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 //
  24 
  25 // AMD64 Architecture Description File
  26 
  27 //----------REGISTER DEFINITION BLOCK------------------------------------------
  28 // This information is used by the matcher and the register allocator to
  29 // describe individual registers and classes of registers within the target
  30 // archtecture.
  31 
  32 register %{
  33 //----------Architecture Description Register Definitions----------------------
  34 // General Registers
  35 // "reg_def"  name ( register save type, C convention save type,
  36 //                   ideal register type, encoding );
  37 // Register Save Types:
  38 //
  39 // NS  = No-Save:       The register allocator assumes that these registers
  40 //                      can be used without saving upon entry to the method, &
  41 //                      that they do not need to be saved at call sites.
  42 //
  43 // SOC = Save-On-Call:  The register allocator assumes that these registers
  44 //                      can be used without saving upon entry to the method,
  45 //                      but that they must be saved at call sites.
  46 //
  47 // SOE = Save-On-Entry: The register allocator assumes that these registers
  48 //                      must be saved before using them upon entry to the
  49 //                      method, but they do not need to be saved at call
  50 //                      sites.
  51 //
  52 // AS  = Always-Save:   The register allocator assumes that these registers
  53 //                      must be saved before using them upon entry to the
  54 //                      method, & that they must be saved at call sites.
  55 //
  56 // Ideal Register Type is used to determine how to save & restore a
  57 // register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
  58 // spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
  59 //
  60 // The encoding number is the actual bit-pattern placed into the opcodes.
  61 
  62 // General Registers
  63 // R8-R15 must be encoded with REX.  (RSP, RBP, RSI, RDI need REX when
  64 // used as byte registers)
  65 
  66 // Previously set RBX, RSI, and RDI as save-on-entry for java code
  67 // Turn off SOE in java-code due to frequent use of uncommon-traps.
  68 // Now that allocator is better, turn on RSI and RDI as SOE registers.
  69 
  70 reg_def RAX  (SOC, SOC, Op_RegI,  0, rax->as_VMReg());
  71 reg_def RAX_H(SOC, SOC, Op_RegI,  0, rax->as_VMReg()->next());
  72 
  73 reg_def RCX  (SOC, SOC, Op_RegI,  1, rcx->as_VMReg());
  74 reg_def RCX_H(SOC, SOC, Op_RegI,  1, rcx->as_VMReg()->next());
  75 
  76 reg_def RDX  (SOC, SOC, Op_RegI,  2, rdx->as_VMReg());
  77 reg_def RDX_H(SOC, SOC, Op_RegI,  2, rdx->as_VMReg()->next());
  78 
  79 reg_def RBX  (SOC, SOE, Op_RegI,  3, rbx->as_VMReg());
  80 reg_def RBX_H(SOC, SOE, Op_RegI,  3, rbx->as_VMReg()->next());
  81 
  82 reg_def RSP  (NS,  NS,  Op_RegI,  4, rsp->as_VMReg());
  83 reg_def RSP_H(NS,  NS,  Op_RegI,  4, rsp->as_VMReg()->next());
  84 
  85 // now that adapter frames are gone RBP is always saved and restored by the prolog/epilog code
  86 reg_def RBP  (NS, SOE, Op_RegI,  5, rbp->as_VMReg());
  87 reg_def RBP_H(NS, SOE, Op_RegI,  5, rbp->as_VMReg()->next());
  88 
  89 #ifdef _WIN64
  90 
  91 reg_def RSI  (SOC, SOE, Op_RegI,  6, rsi->as_VMReg());
  92 reg_def RSI_H(SOC, SOE, Op_RegI,  6, rsi->as_VMReg()->next());
  93 
  94 reg_def RDI  (SOC, SOE, Op_RegI,  7, rdi->as_VMReg());
  95 reg_def RDI_H(SOC, SOE, Op_RegI,  7, rdi->as_VMReg()->next());
  96 
  97 #else
  98 
  99 reg_def RSI  (SOC, SOC, Op_RegI,  6, rsi->as_VMReg());
 100 reg_def RSI_H(SOC, SOC, Op_RegI,  6, rsi->as_VMReg()->next());
 101 
 102 reg_def RDI  (SOC, SOC, Op_RegI,  7, rdi->as_VMReg());
 103 reg_def RDI_H(SOC, SOC, Op_RegI,  7, rdi->as_VMReg()->next());
 104 
 105 #endif
 106 
 107 reg_def R8   (SOC, SOC, Op_RegI,  8, r8->as_VMReg());
 108 reg_def R8_H (SOC, SOC, Op_RegI,  8, r8->as_VMReg()->next());
 109 
 110 reg_def R9   (SOC, SOC, Op_RegI,  9, r9->as_VMReg());
 111 reg_def R9_H (SOC, SOC, Op_RegI,  9, r9->as_VMReg()->next());
 112 
 113 reg_def R10  (SOC, SOC, Op_RegI, 10, r10->as_VMReg());
 114 reg_def R10_H(SOC, SOC, Op_RegI, 10, r10->as_VMReg()->next());
 115 
 116 reg_def R11  (SOC, SOC, Op_RegI, 11, r11->as_VMReg());
 117 reg_def R11_H(SOC, SOC, Op_RegI, 11, r11->as_VMReg()->next());
 118 
 119 reg_def R12  (SOC, SOE, Op_RegI, 12, r12->as_VMReg());
 120 reg_def R12_H(SOC, SOE, Op_RegI, 12, r12->as_VMReg()->next());
 121 
 122 reg_def R13  (SOC, SOE, Op_RegI, 13, r13->as_VMReg());
 123 reg_def R13_H(SOC, SOE, Op_RegI, 13, r13->as_VMReg()->next());
 124 
 125 reg_def R14  (SOC, SOE, Op_RegI, 14, r14->as_VMReg());
 126 reg_def R14_H(SOC, SOE, Op_RegI, 14, r14->as_VMReg()->next());
 127 
 128 reg_def R15  (SOC, SOE, Op_RegI, 15, r15->as_VMReg());
 129 reg_def R15_H(SOC, SOE, Op_RegI, 15, r15->as_VMReg()->next());
 130 
 131 
 132 // Floating Point Registers
 133 
 134 // XMM registers.  128-bit registers or 4 words each, labeled (a)-d.
 135 // Word a in each register holds a Float, words ab hold a Double.  We
 136 // currently do not use the SIMD capabilities, so registers cd are
 137 // unused at the moment.
 138 // XMM8-XMM15 must be encoded with REX.
 139 // Linux ABI:   No register preserved across function calls
 140 //              XMM0-XMM7 might hold parameters
 141 // Windows ABI: XMM6-XMM15 preserved across function calls
 142 //              XMM0-XMM3 might hold parameters
 143 
 144 reg_def XMM0   (SOC, SOC, Op_RegF,  0, xmm0->as_VMReg());
 145 reg_def XMM0_H (SOC, SOC, Op_RegF,  0, xmm0->as_VMReg()->next());
 146 
 147 reg_def XMM1   (SOC, SOC, Op_RegF,  1, xmm1->as_VMReg());
 148 reg_def XMM1_H (SOC, SOC, Op_RegF,  1, xmm1->as_VMReg()->next());
 149 
 150 reg_def XMM2   (SOC, SOC, Op_RegF,  2, xmm2->as_VMReg());
 151 reg_def XMM2_H (SOC, SOC, Op_RegF,  2, xmm2->as_VMReg()->next());
 152 
 153 reg_def XMM3   (SOC, SOC, Op_RegF,  3, xmm3->as_VMReg());
 154 reg_def XMM3_H (SOC, SOC, Op_RegF,  3, xmm3->as_VMReg()->next());
 155 
 156 reg_def XMM4   (SOC, SOC, Op_RegF,  4, xmm4->as_VMReg());
 157 reg_def XMM4_H (SOC, SOC, Op_RegF,  4, xmm4->as_VMReg()->next());
 158 
 159 reg_def XMM5   (SOC, SOC, Op_RegF,  5, xmm5->as_VMReg());
 160 reg_def XMM5_H (SOC, SOC, Op_RegF,  5, xmm5->as_VMReg()->next());
 161 
 162 #ifdef _WIN64
 163 
 164 reg_def XMM6   (SOC, SOE, Op_RegF,  6, xmm6->as_VMReg());
 165 reg_def XMM6_H (SOC, SOE, Op_RegF,  6, xmm6->as_VMReg()->next());
 166 
 167 reg_def XMM7   (SOC, SOE, Op_RegF,  7, xmm7->as_VMReg());
 168 reg_def XMM7_H (SOC, SOE, Op_RegF,  7, xmm7->as_VMReg()->next());
 169 
 170 reg_def XMM8   (SOC, SOE, Op_RegF,  8, xmm8->as_VMReg());
 171 reg_def XMM8_H (SOC, SOE, Op_RegF,  8, xmm8->as_VMReg()->next());
 172 
 173 reg_def XMM9   (SOC, SOE, Op_RegF,  9, xmm9->as_VMReg());
 174 reg_def XMM9_H (SOC, SOE, Op_RegF,  9, xmm9->as_VMReg()->next());
 175 
 176 reg_def XMM10  (SOC, SOE, Op_RegF, 10, xmm10->as_VMReg());
 177 reg_def XMM10_H(SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()->next());
 178 
 179 reg_def XMM11  (SOC, SOE, Op_RegF, 11, xmm11->as_VMReg());
 180 reg_def XMM11_H(SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()->next());
 181 
 182 reg_def XMM12  (SOC, SOE, Op_RegF, 12, xmm12->as_VMReg());
 183 reg_def XMM12_H(SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()->next());
 184 
 185 reg_def XMM13  (SOC, SOE, Op_RegF, 13, xmm13->as_VMReg());
 186 reg_def XMM13_H(SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()->next());
 187 
 188 reg_def XMM14  (SOC, SOE, Op_RegF, 14, xmm14->as_VMReg());
 189 reg_def XMM14_H(SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()->next());
 190 
 191 reg_def XMM15  (SOC, SOE, Op_RegF, 15, xmm15->as_VMReg());
 192 reg_def XMM15_H(SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()->next());
 193 
 194 #else
 195 
 196 reg_def XMM6   (SOC, SOC, Op_RegF,  6, xmm6->as_VMReg());
 197 reg_def XMM6_H (SOC, SOC, Op_RegF,  6, xmm6->as_VMReg()->next());
 198 
 199 reg_def XMM7   (SOC, SOC, Op_RegF,  7, xmm7->as_VMReg());
 200 reg_def XMM7_H (SOC, SOC, Op_RegF,  7, xmm7->as_VMReg()->next());
 201 
 202 reg_def XMM8   (SOC, SOC, Op_RegF,  8, xmm8->as_VMReg());
 203 reg_def XMM8_H (SOC, SOC, Op_RegF,  8, xmm8->as_VMReg()->next());
 204 
 205 reg_def XMM9   (SOC, SOC, Op_RegF,  9, xmm9->as_VMReg());
 206 reg_def XMM9_H (SOC, SOC, Op_RegF,  9, xmm9->as_VMReg()->next());
 207 
 208 reg_def XMM10  (SOC, SOC, Op_RegF, 10, xmm10->as_VMReg());
 209 reg_def XMM10_H(SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next());
 210 
 211 reg_def XMM11  (SOC, SOC, Op_RegF, 11, xmm11->as_VMReg());
 212 reg_def XMM11_H(SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next());
 213 
 214 reg_def XMM12  (SOC, SOC, Op_RegF, 12, xmm12->as_VMReg());
 215 reg_def XMM12_H(SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next());
 216 
 217 reg_def XMM13  (SOC, SOC, Op_RegF, 13, xmm13->as_VMReg());
 218 reg_def XMM13_H(SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next());
 219 
 220 reg_def XMM14  (SOC, SOC, Op_RegF, 14, xmm14->as_VMReg());
 221 reg_def XMM14_H(SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next());
 222 
 223 reg_def XMM15  (SOC, SOC, Op_RegF, 15, xmm15->as_VMReg());
 224 reg_def XMM15_H(SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next());
 225 
 226 #endif // _WIN64
 227 
 228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
 229 
 230 // Specify priority of register selection within phases of register
 231 // allocation.  Highest priority is first.  A useful heuristic is to
 232 // give registers a low priority when they are required by machine
 233 // instructions, like EAX and EDX on I486, and choose no-save registers
 234 // before save-on-call, & save-on-call before save-on-entry.  Registers
 235 // which participate in fixed calling sequences should come last.
 236 // Registers which are used as pairs must fall on an even boundary.
 237 
 238 alloc_class chunk0(R10,         R10_H,
 239                    R11,         R11_H,
 240                    R8,          R8_H,
 241                    R9,          R9_H,
 242                    R12,         R12_H,
 243                    RCX,         RCX_H,
 244                    RBX,         RBX_H,
 245                    RDI,         RDI_H,
 246                    RDX,         RDX_H,
 247                    RSI,         RSI_H,
 248                    RAX,         RAX_H,
 249                    RBP,         RBP_H,
 250                    R13,         R13_H,
 251                    R14,         R14_H,
 252                    R15,         R15_H,
 253                    RSP,         RSP_H);
 254 
 255 // XXX probably use 8-15 first on Linux
 256 alloc_class chunk1(XMM0,  XMM0_H,
 257                    XMM1,  XMM1_H,
 258                    XMM2,  XMM2_H,
 259                    XMM3,  XMM3_H,
 260                    XMM4,  XMM4_H,
 261                    XMM5,  XMM5_H,
 262                    XMM6,  XMM6_H,
 263                    XMM7,  XMM7_H,
 264                    XMM8,  XMM8_H,
 265                    XMM9,  XMM9_H,
 266                    XMM10, XMM10_H,
 267                    XMM11, XMM11_H,
 268                    XMM12, XMM12_H,
 269                    XMM13, XMM13_H,
 270                    XMM14, XMM14_H,
 271                    XMM15, XMM15_H);
 272 
 273 alloc_class chunk2(RFLAGS);
 274 
 275 
 276 //----------Architecture Description Register Classes--------------------------
 277 // Several register classes are automatically defined based upon information in
 278 // this architecture description.
 279 // 1) reg_class inline_cache_reg           ( /* as def'd in frame section */ )
 280 // 2) reg_class compiler_method_oop_reg    ( /* as def'd in frame section */ )
 281 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
 282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
 283 //
 284 
 285 // Class for all pointer registers (including RSP)
 286 reg_class any_reg(RAX, RAX_H,
 287                   RDX, RDX_H,
 288                   RBP, RBP_H,
 289                   RDI, RDI_H,
 290                   RSI, RSI_H,
 291                   RCX, RCX_H,
 292                   RBX, RBX_H,
 293                   RSP, RSP_H,
 294                   R8,  R8_H,
 295                   R9,  R9_H,
 296                   R10, R10_H,
 297                   R11, R11_H,
 298                   R12, R12_H,
 299                   R13, R13_H,
 300                   R14, R14_H,
 301                   R15, R15_H);
 302 
 303 // Class for all pointer registers except RSP
 304 reg_class ptr_reg(RAX, RAX_H,
 305                   RDX, RDX_H,
 306                   RBP, RBP_H,
 307                   RDI, RDI_H,
 308                   RSI, RSI_H,
 309                   RCX, RCX_H,
 310                   RBX, RBX_H,
 311                   R8,  R8_H,
 312                   R9,  R9_H,
 313                   R10, R10_H,
 314                   R11, R11_H,
 315                   R13, R13_H,
 316                   R14, R14_H);
 317 
 318 // Class for all pointer registers except RAX and RSP
 319 reg_class ptr_no_rax_reg(RDX, RDX_H,
 320                          RBP, RBP_H,
 321                          RDI, RDI_H,
 322                          RSI, RSI_H,
 323                          RCX, RCX_H,
 324                          RBX, RBX_H,
 325                          R8,  R8_H,
 326                          R9,  R9_H,
 327                          R10, R10_H,
 328                          R11, R11_H,
 329                          R12, R12_H,
 330                          R13, R13_H,
 331                          R14, R14_H);
 332 
 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);
 346 
 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);
 360 
 361 // Singleton class for RAX pointer register
 362 reg_class ptr_rax_reg(RAX, RAX_H);
 363 
 364 // Singleton class for RBX pointer register
 365 reg_class ptr_rbx_reg(RBX, RBX_H);
 366 
 367 // Singleton class for RSI pointer register
 368 reg_class ptr_rsi_reg(RSI, RSI_H);
 369 
 370 // Singleton class for RDI pointer register
 371 reg_class ptr_rdi_reg(RDI, RDI_H);
 372 
 373 // Singleton class for RBP pointer register
 374 reg_class ptr_rbp_reg(RBP, RBP_H);
 375 
 376 // Singleton class for stack pointer
 377 reg_class ptr_rsp_reg(RSP, RSP_H);
 378 
 379 // Singleton class for TLS pointer
 380 reg_class ptr_r15_reg(R15, R15_H);
 381 
 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);
 396 
 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);
 409 
 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);
 423 
 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);
 437 
 438 // Singleton class for RAX long register
 439 reg_class long_rax_reg(RAX, RAX_H);
 440 
 441 // Singleton class for RCX long register
 442 reg_class long_rcx_reg(RCX, RCX_H);
 443 
 444 // Singleton class for RDX long register
 445 reg_class long_rdx_reg(RDX, RDX_H);
 446 
 447 // Singleton class for R12 long register
 448 reg_class long_r12_reg(R12, R12_H);
 449 
 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);
 464 
 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);
 478 
 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);
 491 
 492 // Singleton class for RAX int register
 493 reg_class int_rax_reg(RAX);
 494 
 495 // Singleton class for RBX int register
 496 reg_class int_rbx_reg(RBX);
 497 
 498 // Singleton class for RCX int register
 499 reg_class int_rcx_reg(RCX);
 500 
 501 // Singleton class for RCX int register
 502 reg_class int_rdx_reg(RDX);
 503 
 504 // Singleton class for RCX int register
 505 reg_class int_rdi_reg(RDI);
 506 
 507 // Singleton class for instruction pointer
 508 // reg_class ip_reg(RIP);
 509 
 510 // Singleton class for condition codes
 511 reg_class int_flags(RFLAGS);
 512 
 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);
 530 
 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 
 549 // XMM6 and XMM7 could be used as temporary registers for long, float and
 550 // double values for SSE2.
 551 reg_class double_reg6( XMM6,XMM6_H );
 552 reg_class double_reg7( XMM7,XMM7_H );
 553 %}
 554 
 555 
 556 //----------SOURCE BLOCK-------------------------------------------------------
 557 // This is a block of C++ code which provides values, functions, and
 558 // definitions necessary in the rest of the architecture description
 559 source %{
 560 #define   RELOC_IMM64    Assembler::imm_operand
 561 #define   RELOC_DISP32   Assembler::disp32_operand
 562 
 563 #define __ _masm.
 564 
 565 // !!!!! Special hack to get all types of calls to specify the byte offset
 566 //       from the start of the call to the point where the return address
 567 //       will point.
 568 int MachCallStaticJavaNode::ret_addr_offset()
 569 {
 570   return 5; // 5 bytes from start of call to where return address points
 571 }
 572 
 573 int MachCallDynamicJavaNode::ret_addr_offset()
 574 {
 575   return 15; // 15 bytes from start of call to where return address points
 576 }
 577 
 578 // In os_cpu .ad file
 579 // int MachCallRuntimeNode::ret_addr_offset()
 580 
 581 // Indicate if the safepoint node needs the polling page as an input.
 582 // Since amd64 does not have absolute addressing but RIP-relative
 583 // addressing and the polling page is within 2G, it doesn't.
 584 bool SafePointNode::needs_polling_address_input()
 585 {
 586   return false;
 587 }
 588 
 589 //
 590 // Compute padding required for nodes which need alignment
 591 //
 592 
 593 // The address of the call instruction needs to be 4-byte aligned to
 594 // ensure that it does not span a cache line so that it can be patched.
 595 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
 596 {
 597   current_offset += 1; // skip call opcode byte
 598   return round_to(current_offset, alignment_required()) - current_offset;
 599 }
 600 
 601 // The address of the call instruction needs to be 4-byte aligned to
 602 // ensure that it does not span a cache line so that it can be patched.
 603 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
 604 {
 605   current_offset += 11; // skip movq instruction + call opcode byte
 606   return round_to(current_offset, alignment_required()) - current_offset;
 607 }
 608 
 609 #ifndef PRODUCT
 610 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
 611 {
 612   st->print("INT3");
 613 }
 614 #endif
 615 
 616 // EMIT_RM()
 617 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3)
 618 {
 619   unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
 620   *(cbuf.code_end()) = c;
 621   cbuf.set_code_end(cbuf.code_end() + 1);
 622 }
 623 
 624 // EMIT_CC()
 625 void emit_cc(CodeBuffer &cbuf, int f1, int f2)
 626 {
 627   unsigned char c = (unsigned char) (f1 | f2);
 628   *(cbuf.code_end()) = c;
 629   cbuf.set_code_end(cbuf.code_end() + 1);
 630 }
 631 
 632 // EMIT_OPCODE()
 633 void emit_opcode(CodeBuffer &cbuf, int code)
 634 {
 635   *(cbuf.code_end()) = (unsigned char) code;
 636   cbuf.set_code_end(cbuf.code_end() + 1);
 637 }
 638 
 639 // EMIT_OPCODE() w/ relocation information
 640 void emit_opcode(CodeBuffer &cbuf,
 641                  int code, relocInfo::relocType reloc, int offset, int format)
 642 {
 643   cbuf.relocate(cbuf.inst_mark() + offset, reloc, format);
 644   emit_opcode(cbuf, code);
 645 }
 646 
 647 // EMIT_D8()
 648 void emit_d8(CodeBuffer &cbuf, int d8)
 649 {
 650   *(cbuf.code_end()) = (unsigned char) d8;
 651   cbuf.set_code_end(cbuf.code_end() + 1);
 652 }
 653 
 654 // EMIT_D16()
 655 void emit_d16(CodeBuffer &cbuf, int d16)
 656 {
 657   *((short *)(cbuf.code_end())) = d16;
 658   cbuf.set_code_end(cbuf.code_end() + 2);
 659 }
 660 
 661 // EMIT_D32()
 662 void emit_d32(CodeBuffer &cbuf, int d32)
 663 {
 664   *((int *)(cbuf.code_end())) = d32;
 665   cbuf.set_code_end(cbuf.code_end() + 4);
 666 }
 667 
 668 // EMIT_D64()
 669 void emit_d64(CodeBuffer &cbuf, int64_t d64)
 670 {
 671   *((int64_t*) (cbuf.code_end())) = d64;
 672   cbuf.set_code_end(cbuf.code_end() + 8);
 673 }
 674 
 675 // emit 32 bit value and construct relocation entry from relocInfo::relocType
 676 void emit_d32_reloc(CodeBuffer& cbuf,
 677                     int d32,
 678                     relocInfo::relocType reloc,
 679                     int format)
 680 {
 681   assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
 682   cbuf.relocate(cbuf.inst_mark(), reloc, format);
 683 
 684   *((int*) (cbuf.code_end())) = d32;
 685   cbuf.set_code_end(cbuf.code_end() + 4);
 686 }
 687 
 688 // emit 32 bit value and construct relocation entry from RelocationHolder
 689 void emit_d32_reloc(CodeBuffer& cbuf,
 690                     int d32,
 691                     RelocationHolder const& rspec,
 692                     int format)
 693 {
 694 #ifdef ASSERT
 695   if (rspec.reloc()->type() == relocInfo::oop_type &&
 696       d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
 697     assert(oop((intptr_t)d32)->is_oop() && oop((intptr_t)d32)->is_perm(), "cannot embed non-perm oops in code");
 698   }
 699 #endif
 700   cbuf.relocate(cbuf.inst_mark(), rspec, format);
 701 
 702   *((int* )(cbuf.code_end())) = d32;
 703   cbuf.set_code_end(cbuf.code_end() + 4);
 704 }
 705 
 706 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
 707   address next_ip = cbuf.code_end() + 4;
 708   emit_d32_reloc(cbuf, (int) (addr - next_ip),
 709                  external_word_Relocation::spec(addr),
 710                  RELOC_DISP32);
 711 }
 712 
 713 
 714 // emit 64 bit value and construct relocation entry from relocInfo::relocType
 715 void emit_d64_reloc(CodeBuffer& cbuf,
 716                     int64_t d64,
 717                     relocInfo::relocType reloc,
 718                     int format)
 719 {
 720   cbuf.relocate(cbuf.inst_mark(), reloc, format);
 721 
 722   *((int64_t*) (cbuf.code_end())) = d64;
 723   cbuf.set_code_end(cbuf.code_end() + 8);
 724 }
 725 
 726 // emit 64 bit value and construct relocation entry from RelocationHolder
 727 void emit_d64_reloc(CodeBuffer& cbuf,
 728                     int64_t d64,
 729                     RelocationHolder const& rspec,
 730                     int format)
 731 {
 732 #ifdef ASSERT
 733   if (rspec.reloc()->type() == relocInfo::oop_type &&
 734       d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
 735     assert(oop(d64)->is_oop() && oop(d64)->is_perm(),
 736            "cannot embed non-perm oops in code");
 737   }
 738 #endif
 739   cbuf.relocate(cbuf.inst_mark(), rspec, format);
 740 
 741   *((int64_t*) (cbuf.code_end())) = d64;
 742   cbuf.set_code_end(cbuf.code_end() + 8);
 743 }
 744 
 745 // Access stack slot for load or store
 746 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
 747 {
 748   emit_opcode(cbuf, opcode);                  // (e.g., FILD   [RSP+src])
 749   if (-0x80 <= disp && disp < 0x80) {
 750     emit_rm(cbuf, 0x01, rm_field, RSP_enc);   // R/M byte
 751     emit_rm(cbuf, 0x00, RSP_enc, RSP_enc);    // SIB byte
 752     emit_d8(cbuf, disp);     // Displacement  // R/M byte
 753   } else {
 754     emit_rm(cbuf, 0x02, rm_field, RSP_enc);   // R/M byte
 755     emit_rm(cbuf, 0x00, RSP_enc, RSP_enc);    // SIB byte
 756     emit_d32(cbuf, disp);     // Displacement // R/M byte
 757   }
 758 }
 759 
 760    // rRegI ereg, memory mem) %{    // emit_reg_mem
 761 void encode_RegMem(CodeBuffer &cbuf,
 762                    int reg,
 763                    int base, int index, int scale, int disp, bool disp_is_oop)
 764 {
 765   assert(!disp_is_oop, "cannot have disp");
 766   int regenc = reg & 7;
 767   int baseenc = base & 7;
 768   int indexenc = index & 7;
 769 
 770   // There is no index & no scale, use form without SIB byte
 771   if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
 772     // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
 773     if (disp == 0 && base != RBP_enc && base != R13_enc) {
 774       emit_rm(cbuf, 0x0, regenc, baseenc); // *
 775     } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
 776       // If 8-bit displacement, mode 0x1
 777       emit_rm(cbuf, 0x1, regenc, baseenc); // *
 778       emit_d8(cbuf, disp);
 779     } else {
 780       // If 32-bit displacement
 781       if (base == -1) { // Special flag for absolute address
 782         emit_rm(cbuf, 0x0, regenc, 0x5); // *
 783         if (disp_is_oop) {
 784           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
 785         } else {
 786           emit_d32(cbuf, disp);
 787         }
 788       } else {
 789         // Normal base + offset
 790         emit_rm(cbuf, 0x2, regenc, baseenc); // *
 791         if (disp_is_oop) {
 792           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
 793         } else {
 794           emit_d32(cbuf, disp);
 795         }
 796       }
 797     }
 798   } else {
 799     // Else, encode with the SIB byte
 800     // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
 801     if (disp == 0 && base != RBP_enc && base != R13_enc) {
 802       // If no displacement
 803       emit_rm(cbuf, 0x0, regenc, 0x4); // *
 804       emit_rm(cbuf, scale, indexenc, baseenc);
 805     } else {
 806       if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
 807         // If 8-bit displacement, mode 0x1
 808         emit_rm(cbuf, 0x1, regenc, 0x4); // *
 809         emit_rm(cbuf, scale, indexenc, baseenc);
 810         emit_d8(cbuf, disp);
 811       } else {
 812         // If 32-bit displacement
 813         if (base == 0x04 ) {
 814           emit_rm(cbuf, 0x2, regenc, 0x4);
 815           emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
 816         } else {
 817           emit_rm(cbuf, 0x2, regenc, 0x4);
 818           emit_rm(cbuf, scale, indexenc, baseenc); // *
 819         }
 820         if (disp_is_oop) {
 821           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
 822         } else {
 823           emit_d32(cbuf, disp);
 824         }
 825       }
 826     }
 827   }
 828 }
 829 
 830 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
 831 {
 832   if (dstenc != srcenc) {
 833     if (dstenc < 8) {
 834       if (srcenc >= 8) {
 835         emit_opcode(cbuf, Assembler::REX_B);
 836         srcenc -= 8;
 837       }
 838     } else {
 839       if (srcenc < 8) {
 840         emit_opcode(cbuf, Assembler::REX_R);
 841       } else {
 842         emit_opcode(cbuf, Assembler::REX_RB);
 843         srcenc -= 8;
 844       }
 845       dstenc -= 8;
 846     }
 847 
 848     emit_opcode(cbuf, 0x8B);
 849     emit_rm(cbuf, 0x3, dstenc, srcenc);
 850   }
 851 }
 852 
 853 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
 854   if( dst_encoding == src_encoding ) {
 855     // reg-reg copy, use an empty encoding
 856   } else {
 857     MacroAssembler _masm(&cbuf);
 858 
 859     __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
 860   }
 861 }
 862 
 863 
 864 //=============================================================================
 865 #ifndef PRODUCT
 866 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
 867 {
 868   Compile* C = ra_->C;
 869 
 870   int framesize = C->frame_slots() << LogBytesPerInt;
 871   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 872   // Remove wordSize for return adr already pushed
 873   // and another for the RBP we are going to save
 874   framesize -= 2*wordSize;
 875   bool need_nop = true;
 876 
 877   // Calls to C2R adapters often do not accept exceptional returns.
 878   // We require that their callers must bang for them.  But be
 879   // careful, because some VM calls (such as call site linkage) can
 880   // use several kilobytes of stack.  But the stack safety zone should
 881   // account for that.  See bugs 4446381, 4468289, 4497237.
 882   if (C->need_stack_bang(framesize)) {
 883     st->print_cr("# stack bang"); st->print("\t");
 884     need_nop = false;
 885   }
 886   st->print_cr("pushq   rbp"); st->print("\t");
 887 
 888   if (VerifyStackAtCalls) {
 889     // Majik cookie to verify stack depth
 890     st->print_cr("pushq   0xffffffffbadb100d"
 891                   "\t# Majik cookie for stack depth check");
 892     st->print("\t");
 893     framesize -= wordSize; // Remove 2 for cookie
 894     need_nop = false;
 895   }
 896 
 897   if (framesize) {
 898     st->print("subq    rsp, #%d\t# Create frame", framesize);
 899     if (framesize < 0x80 && need_nop) {
 900       st->print("\n\tnop\t# nop for patch_verified_entry");
 901     }
 902   }
 903 }
 904 #endif
 905 
 906 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
 907 {
 908   Compile* C = ra_->C;
 909 
 910   // WARNING: Initial instruction MUST be 5 bytes or longer so that
 911   // NativeJump::patch_verified_entry will be able to patch out the entry
 912   // code safely. The fldcw is ok at 6 bytes, the push to verify stack
 913   // depth is ok at 5 bytes, the frame allocation can be either 3 or
 914   // 6 bytes. So if we don't do the fldcw or the push then we must
 915   // use the 6 byte frame allocation even if we have no frame. :-(
 916   // If method sets FPU control word do it now
 917 
 918   int framesize = C->frame_slots() << LogBytesPerInt;
 919   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 920   // Remove wordSize for return adr already pushed
 921   // and another for the RBP we are going to save
 922   framesize -= 2*wordSize;
 923   bool need_nop = true;
 924 
 925   // Calls to C2R adapters often do not accept exceptional returns.
 926   // We require that their callers must bang for them.  But be
 927   // careful, because some VM calls (such as call site linkage) can
 928   // use several kilobytes of stack.  But the stack safety zone should
 929   // account for that.  See bugs 4446381, 4468289, 4497237.
 930   if (C->need_stack_bang(framesize)) {
 931     MacroAssembler masm(&cbuf);
 932     masm.generate_stack_overflow_check(framesize);
 933     need_nop = false;
 934   }
 935 
 936   // We always push rbp so that on return to interpreter rbp will be
 937   // restored correctly and we can correct the stack.
 938   emit_opcode(cbuf, 0x50 | RBP_enc);
 939 
 940   if (VerifyStackAtCalls) {
 941     // Majik cookie to verify stack depth
 942     emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
 943     emit_d32(cbuf, 0xbadb100d);
 944     framesize -= wordSize; // Remove 2 for cookie
 945     need_nop = false;
 946   }
 947 
 948   if (framesize) {
 949     emit_opcode(cbuf, Assembler::REX_W);
 950     if (framesize < 0x80) {
 951       emit_opcode(cbuf, 0x83);   // sub  SP,#framesize
 952       emit_rm(cbuf, 0x3, 0x05, RSP_enc);
 953       emit_d8(cbuf, framesize);
 954       if (need_nop) {
 955         emit_opcode(cbuf, 0x90); // nop
 956       }
 957     } else {
 958       emit_opcode(cbuf, 0x81);   // sub  SP,#framesize
 959       emit_rm(cbuf, 0x3, 0x05, RSP_enc);
 960       emit_d32(cbuf, framesize);
 961     }
 962   }
 963 
 964   C->set_frame_complete(cbuf.code_end() - cbuf.code_begin());
 965 
 966 #ifdef ASSERT
 967   if (VerifyStackAtCalls) {
 968     Label L;
 969     MacroAssembler masm(&cbuf);
 970     masm.push(rax);
 971     masm.mov(rax, rsp);
 972     masm.andptr(rax, StackAlignmentInBytes-1);
 973     masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
 974     masm.pop(rax);
 975     masm.jcc(Assembler::equal, L);
 976     masm.stop("Stack is not properly aligned!");
 977     masm.bind(L);
 978   }
 979 #endif
 980 }
 981 
 982 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
 983 {
 984   return MachNode::size(ra_); // too many variables; just compute it
 985                               // the hard way
 986 }
 987 
 988 int MachPrologNode::reloc() const
 989 {
 990   return 0; // a large enough number
 991 }
 992 
 993 //=============================================================================
 994 #ifndef PRODUCT
 995 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
 996 {
 997   Compile* C = ra_->C;
 998   int framesize = C->frame_slots() << LogBytesPerInt;
 999   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1000   // Remove word for return adr already pushed
1001   // and RBP
1002   framesize -= 2*wordSize;
1003 
1004   if (framesize) {
1005     st->print_cr("addq\trsp, %d\t# Destroy frame", framesize);
1006     st->print("\t");
1007   }
1008 
1009   st->print_cr("popq\trbp");
1010   if (do_polling() && C->is_method_compilation()) {
1011     st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t"
1012                   "# Safepoint: poll for GC");
1013     st->print("\t");
1014   }
1015 }
1016 #endif
1017 
1018 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1019 {
1020   Compile* C = ra_->C;
1021   int framesize = C->frame_slots() << LogBytesPerInt;
1022   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1023   // Remove word for return adr already pushed
1024   // and RBP
1025   framesize -= 2*wordSize;
1026 
1027   // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1028 
1029   if (framesize) {
1030     emit_opcode(cbuf, Assembler::REX_W);
1031     if (framesize < 0x80) {
1032       emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1033       emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1034       emit_d8(cbuf, framesize);
1035     } else {
1036       emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1037       emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1038       emit_d32(cbuf, framesize);
1039     }
1040   }
1041 
1042   // popq rbp
1043   emit_opcode(cbuf, 0x58 | RBP_enc);
1044 
1045   if (do_polling() && C->is_method_compilation()) {
1046     // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
1047     // XXX reg_mem doesn't support RIP-relative addressing yet
1048     cbuf.set_inst_mark();
1049     cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX
1050     emit_opcode(cbuf, 0x85); // testl
1051     emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
1052     // cbuf.inst_mark() is beginning of instruction
1053     emit_d32_reloc(cbuf, os::get_polling_page());
1054 //                    relocInfo::poll_return_type,
1055   }
1056 }
1057 
1058 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1059 {
1060   Compile* C = ra_->C;
1061   int framesize = C->frame_slots() << LogBytesPerInt;
1062   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1063   // Remove word for return adr already pushed
1064   // and RBP
1065   framesize -= 2*wordSize;
1066 
1067   uint size = 0;
1068 
1069   if (do_polling() && C->is_method_compilation()) {
1070     size += 6;
1071   }
1072 
1073   // count popq rbp
1074   size++;
1075 
1076   if (framesize) {
1077     if (framesize < 0x80) {
1078       size += 4;
1079     } else if (framesize) {
1080       size += 7;
1081     }
1082   }
1083 
1084   return size;
1085 }
1086 
1087 int MachEpilogNode::reloc() const
1088 {
1089   return 2; // a large enough number
1090 }
1091 
1092 const Pipeline* MachEpilogNode::pipeline() const
1093 {
1094   return MachNode::pipeline_class();
1095 }
1096 
1097 int MachEpilogNode::safepoint_offset() const
1098 {
1099   return 0;
1100 }
1101 
1102 //=============================================================================
1103 
1104 enum RC {
1105   rc_bad,
1106   rc_int,
1107   rc_float,
1108   rc_stack
1109 };
1110 
1111 static enum RC rc_class(OptoReg::Name reg)
1112 {
1113   if( !OptoReg::is_valid(reg)  ) return rc_bad;
1114 
1115   if (OptoReg::is_stack(reg)) return rc_stack;
1116 
1117   VMReg r = OptoReg::as_VMReg(reg);
1118 
1119   if (r->is_Register()) return rc_int;
1120 
1121   assert(r->is_XMMRegister(), "must be");
1122   return rc_float;
1123 }
1124 
1125 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1126                                        PhaseRegAlloc* ra_,
1127                                        bool do_size,
1128                                        outputStream* st) const
1129 {
1130 
1131   // Get registers to move
1132   OptoReg::Name src_second = ra_->get_reg_second(in(1));
1133   OptoReg::Name src_first = ra_->get_reg_first(in(1));
1134   OptoReg::Name dst_second = ra_->get_reg_second(this);
1135   OptoReg::Name dst_first = ra_->get_reg_first(this);
1136 
1137   enum RC src_second_rc = rc_class(src_second);
1138   enum RC src_first_rc = rc_class(src_first);
1139   enum RC dst_second_rc = rc_class(dst_second);
1140   enum RC dst_first_rc = rc_class(dst_first);
1141 
1142   assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1143          "must move at least 1 register" );
1144 
1145   if (src_first == dst_first && src_second == dst_second) {
1146     // Self copy, no move
1147     return 0;
1148   } else if (src_first_rc == rc_stack) {
1149     // mem ->
1150     if (dst_first_rc == rc_stack) {
1151       // mem -> mem
1152       assert(src_second != dst_first, "overlap");
1153       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1154           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1155         // 64-bit
1156         int src_offset = ra_->reg2offset(src_first);
1157         int dst_offset = ra_->reg2offset(dst_first);
1158         if (cbuf) {
1159           emit_opcode(*cbuf, 0xFF);
1160           encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1161 
1162           emit_opcode(*cbuf, 0x8F);
1163           encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1164 
1165 #ifndef PRODUCT
1166         } else if (!do_size) {
1167           st->print("pushq   [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1168                      "popq    [rsp + #%d]",
1169                      src_offset,
1170                      dst_offset);
1171 #endif
1172         }
1173         return
1174           3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1175           3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1176       } else {
1177         // 32-bit
1178         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1179         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1180         // No pushl/popl, so:
1181         int src_offset = ra_->reg2offset(src_first);
1182         int dst_offset = ra_->reg2offset(dst_first);
1183         if (cbuf) {
1184           emit_opcode(*cbuf, Assembler::REX_W);
1185           emit_opcode(*cbuf, 0x89);
1186           emit_opcode(*cbuf, 0x44);
1187           emit_opcode(*cbuf, 0x24);
1188           emit_opcode(*cbuf, 0xF8);
1189 
1190           emit_opcode(*cbuf, 0x8B);
1191           encode_RegMem(*cbuf,
1192                         RAX_enc,
1193                         RSP_enc, 0x4, 0, src_offset,
1194                         false);
1195 
1196           emit_opcode(*cbuf, 0x89);
1197           encode_RegMem(*cbuf,
1198                         RAX_enc,
1199                         RSP_enc, 0x4, 0, dst_offset,
1200                         false);
1201 
1202           emit_opcode(*cbuf, Assembler::REX_W);
1203           emit_opcode(*cbuf, 0x8B);
1204           emit_opcode(*cbuf, 0x44);
1205           emit_opcode(*cbuf, 0x24);
1206           emit_opcode(*cbuf, 0xF8);
1207 
1208 #ifndef PRODUCT
1209         } else if (!do_size) {
1210           st->print("movq    [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1211                      "movl    rax, [rsp + #%d]\n\t"
1212                      "movl    [rsp + #%d], rax\n\t"
1213                      "movq    rax, [rsp - #8]",
1214                      src_offset,
1215                      dst_offset);
1216 #endif
1217         }
1218         return
1219           5 + // movq
1220           3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1221           3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1222           5; // movq
1223       }
1224     } else if (dst_first_rc == rc_int) {
1225       // mem -> gpr
1226       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1227           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1228         // 64-bit
1229         int offset = ra_->reg2offset(src_first);
1230         if (cbuf) {
1231           if (Matcher::_regEncode[dst_first] < 8) {
1232             emit_opcode(*cbuf, Assembler::REX_W);
1233           } else {
1234             emit_opcode(*cbuf, Assembler::REX_WR);
1235           }
1236           emit_opcode(*cbuf, 0x8B);
1237           encode_RegMem(*cbuf,
1238                         Matcher::_regEncode[dst_first],
1239                         RSP_enc, 0x4, 0, offset,
1240                         false);
1241 #ifndef PRODUCT
1242         } else if (!do_size) {
1243           st->print("movq    %s, [rsp + #%d]\t# spill",
1244                      Matcher::regName[dst_first],
1245                      offset);
1246 #endif
1247         }
1248         return
1249           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1250       } else {
1251         // 32-bit
1252         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1253         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1254         int offset = ra_->reg2offset(src_first);
1255         if (cbuf) {
1256           if (Matcher::_regEncode[dst_first] >= 8) {
1257             emit_opcode(*cbuf, Assembler::REX_R);
1258           }
1259           emit_opcode(*cbuf, 0x8B);
1260           encode_RegMem(*cbuf,
1261                         Matcher::_regEncode[dst_first],
1262                         RSP_enc, 0x4, 0, offset,
1263                         false);
1264 #ifndef PRODUCT
1265         } else if (!do_size) {
1266           st->print("movl    %s, [rsp + #%d]\t# spill",
1267                      Matcher::regName[dst_first],
1268                      offset);
1269 #endif
1270         }
1271         return
1272           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1273           ((Matcher::_regEncode[dst_first] < 8)
1274            ? 3
1275            : 4); // REX
1276       }
1277     } else if (dst_first_rc == rc_float) {
1278       // mem-> xmm
1279       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1280           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1281         // 64-bit
1282         int offset = ra_->reg2offset(src_first);
1283         if (cbuf) {
1284           emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1285           if (Matcher::_regEncode[dst_first] >= 8) {
1286             emit_opcode(*cbuf, Assembler::REX_R);
1287           }
1288           emit_opcode(*cbuf, 0x0F);
1289           emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1290           encode_RegMem(*cbuf,
1291                         Matcher::_regEncode[dst_first],
1292                         RSP_enc, 0x4, 0, offset,
1293                         false);
1294 #ifndef PRODUCT
1295         } else if (!do_size) {
1296           st->print("%s  %s, [rsp + #%d]\t# spill",
1297                      UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1298                      Matcher::regName[dst_first],
1299                      offset);
1300 #endif
1301         }
1302         return
1303           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1304           ((Matcher::_regEncode[dst_first] < 8)
1305            ? 5
1306            : 6); // REX
1307       } else {
1308         // 32-bit
1309         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1310         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1311         int offset = ra_->reg2offset(src_first);
1312         if (cbuf) {
1313           emit_opcode(*cbuf, 0xF3);
1314           if (Matcher::_regEncode[dst_first] >= 8) {
1315             emit_opcode(*cbuf, Assembler::REX_R);
1316           }
1317           emit_opcode(*cbuf, 0x0F);
1318           emit_opcode(*cbuf, 0x10);
1319           encode_RegMem(*cbuf,
1320                         Matcher::_regEncode[dst_first],
1321                         RSP_enc, 0x4, 0, offset,
1322                         false);
1323 #ifndef PRODUCT
1324         } else if (!do_size) {
1325           st->print("movss   %s, [rsp + #%d]\t# spill",
1326                      Matcher::regName[dst_first],
1327                      offset);
1328 #endif
1329         }
1330         return
1331           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1332           ((Matcher::_regEncode[dst_first] < 8)
1333            ? 5
1334            : 6); // REX
1335       }
1336     }
1337   } else if (src_first_rc == rc_int) {
1338     // gpr ->
1339     if (dst_first_rc == rc_stack) {
1340       // gpr -> mem
1341       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1342           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1343         // 64-bit
1344         int offset = ra_->reg2offset(dst_first);
1345         if (cbuf) {
1346           if (Matcher::_regEncode[src_first] < 8) {
1347             emit_opcode(*cbuf, Assembler::REX_W);
1348           } else {
1349             emit_opcode(*cbuf, Assembler::REX_WR);
1350           }
1351           emit_opcode(*cbuf, 0x89);
1352           encode_RegMem(*cbuf,
1353                         Matcher::_regEncode[src_first],
1354                         RSP_enc, 0x4, 0, offset,
1355                         false);
1356 #ifndef PRODUCT
1357         } else if (!do_size) {
1358           st->print("movq    [rsp + #%d], %s\t# spill",
1359                      offset,
1360                      Matcher::regName[src_first]);
1361 #endif
1362         }
1363         return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1364       } else {
1365         // 32-bit
1366         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1367         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1368         int offset = ra_->reg2offset(dst_first);
1369         if (cbuf) {
1370           if (Matcher::_regEncode[src_first] >= 8) {
1371             emit_opcode(*cbuf, Assembler::REX_R);
1372           }
1373           emit_opcode(*cbuf, 0x89);
1374           encode_RegMem(*cbuf,
1375                         Matcher::_regEncode[src_first],
1376                         RSP_enc, 0x4, 0, offset,
1377                         false);
1378 #ifndef PRODUCT
1379         } else if (!do_size) {
1380           st->print("movl    [rsp + #%d], %s\t# spill",
1381                      offset,
1382                      Matcher::regName[src_first]);
1383 #endif
1384         }
1385         return
1386           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1387           ((Matcher::_regEncode[src_first] < 8)
1388            ? 3
1389            : 4); // REX
1390       }
1391     } else if (dst_first_rc == rc_int) {
1392       // gpr -> gpr
1393       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1394           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1395         // 64-bit
1396         if (cbuf) {
1397           if (Matcher::_regEncode[dst_first] < 8) {
1398             if (Matcher::_regEncode[src_first] < 8) {
1399               emit_opcode(*cbuf, Assembler::REX_W);
1400             } else {
1401               emit_opcode(*cbuf, Assembler::REX_WB);
1402             }
1403           } else {
1404             if (Matcher::_regEncode[src_first] < 8) {
1405               emit_opcode(*cbuf, Assembler::REX_WR);
1406             } else {
1407               emit_opcode(*cbuf, Assembler::REX_WRB);
1408             }
1409           }
1410           emit_opcode(*cbuf, 0x8B);
1411           emit_rm(*cbuf, 0x3,
1412                   Matcher::_regEncode[dst_first] & 7,
1413                   Matcher::_regEncode[src_first] & 7);
1414 #ifndef PRODUCT
1415         } else if (!do_size) {
1416           st->print("movq    %s, %s\t# spill",
1417                      Matcher::regName[dst_first],
1418                      Matcher::regName[src_first]);
1419 #endif
1420         }
1421         return 3; // REX
1422       } else {
1423         // 32-bit
1424         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1425         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1426         if (cbuf) {
1427           if (Matcher::_regEncode[dst_first] < 8) {
1428             if (Matcher::_regEncode[src_first] >= 8) {
1429               emit_opcode(*cbuf, Assembler::REX_B);
1430             }
1431           } else {
1432             if (Matcher::_regEncode[src_first] < 8) {
1433               emit_opcode(*cbuf, Assembler::REX_R);
1434             } else {
1435               emit_opcode(*cbuf, Assembler::REX_RB);
1436             }
1437           }
1438           emit_opcode(*cbuf, 0x8B);
1439           emit_rm(*cbuf, 0x3,
1440                   Matcher::_regEncode[dst_first] & 7,
1441                   Matcher::_regEncode[src_first] & 7);
1442 #ifndef PRODUCT
1443         } else if (!do_size) {
1444           st->print("movl    %s, %s\t# spill",
1445                      Matcher::regName[dst_first],
1446                      Matcher::regName[src_first]);
1447 #endif
1448         }
1449         return
1450           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1451           ? 2
1452           : 3; // REX
1453       }
1454     } else if (dst_first_rc == rc_float) {
1455       // gpr -> xmm
1456       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1457           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1458         // 64-bit
1459         if (cbuf) {
1460           emit_opcode(*cbuf, 0x66);
1461           if (Matcher::_regEncode[dst_first] < 8) {
1462             if (Matcher::_regEncode[src_first] < 8) {
1463               emit_opcode(*cbuf, Assembler::REX_W);
1464             } else {
1465               emit_opcode(*cbuf, Assembler::REX_WB);
1466             }
1467           } else {
1468             if (Matcher::_regEncode[src_first] < 8) {
1469               emit_opcode(*cbuf, Assembler::REX_WR);
1470             } else {
1471               emit_opcode(*cbuf, Assembler::REX_WRB);
1472             }
1473           }
1474           emit_opcode(*cbuf, 0x0F);
1475           emit_opcode(*cbuf, 0x6E);
1476           emit_rm(*cbuf, 0x3,
1477                   Matcher::_regEncode[dst_first] & 7,
1478                   Matcher::_regEncode[src_first] & 7);
1479 #ifndef PRODUCT
1480         } else if (!do_size) {
1481           st->print("movdq   %s, %s\t# spill",
1482                      Matcher::regName[dst_first],
1483                      Matcher::regName[src_first]);
1484 #endif
1485         }
1486         return 5; // REX
1487       } else {
1488         // 32-bit
1489         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1490         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1491         if (cbuf) {
1492           emit_opcode(*cbuf, 0x66);
1493           if (Matcher::_regEncode[dst_first] < 8) {
1494             if (Matcher::_regEncode[src_first] >= 8) {
1495               emit_opcode(*cbuf, Assembler::REX_B);
1496             }
1497           } else {
1498             if (Matcher::_regEncode[src_first] < 8) {
1499               emit_opcode(*cbuf, Assembler::REX_R);
1500             } else {
1501               emit_opcode(*cbuf, Assembler::REX_RB);
1502             }
1503           }
1504           emit_opcode(*cbuf, 0x0F);
1505           emit_opcode(*cbuf, 0x6E);
1506           emit_rm(*cbuf, 0x3,
1507                   Matcher::_regEncode[dst_first] & 7,
1508                   Matcher::_regEncode[src_first] & 7);
1509 #ifndef PRODUCT
1510         } else if (!do_size) {
1511           st->print("movdl   %s, %s\t# spill",
1512                      Matcher::regName[dst_first],
1513                      Matcher::regName[src_first]);
1514 #endif
1515         }
1516         return
1517           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1518           ? 4
1519           : 5; // REX
1520       }
1521     }
1522   } else if (src_first_rc == rc_float) {
1523     // xmm ->
1524     if (dst_first_rc == rc_stack) {
1525       // xmm -> mem
1526       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1527           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1528         // 64-bit
1529         int offset = ra_->reg2offset(dst_first);
1530         if (cbuf) {
1531           emit_opcode(*cbuf, 0xF2);
1532           if (Matcher::_regEncode[src_first] >= 8) {
1533               emit_opcode(*cbuf, Assembler::REX_R);
1534           }
1535           emit_opcode(*cbuf, 0x0F);
1536           emit_opcode(*cbuf, 0x11);
1537           encode_RegMem(*cbuf,
1538                         Matcher::_regEncode[src_first],
1539                         RSP_enc, 0x4, 0, offset,
1540                         false);
1541 #ifndef PRODUCT
1542         } else if (!do_size) {
1543           st->print("movsd   [rsp + #%d], %s\t# spill",
1544                      offset,
1545                      Matcher::regName[src_first]);
1546 #endif
1547         }
1548         return
1549           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1550           ((Matcher::_regEncode[src_first] < 8)
1551            ? 5
1552            : 6); // REX
1553       } else {
1554         // 32-bit
1555         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1556         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1557         int offset = ra_->reg2offset(dst_first);
1558         if (cbuf) {
1559           emit_opcode(*cbuf, 0xF3);
1560           if (Matcher::_regEncode[src_first] >= 8) {
1561               emit_opcode(*cbuf, Assembler::REX_R);
1562           }
1563           emit_opcode(*cbuf, 0x0F);
1564           emit_opcode(*cbuf, 0x11);
1565           encode_RegMem(*cbuf,
1566                         Matcher::_regEncode[src_first],
1567                         RSP_enc, 0x4, 0, offset,
1568                         false);
1569 #ifndef PRODUCT
1570         } else if (!do_size) {
1571           st->print("movss   [rsp + #%d], %s\t# spill",
1572                      offset,
1573                      Matcher::regName[src_first]);
1574 #endif
1575         }
1576         return
1577           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1578           ((Matcher::_regEncode[src_first] < 8)
1579            ? 5
1580            : 6); // REX
1581       }
1582     } else if (dst_first_rc == rc_int) {
1583       // xmm -> gpr
1584       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1585           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1586         // 64-bit
1587         if (cbuf) {
1588           emit_opcode(*cbuf, 0x66);
1589           if (Matcher::_regEncode[dst_first] < 8) {
1590             if (Matcher::_regEncode[src_first] < 8) {
1591               emit_opcode(*cbuf, Assembler::REX_W);
1592             } else {
1593               emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1594             }
1595           } else {
1596             if (Matcher::_regEncode[src_first] < 8) {
1597               emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1598             } else {
1599               emit_opcode(*cbuf, Assembler::REX_WRB);
1600             }
1601           }
1602           emit_opcode(*cbuf, 0x0F);
1603           emit_opcode(*cbuf, 0x7E);
1604           emit_rm(*cbuf, 0x3,
1605                   Matcher::_regEncode[dst_first] & 7,
1606                   Matcher::_regEncode[src_first] & 7);
1607 #ifndef PRODUCT
1608         } else if (!do_size) {
1609           st->print("movdq   %s, %s\t# spill",
1610                      Matcher::regName[dst_first],
1611                      Matcher::regName[src_first]);
1612 #endif
1613         }
1614         return 5; // REX
1615       } else {
1616         // 32-bit
1617         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1618         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1619         if (cbuf) {
1620           emit_opcode(*cbuf, 0x66);
1621           if (Matcher::_regEncode[dst_first] < 8) {
1622             if (Matcher::_regEncode[src_first] >= 8) {
1623               emit_opcode(*cbuf, Assembler::REX_R); // attention!
1624             }
1625           } else {
1626             if (Matcher::_regEncode[src_first] < 8) {
1627               emit_opcode(*cbuf, Assembler::REX_B); // attention!
1628             } else {
1629               emit_opcode(*cbuf, Assembler::REX_RB);
1630             }
1631           }
1632           emit_opcode(*cbuf, 0x0F);
1633           emit_opcode(*cbuf, 0x7E);
1634           emit_rm(*cbuf, 0x3,
1635                   Matcher::_regEncode[dst_first] & 7,
1636                   Matcher::_regEncode[src_first] & 7);
1637 #ifndef PRODUCT
1638         } else if (!do_size) {
1639           st->print("movdl   %s, %s\t# spill",
1640                      Matcher::regName[dst_first],
1641                      Matcher::regName[src_first]);
1642 #endif
1643         }
1644         return
1645           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1646           ? 4
1647           : 5; // REX
1648       }
1649     } else if (dst_first_rc == rc_float) {
1650       // xmm -> xmm
1651       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1652           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1653         // 64-bit
1654         if (cbuf) {
1655           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1656           if (Matcher::_regEncode[dst_first] < 8) {
1657             if (Matcher::_regEncode[src_first] >= 8) {
1658               emit_opcode(*cbuf, Assembler::REX_B);
1659             }
1660           } else {
1661             if (Matcher::_regEncode[src_first] < 8) {
1662               emit_opcode(*cbuf, Assembler::REX_R);
1663             } else {
1664               emit_opcode(*cbuf, Assembler::REX_RB);
1665             }
1666           }
1667           emit_opcode(*cbuf, 0x0F);
1668           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1669           emit_rm(*cbuf, 0x3,
1670                   Matcher::_regEncode[dst_first] & 7,
1671                   Matcher::_regEncode[src_first] & 7);
1672 #ifndef PRODUCT
1673         } else if (!do_size) {
1674           st->print("%s  %s, %s\t# spill",
1675                      UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1676                      Matcher::regName[dst_first],
1677                      Matcher::regName[src_first]);
1678 #endif
1679         }
1680         return
1681           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1682           ? 4
1683           : 5; // REX
1684       } else {
1685         // 32-bit
1686         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1687         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1688         if (cbuf) {
1689           if (!UseXmmRegToRegMoveAll)
1690             emit_opcode(*cbuf, 0xF3);
1691           if (Matcher::_regEncode[dst_first] < 8) {
1692             if (Matcher::_regEncode[src_first] >= 8) {
1693               emit_opcode(*cbuf, Assembler::REX_B);
1694             }
1695           } else {
1696             if (Matcher::_regEncode[src_first] < 8) {
1697               emit_opcode(*cbuf, Assembler::REX_R);
1698             } else {
1699               emit_opcode(*cbuf, Assembler::REX_RB);
1700             }
1701           }
1702           emit_opcode(*cbuf, 0x0F);
1703           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1704           emit_rm(*cbuf, 0x3,
1705                   Matcher::_regEncode[dst_first] & 7,
1706                   Matcher::_regEncode[src_first] & 7);
1707 #ifndef PRODUCT
1708         } else if (!do_size) {
1709           st->print("%s  %s, %s\t# spill",
1710                      UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1711                      Matcher::regName[dst_first],
1712                      Matcher::regName[src_first]);
1713 #endif
1714         }
1715         return
1716           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1717           ? (UseXmmRegToRegMoveAll ? 3 : 4)
1718           : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1719       }
1720     }
1721   }
1722 
1723   assert(0," foo ");
1724   Unimplemented();
1725 
1726   return 0;
1727 }
1728 
1729 #ifndef PRODUCT
1730 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1731 {
1732   implementation(NULL, ra_, false, st);
1733 }
1734 #endif
1735 
1736 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1737 {
1738   implementation(&cbuf, ra_, false, NULL);
1739 }
1740 
1741 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1742 {
1743   return implementation(NULL, ra_, true, NULL);
1744 }
1745 
1746 //=============================================================================
1747 #ifndef PRODUCT
1748 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1749 {
1750   st->print("nop \t# %d bytes pad for loops and calls", _count);
1751 }
1752 #endif
1753 
1754 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1755 {
1756   MacroAssembler _masm(&cbuf);
1757   __ nop(_count);
1758 }
1759 
1760 uint MachNopNode::size(PhaseRegAlloc*) const
1761 {
1762   return _count;
1763 }
1764 
1765 
1766 //=============================================================================
1767 #ifndef PRODUCT
1768 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1769 {
1770   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1771   int reg = ra_->get_reg_first(this);
1772   st->print("leaq    %s, [rsp + #%d]\t# box lock",
1773             Matcher::regName[reg], offset);
1774 }
1775 #endif
1776 
1777 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1778 {
1779   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1780   int reg = ra_->get_encode(this);
1781   if (offset >= 0x80) {
1782     emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1783     emit_opcode(cbuf, 0x8D); // LEA  reg,[SP+offset]
1784     emit_rm(cbuf, 0x2, reg & 7, 0x04);
1785     emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1786     emit_d32(cbuf, offset);
1787   } else {
1788     emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1789     emit_opcode(cbuf, 0x8D); // LEA  reg,[SP+offset]
1790     emit_rm(cbuf, 0x1, reg & 7, 0x04);
1791     emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1792     emit_d8(cbuf, offset);
1793   }
1794 }
1795 
1796 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1797 {
1798   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1799   return (offset < 0x80) ? 5 : 8; // REX
1800 }
1801 
1802 //=============================================================================
1803 
1804 // emit call stub, compiled java to interpreter
1805 void emit_java_to_interp(CodeBuffer& cbuf)
1806 {
1807   // Stub is fixed up when the corresponding call is converted from
1808   // calling compiled code to calling interpreted code.
1809   // movq rbx, 0
1810   // jmp -5 # to self
1811 
1812   address mark = cbuf.inst_mark();  // get mark within main instrs section
1813 
1814   // Note that the code buffer's inst_mark is always relative to insts.
1815   // That's why we must use the macroassembler to generate a stub.
1816   MacroAssembler _masm(&cbuf);
1817 
1818   address base =
1819   __ start_a_stub(Compile::MAX_stubs_size);
1820   if (base == NULL)  return;  // CodeBuffer::expand failed
1821   // static stub relocation stores the instruction address of the call
1822   __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1823   // static stub relocation also tags the methodOop in the code-stream.
1824   __ movoop(rbx, (jobject) NULL);  // method is zapped till fixup time
1825   // This is recognized as unresolved by relocs/nativeinst/ic code
1826   __ jump(RuntimeAddress(__ pc()));
1827 
1828   // Update current stubs pointer and restore code_end.
1829   __ end_a_stub();
1830 }
1831 
1832 // size of call stub, compiled java to interpretor
1833 uint size_java_to_interp()
1834 {
1835   return 15;  // movq (1+1+8); jmp (1+4)
1836 }
1837 
1838 // relocation entries for call stub, compiled java to interpretor
1839 uint reloc_java_to_interp()
1840 {
1841   return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1842 }
1843 
1844 //=============================================================================
1845 #ifndef PRODUCT
1846 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1847 {
1848   if (UseCompressedOops) {
1849     st->print_cr("movl    rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
1850     st->print_cr("leaq    rscratch1, [r12_heapbase, r, Address::times_8, 0]");
1851     st->print_cr("cmpq    rax, rscratch1\t # Inline cache check");
1852   } else {
1853     st->print_cr("cmpq    rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
1854                  "# Inline cache check", oopDesc::klass_offset_in_bytes());
1855   }
1856   st->print_cr("\tjne     SharedRuntime::_ic_miss_stub");
1857   st->print_cr("\tnop");
1858   if (!OptoBreakpoint) {
1859     st->print_cr("\tnop");
1860   }
1861 }
1862 #endif
1863 
1864 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1865 {
1866   MacroAssembler masm(&cbuf);
1867 #ifdef ASSERT
1868   uint code_size = cbuf.code_size();
1869 #endif
1870   if (UseCompressedOops) {
1871     masm.load_klass(rscratch1, j_rarg0);
1872     masm.cmpptr(rax, rscratch1);
1873   } else {
1874     masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1875   }
1876 
1877   masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1878 
1879   /* WARNING these NOPs are critical so that verified entry point is properly
1880      aligned for patching by NativeJump::patch_verified_entry() */
1881   int nops_cnt = 1;
1882   if (!OptoBreakpoint) {
1883     // Leave space for int3
1884      nops_cnt += 1;
1885   }
1886   if (UseCompressedOops) {
1887     // ??? divisible by 4 is aligned?
1888     nops_cnt += 1;
1889   }
1890   masm.nop(nops_cnt);
1891 
1892   assert(cbuf.code_size() - code_size == size(ra_),
1893          "checking code size of inline cache node");
1894 }
1895 
1896 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1897 {
1898   if (UseCompressedOops) {
1899     return OptoBreakpoint ? 19 : 20;
1900   } else {
1901     return OptoBreakpoint ? 11 : 12;
1902   }
1903 }
1904 
1905 
1906 //=============================================================================
1907 uint size_exception_handler()
1908 {
1909   // NativeCall instruction size is the same as NativeJump.
1910   // Note that this value is also credited (in output.cpp) to
1911   // the size of the code section.
1912   return NativeJump::instruction_size;
1913 }
1914 
1915 // Emit exception handler code.
1916 int emit_exception_handler(CodeBuffer& cbuf)
1917 {
1918 
1919   // Note that the code buffer's inst_mark is always relative to insts.
1920   // That's why we must use the macroassembler to generate a handler.
1921   MacroAssembler _masm(&cbuf);
1922   address base =
1923   __ start_a_stub(size_exception_handler());
1924   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1925   int offset = __ offset();
1926   __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
1927   assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1928   __ end_a_stub();
1929   return offset;
1930 }
1931 
1932 uint size_deopt_handler()
1933 {
1934   // three 5 byte instructions
1935   return 15;
1936 }
1937 
1938 // Emit deopt handler code.
1939 int emit_deopt_handler(CodeBuffer& cbuf)
1940 {
1941 
1942   // Note that the code buffer's inst_mark is always relative to insts.
1943   // That's why we must use the macroassembler to generate a handler.
1944   MacroAssembler _masm(&cbuf);
1945   address base =
1946   __ start_a_stub(size_deopt_handler());
1947   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1948   int offset = __ offset();
1949   address the_pc = (address) __ pc();
1950   Label next;
1951   // push a "the_pc" on the stack without destroying any registers
1952   // as they all may be live.
1953 
1954   // push address of "next"
1955   __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1956   __ bind(next);
1957   // adjust it so it matches "the_pc"
1958   __ subptr(Address(rsp, 0), __ offset() - offset);
1959   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1960   assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1961   __ end_a_stub();
1962   return offset;
1963 }
1964 
1965 static void emit_double_constant(CodeBuffer& cbuf, double x) {
1966   int mark = cbuf.insts()->mark_off();
1967   MacroAssembler _masm(&cbuf);
1968   address double_address = __ double_constant(x);
1969   cbuf.insts()->set_mark_off(mark);  // preserve mark across masm shift
1970   emit_d32_reloc(cbuf,
1971                  (int) (double_address - cbuf.code_end() - 4),
1972                  internal_word_Relocation::spec(double_address),
1973                  RELOC_DISP32);
1974 }
1975 
1976 static void emit_float_constant(CodeBuffer& cbuf, float x) {
1977   int mark = cbuf.insts()->mark_off();
1978   MacroAssembler _masm(&cbuf);
1979   address float_address = __ float_constant(x);
1980   cbuf.insts()->set_mark_off(mark);  // preserve mark across masm shift
1981   emit_d32_reloc(cbuf,
1982                  (int) (float_address - cbuf.code_end() - 4),
1983                  internal_word_Relocation::spec(float_address),
1984                  RELOC_DISP32);
1985 }
1986 
1987 
1988 int Matcher::regnum_to_fpu_offset(int regnum)
1989 {
1990   return regnum - 32; // The FP registers are in the second chunk
1991 }
1992 
1993 // This is UltraSparc specific, true just means we have fast l2f conversion
1994 const bool Matcher::convL2FSupported(void) {
1995   return true;
1996 }
1997 
1998 // Vector width in bytes
1999 const uint Matcher::vector_width_in_bytes(void) {
2000   return 8;
2001 }
2002 
2003 // Vector ideal reg
2004 const uint Matcher::vector_ideal_reg(void) {
2005   return Op_RegD;
2006 }
2007 
2008 // Is this branch offset short enough that a short branch can be used?
2009 //
2010 // NOTE: If the platform does not provide any short branch variants, then
2011 //       this method should return false for offset 0.
2012 bool Matcher::is_short_branch_offset(int rule, int offset) {
2013   // the short version of jmpConUCF2 contains multiple branches,
2014   // making the reach slightly less
2015   if (rule == jmpConUCF2_rule)
2016     return (-126 <= offset && offset <= 125);
2017   return (-128 <= offset && offset <= 127);
2018 }
2019 
2020 const bool Matcher::isSimpleConstant64(jlong value) {
2021   // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
2022   //return value == (int) value;  // Cf. storeImmL and immL32.
2023 
2024   // Probably always true, even if a temp register is required.
2025   return true;
2026 }
2027 
2028 // The ecx parameter to rep stosq for the ClearArray node is in words.
2029 const bool Matcher::init_array_count_is_in_bytes = false;
2030 
2031 // Threshold size for cleararray.
2032 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2033 
2034 // Should the Matcher clone shifts on addressing modes, expecting them
2035 // to be subsumed into complex addressing expressions or compute them
2036 // into registers?  True for Intel but false for most RISCs
2037 const bool Matcher::clone_shift_expressions = true;
2038 
2039 // Is it better to copy float constants, or load them directly from
2040 // memory?  Intel can load a float constant from a direct address,
2041 // requiring no extra registers.  Most RISCs will have to materialize
2042 // an address into a register first, so they would do better to copy
2043 // the constant from stack.
2044 const bool Matcher::rematerialize_float_constants = true; // XXX
2045 
2046 // If CPU can load and store mis-aligned doubles directly then no
2047 // fixup is needed.  Else we split the double into 2 integer pieces
2048 // and move it piece-by-piece.  Only happens when passing doubles into
2049 // C code as the Java calling convention forces doubles to be aligned.
2050 const bool Matcher::misaligned_doubles_ok = true;
2051 
2052 // No-op on amd64
2053 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2054 
2055 // Advertise here if the CPU requires explicit rounding operations to
2056 // implement the UseStrictFP mode.
2057 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2058 
2059 // Do floats take an entire double register or just half?
2060 const bool Matcher::float_in_double = true;
2061 // Do ints take an entire long register or just half?
2062 const bool Matcher::int_in_long = true;
2063 
2064 // Return whether or not this register is ever used as an argument.
2065 // This function is used on startup to build the trampoline stubs in
2066 // generateOptoStub.  Registers not mentioned will be killed by the VM
2067 // call in the trampoline, and arguments in those registers not be
2068 // available to the callee.
2069 bool Matcher::can_be_java_arg(int reg)
2070 {
2071   return
2072     reg ==  RDI_num || reg ==  RDI_H_num ||
2073     reg ==  RSI_num || reg ==  RSI_H_num ||
2074     reg ==  RDX_num || reg ==  RDX_H_num ||
2075     reg ==  RCX_num || reg ==  RCX_H_num ||
2076     reg ==   R8_num || reg ==   R8_H_num ||
2077     reg ==   R9_num || reg ==   R9_H_num ||
2078     reg ==  R12_num || reg ==  R12_H_num ||
2079     reg == XMM0_num || reg == XMM0_H_num ||
2080     reg == XMM1_num || reg == XMM1_H_num ||
2081     reg == XMM2_num || reg == XMM2_H_num ||
2082     reg == XMM3_num || reg == XMM3_H_num ||
2083     reg == XMM4_num || reg == XMM4_H_num ||
2084     reg == XMM5_num || reg == XMM5_H_num ||
2085     reg == XMM6_num || reg == XMM6_H_num ||
2086     reg == XMM7_num || reg == XMM7_H_num;
2087 }
2088 
2089 bool Matcher::is_spillable_arg(int reg)
2090 {
2091   return can_be_java_arg(reg);
2092 }
2093 
2094 // Register for DIVI projection of divmodI
2095 RegMask Matcher::divI_proj_mask() {
2096   return INT_RAX_REG_mask;
2097 }
2098 
2099 // Register for MODI projection of divmodI
2100 RegMask Matcher::modI_proj_mask() {
2101   return INT_RDX_REG_mask;
2102 }
2103 
2104 // Register for DIVL projection of divmodL
2105 RegMask Matcher::divL_proj_mask() {
2106   return LONG_RAX_REG_mask;
2107 }
2108 
2109 // Register for MODL projection of divmodL
2110 RegMask Matcher::modL_proj_mask() {
2111   return LONG_RDX_REG_mask;
2112 }
2113 
2114 static Address build_address(int b, int i, int s, int d) {
2115   Register index = as_Register(i);
2116   Address::ScaleFactor scale = (Address::ScaleFactor)s;
2117   if (index == rsp) {
2118     index = noreg;
2119     scale = Address::no_scale;
2120   }
2121   Address addr(as_Register(b), index, scale, d);
2122   return addr;
2123 }
2124 
2125 %}
2126 
2127 //----------ENCODING BLOCK-----------------------------------------------------
2128 // This block specifies the encoding classes used by the compiler to
2129 // output byte streams.  Encoding classes are parameterized macros
2130 // used by Machine Instruction Nodes in order to generate the bit
2131 // encoding of the instruction.  Operands specify their base encoding
2132 // interface with the interface keyword.  There are currently
2133 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2134 // COND_INTER.  REG_INTER causes an operand to generate a function
2135 // which returns its register number when queried.  CONST_INTER causes
2136 // an operand to generate a function which returns the value of the
2137 // constant when queried.  MEMORY_INTER causes an operand to generate
2138 // four functions which return the Base Register, the Index Register,
2139 // the Scale Value, and the Offset Value of the operand when queried.
2140 // COND_INTER causes an operand to generate six functions which return
2141 // the encoding code (ie - encoding bits for the instruction)
2142 // associated with each basic boolean condition for a conditional
2143 // instruction.
2144 //
2145 // Instructions specify two basic values for encoding.  Again, a
2146 // function is available to check if the constant displacement is an
2147 // oop. They use the ins_encode keyword to specify their encoding
2148 // classes (which must be a sequence of enc_class names, and their
2149 // parameters, specified in the encoding block), and they use the
2150 // opcode keyword to specify, in order, their primary, secondary, and
2151 // tertiary opcode.  Only the opcode sections which a particular
2152 // instruction needs for encoding need to be specified.
2153 encode %{
2154   // Build emit functions for each basic byte or larger field in the
2155   // intel encoding scheme (opcode, rm, sib, immediate), and call them
2156   // from C++ code in the enc_class source block.  Emit functions will
2157   // live in the main source block for now.  In future, we can
2158   // generalize this by adding a syntax that specifies the sizes of
2159   // fields in an order, so that the adlc can build the emit functions
2160   // automagically
2161 
2162   // Emit primary opcode
2163   enc_class OpcP
2164   %{
2165     emit_opcode(cbuf, $primary);
2166   %}
2167 
2168   // Emit secondary opcode
2169   enc_class OpcS
2170   %{
2171     emit_opcode(cbuf, $secondary);
2172   %}
2173 
2174   // Emit tertiary opcode
2175   enc_class OpcT
2176   %{
2177     emit_opcode(cbuf, $tertiary);
2178   %}
2179 
2180   // Emit opcode directly
2181   enc_class Opcode(immI d8)
2182   %{
2183     emit_opcode(cbuf, $d8$$constant);
2184   %}
2185 
2186   // Emit size prefix
2187   enc_class SizePrefix
2188   %{
2189     emit_opcode(cbuf, 0x66);
2190   %}
2191 
2192   enc_class reg(rRegI reg)
2193   %{
2194     emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2195   %}
2196 
2197   enc_class reg_reg(rRegI dst, rRegI src)
2198   %{
2199     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2200   %}
2201 
2202   enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2203   %{
2204     emit_opcode(cbuf, $opcode$$constant);
2205     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2206   %}
2207 
2208   enc_class cmpfp_fixup()
2209   %{
2210     // jnp,s exit
2211     emit_opcode(cbuf, 0x7B);
2212     emit_d8(cbuf, 0x0A);
2213 
2214     // pushfq
2215     emit_opcode(cbuf, 0x9C);
2216 
2217     // andq $0xffffff2b, (%rsp)
2218     emit_opcode(cbuf, Assembler::REX_W);
2219     emit_opcode(cbuf, 0x81);
2220     emit_opcode(cbuf, 0x24);
2221     emit_opcode(cbuf, 0x24);
2222     emit_d32(cbuf, 0xffffff2b);
2223 
2224     // popfq
2225     emit_opcode(cbuf, 0x9D);
2226 
2227     // nop (target for branch to avoid branch to branch)
2228     emit_opcode(cbuf, 0x90);
2229   %}
2230 
2231   enc_class cmpfp3(rRegI dst)
2232   %{
2233     int dstenc = $dst$$reg;
2234 
2235     // movl $dst, -1
2236     if (dstenc >= 8) {
2237       emit_opcode(cbuf, Assembler::REX_B);
2238     }
2239     emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2240     emit_d32(cbuf, -1);
2241 
2242     // jp,s done
2243     emit_opcode(cbuf, 0x7A);
2244     emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2245 
2246     // jb,s done
2247     emit_opcode(cbuf, 0x72);
2248     emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2249 
2250     // setne $dst
2251     if (dstenc >= 4) {
2252       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2253     }
2254     emit_opcode(cbuf, 0x0F);
2255     emit_opcode(cbuf, 0x95);
2256     emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2257 
2258     // movzbl $dst, $dst
2259     if (dstenc >= 4) {
2260       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2261     }
2262     emit_opcode(cbuf, 0x0F);
2263     emit_opcode(cbuf, 0xB6);
2264     emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2265   %}
2266 
2267   enc_class cdql_enc(no_rax_rdx_RegI div)
2268   %{
2269     // Full implementation of Java idiv and irem; checks for
2270     // special case as described in JVM spec., p.243 & p.271.
2271     //
2272     //         normal case                           special case
2273     //
2274     // input : rax: dividend                         min_int
2275     //         reg: divisor                          -1
2276     //
2277     // output: rax: quotient  (= rax idiv reg)       min_int
2278     //         rdx: remainder (= rax irem reg)       0
2279     //
2280     //  Code sequnce:
2281     //
2282     //    0:   3d 00 00 00 80          cmp    $0x80000000,%eax
2283     //    5:   75 07/08                jne    e <normal>
2284     //    7:   33 d2                   xor    %edx,%edx
2285     //  [div >= 8 -> offset + 1]
2286     //  [REX_B]
2287     //    9:   83 f9 ff                cmp    $0xffffffffffffffff,$div
2288     //    c:   74 03/04                je     11 <done>
2289     // 000000000000000e <normal>:
2290     //    e:   99                      cltd
2291     //  [div >= 8 -> offset + 1]
2292     //  [REX_B]
2293     //    f:   f7 f9                   idiv   $div
2294     // 0000000000000011 <done>:
2295 
2296     // cmp    $0x80000000,%eax
2297     emit_opcode(cbuf, 0x3d);
2298     emit_d8(cbuf, 0x00);
2299     emit_d8(cbuf, 0x00);
2300     emit_d8(cbuf, 0x00);
2301     emit_d8(cbuf, 0x80);
2302 
2303     // jne    e <normal>
2304     emit_opcode(cbuf, 0x75);
2305     emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2306 
2307     // xor    %edx,%edx
2308     emit_opcode(cbuf, 0x33);
2309     emit_d8(cbuf, 0xD2);
2310 
2311     // cmp    $0xffffffffffffffff,%ecx
2312     if ($div$$reg >= 8) {
2313       emit_opcode(cbuf, Assembler::REX_B);
2314     }
2315     emit_opcode(cbuf, 0x83);
2316     emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2317     emit_d8(cbuf, 0xFF);
2318 
2319     // je     11 <done>
2320     emit_opcode(cbuf, 0x74);
2321     emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2322 
2323     // <normal>
2324     // cltd
2325     emit_opcode(cbuf, 0x99);
2326 
2327     // idivl (note: must be emitted by the user of this rule)
2328     // <done>
2329   %}
2330 
2331   enc_class cdqq_enc(no_rax_rdx_RegL div)
2332   %{
2333     // Full implementation of Java ldiv and lrem; checks for
2334     // special case as described in JVM spec., p.243 & p.271.
2335     //
2336     //         normal case                           special case
2337     //
2338     // input : rax: dividend                         min_long
2339     //         reg: divisor                          -1
2340     //
2341     // output: rax: quotient  (= rax idiv reg)       min_long
2342     //         rdx: remainder (= rax irem reg)       0
2343     //
2344     //  Code sequnce:
2345     //
2346     //    0:   48 ba 00 00 00 00 00    mov    $0x8000000000000000,%rdx
2347     //    7:   00 00 80
2348     //    a:   48 39 d0                cmp    %rdx,%rax
2349     //    d:   75 08                   jne    17 <normal>
2350     //    f:   33 d2                   xor    %edx,%edx
2351     //   11:   48 83 f9 ff             cmp    $0xffffffffffffffff,$div
2352     //   15:   74 05                   je     1c <done>
2353     // 0000000000000017 <normal>:
2354     //   17:   48 99                   cqto
2355     //   19:   48 f7 f9                idiv   $div
2356     // 000000000000001c <done>:
2357 
2358     // mov    $0x8000000000000000,%rdx
2359     emit_opcode(cbuf, Assembler::REX_W);
2360     emit_opcode(cbuf, 0xBA);
2361     emit_d8(cbuf, 0x00);
2362     emit_d8(cbuf, 0x00);
2363     emit_d8(cbuf, 0x00);
2364     emit_d8(cbuf, 0x00);
2365     emit_d8(cbuf, 0x00);
2366     emit_d8(cbuf, 0x00);
2367     emit_d8(cbuf, 0x00);
2368     emit_d8(cbuf, 0x80);
2369 
2370     // cmp    %rdx,%rax
2371     emit_opcode(cbuf, Assembler::REX_W);
2372     emit_opcode(cbuf, 0x39);
2373     emit_d8(cbuf, 0xD0);
2374 
2375     // jne    17 <normal>
2376     emit_opcode(cbuf, 0x75);
2377     emit_d8(cbuf, 0x08);
2378 
2379     // xor    %edx,%edx
2380     emit_opcode(cbuf, 0x33);
2381     emit_d8(cbuf, 0xD2);
2382 
2383     // cmp    $0xffffffffffffffff,$div
2384     emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2385     emit_opcode(cbuf, 0x83);
2386     emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2387     emit_d8(cbuf, 0xFF);
2388 
2389     // je     1e <done>
2390     emit_opcode(cbuf, 0x74);
2391     emit_d8(cbuf, 0x05);
2392 
2393     // <normal>
2394     // cqto
2395     emit_opcode(cbuf, Assembler::REX_W);
2396     emit_opcode(cbuf, 0x99);
2397 
2398     // idivq (note: must be emitted by the user of this rule)
2399     // <done>
2400   %}
2401 
2402   // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2403   enc_class OpcSE(immI imm)
2404   %{
2405     // Emit primary opcode and set sign-extend bit
2406     // Check for 8-bit immediate, and set sign extend bit in opcode
2407     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2408       emit_opcode(cbuf, $primary | 0x02);
2409     } else {
2410       // 32-bit immediate
2411       emit_opcode(cbuf, $primary);
2412     }
2413   %}
2414 
2415   enc_class OpcSErm(rRegI dst, immI imm)
2416   %{
2417     // OpcSEr/m
2418     int dstenc = $dst$$reg;
2419     if (dstenc >= 8) {
2420       emit_opcode(cbuf, Assembler::REX_B);
2421       dstenc -= 8;
2422     }
2423     // Emit primary opcode and set sign-extend bit
2424     // Check for 8-bit immediate, and set sign extend bit in opcode
2425     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2426       emit_opcode(cbuf, $primary | 0x02);
2427     } else {
2428       // 32-bit immediate
2429       emit_opcode(cbuf, $primary);
2430     }
2431     // Emit r/m byte with secondary opcode, after primary opcode.
2432     emit_rm(cbuf, 0x3, $secondary, dstenc);
2433   %}
2434 
2435   enc_class OpcSErm_wide(rRegL dst, immI imm)
2436   %{
2437     // OpcSEr/m
2438     int dstenc = $dst$$reg;
2439     if (dstenc < 8) {
2440       emit_opcode(cbuf, Assembler::REX_W);
2441     } else {
2442       emit_opcode(cbuf, Assembler::REX_WB);
2443       dstenc -= 8;
2444     }
2445     // Emit primary opcode and set sign-extend bit
2446     // Check for 8-bit immediate, and set sign extend bit in opcode
2447     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2448       emit_opcode(cbuf, $primary | 0x02);
2449     } else {
2450       // 32-bit immediate
2451       emit_opcode(cbuf, $primary);
2452     }
2453     // Emit r/m byte with secondary opcode, after primary opcode.
2454     emit_rm(cbuf, 0x3, $secondary, dstenc);
2455   %}
2456 
2457   enc_class Con8or32(immI imm)
2458   %{
2459     // Check for 8-bit immediate, and set sign extend bit in opcode
2460     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2461       $$$emit8$imm$$constant;
2462     } else {
2463       // 32-bit immediate
2464       $$$emit32$imm$$constant;
2465     }
2466   %}
2467 
2468   enc_class Lbl(label labl)
2469   %{
2470     // JMP, CALL
2471     Label* l = $labl$$label;
2472     emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2473   %}
2474 
2475   enc_class LblShort(label labl)
2476   %{
2477     // JMP, CALL
2478     Label* l = $labl$$label;
2479     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2480     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2481     emit_d8(cbuf, disp);
2482   %}
2483 
2484   enc_class opc2_reg(rRegI dst)
2485   %{
2486     // BSWAP
2487     emit_cc(cbuf, $secondary, $dst$$reg);
2488   %}
2489 
2490   enc_class opc3_reg(rRegI dst)
2491   %{
2492     // BSWAP
2493     emit_cc(cbuf, $tertiary, $dst$$reg);
2494   %}
2495 
2496   enc_class reg_opc(rRegI div)
2497   %{
2498     // INC, DEC, IDIV, IMOD, JMP indirect, ...
2499     emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2500   %}
2501 
2502   enc_class Jcc(cmpOp cop, label labl)
2503   %{
2504     // JCC
2505     Label* l = $labl$$label;
2506     $$$emit8$primary;
2507     emit_cc(cbuf, $secondary, $cop$$cmpcode);
2508     emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2509   %}
2510 
2511   enc_class JccShort (cmpOp cop, label labl)
2512   %{
2513   // JCC
2514     Label *l = $labl$$label;
2515     emit_cc(cbuf, $primary, $cop$$cmpcode);
2516     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2517     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2518     emit_d8(cbuf, disp);
2519   %}
2520 
2521   enc_class enc_cmov(cmpOp cop)
2522   %{
2523     // CMOV
2524     $$$emit8$primary;
2525     emit_cc(cbuf, $secondary, $cop$$cmpcode);
2526   %}
2527 
2528   enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2529   %{
2530     // Invert sense of branch from sense of cmov
2531     emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2532     emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2533                   ? (UseXmmRegToRegMoveAll ? 3 : 4)
2534                   : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2535     // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2536     if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2537     if ($dst$$reg < 8) {
2538       if ($src$$reg >= 8) {
2539         emit_opcode(cbuf, Assembler::REX_B);
2540       }
2541     } else {
2542       if ($src$$reg < 8) {
2543         emit_opcode(cbuf, Assembler::REX_R);
2544       } else {
2545         emit_opcode(cbuf, Assembler::REX_RB);
2546       }
2547     }
2548     emit_opcode(cbuf, 0x0F);
2549     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2550     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2551   %}
2552 
2553   enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2554   %{
2555     // Invert sense of branch from sense of cmov
2556     emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2557     emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2558 
2559     //  UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2560     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2561     if ($dst$$reg < 8) {
2562       if ($src$$reg >= 8) {
2563         emit_opcode(cbuf, Assembler::REX_B);
2564       }
2565     } else {
2566       if ($src$$reg < 8) {
2567         emit_opcode(cbuf, Assembler::REX_R);
2568       } else {
2569         emit_opcode(cbuf, Assembler::REX_RB);
2570       }
2571     }
2572     emit_opcode(cbuf, 0x0F);
2573     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2574     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2575   %}
2576 
2577   enc_class enc_PartialSubtypeCheck()
2578   %{
2579     Register Rrdi = as_Register(RDI_enc); // result register
2580     Register Rrax = as_Register(RAX_enc); // super class
2581     Register Rrcx = as_Register(RCX_enc); // killed
2582     Register Rrsi = as_Register(RSI_enc); // sub class
2583     Label hit, miss, cmiss;
2584 
2585     MacroAssembler _masm(&cbuf);
2586     // Compare super with sub directly, since super is not in its own SSA.
2587     // The compiler used to emit this test, but we fold it in here,
2588     // to allow platform-specific tweaking on sparc.
2589     __ cmpptr(Rrax, Rrsi);
2590     __ jcc(Assembler::equal, hit);
2591 #ifndef PRODUCT
2592     __ lea(Rrcx, ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
2593     __ incrementl(Address(Rrcx, 0));
2594 #endif //PRODUCT
2595     __ movptr(Rrdi, Address(Rrsi, 
2596                           sizeof(oopDesc) + 
2597                           Klass::secondary_supers_offset_in_bytes()));
2598     __ movl(Rrcx, Address(Rrdi, arrayOopDesc::length_offset_in_bytes()));
2599     __ addptr(Rrdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
2600     if (UseCompressedOops) {
2601       __ encode_heap_oop(Rrax);
2602       __ repne_scanl();
2603       __ jcc(Assembler::notEqual, cmiss);
2604       __ decode_heap_oop(Rrax);
2605       __ movptr(Address(Rrsi,
2606                       sizeof(oopDesc) +
2607                       Klass::secondary_super_cache_offset_in_bytes()),
2608               Rrax);
2609       __ jmp(hit);
2610       __ bind(cmiss);
2611       __ decode_heap_oop(Rrax);
2612       __ jmp(miss);
2613     } else {
2614       __ repne_scan();
2615       __ jcc(Assembler::notEqual, miss);
2616       __ movptr(Address(Rrsi,
2617                       sizeof(oopDesc) +
2618                       Klass::secondary_super_cache_offset_in_bytes()),
2619               Rrax);
2620     }
2621     __ bind(hit);
2622     if ($primary) {
2623       __ xorptr(Rrdi, Rrdi);
2624     }
2625     __ bind(miss);
2626   %}
2627 
2628   enc_class Java_To_Interpreter(method meth)
2629   %{
2630     // CALL Java_To_Interpreter
2631     // This is the instruction starting address for relocation info.
2632     cbuf.set_inst_mark();
2633     $$$emit8$primary;
2634     // CALL directly to the runtime
2635     emit_d32_reloc(cbuf,
2636                    (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2637                    runtime_call_Relocation::spec(),
2638                    RELOC_DISP32);
2639   %}
2640 
2641   enc_class Java_Static_Call(method meth)
2642   %{
2643     // JAVA STATIC CALL
2644     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to
2645     // determine who we intended to call.
2646     cbuf.set_inst_mark();
2647     $$$emit8$primary;
2648 
2649     if (!_method) {
2650       emit_d32_reloc(cbuf,
2651                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2652                      runtime_call_Relocation::spec(),
2653                      RELOC_DISP32);
2654     } else if (_optimized_virtual) {
2655       emit_d32_reloc(cbuf,
2656                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2657                      opt_virtual_call_Relocation::spec(),
2658                      RELOC_DISP32);
2659     } else {
2660       emit_d32_reloc(cbuf,
2661                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2662                      static_call_Relocation::spec(),
2663                      RELOC_DISP32);
2664     }
2665     if (_method) {
2666       // Emit stub for static call
2667       emit_java_to_interp(cbuf);
2668     }
2669   %}
2670 
2671   enc_class Java_Dynamic_Call(method meth)
2672   %{
2673     // JAVA DYNAMIC CALL
2674     // !!!!!
2675     // Generate  "movq rax, -1", placeholder instruction to load oop-info
2676     // emit_call_dynamic_prologue( cbuf );
2677     cbuf.set_inst_mark();
2678 
2679     // movq rax, -1
2680     emit_opcode(cbuf, Assembler::REX_W);
2681     emit_opcode(cbuf, 0xB8 | RAX_enc);
2682     emit_d64_reloc(cbuf,
2683                    (int64_t) Universe::non_oop_word(),
2684                    oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2685     address virtual_call_oop_addr = cbuf.inst_mark();
2686     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
2687     // who we intended to call.
2688     cbuf.set_inst_mark();
2689     $$$emit8$primary;
2690     emit_d32_reloc(cbuf,
2691                    (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2692                    virtual_call_Relocation::spec(virtual_call_oop_addr),
2693                    RELOC_DISP32);
2694   %}
2695 
2696   enc_class Java_Compiled_Call(method meth)
2697   %{
2698     // JAVA COMPILED CALL
2699     int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2700 
2701     // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2702     // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2703 
2704     // callq *disp(%rax)
2705     cbuf.set_inst_mark();
2706     $$$emit8$primary;
2707     if (disp < 0x80) {
2708       emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2709       emit_d8(cbuf, disp); // Displacement
2710     } else {
2711       emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2712       emit_d32(cbuf, disp); // Displacement
2713     }
2714   %}
2715 
2716   enc_class reg_opc_imm(rRegI dst, immI8 shift)
2717   %{
2718     // SAL, SAR, SHR
2719     int dstenc = $dst$$reg;
2720     if (dstenc >= 8) {
2721       emit_opcode(cbuf, Assembler::REX_B);
2722       dstenc -= 8;
2723     }
2724     $$$emit8$primary;
2725     emit_rm(cbuf, 0x3, $secondary, dstenc);
2726     $$$emit8$shift$$constant;
2727   %}
2728 
2729   enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2730   %{
2731     // SAL, SAR, SHR
2732     int dstenc = $dst$$reg;
2733     if (dstenc < 8) {
2734       emit_opcode(cbuf, Assembler::REX_W);
2735     } else {
2736       emit_opcode(cbuf, Assembler::REX_WB);
2737       dstenc -= 8;
2738     }
2739     $$$emit8$primary;
2740     emit_rm(cbuf, 0x3, $secondary, dstenc);
2741     $$$emit8$shift$$constant;
2742   %}
2743 
2744   enc_class load_immI(rRegI dst, immI src)
2745   %{
2746     int dstenc = $dst$$reg;
2747     if (dstenc >= 8) {
2748       emit_opcode(cbuf, Assembler::REX_B);
2749       dstenc -= 8;
2750     }
2751     emit_opcode(cbuf, 0xB8 | dstenc);
2752     $$$emit32$src$$constant;
2753   %}
2754 
2755   enc_class load_immL(rRegL dst, immL src)
2756   %{
2757     int dstenc = $dst$$reg;
2758     if (dstenc < 8) {
2759       emit_opcode(cbuf, Assembler::REX_W);
2760     } else {
2761       emit_opcode(cbuf, Assembler::REX_WB);
2762       dstenc -= 8;
2763     }
2764     emit_opcode(cbuf, 0xB8 | dstenc);
2765     emit_d64(cbuf, $src$$constant);
2766   %}
2767 
2768   enc_class load_immUL32(rRegL dst, immUL32 src)
2769   %{
2770     // same as load_immI, but this time we care about zeroes in the high word
2771     int dstenc = $dst$$reg;
2772     if (dstenc >= 8) {
2773       emit_opcode(cbuf, Assembler::REX_B);
2774       dstenc -= 8;
2775     }
2776     emit_opcode(cbuf, 0xB8 | dstenc);
2777     $$$emit32$src$$constant;
2778   %}
2779 
2780   enc_class load_immL32(rRegL dst, immL32 src)
2781   %{
2782     int dstenc = $dst$$reg;
2783     if (dstenc < 8) {
2784       emit_opcode(cbuf, Assembler::REX_W);
2785     } else {
2786       emit_opcode(cbuf, Assembler::REX_WB);
2787       dstenc -= 8;
2788     }
2789     emit_opcode(cbuf, 0xC7);
2790     emit_rm(cbuf, 0x03, 0x00, dstenc);
2791     $$$emit32$src$$constant;
2792   %}
2793 
2794   enc_class load_immP31(rRegP dst, immP32 src)
2795   %{
2796     // same as load_immI, but this time we care about zeroes in the high word
2797     int dstenc = $dst$$reg;
2798     if (dstenc >= 8) {
2799       emit_opcode(cbuf, Assembler::REX_B);
2800       dstenc -= 8;
2801     }
2802     emit_opcode(cbuf, 0xB8 | dstenc);
2803     $$$emit32$src$$constant;
2804   %}
2805 
2806   enc_class load_immP(rRegP dst, immP src)
2807   %{
2808     int dstenc = $dst$$reg;
2809     if (dstenc < 8) {
2810       emit_opcode(cbuf, Assembler::REX_W);
2811     } else {
2812       emit_opcode(cbuf, Assembler::REX_WB);
2813       dstenc -= 8;
2814     }
2815     emit_opcode(cbuf, 0xB8 | dstenc);
2816     // This next line should be generated from ADLC
2817     if ($src->constant_is_oop()) {
2818       emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2819     } else {
2820       emit_d64(cbuf, $src$$constant);
2821     }
2822   %}
2823 
2824   enc_class load_immF(regF dst, immF con)
2825   %{
2826     // XXX reg_mem doesn't support RIP-relative addressing yet
2827     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2828     emit_float_constant(cbuf, $con$$constant);
2829   %}
2830 
2831   enc_class load_immD(regD dst, immD con)
2832   %{
2833     // XXX reg_mem doesn't support RIP-relative addressing yet
2834     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2835     emit_double_constant(cbuf, $con$$constant);
2836   %}
2837 
2838   enc_class load_conF (regF dst, immF con) %{    // Load float constant
2839     emit_opcode(cbuf, 0xF3);
2840     if ($dst$$reg >= 8) {
2841       emit_opcode(cbuf, Assembler::REX_R);
2842     }
2843     emit_opcode(cbuf, 0x0F);
2844     emit_opcode(cbuf, 0x10);
2845     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2846     emit_float_constant(cbuf, $con$$constant);
2847   %}
2848 
2849   enc_class load_conD (regD dst, immD con) %{    // Load double constant
2850     // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
2851     emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
2852     if ($dst$$reg >= 8) {
2853       emit_opcode(cbuf, Assembler::REX_R);
2854     }
2855     emit_opcode(cbuf, 0x0F);
2856     emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
2857     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2858     emit_double_constant(cbuf, $con$$constant);
2859   %}
2860 
2861   // Encode a reg-reg copy.  If it is useless, then empty encoding.
2862   enc_class enc_copy(rRegI dst, rRegI src)
2863   %{
2864     encode_copy(cbuf, $dst$$reg, $src$$reg);
2865   %}
2866 
2867   // Encode xmm reg-reg copy.  If it is useless, then empty encoding.
2868   enc_class enc_CopyXD( RegD dst, RegD src ) %{
2869     encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2870   %}
2871 
2872   enc_class enc_copy_always(rRegI dst, rRegI src)
2873   %{
2874     int srcenc = $src$$reg;
2875     int dstenc = $dst$$reg;
2876 
2877     if (dstenc < 8) {
2878       if (srcenc >= 8) {
2879         emit_opcode(cbuf, Assembler::REX_B);
2880         srcenc -= 8;
2881       }
2882     } else {
2883       if (srcenc < 8) {
2884         emit_opcode(cbuf, Assembler::REX_R);
2885       } else {
2886         emit_opcode(cbuf, Assembler::REX_RB);
2887         srcenc -= 8;
2888       }
2889       dstenc -= 8;
2890     }
2891 
2892     emit_opcode(cbuf, 0x8B);
2893     emit_rm(cbuf, 0x3, dstenc, srcenc);
2894   %}
2895 
2896   enc_class enc_copy_wide(rRegL dst, rRegL src)
2897   %{
2898     int srcenc = $src$$reg;
2899     int dstenc = $dst$$reg;
2900 
2901     if (dstenc != srcenc) {
2902       if (dstenc < 8) {
2903         if (srcenc < 8) {
2904           emit_opcode(cbuf, Assembler::REX_W);
2905         } else {
2906           emit_opcode(cbuf, Assembler::REX_WB);
2907           srcenc -= 8;
2908         }
2909       } else {
2910         if (srcenc < 8) {
2911           emit_opcode(cbuf, Assembler::REX_WR);
2912         } else {
2913           emit_opcode(cbuf, Assembler::REX_WRB);
2914           srcenc -= 8;
2915         }
2916         dstenc -= 8;
2917       }
2918       emit_opcode(cbuf, 0x8B);
2919       emit_rm(cbuf, 0x3, dstenc, srcenc);
2920     }
2921   %}
2922 
2923   enc_class Con32(immI src)
2924   %{
2925     // Output immediate
2926     $$$emit32$src$$constant;
2927   %}
2928 
2929   enc_class Con64(immL src)
2930   %{
2931     // Output immediate
2932     emit_d64($src$$constant);
2933   %}
2934 
2935   enc_class Con32F_as_bits(immF src)
2936   %{
2937     // Output Float immediate bits
2938     jfloat jf = $src$$constant;
2939     jint jf_as_bits = jint_cast(jf);
2940     emit_d32(cbuf, jf_as_bits);
2941   %}
2942 
2943   enc_class Con16(immI src)
2944   %{
2945     // Output immediate
2946     $$$emit16$src$$constant;
2947   %}
2948 
2949   // How is this different from Con32??? XXX
2950   enc_class Con_d32(immI src)
2951   %{
2952     emit_d32(cbuf,$src$$constant);
2953   %}
2954 
2955   enc_class conmemref (rRegP t1) %{    // Con32(storeImmI)
2956     // Output immediate memory reference
2957     emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2958     emit_d32(cbuf, 0x00);
2959   %}
2960 
2961   enc_class jump_enc(rRegL switch_val, rRegI dest) %{
2962     MacroAssembler masm(&cbuf);
2963 
2964     Register switch_reg = as_Register($switch_val$$reg);
2965     Register dest_reg   = as_Register($dest$$reg);
2966     address table_base  = masm.address_table_constant(_index2label);
2967 
2968     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2969     // to do that and the compiler is using that register as one it can allocate.
2970     // So we build it all by hand.
2971     // Address index(noreg, switch_reg, Address::times_1);
2972     // ArrayAddress dispatch(table, index);
2973 
2974     Address dispatch(dest_reg, switch_reg, Address::times_1);
2975 
2976     masm.lea(dest_reg, InternalAddress(table_base));
2977     masm.jmp(dispatch);
2978   %}
2979 
2980   enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
2981     MacroAssembler masm(&cbuf);
2982 
2983     Register switch_reg = as_Register($switch_val$$reg);
2984     Register dest_reg   = as_Register($dest$$reg);
2985     address table_base  = masm.address_table_constant(_index2label);
2986 
2987     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2988     // to do that and the compiler is using that register as one it can allocate.
2989     // So we build it all by hand.
2990     // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2991     // ArrayAddress dispatch(table, index);
2992 
2993     Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2994 
2995     masm.lea(dest_reg, InternalAddress(table_base));
2996     masm.jmp(dispatch);
2997   %}
2998 
2999   enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
3000     MacroAssembler masm(&cbuf);
3001 
3002     Register switch_reg = as_Register($switch_val$$reg);
3003     Register dest_reg   = as_Register($dest$$reg);
3004     address table_base  = masm.address_table_constant(_index2label);
3005 
3006     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
3007     // to do that and the compiler is using that register as one it can allocate.
3008     // So we build it all by hand.
3009     // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3010     // ArrayAddress dispatch(table, index);
3011 
3012     Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3013     masm.lea(dest_reg, InternalAddress(table_base));
3014     masm.jmp(dispatch);
3015 
3016   %}
3017 
3018   enc_class lock_prefix()
3019   %{
3020     if (os::is_MP()) {
3021       emit_opcode(cbuf, 0xF0); // lock
3022     }
3023   %}
3024 
3025   enc_class REX_mem(memory mem)
3026   %{
3027     if ($mem$$base >= 8) {
3028       if ($mem$$index < 8) {
3029         emit_opcode(cbuf, Assembler::REX_B);
3030       } else {
3031         emit_opcode(cbuf, Assembler::REX_XB);
3032       }
3033     } else {
3034       if ($mem$$index >= 8) {
3035         emit_opcode(cbuf, Assembler::REX_X);
3036       }
3037     }
3038   %}
3039 
3040   enc_class REX_mem_wide(memory mem)
3041   %{
3042     if ($mem$$base >= 8) {
3043       if ($mem$$index < 8) {
3044         emit_opcode(cbuf, Assembler::REX_WB);
3045       } else {
3046         emit_opcode(cbuf, Assembler::REX_WXB);
3047       }
3048     } else {
3049       if ($mem$$index < 8) {
3050         emit_opcode(cbuf, Assembler::REX_W);
3051       } else {
3052         emit_opcode(cbuf, Assembler::REX_WX);
3053       }
3054     }
3055   %}
3056 
3057   // for byte regs
3058   enc_class REX_breg(rRegI reg)
3059   %{
3060     if ($reg$$reg >= 4) {
3061       emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3062     }
3063   %}
3064 
3065   // for byte regs
3066   enc_class REX_reg_breg(rRegI dst, rRegI src)
3067   %{
3068     if ($dst$$reg < 8) {
3069       if ($src$$reg >= 4) {
3070         emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3071       }
3072     } else {
3073       if ($src$$reg < 8) {
3074         emit_opcode(cbuf, Assembler::REX_R);
3075       } else {
3076         emit_opcode(cbuf, Assembler::REX_RB);
3077       }
3078     }
3079   %}
3080 
3081   // for byte regs
3082   enc_class REX_breg_mem(rRegI reg, memory mem)
3083   %{
3084     if ($reg$$reg < 8) {
3085       if ($mem$$base < 8) {
3086         if ($mem$$index >= 8) {
3087           emit_opcode(cbuf, Assembler::REX_X);
3088         } else if ($reg$$reg >= 4) {
3089           emit_opcode(cbuf, Assembler::REX);
3090         }
3091       } else {
3092         if ($mem$$index < 8) {
3093           emit_opcode(cbuf, Assembler::REX_B);
3094         } else {
3095           emit_opcode(cbuf, Assembler::REX_XB);
3096         }
3097       }
3098     } else {
3099       if ($mem$$base < 8) {
3100         if ($mem$$index < 8) {
3101           emit_opcode(cbuf, Assembler::REX_R);
3102         } else {
3103           emit_opcode(cbuf, Assembler::REX_RX);
3104         }
3105       } else {
3106         if ($mem$$index < 8) {
3107           emit_opcode(cbuf, Assembler::REX_RB);
3108         } else {
3109           emit_opcode(cbuf, Assembler::REX_RXB);
3110         }
3111       }
3112     }
3113   %}
3114 
3115   enc_class REX_reg(rRegI reg)
3116   %{
3117     if ($reg$$reg >= 8) {
3118       emit_opcode(cbuf, Assembler::REX_B);
3119     }
3120   %}
3121 
3122   enc_class REX_reg_wide(rRegI reg)
3123   %{
3124     if ($reg$$reg < 8) {
3125       emit_opcode(cbuf, Assembler::REX_W);
3126     } else {
3127       emit_opcode(cbuf, Assembler::REX_WB);
3128     }
3129   %}
3130 
3131   enc_class REX_reg_reg(rRegI dst, rRegI src)
3132   %{
3133     if ($dst$$reg < 8) {
3134       if ($src$$reg >= 8) {
3135         emit_opcode(cbuf, Assembler::REX_B);
3136       }
3137     } else {
3138       if ($src$$reg < 8) {
3139         emit_opcode(cbuf, Assembler::REX_R);
3140       } else {
3141         emit_opcode(cbuf, Assembler::REX_RB);
3142       }
3143     }
3144   %}
3145 
3146   enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3147   %{
3148     if ($dst$$reg < 8) {
3149       if ($src$$reg < 8) {
3150         emit_opcode(cbuf, Assembler::REX_W);
3151       } else {
3152         emit_opcode(cbuf, Assembler::REX_WB);
3153       }
3154     } else {
3155       if ($src$$reg < 8) {
3156         emit_opcode(cbuf, Assembler::REX_WR);
3157       } else {
3158         emit_opcode(cbuf, Assembler::REX_WRB);
3159       }
3160     }
3161   %}
3162 
3163   enc_class REX_reg_mem(rRegI reg, memory mem)
3164   %{
3165     if ($reg$$reg < 8) {
3166       if ($mem$$base < 8) {
3167         if ($mem$$index >= 8) {
3168           emit_opcode(cbuf, Assembler::REX_X);
3169         }
3170       } else {
3171         if ($mem$$index < 8) {
3172           emit_opcode(cbuf, Assembler::REX_B);
3173         } else {
3174           emit_opcode(cbuf, Assembler::REX_XB);
3175         }
3176       }
3177     } else {
3178       if ($mem$$base < 8) {
3179         if ($mem$$index < 8) {
3180           emit_opcode(cbuf, Assembler::REX_R);
3181         } else {
3182           emit_opcode(cbuf, Assembler::REX_RX);
3183         }
3184       } else {
3185         if ($mem$$index < 8) {
3186           emit_opcode(cbuf, Assembler::REX_RB);
3187         } else {
3188           emit_opcode(cbuf, Assembler::REX_RXB);
3189         }
3190       }
3191     }
3192   %}
3193 
3194   enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3195   %{
3196     if ($reg$$reg < 8) {
3197       if ($mem$$base < 8) {
3198         if ($mem$$index < 8) {
3199           emit_opcode(cbuf, Assembler::REX_W);
3200         } else {
3201           emit_opcode(cbuf, Assembler::REX_WX);
3202         }
3203       } else {
3204         if ($mem$$index < 8) {
3205           emit_opcode(cbuf, Assembler::REX_WB);
3206         } else {
3207           emit_opcode(cbuf, Assembler::REX_WXB);
3208         }
3209       }
3210     } else {
3211       if ($mem$$base < 8) {
3212         if ($mem$$index < 8) {
3213           emit_opcode(cbuf, Assembler::REX_WR);
3214         } else {
3215           emit_opcode(cbuf, Assembler::REX_WRX);
3216         }
3217       } else {
3218         if ($mem$$index < 8) {
3219           emit_opcode(cbuf, Assembler::REX_WRB);
3220         } else {
3221           emit_opcode(cbuf, Assembler::REX_WRXB);
3222         }
3223       }
3224     }
3225   %}
3226 
3227   enc_class reg_mem(rRegI ereg, memory mem)
3228   %{
3229     // High registers handle in encode_RegMem
3230     int reg = $ereg$$reg;
3231     int base = $mem$$base;
3232     int index = $mem$$index;
3233     int scale = $mem$$scale;
3234     int disp = $mem$$disp;
3235     bool disp_is_oop = $mem->disp_is_oop();
3236 
3237     encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3238   %}
3239 
3240   enc_class RM_opc_mem(immI rm_opcode, memory mem)
3241   %{
3242     int rm_byte_opcode = $rm_opcode$$constant;
3243 
3244     // High registers handle in encode_RegMem
3245     int base = $mem$$base;
3246     int index = $mem$$index;
3247     int scale = $mem$$scale;
3248     int displace = $mem$$disp;
3249 
3250     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3251                                             // working with static
3252                                             // globals
3253     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3254                   disp_is_oop);
3255   %}
3256 
3257   enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3258   %{
3259     int reg_encoding = $dst$$reg;
3260     int base         = $src0$$reg;      // 0xFFFFFFFF indicates no base
3261     int index        = 0x04;            // 0x04 indicates no index
3262     int scale        = 0x00;            // 0x00 indicates no scale
3263     int displace     = $src1$$constant; // 0x00 indicates no displacement
3264     bool disp_is_oop = false;
3265     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3266                   disp_is_oop);
3267   %}
3268 
3269   enc_class neg_reg(rRegI dst)
3270   %{
3271     int dstenc = $dst$$reg;
3272     if (dstenc >= 8) {
3273       emit_opcode(cbuf, Assembler::REX_B);
3274       dstenc -= 8;
3275     }
3276     // NEG $dst
3277     emit_opcode(cbuf, 0xF7);
3278     emit_rm(cbuf, 0x3, 0x03, dstenc);
3279   %}
3280 
3281   enc_class neg_reg_wide(rRegI dst)
3282   %{
3283     int dstenc = $dst$$reg;
3284     if (dstenc < 8) {
3285       emit_opcode(cbuf, Assembler::REX_W);
3286     } else {
3287       emit_opcode(cbuf, Assembler::REX_WB);
3288       dstenc -= 8;
3289     }
3290     // NEG $dst
3291     emit_opcode(cbuf, 0xF7);
3292     emit_rm(cbuf, 0x3, 0x03, dstenc);
3293   %}
3294 
3295   enc_class setLT_reg(rRegI dst)
3296   %{
3297     int dstenc = $dst$$reg;
3298     if (dstenc >= 8) {
3299       emit_opcode(cbuf, Assembler::REX_B);
3300       dstenc -= 8;
3301     } else if (dstenc >= 4) {
3302       emit_opcode(cbuf, Assembler::REX);
3303     }
3304     // SETLT $dst
3305     emit_opcode(cbuf, 0x0F);
3306     emit_opcode(cbuf, 0x9C);
3307     emit_rm(cbuf, 0x3, 0x0, dstenc);
3308   %}
3309 
3310   enc_class setNZ_reg(rRegI dst)
3311   %{
3312     int dstenc = $dst$$reg;
3313     if (dstenc >= 8) {
3314       emit_opcode(cbuf, Assembler::REX_B);
3315       dstenc -= 8;
3316     } else if (dstenc >= 4) {
3317       emit_opcode(cbuf, Assembler::REX);
3318     }
3319     // SETNZ $dst
3320     emit_opcode(cbuf, 0x0F);
3321     emit_opcode(cbuf, 0x95);
3322     emit_rm(cbuf, 0x3, 0x0, dstenc);
3323   %}
3324 
3325   enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3326                        rcx_RegI tmp)
3327   %{
3328     // cadd_cmpLT
3329 
3330     int tmpReg = $tmp$$reg;
3331 
3332     int penc = $p$$reg;
3333     int qenc = $q$$reg;
3334     int yenc = $y$$reg;
3335 
3336     // subl $p,$q
3337     if (penc < 8) {
3338       if (qenc >= 8) {
3339         emit_opcode(cbuf, Assembler::REX_B);
3340       }
3341     } else {
3342       if (qenc < 8) {
3343         emit_opcode(cbuf, Assembler::REX_R);
3344       } else {
3345         emit_opcode(cbuf, Assembler::REX_RB);
3346       }
3347     }
3348     emit_opcode(cbuf, 0x2B);
3349     emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3350 
3351     // sbbl $tmp, $tmp
3352     emit_opcode(cbuf, 0x1B);
3353     emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3354 
3355     // andl $tmp, $y
3356     if (yenc >= 8) {
3357       emit_opcode(cbuf, Assembler::REX_B);
3358     }
3359     emit_opcode(cbuf, 0x23);
3360     emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3361 
3362     // addl $p,$tmp
3363     if (penc >= 8) {
3364         emit_opcode(cbuf, Assembler::REX_R);
3365     }
3366     emit_opcode(cbuf, 0x03);
3367     emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3368   %}
3369 
3370   // Compare the lonogs and set -1, 0, or 1 into dst
3371   enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3372   %{
3373     int src1enc = $src1$$reg;
3374     int src2enc = $src2$$reg;
3375     int dstenc = $dst$$reg;
3376 
3377     // cmpq $src1, $src2
3378     if (src1enc < 8) {
3379       if (src2enc < 8) {
3380         emit_opcode(cbuf, Assembler::REX_W);
3381       } else {
3382         emit_opcode(cbuf, Assembler::REX_WB);
3383       }
3384     } else {
3385       if (src2enc < 8) {
3386         emit_opcode(cbuf, Assembler::REX_WR);
3387       } else {
3388         emit_opcode(cbuf, Assembler::REX_WRB);
3389       }
3390     }
3391     emit_opcode(cbuf, 0x3B);
3392     emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3393 
3394     // movl $dst, -1
3395     if (dstenc >= 8) {
3396       emit_opcode(cbuf, Assembler::REX_B);
3397     }
3398     emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3399     emit_d32(cbuf, -1);
3400 
3401     // jl,s done
3402     emit_opcode(cbuf, 0x7C);
3403     emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3404 
3405     // setne $dst
3406     if (dstenc >= 4) {
3407       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3408     }
3409     emit_opcode(cbuf, 0x0F);
3410     emit_opcode(cbuf, 0x95);
3411     emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3412 
3413     // movzbl $dst, $dst
3414     if (dstenc >= 4) {
3415       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3416     }
3417     emit_opcode(cbuf, 0x0F);
3418     emit_opcode(cbuf, 0xB6);
3419     emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3420   %}
3421 
3422   enc_class Push_ResultXD(regD dst) %{
3423     int dstenc = $dst$$reg;
3424 
3425     store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3426 
3427     // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3428     emit_opcode  (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3429     if (dstenc >= 8) {
3430       emit_opcode(cbuf, Assembler::REX_R);
3431     }
3432     emit_opcode  (cbuf, 0x0F );
3433     emit_opcode  (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3434     encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3435 
3436     // add rsp,8
3437     emit_opcode(cbuf, Assembler::REX_W);
3438     emit_opcode(cbuf,0x83);
3439     emit_rm(cbuf,0x3, 0x0, RSP_enc);
3440     emit_d8(cbuf,0x08);
3441   %}
3442 
3443   enc_class Push_SrcXD(regD src) %{
3444     int srcenc = $src$$reg;
3445 
3446     // subq rsp,#8
3447     emit_opcode(cbuf, Assembler::REX_W);
3448     emit_opcode(cbuf, 0x83);
3449     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3450     emit_d8(cbuf, 0x8);
3451 
3452     // movsd [rsp],src
3453     emit_opcode(cbuf, 0xF2);
3454     if (srcenc >= 8) {
3455       emit_opcode(cbuf, Assembler::REX_R);
3456     }
3457     emit_opcode(cbuf, 0x0F);
3458     emit_opcode(cbuf, 0x11);
3459     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3460 
3461     // fldd [rsp]
3462     emit_opcode(cbuf, 0x66);
3463     emit_opcode(cbuf, 0xDD);
3464     encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3465   %}
3466 
3467 
3468   enc_class movq_ld(regD dst, memory mem) %{
3469     MacroAssembler _masm(&cbuf);
3470     Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3471     __ movq(as_XMMRegister($dst$$reg), madr);
3472   %}
3473 
3474   enc_class movq_st(memory mem, regD src) %{
3475     MacroAssembler _masm(&cbuf);
3476     Address madr = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
3477     __ movq(madr, as_XMMRegister($src$$reg));
3478   %}
3479 
3480   enc_class pshufd_8x8(regF dst, regF src) %{
3481     MacroAssembler _masm(&cbuf);
3482 
3483     encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3484     __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3485     __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3486   %}
3487 
3488   enc_class pshufd_4x16(regF dst, regF src) %{
3489     MacroAssembler _masm(&cbuf);
3490 
3491     __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3492   %}
3493 
3494   enc_class pshufd(regD dst, regD src, int mode) %{
3495     MacroAssembler _masm(&cbuf);
3496 
3497     __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3498   %}
3499 
3500   enc_class pxor(regD dst, regD src) %{
3501     MacroAssembler _masm(&cbuf);
3502 
3503     __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3504   %}
3505 
3506   enc_class mov_i2x(regD dst, rRegI src) %{
3507     MacroAssembler _masm(&cbuf);
3508 
3509     __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3510   %}
3511 
3512   // obj: object to lock
3513   // box: box address (header location) -- killed
3514   // tmp: rax -- killed
3515   // scr: rbx -- killed
3516   //
3517   // What follows is a direct transliteration of fast_lock() and fast_unlock()
3518   // from i486.ad.  See that file for comments.
3519   // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3520   // use the shorter encoding.  (Movl clears the high-order 32-bits).
3521 
3522 
3523   enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3524   %{
3525     Register objReg = as_Register((int)$obj$$reg);
3526     Register boxReg = as_Register((int)$box$$reg);
3527     Register tmpReg = as_Register($tmp$$reg);
3528     Register scrReg = as_Register($scr$$reg);
3529     MacroAssembler masm(&cbuf);
3530 
3531     // Verify uniqueness of register assignments -- necessary but not sufficient
3532     assert (objReg != boxReg && objReg != tmpReg &&
3533             objReg != scrReg && tmpReg != scrReg, "invariant") ;
3534 
3535     if (_counters != NULL) {
3536       masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3537     }
3538     if (EmitSync & 1) {
3539         // Without cast to int32_t a movptr will destroy r10 which is typically obj
3540         masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; 
3541         masm.cmpptr(rsp, (int32_t)NULL_WORD) ; 
3542     } else
3543     if (EmitSync & 2) {
3544         Label DONE_LABEL;
3545         if (UseBiasedLocking) {
3546            // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3547           masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3548         }
3549         // QQQ was movl...
3550         masm.movptr(tmpReg, 0x1);
3551         masm.orptr(tmpReg, Address(objReg, 0));
3552         masm.movptr(Address(boxReg, 0), tmpReg);
3553         if (os::is_MP()) {
3554           masm.lock();
3555         }
3556         masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3557         masm.jcc(Assembler::equal, DONE_LABEL);
3558 
3559         // Recursive locking
3560         masm.subptr(tmpReg, rsp);
3561         masm.andptr(tmpReg, 7 - os::vm_page_size());
3562         masm.movptr(Address(boxReg, 0), tmpReg);
3563 
3564         masm.bind(DONE_LABEL);
3565         masm.nop(); // avoid branch to branch
3566     } else {
3567         Label DONE_LABEL, IsInflated, Egress;
3568 
3569         masm.movptr(tmpReg, Address(objReg, 0)) ; 
3570         masm.testl (tmpReg, 0x02) ;         // inflated vs stack-locked|neutral|biased
3571         masm.jcc   (Assembler::notZero, IsInflated) ; 
3572          
3573         // it's stack-locked, biased or neutral
3574         // TODO: optimize markword triage order to reduce the number of
3575         // conditional branches in the most common cases.
3576         // Beware -- there's a subtle invariant that fetch of the markword
3577         // at [FETCH], below, will never observe a biased encoding (*101b).
3578         // If this invariant is not held we'll suffer exclusion (safety) failure.
3579 
3580         if (UseBiasedLocking && !UseOptoBiasInlining) {
3581           masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3582           masm.movptr(tmpReg, Address(objReg, 0)) ;        // [FETCH]
3583         }
3584 
3585         // was q will it destroy high?
3586         masm.orl   (tmpReg, 1) ; 
3587         masm.movptr(Address(boxReg, 0), tmpReg) ;  
3588         if (os::is_MP()) { masm.lock(); } 
3589         masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3590         if (_counters != NULL) {
3591            masm.cond_inc32(Assembler::equal,
3592                            ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3593         }
3594         masm.jcc   (Assembler::equal, DONE_LABEL);
3595 
3596         // Recursive locking
3597         masm.subptr(tmpReg, rsp);
3598         masm.andptr(tmpReg, 7 - os::vm_page_size());
3599         masm.movptr(Address(boxReg, 0), tmpReg);
3600         if (_counters != NULL) {
3601            masm.cond_inc32(Assembler::equal,
3602                            ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3603         }
3604         masm.jmp   (DONE_LABEL) ;
3605 
3606         masm.bind  (IsInflated) ;
3607         // It's inflated
3608 
3609         // TODO: someday avoid the ST-before-CAS penalty by
3610         // relocating (deferring) the following ST.
3611         // We should also think about trying a CAS without having
3612         // fetched _owner.  If the CAS is successful we may
3613         // avoid an RTO->RTS upgrade on the $line.
3614         // Without cast to int32_t a movptr will destroy r10 which is typically obj
3615         masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; 
3616 
3617         masm.mov    (boxReg, tmpReg) ; 
3618         masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3619         masm.testptr(tmpReg, tmpReg) ;   
3620         masm.jcc    (Assembler::notZero, DONE_LABEL) ; 
3621 
3622         // It's inflated and appears unlocked
3623         if (os::is_MP()) { masm.lock(); } 
3624         masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3625         // Intentional fall-through into DONE_LABEL ...
3626 
3627         masm.bind  (DONE_LABEL) ;
3628         masm.nop   () ;                 // avoid jmp to jmp
3629     }
3630   %}
3631 
3632   // obj: object to unlock
3633   // box: box address (displaced header location), killed
3634   // RBX: killed tmp; cannot be obj nor box
3635   enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3636   %{
3637 
3638     Register objReg = as_Register($obj$$reg);
3639     Register boxReg = as_Register($box$$reg);
3640     Register tmpReg = as_Register($tmp$$reg);
3641     MacroAssembler masm(&cbuf);
3642 
3643     if (EmitSync & 4) { 
3644        masm.cmpptr(rsp, 0) ; 
3645     } else
3646     if (EmitSync & 8) {
3647        Label DONE_LABEL;
3648        if (UseBiasedLocking) {
3649          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3650        }
3651 
3652        // Check whether the displaced header is 0
3653        //(=> recursive unlock)
3654        masm.movptr(tmpReg, Address(boxReg, 0));
3655        masm.testptr(tmpReg, tmpReg);
3656        masm.jcc(Assembler::zero, DONE_LABEL);
3657 
3658        // If not recursive lock, reset the header to displaced header
3659        if (os::is_MP()) {
3660          masm.lock();
3661        }
3662        masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3663        masm.bind(DONE_LABEL);
3664        masm.nop(); // avoid branch to branch
3665     } else {
3666        Label DONE_LABEL, Stacked, CheckSucc ;
3667 
3668        if (UseBiasedLocking && !UseOptoBiasInlining) {
3669          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3670        }
3671         
3672        masm.movptr(tmpReg, Address(objReg, 0)) ; 
3673        masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ; 
3674        masm.jcc   (Assembler::zero, DONE_LABEL) ; 
3675        masm.testl (tmpReg, 0x02) ; 
3676        masm.jcc   (Assembler::zero, Stacked) ; 
3677         
3678        // It's inflated
3679        masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3680        masm.xorptr(boxReg, r15_thread) ; 
3681        masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ; 
3682        masm.jcc   (Assembler::notZero, DONE_LABEL) ; 
3683        masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
3684        masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
3685        masm.jcc   (Assembler::notZero, CheckSucc) ; 
3686        masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ; 
3687        masm.jmp   (DONE_LABEL) ; 
3688         
3689        if ((EmitSync & 65536) == 0) { 
3690          Label LSuccess, LGoSlowPath ;
3691          masm.bind  (CheckSucc) ;
3692          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3693          masm.jcc   (Assembler::zero, LGoSlowPath) ;
3694 
3695          // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3696          // the explicit ST;MEMBAR combination, but masm doesn't currently support
3697          // "ANDQ M,IMM".  Don't use MFENCE here.  lock:add to TOS, xchg, etc
3698          // are all faster when the write buffer is populated.
3699          masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3700          if (os::is_MP()) {
3701             masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3702          }
3703          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3704          masm.jcc   (Assembler::notZero, LSuccess) ;
3705 
3706          masm.movptr (boxReg, (int32_t)NULL_WORD) ;                   // box is really EAX
3707          if (os::is_MP()) { masm.lock(); }
3708          masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3709          masm.jcc   (Assembler::notEqual, LSuccess) ;
3710          // Intentional fall-through into slow-path
3711 
3712          masm.bind  (LGoSlowPath) ;
3713          masm.orl   (boxReg, 1) ;                      // set ICC.ZF=0 to indicate failure
3714          masm.jmp   (DONE_LABEL) ;
3715 
3716          masm.bind  (LSuccess) ;
3717          masm.testl (boxReg, 0) ;                      // set ICC.ZF=1 to indicate success
3718          masm.jmp   (DONE_LABEL) ;
3719        }
3720 
3721        masm.bind  (Stacked) ; 
3722        masm.movptr(tmpReg, Address (boxReg, 0)) ;      // re-fetch
3723        if (os::is_MP()) { masm.lock(); } 
3724        masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3725 
3726        if (EmitSync & 65536) {
3727           masm.bind (CheckSucc) ;
3728        }
3729        masm.bind(DONE_LABEL);
3730        if (EmitSync & 32768) {
3731           masm.nop();                      // avoid branch to branch
3732        }
3733     }
3734   %}
3735 
3736   enc_class enc_String_Compare()
3737   %{
3738     Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
3739           POP_LABEL, DONE_LABEL, CONT_LABEL,
3740           WHILE_HEAD_LABEL;
3741     MacroAssembler masm(&cbuf);
3742 
3743     // Get the first character position in both strings
3744     //         [8] char array, [12] offset, [16] count
3745     int value_offset  = java_lang_String::value_offset_in_bytes();
3746     int offset_offset = java_lang_String::offset_offset_in_bytes();
3747     int count_offset  = java_lang_String::count_offset_in_bytes();
3748     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3749 
3750     masm.load_heap_oop(rax, Address(rsi, value_offset));
3751     masm.movl(rcx, Address(rsi, offset_offset));
3752     masm.lea(rax, Address(rax, rcx, Address::times_2, base_offset));
3753     masm.load_heap_oop(rbx, Address(rdi, value_offset));
3754     masm.movl(rcx, Address(rdi, offset_offset));
3755     masm.lea(rbx, Address(rbx, rcx, Address::times_2, base_offset));
3756 
3757     // Compute the minimum of the string lengths(rsi) and the
3758     // difference of the string lengths (stack)
3759 
3760     // do the conditional move stuff
3761     masm.movl(rdi, Address(rdi, count_offset));
3762     masm.movl(rsi, Address(rsi, count_offset));
3763     masm.movl(rcx, rdi);
3764     masm.subl(rdi, rsi);
3765     masm.push(rdi);
3766     masm.cmov(Assembler::lessEqual, rsi, rcx);
3767 
3768     // Is the minimum length zero?
3769     masm.bind(RCX_GOOD_LABEL);
3770     masm.testl(rsi, rsi);
3771     masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
3772 
3773     // Load first characters
3774     masm.load_unsigned_word(rcx, Address(rbx, 0));
3775     masm.load_unsigned_word(rdi, Address(rax, 0));
3776 
3777     // Compare first characters
3778     masm.subl(rcx, rdi);
3779     masm.jccb(Assembler::notZero,  POP_LABEL);
3780     masm.decrementl(rsi);
3781     masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
3782 
3783     {
3784       // Check after comparing first character to see if strings are equivalent
3785       Label LSkip2;
3786       // Check if the strings start at same location
3787       masm.cmpptr(rbx, rax);
3788       masm.jccb(Assembler::notEqual, LSkip2);
3789 
3790       // Check if the length difference is zero (from stack)
3791       masm.cmpl(Address(rsp, 0), 0x0);
3792       masm.jccb(Assembler::equal,  LENGTH_DIFF_LABEL);
3793 
3794       // Strings might not be equivalent
3795       masm.bind(LSkip2);
3796     }
3797 
3798     // Advance to next character
3799     masm.addptr(rax, 2);
3800     masm.addptr(rbx, 2);
3801 
3802     if (UseSSE>=4 && UseSSE42Intrinsics) {
3803       // With SSE4.2, use double quad vector compare
3804       Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
3805       // Setup to compare 16-byte vectors
3806       masm.movl(rdi, rsi);
3807       masm.andl(rsi, 0xfffffff8); // rsi holds the vector count
3808       masm.andl(rdi, 0x00000007); // rdi holds the tail count
3809       masm.testl(rsi, rsi);
3810       masm.jccb(Assembler::zero, COMPARE_TAIL);
3811       
3812       masm.lea(rax, Address(rax, rsi, Address::times_2));
3813       masm.lea(rbx, Address(rbx, rsi, Address::times_2));
3814       masm.negptr(rsi);
3815       
3816       masm.bind(COMPARE_VECTORS);
3817       masm.movdqu(xmm6, Address(rax, rsi, Address::times_2));
3818       masm.movdqu(xmm7, Address(rbx, rsi, Address::times_2));
3819       masm.pxor(xmm6, xmm7);
3820       masm.ptest(xmm6, xmm6);
3821       masm.jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
3822       masm.addptr(rsi, 8);
3823       masm.jcc(Assembler::notZero, COMPARE_VECTORS);
3824       masm.jmpb(COMPARE_TAIL);
3825       
3826       // Mismatched characters in the vectors
3827       masm.bind(VECTOR_NOT_EQUAL);
3828       masm.lea(rax, Address(rax, rsi, Address::times_2));
3829       masm.lea(rbx, Address(rbx, rsi, Address::times_2));
3830       masm.movl(rdi, 8);
3831   
3832       // Compare tail (< 8 chars), or rescan last vectors to 
3833       // find 1st mismatched characters
3834       masm.bind(COMPARE_TAIL);
3835       masm.testl(rdi, rdi);
3836       masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
3837       masm.movl(rsi, rdi);
3838       // Fallthru to tail compare
3839     } 
3840 
3841     // Shift RAX and RBX to the end of the arrays, negate min
3842     masm.lea(rax, Address(rax, rsi, Address::times_2, 0));
3843     masm.lea(rbx, Address(rbx, rsi, Address::times_2, 0));
3844     masm.negptr(rsi);
3845 
3846     // Compare the rest of the characters
3847     masm.bind(WHILE_HEAD_LABEL);
3848     masm.load_unsigned_word(rcx, Address(rbx, rsi, Address::times_2, 0));
3849     masm.load_unsigned_word(rdi, Address(rax, rsi, Address::times_2, 0));
3850     masm.subl(rcx, rdi);
3851     masm.jccb(Assembler::notZero, POP_LABEL);
3852     masm.increment(rsi);
3853     masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
3854 
3855     // Strings are equal up to min length.  Return the length difference.
3856     masm.bind(LENGTH_DIFF_LABEL);
3857     masm.pop(rcx);
3858     masm.jmpb(DONE_LABEL);
3859 
3860     // Discard the stored length difference
3861     masm.bind(POP_LABEL);
3862     masm.addptr(rsp, 8);
3863        
3864     // That's it
3865     masm.bind(DONE_LABEL);
3866   %}
3867 
3868  enc_class enc_String_IndexOf() %{
3869     // SSE4.2 version 
3870     Label LOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR, 
3871           SCAN_SUBSTR, RET_NEG_ONE, RET_NOT_FOUND, CLEANUP, DONE;
3872     MacroAssembler masm(&cbuf);
3873     
3874     // Get the first character position in both strings
3875     //         [8] char array, [12] offset, [16] count
3876     int value_offset  = java_lang_String::value_offset_in_bytes();
3877     int offset_offset = java_lang_String::offset_offset_in_bytes();
3878     int count_offset  = java_lang_String::count_offset_in_bytes();
3879     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3880    
3881     // Get counts for string and substr
3882     masm.movl(rdx, Address(rsi, count_offset));
3883     masm.movl(rax, Address(rdi, count_offset));
3884     // Check for substr count > string count
3885     masm.cmpl(rax, rdx);
3886     masm.jccb(Assembler::greater, RET_NEG_ONE);
3887     
3888     // Start the indexOf operation
3889     // Get start addr of string
3890     masm.load_heap_oop(rbx, Address(rsi, value_offset));
3891     masm.movl(rcx, Address(rsi, offset_offset));
3892     masm.lea(rsi, Address(rbx, rcx, Address::times_2, base_offset));
3893     masm.push(rsi);
3894 
3895     // Get start addr of substr
3896     masm.load_heap_oop(rbx, Address(rdi, value_offset));
3897     masm.movl(rcx, Address(rdi, offset_offset));
3898     masm.lea(rdi, Address(rbx, rcx, Address::times_2, base_offset));
3899     masm.push(rdi);
3900     masm.push(rax);
3901     masm.jmpb(PREP_FOR_SCAN);
3902 
3903     // Substr count saved at sp
3904     // Substr saved at sp+8
3905     // String saved at sp+16
3906 
3907     // Prep to load substr for scan
3908     masm.bind(LOAD_SUBSTR);
3909     masm.movptr(rdi, Address(rsp, 8));
3910     masm.movl(rax, Address(rsp, 0));
3911 
3912     // Load substr
3913     masm.bind(PREP_FOR_SCAN);
3914     masm.movdqu(xmm6, Address(rdi, 0));
3915     masm.addq(rdx, 8);    // prime the loop
3916     masm.subptr(rsi, 16);    
3917 
3918 
3919     // Scan string for substr in 16-byte vectors
3920     masm.bind(SCAN_TO_SUBSTR);
3921     masm.subq(rdx, 8);
3922     masm.addptr(rsi, 16);
3923     masm.pcmpestri(xmm6, Address(rsi, 0), 0x0d);
3924     masm.jcc(Assembler::above, SCAN_TO_SUBSTR);
3925     masm.jccb(Assembler::aboveEqual, RET_NOT_FOUND);    
3926     
3927     // Fallthru: found a potential substr
3928 
3929     //Make sure string is still long enough
3930     masm.subl(rdx, rcx);
3931     masm.cmpl(rdx, rax);
3932     masm.jccb(Assembler::negative, RET_NOT_FOUND);
3933     // Compute start addr of substr
3934     masm.lea(rsi, Address(rsi, rcx, Address::times_2));
3935     masm.movptr(rbx, rsi);
3936     
3937     // Compare potential substr 
3938     masm.addq(rdx, 8);        // prime the loop
3939     masm.addq(rax, 8);
3940     masm.subptr(rsi, 16);
3941     masm.subptr(rdi, 16);
3942 
3943     // Scan 16-byte vectors of string and substr
3944     masm.bind(SCAN_SUBSTR);
3945     masm.subq(rax, 8);
3946     masm.subq(rdx, 8);
3947     masm.addptr(rsi, 16);
3948     masm.addptr(rdi, 16);
3949     masm.movdqu(xmm6, Address(rdi, 0));
3950     masm.pcmpestri(xmm6, Address(rsi, 0), 0x0d);
3951     masm.jcc(Assembler::noOverflow, LOAD_SUBSTR);   // OF == 0
3952     masm.jcc(Assembler::positive, SCAN_SUBSTR);     // SF == 0
3953     
3954     // Compute substr offset
3955     masm.movptr(rsi, Address(rsp, 16));
3956     masm.subptr(rbx, rsi);
3957     masm.shrl(rbx, 1);
3958     masm.jmpb(CLEANUP);
3959 
3960     masm.bind(RET_NEG_ONE);
3961     masm.movl(rbx, -1);
3962     masm.jmpb(DONE);
3963 
3964     masm.bind(RET_NOT_FOUND);
3965     masm.movl(rbx, -1);
3966 
3967     masm.bind(CLEANUP);
3968     masm.addptr(rsp, 24);
3969 
3970     masm.bind(DONE);
3971   %}
3972 
3973   enc_class enc_String_Equals() %{
3974     Label RET_TRUE, RET_FALSE, DONE, COMPARE_VECTORS, COMPARE_CHAR;
3975     MacroAssembler masm(&cbuf);
3976 
3977     int value_offset  = java_lang_String::value_offset_in_bytes();
3978     int offset_offset = java_lang_String::offset_offset_in_bytes();
3979     int count_offset  = java_lang_String::count_offset_in_bytes();
3980     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3981     
3982     // does source == target string?
3983     masm.cmpptr(rdi, rsi);
3984     masm.jccb(Assembler::equal, RET_TRUE);
3985 
3986     // get and compare counts
3987     masm.movl(rcx, Address(rdi, count_offset));
3988     masm.movl(rax, Address(rsi, count_offset));
3989     masm.cmpl(rcx, rax);
3990     masm.jccb(Assembler::notEqual, RET_FALSE);
3991     masm.testl(rax, rax);
3992     masm.jccb(Assembler::zero, RET_TRUE);
3993 
3994     // get source string offset and value
3995     masm.load_heap_oop(rbx, Address(rsi, value_offset));
3996     masm.movl(rax, Address(rsi, offset_offset));
3997     masm.lea(rsi, Address(rbx, rax, Address::times_2, base_offset));
3998     
3999     // get compare string offset and value 
4000     masm.load_heap_oop(rbx, Address(rdi, value_offset));
4001     masm.movl(rax, Address(rdi, offset_offset));
4002     masm.lea(rdi, Address(rbx, rax, Address::times_2, base_offset));
4003 
4004     // Set byte count
4005     masm.shll(rcx, 1);
4006     masm.movl(rax, rcx);
4007 
4008     if (UseSSE>=4 && UseSSE42Intrinsics) {
4009       // With SSE4.2, use double quad vector compare
4010       Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
4011       // Compare 16-byte vectors
4012       masm.andl(rcx, 0xfffffff0);  // vector count (in bytes)
4013       masm.andl(rax, 0x0000000e);  // tail count (in bytes)
4014       masm.testl(rcx, rcx);
4015       masm.jccb(Assembler::zero, COMPARE_TAIL);
4016       masm.lea(rdi, Address(rdi, rcx, Address::times_1));
4017       masm.lea(rsi, Address(rsi, rcx, Address::times_1));
4018       masm.negptr(rcx);
4019 
4020       masm.bind(COMPARE_WIDE_VECTORS);
4021       masm.movdqu(xmm6, Address(rdi, rcx, Address::times_1));
4022       masm.movdqu(xmm7, Address(rsi, rcx, Address::times_1));
4023       masm.pxor(xmm6, xmm7);
4024       masm.ptest(xmm6, xmm6);
4025       masm.jccb(Assembler::notZero, RET_FALSE);
4026       masm.addptr(rcx, 16);
4027       masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
4028       masm.bind(COMPARE_TAIL);
4029       masm.movl(rcx, rax);
4030       // Fallthru to tail compare
4031     }
4032 
4033     // Compare 4-byte vectors
4034     masm.andl(rcx, 0xfffffffc);  // vector count (in bytes)
4035     masm.andl(rax, 0x00000002);  // tail char (in bytes)
4036     masm.testl(rcx, rcx);
4037     masm.jccb(Assembler::zero, COMPARE_CHAR);
4038     masm.lea(rdi, Address(rdi, rcx, Address::times_1));
4039     masm.lea(rsi, Address(rsi, rcx, Address::times_1));
4040     masm.negptr(rcx);
4041       
4042     masm.bind(COMPARE_VECTORS);
4043     masm.movl(rbx, Address(rdi, rcx, Address::times_1));
4044     masm.cmpl(rbx, Address(rsi, rcx, Address::times_1));
4045     masm.jccb(Assembler::notEqual, RET_FALSE);
4046     masm.addptr(rcx, 4);
4047     masm.jcc(Assembler::notZero, COMPARE_VECTORS);
4048 
4049     // Compare trailing char (final 2 bytes), if any
4050     masm.bind(COMPARE_CHAR);
4051     masm.testl(rax, rax);
4052     masm.jccb(Assembler::zero, RET_TRUE);
4053     masm.load_unsigned_word(rbx, Address(rdi, 0));
4054     masm.load_unsigned_word(rcx, Address(rsi, 0));
4055     masm.cmpl(rbx, rcx);
4056     masm.jccb(Assembler::notEqual, RET_FALSE);
4057 
4058     masm.bind(RET_TRUE);
4059     masm.movl(rax, 1);   // return true
4060     masm.jmpb(DONE);
4061 
4062     masm.bind(RET_FALSE);
4063     masm.xorl(rax, rax); // return false
4064 
4065     masm.bind(DONE);
4066     %}
4067 
4068 
4069   enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, rbx_RegI tmp2, regD6 tmp3, regD7 tmp4, rcx_RegI result) %{
4070     Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
4071     MacroAssembler masm(&cbuf);
4072 
4073     Register ary1Reg   = as_Register($ary1$$reg);
4074     Register ary2Reg   = as_Register($ary2$$reg);
4075     Register tmp1Reg   = as_Register($tmp1$$reg);
4076     Register tmp2Reg   = as_Register($tmp2$$reg);
4077     Register resultReg = as_Register($result$$reg);
4078 
4079     int length_offset  = arrayOopDesc::length_offset_in_bytes();
4080     int base_offset    = arrayOopDesc::base_offset_in_bytes(T_CHAR);
4081 
4082     // Check the input args
4083     masm.cmpq(ary1Reg, ary2Reg);                        
4084     masm.jccb(Assembler::equal, TRUE_LABEL);
4085     masm.testq(ary1Reg, ary1Reg);                       
4086     masm.jccb(Assembler::zero, FALSE_LABEL);
4087     masm.testq(ary2Reg, ary2Reg);                       
4088     masm.jccb(Assembler::zero, FALSE_LABEL);
4089 
4090     // Check the lengths
4091     masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
4092     masm.movl(resultReg, Address(ary2Reg, length_offset));
4093     masm.cmpl(tmp2Reg, resultReg);
4094     masm.jccb(Assembler::notEqual, FALSE_LABEL);
4095     masm.testl(resultReg, resultReg);
4096     masm.jccb(Assembler::zero, TRUE_LABEL);
4097 
4098     //load array address
4099     masm.lea(ary1Reg, Address(ary1Reg, base_offset));
4100     masm.lea(ary2Reg, Address(ary2Reg, base_offset));
4101 
4102     //set byte count
4103     masm.shll(tmp2Reg, 1);
4104     masm.movl(resultReg,tmp2Reg);
4105 
4106     if (UseSSE>=4 && UseSSE42Intrinsics){  
4107       // With SSE4.2, use double quad vector compare
4108       Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
4109       // Compare 16-byte vectors
4110       masm.andl(tmp2Reg, 0xfffffff0);    // vector count (in bytes)
4111       masm.andl(resultReg, 0x0000000e);  // tail count (in bytes)
4112       masm.testl(tmp2Reg, tmp2Reg);
4113       masm.jccb(Assembler::zero, COMPARE_TAIL);
4114       masm.lea(ary1Reg, Address(ary1Reg, tmp2Reg, Address::times_1));
4115       masm.lea(ary2Reg, Address(ary2Reg, tmp2Reg, Address::times_1));
4116       masm.negptr(tmp2Reg);
4117       
4118       masm.bind(COMPARE_WIDE_VECTORS);
4119       masm.movdqu(xmm6, Address(ary1Reg, tmp2Reg, Address::times_1));
4120       masm.movdqu(xmm7, Address(ary2Reg, tmp2Reg, Address::times_1));
4121       masm.pxor(xmm6, xmm7);
4122       masm.ptest(xmm6, xmm6);
4123       
4124       masm.jccb(Assembler::notZero, FALSE_LABEL);
4125       masm.addptr(tmp2Reg, 16);
4126       masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
4127       masm.bind(COMPARE_TAIL);
4128       masm.movl(tmp2Reg, resultReg);
4129       // Fallthru to tail compare
4130     }
4131 
4132    // Compare 4-byte vectors
4133     masm.andl(tmp2Reg, 0xfffffffc);    // vector count (in bytes)
4134     masm.andl(resultReg, 0x00000002);  // tail char (in bytes)
4135     masm.testl(tmp2Reg, tmp2Reg); //if tmp2 == 0, only compare char
4136     masm.jccb(Assembler::zero, COMPARE_CHAR);
4137     masm.lea(ary1Reg, Address(ary1Reg, tmp2Reg, Address::times_1));
4138     masm.lea(ary2Reg, Address(ary2Reg, tmp2Reg, Address::times_1));
4139     masm.negptr(tmp2Reg);
4140       
4141     masm.bind(COMPARE_VECTORS);
4142     masm.movl(tmp1Reg, Address(ary1Reg, tmp2Reg, Address::times_1));
4143     masm.cmpl(tmp1Reg, Address(ary2Reg, tmp2Reg, Address::times_1));
4144     masm.jccb(Assembler::notEqual, FALSE_LABEL);
4145     masm.addptr(tmp2Reg, 4);
4146     masm.jcc(Assembler::notZero, COMPARE_VECTORS);
4147 
4148     // Compare trailing char (final 2 bytes), if any
4149     masm.bind(COMPARE_CHAR);
4150     masm.testl(resultReg, resultReg);
4151     masm.jccb(Assembler::zero, TRUE_LABEL);
4152     masm.load_unsigned_word(tmp1Reg, Address(ary1Reg, 0));
4153     masm.load_unsigned_word(tmp2Reg, Address(ary2Reg, 0));
4154     masm.cmpl(tmp1Reg, tmp2Reg);
4155     masm.jccb(Assembler::notEqual, FALSE_LABEL);
4156 
4157     masm.bind(TRUE_LABEL);
4158     masm.movl(resultReg, 1);   // return true
4159     masm.jmpb(DONE);
4160 
4161     masm.bind(FALSE_LABEL);
4162     masm.xorl(resultReg, resultReg); // return false
4163 
4164     // That's it
4165     masm.bind(DONE);
4166   %}
4167 
4168   enc_class enc_rethrow()
4169   %{
4170     cbuf.set_inst_mark();
4171     emit_opcode(cbuf, 0xE9); // jmp entry
4172     emit_d32_reloc(cbuf,
4173                    (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
4174                    runtime_call_Relocation::spec(),
4175                    RELOC_DISP32);
4176   %}
4177 
4178   enc_class absF_encoding(regF dst)
4179   %{
4180     int dstenc = $dst$$reg;
4181     address signmask_address = (address) StubRoutines::x86::float_sign_mask();
4182 
4183     cbuf.set_inst_mark();
4184     if (dstenc >= 8) {
4185       emit_opcode(cbuf, Assembler::REX_R);
4186       dstenc -= 8;
4187     }
4188     // XXX reg_mem doesn't support RIP-relative addressing yet
4189     emit_opcode(cbuf, 0x0F);
4190     emit_opcode(cbuf, 0x54);
4191     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
4192     emit_d32_reloc(cbuf, signmask_address);
4193   %}
4194 
4195   enc_class absD_encoding(regD dst)
4196   %{
4197     int dstenc = $dst$$reg;
4198     address signmask_address = (address) StubRoutines::x86::double_sign_mask();
4199 
4200     cbuf.set_inst_mark();
4201     emit_opcode(cbuf, 0x66);
4202     if (dstenc >= 8) {
4203       emit_opcode(cbuf, Assembler::REX_R);
4204       dstenc -= 8;
4205     }
4206     // XXX reg_mem doesn't support RIP-relative addressing yet
4207     emit_opcode(cbuf, 0x0F);
4208     emit_opcode(cbuf, 0x54);
4209     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
4210     emit_d32_reloc(cbuf, signmask_address);
4211   %}
4212 
4213   enc_class negF_encoding(regF dst)
4214   %{
4215     int dstenc = $dst$$reg;
4216     address signflip_address = (address) StubRoutines::x86::float_sign_flip();
4217 
4218     cbuf.set_inst_mark();
4219     if (dstenc >= 8) {
4220       emit_opcode(cbuf, Assembler::REX_R);
4221       dstenc -= 8;
4222     }
4223     // XXX reg_mem doesn't support RIP-relative addressing yet
4224     emit_opcode(cbuf, 0x0F);
4225     emit_opcode(cbuf, 0x57);
4226     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
4227     emit_d32_reloc(cbuf, signflip_address);
4228   %}
4229 
4230   enc_class negD_encoding(regD dst)
4231   %{
4232     int dstenc = $dst$$reg;
4233     address signflip_address = (address) StubRoutines::x86::double_sign_flip();
4234 
4235     cbuf.set_inst_mark();
4236     emit_opcode(cbuf, 0x66);
4237     if (dstenc >= 8) {
4238       emit_opcode(cbuf, Assembler::REX_R);
4239       dstenc -= 8;
4240     }
4241     // XXX reg_mem doesn't support RIP-relative addressing yet
4242     emit_opcode(cbuf, 0x0F);
4243     emit_opcode(cbuf, 0x57);
4244     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
4245     emit_d32_reloc(cbuf, signflip_address);
4246   %}
4247 
4248   enc_class f2i_fixup(rRegI dst, regF src)
4249   %{
4250     int dstenc = $dst$$reg;
4251     int srcenc = $src$$reg;
4252 
4253     // cmpl $dst, #0x80000000
4254     if (dstenc >= 8) {
4255       emit_opcode(cbuf, Assembler::REX_B);
4256     }
4257     emit_opcode(cbuf, 0x81);
4258     emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
4259     emit_d32(cbuf, 0x80000000);
4260 
4261     // jne,s done
4262     emit_opcode(cbuf, 0x75);
4263     if (srcenc < 8 && dstenc < 8) {
4264       emit_d8(cbuf, 0xF);
4265     } else if (srcenc >= 8 && dstenc >= 8) {
4266       emit_d8(cbuf, 0x11);
4267     } else {
4268       emit_d8(cbuf, 0x10);
4269     }
4270 
4271     // subq rsp, #8
4272     emit_opcode(cbuf, Assembler::REX_W);
4273     emit_opcode(cbuf, 0x83);
4274     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4275     emit_d8(cbuf, 8);
4276 
4277     // movss [rsp], $src
4278     emit_opcode(cbuf, 0xF3);
4279     if (srcenc >= 8) {
4280       emit_opcode(cbuf, Assembler::REX_R);
4281     }
4282     emit_opcode(cbuf, 0x0F);
4283     emit_opcode(cbuf, 0x11);
4284     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4285 
4286     // call f2i_fixup
4287     cbuf.set_inst_mark();
4288     emit_opcode(cbuf, 0xE8);
4289     emit_d32_reloc(cbuf,
4290                    (int)
4291                    (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4),
4292                    runtime_call_Relocation::spec(),
4293                    RELOC_DISP32);
4294 
4295     // popq $dst
4296     if (dstenc >= 8) {
4297       emit_opcode(cbuf, Assembler::REX_B);
4298     }
4299     emit_opcode(cbuf, 0x58 | (dstenc & 7));
4300 
4301     // done:
4302   %}
4303 
4304   enc_class f2l_fixup(rRegL dst, regF src)
4305   %{
4306     int dstenc = $dst$$reg;
4307     int srcenc = $src$$reg;
4308     address const_address = (address) StubRoutines::x86::double_sign_flip();
4309 
4310     // cmpq $dst, [0x8000000000000000]
4311     cbuf.set_inst_mark();
4312     emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4313     emit_opcode(cbuf, 0x39);
4314     // XXX reg_mem doesn't support RIP-relative addressing yet
4315     emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4316     emit_d32_reloc(cbuf, const_address);
4317 
4318 
4319     // jne,s done
4320     emit_opcode(cbuf, 0x75);
4321     if (srcenc < 8 && dstenc < 8) {
4322       emit_d8(cbuf, 0xF);
4323     } else if (srcenc >= 8 && dstenc >= 8) {
4324       emit_d8(cbuf, 0x11);
4325     } else {
4326       emit_d8(cbuf, 0x10);
4327     }
4328 
4329     // subq rsp, #8
4330     emit_opcode(cbuf, Assembler::REX_W);
4331     emit_opcode(cbuf, 0x83);
4332     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4333     emit_d8(cbuf, 8);
4334 
4335     // movss [rsp], $src
4336     emit_opcode(cbuf, 0xF3);
4337     if (srcenc >= 8) {
4338       emit_opcode(cbuf, Assembler::REX_R);
4339     }
4340     emit_opcode(cbuf, 0x0F);
4341     emit_opcode(cbuf, 0x11);
4342     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4343 
4344     // call f2l_fixup
4345     cbuf.set_inst_mark();
4346     emit_opcode(cbuf, 0xE8);
4347     emit_d32_reloc(cbuf,
4348                    (int)
4349                    (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4),
4350                    runtime_call_Relocation::spec(),
4351                    RELOC_DISP32);
4352 
4353     // popq $dst
4354     if (dstenc >= 8) {
4355       emit_opcode(cbuf, Assembler::REX_B);
4356     }
4357     emit_opcode(cbuf, 0x58 | (dstenc & 7));
4358 
4359     // done:
4360   %}
4361 
4362   enc_class d2i_fixup(rRegI dst, regD src)
4363   %{
4364     int dstenc = $dst$$reg;
4365     int srcenc = $src$$reg;
4366 
4367     // cmpl $dst, #0x80000000
4368     if (dstenc >= 8) {
4369       emit_opcode(cbuf, Assembler::REX_B);
4370     }
4371     emit_opcode(cbuf, 0x81);
4372     emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
4373     emit_d32(cbuf, 0x80000000);
4374 
4375     // jne,s done
4376     emit_opcode(cbuf, 0x75);
4377     if (srcenc < 8 && dstenc < 8) {
4378       emit_d8(cbuf, 0xF);
4379     } else if (srcenc >= 8 && dstenc >= 8) {
4380       emit_d8(cbuf, 0x11);
4381     } else {
4382       emit_d8(cbuf, 0x10);
4383     }
4384 
4385     // subq rsp, #8
4386     emit_opcode(cbuf, Assembler::REX_W);
4387     emit_opcode(cbuf, 0x83);
4388     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4389     emit_d8(cbuf, 8);
4390 
4391     // movsd [rsp], $src
4392     emit_opcode(cbuf, 0xF2);
4393     if (srcenc >= 8) {
4394       emit_opcode(cbuf, Assembler::REX_R);
4395     }
4396     emit_opcode(cbuf, 0x0F);
4397     emit_opcode(cbuf, 0x11);
4398     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4399 
4400     // call d2i_fixup
4401     cbuf.set_inst_mark();
4402     emit_opcode(cbuf, 0xE8);
4403     emit_d32_reloc(cbuf,
4404                    (int)
4405                    (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4),
4406                    runtime_call_Relocation::spec(),
4407                    RELOC_DISP32);
4408 
4409     // popq $dst
4410     if (dstenc >= 8) {
4411       emit_opcode(cbuf, Assembler::REX_B);
4412     }
4413     emit_opcode(cbuf, 0x58 | (dstenc & 7));
4414 
4415     // done:
4416   %}
4417 
4418   enc_class d2l_fixup(rRegL dst, regD src)
4419   %{
4420     int dstenc = $dst$$reg;
4421     int srcenc = $src$$reg;
4422     address const_address = (address) StubRoutines::x86::double_sign_flip();
4423 
4424     // cmpq $dst, [0x8000000000000000]
4425     cbuf.set_inst_mark();
4426     emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4427     emit_opcode(cbuf, 0x39);
4428     // XXX reg_mem doesn't support RIP-relative addressing yet
4429     emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4430     emit_d32_reloc(cbuf, const_address);
4431 
4432 
4433     // jne,s done
4434     emit_opcode(cbuf, 0x75);
4435     if (srcenc < 8 && dstenc < 8) {
4436       emit_d8(cbuf, 0xF);
4437     } else if (srcenc >= 8 && dstenc >= 8) {
4438       emit_d8(cbuf, 0x11);
4439     } else {
4440       emit_d8(cbuf, 0x10);
4441     }
4442 
4443     // subq rsp, #8
4444     emit_opcode(cbuf, Assembler::REX_W);
4445     emit_opcode(cbuf, 0x83);
4446     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4447     emit_d8(cbuf, 8);
4448 
4449     // movsd [rsp], $src
4450     emit_opcode(cbuf, 0xF2);
4451     if (srcenc >= 8) {
4452       emit_opcode(cbuf, Assembler::REX_R);
4453     }
4454     emit_opcode(cbuf, 0x0F);
4455     emit_opcode(cbuf, 0x11);
4456     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4457 
4458     // call d2l_fixup
4459     cbuf.set_inst_mark();
4460     emit_opcode(cbuf, 0xE8);
4461     emit_d32_reloc(cbuf,
4462                    (int)
4463                    (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4),
4464                    runtime_call_Relocation::spec(),
4465                    RELOC_DISP32);
4466 
4467     // popq $dst
4468     if (dstenc >= 8) {
4469       emit_opcode(cbuf, Assembler::REX_B);
4470     }
4471     emit_opcode(cbuf, 0x58 | (dstenc & 7));
4472 
4473     // done:
4474   %}
4475 
4476   enc_class enc_membar_acquire
4477   %{
4478     // [jk] not needed currently, if you enable this and it really
4479     // emits code don't forget to the remove the "size(0)" line in
4480     // membar_acquire()
4481     // MacroAssembler masm(&cbuf);
4482     // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4483     //                                         Assembler::LoadLoad));
4484   %}
4485 
4486   enc_class enc_membar_release
4487   %{
4488     // [jk] not needed currently, if you enable this and it really
4489     // emits code don't forget to the remove the "size(0)" line in
4490     // membar_release()
4491     // MacroAssembler masm(&cbuf);
4492     // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
4493     //                                         Assembler::StoreStore));
4494   %}
4495 
4496   enc_class enc_membar_volatile
4497   %{
4498     MacroAssembler masm(&cbuf);
4499     masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
4500                                             Assembler::StoreStore));
4501   %}
4502 
4503   // Safepoint Poll.  This polls the safepoint page, and causes an
4504   // exception if it is not readable. Unfortunately, it kills
4505   // RFLAGS in the process.
4506   enc_class enc_safepoint_poll
4507   %{
4508     // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
4509     // XXX reg_mem doesn't support RIP-relative addressing yet
4510     cbuf.set_inst_mark();
4511     cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
4512     emit_opcode(cbuf, 0x85); // testl
4513     emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
4514     // cbuf.inst_mark() is beginning of instruction
4515     emit_d32_reloc(cbuf, os::get_polling_page());
4516 //                    relocInfo::poll_type,
4517   %}
4518 %}
4519 
4520 
4521 
4522 //----------FRAME--------------------------------------------------------------
4523 // Definition of frame structure and management information.
4524 //
4525 //  S T A C K   L A Y O U T    Allocators stack-slot number
4526 //                             |   (to get allocators register number
4527 //  G  Owned by    |        |  v    add OptoReg::stack0())
4528 //  r   CALLER     |        |
4529 //  o     |        +--------+      pad to even-align allocators stack-slot
4530 //  w     V        |  pad0  |        numbers; owned by CALLER
4531 //  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
4532 //  h     ^        |   in   |  5
4533 //        |        |  args  |  4   Holes in incoming args owned by SELF
4534 //  |     |        |        |  3
4535 //  |     |        +--------+
4536 //  V     |        | old out|      Empty on Intel, window on Sparc
4537 //        |    old |preserve|      Must be even aligned.
4538 //        |     SP-+--------+----> Matcher::_old_SP, even aligned
4539 //        |        |   in   |  3   area for Intel ret address
4540 //     Owned by    |preserve|      Empty on Sparc.
4541 //       SELF      +--------+
4542 //        |        |  pad2  |  2   pad to align old SP
4543 //        |        +--------+  1
4544 //        |        | locks  |  0
4545 //        |        +--------+----> OptoReg::stack0(), even aligned
4546 //        |        |  pad1  | 11   pad to align new SP
4547 //        |        +--------+
4548 //        |        |        | 10
4549 //        |        | spills |  9   spills
4550 //        V        |        |  8   (pad0 slot for callee)
4551 //      -----------+--------+----> Matcher::_out_arg_limit, unaligned
4552 //        ^        |  out   |  7
4553 //        |        |  args  |  6   Holes in outgoing args owned by CALLEE
4554 //     Owned by    +--------+
4555 //      CALLEE     | new out|  6   Empty on Intel, window on Sparc
4556 //        |    new |preserve|      Must be even-aligned.
4557 //        |     SP-+--------+----> Matcher::_new_SP, even aligned
4558 //        |        |        |
4559 //
4560 // Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
4561 //         known from SELF's arguments and the Java calling convention.
4562 //         Region 6-7 is determined per call site.
4563 // Note 2: If the calling convention leaves holes in the incoming argument
4564 //         area, those holes are owned by SELF.  Holes in the outgoing area
4565 //         are owned by the CALLEE.  Holes should not be nessecary in the
4566 //         incoming area, as the Java calling convention is completely under
4567 //         the control of the AD file.  Doubles can be sorted and packed to
4568 //         avoid holes.  Holes in the outgoing arguments may be nessecary for
4569 //         varargs C calling conventions.
4570 // Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
4571 //         even aligned with pad0 as needed.
4572 //         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
4573 //         region 6-11 is even aligned; it may be padded out more so that
4574 //         the region from SP to FP meets the minimum stack alignment.
4575 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
4576 //         alignment.  Region 11, pad1, may be dynamically extended so that
4577 //         SP meets the minimum alignment.
4578 
4579 frame
4580 %{
4581   // What direction does stack grow in (assumed to be same for C & Java)
4582   stack_direction(TOWARDS_LOW);
4583 
4584   // These three registers define part of the calling convention
4585   // between compiled code and the interpreter.
4586   inline_cache_reg(RAX);                // Inline Cache Register
4587   interpreter_method_oop_reg(RBX);      // Method Oop Register when
4588                                         // calling interpreter
4589 
4590   // Optional: name the operand used by cisc-spilling to access
4591   // [stack_pointer + offset]
4592   cisc_spilling_operand_name(indOffset32);
4593 
4594   // Number of stack slots consumed by locking an object
4595   sync_stack_slots(2);
4596 
4597   // Compiled code's Frame Pointer
4598   frame_pointer(RSP);
4599 
4600   // Interpreter stores its frame pointer in a register which is
4601   // stored to the stack by I2CAdaptors.
4602   // I2CAdaptors convert from interpreted java to compiled java.
4603   interpreter_frame_pointer(RBP);
4604 
4605   // Stack alignment requirement
4606   stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
4607 
4608   // Number of stack slots between incoming argument block and the start of
4609   // a new frame.  The PROLOG must add this many slots to the stack.  The
4610   // EPILOG must remove this many slots.  amd64 needs two slots for
4611   // return address.
4612   in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
4613 
4614   // Number of outgoing stack slots killed above the out_preserve_stack_slots
4615   // for calls to C.  Supports the var-args backing area for register parms.
4616   varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
4617 
4618   // The after-PROLOG location of the return address.  Location of
4619   // return address specifies a type (REG or STACK) and a number
4620   // representing the register number (i.e. - use a register name) or
4621   // stack slot.
4622   // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4623   // Otherwise, it is above the locks and verification slot and alignment word
4624   return_addr(STACK - 2 +
4625               round_to(2 + 2 * VerifyStackAtCalls +
4626                        Compile::current()->fixed_slots(),
4627                        WordsPerLong * 2));
4628 
4629   // Body of function which returns an integer array locating
4630   // arguments either in registers or in stack slots.  Passed an array
4631   // of ideal registers called "sig" and a "length" count.  Stack-slot
4632   // offsets are based on outgoing arguments, i.e. a CALLER setting up
4633   // arguments for a CALLEE.  Incoming stack arguments are
4634   // automatically biased by the preserve_stack_slots field above.
4635 
4636   calling_convention
4637   %{
4638     // No difference between ingoing/outgoing just pass false
4639     SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4640   %}
4641 
4642   c_calling_convention
4643   %{
4644     // This is obviously always outgoing
4645     (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4646   %}
4647 
4648   // Location of compiled Java return values.  Same as C for now.
4649   return_value
4650   %{
4651     assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4652            "only return normal values");
4653 
4654     static const int lo[Op_RegL + 1] = {
4655       0,
4656       0,
4657       RAX_num,  // Op_RegN
4658       RAX_num,  // Op_RegI
4659       RAX_num,  // Op_RegP
4660       XMM0_num, // Op_RegF
4661       XMM0_num, // Op_RegD
4662       RAX_num   // Op_RegL
4663     };
4664     static const int hi[Op_RegL + 1] = {
4665       0,
4666       0,
4667       OptoReg::Bad, // Op_RegN
4668       OptoReg::Bad, // Op_RegI
4669       RAX_H_num,    // Op_RegP
4670       OptoReg::Bad, // Op_RegF
4671       XMM0_H_num,   // Op_RegD
4672       RAX_H_num     // Op_RegL
4673     };
4674     assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4675     return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4676   %}
4677 %}
4678 
4679 //----------ATTRIBUTES---------------------------------------------------------
4680 //----------Operand Attributes-------------------------------------------------
4681 op_attrib op_cost(0);        // Required cost attribute
4682 
4683 //----------Instruction Attributes---------------------------------------------
4684 ins_attrib ins_cost(100);       // Required cost attribute
4685 ins_attrib ins_size(8);         // Required size attribute (in bits)
4686 ins_attrib ins_pc_relative(0);  // Required PC Relative flag
4687 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4688                                 // a non-matching short branch variant
4689                                 // of some long branch?
4690 ins_attrib ins_alignment(1);    // Required alignment attribute (must
4691                                 // be a power of 2) specifies the
4692                                 // alignment that some part of the
4693                                 // instruction (not necessarily the
4694                                 // start) requires.  If > 1, a
4695                                 // compute_padding() function must be
4696                                 // provided for the instruction
4697 
4698 //----------OPERANDS-----------------------------------------------------------
4699 // Operand definitions must precede instruction definitions for correct parsing
4700 // in the ADLC because operands constitute user defined types which are used in
4701 // instruction definitions.
4702 
4703 //----------Simple Operands----------------------------------------------------
4704 // Immediate Operands
4705 // Integer Immediate
4706 operand immI()
4707 %{
4708   match(ConI);
4709 
4710   op_cost(10);
4711   format %{ %}
4712   interface(CONST_INTER);
4713 %}
4714 
4715 // Constant for test vs zero
4716 operand immI0()
4717 %{
4718   predicate(n->get_int() == 0);
4719   match(ConI);
4720 
4721   op_cost(0);
4722   format %{ %}
4723   interface(CONST_INTER);
4724 %}
4725 
4726 // Constant for increment
4727 operand immI1()
4728 %{
4729   predicate(n->get_int() == 1);
4730   match(ConI);
4731 
4732   op_cost(0);
4733   format %{ %}
4734   interface(CONST_INTER);
4735 %}
4736 
4737 // Constant for decrement
4738 operand immI_M1()
4739 %{
4740   predicate(n->get_int() == -1);
4741   match(ConI);
4742 
4743   op_cost(0);
4744   format %{ %}
4745   interface(CONST_INTER);
4746 %}
4747 
4748 // Valid scale values for addressing modes
4749 operand immI2()
4750 %{
4751   predicate(0 <= n->get_int() && (n->get_int() <= 3));
4752   match(ConI);
4753 
4754   format %{ %}
4755   interface(CONST_INTER);
4756 %}
4757 
4758 operand immI8()
4759 %{
4760   predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4761   match(ConI);
4762 
4763   op_cost(5);
4764   format %{ %}
4765   interface(CONST_INTER);
4766 %}
4767 
4768 operand immI16()
4769 %{
4770   predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4771   match(ConI);
4772 
4773   op_cost(10);
4774   format %{ %}
4775   interface(CONST_INTER);
4776 %}
4777 
4778 // Constant for long shifts
4779 operand immI_32()
4780 %{
4781   predicate( n->get_int() == 32 );
4782   match(ConI);
4783 
4784   op_cost(0);
4785   format %{ %}
4786   interface(CONST_INTER);
4787 %}
4788 
4789 // Constant for long shifts
4790 operand immI_64()
4791 %{
4792   predicate( n->get_int() == 64 );
4793   match(ConI);
4794 
4795   op_cost(0);
4796   format %{ %}
4797   interface(CONST_INTER);
4798 %}
4799 
4800 // Pointer Immediate
4801 operand immP()
4802 %{
4803   match(ConP);
4804 
4805   op_cost(10);
4806   format %{ %}
4807   interface(CONST_INTER);
4808 %}
4809 
4810 // NULL Pointer Immediate
4811 operand immP0()
4812 %{
4813   predicate(n->get_ptr() == 0);
4814   match(ConP);
4815 
4816   op_cost(5);
4817   format %{ %}
4818   interface(CONST_INTER);
4819 %}
4820 
4821 // Pointer Immediate
4822 operand immN() %{
4823   match(ConN);
4824 
4825   op_cost(10);
4826   format %{ %}
4827   interface(CONST_INTER);
4828 %}
4829 
4830 // NULL Pointer Immediate
4831 operand immN0() %{
4832   predicate(n->get_narrowcon() == 0);
4833   match(ConN);
4834 
4835   op_cost(5);
4836   format %{ %}
4837   interface(CONST_INTER);
4838 %}
4839 
4840 operand immP31()
4841 %{
4842   predicate(!n->as_Type()->type()->isa_oopptr()
4843             && (n->get_ptr() >> 31) == 0);
4844   match(ConP);
4845 
4846   op_cost(5);
4847   format %{ %}
4848   interface(CONST_INTER);
4849 %}
4850 
4851 
4852 // Long Immediate
4853 operand immL()
4854 %{
4855   match(ConL);
4856 
4857   op_cost(20);
4858   format %{ %}
4859   interface(CONST_INTER);
4860 %}
4861 
4862 // Long Immediate 8-bit
4863 operand immL8()
4864 %{
4865   predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4866   match(ConL);
4867 
4868   op_cost(5);
4869   format %{ %}
4870   interface(CONST_INTER);
4871 %}
4872 
4873 // Long Immediate 32-bit unsigned
4874 operand immUL32()
4875 %{
4876   predicate(n->get_long() == (unsigned int) (n->get_long()));
4877   match(ConL);
4878 
4879   op_cost(10);
4880   format %{ %}
4881   interface(CONST_INTER);
4882 %}
4883 
4884 // Long Immediate 32-bit signed
4885 operand immL32()
4886 %{
4887   predicate(n->get_long() == (int) (n->get_long()));
4888   match(ConL);
4889 
4890   op_cost(15);
4891   format %{ %}
4892   interface(CONST_INTER);
4893 %}
4894 
4895 // Long Immediate zero
4896 operand immL0()
4897 %{
4898   predicate(n->get_long() == 0L);
4899   match(ConL);
4900 
4901   op_cost(10);
4902   format %{ %}
4903   interface(CONST_INTER);
4904 %}
4905 
4906 // Constant for increment
4907 operand immL1()
4908 %{
4909   predicate(n->get_long() == 1);
4910   match(ConL);
4911 
4912   format %{ %}
4913   interface(CONST_INTER);
4914 %}
4915 
4916 // Constant for decrement
4917 operand immL_M1()
4918 %{
4919   predicate(n->get_long() == -1);
4920   match(ConL);
4921 
4922   format %{ %}
4923   interface(CONST_INTER);
4924 %}
4925 
4926 // Long Immediate: the value 10
4927 operand immL10()
4928 %{
4929   predicate(n->get_long() == 10);
4930   match(ConL);
4931 
4932   format %{ %}
4933   interface(CONST_INTER);
4934 %}
4935 
4936 // Long immediate from 0 to 127.
4937 // Used for a shorter form of long mul by 10.
4938 operand immL_127()
4939 %{
4940   predicate(0 <= n->get_long() && n->get_long() < 0x80);
4941   match(ConL);
4942 
4943   op_cost(10);
4944   format %{ %}
4945   interface(CONST_INTER);
4946 %}
4947 
4948 // Long Immediate: low 32-bit mask
4949 operand immL_32bits()
4950 %{
4951   predicate(n->get_long() == 0xFFFFFFFFL);
4952   match(ConL);
4953   op_cost(20);
4954 
4955   format %{ %}
4956   interface(CONST_INTER);
4957 %}
4958 
4959 // Float Immediate zero
4960 operand immF0()
4961 %{
4962   predicate(jint_cast(n->getf()) == 0);
4963   match(ConF);
4964 
4965   op_cost(5);
4966   format %{ %}
4967   interface(CONST_INTER);
4968 %}
4969 
4970 // Float Immediate
4971 operand immF()
4972 %{
4973   match(ConF);
4974 
4975   op_cost(15);
4976   format %{ %}
4977   interface(CONST_INTER);
4978 %}
4979 
4980 // Double Immediate zero
4981 operand immD0()
4982 %{
4983   predicate(jlong_cast(n->getd()) == 0);
4984   match(ConD);
4985 
4986   op_cost(5);
4987   format %{ %}
4988   interface(CONST_INTER);
4989 %}
4990 
4991 // Double Immediate
4992 operand immD()
4993 %{
4994   match(ConD);
4995 
4996   op_cost(15);
4997   format %{ %}
4998   interface(CONST_INTER);
4999 %}
5000 
5001 // Immediates for special shifts (sign extend)
5002 
5003 // Constants for increment
5004 operand immI_16()
5005 %{
5006   predicate(n->get_int() == 16);
5007   match(ConI);
5008 
5009   format %{ %}
5010   interface(CONST_INTER);
5011 %}
5012 
5013 operand immI_24()
5014 %{
5015   predicate(n->get_int() == 24);
5016   match(ConI);
5017 
5018   format %{ %}
5019   interface(CONST_INTER);
5020 %}
5021 
5022 // Constant for byte-wide masking
5023 operand immI_255()
5024 %{
5025   predicate(n->get_int() == 255);
5026   match(ConI);
5027 
5028   format %{ %}
5029   interface(CONST_INTER);
5030 %}
5031 
5032 // Constant for short-wide masking
5033 operand immI_65535()
5034 %{
5035   predicate(n->get_int() == 65535);
5036   match(ConI);
5037 
5038   format %{ %}
5039   interface(CONST_INTER);
5040 %}
5041 
5042 // Constant for byte-wide masking
5043 operand immL_255()
5044 %{
5045   predicate(n->get_long() == 255);
5046   match(ConL);
5047 
5048   format %{ %}
5049   interface(CONST_INTER);
5050 %}
5051 
5052 // Constant for short-wide masking
5053 operand immL_65535()
5054 %{
5055   predicate(n->get_long() == 65535);
5056   match(ConL);
5057 
5058   format %{ %}
5059   interface(CONST_INTER);
5060 %}
5061 
5062 // Register Operands
5063 // Integer Register
5064 operand rRegI()
5065 %{
5066   constraint(ALLOC_IN_RC(int_reg));
5067   match(RegI);
5068 
5069   match(rax_RegI);
5070   match(rbx_RegI);
5071   match(rcx_RegI);
5072   match(rdx_RegI);
5073   match(rdi_RegI);
5074 
5075   format %{ %}
5076   interface(REG_INTER);
5077 %}
5078 
5079 // Special Registers
5080 operand rax_RegI()
5081 %{
5082   constraint(ALLOC_IN_RC(int_rax_reg));
5083   match(RegI);
5084   match(rRegI);
5085 
5086   format %{ "RAX" %}
5087   interface(REG_INTER);
5088 %}
5089 
5090 // Special Registers
5091 operand rbx_RegI()
5092 %{
5093   constraint(ALLOC_IN_RC(int_rbx_reg));
5094   match(RegI);
5095   match(rRegI);
5096 
5097   format %{ "RBX" %}
5098   interface(REG_INTER);
5099 %}
5100 
5101 operand rcx_RegI()
5102 %{
5103   constraint(ALLOC_IN_RC(int_rcx_reg));
5104   match(RegI);
5105   match(rRegI);
5106 
5107   format %{ "RCX" %}
5108   interface(REG_INTER);
5109 %}
5110 
5111 operand rdx_RegI()
5112 %{
5113   constraint(ALLOC_IN_RC(int_rdx_reg));
5114   match(RegI);
5115   match(rRegI);
5116 
5117   format %{ "RDX" %}
5118   interface(REG_INTER);
5119 %}
5120 
5121 operand rdi_RegI()
5122 %{
5123   constraint(ALLOC_IN_RC(int_rdi_reg));
5124   match(RegI);
5125   match(rRegI);
5126 
5127   format %{ "RDI" %}
5128   interface(REG_INTER);
5129 %}
5130 
5131 operand no_rcx_RegI()
5132 %{
5133   constraint(ALLOC_IN_RC(int_no_rcx_reg));
5134   match(RegI);
5135   match(rax_RegI);
5136   match(rbx_RegI);
5137   match(rdx_RegI);
5138   match(rdi_RegI);
5139 
5140   format %{ %}
5141   interface(REG_INTER);
5142 %}
5143 
5144 operand no_rax_rdx_RegI()
5145 %{
5146   constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
5147   match(RegI);
5148   match(rbx_RegI);
5149   match(rcx_RegI);
5150   match(rdi_RegI);
5151 
5152   format %{ %}
5153   interface(REG_INTER);
5154 %}
5155 
5156 // Pointer Register
5157 operand any_RegP()
5158 %{
5159   constraint(ALLOC_IN_RC(any_reg));
5160   match(RegP);
5161   match(rax_RegP);
5162   match(rbx_RegP);
5163   match(rdi_RegP);
5164   match(rsi_RegP);
5165   match(rbp_RegP);
5166   match(r15_RegP);
5167   match(rRegP);
5168 
5169   format %{ %}
5170   interface(REG_INTER);
5171 %}
5172 
5173 operand rRegP()
5174 %{
5175   constraint(ALLOC_IN_RC(ptr_reg));
5176   match(RegP);
5177   match(rax_RegP);
5178   match(rbx_RegP);
5179   match(rdi_RegP);
5180   match(rsi_RegP);
5181   match(rbp_RegP);
5182   match(r15_RegP);  // See Q&A below about r15_RegP.
5183 
5184   format %{ %}
5185   interface(REG_INTER);
5186 %}
5187 
5188 
5189 operand r12RegL() %{
5190   constraint(ALLOC_IN_RC(long_r12_reg));
5191   match(RegL);
5192 
5193   format %{ %}
5194   interface(REG_INTER);
5195 %}
5196 
5197 operand rRegN() %{
5198   constraint(ALLOC_IN_RC(int_reg));
5199   match(RegN);
5200 
5201   format %{ %}
5202   interface(REG_INTER);
5203 %}
5204 
5205 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
5206 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
5207 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
5208 // The output of an instruction is controlled by the allocator, which respects
5209 // register class masks, not match rules.  Unless an instruction mentions
5210 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
5211 // by the allocator as an input.
5212 
5213 operand no_rax_RegP()
5214 %{
5215   constraint(ALLOC_IN_RC(ptr_no_rax_reg));
5216   match(RegP);
5217   match(rbx_RegP);
5218   match(rsi_RegP);
5219   match(rdi_RegP);
5220 
5221   format %{ %}
5222   interface(REG_INTER);
5223 %}
5224 
5225 operand no_rbp_RegP()
5226 %{
5227   constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
5228   match(RegP);
5229   match(rbx_RegP);
5230   match(rsi_RegP);
5231   match(rdi_RegP);
5232 
5233   format %{ %}
5234   interface(REG_INTER);
5235 %}
5236 
5237 operand no_rax_rbx_RegP()
5238 %{
5239   constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
5240   match(RegP);
5241   match(rsi_RegP);
5242   match(rdi_RegP);
5243 
5244   format %{ %}
5245   interface(REG_INTER);
5246 %}
5247 
5248 // Special Registers
5249 // Return a pointer value
5250 operand rax_RegP()
5251 %{
5252   constraint(ALLOC_IN_RC(ptr_rax_reg));
5253   match(RegP);
5254   match(rRegP);
5255 
5256   format %{ %}
5257   interface(REG_INTER);
5258 %}
5259 
5260 // Special Registers
5261 // Return a compressed pointer value
5262 operand rax_RegN()
5263 %{
5264   constraint(ALLOC_IN_RC(int_rax_reg));
5265   match(RegN);
5266   match(rRegN);
5267 
5268   format %{ %}
5269   interface(REG_INTER);
5270 %}
5271 
5272 // Used in AtomicAdd
5273 operand rbx_RegP()
5274 %{
5275   constraint(ALLOC_IN_RC(ptr_rbx_reg));
5276   match(RegP);
5277   match(rRegP);
5278 
5279   format %{ %}
5280   interface(REG_INTER);
5281 %}
5282 
5283 operand rsi_RegP()
5284 %{
5285   constraint(ALLOC_IN_RC(ptr_rsi_reg));
5286   match(RegP);
5287   match(rRegP);
5288 
5289   format %{ %}
5290   interface(REG_INTER);
5291 %}
5292 
5293 // Used in rep stosq
5294 operand rdi_RegP()
5295 %{
5296   constraint(ALLOC_IN_RC(ptr_rdi_reg));
5297   match(RegP);
5298   match(rRegP);
5299 
5300   format %{ %}
5301   interface(REG_INTER);
5302 %}
5303 
5304 operand rbp_RegP()
5305 %{
5306   constraint(ALLOC_IN_RC(ptr_rbp_reg));
5307   match(RegP);
5308   match(rRegP);
5309 
5310   format %{ %}
5311   interface(REG_INTER);
5312 %}
5313 
5314 operand r15_RegP()
5315 %{
5316   constraint(ALLOC_IN_RC(ptr_r15_reg));
5317   match(RegP);
5318   match(rRegP);
5319 
5320   format %{ %}
5321   interface(REG_INTER);
5322 %}
5323 
5324 operand rRegL()
5325 %{
5326   constraint(ALLOC_IN_RC(long_reg));
5327   match(RegL);
5328   match(rax_RegL);
5329   match(rdx_RegL);
5330 
5331   format %{ %}
5332   interface(REG_INTER);
5333 %}
5334 
5335 // Special Registers
5336 operand no_rax_rdx_RegL()
5337 %{
5338   constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5339   match(RegL);
5340   match(rRegL);
5341 
5342   format %{ %}
5343   interface(REG_INTER);
5344 %}
5345 
5346 operand no_rax_RegL()
5347 %{
5348   constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
5349   match(RegL);
5350   match(rRegL);
5351   match(rdx_RegL);
5352 
5353   format %{ %}
5354   interface(REG_INTER);
5355 %}
5356 
5357 operand no_rcx_RegL()
5358 %{
5359   constraint(ALLOC_IN_RC(long_no_rcx_reg));
5360   match(RegL);
5361   match(rRegL);
5362 
5363   format %{ %}
5364   interface(REG_INTER);
5365 %}
5366 
5367 operand rax_RegL()
5368 %{
5369   constraint(ALLOC_IN_RC(long_rax_reg));
5370   match(RegL);
5371   match(rRegL);
5372 
5373   format %{ "RAX" %}
5374   interface(REG_INTER);
5375 %}
5376 
5377 operand rcx_RegL()
5378 %{
5379   constraint(ALLOC_IN_RC(long_rcx_reg));
5380   match(RegL);
5381   match(rRegL);
5382 
5383   format %{ %}
5384   interface(REG_INTER);
5385 %}
5386 
5387 operand rdx_RegL()
5388 %{
5389   constraint(ALLOC_IN_RC(long_rdx_reg));
5390   match(RegL);
5391   match(rRegL);
5392 
5393   format %{ %}
5394   interface(REG_INTER);
5395 %}
5396 
5397 // Flags register, used as output of compare instructions
5398 operand rFlagsReg()
5399 %{
5400   constraint(ALLOC_IN_RC(int_flags));
5401   match(RegFlags);
5402 
5403   format %{ "RFLAGS" %}
5404   interface(REG_INTER);
5405 %}
5406 
5407 // Flags register, used as output of FLOATING POINT compare instructions
5408 operand rFlagsRegU()
5409 %{
5410   constraint(ALLOC_IN_RC(int_flags));
5411   match(RegFlags);
5412 
5413   format %{ "RFLAGS_U" %}
5414   interface(REG_INTER);
5415 %}
5416 
5417 operand rFlagsRegUCF() %{
5418   constraint(ALLOC_IN_RC(int_flags));
5419   match(RegFlags);
5420   predicate(false);
5421 
5422   format %{ "RFLAGS_U_CF" %}
5423   interface(REG_INTER);
5424 %}
5425 
5426 // Float register operands
5427 operand regF()
5428 %{
5429   constraint(ALLOC_IN_RC(float_reg));
5430   match(RegF);
5431 
5432   format %{ %}
5433   interface(REG_INTER);
5434 %}
5435 
5436 // XMM Double register operands
5437 operand regD() %{
5438   constraint(ALLOC_IN_RC(double_reg));
5439   match(RegD);
5440   match(regD6);
5441   match(regD7);
5442   format %{ %}
5443   interface(REG_INTER);
5444 %}
5445 
5446 // XMM6 double register operands
5447 operand regD6(regD reg) %{
5448   constraint(ALLOC_IN_RC(double_reg6));
5449   match(reg);
5450   format %{ "XMM6" %}
5451   interface(REG_INTER);
5452 %}
5453 
5454 // XMM7 double register operands
5455 operand regD7(regD reg) %{
5456   constraint(ALLOC_IN_RC(double_reg7));
5457   match(reg);
5458 
5459   format %{ "XMM7" %}
5460   interface(REG_INTER);
5461 %}
5462 
5463 //----------Memory Operands----------------------------------------------------
5464 // Direct Memory Operand
5465 // operand direct(immP addr)
5466 // %{
5467 //   match(addr);
5468 
5469 //   format %{ "[$addr]" %}
5470 //   interface(MEMORY_INTER) %{
5471 //     base(0xFFFFFFFF);
5472 //     index(0x4);
5473 //     scale(0x0);
5474 //     disp($addr);
5475 //   %}
5476 // %}
5477 
5478 // Indirect Memory Operand
5479 operand indirect(any_RegP reg)
5480 %{
5481   constraint(ALLOC_IN_RC(ptr_reg));
5482   match(reg);
5483 
5484   format %{ "[$reg]" %}
5485   interface(MEMORY_INTER) %{
5486     base($reg);
5487     index(0x4);
5488     scale(0x0);
5489     disp(0x0);
5490   %}
5491 %}
5492 
5493 // Indirect Memory Plus Short Offset Operand
5494 operand indOffset8(any_RegP reg, immL8 off)
5495 %{
5496   constraint(ALLOC_IN_RC(ptr_reg));
5497   match(AddP reg off);
5498 
5499   format %{ "[$reg + $off (8-bit)]" %}
5500   interface(MEMORY_INTER) %{
5501     base($reg);
5502     index(0x4);
5503     scale(0x0);
5504     disp($off);
5505   %}
5506 %}
5507 
5508 // Indirect Memory Plus Long Offset Operand
5509 operand indOffset32(any_RegP reg, immL32 off)
5510 %{
5511   constraint(ALLOC_IN_RC(ptr_reg));
5512   match(AddP reg off);
5513 
5514   format %{ "[$reg + $off (32-bit)]" %}
5515   interface(MEMORY_INTER) %{
5516     base($reg);
5517     index(0x4);
5518     scale(0x0);
5519     disp($off);
5520   %}
5521 %}
5522 
5523 // Indirect Memory Plus Index Register Plus Offset Operand
5524 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
5525 %{
5526   constraint(ALLOC_IN_RC(ptr_reg));
5527   match(AddP (AddP reg lreg) off);
5528 
5529   op_cost(10);
5530   format %{"[$reg + $off + $lreg]" %}
5531   interface(MEMORY_INTER) %{
5532     base($reg);
5533     index($lreg);
5534     scale(0x0);
5535     disp($off);
5536   %}
5537 %}
5538 
5539 // Indirect Memory Plus Index Register Plus Offset Operand
5540 operand indIndex(any_RegP reg, rRegL lreg)
5541 %{
5542   constraint(ALLOC_IN_RC(ptr_reg));
5543   match(AddP reg lreg);
5544 
5545   op_cost(10);
5546   format %{"[$reg + $lreg]" %}
5547   interface(MEMORY_INTER) %{
5548     base($reg);
5549     index($lreg);
5550     scale(0x0);
5551     disp(0x0);
5552   %}
5553 %}
5554 
5555 // Indirect Memory Times Scale Plus Index Register
5556 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
5557 %{
5558   constraint(ALLOC_IN_RC(ptr_reg));
5559   match(AddP reg (LShiftL lreg scale));
5560 
5561   op_cost(10);
5562   format %{"[$reg + $lreg << $scale]" %}
5563   interface(MEMORY_INTER) %{
5564     base($reg);
5565     index($lreg);
5566     scale($scale);
5567     disp(0x0);
5568   %}
5569 %}
5570 
5571 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5572 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
5573 %{
5574   constraint(ALLOC_IN_RC(ptr_reg));
5575   match(AddP (AddP reg (LShiftL lreg scale)) off);
5576 
5577   op_cost(10);
5578   format %{"[$reg + $off + $lreg << $scale]" %}
5579   interface(MEMORY_INTER) %{
5580     base($reg);
5581     index($lreg);
5582     scale($scale);
5583     disp($off);
5584   %}
5585 %}
5586 
5587 // Indirect Narrow Oop Plus Offset Operand
5588 operand indNarrowOopOffset(rRegN src, immL32 off) %{
5589   constraint(ALLOC_IN_RC(ptr_reg));
5590   match(AddP (DecodeN src) off);
5591 
5592   op_cost(10);
5593   format %{"[R12 + $src << 3 + $off] (compressed oop addressing)" %}
5594   interface(MEMORY_INTER) %{
5595     base(0xc); // R12
5596     index($src);
5597     scale(0x3);
5598     disp($off);
5599   %}
5600 %}
5601 
5602 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5603 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
5604 %{
5605   constraint(ALLOC_IN_RC(ptr_reg));
5606   predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5607   match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
5608 
5609   op_cost(10);
5610   format %{"[$reg + $off + $idx << $scale]" %}
5611   interface(MEMORY_INTER) %{
5612     base($reg);
5613     index($idx);
5614     scale($scale);
5615     disp($off);
5616   %}
5617 %}
5618 
5619 //----------Special Memory Operands--------------------------------------------
5620 // Stack Slot Operand - This operand is used for loading and storing temporary
5621 //                      values on the stack where a match requires a value to
5622 //                      flow through memory.
5623 operand stackSlotP(sRegP reg)
5624 %{
5625   constraint(ALLOC_IN_RC(stack_slots));
5626   // No match rule because this operand is only generated in matching
5627 
5628   format %{ "[$reg]" %}
5629   interface(MEMORY_INTER) %{
5630     base(0x4);   // RSP
5631     index(0x4);  // No Index
5632     scale(0x0);  // No Scale
5633     disp($reg);  // Stack Offset
5634   %}
5635 %}
5636 
5637 operand stackSlotI(sRegI reg)
5638 %{
5639   constraint(ALLOC_IN_RC(stack_slots));
5640   // No match rule because this operand is only generated in matching
5641 
5642   format %{ "[$reg]" %}
5643   interface(MEMORY_INTER) %{
5644     base(0x4);   // RSP
5645     index(0x4);  // No Index
5646     scale(0x0);  // No Scale
5647     disp($reg);  // Stack Offset
5648   %}
5649 %}
5650 
5651 operand stackSlotF(sRegF reg)
5652 %{
5653   constraint(ALLOC_IN_RC(stack_slots));
5654   // No match rule because this operand is only generated in matching
5655 
5656   format %{ "[$reg]" %}
5657   interface(MEMORY_INTER) %{
5658     base(0x4);   // RSP
5659     index(0x4);  // No Index
5660     scale(0x0);  // No Scale
5661     disp($reg);  // Stack Offset
5662   %}
5663 %}
5664 
5665 operand stackSlotD(sRegD reg)
5666 %{
5667   constraint(ALLOC_IN_RC(stack_slots));
5668   // No match rule because this operand is only generated in matching
5669 
5670   format %{ "[$reg]" %}
5671   interface(MEMORY_INTER) %{
5672     base(0x4);   // RSP
5673     index(0x4);  // No Index
5674     scale(0x0);  // No Scale
5675     disp($reg);  // Stack Offset
5676   %}
5677 %}
5678 operand stackSlotL(sRegL reg)
5679 %{
5680   constraint(ALLOC_IN_RC(stack_slots));
5681   // No match rule because this operand is only generated in matching
5682 
5683   format %{ "[$reg]" %}
5684   interface(MEMORY_INTER) %{
5685     base(0x4);   // RSP
5686     index(0x4);  // No Index
5687     scale(0x0);  // No Scale
5688     disp($reg);  // Stack Offset
5689   %}
5690 %}
5691 
5692 //----------Conditional Branch Operands----------------------------------------
5693 // Comparison Op  - This is the operation of the comparison, and is limited to
5694 //                  the following set of codes:
5695 //                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5696 //
5697 // Other attributes of the comparison, such as unsignedness, are specified
5698 // by the comparison instruction that sets a condition code flags register.
5699 // That result is represented by a flags operand whose subtype is appropriate
5700 // to the unsignedness (etc.) of the comparison.
5701 //
5702 // Later, the instruction which matches both the Comparison Op (a Bool) and
5703 // the flags (produced by the Cmp) specifies the coding of the comparison op
5704 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5705 
5706 // Comparision Code
5707 operand cmpOp()
5708 %{
5709   match(Bool);
5710 
5711   format %{ "" %}
5712   interface(COND_INTER) %{
5713     equal(0x4, "e");
5714     not_equal(0x5, "ne");
5715     less(0xC, "l");
5716     greater_equal(0xD, "ge");
5717     less_equal(0xE, "le");
5718     greater(0xF, "g");
5719   %}
5720 %}
5721 
5722 // Comparison Code, unsigned compare.  Used by FP also, with
5723 // C2 (unordered) turned into GT or LT already.  The other bits
5724 // C0 and C3 are turned into Carry & Zero flags.
5725 operand cmpOpU()
5726 %{
5727   match(Bool);
5728 
5729   format %{ "" %}
5730   interface(COND_INTER) %{
5731     equal(0x4, "e");
5732     not_equal(0x5, "ne");
5733     less(0x2, "b");
5734     greater_equal(0x3, "nb");
5735     less_equal(0x6, "be");
5736     greater(0x7, "nbe");
5737   %}
5738 %}
5739 
5740 
5741 // Floating comparisons that don't require any fixup for the unordered case
5742 operand cmpOpUCF() %{
5743   match(Bool);
5744   predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5745             n->as_Bool()->_test._test == BoolTest::ge ||
5746             n->as_Bool()->_test._test == BoolTest::le ||
5747             n->as_Bool()->_test._test == BoolTest::gt);
5748   format %{ "" %}
5749   interface(COND_INTER) %{
5750     equal(0x4, "e");
5751     not_equal(0x5, "ne");
5752     less(0x2, "b");
5753     greater_equal(0x3, "nb");
5754     less_equal(0x6, "be");
5755     greater(0x7, "nbe");
5756   %}
5757 %}
5758 
5759 
5760 // Floating comparisons that can be fixed up with extra conditional jumps
5761 operand cmpOpUCF2() %{
5762   match(Bool);
5763   predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5764             n->as_Bool()->_test._test == BoolTest::eq);
5765   format %{ "" %}
5766   interface(COND_INTER) %{
5767     equal(0x4, "e");
5768     not_equal(0x5, "ne");
5769     less(0x2, "b");
5770     greater_equal(0x3, "nb");
5771     less_equal(0x6, "be");
5772     greater(0x7, "nbe");
5773   %}
5774 %}
5775 
5776 
5777 //----------OPERAND CLASSES----------------------------------------------------
5778 // Operand Classes are groups of operands that are used as to simplify
5779 // instruction definitions by not requiring the AD writer to specify seperate
5780 // instructions for every form of operand when the instruction accepts
5781 // multiple operand types with the same basic encoding and format.  The classic
5782 // case of this is memory operands.
5783 
5784 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5785                indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5786                indNarrowOopOffset);
5787 
5788 //----------PIPELINE-----------------------------------------------------------
5789 // Rules which define the behavior of the target architectures pipeline.
5790 pipeline %{
5791 
5792 //----------ATTRIBUTES---------------------------------------------------------
5793 attributes %{
5794   variable_size_instructions;        // Fixed size instructions
5795   max_instructions_per_bundle = 3;   // Up to 3 instructions per bundle
5796   instruction_unit_size = 1;         // An instruction is 1 bytes long
5797   instruction_fetch_unit_size = 16;  // The processor fetches one line
5798   instruction_fetch_units = 1;       // of 16 bytes
5799 
5800   // List of nop instructions
5801   nops( MachNop );
5802 %}
5803 
5804 //----------RESOURCES----------------------------------------------------------
5805 // Resources are the functional units available to the machine
5806 
5807 // Generic P2/P3 pipeline
5808 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5809 // 3 instructions decoded per cycle.
5810 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5811 // 3 ALU op, only ALU0 handles mul instructions.
5812 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5813            MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5814            BR, FPU,
5815            ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5816 
5817 //----------PIPELINE DESCRIPTION-----------------------------------------------
5818 // Pipeline Description specifies the stages in the machine's pipeline
5819 
5820 // Generic P2/P3 pipeline
5821 pipe_desc(S0, S1, S2, S3, S4, S5);
5822 
5823 //----------PIPELINE CLASSES---------------------------------------------------
5824 // Pipeline Classes describe the stages in which input and output are
5825 // referenced by the hardware pipeline.
5826 
5827 // Naming convention: ialu or fpu
5828 // Then: _reg
5829 // Then: _reg if there is a 2nd register
5830 // Then: _long if it's a pair of instructions implementing a long
5831 // Then: _fat if it requires the big decoder
5832 //   Or: _mem if it requires the big decoder and a memory unit.
5833 
5834 // Integer ALU reg operation
5835 pipe_class ialu_reg(rRegI dst)
5836 %{
5837     single_instruction;
5838     dst    : S4(write);
5839     dst    : S3(read);
5840     DECODE : S0;        // any decoder
5841     ALU    : S3;        // any alu
5842 %}
5843 
5844 // Long ALU reg operation
5845 pipe_class ialu_reg_long(rRegL dst)
5846 %{
5847     instruction_count(2);
5848     dst    : S4(write);
5849     dst    : S3(read);
5850     DECODE : S0(2);     // any 2 decoders
5851     ALU    : S3(2);     // both alus
5852 %}
5853 
5854 // Integer ALU reg operation using big decoder
5855 pipe_class ialu_reg_fat(rRegI dst)
5856 %{
5857     single_instruction;
5858     dst    : S4(write);
5859     dst    : S3(read);
5860     D0     : S0;        // big decoder only
5861     ALU    : S3;        // any alu
5862 %}
5863 
5864 // Long ALU reg operation using big decoder
5865 pipe_class ialu_reg_long_fat(rRegL dst)
5866 %{
5867     instruction_count(2);
5868     dst    : S4(write);
5869     dst    : S3(read);
5870     D0     : S0(2);     // big decoder only; twice
5871     ALU    : S3(2);     // any 2 alus
5872 %}
5873 
5874 // Integer ALU reg-reg operation
5875 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5876 %{
5877     single_instruction;
5878     dst    : S4(write);
5879     src    : S3(read);
5880     DECODE : S0;        // any decoder
5881     ALU    : S3;        // any alu
5882 %}
5883 
5884 // Long ALU reg-reg operation
5885 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5886 %{
5887     instruction_count(2);
5888     dst    : S4(write);
5889     src    : S3(read);
5890     DECODE : S0(2);     // any 2 decoders
5891     ALU    : S3(2);     // both alus
5892 %}
5893 
5894 // Integer ALU reg-reg operation
5895 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5896 %{
5897     single_instruction;
5898     dst    : S4(write);
5899     src    : S3(read);
5900     D0     : S0;        // big decoder only
5901     ALU    : S3;        // any alu
5902 %}
5903 
5904 // Long ALU reg-reg operation
5905 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5906 %{
5907     instruction_count(2);
5908     dst    : S4(write);
5909     src    : S3(read);
5910     D0     : S0(2);     // big decoder only; twice
5911     ALU    : S3(2);     // both alus
5912 %}
5913 
5914 // Integer ALU reg-mem operation
5915 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5916 %{
5917     single_instruction;
5918     dst    : S5(write);
5919     mem    : S3(read);
5920     D0     : S0;        // big decoder only
5921     ALU    : S4;        // any alu
5922     MEM    : S3;        // any mem
5923 %}
5924 
5925 // Integer mem operation (prefetch)
5926 pipe_class ialu_mem(memory mem)
5927 %{
5928     single_instruction;
5929     mem    : S3(read);
5930     D0     : S0;        // big decoder only
5931     MEM    : S3;        // any mem
5932 %}
5933 
5934 // Integer Store to Memory
5935 pipe_class ialu_mem_reg(memory mem, rRegI src)
5936 %{
5937     single_instruction;
5938     mem    : S3(read);
5939     src    : S5(read);
5940     D0     : S0;        // big decoder only
5941     ALU    : S4;        // any alu
5942     MEM    : S3;
5943 %}
5944 
5945 // // Long Store to Memory
5946 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5947 // %{
5948 //     instruction_count(2);
5949 //     mem    : S3(read);
5950 //     src    : S5(read);
5951 //     D0     : S0(2);          // big decoder only; twice
5952 //     ALU    : S4(2);     // any 2 alus
5953 //     MEM    : S3(2);  // Both mems
5954 // %}
5955 
5956 // Integer Store to Memory
5957 pipe_class ialu_mem_imm(memory mem)
5958 %{
5959     single_instruction;
5960     mem    : S3(read);
5961     D0     : S0;        // big decoder only
5962     ALU    : S4;        // any alu
5963     MEM    : S3;
5964 %}
5965 
5966 // Integer ALU0 reg-reg operation
5967 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5968 %{
5969     single_instruction;
5970     dst    : S4(write);
5971     src    : S3(read);
5972     D0     : S0;        // Big decoder only
5973     ALU0   : S3;        // only alu0
5974 %}
5975 
5976 // Integer ALU0 reg-mem operation
5977 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5978 %{
5979     single_instruction;
5980     dst    : S5(write);
5981     mem    : S3(read);
5982     D0     : S0;        // big decoder only
5983     ALU0   : S4;        // ALU0 only
5984     MEM    : S3;        // any mem
5985 %}
5986 
5987 // Integer ALU reg-reg operation
5988 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5989 %{
5990     single_instruction;
5991     cr     : S4(write);
5992     src1   : S3(read);
5993     src2   : S3(read);
5994     DECODE : S0;        // any decoder
5995     ALU    : S3;        // any alu
5996 %}
5997 
5998 // Integer ALU reg-imm operation
5999 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
6000 %{
6001     single_instruction;
6002     cr     : S4(write);
6003     src1   : S3(read);
6004     DECODE : S0;        // any decoder
6005     ALU    : S3;        // any alu
6006 %}
6007 
6008 // Integer ALU reg-mem operation
6009 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
6010 %{
6011     single_instruction;
6012     cr     : S4(write);
6013     src1   : S3(read);
6014     src2   : S3(read);
6015     D0     : S0;        // big decoder only
6016     ALU    : S4;        // any alu
6017     MEM    : S3;
6018 %}
6019 
6020 // Conditional move reg-reg
6021 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
6022 %{
6023     instruction_count(4);
6024     y      : S4(read);
6025     q      : S3(read);
6026     p      : S3(read);
6027     DECODE : S0(4);     // any decoder
6028 %}
6029 
6030 // Conditional move reg-reg
6031 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
6032 %{
6033     single_instruction;
6034     dst    : S4(write);
6035     src    : S3(read);
6036     cr     : S3(read);
6037     DECODE : S0;        // any decoder
6038 %}
6039 
6040 // Conditional move reg-mem
6041 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
6042 %{
6043     single_instruction;
6044     dst    : S4(write);
6045     src    : S3(read);
6046     cr     : S3(read);
6047     DECODE : S0;        // any decoder
6048     MEM    : S3;
6049 %}
6050 
6051 // Conditional move reg-reg long
6052 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
6053 %{
6054     single_instruction;
6055     dst    : S4(write);
6056     src    : S3(read);
6057     cr     : S3(read);
6058     DECODE : S0(2);     // any 2 decoders
6059 %}
6060 
6061 // XXX
6062 // // Conditional move double reg-reg
6063 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
6064 // %{
6065 //     single_instruction;
6066 //     dst    : S4(write);
6067 //     src    : S3(read);
6068 //     cr     : S3(read);
6069 //     DECODE : S0;     // any decoder
6070 // %}
6071 
6072 // Float reg-reg operation
6073 pipe_class fpu_reg(regD dst)
6074 %{
6075     instruction_count(2);
6076     dst    : S3(read);
6077     DECODE : S0(2);     // any 2 decoders
6078     FPU    : S3;
6079 %}
6080 
6081 // Float reg-reg operation
6082 pipe_class fpu_reg_reg(regD dst, regD src)
6083 %{
6084     instruction_count(2);
6085     dst    : S4(write);
6086     src    : S3(read);
6087     DECODE : S0(2);     // any 2 decoders
6088     FPU    : S3;
6089 %}
6090 
6091 // Float reg-reg operation
6092 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
6093 %{
6094     instruction_count(3);
6095     dst    : S4(write);
6096     src1   : S3(read);
6097     src2   : S3(read);
6098     DECODE : S0(3);     // any 3 decoders
6099     FPU    : S3(2);
6100 %}
6101 
6102 // Float reg-reg operation
6103 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
6104 %{
6105     instruction_count(4);
6106     dst    : S4(write);
6107     src1   : S3(read);
6108     src2   : S3(read);
6109     src3   : S3(read);
6110     DECODE : S0(4);     // any 3 decoders
6111     FPU    : S3(2);
6112 %}
6113 
6114 // Float reg-reg operation
6115 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
6116 %{
6117     instruction_count(4);
6118     dst    : S4(write);
6119     src1   : S3(read);
6120     src2   : S3(read);
6121     src3   : S3(read);
6122     DECODE : S1(3);     // any 3 decoders
6123     D0     : S0;        // Big decoder only
6124     FPU    : S3(2);
6125     MEM    : S3;
6126 %}
6127 
6128 // Float reg-mem operation
6129 pipe_class fpu_reg_mem(regD dst, memory mem)
6130 %{
6131     instruction_count(2);
6132     dst    : S5(write);
6133     mem    : S3(read);
6134     D0     : S0;        // big decoder only
6135     DECODE : S1;        // any decoder for FPU POP
6136     FPU    : S4;
6137     MEM    : S3;        // any mem
6138 %}
6139 
6140 // Float reg-mem operation
6141 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
6142 %{
6143     instruction_count(3);
6144     dst    : S5(write);
6145     src1   : S3(read);
6146     mem    : S3(read);
6147     D0     : S0;        // big decoder only
6148     DECODE : S1(2);     // any decoder for FPU POP
6149     FPU    : S4;
6150     MEM    : S3;        // any mem
6151 %}
6152 
6153 // Float mem-reg operation
6154 pipe_class fpu_mem_reg(memory mem, regD src)
6155 %{
6156     instruction_count(2);
6157     src    : S5(read);
6158     mem    : S3(read);
6159     DECODE : S0;        // any decoder for FPU PUSH
6160     D0     : S1;        // big decoder only
6161     FPU    : S4;
6162     MEM    : S3;        // any mem
6163 %}
6164 
6165 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
6166 %{
6167     instruction_count(3);
6168     src1   : S3(read);
6169     src2   : S3(read);
6170     mem    : S3(read);
6171     DECODE : S0(2);     // any decoder for FPU PUSH
6172     D0     : S1;        // big decoder only
6173     FPU    : S4;
6174     MEM    : S3;        // any mem
6175 %}
6176 
6177 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
6178 %{
6179     instruction_count(3);
6180     src1   : S3(read);
6181     src2   : S3(read);
6182     mem    : S4(read);
6183     DECODE : S0;        // any decoder for FPU PUSH
6184     D0     : S0(2);     // big decoder only
6185     FPU    : S4;
6186     MEM    : S3(2);     // any mem
6187 %}
6188 
6189 pipe_class fpu_mem_mem(memory dst, memory src1)
6190 %{
6191     instruction_count(2);
6192     src1   : S3(read);
6193     dst    : S4(read);
6194     D0     : S0(2);     // big decoder only
6195     MEM    : S3(2);     // any mem
6196 %}
6197 
6198 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
6199 %{
6200     instruction_count(3);
6201     src1   : S3(read);
6202     src2   : S3(read);
6203     dst    : S4(read);
6204     D0     : S0(3);     // big decoder only
6205     FPU    : S4;
6206     MEM    : S3(3);     // any mem
6207 %}
6208 
6209 pipe_class fpu_mem_reg_con(memory mem, regD src1)
6210 %{
6211     instruction_count(3);
6212     src1   : S4(read);
6213     mem    : S4(read);
6214     DECODE : S0;        // any decoder for FPU PUSH
6215     D0     : S0(2);     // big decoder only
6216     FPU    : S4;
6217     MEM    : S3(2);     // any mem
6218 %}
6219 
6220 // Float load constant
6221 pipe_class fpu_reg_con(regD dst)
6222 %{
6223     instruction_count(2);
6224     dst    : S5(write);
6225     D0     : S0;        // big decoder only for the load
6226     DECODE : S1;        // any decoder for FPU POP
6227     FPU    : S4;
6228     MEM    : S3;        // any mem
6229 %}
6230 
6231 // Float load constant
6232 pipe_class fpu_reg_reg_con(regD dst, regD src)
6233 %{
6234     instruction_count(3);
6235     dst    : S5(write);
6236     src    : S3(read);
6237     D0     : S0;        // big decoder only for the load
6238     DECODE : S1(2);     // any decoder for FPU POP
6239     FPU    : S4;
6240     MEM    : S3;        // any mem
6241 %}
6242 
6243 // UnConditional branch
6244 pipe_class pipe_jmp(label labl)
6245 %{
6246     single_instruction;
6247     BR   : S3;
6248 %}
6249 
6250 // Conditional branch
6251 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
6252 %{
6253     single_instruction;
6254     cr    : S1(read);
6255     BR    : S3;
6256 %}
6257 
6258 // Allocation idiom
6259 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
6260 %{
6261     instruction_count(1); force_serialization;
6262     fixed_latency(6);
6263     heap_ptr : S3(read);
6264     DECODE   : S0(3);
6265     D0       : S2;
6266     MEM      : S3;
6267     ALU      : S3(2);
6268     dst      : S5(write);
6269     BR       : S5;
6270 %}
6271 
6272 // Generic big/slow expanded idiom
6273 pipe_class pipe_slow()
6274 %{
6275     instruction_count(10); multiple_bundles; force_serialization;
6276     fixed_latency(100);
6277     D0  : S0(2);
6278     MEM : S3(2);
6279 %}
6280 
6281 // The real do-nothing guy
6282 pipe_class empty()
6283 %{
6284     instruction_count(0);
6285 %}
6286 
6287 // Define the class for the Nop node
6288 define
6289 %{
6290    MachNop = empty;
6291 %}
6292 
6293 %}
6294 
6295 //----------INSTRUCTIONS-------------------------------------------------------
6296 //
6297 // match      -- States which machine-independent subtree may be replaced
6298 //               by this instruction.
6299 // ins_cost   -- The estimated cost of this instruction is used by instruction
6300 //               selection to identify a minimum cost tree of machine
6301 //               instructions that matches a tree of machine-independent
6302 //               instructions.
6303 // format     -- A string providing the disassembly for this instruction.
6304 //               The value of an instruction's operand may be inserted
6305 //               by referring to it with a '$' prefix.
6306 // opcode     -- Three instruction opcodes may be provided.  These are referred
6307 //               to within an encode class as $primary, $secondary, and $tertiary
6308 //               rrspectively.  The primary opcode is commonly used to
6309 //               indicate the type of machine instruction, while secondary
6310 //               and tertiary are often used for prefix options or addressing
6311 //               modes.
6312 // ins_encode -- A list of encode classes with parameters. The encode class
6313 //               name must have been defined in an 'enc_class' specification
6314 //               in the encode section of the architecture description.
6315 
6316 
6317 //----------Load/Store/Move Instructions---------------------------------------
6318 //----------Load Instructions--------------------------------------------------
6319 
6320 // Load Byte (8 bit signed)
6321 instruct loadB(rRegI dst, memory mem)
6322 %{
6323   match(Set dst (LoadB mem));
6324 
6325   ins_cost(125);
6326   format %{ "movsbl  $dst, $mem\t# byte" %}
6327   opcode(0x0F, 0xBE);
6328   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6329   ins_pipe(ialu_reg_mem);
6330 %}
6331 
6332 // Load Byte (8 bit signed) into long
6333 // instruct loadB2L(rRegL dst, memory mem)
6334 // %{
6335 //   match(Set dst (ConvI2L (LoadB mem)));
6336 
6337 //   ins_cost(125);
6338 //   format %{ "movsbq  $dst, $mem\t# byte -> long" %}
6339 //   opcode(0x0F, 0xBE);
6340 //   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6341 //   ins_pipe(ialu_reg_mem);
6342 // %}
6343 
6344 // Load Byte (8 bit UNsigned)
6345 instruct loadUB(rRegI dst, memory mem, immI_255 bytemask)
6346 %{
6347   match(Set dst (AndI (LoadB mem) bytemask));
6348 
6349   ins_cost(125);
6350   format %{ "movzbl  $dst, $mem\t# ubyte" %}
6351   opcode(0x0F, 0xB6);
6352   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6353   ins_pipe(ialu_reg_mem);
6354 %}
6355 
6356 // Load Byte (8 bit UNsigned) into long
6357 // instruct loadUB2L(rRegL dst, memory mem, immI_255 bytemask)
6358 // %{
6359 //   match(Set dst (ConvI2L (AndI (LoadB mem) bytemask)));
6360 
6361 //   ins_cost(125);
6362 //   format %{ "movzbl  $dst, $mem\t# ubyte -> long" %}
6363 //   opcode(0x0F, 0xB6);
6364 //   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6365 //   ins_pipe(ialu_reg_mem);
6366 // %}
6367 
6368 // Load Short (16 bit signed)
6369 instruct loadS(rRegI dst, memory mem)
6370 %{
6371   match(Set dst (LoadS mem));
6372 
6373   ins_cost(125); // XXX
6374   format %{ "movswl $dst, $mem\t# short" %}
6375   opcode(0x0F, 0xBF);
6376   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6377   ins_pipe(ialu_reg_mem);
6378 %}
6379 
6380 // Load Short (16 bit signed) into long
6381 // instruct loadS2L(rRegL dst, memory mem)
6382 // %{
6383 //   match(Set dst (ConvI2L (LoadS mem)));
6384 
6385 //   ins_cost(125); // XXX
6386 //   format %{ "movswq $dst, $mem\t# short -> long" %}
6387 //   opcode(0x0F, 0xBF);
6388 //   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6389 //   ins_pipe(ialu_reg_mem);
6390 // %}
6391 
6392 // Load Unsigned Short/Char (16 bit UNsigned)
6393 instruct loadUS(rRegI dst, memory mem)
6394 %{
6395   match(Set dst (LoadUS mem));
6396 
6397   ins_cost(125);
6398   format %{ "movzwl  $dst, $mem\t# ushort/char" %}
6399   opcode(0x0F, 0xB7);
6400   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6401   ins_pipe(ialu_reg_mem);
6402 %}
6403 
6404 // Load Unsigned Short/Char (16 bit UNsigned) into long
6405 // instruct loadUS2L(rRegL dst, memory mem)
6406 // %{
6407 //   match(Set dst (ConvI2L (LoadUS mem)));
6408 
6409 //   ins_cost(125);
6410 //   format %{ "movzwl  $dst, $mem\t# ushort/char -> long" %}
6411 //   opcode(0x0F, 0xB7);
6412 //   ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem));
6413 //   ins_pipe(ialu_reg_mem);
6414 // %}
6415 
6416 // Load Integer
6417 instruct loadI(rRegI dst, memory mem)
6418 %{
6419   match(Set dst (LoadI mem));
6420 
6421   ins_cost(125); // XXX
6422   format %{ "movl    $dst, $mem\t# int" %}
6423   opcode(0x8B);
6424   ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6425   ins_pipe(ialu_reg_mem);
6426 %}
6427 
6428 // Load Long
6429 instruct loadL(rRegL dst, memory mem)
6430 %{
6431   match(Set dst (LoadL mem));
6432 
6433   ins_cost(125); // XXX
6434   format %{ "movq    $dst, $mem\t# long" %}
6435   opcode(0x8B);
6436   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6437   ins_pipe(ialu_reg_mem); // XXX
6438 %}
6439 
6440 // Load Range
6441 instruct loadRange(rRegI dst, memory mem)
6442 %{
6443   match(Set dst (LoadRange mem));
6444 
6445   ins_cost(125); // XXX
6446   format %{ "movl    $dst, $mem\t# range" %}
6447   opcode(0x8B);
6448   ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6449   ins_pipe(ialu_reg_mem);
6450 %}
6451 
6452 // Load Pointer
6453 instruct loadP(rRegP dst, memory mem)
6454 %{
6455   match(Set dst (LoadP mem));
6456 
6457   ins_cost(125); // XXX
6458   format %{ "movq    $dst, $mem\t# ptr" %}
6459   opcode(0x8B);
6460   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6461   ins_pipe(ialu_reg_mem); // XXX
6462 %}
6463 
6464 // Load Compressed Pointer
6465 instruct loadN(rRegN dst, memory mem)
6466 %{
6467    match(Set dst (LoadN mem));
6468 
6469    ins_cost(125); // XXX
6470    format %{ "movl    $dst, $mem\t# compressed ptr" %}
6471    ins_encode %{
6472      Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6473      Register dst = as_Register($dst$$reg);
6474      __ movl(dst, addr);
6475    %}
6476    ins_pipe(ialu_reg_mem); // XXX
6477 %}
6478 
6479 
6480 // Load Klass Pointer
6481 instruct loadKlass(rRegP dst, memory mem)
6482 %{
6483   match(Set dst (LoadKlass mem));
6484 
6485   ins_cost(125); // XXX
6486   format %{ "movq    $dst, $mem\t# class" %}
6487   opcode(0x8B);
6488   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6489   ins_pipe(ialu_reg_mem); // XXX
6490 %}
6491 
6492 // Load narrow Klass Pointer
6493 instruct loadNKlass(rRegN dst, memory mem)
6494 %{
6495   match(Set dst (LoadNKlass mem));
6496 
6497   ins_cost(125); // XXX
6498   format %{ "movl    $dst, $mem\t# compressed klass ptr" %}
6499   ins_encode %{
6500     Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
6501     Register dst = as_Register($dst$$reg);
6502     __ movl(dst, addr);
6503   %}
6504   ins_pipe(ialu_reg_mem); // XXX
6505 %}
6506 
6507 // Load Float
6508 instruct loadF(regF dst, memory mem)
6509 %{
6510   match(Set dst (LoadF mem));
6511 
6512   ins_cost(145); // XXX
6513   format %{ "movss   $dst, $mem\t# float" %}
6514   opcode(0xF3, 0x0F, 0x10);
6515   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6516   ins_pipe(pipe_slow); // XXX
6517 %}
6518 
6519 // Load Double
6520 instruct loadD_partial(regD dst, memory mem)
6521 %{
6522   predicate(!UseXmmLoadAndClearUpper);
6523   match(Set dst (LoadD mem));
6524 
6525   ins_cost(145); // XXX
6526   format %{ "movlpd  $dst, $mem\t# double" %}
6527   opcode(0x66, 0x0F, 0x12);
6528   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6529   ins_pipe(pipe_slow); // XXX
6530 %}
6531 
6532 instruct loadD(regD dst, memory mem)
6533 %{
6534   predicate(UseXmmLoadAndClearUpper);
6535   match(Set dst (LoadD mem));
6536 
6537   ins_cost(145); // XXX
6538   format %{ "movsd   $dst, $mem\t# double" %}
6539   opcode(0xF2, 0x0F, 0x10);
6540   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6541   ins_pipe(pipe_slow); // XXX
6542 %}
6543 
6544 // Load Aligned Packed Byte to XMM register
6545 instruct loadA8B(regD dst, memory mem) %{
6546   match(Set dst (Load8B mem));
6547   ins_cost(125);
6548   format %{ "MOVQ  $dst,$mem\t! packed8B" %}
6549   ins_encode( movq_ld(dst, mem));
6550   ins_pipe( pipe_slow );
6551 %}
6552 
6553 // Load Aligned Packed Short to XMM register
6554 instruct loadA4S(regD dst, memory mem) %{
6555   match(Set dst (Load4S mem));
6556   ins_cost(125);
6557   format %{ "MOVQ  $dst,$mem\t! packed4S" %}
6558   ins_encode( movq_ld(dst, mem));
6559   ins_pipe( pipe_slow );
6560 %}
6561 
6562 // Load Aligned Packed Char to XMM register
6563 instruct loadA4C(regD dst, memory mem) %{
6564   match(Set dst (Load4C mem));
6565   ins_cost(125);
6566   format %{ "MOVQ  $dst,$mem\t! packed4C" %}
6567   ins_encode( movq_ld(dst, mem));
6568   ins_pipe( pipe_slow );
6569 %}
6570 
6571 // Load Aligned Packed Integer to XMM register
6572 instruct load2IU(regD dst, memory mem) %{
6573   match(Set dst (Load2I mem));
6574   ins_cost(125);
6575   format %{ "MOVQ  $dst,$mem\t! packed2I" %}
6576   ins_encode( movq_ld(dst, mem));
6577   ins_pipe( pipe_slow );
6578 %}
6579 
6580 // Load Aligned Packed Single to XMM
6581 instruct loadA2F(regD dst, memory mem) %{
6582   match(Set dst (Load2F mem));
6583   ins_cost(145);
6584   format %{ "MOVQ  $dst,$mem\t! packed2F" %}
6585   ins_encode( movq_ld(dst, mem));
6586   ins_pipe( pipe_slow );
6587 %}
6588 
6589 // Load Effective Address
6590 instruct leaP8(rRegP dst, indOffset8 mem)
6591 %{
6592   match(Set dst mem);
6593 
6594   ins_cost(110); // XXX
6595   format %{ "leaq    $dst, $mem\t# ptr 8" %}
6596   opcode(0x8D);
6597   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6598   ins_pipe(ialu_reg_reg_fat);
6599 %}
6600 
6601 instruct leaP32(rRegP dst, indOffset32 mem)
6602 %{
6603   match(Set dst mem);
6604 
6605   ins_cost(110);
6606   format %{ "leaq    $dst, $mem\t# ptr 32" %}
6607   opcode(0x8D);
6608   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6609   ins_pipe(ialu_reg_reg_fat);
6610 %}
6611 
6612 // instruct leaPIdx(rRegP dst, indIndex mem)
6613 // %{
6614 //   match(Set dst mem);
6615 
6616 //   ins_cost(110);
6617 //   format %{ "leaq    $dst, $mem\t# ptr idx" %}
6618 //   opcode(0x8D);
6619 //   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6620 //   ins_pipe(ialu_reg_reg_fat);
6621 // %}
6622 
6623 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6624 %{
6625   match(Set dst mem);
6626 
6627   ins_cost(110);
6628   format %{ "leaq    $dst, $mem\t# ptr idxoff" %}
6629   opcode(0x8D);
6630   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6631   ins_pipe(ialu_reg_reg_fat);
6632 %}
6633 
6634 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6635 %{
6636   match(Set dst mem);
6637 
6638   ins_cost(110);
6639   format %{ "leaq    $dst, $mem\t# ptr idxscale" %}
6640   opcode(0x8D);
6641   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6642   ins_pipe(ialu_reg_reg_fat);
6643 %}
6644 
6645 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6646 %{
6647   match(Set dst mem);
6648 
6649   ins_cost(110);
6650   format %{ "leaq    $dst, $mem\t# ptr idxscaleoff" %}
6651   opcode(0x8D);
6652   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6653   ins_pipe(ialu_reg_reg_fat);
6654 %}
6655 
6656 instruct loadConI(rRegI dst, immI src)
6657 %{
6658   match(Set dst src);
6659 
6660   format %{ "movl    $dst, $src\t# int" %}
6661   ins_encode(load_immI(dst, src));
6662   ins_pipe(ialu_reg_fat); // XXX
6663 %}
6664 
6665 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6666 %{
6667   match(Set dst src);
6668   effect(KILL cr);
6669 
6670   ins_cost(50);
6671   format %{ "xorl    $dst, $dst\t# int" %}
6672   opcode(0x33); /* + rd */
6673   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6674   ins_pipe(ialu_reg);
6675 %}
6676 
6677 instruct loadConL(rRegL dst, immL src)
6678 %{
6679   match(Set dst src);
6680 
6681   ins_cost(150);
6682   format %{ "movq    $dst, $src\t# long" %}
6683   ins_encode(load_immL(dst, src));
6684   ins_pipe(ialu_reg);
6685 %}
6686 
6687 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6688 %{
6689   match(Set dst src);
6690   effect(KILL cr);
6691 
6692   ins_cost(50);
6693   format %{ "xorl    $dst, $dst\t# long" %}
6694   opcode(0x33); /* + rd */
6695   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6696   ins_pipe(ialu_reg); // XXX
6697 %}
6698 
6699 instruct loadConUL32(rRegL dst, immUL32 src)
6700 %{
6701   match(Set dst src);
6702 
6703   ins_cost(60);
6704   format %{ "movl    $dst, $src\t# long (unsigned 32-bit)" %}
6705   ins_encode(load_immUL32(dst, src));
6706   ins_pipe(ialu_reg);
6707 %}
6708 
6709 instruct loadConL32(rRegL dst, immL32 src)
6710 %{
6711   match(Set dst src);
6712 
6713   ins_cost(70);
6714   format %{ "movq    $dst, $src\t# long (32-bit)" %}
6715   ins_encode(load_immL32(dst, src));
6716   ins_pipe(ialu_reg);
6717 %}
6718 
6719 instruct loadConP(rRegP dst, immP src)
6720 %{
6721   match(Set dst src);
6722 
6723   format %{ "movq    $dst, $src\t# ptr" %}
6724   ins_encode(load_immP(dst, src));
6725   ins_pipe(ialu_reg_fat); // XXX
6726 %}
6727 
6728 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6729 %{
6730   match(Set dst src);
6731   effect(KILL cr);
6732 
6733   ins_cost(50);
6734   format %{ "xorl    $dst, $dst\t# ptr" %}
6735   opcode(0x33); /* + rd */
6736   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6737   ins_pipe(ialu_reg);
6738 %}
6739 
6740 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6741 %{
6742   match(Set dst src);
6743   effect(KILL cr);
6744 
6745   ins_cost(60);
6746   format %{ "movl    $dst, $src\t# ptr (positive 32-bit)" %}
6747   ins_encode(load_immP31(dst, src));
6748   ins_pipe(ialu_reg);
6749 %}
6750 
6751 instruct loadConF(regF dst, immF src)
6752 %{
6753   match(Set dst src);
6754   ins_cost(125);
6755 
6756   format %{ "movss   $dst, [$src]" %}
6757   ins_encode(load_conF(dst, src));
6758   ins_pipe(pipe_slow);
6759 %}
6760 
6761 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6762   match(Set dst src);
6763   effect(KILL cr);
6764   format %{ "xorq    $dst, $src\t# compressed NULL ptr" %}
6765   ins_encode %{
6766     Register dst = $dst$$Register;
6767     __ xorq(dst, dst);
6768   %}
6769   ins_pipe(ialu_reg);
6770 %}
6771 
6772 instruct loadConN(rRegN dst, immN src) %{
6773   match(Set dst src);
6774 
6775   ins_cost(125);
6776   format %{ "movl    $dst, $src\t# compressed ptr" %}
6777   ins_encode %{
6778     address con = (address)$src$$constant;
6779     Register dst = $dst$$Register;
6780     if (con == NULL) {
6781       ShouldNotReachHere();
6782     } else {
6783       __ set_narrow_oop(dst, (jobject)$src$$constant);
6784     }
6785   %}
6786   ins_pipe(ialu_reg_fat); // XXX
6787 %}
6788 
6789 instruct loadConF0(regF dst, immF0 src)
6790 %{
6791   match(Set dst src);
6792   ins_cost(100);
6793 
6794   format %{ "xorps   $dst, $dst\t# float 0.0" %}
6795   opcode(0x0F, 0x57);
6796   ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6797   ins_pipe(pipe_slow);
6798 %}
6799 
6800 // Use the same format since predicate() can not be used here.
6801 instruct loadConD(regD dst, immD src)
6802 %{
6803   match(Set dst src);
6804   ins_cost(125);
6805 
6806   format %{ "movsd   $dst, [$src]" %}
6807   ins_encode(load_conD(dst, src));
6808   ins_pipe(pipe_slow);
6809 %}
6810 
6811 instruct loadConD0(regD dst, immD0 src)
6812 %{
6813   match(Set dst src);
6814   ins_cost(100);
6815 
6816   format %{ "xorpd   $dst, $dst\t# double 0.0" %}
6817   opcode(0x66, 0x0F, 0x57);
6818   ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6819   ins_pipe(pipe_slow);
6820 %}
6821 
6822 instruct loadSSI(rRegI dst, stackSlotI src)
6823 %{
6824   match(Set dst src);
6825 
6826   ins_cost(125);
6827   format %{ "movl    $dst, $src\t# int stk" %}
6828   opcode(0x8B);
6829   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6830   ins_pipe(ialu_reg_mem);
6831 %}
6832 
6833 instruct loadSSL(rRegL dst, stackSlotL src)
6834 %{
6835   match(Set dst src);
6836 
6837   ins_cost(125);
6838   format %{ "movq    $dst, $src\t# long stk" %}
6839   opcode(0x8B);
6840   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6841   ins_pipe(ialu_reg_mem);
6842 %}
6843 
6844 instruct loadSSP(rRegP dst, stackSlotP src)
6845 %{
6846   match(Set dst src);
6847 
6848   ins_cost(125);
6849   format %{ "movq    $dst, $src\t# ptr stk" %}
6850   opcode(0x8B);
6851   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6852   ins_pipe(ialu_reg_mem);
6853 %}
6854 
6855 instruct loadSSF(regF dst, stackSlotF src)
6856 %{
6857   match(Set dst src);
6858 
6859   ins_cost(125);
6860   format %{ "movss   $dst, $src\t# float stk" %}
6861   opcode(0xF3, 0x0F, 0x10);
6862   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6863   ins_pipe(pipe_slow); // XXX
6864 %}
6865 
6866 // Use the same format since predicate() can not be used here.
6867 instruct loadSSD(regD dst, stackSlotD src)
6868 %{
6869   match(Set dst src);
6870 
6871   ins_cost(125);
6872   format %{ "movsd   $dst, $src\t# double stk" %}
6873   ins_encode  %{
6874     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6875   %}
6876   ins_pipe(pipe_slow); // XXX
6877 %}
6878 
6879 // Prefetch instructions.
6880 // Must be safe to execute with invalid address (cannot fault).
6881 
6882 instruct prefetchr( memory mem ) %{
6883   predicate(ReadPrefetchInstr==3);
6884   match(PrefetchRead mem);
6885   ins_cost(125);
6886 
6887   format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6888   opcode(0x0F, 0x0D);     /* Opcode 0F 0D /0 */
6889   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6890   ins_pipe(ialu_mem);
6891 %}
6892 
6893 instruct prefetchrNTA( memory mem ) %{
6894   predicate(ReadPrefetchInstr==0);
6895   match(PrefetchRead mem);
6896   ins_cost(125);
6897 
6898   format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6899   opcode(0x0F, 0x18);     /* Opcode 0F 18 /0 */
6900   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6901   ins_pipe(ialu_mem);
6902 %}
6903 
6904 instruct prefetchrT0( memory mem ) %{
6905   predicate(ReadPrefetchInstr==1);
6906   match(PrefetchRead mem);
6907   ins_cost(125);
6908 
6909   format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6910   opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6911   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6912   ins_pipe(ialu_mem);
6913 %}
6914 
6915 instruct prefetchrT2( memory mem ) %{
6916   predicate(ReadPrefetchInstr==2);
6917   match(PrefetchRead mem);
6918   ins_cost(125);
6919 
6920   format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6921   opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6922   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6923   ins_pipe(ialu_mem);
6924 %}
6925 
6926 instruct prefetchw( memory mem ) %{
6927   predicate(AllocatePrefetchInstr==3);
6928   match(PrefetchWrite mem);
6929   ins_cost(125);
6930 
6931   format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6932   opcode(0x0F, 0x0D);     /* Opcode 0F 0D /1 */
6933   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6934   ins_pipe(ialu_mem);
6935 %}
6936 
6937 instruct prefetchwNTA( memory mem ) %{
6938   predicate(AllocatePrefetchInstr==0);
6939   match(PrefetchWrite mem);
6940   ins_cost(125);
6941 
6942   format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6943   opcode(0x0F, 0x18);     /* Opcode 0F 18 /0 */
6944   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6945   ins_pipe(ialu_mem);
6946 %}
6947 
6948 instruct prefetchwT0( memory mem ) %{
6949   predicate(AllocatePrefetchInstr==1);
6950   match(PrefetchWrite mem);
6951   ins_cost(125);
6952 
6953   format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6954   opcode(0x0F, 0x18);     /* Opcode 0F 18 /1 */
6955   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6956   ins_pipe(ialu_mem);
6957 %}
6958 
6959 instruct prefetchwT2( memory mem ) %{
6960   predicate(AllocatePrefetchInstr==2);
6961   match(PrefetchWrite mem);
6962   ins_cost(125);
6963 
6964   format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6965   opcode(0x0F, 0x18);     /* Opcode 0F 18 /3 */
6966   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6967   ins_pipe(ialu_mem);
6968 %}
6969 
6970 //----------Store Instructions-------------------------------------------------
6971 
6972 // Store Byte
6973 instruct storeB(memory mem, rRegI src)
6974 %{
6975   match(Set mem (StoreB mem src));
6976 
6977   ins_cost(125); // XXX
6978   format %{ "movb    $mem, $src\t# byte" %}
6979   opcode(0x88);
6980   ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6981   ins_pipe(ialu_mem_reg);
6982 %}
6983 
6984 // Store Char/Short
6985 instruct storeC(memory mem, rRegI src)
6986 %{
6987   match(Set mem (StoreC mem src));
6988 
6989   ins_cost(125); // XXX
6990   format %{ "movw    $mem, $src\t# char/short" %}
6991   opcode(0x89);
6992   ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6993   ins_pipe(ialu_mem_reg);
6994 %}
6995 
6996 // Store Integer
6997 instruct storeI(memory mem, rRegI src)
6998 %{
6999   match(Set mem (StoreI mem src));
7000 
7001   ins_cost(125); // XXX
7002   format %{ "movl    $mem, $src\t# int" %}
7003   opcode(0x89);
7004   ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
7005   ins_pipe(ialu_mem_reg);
7006 %}
7007 
7008 // Store Long
7009 instruct storeL(memory mem, rRegL src)
7010 %{
7011   match(Set mem (StoreL mem src));
7012 
7013   ins_cost(125); // XXX
7014   format %{ "movq    $mem, $src\t# long" %}
7015   opcode(0x89);
7016   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
7017   ins_pipe(ialu_mem_reg); // XXX
7018 %}
7019 
7020 // Store Pointer
7021 instruct storeP(memory mem, any_RegP src)
7022 %{
7023   match(Set mem (StoreP mem src));
7024 
7025   ins_cost(125); // XXX
7026   format %{ "movq    $mem, $src\t# ptr" %}
7027   opcode(0x89);
7028   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
7029   ins_pipe(ialu_mem_reg);
7030 %}
7031 
7032 // Store NULL Pointer, mark word, or other simple pointer constant.
7033 instruct storeImmP(memory mem, immP31 src)
7034 %{
7035   match(Set mem (StoreP mem src));
7036 
7037   ins_cost(125); // XXX
7038   format %{ "movq    $mem, $src\t# ptr" %}
7039   opcode(0xC7); /* C7 /0 */
7040   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7041   ins_pipe(ialu_mem_imm);
7042 %}
7043 
7044 // Store Compressed Pointer
7045 instruct storeN(memory mem, rRegN src)
7046 %{
7047   match(Set mem (StoreN mem src));
7048 
7049   ins_cost(125); // XXX
7050   format %{ "movl    $mem, $src\t# compressed ptr" %}
7051   ins_encode %{
7052     Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
7053     Register src = as_Register($src$$reg);
7054     __ movl(addr, src);
7055   %}
7056   ins_pipe(ialu_mem_reg);
7057 %}
7058 
7059 // Store Integer Immediate
7060 instruct storeImmI(memory mem, immI src)
7061 %{
7062   match(Set mem (StoreI mem src));
7063 
7064   ins_cost(150);
7065   format %{ "movl    $mem, $src\t# int" %}
7066   opcode(0xC7); /* C7 /0 */
7067   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7068   ins_pipe(ialu_mem_imm);
7069 %}
7070 
7071 // Store Long Immediate
7072 instruct storeImmL(memory mem, immL32 src)
7073 %{
7074   match(Set mem (StoreL mem src));
7075 
7076   ins_cost(150);
7077   format %{ "movq    $mem, $src\t# long" %}
7078   opcode(0xC7); /* C7 /0 */
7079   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7080   ins_pipe(ialu_mem_imm);
7081 %}
7082 
7083 // Store Short/Char Immediate
7084 instruct storeImmI16(memory mem, immI16 src)
7085 %{
7086   predicate(UseStoreImmI16);
7087   match(Set mem (StoreC mem src));
7088 
7089   ins_cost(150);
7090   format %{ "movw    $mem, $src\t# short/char" %}
7091   opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
7092   ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
7093   ins_pipe(ialu_mem_imm);
7094 %}
7095 
7096 // Store Byte Immediate
7097 instruct storeImmB(memory mem, immI8 src)
7098 %{
7099   match(Set mem (StoreB mem src));
7100 
7101   ins_cost(150); // XXX
7102   format %{ "movb    $mem, $src\t# byte" %}
7103   opcode(0xC6); /* C6 /0 */
7104   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7105   ins_pipe(ialu_mem_imm);
7106 %}
7107 
7108 // Store Aligned Packed Byte XMM register to memory
7109 instruct storeA8B(memory mem, regD src) %{
7110   match(Set mem (Store8B mem src));
7111   ins_cost(145);
7112   format %{ "MOVQ  $mem,$src\t! packed8B" %}
7113   ins_encode( movq_st(mem, src));
7114   ins_pipe( pipe_slow );
7115 %}
7116 
7117 // Store Aligned Packed Char/Short XMM register to memory
7118 instruct storeA4C(memory mem, regD src) %{
7119   match(Set mem (Store4C mem src));
7120   ins_cost(145);
7121   format %{ "MOVQ  $mem,$src\t! packed4C" %}
7122   ins_encode( movq_st(mem, src));
7123   ins_pipe( pipe_slow );
7124 %}
7125 
7126 // Store Aligned Packed Integer XMM register to memory
7127 instruct storeA2I(memory mem, regD src) %{
7128   match(Set mem (Store2I mem src));
7129   ins_cost(145);
7130   format %{ "MOVQ  $mem,$src\t! packed2I" %}
7131   ins_encode( movq_st(mem, src));
7132   ins_pipe( pipe_slow );
7133 %}
7134 
7135 // Store CMS card-mark Immediate
7136 instruct storeImmCM0(memory mem, immI0 src)
7137 %{
7138   match(Set mem (StoreCM mem src));
7139 
7140   ins_cost(150); // XXX
7141   format %{ "movb    $mem, $src\t# CMS card-mark byte 0" %}
7142   opcode(0xC6); /* C6 /0 */
7143   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7144   ins_pipe(ialu_mem_imm);
7145 %}
7146 
7147 // Store Aligned Packed Single Float XMM register to memory
7148 instruct storeA2F(memory mem, regD src) %{
7149   match(Set mem (Store2F mem src));
7150   ins_cost(145);
7151   format %{ "MOVQ  $mem,$src\t! packed2F" %}
7152   ins_encode( movq_st(mem, src));
7153   ins_pipe( pipe_slow );
7154 %}
7155 
7156 // Store Float
7157 instruct storeF(memory mem, regF src)
7158 %{
7159   match(Set mem (StoreF mem src));
7160 
7161   ins_cost(95); // XXX
7162   format %{ "movss   $mem, $src\t# float" %}
7163   opcode(0xF3, 0x0F, 0x11);
7164   ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7165   ins_pipe(pipe_slow); // XXX
7166 %}
7167 
7168 // Store immediate Float value (it is faster than store from XMM register)
7169 instruct storeF_imm(memory mem, immF src)
7170 %{
7171   match(Set mem (StoreF mem src));
7172 
7173   ins_cost(50);
7174   format %{ "movl    $mem, $src\t# float" %}
7175   opcode(0xC7); /* C7 /0 */
7176   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7177   ins_pipe(ialu_mem_imm);
7178 %}
7179 
7180 // Store Double
7181 instruct storeD(memory mem, regD src)
7182 %{
7183   match(Set mem (StoreD mem src));
7184 
7185   ins_cost(95); // XXX
7186   format %{ "movsd   $mem, $src\t# double" %}
7187   opcode(0xF2, 0x0F, 0x11);
7188   ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7189   ins_pipe(pipe_slow); // XXX
7190 %}
7191 
7192 // Store immediate double 0.0 (it is faster than store from XMM register)
7193 instruct storeD0_imm(memory mem, immD0 src)
7194 %{
7195   match(Set mem (StoreD mem src));
7196 
7197   ins_cost(50);
7198   format %{ "movq    $mem, $src\t# double 0." %}
7199   opcode(0xC7); /* C7 /0 */
7200   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7201   ins_pipe(ialu_mem_imm);
7202 %}
7203 
7204 instruct storeSSI(stackSlotI dst, rRegI src)
7205 %{
7206   match(Set dst src);
7207 
7208   ins_cost(100);
7209   format %{ "movl    $dst, $src\t# int stk" %}
7210   opcode(0x89);
7211   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7212   ins_pipe( ialu_mem_reg );
7213 %}
7214 
7215 instruct storeSSL(stackSlotL dst, rRegL src)
7216 %{
7217   match(Set dst src);
7218 
7219   ins_cost(100);
7220   format %{ "movq    $dst, $src\t# long stk" %}
7221   opcode(0x89);
7222   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7223   ins_pipe(ialu_mem_reg);
7224 %}
7225 
7226 instruct storeSSP(stackSlotP dst, rRegP src)
7227 %{
7228   match(Set dst src);
7229 
7230   ins_cost(100);
7231   format %{ "movq    $dst, $src\t# ptr stk" %}
7232   opcode(0x89);
7233   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7234   ins_pipe(ialu_mem_reg);
7235 %}
7236 
7237 instruct storeSSF(stackSlotF dst, regF src)
7238 %{
7239   match(Set dst src);
7240 
7241   ins_cost(95); // XXX
7242   format %{ "movss   $dst, $src\t# float stk" %}
7243   opcode(0xF3, 0x0F, 0x11);
7244   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7245   ins_pipe(pipe_slow); // XXX
7246 %}
7247 
7248 instruct storeSSD(stackSlotD dst, regD src)
7249 %{
7250   match(Set dst src);
7251 
7252   ins_cost(95); // XXX
7253   format %{ "movsd   $dst, $src\t# double stk" %}
7254   opcode(0xF2, 0x0F, 0x11);
7255   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7256   ins_pipe(pipe_slow); // XXX
7257 %}
7258 
7259 //----------BSWAP Instructions-------------------------------------------------
7260 instruct bytes_reverse_int(rRegI dst) %{
7261   match(Set dst (ReverseBytesI dst));
7262 
7263   format %{ "bswapl  $dst" %}
7264   opcode(0x0F, 0xC8);  /*Opcode 0F /C8 */
7265   ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
7266   ins_pipe( ialu_reg );
7267 %}
7268 
7269 instruct bytes_reverse_long(rRegL dst) %{
7270   match(Set dst (ReverseBytesL dst));
7271 
7272   format %{ "bswapq  $dst" %}
7273 
7274   opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
7275   ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
7276   ins_pipe( ialu_reg);
7277 %}
7278 
7279 instruct loadI_reversed(rRegI dst, memory src) %{
7280   match(Set dst (ReverseBytesI (LoadI src)));
7281 
7282   format %{ "bswap_movl $dst, $src" %}
7283   opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7284   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
7285   ins_pipe( ialu_reg_mem );
7286 %}
7287 
7288 instruct loadL_reversed(rRegL dst, memory src) %{
7289   match(Set dst (ReverseBytesL (LoadL src)));
7290 
7291   format %{ "bswap_movq $dst, $src" %}
7292   opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7293   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
7294   ins_pipe( ialu_reg_mem );
7295 %}
7296 
7297 instruct storeI_reversed(memory dst, rRegI src) %{
7298   match(Set dst (StoreI dst (ReverseBytesI  src)));
7299 
7300   format %{ "movl_bswap $dst, $src" %}
7301   opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7302   ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
7303   ins_pipe( ialu_mem_reg );
7304 %}
7305 
7306 instruct storeL_reversed(memory dst, rRegL src) %{
7307   match(Set dst (StoreL dst (ReverseBytesL  src)));
7308 
7309   format %{ "movq_bswap $dst, $src" %}
7310   opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7311   ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
7312   ins_pipe( ialu_mem_reg );
7313 %}
7314 
7315 //----------MemBar Instructions-----------------------------------------------
7316 // Memory barrier flavors
7317 
7318 instruct membar_acquire()
7319 %{
7320   match(MemBarAcquire);
7321   ins_cost(0);
7322 
7323   size(0);
7324   format %{ "MEMBAR-acquire" %}
7325   ins_encode();
7326   ins_pipe(empty);
7327 %}
7328 
7329 instruct membar_acquire_lock()
7330 %{
7331   match(MemBarAcquire);
7332   predicate(Matcher::prior_fast_lock(n));
7333   ins_cost(0);
7334 
7335   size(0);
7336   format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7337   ins_encode();
7338   ins_pipe(empty);
7339 %}
7340 
7341 instruct membar_release()
7342 %{
7343   match(MemBarRelease);
7344   ins_cost(0);
7345 
7346   size(0);
7347   format %{ "MEMBAR-release" %}
7348   ins_encode();
7349   ins_pipe(empty);
7350 %}
7351 
7352 instruct membar_release_lock()
7353 %{
7354   match(MemBarRelease);
7355   predicate(Matcher::post_fast_unlock(n));
7356   ins_cost(0);
7357 
7358   size(0);
7359   format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7360   ins_encode();
7361   ins_pipe(empty);
7362 %}
7363 
7364 instruct membar_volatile()
7365 %{
7366   match(MemBarVolatile);
7367   ins_cost(400);
7368 
7369   format %{ "MEMBAR-volatile" %}
7370   ins_encode(enc_membar_volatile);
7371   ins_pipe(pipe_slow);
7372 %}
7373 
7374 instruct unnecessary_membar_volatile()
7375 %{
7376   match(MemBarVolatile);
7377   predicate(Matcher::post_store_load_barrier(n));
7378   ins_cost(0);
7379 
7380   size(0);
7381   format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7382   ins_encode();
7383   ins_pipe(empty);
7384 %}
7385 
7386 //----------Move Instructions--------------------------------------------------
7387 
7388 instruct castX2P(rRegP dst, rRegL src)
7389 %{
7390   match(Set dst (CastX2P src));
7391 
7392   format %{ "movq    $dst, $src\t# long->ptr" %}
7393   ins_encode(enc_copy_wide(dst, src));
7394   ins_pipe(ialu_reg_reg); // XXX
7395 %}
7396 
7397 instruct castP2X(rRegL dst, rRegP src)
7398 %{
7399   match(Set dst (CastP2X src));
7400 
7401   format %{ "movq    $dst, $src\t# ptr -> long" %}
7402   ins_encode(enc_copy_wide(dst, src));
7403   ins_pipe(ialu_reg_reg); // XXX
7404 %}
7405 
7406 
7407 // Convert oop pointer into compressed form
7408 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7409   predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7410   match(Set dst (EncodeP src));
7411   effect(KILL cr);
7412   format %{ "encode_heap_oop $dst,$src" %}
7413   ins_encode %{
7414     Register s = $src$$Register;
7415     Register d = $dst$$Register;
7416     if (s != d) {
7417       __ movq(d, s);
7418     }
7419     __ encode_heap_oop(d);
7420   %}
7421   ins_pipe(ialu_reg_long);
7422 %}
7423 
7424 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7425   predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7426   match(Set dst (EncodeP src));
7427   effect(KILL cr);
7428   format %{ "encode_heap_oop_not_null $dst,$src" %}
7429   ins_encode %{
7430     Register s = $src$$Register;
7431     Register d = $dst$$Register;
7432     __ encode_heap_oop_not_null(d, s);
7433   %}
7434   ins_pipe(ialu_reg_long);
7435 %}
7436 
7437 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7438   predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7439             n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7440   match(Set dst (DecodeN src));
7441   effect(KILL cr);
7442   format %{ "decode_heap_oop $dst,$src" %}
7443   ins_encode %{
7444     Register s = $src$$Register;
7445     Register d = $dst$$Register;
7446     if (s != d) {
7447       __ movq(d, s);
7448     }
7449     __ decode_heap_oop(d);
7450   %}
7451   ins_pipe(ialu_reg_long);
7452 %}
7453 
7454 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{
7455   predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7456             n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7457   match(Set dst (DecodeN src));
7458   format %{ "decode_heap_oop_not_null $dst,$src" %}
7459   ins_encode %{
7460     Register s = $src$$Register;
7461     Register d = $dst$$Register;
7462     __ decode_heap_oop_not_null(d, s);
7463   %}
7464   ins_pipe(ialu_reg_long);
7465 %}
7466 
7467 
7468 //----------Conditional Move---------------------------------------------------
7469 // Jump
7470 // dummy instruction for generating temp registers
7471 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7472   match(Jump (LShiftL switch_val shift));
7473   ins_cost(350);
7474   predicate(false);
7475   effect(TEMP dest);
7476 
7477   format %{ "leaq    $dest, table_base\n\t"
7478             "jmp     [$dest + $switch_val << $shift]\n\t" %}
7479   ins_encode(jump_enc_offset(switch_val, shift, dest));
7480   ins_pipe(pipe_jmp);
7481   ins_pc_relative(1);
7482 %}
7483 
7484 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7485   match(Jump (AddL (LShiftL switch_val shift) offset));
7486   ins_cost(350);
7487   effect(TEMP dest);
7488 
7489   format %{ "leaq    $dest, table_base\n\t"
7490             "jmp     [$dest + $switch_val << $shift + $offset]\n\t" %}
7491   ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
7492   ins_pipe(pipe_jmp);
7493   ins_pc_relative(1);
7494 %}
7495 
7496 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7497   match(Jump switch_val);
7498   ins_cost(350);
7499   effect(TEMP dest);
7500 
7501   format %{ "leaq    $dest, table_base\n\t"
7502             "jmp     [$dest + $switch_val]\n\t" %}
7503   ins_encode(jump_enc(switch_val, dest));
7504   ins_pipe(pipe_jmp);
7505   ins_pc_relative(1);
7506 %}
7507 
7508 // Conditional move
7509 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7510 %{
7511   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7512 
7513   ins_cost(200); // XXX
7514   format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7515   opcode(0x0F, 0x40);
7516   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7517   ins_pipe(pipe_cmov_reg);
7518 %}
7519 
7520 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7521   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7522 
7523   ins_cost(200); // XXX
7524   format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7525   opcode(0x0F, 0x40);
7526   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7527   ins_pipe(pipe_cmov_reg);
7528 %}
7529 
7530 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7531   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7532   ins_cost(200);
7533   expand %{
7534     cmovI_regU(cop, cr, dst, src);
7535   %}
7536 %}
7537 
7538 // Conditional move
7539 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7540   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7541 
7542   ins_cost(250); // XXX
7543   format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7544   opcode(0x0F, 0x40);
7545   ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7546   ins_pipe(pipe_cmov_mem);
7547 %}
7548 
7549 // Conditional move
7550 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7551 %{
7552   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7553 
7554   ins_cost(250); // XXX
7555   format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7556   opcode(0x0F, 0x40);
7557   ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7558   ins_pipe(pipe_cmov_mem);
7559 %}
7560 
7561 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7562   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7563   ins_cost(250);
7564   expand %{
7565     cmovI_memU(cop, cr, dst, src);
7566   %}
7567 %}
7568 
7569 // Conditional move
7570 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7571 %{
7572   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7573 
7574   ins_cost(200); // XXX
7575   format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7576   opcode(0x0F, 0x40);
7577   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7578   ins_pipe(pipe_cmov_reg);
7579 %}
7580 
7581 // Conditional move
7582 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7583 %{
7584   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7585 
7586   ins_cost(200); // XXX
7587   format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7588   opcode(0x0F, 0x40);
7589   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7590   ins_pipe(pipe_cmov_reg);
7591 %}
7592 
7593 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7594   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7595   ins_cost(200);
7596   expand %{
7597     cmovN_regU(cop, cr, dst, src);
7598   %}
7599 %}
7600 
7601 // Conditional move
7602 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7603 %{
7604   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7605 
7606   ins_cost(200); // XXX
7607   format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7608   opcode(0x0F, 0x40);
7609   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7610   ins_pipe(pipe_cmov_reg);  // XXX
7611 %}
7612 
7613 // Conditional move
7614 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7615 %{
7616   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7617 
7618   ins_cost(200); // XXX
7619   format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7620   opcode(0x0F, 0x40);
7621   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7622   ins_pipe(pipe_cmov_reg); // XXX
7623 %}
7624 
7625 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7626   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7627   ins_cost(200);
7628   expand %{
7629     cmovP_regU(cop, cr, dst, src);
7630   %}
7631 %}
7632 
7633 // DISABLED: Requires the ADLC to emit a bottom_type call that
7634 // correctly meets the two pointer arguments; one is an incoming
7635 // register but the other is a memory operand.  ALSO appears to
7636 // be buggy with implicit null checks.
7637 //
7638 //// Conditional move
7639 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7640 //%{
7641 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7642 //  ins_cost(250);
7643 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7644 //  opcode(0x0F,0x40);
7645 //  ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7646 //  ins_pipe( pipe_cmov_mem );
7647 //%}
7648 //
7649 //// Conditional move
7650 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7651 //%{
7652 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7653 //  ins_cost(250);
7654 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7655 //  opcode(0x0F,0x40);
7656 //  ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7657 //  ins_pipe( pipe_cmov_mem );
7658 //%}
7659 
7660 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7661 %{
7662   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7663 
7664   ins_cost(200); // XXX
7665   format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7666   opcode(0x0F, 0x40);
7667   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7668   ins_pipe(pipe_cmov_reg);  // XXX
7669 %}
7670 
7671 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7672 %{
7673   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7674 
7675   ins_cost(200); // XXX
7676   format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7677   opcode(0x0F, 0x40);
7678   ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7679   ins_pipe(pipe_cmov_mem);  // XXX
7680 %}
7681 
7682 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7683 %{
7684   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7685 
7686   ins_cost(200); // XXX
7687   format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7688   opcode(0x0F, 0x40);
7689   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7690   ins_pipe(pipe_cmov_reg); // XXX
7691 %}
7692 
7693 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7694   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7695   ins_cost(200);
7696   expand %{
7697     cmovL_regU(cop, cr, dst, src);
7698   %}
7699 %}
7700 
7701 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7702 %{
7703   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7704 
7705   ins_cost(200); // XXX
7706   format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7707   opcode(0x0F, 0x40);
7708   ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7709   ins_pipe(pipe_cmov_mem); // XXX
7710 %}
7711 
7712 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
7713   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7714   ins_cost(200);
7715   expand %{
7716     cmovL_memU(cop, cr, dst, src);
7717   %}
7718 %}
7719 
7720 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7721 %{
7722   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7723 
7724   ins_cost(200); // XXX
7725   format %{ "jn$cop    skip\t# signed cmove float\n\t"
7726             "movss     $dst, $src\n"
7727     "skip:" %}
7728   ins_encode(enc_cmovf_branch(cop, dst, src));
7729   ins_pipe(pipe_slow);
7730 %}
7731 
7732 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7733 // %{
7734 //   match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7735 
7736 //   ins_cost(200); // XXX
7737 //   format %{ "jn$cop    skip\t# signed cmove float\n\t"
7738 //             "movss     $dst, $src\n"
7739 //     "skip:" %}
7740 //   ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7741 //   ins_pipe(pipe_slow);
7742 // %}
7743 
7744 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7745 %{
7746   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7747 
7748   ins_cost(200); // XXX
7749   format %{ "jn$cop    skip\t# unsigned cmove float\n\t"
7750             "movss     $dst, $src\n"
7751     "skip:" %}
7752   ins_encode(enc_cmovf_branch(cop, dst, src));
7753   ins_pipe(pipe_slow);
7754 %}
7755 
7756 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
7757   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7758   ins_cost(200);
7759   expand %{
7760     cmovF_regU(cop, cr, dst, src);
7761   %}
7762 %}
7763 
7764 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7765 %{
7766   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7767 
7768   ins_cost(200); // XXX
7769   format %{ "jn$cop    skip\t# signed cmove double\n\t"
7770             "movsd     $dst, $src\n"
7771     "skip:" %}
7772   ins_encode(enc_cmovd_branch(cop, dst, src));
7773   ins_pipe(pipe_slow);
7774 %}
7775 
7776 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7777 %{
7778   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7779 
7780   ins_cost(200); // XXX
7781   format %{ "jn$cop    skip\t# unsigned cmove double\n\t"
7782             "movsd     $dst, $src\n"
7783     "skip:" %}
7784   ins_encode(enc_cmovd_branch(cop, dst, src));
7785   ins_pipe(pipe_slow);
7786 %}
7787 
7788 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
7789   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7790   ins_cost(200);
7791   expand %{
7792     cmovD_regU(cop, cr, dst, src);
7793   %}
7794 %}
7795 
7796 //----------Arithmetic Instructions--------------------------------------------
7797 //----------Addition Instructions----------------------------------------------
7798 
7799 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7800 %{
7801   match(Set dst (AddI dst src));
7802   effect(KILL cr);
7803 
7804   format %{ "addl    $dst, $src\t# int" %}
7805   opcode(0x03);
7806   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7807   ins_pipe(ialu_reg_reg);
7808 %}
7809 
7810 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7811 %{
7812   match(Set dst (AddI dst src));
7813   effect(KILL cr);
7814 
7815   format %{ "addl    $dst, $src\t# int" %}
7816   opcode(0x81, 0x00); /* /0 id */
7817   ins_encode(OpcSErm(dst, src), Con8or32(src));
7818   ins_pipe( ialu_reg );
7819 %}
7820 
7821 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7822 %{
7823   match(Set dst (AddI dst (LoadI src)));
7824   effect(KILL cr);
7825 
7826   ins_cost(125); // XXX
7827   format %{ "addl    $dst, $src\t# int" %}
7828   opcode(0x03);
7829   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7830   ins_pipe(ialu_reg_mem);
7831 %}
7832 
7833 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7834 %{
7835   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7836   effect(KILL cr);
7837 
7838   ins_cost(150); // XXX
7839   format %{ "addl    $dst, $src\t# int" %}
7840   opcode(0x01); /* Opcode 01 /r */
7841   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7842   ins_pipe(ialu_mem_reg);
7843 %}
7844 
7845 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7846 %{
7847   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7848   effect(KILL cr);
7849 
7850   ins_cost(125); // XXX
7851   format %{ "addl    $dst, $src\t# int" %}
7852   opcode(0x81); /* Opcode 81 /0 id */
7853   ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7854   ins_pipe(ialu_mem_imm);
7855 %}
7856 
7857 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7858 %{
7859   predicate(UseIncDec);
7860   match(Set dst (AddI dst src));
7861   effect(KILL cr);
7862 
7863   format %{ "incl    $dst\t# int" %}
7864   opcode(0xFF, 0x00); // FF /0
7865   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7866   ins_pipe(ialu_reg);
7867 %}
7868 
7869 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
7870 %{
7871   predicate(UseIncDec);
7872   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7873   effect(KILL cr);
7874 
7875   ins_cost(125); // XXX
7876   format %{ "incl    $dst\t# int" %}
7877   opcode(0xFF); /* Opcode FF /0 */
7878   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
7879   ins_pipe(ialu_mem_imm);
7880 %}
7881 
7882 // XXX why does that use AddI
7883 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
7884 %{
7885   predicate(UseIncDec);
7886   match(Set dst (AddI dst src));
7887   effect(KILL cr);
7888 
7889   format %{ "decl    $dst\t# int" %}
7890   opcode(0xFF, 0x01); // FF /1
7891   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
7892   ins_pipe(ialu_reg);
7893 %}
7894 
7895 // XXX why does that use AddI
7896 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
7897 %{
7898   predicate(UseIncDec);
7899   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7900   effect(KILL cr);
7901 
7902   ins_cost(125); // XXX
7903   format %{ "decl    $dst\t# int" %}
7904   opcode(0xFF); /* Opcode FF /1 */
7905   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
7906   ins_pipe(ialu_mem_imm);
7907 %}
7908 
7909 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
7910 %{
7911   match(Set dst (AddI src0 src1));
7912 
7913   ins_cost(110);
7914   format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
7915   opcode(0x8D); /* 0x8D /r */
7916   ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
7917   ins_pipe(ialu_reg_reg);
7918 %}
7919 
7920 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
7921 %{
7922   match(Set dst (AddL dst src));
7923   effect(KILL cr);
7924 
7925   format %{ "addq    $dst, $src\t# long" %}
7926   opcode(0x03);
7927   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
7928   ins_pipe(ialu_reg_reg);
7929 %}
7930 
7931 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
7932 %{
7933   match(Set dst (AddL dst src));
7934   effect(KILL cr);
7935 
7936   format %{ "addq    $dst, $src\t# long" %}
7937   opcode(0x81, 0x00); /* /0 id */
7938   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
7939   ins_pipe( ialu_reg );
7940 %}
7941 
7942 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
7943 %{
7944   match(Set dst (AddL dst (LoadL src)));
7945   effect(KILL cr);
7946 
7947   ins_cost(125); // XXX
7948   format %{ "addq    $dst, $src\t# long" %}
7949   opcode(0x03);
7950   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
7951   ins_pipe(ialu_reg_mem);
7952 %}
7953 
7954 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
7955 %{
7956   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7957   effect(KILL cr);
7958 
7959   ins_cost(150); // XXX
7960   format %{ "addq    $dst, $src\t# long" %}
7961   opcode(0x01); /* Opcode 01 /r */
7962   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7963   ins_pipe(ialu_mem_reg);
7964 %}
7965 
7966 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
7967 %{
7968   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7969   effect(KILL cr);
7970 
7971   ins_cost(125); // XXX
7972   format %{ "addq    $dst, $src\t# long" %}
7973   opcode(0x81); /* Opcode 81 /0 id */
7974   ins_encode(REX_mem_wide(dst),
7975              OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7976   ins_pipe(ialu_mem_imm);
7977 %}
7978 
7979 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
7980 %{
7981   predicate(UseIncDec);
7982   match(Set dst (AddL dst src));
7983   effect(KILL cr);
7984 
7985   format %{ "incq    $dst\t# long" %}
7986   opcode(0xFF, 0x00); // FF /0
7987   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
7988   ins_pipe(ialu_reg);
7989 %}
7990 
7991 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
7992 %{
7993   predicate(UseIncDec);
7994   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
7995   effect(KILL cr);
7996 
7997   ins_cost(125); // XXX
7998   format %{ "incq    $dst\t# long" %}
7999   opcode(0xFF); /* Opcode FF /0 */
8000   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
8001   ins_pipe(ialu_mem_imm);
8002 %}
8003 
8004 // XXX why does that use AddL
8005 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
8006 %{
8007   predicate(UseIncDec);
8008   match(Set dst (AddL dst src));
8009   effect(KILL cr);
8010 
8011   format %{ "decq    $dst\t# long" %}
8012   opcode(0xFF, 0x01); // FF /1
8013   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8014   ins_pipe(ialu_reg);
8015 %}
8016 
8017 // XXX why does that use AddL
8018 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
8019 %{
8020   predicate(UseIncDec);
8021   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8022   effect(KILL cr);
8023 
8024   ins_cost(125); // XXX
8025   format %{ "decq    $dst\t# long" %}
8026   opcode(0xFF); /* Opcode FF /1 */
8027   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
8028   ins_pipe(ialu_mem_imm);
8029 %}
8030 
8031 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
8032 %{
8033   match(Set dst (AddL src0 src1));
8034 
8035   ins_cost(110);
8036   format %{ "leaq    $dst, [$src0 + $src1]\t# long" %}
8037   opcode(0x8D); /* 0x8D /r */
8038   ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8039   ins_pipe(ialu_reg_reg);
8040 %}
8041 
8042 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
8043 %{
8044   match(Set dst (AddP dst src));
8045   effect(KILL cr);
8046 
8047   format %{ "addq    $dst, $src\t# ptr" %}
8048   opcode(0x03);
8049   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8050   ins_pipe(ialu_reg_reg);
8051 %}
8052 
8053 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
8054 %{
8055   match(Set dst (AddP dst src));
8056   effect(KILL cr);
8057 
8058   format %{ "addq    $dst, $src\t# ptr" %}
8059   opcode(0x81, 0x00); /* /0 id */
8060   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8061   ins_pipe( ialu_reg );
8062 %}
8063 
8064 // XXX addP mem ops ????
8065 
8066 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
8067 %{
8068   match(Set dst (AddP src0 src1));
8069 
8070   ins_cost(110);
8071   format %{ "leaq    $dst, [$src0 + $src1]\t# ptr" %}
8072   opcode(0x8D); /* 0x8D /r */
8073   ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
8074   ins_pipe(ialu_reg_reg);
8075 %}
8076 
8077 instruct checkCastPP(rRegP dst)
8078 %{
8079   match(Set dst (CheckCastPP dst));
8080 
8081   size(0);
8082   format %{ "# checkcastPP of $dst" %}
8083   ins_encode(/* empty encoding */);
8084   ins_pipe(empty);
8085 %}
8086 
8087 instruct castPP(rRegP dst)
8088 %{
8089   match(Set dst (CastPP dst));
8090 
8091   size(0);
8092   format %{ "# castPP of $dst" %}
8093   ins_encode(/* empty encoding */);
8094   ins_pipe(empty);
8095 %}
8096 
8097 instruct castII(rRegI dst)
8098 %{
8099   match(Set dst (CastII dst));
8100 
8101   size(0);
8102   format %{ "# castII of $dst" %}
8103   ins_encode(/* empty encoding */);
8104   ins_cost(0);
8105   ins_pipe(empty);
8106 %}
8107 
8108 // LoadP-locked same as a regular LoadP when used with compare-swap
8109 instruct loadPLocked(rRegP dst, memory mem)
8110 %{
8111   match(Set dst (LoadPLocked mem));
8112 
8113   ins_cost(125); // XXX
8114   format %{ "movq    $dst, $mem\t# ptr locked" %}
8115   opcode(0x8B);
8116   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8117   ins_pipe(ialu_reg_mem); // XXX
8118 %}
8119 
8120 // LoadL-locked - same as a regular LoadL when used with compare-swap
8121 instruct loadLLocked(rRegL dst, memory mem)
8122 %{
8123   match(Set dst (LoadLLocked mem));
8124 
8125   ins_cost(125); // XXX
8126   format %{ "movq    $dst, $mem\t# long locked" %}
8127   opcode(0x8B);
8128   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8129   ins_pipe(ialu_reg_mem); // XXX
8130 %}
8131 
8132 // Conditional-store of the updated heap-top.
8133 // Used during allocation of the shared heap.
8134 // Sets flags (EQ) on success.  Implemented with a CMPXCHG on Intel.
8135 
8136 instruct storePConditional(memory heap_top_ptr,
8137                            rax_RegP oldval, rRegP newval,
8138                            rFlagsReg cr)
8139 %{
8140   match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
8141  
8142   format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
8143             "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
8144   opcode(0x0F, 0xB1);
8145   ins_encode(lock_prefix,
8146              REX_reg_mem_wide(newval, heap_top_ptr),
8147              OpcP, OpcS,
8148              reg_mem(newval, heap_top_ptr));
8149   ins_pipe(pipe_cmpxchg);
8150 %}
8151 
8152 // Conditional-store of an int value.
8153 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG.
8154 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
8155 %{
8156   match(Set cr (StoreIConditional mem (Binary oldval newval)));
8157   effect(KILL oldval);
8158 
8159   format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8160   opcode(0x0F, 0xB1);
8161   ins_encode(lock_prefix,
8162              REX_reg_mem(newval, mem),
8163              OpcP, OpcS,
8164              reg_mem(newval, mem));
8165   ins_pipe(pipe_cmpxchg);
8166 %}
8167 
8168 // Conditional-store of a long value.
8169 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG.
8170 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
8171 %{
8172   match(Set cr (StoreLConditional mem (Binary oldval newval)));
8173   effect(KILL oldval);
8174 
8175   format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8176   opcode(0x0F, 0xB1);
8177   ins_encode(lock_prefix,
8178              REX_reg_mem_wide(newval, mem),
8179              OpcP, OpcS,
8180              reg_mem(newval, mem));
8181   ins_pipe(pipe_cmpxchg);
8182 %}
8183 
8184 
8185 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
8186 instruct compareAndSwapP(rRegI res,
8187                          memory mem_ptr,
8188                          rax_RegP oldval, rRegP newval,
8189                          rFlagsReg cr)
8190 %{
8191   match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
8192   effect(KILL cr, KILL oldval);
8193 
8194   format %{ "cmpxchgq $mem_ptr,$newval\t# "
8195             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8196             "sete    $res\n\t"
8197             "movzbl  $res, $res" %}
8198   opcode(0x0F, 0xB1);
8199   ins_encode(lock_prefix,
8200              REX_reg_mem_wide(newval, mem_ptr),
8201              OpcP, OpcS,
8202              reg_mem(newval, mem_ptr),
8203              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8204              REX_reg_breg(res, res), // movzbl
8205              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8206   ins_pipe( pipe_cmpxchg );
8207 %}
8208 
8209 instruct compareAndSwapL(rRegI res,
8210                          memory mem_ptr,
8211                          rax_RegL oldval, rRegL newval,
8212                          rFlagsReg cr)
8213 %{
8214   match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
8215   effect(KILL cr, KILL oldval);
8216 
8217   format %{ "cmpxchgq $mem_ptr,$newval\t# "
8218             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8219             "sete    $res\n\t"
8220             "movzbl  $res, $res" %}
8221   opcode(0x0F, 0xB1);
8222   ins_encode(lock_prefix,
8223              REX_reg_mem_wide(newval, mem_ptr),
8224              OpcP, OpcS,
8225              reg_mem(newval, mem_ptr),
8226              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8227              REX_reg_breg(res, res), // movzbl
8228              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8229   ins_pipe( pipe_cmpxchg );
8230 %}
8231 
8232 instruct compareAndSwapI(rRegI res,
8233                          memory mem_ptr,
8234                          rax_RegI oldval, rRegI newval,
8235                          rFlagsReg cr)
8236 %{
8237   match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
8238   effect(KILL cr, KILL oldval);
8239 
8240   format %{ "cmpxchgl $mem_ptr,$newval\t# "
8241             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8242             "sete    $res\n\t"
8243             "movzbl  $res, $res" %}
8244   opcode(0x0F, 0xB1);
8245   ins_encode(lock_prefix,
8246              REX_reg_mem(newval, mem_ptr),
8247              OpcP, OpcS,
8248              reg_mem(newval, mem_ptr),
8249              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8250              REX_reg_breg(res, res), // movzbl
8251              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8252   ins_pipe( pipe_cmpxchg );
8253 %}
8254 
8255 
8256 instruct compareAndSwapN(rRegI res,
8257                           memory mem_ptr,
8258                           rax_RegN oldval, rRegN newval,
8259                           rFlagsReg cr) %{
8260   match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
8261   effect(KILL cr, KILL oldval);
8262 
8263   format %{ "cmpxchgl $mem_ptr,$newval\t# "
8264             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8265             "sete    $res\n\t"
8266             "movzbl  $res, $res" %}
8267   opcode(0x0F, 0xB1);
8268   ins_encode(lock_prefix,
8269              REX_reg_mem(newval, mem_ptr),
8270              OpcP, OpcS,
8271              reg_mem(newval, mem_ptr),
8272              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8273              REX_reg_breg(res, res), // movzbl
8274              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8275   ins_pipe( pipe_cmpxchg );
8276 %}
8277 
8278 //----------Subtraction Instructions-------------------------------------------
8279 
8280 // Integer Subtraction Instructions
8281 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8282 %{
8283   match(Set dst (SubI dst src));
8284   effect(KILL cr);
8285 
8286   format %{ "subl    $dst, $src\t# int" %}
8287   opcode(0x2B);
8288   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8289   ins_pipe(ialu_reg_reg);
8290 %}
8291 
8292 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8293 %{
8294   match(Set dst (SubI dst src));
8295   effect(KILL cr);
8296 
8297   format %{ "subl    $dst, $src\t# int" %}
8298   opcode(0x81, 0x05);  /* Opcode 81 /5 */
8299   ins_encode(OpcSErm(dst, src), Con8or32(src));
8300   ins_pipe(ialu_reg);
8301 %}
8302 
8303 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8304 %{
8305   match(Set dst (SubI dst (LoadI src)));
8306   effect(KILL cr);
8307 
8308   ins_cost(125);
8309   format %{ "subl    $dst, $src\t# int" %}
8310   opcode(0x2B);
8311   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8312   ins_pipe(ialu_reg_mem);
8313 %}
8314 
8315 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8316 %{
8317   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8318   effect(KILL cr);
8319 
8320   ins_cost(150);
8321   format %{ "subl    $dst, $src\t# int" %}
8322   opcode(0x29); /* Opcode 29 /r */
8323   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8324   ins_pipe(ialu_mem_reg);
8325 %}
8326 
8327 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8328 %{
8329   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8330   effect(KILL cr);
8331 
8332   ins_cost(125); // XXX
8333   format %{ "subl    $dst, $src\t# int" %}
8334   opcode(0x81); /* Opcode 81 /5 id */
8335   ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8336   ins_pipe(ialu_mem_imm);
8337 %}
8338 
8339 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8340 %{
8341   match(Set dst (SubL dst src));
8342   effect(KILL cr);
8343 
8344   format %{ "subq    $dst, $src\t# long" %}
8345   opcode(0x2B);
8346   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8347   ins_pipe(ialu_reg_reg);
8348 %}
8349 
8350 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8351 %{
8352   match(Set dst (SubL dst src));
8353   effect(KILL cr);
8354 
8355   format %{ "subq    $dst, $src\t# long" %}
8356   opcode(0x81, 0x05);  /* Opcode 81 /5 */
8357   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8358   ins_pipe(ialu_reg);
8359 %}
8360 
8361 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8362 %{
8363   match(Set dst (SubL dst (LoadL src)));
8364   effect(KILL cr);
8365 
8366   ins_cost(125);
8367   format %{ "subq    $dst, $src\t# long" %}
8368   opcode(0x2B);
8369   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8370   ins_pipe(ialu_reg_mem);
8371 %}
8372 
8373 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8374 %{
8375   match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8376   effect(KILL cr);
8377 
8378   ins_cost(150);
8379   format %{ "subq    $dst, $src\t# long" %}
8380   opcode(0x29); /* Opcode 29 /r */
8381   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8382   ins_pipe(ialu_mem_reg);
8383 %}
8384 
8385 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8386 %{
8387   match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8388   effect(KILL cr);
8389 
8390   ins_cost(125); // XXX
8391   format %{ "subq    $dst, $src\t# long" %}
8392   opcode(0x81); /* Opcode 81 /5 id */
8393   ins_encode(REX_mem_wide(dst),
8394              OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8395   ins_pipe(ialu_mem_imm);
8396 %}
8397 
8398 // Subtract from a pointer
8399 // XXX hmpf???
8400 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8401 %{
8402   match(Set dst (AddP dst (SubI zero src)));
8403   effect(KILL cr);
8404 
8405   format %{ "subq    $dst, $src\t# ptr - int" %}
8406   opcode(0x2B);
8407   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8408   ins_pipe(ialu_reg_reg);
8409 %}
8410 
8411 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8412 %{
8413   match(Set dst (SubI zero dst));
8414   effect(KILL cr);
8415 
8416   format %{ "negl    $dst\t# int" %}
8417   opcode(0xF7, 0x03);  // Opcode F7 /3
8418   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8419   ins_pipe(ialu_reg);
8420 %}
8421 
8422 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8423 %{
8424   match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8425   effect(KILL cr);
8426 
8427   format %{ "negl    $dst\t# int" %}
8428   opcode(0xF7, 0x03);  // Opcode F7 /3
8429   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8430   ins_pipe(ialu_reg);
8431 %}
8432 
8433 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8434 %{
8435   match(Set dst (SubL zero dst));
8436   effect(KILL cr);
8437 
8438   format %{ "negq    $dst\t# long" %}
8439   opcode(0xF7, 0x03);  // Opcode F7 /3
8440   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8441   ins_pipe(ialu_reg);
8442 %}
8443 
8444 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8445 %{
8446   match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8447   effect(KILL cr);
8448 
8449   format %{ "negq    $dst\t# long" %}
8450   opcode(0xF7, 0x03);  // Opcode F7 /3
8451   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8452   ins_pipe(ialu_reg);
8453 %}
8454 
8455 
8456 //----------Multiplication/Division Instructions-------------------------------
8457 // Integer Multiplication Instructions
8458 // Multiply Register
8459 
8460 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8461 %{
8462   match(Set dst (MulI dst src));
8463   effect(KILL cr);
8464 
8465   ins_cost(300);
8466   format %{ "imull   $dst, $src\t# int" %}
8467   opcode(0x0F, 0xAF);
8468   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8469   ins_pipe(ialu_reg_reg_alu0);
8470 %}
8471 
8472 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8473 %{
8474   match(Set dst (MulI src imm));
8475   effect(KILL cr);
8476 
8477   ins_cost(300);
8478   format %{ "imull   $dst, $src, $imm\t# int" %}
8479   opcode(0x69); /* 69 /r id */
8480   ins_encode(REX_reg_reg(dst, src),
8481              OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8482   ins_pipe(ialu_reg_reg_alu0);
8483 %}
8484 
8485 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8486 %{
8487   match(Set dst (MulI dst (LoadI src)));
8488   effect(KILL cr);
8489 
8490   ins_cost(350);
8491   format %{ "imull   $dst, $src\t# int" %}
8492   opcode(0x0F, 0xAF);
8493   ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8494   ins_pipe(ialu_reg_mem_alu0);
8495 %}
8496 
8497 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8498 %{
8499   match(Set dst (MulI (LoadI src) imm));
8500   effect(KILL cr);
8501 
8502   ins_cost(300);
8503   format %{ "imull   $dst, $src, $imm\t# int" %}
8504   opcode(0x69); /* 69 /r id */
8505   ins_encode(REX_reg_mem(dst, src),
8506              OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8507   ins_pipe(ialu_reg_mem_alu0);
8508 %}
8509 
8510 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8511 %{
8512   match(Set dst (MulL dst src));
8513   effect(KILL cr);
8514 
8515   ins_cost(300);
8516   format %{ "imulq   $dst, $src\t# long" %}
8517   opcode(0x0F, 0xAF);
8518   ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8519   ins_pipe(ialu_reg_reg_alu0);
8520 %}
8521 
8522 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8523 %{
8524   match(Set dst (MulL src imm));
8525   effect(KILL cr);
8526 
8527   ins_cost(300);
8528   format %{ "imulq   $dst, $src, $imm\t# long" %}
8529   opcode(0x69); /* 69 /r id */
8530   ins_encode(REX_reg_reg_wide(dst, src),
8531              OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8532   ins_pipe(ialu_reg_reg_alu0);
8533 %}
8534 
8535 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8536 %{
8537   match(Set dst (MulL dst (LoadL src)));
8538   effect(KILL cr);
8539 
8540   ins_cost(350);
8541   format %{ "imulq   $dst, $src\t# long" %}
8542   opcode(0x0F, 0xAF);
8543   ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8544   ins_pipe(ialu_reg_mem_alu0);
8545 %}
8546 
8547 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8548 %{
8549   match(Set dst (MulL (LoadL src) imm));
8550   effect(KILL cr);
8551 
8552   ins_cost(300);
8553   format %{ "imulq   $dst, $src, $imm\t# long" %}
8554   opcode(0x69); /* 69 /r id */
8555   ins_encode(REX_reg_mem_wide(dst, src),
8556              OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8557   ins_pipe(ialu_reg_mem_alu0);
8558 %}
8559 
8560 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8561 %{
8562   match(Set dst (MulHiL src rax));
8563   effect(USE_KILL rax, KILL cr);
8564 
8565   ins_cost(300);
8566   format %{ "imulq   RDX:RAX, RAX, $src\t# mulhi" %}
8567   opcode(0xF7, 0x5); /* Opcode F7 /5 */
8568   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8569   ins_pipe(ialu_reg_reg_alu0);
8570 %}
8571 
8572 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8573                    rFlagsReg cr)
8574 %{
8575   match(Set rax (DivI rax div));
8576   effect(KILL rdx, KILL cr);
8577 
8578   ins_cost(30*100+10*100); // XXX
8579   format %{ "cmpl    rax, 0x80000000\t# idiv\n\t"
8580             "jne,s   normal\n\t"
8581             "xorl    rdx, rdx\n\t"
8582             "cmpl    $div, -1\n\t"
8583             "je,s    done\n"
8584     "normal: cdql\n\t"
8585             "idivl   $div\n"
8586     "done:"        %}
8587   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8588   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8589   ins_pipe(ialu_reg_reg_alu0);
8590 %}
8591 
8592 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8593                    rFlagsReg cr)
8594 %{
8595   match(Set rax (DivL rax div));
8596   effect(KILL rdx, KILL cr);
8597 
8598   ins_cost(30*100+10*100); // XXX
8599   format %{ "movq    rdx, 0x8000000000000000\t# ldiv\n\t"
8600             "cmpq    rax, rdx\n\t"
8601             "jne,s   normal\n\t"
8602             "xorl    rdx, rdx\n\t"
8603             "cmpq    $div, -1\n\t"
8604             "je,s    done\n"
8605     "normal: cdqq\n\t"
8606             "idivq   $div\n"
8607     "done:"        %}
8608   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8609   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8610   ins_pipe(ialu_reg_reg_alu0);
8611 %}
8612 
8613 // Integer DIVMOD with Register, both quotient and mod results
8614 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8615                              rFlagsReg cr)
8616 %{
8617   match(DivModI rax div);
8618   effect(KILL cr);
8619 
8620   ins_cost(30*100+10*100); // XXX
8621   format %{ "cmpl    rax, 0x80000000\t# idiv\n\t"
8622             "jne,s   normal\n\t"
8623             "xorl    rdx, rdx\n\t"
8624             "cmpl    $div, -1\n\t"
8625             "je,s    done\n"
8626     "normal: cdql\n\t"
8627             "idivl   $div\n"
8628     "done:"        %}
8629   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8630   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8631   ins_pipe(pipe_slow);
8632 %}
8633 
8634 // Long DIVMOD with Register, both quotient and mod results
8635 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8636                              rFlagsReg cr)
8637 %{
8638   match(DivModL rax div);
8639   effect(KILL cr);
8640 
8641   ins_cost(30*100+10*100); // XXX
8642   format %{ "movq    rdx, 0x8000000000000000\t# ldiv\n\t"
8643             "cmpq    rax, rdx\n\t"
8644             "jne,s   normal\n\t"
8645             "xorl    rdx, rdx\n\t"
8646             "cmpq    $div, -1\n\t"
8647             "je,s    done\n"
8648     "normal: cdqq\n\t"
8649             "idivq   $div\n"
8650     "done:"        %}
8651   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8652   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8653   ins_pipe(pipe_slow);
8654 %}
8655 
8656 //----------- DivL-By-Constant-Expansions--------------------------------------
8657 // DivI cases are handled by the compiler
8658 
8659 // Magic constant, reciprical of 10
8660 instruct loadConL_0x6666666666666667(rRegL dst)
8661 %{
8662   effect(DEF dst);
8663 
8664   format %{ "movq    $dst, #0x666666666666667\t# Used in div-by-10" %}
8665   ins_encode(load_immL(dst, 0x6666666666666667));
8666   ins_pipe(ialu_reg);
8667 %}
8668 
8669 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8670 %{
8671   effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8672 
8673   format %{ "imulq   rdx:rax, rax, $src\t# Used in div-by-10" %}
8674   opcode(0xF7, 0x5); /* Opcode F7 /5 */
8675   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8676   ins_pipe(ialu_reg_reg_alu0);
8677 %}
8678 
8679 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8680 %{
8681   effect(USE_DEF dst, KILL cr);
8682 
8683   format %{ "sarq    $dst, #63\t# Used in div-by-10" %}
8684   opcode(0xC1, 0x7); /* C1 /7 ib */
8685   ins_encode(reg_opc_imm_wide(dst, 0x3F));
8686   ins_pipe(ialu_reg);
8687 %}
8688 
8689 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8690 %{
8691   effect(USE_DEF dst, KILL cr);
8692 
8693   format %{ "sarq    $dst, #2\t# Used in div-by-10" %}
8694   opcode(0xC1, 0x7); /* C1 /7 ib */
8695   ins_encode(reg_opc_imm_wide(dst, 0x2));
8696   ins_pipe(ialu_reg);
8697 %}
8698 
8699 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8700 %{
8701   match(Set dst (DivL src div));
8702 
8703   ins_cost((5+8)*100);
8704   expand %{
8705     rax_RegL rax;                     // Killed temp
8706     rFlagsReg cr;                     // Killed
8707     loadConL_0x6666666666666667(rax); // movq  rax, 0x6666666666666667
8708     mul_hi(dst, src, rax, cr);        // mulq  rdx:rax <= rax * $src
8709     sarL_rReg_63(src, cr);            // sarq  src, 63
8710     sarL_rReg_2(dst, cr);             // sarq  rdx, 2
8711     subL_rReg(dst, src, cr);          // subl  rdx, src
8712   %}
8713 %}
8714 
8715 //-----------------------------------------------------------------------------
8716 
8717 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8718                    rFlagsReg cr)
8719 %{
8720   match(Set rdx (ModI rax div));
8721   effect(KILL rax, KILL cr);
8722 
8723   ins_cost(300); // XXX
8724   format %{ "cmpl    rax, 0x80000000\t# irem\n\t"
8725             "jne,s   normal\n\t"
8726             "xorl    rdx, rdx\n\t"
8727             "cmpl    $div, -1\n\t"
8728             "je,s    done\n"
8729     "normal: cdql\n\t"
8730             "idivl   $div\n"
8731     "done:"        %}
8732   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8733   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8734   ins_pipe(ialu_reg_reg_alu0);
8735 %}
8736 
8737 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8738                    rFlagsReg cr)
8739 %{
8740   match(Set rdx (ModL rax div));
8741   effect(KILL rax, KILL cr);
8742 
8743   ins_cost(300); // XXX
8744   format %{ "movq    rdx, 0x8000000000000000\t# lrem\n\t"
8745             "cmpq    rax, rdx\n\t"
8746             "jne,s   normal\n\t"
8747             "xorl    rdx, rdx\n\t"
8748             "cmpq    $div, -1\n\t"
8749             "je,s    done\n"
8750     "normal: cdqq\n\t"
8751             "idivq   $div\n"
8752     "done:"        %}
8753   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8754   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8755   ins_pipe(ialu_reg_reg_alu0);
8756 %}
8757 
8758 // Integer Shift Instructions
8759 // Shift Left by one
8760 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8761 %{
8762   match(Set dst (LShiftI dst shift));
8763   effect(KILL cr);
8764 
8765   format %{ "sall    $dst, $shift" %}
8766   opcode(0xD1, 0x4); /* D1 /4 */
8767   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8768   ins_pipe(ialu_reg);
8769 %}
8770 
8771 // Shift Left by one
8772 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8773 %{
8774   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8775   effect(KILL cr);
8776 
8777   format %{ "sall    $dst, $shift\t" %}
8778   opcode(0xD1, 0x4); /* D1 /4 */
8779   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8780   ins_pipe(ialu_mem_imm);
8781 %}
8782 
8783 // Shift Left by 8-bit immediate
8784 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8785 %{
8786   match(Set dst (LShiftI dst shift));
8787   effect(KILL cr);
8788 
8789   format %{ "sall    $dst, $shift" %}
8790   opcode(0xC1, 0x4); /* C1 /4 ib */
8791   ins_encode(reg_opc_imm(dst, shift));
8792   ins_pipe(ialu_reg);
8793 %}
8794 
8795 // Shift Left by 8-bit immediate
8796 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8797 %{
8798   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8799   effect(KILL cr);
8800 
8801   format %{ "sall    $dst, $shift" %}
8802   opcode(0xC1, 0x4); /* C1 /4 ib */
8803   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8804   ins_pipe(ialu_mem_imm);
8805 %}
8806 
8807 // Shift Left by variable
8808 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8809 %{
8810   match(Set dst (LShiftI dst shift));
8811   effect(KILL cr);
8812 
8813   format %{ "sall    $dst, $shift" %}
8814   opcode(0xD3, 0x4); /* D3 /4 */
8815   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8816   ins_pipe(ialu_reg_reg);
8817 %}
8818 
8819 // Shift Left by variable
8820 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8821 %{
8822   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8823   effect(KILL cr);
8824 
8825   format %{ "sall    $dst, $shift" %}
8826   opcode(0xD3, 0x4); /* D3 /4 */
8827   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8828   ins_pipe(ialu_mem_reg);
8829 %}
8830 
8831 // Arithmetic shift right by one
8832 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8833 %{
8834   match(Set dst (RShiftI dst shift));
8835   effect(KILL cr);
8836 
8837   format %{ "sarl    $dst, $shift" %}
8838   opcode(0xD1, 0x7); /* D1 /7 */
8839   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8840   ins_pipe(ialu_reg);
8841 %}
8842 
8843 // Arithmetic shift right by one
8844 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8845 %{
8846   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8847   effect(KILL cr);
8848 
8849   format %{ "sarl    $dst, $shift" %}
8850   opcode(0xD1, 0x7); /* D1 /7 */
8851   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8852   ins_pipe(ialu_mem_imm);
8853 %}
8854 
8855 // Arithmetic Shift Right by 8-bit immediate
8856 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8857 %{
8858   match(Set dst (RShiftI dst shift));
8859   effect(KILL cr);
8860 
8861   format %{ "sarl    $dst, $shift" %}
8862   opcode(0xC1, 0x7); /* C1 /7 ib */
8863   ins_encode(reg_opc_imm(dst, shift));
8864   ins_pipe(ialu_mem_imm);
8865 %}
8866 
8867 // Arithmetic Shift Right by 8-bit immediate
8868 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8869 %{
8870   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8871   effect(KILL cr);
8872 
8873   format %{ "sarl    $dst, $shift" %}
8874   opcode(0xC1, 0x7); /* C1 /7 ib */
8875   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8876   ins_pipe(ialu_mem_imm);
8877 %}
8878 
8879 // Arithmetic Shift Right by variable
8880 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8881 %{
8882   match(Set dst (RShiftI dst shift));
8883   effect(KILL cr);
8884 
8885   format %{ "sarl    $dst, $shift" %}
8886   opcode(0xD3, 0x7); /* D3 /7 */
8887   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8888   ins_pipe(ialu_reg_reg);
8889 %}
8890 
8891 // Arithmetic Shift Right by variable
8892 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8893 %{
8894   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8895   effect(KILL cr);
8896 
8897   format %{ "sarl    $dst, $shift" %}
8898   opcode(0xD3, 0x7); /* D3 /7 */
8899   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8900   ins_pipe(ialu_mem_reg);
8901 %}
8902 
8903 // Logical shift right by one
8904 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8905 %{
8906   match(Set dst (URShiftI dst shift));
8907   effect(KILL cr);
8908 
8909   format %{ "shrl    $dst, $shift" %}
8910   opcode(0xD1, 0x5); /* D1 /5 */
8911   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8912   ins_pipe(ialu_reg);
8913 %}
8914 
8915 // Logical shift right by one
8916 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8917 %{
8918   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8919   effect(KILL cr);
8920 
8921   format %{ "shrl    $dst, $shift" %}
8922   opcode(0xD1, 0x5); /* D1 /5 */
8923   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8924   ins_pipe(ialu_mem_imm);
8925 %}
8926 
8927 // Logical Shift Right by 8-bit immediate
8928 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8929 %{
8930   match(Set dst (URShiftI dst shift));
8931   effect(KILL cr);
8932 
8933   format %{ "shrl    $dst, $shift" %}
8934   opcode(0xC1, 0x5); /* C1 /5 ib */
8935   ins_encode(reg_opc_imm(dst, shift));
8936   ins_pipe(ialu_reg);
8937 %}
8938 
8939 // Logical Shift Right by 8-bit immediate
8940 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8941 %{
8942   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8943   effect(KILL cr);
8944 
8945   format %{ "shrl    $dst, $shift" %}
8946   opcode(0xC1, 0x5); /* C1 /5 ib */
8947   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8948   ins_pipe(ialu_mem_imm);
8949 %}
8950 
8951 // Logical Shift Right by variable
8952 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8953 %{
8954   match(Set dst (URShiftI dst shift));
8955   effect(KILL cr);
8956 
8957   format %{ "shrl    $dst, $shift" %}
8958   opcode(0xD3, 0x5); /* D3 /5 */
8959   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8960   ins_pipe(ialu_reg_reg);
8961 %}
8962 
8963 // Logical Shift Right by variable
8964 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8965 %{
8966   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
8967   effect(KILL cr);
8968 
8969   format %{ "shrl    $dst, $shift" %}
8970   opcode(0xD3, 0x5); /* D3 /5 */
8971   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8972   ins_pipe(ialu_mem_reg);
8973 %}
8974 
8975 // Long Shift Instructions
8976 // Shift Left by one
8977 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
8978 %{
8979   match(Set dst (LShiftL dst shift));
8980   effect(KILL cr);
8981 
8982   format %{ "salq    $dst, $shift" %}
8983   opcode(0xD1, 0x4); /* D1 /4 */
8984   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8985   ins_pipe(ialu_reg);
8986 %}
8987 
8988 // Shift Left by one
8989 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8990 %{
8991   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
8992   effect(KILL cr);
8993 
8994   format %{ "salq    $dst, $shift" %}
8995   opcode(0xD1, 0x4); /* D1 /4 */
8996   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8997   ins_pipe(ialu_mem_imm);
8998 %}
8999 
9000 // Shift Left by 8-bit immediate
9001 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9002 %{
9003   match(Set dst (LShiftL dst shift));
9004   effect(KILL cr);
9005 
9006   format %{ "salq    $dst, $shift" %}
9007   opcode(0xC1, 0x4); /* C1 /4 ib */
9008   ins_encode(reg_opc_imm_wide(dst, shift));
9009   ins_pipe(ialu_reg);
9010 %}
9011 
9012 // Shift Left by 8-bit immediate
9013 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9014 %{
9015   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9016   effect(KILL cr);
9017 
9018   format %{ "salq    $dst, $shift" %}
9019   opcode(0xC1, 0x4); /* C1 /4 ib */
9020   ins_encode(REX_mem_wide(dst), OpcP,
9021              RM_opc_mem(secondary, dst), Con8or32(shift));
9022   ins_pipe(ialu_mem_imm);
9023 %}
9024 
9025 // Shift Left by variable
9026 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9027 %{
9028   match(Set dst (LShiftL dst shift));
9029   effect(KILL cr);
9030 
9031   format %{ "salq    $dst, $shift" %}
9032   opcode(0xD3, 0x4); /* D3 /4 */
9033   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9034   ins_pipe(ialu_reg_reg);
9035 %}
9036 
9037 // Shift Left by variable
9038 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9039 %{
9040   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9041   effect(KILL cr);
9042 
9043   format %{ "salq    $dst, $shift" %}
9044   opcode(0xD3, 0x4); /* D3 /4 */
9045   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9046   ins_pipe(ialu_mem_reg);
9047 %}
9048 
9049 // Arithmetic shift right by one
9050 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9051 %{
9052   match(Set dst (RShiftL dst shift));
9053   effect(KILL cr);
9054 
9055   format %{ "sarq    $dst, $shift" %}
9056   opcode(0xD1, 0x7); /* D1 /7 */
9057   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9058   ins_pipe(ialu_reg);
9059 %}
9060 
9061 // Arithmetic shift right by one
9062 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9063 %{
9064   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9065   effect(KILL cr);
9066 
9067   format %{ "sarq    $dst, $shift" %}
9068   opcode(0xD1, 0x7); /* D1 /7 */
9069   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9070   ins_pipe(ialu_mem_imm);
9071 %}
9072 
9073 // Arithmetic Shift Right by 8-bit immediate
9074 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9075 %{
9076   match(Set dst (RShiftL dst shift));
9077   effect(KILL cr);
9078 
9079   format %{ "sarq    $dst, $shift" %}
9080   opcode(0xC1, 0x7); /* C1 /7 ib */
9081   ins_encode(reg_opc_imm_wide(dst, shift));
9082   ins_pipe(ialu_mem_imm);
9083 %}
9084 
9085 // Arithmetic Shift Right by 8-bit immediate
9086 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9087 %{
9088   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9089   effect(KILL cr);
9090 
9091   format %{ "sarq    $dst, $shift" %}
9092   opcode(0xC1, 0x7); /* C1 /7 ib */
9093   ins_encode(REX_mem_wide(dst), OpcP,
9094              RM_opc_mem(secondary, dst), Con8or32(shift));
9095   ins_pipe(ialu_mem_imm);
9096 %}
9097 
9098 // Arithmetic Shift Right by variable
9099 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9100 %{
9101   match(Set dst (RShiftL dst shift));
9102   effect(KILL cr);
9103 
9104   format %{ "sarq    $dst, $shift" %}
9105   opcode(0xD3, 0x7); /* D3 /7 */
9106   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9107   ins_pipe(ialu_reg_reg);
9108 %}
9109 
9110 // Arithmetic Shift Right by variable
9111 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9112 %{
9113   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9114   effect(KILL cr);
9115 
9116   format %{ "sarq    $dst, $shift" %}
9117   opcode(0xD3, 0x7); /* D3 /7 */
9118   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9119   ins_pipe(ialu_mem_reg);
9120 %}
9121 
9122 // Logical shift right by one
9123 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9124 %{
9125   match(Set dst (URShiftL dst shift));
9126   effect(KILL cr);
9127 
9128   format %{ "shrq    $dst, $shift" %}
9129   opcode(0xD1, 0x5); /* D1 /5 */
9130   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
9131   ins_pipe(ialu_reg);
9132 %}
9133 
9134 // Logical shift right by one
9135 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9136 %{
9137   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9138   effect(KILL cr);
9139 
9140   format %{ "shrq    $dst, $shift" %}
9141   opcode(0xD1, 0x5); /* D1 /5 */
9142   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9143   ins_pipe(ialu_mem_imm);
9144 %}
9145 
9146 // Logical Shift Right by 8-bit immediate
9147 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9148 %{
9149   match(Set dst (URShiftL dst shift));
9150   effect(KILL cr);
9151 
9152   format %{ "shrq    $dst, $shift" %}
9153   opcode(0xC1, 0x5); /* C1 /5 ib */
9154   ins_encode(reg_opc_imm_wide(dst, shift));
9155   ins_pipe(ialu_reg);
9156 %}
9157 
9158 
9159 // Logical Shift Right by 8-bit immediate
9160 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9161 %{
9162   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9163   effect(KILL cr);
9164 
9165   format %{ "shrq    $dst, $shift" %}
9166   opcode(0xC1, 0x5); /* C1 /5 ib */
9167   ins_encode(REX_mem_wide(dst), OpcP,
9168              RM_opc_mem(secondary, dst), Con8or32(shift));
9169   ins_pipe(ialu_mem_imm);
9170 %}
9171 
9172 // Logical Shift Right by variable
9173 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9174 %{
9175   match(Set dst (URShiftL dst shift));
9176   effect(KILL cr);
9177 
9178   format %{ "shrq    $dst, $shift" %}
9179   opcode(0xD3, 0x5); /* D3 /5 */
9180   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9181   ins_pipe(ialu_reg_reg);
9182 %}
9183 
9184 // Logical Shift Right by variable
9185 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9186 %{
9187   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9188   effect(KILL cr);
9189 
9190   format %{ "shrq    $dst, $shift" %}
9191   opcode(0xD3, 0x5); /* D3 /5 */
9192   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9193   ins_pipe(ialu_mem_reg);
9194 %}
9195 
9196 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
9197 // This idiom is used by the compiler for the i2b bytecode.
9198 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
9199 %{
9200   match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
9201 
9202   format %{ "movsbl  $dst, $src\t# i2b" %}
9203   opcode(0x0F, 0xBE);
9204   ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9205   ins_pipe(ialu_reg_reg);
9206 %}
9207 
9208 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
9209 // This idiom is used by the compiler the i2s bytecode.
9210 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
9211 %{
9212   match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
9213 
9214   format %{ "movswl  $dst, $src\t# i2s" %}
9215   opcode(0x0F, 0xBF);
9216   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9217   ins_pipe(ialu_reg_reg);
9218 %}
9219 
9220 // ROL/ROR instructions
9221 
9222 // ROL expand
9223 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
9224   effect(KILL cr, USE_DEF dst);
9225 
9226   format %{ "roll    $dst" %}
9227   opcode(0xD1, 0x0); /* Opcode  D1 /0 */
9228   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9229   ins_pipe(ialu_reg);
9230 %}
9231 
9232 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
9233   effect(USE_DEF dst, USE shift, KILL cr);
9234 
9235   format %{ "roll    $dst, $shift" %}
9236   opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9237   ins_encode( reg_opc_imm(dst, shift) );
9238   ins_pipe(ialu_reg);
9239 %}
9240 
9241 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9242 %{
9243   effect(USE_DEF dst, USE shift, KILL cr);
9244 
9245   format %{ "roll    $dst, $shift" %}
9246   opcode(0xD3, 0x0); /* Opcode D3 /0 */
9247   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9248   ins_pipe(ialu_reg_reg);
9249 %}
9250 // end of ROL expand
9251 
9252 // Rotate Left by one
9253 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9254 %{
9255   match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9256 
9257   expand %{
9258     rolI_rReg_imm1(dst, cr);
9259   %}
9260 %}
9261 
9262 // Rotate Left by 8-bit immediate
9263 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9264 %{
9265   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9266   match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9267 
9268   expand %{
9269     rolI_rReg_imm8(dst, lshift, cr);
9270   %}
9271 %}
9272 
9273 // Rotate Left by variable
9274 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9275 %{
9276   match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
9277 
9278   expand %{
9279     rolI_rReg_CL(dst, shift, cr);
9280   %}
9281 %}
9282 
9283 // Rotate Left by variable
9284 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9285 %{
9286   match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9287 
9288   expand %{
9289     rolI_rReg_CL(dst, shift, cr);
9290   %}
9291 %}
9292 
9293 // ROR expand
9294 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9295 %{
9296   effect(USE_DEF dst, KILL cr);
9297 
9298   format %{ "rorl    $dst" %}
9299   opcode(0xD1, 0x1); /* D1 /1 */
9300   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9301   ins_pipe(ialu_reg);
9302 %}
9303 
9304 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9305 %{
9306   effect(USE_DEF dst, USE shift, KILL cr);
9307 
9308   format %{ "rorl    $dst, $shift" %}
9309   opcode(0xC1, 0x1); /* C1 /1 ib */
9310   ins_encode(reg_opc_imm(dst, shift));
9311   ins_pipe(ialu_reg);
9312 %}
9313 
9314 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9315 %{
9316   effect(USE_DEF dst, USE shift, KILL cr);
9317 
9318   format %{ "rorl    $dst, $shift" %}
9319   opcode(0xD3, 0x1); /* D3 /1 */
9320   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9321   ins_pipe(ialu_reg_reg);
9322 %}
9323 // end of ROR expand
9324 
9325 // Rotate Right by one
9326 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9327 %{
9328   match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9329 
9330   expand %{
9331     rorI_rReg_imm1(dst, cr);
9332   %}
9333 %}
9334 
9335 // Rotate Right by 8-bit immediate
9336 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9337 %{
9338   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9339   match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9340 
9341   expand %{
9342     rorI_rReg_imm8(dst, rshift, cr);
9343   %}
9344 %}
9345 
9346 // Rotate Right by variable
9347 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9348 %{
9349   match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9350 
9351   expand %{
9352     rorI_rReg_CL(dst, shift, cr);
9353   %}
9354 %}
9355 
9356 // Rotate Right by variable
9357 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9358 %{
9359   match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9360 
9361   expand %{
9362     rorI_rReg_CL(dst, shift, cr);
9363   %}
9364 %}
9365 
9366 // for long rotate
9367 // ROL expand
9368 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9369   effect(USE_DEF dst, KILL cr);
9370 
9371   format %{ "rolq    $dst" %}
9372   opcode(0xD1, 0x0); /* Opcode  D1 /0 */
9373   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9374   ins_pipe(ialu_reg);
9375 %}
9376 
9377 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9378   effect(USE_DEF dst, USE shift, KILL cr);
9379 
9380   format %{ "rolq    $dst, $shift" %}
9381   opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9382   ins_encode( reg_opc_imm_wide(dst, shift) );
9383   ins_pipe(ialu_reg);
9384 %}
9385 
9386 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9387 %{
9388   effect(USE_DEF dst, USE shift, KILL cr);
9389 
9390   format %{ "rolq    $dst, $shift" %}
9391   opcode(0xD3, 0x0); /* Opcode D3 /0 */
9392   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9393   ins_pipe(ialu_reg_reg);
9394 %}
9395 // end of ROL expand
9396 
9397 // Rotate Left by one
9398 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9399 %{
9400   match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9401 
9402   expand %{
9403     rolL_rReg_imm1(dst, cr);
9404   %}
9405 %}
9406 
9407 // Rotate Left by 8-bit immediate
9408 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9409 %{
9410   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9411   match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9412 
9413   expand %{
9414     rolL_rReg_imm8(dst, lshift, cr);
9415   %}
9416 %}
9417 
9418 // Rotate Left by variable
9419 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9420 %{
9421   match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9422 
9423   expand %{
9424     rolL_rReg_CL(dst, shift, cr);
9425   %}
9426 %}
9427 
9428 // Rotate Left by variable
9429 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9430 %{
9431   match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9432 
9433   expand %{
9434     rolL_rReg_CL(dst, shift, cr);
9435   %}
9436 %}
9437 
9438 // ROR expand
9439 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9440 %{
9441   effect(USE_DEF dst, KILL cr);
9442 
9443   format %{ "rorq    $dst" %}
9444   opcode(0xD1, 0x1); /* D1 /1 */
9445   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9446   ins_pipe(ialu_reg);
9447 %}
9448 
9449 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9450 %{
9451   effect(USE_DEF dst, USE shift, KILL cr);
9452 
9453   format %{ "rorq    $dst, $shift" %}
9454   opcode(0xC1, 0x1); /* C1 /1 ib */
9455   ins_encode(reg_opc_imm_wide(dst, shift));
9456   ins_pipe(ialu_reg);
9457 %}
9458 
9459 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9460 %{
9461   effect(USE_DEF dst, USE shift, KILL cr);
9462 
9463   format %{ "rorq    $dst, $shift" %}
9464   opcode(0xD3, 0x1); /* D3 /1 */
9465   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9466   ins_pipe(ialu_reg_reg);
9467 %}
9468 // end of ROR expand
9469 
9470 // Rotate Right by one
9471 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9472 %{
9473   match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9474 
9475   expand %{
9476     rorL_rReg_imm1(dst, cr);
9477   %}
9478 %}
9479 
9480 // Rotate Right by 8-bit immediate
9481 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9482 %{
9483   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9484   match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9485 
9486   expand %{
9487     rorL_rReg_imm8(dst, rshift, cr);
9488   %}
9489 %}
9490 
9491 // Rotate Right by variable
9492 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9493 %{
9494   match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9495 
9496   expand %{
9497     rorL_rReg_CL(dst, shift, cr);
9498   %}
9499 %}
9500 
9501 // Rotate Right by variable
9502 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9503 %{
9504   match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9505 
9506   expand %{
9507     rorL_rReg_CL(dst, shift, cr);
9508   %}
9509 %}
9510 
9511 // Logical Instructions
9512 
9513 // Integer Logical Instructions
9514 
9515 // And Instructions
9516 // And Register with Register
9517 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9518 %{
9519   match(Set dst (AndI dst src));
9520   effect(KILL cr);
9521 
9522   format %{ "andl    $dst, $src\t# int" %}
9523   opcode(0x23);
9524   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9525   ins_pipe(ialu_reg_reg);
9526 %}
9527 
9528 // And Register with Immediate 255
9529 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9530 %{
9531   match(Set dst (AndI dst src));
9532 
9533   format %{ "movzbl  $dst, $dst\t# int & 0xFF" %}
9534   opcode(0x0F, 0xB6);
9535   ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9536   ins_pipe(ialu_reg);
9537 %}
9538 
9539 // And Register with Immediate 255 and promote to long
9540 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9541 %{
9542   match(Set dst (ConvI2L (AndI src mask)));
9543 
9544   format %{ "movzbl  $dst, $src\t# int & 0xFF -> long" %}
9545   opcode(0x0F, 0xB6);
9546   ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9547   ins_pipe(ialu_reg);
9548 %}
9549 
9550 // And Register with Immediate 65535
9551 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9552 %{
9553   match(Set dst (AndI dst src));
9554 
9555   format %{ "movzwl  $dst, $dst\t# int & 0xFFFF" %}
9556   opcode(0x0F, 0xB7);
9557   ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9558   ins_pipe(ialu_reg);
9559 %}
9560 
9561 // And Register with Immediate 65535 and promote to long
9562 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9563 %{
9564   match(Set dst (ConvI2L (AndI src mask)));
9565 
9566   format %{ "movzwl  $dst, $src\t# int & 0xFFFF -> long" %}
9567   opcode(0x0F, 0xB7);
9568   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9569   ins_pipe(ialu_reg);
9570 %}
9571 
9572 // And Register with Immediate
9573 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9574 %{
9575   match(Set dst (AndI dst src));
9576   effect(KILL cr);
9577 
9578   format %{ "andl    $dst, $src\t# int" %}
9579   opcode(0x81, 0x04); /* Opcode 81 /4 */
9580   ins_encode(OpcSErm(dst, src), Con8or32(src));
9581   ins_pipe(ialu_reg);
9582 %}
9583 
9584 // And Register with Memory
9585 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9586 %{
9587   match(Set dst (AndI dst (LoadI src)));
9588   effect(KILL cr);
9589 
9590   ins_cost(125);
9591   format %{ "andl    $dst, $src\t# int" %}
9592   opcode(0x23);
9593   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9594   ins_pipe(ialu_reg_mem);
9595 %}
9596 
9597 // And Memory with Register
9598 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9599 %{
9600   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9601   effect(KILL cr);
9602 
9603   ins_cost(150);
9604   format %{ "andl    $dst, $src\t# int" %}
9605   opcode(0x21); /* Opcode 21 /r */
9606   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9607   ins_pipe(ialu_mem_reg);
9608 %}
9609 
9610 // And Memory with Immediate
9611 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9612 %{
9613   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9614   effect(KILL cr);
9615 
9616   ins_cost(125);
9617   format %{ "andl    $dst, $src\t# int" %}
9618   opcode(0x81, 0x4); /* Opcode 81 /4 id */
9619   ins_encode(REX_mem(dst), OpcSE(src),
9620              RM_opc_mem(secondary, dst), Con8or32(src));
9621   ins_pipe(ialu_mem_imm);
9622 %}
9623 
9624 // Or Instructions
9625 // Or Register with Register
9626 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9627 %{
9628   match(Set dst (OrI dst src));
9629   effect(KILL cr);
9630 
9631   format %{ "orl     $dst, $src\t# int" %}
9632   opcode(0x0B);
9633   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9634   ins_pipe(ialu_reg_reg);
9635 %}
9636 
9637 // Or Register with Immediate
9638 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9639 %{
9640   match(Set dst (OrI dst src));
9641   effect(KILL cr);
9642 
9643   format %{ "orl     $dst, $src\t# int" %}
9644   opcode(0x81, 0x01); /* Opcode 81 /1 id */
9645   ins_encode(OpcSErm(dst, src), Con8or32(src));
9646   ins_pipe(ialu_reg);
9647 %}
9648 
9649 // Or Register with Memory
9650 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9651 %{
9652   match(Set dst (OrI dst (LoadI src)));
9653   effect(KILL cr);
9654 
9655   ins_cost(125);
9656   format %{ "orl     $dst, $src\t# int" %}
9657   opcode(0x0B);
9658   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9659   ins_pipe(ialu_reg_mem);
9660 %}
9661 
9662 // Or Memory with Register
9663 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9664 %{
9665   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9666   effect(KILL cr);
9667 
9668   ins_cost(150);
9669   format %{ "orl     $dst, $src\t# int" %}
9670   opcode(0x09); /* Opcode 09 /r */
9671   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9672   ins_pipe(ialu_mem_reg);
9673 %}
9674 
9675 // Or Memory with Immediate
9676 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9677 %{
9678   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9679   effect(KILL cr);
9680 
9681   ins_cost(125);
9682   format %{ "orl     $dst, $src\t# int" %}
9683   opcode(0x81, 0x1); /* Opcode 81 /1 id */
9684   ins_encode(REX_mem(dst), OpcSE(src),
9685              RM_opc_mem(secondary, dst), Con8or32(src));
9686   ins_pipe(ialu_mem_imm);
9687 %}
9688 
9689 // Xor Instructions
9690 // Xor Register with Register
9691 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9692 %{
9693   match(Set dst (XorI dst src));
9694   effect(KILL cr);
9695 
9696   format %{ "xorl    $dst, $src\t# int" %}
9697   opcode(0x33);
9698   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9699   ins_pipe(ialu_reg_reg);
9700 %}
9701 
9702 // Xor Register with Immediate -1
9703 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9704   match(Set dst (XorI dst imm));  
9705 
9706   format %{ "not    $dst" %}  
9707   ins_encode %{
9708      __ notl($dst$$Register);
9709   %}
9710   ins_pipe(ialu_reg);
9711 %}
9712 
9713 // Xor Register with Immediate
9714 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9715 %{
9716   match(Set dst (XorI dst src));
9717   effect(KILL cr);
9718 
9719   format %{ "xorl    $dst, $src\t# int" %}
9720   opcode(0x81, 0x06); /* Opcode 81 /6 id */
9721   ins_encode(OpcSErm(dst, src), Con8or32(src));
9722   ins_pipe(ialu_reg);
9723 %}
9724 
9725 // Xor Register with Memory
9726 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9727 %{
9728   match(Set dst (XorI dst (LoadI src)));
9729   effect(KILL cr);
9730 
9731   ins_cost(125);
9732   format %{ "xorl    $dst, $src\t# int" %}
9733   opcode(0x33);
9734   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9735   ins_pipe(ialu_reg_mem);
9736 %}
9737 
9738 // Xor Memory with Register
9739 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9740 %{
9741   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9742   effect(KILL cr);
9743 
9744   ins_cost(150);
9745   format %{ "xorl    $dst, $src\t# int" %}
9746   opcode(0x31); /* Opcode 31 /r */
9747   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9748   ins_pipe(ialu_mem_reg);
9749 %}
9750 
9751 // Xor Memory with Immediate
9752 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9753 %{
9754   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9755   effect(KILL cr);
9756 
9757   ins_cost(125);
9758   format %{ "xorl    $dst, $src\t# int" %}
9759   opcode(0x81, 0x6); /* Opcode 81 /6 id */
9760   ins_encode(REX_mem(dst), OpcSE(src),
9761              RM_opc_mem(secondary, dst), Con8or32(src));
9762   ins_pipe(ialu_mem_imm);
9763 %}
9764 
9765 
9766 // Long Logical Instructions
9767 
9768 // And Instructions
9769 // And Register with Register
9770 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9771 %{
9772   match(Set dst (AndL dst src));
9773   effect(KILL cr);
9774 
9775   format %{ "andq    $dst, $src\t# long" %}
9776   opcode(0x23);
9777   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9778   ins_pipe(ialu_reg_reg);
9779 %}
9780 
9781 // And Register with Immediate 255
9782 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9783 %{
9784   match(Set dst (AndL dst src));
9785 
9786   format %{ "movzbq  $dst, $src\t# long & 0xFF" %}
9787   opcode(0x0F, 0xB6);
9788   ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9789   ins_pipe(ialu_reg);
9790 %}
9791 
9792 // And Register with Immediate 65535
9793 instruct andL_rReg_imm65535(rRegI dst, immL_65535 src)
9794 %{
9795   match(Set dst (AndL dst src));
9796 
9797   format %{ "movzwq  $dst, $dst\t# long & 0xFFFF" %}
9798   opcode(0x0F, 0xB7);
9799   ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9800   ins_pipe(ialu_reg);
9801 %}
9802 
9803 // And Register with Immediate
9804 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9805 %{
9806   match(Set dst (AndL dst src));
9807   effect(KILL cr);
9808 
9809   format %{ "andq    $dst, $src\t# long" %}
9810   opcode(0x81, 0x04); /* Opcode 81 /4 */
9811   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9812   ins_pipe(ialu_reg);
9813 %}
9814 
9815 // And Register with Memory
9816 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9817 %{
9818   match(Set dst (AndL dst (LoadL src)));
9819   effect(KILL cr);
9820 
9821   ins_cost(125);
9822   format %{ "andq    $dst, $src\t# long" %}
9823   opcode(0x23);
9824   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9825   ins_pipe(ialu_reg_mem);
9826 %}
9827 
9828 // And Memory with Register
9829 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9830 %{
9831   match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9832   effect(KILL cr);
9833 
9834   ins_cost(150);
9835   format %{ "andq    $dst, $src\t# long" %}
9836   opcode(0x21); /* Opcode 21 /r */
9837   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9838   ins_pipe(ialu_mem_reg);
9839 %}
9840 
9841 // And Memory with Immediate
9842 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9843 %{
9844   match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9845   effect(KILL cr);
9846 
9847   ins_cost(125);
9848   format %{ "andq    $dst, $src\t# long" %}
9849   opcode(0x81, 0x4); /* Opcode 81 /4 id */
9850   ins_encode(REX_mem_wide(dst), OpcSE(src),
9851              RM_opc_mem(secondary, dst), Con8or32(src));
9852   ins_pipe(ialu_mem_imm);
9853 %}
9854 
9855 // Or Instructions
9856 // Or Register with Register
9857 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9858 %{
9859   match(Set dst (OrL dst src));
9860   effect(KILL cr);
9861 
9862   format %{ "orq     $dst, $src\t# long" %}
9863   opcode(0x0B);
9864   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9865   ins_pipe(ialu_reg_reg);
9866 %}
9867 
9868 // Use any_RegP to match R15 (TLS register) without spilling.
9869 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
9870   match(Set dst (OrL dst (CastP2X src)));
9871   effect(KILL cr);
9872 
9873   format %{ "orq     $dst, $src\t# long" %}
9874   opcode(0x0B);
9875   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9876   ins_pipe(ialu_reg_reg);
9877 %}
9878 
9879 
9880 // Or Register with Immediate
9881 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9882 %{
9883   match(Set dst (OrL dst src));
9884   effect(KILL cr);
9885 
9886   format %{ "orq     $dst, $src\t# long" %}
9887   opcode(0x81, 0x01); /* Opcode 81 /1 id */
9888   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9889   ins_pipe(ialu_reg);
9890 %}
9891 
9892 // Or Register with Memory
9893 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9894 %{
9895   match(Set dst (OrL dst (LoadL src)));
9896   effect(KILL cr);
9897 
9898   ins_cost(125);
9899   format %{ "orq     $dst, $src\t# long" %}
9900   opcode(0x0B);
9901   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9902   ins_pipe(ialu_reg_mem);
9903 %}
9904 
9905 // Or Memory with Register
9906 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9907 %{
9908   match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9909   effect(KILL cr);
9910 
9911   ins_cost(150);
9912   format %{ "orq     $dst, $src\t# long" %}
9913   opcode(0x09); /* Opcode 09 /r */
9914   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9915   ins_pipe(ialu_mem_reg);
9916 %}
9917 
9918 // Or Memory with Immediate
9919 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9920 %{
9921   match(Set dst (StoreL dst (OrL (LoadL dst) src)));
9922   effect(KILL cr);
9923 
9924   ins_cost(125);
9925   format %{ "orq     $dst, $src\t# long" %}
9926   opcode(0x81, 0x1); /* Opcode 81 /1 id */
9927   ins_encode(REX_mem_wide(dst), OpcSE(src),
9928              RM_opc_mem(secondary, dst), Con8or32(src));
9929   ins_pipe(ialu_mem_imm);
9930 %}
9931 
9932 // Xor Instructions
9933 // Xor Register with Register
9934 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9935 %{
9936   match(Set dst (XorL dst src));
9937   effect(KILL cr);
9938 
9939   format %{ "xorq    $dst, $src\t# long" %}
9940   opcode(0x33);
9941   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9942   ins_pipe(ialu_reg_reg);
9943 %}
9944 
9945 // Xor Register with Immediate -1
9946 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
9947   match(Set dst (XorL dst imm));  
9948 
9949   format %{ "notq   $dst" %}  
9950   ins_encode %{
9951      __ notq($dst$$Register);
9952   %}
9953   ins_pipe(ialu_reg);
9954 %}
9955 
9956 // Xor Register with Immediate
9957 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9958 %{
9959   match(Set dst (XorL dst src));
9960   effect(KILL cr);
9961 
9962   format %{ "xorq    $dst, $src\t# long" %}
9963   opcode(0x81, 0x06); /* Opcode 81 /6 id */
9964   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9965   ins_pipe(ialu_reg);
9966 %}
9967 
9968 // Xor Register with Memory
9969 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9970 %{
9971   match(Set dst (XorL dst (LoadL src)));
9972   effect(KILL cr);
9973 
9974   ins_cost(125);
9975   format %{ "xorq    $dst, $src\t# long" %}
9976   opcode(0x33);
9977   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9978   ins_pipe(ialu_reg_mem);
9979 %}
9980 
9981 // Xor Memory with Register
9982 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9983 %{
9984   match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9985   effect(KILL cr);
9986 
9987   ins_cost(150);
9988   format %{ "xorq    $dst, $src\t# long" %}
9989   opcode(0x31); /* Opcode 31 /r */
9990   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9991   ins_pipe(ialu_mem_reg);
9992 %}
9993 
9994 // Xor Memory with Immediate
9995 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9996 %{
9997   match(Set dst (StoreL dst (XorL (LoadL dst) src)));
9998   effect(KILL cr);
9999 
10000   ins_cost(125);
10001   format %{ "xorq    $dst, $src\t# long" %}
10002   opcode(0x81, 0x6); /* Opcode 81 /6 id */
10003   ins_encode(REX_mem_wide(dst), OpcSE(src),
10004              RM_opc_mem(secondary, dst), Con8or32(src));
10005   ins_pipe(ialu_mem_imm);
10006 %}
10007 
10008 // Convert Int to Boolean
10009 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
10010 %{
10011   match(Set dst (Conv2B src));
10012   effect(KILL cr);
10013 
10014   format %{ "testl   $src, $src\t# ci2b\n\t"
10015             "setnz   $dst\n\t"
10016             "movzbl  $dst, $dst" %}
10017   ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
10018              setNZ_reg(dst),
10019              REX_reg_breg(dst, dst), // movzbl
10020              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10021   ins_pipe(pipe_slow); // XXX
10022 %}
10023 
10024 // Convert Pointer to Boolean
10025 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
10026 %{
10027   match(Set dst (Conv2B src));
10028   effect(KILL cr);
10029 
10030   format %{ "testq   $src, $src\t# cp2b\n\t"
10031             "setnz   $dst\n\t"
10032             "movzbl  $dst, $dst" %}
10033   ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
10034              setNZ_reg(dst),
10035              REX_reg_breg(dst, dst), // movzbl
10036              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10037   ins_pipe(pipe_slow); // XXX
10038 %}
10039 
10040 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
10041 %{
10042   match(Set dst (CmpLTMask p q));
10043   effect(KILL cr);
10044 
10045   ins_cost(400); // XXX
10046   format %{ "cmpl    $p, $q\t# cmpLTMask\n\t"
10047             "setlt   $dst\n\t"
10048             "movzbl  $dst, $dst\n\t"
10049             "negl    $dst" %}
10050   ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
10051              setLT_reg(dst),
10052              REX_reg_breg(dst, dst), // movzbl
10053              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
10054              neg_reg(dst));
10055   ins_pipe(pipe_slow);
10056 %}
10057 
10058 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
10059 %{
10060   match(Set dst (CmpLTMask dst zero));
10061   effect(KILL cr);
10062 
10063   ins_cost(100); // XXX
10064   format %{ "sarl    $dst, #31\t# cmpLTMask0" %}
10065   opcode(0xC1, 0x7);  /* C1 /7 ib */
10066   ins_encode(reg_opc_imm(dst, 0x1F));
10067   ins_pipe(ialu_reg);
10068 %}
10069 
10070 
10071 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
10072                          rRegI tmp,
10073                          rFlagsReg cr)
10074 %{
10075   match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
10076   effect(TEMP tmp, KILL cr);
10077 
10078   ins_cost(400); // XXX
10079   format %{ "subl    $p, $q\t# cadd_cmpLTMask1\n\t"
10080             "sbbl    $tmp, $tmp\n\t"
10081             "andl    $tmp, $y\n\t"
10082             "addl    $p, $tmp" %}
10083   ins_encode(enc_cmpLTP(p, q, y, tmp));
10084   ins_pipe(pipe_cmplt);
10085 %}
10086 
10087 /* If I enable this, I encourage spilling in the inner loop of compress.
10088 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
10089 %{
10090   match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
10091   effect( TEMP tmp, KILL cr );
10092   ins_cost(400);
10093 
10094   format %{ "SUB    $p,$q\n\t"
10095             "SBB    RCX,RCX\n\t"
10096             "AND    RCX,$y\n\t"
10097             "ADD    $p,RCX" %}
10098   ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
10099 %}
10100 */
10101 
10102 //---------- FP Instructions------------------------------------------------
10103 
10104 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
10105 %{
10106   match(Set cr (CmpF src1 src2));
10107 
10108   ins_cost(145);
10109   format %{ "ucomiss $src1, $src2\n\t"
10110             "jnp,s   exit\n\t"
10111             "pushfq\t# saw NaN, set CF\n\t"
10112             "andq    [rsp], #0xffffff2b\n\t"
10113             "popfq\n"
10114     "exit:   nop\t# avoid branch to branch" %}
10115   opcode(0x0F, 0x2E);
10116   ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10117              cmpfp_fixup);
10118   ins_pipe(pipe_slow);
10119 %}
10120 
10121 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
10122   match(Set cr (CmpF src1 src2));
10123 
10124   ins_cost(145);
10125   format %{ "ucomiss $src1, $src2" %}
10126   ins_encode %{
10127     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10128   %}
10129   ins_pipe(pipe_slow);
10130 %}
10131 
10132 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
10133 %{
10134   match(Set cr (CmpF src1 (LoadF src2)));
10135 
10136   ins_cost(145);
10137   format %{ "ucomiss $src1, $src2\n\t"
10138             "jnp,s   exit\n\t"
10139             "pushfq\t# saw NaN, set CF\n\t"
10140             "andq    [rsp], #0xffffff2b\n\t"
10141             "popfq\n"
10142     "exit:   nop\t# avoid branch to branch" %}
10143   opcode(0x0F, 0x2E);
10144   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10145              cmpfp_fixup);
10146   ins_pipe(pipe_slow);
10147 %}
10148 
10149 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
10150   match(Set cr (CmpF src1 (LoadF src2)));
10151 
10152   ins_cost(100);
10153   format %{ "ucomiss $src1, $src2" %}
10154   opcode(0x0F, 0x2E);
10155   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
10156   ins_pipe(pipe_slow);
10157 %}
10158 
10159 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
10160 %{
10161   match(Set cr (CmpF src1 src2));
10162 
10163   ins_cost(145);
10164   format %{ "ucomiss $src1, $src2\n\t"
10165             "jnp,s   exit\n\t"
10166             "pushfq\t# saw NaN, set CF\n\t"
10167             "andq    [rsp], #0xffffff2b\n\t"
10168             "popfq\n"
10169     "exit:   nop\t# avoid branch to branch" %}
10170   opcode(0x0F, 0x2E);
10171   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10172              cmpfp_fixup);
10173   ins_pipe(pipe_slow);
10174 %}
10175 
10176 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{
10177   match(Set cr (CmpF src1 src2));
10178 
10179   ins_cost(100);
10180   format %{ "ucomiss $src1, $src2" %}
10181   opcode(0x0F, 0x2E);
10182   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2));
10183   ins_pipe(pipe_slow);
10184 %}
10185 
10186 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
10187 %{
10188   match(Set cr (CmpD src1 src2));
10189 
10190   ins_cost(145);
10191   format %{ "ucomisd $src1, $src2\n\t"
10192             "jnp,s   exit\n\t"
10193             "pushfq\t# saw NaN, set CF\n\t"
10194             "andq    [rsp], #0xffffff2b\n\t"
10195             "popfq\n"
10196     "exit:   nop\t# avoid branch to branch" %}
10197   opcode(0x66, 0x0F, 0x2E);
10198   ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10199              cmpfp_fixup);
10200   ins_pipe(pipe_slow);
10201 %}
10202 
10203 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
10204   match(Set cr (CmpD src1 src2));
10205 
10206   ins_cost(100);
10207   format %{ "ucomisd $src1, $src2 test" %}
10208   ins_encode %{
10209     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
10210   %}
10211   ins_pipe(pipe_slow);
10212 %}
10213 
10214 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
10215 %{
10216   match(Set cr (CmpD src1 (LoadD src2)));
10217 
10218   ins_cost(145);
10219   format %{ "ucomisd $src1, $src2\n\t"
10220             "jnp,s   exit\n\t"
10221             "pushfq\t# saw NaN, set CF\n\t"
10222             "andq    [rsp], #0xffffff2b\n\t"
10223             "popfq\n"
10224     "exit:   nop\t# avoid branch to branch" %}
10225   opcode(0x66, 0x0F, 0x2E);
10226   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10227              cmpfp_fixup);
10228   ins_pipe(pipe_slow);
10229 %}
10230 
10231 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
10232   match(Set cr (CmpD src1 (LoadD src2)));
10233 
10234   ins_cost(100);
10235   format %{ "ucomisd $src1, $src2" %}
10236   opcode(0x66, 0x0F, 0x2E);
10237   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
10238   ins_pipe(pipe_slow);
10239 %}
10240 
10241 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
10242 %{
10243   match(Set cr (CmpD src1 src2));
10244 
10245   ins_cost(145);
10246   format %{ "ucomisd $src1, [$src2]\n\t"
10247             "jnp,s   exit\n\t"
10248             "pushfq\t# saw NaN, set CF\n\t"
10249             "andq    [rsp], #0xffffff2b\n\t"
10250             "popfq\n"
10251     "exit:   nop\t# avoid branch to branch" %}
10252   opcode(0x66, 0x0F, 0x2E);
10253   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10254              cmpfp_fixup);
10255   ins_pipe(pipe_slow);
10256 %}
10257 
10258 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{
10259   match(Set cr (CmpD src1 src2));
10260 
10261   ins_cost(100);
10262   format %{ "ucomisd $src1, [$src2]" %}
10263   opcode(0x66, 0x0F, 0x2E);
10264   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2));
10265   ins_pipe(pipe_slow);
10266 %}
10267 
10268 // Compare into -1,0,1
10269 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
10270 %{
10271   match(Set dst (CmpF3 src1 src2));
10272   effect(KILL cr);
10273 
10274   ins_cost(275);
10275   format %{ "ucomiss $src1, $src2\n\t"
10276             "movl    $dst, #-1\n\t"
10277             "jp,s    done\n\t"
10278             "jb,s    done\n\t"
10279             "setne   $dst\n\t"
10280             "movzbl  $dst, $dst\n"
10281     "done:" %}
10282 
10283   opcode(0x0F, 0x2E);
10284   ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10285              cmpfp3(dst));
10286   ins_pipe(pipe_slow);
10287 %}
10288 
10289 // Compare into -1,0,1
10290 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
10291 %{
10292   match(Set dst (CmpF3 src1 (LoadF src2)));
10293   effect(KILL cr);
10294 
10295   ins_cost(275);
10296   format %{ "ucomiss $src1, $src2\n\t"
10297             "movl    $dst, #-1\n\t"
10298             "jp,s    done\n\t"
10299             "jb,s    done\n\t"
10300             "setne   $dst\n\t"
10301             "movzbl  $dst, $dst\n"
10302     "done:" %}
10303 
10304   opcode(0x0F, 0x2E);
10305   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10306              cmpfp3(dst));
10307   ins_pipe(pipe_slow);
10308 %}
10309 
10310 // Compare into -1,0,1
10311 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
10312 %{
10313   match(Set dst (CmpF3 src1 src2));
10314   effect(KILL cr);
10315 
10316   ins_cost(275);
10317   format %{ "ucomiss $src1, [$src2]\n\t"
10318             "movl    $dst, #-1\n\t"
10319             "jp,s    done\n\t"
10320             "jb,s    done\n\t"
10321             "setne   $dst\n\t"
10322             "movzbl  $dst, $dst\n"
10323     "done:" %}
10324 
10325   opcode(0x0F, 0x2E);
10326   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10327              cmpfp3(dst));
10328   ins_pipe(pipe_slow);
10329 %}
10330 
10331 // Compare into -1,0,1
10332 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10333 %{
10334   match(Set dst (CmpD3 src1 src2));
10335   effect(KILL cr);
10336 
10337   ins_cost(275);
10338   format %{ "ucomisd $src1, $src2\n\t"
10339             "movl    $dst, #-1\n\t"
10340             "jp,s    done\n\t"
10341             "jb,s    done\n\t"
10342             "setne   $dst\n\t"
10343             "movzbl  $dst, $dst\n"
10344     "done:" %}
10345 
10346   opcode(0x66, 0x0F, 0x2E);
10347   ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10348              cmpfp3(dst));
10349   ins_pipe(pipe_slow);
10350 %}
10351 
10352 // Compare into -1,0,1
10353 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10354 %{
10355   match(Set dst (CmpD3 src1 (LoadD src2)));
10356   effect(KILL cr);
10357 
10358   ins_cost(275);
10359   format %{ "ucomisd $src1, $src2\n\t"
10360             "movl    $dst, #-1\n\t"
10361             "jp,s    done\n\t"
10362             "jb,s    done\n\t"
10363             "setne   $dst\n\t"
10364             "movzbl  $dst, $dst\n"
10365     "done:" %}
10366 
10367   opcode(0x66, 0x0F, 0x2E);
10368   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10369              cmpfp3(dst));
10370   ins_pipe(pipe_slow);
10371 %}
10372 
10373 // Compare into -1,0,1
10374 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
10375 %{
10376   match(Set dst (CmpD3 src1 src2));
10377   effect(KILL cr);
10378 
10379   ins_cost(275);
10380   format %{ "ucomisd $src1, [$src2]\n\t"
10381             "movl    $dst, #-1\n\t"
10382             "jp,s    done\n\t"
10383             "jb,s    done\n\t"
10384             "setne   $dst\n\t"
10385             "movzbl  $dst, $dst\n"
10386     "done:" %}
10387 
10388   opcode(0x66, 0x0F, 0x2E);
10389   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10390              cmpfp3(dst));
10391   ins_pipe(pipe_slow);
10392 %}
10393 
10394 instruct addF_reg(regF dst, regF src)
10395 %{
10396   match(Set dst (AddF dst src));
10397 
10398   format %{ "addss   $dst, $src" %}
10399   ins_cost(150); // XXX
10400   opcode(0xF3, 0x0F, 0x58);
10401   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10402   ins_pipe(pipe_slow);
10403 %}
10404 
10405 instruct addF_mem(regF dst, memory src)
10406 %{
10407   match(Set dst (AddF dst (LoadF src)));
10408 
10409   format %{ "addss   $dst, $src" %}
10410   ins_cost(150); // XXX
10411   opcode(0xF3, 0x0F, 0x58);
10412   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10413   ins_pipe(pipe_slow);
10414 %}
10415 
10416 instruct addF_imm(regF dst, immF src)
10417 %{
10418   match(Set dst (AddF dst src));
10419 
10420   format %{ "addss   $dst, [$src]" %}
10421   ins_cost(150); // XXX
10422   opcode(0xF3, 0x0F, 0x58);
10423   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10424   ins_pipe(pipe_slow);
10425 %}
10426 
10427 instruct addD_reg(regD dst, regD src)
10428 %{
10429   match(Set dst (AddD dst src));
10430 
10431   format %{ "addsd   $dst, $src" %}
10432   ins_cost(150); // XXX
10433   opcode(0xF2, 0x0F, 0x58);
10434   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10435   ins_pipe(pipe_slow);
10436 %}
10437 
10438 instruct addD_mem(regD dst, memory src)
10439 %{
10440   match(Set dst (AddD dst (LoadD src)));
10441 
10442   format %{ "addsd   $dst, $src" %}
10443   ins_cost(150); // XXX
10444   opcode(0xF2, 0x0F, 0x58);
10445   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10446   ins_pipe(pipe_slow);
10447 %}
10448 
10449 instruct addD_imm(regD dst, immD src)
10450 %{
10451   match(Set dst (AddD dst src));
10452 
10453   format %{ "addsd   $dst, [$src]" %}
10454   ins_cost(150); // XXX
10455   opcode(0xF2, 0x0F, 0x58);
10456   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10457   ins_pipe(pipe_slow);
10458 %}
10459 
10460 instruct subF_reg(regF dst, regF src)
10461 %{
10462   match(Set dst (SubF dst src));
10463 
10464   format %{ "subss   $dst, $src" %}
10465   ins_cost(150); // XXX
10466   opcode(0xF3, 0x0F, 0x5C);
10467   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10468   ins_pipe(pipe_slow);
10469 %}
10470 
10471 instruct subF_mem(regF dst, memory src)
10472 %{
10473   match(Set dst (SubF dst (LoadF src)));
10474 
10475   format %{ "subss   $dst, $src" %}
10476   ins_cost(150); // XXX
10477   opcode(0xF3, 0x0F, 0x5C);
10478   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10479   ins_pipe(pipe_slow);
10480 %}
10481 
10482 instruct subF_imm(regF dst, immF src)
10483 %{
10484   match(Set dst (SubF dst src));
10485 
10486   format %{ "subss   $dst, [$src]" %}
10487   ins_cost(150); // XXX
10488   opcode(0xF3, 0x0F, 0x5C);
10489   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10490   ins_pipe(pipe_slow);
10491 %}
10492 
10493 instruct subD_reg(regD dst, regD src)
10494 %{
10495   match(Set dst (SubD dst src));
10496 
10497   format %{ "subsd   $dst, $src" %}
10498   ins_cost(150); // XXX
10499   opcode(0xF2, 0x0F, 0x5C);
10500   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10501   ins_pipe(pipe_slow);
10502 %}
10503 
10504 instruct subD_mem(regD dst, memory src)
10505 %{
10506   match(Set dst (SubD dst (LoadD src)));
10507 
10508   format %{ "subsd   $dst, $src" %}
10509   ins_cost(150); // XXX
10510   opcode(0xF2, 0x0F, 0x5C);
10511   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10512   ins_pipe(pipe_slow);
10513 %}
10514 
10515 instruct subD_imm(regD dst, immD src)
10516 %{
10517   match(Set dst (SubD dst src));
10518 
10519   format %{ "subsd   $dst, [$src]" %}
10520   ins_cost(150); // XXX
10521   opcode(0xF2, 0x0F, 0x5C);
10522   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10523   ins_pipe(pipe_slow);
10524 %}
10525 
10526 instruct mulF_reg(regF dst, regF src)
10527 %{
10528   match(Set dst (MulF dst src));
10529 
10530   format %{ "mulss   $dst, $src" %}
10531   ins_cost(150); // XXX
10532   opcode(0xF3, 0x0F, 0x59);
10533   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10534   ins_pipe(pipe_slow);
10535 %}
10536 
10537 instruct mulF_mem(regF dst, memory src)
10538 %{
10539   match(Set dst (MulF dst (LoadF src)));
10540 
10541   format %{ "mulss   $dst, $src" %}
10542   ins_cost(150); // XXX
10543   opcode(0xF3, 0x0F, 0x59);
10544   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10545   ins_pipe(pipe_slow);
10546 %}
10547 
10548 instruct mulF_imm(regF dst, immF src)
10549 %{
10550   match(Set dst (MulF dst src));
10551 
10552   format %{ "mulss   $dst, [$src]" %}
10553   ins_cost(150); // XXX
10554   opcode(0xF3, 0x0F, 0x59);
10555   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10556   ins_pipe(pipe_slow);
10557 %}
10558 
10559 instruct mulD_reg(regD dst, regD src)
10560 %{
10561   match(Set dst (MulD dst src));
10562 
10563   format %{ "mulsd   $dst, $src" %}
10564   ins_cost(150); // XXX
10565   opcode(0xF2, 0x0F, 0x59);
10566   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10567   ins_pipe(pipe_slow);
10568 %}
10569 
10570 instruct mulD_mem(regD dst, memory src)
10571 %{
10572   match(Set dst (MulD dst (LoadD src)));
10573 
10574   format %{ "mulsd   $dst, $src" %}
10575   ins_cost(150); // XXX
10576   opcode(0xF2, 0x0F, 0x59);
10577   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10578   ins_pipe(pipe_slow);
10579 %}
10580 
10581 instruct mulD_imm(regD dst, immD src)
10582 %{
10583   match(Set dst (MulD dst src));
10584 
10585   format %{ "mulsd   $dst, [$src]" %}
10586   ins_cost(150); // XXX
10587   opcode(0xF2, 0x0F, 0x59);
10588   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10589   ins_pipe(pipe_slow);
10590 %}
10591 
10592 instruct divF_reg(regF dst, regF src)
10593 %{
10594   match(Set dst (DivF dst src));
10595 
10596   format %{ "divss   $dst, $src" %}
10597   ins_cost(150); // XXX
10598   opcode(0xF3, 0x0F, 0x5E);
10599   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10600   ins_pipe(pipe_slow);
10601 %}
10602 
10603 instruct divF_mem(regF dst, memory src)
10604 %{
10605   match(Set dst (DivF dst (LoadF src)));
10606 
10607   format %{ "divss   $dst, $src" %}
10608   ins_cost(150); // XXX
10609   opcode(0xF3, 0x0F, 0x5E);
10610   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10611   ins_pipe(pipe_slow);
10612 %}
10613 
10614 instruct divF_imm(regF dst, immF src)
10615 %{
10616   match(Set dst (DivF dst src));
10617 
10618   format %{ "divss   $dst, [$src]" %}
10619   ins_cost(150); // XXX
10620   opcode(0xF3, 0x0F, 0x5E);
10621   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10622   ins_pipe(pipe_slow);
10623 %}
10624 
10625 instruct divD_reg(regD dst, regD src)
10626 %{
10627   match(Set dst (DivD dst src));
10628 
10629   format %{ "divsd   $dst, $src" %}
10630   ins_cost(150); // XXX
10631   opcode(0xF2, 0x0F, 0x5E);
10632   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10633   ins_pipe(pipe_slow);
10634 %}
10635 
10636 instruct divD_mem(regD dst, memory src)
10637 %{
10638   match(Set dst (DivD dst (LoadD src)));
10639 
10640   format %{ "divsd   $dst, $src" %}
10641   ins_cost(150); // XXX
10642   opcode(0xF2, 0x0F, 0x5E);
10643   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10644   ins_pipe(pipe_slow);
10645 %}
10646 
10647 instruct divD_imm(regD dst, immD src)
10648 %{
10649   match(Set dst (DivD dst src));
10650 
10651   format %{ "divsd   $dst, [$src]" %}
10652   ins_cost(150); // XXX
10653   opcode(0xF2, 0x0F, 0x5E);
10654   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10655   ins_pipe(pipe_slow);
10656 %}
10657 
10658 instruct sqrtF_reg(regF dst, regF src)
10659 %{
10660   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10661 
10662   format %{ "sqrtss  $dst, $src" %}
10663   ins_cost(150); // XXX
10664   opcode(0xF3, 0x0F, 0x51);
10665   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10666   ins_pipe(pipe_slow);
10667 %}
10668 
10669 instruct sqrtF_mem(regF dst, memory src)
10670 %{
10671   match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10672 
10673   format %{ "sqrtss  $dst, $src" %}
10674   ins_cost(150); // XXX
10675   opcode(0xF3, 0x0F, 0x51);
10676   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10677   ins_pipe(pipe_slow);
10678 %}
10679 
10680 instruct sqrtF_imm(regF dst, immF src)
10681 %{
10682   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10683 
10684   format %{ "sqrtss  $dst, [$src]" %}
10685   ins_cost(150); // XXX
10686   opcode(0xF3, 0x0F, 0x51);
10687   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10688   ins_pipe(pipe_slow);
10689 %}
10690 
10691 instruct sqrtD_reg(regD dst, regD src)
10692 %{
10693   match(Set dst (SqrtD src));
10694 
10695   format %{ "sqrtsd  $dst, $src" %}
10696   ins_cost(150); // XXX
10697   opcode(0xF2, 0x0F, 0x51);
10698   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10699   ins_pipe(pipe_slow);
10700 %}
10701 
10702 instruct sqrtD_mem(regD dst, memory src)
10703 %{
10704   match(Set dst (SqrtD (LoadD src)));
10705 
10706   format %{ "sqrtsd  $dst, $src" %}
10707   ins_cost(150); // XXX
10708   opcode(0xF2, 0x0F, 0x51);
10709   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10710   ins_pipe(pipe_slow);
10711 %}
10712 
10713 instruct sqrtD_imm(regD dst, immD src)
10714 %{
10715   match(Set dst (SqrtD src));
10716 
10717   format %{ "sqrtsd  $dst, [$src]" %}
10718   ins_cost(150); // XXX
10719   opcode(0xF2, 0x0F, 0x51);
10720   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10721   ins_pipe(pipe_slow);
10722 %}
10723 
10724 instruct absF_reg(regF dst)
10725 %{
10726   match(Set dst (AbsF dst));
10727 
10728   format %{ "andps   $dst, [0x7fffffff]\t# abs float by sign masking" %}
10729   ins_encode(absF_encoding(dst));
10730   ins_pipe(pipe_slow);
10731 %}
10732 
10733 instruct absD_reg(regD dst)
10734 %{
10735   match(Set dst (AbsD dst));
10736 
10737   format %{ "andpd   $dst, [0x7fffffffffffffff]\t"
10738             "# abs double by sign masking" %}
10739   ins_encode(absD_encoding(dst));
10740   ins_pipe(pipe_slow);
10741 %}
10742 
10743 instruct negF_reg(regF dst)
10744 %{
10745   match(Set dst (NegF dst));
10746 
10747   format %{ "xorps   $dst, [0x80000000]\t# neg float by sign flipping" %}
10748   ins_encode(negF_encoding(dst));
10749   ins_pipe(pipe_slow);
10750 %}
10751 
10752 instruct negD_reg(regD dst)
10753 %{
10754   match(Set dst (NegD dst));
10755 
10756   format %{ "xorpd   $dst, [0x8000000000000000]\t"
10757             "# neg double by sign flipping" %}
10758   ins_encode(negD_encoding(dst));
10759   ins_pipe(pipe_slow);
10760 %}
10761 
10762 // -----------Trig and Trancendental Instructions------------------------------
10763 instruct cosD_reg(regD dst) %{
10764   match(Set dst (CosD dst));
10765 
10766   format %{ "dcos   $dst\n\t" %}
10767   opcode(0xD9, 0xFF);
10768   ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10769   ins_pipe( pipe_slow );
10770 %}
10771 
10772 instruct sinD_reg(regD dst) %{
10773   match(Set dst (SinD dst));
10774 
10775   format %{ "dsin   $dst\n\t" %}
10776   opcode(0xD9, 0xFE);
10777   ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10778   ins_pipe( pipe_slow );
10779 %}
10780 
10781 instruct tanD_reg(regD dst) %{
10782   match(Set dst (TanD dst));
10783 
10784   format %{ "dtan   $dst\n\t" %}
10785   ins_encode( Push_SrcXD(dst),
10786               Opcode(0xD9), Opcode(0xF2),   //fptan
10787               Opcode(0xDD), Opcode(0xD8),   //fstp st
10788               Push_ResultXD(dst) );
10789   ins_pipe( pipe_slow );
10790 %}
10791 
10792 instruct log10D_reg(regD dst) %{
10793   // The source and result Double operands in XMM registers
10794   match(Set dst (Log10D dst));
10795   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
10796   // fyl2x        ; compute log_10(2) * log_2(x)
10797   format %{ "fldlg2\t\t\t#Log10\n\t"
10798             "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10799          %}
10800    ins_encode(Opcode(0xD9), Opcode(0xEC),   // fldlg2
10801               Push_SrcXD(dst),
10802               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10803               Push_ResultXD(dst));
10804 
10805   ins_pipe( pipe_slow );
10806 %}
10807 
10808 instruct logD_reg(regD dst) %{
10809   // The source and result Double operands in XMM registers
10810   match(Set dst (LogD dst));
10811   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
10812   // fyl2x        ; compute log_e(2) * log_2(x)
10813   format %{ "fldln2\t\t\t#Log_e\n\t"
10814             "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10815          %}
10816   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
10817               Push_SrcXD(dst),
10818               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10819               Push_ResultXD(dst));
10820   ins_pipe( pipe_slow );
10821 %}
10822 
10823 
10824 
10825 //----------Arithmetic Conversion Instructions---------------------------------
10826 
10827 instruct roundFloat_nop(regF dst)
10828 %{
10829   match(Set dst (RoundFloat dst));
10830 
10831   ins_cost(0);
10832   ins_encode();
10833   ins_pipe(empty);
10834 %}
10835 
10836 instruct roundDouble_nop(regD dst)
10837 %{
10838   match(Set dst (RoundDouble dst));
10839 
10840   ins_cost(0);
10841   ins_encode();
10842   ins_pipe(empty);
10843 %}
10844 
10845 instruct convF2D_reg_reg(regD dst, regF src)
10846 %{
10847   match(Set dst (ConvF2D src));
10848 
10849   format %{ "cvtss2sd $dst, $src" %}
10850   opcode(0xF3, 0x0F, 0x5A);
10851   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10852   ins_pipe(pipe_slow); // XXX
10853 %}
10854 
10855 instruct convF2D_reg_mem(regD dst, memory src)
10856 %{
10857   match(Set dst (ConvF2D (LoadF src)));
10858 
10859   format %{ "cvtss2sd $dst, $src" %}
10860   opcode(0xF3, 0x0F, 0x5A);
10861   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10862   ins_pipe(pipe_slow); // XXX
10863 %}
10864 
10865 instruct convD2F_reg_reg(regF dst, regD src)
10866 %{
10867   match(Set dst (ConvD2F src));
10868 
10869   format %{ "cvtsd2ss $dst, $src" %}
10870   opcode(0xF2, 0x0F, 0x5A);
10871   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10872   ins_pipe(pipe_slow); // XXX
10873 %}
10874 
10875 instruct convD2F_reg_mem(regF dst, memory src)
10876 %{
10877   match(Set dst (ConvD2F (LoadD src)));
10878 
10879   format %{ "cvtsd2ss $dst, $src" %}
10880   opcode(0xF2, 0x0F, 0x5A);
10881   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10882   ins_pipe(pipe_slow); // XXX
10883 %}
10884 
10885 // XXX do mem variants
10886 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
10887 %{
10888   match(Set dst (ConvF2I src));
10889   effect(KILL cr);
10890 
10891   format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
10892             "cmpl    $dst, #0x80000000\n\t"
10893             "jne,s   done\n\t"
10894             "subq    rsp, #8\n\t"
10895             "movss   [rsp], $src\n\t"
10896             "call    f2i_fixup\n\t"
10897             "popq    $dst\n"
10898     "done:   "%}
10899   opcode(0xF3, 0x0F, 0x2C);
10900   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10901              f2i_fixup(dst, src));
10902   ins_pipe(pipe_slow);
10903 %}
10904 
10905 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
10906 %{
10907   match(Set dst (ConvF2L src));
10908   effect(KILL cr);
10909 
10910   format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
10911             "cmpq    $dst, [0x8000000000000000]\n\t"
10912             "jne,s   done\n\t"
10913             "subq    rsp, #8\n\t"
10914             "movss   [rsp], $src\n\t"
10915             "call    f2l_fixup\n\t"
10916             "popq    $dst\n"
10917     "done:   "%}
10918   opcode(0xF3, 0x0F, 0x2C);
10919   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10920              f2l_fixup(dst, src));
10921   ins_pipe(pipe_slow);
10922 %}
10923 
10924 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
10925 %{
10926   match(Set dst (ConvD2I src));
10927   effect(KILL cr);
10928 
10929   format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
10930             "cmpl    $dst, #0x80000000\n\t"
10931             "jne,s   done\n\t"
10932             "subq    rsp, #8\n\t"
10933             "movsd   [rsp], $src\n\t"
10934             "call    d2i_fixup\n\t"
10935             "popq    $dst\n"
10936     "done:   "%}
10937   opcode(0xF2, 0x0F, 0x2C);
10938   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
10939              d2i_fixup(dst, src));
10940   ins_pipe(pipe_slow);
10941 %}
10942 
10943 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
10944 %{
10945   match(Set dst (ConvD2L src));
10946   effect(KILL cr);
10947 
10948   format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
10949             "cmpq    $dst, [0x8000000000000000]\n\t"
10950             "jne,s   done\n\t"
10951             "subq    rsp, #8\n\t"
10952             "movsd   [rsp], $src\n\t"
10953             "call    d2l_fixup\n\t"
10954             "popq    $dst\n"
10955     "done:   "%}
10956   opcode(0xF2, 0x0F, 0x2C);
10957   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
10958              d2l_fixup(dst, src));
10959   ins_pipe(pipe_slow);
10960 %}
10961 
10962 instruct convI2F_reg_reg(regF dst, rRegI src)
10963 %{
10964   predicate(!UseXmmI2F);
10965   match(Set dst (ConvI2F src));
10966 
10967   format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10968   opcode(0xF3, 0x0F, 0x2A);
10969   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10970   ins_pipe(pipe_slow); // XXX
10971 %}
10972 
10973 instruct convI2F_reg_mem(regF dst, memory src)
10974 %{
10975   match(Set dst (ConvI2F (LoadI src)));
10976 
10977   format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
10978   opcode(0xF3, 0x0F, 0x2A);
10979   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10980   ins_pipe(pipe_slow); // XXX
10981 %}
10982 
10983 instruct convI2D_reg_reg(regD dst, rRegI src)
10984 %{
10985   predicate(!UseXmmI2D);
10986   match(Set dst (ConvI2D src));
10987 
10988   format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10989   opcode(0xF2, 0x0F, 0x2A);
10990   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10991   ins_pipe(pipe_slow); // XXX
10992 %}
10993 
10994 instruct convI2D_reg_mem(regD dst, memory src)
10995 %{
10996   match(Set dst (ConvI2D (LoadI src)));
10997 
10998   format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
10999   opcode(0xF2, 0x0F, 0x2A);
11000   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11001   ins_pipe(pipe_slow); // XXX
11002 %}
11003 
11004 instruct convXI2F_reg(regF dst, rRegI src)
11005 %{
11006   predicate(UseXmmI2F);
11007   match(Set dst (ConvI2F src));
11008 
11009   format %{ "movdl $dst, $src\n\t"
11010             "cvtdq2psl $dst, $dst\t# i2f" %}
11011   ins_encode %{
11012     __ movdl($dst$$XMMRegister, $src$$Register);
11013     __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11014   %}
11015   ins_pipe(pipe_slow); // XXX
11016 %}
11017 
11018 instruct convXI2D_reg(regD dst, rRegI src)
11019 %{
11020   predicate(UseXmmI2D);
11021   match(Set dst (ConvI2D src));
11022 
11023   format %{ "movdl $dst, $src\n\t"
11024             "cvtdq2pdl $dst, $dst\t# i2d" %}
11025   ins_encode %{
11026     __ movdl($dst$$XMMRegister, $src$$Register);
11027     __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11028   %}
11029   ins_pipe(pipe_slow); // XXX
11030 %}
11031 
11032 instruct convL2F_reg_reg(regF dst, rRegL src)
11033 %{
11034   match(Set dst (ConvL2F src));
11035 
11036   format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11037   opcode(0xF3, 0x0F, 0x2A);
11038   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11039   ins_pipe(pipe_slow); // XXX
11040 %}
11041 
11042 instruct convL2F_reg_mem(regF dst, memory src)
11043 %{
11044   match(Set dst (ConvL2F (LoadL src)));
11045 
11046   format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11047   opcode(0xF3, 0x0F, 0x2A);
11048   ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11049   ins_pipe(pipe_slow); // XXX
11050 %}
11051 
11052 instruct convL2D_reg_reg(regD dst, rRegL src)
11053 %{
11054   match(Set dst (ConvL2D src));
11055 
11056   format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11057   opcode(0xF2, 0x0F, 0x2A);
11058   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11059   ins_pipe(pipe_slow); // XXX
11060 %}
11061 
11062 instruct convL2D_reg_mem(regD dst, memory src)
11063 %{
11064   match(Set dst (ConvL2D (LoadL src)));
11065 
11066   format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11067   opcode(0xF2, 0x0F, 0x2A);
11068   ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11069   ins_pipe(pipe_slow); // XXX
11070 %}
11071 
11072 instruct convI2L_reg_reg(rRegL dst, rRegI src)
11073 %{
11074   match(Set dst (ConvI2L src));
11075 
11076   ins_cost(125);
11077   format %{ "movslq  $dst, $src\t# i2l" %}
11078   opcode(0x63); // needs REX.W
11079   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11080   ins_pipe(ialu_reg_reg);
11081 %}
11082 
11083 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
11084 // %{
11085 //   match(Set dst (ConvI2L src));
11086 // //   predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
11087 // //             _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
11088 //   predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
11089 //             (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
11090 //             ((const TypeNode*) n)->type()->is_long()->_lo ==
11091 //             (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
11092 
11093 //   format %{ "movl    $dst, $src\t# unsigned i2l" %}
11094 //   ins_encode(enc_copy(dst, src));
11095 // //   opcode(0x63); // needs REX.W
11096 // //   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11097 //   ins_pipe(ialu_reg_reg);
11098 // %}
11099 
11100 instruct convI2L_reg_mem(rRegL dst, memory src)
11101 %{
11102   match(Set dst (ConvI2L (LoadI src)));
11103 
11104   format %{ "movslq  $dst, $src\t# i2l" %}
11105   opcode(0x63); // needs REX.W
11106   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst,src));
11107   ins_pipe(ialu_reg_mem);
11108 %}
11109 
11110 // Zero-extend convert int to long
11111 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
11112 %{
11113   match(Set dst (AndL (ConvI2L src) mask));
11114 
11115   format %{ "movl    $dst, $src\t# i2l zero-extend\n\t" %}
11116   ins_encode(enc_copy(dst, src));
11117   ins_pipe(ialu_reg_reg);
11118 %}
11119 
11120 // Zero-extend convert int to long
11121 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
11122 %{
11123   match(Set dst (AndL (ConvI2L (LoadI src)) mask));
11124 
11125   format %{ "movl    $dst, $src\t# i2l zero-extend\n\t" %}
11126   opcode(0x8B);
11127   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11128   ins_pipe(ialu_reg_mem);
11129 %}
11130 
11131 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
11132 %{
11133   match(Set dst (AndL src mask));
11134 
11135   format %{ "movl    $dst, $src\t# zero-extend long" %}
11136   ins_encode(enc_copy_always(dst, src));
11137   ins_pipe(ialu_reg_reg);
11138 %}
11139 
11140 instruct convL2I_reg_reg(rRegI dst, rRegL src)
11141 %{
11142   match(Set dst (ConvL2I src));
11143 
11144   format %{ "movl    $dst, $src\t# l2i" %}
11145   ins_encode(enc_copy_always(dst, src));
11146   ins_pipe(ialu_reg_reg);
11147 %}
11148 
11149 
11150 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11151   match(Set dst (MoveF2I src));
11152   effect(DEF dst, USE src);
11153 
11154   ins_cost(125);
11155   format %{ "movl    $dst, $src\t# MoveF2I_stack_reg" %}
11156   opcode(0x8B);
11157   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11158   ins_pipe(ialu_reg_mem);
11159 %}
11160 
11161 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
11162   match(Set dst (MoveI2F src));
11163   effect(DEF dst, USE src);
11164 
11165   ins_cost(125);
11166   format %{ "movss   $dst, $src\t# MoveI2F_stack_reg" %}
11167   opcode(0xF3, 0x0F, 0x10);
11168   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11169   ins_pipe(pipe_slow);
11170 %}
11171 
11172 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
11173   match(Set dst (MoveD2L src));
11174   effect(DEF dst, USE src);
11175 
11176   ins_cost(125);
11177   format %{ "movq    $dst, $src\t# MoveD2L_stack_reg" %}
11178   opcode(0x8B);
11179   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
11180   ins_pipe(ialu_reg_mem);
11181 %}
11182 
11183 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
11184   predicate(!UseXmmLoadAndClearUpper);
11185   match(Set dst (MoveL2D src));
11186   effect(DEF dst, USE src);
11187 
11188   ins_cost(125);
11189   format %{ "movlpd  $dst, $src\t# MoveL2D_stack_reg" %}
11190   opcode(0x66, 0x0F, 0x12);
11191   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11192   ins_pipe(pipe_slow);
11193 %}
11194 
11195 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
11196   predicate(UseXmmLoadAndClearUpper);
11197   match(Set dst (MoveL2D src));
11198   effect(DEF dst, USE src);
11199 
11200   ins_cost(125);
11201   format %{ "movsd   $dst, $src\t# MoveL2D_stack_reg" %}
11202   opcode(0xF2, 0x0F, 0x10);
11203   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11204   ins_pipe(pipe_slow);
11205 %}
11206 
11207 
11208 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
11209   match(Set dst (MoveF2I src));
11210   effect(DEF dst, USE src);
11211 
11212   ins_cost(95); // XXX
11213   format %{ "movss   $dst, $src\t# MoveF2I_reg_stack" %}
11214   opcode(0xF3, 0x0F, 0x11);
11215   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11216   ins_pipe(pipe_slow);
11217 %}
11218 
11219 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11220   match(Set dst (MoveI2F src));
11221   effect(DEF dst, USE src);
11222 
11223   ins_cost(100);
11224   format %{ "movl    $dst, $src\t# MoveI2F_reg_stack" %}
11225   opcode(0x89);
11226   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
11227   ins_pipe( ialu_mem_reg );
11228 %}
11229 
11230 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
11231   match(Set dst (MoveD2L src));
11232   effect(DEF dst, USE src);
11233 
11234   ins_cost(95); // XXX
11235   format %{ "movsd   $dst, $src\t# MoveL2D_reg_stack" %}
11236   opcode(0xF2, 0x0F, 0x11);
11237   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11238   ins_pipe(pipe_slow);
11239 %}
11240 
11241 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
11242   match(Set dst (MoveL2D src));
11243   effect(DEF dst, USE src);
11244 
11245   ins_cost(100);
11246   format %{ "movq    $dst, $src\t# MoveL2D_reg_stack" %}
11247   opcode(0x89);
11248   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
11249   ins_pipe(ialu_mem_reg);
11250 %}
11251 
11252 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
11253   match(Set dst (MoveF2I src));
11254   effect(DEF dst, USE src);
11255   ins_cost(85);
11256   format %{ "movd    $dst,$src\t# MoveF2I" %}
11257   ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
11258   ins_pipe( pipe_slow );
11259 %}
11260 
11261 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
11262   match(Set dst (MoveD2L src));
11263   effect(DEF dst, USE src);
11264   ins_cost(85);
11265   format %{ "movd    $dst,$src\t# MoveD2L" %}
11266   ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
11267   ins_pipe( pipe_slow );
11268 %}
11269 
11270 // The next instructions have long latency and use Int unit. Set high cost.
11271 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
11272   match(Set dst (MoveI2F src));
11273   effect(DEF dst, USE src);
11274   ins_cost(300);
11275   format %{ "movd    $dst,$src\t# MoveI2F" %}
11276   ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
11277   ins_pipe( pipe_slow );
11278 %}
11279 
11280 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
11281   match(Set dst (MoveL2D src));
11282   effect(DEF dst, USE src);
11283   ins_cost(300);
11284   format %{ "movd    $dst,$src\t# MoveL2D" %}
11285   ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
11286   ins_pipe( pipe_slow );
11287 %}
11288 
11289 // Replicate scalar to packed byte (1 byte) values in xmm
11290 instruct Repl8B_reg(regD dst, regD src) %{
11291   match(Set dst (Replicate8B src));
11292   format %{ "MOVDQA  $dst,$src\n\t"
11293             "PUNPCKLBW $dst,$dst\n\t"
11294             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11295   ins_encode( pshufd_8x8(dst, src));
11296   ins_pipe( pipe_slow );
11297 %}
11298 
11299 // Replicate scalar to packed byte (1 byte) values in xmm
11300 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11301   match(Set dst (Replicate8B src));
11302   format %{ "MOVD    $dst,$src\n\t"
11303             "PUNPCKLBW $dst,$dst\n\t"
11304             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11305   ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11306   ins_pipe( pipe_slow );
11307 %}
11308 
11309 // Replicate scalar zero to packed byte (1 byte) values in xmm
11310 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11311   match(Set dst (Replicate8B zero));
11312   format %{ "PXOR  $dst,$dst\t! replicate8B" %}
11313   ins_encode( pxor(dst, dst));
11314   ins_pipe( fpu_reg_reg );
11315 %}
11316 
11317 // Replicate scalar to packed shore (2 byte) values in xmm
11318 instruct Repl4S_reg(regD dst, regD src) %{
11319   match(Set dst (Replicate4S src));
11320   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11321   ins_encode( pshufd_4x16(dst, src));
11322   ins_pipe( fpu_reg_reg );
11323 %}
11324 
11325 // Replicate scalar to packed shore (2 byte) values in xmm
11326 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11327   match(Set dst (Replicate4S src));
11328   format %{ "MOVD    $dst,$src\n\t"
11329             "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11330   ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11331   ins_pipe( fpu_reg_reg );
11332 %}
11333 
11334 // Replicate scalar zero to packed short (2 byte) values in xmm
11335 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11336   match(Set dst (Replicate4S zero));
11337   format %{ "PXOR  $dst,$dst\t! replicate4S" %}
11338   ins_encode( pxor(dst, dst));
11339   ins_pipe( fpu_reg_reg );
11340 %}
11341 
11342 // Replicate scalar to packed char (2 byte) values in xmm
11343 instruct Repl4C_reg(regD dst, regD src) %{
11344   match(Set dst (Replicate4C src));
11345   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11346   ins_encode( pshufd_4x16(dst, src));
11347   ins_pipe( fpu_reg_reg );
11348 %}
11349 
11350 // Replicate scalar to packed char (2 byte) values in xmm
11351 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11352   match(Set dst (Replicate4C src));
11353   format %{ "MOVD    $dst,$src\n\t"
11354             "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11355   ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11356   ins_pipe( fpu_reg_reg );
11357 %}
11358 
11359 // Replicate scalar zero to packed char (2 byte) values in xmm
11360 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11361   match(Set dst (Replicate4C zero));
11362   format %{ "PXOR  $dst,$dst\t! replicate4C" %}
11363   ins_encode( pxor(dst, dst));
11364   ins_pipe( fpu_reg_reg );
11365 %}
11366 
11367 // Replicate scalar to packed integer (4 byte) values in xmm
11368 instruct Repl2I_reg(regD dst, regD src) %{
11369   match(Set dst (Replicate2I src));
11370   format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11371   ins_encode( pshufd(dst, src, 0x00));
11372   ins_pipe( fpu_reg_reg );
11373 %}
11374 
11375 // Replicate scalar to packed integer (4 byte) values in xmm
11376 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11377   match(Set dst (Replicate2I src));
11378   format %{ "MOVD   $dst,$src\n\t"
11379             "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11380   ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11381   ins_pipe( fpu_reg_reg );
11382 %}
11383 
11384 // Replicate scalar zero to packed integer (2 byte) values in xmm
11385 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11386   match(Set dst (Replicate2I zero));
11387   format %{ "PXOR  $dst,$dst\t! replicate2I" %}
11388   ins_encode( pxor(dst, dst));
11389   ins_pipe( fpu_reg_reg );
11390 %}
11391 
11392 // Replicate scalar to packed single precision floating point values in xmm
11393 instruct Repl2F_reg(regD dst, regD src) %{
11394   match(Set dst (Replicate2F src));
11395   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11396   ins_encode( pshufd(dst, src, 0xe0));
11397   ins_pipe( fpu_reg_reg );
11398 %}
11399 
11400 // Replicate scalar to packed single precision floating point values in xmm
11401 instruct Repl2F_regF(regD dst, regF src) %{
11402   match(Set dst (Replicate2F src));
11403   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11404   ins_encode( pshufd(dst, src, 0xe0));
11405   ins_pipe( fpu_reg_reg );
11406 %}
11407 
11408 // Replicate scalar to packed single precision floating point values in xmm
11409 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11410   match(Set dst (Replicate2F zero));
11411   format %{ "PXOR  $dst,$dst\t! replicate2F" %}
11412   ins_encode( pxor(dst, dst));
11413   ins_pipe( fpu_reg_reg );
11414 %}
11415 
11416 
11417 // =======================================================================
11418 // fast clearing of an array
11419 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11420                   rFlagsReg cr)
11421 %{
11422   match(Set dummy (ClearArray cnt base));
11423   effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11424 
11425   format %{ "xorl    rax, rax\t# ClearArray:\n\t"
11426             "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11427   ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11428              Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11429   ins_pipe(pipe_slow);
11430 %}
11431 
11432 instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1,
11433                         rbx_RegI tmp2, regD6 tmp3, regD7 tmp4, rcx_RegI result, rFlagsReg cr)
11434 %{
11435   match(Set result (StrComp str1 str2));
11436   effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
11437   //ins_cost(300);
11438 
11439   format %{ "String Compare $str1, $str2 -> $result    // XXX KILL RAX, RBX" %}
11440   ins_encode( enc_String_Compare() );
11441   ins_pipe( pipe_slow );
11442 %}
11443 
11444 instruct string_indexof(rsi_RegP str1,rdi_RegP str2, rax_RegI tmp1, rcx_RegI tmp2, rdx_RegI tmp3, regD6 tmp4, rbx_RegI result, rFlagsReg cr) %{
11445   predicate(UseSSE>=4 && UseSSE42Intrinsics);
11446   match(Set result (StrIndexOf str1 str2));
11447   effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
11448 
11449   format %{ "String IndexOf $str1,$str2 -> $result   // KILL EAX, ECX, EDX, XMM6" %}
11450   ins_encode( enc_String_IndexOf() );
11451   ins_pipe( pipe_slow );
11452 %}
11453 
11454 // fast string equals
11455 instruct string_equals(rdi_RegP str1, rsi_RegP str2, rbx_RegI tmp1, rcx_RegI tmp2, regD6 tmp3, regD7 tmp4, rax_RegI result, rFlagsReg cr) %{
11456   match(Set result (StrEquals str1 str2));
11457   effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
11458 
11459   format %{ "String Equals $str1,$str2 -> $result    // KILL EBX, ECX, EDX, XMM6, XMM7" %}
11460   ins_encode( enc_String_Equals() );
11461   ins_pipe( pipe_slow );
11462 %}
11463 
11464 // fast array equals
11465 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, 
11466                       rbx_RegI tmp2, regD6 tmp3, regD7 tmp4, rcx_RegI result, rFlagsReg cr) %{
11467   match(Set result (AryEq ary1 ary2));
11468   effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
11469   //ins_cost(300);
11470 
11471   format %{ "Array Equals $ary1,$ary2 -> $result    // KILL RAX, RBX" %}
11472   ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, tmp3, tmp4, result) );
11473   ins_pipe( pipe_slow );
11474 %}
11475 
11476 //----------Control Flow Instructions------------------------------------------
11477 // Signed compare Instructions
11478 
11479 // XXX more variants!!
11480 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11481 %{
11482   match(Set cr (CmpI op1 op2));
11483   effect(DEF cr, USE op1, USE op2);
11484 
11485   format %{ "cmpl    $op1, $op2" %}
11486   opcode(0x3B);  /* Opcode 3B /r */
11487   ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11488   ins_pipe(ialu_cr_reg_reg);
11489 %}
11490 
11491 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11492 %{
11493   match(Set cr (CmpI op1 op2));
11494 
11495   format %{ "cmpl    $op1, $op2" %}
11496   opcode(0x81, 0x07); /* Opcode 81 /7 */
11497   ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11498   ins_pipe(ialu_cr_reg_imm);
11499 %}
11500 
11501 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11502 %{
11503   match(Set cr (CmpI op1 (LoadI op2)));
11504 
11505   ins_cost(500); // XXX
11506   format %{ "cmpl    $op1, $op2" %}
11507   opcode(0x3B); /* Opcode 3B /r */
11508   ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11509   ins_pipe(ialu_cr_reg_mem);
11510 %}
11511 
11512 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11513 %{
11514   match(Set cr (CmpI src zero));
11515 
11516   format %{ "testl   $src, $src" %}
11517   opcode(0x85);
11518   ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11519   ins_pipe(ialu_cr_reg_imm);
11520 %}
11521 
11522 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11523 %{
11524   match(Set cr (CmpI (AndI src con) zero));
11525 
11526   format %{ "testl   $src, $con" %}
11527   opcode(0xF7, 0x00);
11528   ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11529   ins_pipe(ialu_cr_reg_imm);
11530 %}
11531 
11532 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11533 %{
11534   match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11535 
11536   format %{ "testl   $src, $mem" %}
11537   opcode(0x85);
11538   ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11539   ins_pipe(ialu_cr_reg_mem);
11540 %}
11541 
11542 // Unsigned compare Instructions; really, same as signed except they
11543 // produce an rFlagsRegU instead of rFlagsReg.
11544 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11545 %{
11546   match(Set cr (CmpU op1 op2));
11547 
11548   format %{ "cmpl    $op1, $op2\t# unsigned" %}
11549   opcode(0x3B); /* Opcode 3B /r */
11550   ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11551   ins_pipe(ialu_cr_reg_reg);
11552 %}
11553 
11554 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11555 %{
11556   match(Set cr (CmpU op1 op2));
11557 
11558   format %{ "cmpl    $op1, $op2\t# unsigned" %}
11559   opcode(0x81,0x07); /* Opcode 81 /7 */
11560   ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11561   ins_pipe(ialu_cr_reg_imm);
11562 %}
11563 
11564 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11565 %{
11566   match(Set cr (CmpU op1 (LoadI op2)));
11567 
11568   ins_cost(500); // XXX
11569   format %{ "cmpl    $op1, $op2\t# unsigned" %}
11570   opcode(0x3B); /* Opcode 3B /r */
11571   ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11572   ins_pipe(ialu_cr_reg_mem);
11573 %}
11574 
11575 // // // Cisc-spilled version of cmpU_rReg
11576 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11577 // //%{
11578 // //  match(Set cr (CmpU (LoadI op1) op2));
11579 // //
11580 // //  format %{ "CMPu   $op1,$op2" %}
11581 // //  ins_cost(500);
11582 // //  opcode(0x39);  /* Opcode 39 /r */
11583 // //  ins_encode( OpcP, reg_mem( op1, op2) );
11584 // //%}
11585 
11586 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11587 %{
11588   match(Set cr (CmpU src zero));
11589 
11590   format %{ "testl  $src, $src\t# unsigned" %}
11591   opcode(0x85);
11592   ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11593   ins_pipe(ialu_cr_reg_imm);
11594 %}
11595 
11596 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11597 %{
11598   match(Set cr (CmpP op1 op2));
11599 
11600   format %{ "cmpq    $op1, $op2\t# ptr" %}
11601   opcode(0x3B); /* Opcode 3B /r */
11602   ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11603   ins_pipe(ialu_cr_reg_reg);
11604 %}
11605 
11606 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11607 %{
11608   match(Set cr (CmpP op1 (LoadP op2)));
11609 
11610   ins_cost(500); // XXX
11611   format %{ "cmpq    $op1, $op2\t# ptr" %}
11612   opcode(0x3B); /* Opcode 3B /r */
11613   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11614   ins_pipe(ialu_cr_reg_mem);
11615 %}
11616 
11617 // // // Cisc-spilled version of cmpP_rReg
11618 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11619 // //%{
11620 // //  match(Set cr (CmpP (LoadP op1) op2));
11621 // //
11622 // //  format %{ "CMPu   $op1,$op2" %}
11623 // //  ins_cost(500);
11624 // //  opcode(0x39);  /* Opcode 39 /r */
11625 // //  ins_encode( OpcP, reg_mem( op1, op2) );
11626 // //%}
11627 
11628 // XXX this is generalized by compP_rReg_mem???
11629 // Compare raw pointer (used in out-of-heap check).
11630 // Only works because non-oop pointers must be raw pointers
11631 // and raw pointers have no anti-dependencies.
11632 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11633 %{
11634   predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11635   match(Set cr (CmpP op1 (LoadP op2)));
11636 
11637   format %{ "cmpq    $op1, $op2\t# raw ptr" %}
11638   opcode(0x3B); /* Opcode 3B /r */
11639   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11640   ins_pipe(ialu_cr_reg_mem);
11641 %}
11642 
11643 // This will generate a signed flags result. This should be OK since
11644 // any compare to a zero should be eq/neq.
11645 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11646 %{
11647   match(Set cr (CmpP src zero));
11648 
11649   format %{ "testq   $src, $src\t# ptr" %}
11650   opcode(0x85);
11651   ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11652   ins_pipe(ialu_cr_reg_imm);
11653 %}
11654 
11655 // This will generate a signed flags result. This should be OK since
11656 // any compare to a zero should be eq/neq.
11657 instruct testP_reg_mem(rFlagsReg cr, memory op, immP0 zero)
11658 %{
11659   match(Set cr (CmpP (LoadP op) zero));
11660 
11661   ins_cost(500); // XXX
11662   format %{ "testq   $op, 0xffffffffffffffff\t# ptr" %}
11663   opcode(0xF7); /* Opcode F7 /0 */
11664   ins_encode(REX_mem_wide(op),
11665              OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11666   ins_pipe(ialu_cr_reg_imm);
11667 %}
11668 
11669 
11670 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11671 %{
11672   match(Set cr (CmpN op1 op2));
11673 
11674   format %{ "cmpl    $op1, $op2\t# compressed ptr" %}
11675   ins_encode %{ __ cmpl(as_Register($op1$$reg), as_Register($op2$$reg)); %}
11676   ins_pipe(ialu_cr_reg_reg);
11677 %}
11678 
11679 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11680 %{
11681   match(Set cr (CmpN src (LoadN mem)));
11682 
11683   ins_cost(500); // XXX
11684   format %{ "cmpl    $src, mem\t# compressed ptr" %}
11685   ins_encode %{
11686     Address adr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
11687     __ cmpl(as_Register($src$$reg), adr);
11688   %}
11689   ins_pipe(ialu_cr_reg_mem);
11690 %}
11691 
11692 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11693   match(Set cr (CmpN src zero));
11694 
11695   format %{ "testl   $src, $src\t# compressed ptr" %}
11696   ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11697   ins_pipe(ialu_cr_reg_imm);
11698 %}
11699 
11700 instruct testN_reg_mem(rFlagsReg cr, memory mem, immN0 zero)
11701 %{
11702   match(Set cr (CmpN (LoadN mem) zero));
11703 
11704   ins_cost(500); // XXX
11705   format %{ "testl   $mem, 0xffffffff\t# compressed ptr" %}
11706   ins_encode %{
11707     Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp);
11708     __ cmpl(addr, (int)0xFFFFFFFF);
11709   %}
11710   ins_pipe(ialu_cr_reg_mem);
11711 %}
11712 
11713 // Yanked all unsigned pointer compare operations.
11714 // Pointer compares are done with CmpP which is already unsigned.
11715 
11716 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11717 %{
11718   match(Set cr (CmpL op1 op2));
11719 
11720   format %{ "cmpq    $op1, $op2" %}
11721   opcode(0x3B);  /* Opcode 3B /r */
11722   ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11723   ins_pipe(ialu_cr_reg_reg);
11724 %}
11725 
11726 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11727 %{
11728   match(Set cr (CmpL op1 op2));
11729 
11730   format %{ "cmpq    $op1, $op2" %}
11731   opcode(0x81, 0x07); /* Opcode 81 /7 */
11732   ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11733   ins_pipe(ialu_cr_reg_imm);
11734 %}
11735 
11736 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11737 %{
11738   match(Set cr (CmpL op1 (LoadL op2)));
11739 
11740   ins_cost(500); // XXX
11741   format %{ "cmpq    $op1, $op2" %}
11742   opcode(0x3B); /* Opcode 3B /r */
11743   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11744   ins_pipe(ialu_cr_reg_mem);
11745 %}
11746 
11747 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11748 %{
11749   match(Set cr (CmpL src zero));
11750 
11751   format %{ "testq   $src, $src" %}
11752   opcode(0x85);
11753   ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11754   ins_pipe(ialu_cr_reg_imm);
11755 %}
11756 
11757 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11758 %{
11759   match(Set cr (CmpL (AndL src con) zero));
11760 
11761   format %{ "testq   $src, $con\t# long" %}
11762   opcode(0xF7, 0x00);
11763   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11764   ins_pipe(ialu_cr_reg_imm);
11765 %}
11766 
11767 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
11768 %{
11769   match(Set cr (CmpL (AndL src (LoadL mem)) zero));
11770 
11771   format %{ "testq   $src, $mem" %}
11772   opcode(0x85);
11773   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
11774   ins_pipe(ialu_cr_reg_mem);
11775 %}
11776 
11777 // Manifest a CmpL result in an integer register.  Very painful.
11778 // This is the test to avoid.
11779 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
11780 %{
11781   match(Set dst (CmpL3 src1 src2));
11782   effect(KILL flags);
11783 
11784   ins_cost(275); // XXX
11785   format %{ "cmpq    $src1, $src2\t# CmpL3\n\t"
11786             "movl    $dst, -1\n\t"
11787             "jl,s    done\n\t"
11788             "setne   $dst\n\t"
11789             "movzbl  $dst, $dst\n\t"
11790     "done:" %}
11791   ins_encode(cmpl3_flag(src1, src2, dst));
11792   ins_pipe(pipe_slow);
11793 %}
11794 
11795 //----------Max and Min--------------------------------------------------------
11796 // Min Instructions
11797 
11798 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
11799 %{
11800   effect(USE_DEF dst, USE src, USE cr);
11801 
11802   format %{ "cmovlgt $dst, $src\t# min" %}
11803   opcode(0x0F, 0x4F);
11804   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11805   ins_pipe(pipe_cmov_reg);
11806 %}
11807 
11808 
11809 instruct minI_rReg(rRegI dst, rRegI src)
11810 %{
11811   match(Set dst (MinI dst src));
11812 
11813   ins_cost(200);
11814   expand %{
11815     rFlagsReg cr;
11816     compI_rReg(cr, dst, src);
11817     cmovI_reg_g(dst, src, cr);
11818   %}
11819 %}
11820 
11821 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
11822 %{
11823   effect(USE_DEF dst, USE src, USE cr);
11824 
11825   format %{ "cmovllt $dst, $src\t# max" %}
11826   opcode(0x0F, 0x4C);
11827   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
11828   ins_pipe(pipe_cmov_reg);
11829 %}
11830 
11831 
11832 instruct maxI_rReg(rRegI dst, rRegI src)
11833 %{
11834   match(Set dst (MaxI dst src));
11835 
11836   ins_cost(200);
11837   expand %{
11838     rFlagsReg cr;
11839     compI_rReg(cr, dst, src);
11840     cmovI_reg_l(dst, src, cr);
11841   %}
11842 %}
11843 
11844 // ============================================================================
11845 // Branch Instructions
11846 
11847 // Jump Direct - Label defines a relative address from JMP+1
11848 instruct jmpDir(label labl)
11849 %{
11850   match(Goto);
11851   effect(USE labl);
11852 
11853   ins_cost(300);
11854   format %{ "jmp     $labl" %}
11855   size(5);
11856   opcode(0xE9);
11857   ins_encode(OpcP, Lbl(labl));
11858   ins_pipe(pipe_jmp);
11859   ins_pc_relative(1);
11860 %}
11861 
11862 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11863 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
11864 %{
11865   match(If cop cr);
11866   effect(USE labl);
11867 
11868   ins_cost(300);
11869   format %{ "j$cop     $labl" %}
11870   size(6);
11871   opcode(0x0F, 0x80);
11872   ins_encode(Jcc(cop, labl));
11873   ins_pipe(pipe_jcc);
11874   ins_pc_relative(1);
11875 %}
11876 
11877 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11878 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
11879 %{
11880   match(CountedLoopEnd cop cr);
11881   effect(USE labl);
11882 
11883   ins_cost(300);
11884   format %{ "j$cop     $labl\t# loop end" %}
11885   size(6);
11886   opcode(0x0F, 0x80);
11887   ins_encode(Jcc(cop, labl));
11888   ins_pipe(pipe_jcc);
11889   ins_pc_relative(1);
11890 %}
11891 
11892 // Jump Direct Conditional - Label defines a relative address from Jcc+1
11893 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11894   match(CountedLoopEnd cop cmp);
11895   effect(USE labl);
11896 
11897   ins_cost(300);
11898   format %{ "j$cop,u   $labl\t# loop end" %}
11899   size(6);
11900   opcode(0x0F, 0x80);
11901   ins_encode(Jcc(cop, labl));
11902   ins_pipe(pipe_jcc);
11903   ins_pc_relative(1);
11904 %}
11905 
11906 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11907   match(CountedLoopEnd cop cmp);
11908   effect(USE labl);
11909 
11910   ins_cost(200);
11911   format %{ "j$cop,u   $labl\t# loop end" %}
11912   size(6);
11913   opcode(0x0F, 0x80);
11914   ins_encode(Jcc(cop, labl));
11915   ins_pipe(pipe_jcc);
11916   ins_pc_relative(1);
11917 %}
11918 
11919 // Jump Direct Conditional - using unsigned comparison
11920 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
11921   match(If cop cmp);
11922   effect(USE labl);
11923 
11924   ins_cost(300);
11925   format %{ "j$cop,u  $labl" %}
11926   size(6);
11927   opcode(0x0F, 0x80);
11928   ins_encode(Jcc(cop, labl));
11929   ins_pipe(pipe_jcc);
11930   ins_pc_relative(1);
11931 %}
11932 
11933 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
11934   match(If cop cmp);
11935   effect(USE labl);
11936 
11937   ins_cost(200);
11938   format %{ "j$cop,u  $labl" %}
11939   size(6);
11940   opcode(0x0F, 0x80);
11941   ins_encode(Jcc(cop, labl));
11942   ins_pipe(pipe_jcc);
11943   ins_pc_relative(1);
11944 %}
11945 
11946 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
11947   match(If cop cmp);
11948   effect(USE labl);
11949 
11950   ins_cost(200);
11951   format %{ $$template
11952     if ($cop$$cmpcode == Assembler::notEqual) {
11953       $$emit$$"jp,u   $labl\n\t"
11954       $$emit$$"j$cop,u   $labl"
11955     } else {
11956       $$emit$$"jp,u   done\n\t"
11957       $$emit$$"j$cop,u   $labl\n\t"
11958       $$emit$$"done:"
11959     }
11960   %}
11961   size(12);
11962   opcode(0x0F, 0x80);
11963   ins_encode %{
11964     Label* l = $labl$$label;
11965     $$$emit8$primary;
11966     emit_cc(cbuf, $secondary, Assembler::parity);
11967     int parity_disp = -1;
11968     if ($cop$$cmpcode == Assembler::notEqual) {
11969        // the two jumps 6 bytes apart so the jump distances are too
11970        parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
11971     } else if ($cop$$cmpcode == Assembler::equal) {
11972        parity_disp = 6;
11973     } else {
11974        ShouldNotReachHere();
11975     }
11976     emit_d32(cbuf, parity_disp);
11977     $$$emit8$primary;
11978     emit_cc(cbuf, $secondary, $cop$$cmpcode);
11979     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
11980     emit_d32(cbuf, disp);
11981   %}
11982   ins_pipe(pipe_jcc);
11983   ins_pc_relative(1);
11984 %}
11985 
11986 // ============================================================================
11987 // The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary
11988 // superklass array for an instance of the superklass.  Set a hidden
11989 // internal cache on a hit (cache is checked with exposed code in
11990 // gen_subtype_check()).  Return NZ for a miss or zero for a hit.  The
11991 // encoding ALSO sets flags.
11992 
11993 instruct partialSubtypeCheck(rdi_RegP result,
11994                              rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
11995                              rFlagsReg cr)
11996 %{
11997   match(Set result (PartialSubtypeCheck sub super));
11998   effect(KILL rcx, KILL cr);
11999 
12000   ins_cost(1100);  // slightly larger than the next version
12001   format %{ "cmpq    rax, rsi\n\t"
12002             "jeq,s   hit\n\t"
12003             "movq    rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12004             "movl    rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12005             "addq    rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12006             "repne   scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
12007             "jne,s   miss\t\t# Missed: rdi not-zero\n\t"
12008             "movq    [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12009     "hit:\n\t"
12010             "xorq    $result, $result\t\t Hit: rdi zero\n\t"
12011     "miss:\t" %}
12012 
12013   opcode(0x1); // Force a XOR of RDI
12014   ins_encode(enc_PartialSubtypeCheck());
12015   ins_pipe(pipe_slow);
12016 %}
12017 
12018 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
12019                                      rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12020                                      immP0 zero,
12021                                      rdi_RegP result)
12022 %{
12023   match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12024   predicate(!UseCompressedOops); // decoding oop kills condition codes
12025   effect(KILL rcx, KILL result);
12026 
12027   ins_cost(1000);
12028   format %{ "cmpq    rax, rsi\n\t"
12029             "jeq,s   miss\t# Actually a hit; we are done.\n\t"
12030             "movq    rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12031             "movl    rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12032             "addq    rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12033             "repne   scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
12034             "jne,s   miss\t\t# Missed: flags nz\n\t"
12035             "movq    [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12036     "miss:\t" %}
12037 
12038   opcode(0x0); // No need to XOR RDI
12039   ins_encode(enc_PartialSubtypeCheck());
12040   ins_pipe(pipe_slow);
12041 %}
12042 
12043 // ============================================================================
12044 // Branch Instructions -- short offset versions
12045 //
12046 // These instructions are used to replace jumps of a long offset (the default
12047 // match) with jumps of a shorter offset.  These instructions are all tagged
12048 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12049 // match rules in general matching.  Instead, the ADLC generates a conversion
12050 // method in the MachNode which can be used to do in-place replacement of the
12051 // long variant with the shorter variant.  The compiler will determine if a
12052 // branch can be taken by the is_short_branch_offset() predicate in the machine
12053 // specific code section of the file.
12054 
12055 // Jump Direct - Label defines a relative address from JMP+1
12056 instruct jmpDir_short(label labl) %{
12057   match(Goto);
12058   effect(USE labl);
12059 
12060   ins_cost(300);
12061   format %{ "jmp,s   $labl" %}
12062   size(2);
12063   opcode(0xEB);
12064   ins_encode(OpcP, LblShort(labl));
12065   ins_pipe(pipe_jmp);
12066   ins_pc_relative(1);
12067   ins_short_branch(1);
12068 %}
12069 
12070 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12071 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
12072   match(If cop cr);
12073   effect(USE labl);
12074 
12075   ins_cost(300);
12076   format %{ "j$cop,s   $labl" %}
12077   size(2);
12078   opcode(0x70);
12079   ins_encode(JccShort(cop, labl));
12080   ins_pipe(pipe_jcc);
12081   ins_pc_relative(1);
12082   ins_short_branch(1);
12083 %}
12084 
12085 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12086 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
12087   match(CountedLoopEnd cop cr);
12088   effect(USE labl);
12089 
12090   ins_cost(300);
12091   format %{ "j$cop,s   $labl\t# loop end" %}
12092   size(2);
12093   opcode(0x70);
12094   ins_encode(JccShort(cop, labl));
12095   ins_pipe(pipe_jcc);
12096   ins_pc_relative(1);
12097   ins_short_branch(1);
12098 %}
12099 
12100 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12101 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12102   match(CountedLoopEnd cop cmp);
12103   effect(USE labl);
12104 
12105   ins_cost(300);
12106   format %{ "j$cop,us  $labl\t# loop end" %}
12107   size(2);
12108   opcode(0x70);
12109   ins_encode(JccShort(cop, labl));
12110   ins_pipe(pipe_jcc);
12111   ins_pc_relative(1);
12112   ins_short_branch(1);
12113 %}
12114 
12115 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12116   match(CountedLoopEnd cop cmp);
12117   effect(USE labl);
12118 
12119   ins_cost(300);
12120   format %{ "j$cop,us  $labl\t# loop end" %}
12121   size(2);
12122   opcode(0x70);
12123   ins_encode(JccShort(cop, labl));
12124   ins_pipe(pipe_jcc);
12125   ins_pc_relative(1);
12126   ins_short_branch(1);
12127 %}
12128 
12129 // Jump Direct Conditional - using unsigned comparison
12130 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12131   match(If cop cmp);
12132   effect(USE labl);
12133 
12134   ins_cost(300);
12135   format %{ "j$cop,us  $labl" %}
12136   size(2);
12137   opcode(0x70);
12138   ins_encode(JccShort(cop, labl));
12139   ins_pipe(pipe_jcc);
12140   ins_pc_relative(1);
12141   ins_short_branch(1);
12142 %}
12143 
12144 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12145   match(If cop cmp);
12146   effect(USE labl);
12147 
12148   ins_cost(300);
12149   format %{ "j$cop,us  $labl" %}
12150   size(2);
12151   opcode(0x70);
12152   ins_encode(JccShort(cop, labl));
12153   ins_pipe(pipe_jcc);
12154   ins_pc_relative(1);
12155   ins_short_branch(1);
12156 %}
12157 
12158 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12159   match(If cop cmp);
12160   effect(USE labl);
12161 
12162   ins_cost(300);
12163   format %{ $$template
12164     if ($cop$$cmpcode == Assembler::notEqual) {
12165       $$emit$$"jp,u,s   $labl\n\t"
12166       $$emit$$"j$cop,u,s   $labl"
12167     } else {
12168       $$emit$$"jp,u,s   done\n\t"
12169       $$emit$$"j$cop,u,s  $labl\n\t"
12170       $$emit$$"done:"
12171     }
12172   %}
12173   size(4);
12174   opcode(0x70);
12175   ins_encode %{
12176     Label* l = $labl$$label;
12177     emit_cc(cbuf, $primary, Assembler::parity);
12178     int parity_disp = -1;
12179     if ($cop$$cmpcode == Assembler::notEqual) {
12180       parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12181     } else if ($cop$$cmpcode == Assembler::equal) {
12182       parity_disp = 2;
12183     } else {
12184       ShouldNotReachHere();
12185     }
12186     emit_d8(cbuf, parity_disp);
12187     emit_cc(cbuf, $primary, $cop$$cmpcode);
12188     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12189     emit_d8(cbuf, disp);
12190     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
12191     assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
12192   %}
12193   ins_pipe(pipe_jcc);
12194   ins_pc_relative(1);
12195   ins_short_branch(1);
12196 %}
12197 
12198 // ============================================================================
12199 // inlined locking and unlocking
12200 
12201 instruct cmpFastLock(rFlagsReg cr,
12202                      rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
12203 %{
12204   match(Set cr (FastLock object box));
12205   effect(TEMP tmp, TEMP scr);
12206 
12207   ins_cost(300);
12208   format %{ "fastlock $object,$box,$tmp,$scr" %}
12209   ins_encode(Fast_Lock(object, box, tmp, scr));
12210   ins_pipe(pipe_slow);
12211   ins_pc_relative(1);
12212 %}
12213 
12214 instruct cmpFastUnlock(rFlagsReg cr,
12215                        rRegP object, rax_RegP box, rRegP tmp)
12216 %{
12217   match(Set cr (FastUnlock object box));
12218   effect(TEMP tmp);
12219 
12220   ins_cost(300);
12221   format %{ "fastunlock $object, $box, $tmp" %}
12222   ins_encode(Fast_Unlock(object, box, tmp));
12223   ins_pipe(pipe_slow);
12224   ins_pc_relative(1);
12225 %}
12226 
12227 
12228 // ============================================================================
12229 // Safepoint Instructions
12230 instruct safePoint_poll(rFlagsReg cr)
12231 %{
12232   match(SafePoint);
12233   effect(KILL cr);
12234 
12235   format %{ "testl   rax, [rip + #offset_to_poll_page]\t"
12236             "# Safepoint: poll for GC" %}
12237   size(6); // Opcode + ModRM + Disp32 == 6 bytes
12238   ins_cost(125);
12239   ins_encode(enc_safepoint_poll);
12240   ins_pipe(ialu_reg_mem);
12241 %}
12242 
12243 // ============================================================================
12244 // Procedure Call/Return Instructions
12245 // Call Java Static Instruction
12246 // Note: If this code changes, the corresponding ret_addr_offset() and
12247 //       compute_padding() functions will have to be adjusted.
12248 instruct CallStaticJavaDirect(method meth)
12249 %{
12250   match(CallStaticJava);
12251   effect(USE meth);
12252 
12253   ins_cost(300);
12254   format %{ "call,static " %}
12255   opcode(0xE8); /* E8 cd */
12256   ins_encode(Java_Static_Call(meth), call_epilog);
12257   ins_pipe(pipe_slow);
12258   ins_pc_relative(1);
12259   ins_alignment(4);
12260 %}
12261 
12262 // Call Java Dynamic Instruction
12263 // Note: If this code changes, the corresponding ret_addr_offset() and
12264 //       compute_padding() functions will have to be adjusted.
12265 instruct CallDynamicJavaDirect(method meth)
12266 %{
12267   match(CallDynamicJava);
12268   effect(USE meth);
12269 
12270   ins_cost(300);
12271   format %{ "movq    rax, #Universe::non_oop_word()\n\t"
12272             "call,dynamic " %}
12273   opcode(0xE8); /* E8 cd */
12274   ins_encode(Java_Dynamic_Call(meth), call_epilog);
12275   ins_pipe(pipe_slow);
12276   ins_pc_relative(1);
12277   ins_alignment(4);
12278 %}
12279 
12280 // Call Runtime Instruction
12281 instruct CallRuntimeDirect(method meth)
12282 %{
12283   match(CallRuntime);
12284   effect(USE meth);
12285 
12286   ins_cost(300);
12287   format %{ "call,runtime " %}
12288   opcode(0xE8); /* E8 cd */
12289   ins_encode(Java_To_Runtime(meth));
12290   ins_pipe(pipe_slow);
12291   ins_pc_relative(1);
12292 %}
12293 
12294 // Call runtime without safepoint
12295 instruct CallLeafDirect(method meth)
12296 %{
12297   match(CallLeaf);
12298   effect(USE meth);
12299 
12300   ins_cost(300);
12301   format %{ "call_leaf,runtime " %}
12302   opcode(0xE8); /* E8 cd */
12303   ins_encode(Java_To_Runtime(meth));
12304   ins_pipe(pipe_slow);
12305   ins_pc_relative(1);
12306 %}
12307 
12308 // Call runtime without safepoint
12309 instruct CallLeafNoFPDirect(method meth)
12310 %{
12311   match(CallLeafNoFP);
12312   effect(USE meth);
12313 
12314   ins_cost(300);
12315   format %{ "call_leaf_nofp,runtime " %}
12316   opcode(0xE8); /* E8 cd */
12317   ins_encode(Java_To_Runtime(meth));
12318   ins_pipe(pipe_slow);
12319   ins_pc_relative(1);
12320 %}
12321 
12322 // Return Instruction
12323 // Remove the return address & jump to it.
12324 // Notice: We always emit a nop after a ret to make sure there is room
12325 // for safepoint patching
12326 instruct Ret()
12327 %{
12328   match(Return);
12329 
12330   format %{ "ret" %}
12331   opcode(0xC3);
12332   ins_encode(OpcP);
12333   ins_pipe(pipe_jmp);
12334 %}
12335 
12336 // Tail Call; Jump from runtime stub to Java code.
12337 // Also known as an 'interprocedural jump'.
12338 // Target of jump will eventually return to caller.
12339 // TailJump below removes the return address.
12340 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12341 %{
12342   match(TailCall jump_target method_oop);
12343 
12344   ins_cost(300);
12345   format %{ "jmp     $jump_target\t# rbx holds method oop" %}
12346   opcode(0xFF, 0x4); /* Opcode FF /4 */
12347   ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12348   ins_pipe(pipe_jmp);
12349 %}
12350 
12351 // Tail Jump; remove the return address; jump to target.
12352 // TailCall above leaves the return address around.
12353 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12354 %{
12355   match(TailJump jump_target ex_oop);
12356 
12357   ins_cost(300);
12358   format %{ "popq    rdx\t# pop return address\n\t"
12359             "jmp     $jump_target" %}
12360   opcode(0xFF, 0x4); /* Opcode FF /4 */
12361   ins_encode(Opcode(0x5a), // popq rdx
12362              REX_reg(jump_target), OpcP, reg_opc(jump_target));
12363   ins_pipe(pipe_jmp);
12364 %}
12365 
12366 // Create exception oop: created by stack-crawling runtime code.
12367 // Created exception is now available to this handler, and is setup
12368 // just prior to jumping to this handler.  No code emitted.
12369 instruct CreateException(rax_RegP ex_oop)
12370 %{
12371   match(Set ex_oop (CreateEx));
12372 
12373   size(0);
12374   // use the following format syntax
12375   format %{ "# exception oop is in rax; no code emitted" %}
12376   ins_encode();
12377   ins_pipe(empty);
12378 %}
12379 
12380 // Rethrow exception:
12381 // The exception oop will come in the first argument position.
12382 // Then JUMP (not call) to the rethrow stub code.
12383 instruct RethrowException()
12384 %{
12385   match(Rethrow);
12386 
12387   // use the following format syntax
12388   format %{ "jmp     rethrow_stub" %}
12389   ins_encode(enc_rethrow);
12390   ins_pipe(pipe_jmp);
12391 %}
12392 
12393 
12394 //----------PEEPHOLE RULES-----------------------------------------------------
12395 // These must follow all instruction definitions as they use the names
12396 // defined in the instructions definitions.
12397 //
12398 // peepmatch ( root_instr_name [precerding_instruction]* );
12399 //
12400 // peepconstraint %{
12401 // (instruction_number.operand_name relational_op instruction_number.operand_name
12402 //  [, ...] );
12403 // // instruction numbers are zero-based using left to right order in peepmatch
12404 //
12405 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
12406 // // provide an instruction_number.operand_name for each operand that appears
12407 // // in the replacement instruction's match rule
12408 //
12409 // ---------VM FLAGS---------------------------------------------------------
12410 //
12411 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12412 //
12413 // Each peephole rule is given an identifying number starting with zero and
12414 // increasing by one in the order seen by the parser.  An individual peephole
12415 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12416 // on the command-line.
12417 //
12418 // ---------CURRENT LIMITATIONS----------------------------------------------
12419 //
12420 // Only match adjacent instructions in same basic block
12421 // Only equality constraints
12422 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12423 // Only one replacement instruction
12424 //
12425 // ---------EXAMPLE----------------------------------------------------------
12426 //
12427 // // pertinent parts of existing instructions in architecture description
12428 // instruct movI(rRegI dst, rRegI src)
12429 // %{
12430 //   match(Set dst (CopyI src));
12431 // %}
12432 //
12433 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12434 // %{
12435 //   match(Set dst (AddI dst src));
12436 //   effect(KILL cr);
12437 // %}
12438 //
12439 // // Change (inc mov) to lea
12440 // peephole %{
12441 //   // increment preceeded by register-register move
12442 //   peepmatch ( incI_rReg movI );
12443 //   // require that the destination register of the increment
12444 //   // match the destination register of the move
12445 //   peepconstraint ( 0.dst == 1.dst );
12446 //   // construct a replacement instruction that sets
12447 //   // the destination to ( move's source register + one )
12448 //   peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12449 // %}
12450 //
12451 
12452 // Implementation no longer uses movX instructions since
12453 // machine-independent system no longer uses CopyX nodes.
12454 //
12455 // peephole
12456 // %{
12457 //   peepmatch (incI_rReg movI);
12458 //   peepconstraint (0.dst == 1.dst);
12459 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12460 // %}
12461 
12462 // peephole
12463 // %{
12464 //   peepmatch (decI_rReg movI);
12465 //   peepconstraint (0.dst == 1.dst);
12466 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12467 // %}
12468 
12469 // peephole
12470 // %{
12471 //   peepmatch (addI_rReg_imm movI);
12472 //   peepconstraint (0.dst == 1.dst);
12473 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12474 // %}
12475 
12476 // peephole
12477 // %{
12478 //   peepmatch (incL_rReg movL);
12479 //   peepconstraint (0.dst == 1.dst);
12480 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12481 // %}
12482 
12483 // peephole
12484 // %{
12485 //   peepmatch (decL_rReg movL);
12486 //   peepconstraint (0.dst == 1.dst);
12487 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12488 // %}
12489 
12490 // peephole
12491 // %{
12492 //   peepmatch (addL_rReg_imm movL);
12493 //   peepconstraint (0.dst == 1.dst);
12494 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12495 // %}
12496 
12497 // peephole
12498 // %{
12499 //   peepmatch (addP_rReg_imm movP);
12500 //   peepconstraint (0.dst == 1.dst);
12501 //   peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12502 // %}
12503 
12504 // // Change load of spilled value to only a spill
12505 // instruct storeI(memory mem, rRegI src)
12506 // %{
12507 //   match(Set mem (StoreI mem src));
12508 // %}
12509 //
12510 // instruct loadI(rRegI dst, memory mem)
12511 // %{
12512 //   match(Set dst (LoadI mem));
12513 // %}
12514 //
12515 
12516 peephole
12517 %{
12518   peepmatch (loadI storeI);
12519   peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12520   peepreplace (storeI(1.mem 1.mem 1.src));
12521 %}
12522 
12523 peephole
12524 %{
12525   peepmatch (loadL storeL);
12526   peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12527   peepreplace (storeL(1.mem 1.mem 1.src));
12528 %}
12529 
12530 //----------SMARTSPILL RULES---------------------------------------------------
12531 // These must follow all instruction definitions as they use the names
12532 // defined in the instructions definitions.