuf, Assembler::REX_B);
2717 dstenc -= 8;
2718 }
2719 $$$emit8$primary;
2720 emit_rm(cbuf, 0x3, $secondary, dstenc);
2721 $$$emit8$shift$$constant;
2722 %}
2724 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2725 %{
2726 // SAL, SAR, SHR
2727 int dstenc = $dst$$reg;
2728 if (dstenc < 8) {
2729 emit_opcode(cbuf, Assembler::REX_W);
2730 } else {
2731 emit_opcode(cbuf, Assembler::REX_WB);
2732 dstenc -= 8;
2733 }
2734 $$$emit8$primary;
2735 emit_rm(cbuf, 0x3, $secondary, dstenc);
2736 $$$emit8$shift$$constant;
2737 %}
2739 enc_class load_immI(rRegI dst, immI src)
2740 %{
2741 int dstenc = $dst$$reg;
2742 if (dstenc >= 8) {
2743 emit_opcode(cbuf, Assembler::REX_B);
2744 dstenc -= 8;
2745 }
2746 emit_opcode(cbuf, 0xB8 | dstenc);
2747 $$$emit32$src$$constant;
2748 %}
2750 enc_class load_immL(rRegL dst, immL src)
2751 %{
2752 int dstenc = $dst$$reg;
2753 if (dstenc < 8) {
2754 emit_opcode(cbuf, Assembler::REX_W);
2755 } else {
2756 emit_opcode(cbuf, Assembler::REX_WB);
2757 dstenc -= 8;
2758 }
2759 emit_opcode(cbuf, 0xB8 | dstenc);
2760 emit_d64(cbuf, $src$$constant);
2761 %}
2763 enc_class load_immUL32(rRegL dst, immUL32 src)
2764 %{
2765 // same as load_immI, but this time we care about zeroes in the high word
2766 int dstenc = $dst$$reg;
2767 if (dstenc >= 8) {
2768 emit_opcode(cbuf, Assembler::REX_B);
2769 dstenc -= 8;
2770 }
2771 emit_opcode(cbuf, 0xB8 | dstenc);
2772 $$$emit32$src$$constant;
2773 %}
2775 enc_class load_immL32(rRegL dst, immL32 src)
2776 %{
2777 int dstenc = $dst$$reg;
2778 if (dstenc < 8) {
2779 emit_opcode(cbuf, Assembler::REX_W);
2780 } else {
2781 emit_opcode(cbuf, Assembler::REX_WB);
2782 dstenc -= 8;
2783 }
2784 emit_opcode(cbuf, 0xC7);
2785 emit_rm(cbuf, 0x03, 0x00, dstenc);
2786 $$$emit32$src$$constant;
2787 %}
2789 enc_class load_immP31(rRegP dst, immP32 src)
2790 %{
2791 // same as load_immI, but this time we care about zeroes in the high word
2792 int dstenc = $dst$$reg;
2793 if (dstenc >= 8) {
2794 emit_opcode(cbuf, Assembler::REX_B);
2795 dstenc -= 8;
2796 }
2797 emit_opcode(cbuf, 0xB8 | dstenc);
2798 $$$emit32$src$$constant;
2799 %}
2801 enc_class load_immP(rRegP dst, immP src)
2802 %{
2803 int dstenc = $dst$$reg;
2804 if (dstenc < 8) {
2805 emit_opcode(cbuf, Assembler::REX_W);
2806 } else {
2807 emit_opcode(cbuf, Assembler::REX_WB);
2808 dstenc -= 8;
2809 }
2810 emit_opcode(cbuf, 0xB8 | dstenc);
2811 // This next line should be generated from ADLC
2812 if ($src->constant_is_oop()) {
2813 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2814 } else {
2815 emit_d64(cbuf, $src$$constant);
2816 }
2817 %}
2819 enc_class load_immF(regF dst, immF con)
2820 %{
2821 // XXX reg_mem doesn't support RIP-relative addressing yet
2822 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2823 emit_float_constant(cbuf, $con$$constant);
2824 %}
2826 enc_class load_immD(regD dst, immD con)
2827 %{
2828 // XXX reg_mem doesn't support RIP-relative addressing yet
2829 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2830 emit_double_constant(cbuf, $con$$constant);
2831 %}
2833 enc_class load_conF (regF dst, immF con) %{ // Load float constant
2834 emit_opcode(cbuf, 0xF3);
2835 if ($dst$$reg >= 8) {
2836 emit_opcode(cbuf, Assembler::REX_R);
2837 }
2838 emit_opcode(cbuf, 0x0F);
2839 emit_opcode(cbuf, 0x10);
2840 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2841 emit_float_constant(cbuf, $con$$constant);
2842 %}
2844 enc_class load_conD (regD dst, immD con) %{ // Load double constant
2845 // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
2846 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
2847 if ($dst$$reg >= 8) {
2848 emit_opcode(cbuf, Assembler::REX_R);
2849 }
2850 emit_opcode(cbuf, 0x0F);
2851 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
2852 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2853 emit_double_constant(cbuf, $con$$constant);
2854 %}
2856 // Encode a reg-reg copy. If it is useless, then empty encoding.
2857 enc_class enc_copy(rRegI dst, rRegI src)
2858 %{
2859 encode_copy(cbuf, $dst$$reg, $src$$reg);
2860 %}
2862 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2863 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2864 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2865 %}
2867 enc_class enc_copy_always(rRegI dst, rRegI src)
2868 %{
2869 int srcenc = $src$$reg;
2870 int dstenc = $dst$$reg;
2872 if (dstenc < 8) {
2873 if (srcenc >= 8) {
2874 emit_opcode(cbuf, Assembler::REX_B);
2875 srcenc -= 8;
2876 }
2877 } else {
2878 if (srcenc < 8) {
2879 emit_opcode(cbuf, Assembler::REX_R);
2880 } else {
2881 emit_opcode(cbuf, Assembler::REX_RB);
2882 srcenc -= 8;
2883 }
2884 dstenc -= 8;
2885 }
2887 emit_opcode(cbuf, 0x8B);
2888 emit_rm(cbuf, 0x3, dstenc, srcenc);
2889 %}
2891 enc_class enc_copy_wide(rRegL dst, rRegL src)
2892 %{
2893 int srcenc = $src$$reg;
2894 int dstenc = $dst$$reg;
2896 if (dstenc != srcenc) {
2897 if (dstenc < 8) {
2898 if (srcenc < 8) {
2899 emit_opcode(cbuf, Assembler::REX_W);
2900 } else {
2901 emit_opcode(cbuf, Assembler::REX_WB);
2902 srcenc -= 8;
2903 }
2904 } else {
2905 if (srcenc < 8) {
2906 emit_opcode(cbuf, Assembler::REX_WR);
2907 } else {
2908 emit_opcode(cbuf, Assembler::REX_WRB);
2909 srcenc -= 8;
2910 }
2911 dstenc -= 8;
2912 }
2913 emit_opcode(cbuf, 0x8B);
2914 emit_rm(cbuf, 0x3, dstenc, srcenc);
2915 }
2916 %}
2918 enc_class Con32(immI src)
2919 %{
2920 // Output immediate
2921 $$$emit32$src$$constant;
2922 %}
2924 enc_class Con64(immL src)
2925 %{
2926 // Output immediate
2927 emit_d64($src$$constant);
2928 %}
2930 enc_class Con32F_as_bits(immF src)
2931 %{
2932 // Output Float immediate bits
2933 jfloat jf = $src$$constant;
2934 jint jf_as_bits = jint_cast(jf);
2935 emit_d32(cbuf, jf_as_bits);
2936 %}
2938 enc_class Con16(immI src)
2939 %{
2940 // Output immediate
2941 $$$emit16$src$$constant;
2942 %}
2944 // How is this different from Con32??? XXX
2945 enc_class Con_d32(immI src)
2946 %{
2947 emit_d32(cbuf,$src$$constant);
2948 %}
2950 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2951 // Output immediate memory reference
2952 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2953 emit_d32(cbuf, 0x00);
2954 %}
2956 enc_class jump_enc(rRegL switch_val, rRegI dest) %{
2957 MacroAssembler masm(&cbuf);
2959 Register switch_reg = as_Register($switch_val$$reg);
2960 Register dest_reg = as_Register($dest$$reg);
2961 address table_base = masm.address_table_constant(_index2label);
2963 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2964 // to do that and the compiler is using that register as one it can allocate.
2965 // So we build it all by hand.
2966 // Address index(noreg, switch_reg, Address::times_1);
2967 // ArrayAddress dispatch(table, index);
2969 Address dispatch(dest_reg, switch_reg, Address::times_1);
2971 masm.lea(dest_reg, InternalAddress(table_base));
2972 masm.jmp(dispatch);
2973 %}
2975 enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
2976 MacroAssembler masm(&cbuf);
2978 Register switch_reg = as_Register($switch_val$$reg);
2979 Register dest_reg = as_Register($dest$$reg);
2980 address table_base = masm.address_table_constant(_index2label);
2982 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2983 // to do that and the compiler is using that register as one it can allocate.
2984 // So we build it all by hand.
2985 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2986 // ArrayAddress dispatch(table, index);
2988 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2990 masm.lea(dest_reg, InternalAddress(table_base));
2991 masm.jmp(dispatch);
2992 %}
2994 enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
2995 MacroAssembler masm(&cbuf);
2997 Register switch_reg = as_Register($switch_val$$reg);
2998 Register dest_reg = as_Register($dest$$reg);
2999 address table_base = masm.address_table_constant(_index2label);
3001 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
3002 // to do that and the compiler is using that register as one it can allocate.
3003 // So we build it all by hand.
3004 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3005 // ArrayAddress dispatch(table, index);
3007 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3008 masm.lea(dest_reg, InternalAddress(table_base));
3009 masm.jmp(dispatch);
3011 %}
3013 enc_class lock_prefix()
3014 %{
3015 if (os::is_MP()) {
3016 emit_opcode(cbuf, 0xF0); // lock
3017 }
3018 %}
3020 enc_class REX_mem(memory mem)
3021 %{
3022 if ($mem$$base >= 8) {
3023 if ($mem$$index < 8) {
3024 emit_opcode(cbuf, Assembler::REX_B);
3025 } else {
3026 emit_opcode(cbuf, Assembler::REX_XB);
3027 }
3028 } else {
3029 if ($mem$$index >= 8) {
3030 emit_opcode(cbuf, Assembler::REX_X);
3031 }
3032 }
3033 %}
3035 enc_class REX_mem_wide(memory mem)
3036 %{
3037 if ($mem$$base >= 8) {
3038 if ($mem$$index < 8) {
3039 emit_opcode(cbuf, Assembler::REX_WB);
3040 } else {
3041 emit_opcode(cbuf, Assembler::REX_WXB);
3042 }
3043 } else {
3044 if ($mem$$index < 8) {
3045 emit_opcode(cbuf, Assembler::REX_W);
3046 } else {
3047 emit_opcode(cbuf, Assembler::REX_WX);
3048 }
3049 }
3050 %}
3052 // for byte regs
3053 enc_class REX_breg(rRegI reg)
3054 %{
3055 if ($reg$$reg >= 4) {
3056 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3057 }
3058 %}
3060 // for byte regs
3061 enc_class REX_reg_breg(rRegI dst, rRegI src)
3062 %{
3063 if ($dst$$reg < 8) {
3064 if ($src$$reg >= 4) {
3065 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3066 }
3067 } else {
3068 if ($src$$reg < 8) {
3069 emit_opcode(cbuf, Assembler::REX_R);
3070 } else {
3071 emit_opcode(cbuf, Assembler::REX_RB);
3072 }
3073 }
3074 %}
3076 // for byte regs
3077 enc_class REX_breg_mem(rRegI reg, memory mem)
3078 %{
3079 if ($reg$$reg < 8) {
3080 if ($mem$$base < 8) {
3081 if ($mem$$index >= 8) {
3082 emit_opcode(cbuf, Assembler::REX_X);
3083 } else if ($reg$$reg >= 4) {
3084 emit_opcode(cbuf, Assembler::REX);
3085 }
3086 } else {
3087 if ($mem$$index < 8) {
3088 emit_opcode(cbuf, Assembler::REX_B);
3089 } else {
3090 emit_opcode(cbuf, Assembler::REX_XB);
3091 }
3092 }
3093 } else {
3094 if ($mem$$base < 8) {
3095 if ($mem$$index < 8) {
3096 emit_opcode(cbuf, Assembler::REX_R);
3097 } else {
3098 emit_opcode(cbuf, Assembler::REX_RX);
3099 }
3100 } else {
3101 if ($mem$$index < 8) {
3102 emit_opcode(cbuf, Assembler::REX_RB);
3103 } else {
3104 emit_opcode(cbuf, Assembler::REX_RXB);
3105 }
3106 }
3107 }
3108 %}
3110 enc_class REX_reg(rRegI reg)
3111 %{
3112 if ($reg$$reg >= 8) {
3113 emit_opcode(cbuf, Assembler::REX_B);
3114 }
3115 %}
3117 enc_class REX_reg_wide(rRegI reg)
3118 %{
3119 if ($reg$$reg < 8) {
3120 emit_opcode(cbuf, Assembler::REX_W);
3121 } else {
3122 emit_opcode(cbuf, Assembler::REX_WB);
3123 }
3124 %}
3126 enc_class REX_reg_reg(rRegI dst, rRegI src)
3127 %{
3128 if ($dst$$reg < 8) {
3129 if ($src$$reg >= 8) {
3130 emit_opcode(cbuf, Assembler::REX_B);
3131 }
3132 } else {
3133 if ($src$$reg < 8) {
3134 emit_opcode(cbuf, Assembler::REX_R);
3135 } else {
3136 emit_opcode(cbuf, Assembler::REX_RB);
3137 }
3138 }
3139 %}
3141 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3142 %{
3143 if ($dst$$reg < 8) {
3144 if ($src$$reg < 8) {
3145 emit_opcode(cbuf, Assembler::REX_W);
3146 } else {
3147 emit_opcode(cbuf, Assembler::REX_WB);
3148 }
3149 } else {
3150 if ($src$$reg < 8) {
3151 emit_opcode(cbuf, Assembler::REX_WR);
3152 } else {
3153 emit_opcode(cbuf, Assembler::REX_WRB);
3154 }
3155 }
3156 %}
3158 enc_class REX_reg_mem(rRegI reg, memory mem)
3159 %{
3160 if ($reg$$reg < 8) {
3161 if ($mem$$base < 8) {
3162 if ($mem$$index >= 8) {
3163 emit_opcode(cbuf, Assembler::REX_X);
3164 }
3165 } else {
3166 if ($mem$$index < 8) {
3167 emit_opcode(cbuf, Assembler::REX_B);
3168 } else {
3169 emit_opcode(cbuf, Assembler::REX_XB);
3170 }
3171 }
3172 } else {
3173 if ($mem$$base < 8) {
3174 if ($mem$$index < 8) {
3175 emit_opcode(cbuf, Assembler::REX_R);
3176 } else {
3177 emit_opcode(cbuf, Assembler::REX_RX);
3178 }
3179 } else {
3180 if ($mem$$index < 8) {
3181 emit_opcode(cbuf, Assembler::REX_RB);
3182 } else {
3183 emit_opcode(cbuf, Assembler::REX_RXB);
3184 }
3185 }
3186 }
3187 %}
3189 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3190 %{
3191 if ($reg$$reg < 8) {
3192 if ($mem$$base < 8) {
3193 if ($mem$$index < 8) {
3194 emit_opcode(cbuf, Assembler::REX_W);
3195 } else {
3196 emit_opcode(cbuf, Assembler::REX_WX);
3197 }
3198 } else {
3199 if ($mem$$index < 8) {
3200 emit_opcode(cbuf, Assembler::REX_WB);
3201 } else {
3202 emit_opcode(cbuf, Assembler::REX_WXB);
3203 }
3204 }
3205 } else {
3206 if ($mem$$base < 8) {
3207 if ($mem$$index < 8) {
3208 emit_opcode(cbuf, Assembler::REX_WR);
3209 } else {
3210 emit_opcode(cbuf, Assembler::REX_WRX);
3211 }
3212 } else {
3213 if ($mem$$index < 8) {
3214 emit_opcode(cbuf, Assembler::REX_WRB);
3215 } else {
3216 emit_opcode(cbuf, Assembler::REX_WRXB);
3217 }
3218 }
3219 }
3220 %}
3222 enc_class reg_mem(rRegI ereg, memory mem)
3223 %{
3224 // High registers handle in encode_RegMem
3225 int reg = $ereg$$reg;
3226 int base = $mem$$base;
3227 int index = $mem$$index;
3228 int scale = $mem$$scale;
3229 int disp = $mem$$disp;
3230 bool disp_is_oop = $mem->disp_is_oop();
3232 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3233 %}
3235 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3236 %{
3237 int rm_byte_opcode = $rm_opcode$$constant;
3239 // High registers handle in encode_RegMem
3240 int base = $mem$$base;
3241 int index = $mem$$index;
3242 int scale = $mem$$scale;
3243 int displace = $mem$$disp;
3245 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3246 // working with static
3247 // globals
3248 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3249 disp_is_oop);
3250 %}
3252 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3253 %{
3254 int reg_encoding = $dst$$reg;
3255 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3256 int index = 0x04; // 0x04 indicates no index
3257 int scale = 0x00; // 0x00 indicates no scale
3258 int displace = $src1$$constant; // 0x00 indicates no displacement
3259 bool disp_is_oop = false;
3260 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3261 disp_is_oop);
3262 %}
3264 enc_class neg_reg(rRegI dst)
3265 %{
3266 int dstenc = $dst$$reg;
3267 if (dstenc >= 8) {
3268 emit_opcode(cbuf, Assembler::REX_B);
3269 dstenc -= 8;
3270 }
3271 // NEG $dst
3272 emit_opcode(cbuf, 0xF7);
3273 emit_rm(cbuf, 0x3, 0x03, dstenc);
3274 %}
3276 enc_class neg_reg_wide(rRegI dst)
3277 %{
3278 int dstenc = $dst$$reg;
3279 if (dstenc < 8) {
3280 emit_opcode(cbuf, Assembler::REX_W);
3281 } else {
3282 emit_opcode(cbuf, Assembler::REX_WB);
3283 dstenc -= 8;
3284 }
3285 // NEG $dst
3286 emit_opcode(cbuf, 0xF7);
3287 emit_rm(cbuf, 0x3, 0x03, dstenc);
3288 %}
3290 enc_class setLT_reg(rRegI dst)
3291 %{
3292 int dstenc = $dst$$reg;
3293 if (dstenc >= 8) {
3294 emit_opcode(cbuf, Assembler::REX_B);
3295 dstenc -= 8;
3296 } else if (dstenc >= 4) {
3297 emit_opcode(cbuf, Assembler::REX);
3298 }
3299 // SETLT $dst
3300 emit_opcode(cbuf, 0x0F);
3301 emit_opcode(cbuf, 0x9C);
3302 emit_rm(cbuf, 0x3, 0x0, dstenc);
3303 %}
3305 enc_class setNZ_reg(rRegI dst)
3306 %{
3307 int dstenc = $dst$$reg;
3308 if (dstenc >= 8) {
3309 emit_opcode(cbuf, Assembler::REX_B);
3310 dstenc -= 8;
3311 } else if (dstenc >= 4) {
3312 emit_opcode(cbuf, Assembler::REX);
3313 }
3314 // SETNZ $dst
3315 emit_opcode(cbuf, 0x0F);
3316 emit_opcode(cbuf, 0x95);
3317 emit_rm(cbuf, 0x3, 0x0, dstenc);
3318 %}
3320 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3321 rcx_RegI tmp)
3322 %{
3323 // cadd_cmpLT
3325 int tmpReg = $tmp$$reg;
3327 int penc = $p$$reg;
3328 int qenc = $q$$reg;
3329 int yenc = $y$$reg;
3331 // subl $p,$q
3332 if (penc < 8) {
3333 if (qenc >= 8) {
3334 emit_opcode(cbuf, Assembler::REX_B);
3335 }
3336 } else {
3337 if (qenc < 8) {
3338 emit_opcode(cbuf, Assembler::REX_R);
3339 } else {
3340 emit_opcode(cbuf, Assembler::REX_RB);
3341 }
3342 }
3343 emit_opcode(cbuf, 0x2B);
3344 emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3346 // sbbl $tmp, $tmp
3347 emit_opcode(cbuf, 0x1B);
3348 emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3350 // andl $tmp, $y
3351 if (yenc >= 8) {
3352 emit_opcode(cbuf, Assembler::REX_B);
3353 }
3354 emit_opcode(cbuf, 0x23);
3355 emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3357 // addl $p,$tmp
3358 if (penc >= 8) {
3359 emit_opcode(cbuf, Assembler::REX_R);
3360 }
3361 emit_opcode(cbuf, 0x03);
3362 emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3363 %}
3365 // Compare the lonogs and set -1, 0, or 1 into dst
3366 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3367 %{
3368 int src1enc = $src1$$reg;
3369 int src2enc = $src2$$reg;
3370 int dstenc = $dst$$reg;
3372 // cmpq $src1, $src2
3373 if (src1enc < 8) {
3374 if (src2enc < 8) {
3375 emit_opcode(cbuf, Assembler::REX_W);
3376 } else {
3377 emit_opcode(cbuf, Assembler::REX_WB);
3378 }
3379 } else {
3380 if (src2enc < 8) {
3381 emit_opcode(cbuf, Assembler::REX_WR);
3382 } else {
3383 emit_opcode(cbuf, Assembler::REX_WRB);
3384 }
3385 }
3386 emit_opcode(cbuf, 0x3B);
3387 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3389 // movl $dst, -1
3390 if (dstenc >= 8) {
3391 emit_opcode(cbuf, Assembler::REX_B);
3392 }
3393 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3394 emit_d32(cbuf, -1);
3396 // jl,s done
3397 emit_opcode(cbuf, 0x7C);
3398 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3400 // setne $dst
3401 if (dstenc >= 4) {
3402 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3403 }
3404 emit_opcode(cbuf, 0x0F);
3405 emit_opcode(cbuf, 0x95);
3406 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3408 // movzbl $dst, $dst
3409 if (dstenc >= 4) {
3410 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3411 }
3412 emit_opcode(cbuf, 0x0F);
3413 emit_opcode(cbuf, 0xB6);
3414 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3415 %}
3417 enc_class Push_ResultXD(regD dst) %{
3418 int dstenc = $dst$$reg;
3420 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3422 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3423 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3424 if (dstenc >= 8) {
3425 emit_opcode(cbuf, Assembler::REX_R);
3426 }
3427 emit_opcode (cbuf, 0x0F );
3428 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3429 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3431 // add rsp,8
3432 emit_opcode(cbuf, Assembler::REX_W);
3433 emit_opcode(cbuf,0x83);
3434 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3435 emit_d8(cbuf,0x08);
3436 %}
3438 enc_class Push_SrcXD(regD src) %{
3439 int srcenc = $src$$reg;
3441 // subq rsp,#8
3442 emit_opcode(cbuf, Assembler::REX_W);
3443 emit_opcode(cbuf, 0x83);
3444 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3445 emit_d8(cbuf, 0x8);
3447 // movsd [rsp],src
3448 emit_opcode(cbuf, 0xF2);
3449 if (srcenc >= 8) {
3450 emit_opcode(cbuf, Assembler::REX_R);
3451 }
3452 emit_opcode(cbuf, 0x0F);
3453 emit_opcode(cbuf, 0x11);
3454 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3456 // fldd [rsp]
3457 emit_opcode(cbuf, 0x66);
3458 emit_opcode(cbuf, 0xDD);
3459 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3460 %}
3463 enc_class movq_ld(regD dst, memory mem) %{
3464 MacroAssembler _masm(&cbuf);
3465 Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3466 __ movq(as_XMMRegister($dst$$reg), madr);
3467 %}
3469 enc_class movq_st(memory mem, regD src) %{
3470 MacroAssembler _masm(&cbuf);
3471 Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3472 __ movq(madr, as_XMMRegister($src$$reg));
3473 %}
3475 enc_class pshufd_8x8(regF dst, regF src) %{
3476 MacroAssembler _masm(&cbuf);
3478 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3479 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3480 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3481 %}
3483 enc_class pshufd_4x16(regF dst, regF src) %{
3484 MacroAssembler _masm(&cbuf);
3486 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3487 %}
3489 enc_class pshufd(regD dst, regD src, int mode) %{
3490 MacroAssembler _masm(&cbuf);
3492 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3493 %}
3495 enc_class pxor(regD dst, regD src) %{
3496 MacroAssembler _masm(&cbuf);
3498 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3499 %}
3501 enc_class mov_i2x(regD dst, rRegI src) %{
3502 MacroAssembler _masm(&cbuf);
3504 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3505 %}
3507 // obj: object to lock
3508 // box: box address (header location) -- killed
3509 // tmp: rax -- killed
3510 // scr: rbx -- killed
3511 //
3512 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3513 // from i486.ad. See that file for comments.
3514 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3515 // use the shorter encoding. (Movl clears the high-order 32-bits).
3518 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3519 %{
3520 Register objReg = as_Register((int)$obj$$reg);
3521 Register boxReg = as_Register((int)$box$$reg);
3522 Register tmpReg = as_Register($tmp$$reg);
3523 Register scrReg = as_Register($scr$$reg);
3524 MacroAssembler masm(&cbuf);
3526 // Verify uniqueness of register assignments -- necessary but not sufficient
3527 assert (objReg != boxReg && objReg != tmpReg &&
3528 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3530 if (_counters != NULL) {
3531 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3532 }
3533 if (EmitSync & 1) {
3534 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3535 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3536 masm.cmpptr(rsp, (int32_t)NULL_WORD) ;
3537 } else
3538 if (EmitSync & 2) {
3539 Label DONE_LABEL;
3540 if (UseBiasedLocking) {
3541 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3542 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3543 }
3544 // QQQ was movl...
3545 masm.movptr(tmpReg, 0x1);
3546 masm.orptr(tmpReg, Address(objReg, 0));
3547 masm.movptr(Address(boxReg, 0), tmpReg);
3548 if (os::is_MP()) {
3549 masm.lock();
3550 }
3551 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3552 masm.jcc(Assembler::equal, DONE_LABEL);
3554 // Recursive locking
3555 masm.subptr(tmpReg, rsp);
3556 masm.andptr(tmpReg, 7 - os::vm_page_size());
3557 masm.movptr(Address(boxReg, 0), tmpReg);
3559 masm.bind(DONE_LABEL);
3560 masm.nop(); // avoid branch to branch
3561 } else {
3562 Label DONE_LABEL, IsInflated, Egress;
3564 masm.movptr(tmpReg, Address(objReg, 0)) ;
3565 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3566 masm.jcc (Assembler::notZero, IsInflated) ;
3568 // it's stack-locked, biased or neutral
3569 // TODO: optimize markword triage order to reduce the number of
3570 // conditional branches in the most common cases.
3571 // Beware -- there's a subtle invariant that fetch of the markword
3572 // at [FETCH], below, will never observe a biased encoding (*101b).
3573 // If this invariant is not held we'll suffer exclusion (safety) failure.
3575 if (UseBiasedLocking) {
3576 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3577 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH]
3578 }
3580 // was q will it destroy high?
3581 masm.orl (tmpReg, 1) ;
3582 masm.movptr(Address(boxReg, 0), tmpReg) ;
3583 if (os::is_MP()) { masm.lock(); }
3584 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3585 if (_counters != NULL) {
3586 masm.cond_inc32(Assembler::equal,
3587 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3588 }
3589 masm.jcc (Assembler::equal, DONE_LABEL);
3591 // Recursive locking
3592 masm.subptr(tmpReg, rsp);
3593 masm.andptr(tmpReg, 7 - os::vm_page_size());
3594 masm.movptr(Address(boxReg, 0), tmpReg);
3595 if (_counters != NULL) {
3596 masm.cond_inc32(Assembler::equal,
3597 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3598 }
3599 masm.jmp (DONE_LABEL) ;
3601 masm.bind (IsInflated) ;
3602 // It's inflated
3604 // TODO: someday avoid the ST-before-CAS penalty by
3605 // relocating (deferring) the following ST.
3606 // We should also think about trying a CAS without having
3607 // fetched _owner. If the CAS is successful we may
3608 // avoid an RTO->RTS upgrade on the $line.
3609 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3610 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3612 masm.mov (boxReg, tmpReg) ;
3613 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3614 masm.testptr(tmpReg, tmpReg) ;
3615 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3617 // It's inflated and appears unlocked
3618 if (os::is_MP()) { masm.lock(); }
3619 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3620 // Intentional fall-through into DONE_LABEL ...
3622 masm.bind (DONE_LABEL) ;
3623 masm.nop () ; // avoid jmp to jmp
3624 }
3625 %}
3627 // obj: object to unlock
3628 // box: box address (displaced header location), killed
3629 // RBX: killed tmp; cannot be obj nor box
3630 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3631 %{
3633 Register objReg = as_Register($obj$$reg);
3634 Register boxReg = as_Register($box$$reg);
3635 Register tmpReg = as_Register($tmp$$reg);
3636 MacroAssembler masm(&cbuf);
3638 if (EmitSync & 4) {
3639 masm.cmpptr(rsp, 0) ;
3640 } else
3641 if (EmitSync & 8) {
3642 Label DONE_LABEL;
3643 if (UseBiasedLocking) {
3644 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3645 }
3647 // Check whether the displaced header is 0
3648 //(=> recursive unlock)
3649 masm.movptr(tmpReg, Address(boxReg, 0));
3650 masm.testptr(tmpReg, tmpReg);
3651 masm.jcc(Assembler::zero, DONE_LABEL);
3653 // If not recursive lock, reset the header to displaced header
3654 if (os::is_MP()) {
3655 masm.lock();
3656 }
3657 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3658 masm.bind(DONE_LABEL);
3659 masm.nop(); // avoid branch to branch
3660 } else {
3661 Label DONE_LABEL, Stacked, CheckSucc ;
3663 if (UseBiasedLocking) {
3664 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3665 }
3667 masm.movptr(tmpReg, Address(objReg, 0)) ;
3668 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ;
3669 masm.jcc (Assembler::zero, DONE_LABEL) ;
3670 masm.testl (tmpReg, 0x02) ;
3671 masm.jcc (Assembler::zero, Stacked) ;
3673 // It's inflated
3674 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3675 masm.xorptr(boxReg, r15_thread) ;
3676 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3677 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3678 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3679 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3680 masm.jcc (Assembler::notZero, CheckSucc) ;
3681 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3682 masm.jmp (DONE_LABEL) ;
3684 if ((EmitSync & 65536) == 0) {
3685 Label LSuccess, LGoSlowPath ;
3686 masm.bind (CheckSucc) ;
3687 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3688 masm.jcc (Assembler::zero, LGoSlowPath) ;
3690 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3691 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3692 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3693 // are all faster when the write buffer is populated.
3694 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3695 if (os::is_MP()) {
3696 masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3697 }
3698 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3699 masm.jcc (Assembler::notZero, LSuccess) ;
3701 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX
3702 if (os::is_MP()) { masm.lock(); }
3703 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3704 masm.jcc (Assembler::notEqual, LSuccess) ;
3705 // Intentional fall-through into slow-path
3707 masm.bind (LGoSlowPath) ;
3708 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3709 masm.jmp (DONE_LABEL) ;
3711 masm.bind (LSuccess) ;
3712 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3713 masm.jmp (DONE_LABEL) ;
3714 }
3716 masm.bind (Stacked) ;
3717 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch
3718 if (os::is_MP()) { masm.lock(); }
3719 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3721 if (EmitSync & 65536) {
3722 masm.bind (CheckSucc) ;
3723 }
3724 masm.bind(DONE_LABEL);
3725 if (EmitSync & 32768) {
3726 masm.nop(); // avoid branch to branch
3727 }
3728 }
3729 %}
3731 enc_class enc_String_Compare()
3732 %{
3733 Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
3734 POP_LABEL, DONE_LABEL, CONT_LABEL,
3735 WHILE_HEAD_LABEL;
3736 MacroAssembler masm(&cbuf);
3738 // Get the first character position in both strings
3739 // [8] char array, [12] offset, [16] count
3740 int value_offset = java_lang_String::value_offset_in_bytes();
3741 int offset_offset = java_lang_String::offset_offset_in_bytes();
3742 int count_offset = java_lang_String::count_offset_in_bytes();
3743 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3745 masm.load_heap_oop(rax, Address(rsi, value_offset));
3746 masm.movl(rcx, Address(rsi, offset_offset));
3747 masm.lea(rax, Address(rax, rcx, Address::times_2, base_offset));
3748 masm.load_heap_oop(rbx, Address(rdi, value_offset));
3749 masm.movl(rcx, Address(rdi, offset_offset));
3750 masm.lea(rbx, Address(rbx, rcx, Address::times_2, base_offset));
3752 // Compute the minimum of the string lengths(rsi) and the
3753 // difference of the string lengths (stack)
3755 masm.movl(rdi, Address(rdi, count_offset));
3756 masm.movl(rsi, Address(rsi, count_offset));
3757 masm.movl(rcx, rdi);
3758 masm.subl(rdi, rsi);
3759 masm.push(rdi);
3760 masm.cmov(Assembler::lessEqual, rsi, rcx);
3762 // Is the minimum length zero?
3763 masm.bind(RCX_GOOD_LABEL);
3764 masm.testl(rsi, rsi);
3765 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3767 // Load first characters
3768 masm.load_unsigned_word(rcx, Address(rbx, 0));
3769 masm.load_unsigned_word(rdi, Address(rax, 0));
3771 // Compare first characters
3772 masm.subl(rcx, rdi);
3773 masm.jcc(Assembler::notZero, POP_LABEL);
3774 masm.decrementl(rsi);
3775 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
3777 {
3778 // Check after comparing first character to see if strings are equivalent
3779 Label LSkip2;
3780 // Check if the strings start at same location
3781 masm.cmpptr(rbx, rax);
3782 masm.jcc(Assembler::notEqual, LSkip2);
3784 // Check if the length difference is zero (from stack)
3785 masm.cmpl(Address(rsp, 0), 0x0);
3786 masm.jcc(Assembler::equal, LENGTH_DIFF_LABEL);
3788 // Strings might not be equivalent
3789 masm.bind(LSkip2);
3790 }
3792 // Shift RAX and RBX to the end of the arrays, negate min
3793 masm.lea(rax, Address(rax, rsi, Address::times_2, 2));
3794 masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2));
3795 masm.negptr(rsi);
3797 // Compare the rest of the characters
3798 masm.bind(WHILE_HEAD_LABEL);
3799 masm.load_unsigned_word(rcx, Address(rbx, rsi, Address::times_2, 0));
3800 masm.load_unsigned_word(rdi, Address(rax, rsi, Address::times_2, 0));
3801 masm.subl(rcx, rdi);
3802 masm.jcc(Assembler::notZero, POP_LABEL);
3803 masm.increment(rsi);
3804 masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
3806 // Strings are equal up to min length. Return the length difference.
3807 masm.bind(LENGTH_DIFF_LABEL);
3808 masm.pop(rcx);
3809 masm.jmp(DONE_LABEL);
3811 // Discard the stored length difference
3812 masm.bind(POP_LABEL);
3813 masm.addptr(rsp, 8);
3815 // That's it
3816 masm.bind(DONE_LABEL);
3817 %}
3819 enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, rbx_RegI tmp2, rcx_RegI result) %{
3820 Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP;
3821 MacroAssembler masm(&cbuf);
3823 Register ary1Reg = as_Register($ary1$$reg);
3824 Register ary2Reg = as_Register($ary2$$reg);
3825 Register tmp1Reg = as_Register($tmp1$$reg);
3826 Register tmp2Reg = as_Register($tmp2$$reg);
3827 Register resultReg = as_Register($result$$reg);
3829 int length_offset = arrayOopDesc::length_offset_in_bytes();
3830 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3832 // Check the input args
3833 masm.cmpq(ary1Reg, ary2Reg);
3834 masm.jcc(Assembler::equal, TRUE_LABEL);
3835 masm.testq(ary1Reg, ary1Reg);
3836 masm.jcc(Assembler::zero, FALSE_LABEL);
3837 masm.testq(ary2Reg, ary2Reg);
3838 masm.jcc(Assembler::zero, FALSE_LABEL);
3840 // Check the lengths
3841 masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
3842 masm.movl(resultReg, Address(ary2Reg, length_offset));
3843 masm.cmpl(tmp2Reg, resultReg);
3844 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3845 masm.testl(resultReg, resultReg);
3846 masm.jcc(Assembler::zero, TRUE_LABEL);
3848 // Get the number of 4 byte vectors to compare
3849 masm.shrl(resultReg, 1);
3851 // Check for odd-length arrays
3852 masm.andl(tmp2Reg, 1);
3853 masm.testl(tmp2Reg, tmp2Reg);
3854 masm.jcc(Assembler::zero, COMPARE_LOOP_HDR);
3856 // Compare 2-byte "tail" at end of arrays
3857 masm.load_unsigned_word(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
3858 masm.load_unsigned_word(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
3859 masm.cmpl(tmp1Reg, tmp2Reg);
3860 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3861 masm.testl(resultReg, resultReg);
3862 masm.jcc(Assembler::zero, TRUE_LABEL);
3864 // Setup compare loop
3865 masm.bind(COMPARE_LOOP_HDR);
3866 // Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays
3867 masm.leaq(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
3868 masm.leaq(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
3869 masm.negq(resultReg);
3871 // 4-byte-wide compare loop
3872 masm.bind(COMPARE_LOOP);
3873 masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0));
3874 masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0));
3875 masm.cmpl(ary1Reg, ary2Reg);
3876 masm.jcc(Assembler::notEqual, FALSE_LABEL);
3877 masm.incrementq(resultReg);
3878 masm.jcc(Assembler::notZero, COMPARE_LOOP);
3880 masm.bind(TRUE_LABEL);
3881 masm.movl(resultReg, 1); // return true
3882 masm.jmp(DONE_LABEL);
3884 masm.bind(FALSE_LABEL);
3885 masm.xorl(resultReg, resultReg); // return false
3887 // That's it
3888 masm.bind(DONE_LABEL);
3889 %}
3891 enc_class enc_rethrow()
3892 %{
3893 cbuf.set_inst_mark();
3894 emit_opcode(cbuf, 0xE9); // jmp entry
3895 emit_d32_reloc(cbuf,
3896 (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
3897 runtime_call_Relocation::spec(),
3898 RELOC_DISP32);
3899 %}
3901 enc_class absF_encoding(regF dst)
3902 %{
3903 int dstenc = $dst$$reg;
3904 address signmask_address = (address) StubRoutines::x86::float_sign_mask();
3906 cbuf.set_inst_mark();
3907 if (dstenc >= 8) {
3908 emit_opcode(cbuf, Assembler::REX_R);
3909 dstenc -= 8;
3910 }
3911 // XXX reg_mem doesn't support RIP-relative addressing yet
3912 emit_opcode(cbuf, 0x0F);
3913 emit_opcode(cbuf, 0x54);
3914 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3915 emit_d32_reloc(cbuf, signmask_address);
3916 %}
3918 enc_class absD_encoding(regD dst)
3919 %{
3920 int dstenc = $dst$$reg;
3921 address signmask_address = (address) StubRoutines::x86::double_sign_mask();
3923 cbuf.set_inst_mark();
3924 emit_opcode(cbuf, 0x66);
3925 if (dstenc >= 8) {
3926 emit_opcode(cbuf, Assembler::REX_R);
3927 dstenc -= 8;
3928 }
3929 // XXX reg_mem doesn't support RIP-relative addressing yet
3930 emit_opcode(cbuf, 0x0F);
3931 emit_opcode(cbuf, 0x54);
3932 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3933 emit_d32_reloc(cbuf, signmask_address);
3934 %}
3936 enc_class negF_encoding(regF dst)
3937 %{
3938 int dstenc = $dst$$reg;
3939 address signflip_address = (address) StubRoutines::x86::float_sign_flip();
3941 cbuf.set_inst_mark();
3942 if (dstenc >= 8) {
3943 emit_opcode(cbuf, Assembler::REX_R);
3944 dstenc -= 8;
3945 }
3946 // XXX reg_mem doesn't support RIP-relative addressing yet
3947 emit_opcode(cbuf, 0x0F);
3948 emit_opcode(cbuf, 0x57);
3949 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3950 emit_d32_reloc(cbuf, signflip_address);
3951 %}
3953 enc_class negD_encoding(regD dst)
3954 %{
3955 int dstenc = $dst$$reg;
3956 address signflip_address = (address) StubRoutines::x86::double_sign_flip();
3958 cbuf.set_inst_mark();
3959 emit_opcode(cbuf, 0x66);
3960 if (dstenc >= 8) {
3961 emit_opcode(cbuf, Assembler::REX_R);
3962 dstenc -= 8;
3963 }
3964 // XXX reg_mem doesn't support RIP-relative addressing yet
3965 emit_opcode(cbuf, 0x0F);
3966 emit_opcode(cbuf, 0x57);
3967 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3968 emit_d32_reloc(cbuf, signflip_address);
3969 %}
3971 enc_class f2i_fixup(rRegI dst, regF src)
3972 %{
3973 int dstenc = $dst$$reg;
3974 int srcenc = $src$$reg;
3976 // cmpl $dst, #0x80000000
3977 if (dstenc >= 8) {
3978 emit_opcode(cbuf, Assembler::REX_B);
3979 }
3980 emit_opcode(cbuf, 0x81);
3981 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3982 emit_d32(cbuf, 0x80000000);
3984 // jne,s done
3985 emit_opcode(cbuf, 0x75);
3986 if (srcenc < 8 && dstenc < 8) {
3987 emit_d8(cbuf, 0xF);
3988 } else if (srcenc >= 8 && dstenc >= 8) {
3989 emit_d8(cbuf, 0x11);
3990 } else {
3991 emit_d8(cbuf, 0x10);
3992 }
3994 // subq rsp, #8
3995 emit_opcode(cbuf, Assembler::REX_W);
3996 emit_opcode(cbuf, 0x83);
3997 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3998 emit_d8(cbuf, 8);
4000 // movss [rsp], $src
4001 emit_opcode(cbuf, 0xF3);
4002 if (srcenc >= 8) {
4003 emit_opcode(cbuf, Assembler::REX_R);
4004 }
4005 emit_opcode(cbuf, 0x0F);
4006 emit_opcode(cbuf, 0x11);
4007 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4009 // call f2i_fixup
4010 cbuf.set_inst_mark();
4011 emit_opcode(cbuf, 0xE8);
4012 emit_d32_reloc(cbuf,
4013 (int)
4014 (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4),
4015 runtime_call_Relocation::spec(),
4016 RELOC_DISP32);
4018 // popq $dst
4019 if (dstenc >= 8) {
4020 emit_opcode(cbuf, Assembler::REX_B);
4021 }
4022 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4024 // done:
4025 %}
4027 enc_class f2l_fixup(rRegL dst, regF src)
4028 %{
4029 int dstenc = $dst$$reg;
4030 int srcenc = $src$$reg;
4031 address const_address = (address) StubRoutines::x86::double_sign_flip();
4033 // cmpq $dst, [0x8000000000000000]
4034 cbuf.set_inst_mark();
4035 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4036 emit_opcode(cbuf, 0x39);
4037 // XXX reg_mem doesn't support RIP-relative addressing yet
4038 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4039 emit_d32_reloc(cbuf, const_address);
4042 // jne,s done
4043 emit_opcode(cbuf, 0x75);
4044 if (srcenc < 8 && dstenc < 8) {
4045 emit_d8(cbuf, 0xF);
4046 } else if (srcenc >= 8 && dstenc >= 8) {
4047 emit_d8(cbuf, 0x11);
4048 } else {
4049 emit_d8(cbuf, 0x10);
4050 }
4052 // subq rsp, #8
4053 emit_opcode(cbuf, Assembler::REX_W);
4054 emit_opcode(cbuf, 0x83);
4055 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4056 emit_d8(cbuf, 8);
4058 // movss [rsp], $src
4059 emit_opcode(cbuf, 0xF3);
4060 if (srcenc >= 8) {
4061 emit_opcode(cbuf, Assembler::REX_R);
4062 }
4063 emit_opcode(cbuf, 0x0F);
4064 emit_opcode(cbuf, 0x11);
4065 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4067 // call f2l_fixup
4068 cbuf.set_inst_mark();
4069 emit_opcode(cbuf, 0xE8);
4070 emit_d32_reloc(cbuf,
4071 (int)
4072 (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4),
4073 runtime_call_Relocation::spec(),
4074 RELOC_DISP32);
4076 // popq $dst
4077 if (dstenc >= 8) {
4078 emit_opcode(cbuf, Assembler::REX_B);
4079 }
4080 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4082 // done:
4083 %}
4085 enc_class d2i_fixup(rRegI dst, regD src)
4086 %{
4087 int dstenc = $dst$$reg;
4088 int srcenc = $src$$reg;
4090 // cmpl $dst, #0x80000000
4091 if (dstenc >= 8) {
4092 emit_opcode(cbuf, Assembler::REX_B);
4093 }
4094 emit_opcode(cbuf, 0x81);
4095 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
4096 emit_d32(cbuf, 0x80000000);
4098 // jne,s done
4099 emit_opcode(cbuf, 0x75);
4100 if (srcenc < 8 && dstenc < 8) {
4101 emit_d8(cbuf, 0xF);
4102 } else if (srcenc >= 8 && dstenc >= 8) {
4103 emit_d8(cbuf, 0x11);
4104 } else {
4105 emit_d8(cbuf, 0x10);
4106 }
4108 // subq rsp, #8
4109 emit_opcode(cbuf, Assembler::REX_W);
4110 emit_opcode(cbuf, 0x83);
4111 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4112 emit_d8(cbuf, 8);
4114 // movsd [rsp], $src
4115 emit_opcode(cbuf, 0xF2);
4116 if (srcenc >= 8) {
4117 emit_opcode(cbuf, Assembler::REX_R);
4118 }
4119 emit_opcode(cbuf, 0x0F);
4120 emit_opcode(cbuf, 0x11);
4121 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4123 // call d2i_fixup
4124 cbuf.set_inst_mark();
4125 emit_opcode(cbuf, 0xE8);
4126 emit_d32_reloc(cbuf,
4127 (int)
4128 (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4),
4129 runtime_call_Relocation::spec(),
4130 RELOC_DISP32);
4132 // popq $dst
4133 if (dstenc >= 8) {
4134 emit_opcode(cbuf, Assembler::REX_B);
4135 }
4136 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4138 // done:
4139 %}
4141 enc_class d2l_fixup(rRegL dst, regD src)
4142 %{
4143 int dstenc = $dst$$reg;
4144 int srcenc = $src$$reg;
4145 address const_address = (address) StubRoutines::x86::double_sign_flip();
4147 // cmpq $dst, [0x8000000000000000]
4148 cbuf.set_inst_mark();
4149 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4150 emit_opcode(cbuf, 0x39);
4151 // XXX reg_mem doesn't support RIP-relative addressing yet
4152 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4153 emit_d32_reloc(cbuf, const_address);
4156 // jne,s done
4157 emit_opcode(cbuf, 0x75);
4158 if (srcenc < 8 && dstenc < 8) {
4159 emit_d8(cbuf, 0xF);
4160 } else if (srcenc >= 8 && dstenc >= 8) {
4161 emit_d8(cbuf, 0x11);
4162 } else {
4163 emit_d8(cbuf, 0x10);
4164 }
4166 // subq rsp, #8
4167 emit_opcode(cbuf, Assembler::REX_W);
4168 emit_opcode(cbuf, 0x83);
4169 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4170 emit_d8(cbuf, 8);
4172 // movsd [rsp], $src
4173 emit_opcode(cbuf, 0xF2);
4174 if (srcenc >= 8) {
4175 emit_opcode(cbuf, Assembler::REX_R);
4176 }
4177 emit_opcode(cbuf, 0x0F);
4178 emit_opcode(cbuf, 0x11);
4179 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4181 // call d2l_fixup
4182 cbuf.set_inst_mark();
4183 emit_opcode(cbuf, 0xE8);
4184 emit_d32_reloc(cbuf,
4185 (int)
4186 (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4),
4187 runtime_call_Relocation::spec(),
4188 RELOC_DISP32);
4190 // popq $dst
4191 if (dstenc >= 8) {
4192 emit_opcode(cbuf, Assembler::REX_B);
4193 }
4194 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4196 // done:
4197 %}
4199 enc_class enc_membar_acquire
4200 %{
4201 // [jk] not needed currently, if you enable this and it really
4202 // emits code don't forget to the remove the "size(0)" line in
4203 // membar_acquire()
4204 // MacroAssembler masm(&cbuf);
4205 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4206 // Assembler::LoadLoad));
4207 %}
4209 enc_class enc_membar_release
4210 %{
4211 // [jk] not needed currently, if you enable this and it really
4212 // emits code don't forget to the remove the "size(0)" line in
4213 // membar_release()
4214 // MacroAssembler masm(&cbuf);
4215 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4216 // Assembler::StoreStore));
4217 %}
4219 enc_class enc_membar_volatile
4220 %{
4221 MacroAssembler masm(&cbuf);
4222 masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
4223 Assembler::StoreStore));
4224 %}
4226 // Safepoint Poll. This polls the safepoint page, and causes an
4227 // exception if it is not readable. Unfortunately, it kills
4228 // RFLAGS in the process.
4229 enc_class enc_safepoint_poll
4230 %{
4231 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
4232 // XXX reg_mem doesn't support RIP-relative addressing yet
4233 cbuf.set_inst_mark();
4234 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
4235 emit_opcode(cbuf, 0x85); // testl
4236 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
4237 // cbuf.inst_mark() is beginning of instruction
4238 emit_d32_reloc(cbuf, os::get_polling_page());
4239 // relocInfo::poll_type,
4240 %}
4241 %}
4245 //----------FRAME--------------------------------------------------------------
4246 // Definition of frame structure and management information.
4247 //
4248 // S T A C K L A Y O U T Allocators stack-slot number
4249 // | (to get allocators register number
4250 // G Owned by | | v add OptoReg::stack0())
4251 // r CALLER | |
4252 // o | +--------+ pad to even-align allocators stack-slot
4253 // w V | pad0 | numbers; owned by CALLER
4254 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
4255 // h ^ | in | 5
4256 // | | args | 4 Holes in incoming args owned by SELF
4257 // | | | | 3
4258 // | | +--------+
4259 // V | | old out| Empty on Intel, window on Sparc
4260 // | old |preserve| Must be even aligned.
4261 // | SP-+--------+----> Matcher::_old_SP, even aligned
4262 // | | in | 3 area for Intel ret address
4263 // Owned by |preserve| Empty on Sparc.
4264 // SELF +--------+
4265 // | | pad2 | 2 pad to align old SP
4266 // | +--------+ 1
4267 // | | locks | 0
4268 // | +--------+----> OptoReg::stack0(), even aligned
4269 // | | pad1 | 11 pad to align new SP
4270 // | +--------+
4271 // | | | 10
4272 // | | spills | 9 spills
4273 // V | | 8 (pad0 slot for callee)
4274 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
4275 // ^ | out | 7
4276 // | | args | 6 Holes in outgoing args owned by CALLEE
4277 // Owned by +--------+
4278 // CALLEE | new out| 6 Empty on Intel, window on Sparc
4279 // | new |preserve| Must be even-aligned.
4280 // | SP-+--------+----> Matcher::_new_SP, even aligned
4281 // | | |
4282 //
4283 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
4284 // known from SELF's arguments and the Java calling convention.
4285 // Region 6-7 is determined per call site.
4286 // Note 2: If the calling convention leaves holes in the incoming argument
4287 // area, those holes are owned by SELF. Holes in the outgoing area
4288 // are owned by the CALLEE. Holes should not be nessecary in the
4289 // incoming area, as the Java calling convention is completely under
4290 // the control of the AD file. Doubles can be sorted and packed to
4291 // avoid holes. Holes in the outgoing arguments may be nessecary for
4292 // varargs C calling conventions.
4293 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
4294 // even aligned with pad0 as needed.
4295 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
4296 // region 6-11 is even aligned; it may be padded out more so that
4297 // the region from SP to FP meets the minimum stack alignment.
4298 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
4299 // alignment. Region 11, pad1, may be dynamically extended so that
4300 // SP meets the minimum alignment.
4302 frame
4303 %{
4304 // What direction does stack grow in (assumed to be same for C & Java)
4305 stack_direction(TOWARDS_LOW);
4307 // These three registers define part of the calling convention
4308 // between compiled code and the interpreter.
4309 inline_cache_reg(RAX); // Inline Cache Register
4310 interpreter_method_oop_reg(RBX); // Method Oop Register when
4311 // calling interpreter
4313 // Optional: name the operand used by cisc-spilling to access
4314 // [stack_pointer + offset]
4315 cisc_spilling_operand_name(indOffset32);
4317 // Number of stack slots consumed by locking an object
4318 sync_stack_slots(2);
4320 // Compiled code's Frame Pointer
4321 frame_pointer(RSP);
4323 // Interpreter stores its frame pointer in a register which is
4324 // stored to the stack by I2CAdaptors.
4325 // I2CAdaptors convert from interpreted java to compiled java.
4326 interpreter_frame_pointer(RBP);
4328 // Stack alignment requirement
4329 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
4331 // Number of stack slots between incoming argument block and the start of
4332 // a new frame. The PROLOG must add this many slots to the stack. The
4333 // EPILOG must remove this many slots. amd64 needs two slots for
4334 // return address.
4335 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
4337 // Number of outgoing stack slots killed above the out_preserve_stack_slots
4338 // for calls to C. Supports the var-args backing area for register parms.
4339 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
4341 // The after-PROLOG location of the return address. Location of
4342 // return address specifies a type (REG or STACK) and a number
4343 // representing the register number (i.e. - use a register name) or
4344 // stack slot.
4345 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4346 // Otherwise, it is above the locks and verification slot and alignment word
4347 return_addr(STACK - 2 +
4348 round_to(2 + 2 * VerifyStackAtCalls +
4349 Compile::current()->fixed_slots(),
4350 WordsPerLong * 2));
4352 // Body of function which returns an integer array locating
4353 // arguments either in registers or in stack slots. Passed an array
4354 // of ideal registers called "sig" and a "length" count. Stack-slot
4355 // offsets are based on outgoing arguments, i.e. a CALLER setting up
4356 // arguments for a CALLEE. Incoming stack arguments are
4357 // automatically biased by the preserve_stack_slots field above.
4359 calling_convention
4360 %{
4361 // No difference between ingoing/outgoing just pass false
4362 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4363 %}
4365 c_calling_convention
4366 %{
4367 // This is obviously always outgoing
4368 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4369 %}
4371 // Location of compiled Java return values. Same as C for now.
4372 return_value
4373 %{
4374 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4375 "only return normal values");
4377 static const int lo[Op_RegL + 1] = {
4378 0,
4379 0,
4380 RAX_num, // Op_RegN
4381 RAX_num, // Op_RegI
4382 RAX_num, // Op_RegP
4383 XMM0_num, // Op_RegF
4384 XMM0_num, // Op_RegD
4385 RAX_num // Op_RegL
4386 };
4387 static const int hi[Op_RegL + 1] = {
4388 0,
4389 0,
4390 OptoReg::Bad, // Op_RegN
4391 OptoReg::Bad, // Op_RegI
4392 RAX_H_num, // Op_RegP
4393 OptoReg::Bad, // Op_RegF
4394 XMM0_H_num, // Op_RegD
4395 RAX_H_num // Op_RegL
4396 };
4397 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4398 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4399 %}
4400 %}
4402 //----------ATTRIBUTES---------------------------------------------------------
4403 //----------Operand Attributes-------------------------------------------------
4404 op_attrib op_cost(0); // Required cost attribute
4406 //----------Instruction Attributes---------------------------------------------
4407 ins_attrib ins_cost(100); // Required cost attribute
4408 ins_attrib ins_size(8); // Required size attribute (in bits)
4409 ins_attrib ins_pc_relative(0); // Required PC Relative flag
4410 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4411 // a non-matching short branch variant
4412 // of some long branch?
4413 ins_attrib ins_alignment(1); // Required alignment attribute (must
4414 // be a power of 2) specifies the
4415 // alignment that some part of the
4416 // instruction (not necessarily the
4417 // start) requires. If > 1, a
4418 // compute_padding() function must be
4419 // provided for the instruction
4421 //----------OPERANDS-----------------------------------------------------------
4422 // Operand definitions must precede instruction definitions for correct parsing
4423 // in the ADLC because operands constitute user defined types which are used in
4424 // instruction definitions.
4426 //----------Simple Operands----------------------------------------------------
4427 // Immediate Operands
4428 // Integer Immediate
4429 operand immI()
4430 %{
4431 match(ConI);
4433 op_cost(10);
4434 format %{ %}
4435 interface(CONST_INTER);
4436 %}
4438 // Constant for test vs zero
4439 operand immI0()
4440 %{
4441 predicate(n->get_int() == 0);
4442 match(ConI);
4444 op_cost(0);
4445 format %{ %}
4446 interface(CONST_INTER);
4447 %}
4449 // Constant for increment
4450 operand immI1()
4451 %{
4452 predicate(n->get_int() == 1);
4453 match(ConI);
4455 op_cost(0);
4456 format %{ %}
4457 interface(CONST_INTER);
4458 %}
4460 // Constant for decrement
4461 operand immI_M1()
4462 %{
4463 predicate(n->get_int() == -1);
4464 match(ConI);
4466 op_cost(0);
4467 format %{ %}
4468 interface(CONST_INTER);
4469 %}
4471 // Valid scale values for addressing modes
4472 operand immI2()
4473 %{
4474 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4475 match(ConI);
4477 format %{ %}
4478 interface(CONST_INTER);
4479 %}
4481 operand immI8()
4482 %{
4483 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4484 match(ConI);
4486 op_cost(5);
4487 format %{ %}
4488 interface(CONST_INTER);
4489 %}
4491 operand immI16()
4492 %{
4493 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4494 match(ConI);
4496 op_cost(10);
4497 format %{ %}
4498 interface(CONST_INTER);
4499 %}
4501 // Constant for long shifts
4502 operand immI_32()
4503 %{
4504 predicate( n->get_int() == 32 );
4505 match(ConI);
4507 op_cost(0);
4508 format %{ %}
4509 interface(CONST_INTER);
4510 %}
4512 // Constant for long shifts
4513 operand immI_64()
4514 %{
4515 predicate( n->get_int() == 64 );
4516 match(ConI);
4518 op_cost(0);
4519 format %{ %}
4520 interface(CONST_INTER);
4521 %}
4523 // Pointer Immediate
4524 operand immP()
4525 %{
4526 match(ConP);
4528 op_cost(10);
4529 format %{ %}
4530 interface(CONST_INTER);
4531 %}
4533 // NULL Pointer Immediate
4534 operand immP0()
4535 %{
4536 predicate(n->get_ptr() == 0);
4537 match(ConP);
4539 op_cost(5);
4540 format %{ %}
4541 interface(CONST_INTER);
4542 %}
4544 // Pointer Immediate
4545 operand immN() %{
4546 match(ConN);
4548 op_cost(10);
4549 format %{ %}
4550 interface(CONST_INTER);
4551 %}
4553 // NULL Pointer Immediate
4554 operand immN0() %{
4555 predicate(n->get_narrowcon() == 0);
4556 match(ConN);
4558 op_cost(5);
4559 format %{ %}
4560 interface(CONST_INTER);
4561 %}
4563 operand immP31()
4564 %{
4565 predicate(!n->as_Type()->type()->isa_oopptr()
4566 && (n->get_ptr() >> 31) == 0);
4567 match(ConP);
4569 op_cost(5);
4570 format %{ %}
4571 interface(CONST_INTER);
4572 %}
4575 // Long Immediate
4576 operand immL()
4577 %{
4578 match(ConL);
4580 op_cost(20);
4581 format %{ %}
4582 interface(CONST_INTER);
4583 %}
4585 // Long Immediate 8-bit
4586 operand immL8()
4587 %{
4588 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4589 match(ConL);
4591 op_cost(5);
4592 format %{ %}
4593 interface(CONST_INTER);
4594 %}
4596 // Long Immediate 32-bit unsigned
4597 operand immUL32()
4598 %{
4599 predicate(n->get_long() == (unsigned int) (n->get_long()));
4600 match(ConL);
4602 op_cost(10);
4603 format %{ %}
4604 interface(CONST_INTER);
4605 %}
4607 // Long Immediate 32-bit signed
4608 operand immL32()
4609 %{
4610 predicate(n->get_long() == (int) (n->get_long()));
4611 match(ConL);
4613 op_cost(15);
4614 format %{ %}
4615 interface(CONST_INTER);
4616 %}
4618 // Long Immediate zero
4619 operand immL0()
4620 %{
4621 predicate(n->get_long() == 0L);
4622 match(ConL);
4624 op_cost(10);
4625 format %{ %}
4626 interface(CONST_INTER);
4627 %}
4629 // Constant for increment
4630 operand immL1()
4631 %{
4632 predicate(n->get_long() == 1);
4633 match(ConL);
4635 format %{ %}
4636 interface(CONST_INTER);
4637 %}
4639 // Constant for decrement
4640 operand immL_M1()
4641 %{
4642 predicate(n->get_long() == -1);
4643 match(ConL);
4645 format %{ %}
4646 interface(CONST_INTER);
4647 %}
4649 // Long Immediate: the value 10
4650 operand immL10()
4651 %{
4652 predicate(n->get_long() == 10);
4653 match(ConL);
4655 format %{ %}
4656 interface(CONST_INTER);
4657 %}
4659 // Long immediate from 0 to 127.
4660 // Used for a shorter form of long mul by 10.
4661 operand immL_127()
4662 %{
4663 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4664 match(ConL);
4666 op_cost(10);
4667 format %{ %}
4668 interface(CONST_INTER);
4669 %}
4671 // Long Immediate: low 32-bit mask
4672 operand immL_32bits()
4673 %{
4674 predicate(n->get_long() == 0xFFFFFFFFL);
4675 match(ConL);
4676 op_cost(20);
4678 format %{ %}
4679 interface(CONST_INTER);
4680 %}
4682 // Float Immediate zero
4683 operand immF0()
4684 %{
4685 predicate(jint_cast(n->getf()) == 0);
4686 match(ConF);
4688 op_cost(5);
4689 format %{ %}
4690 interface(CONST_INTER);
4691 %}
4693 // Float Immediate
4694 operand immF()
4695 %{
4696 match(ConF);
4698 op_cost(15);
4699 format %{ %}
4700 interface(CONST_INTER);
4701 %}
4703 // Double Immediate zero
4704 operand immD0()
4705 %{
4706 predicate(jlong_cast(n->getd()) == 0);
4707 match(ConD);
4709 op_cost(5);
4710 format %{ %}
4711 interface(CONST_INTER);
4712 %}
4714 // Double Immediate
4715 operand immD()
4716 %{
4717 match(ConD);
4719 op_cost(15);
4720 format %{ %}
4721 interface(CONST_INTER);
4722 %}
4724 // Immediates for special shifts (sign extend)
4726 // Constants for increment
4727 operand immI_16()
4728 %{
4729 predicate(n->get_int() == 16);
4730 match(ConI);
4732 format %{ %}
4733 interface(CONST_INTER);
4734 %}
4736 operand immI_24()
4737 %{
4738 predicate(n->get_int() == 24);
4739 match(ConI);
4741 format %{ %}
4742 interface(CONST_INTER);
4743 %}
4745 // Constant for byte-wide masking
4746 operand immI_255()
4747 %{
4748 predicate(n->get_int() == 255);
4749 match(ConI);
4751 format %{ %}
4752 interface(CONST_INTER);
4753 %}
4755 // Constant for short-wide masking
4756 operand immI_65535()
4757 %{
4758 predicate(n->get_int() == 65535);
4759 match(ConI);
4761 format %{ %}
4762 interface(CONST_INTER);
4763 %}
4765 // Constant for byte-wide masking
4766 operand immL_255()
4767 %{
4768 predicate(n->get_long() == 255);
4769 match(ConL);
4771 format %{ %}
4772 interface(CONST_INTER);
4773 %}
4775 // Constant for short-wide masking
4776 operand immL_65535()
4777 %{
4778 predicate(n->get_long() == 65535);
4779 match(ConL);
4781 format %{ %}
4782 interface(CONST_INTER);
4783 %}
4785 // Register Operands
4786 // Integer Register
4787 operand rRegI()
4788 %{
4789 constraint(ALLOC_IN_RC(int_reg));
4790 match(RegI);
4792 match(rax_RegI);
4793 match(rbx_RegI);
4794 match(rcx_RegI);
4795 match(rdx_RegI);
4796 match(rdi_RegI);
4798 format %{ %}
4799 interface(REG_INTER);
4800 %}
4802 // Special Registers
4803 operand rax_RegI()
4804 %{
4805 constraint(ALLOC_IN_RC(int_rax_reg));
4806 match(RegI);
4807 match(rRegI);
4809 format %{ "RAX" %}
4810 interface(REG_INTER);
4811 %}
4813 // Special Registers
4814 operand rbx_RegI()
4815 %{
4816 constraint(ALLOC_IN_RC(int_rbx_reg));
4817 match(RegI);
4818 match(rRegI);
4820 format %{ "RBX" %}
4821 interface(REG_INTER);
4822 %}
4824 operand rcx_RegI()
4825 %{
4826 constraint(ALLOC_IN_RC(int_rcx_reg));
4827 match(RegI);
4828 match(rRegI);
4830 format %{ "RCX" %}
4831 interface(REG_INTER);
4832 %}
4834 operand rdx_RegI()
4835 %{
4836 constraint(ALLOC_IN_RC(int_rdx_reg));
4837 match(RegI);
4838 match(rRegI);
4840 format %{ "RDX" %}
4841 interface(REG_INTER);
4842 %}
4844 operand rdi_RegI()
4845 %{
4846 constraint(ALLOC_IN_RC(int_rdi_reg));
4847 match(RegI);
4848 match(rRegI);
4850 format %{ "RDI" %}
4851 interface(REG_INTER);
4852 %}
4854 operand no_rcx_RegI()
4855 %{
4856 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4857 match(RegI);
4858 match(rax_RegI);
4859 match(rbx_RegI);
4860 match(rdx_RegI);
4861 match(rdi_RegI);
4863 format %{ %}
4864 interface(REG_INTER);
4865 %}
4867 operand no_rax_rdx_RegI()
4868 %{
4869 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4870 match(RegI);
4871 match(rbx_RegI);
4872 match(rcx_RegI);
4873 match(rdi_RegI);
4875 format %{ %}
4876 interface(REG_INTER);
4877 %}
4879 // Pointer Register
4880 operand any_RegP()
4881 %{
4882 constraint(ALLOC_IN_RC(any_reg));
4883 match(RegP);
4884 match(rax_RegP);
4885 match(rbx_RegP);
4886 match(rdi_RegP);
4887 match(rsi_RegP);
4888 match(rbp_RegP);
4889 match(r15_RegP);
4890 match(rRegP);
4892 format %{ %}
4893 interface(REG_INTER);
4894 %}
4896 operand rRegP()
4897 %{
4898 constraint(ALLOC_IN_RC(ptr_reg));
4899 match(RegP);
4900 match(rax_RegP);
4901 match(rbx_RegP);
4902 match(rdi_RegP);
4903 match(rsi_RegP);
4904 match(rbp_RegP);
4905 match(r15_RegP); // See Q&A below about r15_RegP.
4907 format %{ %}
4908 interface(REG_INTER);
4909 %}
4912 operand r12RegL() %{
4913 constraint(ALLOC_IN_RC(long_r12_reg));
4914 match(RegL);
4916 format %{ %}
4917 interface(REG_INTER);
4918 %}
4920 operand rRegN() %{
4921 constraint(ALLOC_IN_RC(int_reg));
4922 match(RegN);
4924 format %{ %}
4925 interface(REG_INTER);
4926 %}
4928 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4929 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4930 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4931 // The output of an instruction is controlled by the allocator, which respects
4932 // register class masks, not match rules. Unless an instruction mentions
4933 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4934 // by the allocator as an input.
4936 operand no_rax_RegP()
4937 %{
4938 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4939 match(RegP);
4940 match(rbx_RegP);
4941 match(rsi_RegP);
4942 match(rdi_RegP);
4944 format %{ %}
4945 interface(REG_INTER);
4946 %}
4948 operand no_rbp_RegP()
4949 %{
4950 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4951 match(RegP);
4952 match(rbx_RegP);
4953 match(rsi_RegP);
4954 match(rdi_RegP);
4956 format %{ %}
4957 interface(REG_INTER);
4958 %}
4960 operand no_rax_rbx_RegP()
4961 %{
4962 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4963 match(RegP);
4964 match(rsi_RegP);
4965 match(rdi_RegP);
4967 format %{ %}
4968 interface(REG_INTER);
4969 %}
4971 // Special Registers
4972 // Return a pointer value
4973 operand rax_RegP()
4974 %{
4975 constraint(ALLOC_IN_RC(ptr_rax_reg));
4976 match(RegP);
4977 match(rRegP);
4979 format %{ %}
4980 interface(REG_INTER);
4981 %}
4983 // Special Registers
4984 // Return a compressed pointer value
4985 operand rax_RegN()
4986 %{
4987 constraint(ALLOC_IN_RC(int_rax_reg));
4988 match(RegN);
4989 match(rRegN);
4991 format %{ %}
4992 interface(REG_INTER);
4993 %}
4995 // Used in AtomicAdd
4996 operand rbx_RegP()
4997 %{
4998 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4999 match(RegP);
5000 match(rRegP);
5002 format %{ %}
5003 interface(REG_INTER);
5004 %}
5006 operand rsi_RegP()
5007 %{
5008 constraint(ALLOC_IN_RC(ptr_rsi_reg));
5009 match(RegP);
5010 match(rRegP);
5012 format %{ %}
5013 interface(REG_INTER);
5014 %}
5016 // Used in rep stosq
5017 operand rdi_RegP()
5018 %{
5019 constraint(ALLOC_IN_RC(ptr_rdi_reg));
5020 match(RegP);
5021 match(rRegP);
5023 format %{ %}
5024 interface(REG_INTER);
5025 %}
5027 operand rbp_RegP()
5028 %{
5029 constraint(ALLOC_IN_RC(ptr_rbp_reg));
5030 match(RegP);
5031 match(rRegP);
5033 format %{ %}
5034 interface(REG_INTER);
5035 %}
5037 operand r15_RegP()
5038 %{
5039 constraint(ALLOC_IN_RC(ptr_r15_reg));
5040 match(RegP);
5041 match(rRegP);
5043 format %{ %}
5044 interface(REG_INTER);
5045 %}
5047 operand rRegL()
5048 %{
5049 constraint(ALLOC_IN_RC(long_reg));
5050 match(RegL);
5051 match(rax_RegL);
5052 match(rdx_RegL);
5054 format %{ %}
5055 interface(REG_INTER);
5056 %}
5058 // Special Registers
5059 operand no_rax_rdx_RegL()
5060 %{
5061 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5062 match(RegL);
5063 match(rRegL);
5065 format %{ %}
5066 interface(REG_INTER);
5067 %}
5069 operand no_rax_RegL()
5070 %{
5071 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5072 match(RegL);
5073 match(rRegL);
5074 match(rdx_RegL);
5076 format %{ %}
5077 interface(REG_INTER);
5078 %}
5080 operand no_rcx_RegL()
5081 %{
5082 constraint(ALLOC_IN_RC(long_no_rcx_reg));
5083 match(RegL);
5084 match(rRegL);
5086 format %{ %}
5087 interface(REG_INTER);
5088 %}
5090 operand rax_RegL()
5091 %{
5092 constraint(ALLOC_IN_RC(long_rax_reg));
5093 match(RegL);
5094 match(rRegL);
5096 format %{ "RAX" %}
5097 interface(REG_INTER);
5098 %}
5100 operand rcx_RegL()
5101 %{
5102 constraint(ALLOC_IN_RC(long_rcx_reg));
5103 match(RegL);
5104 match(rRegL);
5106 format %{ %}
5107 interface(REG_INTER);
5108 %}
5110 operand rdx_RegL()
5111 %{
5112 constraint(ALLOC_IN_RC(long_rdx_reg));
5113 match(RegL);
5114 match(rRegL);
5116 format %{ %}
5117 interface(REG_INTER);
5118 %}
5120 // Flags register, used as output of compare instructions
5121 operand rFlagsReg()
5122 %{
5123 constraint(ALLOC_IN_RC(int_flags));
5124 match(RegFlags);
5126 format %{ "RFLAGS" %}
5127 interface(REG_INTER);
5128 %}
5130 // Flags register, used as output of FLOATING POINT compare instructions
5131 operand rFlagsRegU()
5132 %{
5133 constraint(ALLOC_IN_RC(int_flags));
5134 match(RegFlags);
5136 format %{ "RFLAGS_U" %}
5137 interface(REG_INTER);
5138 %}
5140 operand rFlagsRegUCF() %{
5141 constraint(ALLOC_IN_RC(int_flags));
5142 match(RegFlags);
5143 predicate(false);
5145 format %{ "RFLAGS_U_CF" %}
5146 interface(REG_INTER);
5147 %}
5149 // Float register operands
5150 operand regF()
5151 %{
5152 constraint(ALLOC_IN_RC(float_reg));
5153 match(RegF);
5155 format %{ %}
5156 interface(REG_INTER);
5157 %}
5159 // Double register operands
5160 operand regD()
5161 %{
5162 constraint(ALLOC_IN_RC(double_reg));
5163 match(RegD);
5165 format %{ %}
5166 interface(REG_INTER);
5167 %}
5170 //----------Memory Operands----------------------------------------------------
5171 // Direct Memory Operand
5172 // operand direct(immP addr)
5173 // %{
5174 // match(addr);
5176 // format %{ "[$addr]" %}
5177 // interface(MEMORY_INTER) %{
5178 // base(0xFFFFFFFF);
5179 // index(0x4);
5180 // scale(0x0);
5181 // disp($addr);
5182 // %}
5183 // %}
5185 // Indirect Memory Operand
5186 operand indirect(any_RegP reg)
5187 %{
5188 constraint(ALLOC_IN_RC(ptr_reg));
5189 match(reg);
5191 format %{ "[$reg]" %}
5192 interface(MEMORY_INTER) %{
5193 base($reg);
5194 index(0x4);
5195 scale(0x0);
5196 disp(0x0);
5197 %}
5198 %}
5200 // Indirect Memory Plus Short Offset Operand
5201 operand indOffset8(any_RegP reg, immL8 off)
5202 %{
5203 constraint(ALLOC_IN_RC(ptr_reg));
5204 match(AddP reg off);
5206 format %{ "[$reg + $off (8-bit)]" %}
5207 interface(MEMORY_INTER) %{
5208 base($reg);
5209 index(0x4);
5210 scale(0x0);
5211 disp($off);
5212 %}
5213 %}
5215 // Indirect Memory Plus Long Offset Operand
5216 operand indOffset32(any_RegP reg, immL32 off)
5217 %{
5218 constraint(ALLOC_IN_RC(ptr_reg));
5219 match(AddP reg off);
5221 format %{ "[$reg + $off (32-bit)]" %}
5222 interface(MEMORY_INTER) %{
5223 base($reg);
5224 index(0x4);
5225 scale(0x0);
5226 disp($off);
5227 %}
5228 %}
5230 // Indirect Memory Plus Index Register Plus Offset Operand
5231 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
5232 %{
5233 constraint(ALLOC_IN_RC(ptr_reg));
5234 match(AddP (AddP reg lreg) off);
5236 op_cost(10);
5237 format %{"[$reg + $off + $lreg]" %}
5238 interface(MEMORY_INTER) %{
5239 base($reg);
5240 index($lreg);
5241 scale(0x0);
5242 disp($off);
5243 %}
5244 %}
5246 // Indirect Memory Plus Index Register Plus Offset Operand
5247 operand indIndex(any_RegP reg, rRegL lreg)
5248 %{
5249 constraint(ALLOC_IN_RC(ptr_reg));
5250 match(AddP reg lreg);
5252 op_cost(10);
5253 format %{"[$reg + $lreg]" %}
5254 interface(MEMORY_INTER) %{
5255 base($reg);
5256 index($lreg);
5257 scale(0x0);
5258 disp(0x0);
5259 %}
5260 %}
5262 // Indirect Memory Times Scale Plus Index Register
5263 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
5264 %{
5265 constraint(ALLOC_IN_RC(ptr_reg));
5266 match(AddP reg (LShiftL lreg scale));
5268 op_cost(10);
5269 format %{"[$reg + $lreg << $scale]" %}
5270 interface(MEMORY_INTER) %{
5271 base($reg);
5272 index($lreg);
5273 scale($scale);
5274 disp(0x0);
5275 %}
5276 %}
5278 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5279 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
5280 %{
5281 constraint(ALLOC_IN_RC(ptr_reg));
5282 match(AddP (AddP reg (LShiftL lreg scale)) off);
5284 op_cost(10);
5285 format %{"[$reg + $off + $lreg << $scale]" %}
5286 interface(MEMORY_INTER) %{
5287 base($reg);
5288 index($lreg);
5289 scale($scale);
5290 disp($off);
5291 %}
5292 %}
5294 // Indirect Narrow Oop Plus Offset Operand
5295 operand indNarrowOopOffset(rRegN src, immL32 off) %{
5296 constraint(ALLOC_IN_RC(ptr_reg));
5297 match(AddP (DecodeN src) off);
5299 op_cost(10);
5300 format %{"[R12 + $src << 3 + $off] (compressed oop addressing)" %}
5301 interface(MEMORY_INTER) %{
5302 base(0xc); // R12
5303 index($src);
5304 scale(0x3);
5305 disp($off);
5306 %}
5307 %}
5309 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5310 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
5311 %{
5312 constraint(ALLOC_IN_RC(ptr_reg));
5313 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5314 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
5316 op_cost(10);
5317 format %{"[$reg + $off + $idx << $scale]" %}
5318 interface(MEMORY_INTER) %{
5319 base($reg);
5320 index($idx);
5321 scale($scale);
5322 disp($off);
5323 %}
5324 %}
5326 //----------Special Memory Operands--------------------------------------------
5327 // Stack Slot Operand - This operand is used for loading and storing temporary
5328 // values on the stack where a match requires a value to
5329 // flow through memory.
5330 operand stackSlotP(sRegP reg)
5331 %{
5332 constraint(ALLOC_IN_RC(stack_slots));
5333 // No match rule because this operand is only generated in matching
5335 format %{ "[$reg]" %}
5336 interface(MEMORY_INTER) %{
5337 base(0x4); // RSP
5338 index(0x4); // No Index
5339 scale(0x0); // No Scale
5340 disp($reg); // Stack Offset
5341 %}
5342 %}
5344 operand stackSlotI(sRegI reg)
5345 %{
5346 constraint(ALLOC_IN_RC(stack_slots));
5347 // No match rule because this operand is only generated in matching
5349 format %{ "[$reg]" %}
5350 interface(MEMORY_INTER) %{
5351 base(0x4); // RSP
5352 index(0x4); // No Index
5353 scale(0x0); // No Scale
5354 disp($reg); // Stack Offset
5355 %}
5356 %}
5358 operand stackSlotF(sRegF reg)
5359 %{
5360 constraint(ALLOC_IN_RC(stack_slots));
5361 // No match rule because this operand is only generated in matching
5363 format %{ "[$reg]" %}
5364 interface(MEMORY_INTER) %{
5365 base(0x4); // RSP
5366 index(0x4); // No Index
5367 scale(0x0); // No Scale
5368 disp($reg); // Stack Offset
5369 %}
5370 %}
5372 operand stackSlotD(sRegD reg)
5373 %{
5374 constraint(ALLOC_IN_RC(stack_slots));
5375 // No match rule because this operand is only generated in matching
5377 format %{ "[$reg]" %}
5378 interface(MEMORY_INTER) %{
5379 base(0x4); // RSP
5380 index(0x4); // No Index
5381 scale(0x0); // No Scale
5382 disp($reg); // Stack Offset
5383 %}
5384 %}
5385 operand stackSlotL(sRegL reg)
5386 %{
5387 constraint(ALLOC_IN_RC(stack_slots));
5388 // No match rule because this operand is only generated in matching
5390 format %{ "[$reg]" %}
5391 interface(MEMORY_INTER) %{
5392 base(0x4); // RSP
5393 index(0x4); // No Index
5394 scale(0x0); // No Scale
5395 disp($reg); // Stack Offset
5396 %}
5397 %}
5399 //----------Conditional Branch Operands----------------------------------------
5400 // Comparison Op - This is the operation of the comparison, and is limited to
5401 // the following set of codes:
5402 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5403 //
5404 // Other attributes of the comparison, such as unsignedness, are specified
5405 // by the comparison instruction that sets a condition code flags register.
5406 // That result is represented by a flags operand whose subtype is appropriate
5407 // to the unsignedness (etc.) of the comparison.
5408 //
5409 // Later, the instruction which matches both the Comparison Op (a Bool) and
5410 // the flags (produced by the Cmp) specifies the coding of the comparison op
5411 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5413 // Comparision Code
5414 operand cmpOp()
5415 %{
5416 match(Bool);
5418 format %{ "" %}
5419 interface(COND_INTER) %{
5420 equal(0x4, "e");
5421 not_equal(0x5, "ne");
5422 less(0xC, "l");
5423 greater_equal(0xD, "ge");
5424 less_equal(0xE, "le");
5425 greater(0xF, "g");
5426 %}
5427 %}
5429 // Comparison Code, unsigned compare. Used by FP also, with
5430 // C2 (unordered) turned into GT or LT already. The other bits
5431 // C0 and C3 are turned into Carry & Zero flags.
5432 operand cmpOpU()
5433 %{
5434 match(Bool);
5436 format %{ "" %}
5437 interface(COND_INTER) %{
5438 equal(0x4, "e");
5439 not_equal(0x5, "ne");
5440 less(0x2, "b");
5441 greater_equal(0x3, "nb");
5442 less_equal(0x6, "be");
5443 greater(0x7, "nbe");
5444 %}
5445 %}
5448 // Floating comparisons that don't require any fixup for the unordered case
5449 operand cmpOpUCF() %{
5450 match(Bool);
5451 predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5452 n->as_Bool()->_test._test == BoolTest::ge ||
5453 n->as_Bool()->_test._test == BoolTest::le ||
5454 n->as_Bool()->_test._test == BoolTest::gt);
5455 format %{ "" %}
5456 interface(COND_INTER) %{
5457 equal(0x4, "e");
5458 not_equal(0x5, "ne");
5459 less(0x2, "b");
5460 greater_equal(0x3, "nb");
5461 less_equal(0x6, "be");
5462 greater(0x7, "nbe");
5463 %}
5464 %}
5467 // Floating comparisons that can be fixed up with extra conditional jumps
5468 operand cmpOpUCF2() %{
5469 match(Bool);
5470 predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5471 n->as_Bool()->_test._test == BoolTest::eq);
5472 format %{ "" %}
5473 interface(COND_INTER) %{
5474 equal(0x4, "e");
5475 not_equal(0x5, "ne");
5476 less(0x2, "b");
5477 greater_equal(0x3, "nb");
5478 less_equal(0x6, "be");
5479 greater(0x7, "nbe");
5480 %}
5481 %}
5484 //----------OPERAND CLASSES----------------------------------------------------
5485 // Operand Classes are groups of operands that are used as to simplify
5486 // instruction definitions by not requiring the AD writer to specify seperate
5487 // instructions for every form of operand when the instruction accepts
5488 // multiple operand types with the same basic encoding and format. The classic
5489 // case of this is memory operands.
5491 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5492 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5493 indNarrowOopOffset);
5495 //----------PIPELINE-----------------------------------------------------------
5496 // Rules which define the behavior of the target architectures pipeline.
5497 pipeline %{
5499 //----------ATTRIBUTES---------------------------------------------------------
5500 attributes %{
5501 variable_size_instructions; // Fixed size instructions
5502 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5503 instruction_unit_size = 1; // An instruction is 1 bytes long
5504 instruction_fetch_unit_size = 16; // The processor fetches one line
5505 instruction_fetch_units = 1; // of 16 bytes
5507 // List of nop instructions
5508 nops( MachNop );
5509 %}
5511 //----------RESOURCES----------------------------------------------------------
5512 // Resources are the functional units available to the machine
5514 // Generic P2/P3 pipeline
5515 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5516 // 3 instructions decoded per cycle.
5517 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5518 // 3 ALU op, only ALU0 handles mul instructions.
5519 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5520 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5521 BR, FPU,
5522 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5524 //----------PIPELINE DESCRIPTION-----------------------------------------------
5525 // Pipeline Description specifies the stages in the machine's pipeline
5527 // Generic P2/P3 pipeline
5528 pipe_desc(S0, S1, S2, S3, S4, S5);
5530 //----------PIPELINE CLASSES---------------------------------------------------
5531 // Pipeline Classes describe the stages in which input and output are
5532 // referenced by the hardware pipeline.
5534 // Naming convention: ialu or fpu
5535 // Then: _reg
5536 // Then: _reg if there is a 2nd register
5537 // Then: _long if it's a pair of instructions implementing a long
5538 // Then: _fat if it requires the big decoder
5539 // Or: _mem if it requires the big decoder and a memory unit.
5541 // Integer ALU reg operation
5542 pipe_class ialu_reg(rRegI dst)
5543 %{
5544 single_instruction;
5545 dst : S4(write);
5546 dst : S3(read);
5547 DECODE : S0; // any decoder
5548 ALU : S3; // any alu
5549 %}
5551 // Long ALU reg operation
5552 pipe_class ialu_reg_long(rRegL dst)
5553 %{
5554 instruction_count(2);
5555 dst : S4(write);
5556 dst : S3(read);
5557 DECODE : S0(2); // any 2 decoders
5558 ALU : S3(2); // both alus
5559 %}
5561 // Integer ALU reg operation using big decoder
5562 pipe_class ialu_reg_fat(rRegI dst)
5563 %{
5564 single_instruction;
5565 dst : S4(write);
5566 dst : S3(read);
5567 D0 : S0; // big decoder only
5568 ALU : S3; // any alu
5569 %}
5571 // Long ALU reg operation using big decoder
5572 pipe_class ialu_reg_long_fat(rRegL dst)
5573 %{
5574 instruction_count(2);
5575 dst : S4(write);
5576 dst : S3(read);
5577 D0 : S0(2); // big decoder only; twice
5578 ALU : S3(2); // any 2 alus
5579 %}
5581 // Integer ALU reg-reg operation
5582 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5583 %{
5584 single_instruction;
5585 dst : S4(write);
5586 src : S3(read);
5587 DECODE : S0; // any decoder
5588 ALU : S3; // any alu
5589 %}
5591 // Long ALU reg-reg operation
5592 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5593 %{
5594 instruction_count(2);
5595 dst : S4(write);
5596 src : S3(read);
5597 DECODE : S0(2); // any 2 decoders
5598 ALU : S3(2); // both alus
5599 %}
5601 // Integer ALU reg-reg operation
5602 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5603 %{
5604 single_instruction;
5605 dst : S4(write);
5606 src : S3(read);
5607 D0 : S0; // big decoder only
5608 ALU : S3; // any alu
5609 %}
5611 // Long ALU reg-reg operation
5612 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5613 %{
5614 instruction_count(2);
5615 dst : S4(write);
5616 src : S3(read);
5617 D0 : S0(2); // big decoder only; twice
5618 ALU : S3(2); // both alus
5619 %}
5621 // Integer ALU reg-mem operation
5622 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5623 %{
5624 single_instruction;
5625 dst : S5(write);
5626 mem : S3(read);
5627 D0 : S0; // big decoder only
5628 ALU : S4; // any alu
5629 MEM : S3; // any mem
5630 %}
5632 // Integer mem operation (prefetch)
5633 pipe_class ialu_mem(memory mem)
5634 %{
5635 single_instruction;
5636 mem : S3(read);
5637 D0 : S0; // big decoder only
5638 MEM : S3; // any mem
5639 %}
5641 // Integer Store to Memory
5642 pipe_class ialu_mem_reg(memory mem, rRegI src)
5643 %{
5644 single_instruction;
5645 mem : S3(read);
5646 src : S5(read);
5647 D0 : S0; // big decoder only
5648 ALU : S4; // any alu
5649 MEM : S3;
5650 %}
5652 // // Long Store to Memory
5653 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5654 // %{
5655 // instruction_count(2);
5656 // mem : S3(read);
5657 // src : S5(read);
5658 // D0 : S0(2); // big decoder only; twice
5659 // ALU : S4(2); // any 2 alus
5660 // MEM : S3(2); // Both mems
5661 // %}
5663 // Integer Store to Memory
5664 pipe_class ialu_mem_imm(memory mem)
5665 %{
5666 single_instruction;
5667 mem : S3(read);
5668 D0 : S0; // big decoder only
5669 ALU : S4; // any alu
5670 MEM : S3;
5671 %}
5673 // Integer ALU0 reg-reg operation
5674 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5675 %{
5676 single_instruction;
5677 dst : S4(write);
5678 src : S3(read);
5679 D0 : S0; // Big decoder only
5680 ALU0 : S3; // only alu0
5681 %}
5683 // Integer ALU0 reg-mem operation
5684 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5685 %{
5686 single_instruction;
5687 dst : S5(write);
5688 mem : S3(read);
5689 D0 : S0; // big decoder only
5690 ALU0 : S4; // ALU0 only
5691 MEM : S3; // any mem
5692 %}
5694 // Integer ALU reg-reg operation
5695 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5696 %{
5697 single_instruction;
5698 cr : S4(write);
5699 src1 : S3(read);
5700 src2 : S3(read);
5701 DECODE : S0; // any decoder
5702 ALU : S3; // any alu
5703 %}
5705 // Integer ALU reg-imm operation
5706 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5707 %{
5708 single_instruction;
5709 cr : S4(write);
5710 src1 : S3(read);
5711 DECODE : S0; // any decoder
5712 ALU : S3; // any alu
5713 %}
5715 // Integer ALU reg-mem operation
5716 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5717 %{
5718 single_instruction;
5719 cr : S4(write);
5720 src1 : S3(read);
5721 src2 : S3(read);
5722 D0 : S0; // big decoder only
5723 ALU : S4; // any alu
5724 MEM : S3;
5725 %}
5727 // Conditional move reg-reg
5728 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5729 %{
5730 instruction_count(4);
5731 y : S4(read);
5732 q : S3(read);
5733 p : S3(read);
5734 DECODE : S0(4); // any decoder
5735 %}
5737 // Conditional move reg-reg
5738 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5739 %{
5740 single_instruction;
5741 dst : S4(write);
5742 src : S3(read);
5743 cr : S3(read);
5744 DECODE : S0; // any decoder
5745 %}
5747 // Conditional move reg-mem
5748 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5749 %{
5750 single_instruction;
5751 dst : S4(write);
5752 src : S3(read);
5753 cr : S3(read);
5754 DECODE : S0; // any decoder
5755 MEM : S3;
5756 %}
5758 // Conditional move reg-reg long
5759 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5760 %{
5761 single_instruction;
5762 dst : S4(write);
5763 src : S3(read);
5764 cr : S3(read);
5765 DECODE : S0(2); // any 2 decoders
5766 %}
5768 // XXX
5769 // // Conditional move double reg-reg
5770 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5771 // %{
5772 // single_instruction;
5773 // dst : S4(write);
5774 // src : S3(read);
5775 // cr : S3(read);
5776 // DECODE : S0; // any decoder
5777 // %}
5779 // Float reg-reg operation
5780 pipe_class fpu_reg(regD dst)
5781 %{
5782 instruction_count(2);
5783 dst : S3(read);
5784 DECODE : S0(2); // any 2 decoders
5785 FPU : S3;
5786 %}
5788 // Float reg-reg operation
5789 pipe_class fpu_reg_reg(regD dst, regD src)
5790 %{
5791 instruction_count(2);
5792 dst : S4(write);
5793 src : S3(read);
5794 DECODE : S0(2); // any 2 decoders
5795 FPU : S3;
5796 %}
5798 // Float reg-reg operation
5799 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5800 %{
5801 instruction_count(3);
5802 dst : S4(write);
5803 src1 : S3(read);
5804 src2 : S3(read);
5805 DECODE : S0(3); // any 3 decoders
5806 FPU : S3(2);
5807 %}
5809 // Float reg-reg operation
5810 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5811 %{
5812 instruction_count(4);
5813 dst : S4(write);
5814 src1 : S3(read);
5815 src2 : S3(read);
5816 src3 : S3(read);
5817 DECODE : S0(4); // any 3 decoders
5818 FPU : S3(2);
5819 %}
5821 // Float reg-reg operation
5822 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5823 %{
5824 instruction_count(4);
5825 dst : S4(write);
5826 src1 : S3(read);
5827 src2 : S3(read);
5828 src3 : S3(read);
5829 DECODE : S1(3); // any 3 decoders
5830 D0 : S0; // Big decoder only
5831 FPU : S3(2);
5832 MEM : S3;
5833 %}
5835 // Float reg-mem operation
5836 pipe_class fpu_reg_mem(regD dst, memory mem)
5837 %{
5838 instruction_count(2);
5839 dst : S5(write);
5840 mem : S3(read);
5841 D0 : S0; // big decoder only
5842 DECODE : S1; // any decoder for FPU POP
5843 FPU : S4;
5844 MEM : S3; // any mem
5845 %}
5847 // Float reg-mem operation
5848 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5849 %{
5850 instruction_count(3);
5851 dst : S5(write);
5852 src1 : S3(read);
5853 mem : S3(read);
5854 D0 : S0; // big decoder only
5855 DECODE : S1(2); // any decoder for FPU POP
5856 FPU : S4;
5857 MEM : S3; // any mem
5858 %}
5860 // Float mem-reg operation
5861 pipe_class fpu_mem_reg(memory mem, regD src)
5862 %{
5863 instruction_count(2);
5864 src : S5(read);
5865 mem : S3(read);
5866 DECODE : S0; // any decoder for FPU PUSH
5867 D0 : S1; // big decoder only
5868 FPU : S4;
5869 MEM : S3; // any mem
5870 %}
5872 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5873 %{
5874 instruction_count(3);
5875 src1 : S3(read);
5876 src2 : S3(read);
5877 mem : S3(read);
5878 DECODE : S0(2); // any decoder for FPU PUSH
5879 D0 : S1; // big decoder only
5880 FPU : S4;
5881 MEM : S3; // any mem
5882 %}
5884 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5885 %{
5886 instruction_count(3);
5887 src1 : S3(read);
5888 src2 : S3(read);
5889 mem : S4(read);
5890 DECODE : S0; // any decoder for FPU PUSH
5891 D0 : S0(2); // big decoder only
5892 FPU : S4;
5893 MEM : S3(2); // any mem
5894 %}
5896 pipe_class fpu_mem_mem(memory dst, memory src1)
5897 %{
5898 instruction_count(2);
5899 src1 : S3(read);
5900 dst : S4(read);
5901 D0 : S0(2); // big decoder only
5902 MEM : S3(2); // any mem
5903 %}
5905 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
5906 %{
5907 instruction_count(3);
5908 src1 : S3(read);
5909 src2 : S3(read);
5910 dst : S4(read);
5911 D0 : S0(3); // big decoder only
5912 FPU : S4;
5913 MEM : S3(3); // any mem
5914 %}
5916 pipe_class fpu_mem_reg_con(memory mem, regD src1)
5917 %{
5918 instruction_count(3);
5919 src1 : S4(read);
5920 mem : S4(read);
5921 DECODE : S0; // any decoder for FPU PUSH
5922 D0 : S0(2); // big decoder only
5923 FPU : S4;
5924 MEM : S3(2); // any mem
5925 %}
5927 // Float load constant
5928 pipe_class fpu_reg_con(regD dst)
5929 %{
5930 instruction_count(2);
5931 dst : S5(write);
5932 D0 : S0; // big decoder only for the load
5933 DECODE : S1; // any decoder for FPU POP
5934 FPU : S4;
5935 MEM : S3; // any mem
5936 %}
5938 // Float load constant
5939 pipe_class fpu_reg_reg_con(regD dst, regD src)
5940 %{
5941 instruction_count(3);
5942 dst : S5(write);
5943 src : S3(read);
5944 D0 : S0; // big decoder only for the load
5945 DECODE : S1(2); // any decoder for FPU POP
5946 FPU : S4;
5947 MEM : S3; // any mem
5948 %}
5950 // UnConditional branch
5951 pipe_class pipe_jmp(label labl)
5952 %{
5953 single_instruction;
5954 BR : S3;
5955 %}
5957 // Conditional branch
5958 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
5959 %{
5960 single_instruction;
5961 cr : S1(read);
5962 BR : S3;
5963 %}
5965 // Allocation idiom
5966 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
5967 %{
5968 instruction_count(1); force_serialization;
5969 fixed_latency(6);
5970 heap_ptr : S3(read);
5971 DECODE : S0(3);
5972 D0 : S2;
5973 MEM : S3;
5974 ALU : S3(2);
5975 dst : S5(write);
5976 BR : S5;
5977 %}
5979 // Generic big/slow expanded idiom
5980 pipe_class pipe_slow()
5981 %{
5982 instruction_count(10); multiple_bundles; force_serialization;
5983 fixed_latency(100);
5984 D0 : S0(2);
5985 MEM : S3(2);
5986 %}
5988 // The real do-nothing guy
5989 pipe_class empty()
5990 %{
5991 instruction_count(0);
5992 %}
5994 // Define the class for the Nop node
5995 define
5996 %{
5997 MachNop = empty;
5998 %}
6000 %}
6002 //----------INSTRUCTIONS-------------------------------------------------------
6003 //
6004 // match -- States which machine-independent subtree may be replaced
6005 // by this instruction.
6006 // ins_cost -- The estimated cost of this instruction is used by instruction
6007 // selection to identify a minimum cost tree of machine
6008 // instructions that matches a tree of machine-independent
6009 // instructions.
6010 // format -- A string providing the disassembly for this instruction.
6011 // The value of an instruction's operand may be inserted
6012 // by referring to it with a '$' prefix.
6013 // opcode -- Three instruction opcodes may be provided. These are referred
6014 // to within an encode class as $primary, $secondary, and $tertiary
6015 // rrspectively. The primary opcode is commonly used to
6016 // indicate the type of machine instruction, while secondary
6017 // and tertiary are often used for prefix options or addressing
6018 // modes.
6019 // ins_encode -- A list of encode classes with parameters. The encode class
6020 // name must have been defined in an 'enc_class' specification
6021 // in the encode section of the architecture description.
6024 //----------Load/Store/Move Instructions---------------------------------------
6025 //----------Load Instructions--------------------------------------------------
6027 // Load Byte (8 bit signed)
6028 instruct loadB(rRegI dst, memory mem)
6029 %{
6030 match(Set dst (LoadB mem));
6032 ins_cost(125);
6033 format %{ "movsbl $dst, $mem\t# byte" %}
6034 opcode(0x0F, 0xBE);
6035 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6036 ins_pipe(ialu_reg_mem);
6037 %}
6039 // Load Byte (8 bit signed) into long
6040 // instruct loadB2L(rRegL dst, memory mem)
6041 // %{
6042 // match(Set dst (ConvI2L (LoadB mem)));
6044 // ins_cost(125);
6045 // format %{ "movsbq $dst, $mem\t# byte -> long" %}
6046 // opcode(0x0F, 0xBE);
6047 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6048 // ins_pipe(ialu_reg_mem);
6049 // %}
6051 // Load Byte (8 bit UNsigned)
6052 instruct loadUB(rRegI dst, memory mem, immI_255 bytemask)
6053 %{
6054 match(Set dst (AndI (LoadB mem) bytemask));
6056 ins_cost(125);
6057 format %{ "movzbl $dst, $mem\t# ubyte" %}
6058 opcode(0x0F, 0xB6);
6059 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6060 ins_pipe(ialu_reg_mem);
6061 %}
6063 // Load Byte (8 bit UNsigned) into long
6064 // instruct loadUB2L(rRegL dst, memory mem, immI_255 bytemask)
6065 // %{
6066 // match(Set dst (ConvI2L (AndI (LoadB mem) bytemask)));
6068 // ins_cost(125);
6069 // format %{ "movzbl $dst, $mem\t# ubyte -> long" %}
6070 // opcode(0x0F, 0xB6);
6071 // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6072 // ins_pipe(ialu_reg_mem);
6073 // %}
6075 // Load Short (16 bit signed)
6076 instruct loadS(rRegI dst, memory mem)
6077 %{
6078 match(Set dst (LoadS mem));
6080 ins_cost(125); // XXX
6081 format %{ "movswl $dst, $mem\t# short" %}
6082 opcode(0x0F, 0xBF);
6083 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6084 ins_pipe(ialu_reg_mem);
6085 %}
6087 // Load Short (16 bit signed) into long
6088 // instruct loadS2L(rRegL dst, memory mem)
6089 // %{
6090 // match(Set dst (ConvI2L (LoadS mem)));
6092 // ins_cost(125); // XXX
6093 // format %{ "movswq $dst, $mem\t# short -> long" %}
6094 // opcode(0x0F, 0xBF);
6095 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6096 // ins_pipe(ialu_reg_mem);
6097 // %}
6099 // Load Char (16 bit UNsigned)
6100 instruct loadC(rRegI dst, memory mem)
6101 %{
6102 match(Set dst (LoadC mem));
6104 ins_cost(125);
6105 format %{ "movzwl $dst, $mem\t# char" %}
6106 opcode(0x0F, 0xB7);
6107 ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6108 ins_pipe(ialu_reg_mem);
6109 %}
6111 // Load Char (16 bit UNsigned) into long
6112 // instruct loadC2L(rRegL dst, memory mem)
6113 // %{
6114 // match(Set dst (ConvI2L (LoadC mem)));
6116 // ins_cost(125);
6117 // format %{ "movzwl $dst, $mem\t# char -> long" %}
6118 // opcode(0x0F, 0xB7);
6119 // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6120 // ins_pipe(ialu_reg_mem);
6121 // %}
6123 // Load Integer
6124 instruct loadI(rRegI dst, memory mem)
6125 %{
6126 match(Set dst (LoadI mem));
6128 ins_cost(125); // XXX
6129 format %{ "movl $dst, $mem\t# int" %}
6130 opcode(0x8B);
6131 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6132 ins_pipe(ialu_reg_mem);
6133 %}
6135 // Load Long
6136 instruct loadL(rRegL dst, memory mem)
6137 %{
6138 match(Set dst (LoadL mem));
6140 ins_cost(125); // XXX
6141 format %{ "movq $dst, $mem\t# long" %}
6142 opcode(0x8B);
6143 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6144 ins_pipe(ialu_reg_mem); // XXX
6145 %}
6147 // Load Range
6148 instruct loadRange(rRegI dst, memory mem)
6149 %{
6150 match(Set dst (LoadRange mem));
6152 ins_cost(125); // XXX
6153 format %{ "movl $dst, $mem\t# range" %}
6154 opcode(0x8B);
6155 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6156 ins_pipe(ialu_reg_mem);
6157 %}
6159 // Load Pointer
6160 instruct loadP(rRegP dst, memory mem)
6161 %{
6162 match(Set dst (LoadP mem));
6164 ins_cost(125); // XXX
6165 format %{ "movq $dst, $mem\t# ptr" %}
6166 opcode(0x8B);
6167 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6168 ins_pipe(ialu_reg_mem); // XXX
6169 %}
6171 // Load Compressed Pointer
6172 instruct loadN(rRegN dst, memory mem)
6173 %{
6174 match(Set dst (LoadN mem));
6176 ins_cost(125); // XXX
6177 format %{ "movl $dst, $mem\t# compressed ptr" %}
6178 ins_encode %{
6179 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6180 Register dst = as_Register($dst$$reg);
6181 __ movl(dst, addr);
6182 %}
6183 ins_pipe(ialu_reg_mem); // XXX
6184 %}
6187 // Load Klass Pointer
6188 instruct loadKlass(rRegP dst, memory mem)
6189 %{
6190 match(Set dst (LoadKlass mem));
6192 ins_cost(125); // XXX
6193 format %{ "movq $dst, $mem\t# class" %}
6194 opcode(0x8B);
6195 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6196 ins_pipe(ialu_reg_mem); // XXX
6197 %}
6199 // Load narrow Klass Pointer
6200 instruct loadNKlass(rRegN dst, memory mem)
6201 %{
6202 match(Set dst (LoadNKlass mem));
6204 ins_cost(125); // XXX
6205 format %{ "movl $dst, $mem\t# compressed klass ptr" %}
6206 ins_encode %{
6207 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6208 Register dst = as_Register($dst$$reg);
6209 __ movl(dst, addr);
6210 %}
6211 ins_pipe(ialu_reg_mem); // XXX
6212 %}
6214 // Load Float
6215 instruct loadF(regF dst, memory mem)
6216 %{
6217 match(Set dst (LoadF mem));
6219 ins_cost(145); // XXX
6220 format %{ "movss $dst, $mem\t# float" %}
6221 opcode(0xF3, 0x0F, 0x10);
6222 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6223 ins_pipe(pipe_slow); // XXX
6224 %}
6226 // Load Double
6227 instruct loadD_partial(regD dst, memory mem)
6228 %{
6229 predicate(!UseXmmLoadAndClearUpper);
6230 match(Set dst (LoadD mem));
6232 ins_cost(145); // XXX
6233 format %{ "movlpd $dst, $mem\t# double" %}
6234 opcode(0x66, 0x0F, 0x12);
6235 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6236 ins_pipe(pipe_slow); // XXX
6237 %}
6239 instruct loadD(regD dst, memory mem)
6240 %{
6241 predicate(UseXmmLoadAndClearUpper);
6242 match(Set dst (LoadD mem));
6244 ins_cost(145); // XXX
6245 format %{ "movsd $dst, $mem\t# double" %}
6246 opcode(0xF2, 0x0F, 0x10);
6247 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6248 ins_pipe(pipe_slow); // XXX
6249 %}
6251 // Load Aligned Packed Byte to XMM register
6252 instruct loadA8B(regD dst, memory mem) %{
6253 match(Set dst (Load8B mem));
6254 ins_cost(125);
6255 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6256 ins_encode( movq_ld(dst, mem));
6257 ins_pipe( pipe_slow );
6258 %}
6260 // Load Aligned Packed Short to XMM register
6261 instruct loadA4S(regD dst, memory mem) %{
6262 match(Set dst (Load4S mem));
6263 ins_cost(125);
6264 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6265 ins_encode( movq_ld(dst, mem));
6266 ins_pipe( pipe_slow );
6267 %}
6269 // Load Aligned Packed Char to XMM register
6270 instruct loadA4C(regD dst, memory mem) %{
6271 match(Set dst (Load4C mem));
6272 ins_cost(125);
6273 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6274 ins_encode( movq_ld(dst, mem));
6275 ins_pipe( pipe_slow );
6276 %}
6278 // Load Aligned Packed Integer to XMM register
6279 instruct load2IU(regD dst, memory mem) %{
6280 match(Set dst (Load2I mem));
6281 ins_cost(125);
6282 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6283 ins_encode( movq_ld(dst, mem));
6284 ins_pipe( pipe_slow );
6285 %}
6287 // Load Aligned Packed Single to XMM
6288 instruct loadA2F(regD dst, memory mem) %{
6289 match(Set dst (Load2F mem));
6290 ins_cost(145);
6291 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6292 ins_encode( movq_ld(dst, mem));
6293 ins_pipe( pipe_slow );
6294 %}
6296 // Load Effective Address
6297 instruct leaP8(rRegP dst, indOffset8 mem)
6298 %{
6299 match(Set dst mem);
6301 ins_cost(110); // XXX
6302 format %{ "leaq $dst, $mem\t# ptr 8" %}
6303 opcode(0x8D);
6304 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6305 ins_pipe(ialu_reg_reg_fat);
6306 %}
6308 instruct leaP32(rRegP dst, indOffset32 mem)
6309 %{
6310 match(Set dst mem);
6312 ins_cost(110);
6313 format %{ "leaq $dst, $mem\t# ptr 32" %}
6314 opcode(0x8D);
6315 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6316 ins_pipe(ialu_reg_reg_fat);
6317 %}
6319 // instruct leaPIdx(rRegP dst, indIndex mem)
6320 // %{
6321 // match(Set dst mem);
6323 // ins_cost(110);
6324 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6325 // opcode(0x8D);
6326 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6327 // ins_pipe(ialu_reg_reg_fat);
6328 // %}
6330 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6331 %{
6332 match(Set dst mem);
6334 ins_cost(110);
6335 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6336 opcode(0x8D);
6337 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6338 ins_pipe(ialu_reg_reg_fat);
6339 %}
6341 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6342 %{
6343 match(Set dst mem);
6345 ins_cost(110);
6346 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6347 opcode(0x8D);
6348 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6349 ins_pipe(ialu_reg_reg_fat);
6350 %}
6352 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6353 %{
6354 match(Set dst mem);
6356 ins_cost(110);
6357 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6358 opcode(0x8D);
6359 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6360 ins_pipe(ialu_reg_reg_fat);
6361 %}
6363 instruct loadConI(rRegI dst, immI src)
6364 %{
6365 match(Set dst src);
6367 format %{ "movl $dst, $src\t# int" %}
6368 ins_encode(load_immI(dst, src));
6369 ins_pipe(ialu_reg_fat); // XXX
6370 %}
6372 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6373 %{
6374 match(Set dst src);
6375 effect(KILL cr);
6377 ins_cost(50);
6378 format %{ "xorl $dst, $dst\t# int" %}
6379 opcode(0x33); /* + rd */
6380 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6381 ins_pipe(ialu_reg);
6382 %}
6384 instruct loadConL(rRegL dst, immL src)
6385 %{
6386 match(Set dst src);
6388 ins_cost(150);
6389 format %{ "movq $dst, $src\t# long" %}
6390 ins_encode(load_immL(dst, src));
6391 ins_pipe(ialu_reg);
6392 %}
6394 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6395 %{
6396 match(Set dst src);
6397 effect(KILL cr);
6399 ins_cost(50);
6400 format %{ "xorl $dst, $dst\t# long" %}
6401 opcode(0x33); /* + rd */
6402 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6403 ins_pipe(ialu_reg); // XXX
6404 %}
6406 instruct loadConUL32(rRegL dst, immUL32 src)
6407 %{
6408 match(Set dst src);
6410 ins_cost(60);
6411 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6412 ins_encode(load_immUL32(dst, src));
6413 ins_pipe(ialu_reg);
6414 %}
6416 instruct loadConL32(rRegL dst, immL32 src)
6417 %{
6418 match(Set dst src);
6420 ins_cost(70);
6421 format %{ "movq $dst, $src\t# long (32-bit)" %}
6422 ins_encode(load_immL32(dst, src));
6423 ins_pipe(ialu_reg);
6424 %}
6426 instruct loadConP(rRegP dst, immP src)
6427 %{
6428 match(Set dst src);
6430 format %{ "movq $dst, $src\t# ptr" %}
6431 ins_encode(load_immP(dst, src));
6432 ins_pipe(ialu_reg_fat); // XXX
6433 %}
6435 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6436 %{
6437 match(Set dst src);
6438 effect(KILL cr);
6440 ins_cost(50);
6441 format %{ "xorl $dst, $dst\t# ptr" %}
6442 opcode(0x33); /* + rd */
6443 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6444 ins_pipe(ialu_reg);
6445 %}
6447 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6448 %{
6449 match(Set dst src);
6450 effect(KILL cr);
6452 ins_cost(60);
6453 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6454 ins_encode(load_immP31(dst, src));
6455 ins_pipe(ialu_reg);
6456 %}
6458 instruct loadConF(regF dst, immF src)
6459 %{
6460 match(Set dst src);
6461 ins_cost(125);
6463 format %{ "movss $dst, [$src]" %}
6464 ins_encode(load_conF(dst, src));
6465 ins_pipe(pipe_slow);
6466 %}
6468 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6469 match(Set dst src);
6470 effect(KILL cr);
6471 format %{ "xorq $dst, $src\t# compressed NULL ptr" %}
6472 ins_encode %{
6473 Register dst = $dst$$Register;
6474 __ xorq(dst, dst);
6475 %}
6476 ins_pipe(ialu_reg);
6477 %}
6479 instruct loadConN(rRegN dst, immN src) %{
6480 match(Set dst src);
6482 ins_cost(125);
6483 format %{ "movl $dst, $src\t# compressed ptr" %}
6484 ins_encode %{
6485 address con = (address)$src$$constant;
6486 Register dst = $dst$$Register;
6487 if (con == NULL) {
6488 ShouldNotReachHere();
6489 } else {
6490 __ set_narrow_oop(dst, (jobject)$src$$constant);
6491 }
6492 %}
6493 ins_pipe(ialu_reg_fat); // XXX
6494 %}
6496 instruct loadConF0(regF dst, immF0 src)
6497 %{
6498 match(Set dst src);
6499 ins_cost(100);
6501 format %{ "xorps $dst, $dst\t# float 0.0" %}
6502 opcode(0x0F, 0x57);
6503 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6504 ins_pipe(pipe_slow);
6505 %}
6507 // Use the same format since predicate() can not be used here.
6508 instruct loadConD(regD dst, immD src)
6509 %{
6510 match(Set dst src);
6511 ins_cost(125);
6513 format %{ "movsd $dst, [$src]" %}
6514 ins_encode(load_conD(dst, src));
6515 ins_pipe(pipe_slow);
6516 %}
6518 instruct loadConD0(regD dst, immD0 src)
6519 %{
6520 match(Set dst src);
6521 ins_cost(100);
6523 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6524 opcode(0x66, 0x0F, 0x57);
6525 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6526 ins_pipe(pipe_slow);
6527 %}
6529 instruct loadSSI(rRegI dst, stackSlotI src)
6530 %{
6531 match(Set dst src);
6533 ins_cost(125);
6534 format %{ "movl $dst, $src\t# int stk" %}
6535 opcode(0x8B);
6536 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6537 ins_pipe(ialu_reg_mem);
6538 %}
6540 instruct loadSSL(rRegL dst, stackSlotL src)
6541 %{
6542 match(Set dst src);
6544 ins_cost(125);
6545 format %{ "movq $dst, $src\t# long stk" %}
6546 opcode(0x8B);
6547 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6548 ins_pipe(ialu_reg_mem);
6549 %}
6551 instruct loadSSP(rRegP dst, stackSlotP src)
6552 %{
6553 match(Set dst src);
6555 ins_cost(125);
6556 format %{ "movq $dst, $src\t# ptr stk" %}
6557 opcode(0x8B);
6558 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6559 ins_pipe(ialu_reg_mem);
6560 %}
6562 instruct loadSSF(regF dst, stackSlotF src)
6563 %{
6564 match(Set dst src);
6566 ins_cost(125);
6567 format %{ "movss $dst, $src\t# float stk" %}
6568 opcode(0xF3, 0x0F, 0x10);
6569 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6570 ins_pipe(pipe_slow); // XXX
6571 %}
6573 // Use the same format since predicate() can not be used here.
6574 instruct loadSSD(regD dst, stackSlotD src)
6575 %{
6576 match(Set dst src);
6578 ins_cost(125);
6579 format %{ "movsd $dst, $src\t# double stk" %}
6580 ins_encode %{
6581 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6582 %}
6583 ins_pipe(pipe_slow); // XXX
6584 %}
6586 // Prefetch instructions.
6587 // Must be safe to execute with invalid address (cannot fault).
6589 instruct prefetchr( memory mem ) %{
6590 predicate(ReadPrefetchInstr==3);
6591 match(PrefetchRead mem);
6592 ins_cost(125);
6594 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6595 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6596 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6597 ins_pipe(ialu_mem);
6598 %}
6600 instruct prefetchrNTA( memory mem ) %{
6601 predicate(ReadPrefetchInstr==0);
6602 match(PrefetchRead mem);
6603 ins_cost(125);
6605 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6606 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6607 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6608 ins_pipe(ialu_mem);
6609 %}
6611 instruct prefetchrT0( memory mem ) %{
6612 predicate(ReadPrefetchInstr==1);
6613 match(PrefetchRead mem);
6614 ins_cost(125);
6616 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6617 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6618 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6619 ins_pipe(ialu_mem);
6620 %}
6622 instruct prefetchrT2( memory mem ) %{
6623 predicate(ReadPrefetchInstr==2);
6624 match(PrefetchRead mem);
6625 ins_cost(125);
6627 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6628 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6629 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6630 ins_pipe(ialu_mem);
6631 %}
6633 instruct prefetchw( memory mem ) %{
6634 predicate(AllocatePrefetchInstr==3);
6635 match(PrefetchWrite mem);
6636 ins_cost(125);
6638 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6639 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6640 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6641 ins_pipe(ialu_mem);
6642 %}
6644 instruct prefetchwNTA( memory mem ) %{
6645 predicate(AllocatePrefetchInstr==0);
6646 match(PrefetchWrite mem);
6647 ins_cost(125);
6649 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6650 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6651 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6652 ins_pipe(ialu_mem);
6653 %}
6655 instruct prefetchwT0( memory mem ) %{
6656 predicate(AllocatePrefetchInstr==1);
6657 match(PrefetchWrite mem);
6658 ins_cost(125);
6660 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6661 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6662 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6663 ins_pipe(ialu_mem);
6664 %}
6666 instruct prefetchwT2( memory mem ) %{
6667 predicate(AllocatePrefetchInstr==2);
6668 match(PrefetchWrite mem);
6669 ins_cost(125);
6671 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6672 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6673 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6674 ins_pipe(ialu_mem);
6675 %}
6677 //----------Store Instructions-------------------------------------------------
6679 // Store Byte
6680 instruct storeB(memory mem, rRegI src)
6681 %{
6682 match(Set mem (StoreB mem src));
6684 ins_cost(125); // XXX
6685 format %{ "movb $mem, $src\t# byte" %}
6686 opcode(0x88);
6687 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6688 ins_pipe(ialu_mem_reg);
6689 %}
6691 // Store Char/Short
6692 instruct storeC(memory mem, rRegI src)
6693 %{
6694 match(Set mem (StoreC mem src));
6696 ins_cost(125); // XXX
6697 format %{ "movw $mem, $src\t# char/short" %}
6698 opcode(0x89);
6699 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6700 ins_pipe(ialu_mem_reg);
6701 %}
6703 // Store Integer
6704 instruct storeI(memory mem, rRegI src)
6705 %{
6706 match(Set mem (StoreI mem src));
6708 ins_cost(125); // XXX
6709 format %{ "movl $mem, $src\t# int" %}
6710 opcode(0x89);
6711 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6712 ins_pipe(ialu_mem_reg);
6713 %}
6715 // Store Long
6716 instruct storeL(memory mem, rRegL src)
6717 %{
6718 match(Set mem (StoreL mem src));
6720 ins_cost(125); // XXX
6721 format %{ "movq $mem, $src\t# long" %}
6722 opcode(0x89);
6723 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6724 ins_pipe(ialu_mem_reg); // XXX
6725 %}
6727 // Store Pointer
6728 instruct storeP(memory mem, any_RegP src)
6729 %{
6730 match(Set mem (StoreP mem src));
6732 ins_cost(125); // XXX
6733 format %{ "movq $mem, $src\t# ptr" %}
6734 opcode(0x89);
6735 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6736 ins_pipe(ialu_mem_reg);
6737 %}
6739 // Store NULL Pointer, mark word, or other simple pointer constant.
6740 instruct storeImmP(memory mem, immP31 src)
6741 %{
6742 match(Set mem (StoreP mem src));
6744 ins_cost(125); // XXX
6745 format %{ "movq $mem, $src\t# ptr" %}
6746 opcode(0xC7); /* C7 /0 */
6747 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6748 ins_pipe(ialu_mem_imm);
6749 %}
6751 // Store Compressed Pointer
6752 instruct storeN(memory mem, rRegN src)
6753 %{
6754 match(Set mem (StoreN mem src));
6756 ins_cost(125); // XXX
6757 format %{ "movl $mem, $src\t# compressed ptr" %}
6758 ins_encode %{
6759 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6760 Register src = as_Register($src$$reg);
6761 __ movl(addr, src);
6762 %}
6763 ins_pipe(ialu_mem_reg);
6764 %}
6766 // Store Integer Immediate
6767 instruct storeImmI(memory mem, immI src)
6768 %{
6769 match(Set mem (StoreI mem src));
6771 ins_cost(150);
6772 format %{ "movl $mem, $src\t# int" %}
6773 opcode(0xC7); /* C7 /0 */
6774 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6775 ins_pipe(ialu_mem_imm);
6776 %}
6778 // Store Long Immediate
6779 instruct storeImmL(memory mem, immL32 src)
6780 %{
6781 match(Set mem (StoreL mem src));
6783 ins_cost(150);
6784 format %{ "movq $mem, $src\t# long" %}
6785 opcode(0xC7); /* C7 /0 */
6786 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6787 ins_pipe(ialu_mem_imm);
6788 %}
6790 // Store Short/Char Immediate
6791 instruct storeImmI16(memory mem, immI16 src)
6792 %{
6793 predicate(UseStoreImmI16);
6794 match(Set mem (StoreC mem src));
6796 ins_cost(150);
6797 format %{ "movw $mem, $src\t# short/char" %}
6798 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
6799 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
6800 ins_pipe(ialu_mem_imm);
6801 %}
6803 // Store Byte Immediate
6804 instruct storeImmB(memory mem, immI8 src)
6805 %{
6806 match(Set mem (StoreB mem src));
6808 ins_cost(150); // XXX
6809 format %{ "movb $mem, $src\t# byte" %}
6810 opcode(0xC6); /* C6 /0 */
6811 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
6812 ins_pipe(ialu_mem_imm);
6813 %}
6815 // Store Aligned Packed Byte XMM register to memory
6816 instruct storeA8B(memory mem, regD src) %{
6817 match(Set mem (Store8B mem src));
6818 ins_cost(145);
6819 format %{ "MOVQ $mem,$src\t! packed8B" %}
6820 ins_encode( movq_st(mem, src));
6821 ins_pipe( pipe_slow );
6822 %}
6824 // Store Aligned Packed Char/Short XMM register to memory
6825 instruct storeA4C(memory mem, regD src) %{
6826 match(Set mem (Store4C mem src));
6827 ins_cost(145);
6828 format %{ "MOVQ $mem,$src\t! packed4C" %}
6829 ins_encode( movq_st(mem, src));
6830 ins_pipe( pipe_slow );
6831 %}
6833 // Store Aligned Packed Integer XMM register to memory
6834 instruct storeA2I(memory mem, regD src) %{
6835 match(Set mem (Store2I mem src));
6836 ins_cost(145);
6837 format %{ "MOVQ $mem,$src\t! packed2I" %}
6838 ins_encode( movq_st(mem, src));
6839 ins_pipe( pipe_slow );
6840 %}
6842 // Store CMS card-mark Immediate
6843 instruct storeImmCM0(memory mem, immI0 src)
6844 %{
6845 match(Set mem (StoreCM mem src));
6847 ins_cost(150); // XXX
6848 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
6849 opcode(0xC6); /* C6 /0 */
6850 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
6851 ins_pipe(ialu_mem_imm);
6852 %}
6854 // Store Aligned Packed Single Float XMM register to memory
6855 instruct storeA2F(memory mem, regD src) %{
6856 match(Set mem (Store2F mem src));
6857 ins_cost(145);
6858 format %{ "MOVQ $mem,$src\t! packed2F" %}
6859 ins_encode( movq_st(mem, src));
6860 ins_pipe( pipe_slow );
6861 %}
6863 // Store Float
6864 instruct storeF(memory mem, regF src)
6865 %{
6866 match(Set mem (StoreF mem src));
6868 ins_cost(95); // XXX
6869 format %{ "movss $mem, $src\t# float" %}
6870 opcode(0xF3, 0x0F, 0x11);
6871 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
6872 ins_pipe(pipe_slow); // XXX
6873 %}
6875 // Store immediate Float value (it is faster than store from XMM register)
6876 instruct storeF_imm(memory mem, immF src)
6877 %{
6878 match(Set mem (StoreF mem src));
6880 ins_cost(50);
6881 format %{ "movl $mem, $src\t# float" %}
6882 opcode(0xC7); /* C7 /0 */
6883 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
6884 ins_pipe(ialu_mem_imm);
6885 %}
6887 // Store Double
6888 instruct storeD(memory mem, regD src)
6889 %{
6890 match(Set mem (StoreD mem src));
6892 ins_cost(95); // XXX
6893 format %{ "movsd $mem, $src\t# double" %}
6894 opcode(0xF2, 0x0F, 0x11);
6895 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
6896 ins_pipe(pipe_slow); // XXX
6897 %}
6899 // Store immediate double 0.0 (it is faster than store from XMM register)
6900 instruct storeD0_imm(memory mem, immD0 src)
6901 %{
6902 match(Set mem (StoreD mem src));
6904 ins_cost(50);
6905 format %{ "movq $mem, $src\t# double 0." %}
6906 opcode(0xC7); /* C7 /0 */
6907 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
6908 ins_pipe(ialu_mem_imm);
6909 %}
6911 instruct storeSSI(stackSlotI dst, rRegI src)
6912 %{
6913 match(Set dst src);
6915 ins_cost(100);
6916 format %{ "movl $dst, $src\t# int stk" %}
6917 opcode(0x89);
6918 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
6919 ins_pipe( ialu_mem_reg );
6920 %}
6922 instruct storeSSL(stackSlotL dst, rRegL src)
6923 %{
6924 match(Set dst src);
6926 ins_cost(100);
6927 format %{ "movq $dst, $src\t# long stk" %}
6928 opcode(0x89);
6929 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
6930 ins_pipe(ialu_mem_reg);
6931 %}
6933 instruct storeSSP(stackSlotP dst, rRegP src)
6934 %{
6935 match(Set dst src);
6937 ins_cost(100);
6938 format %{ "movq $dst, $src\t# ptr stk" %}
6939 opcode(0x89);
6940 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
6941 ins_pipe(ialu_mem_reg);
6942 %}
6944 instruct storeSSF(stackSlotF dst, regF src)
6945 %{
6946 match(Set dst src);
6948 ins_cost(95); // XXX
6949 format %{ "movss $dst, $src\t# float stk" %}
6950 opcode(0xF3, 0x0F, 0x11);
6951 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
6952 ins_pipe(pipe_slow); // XXX
6953 %}
6955 instruct storeSSD(stackSlotD dst, regD src)
6956 %{
6957 match(Set dst src);
6959 ins_cost(95); // XXX
6960 format %{ "movsd $dst, $src\t# double stk" %}
6961 opcode(0xF2, 0x0F, 0x11);
6962 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
6963 ins_pipe(pipe_slow); // XXX
6964 %}
6966 //----------BSWAP Instructions-------------------------------------------------
6967 instruct bytes_reverse_int(rRegI dst) %{
6968 match(Set dst (ReverseBytesI dst));
6970 format %{ "bswapl $dst" %}
6971 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
6972 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
6973 ins_pipe( ialu_reg );
6974 %}
6976 instruct bytes_reverse_long(rRegL dst) %{
6977 match(Set dst (ReverseBytesL dst));
6979 format %{ "bswapq $dst" %}
6981 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
6982 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
6983 ins_pipe( ialu_reg);
6984 %}
6986 instruct loadI_reversed(rRegI dst, memory src) %{
6987 match(Set dst (ReverseBytesI (LoadI src)));
6989 format %{ "bswap_movl $dst, $src" %}
6990 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
6991 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
6992 ins_pipe( ialu_reg_mem );
6993 %}
6995 instruct loadL_reversed(rRegL dst, memory src) %{
6996 match(Set dst (ReverseBytesL (LoadL src)));
6998 format %{ "bswap_movq $dst, $src" %}
6999 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7000 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
7001 ins_pipe( ialu_reg_mem );
7002 %}
7004 instruct storeI_reversed(memory dst, rRegI src) %{
7005 match(Set dst (StoreI dst (ReverseBytesI src)));
7007 format %{ "movl_bswap $dst, $src" %}
7008 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7009 ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
7010 ins_pipe( ialu_mem_reg );
7011 %}
7013 instruct storeL_reversed(memory dst, rRegL src) %{
7014 match(Set dst (StoreL dst (ReverseBytesL src)));
7016 format %{ "movq_bswap $dst, $src" %}
7017 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7018 ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
7019 ins_pipe( ialu_mem_reg );
7020 %}
7022 //----------MemBar Instructions-----------------------------------------------
7023 // Memory barrier flavors
7025 instruct membar_acquire()
7026 %{
7027 match(MemBarAcquire);
7028 ins_cost(0);
7030 size(0);
7031 format %{ "MEMBAR-acquire" %}
7032 ins_encode();
7033 ins_pipe(empty);
7034 %}
7036 instruct membar_acquire_lock()
7037 %{
7038 match(MemBarAcquire);
7039 predicate(Matcher::prior_fast_lock(n));
7040 ins_cost(0);
7042 size(0);
7043 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7044 ins_encode();
7045 ins_pipe(empty);
7046 %}
7048 instruct membar_release()
7049 %{
7050 match(MemBarRelease);
7051 ins_cost(0);
7053 size(0);
7054 format %{ "MEMBAR-release" %}
7055 ins_encode();
7056 ins_pipe(empty);
7057 %}
7059 instruct membar_release_lock()
7060 %{
7061 match(MemBarRelease);
7062 predicate(Matcher::post_fast_unlock(n));
7063 ins_cost(0);
7065 size(0);
7066 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7067 ins_encode();
7068 ins_pipe(empty);
7069 %}
7071 instruct membar_volatile()
7072 %{
7073 match(MemBarVolatile);
7074 ins_cost(400);
7076 format %{ "MEMBAR-volatile" %}
7077 ins_encode(enc_membar_volatile);
7078 ins_pipe(pipe_slow);
7079 %}
7081 instruct unnecessary_membar_volatile()
7082 %{
7083 match(MemBarVolatile);
7084 predicate(Matcher::post_store_load_barrier(n));
7085 ins_cost(0);
7087 size(0);
7088 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7089 ins_encode();
7090 ins_pipe(empty);
7091 %}
7093 //----------Move Instructions--------------------------------------------------
7095 instruct castX2P(rRegP dst, rRegL src)
7096 %{
7097 match(Set dst (CastX2P src));
7099 format %{ "movq $dst, $src\t# long->ptr" %}
7100 ins_encode(enc_copy_wide(dst, src));
7101 ins_pipe(ialu_reg_reg); // XXX
7102 %}
7104 instruct castP2X(rRegL dst, rRegP src)
7105 %{
7106 match(Set dst (CastP2X src));
7108 format %{ "movq $dst, $src\t# ptr -> long" %}
7109 ins_encode(enc_copy_wide(dst, src));
7110 ins_pipe(ialu_reg_reg); // XXX
7111 %}
7114 // Convert oop pointer into compressed form
7115 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7116 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7117 match(Set dst (EncodeP src));
7118 effect(KILL cr);
7119 format %{ "encode_heap_oop $dst,$src" %}
7120 ins_encode %{
7121 Register s = $src$$Register;
7122 Register d = $dst$$Register;
7123 if (s != d) {
7124 __ movq(d, s);
7125 }
7126 __ encode_heap_oop(d);
7127 %}
7128 ins_pipe(ialu_reg_long);
7129 %}
7131 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7132 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7133 match(Set dst (EncodeP src));
7134 effect(KILL cr);
7135 format %{ "encode_heap_oop_not_null $dst,$src" %}
7136 ins_encode %{
7137 Register s = $src$$Register;
7138 Register d = $dst$$Register;
7139 __ encode_heap_oop_not_null(d, s);
7140 %}
7141 ins_pipe(ialu_reg_long);
7142 %}
7144 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7145 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7146 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7147 match(Set dst (DecodeN src));
7148 effect(KILL cr);
7149 format %{ "decode_heap_oop $dst,$src" %}
7150 ins_encode %{
7151 Register s = $src$$Register;
7152 Register d = $dst$$Register;
7153 if (s != d) {
7154 __ movq(d, s);
7155 }
7156 __ decode_heap_oop(d);
7157 %}
7158 ins_pipe(ialu_reg_long);
7159 %}
7161 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{
7162 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7163 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7164 match(Set dst (DecodeN src));
7165 format %{ "decode_heap_oop_not_null $dst,$src" %}
7166 ins_encode %{
7167 Register s = $src$$Register;
7168 Register d = $dst$$Register;
7169 __ decode_heap_oop_not_null(d, s);
7170 %}
7171 ins_pipe(ialu_reg_long);
7172 %}
7175 //----------Conditional Move---------------------------------------------------
7176 // Jump
7177 // dummy instruction for generating temp registers
7178 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7179 match(Jump (LShiftL switch_val shift));
7180 ins_cost(350);
7181 predicate(false);
7182 effect(TEMP dest);
7184 format %{ "leaq $dest, table_base\n\t"
7185 "jmp [$dest + $switch_val << $shift]\n\t" %}
7186 ins_encode(jump_enc_offset(switch_val, shift, dest));
7187 ins_pipe(pipe_jmp);
7188 ins_pc_relative(1);
7189 %}
7191 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7192 match(Jump (AddL (LShiftL switch_val shift) offset));
7193 ins_cost(350);
7194 effect(TEMP dest);
7196 format %{ "leaq $dest, table_base\n\t"
7197 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7198 ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
7199 ins_pipe(pipe_jmp);
7200 ins_pc_relative(1);
7201 %}
7203 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7204 match(Jump switch_val);
7205 ins_cost(350);
7206 effect(TEMP dest);
7208 format %{ "leaq $dest, table_base\n\t"
7209 "jmp [$dest + $switch_val]\n\t" %}
7210 ins_encode(jump_enc(switch_val, dest));
7211 ins_pipe(pipe_jmp);
7212 ins_pc_relative(1);
7213 %}
7215 // Conditional move
7216 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7217 %{
7218 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7220 ins_cost(200); // XXX
7221 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7222 opcode(0x0F, 0x40);
7223 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7224 ins_pipe(pipe_cmov_reg);
7225 %}
7227 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7228 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7230 ins_cost(200); // XXX
7231 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7232 opcode(0x0F, 0x40);
7233 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7234 ins_pipe(pipe_cmov_reg);
7235 %}
7237 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7238 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7239 ins_cost(200);
7240 expand %{
7241 cmovI_regU(cop, cr, dst, src);
7242 %}
7243 %}
7245 // Conditional move
7246 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7247 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7249 ins_cost(250); // XXX
7250 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7251 opcode(0x0F, 0x40);
7252 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7253 ins_pipe(pipe_cmov_mem);
7254 %}
7256 // Conditional move
7257 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7258 %{
7259 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7261 ins_cost(250); // XXX
7262 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7263 opcode(0x0F, 0x40);
7264 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7265 ins_pipe(pipe_cmov_mem);
7266 %}
7268 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7269 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7270 ins_cost(250);
7271 expand %{
7272 cmovI_memU(cop, cr, dst, src);
7273 %}
7274 %}
7276 // Conditional move
7277 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7278 %{
7279 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7281 ins_cost(200); // XXX
7282 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7283 opcode(0x0F, 0x40);
7284 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7285 ins_pipe(pipe_cmov_reg);
7286 %}
7288 // Conditional move
7289 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7290 %{
7291 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7293 ins_cost(200); // XXX
7294 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7295 opcode(0x0F, 0x40);
7296 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7297 ins_pipe(pipe_cmov_reg);
7298 %}
7300 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7301 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7302 ins_cost(200);
7303 expand %{
7304 cmovN_regU(cop, cr, dst, src);
7305 %}
7306 %}
7308 // Conditional move
7309 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7310 %{
7311 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7313 ins_cost(200); // XXX
7314 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7315 opcode(0x0F, 0x40);
7316 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7317 ins_pipe(pipe_cmov_reg); // XXX
7318 %}
7320 // Conditional move
7321 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7322 %{
7323 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7325 ins_cost(200); // XXX
7326 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7327 opcode(0x0F, 0x40);
7328 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7329 ins_pipe(pipe_cmov_reg); // XXX
7330 %}
7332 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7333 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7334 ins_cost(200);
7335 expand %{
7336 cmovP_regU(cop, cr, dst, src);
7337 %}
7338 %}
7340 // DISABLED: Requires the ADLC to emit a bottom_type call that
7341 // correctly meets the two pointer arguments; one is an incoming
7342 // register but the other is a memory operand. ALSO appears to
7343 // be buggy with implicit null checks.
7344 //
7345 //// Conditional move
7346 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7347 //%{
7348 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7349 // ins_cost(250);
7350 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7351 // opcode(0x0F,0x40);
7352 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7353 // ins_pipe( pipe_cmov_mem );
7354 //%}
7355 //
7356 //// Conditional move
7357 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7358 //%{
7359 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7360 // ins_cost(250);
7361 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7362 // opcode(0x0F,0x40);
7363 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7364 // ins_pipe( pipe_cmov_mem );
7365 //%}
7367 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7368 %{
7369 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7371 ins_cost(200); // XXX
7372 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7373 opcode(0x0F, 0x40);
7374 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7375 ins_pipe(pipe_cmov_reg); // XXX
7376 %}
7378 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7379 %{
7380 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7382 ins_cost(200); // XXX
7383 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7384 opcode(0x0F, 0x40);
7385 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7386 ins_pipe(pipe_cmov_mem); // XXX
7387 %}
7389 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7390 %{
7391 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7393 ins_cost(200); // XXX
7394 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7395 opcode(0x0F, 0x40);
7396 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7397 ins_pipe(pipe_cmov_reg); // XXX
7398 %}
7400 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7401 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7402 ins_cost(200);
7403 expand %{
7404 cmovL_regU(cop, cr, dst, src);
7405 %}
7406 %}
7408 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7409 %{
7410 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7412 ins_cost(200); // XXX
7413 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7414 opcode(0x0F, 0x40);
7415 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7416 ins_pipe(pipe_cmov_mem); // XXX
7417 %}
7419 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
7420 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7421 ins_cost(200);
7422 expand %{
7423 cmovL_memU(cop, cr, dst, src);
7424 %}
7425 %}
7427 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7428 %{
7429 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7431 ins_cost(200); // XXX
7432 format %{ "jn$cop skip\t# signed cmove float\n\t"
7433 "movss $dst, $src\n"
7434 "skip:" %}
7435 ins_encode(enc_cmovf_branch(cop, dst, src));
7436 ins_pipe(pipe_slow);
7437 %}
7439 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7440 // %{
7441 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7443 // ins_cost(200); // XXX
7444 // format %{ "jn$cop skip\t# signed cmove float\n\t"
7445 // "movss $dst, $src\n"
7446 // "skip:" %}
7447 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7448 // ins_pipe(pipe_slow);
7449 // %}
7451 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7452 %{
7453 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7455 ins_cost(200); // XXX
7456 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
7457 "movss $dst, $src\n"
7458 "skip:" %}
7459 ins_encode(enc_cmovf_branch(cop, dst, src));
7460 ins_pipe(pipe_slow);
7461 %}
7463 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
7464 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7465 ins_cost(200);
7466 expand %{
7467 cmovF_regU(cop, cr, dst, src);
7468 %}
7469 %}
7471 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7472 %{
7473 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7475 ins_cost(200); // XXX
7476 format %{ "jn$cop skip\t# signed cmove double\n\t"
7477 "movsd $dst, $src\n"
7478 "skip:" %}
7479 ins_encode(enc_cmovd_branch(cop, dst, src));
7480 ins_pipe(pipe_slow);
7481 %}
7483 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7484 %{
7485 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7487 ins_cost(200); // XXX
7488 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
7489 "movsd $dst, $src\n"
7490 "skip:" %}
7491 ins_encode(enc_cmovd_branch(cop, dst, src));
7492 ins_pipe(pipe_slow);
7493 %}
7495 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
7496 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7497 ins_cost(200);
7498 expand %{
7499 cmovD_regU(cop, cr, dst, src);
7500 %}
7501 %}
7503 //----------Arithmetic Instructions--------------------------------------------
7504 //----------Addition Instructions----------------------------------------------
7506 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7507 %{
7508 match(Set dst (AddI dst src));
7509 effect(KILL cr);
7511 format %{ "addl $dst, $src\t# int" %}
7512 opcode(0x03);
7513 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7514 ins_pipe(ialu_reg_reg);
7515 %}
7517 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7518 %{
7519 match(Set dst (AddI dst src));
7520 effect(KILL cr);
7522 format %{ "addl $dst, $src\t# int" %}
7523 opcode(0x81, 0x00); /* /0 id */
7524 ins_encode(OpcSErm(dst, src), Con8or32(src));
7525 ins_pipe( ialu_reg );
7526 %}
7528 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7529 %{
7530 match(Set dst (AddI dst (LoadI src)));
7531 effect(KILL cr);
7533 ins_cost(125); // XXX
7534 format %{ "addl $dst, $src\t# int" %}
7535 opcode(0x03);
7536 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7537 ins_pipe(ialu_reg_mem);
7538 %}
7540 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7541 %{
7542 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7543 effect(KILL cr);
7545 ins_cost(150); // XXX
7546 format %{ "addl $dst, $src\t# int" %}
7547 opcode(0x01); /* Opcode 01 /r */
7548 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7549 ins_pipe(ialu_mem_reg);
7550 %}
7552 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7553 %{
7554 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7555 effect(KILL cr);
7557 ins_cost(125); // XXX
7558 format %{ "addl $dst, $src\t# int" %}
7559 opcode(0x81); /* Opcode 81 /0 id */
7560 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7561 ins_pipe(ialu_mem_imm);
7562 %}
7564 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7565 %{
7566 predicate(UseIncDec);
7567 match(Set dst (AddI dst src));
7568 effect(KILL cr);
7570 format %{ "incl $dst\t# int" %}
7571 opcode(0xFF, 0x00); // FF /0
7572 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7573 ins_pipe(ialu_reg);
7574 %}
7576 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
7577 %{
7578 predicate(UseIncDec);
7579 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7580 effect(KILL cr);
7582 ins_cost(125); // XXX
7583 format %{ "incl $dst\t# int" %}
7584 opcode(0xFF); /* Opcode FF /0 */
7585 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
7586 ins_pipe(ialu_mem_imm);
7587 %}
7589 // XXX why does that use AddI
7590 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
7591 %{
7592 predicate(UseIncDec);
7593 match(Set dst (AddI dst src));
7594 effect(KILL cr);
7596 format %{ "decl $dst\t# int" %}
7597 opcode(0xFF, 0x01); // FF /1
7598 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7599 ins_pipe(ialu_reg);
7600 %}
7602 // XXX why does that use AddI
7603 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
7604 %{
7605 predicate(UseIncDec);
7606 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7607 effect(KILL cr);
7609 ins_cost(125); // XXX
7610 format %{ "decl $dst\t# int" %}
7611 opcode(0xFF); /* Opcode FF /1 */
7612 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
7613 ins_pipe(ialu_mem_imm);
7614 %}
7616 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
7617 %{
7618 match(Set dst (AddI src0 src1));
7620 ins_cost(110);
7621 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
7622 opcode(0x8D); /* 0x8D /r */
7623 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7624 ins_pipe(ialu_reg_reg);
7625 %}
7627 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
7628 %{
7629 match(Set dst (AddL dst src));
7630 effect(KILL cr);
7632 format %{ "addq $dst, $src\t# long" %}
7633 opcode(0x03);
7634 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7635 ins_pipe(ialu_reg_reg);
7636 %}
7638 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
7639 %{
7640 match(Set dst (AddL dst src));
7641 effect(KILL cr);
7643 format %{ "addq $dst, $src\t# long" %}
7644 opcode(0x81, 0x00); /* /0 id */
7645 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7646 ins_pipe( ialu_reg );
7647 %}
7649 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
7650 %{
7651 match(Set dst (AddL dst (LoadL src)));
7652 effect(KILL cr);
7654 ins_cost(125); // XXX
7655 format %{ "addq $dst, $src\t# long" %}
7656 opcode(0x03);
7657 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
7658 ins_pipe(ialu_reg_mem);
7659 %}
7661 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
7662 %{
7663 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7664 effect(KILL cr);
7666 ins_cost(150); // XXX
7667 format %{ "addq $dst, $src\t# long" %}
7668 opcode(0x01); /* Opcode 01 /r */
7669 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7670 ins_pipe(ialu_mem_reg);
7671 %}
7673 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
7674 %{
7675 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7676 effect(KILL cr);
7678 ins_cost(125); // XXX
7679 format %{ "addq $dst, $src\t# long" %}
7680 opcode(0x81); /* Opcode 81 /0 id */
7681 ins_encode(REX_mem_wide(dst),
7682 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7683 ins_pipe(ialu_mem_imm);
7684 %}
7686 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
7687 %{
7688 predicate(UseIncDec);
7689 match(Set dst (AddL dst src));
7690 effect(KILL cr);
7692 format %{ "incq $dst\t# long" %}
7693 opcode(0xFF, 0x00); // FF /0
7694 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7695 ins_pipe(ialu_reg);
7696 %}
7698 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
7699 %{
7700 predicate(UseIncDec);
7701 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7702 effect(KILL cr);
7704 ins_cost(125); // XXX
7705 format %{ "incq $dst\t# long" %}
7706 opcode(0xFF); /* Opcode FF /0 */
7707 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
7708 ins_pipe(ialu_mem_imm);
7709 %}
7711 // XXX why does that use AddL
7712 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
7713 %{
7714 predicate(UseIncDec);
7715 match(Set dst (AddL dst src));
7716 effect(KILL cr);
7718 format %{ "decq $dst\t# long" %}
7719 opcode(0xFF, 0x01); // FF /1
7720 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7721 ins_pipe(ialu_reg);
7722 %}
7724 // XXX why does that use AddL
7725 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
7726 %{
7727 predicate(UseIncDec);
7728 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7729 effect(KILL cr);
7731 ins_cost(125); // XXX
7732 format %{ "decq $dst\t# long" %}
7733 opcode(0xFF); /* Opcode FF /1 */
7734 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
7735 ins_pipe(ialu_mem_imm);
7736 %}
7738 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
7739 %{
7740 match(Set dst (AddL src0 src1));
7742 ins_cost(110);
7743 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
7744 opcode(0x8D); /* 0x8D /r */
7745 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7746 ins_pipe(ialu_reg_reg);
7747 %}
7749 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
7750 %{
7751 match(Set dst (AddP dst src));
7752 effect(KILL cr);
7754 format %{ "addq $dst, $src\t# ptr" %}
7755 opcode(0x03);
7756 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7757 ins_pipe(ialu_reg_reg);
7758 %}
7760 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
7761 %{
7762 match(Set dst (AddP dst src));
7763 effect(KILL cr);
7765 format %{ "addq $dst, $src\t# ptr" %}
7766 opcode(0x81, 0x00); /* /0 id */
7767 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7768 ins_pipe( ialu_reg );
7769 %}
7771 // XXX addP mem ops ????
7773 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
7774 %{
7775 match(Set dst (AddP src0 src1));
7777 ins_cost(110);
7778 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
7779 opcode(0x8D); /* 0x8D /r */
7780 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
7781 ins_pipe(ialu_reg_reg);
7782 %}
7784 instruct checkCastPP(rRegP dst)
7785 %{
7786 match(Set dst (CheckCastPP dst));
7788 size(0);
7789 format %{ "# checkcastPP of $dst" %}
7790 ins_encode(/* empty encoding */);
7791 ins_pipe(empty);
7792 %}
7794 instruct castPP(rRegP dst)
7795 %{
7796 match(Set dst (CastPP dst));
7798 size(0);
7799 format %{ "# castPP of $dst" %}
7800 ins_encode(/* empty encoding */);
7801 ins_pipe(empty);
7802 %}
7804 instruct castII(rRegI dst)
7805 %{
7806 match(Set dst (CastII dst));
7808 size(0);
7809 format %{ "# castII of $dst" %}
7810 ins_encode(/* empty encoding */);
7811 ins_cost(0);
7812 ins_pipe(empty);
7813 %}
7815 // LoadP-locked same as a regular LoadP when used with compare-swap
7816 instruct loadPLocked(rRegP dst, memory mem)
7817 %{
7818 match(Set dst (LoadPLocked mem));
7820 ins_cost(125); // XXX
7821 format %{ "movq $dst, $mem\t# ptr locked" %}
7822 opcode(0x8B);
7823 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
7824 ins_pipe(ialu_reg_mem); // XXX
7825 %}
7827 // LoadL-locked - same as a regular LoadL when used with compare-swap
7828 instruct loadLLocked(rRegL dst, memory mem)
7829 %{
7830 match(Set dst (LoadLLocked mem));
7832 ins_cost(125); // XXX
7833 format %{ "movq $dst, $mem\t# long locked" %}
7834 opcode(0x8B);
7835 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
7836 ins_pipe(ialu_reg_mem); // XXX
7837 %}
7839 // Conditional-store of the updated heap-top.
7840 // Used during allocation of the shared heap.
7841 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
7843 instruct storePConditional(memory heap_top_ptr,
7844 rax_RegP oldval, rRegP newval,
7845 rFlagsReg cr)
7846 %{
7847 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
7849 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
7850 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
7851 opcode(0x0F, 0xB1);
7852 ins_encode(lock_prefix,
7853 REX_reg_mem_wide(newval, heap_top_ptr),
7854 OpcP, OpcS,
7855 reg_mem(newval, heap_top_ptr));
7856 ins_pipe(pipe_cmpxchg);
7857 %}
7859 // Conditional-store of a long value
7860 // Returns a boolean value (0/1) on success. Implemented with a
7861 // CMPXCHG8 on Intel. mem_ptr can actually be in either RSI or RDI
7863 instruct storeLConditional(rRegI res,
7864 memory mem_ptr,
7865 rax_RegL oldval, rRegL newval,
7866 rFlagsReg cr)
7867 %{
7868 match(Set res (StoreLConditional mem_ptr (Binary oldval newval)));
7869 effect(KILL cr);
7871 format %{ "cmpxchgq $mem_ptr, $newval\t# (long) "
7872 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7873 "sete $res\n\t"
7874 "movzbl $res, $res" %}
7875 opcode(0x0F, 0xB1);
7876 ins_encode(lock_prefix,
7877 REX_reg_mem_wide(newval, mem_ptr),
7878 OpcP, OpcS,
7879 reg_mem(newval, mem_ptr),
7880 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7881 REX_reg_breg(res, res), // movzbl
7882 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7883 ins_pipe(pipe_cmpxchg);
7884 %}
7886 // Conditional-store of a long value
7887 // ZF flag is set on success, reset otherwise. Implemented with a
7888 // CMPXCHG8 on Intel. mem_ptr can actually be in either RSI or RDI
7889 instruct storeLConditional_flags(memory mem_ptr,
7890 rax_RegL oldval, rRegL newval,
7891 rFlagsReg cr,
7892 immI0 zero)
7893 %{
7894 match(Set cr (CmpI (StoreLConditional mem_ptr (Binary oldval newval)) zero));
7896 format %{ "cmpxchgq $mem_ptr, $newval\t# (long) "
7897 "If rax == $mem_ptr then store $newval into $mem_ptr" %}
7898 opcode(0x0F, 0xB1);
7899 ins_encode(lock_prefix,
7900 REX_reg_mem_wide(newval, mem_ptr),
7901 OpcP, OpcS,
7902 reg_mem(newval, mem_ptr));
7903 ins_pipe(pipe_cmpxchg);
7904 %}
7906 instruct compareAndSwapP(rRegI res,
7907 memory mem_ptr,
7908 rax_RegP oldval, rRegP newval,
7909 rFlagsReg cr)
7910 %{
7911 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7912 effect(KILL cr, KILL oldval);
7914 format %{ "cmpxchgq $mem_ptr,$newval\t# "
7915 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7916 "sete $res\n\t"
7917 "movzbl $res, $res" %}
7918 opcode(0x0F, 0xB1);
7919 ins_encode(lock_prefix,
7920 REX_reg_mem_wide(newval, mem_ptr),
7921 OpcP, OpcS,
7922 reg_mem(newval, mem_ptr),
7923 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7924 REX_reg_breg(res, res), // movzbl
7925 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7926 ins_pipe( pipe_cmpxchg );
7927 %}
7929 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7930 instruct compareAndSwapL(rRegI res,
7931 memory mem_ptr,
7932 rax_RegL oldval, rRegL newval,
7933 rFlagsReg cr)
7934 %{
7935 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7936 effect(KILL cr, KILL oldval);
7938 format %{ "cmpxchgq $mem_ptr,$newval\t# "
7939 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7940 "sete $res\n\t"
7941 "movzbl $res, $res" %}
7942 opcode(0x0F, 0xB1);
7943 ins_encode(lock_prefix,
7944 REX_reg_mem_wide(newval, mem_ptr),
7945 OpcP, OpcS,
7946 reg_mem(newval, mem_ptr),
7947 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7948 REX_reg_breg(res, res), // movzbl
7949 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7950 ins_pipe( pipe_cmpxchg );
7951 %}
7953 instruct compareAndSwapI(rRegI res,
7954 memory mem_ptr,
7955 rax_RegI oldval, rRegI newval,
7956 rFlagsReg cr)
7957 %{
7958 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7959 effect(KILL cr, KILL oldval);
7961 format %{ "cmpxchgl $mem_ptr,$newval\t# "
7962 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7963 "sete $res\n\t"
7964 "movzbl $res, $res" %}
7965 opcode(0x0F, 0xB1);
7966 ins_encode(lock_prefix,
7967 REX_reg_mem(newval, mem_ptr),
7968 OpcP, OpcS,
7969 reg_mem(newval, mem_ptr),
7970 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7971 REX_reg_breg(res, res), // movzbl
7972 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7973 ins_pipe( pipe_cmpxchg );
7974 %}
7977 instruct compareAndSwapN(rRegI res,
7978 memory mem_ptr,
7979 rax_RegN oldval, rRegN newval,
7980 rFlagsReg cr) %{
7981 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
7982 effect(KILL cr, KILL oldval);
7984 format %{ "cmpxchgl $mem_ptr,$newval\t# "
7985 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
7986 "sete $res\n\t"
7987 "movzbl $res, $res" %}
7988 opcode(0x0F, 0xB1);
7989 ins_encode(lock_prefix,
7990 REX_reg_mem(newval, mem_ptr),
7991 OpcP, OpcS,
7992 reg_mem(newval, mem_ptr),
7993 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
7994 REX_reg_breg(res, res), // movzbl
7995 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
7996 ins_pipe( pipe_cmpxchg );
7997 %}
7999 //----------Subtraction Instructions-------------------------------------------
8001 // Integer Subtraction Instructions
8002 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8003 %{
8004 match(Set dst (SubI dst src));
8005 effect(KILL cr);
8007 format %{ "subl $dst, $src\t# int" %}
8008 opcode(0x2B);
8009 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8010 ins_pipe(ialu_reg_reg);
8011 %}
8013 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8014 %{
8015 match(Set dst (SubI dst src));
8016 effect(KILL cr);
8018 format %{ "subl $dst, $src\t# int" %}
8019 opcode(0x81, 0x05); /* Opcode 81 /5 */
8020 ins_encode(OpcSErm(dst, src), Con8or32(src));
8021 ins_pipe(ialu_reg);
8022 %}
8024 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8025 %{
8026 match(Set dst (SubI dst (LoadI src)));
8027 effect(KILL cr);
8029 ins_cost(125);
8030 format %{ "subl $dst, $src\t# int" %}
8031 opcode(0x2B);
8032 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8033 ins_pipe(ialu_reg_mem);
8034 %}
8036 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8037 %{
8038 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8039 effect(KILL cr);
8041 ins_cost(150);
8042 format %{ "subl $dst, $src\t# int" %}
8043 opcode(0x29); /* Opcode 29 /r */
8044 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8045 ins_pipe(ialu_mem_reg);
8046 %}
8048 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8049 %{
8050 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8051 effect(KILL cr);
8053 ins_cost(125); // XXX
8054 format %{ "subl $dst, $src\t# int" %}
8055 opcode(0x81); /* Opcode 81 /5 id */
8056 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8057 ins_pipe(ialu_mem_imm);
8058 %}
8060 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8061 %{
8062 match(Set dst (SubL dst src));
8063 effect(KILL cr);
8065 format %{ "subq $dst, $src\t# long" %}
8066 opcode(0x2B);
8067 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8068 ins_pipe(ialu_reg_reg);
8069 %}
8071 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8072 %{
8073 match(Set dst (SubL dst src));
8074 effect(KILL cr);
8076 format %{ "subq $dst, $src\t# long" %}
8077 opcode(0x81, 0x05); /* Opcode 81 /5 */
8078 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8079 ins_pipe(ialu_reg);
8080 %}
8082 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8083 %{
8084 match(Set dst (SubL dst (LoadL src)));
8085 effect(KILL cr);
8087 ins_cost(125);
8088 format %{ "subq $dst, $src\t# long" %}
8089 opcode(0x2B);
8090 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8091 ins_pipe(ialu_reg_mem);
8092 %}
8094 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8095 %{
8096 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8097 effect(KILL cr);
8099 ins_cost(150);
8100 format %{ "subq $dst, $src\t# long" %}
8101 opcode(0x29); /* Opcode 29 /r */
8102 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8103 ins_pipe(ialu_mem_reg);
8104 %}
8106 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8107 %{
8108 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8109 effect(KILL cr);
8111 ins_cost(125); // XXX
8112 format %{ "subq $dst, $src\t# long" %}
8113 opcode(0x81); /* Opcode 81 /5 id */
8114 ins_encode(REX_mem_wide(dst),
8115 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8116 ins_pipe(ialu_mem_imm);
8117 %}
8119 // Subtract from a pointer
8120 // XXX hmpf???
8121 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8122 %{
8123 match(Set dst (AddP dst (SubI zero src)));
8124 effect(KILL cr);
8126 format %{ "subq $dst, $src\t# ptr - int" %}
8127 opcode(0x2B);
8128 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8129 ins_pipe(ialu_reg_reg);
8130 %}
8132 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8133 %{
8134 match(Set dst (SubI zero dst));
8135 effect(KILL cr);
8137 format %{ "negl $dst\t# int" %}
8138 opcode(0xF7, 0x03); // Opcode F7 /3
8139 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8140 ins_pipe(ialu_reg);
8141 %}
8143 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8144 %{
8145 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8146 effect(KILL cr);
8148 format %{ "negl $dst\t# int" %}
8149 opcode(0xF7, 0x03); // Opcode F7 /3
8150 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8151 ins_pipe(ialu_reg);
8152 %}
8154 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8155 %{
8156 match(Set dst (SubL zero dst));
8157 effect(KILL cr);
8159 format %{ "negq $dst\t# long" %}
8160 opcode(0xF7, 0x03); // Opcode F7 /3
8161 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8162 ins_pipe(ialu_reg);
8163 %}
8165 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8166 %{
8167 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8168 effect(KILL cr);
8170 format %{ "negq $dst\t# long" %}
8171 opcode(0xF7, 0x03); // Opcode F7 /3
8172 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8173 ins_pipe(ialu_reg);
8174 %}
8177 //----------Multiplication/Division Instructions-------------------------------
8178 // Integer Multiplication Instructions
8179 // Multiply Register
8181 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8182 %{
8183 match(Set dst (MulI dst src));
8184 effect(KILL cr);
8186 ins_cost(300);
8187 format %{ "imull $dst, $src\t# int" %}
8188 opcode(0x0F, 0xAF);
8189 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8190 ins_pipe(ialu_reg_reg_alu0);
8191 %}
8193 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8194 %{
8195 match(Set dst (MulI src imm));
8196 effect(KILL cr);
8198 ins_cost(300);
8199 format %{ "imull $dst, $src, $imm\t# int" %}
8200 opcode(0x69); /* 69 /r id */
8201 ins_encode(REX_reg_reg(dst, src),
8202 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8203 ins_pipe(ialu_reg_reg_alu0);
8204 %}
8206 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8207 %{
8208 match(Set dst (MulI dst (LoadI src)));
8209 effect(KILL cr);
8211 ins_cost(350);
8212 format %{ "imull $dst, $src\t# int" %}
8213 opcode(0x0F, 0xAF);
8214 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8215 ins_pipe(ialu_reg_mem_alu0);
8216 %}
8218 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8219 %{
8220 match(Set dst (MulI (LoadI src) imm));
8221 effect(KILL cr);
8223 ins_cost(300);
8224 format %{ "imull $dst, $src, $imm\t# int" %}
8225 opcode(0x69); /* 69 /r id */
8226 ins_encode(REX_reg_mem(dst, src),
8227 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8228 ins_pipe(ialu_reg_mem_alu0);
8229 %}
8231 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8232 %{
8233 match(Set dst (MulL dst src));
8234 effect(KILL cr);
8236 ins_cost(300);
8237 format %{ "imulq $dst, $src\t# long" %}
8238 opcode(0x0F, 0xAF);
8239 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8240 ins_pipe(ialu_reg_reg_alu0);
8241 %}
8243 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8244 %{
8245 match(Set dst (MulL src imm));
8246 effect(KILL cr);
8248 ins_cost(300);
8249 format %{ "imulq $dst, $src, $imm\t# long" %}
8250 opcode(0x69); /* 69 /r id */
8251 ins_encode(REX_reg_reg_wide(dst, src),
8252 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8253 ins_pipe(ialu_reg_reg_alu0);
8254 %}
8256 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8257 %{
8258 match(Set dst (MulL dst (LoadL src)));
8259 effect(KILL cr);
8261 ins_cost(350);
8262 format %{ "imulq $dst, $src\t# long" %}
8263 opcode(0x0F, 0xAF);
8264 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8265 ins_pipe(ialu_reg_mem_alu0);
8266 %}
8268 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8269 %{
8270 match(Set dst (MulL (LoadL src) imm));
8271 effect(KILL cr);
8273 ins_cost(300);
8274 format %{ "imulq $dst, $src, $imm\t# long" %}
8275 opcode(0x69); /* 69 /r id */
8276 ins_encode(REX_reg_mem_wide(dst, src),
8277 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8278 ins_pipe(ialu_reg_mem_alu0);
8279 %}
8281 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8282 %{
8283 match(Set dst (MulHiL src rax));
8284 effect(USE_KILL rax, KILL cr);
8286 ins_cost(300);
8287 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %}
8288 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8289 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8290 ins_pipe(ialu_reg_reg_alu0);
8291 %}
8293 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8294 rFlagsReg cr)
8295 %{
8296 match(Set rax (DivI rax div));
8297 effect(KILL rdx, KILL cr);
8299 ins_cost(30*100+10*100); // XXX
8300 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8301 "jne,s normal\n\t"
8302 "xorl rdx, rdx\n\t"
8303 "cmpl $div, -1\n\t"
8304 "je,s done\n"
8305 "normal: cdql\n\t"
8306 "idivl $div\n"
8307 "done:" %}
8308 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8309 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8310 ins_pipe(ialu_reg_reg_alu0);
8311 %}
8313 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8314 rFlagsReg cr)
8315 %{
8316 match(Set rax (DivL rax div));
8317 effect(KILL rdx, KILL cr);
8319 ins_cost(30*100+10*100); // XXX
8320 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8321 "cmpq rax, rdx\n\t"
8322 "jne,s normal\n\t"
8323 "xorl rdx, rdx\n\t"
8324 "cmpq $div, -1\n\t"
8325 "je,s done\n"
8326 "normal: cdqq\n\t"
8327 "idivq $div\n"
8328 "done:" %}
8329 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8330 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8331 ins_pipe(ialu_reg_reg_alu0);
8332 %}
8334 // Integer DIVMOD with Register, both quotient and mod results
8335 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8336 rFlagsReg cr)
8337 %{
8338 match(DivModI rax div);
8339 effect(KILL cr);
8341 ins_cost(30*100+10*100); // XXX
8342 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8343 "jne,s normal\n\t"
8344 "xorl rdx, rdx\n\t"
8345 "cmpl $div, -1\n\t"
8346 "je,s done\n"
8347 "normal: cdql\n\t"
8348 "idivl $div\n"
8349 "done:" %}
8350 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8351 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8352 ins_pipe(pipe_slow);
8353 %}
8355 // Long DIVMOD with Register, both quotient and mod results
8356 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8357 rFlagsReg cr)
8358 %{
8359 match(DivModL rax div);
8360 effect(KILL cr);
8362 ins_cost(30*100+10*100); // XXX
8363 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8364 "cmpq rax, rdx\n\t"
8365 "jne,s normal\n\t"
8366 "xorl rdx, rdx\n\t"
8367 "cmpq $div, -1\n\t"
8368 "je,s done\n"
8369 "normal: cdqq\n\t"
8370 "idivq $div\n"
8371 "done:" %}
8372 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8373 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8374 ins_pipe(pipe_slow);
8375 %}
8377 //----------- DivL-By-Constant-Expansions--------------------------------------
8378 // DivI cases are handled by the compiler
8380 // Magic constant, reciprical of 10
8381 instruct loadConL_0x6666666666666667(rRegL dst)
8382 %{
8383 effect(DEF dst);
8385 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8386 ins_encode(load_immL(dst, 0x6666666666666667));
8387 ins_pipe(ialu_reg);
8388 %}
8390 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8391 %{
8392 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8394 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8395 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8396 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8397 ins_pipe(ialu_reg_reg_alu0);
8398 %}
8400 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8401 %{
8402 effect(USE_DEF dst, KILL cr);
8404 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8405 opcode(0xC1, 0x7); /* C1 /7 ib */
8406 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8407 ins_pipe(ialu_reg);
8408 %}
8410 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8411 %{
8412 effect(USE_DEF dst, KILL cr);
8414 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8415 opcode(0xC1, 0x7); /* C1 /7 ib */
8416 ins_encode(reg_opc_imm_wide(dst, 0x2));
8417 ins_pipe(ialu_reg);
8418 %}
8420 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8421 %{
8422 match(Set dst (DivL src div));
8424 ins_cost((5+8)*100);
8425 expand %{
8426 rax_RegL rax; // Killed temp
8427 rFlagsReg cr; // Killed
8428 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
8429 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
8430 sarL_rReg_63(src, cr); // sarq src, 63
8431 sarL_rReg_2(dst, cr); // sarq rdx, 2
8432 subL_rReg(dst, src, cr); // subl rdx, src
8433 %}
8434 %}
8436 //-----------------------------------------------------------------------------
8438 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8439 rFlagsReg cr)
8440 %{
8441 match(Set rdx (ModI rax div));
8442 effect(KILL rax, KILL cr);
8444 ins_cost(300); // XXX
8445 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
8446 "jne,s normal\n\t"
8447 "xorl rdx, rdx\n\t"
8448 "cmpl $div, -1\n\t"
8449 "je,s done\n"
8450 "normal: cdql\n\t"
8451 "idivl $div\n"
8452 "done:" %}
8453 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8454 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8455 ins_pipe(ialu_reg_reg_alu0);
8456 %}
8458 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8459 rFlagsReg cr)
8460 %{
8461 match(Set rdx (ModL rax div));
8462 effect(KILL rax, KILL cr);
8464 ins_cost(300); // XXX
8465 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
8466 "cmpq rax, rdx\n\t"
8467 "jne,s normal\n\t"
8468 "xorl rdx, rdx\n\t"
8469 "cmpq $div, -1\n\t"
8470 "je,s done\n"
8471 "normal: cdqq\n\t"
8472 "idivq $div\n"
8473 "done:" %}
8474 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8475 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8476 ins_pipe(ialu_reg_reg_alu0);
8477 %}
8479 // Integer Shift Instructions
8480 // Shift Left by one
8481 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8482 %{
8483 match(Set dst (LShiftI dst shift));
8484 effect(KILL cr);
8486 format %{ "sall $dst, $shift" %}
8487 opcode(0xD1, 0x4); /* D1 /4 */
8488 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8489 ins_pipe(ialu_reg);
8490 %}
8492 // Shift Left by one
8493 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8494 %{
8495 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8496 effect(KILL cr);
8498 format %{ "sall $dst, $shift\t" %}
8499 opcode(0xD1, 0x4); /* D1 /4 */
8500 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8501 ins_pipe(ialu_mem_imm);
8502 %}
8504 // Shift Left by 8-bit immediate
8505 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8506 %{
8507 match(Set dst (LShiftI dst shift));
8508 effect(KILL cr);
8510 format %{ "sall $dst, $shift" %}
8511 opcode(0xC1, 0x4); /* C1 /4 ib */
8512 ins_encode(reg_opc_imm(dst, shift));
8513 ins_pipe(ialu_reg);
8514 %}
8516 // Shift Left by 8-bit immediate
8517 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8518 %{
8519 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8520 effect(KILL cr);
8522 format %{ "sall $dst, $shift" %}
8523 opcode(0xC1, 0x4); /* C1 /4 ib */
8524 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8525 ins_pipe(ialu_mem_imm);
8526 %}
8528 // Shift Left by variable
8529 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8530 %{
8531 match(Set dst (LShiftI dst shift));
8532 effect(KILL cr);
8534 format %{ "sall $dst, $shift" %}
8535 opcode(0xD3, 0x4); /* D3 /4 */
8536 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8537 ins_pipe(ialu_reg_reg);
8538 %}
8540 // Shift Left by variable
8541 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8542 %{
8543 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8544 effect(KILL cr);
8546 format %{ "sall $dst, $shift" %}
8547 opcode(0xD3, 0x4); /* D3 /4 */
8548 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8549 ins_pipe(ialu_mem_reg);
8550 %}
8552 // Arithmetic shift right by one
8553 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8554 %{
8555 match(Set dst (RShiftI dst shift));
8556 effect(KILL cr);
8558 format %{ "sarl $dst, $shift" %}
8559 opcode(0xD1, 0x7); /* D1 /7 */
8560 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8561 ins_pipe(ialu_reg);
8562 %}
8564 // Arithmetic shift right by one
8565 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8566 %{
8567 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8568 effect(KILL cr);
8570 format %{ "sarl $dst, $shift" %}
8571 opcode(0xD1, 0x7); /* D1 /7 */
8572 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8573 ins_pipe(ialu_mem_imm);
8574 %}
8576 // Arithmetic Shift Right by 8-bit immediate
8577 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8578 %{
8579 match(Set dst (RShiftI dst shift));
8580 effect(KILL cr);
8582 format %{ "sarl $dst, $shift" %}
8583 opcode(0xC1, 0x7); /* C1 /7 ib */
8584 ins_encode(reg_opc_imm(dst, shift));
8585 ins_pipe(ialu_mem_imm);
8586 %}
8588 // Arithmetic Shift Right by 8-bit immediate
8589 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8590 %{
8591 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8592 effect(KILL cr);
8594 format %{ "sarl $dst, $shift" %}
8595 opcode(0xC1, 0x7); /* C1 /7 ib */
8596 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8597 ins_pipe(ialu_mem_imm);
8598 %}
8600 // Arithmetic Shift Right by variable
8601 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8602 %{
8603 match(Set dst (RShiftI dst shift));
8604 effect(KILL cr);
8606 format %{ "sarl $dst, $shift" %}
8607 opcode(0xD3, 0x7); /* D3 /7 */
8608 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8609 ins_pipe(ialu_reg_reg);
8610 %}
8612 // Arithmetic Shift Right by variable
8613 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8614 %{
8615 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8616 effect(KILL cr);
8618 format %{ "sarl $dst, $shift" %}
8619 opcode(0xD3, 0x7); /* D3 /7 */
8620 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8621 ins_pipe(ialu_mem_reg);
8622 %}
8624 // Logical shift right by one
8625 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8626 %{
8627 match(Set dst (URShiftI dst shift));
8628 effect(KILL cr);
8630 format %{ "shrl $dst, $shift" %}
8631 opcode(0xD1, 0x5); /* D1 /5 */
8632 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8633 ins_pipe(ialu_reg);
8634 %}
8636 // Logical shift right by one
8637 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8638 %{
8639 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8640 effect(KILL cr);
8642 format %{ "shrl $dst, $shift" %}
8643 opcode(0xD1, 0x5); /* D1 /5 */
8644 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8645 ins_pipe(ialu_mem_imm);
8646 %}
8648 // Logical Shift Right by 8-bit immediate
8649 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8650 %{
8651 match(Set dst (URShiftI dst shift));
8652 effect(KILL cr);
8654 format %{ "shrl $dst, $shift" %}
8655 opcode(0xC1, 0x5); /* C1 /5 ib */
8656 ins_encode(reg_opc_imm(dst, shift));
8657 ins_pipe(ialu_reg);
8658 %}
8660 // Logical Shift Right by 8-bit immediate
8661 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8662 %{
8663 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8664 effect(KILL cr);
8666 format %{ "shrl $dst, $shift" %}
8667 opcode(0xC1, 0x5); /* C1 /5 ib */
8668 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8669 ins_pipe(ialu_mem_imm);
8670 %}
8672 // Logical Shift Right by variable
8673 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8674 %{
8675 match(Set dst (URShiftI dst shift));
8676 effect(KILL cr);
8678 format %{ "shrl $dst, $shift" %}
8679 opcode(0xD3, 0x5); /* D3 /5 */
8680 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8681 ins_pipe(ialu_reg_reg);
8682 %}
8684 // Logical Shift Right by variable
8685 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8686 %{
8687 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8688 effect(KILL cr);
8690 format %{ "shrl $dst, $shift" %}
8691 opcode(0xD3, 0x5); /* D3 /5 */
8692 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8693 ins_pipe(ialu_mem_reg);
8694 %}
8696 // Long Shift Instructions
8697 // Shift Left by one
8698 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8699 %{
8700 match(Set dst (LShiftL dst shift));
8701 effect(KILL cr);
8703 format %{ "salq $dst, $shift" %}
8704 opcode(0xD1, 0x4); /* D1 /4 */
8705 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8706 ins_pipe(ialu_reg);
8707 %}
8709 // Shift Left by one
8710 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8711 %{
8712 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8713 effect(KILL cr);
8715 format %{ "salq $dst, $shift" %}
8716 opcode(0xD1, 0x4); /* D1 /4 */
8717 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8718 ins_pipe(ialu_mem_imm);
8719 %}
8721 // Shift Left by 8-bit immediate
8722 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8723 %{
8724 match(Set dst (LShiftL dst shift));
8725 effect(KILL cr);
8727 format %{ "salq $dst, $shift" %}
8728 opcode(0xC1, 0x4); /* C1 /4 ib */
8729 ins_encode(reg_opc_imm_wide(dst, shift));
8730 ins_pipe(ialu_reg);
8731 %}
8733 // Shift Left by 8-bit immediate
8734 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8735 %{
8736 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8737 effect(KILL cr);
8739 format %{ "salq $dst, $shift" %}
8740 opcode(0xC1, 0x4); /* C1 /4 ib */
8741 ins_encode(REX_mem_wide(dst), OpcP,
8742 RM_opc_mem(secondary, dst), Con8or32(shift));
8743 ins_pipe(ialu_mem_imm);
8744 %}
8746 // Shift Left by variable
8747 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8748 %{
8749 match(Set dst (LShiftL dst shift));
8750 effect(KILL cr);
8752 format %{ "salq $dst, $shift" %}
8753 opcode(0xD3, 0x4); /* D3 /4 */
8754 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8755 ins_pipe(ialu_reg_reg);
8756 %}
8758 // Shift Left by variable
8759 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8760 %{
8761 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8762 effect(KILL cr);
8764 format %{ "salq $dst, $shift" %}
8765 opcode(0xD3, 0x4); /* D3 /4 */
8766 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8767 ins_pipe(ialu_mem_reg);
8768 %}
8770 // Arithmetic shift right by one
8771 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8772 %{
8773 match(Set dst (RShiftL dst shift));
8774 effect(KILL cr);
8776 format %{ "sarq $dst, $shift" %}
8777 opcode(0xD1, 0x7); /* D1 /7 */
8778 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8779 ins_pipe(ialu_reg);
8780 %}
8782 // Arithmetic shift right by one
8783 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8784 %{
8785 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8786 effect(KILL cr);
8788 format %{ "sarq $dst, $shift" %}
8789 opcode(0xD1, 0x7); /* D1 /7 */
8790 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8791 ins_pipe(ialu_mem_imm);
8792 %}
8794 // Arithmetic Shift Right by 8-bit immediate
8795 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8796 %{
8797 match(Set dst (RShiftL dst shift));
8798 effect(KILL cr);
8800 format %{ "sarq $dst, $shift" %}
8801 opcode(0xC1, 0x7); /* C1 /7 ib */
8802 ins_encode(reg_opc_imm_wide(dst, shift));
8803 ins_pipe(ialu_mem_imm);
8804 %}
8806 // Arithmetic Shift Right by 8-bit immediate
8807 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8808 %{
8809 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8810 effect(KILL cr);
8812 format %{ "sarq $dst, $shift" %}
8813 opcode(0xC1, 0x7); /* C1 /7 ib */
8814 ins_encode(REX_mem_wide(dst), OpcP,
8815 RM_opc_mem(secondary, dst), Con8or32(shift));
8816 ins_pipe(ialu_mem_imm);
8817 %}
8819 // Arithmetic Shift Right by variable
8820 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8821 %{
8822 match(Set dst (RShiftL dst shift));
8823 effect(KILL cr);
8825 format %{ "sarq $dst, $shift" %}
8826 opcode(0xD3, 0x7); /* D3 /7 */
8827 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8828 ins_pipe(ialu_reg_reg);
8829 %}
8831 // Arithmetic Shift Right by variable
8832 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8833 %{
8834 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
8835 effect(KILL cr);
8837 format %{ "sarq $dst, $shift" %}
8838 opcode(0xD3, 0x7); /* D3 /7 */
8839 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8840 ins_pipe(ialu_mem_reg);
8841 %}
8843 // Logical shift right by one
8844 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8845 %{
8846 match(Set dst (URShiftL dst shift));
8847 effect(KILL cr);
8849 format %{ "shrq $dst, $shift" %}
8850 opcode(0xD1, 0x5); /* D1 /5 */
8851 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
8852 ins_pipe(ialu_reg);
8853 %}
8855 // Logical shift right by one
8856 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8857 %{
8858 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8859 effect(KILL cr);
8861 format %{ "shrq $dst, $shift" %}
8862 opcode(0xD1, 0x5); /* D1 /5 */
8863 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8864 ins_pipe(ialu_mem_imm);
8865 %}
8867 // Logical Shift Right by 8-bit immediate
8868 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
8869 %{
8870 match(Set dst (URShiftL dst shift));
8871 effect(KILL cr);
8873 format %{ "shrq $dst, $shift" %}
8874 opcode(0xC1, 0x5); /* C1 /5 ib */
8875 ins_encode(reg_opc_imm_wide(dst, shift));
8876 ins_pipe(ialu_reg);
8877 %}
8879 // Logical Shift Right by 8-bit immediate
8880 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8881 %{
8882 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8883 effect(KILL cr);
8885 format %{ "shrq $dst, $shift" %}
8886 opcode(0xC1, 0x5); /* C1 /5 ib */
8887 ins_encode(REX_mem_wide(dst), OpcP,
8888 RM_opc_mem(secondary, dst), Con8or32(shift));
8889 ins_pipe(ialu_mem_imm);
8890 %}
8892 // Logical Shift Right by variable
8893 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
8894 %{
8895 match(Set dst (URShiftL dst shift));
8896 effect(KILL cr);
8898 format %{ "shrq $dst, $shift" %}
8899 opcode(0xD3, 0x5); /* D3 /5 */
8900 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8901 ins_pipe(ialu_reg_reg);
8902 %}
8904 // Logical Shift Right by variable
8905 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8906 %{
8907 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
8908 effect(KILL cr);
8910 format %{ "shrq $dst, $shift" %}
8911 opcode(0xD3, 0x5); /* D3 /5 */
8912 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8913 ins_pipe(ialu_mem_reg);
8914 %}
8916 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
8917 // This idiom is used by the compiler for the i2b bytecode.
8918 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
8919 %{
8920 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
8922 format %{ "movsbl $dst, $src\t# i2b" %}
8923 opcode(0x0F, 0xBE);
8924 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8925 ins_pipe(ialu_reg_reg);
8926 %}
8928 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
8929 // This idiom is used by the compiler the i2s bytecode.
8930 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
8931 %{
8932 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
8934 format %{ "movswl $dst, $src\t# i2s" %}
8935 opcode(0x0F, 0xBF);
8936 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8937 ins_pipe(ialu_reg_reg);
8938 %}
8940 // ROL/ROR instructions
8942 // ROL expand
8943 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
8944 effect(KILL cr, USE_DEF dst);
8946 format %{ "roll $dst" %}
8947 opcode(0xD1, 0x0); /* Opcode D1 /0 */
8948 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8949 ins_pipe(ialu_reg);
8950 %}
8952 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
8953 effect(USE_DEF dst, USE shift, KILL cr);
8955 format %{ "roll $dst, $shift" %}
8956 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
8957 ins_encode( reg_opc_imm(dst, shift) );
8958 ins_pipe(ialu_reg);
8959 %}
8961 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
8962 %{
8963 effect(USE_DEF dst, USE shift, KILL cr);
8965 format %{ "roll $dst, $shift" %}
8966 opcode(0xD3, 0x0); /* Opcode D3 /0 */
8967 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8968 ins_pipe(ialu_reg_reg);
8969 %}
8970 // end of ROL expand
8972 // Rotate Left by one
8973 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
8974 %{
8975 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8977 expand %{
8978 rolI_rReg_imm1(dst, cr);
8979 %}
8980 %}
8982 // Rotate Left by 8-bit immediate
8983 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
8984 %{
8985 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8986 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8988 expand %{
8989 rolI_rReg_imm8(dst, lshift, cr);
8990 %}
8991 %}
8993 // Rotate Left by variable
8994 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
8995 %{
8996 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
8998 expand %{
8999 rolI_rReg_CL(dst, shift, cr);
9000 %}
9001 %}
9003 // Rotate Left by variable
9004 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9005 %{
9006 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9008 expand %{
9009 rolI_rReg_CL(dst, shift, cr);
9010 %}
9011 %}
9013 // ROR expand
9014 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9015 %{
9016 effect(USE_DEF dst, KILL cr);
9018 format %{ "rorl $dst" %}
9019 opcode(0xD1, 0x1); /* D1 /1 */
9020 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9021 ins_pipe(ialu_reg);
9022 %}
9024 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9025 %{
9026 effect(USE_DEF dst, USE shift, KILL cr);
9028 format %{ "rorl $dst, $shift" %}
9029 opcode(0xC1, 0x1); /* C1 /1 ib */
9030 ins_encode(reg_opc_imm(dst, shift));
9031 ins_pipe(ialu_reg);
9032 %}
9034 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9035 %{
9036 effect(USE_DEF dst, USE shift, KILL cr);
9038 format %{ "rorl $dst, $shift" %}
9039 opcode(0xD3, 0x1); /* D3 /1 */
9040 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9041 ins_pipe(ialu_reg_reg);
9042 %}
9043 // end of ROR expand
9045 // Rotate Right by one
9046 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9047 %{
9048 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9050 expand %{
9051 rorI_rReg_imm1(dst, cr);
9052 %}
9053 %}
9055 // Rotate Right by 8-bit immediate
9056 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9057 %{
9058 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9059 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9061 expand %{
9062 rorI_rReg_imm8(dst, rshift, cr);
9063 %}
9064 %}
9066 // Rotate Right by variable
9067 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9068 %{
9069 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9071 expand %{
9072 rorI_rReg_CL(dst, shift, cr);
9073 %}
9074 %}
9076 // Rotate Right by variable
9077 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9078 %{
9079 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9081 expand %{
9082 rorI_rReg_CL(dst, shift, cr);
9083 %}
9084 %}
9086 // for long rotate
9087 // ROL expand
9088 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9089 effect(USE_DEF dst, KILL cr);
9091 format %{ "rolq $dst" %}
9092 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9093 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9094 ins_pipe(ialu_reg);
9095 %}
9097 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9098 effect(USE_DEF dst, USE shift, KILL cr);
9100 format %{ "rolq $dst, $shift" %}
9101 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9102 ins_encode( reg_opc_imm_wide(dst, shift) );
9103 ins_pipe(ialu_reg);
9104 %}
9106 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9107 %{
9108 effect(USE_DEF dst, USE shift, KILL cr);
9110 format %{ "rolq $dst, $shift" %}
9111 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9112 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9113 ins_pipe(ialu_reg_reg);
9114 %}
9115 // end of ROL expand
9117 // Rotate Left by one
9118 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9119 %{
9120 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9122 expand %{
9123 rolL_rReg_imm1(dst, cr);
9124 %}
9125 %}
9127 // Rotate Left by 8-bit immediate
9128 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9129 %{
9130 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9131 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9133 expand %{
9134 rolL_rReg_imm8(dst, lshift, cr);
9135 %}
9136 %}
9138 // Rotate Left by variable
9139 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9140 %{
9141 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9143 expand %{
9144 rolL_rReg_CL(dst, shift, cr);
9145 %}
9146 %}
9148 // Rotate Left by variable
9149 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9150 %{
9151 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9153 expand %{
9154 rolL_rReg_CL(dst, shift, cr);
9155 %}
9156 %}
9158 // ROR expand
9159 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9160 %{
9161 effect(USE_DEF dst, KILL cr);
9163 format %{ "rorq $dst" %}
9164 opcode(0xD1, 0x1); /* D1 /1 */
9165 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9166 ins_pipe(ialu_reg);
9167 %}
9169 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9170 %{
9171 effect(USE_DEF dst, USE shift, KILL cr);
9173 format %{ "rorq $dst, $shift" %}
9174 opcode(0xC1, 0x1); /* C1 /1 ib */
9175 ins_encode(reg_opc_imm_wide(dst, shift));
9176 ins_pipe(ialu_reg);
9177 %}
9179 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9180 %{
9181 effect(USE_DEF dst, USE shift, KILL cr);
9183 format %{ "rorq $dst, $shift" %}
9184 opcode(0xD3, 0x1); /* D3 /1 */
9185 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9186 ins_pipe(ialu_reg_reg);
9187 %}
9188 // end of ROR expand
9190 // Rotate Right by one
9191 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9192 %{
9193 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9195 expand %{
9196 rorL_rReg_imm1(dst, cr);
9197 %}
9198 %}
9200 // Rotate Right by 8-bit immediate
9201 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9202 %{
9203 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9204 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9206 expand %{
9207 rorL_rReg_imm8(dst, rshift, cr);
9208 %}
9209 %}
9211 // Rotate Right by variable
9212 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9213 %{
9214 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9216 expand %{
9217 rorL_rReg_CL(dst, shift, cr);
9218 %}
9219 %}
9221 // Rotate Right by variable
9222 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9223 %{
9224 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9226 expand %{
9227 rorL_rReg_CL(dst, shift, cr);
9228 %}
9229 %}
9231 // Logical Instructions
9233 // Integer Logical Instructions
9235 // And Instructions
9236 // And Register with Register
9237 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9238 %{
9239 match(Set dst (AndI dst src));
9240 effect(KILL cr);
9242 format %{ "andl $dst, $src\t# int" %}
9243 opcode(0x23);
9244 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9245 ins_pipe(ialu_reg_reg);
9246 %}
9248 // And Register with Immediate 255
9249 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9250 %{
9251 match(Set dst (AndI dst src));
9253 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9254 opcode(0x0F, 0xB6);
9255 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9256 ins_pipe(ialu_reg);
9257 %}
9259 // And Register with Immediate 255 and promote to long
9260 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9261 %{
9262 match(Set dst (ConvI2L (AndI src mask)));
9264 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9265 opcode(0x0F, 0xB6);
9266 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9267 ins_pipe(ialu_reg);
9268 %}
9270 // And Register with Immediate 65535
9271 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9272 %{
9273 match(Set dst (AndI dst src));
9275 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9276 opcode(0x0F, 0xB7);
9277 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9278 ins_pipe(ialu_reg);
9279 %}
9281 // And Register with Immediate 65535 and promote to long
9282 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9283 %{
9284 match(Set dst (ConvI2L (AndI src mask)));
9286 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9287 opcode(0x0F, 0xB7);
9288 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9289 ins_pipe(ialu_reg);
9290 %}
9292 // And Register with Immediate
9293 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9294 %{
9295 match(Set dst (AndI dst src));
9296 effect(KILL cr);
9298 format %{ "andl $dst, $src\t# int" %}
9299 opcode(0x81, 0x04); /* Opcode 81 /4 */
9300 ins_encode(OpcSErm(dst, src), Con8or32(src));
9301 ins_pipe(ialu_reg);
9302 %}
9304 // And Register with Memory
9305 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9306 %{
9307 match(Set dst (AndI dst (LoadI src)));
9308 effect(KILL cr);
9310 ins_cost(125);
9311 format %{ "andl $dst, $src\t# int" %}
9312 opcode(0x23);
9313 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9314 ins_pipe(ialu_reg_mem);
9315 %}
9317 // And Memory with Register
9318 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9319 %{
9320 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9321 effect(KILL cr);
9323 ins_cost(150);
9324 format %{ "andl $dst, $src\t# int" %}
9325 opcode(0x21); /* Opcode 21 /r */
9326 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9327 ins_pipe(ialu_mem_reg);
9328 %}
9330 // And Memory with Immediate
9331 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9332 %{
9333 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9334 effect(KILL cr);
9336 ins_cost(125);
9337 format %{ "andl $dst, $src\t# int" %}
9338 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9339 ins_encode(REX_mem(dst), OpcSE(src),
9340 RM_opc_mem(secondary, dst), Con8or32(src));
9341 ins_pipe(ialu_mem_imm);
9342 %}
9344 // Or Instructions
9345 // Or Register with Register
9346 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9347 %{
9348 match(Set dst (OrI dst src));
9349 effect(KILL cr);
9351 format %{ "orl $dst, $src\t# int" %}
9352 opcode(0x0B);
9353 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9354 ins_pipe(ialu_reg_reg);
9355 %}
9357 // Or Register with Immediate
9358 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9359 %{
9360 match(Set dst (OrI dst src));
9361 effect(KILL cr);
9363 format %{ "orl $dst, $src\t# int" %}
9364 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9365 ins_encode(OpcSErm(dst, src), Con8or32(src));
9366 ins_pipe(ialu_reg);
9367 %}
9369 // Or Register with Memory
9370 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9371 %{
9372 match(Set dst (OrI dst (LoadI src)));
9373 effect(KILL cr);
9375 ins_cost(125);
9376 format %{ "orl $dst, $src\t# int" %}
9377 opcode(0x0B);
9378 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9379 ins_pipe(ialu_reg_mem);
9380 %}
9382 // Or Memory with Register
9383 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9384 %{
9385 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9386 effect(KILL cr);
9388 ins_cost(150);
9389 format %{ "orl $dst, $src\t# int" %}
9390 opcode(0x09); /* Opcode 09 /r */
9391 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9392 ins_pipe(ialu_mem_reg);
9393 %}
9395 // Or Memory with Immediate
9396 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9397 %{
9398 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9399 effect(KILL cr);
9401 ins_cost(125);
9402 format %{ "orl $dst, $src\t# int" %}
9403 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9404 ins_encode(REX_mem(dst), OpcSE(src),
9405 RM_opc_mem(secondary, dst), Con8or32(src));
9406 ins_pipe(ialu_mem_imm);
9407 %}
9409 // Xor Instructions
9410 // Xor Register with Register
9411 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9412 %{
9413 match(Set dst (XorI dst src));
9414 effect(KILL cr);
9416 format %{ "xorl $dst, $src\t# int" %}
9417 opcode(0x33);
9418 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9419 ins_pipe(ialu_reg_reg);
9420 %}
9422 // Xor Register with Immediate -1
9423 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9424 match(Set dst (XorI dst imm));
9426 format %{ "not $dst" %}
9427 ins_encode %{
9428 __ notl($dst$$Register);
9429 %}
9430 ins_pipe(ialu_reg);
9431 %}
9433 // Xor Register with Immediate
9434 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9435 %{
9436 match(Set dst (XorI dst src));
9437 effect(KILL cr);
9439 format %{ "xorl $dst, $src\t# int" %}
9440 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9441 ins_encode(OpcSErm(dst, src), Con8or32(src));
9442 ins_pipe(ialu_reg);
9443 %}
9445 // Xor Register with Memory
9446 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9447 %{
9448 match(Set dst (XorI dst (LoadI src)));
9449 effect(KILL cr);
9451 ins_cost(125);
9452 format %{ "xorl $dst, $src\t# int" %}
9453 opcode(0x33);
9454 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9455 ins_pipe(ialu_reg_mem);
9456 %}
9458 // Xor Memory with Register
9459 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9460 %{
9461 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9462 effect(KILL cr);
9464 ins_cost(150);
9465 format %{ "xorl $dst, $src\t# int" %}
9466 opcode(0x31); /* Opcode 31 /r */
9467 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9468 ins_pipe(ialu_mem_reg);
9469 %}
9471 // Xor Memory with Immediate
9472 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9473 %{
9474 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9475 effect(KILL cr);
9477 ins_cost(125);
9478 format %{ "xorl $dst, $src\t# int" %}
9479 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9480 ins_encode(REX_mem(dst), OpcSE(src),
9481 RM_opc_mem(secondary, dst), Con8or32(src));
9482 ins_pipe(ialu_mem_imm);
9483 %}
9486 // Long Logical Instructions
9488 // And Instructions
9489 // And Register with Register
9490 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9491 %{
9492 match(Set dst (AndL dst src));
9493 effect(KILL cr);
9495 format %{ "andq $dst, $src\t# long" %}
9496 opcode(0x23);
9497 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9498 ins_pipe(ialu_reg_reg);
9499 %}
9501 // And Register with Immediate 255
9502 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9503 %{
9504 match(Set dst (AndL dst src));
9506 format %{ "movzbq $dst, $src\t# long & 0xFF" %}
9507 opcode(0x0F, 0xB6);
9508 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9509 ins_pipe(ialu_reg);
9510 %}
9512 // And Register with Immediate 65535
9513 instruct andL_rReg_imm65535(rRegI dst, immL_65535 src)
9514 %{
9515 match(Set dst (AndL dst src));
9517 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
9518 opcode(0x0F, 0xB7);
9519 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9520 ins_pipe(ialu_reg);
9521 %}
9523 // And Register with Immediate
9524 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9525 %{
9526 match(Set dst (AndL dst src));
9527 effect(KILL cr);
9529 format %{ "andq $dst, $src\t# long" %}
9530 opcode(0x81, 0x04); /* Opcode 81 /4 */
9531 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9532 ins_pipe(ialu_reg);
9533 %}
9535 // And Register with Memory
9536 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9537 %{
9538 match(Set dst (AndL dst (LoadL src)));
9539 effect(KILL cr);
9541 ins_cost(125);
9542 format %{ "andq $dst, $src\t# long" %}
9543 opcode(0x23);
9544 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9545 ins_pipe(ialu_reg_mem);
9546 %}
9548 // And Memory with Register
9549 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9550 %{
9551 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9552 effect(KILL cr);
9554 ins_cost(150);
9555 format %{ "andq $dst, $src\t# long" %}
9556 opcode(0x21); /* Opcode 21 /r */
9557 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9558 ins_pipe(ialu_mem_reg);
9559 %}
9561 // And Memory with Immediate
9562 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9563 %{
9564 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9565 effect(KILL cr);
9567 ins_cost(125);
9568 format %{ "andq $dst, $src\t# long" %}
9569 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9570 ins_encode(REX_mem_wide(dst), OpcSE(src),
9571 RM_opc_mem(secondary, dst), Con8or32(src));
9572 ins_pipe(ialu_mem_imm);
9573 %}
9575 // Or Instructions
9576 // Or Register with Register
9577 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9578 %{
9579 match(Set dst (OrL dst src));
9580 effect(KILL cr);
9582 format %{ "orq $dst, $src\t# long" %}
9583 opcode(0x0B);
9584 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9585 ins_pipe(ialu_reg_reg);
9586 %}
9588 // Or Register with Immediate
9589 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9590 %{
9591 match(Set dst (OrL dst src));
9592 effect(KILL cr);
9594 format %{ "orq $dst, $src\t# long" %}
9595 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9596 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9597 ins_pipe(ialu_reg);
9598 %}
9600 // Or Register with Memory
9601 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9602 %{
9603 match(Set dst (OrL dst (LoadL src)));
9604 effect(KILL cr);
9606 ins_cost(125);
9607 format %{ "orq $dst, $src\t# long" %}
9608 opcode(0x0B);
9609 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9610 ins_pipe(ialu_reg_mem);
9611 %}
9613 // Or Memory with Register
9614 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9615 %{
9616 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9617 effect(KILL cr);
9619 ins_cost(150);
9620 format %{ "orq $dst, $src\t# long" %}
9621 opcode(0x09); /* Opcode 09 /r */
9622 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9623 ins_pipe(ialu_mem_reg);
9624 %}
9626 // Or Memory with Immediate
9627 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9628 %{
9629 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9630 effect(KILL cr);
9632 ins_cost(125);
9633 format %{ "orq $dst, $src\t# long" %}
9634 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9635 ins_encode(REX_mem_wide(dst), OpcSE(src),
9636 RM_opc_mem(secondary, dst), Con8or32(src));
9637 ins_pipe(ialu_mem_imm);
9638 %}
9640 // Xor Instructions
9641 // Xor Register with Register
9642 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9643 %{
9644 match(Set dst (XorL dst src));
9645 effect(KILL cr);
9647 format %{ "xorq $dst, $src\t# long" %}
9648 opcode(0x33);
9649 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9650 ins_pipe(ialu_reg_reg);
9651 %}
9653 // Xor Register with Immediate -1
9654 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
9655 match(Set dst (XorL dst imm));
9657 format %{ "notq $dst" %}
9658 ins_encode %{
9659 __ notq($dst$$Register);
9660 %}
9661 ins_pipe(ialu_reg);
9662 %}
9664 // Xor Register with Immediate
9665 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9666 %{
9667 match(Set dst (XorL dst src));
9668 effect(KILL cr);
9670 format %{ "xorq $dst, $src\t# long" %}
9671 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9672 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9673 ins_pipe(ialu_reg);
9674 %}
9676 // Xor Register with Memory
9677 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9678 %{
9679 match(Set dst (XorL dst (LoadL src)));
9680 effect(KILL cr);
9682 ins_cost(125);
9683 format %{ "xorq $dst, $src\t# long" %}
9684 opcode(0x33);
9685 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9686 ins_pipe(ialu_reg_mem);
9687 %}
9689 // Xor Memory with Register
9690 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9691 %{
9692 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9693 effect(KILL cr);
9695 ins_cost(150);
9696 format %{ "xorq $dst, $src\t# long" %}
9697 opcode(0x31); /* Opcode 31 /r */
9698 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9699 ins_pipe(ialu_mem_reg);
9700 %}
9702 // Xor Memory with Immediate
9703 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9704 %{
9705 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9706 effect(KILL cr);
9708 ins_cost(125);
9709 format %{ "xorq $dst, $src\t# long" %}
9710 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9711 ins_encode(REX_mem_wide(dst), OpcSE(src),
9712 RM_opc_mem(secondary, dst), Con8or32(src));
9713 ins_pipe(ialu_mem_imm);
9714 %}
9716 // Convert Int to Boolean
9717 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
9718 %{
9719 match(Set dst (Conv2B src));
9720 effect(KILL cr);
9722 format %{ "testl $src, $src\t# ci2b\n\t"
9723 "setnz $dst\n\t"
9724 "movzbl $dst, $dst" %}
9725 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
9726 setNZ_reg(dst),
9727 REX_reg_breg(dst, dst), // movzbl
9728 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
9729 ins_pipe(pipe_slow); // XXX
9730 %}
9732 // Convert Pointer to Boolean
9733 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
9734 %{
9735 match(Set dst (Conv2B src));
9736 effect(KILL cr);
9738 format %{ "testq $src, $src\t# cp2b\n\t"
9739 "setnz $dst\n\t"
9740 "movzbl $dst, $dst" %}
9741 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
9742 setNZ_reg(dst),
9743 REX_reg_breg(dst, dst), // movzbl
9744 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
9745 ins_pipe(pipe_slow); // XXX
9746 %}
9748 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
9749 %{
9750 match(Set dst (CmpLTMask p q));
9751 effect(KILL cr);
9753 ins_cost(400); // XXX
9754 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
9755 "setlt $dst\n\t"
9756 "movzbl $dst, $dst\n\t"
9757 "negl $dst" %}
9758 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
9759 setLT_reg(dst),
9760 REX_reg_breg(dst, dst), // movzbl
9761 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
9762 neg_reg(dst));
9763 ins_pipe(pipe_slow);
9764 %}
9766 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
9767 %{
9768 match(Set dst (CmpLTMask dst zero));
9769 effect(KILL cr);
9771 ins_cost(100); // XXX
9772 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
9773 opcode(0xC1, 0x7); /* C1 /7 ib */
9774 ins_encode(reg_opc_imm(dst, 0x1F));
9775 ins_pipe(ialu_reg);
9776 %}
9779 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
9780 rRegI tmp,
9781 rFlagsReg cr)
9782 %{
9783 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
9784 effect(TEMP tmp, KILL cr);
9786 ins_cost(400); // XXX
9787 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
9788 "sbbl $tmp, $tmp\n\t"
9789 "andl $tmp, $y\n\t"
9790 "addl $p, $tmp" %}
9791 ins_encode(enc_cmpLTP(p, q, y, tmp));
9792 ins_pipe(pipe_cmplt);
9793 %}
9795 /* If I enable this, I encourage spilling in the inner loop of compress.
9796 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
9797 %{
9798 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
9799 effect( TEMP tmp, KILL cr );
9800 ins_cost(400);
9802 format %{ "SUB $p,$q\n\t"
9803 "SBB RCX,RCX\n\t"
9804 "AND RCX,$y\n\t"
9805 "ADD $p,RCX" %}
9806 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
9807 %}
9808 */
9810 //---------- FP Instructions------------------------------------------------
9812 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
9813 %{
9814 match(Set cr (CmpF src1 src2));
9816 ins_cost(145);
9817 format %{ "ucomiss $src1, $src2\n\t"
9818 "jnp,s exit\n\t"
9819 "pushfq\t# saw NaN, set CF\n\t"
9820 "andq [rsp], #0xffffff2b\n\t"
9821 "popfq\n"
9822 "exit: nop\t# avoid branch to branch" %}
9823 opcode(0x0F, 0x2E);
9824 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
9825 cmpfp_fixup);
9826 ins_pipe(pipe_slow);
9827 %}
9829 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
9830 match(Set cr (CmpF src1 src2));
9832 ins_cost(145);
9833 format %{ "ucomiss $src1, $src2" %}
9834 ins_encode %{
9835 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
9836 %}
9837 ins_pipe(pipe_slow);
9838 %}
9840 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
9841 %{
9842 match(Set cr (CmpF src1 (LoadF src2)));
9844 ins_cost(145);
9845 format %{ "ucomiss $src1, $src2\n\t"
9846 "jnp,s exit\n\t"
9847 "pushfq\t# saw NaN, set CF\n\t"
9848 "andq [rsp], #0xffffff2b\n\t"
9849 "popfq\n"
9850 "exit: nop\t# avoid branch to branch" %}
9851 opcode(0x0F, 0x2E);
9852 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
9853 cmpfp_fixup);
9854 ins_pipe(pipe_slow);
9855 %}
9857 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
9858 match(Set cr (CmpF src1 (LoadF src2)));
9860 ins_cost(100);
9861 format %{ "ucomiss $src1, $src2" %}
9862 opcode(0x0F, 0x2E);
9863 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
9864 ins_pipe(pipe_slow);
9865 %}
9867 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
9868 %{
9869 match(Set cr (CmpF src1 src2));
9871 ins_cost(145);
9872 format %{ "ucomiss $src1, $src2\n\t"
9873 "jnp,s exit\n\t"
9874 "pushfq\t# saw NaN, set CF\n\t"
9875 "andq [rsp], #0xffffff2b\n\t"
9876 "popfq\n"
9877 "exit: nop\t# avoid branch to branch" %}
9878 opcode(0x0F, 0x2E);
9879 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
9880 cmpfp_fixup);
9881 ins_pipe(pipe_slow);
9882 %}
9884 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{
9885 match(Set cr (CmpF src1 src2));
9887 ins_cost(100);
9888 format %{ "ucomiss $src1, $src2" %}
9889 opcode(0x0F, 0x2E);
9890 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2));
9891 ins_pipe(pipe_slow);
9892 %}
9894 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
9895 %{
9896 match(Set cr (CmpD src1 src2));
9898 ins_cost(145);
9899 format %{ "ucomisd $src1, $src2\n\t"
9900 "jnp,s exit\n\t"
9901 "pushfq\t# saw NaN, set CF\n\t"
9902 "andq [rsp], #0xffffff2b\n\t"
9903 "popfq\n"
9904 "exit: nop\t# avoid branch to branch" %}
9905 opcode(0x66, 0x0F, 0x2E);
9906 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
9907 cmpfp_fixup);
9908 ins_pipe(pipe_slow);
9909 %}
9911 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
9912 match(Set cr (CmpD src1 src2));
9914 ins_cost(100);
9915 format %{ "ucomisd $src1, $src2 test" %}
9916 ins_encode %{
9917 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9918 %}
9919 ins_pipe(pipe_slow);
9920 %}
9922 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
9923 %{
9924 match(Set cr (CmpD src1 (LoadD src2)));
9926 ins_cost(145);
9927 format %{ "ucomisd $src1, $src2\n\t"
9928 "jnp,s exit\n\t"
9929 "pushfq\t# saw NaN, set CF\n\t"
9930 "andq [rsp], #0xffffff2b\n\t"
9931 "popfq\n"
9932 "exit: nop\t# avoid branch to branch" %}
9933 opcode(0x66, 0x0F, 0x2E);
9934 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
9935 cmpfp_fixup);
9936 ins_pipe(pipe_slow);
9937 %}
9939 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
9940 match(Set cr (CmpD src1 (LoadD src2)));
9942 ins_cost(100);
9943 format %{ "ucomisd $src1, $src2" %}
9944 opcode(0x66, 0x0F, 0x2E);
9945 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
9946 ins_pipe(pipe_slow);
9947 %}
9949 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
9950 %{
9951 match(Set cr (CmpD src1 src2));
9953 ins_cost(145);
9954 format %{ "ucomisd $src1, [$src2]\n\t"
9955 "jnp,s exit\n\t"
9956 "pushfq\t# saw NaN, set CF\n\t"
9957 "andq [rsp], #0xffffff2b\n\t"
9958 "popfq\n"
9959 "exit: nop\t# avoid branch to branch" %}
9960 opcode(0x66, 0x0F, 0x2E);
9961 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
9962 cmpfp_fixup);
9963 ins_pipe(pipe_slow);
9964 %}
9966 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{
9967 match(Set cr (CmpD src1 src2));
9969 ins_cost(100);
9970 format %{ "ucomisd $src1, [$src2]" %}
9971 opcode(0x66, 0x0F, 0x2E);
9972 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2));
9973 ins_pipe(pipe_slow);
9974 %}
9976 // Compare into -1,0,1
9977 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
9978 %{
9979 match(Set dst (CmpF3 src1 src2));
9980 effect(KILL cr);
9982 ins_cost(275);
9983 format %{ "ucomiss $src1, $src2\n\t"
9984 "movl $dst, #-1\n\t"
9985 "jp,s done\n\t"
9986 "jb,s done\n\t"
9987 "setne $dst\n\t"
9988 "movzbl $dst, $dst\n"
9989 "done:" %}
9991 opcode(0x0F, 0x2E);
9992 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
9993 cmpfp3(dst));
9994 ins_pipe(pipe_slow);
9995 %}
9997 // Compare into -1,0,1
9998 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
9999 %{
10000 match(Set dst (CmpF3 src1 (LoadF src2)));
10001 effect(KILL cr);
10003 ins_cost(275);
10004 format %{ "ucomiss $src1, $src2\n\t"
10005 "movl $dst, #-1\n\t"
10006 "jp,s done\n\t"
10007 "jb,s done\n\t"
10008 "setne $dst\n\t"
10009 "movzbl $dst, $dst\n"
10010 "done:" %}
10012 opcode(0x0F, 0x2E);
10013 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10014 cmpfp3(dst));
10015 ins_pipe(pipe_slow);
10016 %}
10018 // Compare into -1,0,1
10019 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
10020 %{
10021 match(Set dst (CmpF3 src1 src2));
10022 effect(KILL cr);
10024 ins_cost(275);
10025 format %{ "ucomiss $src1, [$src2]\n\t"
10026 "movl $dst, #-1\n\t"
10027 "jp,s done\n\t"
10028 "jb,s done\n\t"
10029 "setne $dst\n\t"
10030 "movzbl $dst, $dst\n"
10031 "done:" %}
10033 opcode(0x0F, 0x2E);
10034 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10035 cmpfp3(dst));
10036 ins_pipe(pipe_slow);
10037 %}
10039 // Compare into -1,0,1
10040 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10041 %{
10042 match(Set dst (CmpD3 src1 src2));
10043 effect(KILL cr);
10045 ins_cost(275);
10046 format %{ "ucomisd $src1, $src2\n\t"
10047 "movl $dst, #-1\n\t"
10048 "jp,s done\n\t"
10049 "jb,s done\n\t"
10050 "setne $dst\n\t"
10051 "movzbl $dst, $dst\n"
10052 "done:" %}
10054 opcode(0x66, 0x0F, 0x2E);
10055 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10056 cmpfp3(dst));
10057 ins_pipe(pipe_slow);
10058 %}
10060 // Compare into -1,0,1
10061 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10062 %{
10063 match(Set dst (CmpD3 src1 (LoadD src2)));
10064 effect(KILL cr);
10066 ins_cost(275);
10067 format %{ "ucomisd $src1, $src2\n\t"
10068 "movl $dst, #-1\n\t"
10069 "jp,s done\n\t"
10070 "jb,s done\n\t"
10071 "setne $dst\n\t"
10072 "movzbl $dst, $dst\n"
10073 "done:" %}
10075 opcode(0x66, 0x0F, 0x2E);
10076 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10077 cmpfp3(dst));
10078 ins_pipe(pipe_slow);
10079 %}
10081 // Compare into -1,0,1
10082 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
10083 %{
10084 match(Set dst (CmpD3 src1 src2));
10085 effect(KILL cr);
10087 ins_cost(275);
10088 format %{ "ucomisd $src1, [$src2]\n\t"
10089 "movl $dst, #-1\n\t"
10090 "jp,s done\n\t"
10091 "jb,s done\n\t"
10092 "setne $dst\n\t"
10093 "movzbl $dst, $dst\n"
10094 "done:" %}
10096 opcode(0x66, 0x0F, 0x2E);
10097 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10098 cmpfp3(dst));
10099 ins_pipe(pipe_slow);
10100 %}
10102 instruct addF_reg(regF dst, regF src)
10103 %{
10104 match(Set dst (AddF dst src));
10106 format %{ "addss $dst, $src" %}
10107 ins_cost(150); // XXX
10108 opcode(0xF3, 0x0F, 0x58);
10109 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10110 ins_pipe(pipe_slow);
10111 %}
10113 instruct addF_mem(regF dst, memory src)
10114 %{
10115 match(Set dst (AddF dst (LoadF src)));
10117 format %{ "addss $dst, $src" %}
10118 ins_cost(150); // XXX
10119 opcode(0xF3, 0x0F, 0x58);
10120 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10121 ins_pipe(pipe_slow);
10122 %}
10124 instruct addF_imm(regF dst, immF src)
10125 %{
10126 match(Set dst (AddF dst src));
10128 format %{ "addss $dst, [$src]" %}
10129 ins_cost(150); // XXX
10130 opcode(0xF3, 0x0F, 0x58);
10131 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10132 ins_pipe(pipe_slow);
10133 %}
10135 instruct addD_reg(regD dst, regD src)
10136 %{
10137 match(Set dst (AddD dst src));
10139 format %{ "addsd $dst, $src" %}
10140 ins_cost(150); // XXX
10141 opcode(0xF2, 0x0F, 0x58);
10142 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10143 ins_pipe(pipe_slow);
10144 %}
10146 instruct addD_mem(regD dst, memory src)
10147 %{
10148 match(Set dst (AddD dst (LoadD src)));
10150 format %{ "addsd $dst, $src" %}
10151 ins_cost(150); // XXX
10152 opcode(0xF2, 0x0F, 0x58);
10153 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10154 ins_pipe(pipe_slow);
10155 %}
10157 instruct addD_imm(regD dst, immD src)
10158 %{
10159 match(Set dst (AddD dst src));
10161 format %{ "addsd $dst, [$src]" %}
10162 ins_cost(150); // XXX
10163 opcode(0xF2, 0x0F, 0x58);
10164 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10165 ins_pipe(pipe_slow);
10166 %}
10168 instruct subF_reg(regF dst, regF src)
10169 %{
10170 match(Set dst (SubF dst src));
10172 format %{ "subss $dst, $src" %}
10173 ins_cost(150); // XXX
10174 opcode(0xF3, 0x0F, 0x5C);
10175 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10176 ins_pipe(pipe_slow);
10177 %}
10179 instruct subF_mem(regF dst, memory src)
10180 %{
10181 match(Set dst (SubF dst (LoadF src)));
10183 format %{ "subss $dst, $src" %}
10184 ins_cost(150); // XXX
10185 opcode(0xF3, 0x0F, 0x5C);
10186 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10187 ins_pipe(pipe_slow);
10188 %}
10190 instruct subF_imm(regF dst, immF src)
10191 %{
10192 match(Set dst (SubF dst src));
10194 format %{ "subss $dst, [$src]" %}
10195 ins_cost(150); // XXX
10196 opcode(0xF3, 0x0F, 0x5C);
10197 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10198 ins_pipe(pipe_slow);
10199 %}
10201 instruct subD_reg(regD dst, regD src)
10202 %{
10203 match(Set dst (SubD dst src));
10205 format %{ "subsd $dst, $src" %}
10206 ins_cost(150); // XXX
10207 opcode(0xF2, 0x0F, 0x5C);
10208 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10209 ins_pipe(pipe_slow);
10210 %}
10212 instruct subD_mem(regD dst, memory src)
10213 %{
10214 match(Set dst (SubD dst (LoadD src)));
10216 format %{ "subsd $dst, $src" %}
10217 ins_cost(150); // XXX
10218 opcode(0xF2, 0x0F, 0x5C);
10219 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10220 ins_pipe(pipe_slow);
10221 %}
10223 instruct subD_imm(regD dst, immD src)
10224 %{
10225 match(Set dst (SubD dst src));
10227 format %{ "subsd $dst, [$src]" %}
10228 ins_cost(150); // XXX
10229 opcode(0xF2, 0x0F, 0x5C);
10230 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10231 ins_pipe(pipe_slow);
10232 %}
10234 instruct mulF_reg(regF dst, regF src)
10235 %{
10236 match(Set dst (MulF dst src));
10238 format %{ "mulss $dst, $src" %}
10239 ins_cost(150); // XXX
10240 opcode(0xF3, 0x0F, 0x59);
10241 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10242 ins_pipe(pipe_slow);
10243 %}
10245 instruct mulF_mem(regF dst, memory src)
10246 %{
10247 match(Set dst (MulF dst (LoadF src)));
10249 format %{ "mulss $dst, $src" %}
10250 ins_cost(150); // XXX
10251 opcode(0xF3, 0x0F, 0x59);
10252 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10253 ins_pipe(pipe_slow);
10254 %}
10256 instruct mulF_imm(regF dst, immF src)
10257 %{
10258 match(Set dst (MulF dst src));
10260 format %{ "mulss $dst, [$src]" %}
10261 ins_cost(150); // XXX
10262 opcode(0xF3, 0x0F, 0x59);
10263 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10264 ins_pipe(pipe_slow);
10265 %}
10267 instruct mulD_reg(regD dst, regD src)
10268 %{
10269 match(Set dst (MulD dst src));
10271 format %{ "mulsd $dst, $src" %}
10272 ins_cost(150); // XXX
10273 opcode(0xF2, 0x0F, 0x59);
10274 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10275 ins_pipe(pipe_slow);
10276 %}
10278 instruct mulD_mem(regD dst, memory src)
10279 %{
10280 match(Set dst (MulD dst (LoadD src)));
10282 format %{ "mulsd $dst, $src" %}
10283 ins_cost(150); // XXX
10284 opcode(0xF2, 0x0F, 0x59);
10285 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10286 ins_pipe(pipe_slow);
10287 %}
10289 instruct mulD_imm(regD dst, immD src)
10290 %{
10291 match(Set dst (MulD dst src));
10293 format %{ "mulsd $dst, [$src]" %}
10294 ins_cost(150); // XXX
10295 opcode(0xF2, 0x0F, 0x59);
10296 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10297 ins_pipe(pipe_slow);
10298 %}
10300 instruct divF_reg(regF dst, regF src)
10301 %{
10302 match(Set dst (DivF dst src));
10304 format %{ "divss $dst, $src" %}
10305 ins_cost(150); // XXX
10306 opcode(0xF3, 0x0F, 0x5E);
10307 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10308 ins_pipe(pipe_slow);
10309 %}
10311 instruct divF_mem(regF dst, memory src)
10312 %{
10313 match(Set dst (DivF dst (LoadF src)));
10315 format %{ "divss $dst, $src" %}
10316 ins_cost(150); // XXX
10317 opcode(0xF3, 0x0F, 0x5E);
10318 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10319 ins_pipe(pipe_slow);
10320 %}
10322 instruct divF_imm(regF dst, immF src)
10323 %{
10324 match(Set dst (DivF dst src));
10326 format %{ "divss $dst, [$src]" %}
10327 ins_cost(150); // XXX
10328 opcode(0xF3, 0x0F, 0x5E);
10329 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10330 ins_pipe(pipe_slow);
10331 %}
10333 instruct divD_reg(regD dst, regD src)
10334 %{
10335 match(Set dst (DivD dst src));
10337 format %{ "divsd $dst, $src" %}
10338 ins_cost(150); // XXX
10339 opcode(0xF2, 0x0F, 0x5E);
10340 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10341 ins_pipe(pipe_slow);
10342 %}
10344 instruct divD_mem(regD dst, memory src)
10345 %{
10346 match(Set dst (DivD dst (LoadD src)));
10348 format %{ "divsd $dst, $src" %}
10349 ins_cost(150); // XXX
10350 opcode(0xF2, 0x0F, 0x5E);
10351 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10352 ins_pipe(pipe_slow);
10353 %}
10355 instruct divD_imm(regD dst, immD src)
10356 %{
10357 match(Set dst (DivD dst src));
10359 format %{ "divsd $dst, [$src]" %}
10360 ins_cost(150); // XXX
10361 opcode(0xF2, 0x0F, 0x5E);
10362 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10363 ins_pipe(pipe_slow);
10364 %}
10366 instruct sqrtF_reg(regF dst, regF src)
10367 %{
10368 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10370 format %{ "sqrtss $dst, $src" %}
10371 ins_cost(150); // XXX
10372 opcode(0xF3, 0x0F, 0x51);
10373 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10374 ins_pipe(pipe_slow);
10375 %}
10377 instruct sqrtF_mem(regF dst, memory src)
10378 %{
10379 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10381 format %{ "sqrtss $dst, $src" %}
10382 ins_cost(150); // XXX
10383 opcode(0xF3, 0x0F, 0x51);
10384 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10385 ins_pipe(pipe_slow);
10386 %}
10388 instruct sqrtF_imm(regF dst, immF src)
10389 %{
10390 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10392 format %{ "sqrtss $dst, [$src]" %}
10393 ins_cost(150); // XXX
10394 opcode(0xF3, 0x0F, 0x51);
10395 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10396 ins_pipe(pipe_slow);
10397 %}
10399 instruct sqrtD_reg(regD dst, regD src)
10400 %{
10401 match(Set dst (SqrtD src));
10403 format %{ "sqrtsd $dst, $src" %}
10404 ins_cost(150); // XXX
10405 opcode(0xF2, 0x0F, 0x51);
10406 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10407 ins_pipe(pipe_slow);
10408 %}
10410 instruct sqrtD_mem(regD dst, memory src)
10411 %{
10412 match(Set dst (SqrtD (LoadD src)));
10414 format %{ "sqrtsd $dst, $src" %}
10415 ins_cost(150); // XXX
10416 opcode(0xF2, 0x0F, 0x51);
10417 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10418 ins_pipe(pipe_slow);
10419 %}
10421 instruct sqrtD_imm(regD dst, immD src)
10422 %{
10423 match(Set dst (SqrtD src));
10425 format %{ "sqrtsd $dst, [$src]" %}
10426 ins_cost(150); // XXX
10427 opcode(0xF2, 0x0F, 0x51);
10428 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10429 ins_pipe(pipe_slow);
10430 %}
10432 instruct absF_reg(regF dst)
10433 %{
10434 match(Set dst (AbsF dst));
10436 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
10437 ins_encode(absF_encoding(dst));
10438 ins_pipe(pipe_slow);
10439 %}
10441 instruct absD_reg(regD dst)
10442 %{
10443 match(Set dst (AbsD dst));
10445 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
10446 "# abs double by sign masking" %}
10447 ins_encode(absD_encoding(dst));
10448 ins_pipe(pipe_slow);
10449 %}
10451 instruct negF_reg(regF dst)
10452 %{
10453 match(Set dst (NegF dst));
10455 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
10456 ins_encode(negF_encoding(dst));
10457 ins_pipe(pipe_slow);
10458 %}
10460 instruct negD_reg(regD dst)
10461 %{
10462 match(Set dst (NegD dst));
10464 format %{ "xorpd $dst, [0x8000000000000000]\t"
10465 "# neg double by sign flipping" %}
10466 ins_encode(negD_encoding(dst));
10467 ins_pipe(pipe_slow);
10468 %}
10470 // -----------Trig and Trancendental Instructions------------------------------
10471 instruct cosD_reg(regD dst) %{
10472 match(Set dst (CosD dst));
10474 format %{ "dcos $dst\n\t" %}
10475 opcode(0xD9, 0xFF);
10476 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10477 ins_pipe( pipe_slow );
10478 %}
10480 instruct sinD_reg(regD dst) %{
10481 match(Set dst (SinD dst));
10483 format %{ "dsin $dst\n\t" %}
10484 opcode(0xD9, 0xFE);
10485 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10486 ins_pipe( pipe_slow );
10487 %}
10489 instruct tanD_reg(regD dst) %{
10490 match(Set dst (TanD dst));
10492 format %{ "dtan $dst\n\t" %}
10493 ins_encode( Push_SrcXD(dst),
10494 Opcode(0xD9), Opcode(0xF2), //fptan
10495 Opcode(0xDD), Opcode(0xD8), //fstp st
10496 Push_ResultXD(dst) );
10497 ins_pipe( pipe_slow );
10498 %}
10500 instruct log10D_reg(regD dst) %{
10501 // The source and result Double operands in XMM registers
10502 match(Set dst (Log10D dst));
10503 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
10504 // fyl2x ; compute log_10(2) * log_2(x)
10505 format %{ "fldlg2\t\t\t#Log10\n\t"
10506 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10507 %}
10508 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
10509 Push_SrcXD(dst),
10510 Opcode(0xD9), Opcode(0xF1), // fyl2x
10511 Push_ResultXD(dst));
10513 ins_pipe( pipe_slow );
10514 %}
10516 instruct logD_reg(regD dst) %{
10517 // The source and result Double operands in XMM registers
10518 match(Set dst (LogD dst));
10519 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
10520 // fyl2x ; compute log_e(2) * log_2(x)
10521 format %{ "fldln2\t\t\t#Log_e\n\t"
10522 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10523 %}
10524 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
10525 Push_SrcXD(dst),
10526 Opcode(0xD9), Opcode(0xF1), // fyl2x
10527 Push_ResultXD(dst));
10528 ins_pipe( pipe_slow );
10529 %}
10533 //----------Arithmetic Conversion Instructions---------------------------------
10535 instruct roundFloat_nop(regF dst)
10536 %{
10537 match(Set dst (RoundFloat dst));
10539 ins_cost(0);
10540 ins_encode();
10541 ins_pipe(empty);
10542 %}
10544 instruct roundDouble_nop(regD dst)
10545 %{
10546 match(Set dst (RoundDouble dst));
10548 ins_cost(0);
10549 ins_encode();
10550 ins_pipe(empty);
10551 %}
10553 instruct convF2D_reg_reg(regD dst, regF src)
10554 %{
10555 match(Set dst (ConvF2D src));
10557 format %{ "cvtss2sd $dst, $src" %}
10558 opcode(0xF3, 0x0F, 0x5A);
10559 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10560 ins_pipe(pipe_slow); // XXX
10561 %}
10563 instruct convF2D_reg_mem(regD dst, memory src)
10564 %{
10565 match(Set dst (ConvF2D (LoadF src)));
10567 format %{ "cvtss2sd $dst, $src" %}
10568 opcode(0xF3, 0x0F, 0x5A);
10569 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10570 ins_pipe(pipe_slow); // XXX
10571 %}
10573 instruct convD2F_reg_reg(regF dst, regD src)
10574 %{
10575 match(Set dst (ConvD2F src));
10577 format %{ "cvtsd2ss $dst, $src" %}
10578 opcode(0xF2, 0x0F, 0x5A);
10579 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10580 ins_pipe(pipe_slow); // XXX
10581 %}
10583 instruct convD2F_reg_mem(regF dst, memory src)
10584 %{
10585 match(Set dst (ConvD2F (LoadD src)));
10587 format %{ "cvtsd2ss $dst, $src" %}
10588 opcode(0xF2, 0x0F, 0x5A);
10589 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10590 ins_pipe(pipe_slow); // XXX
10591 %}
10593 // XXX do mem variants
10594 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
10595 %{
10596 match(Set dst (ConvF2I src));
10597 effect(KILL cr);
10599 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
10600 "cmpl $dst, #0x80000000\n\t"
10601 "jne,s done\n\t"
10602 "subq rsp, #8\n\t"
10603 "movss [rsp], $src\n\t"
10604 "call f2i_fixup\n\t"
10605 "popq $dst\n"
10606 "done: "%}
10607 opcode(0xF3, 0x0F, 0x2C);
10608 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10609 f2i_fixup(dst, src));
10610 ins_pipe(pipe_slow);
10611 %}
10613 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
10614 %{
10615 match(Set dst (ConvF2L src));
10616 effect(KILL cr);
10618 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
10619 "cmpq $dst, [0x8000000000000000]\n\t"
10620 "jne,s done\n\t"
10621 "subq rsp, #8\n\t"
10622 "movss [rsp], $src\n\t"
10623 "call f2l_fixup\n\t"
10624 "popq $dst\n"
10625 "done: "%}
10626 opcode(0xF3, 0x0F, 0x2C);
10627 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10628 f2l_fixup(dst, src));
10629 ins_pipe(pipe_slow);
10630 %}
10632 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
10633 %{
10634 match(Set dst (ConvD2I src));
10635 effect(KILL cr);
10637 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
10638 "cmpl $dst, #0x80000000\n\t"
10639 "jne,s done\n\t"
10640 "subq rsp, #8\n\t"
10641 "movsd [rsp], $src\n\t"
10642 "call d2i_fixup\n\t"
10643 "popq $dst\n"
10644 "done: "%}
10645 opcode(0xF2, 0x0F, 0x2C);
10646 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10647 d2i_fixup(dst, src));
10648 ins_pipe(pipe_slow);
10649 %}
10651 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
10652 %{
10653 match(Set dst (ConvD2L src));
10654 effect(KILL cr);
10656 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
10657 "cmpq $dst, [0x8000000000000000]\n\t"
10658 "jne,s done\n\t"
10659 "subq rsp, #8\n\t"
10660 "movsd [rsp], $src\n\t"
10661 "call d2l_fixup\n\t"
10662 "popq $dst\n"
10663 "done: "%}
10664 opcode(0xF2, 0x0F, 0x2C);
10665 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10666 d2l_fixup(dst, src));
10667 ins_pipe(pipe_slow);
10668 %}
10670 instruct convI2F_reg_reg(regF dst, rRegI src)
10671 %{
10672 predicate(!UseXmmI2F);
10673 match(Set dst (ConvI2F src));
10675 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10676 opcode(0xF3, 0x0F, 0x2A);
10677 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10678 ins_pipe(pipe_slow); // XXX
10679 %}
10681 instruct convI2F_reg_mem(regF dst, memory src)
10682 %{
10683 match(Set dst (ConvI2F (LoadI src)));
10685 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10686 opcode(0xF3, 0x0F, 0x2A);
10687 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10688 ins_pipe(pipe_slow); // XXX
10689 %}
10691 instruct convI2D_reg_reg(regD dst, rRegI src)
10692 %{
10693 predicate(!UseXmmI2D);
10694 match(Set dst (ConvI2D src));
10696 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10697 opcode(0xF2, 0x0F, 0x2A);
10698 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10699 ins_pipe(pipe_slow); // XXX
10700 %}
10702 instruct convI2D_reg_mem(regD dst, memory src)
10703 %{
10704 match(Set dst (ConvI2D (LoadI src)));
10706 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10707 opcode(0xF2, 0x0F, 0x2A);
10708 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10709 ins_pipe(pipe_slow); // XXX
10710 %}
10712 instruct convXI2F_reg(regF dst, rRegI src)
10713 %{
10714 predicate(UseXmmI2F);
10715 match(Set dst (ConvI2F src));
10717 format %{ "movdl $dst, $src\n\t"
10718 "cvtdq2psl $dst, $dst\t# i2f" %}
10719 ins_encode %{
10720 __ movdl($dst$$XMMRegister, $src$$Register);
10721 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
10722 %}
10723 ins_pipe(pipe_slow); // XXX
10724 %}
10726 instruct convXI2D_reg(regD dst, rRegI src)
10727 %{
10728 predicate(UseXmmI2D);
10729 match(Set dst (ConvI2D src));
10731 format %{ "movdl $dst, $src\n\t"
10732 "cvtdq2pdl $dst, $dst\t# i2d" %}
10733 ins_encode %{
10734 __ movdl($dst$$XMMRegister, $src$$Register);
10735 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
10736 %}
10737 ins_pipe(pipe_slow); // XXX
10738 %}
10740 instruct convL2F_reg_reg(regF dst, rRegL src)
10741 %{
10742 match(Set dst (ConvL2F src));
10744 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
10745 opcode(0xF3, 0x0F, 0x2A);
10746 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
10747 ins_pipe(pipe_slow); // XXX
10748 %}
10750 instruct convL2F_reg_mem(regF dst, memory src)
10751 %{
10752 match(Set dst (ConvL2F (LoadL src)));
10754 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
10755 opcode(0xF3, 0x0F, 0x2A);
10756 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
10757 ins_pipe(pipe_slow); // XXX
10758 %}
10760 instruct convL2D_reg_reg(regD dst, rRegL src)
10761 %{
10762 match(Set dst (ConvL2D src));
10764 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
10765 opcode(0xF2, 0x0F, 0x2A);
10766 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
10767 ins_pipe(pipe_slow); // XXX
10768 %}
10770 instruct convL2D_reg_mem(regD dst, memory src)
10771 %{
10772 match(Set dst (ConvL2D (LoadL src)));
10774 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
10775 opcode(0xF2, 0x0F, 0x2A);
10776 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
10777 ins_pipe(pipe_slow); // XXX
10778 %}
10780 instruct convI2L_reg_reg(rRegL dst, rRegI src)
10781 %{
10782 match(Set dst (ConvI2L src));
10784 ins_cost(125);
10785 format %{ "movslq $dst, $src\t# i2l" %}
10786 opcode(0x63); // needs REX.W
10787 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
10788 ins_pipe(ialu_reg_reg);
10789 %}
10791 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
10792 // %{
10793 // match(Set dst (ConvI2L src));
10794 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
10795 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
10796 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
10797 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
10798 // ((const TypeNode*) n)->type()->is_long()->_lo ==
10799 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
10801 // format %{ "movl $dst, $src\t# unsigned i2l" %}
10802 // ins_encode(enc_copy(dst, src));
10803 // // opcode(0x63); // needs REX.W
10804 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
10805 // ins_pipe(ialu_reg_reg);
10806 // %}
10808 instruct convI2L_reg_mem(rRegL dst, memory src)
10809 %{
10810 match(Set dst (ConvI2L (LoadI src)));
10812 format %{ "movslq $dst, $src\t# i2l" %}
10813 opcode(0x63); // needs REX.W
10814 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst,src));
10815 ins_pipe(ialu_reg_mem);
10816 %}
10818 // Zero-extend convert int to long
10819 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
10820 %{
10821 match(Set dst (AndL (ConvI2L src) mask));
10823 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
10824 ins_encode(enc_copy(dst, src));
10825 ins_pipe(ialu_reg_reg);
10826 %}
10828 // Zero-extend convert int to long
10829 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
10830 %{
10831 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
10833 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
10834 opcode(0x8B);
10835 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
10836 ins_pipe(ialu_reg_mem);
10837 %}
10839 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
10840 %{
10841 match(Set dst (AndL src mask));
10843 format %{ "movl $dst, $src\t# zero-extend long" %}
10844 ins_encode(enc_copy_always(dst, src));
10845 ins_pipe(ialu_reg_reg);
10846 %}
10848 instruct convL2I_reg_reg(rRegI dst, rRegL src)
10849 %{
10850 match(Set dst (ConvL2I src));
10852 format %{ "movl $dst, $src\t# l2i" %}
10853 ins_encode(enc_copy_always(dst, src));
10854 ins_pipe(ialu_reg_reg);
10855 %}
10858 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
10859 match(Set dst (MoveF2I src));
10860 effect(DEF dst, USE src);
10862 ins_cost(125);
10863 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
10864 opcode(0x8B);
10865 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
10866 ins_pipe(ialu_reg_mem);
10867 %}
10869 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
10870 match(Set dst (MoveI2F src));
10871 effect(DEF dst, USE src);
10873 ins_cost(125);
10874 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
10875 opcode(0xF3, 0x0F, 0x10);
10876 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10877 ins_pipe(pipe_slow);
10878 %}
10880 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
10881 match(Set dst (MoveD2L src));
10882 effect(DEF dst, USE src);
10884 ins_cost(125);
10885 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
10886 opcode(0x8B);
10887 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10888 ins_pipe(ialu_reg_mem);
10889 %}
10891 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
10892 predicate(!UseXmmLoadAndClearUpper);
10893 match(Set dst (MoveL2D src));
10894 effect(DEF dst, USE src);
10896 ins_cost(125);
10897 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
10898 opcode(0x66, 0x0F, 0x12);
10899 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10900 ins_pipe(pipe_slow);
10901 %}
10903 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
10904 predicate(UseXmmLoadAndClearUpper);
10905 match(Set dst (MoveL2D src));
10906 effect(DEF dst, USE src);
10908 ins_cost(125);
10909 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
10910 opcode(0xF2, 0x0F, 0x10);
10911 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10912 ins_pipe(pipe_slow);
10913 %}
10916 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
10917 match(Set dst (MoveF2I src));
10918 effect(DEF dst, USE src);
10920 ins_cost(95); // XXX
10921 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
10922 opcode(0xF3, 0x0F, 0x11);
10923 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
10924 ins_pipe(pipe_slow);
10925 %}
10927 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
10928 match(Set dst (MoveI2F src));
10929 effect(DEF dst, USE src);
10931 ins_cost(100);
10932 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
10933 opcode(0x89);
10934 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
10935 ins_pipe( ialu_mem_reg );
10936 %}
10938 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
10939 match(Set dst (MoveD2L src));
10940 effect(DEF dst, USE src);
10942 ins_cost(95); // XXX
10943 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
10944 opcode(0xF2, 0x0F, 0x11);
10945 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
10946 ins_pipe(pipe_slow);
10947 %}
10949 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
10950 match(Set dst (MoveL2D src));
10951 effect(DEF dst, USE src);
10953 ins_cost(100);
10954 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
10955 opcode(0x89);
10956 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10957 ins_pipe(ialu_mem_reg);
10958 %}
10960 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
10961 match(Set dst (MoveF2I src));
10962 effect(DEF dst, USE src);
10963 ins_cost(85);
10964 format %{ "movd $dst,$src\t# MoveF2I" %}
10965 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
10966 ins_pipe( pipe_slow );
10967 %}
10969 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
10970 match(Set dst (MoveD2L src));
10971 effect(DEF dst, USE src);
10972 ins_cost(85);
10973 format %{ "movd $dst,$src\t# MoveD2L" %}
10974 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
10975 ins_pipe( pipe_slow );
10976 %}
10978 // The next instructions have long latency and use Int unit. Set high cost.
10979 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
10980 match(Set dst (MoveI2F src));
10981 effect(DEF dst, USE src);
10982 ins_cost(300);
10983 format %{ "movd $dst,$src\t# MoveI2F" %}
10984 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
10985 ins_pipe( pipe_slow );
10986 %}
10988 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
10989 match(Set dst (MoveL2D src));
10990 effect(DEF dst, USE src);
10991 ins_cost(300);
10992 format %{ "movd $dst,$src\t# MoveL2D" %}
10993 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
10994 ins_pipe( pipe_slow );
10995 %}
10997 // Replicate scalar to packed byte (1 byte) values in xmm
10998 instruct Repl8B_reg(regD dst, regD src) %{
10999 match(Set dst (Replicate8B src));
11000 format %{ "MOVDQA $dst,$src\n\t"
11001 "PUNPCKLBW $dst,$dst\n\t"
11002 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11003 ins_encode( pshufd_8x8(dst, src));
11004 ins_pipe( pipe_slow );
11005 %}
11007 // Replicate scalar to packed byte (1 byte) values in xmm
11008 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11009 match(Set dst (Replicate8B src));
11010 format %{ "MOVD $dst,$src\n\t"
11011 "PUNPCKLBW $dst,$dst\n\t"
11012 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11013 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11014 ins_pipe( pipe_slow );
11015 %}
11017 // Replicate scalar zero to packed byte (1 byte) values in xmm
11018 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11019 match(Set dst (Replicate8B zero));
11020 format %{ "PXOR $dst,$dst\t! replicate8B" %}
11021 ins_encode( pxor(dst, dst));
11022 ins_pipe( fpu_reg_reg );
11023 %}
11025 // Replicate scalar to packed shore (2 byte) values in xmm
11026 instruct Repl4S_reg(regD dst, regD src) %{
11027 match(Set dst (Replicate4S src));
11028 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11029 ins_encode( pshufd_4x16(dst, src));
11030 ins_pipe( fpu_reg_reg );
11031 %}
11033 // Replicate scalar to packed shore (2 byte) values in xmm
11034 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11035 match(Set dst (Replicate4S src));
11036 format %{ "MOVD $dst,$src\n\t"
11037 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11038 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11039 ins_pipe( fpu_reg_reg );
11040 %}
11042 // Replicate scalar zero to packed short (2 byte) values in xmm
11043 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11044 match(Set dst (Replicate4S zero));
11045 format %{ "PXOR $dst,$dst\t! replicate4S" %}
11046 ins_encode( pxor(dst, dst));
11047 ins_pipe( fpu_reg_reg );
11048 %}
11050 // Replicate scalar to packed char (2 byte) values in xmm
11051 instruct Repl4C_reg(regD dst, regD src) %{
11052 match(Set dst (Replicate4C src));
11053 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11054 ins_encode( pshufd_4x16(dst, src));
11055 ins_pipe( fpu_reg_reg );
11056 %}
11058 // Replicate scalar to packed char (2 byte) values in xmm
11059 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11060 match(Set dst (Replicate4C src));
11061 format %{ "MOVD $dst,$src\n\t"
11062 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11063 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11064 ins_pipe( fpu_reg_reg );
11065 %}
11067 // Replicate scalar zero to packed char (2 byte) values in xmm
11068 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11069 match(Set dst (Replicate4C zero));
11070 format %{ "PXOR $dst,$dst\t! replicate4C" %}
11071 ins_encode( pxor(dst, dst));
11072 ins_pipe( fpu_reg_reg );
11073 %}
11075 // Replicate scalar to packed integer (4 byte) values in xmm
11076 instruct Repl2I_reg(regD dst, regD src) %{
11077 match(Set dst (Replicate2I src));
11078 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11079 ins_encode( pshufd(dst, src, 0x00));
11080 ins_pipe( fpu_reg_reg );
11081 %}
11083 // Replicate scalar to packed integer (4 byte) values in xmm
11084 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11085 match(Set dst (Replicate2I src));
11086 format %{ "MOVD $dst,$src\n\t"
11087 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11088 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11089 ins_pipe( fpu_reg_reg );
11090 %}
11092 // Replicate scalar zero to packed integer (2 byte) values in xmm
11093 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11094 match(Set dst (Replicate2I zero));
11095 format %{ "PXOR $dst,$dst\t! replicate2I" %}
11096 ins_encode( pxor(dst, dst));
11097 ins_pipe( fpu_reg_reg );
11098 %}
11100 // Replicate scalar to packed single precision floating point values in xmm
11101 instruct Repl2F_reg(regD dst, regD src) %{
11102 match(Set dst (Replicate2F src));
11103 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11104 ins_encode( pshufd(dst, src, 0xe0));
11105 ins_pipe( fpu_reg_reg );
11106 %}
11108 // Replicate scalar to packed single precision floating point values in xmm
11109 instruct Repl2F_regF(regD dst, regF src) %{
11110 match(Set dst (Replicate2F src));
11111 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11112 ins_encode( pshufd(dst, src, 0xe0));
11113 ins_pipe( fpu_reg_reg );
11114 %}
11116 // Replicate scalar to packed single precision floating point values in xmm
11117 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11118 match(Set dst (Replicate2F zero));
11119 format %{ "PXOR $dst,$dst\t! replicate2F" %}
11120 ins_encode( pxor(dst, dst));
11121 ins_pipe( fpu_reg_reg );
11122 %}
11125 // =======================================================================
11126 // fast clearing of an array
11127 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11128 rFlagsReg cr)
11129 %{
11130 match(Set dummy (ClearArray cnt base));
11131 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11133 format %{ "xorl rax, rax\t# ClearArray:\n\t"
11134 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11135 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11136 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11137 ins_pipe(pipe_slow);
11138 %}
11140 instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1,
11141 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr)
11142 %{
11143 match(Set result (StrComp str1 str2));
11144 effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
11145 //ins_cost(300);
11147 format %{ "String Compare $str1, $str2 -> $result // XXX KILL RAX, RBX" %}
11148 ins_encode( enc_String_Compare() );
11149 ins_pipe( pipe_slow );
11150 %}
11152 // fast array equals
11153 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1,
11154 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr) %{
11155 match(Set result (AryEq ary1 ary2));
11156 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr);
11157 //ins_cost(300);
11159 format %{ "Array Equals $ary1,$ary2 -> $result // KILL RAX, RBX" %}
11160 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) );
11161 ins_pipe( pipe_slow );
11162 %}
11164 //----------Control Flow Instructions------------------------------------------
11165 // Signed compare Instructions
11167 // XXX more variants!!
11168 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11169 %{
11170 match(Set cr (CmpI op1 op2));
11171 effect(DEF cr, USE op1, USE op2);
11173 format %{ "cmpl $op1, $op2" %}
11174 opcode(0x3B); /* Opcode 3B /r */
11175 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11176 ins_pipe(ialu_cr_reg_reg);
11177 %}
11179 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11180 %{
11181 match(Set cr (CmpI op1 op2));
11183 format %{ "cmpl $op1, $op2" %}
11184 opcode(0x81, 0x07); /* Opcode 81 /7 */
11185 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11186 ins_pipe(ialu_cr_reg_imm);
11187 %}
11189 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11190 %{
11191 match(Set cr (CmpI op1 (LoadI op2)));
11193 ins_cost(500); // XXX
11194 format %{ "cmpl $op1, $op2" %}
11195 opcode(0x3B); /* Opcode 3B /r */
11196 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11197 ins_pipe(ialu_cr_reg_mem);
11198 %}
11200 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11201 %{
11202 match(Set cr (CmpI src zero));
11204 format %{ "testl $src, $src" %}
11205 opcode(0x85);
11206 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11207 ins_pipe(ialu_cr_reg_imm);
11208 %}
11210 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11211 %{
11212 match(Set cr (CmpI (AndI src con) zero));
11214 format %{ "testl $src, $con" %}
11215 opcode(0xF7, 0x00);
11216 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11217 ins_pipe(ialu_cr_reg_imm);
11218 %}
11220 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11221 %{
11222 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11224 format %{ "testl $src, $mem" %}
11225 opcode(0x85);
11226 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11227 ins_pipe(ialu_cr_reg_mem);
11228 %}
11230 // Unsigned compare Instructions; really, same as signed except they
11231 // produce an rFlagsRegU instead of rFlagsReg.
11232 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11233 %{
11234 match(Set cr (CmpU op1 op2));
11236 format %{ "cmpl $op1, $op2\t# unsigned" %}
11237 opcode(0x3B); /* Opcode 3B /r */
11238 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11239 ins_pipe(ialu_cr_reg_reg);
11240 %}
11242 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11243 %{
11244 match(Set cr (CmpU op1 op2));
11246 format %{ "cmpl $op1, $op2\t# unsigned" %}
11247 opcode(0x81,0x07); /* Opcode 81 /7 */
11248 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11249 ins_pipe(ialu_cr_reg_imm);
11250 %}
11252 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11253 %{
11254 match(Set cr (CmpU op1 (LoadI op2)));
11256 ins_cost(500); // XXX
11257 format %{ "cmpl $op1, $op2\t# unsigned" %}
11258 opcode(0x3B); /* Opcode 3B /r */
11259 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11260 ins_pipe(ialu_cr_reg_mem);
11261 %}
11263 // // // Cisc-spilled version of cmpU_rReg
11264 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11265 // //%{
11266 // // match(Set cr (CmpU (LoadI op1) op2));
11267 // //
11268 // // format %{ "CMPu $op1,$op2" %}
11269 // // ins_cost(500);
11270 // // opcode(0x39); /* Opcode 39 /r */
11271 // // ins_encode( OpcP, reg_mem( op1, op2) );
11272 // //%}
11274 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11275 %{
11276 match(Set cr (CmpU src zero));
11278 format %{ "testl $src, $src\t# unsigned" %}
11279 opcode(0x85);
11280 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11281 ins_pipe(ialu_cr_reg_imm);
11282 %}
11284 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11285 %{
11286 match(Set cr (CmpP op1 op2));
11288 format %{ "cmpq $op1, $op2\t# ptr" %}
11289 opcode(0x3B); /* Opcode 3B /r */
11290 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11291 ins_pipe(ialu_cr_reg_reg);
11292 %}
11294 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11295 %{
11296 match(Set cr (CmpP op1 (LoadP op2)));
11298 ins_cost(500); // XXX
11299 format %{ "cmpq $op1, $op2\t# ptr" %}
11300 opcode(0x3B); /* Opcode 3B /r */
11301 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11302 ins_pipe(ialu_cr_reg_mem);
11303 %}
11305 // // // Cisc-spilled version of cmpP_rReg
11306 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11307 // //%{
11308 // // match(Set cr (CmpP (LoadP op1) op2));
11309 // //
11310 // // format %{ "CMPu $op1,$op2" %}
11311 // // ins_cost(500);
11312 // // opcode(0x39); /* Opcode 39 /r */
11313 // // ins_encode( OpcP, reg_mem( op1, op2) );
11314 // //%}
11316 // XXX this is generalized by compP_rReg_mem???
11317 // Compare raw pointer (used in out-of-heap check).
11318 // Only works because non-oop pointers must be raw pointers
11319 // and raw pointers have no anti-dependencies.
11320 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11321 %{
11322 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11323 match(Set cr (CmpP op1 (LoadP op2)));
11325 format %{ "cmpq $op1, $op2\t# raw ptr" %}
11326 opcode(0x3B); /* Opcode 3B /r */
11327 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11328 ins_pipe(ialu_cr_reg_mem);
11329 %}
11331 // This will generate a signed flags result. This should be OK since
11332 // any compare to a zero should be eq/neq.
11333 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11334 %{
11335 match(Set cr (CmpP src zero));
11337 format %{ "testq $src, $src\t# ptr" %}
11338 opcode(0x85);
11339 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11340 ins_pipe(ialu_cr_reg_imm);
11341 %}
11343 // This will generate a signed flags result. This should be OK since
11344 // any compare to a zero should be eq/neq.
11345 instruct testP_reg_mem(rFlagsReg cr, memory op, immP0 zero)
11346 %{
11347 match(Set cr (CmpP (LoadP op) zero));
11349 ins_cost(500); // XXX
11350 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11351 opcode(0xF7); /* Opcode F7 /0 */
11352 ins_encode(REX_mem_wide(op),
11353 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11354 ins_pipe(ialu_cr_reg_imm);
11355 %}
11358 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11359 %{
11360 match(Set cr (CmpN op1 op2));
11362 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11363 ins_encode %{ __ cmpl(as_Register($op1$$reg), as_Register($op2$$reg)); %}
11364 ins_pipe(ialu_cr_reg_reg);
11365 %}
11367 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11368 %{
11369 match(Set cr (CmpN src (LoadN mem)));
11371 ins_cost(500); // XXX
11372 format %{ "cmpl $src, mem\t# compressed ptr" %}
11373 ins_encode %{
11374 Address adr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
11375 __ cmpl(as_Register($src$$reg), adr);
11376 %}
11377 ins_pipe(ialu_cr_reg_mem);
11378 %}
11380 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11381 match(Set cr (CmpN src zero));
11383 format %{ "testl $src, $src\t# compressed ptr" %}
11384 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11385 ins_pipe(ialu_cr_reg_imm);
11386 %}
11388 instruct testN_reg_mem(rFlagsReg cr, memory mem, immN0 zero)
11389 %{
11390 match(Set cr (CmpN (LoadN mem) zero));
11392 ins_cost(500); // XXX
11393 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %}
11394 ins_encode %{
11395 Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
11396 __ cmpl(addr, (int)0xFFFFFFFF);
11397 %}
11398 ins_pipe(ialu_cr_reg_mem);
11399 %}
11401 // Yanked all unsigned pointer compare operations.
11402 // Pointer compares are done with CmpP which is already unsigned.
11404 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11405 %{
11406 match(Set cr (CmpL op1 op2));
11408 format %{ "cmpq $op1, $op2" %}
11409 opcode(0x3B); /* Opcode 3B /r */
11410 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11411 ins_pipe(ialu_cr_reg_reg);
11412 %}
11414 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11415 %{
11416 match(Set cr (CmpL op1 op2));
11418 format %{ "cmpq $op1, $op2" %}
11419 opcode(0x81, 0x07); /* Opcode 81 /7 */
11420 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11421 ins_pipe(ialu_cr_reg_imm);
11422 %}
11424 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11425 %{
11426 match(Set cr (CmpL op1 (LoadL op2)));
11428 ins_cost(500); // XXX
11429 format %{ "cmpq $op1, $op2" %}
11430 opcode(0x3B); /* Opcode 3B /r */
11431 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11432 ins_pipe(ialu_cr_reg_mem);
11433 %}
11435 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11436 %{
11437 match(Set cr (CmpL src zero));
11439 format %{ "testq $src, $src" %}
11440 opcode(0x85);
11441 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11442 ins_pipe(ialu_cr_reg_imm);
11443 %}
11445 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11446 %{
11447 match(Set cr (CmpL (AndL src con) zero));
11449 format %{ "testq $src, $con\t# long" %}
11450 opcode(0xF7, 0x00);
11451 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11452 ins_pipe(ialu_cr_reg_imm);
11453 %}
11455 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
11456 %{
11457 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
11459 format %{ "testq $src, $mem" %}
11460 opcode(0x85);
11461 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
11462 ins_pipe(ialu_cr_reg_mem);
11463 %}
11465 // Manifest a CmpL result in an integer register. Very painful.
11466 // This is the test to avoid.
11467 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
11468 %{
11469 match(Set dst (CmpL3 src1 src2));
11470 effect(KILL flags);
11472 ins_cost(275); // XXX
11473 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
11474 "movl $dst, -1\n\t"
11475 "jl,s done\n\t"
11476 "setne $dst\n\t"
11477 "movzbl $dst, $dst\n\t"
11478 "done:" %}
11479 ins_encode(cmpl3_flag(src1, src2, dst));
11480 ins_pipe(pipe_slow);
11481 %}
11483 //----------Max and Min--------------------------------------------------------
11484 // Min Instructions
11486 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
11487 %{
11488 effect(USE_DEF dst, USE src, USE cr);
11490 format %{ "cmovlgt $dst, $src\t# min" %}
11491 opcode(0x0F, 0x4F);
11492 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11493 ins_pipe(pipe_cmov_reg);
11494 %}
11497 instruct minI_rReg(rRegI dst, rRegI src)
11498 %{
11499 match(Set dst (MinI dst src));
11501 ins_cost(200);
11502 expand %{
11503 rFlagsReg cr;
11504 compI_rReg(cr, dst, src);
11505 cmovI_reg_g(dst, src, cr);
11506 %}
11507 %}
11509 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
11510 %{
11511 effect(USE_DEF dst, USE src, USE cr);
11513 format %{ "cmovllt $dst, $src\t# max" %}
11514 opcode(0x0F, 0x4C);
11515 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11516 ins_pipe(pipe_cmov_reg);
11517 %}
11520 instruct maxI_rReg(rRegI dst, rRegI src)
11521 %{
11522 match(Set dst (MaxI dst src));
11524 ins_cost(200);
11525 expand %{
11526 rFlagsReg cr;
11527 compI_rReg(cr, dst, src);
11528 cmovI_reg_l(dst, src, cr);
11529 %}
11530 %}
11532 // ============================================================================
11533 // Branch Instructions
11535 // Jump Direct - Label defines a relative address from JMP+1
11536 instruct jmpDir(label labl)
11537 %{
11538 match(Goto);
11539 effect(USE labl);
11541 ins_cost(300);
11542 format %{ "jmp $labl" %}
11543 size(5);
11544 opcode(0xE9);
11545 ins_encode(OpcP, Lbl(labl));
11546 ins_pipe(pipe_jmp);
11547 ins_pc_relative(1);
11548 %}
11550 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11551 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
11552 %{
11553 match(If cop cr);
11554 effect(USE labl);
11556 ins_cost(300);
11557 format %{ "j$cop $labl" %}
11558 size(6);
11559 opcode(0x0F, 0x80);
11560 ins_encode(Jcc(cop, labl));
11561 ins_pipe(pipe_jcc);
11562 ins_pc_relative(1);
11563 %}
11565 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11566 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
11567 %{
11568 match(CountedLoopEnd cop cr);
11569 effect(USE labl);
11571 ins_cost(300);
11572 format %{ "j$cop $labl\t# loop end" %}
11573 size(6);
11574 opcode(0x0F, 0x80);
11575 ins_encode(Jcc(cop, labl));
11576 ins_pipe(pipe_jcc);
11577 ins_pc_relative(1);
11578 %}
11580 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11581 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11582 match(CountedLoopEnd cop cmp);
11583 effect(USE labl);
11585 ins_cost(300);
11586 format %{ "j$cop,u $labl\t# loop end" %}
11587 size(6);
11588 opcode(0x0F, 0x80);
11589 ins_encode(Jcc(cop, labl));
11590 ins_pipe(pipe_jcc);
11591 ins_pc_relative(1);
11592 %}
11594 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11595 match(CountedLoopEnd cop cmp);
11596 effect(USE labl);
11598 ins_cost(200);
11599 format %{ "j$cop,u $labl\t# loop end" %}
11600 size(6);
11601 opcode(0x0F, 0x80);
11602 ins_encode(Jcc(cop, labl));
11603 ins_pipe(pipe_jcc);
11604 ins_pc_relative(1);
11605 %}
11607 // Jump Direct Conditional - using unsigned comparison
11608 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11609 match(If cop cmp);
11610 effect(USE labl);
11612 ins_cost(300);
11613 format %{ "j$cop,u $labl" %}
11614 size(6);
11615 opcode(0x0F, 0x80);
11616 ins_encode(Jcc(cop, labl));
11617 ins_pipe(pipe_jcc);
11618 ins_pc_relative(1);
11619 %}
11621 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11622 match(If cop cmp);
11623 effect(USE labl);
11625 ins_cost(200);
11626 format %{ "j$cop,u $labl" %}
11627 size(6);
11628 opcode(0x0F, 0x80);
11629 ins_encode(Jcc(cop, labl));
11630 ins_pipe(pipe_jcc);
11631 ins_pc_relative(1);
11632 %}
11634 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
11635 match(If cop cmp);
11636 effect(USE labl);
11638 ins_cost(200);
11639 format %{ $$template
11640 if ($cop$$cmpcode == Assembler::notEqual) {
11641 $$emit$$"jp,u $labl\n\t"
11642 $$emit$$"j$cop,u $labl"
11643 } else {
11644 $$emit$$"jp,u done\n\t"
11645 $$emit$$"j$cop,u $labl\n\t"
11646 $$emit$$"done:"
11647 }
11648 %}
11649 size(12);
11650 opcode(0x0F, 0x80);
11651 ins_encode %{
11652 Label* l = $labl$$label;
11653 $$$emit8$primary;
11654 emit_cc(cbuf, $secondary, Assembler::parity);
11655 int parity_disp = -1;
11656 if ($cop$$cmpcode == Assembler::notEqual) {
11657 // the two jumps 6 bytes apart so the jump distances are too
11658 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
11659 } else if ($cop$$cmpcode == Assembler::equal) {
11660 parity_disp = 6;
11661 } else {
11662 ShouldNotReachHere();
11663 }
11664 emit_d32(cbuf, parity_disp);
11665 $$$emit8$primary;
11666 emit_cc(cbuf, $secondary, $cop$$cmpcode);
11667 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
11668 emit_d32(cbuf, disp);
11669 %}
11670 ins_pipe(pipe_jcc);
11671 ins_pc_relative(1);
11672 %}
11674 // ============================================================================
11675 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
11676 // superklass array for an instance of the superklass. Set a hidden
11677 // internal cache on a hit (cache is checked with exposed code in
11678 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
11679 // encoding ALSO sets flags.
11681 instruct partialSubtypeCheck(rdi_RegP result,
11682 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
11683 rFlagsReg cr)
11684 %{
11685 match(Set result (PartialSubtypeCheck sub super));
11686 effect(KILL rcx, KILL cr);
11688 ins_cost(1100); // slightly larger than the next version
11689 format %{ "cmpq rax, rsi\n\t"
11690 "jeq,s hit\n\t"
11691 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
11692 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
11693 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
11694 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
11695 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
11696 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
11697 "hit:\n\t"
11698 "xorq $result, $result\t\t Hit: rdi zero\n\t"
11699 "miss:\t" %}
11701 opcode(0x1); // Force a XOR of RDI
11702 ins_encode(enc_PartialSubtypeCheck());
11703 ins_pipe(pipe_slow);
11704 %}
11706 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
11707 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
11708 immP0 zero,
11709 rdi_RegP result)
11710 %{
11711 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
11712 predicate(!UseCompressedOops); // decoding oop kills condition codes
11713 effect(KILL rcx, KILL result);
11715 ins_cost(1000);
11716 format %{ "cmpq rax, rsi\n\t"
11717 "jeq,s miss\t# Actually a hit; we are done.\n\t"
11718 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
11719 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
11720 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
11721 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
11722 "jne,s miss\t\t# Missed: flags nz\n\t"
11723 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
11724 "miss:\t" %}
11726 opcode(0x0); // No need to XOR RDI
11727 ins_encode(enc_PartialSubtypeCheck());
11728 ins_pipe(pipe_slow);
11729 %}
11731 // ============================================================================
11732 // Branch Instructions -- short offset versions
11733 //
11734 // These instructions are used to replace jumps of a long offset (the default
11735 // match) with jumps of a shorter offset. These instructions are all tagged
11736 // with the ins_short_branch attribute, which causes the ADLC to suppress the
11737 // match rules in general matching. Instead, the ADLC generates a conversion
11738 // method in the MachNode which can be used to do in-place replacement of the
11739 // long variant with the shorter variant. The compiler will determine if a
11740 // branch can be taken by the is_short_branch_offset() predicate in the machine
11741 // specific code section of the file.
11743 // Jump Direct - Label defines a relative address from JMP+1
11744 instruct jmpDir_short(label labl) %{
11745 match(Goto);
11746 effect(USE labl);
11748 ins_cost(300);
11749 format %{ "jmp,s $labl" %}
11750 size(2);
11751 opcode(0xEB);
11752 ins_encode(OpcP, LblShort(labl));
11753 ins_pipe(pipe_jmp);
11754 ins_pc_relative(1);
11755 ins_short_branch(1);
11756 %}
11758 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11759 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
11760 match(If cop cr);
11761 effect(USE labl);
11763 ins_cost(300);
11764 format %{ "j$cop,s $labl" %}
11765 size(2);
11766 opcode(0x70);
11767 ins_encode(JccShort(cop, labl));
11768 ins_pipe(pipe_jcc);
11769 ins_pc_relative(1);
11770 ins_short_branch(1);
11771 %}
11773 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11774 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
11775 match(CountedLoopEnd cop cr);
11776 effect(USE labl);
11778 ins_cost(300);
11779 format %{ "j$cop,s $labl\t# loop end" %}
11780 size(2);
11781 opcode(0x70);
11782 ins_encode(JccShort(cop, labl));
11783 ins_pipe(pipe_jcc);
11784 ins_pc_relative(1);
11785 ins_short_branch(1);
11786 %}
11788 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11789 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11790 match(CountedLoopEnd cop cmp);
11791 effect(USE labl);
11793 ins_cost(300);
11794 format %{ "j$cop,us $labl\t# loop end" %}
11795 size(2);
11796 opcode(0x70);
11797 ins_encode(JccShort(cop, labl));
11798 ins_pipe(pipe_jcc);
11799 ins_pc_relative(1);
11800 ins_short_branch(1);
11801 %}
11803 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11804 match(CountedLoopEnd cop cmp);
11805 effect(USE labl);
11807 ins_cost(300);
11808 format %{ "j$cop,us $labl\t# loop end" %}
11809 size(2);
11810 opcode(0x70);
11811 ins_encode(JccShort(cop, labl));
11812 ins_pipe(pipe_jcc);
11813 ins_pc_relative(1);
11814 ins_short_branch(1);
11815 %}
11817 // Jump Direct Conditional - using unsigned comparison
11818 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11819 match(If cop cmp);
11820 effect(USE labl);
11822 ins_cost(300);
11823 format %{ "j$cop,us $labl" %}
11824 size(2);
11825 opcode(0x70);
11826 ins_encode(JccShort(cop, labl));
11827 ins_pipe(pipe_jcc);
11828 ins_pc_relative(1);
11829 ins_short_branch(1);
11830 %}
11832 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11833 match(If cop cmp);
11834 effect(USE labl);
11836 ins_cost(300);
11837 format %{ "j$cop,us $labl" %}
11838 size(2);
11839 opcode(0x70);
11840 ins_encode(JccShort(cop, labl));
11841 ins_pipe(pipe_jcc);
11842 ins_pc_relative(1);
11843 ins_short_branch(1);
11844 %}
11846 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
11847 match(If cop cmp);
11848 effect(USE labl);
11850 ins_cost(300);
11851 format %{ $$template
11852 if ($cop$$cmpcode == Assembler::notEqual) {
11853 $$emit$$"jp,u,s $labl\n\t"
11854 $$emit$$"j$cop,u,s $labl"
11855 } else {
11856 $$emit$$"jp,u,s done\n\t"
11857 $$emit$$"j$cop,u,s $labl\n\t"
11858 $$emit$$"done:"
11859 }
11860 %}
11861 size(4);
11862 opcode(0x70);
11863 ins_encode %{
11864 Label* l = $labl$$label;
11865 emit_cc(cbuf, $primary, Assembler::parity);
11866 int parity_disp = -1;
11867 if ($cop$$cmpcode == Assembler::notEqual) {
11868 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
11869 } else if ($cop$$cmpcode == Assembler::equal) {
11870 parity_disp = 2;
11871 } else {
11872 ShouldNotReachHere();
11873 }
11874 emit_d8(cbuf, parity_disp);
11875 emit_cc(cbuf, $primary, $cop$$cmpcode);
11876 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
11877 emit_d8(cbuf, disp);
11878 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
11879 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
11880 %}
11881 ins_pipe(pipe_jcc);
11882 ins_pc_relative(1);
11883 ins_short_branch(1);
11884 %}
11886 // ============================================================================
11887 // inlined locking and unlocking
11889 instruct cmpFastLock(rFlagsReg cr,
11890 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
11891 %{
11892 match(Set cr (FastLock object box));
11893 effect(TEMP tmp, TEMP scr);
11895 ins_cost(300);
11896 format %{ "fastlock $object,$box,$tmp,$scr" %}
11897 ins_encode(Fast_Lock(object, box, tmp, scr));
11898 ins_pipe(pipe_slow);
11899 ins_pc_relative(1);
11900 %}
11902 instruct cmpFastUnlock(rFlagsReg cr,
11903 rRegP object, rax_RegP box, rRegP tmp)
11904 %{
11905 match(Set cr (FastUnlock object box));
11906 effect(TEMP tmp);
11908 ins_cost(300);
11909 format %{ "fastunlock $object, $box, $tmp" %}
11910 ins_encode(Fast_Unlock(object, box, tmp));
11911 ins_pipe(pipe_slow);
11912 ins_pc_relative(1);
11913 %}
11916 // ============================================================================
11917 // Safepoint Instructions
11918 instruct safePoint_poll(rFlagsReg cr)
11919 %{
11920 match(SafePoint);
11921 effect(KILL cr);
11923 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
11924 "# Safepoint: poll for GC" %}
11925 size(6); // Opcode + ModRM + Disp32 == 6 bytes
11926 ins_cost(125);
11927 ins_encode(enc_safepoint_poll);
11928 ins_pipe(ialu_reg_mem);
11929 %}
11931 // ============================================================================
11932 // Procedure Call/Return Instructions
11933 // Call Java Static Instruction
11934 // Note: If this code changes, the corresponding ret_addr_offset() and
11935 // compute_padding() functions will have to be adjusted.
11936 instruct CallStaticJavaDirect(method meth)
11937 %{
11938 match(CallStaticJava);
11939 effect(USE meth);
11941 ins_cost(300);
11942 format %{ "call,static " %}
11943 opcode(0xE8); /* E8 cd */
11944 ins_encode(Java_Static_Call(meth), call_epilog);
11945 ins_pipe(pipe_slow);
11946 ins_pc_relative(1);
11947 ins_alignment(4);
11948 %}
11950 // Call Java Dynamic Instruction
11951 // Note: If this code changes, the corresponding ret_addr_offset() and
11952 // compute_padding() functions will have to be adjusted.
11953 instruct CallDynamicJavaDirect(method meth)
11954 %{
11955 match(CallDynamicJava);
11956 effect(USE meth);
11958 ins_cost(300);
11959 format %{ "movq rax, #Universe::non_oop_word()\n\t"
11960 "call,dynamic " %}
11961 opcode(0xE8); /* E8 cd */
11962 ins_encode(Java_Dynamic_Call(meth), call_epilog);
11963 ins_pipe(pipe_slow);
11964 ins_pc_relative(1);
11965 ins_alignment(4);
11966 %}
11968 // Call Runtime Instruction
11969 instruct CallRuntimeDirect(method meth)
11970 %{
11971 match(CallRuntime);
11972 effect(USE meth);
11974 ins_cost(300);
11975 format %{ "call,runtime " %}
11976 opcode(0xE8); /* E8 cd */
11977 ins_encode(Java_To_Runtime(meth));
11978 ins_pipe(pipe_slow);
11979 ins_pc_relative(1);
11980 %}
11982 // Call runtime without safepoint
11983 instruct CallLeafDirect(method meth)
11984 %{
11985 match(CallLeaf);
11986 effect(USE meth);
11988 ins_cost(300);
11989 format %{ "call_leaf,runtime " %}
11990 opcode(0xE8); /* E8 cd */
11991 ins_encode(Java_To_Runtime(meth));
11992 ins_pipe(pipe_slow);
11993 ins_pc_relative(1);
11994 %}
11996 // Call runtime without safepoint
11997 instruct CallLeafNoFPDirect(method meth)
11998 %{
11999 match(CallLeafNoFP);
12000 effect(USE meth);
12002 ins_cost(300);
12003 format %{ "call_leaf_nofp,runtime " %}
12004 opcode(0xE8); /* E8 cd */
12005 ins_encode(Java_To_Runtime(meth));
12006 ins_pipe(pipe_slow);
12007 ins_pc_relative(1);
12008 %}
12010 // Return Instruction
12011 // Remove the return address & jump to it.
12012 // Notice: We always emit a nop after a ret to make sure there is room
12013 // for safepoint patching
12014 instruct Ret()
12015 %{
12016 match(Return);
12018 format %{ "ret" %}
12019 opcode(0xC3);
12020 ins_encode(OpcP);
12021 ins_pipe(pipe_jmp);
12022 %}
12024 // Tail Call; Jump from runtime stub to Java code.
12025 // Also known as an 'interprocedural jump'.
12026 // Target of jump will eventually return to caller.
12027 // TailJump below removes the return address.
12028 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12029 %{
12030 match(TailCall jump_target method_oop);
12032 ins_cost(300);
12033 format %{ "jmp $jump_target\t# rbx holds method oop" %}
12034 opcode(0xFF, 0x4); /* Opcode FF /4 */
12035 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12036 ins_pipe(pipe_jmp);
12037 %}
12039 // Tail Jump; remove the return address; jump to target.
12040 // TailCall above leaves the return address around.
12041 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12042 %{
12043 match(TailJump jump_target ex_oop);
12045 ins_cost(300);
12046 format %{ "popq rdx\t# pop return address\n\t"
12047 "jmp $jump_target" %}
12048 opcode(0xFF, 0x4); /* Opcode FF /4 */
12049 ins_encode(Opcode(0x5a), // popq rdx
12050 REX_reg(jump_target), OpcP, reg_opc(jump_target));
12051 ins_pipe(pipe_jmp);
12052 %}
12054 // Create exception oop: created by stack-crawling runtime code.
12055 // Created exception is now available to this handler, and is setup
12056 // just prior to jumping to this handler. No code emitted.
12057 instruct CreateException(rax_RegP ex_oop)
12058 %{
12059 match(Set ex_oop (CreateEx));
12061 size(0);
12062 // use the following format syntax
12063 format %{ "# exception oop is in rax; no code emitted" %}
12064 ins_encode();
12065 ins_pipe(empty);
12066 %}
12068 // Rethrow exception:
12069 // The exception oop will come in the first argument position.
12070 // Then JUMP (not call) to the rethrow stub code.
12071 instruct RethrowException()
12072 %{
12073 match(Rethrow);
12075 // use the following format syntax
12076 format %{ "jmp rethrow_stub" %}
12077 ins_encode(enc_rethrow);
12078 ins_pipe(pipe_jmp);
12079 %}
12082 //----------PEEPHOLE RULES-----------------------------------------------------
12083 // These must follow all instruction definitions as they use the names
12084 // defined in the instructions definitions.
12085 //
12086 // peepmatch ( root_instr_name [precerding_instruction]* );
12087 //
12088 // peepconstraint %{
12089 // (instruction_number.operand_name relational_op instruction_number.operand_name
12090 // [, ...] );
12091 // // instruction numbers are zero-based using left to right order in peepmatch
12092 //
12093 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12094 // // provide an instruction_number.operand_name for each operand that appears
12095 // // in the replacement instruction's match rule
12096 //
12097 // ---------VM FLAGS---------------------------------------------------------
12098 //
12099 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12100 //
12101 // Each peephole rule is given an identifying number starting with zero and
12102 // increasing by one in the order seen by the parser. An individual peephole
12103 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12104 // on the command-line.
12105 //
12106 // ---------CURRENT LIMITATIONS----------------------------------------------
12107 //
12108 // Only match adjacent instructions in same basic block
12109 // Only equality constraints
12110 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12111 // Only one replacement instruction
12112 //
12113 // ---------EXAMPLE----------------------------------------------------------
12114 //
12115 // // pertinent parts of existing instructions in architecture description
12116 // instruct movI(rRegI dst, rRegI src)
12117 // %{
12118 // match(Set dst (CopyI src));
12119 // %}
12120 //
12121 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12122 // %{
12123 // match(Set dst (AddI dst src));
12124 // effect(KILL cr);
12125 // %}
12126 //
12127 // // Change (inc mov) to lea
12128 // peephole %{
12129 // // increment preceeded by register-register move
12130 // peepmatch ( incI_rReg movI );
12131 // // require that the destination register of the increment
12132 // // match the destination register of the move
12133 // peepconstraint ( 0.dst == 1.dst );
12134 // // construct a replacement instruction that sets
12135 // // the destination to ( move's source register + one )
12136 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12137 // %}
12138 //
12140 // Implementation no longer uses movX instructions since
12141 // machine-independent system no longer uses CopyX nodes.
12142 //
12143 // peephole
12144 // %{
12145 // peepmatch (incI_rReg movI);
12146 // peepconstraint (0.dst == 1.dst);
12147 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12148 // %}
12150 // peephole
12151 // %{
12152 // peepmatch (decI_rReg movI);
12153 // peepconstraint (0.dst == 1.dst);
12154 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12155 // %}
12157 // peephole
12158 // %{
12159 // peepmatch (addI_rReg_imm movI);
12160 // peepconstraint (0.dst == 1.dst);
12161 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12162 // %}
12164 // peephole
12165 // %{
12166 // peepmatch (incL_rReg movL);
12167 // peepconstraint (0.dst == 1.dst);
12168 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12169 // %}
12171 // peephole
12172 // %{
12173 // peepmatch (decL_rReg movL);
12174 // peepconstraint (0.dst == 1.dst);
12175 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12176 // %}
12178 // peephole
12179 // %{
12180 // peepmatch (addL_rReg_imm movL);
12181 // peepconstraint (0.dst == 1.dst);
12182 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12183 // %}
12185 // peephole
12186 // %{
12187 // peepmatch (addP_rReg_imm movP);
12188 // peepconstraint (0.dst == 1.dst);
12189 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12190 // %}
12192 // // Change load of spilled value to only a spill
12193 // instruct storeI(memory mem, rRegI src)
12194 // %{
12195 // match(Set mem (StoreI mem src));
12196 // %}
12197 //
12198 // instruct loadI(rRegI dst, memory mem)
12199 // %{
12200 // match(Set dst (LoadI mem));
12201 // %}
12202 //
12204 peephole
12205 %{
12206 peepmatch (loadI storeI);
12207 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12208 peepreplace (storeI(1.mem 1.mem 1.src));
12209 %}
12211 peephole
12212 %{
12213 peepmatch (loadL storeL);
12214 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12215 peepreplace (storeL(1.mem 1.mem 1.src));
12216 %}
12218 //----------SMARTSPILL RULES---------------------------------------------------
12219 // These must follow all instruction definitions as they use the names
12220 // defined in the instructions definitions.