src/cpu/x86/vm/x86_64.ad

Fri, 08 May 2009 10:44:20 -0700

author
kvn
date
Fri, 08 May 2009 10:44:20 -0700
changeset 1215
c96bf21b756f
parent 1210
93c14e5562c4
child 1220
2056494941db
permissions
-rw-r--r--

6788527: Server vm intermittently fails with assertion "live value must not be garbage" with fastdebug bits
Summary: Cache Jvmti and DTrace flags used by Compiler.
Reviewed-by: never

     1 //
     2 // Copyright 2003-2009 Sun Microsystems, Inc.  All Rights Reserved.
     3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4 //
     5 // This code is free software; you can redistribute it and/or modify it
     6 // under the terms of the GNU General Public License version 2 only, as
     7 // published by the Free Software Foundation.
     8 //
     9 // This code is distributed in the hope that it will be useful, but WITHOUT
    10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11 // FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12 // version 2 for more details (a copy is included in the LICENSE file that
    13 // accompanied this code).
    14 //
    15 // You should have received a copy of the GNU General Public License version
    16 // 2 along with this work; if not, write to the Free Software Foundation,
    17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18 //
    19 // Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
    20 // CA 95054 USA or visit www.sun.com if you need additional information or
    21 // have any questions.
    22 //
    23 //
    25 // AMD64 Architecture Description File
    27 //----------REGISTER DEFINITION BLOCK------------------------------------------
    28 // This information is used by the matcher and the register allocator to
    29 // describe individual registers and classes of registers within the target
    30 // archtecture.
    32 register %{
    33 //----------Architecture Description Register Definitions----------------------
    34 // General Registers
    35 // "reg_def"  name ( register save type, C convention save type,
    36 //                   ideal register type, encoding );
    37 // Register Save Types:
    38 //
    39 // NS  = No-Save:       The register allocator assumes that these registers
    40 //                      can be used without saving upon entry to the method, &
    41 //                      that they do not need to be saved at call sites.
    42 //
    43 // SOC = Save-On-Call:  The register allocator assumes that these registers
    44 //                      can be used without saving upon entry to the method,
    45 //                      but that they must be saved at call sites.
    46 //
    47 // SOE = Save-On-Entry: The register allocator assumes that these registers
    48 //                      must be saved before using them upon entry to the
    49 //                      method, but they do not need to be saved at call
    50 //                      sites.
    51 //
    52 // AS  = Always-Save:   The register allocator assumes that these registers
    53 //                      must be saved before using them upon entry to the
    54 //                      method, & that they must be saved at call sites.
    55 //
    56 // Ideal Register Type is used to determine how to save & restore a
    57 // register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
    58 // spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
    59 //
    60 // The encoding number is the actual bit-pattern placed into the opcodes.
    62 // General Registers
    63 // R8-R15 must be encoded with REX.  (RSP, RBP, RSI, RDI need REX when
    64 // used as byte registers)
    66 // Previously set RBX, RSI, and RDI as save-on-entry for java code
    67 // Turn off SOE in java-code due to frequent use of uncommon-traps.
    68 // Now that allocator is better, turn on RSI and RDI as SOE registers.
    70 reg_def RAX  (SOC, SOC, Op_RegI,  0, rax->as_VMReg());
    71 reg_def RAX_H(SOC, SOC, Op_RegI,  0, rax->as_VMReg()->next());
    73 reg_def RCX  (SOC, SOC, Op_RegI,  1, rcx->as_VMReg());
    74 reg_def RCX_H(SOC, SOC, Op_RegI,  1, rcx->as_VMReg()->next());
    76 reg_def RDX  (SOC, SOC, Op_RegI,  2, rdx->as_VMReg());
    77 reg_def RDX_H(SOC, SOC, Op_RegI,  2, rdx->as_VMReg()->next());
    79 reg_def RBX  (SOC, SOE, Op_RegI,  3, rbx->as_VMReg());
    80 reg_def RBX_H(SOC, SOE, Op_RegI,  3, rbx->as_VMReg()->next());
    82 reg_def RSP  (NS,  NS,  Op_RegI,  4, rsp->as_VMReg());
    83 reg_def RSP_H(NS,  NS,  Op_RegI,  4, rsp->as_VMReg()->next());
    85 // now that adapter frames are gone RBP is always saved and restored by the prolog/epilog code
    86 reg_def RBP  (NS, SOE, Op_RegI,  5, rbp->as_VMReg());
    87 reg_def RBP_H(NS, SOE, Op_RegI,  5, rbp->as_VMReg()->next());
    89 #ifdef _WIN64
    91 reg_def RSI  (SOC, SOE, Op_RegI,  6, rsi->as_VMReg());
    92 reg_def RSI_H(SOC, SOE, Op_RegI,  6, rsi->as_VMReg()->next());
    94 reg_def RDI  (SOC, SOE, Op_RegI,  7, rdi->as_VMReg());
    95 reg_def RDI_H(SOC, SOE, Op_RegI,  7, rdi->as_VMReg()->next());
    97 #else
    99 reg_def RSI  (SOC, SOC, Op_RegI,  6, rsi->as_VMReg());
   100 reg_def RSI_H(SOC, SOC, Op_RegI,  6, rsi->as_VMReg()->next());
   102 reg_def RDI  (SOC, SOC, Op_RegI,  7, rdi->as_VMReg());
   103 reg_def RDI_H(SOC, SOC, Op_RegI,  7, rdi->as_VMReg()->next());
   105 #endif
   107 reg_def R8   (SOC, SOC, Op_RegI,  8, r8->as_VMReg());
   108 reg_def R8_H (SOC, SOC, Op_RegI,  8, r8->as_VMReg()->next());
   110 reg_def R9   (SOC, SOC, Op_RegI,  9, r9->as_VMReg());
   111 reg_def R9_H (SOC, SOC, Op_RegI,  9, r9->as_VMReg()->next());
   113 reg_def R10  (SOC, SOC, Op_RegI, 10, r10->as_VMReg());
   114 reg_def R10_H(SOC, SOC, Op_RegI, 10, r10->as_VMReg()->next());
   116 reg_def R11  (SOC, SOC, Op_RegI, 11, r11->as_VMReg());
   117 reg_def R11_H(SOC, SOC, Op_RegI, 11, r11->as_VMReg()->next());
   119 reg_def R12  (SOC, SOE, Op_RegI, 12, r12->as_VMReg());
   120 reg_def R12_H(SOC, SOE, Op_RegI, 12, r12->as_VMReg()->next());
   122 reg_def R13  (SOC, SOE, Op_RegI, 13, r13->as_VMReg());
   123 reg_def R13_H(SOC, SOE, Op_RegI, 13, r13->as_VMReg()->next());
   125 reg_def R14  (SOC, SOE, Op_RegI, 14, r14->as_VMReg());
   126 reg_def R14_H(SOC, SOE, Op_RegI, 14, r14->as_VMReg()->next());
   128 reg_def R15  (SOC, SOE, Op_RegI, 15, r15->as_VMReg());
   129 reg_def R15_H(SOC, SOE, Op_RegI, 15, r15->as_VMReg()->next());
   132 // Floating Point Registers
   134 // XMM registers.  128-bit registers or 4 words each, labeled (a)-d.
   135 // Word a in each register holds a Float, words ab hold a Double.  We
   136 // currently do not use the SIMD capabilities, so registers cd are
   137 // unused at the moment.
   138 // XMM8-XMM15 must be encoded with REX.
   139 // Linux ABI:   No register preserved across function calls
   140 //              XMM0-XMM7 might hold parameters
   141 // Windows ABI: XMM6-XMM15 preserved across function calls
   142 //              XMM0-XMM3 might hold parameters
   144 reg_def XMM0   (SOC, SOC, Op_RegF,  0, xmm0->as_VMReg());
   145 reg_def XMM0_H (SOC, SOC, Op_RegF,  0, xmm0->as_VMReg()->next());
   147 reg_def XMM1   (SOC, SOC, Op_RegF,  1, xmm1->as_VMReg());
   148 reg_def XMM1_H (SOC, SOC, Op_RegF,  1, xmm1->as_VMReg()->next());
   150 reg_def XMM2   (SOC, SOC, Op_RegF,  2, xmm2->as_VMReg());
   151 reg_def XMM2_H (SOC, SOC, Op_RegF,  2, xmm2->as_VMReg()->next());
   153 reg_def XMM3   (SOC, SOC, Op_RegF,  3, xmm3->as_VMReg());
   154 reg_def XMM3_H (SOC, SOC, Op_RegF,  3, xmm3->as_VMReg()->next());
   156 reg_def XMM4   (SOC, SOC, Op_RegF,  4, xmm4->as_VMReg());
   157 reg_def XMM4_H (SOC, SOC, Op_RegF,  4, xmm4->as_VMReg()->next());
   159 reg_def XMM5   (SOC, SOC, Op_RegF,  5, xmm5->as_VMReg());
   160 reg_def XMM5_H (SOC, SOC, Op_RegF,  5, xmm5->as_VMReg()->next());
   162 #ifdef _WIN64
   164 reg_def XMM6   (SOC, SOE, Op_RegF,  6, xmm6->as_VMReg());
   165 reg_def XMM6_H (SOC, SOE, Op_RegF,  6, xmm6->as_VMReg()->next());
   167 reg_def XMM7   (SOC, SOE, Op_RegF,  7, xmm7->as_VMReg());
   168 reg_def XMM7_H (SOC, SOE, Op_RegF,  7, xmm7->as_VMReg()->next());
   170 reg_def XMM8   (SOC, SOE, Op_RegF,  8, xmm8->as_VMReg());
   171 reg_def XMM8_H (SOC, SOE, Op_RegF,  8, xmm8->as_VMReg()->next());
   173 reg_def XMM9   (SOC, SOE, Op_RegF,  9, xmm9->as_VMReg());
   174 reg_def XMM9_H (SOC, SOE, Op_RegF,  9, xmm9->as_VMReg()->next());
   176 reg_def XMM10  (SOC, SOE, Op_RegF, 10, xmm10->as_VMReg());
   177 reg_def XMM10_H(SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()->next());
   179 reg_def XMM11  (SOC, SOE, Op_RegF, 11, xmm11->as_VMReg());
   180 reg_def XMM11_H(SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()->next());
   182 reg_def XMM12  (SOC, SOE, Op_RegF, 12, xmm12->as_VMReg());
   183 reg_def XMM12_H(SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()->next());
   185 reg_def XMM13  (SOC, SOE, Op_RegF, 13, xmm13->as_VMReg());
   186 reg_def XMM13_H(SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()->next());
   188 reg_def XMM14  (SOC, SOE, Op_RegF, 14, xmm14->as_VMReg());
   189 reg_def XMM14_H(SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()->next());
   191 reg_def XMM15  (SOC, SOE, Op_RegF, 15, xmm15->as_VMReg());
   192 reg_def XMM15_H(SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()->next());
   194 #else
   196 reg_def XMM6   (SOC, SOC, Op_RegF,  6, xmm6->as_VMReg());
   197 reg_def XMM6_H (SOC, SOC, Op_RegF,  6, xmm6->as_VMReg()->next());
   199 reg_def XMM7   (SOC, SOC, Op_RegF,  7, xmm7->as_VMReg());
   200 reg_def XMM7_H (SOC, SOC, Op_RegF,  7, xmm7->as_VMReg()->next());
   202 reg_def XMM8   (SOC, SOC, Op_RegF,  8, xmm8->as_VMReg());
   203 reg_def XMM8_H (SOC, SOC, Op_RegF,  8, xmm8->as_VMReg()->next());
   205 reg_def XMM9   (SOC, SOC, Op_RegF,  9, xmm9->as_VMReg());
   206 reg_def XMM9_H (SOC, SOC, Op_RegF,  9, xmm9->as_VMReg()->next());
   208 reg_def XMM10  (SOC, SOC, Op_RegF, 10, xmm10->as_VMReg());
   209 reg_def XMM10_H(SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next());
   211 reg_def XMM11  (SOC, SOC, Op_RegF, 11, xmm11->as_VMReg());
   212 reg_def XMM11_H(SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next());
   214 reg_def XMM12  (SOC, SOC, Op_RegF, 12, xmm12->as_VMReg());
   215 reg_def XMM12_H(SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next());
   217 reg_def XMM13  (SOC, SOC, Op_RegF, 13, xmm13->as_VMReg());
   218 reg_def XMM13_H(SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next());
   220 reg_def XMM14  (SOC, SOC, Op_RegF, 14, xmm14->as_VMReg());
   221 reg_def XMM14_H(SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next());
   223 reg_def XMM15  (SOC, SOC, Op_RegF, 15, xmm15->as_VMReg());
   224 reg_def XMM15_H(SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next());
   226 #endif // _WIN64
   228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
   230 // Specify priority of register selection within phases of register
   231 // allocation.  Highest priority is first.  A useful heuristic is to
   232 // give registers a low priority when they are required by machine
   233 // instructions, like EAX and EDX on I486, and choose no-save registers
   234 // before save-on-call, & save-on-call before save-on-entry.  Registers
   235 // which participate in fixed calling sequences should come last.
   236 // Registers which are used as pairs must fall on an even boundary.
   238 alloc_class chunk0(R10,         R10_H,
   239                    R11,         R11_H,
   240                    R8,          R8_H,
   241                    R9,          R9_H,
   242                    R12,         R12_H,
   243                    RCX,         RCX_H,
   244                    RBX,         RBX_H,
   245                    RDI,         RDI_H,
   246                    RDX,         RDX_H,
   247                    RSI,         RSI_H,
   248                    RAX,         RAX_H,
   249                    RBP,         RBP_H,
   250                    R13,         R13_H,
   251                    R14,         R14_H,
   252                    R15,         R15_H,
   253                    RSP,         RSP_H);
   255 // XXX probably use 8-15 first on Linux
   256 alloc_class chunk1(XMM0,  XMM0_H,
   257                    XMM1,  XMM1_H,
   258                    XMM2,  XMM2_H,
   259                    XMM3,  XMM3_H,
   260                    XMM4,  XMM4_H,
   261                    XMM5,  XMM5_H,
   262                    XMM6,  XMM6_H,
   263                    XMM7,  XMM7_H,
   264                    XMM8,  XMM8_H,
   265                    XMM9,  XMM9_H,
   266                    XMM10, XMM10_H,
   267                    XMM11, XMM11_H,
   268                    XMM12, XMM12_H,
   269                    XMM13, XMM13_H,
   270                    XMM14, XMM14_H,
   271                    XMM15, XMM15_H);
   273 alloc_class chunk2(RFLAGS);
   276 //----------Architecture Description Register Classes--------------------------
   277 // Several register classes are automatically defined based upon information in
   278 // this architecture description.
   279 // 1) reg_class inline_cache_reg           ( /* as def'd in frame section */ )
   280 // 2) reg_class compiler_method_oop_reg    ( /* as def'd in frame section */ )
   281 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
   282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
   283 //
   285 // Class for all pointer registers (including RSP)
   286 reg_class any_reg(RAX, RAX_H,
   287                   RDX, RDX_H,
   288                   RBP, RBP_H,
   289                   RDI, RDI_H,
   290                   RSI, RSI_H,
   291                   RCX, RCX_H,
   292                   RBX, RBX_H,
   293                   RSP, RSP_H,
   294                   R8,  R8_H,
   295                   R9,  R9_H,
   296                   R10, R10_H,
   297                   R11, R11_H,
   298                   R12, R12_H,
   299                   R13, R13_H,
   300                   R14, R14_H,
   301                   R15, R15_H);
   303 // Class for all pointer registers except RSP
   304 reg_class ptr_reg(RAX, RAX_H,
   305                   RDX, RDX_H,
   306                   RBP, RBP_H,
   307                   RDI, RDI_H,
   308                   RSI, RSI_H,
   309                   RCX, RCX_H,
   310                   RBX, RBX_H,
   311                   R8,  R8_H,
   312                   R9,  R9_H,
   313                   R10, R10_H,
   314                   R11, R11_H,
   315                   R13, R13_H,
   316                   R14, R14_H);
   318 // Class for all pointer registers except RAX and RSP
   319 reg_class ptr_no_rax_reg(RDX, RDX_H,
   320                          RBP, RBP_H,
   321                          RDI, RDI_H,
   322                          RSI, RSI_H,
   323                          RCX, RCX_H,
   324                          RBX, RBX_H,
   325                          R8,  R8_H,
   326                          R9,  R9_H,
   327                          R10, R10_H,
   328                          R11, R11_H,
   329                          R13, R13_H,
   330                          R14, R14_H);
   332 reg_class ptr_no_rbp_reg(RDX, RDX_H,
   333                          RAX, RAX_H,
   334                          RDI, RDI_H,
   335                          RSI, RSI_H,
   336                          RCX, RCX_H,
   337                          RBX, RBX_H,
   338                          R8,  R8_H,
   339                          R9,  R9_H,
   340                          R10, R10_H,
   341                          R11, R11_H,
   342                          R13, R13_H,
   343                          R14, R14_H);
   345 // Class for all pointer registers except RAX, RBX and RSP
   346 reg_class ptr_no_rax_rbx_reg(RDX, RDX_H,
   347                              RBP, RBP_H,
   348                              RDI, RDI_H,
   349                              RSI, RSI_H,
   350                              RCX, RCX_H,
   351                              R8,  R8_H,
   352                              R9,  R9_H,
   353                              R10, R10_H,
   354                              R11, R11_H,
   355                              R13, R13_H,
   356                              R14, R14_H);
   358 // Singleton class for RAX pointer register
   359 reg_class ptr_rax_reg(RAX, RAX_H);
   361 // Singleton class for RBX pointer register
   362 reg_class ptr_rbx_reg(RBX, RBX_H);
   364 // Singleton class for RSI pointer register
   365 reg_class ptr_rsi_reg(RSI, RSI_H);
   367 // Singleton class for RDI pointer register
   368 reg_class ptr_rdi_reg(RDI, RDI_H);
   370 // Singleton class for RBP pointer register
   371 reg_class ptr_rbp_reg(RBP, RBP_H);
   373 // Singleton class for stack pointer
   374 reg_class ptr_rsp_reg(RSP, RSP_H);
   376 // Singleton class for TLS pointer
   377 reg_class ptr_r15_reg(R15, R15_H);
   379 // Class for all long registers (except RSP)
   380 reg_class long_reg(RAX, RAX_H,
   381                    RDX, RDX_H,
   382                    RBP, RBP_H,
   383                    RDI, RDI_H,
   384                    RSI, RSI_H,
   385                    RCX, RCX_H,
   386                    RBX, RBX_H,
   387                    R8,  R8_H,
   388                    R9,  R9_H,
   389                    R10, R10_H,
   390                    R11, R11_H,
   391                    R13, R13_H,
   392                    R14, R14_H);
   394 // Class for all long registers except RAX, RDX (and RSP)
   395 reg_class long_no_rax_rdx_reg(RBP, RBP_H,
   396                               RDI, RDI_H,
   397                               RSI, RSI_H,
   398                               RCX, RCX_H,
   399                               RBX, RBX_H,
   400                               R8,  R8_H,
   401                               R9,  R9_H,
   402                               R10, R10_H,
   403                               R11, R11_H,
   404                               R13, R13_H,
   405                               R14, R14_H);
   407 // Class for all long registers except RCX (and RSP)
   408 reg_class long_no_rcx_reg(RBP, RBP_H,
   409                           RDI, RDI_H,
   410                           RSI, RSI_H,
   411                           RAX, RAX_H,
   412                           RDX, RDX_H,
   413                           RBX, RBX_H,
   414                           R8,  R8_H,
   415                           R9,  R9_H,
   416                           R10, R10_H,
   417                           R11, R11_H,
   418                           R13, R13_H,
   419                           R14, R14_H);
   421 // Class for all long registers except RAX (and RSP)
   422 reg_class long_no_rax_reg(RBP, RBP_H,
   423                           RDX, RDX_H,
   424                           RDI, RDI_H,
   425                           RSI, RSI_H,
   426                           RCX, RCX_H,
   427                           RBX, RBX_H,
   428                           R8,  R8_H,
   429                           R9,  R9_H,
   430                           R10, R10_H,
   431                           R11, R11_H,
   432                           R13, R13_H,
   433                           R14, R14_H);
   435 // Singleton class for RAX long register
   436 reg_class long_rax_reg(RAX, RAX_H);
   438 // Singleton class for RCX long register
   439 reg_class long_rcx_reg(RCX, RCX_H);
   441 // Singleton class for RDX long register
   442 reg_class long_rdx_reg(RDX, RDX_H);
   444 // Class for all int registers (except RSP)
   445 reg_class int_reg(RAX,
   446                   RDX,
   447                   RBP,
   448                   RDI,
   449                   RSI,
   450                   RCX,
   451                   RBX,
   452                   R8,
   453                   R9,
   454                   R10,
   455                   R11,
   456                   R13,
   457                   R14);
   459 // Class for all int registers except RCX (and RSP)
   460 reg_class int_no_rcx_reg(RAX,
   461                          RDX,
   462                          RBP,
   463                          RDI,
   464                          RSI,
   465                          RBX,
   466                          R8,
   467                          R9,
   468                          R10,
   469                          R11,
   470                          R13,
   471                          R14);
   473 // Class for all int registers except RAX, RDX (and RSP)
   474 reg_class int_no_rax_rdx_reg(RBP,
   475                              RDI,
   476                              RSI,
   477                              RCX,
   478                              RBX,
   479                              R8,
   480                              R9,
   481                              R10,
   482                              R11,
   483                              R13,
   484                              R14);
   486 // Singleton class for RAX int register
   487 reg_class int_rax_reg(RAX);
   489 // Singleton class for RBX int register
   490 reg_class int_rbx_reg(RBX);
   492 // Singleton class for RCX int register
   493 reg_class int_rcx_reg(RCX);
   495 // Singleton class for RCX int register
   496 reg_class int_rdx_reg(RDX);
   498 // Singleton class for RCX int register
   499 reg_class int_rdi_reg(RDI);
   501 // Singleton class for instruction pointer
   502 // reg_class ip_reg(RIP);
   504 // Singleton class for condition codes
   505 reg_class int_flags(RFLAGS);
   507 // Class for all float registers
   508 reg_class float_reg(XMM0,
   509                     XMM1,
   510                     XMM2,
   511                     XMM3,
   512                     XMM4,
   513                     XMM5,
   514                     XMM6,
   515                     XMM7,
   516                     XMM8,
   517                     XMM9,
   518                     XMM10,
   519                     XMM11,
   520                     XMM12,
   521                     XMM13,
   522                     XMM14,
   523                     XMM15);
   525 // Class for all double registers
   526 reg_class double_reg(XMM0,  XMM0_H,
   527                      XMM1,  XMM1_H,
   528                      XMM2,  XMM2_H,
   529                      XMM3,  XMM3_H,
   530                      XMM4,  XMM4_H,
   531                      XMM5,  XMM5_H,
   532                      XMM6,  XMM6_H,
   533                      XMM7,  XMM7_H,
   534                      XMM8,  XMM8_H,
   535                      XMM9,  XMM9_H,
   536                      XMM10, XMM10_H,
   537                      XMM11, XMM11_H,
   538                      XMM12, XMM12_H,
   539                      XMM13, XMM13_H,
   540                      XMM14, XMM14_H,
   541                      XMM15, XMM15_H);
   542 %}
   545 //----------SOURCE BLOCK-------------------------------------------------------
   546 // This is a block of C++ code which provides values, functions, and
   547 // definitions necessary in the rest of the architecture description
   548 source %{
   549 #define   RELOC_IMM64    Assembler::imm_operand
   550 #define   RELOC_DISP32   Assembler::disp32_operand
   552 #define __ _masm.
   554 // !!!!! Special hack to get all types of calls to specify the byte offset
   555 //       from the start of the call to the point where the return address
   556 //       will point.
   557 int MachCallStaticJavaNode::ret_addr_offset()
   558 {
   559   return 5; // 5 bytes from start of call to where return address points
   560 }
   562 int MachCallDynamicJavaNode::ret_addr_offset()
   563 {
   564   return 15; // 15 bytes from start of call to where return address points
   565 }
   567 // In os_cpu .ad file
   568 // int MachCallRuntimeNode::ret_addr_offset()
   570 // Indicate if the safepoint node needs the polling page as an input.
   571 // Since amd64 does not have absolute addressing but RIP-relative
   572 // addressing and the polling page is within 2G, it doesn't.
   573 bool SafePointNode::needs_polling_address_input()
   574 {
   575   return false;
   576 }
   578 //
   579 // Compute padding required for nodes which need alignment
   580 //
   582 // The address of the call instruction needs to be 4-byte aligned to
   583 // ensure that it does not span a cache line so that it can be patched.
   584 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
   585 {
   586   current_offset += 1; // skip call opcode byte
   587   return round_to(current_offset, alignment_required()) - current_offset;
   588 }
   590 // The address of the call instruction needs to be 4-byte aligned to
   591 // ensure that it does not span a cache line so that it can be patched.
   592 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
   593 {
   594   current_offset += 11; // skip movq instruction + call opcode byte
   595   return round_to(current_offset, alignment_required()) - current_offset;
   596 }
   598 #ifndef PRODUCT
   599 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
   600 {
   601   st->print("INT3");
   602 }
   603 #endif
   605 // EMIT_RM()
   606 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3)
   607 {
   608   unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
   609   *(cbuf.code_end()) = c;
   610   cbuf.set_code_end(cbuf.code_end() + 1);
   611 }
   613 // EMIT_CC()
   614 void emit_cc(CodeBuffer &cbuf, int f1, int f2)
   615 {
   616   unsigned char c = (unsigned char) (f1 | f2);
   617   *(cbuf.code_end()) = c;
   618   cbuf.set_code_end(cbuf.code_end() + 1);
   619 }
   621 // EMIT_OPCODE()
   622 void emit_opcode(CodeBuffer &cbuf, int code)
   623 {
   624   *(cbuf.code_end()) = (unsigned char) code;
   625   cbuf.set_code_end(cbuf.code_end() + 1);
   626 }
   628 // EMIT_OPCODE() w/ relocation information
   629 void emit_opcode(CodeBuffer &cbuf,
   630                  int code, relocInfo::relocType reloc, int offset, int format)
   631 {
   632   cbuf.relocate(cbuf.inst_mark() + offset, reloc, format);
   633   emit_opcode(cbuf, code);
   634 }
   636 // EMIT_D8()
   637 void emit_d8(CodeBuffer &cbuf, int d8)
   638 {
   639   *(cbuf.code_end()) = (unsigned char) d8;
   640   cbuf.set_code_end(cbuf.code_end() + 1);
   641 }
   643 // EMIT_D16()
   644 void emit_d16(CodeBuffer &cbuf, int d16)
   645 {
   646   *((short *)(cbuf.code_end())) = d16;
   647   cbuf.set_code_end(cbuf.code_end() + 2);
   648 }
   650 // EMIT_D32()
   651 void emit_d32(CodeBuffer &cbuf, int d32)
   652 {
   653   *((int *)(cbuf.code_end())) = d32;
   654   cbuf.set_code_end(cbuf.code_end() + 4);
   655 }
   657 // EMIT_D64()
   658 void emit_d64(CodeBuffer &cbuf, int64_t d64)
   659 {
   660   *((int64_t*) (cbuf.code_end())) = d64;
   661   cbuf.set_code_end(cbuf.code_end() + 8);
   662 }
   664 // emit 32 bit value and construct relocation entry from relocInfo::relocType
   665 void emit_d32_reloc(CodeBuffer& cbuf,
   666                     int d32,
   667                     relocInfo::relocType reloc,
   668                     int format)
   669 {
   670   assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
   671   cbuf.relocate(cbuf.inst_mark(), reloc, format);
   673   *((int*) (cbuf.code_end())) = d32;
   674   cbuf.set_code_end(cbuf.code_end() + 4);
   675 }
   677 // emit 32 bit value and construct relocation entry from RelocationHolder
   678 void emit_d32_reloc(CodeBuffer& cbuf,
   679                     int d32,
   680                     RelocationHolder const& rspec,
   681                     int format)
   682 {
   683 #ifdef ASSERT
   684   if (rspec.reloc()->type() == relocInfo::oop_type &&
   685       d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
   686     assert(oop((intptr_t)d32)->is_oop() && oop((intptr_t)d32)->is_perm(), "cannot embed non-perm oops in code");
   687   }
   688 #endif
   689   cbuf.relocate(cbuf.inst_mark(), rspec, format);
   691   *((int* )(cbuf.code_end())) = d32;
   692   cbuf.set_code_end(cbuf.code_end() + 4);
   693 }
   695 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
   696   address next_ip = cbuf.code_end() + 4;
   697   emit_d32_reloc(cbuf, (int) (addr - next_ip),
   698                  external_word_Relocation::spec(addr),
   699                  RELOC_DISP32);
   700 }
   703 // emit 64 bit value and construct relocation entry from relocInfo::relocType
   704 void emit_d64_reloc(CodeBuffer& cbuf,
   705                     int64_t d64,
   706                     relocInfo::relocType reloc,
   707                     int format)
   708 {
   709   cbuf.relocate(cbuf.inst_mark(), reloc, format);
   711   *((int64_t*) (cbuf.code_end())) = d64;
   712   cbuf.set_code_end(cbuf.code_end() + 8);
   713 }
   715 // emit 64 bit value and construct relocation entry from RelocationHolder
   716 void emit_d64_reloc(CodeBuffer& cbuf,
   717                     int64_t d64,
   718                     RelocationHolder const& rspec,
   719                     int format)
   720 {
   721 #ifdef ASSERT
   722   if (rspec.reloc()->type() == relocInfo::oop_type &&
   723       d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
   724     assert(oop(d64)->is_oop() && oop(d64)->is_perm(),
   725            "cannot embed non-perm oops in code");
   726   }
   727 #endif
   728   cbuf.relocate(cbuf.inst_mark(), rspec, format);
   730   *((int64_t*) (cbuf.code_end())) = d64;
   731   cbuf.set_code_end(cbuf.code_end() + 8);
   732 }
   734 // Access stack slot for load or store
   735 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
   736 {
   737   emit_opcode(cbuf, opcode);                  // (e.g., FILD   [RSP+src])
   738   if (-0x80 <= disp && disp < 0x80) {
   739     emit_rm(cbuf, 0x01, rm_field, RSP_enc);   // R/M byte
   740     emit_rm(cbuf, 0x00, RSP_enc, RSP_enc);    // SIB byte
   741     emit_d8(cbuf, disp);     // Displacement  // R/M byte
   742   } else {
   743     emit_rm(cbuf, 0x02, rm_field, RSP_enc);   // R/M byte
   744     emit_rm(cbuf, 0x00, RSP_enc, RSP_enc);    // SIB byte
   745     emit_d32(cbuf, disp);     // Displacement // R/M byte
   746   }
   747 }
   749    // rRegI ereg, memory mem) %{    // emit_reg_mem
   750 void encode_RegMem(CodeBuffer &cbuf,
   751                    int reg,
   752                    int base, int index, int scale, int disp, bool disp_is_oop)
   753 {
   754   assert(!disp_is_oop, "cannot have disp");
   755   int regenc = reg & 7;
   756   int baseenc = base & 7;
   757   int indexenc = index & 7;
   759   // There is no index & no scale, use form without SIB byte
   760   if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
   761     // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
   762     if (disp == 0 && base != RBP_enc && base != R13_enc) {
   763       emit_rm(cbuf, 0x0, regenc, baseenc); // *
   764     } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
   765       // If 8-bit displacement, mode 0x1
   766       emit_rm(cbuf, 0x1, regenc, baseenc); // *
   767       emit_d8(cbuf, disp);
   768     } else {
   769       // If 32-bit displacement
   770       if (base == -1) { // Special flag for absolute address
   771         emit_rm(cbuf, 0x0, regenc, 0x5); // *
   772         if (disp_is_oop) {
   773           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
   774         } else {
   775           emit_d32(cbuf, disp);
   776         }
   777       } else {
   778         // Normal base + offset
   779         emit_rm(cbuf, 0x2, regenc, baseenc); // *
   780         if (disp_is_oop) {
   781           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
   782         } else {
   783           emit_d32(cbuf, disp);
   784         }
   785       }
   786     }
   787   } else {
   788     // Else, encode with the SIB byte
   789     // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
   790     if (disp == 0 && base != RBP_enc && base != R13_enc) {
   791       // If no displacement
   792       emit_rm(cbuf, 0x0, regenc, 0x4); // *
   793       emit_rm(cbuf, scale, indexenc, baseenc);
   794     } else {
   795       if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
   796         // If 8-bit displacement, mode 0x1
   797         emit_rm(cbuf, 0x1, regenc, 0x4); // *
   798         emit_rm(cbuf, scale, indexenc, baseenc);
   799         emit_d8(cbuf, disp);
   800       } else {
   801         // If 32-bit displacement
   802         if (base == 0x04 ) {
   803           emit_rm(cbuf, 0x2, regenc, 0x4);
   804           emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
   805         } else {
   806           emit_rm(cbuf, 0x2, regenc, 0x4);
   807           emit_rm(cbuf, scale, indexenc, baseenc); // *
   808         }
   809         if (disp_is_oop) {
   810           emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
   811         } else {
   812           emit_d32(cbuf, disp);
   813         }
   814       }
   815     }
   816   }
   817 }
   819 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
   820 {
   821   if (dstenc != srcenc) {
   822     if (dstenc < 8) {
   823       if (srcenc >= 8) {
   824         emit_opcode(cbuf, Assembler::REX_B);
   825         srcenc -= 8;
   826       }
   827     } else {
   828       if (srcenc < 8) {
   829         emit_opcode(cbuf, Assembler::REX_R);
   830       } else {
   831         emit_opcode(cbuf, Assembler::REX_RB);
   832         srcenc -= 8;
   833       }
   834       dstenc -= 8;
   835     }
   837     emit_opcode(cbuf, 0x8B);
   838     emit_rm(cbuf, 0x3, dstenc, srcenc);
   839   }
   840 }
   842 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
   843   if( dst_encoding == src_encoding ) {
   844     // reg-reg copy, use an empty encoding
   845   } else {
   846     MacroAssembler _masm(&cbuf);
   848     __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
   849   }
   850 }
   853 //=============================================================================
   854 #ifndef PRODUCT
   855 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
   856 {
   857   Compile* C = ra_->C;
   859   int framesize = C->frame_slots() << LogBytesPerInt;
   860   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
   861   // Remove wordSize for return adr already pushed
   862   // and another for the RBP we are going to save
   863   framesize -= 2*wordSize;
   864   bool need_nop = true;
   866   // Calls to C2R adapters often do not accept exceptional returns.
   867   // We require that their callers must bang for them.  But be
   868   // careful, because some VM calls (such as call site linkage) can
   869   // use several kilobytes of stack.  But the stack safety zone should
   870   // account for that.  See bugs 4446381, 4468289, 4497237.
   871   if (C->need_stack_bang(framesize)) {
   872     st->print_cr("# stack bang"); st->print("\t");
   873     need_nop = false;
   874   }
   875   st->print_cr("pushq   rbp"); st->print("\t");
   877   if (VerifyStackAtCalls) {
   878     // Majik cookie to verify stack depth
   879     st->print_cr("pushq   0xffffffffbadb100d"
   880                   "\t# Majik cookie for stack depth check");
   881     st->print("\t");
   882     framesize -= wordSize; // Remove 2 for cookie
   883     need_nop = false;
   884   }
   886   if (framesize) {
   887     st->print("subq    rsp, #%d\t# Create frame", framesize);
   888     if (framesize < 0x80 && need_nop) {
   889       st->print("\n\tnop\t# nop for patch_verified_entry");
   890     }
   891   }
   892 }
   893 #endif
   895 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
   896 {
   897   Compile* C = ra_->C;
   899   // WARNING: Initial instruction MUST be 5 bytes or longer so that
   900   // NativeJump::patch_verified_entry will be able to patch out the entry
   901   // code safely. The fldcw is ok at 6 bytes, the push to verify stack
   902   // depth is ok at 5 bytes, the frame allocation can be either 3 or
   903   // 6 bytes. So if we don't do the fldcw or the push then we must
   904   // use the 6 byte frame allocation even if we have no frame. :-(
   905   // If method sets FPU control word do it now
   907   int framesize = C->frame_slots() << LogBytesPerInt;
   908   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
   909   // Remove wordSize for return adr already pushed
   910   // and another for the RBP we are going to save
   911   framesize -= 2*wordSize;
   912   bool need_nop = true;
   914   // Calls to C2R adapters often do not accept exceptional returns.
   915   // We require that their callers must bang for them.  But be
   916   // careful, because some VM calls (such as call site linkage) can
   917   // use several kilobytes of stack.  But the stack safety zone should
   918   // account for that.  See bugs 4446381, 4468289, 4497237.
   919   if (C->need_stack_bang(framesize)) {
   920     MacroAssembler masm(&cbuf);
   921     masm.generate_stack_overflow_check(framesize);
   922     need_nop = false;
   923   }
   925   // We always push rbp so that on return to interpreter rbp will be
   926   // restored correctly and we can correct the stack.
   927   emit_opcode(cbuf, 0x50 | RBP_enc);
   929   if (VerifyStackAtCalls) {
   930     // Majik cookie to verify stack depth
   931     emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
   932     emit_d32(cbuf, 0xbadb100d);
   933     framesize -= wordSize; // Remove 2 for cookie
   934     need_nop = false;
   935   }
   937   if (framesize) {
   938     emit_opcode(cbuf, Assembler::REX_W);
   939     if (framesize < 0x80) {
   940       emit_opcode(cbuf, 0x83);   // sub  SP,#framesize
   941       emit_rm(cbuf, 0x3, 0x05, RSP_enc);
   942       emit_d8(cbuf, framesize);
   943       if (need_nop) {
   944         emit_opcode(cbuf, 0x90); // nop
   945       }
   946     } else {
   947       emit_opcode(cbuf, 0x81);   // sub  SP,#framesize
   948       emit_rm(cbuf, 0x3, 0x05, RSP_enc);
   949       emit_d32(cbuf, framesize);
   950     }
   951   }
   953   C->set_frame_complete(cbuf.code_end() - cbuf.code_begin());
   955 #ifdef ASSERT
   956   if (VerifyStackAtCalls) {
   957     Label L;
   958     MacroAssembler masm(&cbuf);
   959     masm.push(rax);
   960     masm.mov(rax, rsp);
   961     masm.andptr(rax, StackAlignmentInBytes-1);
   962     masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
   963     masm.pop(rax);
   964     masm.jcc(Assembler::equal, L);
   965     masm.stop("Stack is not properly aligned!");
   966     masm.bind(L);
   967   }
   968 #endif
   969 }
   971 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
   972 {
   973   return MachNode::size(ra_); // too many variables; just compute it
   974                               // the hard way
   975 }
   977 int MachPrologNode::reloc() const
   978 {
   979   return 0; // a large enough number
   980 }
   982 //=============================================================================
   983 #ifndef PRODUCT
   984 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
   985 {
   986   Compile* C = ra_->C;
   987   int framesize = C->frame_slots() << LogBytesPerInt;
   988   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
   989   // Remove word for return adr already pushed
   990   // and RBP
   991   framesize -= 2*wordSize;
   993   if (framesize) {
   994     st->print_cr("addq\trsp, %d\t# Destroy frame", framesize);
   995     st->print("\t");
   996   }
   998   st->print_cr("popq\trbp");
   999   if (do_polling() && C->is_method_compilation()) {
  1000     st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t"
  1001                   "# Safepoint: poll for GC");
  1002     st->print("\t");
  1005 #endif
  1007 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
  1009   Compile* C = ra_->C;
  1010   int framesize = C->frame_slots() << LogBytesPerInt;
  1011   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  1012   // Remove word for return adr already pushed
  1013   // and RBP
  1014   framesize -= 2*wordSize;
  1016   // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
  1018   if (framesize) {
  1019     emit_opcode(cbuf, Assembler::REX_W);
  1020     if (framesize < 0x80) {
  1021       emit_opcode(cbuf, 0x83); // addq rsp, #framesize
  1022       emit_rm(cbuf, 0x3, 0x00, RSP_enc);
  1023       emit_d8(cbuf, framesize);
  1024     } else {
  1025       emit_opcode(cbuf, 0x81); // addq rsp, #framesize
  1026       emit_rm(cbuf, 0x3, 0x00, RSP_enc);
  1027       emit_d32(cbuf, framesize);
  1031   // popq rbp
  1032   emit_opcode(cbuf, 0x58 | RBP_enc);
  1034   if (do_polling() && C->is_method_compilation()) {
  1035     // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
  1036     // XXX reg_mem doesn't support RIP-relative addressing yet
  1037     cbuf.set_inst_mark();
  1038     cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX
  1039     emit_opcode(cbuf, 0x85); // testl
  1040     emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
  1041     // cbuf.inst_mark() is beginning of instruction
  1042     emit_d32_reloc(cbuf, os::get_polling_page());
  1043 //                    relocInfo::poll_return_type,
  1047 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
  1049   Compile* C = ra_->C;
  1050   int framesize = C->frame_slots() << LogBytesPerInt;
  1051   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
  1052   // Remove word for return adr already pushed
  1053   // and RBP
  1054   framesize -= 2*wordSize;
  1056   uint size = 0;
  1058   if (do_polling() && C->is_method_compilation()) {
  1059     size += 6;
  1062   // count popq rbp
  1063   size++;
  1065   if (framesize) {
  1066     if (framesize < 0x80) {
  1067       size += 4;
  1068     } else if (framesize) {
  1069       size += 7;
  1073   return size;
  1076 int MachEpilogNode::reloc() const
  1078   return 2; // a large enough number
  1081 const Pipeline* MachEpilogNode::pipeline() const
  1083   return MachNode::pipeline_class();
  1086 int MachEpilogNode::safepoint_offset() const
  1088   return 0;
  1091 //=============================================================================
  1093 enum RC {
  1094   rc_bad,
  1095   rc_int,
  1096   rc_float,
  1097   rc_stack
  1098 };
  1100 static enum RC rc_class(OptoReg::Name reg)
  1102   if( !OptoReg::is_valid(reg)  ) return rc_bad;
  1104   if (OptoReg::is_stack(reg)) return rc_stack;
  1106   VMReg r = OptoReg::as_VMReg(reg);
  1108   if (r->is_Register()) return rc_int;
  1110   assert(r->is_XMMRegister(), "must be");
  1111   return rc_float;
  1114 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
  1115                                        PhaseRegAlloc* ra_,
  1116                                        bool do_size,
  1117                                        outputStream* st) const
  1120   // Get registers to move
  1121   OptoReg::Name src_second = ra_->get_reg_second(in(1));
  1122   OptoReg::Name src_first = ra_->get_reg_first(in(1));
  1123   OptoReg::Name dst_second = ra_->get_reg_second(this);
  1124   OptoReg::Name dst_first = ra_->get_reg_first(this);
  1126   enum RC src_second_rc = rc_class(src_second);
  1127   enum RC src_first_rc = rc_class(src_first);
  1128   enum RC dst_second_rc = rc_class(dst_second);
  1129   enum RC dst_first_rc = rc_class(dst_first);
  1131   assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
  1132          "must move at least 1 register" );
  1134   if (src_first == dst_first && src_second == dst_second) {
  1135     // Self copy, no move
  1136     return 0;
  1137   } else if (src_first_rc == rc_stack) {
  1138     // mem ->
  1139     if (dst_first_rc == rc_stack) {
  1140       // mem -> mem
  1141       assert(src_second != dst_first, "overlap");
  1142       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1143           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1144         // 64-bit
  1145         int src_offset = ra_->reg2offset(src_first);
  1146         int dst_offset = ra_->reg2offset(dst_first);
  1147         if (cbuf) {
  1148           emit_opcode(*cbuf, 0xFF);
  1149           encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
  1151           emit_opcode(*cbuf, 0x8F);
  1152           encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
  1154 #ifndef PRODUCT
  1155         } else if (!do_size) {
  1156           st->print("pushq   [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
  1157                      "popq    [rsp + #%d]",
  1158                      src_offset,
  1159                      dst_offset);
  1160 #endif
  1162         return
  1163           3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
  1164           3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
  1165       } else {
  1166         // 32-bit
  1167         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1168         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1169         // No pushl/popl, so:
  1170         int src_offset = ra_->reg2offset(src_first);
  1171         int dst_offset = ra_->reg2offset(dst_first);
  1172         if (cbuf) {
  1173           emit_opcode(*cbuf, Assembler::REX_W);
  1174           emit_opcode(*cbuf, 0x89);
  1175           emit_opcode(*cbuf, 0x44);
  1176           emit_opcode(*cbuf, 0x24);
  1177           emit_opcode(*cbuf, 0xF8);
  1179           emit_opcode(*cbuf, 0x8B);
  1180           encode_RegMem(*cbuf,
  1181                         RAX_enc,
  1182                         RSP_enc, 0x4, 0, src_offset,
  1183                         false);
  1185           emit_opcode(*cbuf, 0x89);
  1186           encode_RegMem(*cbuf,
  1187                         RAX_enc,
  1188                         RSP_enc, 0x4, 0, dst_offset,
  1189                         false);
  1191           emit_opcode(*cbuf, Assembler::REX_W);
  1192           emit_opcode(*cbuf, 0x8B);
  1193           emit_opcode(*cbuf, 0x44);
  1194           emit_opcode(*cbuf, 0x24);
  1195           emit_opcode(*cbuf, 0xF8);
  1197 #ifndef PRODUCT
  1198         } else if (!do_size) {
  1199           st->print("movq    [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
  1200                      "movl    rax, [rsp + #%d]\n\t"
  1201                      "movl    [rsp + #%d], rax\n\t"
  1202                      "movq    rax, [rsp - #8]",
  1203                      src_offset,
  1204                      dst_offset);
  1205 #endif
  1207         return
  1208           5 + // movq
  1209           3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
  1210           3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
  1211           5; // movq
  1213     } else if (dst_first_rc == rc_int) {
  1214       // mem -> gpr
  1215       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1216           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1217         // 64-bit
  1218         int offset = ra_->reg2offset(src_first);
  1219         if (cbuf) {
  1220           if (Matcher::_regEncode[dst_first] < 8) {
  1221             emit_opcode(*cbuf, Assembler::REX_W);
  1222           } else {
  1223             emit_opcode(*cbuf, Assembler::REX_WR);
  1225           emit_opcode(*cbuf, 0x8B);
  1226           encode_RegMem(*cbuf,
  1227                         Matcher::_regEncode[dst_first],
  1228                         RSP_enc, 0x4, 0, offset,
  1229                         false);
  1230 #ifndef PRODUCT
  1231         } else if (!do_size) {
  1232           st->print("movq    %s, [rsp + #%d]\t# spill",
  1233                      Matcher::regName[dst_first],
  1234                      offset);
  1235 #endif
  1237         return
  1238           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
  1239       } else {
  1240         // 32-bit
  1241         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1242         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1243         int offset = ra_->reg2offset(src_first);
  1244         if (cbuf) {
  1245           if (Matcher::_regEncode[dst_first] >= 8) {
  1246             emit_opcode(*cbuf, Assembler::REX_R);
  1248           emit_opcode(*cbuf, 0x8B);
  1249           encode_RegMem(*cbuf,
  1250                         Matcher::_regEncode[dst_first],
  1251                         RSP_enc, 0x4, 0, offset,
  1252                         false);
  1253 #ifndef PRODUCT
  1254         } else if (!do_size) {
  1255           st->print("movl    %s, [rsp + #%d]\t# spill",
  1256                      Matcher::regName[dst_first],
  1257                      offset);
  1258 #endif
  1260         return
  1261           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1262           ((Matcher::_regEncode[dst_first] < 8)
  1263            ? 3
  1264            : 4); // REX
  1266     } else if (dst_first_rc == rc_float) {
  1267       // mem-> xmm
  1268       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1269           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1270         // 64-bit
  1271         int offset = ra_->reg2offset(src_first);
  1272         if (cbuf) {
  1273           emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
  1274           if (Matcher::_regEncode[dst_first] >= 8) {
  1275             emit_opcode(*cbuf, Assembler::REX_R);
  1277           emit_opcode(*cbuf, 0x0F);
  1278           emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
  1279           encode_RegMem(*cbuf,
  1280                         Matcher::_regEncode[dst_first],
  1281                         RSP_enc, 0x4, 0, offset,
  1282                         false);
  1283 #ifndef PRODUCT
  1284         } else if (!do_size) {
  1285           st->print("%s  %s, [rsp + #%d]\t# spill",
  1286                      UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
  1287                      Matcher::regName[dst_first],
  1288                      offset);
  1289 #endif
  1291         return
  1292           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1293           ((Matcher::_regEncode[dst_first] < 8)
  1294            ? 5
  1295            : 6); // REX
  1296       } else {
  1297         // 32-bit
  1298         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1299         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1300         int offset = ra_->reg2offset(src_first);
  1301         if (cbuf) {
  1302           emit_opcode(*cbuf, 0xF3);
  1303           if (Matcher::_regEncode[dst_first] >= 8) {
  1304             emit_opcode(*cbuf, Assembler::REX_R);
  1306           emit_opcode(*cbuf, 0x0F);
  1307           emit_opcode(*cbuf, 0x10);
  1308           encode_RegMem(*cbuf,
  1309                         Matcher::_regEncode[dst_first],
  1310                         RSP_enc, 0x4, 0, offset,
  1311                         false);
  1312 #ifndef PRODUCT
  1313         } else if (!do_size) {
  1314           st->print("movss   %s, [rsp + #%d]\t# spill",
  1315                      Matcher::regName[dst_first],
  1316                      offset);
  1317 #endif
  1319         return
  1320           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1321           ((Matcher::_regEncode[dst_first] < 8)
  1322            ? 5
  1323            : 6); // REX
  1326   } else if (src_first_rc == rc_int) {
  1327     // gpr ->
  1328     if (dst_first_rc == rc_stack) {
  1329       // gpr -> mem
  1330       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1331           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1332         // 64-bit
  1333         int offset = ra_->reg2offset(dst_first);
  1334         if (cbuf) {
  1335           if (Matcher::_regEncode[src_first] < 8) {
  1336             emit_opcode(*cbuf, Assembler::REX_W);
  1337           } else {
  1338             emit_opcode(*cbuf, Assembler::REX_WR);
  1340           emit_opcode(*cbuf, 0x89);
  1341           encode_RegMem(*cbuf,
  1342                         Matcher::_regEncode[src_first],
  1343                         RSP_enc, 0x4, 0, offset,
  1344                         false);
  1345 #ifndef PRODUCT
  1346         } else if (!do_size) {
  1347           st->print("movq    [rsp + #%d], %s\t# spill",
  1348                      offset,
  1349                      Matcher::regName[src_first]);
  1350 #endif
  1352         return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
  1353       } else {
  1354         // 32-bit
  1355         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1356         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1357         int offset = ra_->reg2offset(dst_first);
  1358         if (cbuf) {
  1359           if (Matcher::_regEncode[src_first] >= 8) {
  1360             emit_opcode(*cbuf, Assembler::REX_R);
  1362           emit_opcode(*cbuf, 0x89);
  1363           encode_RegMem(*cbuf,
  1364                         Matcher::_regEncode[src_first],
  1365                         RSP_enc, 0x4, 0, offset,
  1366                         false);
  1367 #ifndef PRODUCT
  1368         } else if (!do_size) {
  1369           st->print("movl    [rsp + #%d], %s\t# spill",
  1370                      offset,
  1371                      Matcher::regName[src_first]);
  1372 #endif
  1374         return
  1375           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1376           ((Matcher::_regEncode[src_first] < 8)
  1377            ? 3
  1378            : 4); // REX
  1380     } else if (dst_first_rc == rc_int) {
  1381       // gpr -> gpr
  1382       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1383           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1384         // 64-bit
  1385         if (cbuf) {
  1386           if (Matcher::_regEncode[dst_first] < 8) {
  1387             if (Matcher::_regEncode[src_first] < 8) {
  1388               emit_opcode(*cbuf, Assembler::REX_W);
  1389             } else {
  1390               emit_opcode(*cbuf, Assembler::REX_WB);
  1392           } else {
  1393             if (Matcher::_regEncode[src_first] < 8) {
  1394               emit_opcode(*cbuf, Assembler::REX_WR);
  1395             } else {
  1396               emit_opcode(*cbuf, Assembler::REX_WRB);
  1399           emit_opcode(*cbuf, 0x8B);
  1400           emit_rm(*cbuf, 0x3,
  1401                   Matcher::_regEncode[dst_first] & 7,
  1402                   Matcher::_regEncode[src_first] & 7);
  1403 #ifndef PRODUCT
  1404         } else if (!do_size) {
  1405           st->print("movq    %s, %s\t# spill",
  1406                      Matcher::regName[dst_first],
  1407                      Matcher::regName[src_first]);
  1408 #endif
  1410         return 3; // REX
  1411       } else {
  1412         // 32-bit
  1413         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1414         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1415         if (cbuf) {
  1416           if (Matcher::_regEncode[dst_first] < 8) {
  1417             if (Matcher::_regEncode[src_first] >= 8) {
  1418               emit_opcode(*cbuf, Assembler::REX_B);
  1420           } else {
  1421             if (Matcher::_regEncode[src_first] < 8) {
  1422               emit_opcode(*cbuf, Assembler::REX_R);
  1423             } else {
  1424               emit_opcode(*cbuf, Assembler::REX_RB);
  1427           emit_opcode(*cbuf, 0x8B);
  1428           emit_rm(*cbuf, 0x3,
  1429                   Matcher::_regEncode[dst_first] & 7,
  1430                   Matcher::_regEncode[src_first] & 7);
  1431 #ifndef PRODUCT
  1432         } else if (!do_size) {
  1433           st->print("movl    %s, %s\t# spill",
  1434                      Matcher::regName[dst_first],
  1435                      Matcher::regName[src_first]);
  1436 #endif
  1438         return
  1439           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
  1440           ? 2
  1441           : 3; // REX
  1443     } else if (dst_first_rc == rc_float) {
  1444       // gpr -> xmm
  1445       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1446           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1447         // 64-bit
  1448         if (cbuf) {
  1449           emit_opcode(*cbuf, 0x66);
  1450           if (Matcher::_regEncode[dst_first] < 8) {
  1451             if (Matcher::_regEncode[src_first] < 8) {
  1452               emit_opcode(*cbuf, Assembler::REX_W);
  1453             } else {
  1454               emit_opcode(*cbuf, Assembler::REX_WB);
  1456           } else {
  1457             if (Matcher::_regEncode[src_first] < 8) {
  1458               emit_opcode(*cbuf, Assembler::REX_WR);
  1459             } else {
  1460               emit_opcode(*cbuf, Assembler::REX_WRB);
  1463           emit_opcode(*cbuf, 0x0F);
  1464           emit_opcode(*cbuf, 0x6E);
  1465           emit_rm(*cbuf, 0x3,
  1466                   Matcher::_regEncode[dst_first] & 7,
  1467                   Matcher::_regEncode[src_first] & 7);
  1468 #ifndef PRODUCT
  1469         } else if (!do_size) {
  1470           st->print("movdq   %s, %s\t# spill",
  1471                      Matcher::regName[dst_first],
  1472                      Matcher::regName[src_first]);
  1473 #endif
  1475         return 5; // REX
  1476       } else {
  1477         // 32-bit
  1478         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1479         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1480         if (cbuf) {
  1481           emit_opcode(*cbuf, 0x66);
  1482           if (Matcher::_regEncode[dst_first] < 8) {
  1483             if (Matcher::_regEncode[src_first] >= 8) {
  1484               emit_opcode(*cbuf, Assembler::REX_B);
  1486           } else {
  1487             if (Matcher::_regEncode[src_first] < 8) {
  1488               emit_opcode(*cbuf, Assembler::REX_R);
  1489             } else {
  1490               emit_opcode(*cbuf, Assembler::REX_RB);
  1493           emit_opcode(*cbuf, 0x0F);
  1494           emit_opcode(*cbuf, 0x6E);
  1495           emit_rm(*cbuf, 0x3,
  1496                   Matcher::_regEncode[dst_first] & 7,
  1497                   Matcher::_regEncode[src_first] & 7);
  1498 #ifndef PRODUCT
  1499         } else if (!do_size) {
  1500           st->print("movdl   %s, %s\t# spill",
  1501                      Matcher::regName[dst_first],
  1502                      Matcher::regName[src_first]);
  1503 #endif
  1505         return
  1506           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
  1507           ? 4
  1508           : 5; // REX
  1511   } else if (src_first_rc == rc_float) {
  1512     // xmm ->
  1513     if (dst_first_rc == rc_stack) {
  1514       // xmm -> mem
  1515       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1516           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1517         // 64-bit
  1518         int offset = ra_->reg2offset(dst_first);
  1519         if (cbuf) {
  1520           emit_opcode(*cbuf, 0xF2);
  1521           if (Matcher::_regEncode[src_first] >= 8) {
  1522               emit_opcode(*cbuf, Assembler::REX_R);
  1524           emit_opcode(*cbuf, 0x0F);
  1525           emit_opcode(*cbuf, 0x11);
  1526           encode_RegMem(*cbuf,
  1527                         Matcher::_regEncode[src_first],
  1528                         RSP_enc, 0x4, 0, offset,
  1529                         false);
  1530 #ifndef PRODUCT
  1531         } else if (!do_size) {
  1532           st->print("movsd   [rsp + #%d], %s\t# spill",
  1533                      offset,
  1534                      Matcher::regName[src_first]);
  1535 #endif
  1537         return
  1538           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1539           ((Matcher::_regEncode[src_first] < 8)
  1540            ? 5
  1541            : 6); // REX
  1542       } else {
  1543         // 32-bit
  1544         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1545         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1546         int offset = ra_->reg2offset(dst_first);
  1547         if (cbuf) {
  1548           emit_opcode(*cbuf, 0xF3);
  1549           if (Matcher::_regEncode[src_first] >= 8) {
  1550               emit_opcode(*cbuf, Assembler::REX_R);
  1552           emit_opcode(*cbuf, 0x0F);
  1553           emit_opcode(*cbuf, 0x11);
  1554           encode_RegMem(*cbuf,
  1555                         Matcher::_regEncode[src_first],
  1556                         RSP_enc, 0x4, 0, offset,
  1557                         false);
  1558 #ifndef PRODUCT
  1559         } else if (!do_size) {
  1560           st->print("movss   [rsp + #%d], %s\t# spill",
  1561                      offset,
  1562                      Matcher::regName[src_first]);
  1563 #endif
  1565         return
  1566           ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
  1567           ((Matcher::_regEncode[src_first] < 8)
  1568            ? 5
  1569            : 6); // REX
  1571     } else if (dst_first_rc == rc_int) {
  1572       // xmm -> gpr
  1573       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1574           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1575         // 64-bit
  1576         if (cbuf) {
  1577           emit_opcode(*cbuf, 0x66);
  1578           if (Matcher::_regEncode[dst_first] < 8) {
  1579             if (Matcher::_regEncode[src_first] < 8) {
  1580               emit_opcode(*cbuf, Assembler::REX_W);
  1581             } else {
  1582               emit_opcode(*cbuf, Assembler::REX_WR); // attention!
  1584           } else {
  1585             if (Matcher::_regEncode[src_first] < 8) {
  1586               emit_opcode(*cbuf, Assembler::REX_WB); // attention!
  1587             } else {
  1588               emit_opcode(*cbuf, Assembler::REX_WRB);
  1591           emit_opcode(*cbuf, 0x0F);
  1592           emit_opcode(*cbuf, 0x7E);
  1593           emit_rm(*cbuf, 0x3,
  1594                   Matcher::_regEncode[dst_first] & 7,
  1595                   Matcher::_regEncode[src_first] & 7);
  1596 #ifndef PRODUCT
  1597         } else if (!do_size) {
  1598           st->print("movdq   %s, %s\t# spill",
  1599                      Matcher::regName[dst_first],
  1600                      Matcher::regName[src_first]);
  1601 #endif
  1603         return 5; // REX
  1604       } else {
  1605         // 32-bit
  1606         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1607         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1608         if (cbuf) {
  1609           emit_opcode(*cbuf, 0x66);
  1610           if (Matcher::_regEncode[dst_first] < 8) {
  1611             if (Matcher::_regEncode[src_first] >= 8) {
  1612               emit_opcode(*cbuf, Assembler::REX_R); // attention!
  1614           } else {
  1615             if (Matcher::_regEncode[src_first] < 8) {
  1616               emit_opcode(*cbuf, Assembler::REX_B); // attention!
  1617             } else {
  1618               emit_opcode(*cbuf, Assembler::REX_RB);
  1621           emit_opcode(*cbuf, 0x0F);
  1622           emit_opcode(*cbuf, 0x7E);
  1623           emit_rm(*cbuf, 0x3,
  1624                   Matcher::_regEncode[dst_first] & 7,
  1625                   Matcher::_regEncode[src_first] & 7);
  1626 #ifndef PRODUCT
  1627         } else if (!do_size) {
  1628           st->print("movdl   %s, %s\t# spill",
  1629                      Matcher::regName[dst_first],
  1630                      Matcher::regName[src_first]);
  1631 #endif
  1633         return
  1634           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
  1635           ? 4
  1636           : 5; // REX
  1638     } else if (dst_first_rc == rc_float) {
  1639       // xmm -> xmm
  1640       if ((src_first & 1) == 0 && src_first + 1 == src_second &&
  1641           (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
  1642         // 64-bit
  1643         if (cbuf) {
  1644           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
  1645           if (Matcher::_regEncode[dst_first] < 8) {
  1646             if (Matcher::_regEncode[src_first] >= 8) {
  1647               emit_opcode(*cbuf, Assembler::REX_B);
  1649           } else {
  1650             if (Matcher::_regEncode[src_first] < 8) {
  1651               emit_opcode(*cbuf, Assembler::REX_R);
  1652             } else {
  1653               emit_opcode(*cbuf, Assembler::REX_RB);
  1656           emit_opcode(*cbuf, 0x0F);
  1657           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
  1658           emit_rm(*cbuf, 0x3,
  1659                   Matcher::_regEncode[dst_first] & 7,
  1660                   Matcher::_regEncode[src_first] & 7);
  1661 #ifndef PRODUCT
  1662         } else if (!do_size) {
  1663           st->print("%s  %s, %s\t# spill",
  1664                      UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
  1665                      Matcher::regName[dst_first],
  1666                      Matcher::regName[src_first]);
  1667 #endif
  1669         return
  1670           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
  1671           ? 4
  1672           : 5; // REX
  1673       } else {
  1674         // 32-bit
  1675         assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
  1676         assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
  1677         if (cbuf) {
  1678           if (!UseXmmRegToRegMoveAll)
  1679             emit_opcode(*cbuf, 0xF3);
  1680           if (Matcher::_regEncode[dst_first] < 8) {
  1681             if (Matcher::_regEncode[src_first] >= 8) {
  1682               emit_opcode(*cbuf, Assembler::REX_B);
  1684           } else {
  1685             if (Matcher::_regEncode[src_first] < 8) {
  1686               emit_opcode(*cbuf, Assembler::REX_R);
  1687             } else {
  1688               emit_opcode(*cbuf, Assembler::REX_RB);
  1691           emit_opcode(*cbuf, 0x0F);
  1692           emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
  1693           emit_rm(*cbuf, 0x3,
  1694                   Matcher::_regEncode[dst_first] & 7,
  1695                   Matcher::_regEncode[src_first] & 7);
  1696 #ifndef PRODUCT
  1697         } else if (!do_size) {
  1698           st->print("%s  %s, %s\t# spill",
  1699                      UseXmmRegToRegMoveAll ? "movaps" : "movss ",
  1700                      Matcher::regName[dst_first],
  1701                      Matcher::regName[src_first]);
  1702 #endif
  1704         return
  1705           (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
  1706           ? (UseXmmRegToRegMoveAll ? 3 : 4)
  1707           : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
  1712   assert(0," foo ");
  1713   Unimplemented();
  1715   return 0;
  1718 #ifndef PRODUCT
  1719 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
  1721   implementation(NULL, ra_, false, st);
  1723 #endif
  1725 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
  1727   implementation(&cbuf, ra_, false, NULL);
  1730 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
  1732   return implementation(NULL, ra_, true, NULL);
  1735 //=============================================================================
  1736 #ifndef PRODUCT
  1737 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
  1739   st->print("nop \t# %d bytes pad for loops and calls", _count);
  1741 #endif
  1743 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
  1745   MacroAssembler _masm(&cbuf);
  1746   __ nop(_count);
  1749 uint MachNopNode::size(PhaseRegAlloc*) const
  1751   return _count;
  1755 //=============================================================================
  1756 #ifndef PRODUCT
  1757 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
  1759   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  1760   int reg = ra_->get_reg_first(this);
  1761   st->print("leaq    %s, [rsp + #%d]\t# box lock",
  1762             Matcher::regName[reg], offset);
  1764 #endif
  1766 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
  1768   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  1769   int reg = ra_->get_encode(this);
  1770   if (offset >= 0x80) {
  1771     emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
  1772     emit_opcode(cbuf, 0x8D); // LEA  reg,[SP+offset]
  1773     emit_rm(cbuf, 0x2, reg & 7, 0x04);
  1774     emit_rm(cbuf, 0x0, 0x04, RSP_enc);
  1775     emit_d32(cbuf, offset);
  1776   } else {
  1777     emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
  1778     emit_opcode(cbuf, 0x8D); // LEA  reg,[SP+offset]
  1779     emit_rm(cbuf, 0x1, reg & 7, 0x04);
  1780     emit_rm(cbuf, 0x0, 0x04, RSP_enc);
  1781     emit_d8(cbuf, offset);
  1785 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
  1787   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
  1788   return (offset < 0x80) ? 5 : 8; // REX
  1791 //=============================================================================
  1793 // emit call stub, compiled java to interpreter
  1794 void emit_java_to_interp(CodeBuffer& cbuf)
  1796   // Stub is fixed up when the corresponding call is converted from
  1797   // calling compiled code to calling interpreted code.
  1798   // movq rbx, 0
  1799   // jmp -5 # to self
  1801   address mark = cbuf.inst_mark();  // get mark within main instrs section
  1803   // Note that the code buffer's inst_mark is always relative to insts.
  1804   // That's why we must use the macroassembler to generate a stub.
  1805   MacroAssembler _masm(&cbuf);
  1807   address base =
  1808   __ start_a_stub(Compile::MAX_stubs_size);
  1809   if (base == NULL)  return;  // CodeBuffer::expand failed
  1810   // static stub relocation stores the instruction address of the call
  1811   __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
  1812   // static stub relocation also tags the methodOop in the code-stream.
  1813   __ movoop(rbx, (jobject) NULL);  // method is zapped till fixup time
  1814   // This is recognized as unresolved by relocs/nativeinst/ic code
  1815   __ jump(RuntimeAddress(__ pc()));
  1817   // Update current stubs pointer and restore code_end.
  1818   __ end_a_stub();
  1821 // size of call stub, compiled java to interpretor
  1822 uint size_java_to_interp()
  1824   return 15;  // movq (1+1+8); jmp (1+4)
  1827 // relocation entries for call stub, compiled java to interpretor
  1828 uint reloc_java_to_interp()
  1830   return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
  1833 //=============================================================================
  1834 #ifndef PRODUCT
  1835 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
  1837   if (UseCompressedOops) {
  1838     st->print_cr("movl    rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
  1839     if (Universe::narrow_oop_shift() != 0) {
  1840       st->print_cr("leaq    rscratch1, [r12_heapbase, r, Address::times_8, 0]");
  1842     st->print_cr("cmpq    rax, rscratch1\t # Inline cache check");
  1843   } else {
  1844     st->print_cr("cmpq    rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
  1845                  "# Inline cache check", oopDesc::klass_offset_in_bytes());
  1847   st->print_cr("\tjne     SharedRuntime::_ic_miss_stub");
  1848   st->print_cr("\tnop");
  1849   if (!OptoBreakpoint) {
  1850     st->print_cr("\tnop");
  1853 #endif
  1855 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
  1857   MacroAssembler masm(&cbuf);
  1858 #ifdef ASSERT
  1859   uint code_size = cbuf.code_size();
  1860 #endif
  1861   if (UseCompressedOops) {
  1862     masm.load_klass(rscratch1, j_rarg0);
  1863     masm.cmpptr(rax, rscratch1);
  1864   } else {
  1865     masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
  1868   masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  1870   /* WARNING these NOPs are critical so that verified entry point is properly
  1871      aligned for patching by NativeJump::patch_verified_entry() */
  1872   int nops_cnt = 1;
  1873   if (!OptoBreakpoint) {
  1874     // Leave space for int3
  1875      nops_cnt += 1;
  1877   if (UseCompressedOops) {
  1878     // ??? divisible by 4 is aligned?
  1879     nops_cnt += 1;
  1881   masm.nop(nops_cnt);
  1883   assert(cbuf.code_size() - code_size == size(ra_),
  1884          "checking code size of inline cache node");
  1887 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
  1889   if (UseCompressedOops) {
  1890     if (Universe::narrow_oop_shift() == 0) {
  1891       return OptoBreakpoint ? 15 : 16;
  1892     } else {
  1893       return OptoBreakpoint ? 19 : 20;
  1895   } else {
  1896     return OptoBreakpoint ? 11 : 12;
  1901 //=============================================================================
  1902 uint size_exception_handler()
  1904   // NativeCall instruction size is the same as NativeJump.
  1905   // Note that this value is also credited (in output.cpp) to
  1906   // the size of the code section.
  1907   return NativeJump::instruction_size;
  1910 // Emit exception handler code.
  1911 int emit_exception_handler(CodeBuffer& cbuf)
  1914   // Note that the code buffer's inst_mark is always relative to insts.
  1915   // That's why we must use the macroassembler to generate a handler.
  1916   MacroAssembler _masm(&cbuf);
  1917   address base =
  1918   __ start_a_stub(size_exception_handler());
  1919   if (base == NULL)  return 0;  // CodeBuffer::expand failed
  1920   int offset = __ offset();
  1921   __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
  1922   assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
  1923   __ end_a_stub();
  1924   return offset;
  1927 uint size_deopt_handler()
  1929   // three 5 byte instructions
  1930   return 15;
  1933 // Emit deopt handler code.
  1934 int emit_deopt_handler(CodeBuffer& cbuf)
  1937   // Note that the code buffer's inst_mark is always relative to insts.
  1938   // That's why we must use the macroassembler to generate a handler.
  1939   MacroAssembler _masm(&cbuf);
  1940   address base =
  1941   __ start_a_stub(size_deopt_handler());
  1942   if (base == NULL)  return 0;  // CodeBuffer::expand failed
  1943   int offset = __ offset();
  1944   address the_pc = (address) __ pc();
  1945   Label next;
  1946   // push a "the_pc" on the stack without destroying any registers
  1947   // as they all may be live.
  1949   // push address of "next"
  1950   __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
  1951   __ bind(next);
  1952   // adjust it so it matches "the_pc"
  1953   __ subptr(Address(rsp, 0), __ offset() - offset);
  1954   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
  1955   assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
  1956   __ end_a_stub();
  1957   return offset;
  1960 static void emit_double_constant(CodeBuffer& cbuf, double x) {
  1961   int mark = cbuf.insts()->mark_off();
  1962   MacroAssembler _masm(&cbuf);
  1963   address double_address = __ double_constant(x);
  1964   cbuf.insts()->set_mark_off(mark);  // preserve mark across masm shift
  1965   emit_d32_reloc(cbuf,
  1966                  (int) (double_address - cbuf.code_end() - 4),
  1967                  internal_word_Relocation::spec(double_address),
  1968                  RELOC_DISP32);
  1971 static void emit_float_constant(CodeBuffer& cbuf, float x) {
  1972   int mark = cbuf.insts()->mark_off();
  1973   MacroAssembler _masm(&cbuf);
  1974   address float_address = __ float_constant(x);
  1975   cbuf.insts()->set_mark_off(mark);  // preserve mark across masm shift
  1976   emit_d32_reloc(cbuf,
  1977                  (int) (float_address - cbuf.code_end() - 4),
  1978                  internal_word_Relocation::spec(float_address),
  1979                  RELOC_DISP32);
  1983 const bool Matcher::match_rule_supported(int opcode) {
  1984   if (!has_match_rule(opcode))
  1985     return false;
  1987   return true;  // Per default match rules are supported.
  1990 int Matcher::regnum_to_fpu_offset(int regnum)
  1992   return regnum - 32; // The FP registers are in the second chunk
  1995 // This is UltraSparc specific, true just means we have fast l2f conversion
  1996 const bool Matcher::convL2FSupported(void) {
  1997   return true;
  2000 // Vector width in bytes
  2001 const uint Matcher::vector_width_in_bytes(void) {
  2002   return 8;
  2005 // Vector ideal reg
  2006 const uint Matcher::vector_ideal_reg(void) {
  2007   return Op_RegD;
  2010 // Is this branch offset short enough that a short branch can be used?
  2011 //
  2012 // NOTE: If the platform does not provide any short branch variants, then
  2013 //       this method should return false for offset 0.
  2014 bool Matcher::is_short_branch_offset(int rule, int offset) {
  2015   // the short version of jmpConUCF2 contains multiple branches,
  2016   // making the reach slightly less
  2017   if (rule == jmpConUCF2_rule)
  2018     return (-126 <= offset && offset <= 125);
  2019   return (-128 <= offset && offset <= 127);
  2022 const bool Matcher::isSimpleConstant64(jlong value) {
  2023   // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
  2024   //return value == (int) value;  // Cf. storeImmL and immL32.
  2026   // Probably always true, even if a temp register is required.
  2027   return true;
  2030 // The ecx parameter to rep stosq for the ClearArray node is in words.
  2031 const bool Matcher::init_array_count_is_in_bytes = false;
  2033 // Threshold size for cleararray.
  2034 const int Matcher::init_array_short_size = 8 * BytesPerLong;
  2036 // Should the Matcher clone shifts on addressing modes, expecting them
  2037 // to be subsumed into complex addressing expressions or compute them
  2038 // into registers?  True for Intel but false for most RISCs
  2039 const bool Matcher::clone_shift_expressions = true;
  2041 // Is it better to copy float constants, or load them directly from
  2042 // memory?  Intel can load a float constant from a direct address,
  2043 // requiring no extra registers.  Most RISCs will have to materialize
  2044 // an address into a register first, so they would do better to copy
  2045 // the constant from stack.
  2046 const bool Matcher::rematerialize_float_constants = true; // XXX
  2048 // If CPU can load and store mis-aligned doubles directly then no
  2049 // fixup is needed.  Else we split the double into 2 integer pieces
  2050 // and move it piece-by-piece.  Only happens when passing doubles into
  2051 // C code as the Java calling convention forces doubles to be aligned.
  2052 const bool Matcher::misaligned_doubles_ok = true;
  2054 // No-op on amd64
  2055 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
  2057 // Advertise here if the CPU requires explicit rounding operations to
  2058 // implement the UseStrictFP mode.
  2059 const bool Matcher::strict_fp_requires_explicit_rounding = true;
  2061 // Do floats take an entire double register or just half?
  2062 const bool Matcher::float_in_double = true;
  2063 // Do ints take an entire long register or just half?
  2064 const bool Matcher::int_in_long = true;
  2066 // Return whether or not this register is ever used as an argument.
  2067 // This function is used on startup to build the trampoline stubs in
  2068 // generateOptoStub.  Registers not mentioned will be killed by the VM
  2069 // call in the trampoline, and arguments in those registers not be
  2070 // available to the callee.
  2071 bool Matcher::can_be_java_arg(int reg)
  2073   return
  2074     reg ==  RDI_num || reg ==  RDI_H_num ||
  2075     reg ==  RSI_num || reg ==  RSI_H_num ||
  2076     reg ==  RDX_num || reg ==  RDX_H_num ||
  2077     reg ==  RCX_num || reg ==  RCX_H_num ||
  2078     reg ==   R8_num || reg ==   R8_H_num ||
  2079     reg ==   R9_num || reg ==   R9_H_num ||
  2080     reg ==  R12_num || reg ==  R12_H_num ||
  2081     reg == XMM0_num || reg == XMM0_H_num ||
  2082     reg == XMM1_num || reg == XMM1_H_num ||
  2083     reg == XMM2_num || reg == XMM2_H_num ||
  2084     reg == XMM3_num || reg == XMM3_H_num ||
  2085     reg == XMM4_num || reg == XMM4_H_num ||
  2086     reg == XMM5_num || reg == XMM5_H_num ||
  2087     reg == XMM6_num || reg == XMM6_H_num ||
  2088     reg == XMM7_num || reg == XMM7_H_num;
  2091 bool Matcher::is_spillable_arg(int reg)
  2093   return can_be_java_arg(reg);
  2096 // Register for DIVI projection of divmodI
  2097 RegMask Matcher::divI_proj_mask() {
  2098   return INT_RAX_REG_mask;
  2101 // Register for MODI projection of divmodI
  2102 RegMask Matcher::modI_proj_mask() {
  2103   return INT_RDX_REG_mask;
  2106 // Register for DIVL projection of divmodL
  2107 RegMask Matcher::divL_proj_mask() {
  2108   return LONG_RAX_REG_mask;
  2111 // Register for MODL projection of divmodL
  2112 RegMask Matcher::modL_proj_mask() {
  2113   return LONG_RDX_REG_mask;
  2116 static Address build_address(int b, int i, int s, int d) {
  2117   Register index = as_Register(i);
  2118   Address::ScaleFactor scale = (Address::ScaleFactor)s;
  2119   if (index == rsp) {
  2120     index = noreg;
  2121     scale = Address::no_scale;
  2123   Address addr(as_Register(b), index, scale, d);
  2124   return addr;
  2127 %}
  2129 //----------ENCODING BLOCK-----------------------------------------------------
  2130 // This block specifies the encoding classes used by the compiler to
  2131 // output byte streams.  Encoding classes are parameterized macros
  2132 // used by Machine Instruction Nodes in order to generate the bit
  2133 // encoding of the instruction.  Operands specify their base encoding
  2134 // interface with the interface keyword.  There are currently
  2135 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
  2136 // COND_INTER.  REG_INTER causes an operand to generate a function
  2137 // which returns its register number when queried.  CONST_INTER causes
  2138 // an operand to generate a function which returns the value of the
  2139 // constant when queried.  MEMORY_INTER causes an operand to generate
  2140 // four functions which return the Base Register, the Index Register,
  2141 // the Scale Value, and the Offset Value of the operand when queried.
  2142 // COND_INTER causes an operand to generate six functions which return
  2143 // the encoding code (ie - encoding bits for the instruction)
  2144 // associated with each basic boolean condition for a conditional
  2145 // instruction.
  2146 //
  2147 // Instructions specify two basic values for encoding.  Again, a
  2148 // function is available to check if the constant displacement is an
  2149 // oop. They use the ins_encode keyword to specify their encoding
  2150 // classes (which must be a sequence of enc_class names, and their
  2151 // parameters, specified in the encoding block), and they use the
  2152 // opcode keyword to specify, in order, their primary, secondary, and
  2153 // tertiary opcode.  Only the opcode sections which a particular
  2154 // instruction needs for encoding need to be specified.
  2155 encode %{
  2156   // Build emit functions for each basic byte or larger field in the
  2157   // intel encoding scheme (opcode, rm, sib, immediate), and call them
  2158   // from C++ code in the enc_class source block.  Emit functions will
  2159   // live in the main source block for now.  In future, we can
  2160   // generalize this by adding a syntax that specifies the sizes of
  2161   // fields in an order, so that the adlc can build the emit functions
  2162   // automagically
  2164   // Emit primary opcode
  2165   enc_class OpcP
  2166   %{
  2167     emit_opcode(cbuf, $primary);
  2168   %}
  2170   // Emit secondary opcode
  2171   enc_class OpcS
  2172   %{
  2173     emit_opcode(cbuf, $secondary);
  2174   %}
  2176   // Emit tertiary opcode
  2177   enc_class OpcT
  2178   %{
  2179     emit_opcode(cbuf, $tertiary);
  2180   %}
  2182   // Emit opcode directly
  2183   enc_class Opcode(immI d8)
  2184   %{
  2185     emit_opcode(cbuf, $d8$$constant);
  2186   %}
  2188   // Emit size prefix
  2189   enc_class SizePrefix
  2190   %{
  2191     emit_opcode(cbuf, 0x66);
  2192   %}
  2194   enc_class reg(rRegI reg)
  2195   %{
  2196     emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
  2197   %}
  2199   enc_class reg_reg(rRegI dst, rRegI src)
  2200   %{
  2201     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
  2202   %}
  2204   enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
  2205   %{
  2206     emit_opcode(cbuf, $opcode$$constant);
  2207     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
  2208   %}
  2210   enc_class cmpfp_fixup()
  2211   %{
  2212     // jnp,s exit
  2213     emit_opcode(cbuf, 0x7B);
  2214     emit_d8(cbuf, 0x0A);
  2216     // pushfq
  2217     emit_opcode(cbuf, 0x9C);
  2219     // andq $0xffffff2b, (%rsp)
  2220     emit_opcode(cbuf, Assembler::REX_W);
  2221     emit_opcode(cbuf, 0x81);
  2222     emit_opcode(cbuf, 0x24);
  2223     emit_opcode(cbuf, 0x24);
  2224     emit_d32(cbuf, 0xffffff2b);
  2226     // popfq
  2227     emit_opcode(cbuf, 0x9D);
  2229     // nop (target for branch to avoid branch to branch)
  2230     emit_opcode(cbuf, 0x90);
  2231   %}
  2233   enc_class cmpfp3(rRegI dst)
  2234   %{
  2235     int dstenc = $dst$$reg;
  2237     // movl $dst, -1
  2238     if (dstenc >= 8) {
  2239       emit_opcode(cbuf, Assembler::REX_B);
  2241     emit_opcode(cbuf, 0xB8 | (dstenc & 7));
  2242     emit_d32(cbuf, -1);
  2244     // jp,s done
  2245     emit_opcode(cbuf, 0x7A);
  2246     emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
  2248     // jb,s done
  2249     emit_opcode(cbuf, 0x72);
  2250     emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
  2252     // setne $dst
  2253     if (dstenc >= 4) {
  2254       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
  2256     emit_opcode(cbuf, 0x0F);
  2257     emit_opcode(cbuf, 0x95);
  2258     emit_opcode(cbuf, 0xC0 | (dstenc & 7));
  2260     // movzbl $dst, $dst
  2261     if (dstenc >= 4) {
  2262       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
  2264     emit_opcode(cbuf, 0x0F);
  2265     emit_opcode(cbuf, 0xB6);
  2266     emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
  2267   %}
  2269   enc_class cdql_enc(no_rax_rdx_RegI div)
  2270   %{
  2271     // Full implementation of Java idiv and irem; checks for
  2272     // special case as described in JVM spec., p.243 & p.271.
  2273     //
  2274     //         normal case                           special case
  2275     //
  2276     // input : rax: dividend                         min_int
  2277     //         reg: divisor                          -1
  2278     //
  2279     // output: rax: quotient  (= rax idiv reg)       min_int
  2280     //         rdx: remainder (= rax irem reg)       0
  2281     //
  2282     //  Code sequnce:
  2283     //
  2284     //    0:   3d 00 00 00 80          cmp    $0x80000000,%eax
  2285     //    5:   75 07/08                jne    e <normal>
  2286     //    7:   33 d2                   xor    %edx,%edx
  2287     //  [div >= 8 -> offset + 1]
  2288     //  [REX_B]
  2289     //    9:   83 f9 ff                cmp    $0xffffffffffffffff,$div
  2290     //    c:   74 03/04                je     11 <done>
  2291     // 000000000000000e <normal>:
  2292     //    e:   99                      cltd
  2293     //  [div >= 8 -> offset + 1]
  2294     //  [REX_B]
  2295     //    f:   f7 f9                   idiv   $div
  2296     // 0000000000000011 <done>:
  2298     // cmp    $0x80000000,%eax
  2299     emit_opcode(cbuf, 0x3d);
  2300     emit_d8(cbuf, 0x00);
  2301     emit_d8(cbuf, 0x00);
  2302     emit_d8(cbuf, 0x00);
  2303     emit_d8(cbuf, 0x80);
  2305     // jne    e <normal>
  2306     emit_opcode(cbuf, 0x75);
  2307     emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
  2309     // xor    %edx,%edx
  2310     emit_opcode(cbuf, 0x33);
  2311     emit_d8(cbuf, 0xD2);
  2313     // cmp    $0xffffffffffffffff,%ecx
  2314     if ($div$$reg >= 8) {
  2315       emit_opcode(cbuf, Assembler::REX_B);
  2317     emit_opcode(cbuf, 0x83);
  2318     emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
  2319     emit_d8(cbuf, 0xFF);
  2321     // je     11 <done>
  2322     emit_opcode(cbuf, 0x74);
  2323     emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
  2325     // <normal>
  2326     // cltd
  2327     emit_opcode(cbuf, 0x99);
  2329     // idivl (note: must be emitted by the user of this rule)
  2330     // <done>
  2331   %}
  2333   enc_class cdqq_enc(no_rax_rdx_RegL div)
  2334   %{
  2335     // Full implementation of Java ldiv and lrem; checks for
  2336     // special case as described in JVM spec., p.243 & p.271.
  2337     //
  2338     //         normal case                           special case
  2339     //
  2340     // input : rax: dividend                         min_long
  2341     //         reg: divisor                          -1
  2342     //
  2343     // output: rax: quotient  (= rax idiv reg)       min_long
  2344     //         rdx: remainder (= rax irem reg)       0
  2345     //
  2346     //  Code sequnce:
  2347     //
  2348     //    0:   48 ba 00 00 00 00 00    mov    $0x8000000000000000,%rdx
  2349     //    7:   00 00 80
  2350     //    a:   48 39 d0                cmp    %rdx,%rax
  2351     //    d:   75 08                   jne    17 <normal>
  2352     //    f:   33 d2                   xor    %edx,%edx
  2353     //   11:   48 83 f9 ff             cmp    $0xffffffffffffffff,$div
  2354     //   15:   74 05                   je     1c <done>
  2355     // 0000000000000017 <normal>:
  2356     //   17:   48 99                   cqto
  2357     //   19:   48 f7 f9                idiv   $div
  2358     // 000000000000001c <done>:
  2360     // mov    $0x8000000000000000,%rdx
  2361     emit_opcode(cbuf, Assembler::REX_W);
  2362     emit_opcode(cbuf, 0xBA);
  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, 0x00);
  2369     emit_d8(cbuf, 0x00);
  2370     emit_d8(cbuf, 0x80);
  2372     // cmp    %rdx,%rax
  2373     emit_opcode(cbuf, Assembler::REX_W);
  2374     emit_opcode(cbuf, 0x39);
  2375     emit_d8(cbuf, 0xD0);
  2377     // jne    17 <normal>
  2378     emit_opcode(cbuf, 0x75);
  2379     emit_d8(cbuf, 0x08);
  2381     // xor    %edx,%edx
  2382     emit_opcode(cbuf, 0x33);
  2383     emit_d8(cbuf, 0xD2);
  2385     // cmp    $0xffffffffffffffff,$div
  2386     emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
  2387     emit_opcode(cbuf, 0x83);
  2388     emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
  2389     emit_d8(cbuf, 0xFF);
  2391     // je     1e <done>
  2392     emit_opcode(cbuf, 0x74);
  2393     emit_d8(cbuf, 0x05);
  2395     // <normal>
  2396     // cqto
  2397     emit_opcode(cbuf, Assembler::REX_W);
  2398     emit_opcode(cbuf, 0x99);
  2400     // idivq (note: must be emitted by the user of this rule)
  2401     // <done>
  2402   %}
  2404   // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
  2405   enc_class OpcSE(immI imm)
  2406   %{
  2407     // Emit primary opcode and set sign-extend bit
  2408     // Check for 8-bit immediate, and set sign extend bit in opcode
  2409     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
  2410       emit_opcode(cbuf, $primary | 0x02);
  2411     } else {
  2412       // 32-bit immediate
  2413       emit_opcode(cbuf, $primary);
  2415   %}
  2417   enc_class OpcSErm(rRegI dst, immI imm)
  2418   %{
  2419     // OpcSEr/m
  2420     int dstenc = $dst$$reg;
  2421     if (dstenc >= 8) {
  2422       emit_opcode(cbuf, Assembler::REX_B);
  2423       dstenc -= 8;
  2425     // Emit primary opcode and set sign-extend bit
  2426     // Check for 8-bit immediate, and set sign extend bit in opcode
  2427     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
  2428       emit_opcode(cbuf, $primary | 0x02);
  2429     } else {
  2430       // 32-bit immediate
  2431       emit_opcode(cbuf, $primary);
  2433     // Emit r/m byte with secondary opcode, after primary opcode.
  2434     emit_rm(cbuf, 0x3, $secondary, dstenc);
  2435   %}
  2437   enc_class OpcSErm_wide(rRegL dst, immI imm)
  2438   %{
  2439     // OpcSEr/m
  2440     int dstenc = $dst$$reg;
  2441     if (dstenc < 8) {
  2442       emit_opcode(cbuf, Assembler::REX_W);
  2443     } else {
  2444       emit_opcode(cbuf, Assembler::REX_WB);
  2445       dstenc -= 8;
  2447     // Emit primary opcode and set sign-extend bit
  2448     // Check for 8-bit immediate, and set sign extend bit in opcode
  2449     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
  2450       emit_opcode(cbuf, $primary | 0x02);
  2451     } else {
  2452       // 32-bit immediate
  2453       emit_opcode(cbuf, $primary);
  2455     // Emit r/m byte with secondary opcode, after primary opcode.
  2456     emit_rm(cbuf, 0x3, $secondary, dstenc);
  2457   %}
  2459   enc_class Con8or32(immI imm)
  2460   %{
  2461     // Check for 8-bit immediate, and set sign extend bit in opcode
  2462     if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
  2463       $$$emit8$imm$$constant;
  2464     } else {
  2465       // 32-bit immediate
  2466       $$$emit32$imm$$constant;
  2468   %}
  2470   enc_class Lbl(label labl)
  2471   %{
  2472     // JMP, CALL
  2473     Label* l = $labl$$label;
  2474     emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
  2475   %}
  2477   enc_class LblShort(label labl)
  2478   %{
  2479     // JMP, CALL
  2480     Label* l = $labl$$label;
  2481     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
  2482     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
  2483     emit_d8(cbuf, disp);
  2484   %}
  2486   enc_class opc2_reg(rRegI dst)
  2487   %{
  2488     // BSWAP
  2489     emit_cc(cbuf, $secondary, $dst$$reg);
  2490   %}
  2492   enc_class opc3_reg(rRegI dst)
  2493   %{
  2494     // BSWAP
  2495     emit_cc(cbuf, $tertiary, $dst$$reg);
  2496   %}
  2498   enc_class reg_opc(rRegI div)
  2499   %{
  2500     // INC, DEC, IDIV, IMOD, JMP indirect, ...
  2501     emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
  2502   %}
  2504   enc_class Jcc(cmpOp cop, label labl)
  2505   %{
  2506     // JCC
  2507     Label* l = $labl$$label;
  2508     $$$emit8$primary;
  2509     emit_cc(cbuf, $secondary, $cop$$cmpcode);
  2510     emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
  2511   %}
  2513   enc_class JccShort (cmpOp cop, label labl)
  2514   %{
  2515   // JCC
  2516     Label *l = $labl$$label;
  2517     emit_cc(cbuf, $primary, $cop$$cmpcode);
  2518     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
  2519     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
  2520     emit_d8(cbuf, disp);
  2521   %}
  2523   enc_class enc_cmov(cmpOp cop)
  2524   %{
  2525     // CMOV
  2526     $$$emit8$primary;
  2527     emit_cc(cbuf, $secondary, $cop$$cmpcode);
  2528   %}
  2530   enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
  2531   %{
  2532     // Invert sense of branch from sense of cmov
  2533     emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
  2534     emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
  2535                   ? (UseXmmRegToRegMoveAll ? 3 : 4)
  2536                   : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
  2537     // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
  2538     if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
  2539     if ($dst$$reg < 8) {
  2540       if ($src$$reg >= 8) {
  2541         emit_opcode(cbuf, Assembler::REX_B);
  2543     } else {
  2544       if ($src$$reg < 8) {
  2545         emit_opcode(cbuf, Assembler::REX_R);
  2546       } else {
  2547         emit_opcode(cbuf, Assembler::REX_RB);
  2550     emit_opcode(cbuf, 0x0F);
  2551     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
  2552     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
  2553   %}
  2555   enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
  2556   %{
  2557     // Invert sense of branch from sense of cmov
  2558     emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
  2559     emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
  2561     //  UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
  2562     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
  2563     if ($dst$$reg < 8) {
  2564       if ($src$$reg >= 8) {
  2565         emit_opcode(cbuf, Assembler::REX_B);
  2567     } else {
  2568       if ($src$$reg < 8) {
  2569         emit_opcode(cbuf, Assembler::REX_R);
  2570       } else {
  2571         emit_opcode(cbuf, Assembler::REX_RB);
  2574     emit_opcode(cbuf, 0x0F);
  2575     emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
  2576     emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
  2577   %}
  2579   enc_class enc_PartialSubtypeCheck()
  2580   %{
  2581     Register Rrdi = as_Register(RDI_enc); // result register
  2582     Register Rrax = as_Register(RAX_enc); // super class
  2583     Register Rrcx = as_Register(RCX_enc); // killed
  2584     Register Rrsi = as_Register(RSI_enc); // sub class
  2585     Label miss;
  2586     const bool set_cond_codes = true;
  2588     MacroAssembler _masm(&cbuf);
  2589     __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi,
  2590                                      NULL, &miss,
  2591                                      /*set_cond_codes:*/ true);
  2592     if ($primary) {
  2593       __ xorptr(Rrdi, Rrdi);
  2595     __ bind(miss);
  2596   %}
  2598   enc_class Java_To_Interpreter(method meth)
  2599   %{
  2600     // CALL Java_To_Interpreter
  2601     // This is the instruction starting address for relocation info.
  2602     cbuf.set_inst_mark();
  2603     $$$emit8$primary;
  2604     // CALL directly to the runtime
  2605     emit_d32_reloc(cbuf,
  2606                    (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
  2607                    runtime_call_Relocation::spec(),
  2608                    RELOC_DISP32);
  2609   %}
  2611   enc_class Java_Static_Call(method meth)
  2612   %{
  2613     // JAVA STATIC CALL
  2614     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to
  2615     // determine who we intended to call.
  2616     cbuf.set_inst_mark();
  2617     $$$emit8$primary;
  2619     if (!_method) {
  2620       emit_d32_reloc(cbuf,
  2621                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
  2622                      runtime_call_Relocation::spec(),
  2623                      RELOC_DISP32);
  2624     } else if (_optimized_virtual) {
  2625       emit_d32_reloc(cbuf,
  2626                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
  2627                      opt_virtual_call_Relocation::spec(),
  2628                      RELOC_DISP32);
  2629     } else {
  2630       emit_d32_reloc(cbuf,
  2631                      (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
  2632                      static_call_Relocation::spec(),
  2633                      RELOC_DISP32);
  2635     if (_method) {
  2636       // Emit stub for static call
  2637       emit_java_to_interp(cbuf);
  2639   %}
  2641   enc_class Java_Dynamic_Call(method meth)
  2642   %{
  2643     // JAVA DYNAMIC CALL
  2644     // !!!!!
  2645     // Generate  "movq rax, -1", placeholder instruction to load oop-info
  2646     // emit_call_dynamic_prologue( cbuf );
  2647     cbuf.set_inst_mark();
  2649     // movq rax, -1
  2650     emit_opcode(cbuf, Assembler::REX_W);
  2651     emit_opcode(cbuf, 0xB8 | RAX_enc);
  2652     emit_d64_reloc(cbuf,
  2653                    (int64_t) Universe::non_oop_word(),
  2654                    oop_Relocation::spec_for_immediate(), RELOC_IMM64);
  2655     address virtual_call_oop_addr = cbuf.inst_mark();
  2656     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
  2657     // who we intended to call.
  2658     cbuf.set_inst_mark();
  2659     $$$emit8$primary;
  2660     emit_d32_reloc(cbuf,
  2661                    (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
  2662                    virtual_call_Relocation::spec(virtual_call_oop_addr),
  2663                    RELOC_DISP32);
  2664   %}
  2666   enc_class Java_Compiled_Call(method meth)
  2667   %{
  2668     // JAVA COMPILED CALL
  2669     int disp = in_bytes(methodOopDesc:: from_compiled_offset());
  2671     // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
  2672     // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
  2674     // callq *disp(%rax)
  2675     cbuf.set_inst_mark();
  2676     $$$emit8$primary;
  2677     if (disp < 0x80) {
  2678       emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
  2679       emit_d8(cbuf, disp); // Displacement
  2680     } else {
  2681       emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
  2682       emit_d32(cbuf, disp); // Displacement
  2684   %}
  2686   enc_class reg_opc_imm(rRegI dst, immI8 shift)
  2687   %{
  2688     // SAL, SAR, SHR
  2689     int dstenc = $dst$$reg;
  2690     if (dstenc >= 8) {
  2691       emit_opcode(cbuf, Assembler::REX_B);
  2692       dstenc -= 8;
  2694     $$$emit8$primary;
  2695     emit_rm(cbuf, 0x3, $secondary, dstenc);
  2696     $$$emit8$shift$$constant;
  2697   %}
  2699   enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
  2700   %{
  2701     // SAL, SAR, SHR
  2702     int dstenc = $dst$$reg;
  2703     if (dstenc < 8) {
  2704       emit_opcode(cbuf, Assembler::REX_W);
  2705     } else {
  2706       emit_opcode(cbuf, Assembler::REX_WB);
  2707       dstenc -= 8;
  2709     $$$emit8$primary;
  2710     emit_rm(cbuf, 0x3, $secondary, dstenc);
  2711     $$$emit8$shift$$constant;
  2712   %}
  2714   enc_class load_immI(rRegI dst, immI src)
  2715   %{
  2716     int dstenc = $dst$$reg;
  2717     if (dstenc >= 8) {
  2718       emit_opcode(cbuf, Assembler::REX_B);
  2719       dstenc -= 8;
  2721     emit_opcode(cbuf, 0xB8 | dstenc);
  2722     $$$emit32$src$$constant;
  2723   %}
  2725   enc_class load_immL(rRegL dst, immL src)
  2726   %{
  2727     int dstenc = $dst$$reg;
  2728     if (dstenc < 8) {
  2729       emit_opcode(cbuf, Assembler::REX_W);
  2730     } else {
  2731       emit_opcode(cbuf, Assembler::REX_WB);
  2732       dstenc -= 8;
  2734     emit_opcode(cbuf, 0xB8 | dstenc);
  2735     emit_d64(cbuf, $src$$constant);
  2736   %}
  2738   enc_class load_immUL32(rRegL dst, immUL32 src)
  2739   %{
  2740     // same as load_immI, but this time we care about zeroes in the high word
  2741     int dstenc = $dst$$reg;
  2742     if (dstenc >= 8) {
  2743       emit_opcode(cbuf, Assembler::REX_B);
  2744       dstenc -= 8;
  2746     emit_opcode(cbuf, 0xB8 | dstenc);
  2747     $$$emit32$src$$constant;
  2748   %}
  2750   enc_class load_immL32(rRegL dst, immL32 src)
  2751   %{
  2752     int dstenc = $dst$$reg;
  2753     if (dstenc < 8) {
  2754       emit_opcode(cbuf, Assembler::REX_W);
  2755     } else {
  2756       emit_opcode(cbuf, Assembler::REX_WB);
  2757       dstenc -= 8;
  2759     emit_opcode(cbuf, 0xC7);
  2760     emit_rm(cbuf, 0x03, 0x00, dstenc);
  2761     $$$emit32$src$$constant;
  2762   %}
  2764   enc_class load_immP31(rRegP dst, immP32 src)
  2765   %{
  2766     // same as load_immI, but this time we care about zeroes in the high word
  2767     int dstenc = $dst$$reg;
  2768     if (dstenc >= 8) {
  2769       emit_opcode(cbuf, Assembler::REX_B);
  2770       dstenc -= 8;
  2772     emit_opcode(cbuf, 0xB8 | dstenc);
  2773     $$$emit32$src$$constant;
  2774   %}
  2776   enc_class load_immP(rRegP dst, immP src)
  2777   %{
  2778     int dstenc = $dst$$reg;
  2779     if (dstenc < 8) {
  2780       emit_opcode(cbuf, Assembler::REX_W);
  2781     } else {
  2782       emit_opcode(cbuf, Assembler::REX_WB);
  2783       dstenc -= 8;
  2785     emit_opcode(cbuf, 0xB8 | dstenc);
  2786     // This next line should be generated from ADLC
  2787     if ($src->constant_is_oop()) {
  2788       emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
  2789     } else {
  2790       emit_d64(cbuf, $src$$constant);
  2792   %}
  2794   enc_class load_immF(regF dst, immF con)
  2795   %{
  2796     // XXX reg_mem doesn't support RIP-relative addressing yet
  2797     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
  2798     emit_float_constant(cbuf, $con$$constant);
  2799   %}
  2801   enc_class load_immD(regD dst, immD con)
  2802   %{
  2803     // XXX reg_mem doesn't support RIP-relative addressing yet
  2804     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
  2805     emit_double_constant(cbuf, $con$$constant);
  2806   %}
  2808   enc_class load_conF (regF dst, immF con) %{    // Load float constant
  2809     emit_opcode(cbuf, 0xF3);
  2810     if ($dst$$reg >= 8) {
  2811       emit_opcode(cbuf, Assembler::REX_R);
  2813     emit_opcode(cbuf, 0x0F);
  2814     emit_opcode(cbuf, 0x10);
  2815     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
  2816     emit_float_constant(cbuf, $con$$constant);
  2817   %}
  2819   enc_class load_conD (regD dst, immD con) %{    // Load double constant
  2820     // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
  2821     emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
  2822     if ($dst$$reg >= 8) {
  2823       emit_opcode(cbuf, Assembler::REX_R);
  2825     emit_opcode(cbuf, 0x0F);
  2826     emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
  2827     emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
  2828     emit_double_constant(cbuf, $con$$constant);
  2829   %}
  2831   // Encode a reg-reg copy.  If it is useless, then empty encoding.
  2832   enc_class enc_copy(rRegI dst, rRegI src)
  2833   %{
  2834     encode_copy(cbuf, $dst$$reg, $src$$reg);
  2835   %}
  2837   // Encode xmm reg-reg copy.  If it is useless, then empty encoding.
  2838   enc_class enc_CopyXD( RegD dst, RegD src ) %{
  2839     encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
  2840   %}
  2842   enc_class enc_copy_always(rRegI dst, rRegI src)
  2843   %{
  2844     int srcenc = $src$$reg;
  2845     int dstenc = $dst$$reg;
  2847     if (dstenc < 8) {
  2848       if (srcenc >= 8) {
  2849         emit_opcode(cbuf, Assembler::REX_B);
  2850         srcenc -= 8;
  2852     } else {
  2853       if (srcenc < 8) {
  2854         emit_opcode(cbuf, Assembler::REX_R);
  2855       } else {
  2856         emit_opcode(cbuf, Assembler::REX_RB);
  2857         srcenc -= 8;
  2859       dstenc -= 8;
  2862     emit_opcode(cbuf, 0x8B);
  2863     emit_rm(cbuf, 0x3, dstenc, srcenc);
  2864   %}
  2866   enc_class enc_copy_wide(rRegL dst, rRegL src)
  2867   %{
  2868     int srcenc = $src$$reg;
  2869     int dstenc = $dst$$reg;
  2871     if (dstenc != srcenc) {
  2872       if (dstenc < 8) {
  2873         if (srcenc < 8) {
  2874           emit_opcode(cbuf, Assembler::REX_W);
  2875         } else {
  2876           emit_opcode(cbuf, Assembler::REX_WB);
  2877           srcenc -= 8;
  2879       } else {
  2880         if (srcenc < 8) {
  2881           emit_opcode(cbuf, Assembler::REX_WR);
  2882         } else {
  2883           emit_opcode(cbuf, Assembler::REX_WRB);
  2884           srcenc -= 8;
  2886         dstenc -= 8;
  2888       emit_opcode(cbuf, 0x8B);
  2889       emit_rm(cbuf, 0x3, dstenc, srcenc);
  2891   %}
  2893   enc_class Con32(immI src)
  2894   %{
  2895     // Output immediate
  2896     $$$emit32$src$$constant;
  2897   %}
  2899   enc_class Con64(immL src)
  2900   %{
  2901     // Output immediate
  2902     emit_d64($src$$constant);
  2903   %}
  2905   enc_class Con32F_as_bits(immF src)
  2906   %{
  2907     // Output Float immediate bits
  2908     jfloat jf = $src$$constant;
  2909     jint jf_as_bits = jint_cast(jf);
  2910     emit_d32(cbuf, jf_as_bits);
  2911   %}
  2913   enc_class Con16(immI src)
  2914   %{
  2915     // Output immediate
  2916     $$$emit16$src$$constant;
  2917   %}
  2919   // How is this different from Con32??? XXX
  2920   enc_class Con_d32(immI src)
  2921   %{
  2922     emit_d32(cbuf,$src$$constant);
  2923   %}
  2925   enc_class conmemref (rRegP t1) %{    // Con32(storeImmI)
  2926     // Output immediate memory reference
  2927     emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
  2928     emit_d32(cbuf, 0x00);
  2929   %}
  2931   enc_class jump_enc(rRegL switch_val, rRegI dest) %{
  2932     MacroAssembler masm(&cbuf);
  2934     Register switch_reg = as_Register($switch_val$$reg);
  2935     Register dest_reg   = as_Register($dest$$reg);
  2936     address table_base  = masm.address_table_constant(_index2label);
  2938     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
  2939     // to do that and the compiler is using that register as one it can allocate.
  2940     // So we build it all by hand.
  2941     // Address index(noreg, switch_reg, Address::times_1);
  2942     // ArrayAddress dispatch(table, index);
  2944     Address dispatch(dest_reg, switch_reg, Address::times_1);
  2946     masm.lea(dest_reg, InternalAddress(table_base));
  2947     masm.jmp(dispatch);
  2948   %}
  2950   enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
  2951     MacroAssembler masm(&cbuf);
  2953     Register switch_reg = as_Register($switch_val$$reg);
  2954     Register dest_reg   = as_Register($dest$$reg);
  2955     address table_base  = masm.address_table_constant(_index2label);
  2957     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
  2958     // to do that and the compiler is using that register as one it can allocate.
  2959     // So we build it all by hand.
  2960     // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
  2961     // ArrayAddress dispatch(table, index);
  2963     Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
  2965     masm.lea(dest_reg, InternalAddress(table_base));
  2966     masm.jmp(dispatch);
  2967   %}
  2969   enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
  2970     MacroAssembler masm(&cbuf);
  2972     Register switch_reg = as_Register($switch_val$$reg);
  2973     Register dest_reg   = as_Register($dest$$reg);
  2974     address table_base  = masm.address_table_constant(_index2label);
  2976     // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
  2977     // to do that and the compiler is using that register as one it can allocate.
  2978     // So we build it all by hand.
  2979     // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
  2980     // ArrayAddress dispatch(table, index);
  2982     Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
  2983     masm.lea(dest_reg, InternalAddress(table_base));
  2984     masm.jmp(dispatch);
  2986   %}
  2988   enc_class lock_prefix()
  2989   %{
  2990     if (os::is_MP()) {
  2991       emit_opcode(cbuf, 0xF0); // lock
  2993   %}
  2995   enc_class REX_mem(memory mem)
  2996   %{
  2997     if ($mem$$base >= 8) {
  2998       if ($mem$$index < 8) {
  2999         emit_opcode(cbuf, Assembler::REX_B);
  3000       } else {
  3001         emit_opcode(cbuf, Assembler::REX_XB);
  3003     } else {
  3004       if ($mem$$index >= 8) {
  3005         emit_opcode(cbuf, Assembler::REX_X);
  3008   %}
  3010   enc_class REX_mem_wide(memory mem)
  3011   %{
  3012     if ($mem$$base >= 8) {
  3013       if ($mem$$index < 8) {
  3014         emit_opcode(cbuf, Assembler::REX_WB);
  3015       } else {
  3016         emit_opcode(cbuf, Assembler::REX_WXB);
  3018     } else {
  3019       if ($mem$$index < 8) {
  3020         emit_opcode(cbuf, Assembler::REX_W);
  3021       } else {
  3022         emit_opcode(cbuf, Assembler::REX_WX);
  3025   %}
  3027   // for byte regs
  3028   enc_class REX_breg(rRegI reg)
  3029   %{
  3030     if ($reg$$reg >= 4) {
  3031       emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
  3033   %}
  3035   // for byte regs
  3036   enc_class REX_reg_breg(rRegI dst, rRegI src)
  3037   %{
  3038     if ($dst$$reg < 8) {
  3039       if ($src$$reg >= 4) {
  3040         emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
  3042     } else {
  3043       if ($src$$reg < 8) {
  3044         emit_opcode(cbuf, Assembler::REX_R);
  3045       } else {
  3046         emit_opcode(cbuf, Assembler::REX_RB);
  3049   %}
  3051   // for byte regs
  3052   enc_class REX_breg_mem(rRegI reg, memory mem)
  3053   %{
  3054     if ($reg$$reg < 8) {
  3055       if ($mem$$base < 8) {
  3056         if ($mem$$index >= 8) {
  3057           emit_opcode(cbuf, Assembler::REX_X);
  3058         } else if ($reg$$reg >= 4) {
  3059           emit_opcode(cbuf, Assembler::REX);
  3061       } else {
  3062         if ($mem$$index < 8) {
  3063           emit_opcode(cbuf, Assembler::REX_B);
  3064         } else {
  3065           emit_opcode(cbuf, Assembler::REX_XB);
  3068     } else {
  3069       if ($mem$$base < 8) {
  3070         if ($mem$$index < 8) {
  3071           emit_opcode(cbuf, Assembler::REX_R);
  3072         } else {
  3073           emit_opcode(cbuf, Assembler::REX_RX);
  3075       } else {
  3076         if ($mem$$index < 8) {
  3077           emit_opcode(cbuf, Assembler::REX_RB);
  3078         } else {
  3079           emit_opcode(cbuf, Assembler::REX_RXB);
  3083   %}
  3085   enc_class REX_reg(rRegI reg)
  3086   %{
  3087     if ($reg$$reg >= 8) {
  3088       emit_opcode(cbuf, Assembler::REX_B);
  3090   %}
  3092   enc_class REX_reg_wide(rRegI reg)
  3093   %{
  3094     if ($reg$$reg < 8) {
  3095       emit_opcode(cbuf, Assembler::REX_W);
  3096     } else {
  3097       emit_opcode(cbuf, Assembler::REX_WB);
  3099   %}
  3101   enc_class REX_reg_reg(rRegI dst, rRegI src)
  3102   %{
  3103     if ($dst$$reg < 8) {
  3104       if ($src$$reg >= 8) {
  3105         emit_opcode(cbuf, Assembler::REX_B);
  3107     } else {
  3108       if ($src$$reg < 8) {
  3109         emit_opcode(cbuf, Assembler::REX_R);
  3110       } else {
  3111         emit_opcode(cbuf, Assembler::REX_RB);
  3114   %}
  3116   enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
  3117   %{
  3118     if ($dst$$reg < 8) {
  3119       if ($src$$reg < 8) {
  3120         emit_opcode(cbuf, Assembler::REX_W);
  3121       } else {
  3122         emit_opcode(cbuf, Assembler::REX_WB);
  3124     } else {
  3125       if ($src$$reg < 8) {
  3126         emit_opcode(cbuf, Assembler::REX_WR);
  3127       } else {
  3128         emit_opcode(cbuf, Assembler::REX_WRB);
  3131   %}
  3133   enc_class REX_reg_mem(rRegI reg, memory mem)
  3134   %{
  3135     if ($reg$$reg < 8) {
  3136       if ($mem$$base < 8) {
  3137         if ($mem$$index >= 8) {
  3138           emit_opcode(cbuf, Assembler::REX_X);
  3140       } else {
  3141         if ($mem$$index < 8) {
  3142           emit_opcode(cbuf, Assembler::REX_B);
  3143         } else {
  3144           emit_opcode(cbuf, Assembler::REX_XB);
  3147     } else {
  3148       if ($mem$$base < 8) {
  3149         if ($mem$$index < 8) {
  3150           emit_opcode(cbuf, Assembler::REX_R);
  3151         } else {
  3152           emit_opcode(cbuf, Assembler::REX_RX);
  3154       } else {
  3155         if ($mem$$index < 8) {
  3156           emit_opcode(cbuf, Assembler::REX_RB);
  3157         } else {
  3158           emit_opcode(cbuf, Assembler::REX_RXB);
  3162   %}
  3164   enc_class REX_reg_mem_wide(rRegL reg, memory mem)
  3165   %{
  3166     if ($reg$$reg < 8) {
  3167       if ($mem$$base < 8) {
  3168         if ($mem$$index < 8) {
  3169           emit_opcode(cbuf, Assembler::REX_W);
  3170         } else {
  3171           emit_opcode(cbuf, Assembler::REX_WX);
  3173       } else {
  3174         if ($mem$$index < 8) {
  3175           emit_opcode(cbuf, Assembler::REX_WB);
  3176         } else {
  3177           emit_opcode(cbuf, Assembler::REX_WXB);
  3180     } else {
  3181       if ($mem$$base < 8) {
  3182         if ($mem$$index < 8) {
  3183           emit_opcode(cbuf, Assembler::REX_WR);
  3184         } else {
  3185           emit_opcode(cbuf, Assembler::REX_WRX);
  3187       } else {
  3188         if ($mem$$index < 8) {
  3189           emit_opcode(cbuf, Assembler::REX_WRB);
  3190         } else {
  3191           emit_opcode(cbuf, Assembler::REX_WRXB);
  3195   %}
  3197   enc_class reg_mem(rRegI ereg, memory mem)
  3198   %{
  3199     // High registers handle in encode_RegMem
  3200     int reg = $ereg$$reg;
  3201     int base = $mem$$base;
  3202     int index = $mem$$index;
  3203     int scale = $mem$$scale;
  3204     int disp = $mem$$disp;
  3205     bool disp_is_oop = $mem->disp_is_oop();
  3207     encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
  3208   %}
  3210   enc_class RM_opc_mem(immI rm_opcode, memory mem)
  3211   %{
  3212     int rm_byte_opcode = $rm_opcode$$constant;
  3214     // High registers handle in encode_RegMem
  3215     int base = $mem$$base;
  3216     int index = $mem$$index;
  3217     int scale = $mem$$scale;
  3218     int displace = $mem$$disp;
  3220     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
  3221                                             // working with static
  3222                                             // globals
  3223     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
  3224                   disp_is_oop);
  3225   %}
  3227   enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
  3228   %{
  3229     int reg_encoding = $dst$$reg;
  3230     int base         = $src0$$reg;      // 0xFFFFFFFF indicates no base
  3231     int index        = 0x04;            // 0x04 indicates no index
  3232     int scale        = 0x00;            // 0x00 indicates no scale
  3233     int displace     = $src1$$constant; // 0x00 indicates no displacement
  3234     bool disp_is_oop = false;
  3235     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
  3236                   disp_is_oop);
  3237   %}
  3239   enc_class neg_reg(rRegI dst)
  3240   %{
  3241     int dstenc = $dst$$reg;
  3242     if (dstenc >= 8) {
  3243       emit_opcode(cbuf, Assembler::REX_B);
  3244       dstenc -= 8;
  3246     // NEG $dst
  3247     emit_opcode(cbuf, 0xF7);
  3248     emit_rm(cbuf, 0x3, 0x03, dstenc);
  3249   %}
  3251   enc_class neg_reg_wide(rRegI dst)
  3252   %{
  3253     int dstenc = $dst$$reg;
  3254     if (dstenc < 8) {
  3255       emit_opcode(cbuf, Assembler::REX_W);
  3256     } else {
  3257       emit_opcode(cbuf, Assembler::REX_WB);
  3258       dstenc -= 8;
  3260     // NEG $dst
  3261     emit_opcode(cbuf, 0xF7);
  3262     emit_rm(cbuf, 0x3, 0x03, dstenc);
  3263   %}
  3265   enc_class setLT_reg(rRegI dst)
  3266   %{
  3267     int dstenc = $dst$$reg;
  3268     if (dstenc >= 8) {
  3269       emit_opcode(cbuf, Assembler::REX_B);
  3270       dstenc -= 8;
  3271     } else if (dstenc >= 4) {
  3272       emit_opcode(cbuf, Assembler::REX);
  3274     // SETLT $dst
  3275     emit_opcode(cbuf, 0x0F);
  3276     emit_opcode(cbuf, 0x9C);
  3277     emit_rm(cbuf, 0x3, 0x0, dstenc);
  3278   %}
  3280   enc_class setNZ_reg(rRegI dst)
  3281   %{
  3282     int dstenc = $dst$$reg;
  3283     if (dstenc >= 8) {
  3284       emit_opcode(cbuf, Assembler::REX_B);
  3285       dstenc -= 8;
  3286     } else if (dstenc >= 4) {
  3287       emit_opcode(cbuf, Assembler::REX);
  3289     // SETNZ $dst
  3290     emit_opcode(cbuf, 0x0F);
  3291     emit_opcode(cbuf, 0x95);
  3292     emit_rm(cbuf, 0x3, 0x0, dstenc);
  3293   %}
  3295   enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
  3296                        rcx_RegI tmp)
  3297   %{
  3298     // cadd_cmpLT
  3300     int tmpReg = $tmp$$reg;
  3302     int penc = $p$$reg;
  3303     int qenc = $q$$reg;
  3304     int yenc = $y$$reg;
  3306     // subl $p,$q
  3307     if (penc < 8) {
  3308       if (qenc >= 8) {
  3309         emit_opcode(cbuf, Assembler::REX_B);
  3311     } else {
  3312       if (qenc < 8) {
  3313         emit_opcode(cbuf, Assembler::REX_R);
  3314       } else {
  3315         emit_opcode(cbuf, Assembler::REX_RB);
  3318     emit_opcode(cbuf, 0x2B);
  3319     emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
  3321     // sbbl $tmp, $tmp
  3322     emit_opcode(cbuf, 0x1B);
  3323     emit_rm(cbuf, 0x3, tmpReg, tmpReg);
  3325     // andl $tmp, $y
  3326     if (yenc >= 8) {
  3327       emit_opcode(cbuf, Assembler::REX_B);
  3329     emit_opcode(cbuf, 0x23);
  3330     emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
  3332     // addl $p,$tmp
  3333     if (penc >= 8) {
  3334         emit_opcode(cbuf, Assembler::REX_R);
  3336     emit_opcode(cbuf, 0x03);
  3337     emit_rm(cbuf, 0x3, penc & 7, tmpReg);
  3338   %}
  3340   // Compare the lonogs and set -1, 0, or 1 into dst
  3341   enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
  3342   %{
  3343     int src1enc = $src1$$reg;
  3344     int src2enc = $src2$$reg;
  3345     int dstenc = $dst$$reg;
  3347     // cmpq $src1, $src2
  3348     if (src1enc < 8) {
  3349       if (src2enc < 8) {
  3350         emit_opcode(cbuf, Assembler::REX_W);
  3351       } else {
  3352         emit_opcode(cbuf, Assembler::REX_WB);
  3354     } else {
  3355       if (src2enc < 8) {
  3356         emit_opcode(cbuf, Assembler::REX_WR);
  3357       } else {
  3358         emit_opcode(cbuf, Assembler::REX_WRB);
  3361     emit_opcode(cbuf, 0x3B);
  3362     emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
  3364     // movl $dst, -1
  3365     if (dstenc >= 8) {
  3366       emit_opcode(cbuf, Assembler::REX_B);
  3368     emit_opcode(cbuf, 0xB8 | (dstenc & 7));
  3369     emit_d32(cbuf, -1);
  3371     // jl,s done
  3372     emit_opcode(cbuf, 0x7C);
  3373     emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
  3375     // setne $dst
  3376     if (dstenc >= 4) {
  3377       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
  3379     emit_opcode(cbuf, 0x0F);
  3380     emit_opcode(cbuf, 0x95);
  3381     emit_opcode(cbuf, 0xC0 | (dstenc & 7));
  3383     // movzbl $dst, $dst
  3384     if (dstenc >= 4) {
  3385       emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
  3387     emit_opcode(cbuf, 0x0F);
  3388     emit_opcode(cbuf, 0xB6);
  3389     emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
  3390   %}
  3392   enc_class Push_ResultXD(regD dst) %{
  3393     int dstenc = $dst$$reg;
  3395     store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
  3397     // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
  3398     emit_opcode  (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
  3399     if (dstenc >= 8) {
  3400       emit_opcode(cbuf, Assembler::REX_R);
  3402     emit_opcode  (cbuf, 0x0F );
  3403     emit_opcode  (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
  3404     encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
  3406     // add rsp,8
  3407     emit_opcode(cbuf, Assembler::REX_W);
  3408     emit_opcode(cbuf,0x83);
  3409     emit_rm(cbuf,0x3, 0x0, RSP_enc);
  3410     emit_d8(cbuf,0x08);
  3411   %}
  3413   enc_class Push_SrcXD(regD src) %{
  3414     int srcenc = $src$$reg;
  3416     // subq rsp,#8
  3417     emit_opcode(cbuf, Assembler::REX_W);
  3418     emit_opcode(cbuf, 0x83);
  3419     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
  3420     emit_d8(cbuf, 0x8);
  3422     // movsd [rsp],src
  3423     emit_opcode(cbuf, 0xF2);
  3424     if (srcenc >= 8) {
  3425       emit_opcode(cbuf, Assembler::REX_R);
  3427     emit_opcode(cbuf, 0x0F);
  3428     emit_opcode(cbuf, 0x11);
  3429     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
  3431     // fldd [rsp]
  3432     emit_opcode(cbuf, 0x66);
  3433     emit_opcode(cbuf, 0xDD);
  3434     encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
  3435   %}
  3438   enc_class movq_ld(regD dst, memory mem) %{
  3439     MacroAssembler _masm(&cbuf);
  3440     __ movq($dst$$XMMRegister, $mem$$Address);
  3441   %}
  3443   enc_class movq_st(memory mem, regD src) %{
  3444     MacroAssembler _masm(&cbuf);
  3445     __ movq($mem$$Address, $src$$XMMRegister);
  3446   %}
  3448   enc_class pshufd_8x8(regF dst, regF src) %{
  3449     MacroAssembler _masm(&cbuf);
  3451     encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
  3452     __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
  3453     __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
  3454   %}
  3456   enc_class pshufd_4x16(regF dst, regF src) %{
  3457     MacroAssembler _masm(&cbuf);
  3459     __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
  3460   %}
  3462   enc_class pshufd(regD dst, regD src, int mode) %{
  3463     MacroAssembler _masm(&cbuf);
  3465     __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
  3466   %}
  3468   enc_class pxor(regD dst, regD src) %{
  3469     MacroAssembler _masm(&cbuf);
  3471     __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
  3472   %}
  3474   enc_class mov_i2x(regD dst, rRegI src) %{
  3475     MacroAssembler _masm(&cbuf);
  3477     __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
  3478   %}
  3480   // obj: object to lock
  3481   // box: box address (header location) -- killed
  3482   // tmp: rax -- killed
  3483   // scr: rbx -- killed
  3484   //
  3485   // What follows is a direct transliteration of fast_lock() and fast_unlock()
  3486   // from i486.ad.  See that file for comments.
  3487   // TODO: where possible switch from movq (r, 0) to movl(r,0) and
  3488   // use the shorter encoding.  (Movl clears the high-order 32-bits).
  3491   enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
  3492   %{
  3493     Register objReg = as_Register((int)$obj$$reg);
  3494     Register boxReg = as_Register((int)$box$$reg);
  3495     Register tmpReg = as_Register($tmp$$reg);
  3496     Register scrReg = as_Register($scr$$reg);
  3497     MacroAssembler masm(&cbuf);
  3499     // Verify uniqueness of register assignments -- necessary but not sufficient
  3500     assert (objReg != boxReg && objReg != tmpReg &&
  3501             objReg != scrReg && tmpReg != scrReg, "invariant") ;
  3503     if (_counters != NULL) {
  3504       masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
  3506     if (EmitSync & 1) {
  3507         // Without cast to int32_t a movptr will destroy r10 which is typically obj
  3508         masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; 
  3509         masm.cmpptr(rsp, (int32_t)NULL_WORD) ; 
  3510     } else
  3511     if (EmitSync & 2) {
  3512         Label DONE_LABEL;
  3513         if (UseBiasedLocking) {
  3514            // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
  3515           masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
  3517         // QQQ was movl...
  3518         masm.movptr(tmpReg, 0x1);
  3519         masm.orptr(tmpReg, Address(objReg, 0));
  3520         masm.movptr(Address(boxReg, 0), tmpReg);
  3521         if (os::is_MP()) {
  3522           masm.lock();
  3524         masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
  3525         masm.jcc(Assembler::equal, DONE_LABEL);
  3527         // Recursive locking
  3528         masm.subptr(tmpReg, rsp);
  3529         masm.andptr(tmpReg, 7 - os::vm_page_size());
  3530         masm.movptr(Address(boxReg, 0), tmpReg);
  3532         masm.bind(DONE_LABEL);
  3533         masm.nop(); // avoid branch to branch
  3534     } else {
  3535         Label DONE_LABEL, IsInflated, Egress;
  3537         masm.movptr(tmpReg, Address(objReg, 0)) ; 
  3538         masm.testl (tmpReg, 0x02) ;         // inflated vs stack-locked|neutral|biased
  3539         masm.jcc   (Assembler::notZero, IsInflated) ; 
  3541         // it's stack-locked, biased or neutral
  3542         // TODO: optimize markword triage order to reduce the number of
  3543         // conditional branches in the most common cases.
  3544         // Beware -- there's a subtle invariant that fetch of the markword
  3545         // at [FETCH], below, will never observe a biased encoding (*101b).
  3546         // If this invariant is not held we'll suffer exclusion (safety) failure.
  3548         if (UseBiasedLocking && !UseOptoBiasInlining) {
  3549           masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
  3550           masm.movptr(tmpReg, Address(objReg, 0)) ;        // [FETCH]
  3553         // was q will it destroy high?
  3554         masm.orl   (tmpReg, 1) ; 
  3555         masm.movptr(Address(boxReg, 0), tmpReg) ;  
  3556         if (os::is_MP()) { masm.lock(); } 
  3557         masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
  3558         if (_counters != NULL) {
  3559            masm.cond_inc32(Assembler::equal,
  3560                            ExternalAddress((address) _counters->fast_path_entry_count_addr()));
  3562         masm.jcc   (Assembler::equal, DONE_LABEL);
  3564         // Recursive locking
  3565         masm.subptr(tmpReg, rsp);
  3566         masm.andptr(tmpReg, 7 - os::vm_page_size());
  3567         masm.movptr(Address(boxReg, 0), tmpReg);
  3568         if (_counters != NULL) {
  3569            masm.cond_inc32(Assembler::equal,
  3570                            ExternalAddress((address) _counters->fast_path_entry_count_addr()));
  3572         masm.jmp   (DONE_LABEL) ;
  3574         masm.bind  (IsInflated) ;
  3575         // It's inflated
  3577         // TODO: someday avoid the ST-before-CAS penalty by
  3578         // relocating (deferring) the following ST.
  3579         // We should also think about trying a CAS without having
  3580         // fetched _owner.  If the CAS is successful we may
  3581         // avoid an RTO->RTS upgrade on the $line.
  3582         // Without cast to int32_t a movptr will destroy r10 which is typically obj
  3583         masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; 
  3585         masm.mov    (boxReg, tmpReg) ; 
  3586         masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
  3587         masm.testptr(tmpReg, tmpReg) ;   
  3588         masm.jcc    (Assembler::notZero, DONE_LABEL) ; 
  3590         // It's inflated and appears unlocked
  3591         if (os::is_MP()) { masm.lock(); } 
  3592         masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
  3593         // Intentional fall-through into DONE_LABEL ...
  3595         masm.bind  (DONE_LABEL) ;
  3596         masm.nop   () ;                 // avoid jmp to jmp
  3598   %}
  3600   // obj: object to unlock
  3601   // box: box address (displaced header location), killed
  3602   // RBX: killed tmp; cannot be obj nor box
  3603   enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
  3604   %{
  3606     Register objReg = as_Register($obj$$reg);
  3607     Register boxReg = as_Register($box$$reg);
  3608     Register tmpReg = as_Register($tmp$$reg);
  3609     MacroAssembler masm(&cbuf);
  3611     if (EmitSync & 4) { 
  3612        masm.cmpptr(rsp, 0) ; 
  3613     } else
  3614     if (EmitSync & 8) {
  3615        Label DONE_LABEL;
  3616        if (UseBiasedLocking) {
  3617          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
  3620        // Check whether the displaced header is 0
  3621        //(=> recursive unlock)
  3622        masm.movptr(tmpReg, Address(boxReg, 0));
  3623        masm.testptr(tmpReg, tmpReg);
  3624        masm.jcc(Assembler::zero, DONE_LABEL);
  3626        // If not recursive lock, reset the header to displaced header
  3627        if (os::is_MP()) {
  3628          masm.lock();
  3630        masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
  3631        masm.bind(DONE_LABEL);
  3632        masm.nop(); // avoid branch to branch
  3633     } else {
  3634        Label DONE_LABEL, Stacked, CheckSucc ;
  3636        if (UseBiasedLocking && !UseOptoBiasInlining) {
  3637          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
  3640        masm.movptr(tmpReg, Address(objReg, 0)) ; 
  3641        masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ; 
  3642        masm.jcc   (Assembler::zero, DONE_LABEL) ; 
  3643        masm.testl (tmpReg, 0x02) ; 
  3644        masm.jcc   (Assembler::zero, Stacked) ; 
  3646        // It's inflated
  3647        masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
  3648        masm.xorptr(boxReg, r15_thread) ; 
  3649        masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ; 
  3650        masm.jcc   (Assembler::notZero, DONE_LABEL) ; 
  3651        masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
  3652        masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
  3653        masm.jcc   (Assembler::notZero, CheckSucc) ; 
  3654        masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ; 
  3655        masm.jmp   (DONE_LABEL) ; 
  3657        if ((EmitSync & 65536) == 0) { 
  3658          Label LSuccess, LGoSlowPath ;
  3659          masm.bind  (CheckSucc) ;
  3660          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
  3661          masm.jcc   (Assembler::zero, LGoSlowPath) ;
  3663          // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
  3664          // the explicit ST;MEMBAR combination, but masm doesn't currently support
  3665          // "ANDQ M,IMM".  Don't use MFENCE here.  lock:add to TOS, xchg, etc
  3666          // are all faster when the write buffer is populated.
  3667          masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
  3668          if (os::is_MP()) {
  3669             masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
  3671          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
  3672          masm.jcc   (Assembler::notZero, LSuccess) ;
  3674          masm.movptr (boxReg, (int32_t)NULL_WORD) ;                   // box is really EAX
  3675          if (os::is_MP()) { masm.lock(); }
  3676          masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
  3677          masm.jcc   (Assembler::notEqual, LSuccess) ;
  3678          // Intentional fall-through into slow-path
  3680          masm.bind  (LGoSlowPath) ;
  3681          masm.orl   (boxReg, 1) ;                      // set ICC.ZF=0 to indicate failure
  3682          masm.jmp   (DONE_LABEL) ;
  3684          masm.bind  (LSuccess) ;
  3685          masm.testl (boxReg, 0) ;                      // set ICC.ZF=1 to indicate success
  3686          masm.jmp   (DONE_LABEL) ;
  3689        masm.bind  (Stacked) ; 
  3690        masm.movptr(tmpReg, Address (boxReg, 0)) ;      // re-fetch
  3691        if (os::is_MP()) { masm.lock(); } 
  3692        masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
  3694        if (EmitSync & 65536) {
  3695           masm.bind (CheckSucc) ;
  3697        masm.bind(DONE_LABEL);
  3698        if (EmitSync & 32768) {
  3699           masm.nop();                      // avoid branch to branch
  3702   %}
  3704   enc_class enc_String_Compare(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
  3705                         rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result) %{
  3706     Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
  3707           POP_LABEL, DONE_LABEL, CONT_LABEL,
  3708           WHILE_HEAD_LABEL;
  3709     MacroAssembler masm(&cbuf);
  3711     XMMRegister tmp1Reg   = as_XMMRegister($tmp1$$reg);
  3712     XMMRegister tmp2Reg   = as_XMMRegister($tmp2$$reg);
  3714     // Get the first character position in both strings
  3715     //         [8] char array, [12] offset, [16] count
  3716     int value_offset  = java_lang_String::value_offset_in_bytes();
  3717     int offset_offset = java_lang_String::offset_offset_in_bytes();
  3718     int count_offset  = java_lang_String::count_offset_in_bytes();
  3719     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
  3721     masm.load_heap_oop(rax, Address(rsi, value_offset));
  3722     masm.movl(rcx, Address(rsi, offset_offset));
  3723     masm.lea(rax, Address(rax, rcx, Address::times_2, base_offset));
  3724     masm.load_heap_oop(rbx, Address(rdi, value_offset));
  3725     masm.movl(rcx, Address(rdi, offset_offset));
  3726     masm.lea(rbx, Address(rbx, rcx, Address::times_2, base_offset));
  3728     // Compute the minimum of the string lengths(rsi) and the
  3729     // difference of the string lengths (stack)
  3731     // do the conditional move stuff
  3732     masm.movl(rdi, Address(rdi, count_offset));
  3733     masm.movl(rsi, Address(rsi, count_offset));
  3734     masm.movl(rcx, rdi);
  3735     masm.subl(rdi, rsi);
  3736     masm.push(rdi);
  3737     masm.cmov(Assembler::lessEqual, rsi, rcx);
  3739     // Is the minimum length zero?
  3740     masm.bind(RCX_GOOD_LABEL);
  3741     masm.testl(rsi, rsi);
  3742     masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
  3744     // Load first characters
  3745     masm.load_unsigned_short(rcx, Address(rbx, 0));
  3746     masm.load_unsigned_short(rdi, Address(rax, 0));
  3748     // Compare first characters
  3749     masm.subl(rcx, rdi);
  3750     masm.jcc(Assembler::notZero,  POP_LABEL);
  3751     masm.decrementl(rsi);
  3752     masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL);
  3755       // Check after comparing first character to see if strings are equivalent
  3756       Label LSkip2;
  3757       // Check if the strings start at same location
  3758       masm.cmpptr(rbx, rax);
  3759       masm.jccb(Assembler::notEqual, LSkip2);
  3761       // Check if the length difference is zero (from stack)
  3762       masm.cmpl(Address(rsp, 0), 0x0);
  3763       masm.jcc(Assembler::equal,  LENGTH_DIFF_LABEL);
  3765       // Strings might not be equivalent
  3766       masm.bind(LSkip2);
  3769     // Advance to next character
  3770     masm.addptr(rax, 2);
  3771     masm.addptr(rbx, 2);
  3773     if (UseSSE42Intrinsics) {
  3774       // With SSE4.2, use double quad vector compare
  3775       Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
  3776       // Setup to compare 16-byte vectors
  3777       masm.movl(rdi, rsi);
  3778       masm.andl(rsi, 0xfffffff8); // rsi holds the vector count
  3779       masm.andl(rdi, 0x00000007); // rdi holds the tail count
  3780       masm.testl(rsi, rsi);
  3781       masm.jccb(Assembler::zero, COMPARE_TAIL);
  3783       masm.lea(rax, Address(rax, rsi, Address::times_2));
  3784       masm.lea(rbx, Address(rbx, rsi, Address::times_2));
  3785       masm.negptr(rsi);
  3787       masm.bind(COMPARE_VECTORS);
  3788       masm.movdqu(tmp1Reg, Address(rax, rsi, Address::times_2));
  3789       masm.movdqu(tmp2Reg, Address(rbx, rsi, Address::times_2));
  3790       masm.pxor(tmp1Reg, tmp2Reg);
  3791       masm.ptest(tmp1Reg, tmp1Reg);
  3792       masm.jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
  3793       masm.addptr(rsi, 8);
  3794       masm.jcc(Assembler::notZero, COMPARE_VECTORS);
  3795       masm.jmpb(COMPARE_TAIL);
  3797       // Mismatched characters in the vectors
  3798       masm.bind(VECTOR_NOT_EQUAL);
  3799       masm.lea(rax, Address(rax, rsi, Address::times_2));
  3800       masm.lea(rbx, Address(rbx, rsi, Address::times_2));
  3801       masm.movl(rdi, 8);
  3803       // Compare tail (< 8 chars), or rescan last vectors to
  3804       // find 1st mismatched characters
  3805       masm.bind(COMPARE_TAIL);
  3806       masm.testl(rdi, rdi);
  3807       masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
  3808       masm.movl(rsi, rdi);
  3809       // Fallthru to tail compare
  3812     // Shift RAX and RBX to the end of the arrays, negate min
  3813     masm.lea(rax, Address(rax, rsi, Address::times_2, 0));
  3814     masm.lea(rbx, Address(rbx, rsi, Address::times_2, 0));
  3815     masm.negptr(rsi);
  3817     // Compare the rest of the characters
  3818     masm.bind(WHILE_HEAD_LABEL);
  3819     masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0));
  3820     masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0));
  3821     masm.subl(rcx, rdi);
  3822     masm.jccb(Assembler::notZero, POP_LABEL);
  3823     masm.increment(rsi);
  3824     masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
  3826     // Strings are equal up to min length.  Return the length difference.
  3827     masm.bind(LENGTH_DIFF_LABEL);
  3828     masm.pop(rcx);
  3829     masm.jmpb(DONE_LABEL);
  3831     // Discard the stored length difference
  3832     masm.bind(POP_LABEL);
  3833     masm.addptr(rsp, 8);
  3835     // That's it
  3836     masm.bind(DONE_LABEL);
  3837   %}
  3839  enc_class enc_String_IndexOf(rsi_RegP str1, rdi_RegP str2, regD tmp1, rax_RegI tmp2,
  3840                         rcx_RegI tmp3, rdx_RegI tmp4, rbx_RegI result) %{
  3841     // SSE4.2 version
  3842     Label LOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR,
  3843           SCAN_SUBSTR, RET_NEG_ONE, RET_NOT_FOUND, CLEANUP, DONE;
  3844     MacroAssembler masm(&cbuf);
  3846     XMMRegister tmp1Reg   = as_XMMRegister($tmp1$$reg);
  3848     // Get the first character position in both strings
  3849     //         [8] char array, [12] offset, [16] count
  3850     int value_offset  = java_lang_String::value_offset_in_bytes();
  3851     int offset_offset = java_lang_String::offset_offset_in_bytes();
  3852     int count_offset  = java_lang_String::count_offset_in_bytes();
  3853     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
  3855     // Get counts for string and substr
  3856     masm.movl(rdx, Address(rsi, count_offset));
  3857     masm.movl(rax, Address(rdi, count_offset));
  3858     // Check for substr count > string count
  3859     masm.cmpl(rax, rdx);
  3860     masm.jcc(Assembler::greater, RET_NEG_ONE);
  3862     // Start the indexOf operation
  3863     // Get start addr of string
  3864     masm.load_heap_oop(rbx, Address(rsi, value_offset));
  3865     masm.movl(rcx, Address(rsi, offset_offset));
  3866     masm.lea(rsi, Address(rbx, rcx, Address::times_2, base_offset));
  3867     masm.push(rsi);
  3869     // Get start addr of substr
  3870     masm.load_heap_oop(rbx, Address(rdi, value_offset));
  3871     masm.movl(rcx, Address(rdi, offset_offset));
  3872     masm.lea(rdi, Address(rbx, rcx, Address::times_2, base_offset));
  3873     masm.push(rdi);
  3874     masm.push(rax);
  3875     masm.jmpb(PREP_FOR_SCAN);
  3877     // Substr count saved at sp
  3878     // Substr saved at sp+8
  3879     // String saved at sp+16
  3881     // Prep to load substr for scan
  3882     masm.bind(LOAD_SUBSTR);
  3883     masm.movptr(rdi, Address(rsp, 8));
  3884     masm.movl(rax, Address(rsp, 0));
  3886     // Load substr
  3887     masm.bind(PREP_FOR_SCAN);
  3888     masm.movdqu(tmp1Reg, Address(rdi, 0));
  3889     masm.addq(rdx, 8);    // prime the loop
  3890     masm.subptr(rsi, 16);
  3892     // Scan string for substr in 16-byte vectors
  3893     masm.bind(SCAN_TO_SUBSTR);
  3894     masm.subq(rdx, 8);
  3895     masm.addptr(rsi, 16);
  3896     masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
  3897     masm.jcc(Assembler::above, SCAN_TO_SUBSTR);
  3898     masm.jccb(Assembler::aboveEqual, RET_NOT_FOUND);
  3900     // Fallthru: found a potential substr
  3902     //Make sure string is still long enough
  3903     masm.subl(rdx, rcx);
  3904     masm.cmpl(rdx, rax);
  3905     masm.jccb(Assembler::negative, RET_NOT_FOUND);
  3906     // Compute start addr of substr
  3907     masm.lea(rsi, Address(rsi, rcx, Address::times_2));
  3908     masm.movptr(rbx, rsi);
  3910     // Compare potential substr
  3911     masm.addq(rdx, 8);        // prime the loop
  3912     masm.addq(rax, 8);
  3913     masm.subptr(rsi, 16);
  3914     masm.subptr(rdi, 16);
  3916     // Scan 16-byte vectors of string and substr
  3917     masm.bind(SCAN_SUBSTR);
  3918     masm.subq(rax, 8);
  3919     masm.subq(rdx, 8);
  3920     masm.addptr(rsi, 16);
  3921     masm.addptr(rdi, 16);
  3922     masm.movdqu(tmp1Reg, Address(rdi, 0));
  3923     masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
  3924     masm.jcc(Assembler::noOverflow, LOAD_SUBSTR);   // OF == 0
  3925     masm.jcc(Assembler::positive, SCAN_SUBSTR);     // SF == 0
  3927     // Compute substr offset
  3928     masm.movptr(rsi, Address(rsp, 16));
  3929     masm.subptr(rbx, rsi);
  3930     masm.shrl(rbx, 1);
  3931     masm.jmpb(CLEANUP);
  3933     masm.bind(RET_NEG_ONE);
  3934     masm.movl(rbx, -1);
  3935     masm.jmpb(DONE);
  3937     masm.bind(RET_NOT_FOUND);
  3938     masm.movl(rbx, -1);
  3940     masm.bind(CLEANUP);
  3941     masm.addptr(rsp, 24);
  3943     masm.bind(DONE);
  3944   %}
  3946   enc_class enc_String_Equals(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
  3947                               rbx_RegI tmp3, rcx_RegI tmp2, rax_RegI result) %{
  3948     Label RET_TRUE, RET_FALSE, DONE, COMPARE_VECTORS, COMPARE_CHAR;
  3949     MacroAssembler masm(&cbuf);
  3951     XMMRegister tmp1Reg   = as_XMMRegister($tmp1$$reg);
  3952     XMMRegister tmp2Reg   = as_XMMRegister($tmp2$$reg);
  3954     int value_offset  = java_lang_String::value_offset_in_bytes();
  3955     int offset_offset = java_lang_String::offset_offset_in_bytes();
  3956     int count_offset  = java_lang_String::count_offset_in_bytes();
  3957     int base_offset   = arrayOopDesc::base_offset_in_bytes(T_CHAR);
  3959     // does source == target string?
  3960     masm.cmpptr(rdi, rsi);
  3961     masm.jcc(Assembler::equal, RET_TRUE);
  3963     // get and compare counts
  3964     masm.movl(rcx, Address(rdi, count_offset));
  3965     masm.movl(rax, Address(rsi, count_offset));
  3966     masm.cmpl(rcx, rax);
  3967     masm.jcc(Assembler::notEqual, RET_FALSE);
  3968     masm.testl(rax, rax);
  3969     masm.jcc(Assembler::zero, RET_TRUE);
  3971     // get source string offset and value
  3972     masm.load_heap_oop(rbx, Address(rsi, value_offset));
  3973     masm.movl(rax, Address(rsi, offset_offset));
  3974     masm.lea(rsi, Address(rbx, rax, Address::times_2, base_offset));
  3976     // get compare string offset and value
  3977     masm.load_heap_oop(rbx, Address(rdi, value_offset));
  3978     masm.movl(rax, Address(rdi, offset_offset));
  3979     masm.lea(rdi, Address(rbx, rax, Address::times_2, base_offset));
  3981     // Set byte count
  3982     masm.shll(rcx, 1);
  3983     masm.movl(rax, rcx);
  3985     if (UseSSE42Intrinsics) {
  3986       // With SSE4.2, use double quad vector compare
  3987       Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
  3988       // Compare 16-byte vectors
  3989       masm.andl(rcx, 0xfffffff0);  // vector count (in bytes)
  3990       masm.andl(rax, 0x0000000e);  // tail count (in bytes)
  3991       masm.testl(rcx, rcx);
  3992       masm.jccb(Assembler::zero, COMPARE_TAIL);
  3993       masm.lea(rdi, Address(rdi, rcx, Address::times_1));
  3994       masm.lea(rsi, Address(rsi, rcx, Address::times_1));
  3995       masm.negptr(rcx);
  3997       masm.bind(COMPARE_WIDE_VECTORS);
  3998       masm.movdqu(tmp1Reg, Address(rdi, rcx, Address::times_1));
  3999       masm.movdqu(tmp2Reg, Address(rsi, rcx, Address::times_1));
  4000       masm.pxor(tmp1Reg, tmp2Reg);
  4001       masm.ptest(tmp1Reg, tmp1Reg);
  4002       masm.jccb(Assembler::notZero, RET_FALSE);
  4003       masm.addptr(rcx, 16);
  4004       masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
  4005       masm.bind(COMPARE_TAIL);
  4006       masm.movl(rcx, rax);
  4007       // Fallthru to tail compare
  4010     // Compare 4-byte vectors
  4011     masm.andl(rcx, 0xfffffffc);  // vector count (in bytes)
  4012     masm.andl(rax, 0x00000002);  // tail char (in bytes)
  4013     masm.testl(rcx, rcx);
  4014     masm.jccb(Assembler::zero, COMPARE_CHAR);
  4015     masm.lea(rdi, Address(rdi, rcx, Address::times_1));
  4016     masm.lea(rsi, Address(rsi, rcx, Address::times_1));
  4017     masm.negptr(rcx);
  4019     masm.bind(COMPARE_VECTORS);
  4020     masm.movl(rbx, Address(rdi, rcx, Address::times_1));
  4021     masm.cmpl(rbx, Address(rsi, rcx, Address::times_1));
  4022     masm.jccb(Assembler::notEqual, RET_FALSE);
  4023     masm.addptr(rcx, 4);
  4024     masm.jcc(Assembler::notZero, COMPARE_VECTORS);
  4026     // Compare trailing char (final 2 bytes), if any
  4027     masm.bind(COMPARE_CHAR);
  4028     masm.testl(rax, rax);
  4029     masm.jccb(Assembler::zero, RET_TRUE);
  4030     masm.load_unsigned_short(rbx, Address(rdi, 0));
  4031     masm.load_unsigned_short(rcx, Address(rsi, 0));
  4032     masm.cmpl(rbx, rcx);
  4033     masm.jccb(Assembler::notEqual, RET_FALSE);
  4035     masm.bind(RET_TRUE);
  4036     masm.movl(rax, 1);   // return true
  4037     masm.jmpb(DONE);
  4039     masm.bind(RET_FALSE);
  4040     masm.xorl(rax, rax); // return false
  4042     masm.bind(DONE);
  4043   %}
  4045   enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, regD tmp1, regD tmp2,
  4046                              rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result) %{
  4047     Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
  4048     MacroAssembler masm(&cbuf);
  4050     XMMRegister tmp1Reg   = as_XMMRegister($tmp1$$reg);
  4051     XMMRegister tmp2Reg   = as_XMMRegister($tmp2$$reg);
  4052     Register ary1Reg      = as_Register($ary1$$reg);
  4053     Register ary2Reg      = as_Register($ary2$$reg);
  4054     Register tmp3Reg      = as_Register($tmp3$$reg);
  4055     Register tmp4Reg      = as_Register($tmp4$$reg);
  4056     Register resultReg    = as_Register($result$$reg);
  4058     int length_offset  = arrayOopDesc::length_offset_in_bytes();
  4059     int base_offset    = arrayOopDesc::base_offset_in_bytes(T_CHAR);
  4061     // Check the input args
  4062     masm.cmpq(ary1Reg, ary2Reg);
  4063     masm.jcc(Assembler::equal, TRUE_LABEL);
  4064     masm.testq(ary1Reg, ary1Reg);
  4065     masm.jcc(Assembler::zero, FALSE_LABEL);
  4066     masm.testq(ary2Reg, ary2Reg);
  4067     masm.jcc(Assembler::zero, FALSE_LABEL);
  4069     // Check the lengths
  4070     masm.movl(tmp4Reg, Address(ary1Reg, length_offset));
  4071     masm.movl(resultReg, Address(ary2Reg, length_offset));
  4072     masm.cmpl(tmp4Reg, resultReg);
  4073     masm.jcc(Assembler::notEqual, FALSE_LABEL);
  4074     masm.testl(resultReg, resultReg);
  4075     masm.jcc(Assembler::zero, TRUE_LABEL);
  4077     //load array address
  4078     masm.lea(ary1Reg, Address(ary1Reg, base_offset));
  4079     masm.lea(ary2Reg, Address(ary2Reg, base_offset));
  4081     //set byte count
  4082     masm.shll(tmp4Reg, 1);
  4083     masm.movl(resultReg,tmp4Reg);
  4085     if (UseSSE42Intrinsics){
  4086       // With SSE4.2, use double quad vector compare
  4087       Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
  4088       // Compare 16-byte vectors
  4089       masm.andl(tmp4Reg, 0xfffffff0);    // vector count (in bytes)
  4090       masm.andl(resultReg, 0x0000000e);  // tail count (in bytes)
  4091       masm.testl(tmp4Reg, tmp4Reg);
  4092       masm.jccb(Assembler::zero, COMPARE_TAIL);
  4093       masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
  4094       masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
  4095       masm.negptr(tmp4Reg);
  4097       masm.bind(COMPARE_WIDE_VECTORS);
  4098       masm.movdqu(tmp1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
  4099       masm.movdqu(tmp2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
  4100       masm.pxor(tmp1Reg, tmp2Reg);
  4101       masm.ptest(tmp1Reg, tmp1Reg);
  4103       masm.jccb(Assembler::notZero, FALSE_LABEL);
  4104       masm.addptr(tmp4Reg, 16);
  4105       masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
  4106       masm.bind(COMPARE_TAIL);
  4107       masm.movl(tmp4Reg, resultReg);
  4108       // Fallthru to tail compare
  4111    // Compare 4-byte vectors
  4112     masm.andl(tmp4Reg, 0xfffffffc);    // vector count (in bytes)
  4113     masm.andl(resultReg, 0x00000002);  // tail char (in bytes)
  4114     masm.testl(tmp4Reg, tmp4Reg); //if tmp2 == 0, only compare char
  4115     masm.jccb(Assembler::zero, COMPARE_CHAR);
  4116     masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
  4117     masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
  4118     masm.negptr(tmp4Reg);
  4120     masm.bind(COMPARE_VECTORS);
  4121     masm.movl(tmp3Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
  4122     masm.cmpl(tmp3Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
  4123     masm.jccb(Assembler::notEqual, FALSE_LABEL);
  4124     masm.addptr(tmp4Reg, 4);
  4125     masm.jcc(Assembler::notZero, COMPARE_VECTORS);
  4127     // Compare trailing char (final 2 bytes), if any
  4128     masm.bind(COMPARE_CHAR);
  4129     masm.testl(resultReg, resultReg);
  4130     masm.jccb(Assembler::zero, TRUE_LABEL);
  4131     masm.load_unsigned_short(tmp3Reg, Address(ary1Reg, 0));
  4132     masm.load_unsigned_short(tmp4Reg, Address(ary2Reg, 0));
  4133     masm.cmpl(tmp3Reg, tmp4Reg);
  4134     masm.jccb(Assembler::notEqual, FALSE_LABEL);
  4136     masm.bind(TRUE_LABEL);
  4137     masm.movl(resultReg, 1);   // return true
  4138     masm.jmpb(DONE);
  4140     masm.bind(FALSE_LABEL);
  4141     masm.xorl(resultReg, resultReg); // return false
  4143     // That's it
  4144     masm.bind(DONE);
  4145   %}
  4147   enc_class enc_rethrow()
  4148   %{
  4149     cbuf.set_inst_mark();
  4150     emit_opcode(cbuf, 0xE9); // jmp entry
  4151     emit_d32_reloc(cbuf,
  4152                    (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
  4153                    runtime_call_Relocation::spec(),
  4154                    RELOC_DISP32);
  4155   %}
  4157   enc_class absF_encoding(regF dst)
  4158   %{
  4159     int dstenc = $dst$$reg;
  4160     address signmask_address = (address) StubRoutines::x86::float_sign_mask();
  4162     cbuf.set_inst_mark();
  4163     if (dstenc >= 8) {
  4164       emit_opcode(cbuf, Assembler::REX_R);
  4165       dstenc -= 8;
  4167     // XXX reg_mem doesn't support RIP-relative addressing yet
  4168     emit_opcode(cbuf, 0x0F);
  4169     emit_opcode(cbuf, 0x54);
  4170     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
  4171     emit_d32_reloc(cbuf, signmask_address);
  4172   %}
  4174   enc_class absD_encoding(regD dst)
  4175   %{
  4176     int dstenc = $dst$$reg;
  4177     address signmask_address = (address) StubRoutines::x86::double_sign_mask();
  4179     cbuf.set_inst_mark();
  4180     emit_opcode(cbuf, 0x66);
  4181     if (dstenc >= 8) {
  4182       emit_opcode(cbuf, Assembler::REX_R);
  4183       dstenc -= 8;
  4185     // XXX reg_mem doesn't support RIP-relative addressing yet
  4186     emit_opcode(cbuf, 0x0F);
  4187     emit_opcode(cbuf, 0x54);
  4188     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
  4189     emit_d32_reloc(cbuf, signmask_address);
  4190   %}
  4192   enc_class negF_encoding(regF dst)
  4193   %{
  4194     int dstenc = $dst$$reg;
  4195     address signflip_address = (address) StubRoutines::x86::float_sign_flip();
  4197     cbuf.set_inst_mark();
  4198     if (dstenc >= 8) {
  4199       emit_opcode(cbuf, Assembler::REX_R);
  4200       dstenc -= 8;
  4202     // XXX reg_mem doesn't support RIP-relative addressing yet
  4203     emit_opcode(cbuf, 0x0F);
  4204     emit_opcode(cbuf, 0x57);
  4205     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
  4206     emit_d32_reloc(cbuf, signflip_address);
  4207   %}
  4209   enc_class negD_encoding(regD dst)
  4210   %{
  4211     int dstenc = $dst$$reg;
  4212     address signflip_address = (address) StubRoutines::x86::double_sign_flip();
  4214     cbuf.set_inst_mark();
  4215     emit_opcode(cbuf, 0x66);
  4216     if (dstenc >= 8) {
  4217       emit_opcode(cbuf, Assembler::REX_R);
  4218       dstenc -= 8;
  4220     // XXX reg_mem doesn't support RIP-relative addressing yet
  4221     emit_opcode(cbuf, 0x0F);
  4222     emit_opcode(cbuf, 0x57);
  4223     emit_rm(cbuf, 0x0, dstenc, 0x5);  // 00 reg 101
  4224     emit_d32_reloc(cbuf, signflip_address);
  4225   %}
  4227   enc_class f2i_fixup(rRegI dst, regF src)
  4228   %{
  4229     int dstenc = $dst$$reg;
  4230     int srcenc = $src$$reg;
  4232     // cmpl $dst, #0x80000000
  4233     if (dstenc >= 8) {
  4234       emit_opcode(cbuf, Assembler::REX_B);
  4236     emit_opcode(cbuf, 0x81);
  4237     emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
  4238     emit_d32(cbuf, 0x80000000);
  4240     // jne,s done
  4241     emit_opcode(cbuf, 0x75);
  4242     if (srcenc < 8 && dstenc < 8) {
  4243       emit_d8(cbuf, 0xF);
  4244     } else if (srcenc >= 8 && dstenc >= 8) {
  4245       emit_d8(cbuf, 0x11);
  4246     } else {
  4247       emit_d8(cbuf, 0x10);
  4250     // subq rsp, #8
  4251     emit_opcode(cbuf, Assembler::REX_W);
  4252     emit_opcode(cbuf, 0x83);
  4253     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
  4254     emit_d8(cbuf, 8);
  4256     // movss [rsp], $src
  4257     emit_opcode(cbuf, 0xF3);
  4258     if (srcenc >= 8) {
  4259       emit_opcode(cbuf, Assembler::REX_R);
  4261     emit_opcode(cbuf, 0x0F);
  4262     emit_opcode(cbuf, 0x11);
  4263     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
  4265     // call f2i_fixup
  4266     cbuf.set_inst_mark();
  4267     emit_opcode(cbuf, 0xE8);
  4268     emit_d32_reloc(cbuf,
  4269                    (int)
  4270                    (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4),
  4271                    runtime_call_Relocation::spec(),
  4272                    RELOC_DISP32);
  4274     // popq $dst
  4275     if (dstenc >= 8) {
  4276       emit_opcode(cbuf, Assembler::REX_B);
  4278     emit_opcode(cbuf, 0x58 | (dstenc & 7));
  4280     // done:
  4281   %}
  4283   enc_class f2l_fixup(rRegL dst, regF src)
  4284   %{
  4285     int dstenc = $dst$$reg;
  4286     int srcenc = $src$$reg;
  4287     address const_address = (address) StubRoutines::x86::double_sign_flip();
  4289     // cmpq $dst, [0x8000000000000000]
  4290     cbuf.set_inst_mark();
  4291     emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
  4292     emit_opcode(cbuf, 0x39);
  4293     // XXX reg_mem doesn't support RIP-relative addressing yet
  4294     emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
  4295     emit_d32_reloc(cbuf, const_address);
  4298     // jne,s done
  4299     emit_opcode(cbuf, 0x75);
  4300     if (srcenc < 8 && dstenc < 8) {
  4301       emit_d8(cbuf, 0xF);
  4302     } else if (srcenc >= 8 && dstenc >= 8) {
  4303       emit_d8(cbuf, 0x11);
  4304     } else {
  4305       emit_d8(cbuf, 0x10);
  4308     // subq rsp, #8
  4309     emit_opcode(cbuf, Assembler::REX_W);
  4310     emit_opcode(cbuf, 0x83);
  4311     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
  4312     emit_d8(cbuf, 8);
  4314     // movss [rsp], $src
  4315     emit_opcode(cbuf, 0xF3);
  4316     if (srcenc >= 8) {
  4317       emit_opcode(cbuf, Assembler::REX_R);
  4319     emit_opcode(cbuf, 0x0F);
  4320     emit_opcode(cbuf, 0x11);
  4321     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
  4323     // call f2l_fixup
  4324     cbuf.set_inst_mark();
  4325     emit_opcode(cbuf, 0xE8);
  4326     emit_d32_reloc(cbuf,
  4327                    (int)
  4328                    (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4),
  4329                    runtime_call_Relocation::spec(),
  4330                    RELOC_DISP32);
  4332     // popq $dst
  4333     if (dstenc >= 8) {
  4334       emit_opcode(cbuf, Assembler::REX_B);
  4336     emit_opcode(cbuf, 0x58 | (dstenc & 7));
  4338     // done:
  4339   %}
  4341   enc_class d2i_fixup(rRegI dst, regD src)
  4342   %{
  4343     int dstenc = $dst$$reg;
  4344     int srcenc = $src$$reg;
  4346     // cmpl $dst, #0x80000000
  4347     if (dstenc >= 8) {
  4348       emit_opcode(cbuf, Assembler::REX_B);
  4350     emit_opcode(cbuf, 0x81);
  4351     emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
  4352     emit_d32(cbuf, 0x80000000);
  4354     // jne,s done
  4355     emit_opcode(cbuf, 0x75);
  4356     if (srcenc < 8 && dstenc < 8) {
  4357       emit_d8(cbuf, 0xF);
  4358     } else if (srcenc >= 8 && dstenc >= 8) {
  4359       emit_d8(cbuf, 0x11);
  4360     } else {
  4361       emit_d8(cbuf, 0x10);
  4364     // subq rsp, #8
  4365     emit_opcode(cbuf, Assembler::REX_W);
  4366     emit_opcode(cbuf, 0x83);
  4367     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
  4368     emit_d8(cbuf, 8);
  4370     // movsd [rsp], $src
  4371     emit_opcode(cbuf, 0xF2);
  4372     if (srcenc >= 8) {
  4373       emit_opcode(cbuf, Assembler::REX_R);
  4375     emit_opcode(cbuf, 0x0F);
  4376     emit_opcode(cbuf, 0x11);
  4377     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
  4379     // call d2i_fixup
  4380     cbuf.set_inst_mark();
  4381     emit_opcode(cbuf, 0xE8);
  4382     emit_d32_reloc(cbuf,
  4383                    (int)
  4384                    (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4),
  4385                    runtime_call_Relocation::spec(),
  4386                    RELOC_DISP32);
  4388     // popq $dst
  4389     if (dstenc >= 8) {
  4390       emit_opcode(cbuf, Assembler::REX_B);
  4392     emit_opcode(cbuf, 0x58 | (dstenc & 7));
  4394     // done:
  4395   %}
  4397   enc_class d2l_fixup(rRegL dst, regD src)
  4398   %{
  4399     int dstenc = $dst$$reg;
  4400     int srcenc = $src$$reg;
  4401     address const_address = (address) StubRoutines::x86::double_sign_flip();
  4403     // cmpq $dst, [0x8000000000000000]
  4404     cbuf.set_inst_mark();
  4405     emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
  4406     emit_opcode(cbuf, 0x39);
  4407     // XXX reg_mem doesn't support RIP-relative addressing yet
  4408     emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
  4409     emit_d32_reloc(cbuf, const_address);
  4412     // jne,s done
  4413     emit_opcode(cbuf, 0x75);
  4414     if (srcenc < 8 && dstenc < 8) {
  4415       emit_d8(cbuf, 0xF);
  4416     } else if (srcenc >= 8 && dstenc >= 8) {
  4417       emit_d8(cbuf, 0x11);
  4418     } else {
  4419       emit_d8(cbuf, 0x10);
  4422     // subq rsp, #8
  4423     emit_opcode(cbuf, Assembler::REX_W);
  4424     emit_opcode(cbuf, 0x83);
  4425     emit_rm(cbuf, 0x3, 0x5, RSP_enc);
  4426     emit_d8(cbuf, 8);
  4428     // movsd [rsp], $src
  4429     emit_opcode(cbuf, 0xF2);
  4430     if (srcenc >= 8) {
  4431       emit_opcode(cbuf, Assembler::REX_R);
  4433     emit_opcode(cbuf, 0x0F);
  4434     emit_opcode(cbuf, 0x11);
  4435     encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
  4437     // call d2l_fixup
  4438     cbuf.set_inst_mark();
  4439     emit_opcode(cbuf, 0xE8);
  4440     emit_d32_reloc(cbuf,
  4441                    (int)
  4442                    (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4),
  4443                    runtime_call_Relocation::spec(),
  4444                    RELOC_DISP32);
  4446     // popq $dst
  4447     if (dstenc >= 8) {
  4448       emit_opcode(cbuf, Assembler::REX_B);
  4450     emit_opcode(cbuf, 0x58 | (dstenc & 7));
  4452     // done:
  4453   %}
  4455   // Safepoint Poll.  This polls the safepoint page, and causes an
  4456   // exception if it is not readable. Unfortunately, it kills
  4457   // RFLAGS in the process.
  4458   enc_class enc_safepoint_poll
  4459   %{
  4460     // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
  4461     // XXX reg_mem doesn't support RIP-relative addressing yet
  4462     cbuf.set_inst_mark();
  4463     cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
  4464     emit_opcode(cbuf, 0x85); // testl
  4465     emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
  4466     // cbuf.inst_mark() is beginning of instruction
  4467     emit_d32_reloc(cbuf, os::get_polling_page());
  4468 //                    relocInfo::poll_type,
  4469   %}
  4470 %}
  4474 //----------FRAME--------------------------------------------------------------
  4475 // Definition of frame structure and management information.
  4476 //
  4477 //  S T A C K   L A Y O U T    Allocators stack-slot number
  4478 //                             |   (to get allocators register number
  4479 //  G  Owned by    |        |  v    add OptoReg::stack0())
  4480 //  r   CALLER     |        |
  4481 //  o     |        +--------+      pad to even-align allocators stack-slot
  4482 //  w     V        |  pad0  |        numbers; owned by CALLER
  4483 //  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
  4484 //  h     ^        |   in   |  5
  4485 //        |        |  args  |  4   Holes in incoming args owned by SELF
  4486 //  |     |        |        |  3
  4487 //  |     |        +--------+
  4488 //  V     |        | old out|      Empty on Intel, window on Sparc
  4489 //        |    old |preserve|      Must be even aligned.
  4490 //        |     SP-+--------+----> Matcher::_old_SP, even aligned
  4491 //        |        |   in   |  3   area for Intel ret address
  4492 //     Owned by    |preserve|      Empty on Sparc.
  4493 //       SELF      +--------+
  4494 //        |        |  pad2  |  2   pad to align old SP
  4495 //        |        +--------+  1
  4496 //        |        | locks  |  0
  4497 //        |        +--------+----> OptoReg::stack0(), even aligned
  4498 //        |        |  pad1  | 11   pad to align new SP
  4499 //        |        +--------+
  4500 //        |        |        | 10
  4501 //        |        | spills |  9   spills
  4502 //        V        |        |  8   (pad0 slot for callee)
  4503 //      -----------+--------+----> Matcher::_out_arg_limit, unaligned
  4504 //        ^        |  out   |  7
  4505 //        |        |  args  |  6   Holes in outgoing args owned by CALLEE
  4506 //     Owned by    +--------+
  4507 //      CALLEE     | new out|  6   Empty on Intel, window on Sparc
  4508 //        |    new |preserve|      Must be even-aligned.
  4509 //        |     SP-+--------+----> Matcher::_new_SP, even aligned
  4510 //        |        |        |
  4511 //
  4512 // Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
  4513 //         known from SELF's arguments and the Java calling convention.
  4514 //         Region 6-7 is determined per call site.
  4515 // Note 2: If the calling convention leaves holes in the incoming argument
  4516 //         area, those holes are owned by SELF.  Holes in the outgoing area
  4517 //         are owned by the CALLEE.  Holes should not be nessecary in the
  4518 //         incoming area, as the Java calling convention is completely under
  4519 //         the control of the AD file.  Doubles can be sorted and packed to
  4520 //         avoid holes.  Holes in the outgoing arguments may be nessecary for
  4521 //         varargs C calling conventions.
  4522 // Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
  4523 //         even aligned with pad0 as needed.
  4524 //         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
  4525 //         region 6-11 is even aligned; it may be padded out more so that
  4526 //         the region from SP to FP meets the minimum stack alignment.
  4527 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
  4528 //         alignment.  Region 11, pad1, may be dynamically extended so that
  4529 //         SP meets the minimum alignment.
  4531 frame
  4532 %{
  4533   // What direction does stack grow in (assumed to be same for C & Java)
  4534   stack_direction(TOWARDS_LOW);
  4536   // These three registers define part of the calling convention
  4537   // between compiled code and the interpreter.
  4538   inline_cache_reg(RAX);                // Inline Cache Register
  4539   interpreter_method_oop_reg(RBX);      // Method Oop Register when
  4540                                         // calling interpreter
  4542   // Optional: name the operand used by cisc-spilling to access
  4543   // [stack_pointer + offset]
  4544   cisc_spilling_operand_name(indOffset32);
  4546   // Number of stack slots consumed by locking an object
  4547   sync_stack_slots(2);
  4549   // Compiled code's Frame Pointer
  4550   frame_pointer(RSP);
  4552   // Interpreter stores its frame pointer in a register which is
  4553   // stored to the stack by I2CAdaptors.
  4554   // I2CAdaptors convert from interpreted java to compiled java.
  4555   interpreter_frame_pointer(RBP);
  4557   // Stack alignment requirement
  4558   stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
  4560   // Number of stack slots between incoming argument block and the start of
  4561   // a new frame.  The PROLOG must add this many slots to the stack.  The
  4562   // EPILOG must remove this many slots.  amd64 needs two slots for
  4563   // return address.
  4564   in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
  4566   // Number of outgoing stack slots killed above the out_preserve_stack_slots
  4567   // for calls to C.  Supports the var-args backing area for register parms.
  4568   varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
  4570   // The after-PROLOG location of the return address.  Location of
  4571   // return address specifies a type (REG or STACK) and a number
  4572   // representing the register number (i.e. - use a register name) or
  4573   // stack slot.
  4574   // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
  4575   // Otherwise, it is above the locks and verification slot and alignment word
  4576   return_addr(STACK - 2 +
  4577               round_to(2 + 2 * VerifyStackAtCalls +
  4578                        Compile::current()->fixed_slots(),
  4579                        WordsPerLong * 2));
  4581   // Body of function which returns an integer array locating
  4582   // arguments either in registers or in stack slots.  Passed an array
  4583   // of ideal registers called "sig" and a "length" count.  Stack-slot
  4584   // offsets are based on outgoing arguments, i.e. a CALLER setting up
  4585   // arguments for a CALLEE.  Incoming stack arguments are
  4586   // automatically biased by the preserve_stack_slots field above.
  4588   calling_convention
  4589   %{
  4590     // No difference between ingoing/outgoing just pass false
  4591     SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
  4592   %}
  4594   c_calling_convention
  4595   %{
  4596     // This is obviously always outgoing
  4597     (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
  4598   %}
  4600   // Location of compiled Java return values.  Same as C for now.
  4601   return_value
  4602   %{
  4603     assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
  4604            "only return normal values");
  4606     static const int lo[Op_RegL + 1] = {
  4607       0,
  4608       0,
  4609       RAX_num,  // Op_RegN
  4610       RAX_num,  // Op_RegI
  4611       RAX_num,  // Op_RegP
  4612       XMM0_num, // Op_RegF
  4613       XMM0_num, // Op_RegD
  4614       RAX_num   // Op_RegL
  4615     };
  4616     static const int hi[Op_RegL + 1] = {
  4617       0,
  4618       0,
  4619       OptoReg::Bad, // Op_RegN
  4620       OptoReg::Bad, // Op_RegI
  4621       RAX_H_num,    // Op_RegP
  4622       OptoReg::Bad, // Op_RegF
  4623       XMM0_H_num,   // Op_RegD
  4624       RAX_H_num     // Op_RegL
  4625     };
  4626     assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
  4627     return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
  4628   %}
  4629 %}
  4631 //----------ATTRIBUTES---------------------------------------------------------
  4632 //----------Operand Attributes-------------------------------------------------
  4633 op_attrib op_cost(0);        // Required cost attribute
  4635 //----------Instruction Attributes---------------------------------------------
  4636 ins_attrib ins_cost(100);       // Required cost attribute
  4637 ins_attrib ins_size(8);         // Required size attribute (in bits)
  4638 ins_attrib ins_pc_relative(0);  // Required PC Relative flag
  4639 ins_attrib ins_short_branch(0); // Required flag: is this instruction
  4640                                 // a non-matching short branch variant
  4641                                 // of some long branch?
  4642 ins_attrib ins_alignment(1);    // Required alignment attribute (must
  4643                                 // be a power of 2) specifies the
  4644                                 // alignment that some part of the
  4645                                 // instruction (not necessarily the
  4646                                 // start) requires.  If > 1, a
  4647                                 // compute_padding() function must be
  4648                                 // provided for the instruction
  4650 //----------OPERANDS-----------------------------------------------------------
  4651 // Operand definitions must precede instruction definitions for correct parsing
  4652 // in the ADLC because operands constitute user defined types which are used in
  4653 // instruction definitions.
  4655 //----------Simple Operands----------------------------------------------------
  4656 // Immediate Operands
  4657 // Integer Immediate
  4658 operand immI()
  4659 %{
  4660   match(ConI);
  4662   op_cost(10);
  4663   format %{ %}
  4664   interface(CONST_INTER);
  4665 %}
  4667 // Constant for test vs zero
  4668 operand immI0()
  4669 %{
  4670   predicate(n->get_int() == 0);
  4671   match(ConI);
  4673   op_cost(0);
  4674   format %{ %}
  4675   interface(CONST_INTER);
  4676 %}
  4678 // Constant for increment
  4679 operand immI1()
  4680 %{
  4681   predicate(n->get_int() == 1);
  4682   match(ConI);
  4684   op_cost(0);
  4685   format %{ %}
  4686   interface(CONST_INTER);
  4687 %}
  4689 // Constant for decrement
  4690 operand immI_M1()
  4691 %{
  4692   predicate(n->get_int() == -1);
  4693   match(ConI);
  4695   op_cost(0);
  4696   format %{ %}
  4697   interface(CONST_INTER);
  4698 %}
  4700 // Valid scale values for addressing modes
  4701 operand immI2()
  4702 %{
  4703   predicate(0 <= n->get_int() && (n->get_int() <= 3));
  4704   match(ConI);
  4706   format %{ %}
  4707   interface(CONST_INTER);
  4708 %}
  4710 operand immI8()
  4711 %{
  4712   predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
  4713   match(ConI);
  4715   op_cost(5);
  4716   format %{ %}
  4717   interface(CONST_INTER);
  4718 %}
  4720 operand immI16()
  4721 %{
  4722   predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
  4723   match(ConI);
  4725   op_cost(10);
  4726   format %{ %}
  4727   interface(CONST_INTER);
  4728 %}
  4730 // Constant for long shifts
  4731 operand immI_32()
  4732 %{
  4733   predicate( n->get_int() == 32 );
  4734   match(ConI);
  4736   op_cost(0);
  4737   format %{ %}
  4738   interface(CONST_INTER);
  4739 %}
  4741 // Constant for long shifts
  4742 operand immI_64()
  4743 %{
  4744   predicate( n->get_int() == 64 );
  4745   match(ConI);
  4747   op_cost(0);
  4748   format %{ %}
  4749   interface(CONST_INTER);
  4750 %}
  4752 // Pointer Immediate
  4753 operand immP()
  4754 %{
  4755   match(ConP);
  4757   op_cost(10);
  4758   format %{ %}
  4759   interface(CONST_INTER);
  4760 %}
  4762 // NULL Pointer Immediate
  4763 operand immP0()
  4764 %{
  4765   predicate(n->get_ptr() == 0);
  4766   match(ConP);
  4768   op_cost(5);
  4769   format %{ %}
  4770   interface(CONST_INTER);
  4771 %}
  4773 // Pointer Immediate
  4774 operand immN() %{
  4775   match(ConN);
  4777   op_cost(10);
  4778   format %{ %}
  4779   interface(CONST_INTER);
  4780 %}
  4782 // NULL Pointer Immediate
  4783 operand immN0() %{
  4784   predicate(n->get_narrowcon() == 0);
  4785   match(ConN);
  4787   op_cost(5);
  4788   format %{ %}
  4789   interface(CONST_INTER);
  4790 %}
  4792 operand immP31()
  4793 %{
  4794   predicate(!n->as_Type()->type()->isa_oopptr()
  4795             && (n->get_ptr() >> 31) == 0);
  4796   match(ConP);
  4798   op_cost(5);
  4799   format %{ %}
  4800   interface(CONST_INTER);
  4801 %}
  4804 // Long Immediate
  4805 operand immL()
  4806 %{
  4807   match(ConL);
  4809   op_cost(20);
  4810   format %{ %}
  4811   interface(CONST_INTER);
  4812 %}
  4814 // Long Immediate 8-bit
  4815 operand immL8()
  4816 %{
  4817   predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
  4818   match(ConL);
  4820   op_cost(5);
  4821   format %{ %}
  4822   interface(CONST_INTER);
  4823 %}
  4825 // Long Immediate 32-bit unsigned
  4826 operand immUL32()
  4827 %{
  4828   predicate(n->get_long() == (unsigned int) (n->get_long()));
  4829   match(ConL);
  4831   op_cost(10);
  4832   format %{ %}
  4833   interface(CONST_INTER);
  4834 %}
  4836 // Long Immediate 32-bit signed
  4837 operand immL32()
  4838 %{
  4839   predicate(n->get_long() == (int) (n->get_long()));
  4840   match(ConL);
  4842   op_cost(15);
  4843   format %{ %}
  4844   interface(CONST_INTER);
  4845 %}
  4847 // Long Immediate zero
  4848 operand immL0()
  4849 %{
  4850   predicate(n->get_long() == 0L);
  4851   match(ConL);
  4853   op_cost(10);
  4854   format %{ %}
  4855   interface(CONST_INTER);
  4856 %}
  4858 // Constant for increment
  4859 operand immL1()
  4860 %{
  4861   predicate(n->get_long() == 1);
  4862   match(ConL);
  4864   format %{ %}
  4865   interface(CONST_INTER);
  4866 %}
  4868 // Constant for decrement
  4869 operand immL_M1()
  4870 %{
  4871   predicate(n->get_long() == -1);
  4872   match(ConL);
  4874   format %{ %}
  4875   interface(CONST_INTER);
  4876 %}
  4878 // Long Immediate: the value 10
  4879 operand immL10()
  4880 %{
  4881   predicate(n->get_long() == 10);
  4882   match(ConL);
  4884   format %{ %}
  4885   interface(CONST_INTER);
  4886 %}
  4888 // Long immediate from 0 to 127.
  4889 // Used for a shorter form of long mul by 10.
  4890 operand immL_127()
  4891 %{
  4892   predicate(0 <= n->get_long() && n->get_long() < 0x80);
  4893   match(ConL);
  4895   op_cost(10);
  4896   format %{ %}
  4897   interface(CONST_INTER);
  4898 %}
  4900 // Long Immediate: low 32-bit mask
  4901 operand immL_32bits()
  4902 %{
  4903   predicate(n->get_long() == 0xFFFFFFFFL);
  4904   match(ConL);
  4905   op_cost(20);
  4907   format %{ %}
  4908   interface(CONST_INTER);
  4909 %}
  4911 // Float Immediate zero
  4912 operand immF0()
  4913 %{
  4914   predicate(jint_cast(n->getf()) == 0);
  4915   match(ConF);
  4917   op_cost(5);
  4918   format %{ %}
  4919   interface(CONST_INTER);
  4920 %}
  4922 // Float Immediate
  4923 operand immF()
  4924 %{
  4925   match(ConF);
  4927   op_cost(15);
  4928   format %{ %}
  4929   interface(CONST_INTER);
  4930 %}
  4932 // Double Immediate zero
  4933 operand immD0()
  4934 %{
  4935   predicate(jlong_cast(n->getd()) == 0);
  4936   match(ConD);
  4938   op_cost(5);
  4939   format %{ %}
  4940   interface(CONST_INTER);
  4941 %}
  4943 // Double Immediate
  4944 operand immD()
  4945 %{
  4946   match(ConD);
  4948   op_cost(15);
  4949   format %{ %}
  4950   interface(CONST_INTER);
  4951 %}
  4953 // Immediates for special shifts (sign extend)
  4955 // Constants for increment
  4956 operand immI_16()
  4957 %{
  4958   predicate(n->get_int() == 16);
  4959   match(ConI);
  4961   format %{ %}
  4962   interface(CONST_INTER);
  4963 %}
  4965 operand immI_24()
  4966 %{
  4967   predicate(n->get_int() == 24);
  4968   match(ConI);
  4970   format %{ %}
  4971   interface(CONST_INTER);
  4972 %}
  4974 // Constant for byte-wide masking
  4975 operand immI_255()
  4976 %{
  4977   predicate(n->get_int() == 255);
  4978   match(ConI);
  4980   format %{ %}
  4981   interface(CONST_INTER);
  4982 %}
  4984 // Constant for short-wide masking
  4985 operand immI_65535()
  4986 %{
  4987   predicate(n->get_int() == 65535);
  4988   match(ConI);
  4990   format %{ %}
  4991   interface(CONST_INTER);
  4992 %}
  4994 // Constant for byte-wide masking
  4995 operand immL_255()
  4996 %{
  4997   predicate(n->get_long() == 255);
  4998   match(ConL);
  5000   format %{ %}
  5001   interface(CONST_INTER);
  5002 %}
  5004 // Constant for short-wide masking
  5005 operand immL_65535()
  5006 %{
  5007   predicate(n->get_long() == 65535);
  5008   match(ConL);
  5010   format %{ %}
  5011   interface(CONST_INTER);
  5012 %}
  5014 // Register Operands
  5015 // Integer Register
  5016 operand rRegI()
  5017 %{
  5018   constraint(ALLOC_IN_RC(int_reg));
  5019   match(RegI);
  5021   match(rax_RegI);
  5022   match(rbx_RegI);
  5023   match(rcx_RegI);
  5024   match(rdx_RegI);
  5025   match(rdi_RegI);
  5027   format %{ %}
  5028   interface(REG_INTER);
  5029 %}
  5031 // Special Registers
  5032 operand rax_RegI()
  5033 %{
  5034   constraint(ALLOC_IN_RC(int_rax_reg));
  5035   match(RegI);
  5036   match(rRegI);
  5038   format %{ "RAX" %}
  5039   interface(REG_INTER);
  5040 %}
  5042 // Special Registers
  5043 operand rbx_RegI()
  5044 %{
  5045   constraint(ALLOC_IN_RC(int_rbx_reg));
  5046   match(RegI);
  5047   match(rRegI);
  5049   format %{ "RBX" %}
  5050   interface(REG_INTER);
  5051 %}
  5053 operand rcx_RegI()
  5054 %{
  5055   constraint(ALLOC_IN_RC(int_rcx_reg));
  5056   match(RegI);
  5057   match(rRegI);
  5059   format %{ "RCX" %}
  5060   interface(REG_INTER);
  5061 %}
  5063 operand rdx_RegI()
  5064 %{
  5065   constraint(ALLOC_IN_RC(int_rdx_reg));
  5066   match(RegI);
  5067   match(rRegI);
  5069   format %{ "RDX" %}
  5070   interface(REG_INTER);
  5071 %}
  5073 operand rdi_RegI()
  5074 %{
  5075   constraint(ALLOC_IN_RC(int_rdi_reg));
  5076   match(RegI);
  5077   match(rRegI);
  5079   format %{ "RDI" %}
  5080   interface(REG_INTER);
  5081 %}
  5083 operand no_rcx_RegI()
  5084 %{
  5085   constraint(ALLOC_IN_RC(int_no_rcx_reg));
  5086   match(RegI);
  5087   match(rax_RegI);
  5088   match(rbx_RegI);
  5089   match(rdx_RegI);
  5090   match(rdi_RegI);
  5092   format %{ %}
  5093   interface(REG_INTER);
  5094 %}
  5096 operand no_rax_rdx_RegI()
  5097 %{
  5098   constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
  5099   match(RegI);
  5100   match(rbx_RegI);
  5101   match(rcx_RegI);
  5102   match(rdi_RegI);
  5104   format %{ %}
  5105   interface(REG_INTER);
  5106 %}
  5108 // Pointer Register
  5109 operand any_RegP()
  5110 %{
  5111   constraint(ALLOC_IN_RC(any_reg));
  5112   match(RegP);
  5113   match(rax_RegP);
  5114   match(rbx_RegP);
  5115   match(rdi_RegP);
  5116   match(rsi_RegP);
  5117   match(rbp_RegP);
  5118   match(r15_RegP);
  5119   match(rRegP);
  5121   format %{ %}
  5122   interface(REG_INTER);
  5123 %}
  5125 operand rRegP()
  5126 %{
  5127   constraint(ALLOC_IN_RC(ptr_reg));
  5128   match(RegP);
  5129   match(rax_RegP);
  5130   match(rbx_RegP);
  5131   match(rdi_RegP);
  5132   match(rsi_RegP);
  5133   match(rbp_RegP);
  5134   match(r15_RegP);  // See Q&A below about r15_RegP.
  5136   format %{ %}
  5137   interface(REG_INTER);
  5138 %}
  5140 operand rRegN() %{
  5141   constraint(ALLOC_IN_RC(int_reg));
  5142   match(RegN);
  5144   format %{ %}
  5145   interface(REG_INTER);
  5146 %}
  5148 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
  5149 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
  5150 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
  5151 // The output of an instruction is controlled by the allocator, which respects
  5152 // register class masks, not match rules.  Unless an instruction mentions
  5153 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
  5154 // by the allocator as an input.
  5156 operand no_rax_RegP()
  5157 %{
  5158   constraint(ALLOC_IN_RC(ptr_no_rax_reg));
  5159   match(RegP);
  5160   match(rbx_RegP);
  5161   match(rsi_RegP);
  5162   match(rdi_RegP);
  5164   format %{ %}
  5165   interface(REG_INTER);
  5166 %}
  5168 operand no_rbp_RegP()
  5169 %{
  5170   constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
  5171   match(RegP);
  5172   match(rbx_RegP);
  5173   match(rsi_RegP);
  5174   match(rdi_RegP);
  5176   format %{ %}
  5177   interface(REG_INTER);
  5178 %}
  5180 operand no_rax_rbx_RegP()
  5181 %{
  5182   constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
  5183   match(RegP);
  5184   match(rsi_RegP);
  5185   match(rdi_RegP);
  5187   format %{ %}
  5188   interface(REG_INTER);
  5189 %}
  5191 // Special Registers
  5192 // Return a pointer value
  5193 operand rax_RegP()
  5194 %{
  5195   constraint(ALLOC_IN_RC(ptr_rax_reg));
  5196   match(RegP);
  5197   match(rRegP);
  5199   format %{ %}
  5200   interface(REG_INTER);
  5201 %}
  5203 // Special Registers
  5204 // Return a compressed pointer value
  5205 operand rax_RegN()
  5206 %{
  5207   constraint(ALLOC_IN_RC(int_rax_reg));
  5208   match(RegN);
  5209   match(rRegN);
  5211   format %{ %}
  5212   interface(REG_INTER);
  5213 %}
  5215 // Used in AtomicAdd
  5216 operand rbx_RegP()
  5217 %{
  5218   constraint(ALLOC_IN_RC(ptr_rbx_reg));
  5219   match(RegP);
  5220   match(rRegP);
  5222   format %{ %}
  5223   interface(REG_INTER);
  5224 %}
  5226 operand rsi_RegP()
  5227 %{
  5228   constraint(ALLOC_IN_RC(ptr_rsi_reg));
  5229   match(RegP);
  5230   match(rRegP);
  5232   format %{ %}
  5233   interface(REG_INTER);
  5234 %}
  5236 // Used in rep stosq
  5237 operand rdi_RegP()
  5238 %{
  5239   constraint(ALLOC_IN_RC(ptr_rdi_reg));
  5240   match(RegP);
  5241   match(rRegP);
  5243   format %{ %}
  5244   interface(REG_INTER);
  5245 %}
  5247 operand rbp_RegP()
  5248 %{
  5249   constraint(ALLOC_IN_RC(ptr_rbp_reg));
  5250   match(RegP);
  5251   match(rRegP);
  5253   format %{ %}
  5254   interface(REG_INTER);
  5255 %}
  5257 operand r15_RegP()
  5258 %{
  5259   constraint(ALLOC_IN_RC(ptr_r15_reg));
  5260   match(RegP);
  5261   match(rRegP);
  5263   format %{ %}
  5264   interface(REG_INTER);
  5265 %}
  5267 operand rRegL()
  5268 %{
  5269   constraint(ALLOC_IN_RC(long_reg));
  5270   match(RegL);
  5271   match(rax_RegL);
  5272   match(rdx_RegL);
  5274   format %{ %}
  5275   interface(REG_INTER);
  5276 %}
  5278 // Special Registers
  5279 operand no_rax_rdx_RegL()
  5280 %{
  5281   constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
  5282   match(RegL);
  5283   match(rRegL);
  5285   format %{ %}
  5286   interface(REG_INTER);
  5287 %}
  5289 operand no_rax_RegL()
  5290 %{
  5291   constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
  5292   match(RegL);
  5293   match(rRegL);
  5294   match(rdx_RegL);
  5296   format %{ %}
  5297   interface(REG_INTER);
  5298 %}
  5300 operand no_rcx_RegL()
  5301 %{
  5302   constraint(ALLOC_IN_RC(long_no_rcx_reg));
  5303   match(RegL);
  5304   match(rRegL);
  5306   format %{ %}
  5307   interface(REG_INTER);
  5308 %}
  5310 operand rax_RegL()
  5311 %{
  5312   constraint(ALLOC_IN_RC(long_rax_reg));
  5313   match(RegL);
  5314   match(rRegL);
  5316   format %{ "RAX" %}
  5317   interface(REG_INTER);
  5318 %}
  5320 operand rcx_RegL()
  5321 %{
  5322   constraint(ALLOC_IN_RC(long_rcx_reg));
  5323   match(RegL);
  5324   match(rRegL);
  5326   format %{ %}
  5327   interface(REG_INTER);
  5328 %}
  5330 operand rdx_RegL()
  5331 %{
  5332   constraint(ALLOC_IN_RC(long_rdx_reg));
  5333   match(RegL);
  5334   match(rRegL);
  5336   format %{ %}
  5337   interface(REG_INTER);
  5338 %}
  5340 // Flags register, used as output of compare instructions
  5341 operand rFlagsReg()
  5342 %{
  5343   constraint(ALLOC_IN_RC(int_flags));
  5344   match(RegFlags);
  5346   format %{ "RFLAGS" %}
  5347   interface(REG_INTER);
  5348 %}
  5350 // Flags register, used as output of FLOATING POINT compare instructions
  5351 operand rFlagsRegU()
  5352 %{
  5353   constraint(ALLOC_IN_RC(int_flags));
  5354   match(RegFlags);
  5356   format %{ "RFLAGS_U" %}
  5357   interface(REG_INTER);
  5358 %}
  5360 operand rFlagsRegUCF() %{
  5361   constraint(ALLOC_IN_RC(int_flags));
  5362   match(RegFlags);
  5363   predicate(false);
  5365   format %{ "RFLAGS_U_CF" %}
  5366   interface(REG_INTER);
  5367 %}
  5369 // Float register operands
  5370 operand regF()
  5371 %{
  5372   constraint(ALLOC_IN_RC(float_reg));
  5373   match(RegF);
  5375   format %{ %}
  5376   interface(REG_INTER);
  5377 %}
  5379 // Double register operands
  5380 operand regD() 
  5381 %{
  5382   constraint(ALLOC_IN_RC(double_reg));
  5383   match(RegD);
  5385   format %{ %}
  5386   interface(REG_INTER);
  5387 %}
  5390 //----------Memory Operands----------------------------------------------------
  5391 // Direct Memory Operand
  5392 // operand direct(immP addr)
  5393 // %{
  5394 //   match(addr);
  5396 //   format %{ "[$addr]" %}
  5397 //   interface(MEMORY_INTER) %{
  5398 //     base(0xFFFFFFFF);
  5399 //     index(0x4);
  5400 //     scale(0x0);
  5401 //     disp($addr);
  5402 //   %}
  5403 // %}
  5405 // Indirect Memory Operand
  5406 operand indirect(any_RegP reg)
  5407 %{
  5408   constraint(ALLOC_IN_RC(ptr_reg));
  5409   match(reg);
  5411   format %{ "[$reg]" %}
  5412   interface(MEMORY_INTER) %{
  5413     base($reg);
  5414     index(0x4);
  5415     scale(0x0);
  5416     disp(0x0);
  5417   %}
  5418 %}
  5420 // Indirect Memory Plus Short Offset Operand
  5421 operand indOffset8(any_RegP reg, immL8 off)
  5422 %{
  5423   constraint(ALLOC_IN_RC(ptr_reg));
  5424   match(AddP reg off);
  5426   format %{ "[$reg + $off (8-bit)]" %}
  5427   interface(MEMORY_INTER) %{
  5428     base($reg);
  5429     index(0x4);
  5430     scale(0x0);
  5431     disp($off);
  5432   %}
  5433 %}
  5435 // Indirect Memory Plus Long Offset Operand
  5436 operand indOffset32(any_RegP reg, immL32 off)
  5437 %{
  5438   constraint(ALLOC_IN_RC(ptr_reg));
  5439   match(AddP reg off);
  5441   format %{ "[$reg + $off (32-bit)]" %}
  5442   interface(MEMORY_INTER) %{
  5443     base($reg);
  5444     index(0x4);
  5445     scale(0x0);
  5446     disp($off);
  5447   %}
  5448 %}
  5450 // Indirect Memory Plus Index Register Plus Offset Operand
  5451 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
  5452 %{
  5453   constraint(ALLOC_IN_RC(ptr_reg));
  5454   match(AddP (AddP reg lreg) off);
  5456   op_cost(10);
  5457   format %{"[$reg + $off + $lreg]" %}
  5458   interface(MEMORY_INTER) %{
  5459     base($reg);
  5460     index($lreg);
  5461     scale(0x0);
  5462     disp($off);
  5463   %}
  5464 %}
  5466 // Indirect Memory Plus Index Register Plus Offset Operand
  5467 operand indIndex(any_RegP reg, rRegL lreg)
  5468 %{
  5469   constraint(ALLOC_IN_RC(ptr_reg));
  5470   match(AddP reg lreg);
  5472   op_cost(10);
  5473   format %{"[$reg + $lreg]" %}
  5474   interface(MEMORY_INTER) %{
  5475     base($reg);
  5476     index($lreg);
  5477     scale(0x0);
  5478     disp(0x0);
  5479   %}
  5480 %}
  5482 // Indirect Memory Times Scale Plus Index Register
  5483 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
  5484 %{
  5485   constraint(ALLOC_IN_RC(ptr_reg));
  5486   match(AddP reg (LShiftL lreg scale));
  5488   op_cost(10);
  5489   format %{"[$reg + $lreg << $scale]" %}
  5490   interface(MEMORY_INTER) %{
  5491     base($reg);
  5492     index($lreg);
  5493     scale($scale);
  5494     disp(0x0);
  5495   %}
  5496 %}
  5498 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
  5499 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
  5500 %{
  5501   constraint(ALLOC_IN_RC(ptr_reg));
  5502   match(AddP (AddP reg (LShiftL lreg scale)) off);
  5504   op_cost(10);
  5505   format %{"[$reg + $off + $lreg << $scale]" %}
  5506   interface(MEMORY_INTER) %{
  5507     base($reg);
  5508     index($lreg);
  5509     scale($scale);
  5510     disp($off);
  5511   %}
  5512 %}
  5514 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
  5515 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
  5516 %{
  5517   constraint(ALLOC_IN_RC(ptr_reg));
  5518   predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
  5519   match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
  5521   op_cost(10);
  5522   format %{"[$reg + $off + $idx << $scale]" %}
  5523   interface(MEMORY_INTER) %{
  5524     base($reg);
  5525     index($idx);
  5526     scale($scale);
  5527     disp($off);
  5528   %}
  5529 %}
  5531 // Indirect Narrow Oop Plus Offset Operand
  5532 // Note: x86 architecture doesn't support "scale * index + offset" without a base
  5533 // we can't free r12 even with Universe::narrow_oop_base() == NULL.
  5534 operand indCompressedOopOffset(rRegN reg, immL32 off) %{
  5535   predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
  5536   constraint(ALLOC_IN_RC(ptr_reg));
  5537   match(AddP (DecodeN reg) off);
  5539   op_cost(10);
  5540   format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %}
  5541   interface(MEMORY_INTER) %{
  5542     base(0xc); // R12
  5543     index($reg);
  5544     scale(0x3);
  5545     disp($off);
  5546   %}
  5547 %}
  5549 // Indirect Memory Operand
  5550 operand indirectNarrow(rRegN reg)
  5551 %{
  5552   predicate(Universe::narrow_oop_shift() == 0);
  5553   constraint(ALLOC_IN_RC(ptr_reg));
  5554   match(DecodeN reg);
  5556   format %{ "[$reg]" %}
  5557   interface(MEMORY_INTER) %{
  5558     base($reg);
  5559     index(0x4);
  5560     scale(0x0);
  5561     disp(0x0);
  5562   %}
  5563 %}
  5565 // Indirect Memory Plus Short Offset Operand
  5566 operand indOffset8Narrow(rRegN reg, immL8 off)
  5567 %{
  5568   predicate(Universe::narrow_oop_shift() == 0);
  5569   constraint(ALLOC_IN_RC(ptr_reg));
  5570   match(AddP (DecodeN reg) off);
  5572   format %{ "[$reg + $off (8-bit)]" %}
  5573   interface(MEMORY_INTER) %{
  5574     base($reg);
  5575     index(0x4);
  5576     scale(0x0);
  5577     disp($off);
  5578   %}
  5579 %}
  5581 // Indirect Memory Plus Long Offset Operand
  5582 operand indOffset32Narrow(rRegN reg, immL32 off)
  5583 %{
  5584   predicate(Universe::narrow_oop_shift() == 0);
  5585   constraint(ALLOC_IN_RC(ptr_reg));
  5586   match(AddP (DecodeN reg) off);
  5588   format %{ "[$reg + $off (32-bit)]" %}
  5589   interface(MEMORY_INTER) %{
  5590     base($reg);
  5591     index(0x4);
  5592     scale(0x0);
  5593     disp($off);
  5594   %}
  5595 %}
  5597 // Indirect Memory Plus Index Register Plus Offset Operand
  5598 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off)
  5599 %{
  5600   predicate(Universe::narrow_oop_shift() == 0);
  5601   constraint(ALLOC_IN_RC(ptr_reg));
  5602   match(AddP (AddP (DecodeN reg) lreg) off);
  5604   op_cost(10);
  5605   format %{"[$reg + $off + $lreg]" %}
  5606   interface(MEMORY_INTER) %{
  5607     base($reg);
  5608     index($lreg);
  5609     scale(0x0);
  5610     disp($off);
  5611   %}
  5612 %}
  5614 // Indirect Memory Plus Index Register Plus Offset Operand
  5615 operand indIndexNarrow(rRegN reg, rRegL lreg)
  5616 %{
  5617   predicate(Universe::narrow_oop_shift() == 0);
  5618   constraint(ALLOC_IN_RC(ptr_reg));
  5619   match(AddP (DecodeN reg) lreg);
  5621   op_cost(10);
  5622   format %{"[$reg + $lreg]" %}
  5623   interface(MEMORY_INTER) %{
  5624     base($reg);
  5625     index($lreg);
  5626     scale(0x0);
  5627     disp(0x0);
  5628   %}
  5629 %}
  5631 // Indirect Memory Times Scale Plus Index Register
  5632 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale)
  5633 %{
  5634   predicate(Universe::narrow_oop_shift() == 0);
  5635   constraint(ALLOC_IN_RC(ptr_reg));
  5636   match(AddP (DecodeN reg) (LShiftL lreg scale));
  5638   op_cost(10);
  5639   format %{"[$reg + $lreg << $scale]" %}
  5640   interface(MEMORY_INTER) %{
  5641     base($reg);
  5642     index($lreg);
  5643     scale($scale);
  5644     disp(0x0);
  5645   %}
  5646 %}
  5648 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
  5649 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
  5650 %{
  5651   predicate(Universe::narrow_oop_shift() == 0);
  5652   constraint(ALLOC_IN_RC(ptr_reg));
  5653   match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off);
  5655   op_cost(10);
  5656   format %{"[$reg + $off + $lreg << $scale]" %}
  5657   interface(MEMORY_INTER) %{
  5658     base($reg);
  5659     index($lreg);
  5660     scale($scale);
  5661     disp($off);
  5662   %}
  5663 %}
  5665 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
  5666 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale)
  5667 %{
  5668   constraint(ALLOC_IN_RC(ptr_reg));
  5669   predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
  5670   match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off);
  5672   op_cost(10);
  5673   format %{"[$reg + $off + $idx << $scale]" %}
  5674   interface(MEMORY_INTER) %{
  5675     base($reg);
  5676     index($idx);
  5677     scale($scale);
  5678     disp($off);
  5679   %}
  5680 %}
  5683 //----------Special Memory Operands--------------------------------------------
  5684 // Stack Slot Operand - This operand is used for loading and storing temporary
  5685 //                      values on the stack where a match requires a value to
  5686 //                      flow through memory.
  5687 operand stackSlotP(sRegP reg)
  5688 %{
  5689   constraint(ALLOC_IN_RC(stack_slots));
  5690   // No match rule because this operand is only generated in matching
  5692   format %{ "[$reg]" %}
  5693   interface(MEMORY_INTER) %{
  5694     base(0x4);   // RSP
  5695     index(0x4);  // No Index
  5696     scale(0x0);  // No Scale
  5697     disp($reg);  // Stack Offset
  5698   %}
  5699 %}
  5701 operand stackSlotI(sRegI reg)
  5702 %{
  5703   constraint(ALLOC_IN_RC(stack_slots));
  5704   // No match rule because this operand is only generated in matching
  5706   format %{ "[$reg]" %}
  5707   interface(MEMORY_INTER) %{
  5708     base(0x4);   // RSP
  5709     index(0x4);  // No Index
  5710     scale(0x0);  // No Scale
  5711     disp($reg);  // Stack Offset
  5712   %}
  5713 %}
  5715 operand stackSlotF(sRegF reg)
  5716 %{
  5717   constraint(ALLOC_IN_RC(stack_slots));
  5718   // No match rule because this operand is only generated in matching
  5720   format %{ "[$reg]" %}
  5721   interface(MEMORY_INTER) %{
  5722     base(0x4);   // RSP
  5723     index(0x4);  // No Index
  5724     scale(0x0);  // No Scale
  5725     disp($reg);  // Stack Offset
  5726   %}
  5727 %}
  5729 operand stackSlotD(sRegD reg)
  5730 %{
  5731   constraint(ALLOC_IN_RC(stack_slots));
  5732   // No match rule because this operand is only generated in matching
  5734   format %{ "[$reg]" %}
  5735   interface(MEMORY_INTER) %{
  5736     base(0x4);   // RSP
  5737     index(0x4);  // No Index
  5738     scale(0x0);  // No Scale
  5739     disp($reg);  // Stack Offset
  5740   %}
  5741 %}
  5742 operand stackSlotL(sRegL reg)
  5743 %{
  5744   constraint(ALLOC_IN_RC(stack_slots));
  5745   // No match rule because this operand is only generated in matching
  5747   format %{ "[$reg]" %}
  5748   interface(MEMORY_INTER) %{
  5749     base(0x4);   // RSP
  5750     index(0x4);  // No Index
  5751     scale(0x0);  // No Scale
  5752     disp($reg);  // Stack Offset
  5753   %}
  5754 %}
  5756 //----------Conditional Branch Operands----------------------------------------
  5757 // Comparison Op  - This is the operation of the comparison, and is limited to
  5758 //                  the following set of codes:
  5759 //                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
  5760 //
  5761 // Other attributes of the comparison, such as unsignedness, are specified
  5762 // by the comparison instruction that sets a condition code flags register.
  5763 // That result is represented by a flags operand whose subtype is appropriate
  5764 // to the unsignedness (etc.) of the comparison.
  5765 //
  5766 // Later, the instruction which matches both the Comparison Op (a Bool) and
  5767 // the flags (produced by the Cmp) specifies the coding of the comparison op
  5768 // by matching a specific subtype of Bool operand below, such as cmpOpU.
  5770 // Comparision Code
  5771 operand cmpOp()
  5772 %{
  5773   match(Bool);
  5775   format %{ "" %}
  5776   interface(COND_INTER) %{
  5777     equal(0x4, "e");
  5778     not_equal(0x5, "ne");
  5779     less(0xC, "l");
  5780     greater_equal(0xD, "ge");
  5781     less_equal(0xE, "le");
  5782     greater(0xF, "g");
  5783   %}
  5784 %}
  5786 // Comparison Code, unsigned compare.  Used by FP also, with
  5787 // C2 (unordered) turned into GT or LT already.  The other bits
  5788 // C0 and C3 are turned into Carry & Zero flags.
  5789 operand cmpOpU()
  5790 %{
  5791   match(Bool);
  5793   format %{ "" %}
  5794   interface(COND_INTER) %{
  5795     equal(0x4, "e");
  5796     not_equal(0x5, "ne");
  5797     less(0x2, "b");
  5798     greater_equal(0x3, "nb");
  5799     less_equal(0x6, "be");
  5800     greater(0x7, "nbe");
  5801   %}
  5802 %}
  5805 // Floating comparisons that don't require any fixup for the unordered case
  5806 operand cmpOpUCF() %{
  5807   match(Bool);
  5808   predicate(n->as_Bool()->_test._test == BoolTest::lt ||
  5809             n->as_Bool()->_test._test == BoolTest::ge ||
  5810             n->as_Bool()->_test._test == BoolTest::le ||
  5811             n->as_Bool()->_test._test == BoolTest::gt);
  5812   format %{ "" %}
  5813   interface(COND_INTER) %{
  5814     equal(0x4, "e");
  5815     not_equal(0x5, "ne");
  5816     less(0x2, "b");
  5817     greater_equal(0x3, "nb");
  5818     less_equal(0x6, "be");
  5819     greater(0x7, "nbe");
  5820   %}
  5821 %}
  5824 // Floating comparisons that can be fixed up with extra conditional jumps
  5825 operand cmpOpUCF2() %{
  5826   match(Bool);
  5827   predicate(n->as_Bool()->_test._test == BoolTest::ne ||
  5828             n->as_Bool()->_test._test == BoolTest::eq);
  5829   format %{ "" %}
  5830   interface(COND_INTER) %{
  5831     equal(0x4, "e");
  5832     not_equal(0x5, "ne");
  5833     less(0x2, "b");
  5834     greater_equal(0x3, "nb");
  5835     less_equal(0x6, "be");
  5836     greater(0x7, "nbe");
  5837   %}
  5838 %}
  5841 //----------OPERAND CLASSES----------------------------------------------------
  5842 // Operand Classes are groups of operands that are used as to simplify
  5843 // instruction definitions by not requiring the AD writer to specify separate
  5844 // instructions for every form of operand when the instruction accepts
  5845 // multiple operand types with the same basic encoding and format.  The classic
  5846 // case of this is memory operands.
  5848 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
  5849                indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
  5850                indCompressedOopOffset,
  5851                indirectNarrow, indOffset8Narrow, indOffset32Narrow,
  5852                indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
  5853                indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
  5855 //----------PIPELINE-----------------------------------------------------------
  5856 // Rules which define the behavior of the target architectures pipeline.
  5857 pipeline %{
  5859 //----------ATTRIBUTES---------------------------------------------------------
  5860 attributes %{
  5861   variable_size_instructions;        // Fixed size instructions
  5862   max_instructions_per_bundle = 3;   // Up to 3 instructions per bundle
  5863   instruction_unit_size = 1;         // An instruction is 1 bytes long
  5864   instruction_fetch_unit_size = 16;  // The processor fetches one line
  5865   instruction_fetch_units = 1;       // of 16 bytes
  5867   // List of nop instructions
  5868   nops( MachNop );
  5869 %}
  5871 //----------RESOURCES----------------------------------------------------------
  5872 // Resources are the functional units available to the machine
  5874 // Generic P2/P3 pipeline
  5875 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
  5876 // 3 instructions decoded per cycle.
  5877 // 2 load/store ops per cycle, 1 branch, 1 FPU,
  5878 // 3 ALU op, only ALU0 handles mul instructions.
  5879 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
  5880            MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
  5881            BR, FPU,
  5882            ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
  5884 //----------PIPELINE DESCRIPTION-----------------------------------------------
  5885 // Pipeline Description specifies the stages in the machine's pipeline
  5887 // Generic P2/P3 pipeline
  5888 pipe_desc(S0, S1, S2, S3, S4, S5);
  5890 //----------PIPELINE CLASSES---------------------------------------------------
  5891 // Pipeline Classes describe the stages in which input and output are
  5892 // referenced by the hardware pipeline.
  5894 // Naming convention: ialu or fpu
  5895 // Then: _reg
  5896 // Then: _reg if there is a 2nd register
  5897 // Then: _long if it's a pair of instructions implementing a long
  5898 // Then: _fat if it requires the big decoder
  5899 //   Or: _mem if it requires the big decoder and a memory unit.
  5901 // Integer ALU reg operation
  5902 pipe_class ialu_reg(rRegI dst)
  5903 %{
  5904     single_instruction;
  5905     dst    : S4(write);
  5906     dst    : S3(read);
  5907     DECODE : S0;        // any decoder
  5908     ALU    : S3;        // any alu
  5909 %}
  5911 // Long ALU reg operation
  5912 pipe_class ialu_reg_long(rRegL dst)
  5913 %{
  5914     instruction_count(2);
  5915     dst    : S4(write);
  5916     dst    : S3(read);
  5917     DECODE : S0(2);     // any 2 decoders
  5918     ALU    : S3(2);     // both alus
  5919 %}
  5921 // Integer ALU reg operation using big decoder
  5922 pipe_class ialu_reg_fat(rRegI dst)
  5923 %{
  5924     single_instruction;
  5925     dst    : S4(write);
  5926     dst    : S3(read);
  5927     D0     : S0;        // big decoder only
  5928     ALU    : S3;        // any alu
  5929 %}
  5931 // Long ALU reg operation using big decoder
  5932 pipe_class ialu_reg_long_fat(rRegL dst)
  5933 %{
  5934     instruction_count(2);
  5935     dst    : S4(write);
  5936     dst    : S3(read);
  5937     D0     : S0(2);     // big decoder only; twice
  5938     ALU    : S3(2);     // any 2 alus
  5939 %}
  5941 // Integer ALU reg-reg operation
  5942 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
  5943 %{
  5944     single_instruction;
  5945     dst    : S4(write);
  5946     src    : S3(read);
  5947     DECODE : S0;        // any decoder
  5948     ALU    : S3;        // any alu
  5949 %}
  5951 // Long ALU reg-reg operation
  5952 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
  5953 %{
  5954     instruction_count(2);
  5955     dst    : S4(write);
  5956     src    : S3(read);
  5957     DECODE : S0(2);     // any 2 decoders
  5958     ALU    : S3(2);     // both alus
  5959 %}
  5961 // Integer ALU reg-reg operation
  5962 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
  5963 %{
  5964     single_instruction;
  5965     dst    : S4(write);
  5966     src    : S3(read);
  5967     D0     : S0;        // big decoder only
  5968     ALU    : S3;        // any alu
  5969 %}
  5971 // Long ALU reg-reg operation
  5972 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
  5973 %{
  5974     instruction_count(2);
  5975     dst    : S4(write);
  5976     src    : S3(read);
  5977     D0     : S0(2);     // big decoder only; twice
  5978     ALU    : S3(2);     // both alus
  5979 %}
  5981 // Integer ALU reg-mem operation
  5982 pipe_class ialu_reg_mem(rRegI dst, memory mem)
  5983 %{
  5984     single_instruction;
  5985     dst    : S5(write);
  5986     mem    : S3(read);
  5987     D0     : S0;        // big decoder only
  5988     ALU    : S4;        // any alu
  5989     MEM    : S3;        // any mem
  5990 %}
  5992 // Integer mem operation (prefetch)
  5993 pipe_class ialu_mem(memory mem)
  5994 %{
  5995     single_instruction;
  5996     mem    : S3(read);
  5997     D0     : S0;        // big decoder only
  5998     MEM    : S3;        // any mem
  5999 %}
  6001 // Integer Store to Memory
  6002 pipe_class ialu_mem_reg(memory mem, rRegI src)
  6003 %{
  6004     single_instruction;
  6005     mem    : S3(read);
  6006     src    : S5(read);
  6007     D0     : S0;        // big decoder only
  6008     ALU    : S4;        // any alu
  6009     MEM    : S3;
  6010 %}
  6012 // // Long Store to Memory
  6013 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
  6014 // %{
  6015 //     instruction_count(2);
  6016 //     mem    : S3(read);
  6017 //     src    : S5(read);
  6018 //     D0     : S0(2);          // big decoder only; twice
  6019 //     ALU    : S4(2);     // any 2 alus
  6020 //     MEM    : S3(2);  // Both mems
  6021 // %}
  6023 // Integer Store to Memory
  6024 pipe_class ialu_mem_imm(memory mem)
  6025 %{
  6026     single_instruction;
  6027     mem    : S3(read);
  6028     D0     : S0;        // big decoder only
  6029     ALU    : S4;        // any alu
  6030     MEM    : S3;
  6031 %}
  6033 // Integer ALU0 reg-reg operation
  6034 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
  6035 %{
  6036     single_instruction;
  6037     dst    : S4(write);
  6038     src    : S3(read);
  6039     D0     : S0;        // Big decoder only
  6040     ALU0   : S3;        // only alu0
  6041 %}
  6043 // Integer ALU0 reg-mem operation
  6044 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
  6045 %{
  6046     single_instruction;
  6047     dst    : S5(write);
  6048     mem    : S3(read);
  6049     D0     : S0;        // big decoder only
  6050     ALU0   : S4;        // ALU0 only
  6051     MEM    : S3;        // any mem
  6052 %}
  6054 // Integer ALU reg-reg operation
  6055 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
  6056 %{
  6057     single_instruction;
  6058     cr     : S4(write);
  6059     src1   : S3(read);
  6060     src2   : S3(read);
  6061     DECODE : S0;        // any decoder
  6062     ALU    : S3;        // any alu
  6063 %}
  6065 // Integer ALU reg-imm operation
  6066 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
  6067 %{
  6068     single_instruction;
  6069     cr     : S4(write);
  6070     src1   : S3(read);
  6071     DECODE : S0;        // any decoder
  6072     ALU    : S3;        // any alu
  6073 %}
  6075 // Integer ALU reg-mem operation
  6076 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
  6077 %{
  6078     single_instruction;
  6079     cr     : S4(write);
  6080     src1   : S3(read);
  6081     src2   : S3(read);
  6082     D0     : S0;        // big decoder only
  6083     ALU    : S4;        // any alu
  6084     MEM    : S3;
  6085 %}
  6087 // Conditional move reg-reg
  6088 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
  6089 %{
  6090     instruction_count(4);
  6091     y      : S4(read);
  6092     q      : S3(read);
  6093     p      : S3(read);
  6094     DECODE : S0(4);     // any decoder
  6095 %}
  6097 // Conditional move reg-reg
  6098 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
  6099 %{
  6100     single_instruction;
  6101     dst    : S4(write);
  6102     src    : S3(read);
  6103     cr     : S3(read);
  6104     DECODE : S0;        // any decoder
  6105 %}
  6107 // Conditional move reg-mem
  6108 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
  6109 %{
  6110     single_instruction;
  6111     dst    : S4(write);
  6112     src    : S3(read);
  6113     cr     : S3(read);
  6114     DECODE : S0;        // any decoder
  6115     MEM    : S3;
  6116 %}
  6118 // Conditional move reg-reg long
  6119 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
  6120 %{
  6121     single_instruction;
  6122     dst    : S4(write);
  6123     src    : S3(read);
  6124     cr     : S3(read);
  6125     DECODE : S0(2);     // any 2 decoders
  6126 %}
  6128 // XXX
  6129 // // Conditional move double reg-reg
  6130 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
  6131 // %{
  6132 //     single_instruction;
  6133 //     dst    : S4(write);
  6134 //     src    : S3(read);
  6135 //     cr     : S3(read);
  6136 //     DECODE : S0;     // any decoder
  6137 // %}
  6139 // Float reg-reg operation
  6140 pipe_class fpu_reg(regD dst)
  6141 %{
  6142     instruction_count(2);
  6143     dst    : S3(read);
  6144     DECODE : S0(2);     // any 2 decoders
  6145     FPU    : S3;
  6146 %}
  6148 // Float reg-reg operation
  6149 pipe_class fpu_reg_reg(regD dst, regD src)
  6150 %{
  6151     instruction_count(2);
  6152     dst    : S4(write);
  6153     src    : S3(read);
  6154     DECODE : S0(2);     // any 2 decoders
  6155     FPU    : S3;
  6156 %}
  6158 // Float reg-reg operation
  6159 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
  6160 %{
  6161     instruction_count(3);
  6162     dst    : S4(write);
  6163     src1   : S3(read);
  6164     src2   : S3(read);
  6165     DECODE : S0(3);     // any 3 decoders
  6166     FPU    : S3(2);
  6167 %}
  6169 // Float reg-reg operation
  6170 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
  6171 %{
  6172     instruction_count(4);
  6173     dst    : S4(write);
  6174     src1   : S3(read);
  6175     src2   : S3(read);
  6176     src3   : S3(read);
  6177     DECODE : S0(4);     // any 3 decoders
  6178     FPU    : S3(2);
  6179 %}
  6181 // Float reg-reg operation
  6182 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
  6183 %{
  6184     instruction_count(4);
  6185     dst    : S4(write);
  6186     src1   : S3(read);
  6187     src2   : S3(read);
  6188     src3   : S3(read);
  6189     DECODE : S1(3);     // any 3 decoders
  6190     D0     : S0;        // Big decoder only
  6191     FPU    : S3(2);
  6192     MEM    : S3;
  6193 %}
  6195 // Float reg-mem operation
  6196 pipe_class fpu_reg_mem(regD dst, memory mem)
  6197 %{
  6198     instruction_count(2);
  6199     dst    : S5(write);
  6200     mem    : S3(read);
  6201     D0     : S0;        // big decoder only
  6202     DECODE : S1;        // any decoder for FPU POP
  6203     FPU    : S4;
  6204     MEM    : S3;        // any mem
  6205 %}
  6207 // Float reg-mem operation
  6208 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
  6209 %{
  6210     instruction_count(3);
  6211     dst    : S5(write);
  6212     src1   : S3(read);
  6213     mem    : S3(read);
  6214     D0     : S0;        // big decoder only
  6215     DECODE : S1(2);     // any decoder for FPU POP
  6216     FPU    : S4;
  6217     MEM    : S3;        // any mem
  6218 %}
  6220 // Float mem-reg operation
  6221 pipe_class fpu_mem_reg(memory mem, regD src)
  6222 %{
  6223     instruction_count(2);
  6224     src    : S5(read);
  6225     mem    : S3(read);
  6226     DECODE : S0;        // any decoder for FPU PUSH
  6227     D0     : S1;        // big decoder only
  6228     FPU    : S4;
  6229     MEM    : S3;        // any mem
  6230 %}
  6232 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
  6233 %{
  6234     instruction_count(3);
  6235     src1   : S3(read);
  6236     src2   : S3(read);
  6237     mem    : S3(read);
  6238     DECODE : S0(2);     // any decoder for FPU PUSH
  6239     D0     : S1;        // big decoder only
  6240     FPU    : S4;
  6241     MEM    : S3;        // any mem
  6242 %}
  6244 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
  6245 %{
  6246     instruction_count(3);
  6247     src1   : S3(read);
  6248     src2   : S3(read);
  6249     mem    : S4(read);
  6250     DECODE : S0;        // any decoder for FPU PUSH
  6251     D0     : S0(2);     // big decoder only
  6252     FPU    : S4;
  6253     MEM    : S3(2);     // any mem
  6254 %}
  6256 pipe_class fpu_mem_mem(memory dst, memory src1)
  6257 %{
  6258     instruction_count(2);
  6259     src1   : S3(read);
  6260     dst    : S4(read);
  6261     D0     : S0(2);     // big decoder only
  6262     MEM    : S3(2);     // any mem
  6263 %}
  6265 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
  6266 %{
  6267     instruction_count(3);
  6268     src1   : S3(read);
  6269     src2   : S3(read);
  6270     dst    : S4(read);
  6271     D0     : S0(3);     // big decoder only
  6272     FPU    : S4;
  6273     MEM    : S3(3);     // any mem
  6274 %}
  6276 pipe_class fpu_mem_reg_con(memory mem, regD src1)
  6277 %{
  6278     instruction_count(3);
  6279     src1   : S4(read);
  6280     mem    : S4(read);
  6281     DECODE : S0;        // any decoder for FPU PUSH
  6282     D0     : S0(2);     // big decoder only
  6283     FPU    : S4;
  6284     MEM    : S3(2);     // any mem
  6285 %}
  6287 // Float load constant
  6288 pipe_class fpu_reg_con(regD dst)
  6289 %{
  6290     instruction_count(2);
  6291     dst    : S5(write);
  6292     D0     : S0;        // big decoder only for the load
  6293     DECODE : S1;        // any decoder for FPU POP
  6294     FPU    : S4;
  6295     MEM    : S3;        // any mem
  6296 %}
  6298 // Float load constant
  6299 pipe_class fpu_reg_reg_con(regD dst, regD src)
  6300 %{
  6301     instruction_count(3);
  6302     dst    : S5(write);
  6303     src    : S3(read);
  6304     D0     : S0;        // big decoder only for the load
  6305     DECODE : S1(2);     // any decoder for FPU POP
  6306     FPU    : S4;
  6307     MEM    : S3;        // any mem
  6308 %}
  6310 // UnConditional branch
  6311 pipe_class pipe_jmp(label labl)
  6312 %{
  6313     single_instruction;
  6314     BR   : S3;
  6315 %}
  6317 // Conditional branch
  6318 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
  6319 %{
  6320     single_instruction;
  6321     cr    : S1(read);
  6322     BR    : S3;
  6323 %}
  6325 // Allocation idiom
  6326 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
  6327 %{
  6328     instruction_count(1); force_serialization;
  6329     fixed_latency(6);
  6330     heap_ptr : S3(read);
  6331     DECODE   : S0(3);
  6332     D0       : S2;
  6333     MEM      : S3;
  6334     ALU      : S3(2);
  6335     dst      : S5(write);
  6336     BR       : S5;
  6337 %}
  6339 // Generic big/slow expanded idiom
  6340 pipe_class pipe_slow()
  6341 %{
  6342     instruction_count(10); multiple_bundles; force_serialization;
  6343     fixed_latency(100);
  6344     D0  : S0(2);
  6345     MEM : S3(2);
  6346 %}
  6348 // The real do-nothing guy
  6349 pipe_class empty()
  6350 %{
  6351     instruction_count(0);
  6352 %}
  6354 // Define the class for the Nop node
  6355 define
  6356 %{
  6357    MachNop = empty;
  6358 %}
  6360 %}
  6362 //----------INSTRUCTIONS-------------------------------------------------------
  6363 //
  6364 // match      -- States which machine-independent subtree may be replaced
  6365 //               by this instruction.
  6366 // ins_cost   -- The estimated cost of this instruction is used by instruction
  6367 //               selection to identify a minimum cost tree of machine
  6368 //               instructions that matches a tree of machine-independent
  6369 //               instructions.
  6370 // format     -- A string providing the disassembly for this instruction.
  6371 //               The value of an instruction's operand may be inserted
  6372 //               by referring to it with a '$' prefix.
  6373 // opcode     -- Three instruction opcodes may be provided.  These are referred
  6374 //               to within an encode class as $primary, $secondary, and $tertiary
  6375 //               rrspectively.  The primary opcode is commonly used to
  6376 //               indicate the type of machine instruction, while secondary
  6377 //               and tertiary are often used for prefix options or addressing
  6378 //               modes.
  6379 // ins_encode -- A list of encode classes with parameters. The encode class
  6380 //               name must have been defined in an 'enc_class' specification
  6381 //               in the encode section of the architecture description.
  6384 //----------Load/Store/Move Instructions---------------------------------------
  6385 //----------Load Instructions--------------------------------------------------
  6387 // Load Byte (8 bit signed)
  6388 instruct loadB(rRegI dst, memory mem)
  6389 %{
  6390   match(Set dst (LoadB mem));
  6392   ins_cost(125);
  6393   format %{ "movsbl  $dst, $mem\t# byte" %}
  6395   ins_encode %{
  6396     __ movsbl($dst$$Register, $mem$$Address);
  6397   %}
  6399   ins_pipe(ialu_reg_mem);
  6400 %}
  6402 // Load Byte (8 bit signed) into Long Register
  6403 instruct loadB2L(rRegL dst, memory mem)
  6404 %{
  6405   match(Set dst (ConvI2L (LoadB mem)));
  6407   ins_cost(125);
  6408   format %{ "movsbq  $dst, $mem\t# byte -> long" %}
  6410   ins_encode %{
  6411     __ movsbq($dst$$Register, $mem$$Address);
  6412   %}
  6414   ins_pipe(ialu_reg_mem);
  6415 %}
  6417 // Load Unsigned Byte (8 bit UNsigned)
  6418 instruct loadUB(rRegI dst, memory mem)
  6419 %{
  6420   match(Set dst (LoadUB mem));
  6422   ins_cost(125);
  6423   format %{ "movzbl  $dst, $mem\t# ubyte" %}
  6425   ins_encode %{
  6426     __ movzbl($dst$$Register, $mem$$Address);
  6427   %}
  6429   ins_pipe(ialu_reg_mem);
  6430 %}
  6432 // Load Unsigned Byte (8 bit UNsigned) into Long Register
  6433 instruct loadUB2L(rRegL dst, memory mem)
  6434 %{
  6435   match(Set dst (ConvI2L (LoadUB mem)));
  6437   ins_cost(125);
  6438   format %{ "movzbq  $dst, $mem\t# ubyte -> long" %}
  6440   ins_encode %{
  6441     __ movzbq($dst$$Register, $mem$$Address);
  6442   %}
  6444   ins_pipe(ialu_reg_mem);
  6445 %}
  6447 // Load Short (16 bit signed)
  6448 instruct loadS(rRegI dst, memory mem)
  6449 %{
  6450   match(Set dst (LoadS mem));
  6452   ins_cost(125);
  6453   format %{ "movswl $dst, $mem\t# short" %}
  6455   ins_encode %{
  6456     __ movswl($dst$$Register, $mem$$Address);
  6457   %}
  6459   ins_pipe(ialu_reg_mem);
  6460 %}
  6462 // Load Short (16 bit signed) into Long Register
  6463 instruct loadS2L(rRegL dst, memory mem)
  6464 %{
  6465   match(Set dst (ConvI2L (LoadS mem)));
  6467   ins_cost(125);
  6468   format %{ "movswq $dst, $mem\t# short -> long" %}
  6470   ins_encode %{
  6471     __ movswq($dst$$Register, $mem$$Address);
  6472   %}
  6474   ins_pipe(ialu_reg_mem);
  6475 %}
  6477 // Load Unsigned Short/Char (16 bit UNsigned)
  6478 instruct loadUS(rRegI dst, memory mem)
  6479 %{
  6480   match(Set dst (LoadUS mem));
  6482   ins_cost(125);
  6483   format %{ "movzwl  $dst, $mem\t# ushort/char" %}
  6485   ins_encode %{
  6486     __ movzwl($dst$$Register, $mem$$Address);
  6487   %}
  6489   ins_pipe(ialu_reg_mem);
  6490 %}
  6492 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
  6493 instruct loadUS2L(rRegL dst, memory mem)
  6494 %{
  6495   match(Set dst (ConvI2L (LoadUS mem)));
  6497   ins_cost(125);
  6498   format %{ "movzwq  $dst, $mem\t# ushort/char -> long" %}
  6500   ins_encode %{
  6501     __ movzwq($dst$$Register, $mem$$Address);
  6502   %}
  6504   ins_pipe(ialu_reg_mem);
  6505 %}
  6507 // Load Integer
  6508 instruct loadI(rRegI dst, memory mem)
  6509 %{
  6510   match(Set dst (LoadI mem));
  6512   ins_cost(125);
  6513   format %{ "movl    $dst, $mem\t# int" %}
  6515   ins_encode %{
  6516     __ movl($dst$$Register, $mem$$Address);
  6517   %}
  6519   ins_pipe(ialu_reg_mem);
  6520 %}
  6522 // Load Integer into Long Register
  6523 instruct loadI2L(rRegL dst, memory mem)
  6524 %{
  6525   match(Set dst (ConvI2L (LoadI mem)));
  6527   ins_cost(125);
  6528   format %{ "movslq  $dst, $mem\t# int -> long" %}
  6530   ins_encode %{
  6531     __ movslq($dst$$Register, $mem$$Address);
  6532   %}
  6534   ins_pipe(ialu_reg_mem);
  6535 %}
  6537 // Load Unsigned Integer into Long Register
  6538 instruct loadUI2L(rRegL dst, memory mem)
  6539 %{
  6540   match(Set dst (LoadUI2L mem));
  6542   ins_cost(125);
  6543   format %{ "movl    $dst, $mem\t# uint -> long" %}
  6545   ins_encode %{
  6546     __ movl($dst$$Register, $mem$$Address);
  6547   %}
  6549   ins_pipe(ialu_reg_mem);
  6550 %}
  6552 // Load Long
  6553 instruct loadL(rRegL dst, memory mem)
  6554 %{
  6555   match(Set dst (LoadL mem));
  6557   ins_cost(125);
  6558   format %{ "movq    $dst, $mem\t# long" %}
  6560   ins_encode %{
  6561     __ movq($dst$$Register, $mem$$Address);
  6562   %}
  6564   ins_pipe(ialu_reg_mem); // XXX
  6565 %}
  6567 // Load Range
  6568 instruct loadRange(rRegI dst, memory mem)
  6569 %{
  6570   match(Set dst (LoadRange mem));
  6572   ins_cost(125); // XXX
  6573   format %{ "movl    $dst, $mem\t# range" %}
  6574   opcode(0x8B);
  6575   ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
  6576   ins_pipe(ialu_reg_mem);
  6577 %}
  6579 // Load Pointer
  6580 instruct loadP(rRegP dst, memory mem)
  6581 %{
  6582   match(Set dst (LoadP mem));
  6584   ins_cost(125); // XXX
  6585   format %{ "movq    $dst, $mem\t# ptr" %}
  6586   opcode(0x8B);
  6587   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6588   ins_pipe(ialu_reg_mem); // XXX
  6589 %}
  6591 // Load Compressed Pointer
  6592 instruct loadN(rRegN dst, memory mem)
  6593 %{
  6594    match(Set dst (LoadN mem));
  6596    ins_cost(125); // XXX
  6597    format %{ "movl    $dst, $mem\t# compressed ptr" %}
  6598    ins_encode %{
  6599      __ movl($dst$$Register, $mem$$Address);
  6600    %}
  6601    ins_pipe(ialu_reg_mem); // XXX
  6602 %}
  6605 // Load Klass Pointer
  6606 instruct loadKlass(rRegP dst, memory mem)
  6607 %{
  6608   match(Set dst (LoadKlass mem));
  6610   ins_cost(125); // XXX
  6611   format %{ "movq    $dst, $mem\t# class" %}
  6612   opcode(0x8B);
  6613   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6614   ins_pipe(ialu_reg_mem); // XXX
  6615 %}
  6617 // Load narrow Klass Pointer
  6618 instruct loadNKlass(rRegN dst, memory mem)
  6619 %{
  6620   match(Set dst (LoadNKlass mem));
  6622   ins_cost(125); // XXX
  6623   format %{ "movl    $dst, $mem\t# compressed klass ptr" %}
  6624   ins_encode %{
  6625     __ movl($dst$$Register, $mem$$Address);
  6626   %}
  6627   ins_pipe(ialu_reg_mem); // XXX
  6628 %}
  6630 // Load Float
  6631 instruct loadF(regF dst, memory mem)
  6632 %{
  6633   match(Set dst (LoadF mem));
  6635   ins_cost(145); // XXX
  6636   format %{ "movss   $dst, $mem\t# float" %}
  6637   opcode(0xF3, 0x0F, 0x10);
  6638   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
  6639   ins_pipe(pipe_slow); // XXX
  6640 %}
  6642 // Load Double
  6643 instruct loadD_partial(regD dst, memory mem)
  6644 %{
  6645   predicate(!UseXmmLoadAndClearUpper);
  6646   match(Set dst (LoadD mem));
  6648   ins_cost(145); // XXX
  6649   format %{ "movlpd  $dst, $mem\t# double" %}
  6650   opcode(0x66, 0x0F, 0x12);
  6651   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
  6652   ins_pipe(pipe_slow); // XXX
  6653 %}
  6655 instruct loadD(regD dst, memory mem)
  6656 %{
  6657   predicate(UseXmmLoadAndClearUpper);
  6658   match(Set dst (LoadD mem));
  6660   ins_cost(145); // XXX
  6661   format %{ "movsd   $dst, $mem\t# double" %}
  6662   opcode(0xF2, 0x0F, 0x10);
  6663   ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
  6664   ins_pipe(pipe_slow); // XXX
  6665 %}
  6667 // Load Aligned Packed Byte to XMM register
  6668 instruct loadA8B(regD dst, memory mem) %{
  6669   match(Set dst (Load8B mem));
  6670   ins_cost(125);
  6671   format %{ "MOVQ  $dst,$mem\t! packed8B" %}
  6672   ins_encode( movq_ld(dst, mem));
  6673   ins_pipe( pipe_slow );
  6674 %}
  6676 // Load Aligned Packed Short to XMM register
  6677 instruct loadA4S(regD dst, memory mem) %{
  6678   match(Set dst (Load4S mem));
  6679   ins_cost(125);
  6680   format %{ "MOVQ  $dst,$mem\t! packed4S" %}
  6681   ins_encode( movq_ld(dst, mem));
  6682   ins_pipe( pipe_slow );
  6683 %}
  6685 // Load Aligned Packed Char to XMM register
  6686 instruct loadA4C(regD dst, memory mem) %{
  6687   match(Set dst (Load4C mem));
  6688   ins_cost(125);
  6689   format %{ "MOVQ  $dst,$mem\t! packed4C" %}
  6690   ins_encode( movq_ld(dst, mem));
  6691   ins_pipe( pipe_slow );
  6692 %}
  6694 // Load Aligned Packed Integer to XMM register
  6695 instruct load2IU(regD dst, memory mem) %{
  6696   match(Set dst (Load2I mem));
  6697   ins_cost(125);
  6698   format %{ "MOVQ  $dst,$mem\t! packed2I" %}
  6699   ins_encode( movq_ld(dst, mem));
  6700   ins_pipe( pipe_slow );
  6701 %}
  6703 // Load Aligned Packed Single to XMM
  6704 instruct loadA2F(regD dst, memory mem) %{
  6705   match(Set dst (Load2F mem));
  6706   ins_cost(145);
  6707   format %{ "MOVQ  $dst,$mem\t! packed2F" %}
  6708   ins_encode( movq_ld(dst, mem));
  6709   ins_pipe( pipe_slow );
  6710 %}
  6712 // Load Effective Address
  6713 instruct leaP8(rRegP dst, indOffset8 mem)
  6714 %{
  6715   match(Set dst mem);
  6717   ins_cost(110); // XXX
  6718   format %{ "leaq    $dst, $mem\t# ptr 8" %}
  6719   opcode(0x8D);
  6720   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6721   ins_pipe(ialu_reg_reg_fat);
  6722 %}
  6724 instruct leaP32(rRegP dst, indOffset32 mem)
  6725 %{
  6726   match(Set dst mem);
  6728   ins_cost(110);
  6729   format %{ "leaq    $dst, $mem\t# ptr 32" %}
  6730   opcode(0x8D);
  6731   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6732   ins_pipe(ialu_reg_reg_fat);
  6733 %}
  6735 // instruct leaPIdx(rRegP dst, indIndex mem)
  6736 // %{
  6737 //   match(Set dst mem);
  6739 //   ins_cost(110);
  6740 //   format %{ "leaq    $dst, $mem\t# ptr idx" %}
  6741 //   opcode(0x8D);
  6742 //   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6743 //   ins_pipe(ialu_reg_reg_fat);
  6744 // %}
  6746 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
  6747 %{
  6748   match(Set dst mem);
  6750   ins_cost(110);
  6751   format %{ "leaq    $dst, $mem\t# ptr idxoff" %}
  6752   opcode(0x8D);
  6753   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6754   ins_pipe(ialu_reg_reg_fat);
  6755 %}
  6757 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
  6758 %{
  6759   match(Set dst mem);
  6761   ins_cost(110);
  6762   format %{ "leaq    $dst, $mem\t# ptr idxscale" %}
  6763   opcode(0x8D);
  6764   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6765   ins_pipe(ialu_reg_reg_fat);
  6766 %}
  6768 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
  6769 %{
  6770   match(Set dst mem);
  6772   ins_cost(110);
  6773   format %{ "leaq    $dst, $mem\t# ptr idxscaleoff" %}
  6774   opcode(0x8D);
  6775   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6776   ins_pipe(ialu_reg_reg_fat);
  6777 %}
  6779 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem)
  6780 %{
  6781   match(Set dst mem);
  6783   ins_cost(110);
  6784   format %{ "leaq    $dst, $mem\t# ptr posidxscaleoff" %}
  6785   opcode(0x8D);
  6786   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6787   ins_pipe(ialu_reg_reg_fat);
  6788 %}
  6790 // Load Effective Address which uses Narrow (32-bits) oop
  6791 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem)
  6792 %{
  6793   predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
  6794   match(Set dst mem);
  6796   ins_cost(110);
  6797   format %{ "leaq    $dst, $mem\t# ptr compressedoopoff32" %}
  6798   opcode(0x8D);
  6799   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6800   ins_pipe(ialu_reg_reg_fat);
  6801 %}
  6803 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem)
  6804 %{
  6805   predicate(Universe::narrow_oop_shift() == 0);
  6806   match(Set dst mem);
  6808   ins_cost(110); // XXX
  6809   format %{ "leaq    $dst, $mem\t# ptr off8narrow" %}
  6810   opcode(0x8D);
  6811   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6812   ins_pipe(ialu_reg_reg_fat);
  6813 %}
  6815 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem)
  6816 %{
  6817   predicate(Universe::narrow_oop_shift() == 0);
  6818   match(Set dst mem);
  6820   ins_cost(110);
  6821   format %{ "leaq    $dst, $mem\t# ptr off32narrow" %}
  6822   opcode(0x8D);
  6823   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6824   ins_pipe(ialu_reg_reg_fat);
  6825 %}
  6827 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem)
  6828 %{
  6829   predicate(Universe::narrow_oop_shift() == 0);
  6830   match(Set dst mem);
  6832   ins_cost(110);
  6833   format %{ "leaq    $dst, $mem\t# ptr idxoffnarrow" %}
  6834   opcode(0x8D);
  6835   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6836   ins_pipe(ialu_reg_reg_fat);
  6837 %}
  6839 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem)
  6840 %{
  6841   predicate(Universe::narrow_oop_shift() == 0);
  6842   match(Set dst mem);
  6844   ins_cost(110);
  6845   format %{ "leaq    $dst, $mem\t# ptr idxscalenarrow" %}
  6846   opcode(0x8D);
  6847   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6848   ins_pipe(ialu_reg_reg_fat);
  6849 %}
  6851 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem)
  6852 %{
  6853   predicate(Universe::narrow_oop_shift() == 0);
  6854   match(Set dst mem);
  6856   ins_cost(110);
  6857   format %{ "leaq    $dst, $mem\t# ptr idxscaleoffnarrow" %}
  6858   opcode(0x8D);
  6859   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6860   ins_pipe(ialu_reg_reg_fat);
  6861 %}
  6863 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem)
  6864 %{
  6865   predicate(Universe::narrow_oop_shift() == 0);
  6866   match(Set dst mem);
  6868   ins_cost(110);
  6869   format %{ "leaq    $dst, $mem\t# ptr posidxscaleoffnarrow" %}
  6870   opcode(0x8D);
  6871   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  6872   ins_pipe(ialu_reg_reg_fat);
  6873 %}
  6875 instruct loadConI(rRegI dst, immI src)
  6876 %{
  6877   match(Set dst src);
  6879   format %{ "movl    $dst, $src\t# int" %}
  6880   ins_encode(load_immI(dst, src));
  6881   ins_pipe(ialu_reg_fat); // XXX
  6882 %}
  6884 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
  6885 %{
  6886   match(Set dst src);
  6887   effect(KILL cr);
  6889   ins_cost(50);
  6890   format %{ "xorl    $dst, $dst\t# int" %}
  6891   opcode(0x33); /* + rd */
  6892   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
  6893   ins_pipe(ialu_reg);
  6894 %}
  6896 instruct loadConL(rRegL dst, immL src)
  6897 %{
  6898   match(Set dst src);
  6900   ins_cost(150);
  6901   format %{ "movq    $dst, $src\t# long" %}
  6902   ins_encode(load_immL(dst, src));
  6903   ins_pipe(ialu_reg);
  6904 %}
  6906 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
  6907 %{
  6908   match(Set dst src);
  6909   effect(KILL cr);
  6911   ins_cost(50);
  6912   format %{ "xorl    $dst, $dst\t# long" %}
  6913   opcode(0x33); /* + rd */
  6914   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
  6915   ins_pipe(ialu_reg); // XXX
  6916 %}
  6918 instruct loadConUL32(rRegL dst, immUL32 src)
  6919 %{
  6920   match(Set dst src);
  6922   ins_cost(60);
  6923   format %{ "movl    $dst, $src\t# long (unsigned 32-bit)" %}
  6924   ins_encode(load_immUL32(dst, src));
  6925   ins_pipe(ialu_reg);
  6926 %}
  6928 instruct loadConL32(rRegL dst, immL32 src)
  6929 %{
  6930   match(Set dst src);
  6932   ins_cost(70);
  6933   format %{ "movq    $dst, $src\t# long (32-bit)" %}
  6934   ins_encode(load_immL32(dst, src));
  6935   ins_pipe(ialu_reg);
  6936 %}
  6938 instruct loadConP(rRegP dst, immP src)
  6939 %{
  6940   match(Set dst src);
  6942   format %{ "movq    $dst, $src\t# ptr" %}
  6943   ins_encode(load_immP(dst, src));
  6944   ins_pipe(ialu_reg_fat); // XXX
  6945 %}
  6947 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
  6948 %{
  6949   match(Set dst src);
  6950   effect(KILL cr);
  6952   ins_cost(50);
  6953   format %{ "xorl    $dst, $dst\t# ptr" %}
  6954   opcode(0x33); /* + rd */
  6955   ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
  6956   ins_pipe(ialu_reg);
  6957 %}
  6959 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
  6960 %{
  6961   match(Set dst src);
  6962   effect(KILL cr);
  6964   ins_cost(60);
  6965   format %{ "movl    $dst, $src\t# ptr (positive 32-bit)" %}
  6966   ins_encode(load_immP31(dst, src));
  6967   ins_pipe(ialu_reg);
  6968 %}
  6970 instruct loadConF(regF dst, immF src)
  6971 %{
  6972   match(Set dst src);
  6973   ins_cost(125);
  6975   format %{ "movss   $dst, [$src]" %}
  6976   ins_encode(load_conF(dst, src));
  6977   ins_pipe(pipe_slow);
  6978 %}
  6980 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
  6981   match(Set dst src);
  6982   effect(KILL cr);
  6983   format %{ "xorq    $dst, $src\t# compressed NULL ptr" %}
  6984   ins_encode %{
  6985     __ xorq($dst$$Register, $dst$$Register);
  6986   %}
  6987   ins_pipe(ialu_reg);
  6988 %}
  6990 instruct loadConN(rRegN dst, immN src) %{
  6991   match(Set dst src);
  6993   ins_cost(125);
  6994   format %{ "movl    $dst, $src\t# compressed ptr" %}
  6995   ins_encode %{
  6996     address con = (address)$src$$constant;
  6997     if (con == NULL) {
  6998       ShouldNotReachHere();
  6999     } else {
  7000       __ set_narrow_oop($dst$$Register, (jobject)$src$$constant);
  7002   %}
  7003   ins_pipe(ialu_reg_fat); // XXX
  7004 %}
  7006 instruct loadConF0(regF dst, immF0 src)
  7007 %{
  7008   match(Set dst src);
  7009   ins_cost(100);
  7011   format %{ "xorps   $dst, $dst\t# float 0.0" %}
  7012   opcode(0x0F, 0x57);
  7013   ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
  7014   ins_pipe(pipe_slow);
  7015 %}
  7017 // Use the same format since predicate() can not be used here.
  7018 instruct loadConD(regD dst, immD src)
  7019 %{
  7020   match(Set dst src);
  7021   ins_cost(125);
  7023   format %{ "movsd   $dst, [$src]" %}
  7024   ins_encode(load_conD(dst, src));
  7025   ins_pipe(pipe_slow);
  7026 %}
  7028 instruct loadConD0(regD dst, immD0 src)
  7029 %{
  7030   match(Set dst src);
  7031   ins_cost(100);
  7033   format %{ "xorpd   $dst, $dst\t# double 0.0" %}
  7034   opcode(0x66, 0x0F, 0x57);
  7035   ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
  7036   ins_pipe(pipe_slow);
  7037 %}
  7039 instruct loadSSI(rRegI dst, stackSlotI src)
  7040 %{
  7041   match(Set dst src);
  7043   ins_cost(125);
  7044   format %{ "movl    $dst, $src\t# int stk" %}
  7045   opcode(0x8B);
  7046   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
  7047   ins_pipe(ialu_reg_mem);
  7048 %}
  7050 instruct loadSSL(rRegL dst, stackSlotL src)
  7051 %{
  7052   match(Set dst src);
  7054   ins_cost(125);
  7055   format %{ "movq    $dst, $src\t# long stk" %}
  7056   opcode(0x8B);
  7057   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
  7058   ins_pipe(ialu_reg_mem);
  7059 %}
  7061 instruct loadSSP(rRegP dst, stackSlotP src)
  7062 %{
  7063   match(Set dst src);
  7065   ins_cost(125);
  7066   format %{ "movq    $dst, $src\t# ptr stk" %}
  7067   opcode(0x8B);
  7068   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
  7069   ins_pipe(ialu_reg_mem);
  7070 %}
  7072 instruct loadSSF(regF dst, stackSlotF src)
  7073 %{
  7074   match(Set dst src);
  7076   ins_cost(125);
  7077   format %{ "movss   $dst, $src\t# float stk" %}
  7078   opcode(0xF3, 0x0F, 0x10);
  7079   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
  7080   ins_pipe(pipe_slow); // XXX
  7081 %}
  7083 // Use the same format since predicate() can not be used here.
  7084 instruct loadSSD(regD dst, stackSlotD src)
  7085 %{
  7086   match(Set dst src);
  7088   ins_cost(125);
  7089   format %{ "movsd   $dst, $src\t# double stk" %}
  7090   ins_encode  %{
  7091     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
  7092   %}
  7093   ins_pipe(pipe_slow); // XXX
  7094 %}
  7096 // Prefetch instructions.
  7097 // Must be safe to execute with invalid address (cannot fault).
  7099 instruct prefetchr( memory mem ) %{
  7100   predicate(ReadPrefetchInstr==3);
  7101   match(PrefetchRead mem);
  7102   ins_cost(125);
  7104   format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
  7105   opcode(0x0F, 0x0D);     /* Opcode 0F 0D /0 */
  7106   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
  7107   ins_pipe(ialu_mem);
  7108 %}
  7110 instruct prefetchrNTA( memory mem ) %{
  7111   predicate(ReadPrefetchInstr==0);
  7112   match(PrefetchRead mem);
  7113   ins_cost(125);
  7115   format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
  7116   opcode(0x0F, 0x18);     /* Opcode 0F 18 /0 */
  7117   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
  7118   ins_pipe(ialu_mem);
  7119 %}
  7121 instruct prefetchrT0( memory mem ) %{
  7122   predicate(ReadPrefetchInstr==1);
  7123   match(PrefetchRead mem);
  7124   ins_cost(125);
  7126   format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
  7127   opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
  7128   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
  7129   ins_pipe(ialu_mem);
  7130 %}
  7132 instruct prefetchrT2( memory mem ) %{
  7133   predicate(ReadPrefetchInstr==2);
  7134   match(PrefetchRead mem);
  7135   ins_cost(125);
  7137   format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
  7138   opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
  7139   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
  7140   ins_pipe(ialu_mem);
  7141 %}
  7143 instruct prefetchw( memory mem ) %{
  7144   predicate(AllocatePrefetchInstr==3);
  7145   match(PrefetchWrite mem);
  7146   ins_cost(125);
  7148   format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
  7149   opcode(0x0F, 0x0D);     /* Opcode 0F 0D /1 */
  7150   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
  7151   ins_pipe(ialu_mem);
  7152 %}
  7154 instruct prefetchwNTA( memory mem ) %{
  7155   predicate(AllocatePrefetchInstr==0);
  7156   match(PrefetchWrite mem);
  7157   ins_cost(125);
  7159   format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
  7160   opcode(0x0F, 0x18);     /* Opcode 0F 18 /0 */
  7161   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
  7162   ins_pipe(ialu_mem);
  7163 %}
  7165 instruct prefetchwT0( memory mem ) %{
  7166   predicate(AllocatePrefetchInstr==1);
  7167   match(PrefetchWrite mem);
  7168   ins_cost(125);
  7170   format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
  7171   opcode(0x0F, 0x18);     /* Opcode 0F 18 /1 */
  7172   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
  7173   ins_pipe(ialu_mem);
  7174 %}
  7176 instruct prefetchwT2( memory mem ) %{
  7177   predicate(AllocatePrefetchInstr==2);
  7178   match(PrefetchWrite mem);
  7179   ins_cost(125);
  7181   format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
  7182   opcode(0x0F, 0x18);     /* Opcode 0F 18 /3 */
  7183   ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
  7184   ins_pipe(ialu_mem);
  7185 %}
  7187 //----------Store Instructions-------------------------------------------------
  7189 // Store Byte
  7190 instruct storeB(memory mem, rRegI src)
  7191 %{
  7192   match(Set mem (StoreB mem src));
  7194   ins_cost(125); // XXX
  7195   format %{ "movb    $mem, $src\t# byte" %}
  7196   opcode(0x88);
  7197   ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
  7198   ins_pipe(ialu_mem_reg);
  7199 %}
  7201 // Store Char/Short
  7202 instruct storeC(memory mem, rRegI src)
  7203 %{
  7204   match(Set mem (StoreC mem src));
  7206   ins_cost(125); // XXX
  7207   format %{ "movw    $mem, $src\t# char/short" %}
  7208   opcode(0x89);
  7209   ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
  7210   ins_pipe(ialu_mem_reg);
  7211 %}
  7213 // Store Integer
  7214 instruct storeI(memory mem, rRegI src)
  7215 %{
  7216   match(Set mem (StoreI mem src));
  7218   ins_cost(125); // XXX
  7219   format %{ "movl    $mem, $src\t# int" %}
  7220   opcode(0x89);
  7221   ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
  7222   ins_pipe(ialu_mem_reg);
  7223 %}
  7225 // Store Long
  7226 instruct storeL(memory mem, rRegL src)
  7227 %{
  7228   match(Set mem (StoreL mem src));
  7230   ins_cost(125); // XXX
  7231   format %{ "movq    $mem, $src\t# long" %}
  7232   opcode(0x89);
  7233   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
  7234   ins_pipe(ialu_mem_reg); // XXX
  7235 %}
  7237 // Store Pointer
  7238 instruct storeP(memory mem, any_RegP src)
  7239 %{
  7240   match(Set mem (StoreP mem src));
  7242   ins_cost(125); // XXX
  7243   format %{ "movq    $mem, $src\t# ptr" %}
  7244   opcode(0x89);
  7245   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
  7246   ins_pipe(ialu_mem_reg);
  7247 %}
  7249 instruct storeImmP0(memory mem, immP0 zero)
  7250 %{
  7251   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7252   match(Set mem (StoreP mem zero));
  7254   ins_cost(125); // XXX
  7255   format %{ "movq    $mem, R12\t# ptr (R12_heapbase==0)" %}
  7256   ins_encode %{
  7257     __ movq($mem$$Address, r12);
  7258   %}
  7259   ins_pipe(ialu_mem_reg);
  7260 %}
  7262 // Store NULL Pointer, mark word, or other simple pointer constant.
  7263 instruct storeImmP(memory mem, immP31 src)
  7264 %{
  7265   match(Set mem (StoreP mem src));
  7267   ins_cost(150); // XXX
  7268   format %{ "movq    $mem, $src\t# ptr" %}
  7269   opcode(0xC7); /* C7 /0 */
  7270   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
  7271   ins_pipe(ialu_mem_imm);
  7272 %}
  7274 // Store Compressed Pointer
  7275 instruct storeN(memory mem, rRegN src)
  7276 %{
  7277   match(Set mem (StoreN mem src));
  7279   ins_cost(125); // XXX
  7280   format %{ "movl    $mem, $src\t# compressed ptr" %}
  7281   ins_encode %{
  7282     __ movl($mem$$Address, $src$$Register);
  7283   %}
  7284   ins_pipe(ialu_mem_reg);
  7285 %}
  7287 instruct storeImmN0(memory mem, immN0 zero)
  7288 %{
  7289   predicate(Universe::narrow_oop_base() == NULL);
  7290   match(Set mem (StoreN mem zero));
  7292   ins_cost(125); // XXX
  7293   format %{ "movl    $mem, R12\t# compressed ptr (R12_heapbase==0)" %}
  7294   ins_encode %{
  7295     __ movl($mem$$Address, r12);
  7296   %}
  7297   ins_pipe(ialu_mem_reg);
  7298 %}
  7300 instruct storeImmN(memory mem, immN src)
  7301 %{
  7302   match(Set mem (StoreN mem src));
  7304   ins_cost(150); // XXX
  7305   format %{ "movl    $mem, $src\t# compressed ptr" %}
  7306   ins_encode %{
  7307     address con = (address)$src$$constant;
  7308     if (con == NULL) {
  7309       __ movl($mem$$Address, (int32_t)0);
  7310     } else {
  7311       __ set_narrow_oop($mem$$Address, (jobject)$src$$constant);
  7313   %}
  7314   ins_pipe(ialu_mem_imm);
  7315 %}
  7317 // Store Integer Immediate
  7318 instruct storeImmI0(memory mem, immI0 zero)
  7319 %{
  7320   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7321   match(Set mem (StoreI mem zero));
  7323   ins_cost(125); // XXX
  7324   format %{ "movl    $mem, R12\t# int (R12_heapbase==0)" %}
  7325   ins_encode %{
  7326     __ movl($mem$$Address, r12);
  7327   %}
  7328   ins_pipe(ialu_mem_reg);
  7329 %}
  7331 instruct storeImmI(memory mem, immI src)
  7332 %{
  7333   match(Set mem (StoreI mem src));
  7335   ins_cost(150);
  7336   format %{ "movl    $mem, $src\t# int" %}
  7337   opcode(0xC7); /* C7 /0 */
  7338   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
  7339   ins_pipe(ialu_mem_imm);
  7340 %}
  7342 // Store Long Immediate
  7343 instruct storeImmL0(memory mem, immL0 zero)
  7344 %{
  7345   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7346   match(Set mem (StoreL mem zero));
  7348   ins_cost(125); // XXX
  7349   format %{ "movq    $mem, R12\t# long (R12_heapbase==0)" %}
  7350   ins_encode %{
  7351     __ movq($mem$$Address, r12);
  7352   %}
  7353   ins_pipe(ialu_mem_reg);
  7354 %}
  7356 instruct storeImmL(memory mem, immL32 src)
  7357 %{
  7358   match(Set mem (StoreL mem src));
  7360   ins_cost(150);
  7361   format %{ "movq    $mem, $src\t# long" %}
  7362   opcode(0xC7); /* C7 /0 */
  7363   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
  7364   ins_pipe(ialu_mem_imm);
  7365 %}
  7367 // Store Short/Char Immediate
  7368 instruct storeImmC0(memory mem, immI0 zero)
  7369 %{
  7370   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7371   match(Set mem (StoreC mem zero));
  7373   ins_cost(125); // XXX
  7374   format %{ "movw    $mem, R12\t# short/char (R12_heapbase==0)" %}
  7375   ins_encode %{
  7376     __ movw($mem$$Address, r12);
  7377   %}
  7378   ins_pipe(ialu_mem_reg);
  7379 %}
  7381 instruct storeImmI16(memory mem, immI16 src)
  7382 %{
  7383   predicate(UseStoreImmI16);
  7384   match(Set mem (StoreC mem src));
  7386   ins_cost(150);
  7387   format %{ "movw    $mem, $src\t# short/char" %}
  7388   opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
  7389   ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
  7390   ins_pipe(ialu_mem_imm);
  7391 %}
  7393 // Store Byte Immediate
  7394 instruct storeImmB0(memory mem, immI0 zero)
  7395 %{
  7396   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7397   match(Set mem (StoreB mem zero));
  7399   ins_cost(125); // XXX
  7400   format %{ "movb    $mem, R12\t# short/char (R12_heapbase==0)" %}
  7401   ins_encode %{
  7402     __ movb($mem$$Address, r12);
  7403   %}
  7404   ins_pipe(ialu_mem_reg);
  7405 %}
  7407 instruct storeImmB(memory mem, immI8 src)
  7408 %{
  7409   match(Set mem (StoreB mem src));
  7411   ins_cost(150); // XXX
  7412   format %{ "movb    $mem, $src\t# byte" %}
  7413   opcode(0xC6); /* C6 /0 */
  7414   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
  7415   ins_pipe(ialu_mem_imm);
  7416 %}
  7418 // Store Aligned Packed Byte XMM register to memory
  7419 instruct storeA8B(memory mem, regD src) %{
  7420   match(Set mem (Store8B mem src));
  7421   ins_cost(145);
  7422   format %{ "MOVQ  $mem,$src\t! packed8B" %}
  7423   ins_encode( movq_st(mem, src));
  7424   ins_pipe( pipe_slow );
  7425 %}
  7427 // Store Aligned Packed Char/Short XMM register to memory
  7428 instruct storeA4C(memory mem, regD src) %{
  7429   match(Set mem (Store4C mem src));
  7430   ins_cost(145);
  7431   format %{ "MOVQ  $mem,$src\t! packed4C" %}
  7432   ins_encode( movq_st(mem, src));
  7433   ins_pipe( pipe_slow );
  7434 %}
  7436 // Store Aligned Packed Integer XMM register to memory
  7437 instruct storeA2I(memory mem, regD src) %{
  7438   match(Set mem (Store2I mem src));
  7439   ins_cost(145);
  7440   format %{ "MOVQ  $mem,$src\t! packed2I" %}
  7441   ins_encode( movq_st(mem, src));
  7442   ins_pipe( pipe_slow );
  7443 %}
  7445 // Store CMS card-mark Immediate
  7446 instruct storeImmCM0_reg(memory mem, immI0 zero)
  7447 %{
  7448   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7449   match(Set mem (StoreCM mem zero));
  7451   ins_cost(125); // XXX
  7452   format %{ "movb    $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %}
  7453   ins_encode %{
  7454     __ movb($mem$$Address, r12);
  7455   %}
  7456   ins_pipe(ialu_mem_reg);
  7457 %}
  7459 instruct storeImmCM0(memory mem, immI0 src)
  7460 %{
  7461   match(Set mem (StoreCM mem src));
  7463   ins_cost(150); // XXX
  7464   format %{ "movb    $mem, $src\t# CMS card-mark byte 0" %}
  7465   opcode(0xC6); /* C6 /0 */
  7466   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
  7467   ins_pipe(ialu_mem_imm);
  7468 %}
  7470 // Store Aligned Packed Single Float XMM register to memory
  7471 instruct storeA2F(memory mem, regD src) %{
  7472   match(Set mem (Store2F mem src));
  7473   ins_cost(145);
  7474   format %{ "MOVQ  $mem,$src\t! packed2F" %}
  7475   ins_encode( movq_st(mem, src));
  7476   ins_pipe( pipe_slow );
  7477 %}
  7479 // Store Float
  7480 instruct storeF(memory mem, regF src)
  7481 %{
  7482   match(Set mem (StoreF mem src));
  7484   ins_cost(95); // XXX
  7485   format %{ "movss   $mem, $src\t# float" %}
  7486   opcode(0xF3, 0x0F, 0x11);
  7487   ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
  7488   ins_pipe(pipe_slow); // XXX
  7489 %}
  7491 // Store immediate Float value (it is faster than store from XMM register)
  7492 instruct storeF0(memory mem, immF0 zero)
  7493 %{
  7494   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7495   match(Set mem (StoreF mem zero));
  7497   ins_cost(25); // XXX
  7498   format %{ "movl    $mem, R12\t# float 0. (R12_heapbase==0)" %}
  7499   ins_encode %{
  7500     __ movl($mem$$Address, r12);
  7501   %}
  7502   ins_pipe(ialu_mem_reg);
  7503 %}
  7505 instruct storeF_imm(memory mem, immF src)
  7506 %{
  7507   match(Set mem (StoreF mem src));
  7509   ins_cost(50);
  7510   format %{ "movl    $mem, $src\t# float" %}
  7511   opcode(0xC7); /* C7 /0 */
  7512   ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
  7513   ins_pipe(ialu_mem_imm);
  7514 %}
  7516 // Store Double
  7517 instruct storeD(memory mem, regD src)
  7518 %{
  7519   match(Set mem (StoreD mem src));
  7521   ins_cost(95); // XXX
  7522   format %{ "movsd   $mem, $src\t# double" %}
  7523   opcode(0xF2, 0x0F, 0x11);
  7524   ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
  7525   ins_pipe(pipe_slow); // XXX
  7526 %}
  7528 // Store immediate double 0.0 (it is faster than store from XMM register)
  7529 instruct storeD0_imm(memory mem, immD0 src)
  7530 %{
  7531   predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
  7532   match(Set mem (StoreD mem src));
  7534   ins_cost(50);
  7535   format %{ "movq    $mem, $src\t# double 0." %}
  7536   opcode(0xC7); /* C7 /0 */
  7537   ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
  7538   ins_pipe(ialu_mem_imm);
  7539 %}
  7541 instruct storeD0(memory mem, immD0 zero)
  7542 %{
  7543   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
  7544   match(Set mem (StoreD mem zero));
  7546   ins_cost(25); // XXX
  7547   format %{ "movq    $mem, R12\t# double 0. (R12_heapbase==0)" %}
  7548   ins_encode %{
  7549     __ movq($mem$$Address, r12);
  7550   %}
  7551   ins_pipe(ialu_mem_reg);
  7552 %}
  7554 instruct storeSSI(stackSlotI dst, rRegI src)
  7555 %{
  7556   match(Set dst src);
  7558   ins_cost(100);
  7559   format %{ "movl    $dst, $src\t# int stk" %}
  7560   opcode(0x89);
  7561   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
  7562   ins_pipe( ialu_mem_reg );
  7563 %}
  7565 instruct storeSSL(stackSlotL dst, rRegL src)
  7566 %{
  7567   match(Set dst src);
  7569   ins_cost(100);
  7570   format %{ "movq    $dst, $src\t# long stk" %}
  7571   opcode(0x89);
  7572   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
  7573   ins_pipe(ialu_mem_reg);
  7574 %}
  7576 instruct storeSSP(stackSlotP dst, rRegP src)
  7577 %{
  7578   match(Set dst src);
  7580   ins_cost(100);
  7581   format %{ "movq    $dst, $src\t# ptr stk" %}
  7582   opcode(0x89);
  7583   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
  7584   ins_pipe(ialu_mem_reg);
  7585 %}
  7587 instruct storeSSF(stackSlotF dst, regF src)
  7588 %{
  7589   match(Set dst src);
  7591   ins_cost(95); // XXX
  7592   format %{ "movss   $dst, $src\t# float stk" %}
  7593   opcode(0xF3, 0x0F, 0x11);
  7594   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
  7595   ins_pipe(pipe_slow); // XXX
  7596 %}
  7598 instruct storeSSD(stackSlotD dst, regD src)
  7599 %{
  7600   match(Set dst src);
  7602   ins_cost(95); // XXX
  7603   format %{ "movsd   $dst, $src\t# double stk" %}
  7604   opcode(0xF2, 0x0F, 0x11);
  7605   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
  7606   ins_pipe(pipe_slow); // XXX
  7607 %}
  7609 //----------BSWAP Instructions-------------------------------------------------
  7610 instruct bytes_reverse_int(rRegI dst) %{
  7611   match(Set dst (ReverseBytesI dst));
  7613   format %{ "bswapl  $dst" %}
  7614   opcode(0x0F, 0xC8);  /*Opcode 0F /C8 */
  7615   ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
  7616   ins_pipe( ialu_reg );
  7617 %}
  7619 instruct bytes_reverse_long(rRegL dst) %{
  7620   match(Set dst (ReverseBytesL dst));
  7622   format %{ "bswapq  $dst" %}
  7624   opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
  7625   ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
  7626   ins_pipe( ialu_reg);
  7627 %}
  7629 instruct loadI_reversed(rRegI dst, memory src) %{
  7630   match(Set dst (ReverseBytesI (LoadI src)));
  7632   format %{ "bswap_movl $dst, $src" %}
  7633   opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
  7634   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
  7635   ins_pipe( ialu_reg_mem );
  7636 %}
  7638 instruct loadL_reversed(rRegL dst, memory src) %{
  7639   match(Set dst (ReverseBytesL (LoadL src)));
  7641   format %{ "bswap_movq $dst, $src" %}
  7642   opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
  7643   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
  7644   ins_pipe( ialu_reg_mem );
  7645 %}
  7647 instruct storeI_reversed(memory dst, rRegI src) %{
  7648   match(Set dst (StoreI dst (ReverseBytesI  src)));
  7650   format %{ "movl_bswap $dst, $src" %}
  7651   opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
  7652   ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
  7653   ins_pipe( ialu_mem_reg );
  7654 %}
  7656 instruct storeL_reversed(memory dst, rRegL src) %{
  7657   match(Set dst (StoreL dst (ReverseBytesL  src)));
  7659   format %{ "movq_bswap $dst, $src" %}
  7660   opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
  7661   ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
  7662   ins_pipe( ialu_mem_reg );
  7663 %}
  7666 //---------- Zeros Count Instructions ------------------------------------------
  7668 instruct countLeadingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
  7669   predicate(UseCountLeadingZerosInstruction);
  7670   match(Set dst (CountLeadingZerosI src));
  7671   effect(KILL cr);
  7673   format %{ "lzcntl  $dst, $src\t# count leading zeros (int)" %}
  7674   ins_encode %{
  7675     __ lzcntl($dst$$Register, $src$$Register);
  7676   %}
  7677   ins_pipe(ialu_reg);
  7678 %}
  7680 instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, rFlagsReg cr) %{
  7681   predicate(!UseCountLeadingZerosInstruction);
  7682   match(Set dst (CountLeadingZerosI src));
  7683   effect(KILL cr);
  7685   format %{ "bsrl    $dst, $src\t# count leading zeros (int)\n\t"
  7686             "jnz     skip\n\t"
  7687             "movl    $dst, -1\n"
  7688       "skip:\n\t"
  7689             "negl    $dst\n\t"
  7690             "addl    $dst, 31" %}
  7691   ins_encode %{
  7692     Register Rdst = $dst$$Register;
  7693     Register Rsrc = $src$$Register;
  7694     Label skip;
  7695     __ bsrl(Rdst, Rsrc);
  7696     __ jccb(Assembler::notZero, skip);
  7697     __ movl(Rdst, -1);
  7698     __ bind(skip);
  7699     __ negl(Rdst);
  7700     __ addl(Rdst, BitsPerInt - 1);
  7701   %}
  7702   ins_pipe(ialu_reg);
  7703 %}
  7705 instruct countLeadingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
  7706   predicate(UseCountLeadingZerosInstruction);
  7707   match(Set dst (CountLeadingZerosL src));
  7708   effect(KILL cr);
  7710   format %{ "lzcntq  $dst, $src\t# count leading zeros (long)" %}
  7711   ins_encode %{
  7712     __ lzcntq($dst$$Register, $src$$Register);
  7713   %}
  7714   ins_pipe(ialu_reg);
  7715 %}
  7717 instruct countLeadingZerosL_bsr(rRegI dst, rRegL src, rFlagsReg cr) %{
  7718   predicate(!UseCountLeadingZerosInstruction);
  7719   match(Set dst (CountLeadingZerosL src));
  7720   effect(KILL cr);
  7722   format %{ "bsrq    $dst, $src\t# count leading zeros (long)\n\t"
  7723             "jnz     skip\n\t"
  7724             "movl    $dst, -1\n"
  7725       "skip:\n\t"
  7726             "negl    $dst\n\t"
  7727             "addl    $dst, 63" %}
  7728   ins_encode %{
  7729     Register Rdst = $dst$$Register;
  7730     Register Rsrc = $src$$Register;
  7731     Label skip;
  7732     __ bsrq(Rdst, Rsrc);
  7733     __ jccb(Assembler::notZero, skip);
  7734     __ movl(Rdst, -1);
  7735     __ bind(skip);
  7736     __ negl(Rdst);
  7737     __ addl(Rdst, BitsPerLong - 1);
  7738   %}
  7739   ins_pipe(ialu_reg);
  7740 %}
  7742 instruct countTrailingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
  7743   match(Set dst (CountTrailingZerosI src));
  7744   effect(KILL cr);
  7746   format %{ "bsfl    $dst, $src\t# count trailing zeros (int)\n\t"
  7747             "jnz     done\n\t"
  7748             "movl    $dst, 32\n"
  7749       "done:" %}
  7750   ins_encode %{
  7751     Register Rdst = $dst$$Register;
  7752     Label done;
  7753     __ bsfl(Rdst, $src$$Register);
  7754     __ jccb(Assembler::notZero, done);
  7755     __ movl(Rdst, BitsPerInt);
  7756     __ bind(done);
  7757   %}
  7758   ins_pipe(ialu_reg);
  7759 %}
  7761 instruct countTrailingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
  7762   match(Set dst (CountTrailingZerosL src));
  7763   effect(KILL cr);
  7765   format %{ "bsfq    $dst, $src\t# count trailing zeros (long)\n\t"
  7766             "jnz     done\n\t"
  7767             "movl    $dst, 64\n"
  7768       "done:" %}
  7769   ins_encode %{
  7770     Register Rdst = $dst$$Register;
  7771     Label done;
  7772     __ bsfq(Rdst, $src$$Register);
  7773     __ jccb(Assembler::notZero, done);
  7774     __ movl(Rdst, BitsPerLong);
  7775     __ bind(done);
  7776   %}
  7777   ins_pipe(ialu_reg);
  7778 %}
  7781 //---------- Population Count Instructions -------------------------------------
  7783 instruct popCountI(rRegI dst, rRegI src) %{
  7784   predicate(UsePopCountInstruction);
  7785   match(Set dst (PopCountI src));
  7787   format %{ "popcnt  $dst, $src" %}
  7788   ins_encode %{
  7789     __ popcntl($dst$$Register, $src$$Register);
  7790   %}
  7791   ins_pipe(ialu_reg);
  7792 %}
  7794 instruct popCountI_mem(rRegI dst, memory mem) %{
  7795   predicate(UsePopCountInstruction);
  7796   match(Set dst (PopCountI (LoadI mem)));
  7798   format %{ "popcnt  $dst, $mem" %}
  7799   ins_encode %{
  7800     __ popcntl($dst$$Register, $mem$$Address);
  7801   %}
  7802   ins_pipe(ialu_reg);
  7803 %}
  7805 // Note: Long.bitCount(long) returns an int.
  7806 instruct popCountL(rRegI dst, rRegL src) %{
  7807   predicate(UsePopCountInstruction);
  7808   match(Set dst (PopCountL src));
  7810   format %{ "popcnt  $dst, $src" %}
  7811   ins_encode %{
  7812     __ popcntq($dst$$Register, $src$$Register);
  7813   %}
  7814   ins_pipe(ialu_reg);
  7815 %}
  7817 // Note: Long.bitCount(long) returns an int.
  7818 instruct popCountL_mem(rRegI dst, memory mem) %{
  7819   predicate(UsePopCountInstruction);
  7820   match(Set dst (PopCountL (LoadL mem)));
  7822   format %{ "popcnt  $dst, $mem" %}
  7823   ins_encode %{
  7824     __ popcntq($dst$$Register, $mem$$Address);
  7825   %}
  7826   ins_pipe(ialu_reg);
  7827 %}
  7830 //----------MemBar Instructions-----------------------------------------------
  7831 // Memory barrier flavors
  7833 instruct membar_acquire()
  7834 %{
  7835   match(MemBarAcquire);
  7836   ins_cost(0);
  7838   size(0);
  7839   format %{ "MEMBAR-acquire ! (empty encoding)" %}
  7840   ins_encode();
  7841   ins_pipe(empty);
  7842 %}
  7844 instruct membar_acquire_lock()
  7845 %{
  7846   match(MemBarAcquire);
  7847   predicate(Matcher::prior_fast_lock(n));
  7848   ins_cost(0);
  7850   size(0);
  7851   format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
  7852   ins_encode();
  7853   ins_pipe(empty);
  7854 %}
  7856 instruct membar_release()
  7857 %{
  7858   match(MemBarRelease);
  7859   ins_cost(0);
  7861   size(0);
  7862   format %{ "MEMBAR-release ! (empty encoding)" %}
  7863   ins_encode();
  7864   ins_pipe(empty);
  7865 %}
  7867 instruct membar_release_lock()
  7868 %{
  7869   match(MemBarRelease);
  7870   predicate(Matcher::post_fast_unlock(n));
  7871   ins_cost(0);
  7873   size(0);
  7874   format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
  7875   ins_encode();
  7876   ins_pipe(empty);
  7877 %}
  7879 instruct membar_volatile(rFlagsReg cr) %{
  7880   match(MemBarVolatile);
  7881   effect(KILL cr);
  7882   ins_cost(400);
  7884   format %{ 
  7885     $$template
  7886     if (os::is_MP()) {
  7887       $$emit$$"lock addl [rsp + #0], 0\t! membar_volatile"
  7888     } else {
  7889       $$emit$$"MEMBAR-volatile ! (empty encoding)"
  7891   %}
  7892   ins_encode %{
  7893     __ membar(Assembler::StoreLoad);
  7894   %}
  7895   ins_pipe(pipe_slow);
  7896 %}
  7898 instruct unnecessary_membar_volatile()
  7899 %{
  7900   match(MemBarVolatile);
  7901   predicate(Matcher::post_store_load_barrier(n));
  7902   ins_cost(0);
  7904   size(0);
  7905   format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
  7906   ins_encode();
  7907   ins_pipe(empty);
  7908 %}
  7910 //----------Move Instructions--------------------------------------------------
  7912 instruct castX2P(rRegP dst, rRegL src)
  7913 %{
  7914   match(Set dst (CastX2P src));
  7916   format %{ "movq    $dst, $src\t# long->ptr" %}
  7917   ins_encode(enc_copy_wide(dst, src));
  7918   ins_pipe(ialu_reg_reg); // XXX
  7919 %}
  7921 instruct castP2X(rRegL dst, rRegP src)
  7922 %{
  7923   match(Set dst (CastP2X src));
  7925   format %{ "movq    $dst, $src\t# ptr -> long" %}
  7926   ins_encode(enc_copy_wide(dst, src));
  7927   ins_pipe(ialu_reg_reg); // XXX
  7928 %}
  7931 // Convert oop pointer into compressed form
  7932 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
  7933   predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
  7934   match(Set dst (EncodeP src));
  7935   effect(KILL cr);
  7936   format %{ "encode_heap_oop $dst,$src" %}
  7937   ins_encode %{
  7938     Register s = $src$$Register;
  7939     Register d = $dst$$Register;
  7940     if (s != d) {
  7941       __ movq(d, s);
  7943     __ encode_heap_oop(d);
  7944   %}
  7945   ins_pipe(ialu_reg_long);
  7946 %}
  7948 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
  7949   predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
  7950   match(Set dst (EncodeP src));
  7951   effect(KILL cr);
  7952   format %{ "encode_heap_oop_not_null $dst,$src" %}
  7953   ins_encode %{
  7954     __ encode_heap_oop_not_null($dst$$Register, $src$$Register);
  7955   %}
  7956   ins_pipe(ialu_reg_long);
  7957 %}
  7959 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
  7960   predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
  7961             n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
  7962   match(Set dst (DecodeN src));
  7963   effect(KILL cr);
  7964   format %{ "decode_heap_oop $dst,$src" %}
  7965   ins_encode %{
  7966     Register s = $src$$Register;
  7967     Register d = $dst$$Register;
  7968     if (s != d) {
  7969       __ movq(d, s);
  7971     __ decode_heap_oop(d);
  7972   %}
  7973   ins_pipe(ialu_reg_long);
  7974 %}
  7976 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{
  7977   predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
  7978             n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
  7979   match(Set dst (DecodeN src));
  7980   format %{ "decode_heap_oop_not_null $dst,$src" %}
  7981   ins_encode %{
  7982     Register s = $src$$Register;
  7983     Register d = $dst$$Register;
  7984     if (s != d) {
  7985       __ decode_heap_oop_not_null(d, s);
  7986     } else {
  7987       __ decode_heap_oop_not_null(d);
  7989   %}
  7990   ins_pipe(ialu_reg_long);
  7991 %}
  7994 //----------Conditional Move---------------------------------------------------
  7995 // Jump
  7996 // dummy instruction for generating temp registers
  7997 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
  7998   match(Jump (LShiftL switch_val shift));
  7999   ins_cost(350);
  8000   predicate(false);
  8001   effect(TEMP dest);
  8003   format %{ "leaq    $dest, table_base\n\t"
  8004             "jmp     [$dest + $switch_val << $shift]\n\t" %}
  8005   ins_encode(jump_enc_offset(switch_val, shift, dest));
  8006   ins_pipe(pipe_jmp);
  8007   ins_pc_relative(1);
  8008 %}
  8010 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
  8011   match(Jump (AddL (LShiftL switch_val shift) offset));
  8012   ins_cost(350);
  8013   effect(TEMP dest);
  8015   format %{ "leaq    $dest, table_base\n\t"
  8016             "jmp     [$dest + $switch_val << $shift + $offset]\n\t" %}
  8017   ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
  8018   ins_pipe(pipe_jmp);
  8019   ins_pc_relative(1);
  8020 %}
  8022 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
  8023   match(Jump switch_val);
  8024   ins_cost(350);
  8025   effect(TEMP dest);
  8027   format %{ "leaq    $dest, table_base\n\t"
  8028             "jmp     [$dest + $switch_val]\n\t" %}
  8029   ins_encode(jump_enc(switch_val, dest));
  8030   ins_pipe(pipe_jmp);
  8031   ins_pc_relative(1);
  8032 %}
  8034 // Conditional move
  8035 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
  8036 %{
  8037   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  8039   ins_cost(200); // XXX
  8040   format %{ "cmovl$cop $dst, $src\t# signed, int" %}
  8041   opcode(0x0F, 0x40);
  8042   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8043   ins_pipe(pipe_cmov_reg);
  8044 %}
  8046 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
  8047   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  8049   ins_cost(200); // XXX
  8050   format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
  8051   opcode(0x0F, 0x40);
  8052   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8053   ins_pipe(pipe_cmov_reg);
  8054 %}
  8056 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
  8057   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
  8058   ins_cost(200);
  8059   expand %{
  8060     cmovI_regU(cop, cr, dst, src);
  8061   %}
  8062 %}
  8064 // Conditional move
  8065 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
  8066   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  8068   ins_cost(250); // XXX
  8069   format %{ "cmovl$cop $dst, $src\t# signed, int" %}
  8070   opcode(0x0F, 0x40);
  8071   ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
  8072   ins_pipe(pipe_cmov_mem);
  8073 %}
  8075 // Conditional move
  8076 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
  8077 %{
  8078   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  8080   ins_cost(250); // XXX
  8081   format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
  8082   opcode(0x0F, 0x40);
  8083   ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
  8084   ins_pipe(pipe_cmov_mem);
  8085 %}
  8087 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
  8088   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
  8089   ins_cost(250);
  8090   expand %{
  8091     cmovI_memU(cop, cr, dst, src);
  8092   %}
  8093 %}
  8095 // Conditional move
  8096 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
  8097 %{
  8098   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
  8100   ins_cost(200); // XXX
  8101   format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
  8102   opcode(0x0F, 0x40);
  8103   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8104   ins_pipe(pipe_cmov_reg);
  8105 %}
  8107 // Conditional move
  8108 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
  8109 %{
  8110   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
  8112   ins_cost(200); // XXX
  8113   format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
  8114   opcode(0x0F, 0x40);
  8115   ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8116   ins_pipe(pipe_cmov_reg);
  8117 %}
  8119 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
  8120   match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
  8121   ins_cost(200);
  8122   expand %{
  8123     cmovN_regU(cop, cr, dst, src);
  8124   %}
  8125 %}
  8127 // Conditional move
  8128 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
  8129 %{
  8130   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  8132   ins_cost(200); // XXX
  8133   format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
  8134   opcode(0x0F, 0x40);
  8135   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8136   ins_pipe(pipe_cmov_reg);  // XXX
  8137 %}
  8139 // Conditional move
  8140 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
  8141 %{
  8142   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  8144   ins_cost(200); // XXX
  8145   format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
  8146   opcode(0x0F, 0x40);
  8147   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8148   ins_pipe(pipe_cmov_reg); // XXX
  8149 %}
  8151 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
  8152   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
  8153   ins_cost(200);
  8154   expand %{
  8155     cmovP_regU(cop, cr, dst, src);
  8156   %}
  8157 %}
  8159 // DISABLED: Requires the ADLC to emit a bottom_type call that
  8160 // correctly meets the two pointer arguments; one is an incoming
  8161 // register but the other is a memory operand.  ALSO appears to
  8162 // be buggy with implicit null checks.
  8163 //
  8164 //// Conditional move
  8165 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
  8166 //%{
  8167 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
  8168 //  ins_cost(250);
  8169 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
  8170 //  opcode(0x0F,0x40);
  8171 //  ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
  8172 //  ins_pipe( pipe_cmov_mem );
  8173 //%}
  8174 //
  8175 //// Conditional move
  8176 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
  8177 //%{
  8178 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
  8179 //  ins_cost(250);
  8180 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
  8181 //  opcode(0x0F,0x40);
  8182 //  ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
  8183 //  ins_pipe( pipe_cmov_mem );
  8184 //%}
  8186 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
  8187 %{
  8188   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  8190   ins_cost(200); // XXX
  8191   format %{ "cmovq$cop $dst, $src\t# signed, long" %}
  8192   opcode(0x0F, 0x40);
  8193   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8194   ins_pipe(pipe_cmov_reg);  // XXX
  8195 %}
  8197 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
  8198 %{
  8199   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
  8201   ins_cost(200); // XXX
  8202   format %{ "cmovq$cop $dst, $src\t# signed, long" %}
  8203   opcode(0x0F, 0x40);
  8204   ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
  8205   ins_pipe(pipe_cmov_mem);  // XXX
  8206 %}
  8208 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
  8209 %{
  8210   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  8212   ins_cost(200); // XXX
  8213   format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
  8214   opcode(0x0F, 0x40);
  8215   ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
  8216   ins_pipe(pipe_cmov_reg); // XXX
  8217 %}
  8219 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
  8220   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
  8221   ins_cost(200);
  8222   expand %{
  8223     cmovL_regU(cop, cr, dst, src);
  8224   %}
  8225 %}
  8227 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
  8228 %{
  8229   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
  8231   ins_cost(200); // XXX
  8232   format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
  8233   opcode(0x0F, 0x40);
  8234   ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
  8235   ins_pipe(pipe_cmov_mem); // XXX
  8236 %}
  8238 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
  8239   match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
  8240   ins_cost(200);
  8241   expand %{
  8242     cmovL_memU(cop, cr, dst, src);
  8243   %}
  8244 %}
  8246 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
  8247 %{
  8248   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  8250   ins_cost(200); // XXX
  8251   format %{ "jn$cop    skip\t# signed cmove float\n\t"
  8252             "movss     $dst, $src\n"
  8253     "skip:" %}
  8254   ins_encode(enc_cmovf_branch(cop, dst, src));
  8255   ins_pipe(pipe_slow);
  8256 %}
  8258 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
  8259 // %{
  8260 //   match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
  8262 //   ins_cost(200); // XXX
  8263 //   format %{ "jn$cop    skip\t# signed cmove float\n\t"
  8264 //             "movss     $dst, $src\n"
  8265 //     "skip:" %}
  8266 //   ins_encode(enc_cmovf_mem_branch(cop, dst, src));
  8267 //   ins_pipe(pipe_slow);
  8268 // %}
  8270 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
  8271 %{
  8272   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  8274   ins_cost(200); // XXX
  8275   format %{ "jn$cop    skip\t# unsigned cmove float\n\t"
  8276             "movss     $dst, $src\n"
  8277     "skip:" %}
  8278   ins_encode(enc_cmovf_branch(cop, dst, src));
  8279   ins_pipe(pipe_slow);
  8280 %}
  8282 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
  8283   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
  8284   ins_cost(200);
  8285   expand %{
  8286     cmovF_regU(cop, cr, dst, src);
  8287   %}
  8288 %}
  8290 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
  8291 %{
  8292   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  8294   ins_cost(200); // XXX
  8295   format %{ "jn$cop    skip\t# signed cmove double\n\t"
  8296             "movsd     $dst, $src\n"
  8297     "skip:" %}
  8298   ins_encode(enc_cmovd_branch(cop, dst, src));
  8299   ins_pipe(pipe_slow);
  8300 %}
  8302 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
  8303 %{
  8304   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  8306   ins_cost(200); // XXX
  8307   format %{ "jn$cop    skip\t# unsigned cmove double\n\t"
  8308             "movsd     $dst, $src\n"
  8309     "skip:" %}
  8310   ins_encode(enc_cmovd_branch(cop, dst, src));
  8311   ins_pipe(pipe_slow);
  8312 %}
  8314 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
  8315   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
  8316   ins_cost(200);
  8317   expand %{
  8318     cmovD_regU(cop, cr, dst, src);
  8319   %}
  8320 %}
  8322 //----------Arithmetic Instructions--------------------------------------------
  8323 //----------Addition Instructions----------------------------------------------
  8325 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
  8326 %{
  8327   match(Set dst (AddI dst src));
  8328   effect(KILL cr);
  8330   format %{ "addl    $dst, $src\t# int" %}
  8331   opcode(0x03);
  8332   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
  8333   ins_pipe(ialu_reg_reg);
  8334 %}
  8336 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
  8337 %{
  8338   match(Set dst (AddI dst src));
  8339   effect(KILL cr);
  8341   format %{ "addl    $dst, $src\t# int" %}
  8342   opcode(0x81, 0x00); /* /0 id */
  8343   ins_encode(OpcSErm(dst, src), Con8or32(src));
  8344   ins_pipe( ialu_reg );
  8345 %}
  8347 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
  8348 %{
  8349   match(Set dst (AddI dst (LoadI src)));
  8350   effect(KILL cr);
  8352   ins_cost(125); // XXX
  8353   format %{ "addl    $dst, $src\t# int" %}
  8354   opcode(0x03);
  8355   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
  8356   ins_pipe(ialu_reg_mem);
  8357 %}
  8359 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
  8360 %{
  8361   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  8362   effect(KILL cr);
  8364   ins_cost(150); // XXX
  8365   format %{ "addl    $dst, $src\t# int" %}
  8366   opcode(0x01); /* Opcode 01 /r */
  8367   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
  8368   ins_pipe(ialu_mem_reg);
  8369 %}
  8371 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
  8372 %{
  8373   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  8374   effect(KILL cr);
  8376   ins_cost(125); // XXX
  8377   format %{ "addl    $dst, $src\t# int" %}
  8378   opcode(0x81); /* Opcode 81 /0 id */
  8379   ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
  8380   ins_pipe(ialu_mem_imm);
  8381 %}
  8383 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
  8384 %{
  8385   predicate(UseIncDec);
  8386   match(Set dst (AddI dst src));
  8387   effect(KILL cr);
  8389   format %{ "incl    $dst\t# int" %}
  8390   opcode(0xFF, 0x00); // FF /0
  8391   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  8392   ins_pipe(ialu_reg);
  8393 %}
  8395 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
  8396 %{
  8397   predicate(UseIncDec);
  8398   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  8399   effect(KILL cr);
  8401   ins_cost(125); // XXX
  8402   format %{ "incl    $dst\t# int" %}
  8403   opcode(0xFF); /* Opcode FF /0 */
  8404   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
  8405   ins_pipe(ialu_mem_imm);
  8406 %}
  8408 // XXX why does that use AddI
  8409 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
  8410 %{
  8411   predicate(UseIncDec);
  8412   match(Set dst (AddI dst src));
  8413   effect(KILL cr);
  8415   format %{ "decl    $dst\t# int" %}
  8416   opcode(0xFF, 0x01); // FF /1
  8417   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  8418   ins_pipe(ialu_reg);
  8419 %}
  8421 // XXX why does that use AddI
  8422 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
  8423 %{
  8424   predicate(UseIncDec);
  8425   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
  8426   effect(KILL cr);
  8428   ins_cost(125); // XXX
  8429   format %{ "decl    $dst\t# int" %}
  8430   opcode(0xFF); /* Opcode FF /1 */
  8431   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
  8432   ins_pipe(ialu_mem_imm);
  8433 %}
  8435 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
  8436 %{
  8437   match(Set dst (AddI src0 src1));
  8439   ins_cost(110);
  8440   format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
  8441   opcode(0x8D); /* 0x8D /r */
  8442   ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
  8443   ins_pipe(ialu_reg_reg);
  8444 %}
  8446 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
  8447 %{
  8448   match(Set dst (AddL dst src));
  8449   effect(KILL cr);
  8451   format %{ "addq    $dst, $src\t# long" %}
  8452   opcode(0x03);
  8453   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
  8454   ins_pipe(ialu_reg_reg);
  8455 %}
  8457 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
  8458 %{
  8459   match(Set dst (AddL dst src));
  8460   effect(KILL cr);
  8462   format %{ "addq    $dst, $src\t# long" %}
  8463   opcode(0x81, 0x00); /* /0 id */
  8464   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
  8465   ins_pipe( ialu_reg );
  8466 %}
  8468 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
  8469 %{
  8470   match(Set dst (AddL dst (LoadL src)));
  8471   effect(KILL cr);
  8473   ins_cost(125); // XXX
  8474   format %{ "addq    $dst, $src\t# long" %}
  8475   opcode(0x03);
  8476   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
  8477   ins_pipe(ialu_reg_mem);
  8478 %}
  8480 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
  8481 %{
  8482   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
  8483   effect(KILL cr);
  8485   ins_cost(150); // XXX
  8486   format %{ "addq    $dst, $src\t# long" %}
  8487   opcode(0x01); /* Opcode 01 /r */
  8488   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
  8489   ins_pipe(ialu_mem_reg);
  8490 %}
  8492 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
  8493 %{
  8494   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
  8495   effect(KILL cr);
  8497   ins_cost(125); // XXX
  8498   format %{ "addq    $dst, $src\t# long" %}
  8499   opcode(0x81); /* Opcode 81 /0 id */
  8500   ins_encode(REX_mem_wide(dst),
  8501              OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
  8502   ins_pipe(ialu_mem_imm);
  8503 %}
  8505 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
  8506 %{
  8507   predicate(UseIncDec);
  8508   match(Set dst (AddL dst src));
  8509   effect(KILL cr);
  8511   format %{ "incq    $dst\t# long" %}
  8512   opcode(0xFF, 0x00); // FF /0
  8513   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  8514   ins_pipe(ialu_reg);
  8515 %}
  8517 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
  8518 %{
  8519   predicate(UseIncDec);
  8520   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
  8521   effect(KILL cr);
  8523   ins_cost(125); // XXX
  8524   format %{ "incq    $dst\t# long" %}
  8525   opcode(0xFF); /* Opcode FF /0 */
  8526   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
  8527   ins_pipe(ialu_mem_imm);
  8528 %}
  8530 // XXX why does that use AddL
  8531 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
  8532 %{
  8533   predicate(UseIncDec);
  8534   match(Set dst (AddL dst src));
  8535   effect(KILL cr);
  8537   format %{ "decq    $dst\t# long" %}
  8538   opcode(0xFF, 0x01); // FF /1
  8539   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  8540   ins_pipe(ialu_reg);
  8541 %}
  8543 // XXX why does that use AddL
  8544 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
  8545 %{
  8546   predicate(UseIncDec);
  8547   match(Set dst (StoreL dst (AddL (LoadL dst) src)));
  8548   effect(KILL cr);
  8550   ins_cost(125); // XXX
  8551   format %{ "decq    $dst\t# long" %}
  8552   opcode(0xFF); /* Opcode FF /1 */
  8553   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
  8554   ins_pipe(ialu_mem_imm);
  8555 %}
  8557 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
  8558 %{
  8559   match(Set dst (AddL src0 src1));
  8561   ins_cost(110);
  8562   format %{ "leaq    $dst, [$src0 + $src1]\t# long" %}
  8563   opcode(0x8D); /* 0x8D /r */
  8564   ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
  8565   ins_pipe(ialu_reg_reg);
  8566 %}
  8568 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
  8569 %{
  8570   match(Set dst (AddP dst src));
  8571   effect(KILL cr);
  8573   format %{ "addq    $dst, $src\t# ptr" %}
  8574   opcode(0x03);
  8575   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
  8576   ins_pipe(ialu_reg_reg);
  8577 %}
  8579 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
  8580 %{
  8581   match(Set dst (AddP dst src));
  8582   effect(KILL cr);
  8584   format %{ "addq    $dst, $src\t# ptr" %}
  8585   opcode(0x81, 0x00); /* /0 id */
  8586   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
  8587   ins_pipe( ialu_reg );
  8588 %}
  8590 // XXX addP mem ops ????
  8592 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
  8593 %{
  8594   match(Set dst (AddP src0 src1));
  8596   ins_cost(110);
  8597   format %{ "leaq    $dst, [$src0 + $src1]\t# ptr" %}
  8598   opcode(0x8D); /* 0x8D /r */
  8599   ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
  8600   ins_pipe(ialu_reg_reg);
  8601 %}
  8603 instruct checkCastPP(rRegP dst)
  8604 %{
  8605   match(Set dst (CheckCastPP dst));
  8607   size(0);
  8608   format %{ "# checkcastPP of $dst" %}
  8609   ins_encode(/* empty encoding */);
  8610   ins_pipe(empty);
  8611 %}
  8613 instruct castPP(rRegP dst)
  8614 %{
  8615   match(Set dst (CastPP dst));
  8617   size(0);
  8618   format %{ "# castPP of $dst" %}
  8619   ins_encode(/* empty encoding */);
  8620   ins_pipe(empty);
  8621 %}
  8623 instruct castII(rRegI dst)
  8624 %{
  8625   match(Set dst (CastII dst));
  8627   size(0);
  8628   format %{ "# castII of $dst" %}
  8629   ins_encode(/* empty encoding */);
  8630   ins_cost(0);
  8631   ins_pipe(empty);
  8632 %}
  8634 // LoadP-locked same as a regular LoadP when used with compare-swap
  8635 instruct loadPLocked(rRegP dst, memory mem)
  8636 %{
  8637   match(Set dst (LoadPLocked mem));
  8639   ins_cost(125); // XXX
  8640   format %{ "movq    $dst, $mem\t# ptr locked" %}
  8641   opcode(0x8B);
  8642   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  8643   ins_pipe(ialu_reg_mem); // XXX
  8644 %}
  8646 // LoadL-locked - same as a regular LoadL when used with compare-swap
  8647 instruct loadLLocked(rRegL dst, memory mem)
  8648 %{
  8649   match(Set dst (LoadLLocked mem));
  8651   ins_cost(125); // XXX
  8652   format %{ "movq    $dst, $mem\t# long locked" %}
  8653   opcode(0x8B);
  8654   ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
  8655   ins_pipe(ialu_reg_mem); // XXX
  8656 %}
  8658 // Conditional-store of the updated heap-top.
  8659 // Used during allocation of the shared heap.
  8660 // Sets flags (EQ) on success.  Implemented with a CMPXCHG on Intel.
  8662 instruct storePConditional(memory heap_top_ptr,
  8663                            rax_RegP oldval, rRegP newval,
  8664                            rFlagsReg cr)
  8665 %{
  8666   match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
  8668   format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
  8669             "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
  8670   opcode(0x0F, 0xB1);
  8671   ins_encode(lock_prefix,
  8672              REX_reg_mem_wide(newval, heap_top_ptr),
  8673              OpcP, OpcS,
  8674              reg_mem(newval, heap_top_ptr));
  8675   ins_pipe(pipe_cmpxchg);
  8676 %}
  8678 // Conditional-store of an int value.
  8679 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG.
  8680 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
  8681 %{
  8682   match(Set cr (StoreIConditional mem (Binary oldval newval)));
  8683   effect(KILL oldval);
  8685   format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
  8686   opcode(0x0F, 0xB1);
  8687   ins_encode(lock_prefix,
  8688              REX_reg_mem(newval, mem),
  8689              OpcP, OpcS,
  8690              reg_mem(newval, mem));
  8691   ins_pipe(pipe_cmpxchg);
  8692 %}
  8694 // Conditional-store of a long value.
  8695 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG.
  8696 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
  8697 %{
  8698   match(Set cr (StoreLConditional mem (Binary oldval newval)));
  8699   effect(KILL oldval);
  8701   format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
  8702   opcode(0x0F, 0xB1);
  8703   ins_encode(lock_prefix,
  8704              REX_reg_mem_wide(newval, mem),
  8705              OpcP, OpcS,
  8706              reg_mem(newval, mem));
  8707   ins_pipe(pipe_cmpxchg);
  8708 %}
  8711 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
  8712 instruct compareAndSwapP(rRegI res,
  8713                          memory mem_ptr,
  8714                          rax_RegP oldval, rRegP newval,
  8715                          rFlagsReg cr)
  8716 %{
  8717   match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
  8718   effect(KILL cr, KILL oldval);
  8720   format %{ "cmpxchgq $mem_ptr,$newval\t# "
  8721             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
  8722             "sete    $res\n\t"
  8723             "movzbl  $res, $res" %}
  8724   opcode(0x0F, 0xB1);
  8725   ins_encode(lock_prefix,
  8726              REX_reg_mem_wide(newval, mem_ptr),
  8727              OpcP, OpcS,
  8728              reg_mem(newval, mem_ptr),
  8729              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
  8730              REX_reg_breg(res, res), // movzbl
  8731              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
  8732   ins_pipe( pipe_cmpxchg );
  8733 %}
  8735 instruct compareAndSwapL(rRegI res,
  8736                          memory mem_ptr,
  8737                          rax_RegL oldval, rRegL newval,
  8738                          rFlagsReg cr)
  8739 %{
  8740   match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
  8741   effect(KILL cr, KILL oldval);
  8743   format %{ "cmpxchgq $mem_ptr,$newval\t# "
  8744             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
  8745             "sete    $res\n\t"
  8746             "movzbl  $res, $res" %}
  8747   opcode(0x0F, 0xB1);
  8748   ins_encode(lock_prefix,
  8749              REX_reg_mem_wide(newval, mem_ptr),
  8750              OpcP, OpcS,
  8751              reg_mem(newval, mem_ptr),
  8752              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
  8753              REX_reg_breg(res, res), // movzbl
  8754              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
  8755   ins_pipe( pipe_cmpxchg );
  8756 %}
  8758 instruct compareAndSwapI(rRegI res,
  8759                          memory mem_ptr,
  8760                          rax_RegI oldval, rRegI newval,
  8761                          rFlagsReg cr)
  8762 %{
  8763   match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
  8764   effect(KILL cr, KILL oldval);
  8766   format %{ "cmpxchgl $mem_ptr,$newval\t# "
  8767             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
  8768             "sete    $res\n\t"
  8769             "movzbl  $res, $res" %}
  8770   opcode(0x0F, 0xB1);
  8771   ins_encode(lock_prefix,
  8772              REX_reg_mem(newval, mem_ptr),
  8773              OpcP, OpcS,
  8774              reg_mem(newval, mem_ptr),
  8775              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
  8776              REX_reg_breg(res, res), // movzbl
  8777              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
  8778   ins_pipe( pipe_cmpxchg );
  8779 %}
  8782 instruct compareAndSwapN(rRegI res,
  8783                           memory mem_ptr,
  8784                           rax_RegN oldval, rRegN newval,
  8785                           rFlagsReg cr) %{
  8786   match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
  8787   effect(KILL cr, KILL oldval);
  8789   format %{ "cmpxchgl $mem_ptr,$newval\t# "
  8790             "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
  8791             "sete    $res\n\t"
  8792             "movzbl  $res, $res" %}
  8793   opcode(0x0F, 0xB1);
  8794   ins_encode(lock_prefix,
  8795              REX_reg_mem(newval, mem_ptr),
  8796              OpcP, OpcS,
  8797              reg_mem(newval, mem_ptr),
  8798              REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
  8799              REX_reg_breg(res, res), // movzbl
  8800              Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
  8801   ins_pipe( pipe_cmpxchg );
  8802 %}
  8804 //----------Subtraction Instructions-------------------------------------------
  8806 // Integer Subtraction Instructions
  8807 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
  8808 %{
  8809   match(Set dst (SubI dst src));
  8810   effect(KILL cr);
  8812   format %{ "subl    $dst, $src\t# int" %}
  8813   opcode(0x2B);
  8814   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
  8815   ins_pipe(ialu_reg_reg);
  8816 %}
  8818 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
  8819 %{
  8820   match(Set dst (SubI dst src));
  8821   effect(KILL cr);
  8823   format %{ "subl    $dst, $src\t# int" %}
  8824   opcode(0x81, 0x05);  /* Opcode 81 /5 */
  8825   ins_encode(OpcSErm(dst, src), Con8or32(src));
  8826   ins_pipe(ialu_reg);
  8827 %}
  8829 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
  8830 %{
  8831   match(Set dst (SubI dst (LoadI src)));
  8832   effect(KILL cr);
  8834   ins_cost(125);
  8835   format %{ "subl    $dst, $src\t# int" %}
  8836   opcode(0x2B);
  8837   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
  8838   ins_pipe(ialu_reg_mem);
  8839 %}
  8841 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
  8842 %{
  8843   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
  8844   effect(KILL cr);
  8846   ins_cost(150);
  8847   format %{ "subl    $dst, $src\t# int" %}
  8848   opcode(0x29); /* Opcode 29 /r */
  8849   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
  8850   ins_pipe(ialu_mem_reg);
  8851 %}
  8853 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
  8854 %{
  8855   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
  8856   effect(KILL cr);
  8858   ins_cost(125); // XXX
  8859   format %{ "subl    $dst, $src\t# int" %}
  8860   opcode(0x81); /* Opcode 81 /5 id */
  8861   ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
  8862   ins_pipe(ialu_mem_imm);
  8863 %}
  8865 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
  8866 %{
  8867   match(Set dst (SubL dst src));
  8868   effect(KILL cr);
  8870   format %{ "subq    $dst, $src\t# long" %}
  8871   opcode(0x2B);
  8872   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
  8873   ins_pipe(ialu_reg_reg);
  8874 %}
  8876 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
  8877 %{
  8878   match(Set dst (SubL dst src));
  8879   effect(KILL cr);
  8881   format %{ "subq    $dst, $src\t# long" %}
  8882   opcode(0x81, 0x05);  /* Opcode 81 /5 */
  8883   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
  8884   ins_pipe(ialu_reg);
  8885 %}
  8887 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
  8888 %{
  8889   match(Set dst (SubL dst (LoadL src)));
  8890   effect(KILL cr);
  8892   ins_cost(125);
  8893   format %{ "subq    $dst, $src\t# long" %}
  8894   opcode(0x2B);
  8895   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
  8896   ins_pipe(ialu_reg_mem);
  8897 %}
  8899 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
  8900 %{
  8901   match(Set dst (StoreL dst (SubL (LoadL dst) src)));
  8902   effect(KILL cr);
  8904   ins_cost(150);
  8905   format %{ "subq    $dst, $src\t# long" %}
  8906   opcode(0x29); /* Opcode 29 /r */
  8907   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
  8908   ins_pipe(ialu_mem_reg);
  8909 %}
  8911 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
  8912 %{
  8913   match(Set dst (StoreL dst (SubL (LoadL dst) src)));
  8914   effect(KILL cr);
  8916   ins_cost(125); // XXX
  8917   format %{ "subq    $dst, $src\t# long" %}
  8918   opcode(0x81); /* Opcode 81 /5 id */
  8919   ins_encode(REX_mem_wide(dst),
  8920              OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
  8921   ins_pipe(ialu_mem_imm);
  8922 %}
  8924 // Subtract from a pointer
  8925 // XXX hmpf???
  8926 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
  8927 %{
  8928   match(Set dst (AddP dst (SubI zero src)));
  8929   effect(KILL cr);
  8931   format %{ "subq    $dst, $src\t# ptr - int" %}
  8932   opcode(0x2B);
  8933   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
  8934   ins_pipe(ialu_reg_reg);
  8935 %}
  8937 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
  8938 %{
  8939   match(Set dst (SubI zero dst));
  8940   effect(KILL cr);
  8942   format %{ "negl    $dst\t# int" %}
  8943   opcode(0xF7, 0x03);  // Opcode F7 /3
  8944   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  8945   ins_pipe(ialu_reg);
  8946 %}
  8948 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
  8949 %{
  8950   match(Set dst (StoreI dst (SubI zero (LoadI dst))));
  8951   effect(KILL cr);
  8953   format %{ "negl    $dst\t# int" %}
  8954   opcode(0xF7, 0x03);  // Opcode F7 /3
  8955   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  8956   ins_pipe(ialu_reg);
  8957 %}
  8959 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
  8960 %{
  8961   match(Set dst (SubL zero dst));
  8962   effect(KILL cr);
  8964   format %{ "negq    $dst\t# long" %}
  8965   opcode(0xF7, 0x03);  // Opcode F7 /3
  8966   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  8967   ins_pipe(ialu_reg);
  8968 %}
  8970 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
  8971 %{
  8972   match(Set dst (StoreL dst (SubL zero (LoadL dst))));
  8973   effect(KILL cr);
  8975   format %{ "negq    $dst\t# long" %}
  8976   opcode(0xF7, 0x03);  // Opcode F7 /3
  8977   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  8978   ins_pipe(ialu_reg);
  8979 %}
  8982 //----------Multiplication/Division Instructions-------------------------------
  8983 // Integer Multiplication Instructions
  8984 // Multiply Register
  8986 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
  8987 %{
  8988   match(Set dst (MulI dst src));
  8989   effect(KILL cr);
  8991   ins_cost(300);
  8992   format %{ "imull   $dst, $src\t# int" %}
  8993   opcode(0x0F, 0xAF);
  8994   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
  8995   ins_pipe(ialu_reg_reg_alu0);
  8996 %}
  8998 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
  8999 %{
  9000   match(Set dst (MulI src imm));
  9001   effect(KILL cr);
  9003   ins_cost(300);
  9004   format %{ "imull   $dst, $src, $imm\t# int" %}
  9005   opcode(0x69); /* 69 /r id */
  9006   ins_encode(REX_reg_reg(dst, src),
  9007              OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
  9008   ins_pipe(ialu_reg_reg_alu0);
  9009 %}
  9011 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
  9012 %{
  9013   match(Set dst (MulI dst (LoadI src)));
  9014   effect(KILL cr);
  9016   ins_cost(350);
  9017   format %{ "imull   $dst, $src\t# int" %}
  9018   opcode(0x0F, 0xAF);
  9019   ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
  9020   ins_pipe(ialu_reg_mem_alu0);
  9021 %}
  9023 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
  9024 %{
  9025   match(Set dst (MulI (LoadI src) imm));
  9026   effect(KILL cr);
  9028   ins_cost(300);
  9029   format %{ "imull   $dst, $src, $imm\t# int" %}
  9030   opcode(0x69); /* 69 /r id */
  9031   ins_encode(REX_reg_mem(dst, src),
  9032              OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
  9033   ins_pipe(ialu_reg_mem_alu0);
  9034 %}
  9036 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
  9037 %{
  9038   match(Set dst (MulL dst src));
  9039   effect(KILL cr);
  9041   ins_cost(300);
  9042   format %{ "imulq   $dst, $src\t# long" %}
  9043   opcode(0x0F, 0xAF);
  9044   ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
  9045   ins_pipe(ialu_reg_reg_alu0);
  9046 %}
  9048 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
  9049 %{
  9050   match(Set dst (MulL src imm));
  9051   effect(KILL cr);
  9053   ins_cost(300);
  9054   format %{ "imulq   $dst, $src, $imm\t# long" %}
  9055   opcode(0x69); /* 69 /r id */
  9056   ins_encode(REX_reg_reg_wide(dst, src),
  9057              OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
  9058   ins_pipe(ialu_reg_reg_alu0);
  9059 %}
  9061 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
  9062 %{
  9063   match(Set dst (MulL dst (LoadL src)));
  9064   effect(KILL cr);
  9066   ins_cost(350);
  9067   format %{ "imulq   $dst, $src\t# long" %}
  9068   opcode(0x0F, 0xAF);
  9069   ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
  9070   ins_pipe(ialu_reg_mem_alu0);
  9071 %}
  9073 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
  9074 %{
  9075   match(Set dst (MulL (LoadL src) imm));
  9076   effect(KILL cr);
  9078   ins_cost(300);
  9079   format %{ "imulq   $dst, $src, $imm\t# long" %}
  9080   opcode(0x69); /* 69 /r id */
  9081   ins_encode(REX_reg_mem_wide(dst, src),
  9082              OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
  9083   ins_pipe(ialu_reg_mem_alu0);
  9084 %}
  9086 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
  9087 %{
  9088   match(Set dst (MulHiL src rax));
  9089   effect(USE_KILL rax, KILL cr);
  9091   ins_cost(300);
  9092   format %{ "imulq   RDX:RAX, RAX, $src\t# mulhi" %}
  9093   opcode(0xF7, 0x5); /* Opcode F7 /5 */
  9094   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
  9095   ins_pipe(ialu_reg_reg_alu0);
  9096 %}
  9098 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
  9099                    rFlagsReg cr)
  9100 %{
  9101   match(Set rax (DivI rax div));
  9102   effect(KILL rdx, KILL cr);
  9104   ins_cost(30*100+10*100); // XXX
  9105   format %{ "cmpl    rax, 0x80000000\t# idiv\n\t"
  9106             "jne,s   normal\n\t"
  9107             "xorl    rdx, rdx\n\t"
  9108             "cmpl    $div, -1\n\t"
  9109             "je,s    done\n"
  9110     "normal: cdql\n\t"
  9111             "idivl   $div\n"
  9112     "done:"        %}
  9113   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9114   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
  9115   ins_pipe(ialu_reg_reg_alu0);
  9116 %}
  9118 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
  9119                    rFlagsReg cr)
  9120 %{
  9121   match(Set rax (DivL rax div));
  9122   effect(KILL rdx, KILL cr);
  9124   ins_cost(30*100+10*100); // XXX
  9125   format %{ "movq    rdx, 0x8000000000000000\t# ldiv\n\t"
  9126             "cmpq    rax, rdx\n\t"
  9127             "jne,s   normal\n\t"
  9128             "xorl    rdx, rdx\n\t"
  9129             "cmpq    $div, -1\n\t"
  9130             "je,s    done\n"
  9131     "normal: cdqq\n\t"
  9132             "idivq   $div\n"
  9133     "done:"        %}
  9134   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9135   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
  9136   ins_pipe(ialu_reg_reg_alu0);
  9137 %}
  9139 // Integer DIVMOD with Register, both quotient and mod results
  9140 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
  9141                              rFlagsReg cr)
  9142 %{
  9143   match(DivModI rax div);
  9144   effect(KILL cr);
  9146   ins_cost(30*100+10*100); // XXX
  9147   format %{ "cmpl    rax, 0x80000000\t# idiv\n\t"
  9148             "jne,s   normal\n\t"
  9149             "xorl    rdx, rdx\n\t"
  9150             "cmpl    $div, -1\n\t"
  9151             "je,s    done\n"
  9152     "normal: cdql\n\t"
  9153             "idivl   $div\n"
  9154     "done:"        %}
  9155   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9156   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
  9157   ins_pipe(pipe_slow);
  9158 %}
  9160 // Long DIVMOD with Register, both quotient and mod results
  9161 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
  9162                              rFlagsReg cr)
  9163 %{
  9164   match(DivModL rax div);
  9165   effect(KILL cr);
  9167   ins_cost(30*100+10*100); // XXX
  9168   format %{ "movq    rdx, 0x8000000000000000\t# ldiv\n\t"
  9169             "cmpq    rax, rdx\n\t"
  9170             "jne,s   normal\n\t"
  9171             "xorl    rdx, rdx\n\t"
  9172             "cmpq    $div, -1\n\t"
  9173             "je,s    done\n"
  9174     "normal: cdqq\n\t"
  9175             "idivq   $div\n"
  9176     "done:"        %}
  9177   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9178   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
  9179   ins_pipe(pipe_slow);
  9180 %}
  9182 //----------- DivL-By-Constant-Expansions--------------------------------------
  9183 // DivI cases are handled by the compiler
  9185 // Magic constant, reciprocal of 10
  9186 instruct loadConL_0x6666666666666667(rRegL dst)
  9187 %{
  9188   effect(DEF dst);
  9190   format %{ "movq    $dst, #0x666666666666667\t# Used in div-by-10" %}
  9191   ins_encode(load_immL(dst, 0x6666666666666667));
  9192   ins_pipe(ialu_reg);
  9193 %}
  9195 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
  9196 %{
  9197   effect(DEF dst, USE src, USE_KILL rax, KILL cr);
  9199   format %{ "imulq   rdx:rax, rax, $src\t# Used in div-by-10" %}
  9200   opcode(0xF7, 0x5); /* Opcode F7 /5 */
  9201   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
  9202   ins_pipe(ialu_reg_reg_alu0);
  9203 %}
  9205 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
  9206 %{
  9207   effect(USE_DEF dst, KILL cr);
  9209   format %{ "sarq    $dst, #63\t# Used in div-by-10" %}
  9210   opcode(0xC1, 0x7); /* C1 /7 ib */
  9211   ins_encode(reg_opc_imm_wide(dst, 0x3F));
  9212   ins_pipe(ialu_reg);
  9213 %}
  9215 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
  9216 %{
  9217   effect(USE_DEF dst, KILL cr);
  9219   format %{ "sarq    $dst, #2\t# Used in div-by-10" %}
  9220   opcode(0xC1, 0x7); /* C1 /7 ib */
  9221   ins_encode(reg_opc_imm_wide(dst, 0x2));
  9222   ins_pipe(ialu_reg);
  9223 %}
  9225 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
  9226 %{
  9227   match(Set dst (DivL src div));
  9229   ins_cost((5+8)*100);
  9230   expand %{
  9231     rax_RegL rax;                     // Killed temp
  9232     rFlagsReg cr;                     // Killed
  9233     loadConL_0x6666666666666667(rax); // movq  rax, 0x6666666666666667
  9234     mul_hi(dst, src, rax, cr);        // mulq  rdx:rax <= rax * $src
  9235     sarL_rReg_63(src, cr);            // sarq  src, 63
  9236     sarL_rReg_2(dst, cr);             // sarq  rdx, 2
  9237     subL_rReg(dst, src, cr);          // subl  rdx, src
  9238   %}
  9239 %}
  9241 //-----------------------------------------------------------------------------
  9243 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
  9244                    rFlagsReg cr)
  9245 %{
  9246   match(Set rdx (ModI rax div));
  9247   effect(KILL rax, KILL cr);
  9249   ins_cost(300); // XXX
  9250   format %{ "cmpl    rax, 0x80000000\t# irem\n\t"
  9251             "jne,s   normal\n\t"
  9252             "xorl    rdx, rdx\n\t"
  9253             "cmpl    $div, -1\n\t"
  9254             "je,s    done\n"
  9255     "normal: cdql\n\t"
  9256             "idivl   $div\n"
  9257     "done:"        %}
  9258   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9259   ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
  9260   ins_pipe(ialu_reg_reg_alu0);
  9261 %}
  9263 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
  9264                    rFlagsReg cr)
  9265 %{
  9266   match(Set rdx (ModL rax div));
  9267   effect(KILL rax, KILL cr);
  9269   ins_cost(300); // XXX
  9270   format %{ "movq    rdx, 0x8000000000000000\t# lrem\n\t"
  9271             "cmpq    rax, rdx\n\t"
  9272             "jne,s   normal\n\t"
  9273             "xorl    rdx, rdx\n\t"
  9274             "cmpq    $div, -1\n\t"
  9275             "je,s    done\n"
  9276     "normal: cdqq\n\t"
  9277             "idivq   $div\n"
  9278     "done:"        %}
  9279   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
  9280   ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
  9281   ins_pipe(ialu_reg_reg_alu0);
  9282 %}
  9284 // Integer Shift Instructions
  9285 // Shift Left by one
  9286 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
  9287 %{
  9288   match(Set dst (LShiftI dst shift));
  9289   effect(KILL cr);
  9291   format %{ "sall    $dst, $shift" %}
  9292   opcode(0xD1, 0x4); /* D1 /4 */
  9293   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9294   ins_pipe(ialu_reg);
  9295 %}
  9297 // Shift Left by one
  9298 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9299 %{
  9300   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
  9301   effect(KILL cr);
  9303   format %{ "sall    $dst, $shift\t" %}
  9304   opcode(0xD1, 0x4); /* D1 /4 */
  9305   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9306   ins_pipe(ialu_mem_imm);
  9307 %}
  9309 // Shift Left by 8-bit immediate
  9310 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
  9311 %{
  9312   match(Set dst (LShiftI dst shift));
  9313   effect(KILL cr);
  9315   format %{ "sall    $dst, $shift" %}
  9316   opcode(0xC1, 0x4); /* C1 /4 ib */
  9317   ins_encode(reg_opc_imm(dst, shift));
  9318   ins_pipe(ialu_reg);
  9319 %}
  9321 // Shift Left by 8-bit immediate
  9322 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9323 %{
  9324   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
  9325   effect(KILL cr);
  9327   format %{ "sall    $dst, $shift" %}
  9328   opcode(0xC1, 0x4); /* C1 /4 ib */
  9329   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
  9330   ins_pipe(ialu_mem_imm);
  9331 %}
  9333 // Shift Left by variable
  9334 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
  9335 %{
  9336   match(Set dst (LShiftI dst shift));
  9337   effect(KILL cr);
  9339   format %{ "sall    $dst, $shift" %}
  9340   opcode(0xD3, 0x4); /* D3 /4 */
  9341   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9342   ins_pipe(ialu_reg_reg);
  9343 %}
  9345 // Shift Left by variable
  9346 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9347 %{
  9348   match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
  9349   effect(KILL cr);
  9351   format %{ "sall    $dst, $shift" %}
  9352   opcode(0xD3, 0x4); /* D3 /4 */
  9353   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9354   ins_pipe(ialu_mem_reg);
  9355 %}
  9357 // Arithmetic shift right by one
  9358 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
  9359 %{
  9360   match(Set dst (RShiftI dst shift));
  9361   effect(KILL cr);
  9363   format %{ "sarl    $dst, $shift" %}
  9364   opcode(0xD1, 0x7); /* D1 /7 */
  9365   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9366   ins_pipe(ialu_reg);
  9367 %}
  9369 // Arithmetic shift right by one
  9370 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9371 %{
  9372   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
  9373   effect(KILL cr);
  9375   format %{ "sarl    $dst, $shift" %}
  9376   opcode(0xD1, 0x7); /* D1 /7 */
  9377   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9378   ins_pipe(ialu_mem_imm);
  9379 %}
  9381 // Arithmetic Shift Right by 8-bit immediate
  9382 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
  9383 %{
  9384   match(Set dst (RShiftI dst shift));
  9385   effect(KILL cr);
  9387   format %{ "sarl    $dst, $shift" %}
  9388   opcode(0xC1, 0x7); /* C1 /7 ib */
  9389   ins_encode(reg_opc_imm(dst, shift));
  9390   ins_pipe(ialu_mem_imm);
  9391 %}
  9393 // Arithmetic Shift Right by 8-bit immediate
  9394 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9395 %{
  9396   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
  9397   effect(KILL cr);
  9399   format %{ "sarl    $dst, $shift" %}
  9400   opcode(0xC1, 0x7); /* C1 /7 ib */
  9401   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
  9402   ins_pipe(ialu_mem_imm);
  9403 %}
  9405 // Arithmetic Shift Right by variable
  9406 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
  9407 %{
  9408   match(Set dst (RShiftI dst shift));
  9409   effect(KILL cr);
  9411   format %{ "sarl    $dst, $shift" %}
  9412   opcode(0xD3, 0x7); /* D3 /7 */
  9413   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9414   ins_pipe(ialu_reg_reg);
  9415 %}
  9417 // Arithmetic Shift Right by variable
  9418 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9419 %{
  9420   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
  9421   effect(KILL cr);
  9423   format %{ "sarl    $dst, $shift" %}
  9424   opcode(0xD3, 0x7); /* D3 /7 */
  9425   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9426   ins_pipe(ialu_mem_reg);
  9427 %}
  9429 // Logical shift right by one
  9430 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
  9431 %{
  9432   match(Set dst (URShiftI dst shift));
  9433   effect(KILL cr);
  9435   format %{ "shrl    $dst, $shift" %}
  9436   opcode(0xD1, 0x5); /* D1 /5 */
  9437   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9438   ins_pipe(ialu_reg);
  9439 %}
  9441 // Logical shift right by one
  9442 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9443 %{
  9444   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
  9445   effect(KILL cr);
  9447   format %{ "shrl    $dst, $shift" %}
  9448   opcode(0xD1, 0x5); /* D1 /5 */
  9449   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9450   ins_pipe(ialu_mem_imm);
  9451 %}
  9453 // Logical Shift Right by 8-bit immediate
  9454 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
  9455 %{
  9456   match(Set dst (URShiftI dst shift));
  9457   effect(KILL cr);
  9459   format %{ "shrl    $dst, $shift" %}
  9460   opcode(0xC1, 0x5); /* C1 /5 ib */
  9461   ins_encode(reg_opc_imm(dst, shift));
  9462   ins_pipe(ialu_reg);
  9463 %}
  9465 // Logical Shift Right by 8-bit immediate
  9466 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9467 %{
  9468   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
  9469   effect(KILL cr);
  9471   format %{ "shrl    $dst, $shift" %}
  9472   opcode(0xC1, 0x5); /* C1 /5 ib */
  9473   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
  9474   ins_pipe(ialu_mem_imm);
  9475 %}
  9477 // Logical Shift Right by variable
  9478 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
  9479 %{
  9480   match(Set dst (URShiftI dst shift));
  9481   effect(KILL cr);
  9483   format %{ "shrl    $dst, $shift" %}
  9484   opcode(0xD3, 0x5); /* D3 /5 */
  9485   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9486   ins_pipe(ialu_reg_reg);
  9487 %}
  9489 // Logical Shift Right by variable
  9490 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9491 %{
  9492   match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
  9493   effect(KILL cr);
  9495   format %{ "shrl    $dst, $shift" %}
  9496   opcode(0xD3, 0x5); /* D3 /5 */
  9497   ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
  9498   ins_pipe(ialu_mem_reg);
  9499 %}
  9501 // Long Shift Instructions
  9502 // Shift Left by one
  9503 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
  9504 %{
  9505   match(Set dst (LShiftL dst shift));
  9506   effect(KILL cr);
  9508   format %{ "salq    $dst, $shift" %}
  9509   opcode(0xD1, 0x4); /* D1 /4 */
  9510   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9511   ins_pipe(ialu_reg);
  9512 %}
  9514 // Shift Left by one
  9515 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9516 %{
  9517   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
  9518   effect(KILL cr);
  9520   format %{ "salq    $dst, $shift" %}
  9521   opcode(0xD1, 0x4); /* D1 /4 */
  9522   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9523   ins_pipe(ialu_mem_imm);
  9524 %}
  9526 // Shift Left by 8-bit immediate
  9527 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
  9528 %{
  9529   match(Set dst (LShiftL dst shift));
  9530   effect(KILL cr);
  9532   format %{ "salq    $dst, $shift" %}
  9533   opcode(0xC1, 0x4); /* C1 /4 ib */
  9534   ins_encode(reg_opc_imm_wide(dst, shift));
  9535   ins_pipe(ialu_reg);
  9536 %}
  9538 // Shift Left by 8-bit immediate
  9539 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9540 %{
  9541   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
  9542   effect(KILL cr);
  9544   format %{ "salq    $dst, $shift" %}
  9545   opcode(0xC1, 0x4); /* C1 /4 ib */
  9546   ins_encode(REX_mem_wide(dst), OpcP,
  9547              RM_opc_mem(secondary, dst), Con8or32(shift));
  9548   ins_pipe(ialu_mem_imm);
  9549 %}
  9551 // Shift Left by variable
  9552 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
  9553 %{
  9554   match(Set dst (LShiftL dst shift));
  9555   effect(KILL cr);
  9557   format %{ "salq    $dst, $shift" %}
  9558   opcode(0xD3, 0x4); /* D3 /4 */
  9559   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9560   ins_pipe(ialu_reg_reg);
  9561 %}
  9563 // Shift Left by variable
  9564 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9565 %{
  9566   match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
  9567   effect(KILL cr);
  9569   format %{ "salq    $dst, $shift" %}
  9570   opcode(0xD3, 0x4); /* D3 /4 */
  9571   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9572   ins_pipe(ialu_mem_reg);
  9573 %}
  9575 // Arithmetic shift right by one
  9576 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
  9577 %{
  9578   match(Set dst (RShiftL dst shift));
  9579   effect(KILL cr);
  9581   format %{ "sarq    $dst, $shift" %}
  9582   opcode(0xD1, 0x7); /* D1 /7 */
  9583   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9584   ins_pipe(ialu_reg);
  9585 %}
  9587 // Arithmetic shift right by one
  9588 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9589 %{
  9590   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
  9591   effect(KILL cr);
  9593   format %{ "sarq    $dst, $shift" %}
  9594   opcode(0xD1, 0x7); /* D1 /7 */
  9595   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9596   ins_pipe(ialu_mem_imm);
  9597 %}
  9599 // Arithmetic Shift Right by 8-bit immediate
  9600 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
  9601 %{
  9602   match(Set dst (RShiftL dst shift));
  9603   effect(KILL cr);
  9605   format %{ "sarq    $dst, $shift" %}
  9606   opcode(0xC1, 0x7); /* C1 /7 ib */
  9607   ins_encode(reg_opc_imm_wide(dst, shift));
  9608   ins_pipe(ialu_mem_imm);
  9609 %}
  9611 // Arithmetic Shift Right by 8-bit immediate
  9612 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9613 %{
  9614   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
  9615   effect(KILL cr);
  9617   format %{ "sarq    $dst, $shift" %}
  9618   opcode(0xC1, 0x7); /* C1 /7 ib */
  9619   ins_encode(REX_mem_wide(dst), OpcP,
  9620              RM_opc_mem(secondary, dst), Con8or32(shift));
  9621   ins_pipe(ialu_mem_imm);
  9622 %}
  9624 // Arithmetic Shift Right by variable
  9625 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
  9626 %{
  9627   match(Set dst (RShiftL dst shift));
  9628   effect(KILL cr);
  9630   format %{ "sarq    $dst, $shift" %}
  9631   opcode(0xD3, 0x7); /* D3 /7 */
  9632   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9633   ins_pipe(ialu_reg_reg);
  9634 %}
  9636 // Arithmetic Shift Right by variable
  9637 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9638 %{
  9639   match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
  9640   effect(KILL cr);
  9642   format %{ "sarq    $dst, $shift" %}
  9643   opcode(0xD3, 0x7); /* D3 /7 */
  9644   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9645   ins_pipe(ialu_mem_reg);
  9646 %}
  9648 // Logical shift right by one
  9649 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
  9650 %{
  9651   match(Set dst (URShiftL dst shift));
  9652   effect(KILL cr);
  9654   format %{ "shrq    $dst, $shift" %}
  9655   opcode(0xD1, 0x5); /* D1 /5 */
  9656   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
  9657   ins_pipe(ialu_reg);
  9658 %}
  9660 // Logical shift right by one
  9661 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
  9662 %{
  9663   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
  9664   effect(KILL cr);
  9666   format %{ "shrq    $dst, $shift" %}
  9667   opcode(0xD1, 0x5); /* D1 /5 */
  9668   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9669   ins_pipe(ialu_mem_imm);
  9670 %}
  9672 // Logical Shift Right by 8-bit immediate
  9673 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
  9674 %{
  9675   match(Set dst (URShiftL dst shift));
  9676   effect(KILL cr);
  9678   format %{ "shrq    $dst, $shift" %}
  9679   opcode(0xC1, 0x5); /* C1 /5 ib */
  9680   ins_encode(reg_opc_imm_wide(dst, shift));
  9681   ins_pipe(ialu_reg);
  9682 %}
  9685 // Logical Shift Right by 8-bit immediate
  9686 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
  9687 %{
  9688   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
  9689   effect(KILL cr);
  9691   format %{ "shrq    $dst, $shift" %}
  9692   opcode(0xC1, 0x5); /* C1 /5 ib */
  9693   ins_encode(REX_mem_wide(dst), OpcP,
  9694              RM_opc_mem(secondary, dst), Con8or32(shift));
  9695   ins_pipe(ialu_mem_imm);
  9696 %}
  9698 // Logical Shift Right by variable
  9699 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
  9700 %{
  9701   match(Set dst (URShiftL dst shift));
  9702   effect(KILL cr);
  9704   format %{ "shrq    $dst, $shift" %}
  9705   opcode(0xD3, 0x5); /* D3 /5 */
  9706   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9707   ins_pipe(ialu_reg_reg);
  9708 %}
  9710 // Logical Shift Right by variable
  9711 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
  9712 %{
  9713   match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
  9714   effect(KILL cr);
  9716   format %{ "shrq    $dst, $shift" %}
  9717   opcode(0xD3, 0x5); /* D3 /5 */
  9718   ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
  9719   ins_pipe(ialu_mem_reg);
  9720 %}
  9722 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
  9723 // This idiom is used by the compiler for the i2b bytecode.
  9724 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
  9725 %{
  9726   match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
  9728   format %{ "movsbl  $dst, $src\t# i2b" %}
  9729   opcode(0x0F, 0xBE);
  9730   ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
  9731   ins_pipe(ialu_reg_reg);
  9732 %}
  9734 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
  9735 // This idiom is used by the compiler the i2s bytecode.
  9736 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
  9737 %{
  9738   match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
  9740   format %{ "movswl  $dst, $src\t# i2s" %}
  9741   opcode(0x0F, 0xBF);
  9742   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
  9743   ins_pipe(ialu_reg_reg);
  9744 %}
  9746 // ROL/ROR instructions
  9748 // ROL expand
  9749 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
  9750   effect(KILL cr, USE_DEF dst);
  9752   format %{ "roll    $dst" %}
  9753   opcode(0xD1, 0x0); /* Opcode  D1 /0 */
  9754   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9755   ins_pipe(ialu_reg);
  9756 %}
  9758 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
  9759   effect(USE_DEF dst, USE shift, KILL cr);
  9761   format %{ "roll    $dst, $shift" %}
  9762   opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
  9763   ins_encode( reg_opc_imm(dst, shift) );
  9764   ins_pipe(ialu_reg);
  9765 %}
  9767 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
  9768 %{
  9769   effect(USE_DEF dst, USE shift, KILL cr);
  9771   format %{ "roll    $dst, $shift" %}
  9772   opcode(0xD3, 0x0); /* Opcode D3 /0 */
  9773   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9774   ins_pipe(ialu_reg_reg);
  9775 %}
  9776 // end of ROL expand
  9778 // Rotate Left by one
  9779 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
  9780 %{
  9781   match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
  9783   expand %{
  9784     rolI_rReg_imm1(dst, cr);
  9785   %}
  9786 %}
  9788 // Rotate Left by 8-bit immediate
  9789 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
  9790 %{
  9791   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  9792   match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
  9794   expand %{
  9795     rolI_rReg_imm8(dst, lshift, cr);
  9796   %}
  9797 %}
  9799 // Rotate Left by variable
  9800 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
  9801 %{
  9802   match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
  9804   expand %{
  9805     rolI_rReg_CL(dst, shift, cr);
  9806   %}
  9807 %}
  9809 // Rotate Left by variable
  9810 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
  9811 %{
  9812   match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
  9814   expand %{
  9815     rolI_rReg_CL(dst, shift, cr);
  9816   %}
  9817 %}
  9819 // ROR expand
  9820 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
  9821 %{
  9822   effect(USE_DEF dst, KILL cr);
  9824   format %{ "rorl    $dst" %}
  9825   opcode(0xD1, 0x1); /* D1 /1 */
  9826   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9827   ins_pipe(ialu_reg);
  9828 %}
  9830 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
  9831 %{
  9832   effect(USE_DEF dst, USE shift, KILL cr);
  9834   format %{ "rorl    $dst, $shift" %}
  9835   opcode(0xC1, 0x1); /* C1 /1 ib */
  9836   ins_encode(reg_opc_imm(dst, shift));
  9837   ins_pipe(ialu_reg);
  9838 %}
  9840 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
  9841 %{
  9842   effect(USE_DEF dst, USE shift, KILL cr);
  9844   format %{ "rorl    $dst, $shift" %}
  9845   opcode(0xD3, 0x1); /* D3 /1 */
  9846   ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
  9847   ins_pipe(ialu_reg_reg);
  9848 %}
  9849 // end of ROR expand
  9851 // Rotate Right by one
  9852 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
  9853 %{
  9854   match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
  9856   expand %{
  9857     rorI_rReg_imm1(dst, cr);
  9858   %}
  9859 %}
  9861 // Rotate Right by 8-bit immediate
  9862 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
  9863 %{
  9864   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
  9865   match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
  9867   expand %{
  9868     rorI_rReg_imm8(dst, rshift, cr);
  9869   %}
  9870 %}
  9872 // Rotate Right by variable
  9873 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
  9874 %{
  9875   match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
  9877   expand %{
  9878     rorI_rReg_CL(dst, shift, cr);
  9879   %}
  9880 %}
  9882 // Rotate Right by variable
  9883 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
  9884 %{
  9885   match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
  9887   expand %{
  9888     rorI_rReg_CL(dst, shift, cr);
  9889   %}
  9890 %}
  9892 // for long rotate
  9893 // ROL expand
  9894 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
  9895   effect(USE_DEF dst, KILL cr);
  9897   format %{ "rolq    $dst" %}
  9898   opcode(0xD1, 0x0); /* Opcode  D1 /0 */
  9899   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9900   ins_pipe(ialu_reg);
  9901 %}
  9903 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
  9904   effect(USE_DEF dst, USE shift, KILL cr);
  9906   format %{ "rolq    $dst, $shift" %}
  9907   opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
  9908   ins_encode( reg_opc_imm_wide(dst, shift) );
  9909   ins_pipe(ialu_reg);
  9910 %}
  9912 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
  9913 %{
  9914   effect(USE_DEF dst, USE shift, KILL cr);
  9916   format %{ "rolq    $dst, $shift" %}
  9917   opcode(0xD3, 0x0); /* Opcode D3 /0 */
  9918   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9919   ins_pipe(ialu_reg_reg);
  9920 %}
  9921 // end of ROL expand
  9923 // Rotate Left by one
  9924 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
  9925 %{
  9926   match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
  9928   expand %{
  9929     rolL_rReg_imm1(dst, cr);
  9930   %}
  9931 %}
  9933 // Rotate Left by 8-bit immediate
  9934 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
  9935 %{
  9936   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
  9937   match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
  9939   expand %{
  9940     rolL_rReg_imm8(dst, lshift, cr);
  9941   %}
  9942 %}
  9944 // Rotate Left by variable
  9945 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
  9946 %{
  9947   match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
  9949   expand %{
  9950     rolL_rReg_CL(dst, shift, cr);
  9951   %}
  9952 %}
  9954 // Rotate Left by variable
  9955 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
  9956 %{
  9957   match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
  9959   expand %{
  9960     rolL_rReg_CL(dst, shift, cr);
  9961   %}
  9962 %}
  9964 // ROR expand
  9965 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
  9966 %{
  9967   effect(USE_DEF dst, KILL cr);
  9969   format %{ "rorq    $dst" %}
  9970   opcode(0xD1, 0x1); /* D1 /1 */
  9971   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9972   ins_pipe(ialu_reg);
  9973 %}
  9975 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
  9976 %{
  9977   effect(USE_DEF dst, USE shift, KILL cr);
  9979   format %{ "rorq    $dst, $shift" %}
  9980   opcode(0xC1, 0x1); /* C1 /1 ib */
  9981   ins_encode(reg_opc_imm_wide(dst, shift));
  9982   ins_pipe(ialu_reg);
  9983 %}
  9985 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
  9986 %{
  9987   effect(USE_DEF dst, USE shift, KILL cr);
  9989   format %{ "rorq    $dst, $shift" %}
  9990   opcode(0xD3, 0x1); /* D3 /1 */
  9991   ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
  9992   ins_pipe(ialu_reg_reg);
  9993 %}
  9994 // end of ROR expand
  9996 // Rotate Right by one
  9997 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
  9998 %{
  9999   match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
 10001   expand %{
 10002     rorL_rReg_imm1(dst, cr);
 10003   %}
 10004 %}
 10006 // Rotate Right by 8-bit immediate
 10007 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
 10008 %{
 10009   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
 10010   match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
 10012   expand %{
 10013     rorL_rReg_imm8(dst, rshift, cr);
 10014   %}
 10015 %}
 10017 // Rotate Right by variable
 10018 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
 10019 %{
 10020   match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
 10022   expand %{
 10023     rorL_rReg_CL(dst, shift, cr);
 10024   %}
 10025 %}
 10027 // Rotate Right by variable
 10028 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
 10029 %{
 10030   match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
 10032   expand %{
 10033     rorL_rReg_CL(dst, shift, cr);
 10034   %}
 10035 %}
 10037 // Logical Instructions
 10039 // Integer Logical Instructions
 10041 // And Instructions
 10042 // And Register with Register
 10043 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
 10044 %{
 10045   match(Set dst (AndI dst src));
 10046   effect(KILL cr);
 10048   format %{ "andl    $dst, $src\t# int" %}
 10049   opcode(0x23);
 10050   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
 10051   ins_pipe(ialu_reg_reg);
 10052 %}
 10054 // And Register with Immediate 255
 10055 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
 10056 %{
 10057   match(Set dst (AndI dst src));
 10059   format %{ "movzbl  $dst, $dst\t# int & 0xFF" %}
 10060   opcode(0x0F, 0xB6);
 10061   ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
 10062   ins_pipe(ialu_reg);
 10063 %}
 10065 // And Register with Immediate 255 and promote to long
 10066 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
 10067 %{
 10068   match(Set dst (ConvI2L (AndI src mask)));
 10070   format %{ "movzbl  $dst, $src\t# int & 0xFF -> long" %}
 10071   opcode(0x0F, 0xB6);
 10072   ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
 10073   ins_pipe(ialu_reg);
 10074 %}
 10076 // And Register with Immediate 65535
 10077 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
 10078 %{
 10079   match(Set dst (AndI dst src));
 10081   format %{ "movzwl  $dst, $dst\t# int & 0xFFFF" %}
 10082   opcode(0x0F, 0xB7);
 10083   ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
 10084   ins_pipe(ialu_reg);
 10085 %}
 10087 // And Register with Immediate 65535 and promote to long
 10088 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
 10089 %{
 10090   match(Set dst (ConvI2L (AndI src mask)));
 10092   format %{ "movzwl  $dst, $src\t# int & 0xFFFF -> long" %}
 10093   opcode(0x0F, 0xB7);
 10094   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
 10095   ins_pipe(ialu_reg);
 10096 %}
 10098 // And Register with Immediate
 10099 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
 10100 %{
 10101   match(Set dst (AndI dst src));
 10102   effect(KILL cr);
 10104   format %{ "andl    $dst, $src\t# int" %}
 10105   opcode(0x81, 0x04); /* Opcode 81 /4 */
 10106   ins_encode(OpcSErm(dst, src), Con8or32(src));
 10107   ins_pipe(ialu_reg);
 10108 %}
 10110 // And Register with Memory
 10111 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
 10112 %{
 10113   match(Set dst (AndI dst (LoadI src)));
 10114   effect(KILL cr);
 10116   ins_cost(125);
 10117   format %{ "andl    $dst, $src\t# int" %}
 10118   opcode(0x23);
 10119   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
 10120   ins_pipe(ialu_reg_mem);
 10121 %}
 10123 // And Memory with Register
 10124 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
 10125 %{
 10126   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
 10127   effect(KILL cr);
 10129   ins_cost(150);
 10130   format %{ "andl    $dst, $src\t# int" %}
 10131   opcode(0x21); /* Opcode 21 /r */
 10132   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
 10133   ins_pipe(ialu_mem_reg);
 10134 %}
 10136 // And Memory with Immediate
 10137 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
 10138 %{
 10139   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
 10140   effect(KILL cr);
 10142   ins_cost(125);
 10143   format %{ "andl    $dst, $src\t# int" %}
 10144   opcode(0x81, 0x4); /* Opcode 81 /4 id */
 10145   ins_encode(REX_mem(dst), OpcSE(src),
 10146              RM_opc_mem(secondary, dst), Con8or32(src));
 10147   ins_pipe(ialu_mem_imm);
 10148 %}
 10150 // Or Instructions
 10151 // Or Register with Register
 10152 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
 10153 %{
 10154   match(Set dst (OrI dst src));
 10155   effect(KILL cr);
 10157   format %{ "orl     $dst, $src\t# int" %}
 10158   opcode(0x0B);
 10159   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
 10160   ins_pipe(ialu_reg_reg);
 10161 %}
 10163 // Or Register with Immediate
 10164 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
 10165 %{
 10166   match(Set dst (OrI dst src));
 10167   effect(KILL cr);
 10169   format %{ "orl     $dst, $src\t# int" %}
 10170   opcode(0x81, 0x01); /* Opcode 81 /1 id */
 10171   ins_encode(OpcSErm(dst, src), Con8or32(src));
 10172   ins_pipe(ialu_reg);
 10173 %}
 10175 // Or Register with Memory
 10176 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
 10177 %{
 10178   match(Set dst (OrI dst (LoadI src)));
 10179   effect(KILL cr);
 10181   ins_cost(125);
 10182   format %{ "orl     $dst, $src\t# int" %}
 10183   opcode(0x0B);
 10184   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
 10185   ins_pipe(ialu_reg_mem);
 10186 %}
 10188 // Or Memory with Register
 10189 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
 10190 %{
 10191   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
 10192   effect(KILL cr);
 10194   ins_cost(150);
 10195   format %{ "orl     $dst, $src\t# int" %}
 10196   opcode(0x09); /* Opcode 09 /r */
 10197   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
 10198   ins_pipe(ialu_mem_reg);
 10199 %}
 10201 // Or Memory with Immediate
 10202 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
 10203 %{
 10204   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
 10205   effect(KILL cr);
 10207   ins_cost(125);
 10208   format %{ "orl     $dst, $src\t# int" %}
 10209   opcode(0x81, 0x1); /* Opcode 81 /1 id */
 10210   ins_encode(REX_mem(dst), OpcSE(src),
 10211              RM_opc_mem(secondary, dst), Con8or32(src));
 10212   ins_pipe(ialu_mem_imm);
 10213 %}
 10215 // Xor Instructions
 10216 // Xor Register with Register
 10217 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
 10218 %{
 10219   match(Set dst (XorI dst src));
 10220   effect(KILL cr);
 10222   format %{ "xorl    $dst, $src\t# int" %}
 10223   opcode(0x33);
 10224   ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
 10225   ins_pipe(ialu_reg_reg);
 10226 %}
 10228 // Xor Register with Immediate -1
 10229 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
 10230   match(Set dst (XorI dst imm));  
 10232   format %{ "not    $dst" %}  
 10233   ins_encode %{
 10234      __ notl($dst$$Register);
 10235   %}
 10236   ins_pipe(ialu_reg);
 10237 %}
 10239 // Xor Register with Immediate
 10240 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
 10241 %{
 10242   match(Set dst (XorI dst src));
 10243   effect(KILL cr);
 10245   format %{ "xorl    $dst, $src\t# int" %}
 10246   opcode(0x81, 0x06); /* Opcode 81 /6 id */
 10247   ins_encode(OpcSErm(dst, src), Con8or32(src));
 10248   ins_pipe(ialu_reg);
 10249 %}
 10251 // Xor Register with Memory
 10252 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
 10253 %{
 10254   match(Set dst (XorI dst (LoadI src)));
 10255   effect(KILL cr);
 10257   ins_cost(125);
 10258   format %{ "xorl    $dst, $src\t# int" %}
 10259   opcode(0x33);
 10260   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
 10261   ins_pipe(ialu_reg_mem);
 10262 %}
 10264 // Xor Memory with Register
 10265 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
 10266 %{
 10267   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
 10268   effect(KILL cr);
 10270   ins_cost(150);
 10271   format %{ "xorl    $dst, $src\t# int" %}
 10272   opcode(0x31); /* Opcode 31 /r */
 10273   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
 10274   ins_pipe(ialu_mem_reg);
 10275 %}
 10277 // Xor Memory with Immediate
 10278 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
 10279 %{
 10280   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
 10281   effect(KILL cr);
 10283   ins_cost(125);
 10284   format %{ "xorl    $dst, $src\t# int" %}
 10285   opcode(0x81, 0x6); /* Opcode 81 /6 id */
 10286   ins_encode(REX_mem(dst), OpcSE(src),
 10287              RM_opc_mem(secondary, dst), Con8or32(src));
 10288   ins_pipe(ialu_mem_imm);
 10289 %}
 10292 // Long Logical Instructions
 10294 // And Instructions
 10295 // And Register with Register
 10296 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
 10297 %{
 10298   match(Set dst (AndL dst src));
 10299   effect(KILL cr);
 10301   format %{ "andq    $dst, $src\t# long" %}
 10302   opcode(0x23);
 10303   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
 10304   ins_pipe(ialu_reg_reg);
 10305 %}
 10307 // And Register with Immediate 255
 10308 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
 10309 %{
 10310   match(Set dst (AndL dst src));
 10312   format %{ "movzbq  $dst, $dst\t# long & 0xFF" %}
 10313   opcode(0x0F, 0xB6);
 10314   ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
 10315   ins_pipe(ialu_reg);
 10316 %}
 10318 // And Register with Immediate 65535
 10319 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src)
 10320 %{
 10321   match(Set dst (AndL dst src));
 10323   format %{ "movzwq  $dst, $dst\t# long & 0xFFFF" %}
 10324   opcode(0x0F, 0xB7);
 10325   ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
 10326   ins_pipe(ialu_reg);
 10327 %}
 10329 // And Register with Immediate
 10330 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
 10331 %{
 10332   match(Set dst (AndL dst src));
 10333   effect(KILL cr);
 10335   format %{ "andq    $dst, $src\t# long" %}
 10336   opcode(0x81, 0x04); /* Opcode 81 /4 */
 10337   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
 10338   ins_pipe(ialu_reg);
 10339 %}
 10341 // And Register with Memory
 10342 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
 10343 %{
 10344   match(Set dst (AndL dst (LoadL src)));
 10345   effect(KILL cr);
 10347   ins_cost(125);
 10348   format %{ "andq    $dst, $src\t# long" %}
 10349   opcode(0x23);
 10350   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
 10351   ins_pipe(ialu_reg_mem);
 10352 %}
 10354 // And Memory with Register
 10355 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
 10356 %{
 10357   match(Set dst (StoreL dst (AndL (LoadL dst) src)));
 10358   effect(KILL cr);
 10360   ins_cost(150);
 10361   format %{ "andq    $dst, $src\t# long" %}
 10362   opcode(0x21); /* Opcode 21 /r */
 10363   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
 10364   ins_pipe(ialu_mem_reg);
 10365 %}
 10367 // And Memory with Immediate
 10368 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
 10369 %{
 10370   match(Set dst (StoreL dst (AndL (LoadL dst) src)));
 10371   effect(KILL cr);
 10373   ins_cost(125);
 10374   format %{ "andq    $dst, $src\t# long" %}
 10375   opcode(0x81, 0x4); /* Opcode 81 /4 id */
 10376   ins_encode(REX_mem_wide(dst), OpcSE(src),
 10377              RM_opc_mem(secondary, dst), Con8or32(src));
 10378   ins_pipe(ialu_mem_imm);
 10379 %}
 10381 // Or Instructions
 10382 // Or Register with Register
 10383 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
 10384 %{
 10385   match(Set dst (OrL dst src));
 10386   effect(KILL cr);
 10388   format %{ "orq     $dst, $src\t# long" %}
 10389   opcode(0x0B);
 10390   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
 10391   ins_pipe(ialu_reg_reg);
 10392 %}
 10394 // Use any_RegP to match R15 (TLS register) without spilling.
 10395 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
 10396   match(Set dst (OrL dst (CastP2X src)));
 10397   effect(KILL cr);
 10399   format %{ "orq     $dst, $src\t# long" %}
 10400   opcode(0x0B);
 10401   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
 10402   ins_pipe(ialu_reg_reg);
 10403 %}
 10406 // Or Register with Immediate
 10407 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
 10408 %{
 10409   match(Set dst (OrL dst src));
 10410   effect(KILL cr);
 10412   format %{ "orq     $dst, $src\t# long" %}
 10413   opcode(0x81, 0x01); /* Opcode 81 /1 id */
 10414   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
 10415   ins_pipe(ialu_reg);
 10416 %}
 10418 // Or Register with Memory
 10419 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
 10420 %{
 10421   match(Set dst (OrL dst (LoadL src)));
 10422   effect(KILL cr);
 10424   ins_cost(125);
 10425   format %{ "orq     $dst, $src\t# long" %}
 10426   opcode(0x0B);
 10427   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
 10428   ins_pipe(ialu_reg_mem);
 10429 %}
 10431 // Or Memory with Register
 10432 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
 10433 %{
 10434   match(Set dst (StoreL dst (OrL (LoadL dst) src)));
 10435   effect(KILL cr);
 10437   ins_cost(150);
 10438   format %{ "orq     $dst, $src\t# long" %}
 10439   opcode(0x09); /* Opcode 09 /r */
 10440   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
 10441   ins_pipe(ialu_mem_reg);
 10442 %}
 10444 // Or Memory with Immediate
 10445 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
 10446 %{
 10447   match(Set dst (StoreL dst (OrL (LoadL dst) src)));
 10448   effect(KILL cr);
 10450   ins_cost(125);
 10451   format %{ "orq     $dst, $src\t# long" %}
 10452   opcode(0x81, 0x1); /* Opcode 81 /1 id */
 10453   ins_encode(REX_mem_wide(dst), OpcSE(src),
 10454              RM_opc_mem(secondary, dst), Con8or32(src));
 10455   ins_pipe(ialu_mem_imm);
 10456 %}
 10458 // Xor Instructions
 10459 // Xor Register with Register
 10460 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
 10461 %{
 10462   match(Set dst (XorL dst src));
 10463   effect(KILL cr);
 10465   format %{ "xorq    $dst, $src\t# long" %}
 10466   opcode(0x33);
 10467   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
 10468   ins_pipe(ialu_reg_reg);
 10469 %}
 10471 // Xor Register with Immediate -1
 10472 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
 10473   match(Set dst (XorL dst imm));  
 10475   format %{ "notq   $dst" %}  
 10476   ins_encode %{
 10477      __ notq($dst$$Register);
 10478   %}
 10479   ins_pipe(ialu_reg);
 10480 %}
 10482 // Xor Register with Immediate
 10483 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
 10484 %{
 10485   match(Set dst (XorL dst src));
 10486   effect(KILL cr);
 10488   format %{ "xorq    $dst, $src\t# long" %}
 10489   opcode(0x81, 0x06); /* Opcode 81 /6 id */
 10490   ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
 10491   ins_pipe(ialu_reg);
 10492 %}
 10494 // Xor Register with Memory
 10495 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
 10496 %{
 10497   match(Set dst (XorL dst (LoadL src)));
 10498   effect(KILL cr);
 10500   ins_cost(125);
 10501   format %{ "xorq    $dst, $src\t# long" %}
 10502   opcode(0x33);
 10503   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
 10504   ins_pipe(ialu_reg_mem);
 10505 %}
 10507 // Xor Memory with Register
 10508 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
 10509 %{
 10510   match(Set dst (StoreL dst (XorL (LoadL dst) src)));
 10511   effect(KILL cr);
 10513   ins_cost(150);
 10514   format %{ "xorq    $dst, $src\t# long" %}
 10515   opcode(0x31); /* Opcode 31 /r */
 10516   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
 10517   ins_pipe(ialu_mem_reg);
 10518 %}
 10520 // Xor Memory with Immediate
 10521 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
 10522 %{
 10523   match(Set dst (StoreL dst (XorL (LoadL dst) src)));
 10524   effect(KILL cr);
 10526   ins_cost(125);
 10527   format %{ "xorq    $dst, $src\t# long" %}
 10528   opcode(0x81, 0x6); /* Opcode 81 /6 id */
 10529   ins_encode(REX_mem_wide(dst), OpcSE(src),
 10530              RM_opc_mem(secondary, dst), Con8or32(src));
 10531   ins_pipe(ialu_mem_imm);
 10532 %}
 10534 // Convert Int to Boolean
 10535 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
 10536 %{
 10537   match(Set dst (Conv2B src));
 10538   effect(KILL cr);
 10540   format %{ "testl   $src, $src\t# ci2b\n\t"
 10541             "setnz   $dst\n\t"
 10542             "movzbl  $dst, $dst" %}
 10543   ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
 10544              setNZ_reg(dst),
 10545              REX_reg_breg(dst, dst), // movzbl
 10546              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
 10547   ins_pipe(pipe_slow); // XXX
 10548 %}
 10550 // Convert Pointer to Boolean
 10551 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
 10552 %{
 10553   match(Set dst (Conv2B src));
 10554   effect(KILL cr);
 10556   format %{ "testq   $src, $src\t# cp2b\n\t"
 10557             "setnz   $dst\n\t"
 10558             "movzbl  $dst, $dst" %}
 10559   ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
 10560              setNZ_reg(dst),
 10561              REX_reg_breg(dst, dst), // movzbl
 10562              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
 10563   ins_pipe(pipe_slow); // XXX
 10564 %}
 10566 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
 10567 %{
 10568   match(Set dst (CmpLTMask p q));
 10569   effect(KILL cr);
 10571   ins_cost(400); // XXX
 10572   format %{ "cmpl    $p, $q\t# cmpLTMask\n\t"
 10573             "setlt   $dst\n\t"
 10574             "movzbl  $dst, $dst\n\t"
 10575             "negl    $dst" %}
 10576   ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
 10577              setLT_reg(dst),
 10578              REX_reg_breg(dst, dst), // movzbl
 10579              Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
 10580              neg_reg(dst));
 10581   ins_pipe(pipe_slow);
 10582 %}
 10584 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
 10585 %{
 10586   match(Set dst (CmpLTMask dst zero));
 10587   effect(KILL cr);
 10589   ins_cost(100); // XXX
 10590   format %{ "sarl    $dst, #31\t# cmpLTMask0" %}
 10591   opcode(0xC1, 0x7);  /* C1 /7 ib */
 10592   ins_encode(reg_opc_imm(dst, 0x1F));
 10593   ins_pipe(ialu_reg);
 10594 %}
 10597 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
 10598                          rRegI tmp,
 10599                          rFlagsReg cr)
 10600 %{
 10601   match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
 10602   effect(TEMP tmp, KILL cr);
 10604   ins_cost(400); // XXX
 10605   format %{ "subl    $p, $q\t# cadd_cmpLTMask1\n\t"
 10606             "sbbl    $tmp, $tmp\n\t"
 10607             "andl    $tmp, $y\n\t"
 10608             "addl    $p, $tmp" %}
 10609   ins_encode(enc_cmpLTP(p, q, y, tmp));
 10610   ins_pipe(pipe_cmplt);
 10611 %}
 10613 /* If I enable this, I encourage spilling in the inner loop of compress.
 10614 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
 10615 %{
 10616   match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
 10617   effect( TEMP tmp, KILL cr );
 10618   ins_cost(400);
 10620   format %{ "SUB    $p,$q\n\t"
 10621             "SBB    RCX,RCX\n\t"
 10622             "AND    RCX,$y\n\t"
 10623             "ADD    $p,RCX" %}
 10624   ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
 10625 %}
 10626 */
 10628 //---------- FP Instructions------------------------------------------------
 10630 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
 10631 %{
 10632   match(Set cr (CmpF src1 src2));
 10634   ins_cost(145);
 10635   format %{ "ucomiss $src1, $src2\n\t"
 10636             "jnp,s   exit\n\t"
 10637             "pushfq\t# saw NaN, set CF\n\t"
 10638             "andq    [rsp], #0xffffff2b\n\t"
 10639             "popfq\n"
 10640     "exit:   nop\t# avoid branch to branch" %}
 10641   opcode(0x0F, 0x2E);
 10642   ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
 10643              cmpfp_fixup);
 10644   ins_pipe(pipe_slow);
 10645 %}
 10647 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
 10648   match(Set cr (CmpF src1 src2));
 10650   ins_cost(145);
 10651   format %{ "ucomiss $src1, $src2" %}
 10652   ins_encode %{
 10653     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
 10654   %}
 10655   ins_pipe(pipe_slow);
 10656 %}
 10658 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
 10659 %{
 10660   match(Set cr (CmpF src1 (LoadF src2)));
 10662   ins_cost(145);
 10663   format %{ "ucomiss $src1, $src2\n\t"
 10664             "jnp,s   exit\n\t"
 10665             "pushfq\t# saw NaN, set CF\n\t"
 10666             "andq    [rsp], #0xffffff2b\n\t"
 10667             "popfq\n"
 10668     "exit:   nop\t# avoid branch to branch" %}
 10669   opcode(0x0F, 0x2E);
 10670   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
 10671              cmpfp_fixup);
 10672   ins_pipe(pipe_slow);
 10673 %}
 10675 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
 10676   match(Set cr (CmpF src1 (LoadF src2)));
 10678   ins_cost(100);
 10679   format %{ "ucomiss $src1, $src2" %}
 10680   opcode(0x0F, 0x2E);
 10681   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
 10682   ins_pipe(pipe_slow);
 10683 %}
 10685 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
 10686 %{
 10687   match(Set cr (CmpF src1 src2));
 10689   ins_cost(145);
 10690   format %{ "ucomiss $src1, $src2\n\t"
 10691             "jnp,s   exit\n\t"
 10692             "pushfq\t# saw NaN, set CF\n\t"
 10693             "andq    [rsp], #0xffffff2b\n\t"
 10694             "popfq\n"
 10695     "exit:   nop\t# avoid branch to branch" %}
 10696   opcode(0x0F, 0x2E);
 10697   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
 10698              cmpfp_fixup);
 10699   ins_pipe(pipe_slow);
 10700 %}
 10702 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{
 10703   match(Set cr (CmpF src1 src2));
 10705   ins_cost(100);
 10706   format %{ "ucomiss $src1, $src2" %}
 10707   opcode(0x0F, 0x2E);
 10708   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2));
 10709   ins_pipe(pipe_slow);
 10710 %}
 10712 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
 10713 %{
 10714   match(Set cr (CmpD src1 src2));
 10716   ins_cost(145);
 10717   format %{ "ucomisd $src1, $src2\n\t"
 10718             "jnp,s   exit\n\t"
 10719             "pushfq\t# saw NaN, set CF\n\t"
 10720             "andq    [rsp], #0xffffff2b\n\t"
 10721             "popfq\n"
 10722     "exit:   nop\t# avoid branch to branch" %}
 10723   opcode(0x66, 0x0F, 0x2E);
 10724   ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
 10725              cmpfp_fixup);
 10726   ins_pipe(pipe_slow);
 10727 %}
 10729 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
 10730   match(Set cr (CmpD src1 src2));
 10732   ins_cost(100);
 10733   format %{ "ucomisd $src1, $src2 test" %}
 10734   ins_encode %{
 10735     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
 10736   %}
 10737   ins_pipe(pipe_slow);
 10738 %}
 10740 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
 10741 %{
 10742   match(Set cr (CmpD src1 (LoadD src2)));
 10744   ins_cost(145);
 10745   format %{ "ucomisd $src1, $src2\n\t"
 10746             "jnp,s   exit\n\t"
 10747             "pushfq\t# saw NaN, set CF\n\t"
 10748             "andq    [rsp], #0xffffff2b\n\t"
 10749             "popfq\n"
 10750     "exit:   nop\t# avoid branch to branch" %}
 10751   opcode(0x66, 0x0F, 0x2E);
 10752   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
 10753              cmpfp_fixup);
 10754   ins_pipe(pipe_slow);
 10755 %}
 10757 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
 10758   match(Set cr (CmpD src1 (LoadD src2)));
 10760   ins_cost(100);
 10761   format %{ "ucomisd $src1, $src2" %}
 10762   opcode(0x66, 0x0F, 0x2E);
 10763   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
 10764   ins_pipe(pipe_slow);
 10765 %}
 10767 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
 10768 %{
 10769   match(Set cr (CmpD src1 src2));
 10771   ins_cost(145);
 10772   format %{ "ucomisd $src1, [$src2]\n\t"
 10773             "jnp,s   exit\n\t"
 10774             "pushfq\t# saw NaN, set CF\n\t"
 10775             "andq    [rsp], #0xffffff2b\n\t"
 10776             "popfq\n"
 10777     "exit:   nop\t# avoid branch to branch" %}
 10778   opcode(0x66, 0x0F, 0x2E);
 10779   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
 10780              cmpfp_fixup);
 10781   ins_pipe(pipe_slow);
 10782 %}
 10784 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{
 10785   match(Set cr (CmpD src1 src2));
 10787   ins_cost(100);
 10788   format %{ "ucomisd $src1, [$src2]" %}
 10789   opcode(0x66, 0x0F, 0x2E);
 10790   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2));
 10791   ins_pipe(pipe_slow);
 10792 %}
 10794 // Compare into -1,0,1
 10795 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
 10796 %{
 10797   match(Set dst (CmpF3 src1 src2));
 10798   effect(KILL cr);
 10800   ins_cost(275);
 10801   format %{ "ucomiss $src1, $src2\n\t"
 10802             "movl    $dst, #-1\n\t"
 10803             "jp,s    done\n\t"
 10804             "jb,s    done\n\t"
 10805             "setne   $dst\n\t"
 10806             "movzbl  $dst, $dst\n"
 10807     "done:" %}
 10809   opcode(0x0F, 0x2E);
 10810   ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
 10811              cmpfp3(dst));
 10812   ins_pipe(pipe_slow);
 10813 %}
 10815 // Compare into -1,0,1
 10816 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
 10817 %{
 10818   match(Set dst (CmpF3 src1 (LoadF src2)));
 10819   effect(KILL cr);
 10821   ins_cost(275);
 10822   format %{ "ucomiss $src1, $src2\n\t"
 10823             "movl    $dst, #-1\n\t"
 10824             "jp,s    done\n\t"
 10825             "jb,s    done\n\t"
 10826             "setne   $dst\n\t"
 10827             "movzbl  $dst, $dst\n"
 10828     "done:" %}
 10830   opcode(0x0F, 0x2E);
 10831   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
 10832              cmpfp3(dst));
 10833   ins_pipe(pipe_slow);
 10834 %}
 10836 // Compare into -1,0,1
 10837 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
 10838 %{
 10839   match(Set dst (CmpF3 src1 src2));
 10840   effect(KILL cr);
 10842   ins_cost(275);
 10843   format %{ "ucomiss $src1, [$src2]\n\t"
 10844             "movl    $dst, #-1\n\t"
 10845             "jp,s    done\n\t"
 10846             "jb,s    done\n\t"
 10847             "setne   $dst\n\t"
 10848             "movzbl  $dst, $dst\n"
 10849     "done:" %}
 10851   opcode(0x0F, 0x2E);
 10852   ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
 10853              cmpfp3(dst));
 10854   ins_pipe(pipe_slow);
 10855 %}
 10857 // Compare into -1,0,1
 10858 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
 10859 %{
 10860   match(Set dst (CmpD3 src1 src2));
 10861   effect(KILL cr);
 10863   ins_cost(275);
 10864   format %{ "ucomisd $src1, $src2\n\t"
 10865             "movl    $dst, #-1\n\t"
 10866             "jp,s    done\n\t"
 10867             "jb,s    done\n\t"
 10868             "setne   $dst\n\t"
 10869             "movzbl  $dst, $dst\n"
 10870     "done:" %}
 10872   opcode(0x66, 0x0F, 0x2E);
 10873   ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
 10874              cmpfp3(dst));
 10875   ins_pipe(pipe_slow);
 10876 %}
 10878 // Compare into -1,0,1
 10879 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
 10880 %{
 10881   match(Set dst (CmpD3 src1 (LoadD src2)));
 10882   effect(KILL cr);
 10884   ins_cost(275);
 10885   format %{ "ucomisd $src1, $src2\n\t"
 10886             "movl    $dst, #-1\n\t"
 10887             "jp,s    done\n\t"
 10888             "jb,s    done\n\t"
 10889             "setne   $dst\n\t"
 10890             "movzbl  $dst, $dst\n"
 10891     "done:" %}
 10893   opcode(0x66, 0x0F, 0x2E);
 10894   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
 10895              cmpfp3(dst));
 10896   ins_pipe(pipe_slow);
 10897 %}
 10899 // Compare into -1,0,1
 10900 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
 10901 %{
 10902   match(Set dst (CmpD3 src1 src2));
 10903   effect(KILL cr);
 10905   ins_cost(275);
 10906   format %{ "ucomisd $src1, [$src2]\n\t"
 10907             "movl    $dst, #-1\n\t"
 10908             "jp,s    done\n\t"
 10909             "jb,s    done\n\t"
 10910             "setne   $dst\n\t"
 10911             "movzbl  $dst, $dst\n"
 10912     "done:" %}
 10914   opcode(0x66, 0x0F, 0x2E);
 10915   ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
 10916              cmpfp3(dst));
 10917   ins_pipe(pipe_slow);
 10918 %}
 10920 instruct addF_reg(regF dst, regF src)
 10921 %{
 10922   match(Set dst (AddF dst src));
 10924   format %{ "addss   $dst, $src" %}
 10925   ins_cost(150); // XXX
 10926   opcode(0xF3, 0x0F, 0x58);
 10927   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 10928   ins_pipe(pipe_slow);
 10929 %}
 10931 instruct addF_mem(regF dst, memory src)
 10932 %{
 10933   match(Set dst (AddF dst (LoadF src)));
 10935   format %{ "addss   $dst, $src" %}
 10936   ins_cost(150); // XXX
 10937   opcode(0xF3, 0x0F, 0x58);
 10938   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 10939   ins_pipe(pipe_slow);
 10940 %}
 10942 instruct addF_imm(regF dst, immF src)
 10943 %{
 10944   match(Set dst (AddF dst src));
 10946   format %{ "addss   $dst, [$src]" %}
 10947   ins_cost(150); // XXX
 10948   opcode(0xF3, 0x0F, 0x58);
 10949   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
 10950   ins_pipe(pipe_slow);
 10951 %}
 10953 instruct addD_reg(regD dst, regD src)
 10954 %{
 10955   match(Set dst (AddD dst src));
 10957   format %{ "addsd   $dst, $src" %}
 10958   ins_cost(150); // XXX
 10959   opcode(0xF2, 0x0F, 0x58);
 10960   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 10961   ins_pipe(pipe_slow);
 10962 %}
 10964 instruct addD_mem(regD dst, memory src)
 10965 %{
 10966   match(Set dst (AddD dst (LoadD src)));
 10968   format %{ "addsd   $dst, $src" %}
 10969   ins_cost(150); // XXX
 10970   opcode(0xF2, 0x0F, 0x58);
 10971   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 10972   ins_pipe(pipe_slow);
 10973 %}
 10975 instruct addD_imm(regD dst, immD src)
 10976 %{
 10977   match(Set dst (AddD dst src));
 10979   format %{ "addsd   $dst, [$src]" %}
 10980   ins_cost(150); // XXX
 10981   opcode(0xF2, 0x0F, 0x58);
 10982   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
 10983   ins_pipe(pipe_slow);
 10984 %}
 10986 instruct subF_reg(regF dst, regF src)
 10987 %{
 10988   match(Set dst (SubF dst src));
 10990   format %{ "subss   $dst, $src" %}
 10991   ins_cost(150); // XXX
 10992   opcode(0xF3, 0x0F, 0x5C);
 10993   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 10994   ins_pipe(pipe_slow);
 10995 %}
 10997 instruct subF_mem(regF dst, memory src)
 10998 %{
 10999   match(Set dst (SubF dst (LoadF src)));
 11001   format %{ "subss   $dst, $src" %}
 11002   ins_cost(150); // XXX
 11003   opcode(0xF3, 0x0F, 0x5C);
 11004   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11005   ins_pipe(pipe_slow);
 11006 %}
 11008 instruct subF_imm(regF dst, immF src)
 11009 %{
 11010   match(Set dst (SubF dst src));
 11012   format %{ "subss   $dst, [$src]" %}
 11013   ins_cost(150); // XXX
 11014   opcode(0xF3, 0x0F, 0x5C);
 11015   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
 11016   ins_pipe(pipe_slow);
 11017 %}
 11019 instruct subD_reg(regD dst, regD src)
 11020 %{
 11021   match(Set dst (SubD dst src));
 11023   format %{ "subsd   $dst, $src" %}
 11024   ins_cost(150); // XXX
 11025   opcode(0xF2, 0x0F, 0x5C);
 11026   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11027   ins_pipe(pipe_slow);
 11028 %}
 11030 instruct subD_mem(regD dst, memory src)
 11031 %{
 11032   match(Set dst (SubD dst (LoadD src)));
 11034   format %{ "subsd   $dst, $src" %}
 11035   ins_cost(150); // XXX
 11036   opcode(0xF2, 0x0F, 0x5C);
 11037   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11038   ins_pipe(pipe_slow);
 11039 %}
 11041 instruct subD_imm(regD dst, immD src)
 11042 %{
 11043   match(Set dst (SubD dst src));
 11045   format %{ "subsd   $dst, [$src]" %}
 11046   ins_cost(150); // XXX
 11047   opcode(0xF2, 0x0F, 0x5C);
 11048   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
 11049   ins_pipe(pipe_slow);
 11050 %}
 11052 instruct mulF_reg(regF dst, regF src)
 11053 %{
 11054   match(Set dst (MulF dst src));
 11056   format %{ "mulss   $dst, $src" %}
 11057   ins_cost(150); // XXX
 11058   opcode(0xF3, 0x0F, 0x59);
 11059   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11060   ins_pipe(pipe_slow);
 11061 %}
 11063 instruct mulF_mem(regF dst, memory src)
 11064 %{
 11065   match(Set dst (MulF dst (LoadF src)));
 11067   format %{ "mulss   $dst, $src" %}
 11068   ins_cost(150); // XXX
 11069   opcode(0xF3, 0x0F, 0x59);
 11070   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11071   ins_pipe(pipe_slow);
 11072 %}
 11074 instruct mulF_imm(regF dst, immF src)
 11075 %{
 11076   match(Set dst (MulF dst src));
 11078   format %{ "mulss   $dst, [$src]" %}
 11079   ins_cost(150); // XXX
 11080   opcode(0xF3, 0x0F, 0x59);
 11081   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
 11082   ins_pipe(pipe_slow);
 11083 %}
 11085 instruct mulD_reg(regD dst, regD src)
 11086 %{
 11087   match(Set dst (MulD dst src));
 11089   format %{ "mulsd   $dst, $src" %}
 11090   ins_cost(150); // XXX
 11091   opcode(0xF2, 0x0F, 0x59);
 11092   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11093   ins_pipe(pipe_slow);
 11094 %}
 11096 instruct mulD_mem(regD dst, memory src)
 11097 %{
 11098   match(Set dst (MulD dst (LoadD src)));
 11100   format %{ "mulsd   $dst, $src" %}
 11101   ins_cost(150); // XXX
 11102   opcode(0xF2, 0x0F, 0x59);
 11103   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11104   ins_pipe(pipe_slow);
 11105 %}
 11107 instruct mulD_imm(regD dst, immD src)
 11108 %{
 11109   match(Set dst (MulD dst src));
 11111   format %{ "mulsd   $dst, [$src]" %}
 11112   ins_cost(150); // XXX
 11113   opcode(0xF2, 0x0F, 0x59);
 11114   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
 11115   ins_pipe(pipe_slow);
 11116 %}
 11118 instruct divF_reg(regF dst, regF src)
 11119 %{
 11120   match(Set dst (DivF dst src));
 11122   format %{ "divss   $dst, $src" %}
 11123   ins_cost(150); // XXX
 11124   opcode(0xF3, 0x0F, 0x5E);
 11125   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11126   ins_pipe(pipe_slow);
 11127 %}
 11129 instruct divF_mem(regF dst, memory src)
 11130 %{
 11131   match(Set dst (DivF dst (LoadF src)));
 11133   format %{ "divss   $dst, $src" %}
 11134   ins_cost(150); // XXX
 11135   opcode(0xF3, 0x0F, 0x5E);
 11136   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11137   ins_pipe(pipe_slow);
 11138 %}
 11140 instruct divF_imm(regF dst, immF src)
 11141 %{
 11142   match(Set dst (DivF dst src));
 11144   format %{ "divss   $dst, [$src]" %}
 11145   ins_cost(150); // XXX
 11146   opcode(0xF3, 0x0F, 0x5E);
 11147   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
 11148   ins_pipe(pipe_slow);
 11149 %}
 11151 instruct divD_reg(regD dst, regD src)
 11152 %{
 11153   match(Set dst (DivD dst src));
 11155   format %{ "divsd   $dst, $src" %}
 11156   ins_cost(150); // XXX
 11157   opcode(0xF2, 0x0F, 0x5E);
 11158   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11159   ins_pipe(pipe_slow);
 11160 %}
 11162 instruct divD_mem(regD dst, memory src)
 11163 %{
 11164   match(Set dst (DivD dst (LoadD src)));
 11166   format %{ "divsd   $dst, $src" %}
 11167   ins_cost(150); // XXX
 11168   opcode(0xF2, 0x0F, 0x5E);
 11169   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11170   ins_pipe(pipe_slow);
 11171 %}
 11173 instruct divD_imm(regD dst, immD src)
 11174 %{
 11175   match(Set dst (DivD dst src));
 11177   format %{ "divsd   $dst, [$src]" %}
 11178   ins_cost(150); // XXX
 11179   opcode(0xF2, 0x0F, 0x5E);
 11180   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
 11181   ins_pipe(pipe_slow);
 11182 %}
 11184 instruct sqrtF_reg(regF dst, regF src)
 11185 %{
 11186   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
 11188   format %{ "sqrtss  $dst, $src" %}
 11189   ins_cost(150); // XXX
 11190   opcode(0xF3, 0x0F, 0x51);
 11191   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11192   ins_pipe(pipe_slow);
 11193 %}
 11195 instruct sqrtF_mem(regF dst, memory src)
 11196 %{
 11197   match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
 11199   format %{ "sqrtss  $dst, $src" %}
 11200   ins_cost(150); // XXX
 11201   opcode(0xF3, 0x0F, 0x51);
 11202   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11203   ins_pipe(pipe_slow);
 11204 %}
 11206 instruct sqrtF_imm(regF dst, immF src)
 11207 %{
 11208   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
 11210   format %{ "sqrtss  $dst, [$src]" %}
 11211   ins_cost(150); // XXX
 11212   opcode(0xF3, 0x0F, 0x51);
 11213   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
 11214   ins_pipe(pipe_slow);
 11215 %}
 11217 instruct sqrtD_reg(regD dst, regD src)
 11218 %{
 11219   match(Set dst (SqrtD src));
 11221   format %{ "sqrtsd  $dst, $src" %}
 11222   ins_cost(150); // XXX
 11223   opcode(0xF2, 0x0F, 0x51);
 11224   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11225   ins_pipe(pipe_slow);
 11226 %}
 11228 instruct sqrtD_mem(regD dst, memory src)
 11229 %{
 11230   match(Set dst (SqrtD (LoadD src)));
 11232   format %{ "sqrtsd  $dst, $src" %}
 11233   ins_cost(150); // XXX
 11234   opcode(0xF2, 0x0F, 0x51);
 11235   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11236   ins_pipe(pipe_slow);
 11237 %}
 11239 instruct sqrtD_imm(regD dst, immD src)
 11240 %{
 11241   match(Set dst (SqrtD src));
 11243   format %{ "sqrtsd  $dst, [$src]" %}
 11244   ins_cost(150); // XXX
 11245   opcode(0xF2, 0x0F, 0x51);
 11246   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
 11247   ins_pipe(pipe_slow);
 11248 %}
 11250 instruct absF_reg(regF dst)
 11251 %{
 11252   match(Set dst (AbsF dst));
 11254   format %{ "andps   $dst, [0x7fffffff]\t# abs float by sign masking" %}
 11255   ins_encode(absF_encoding(dst));
 11256   ins_pipe(pipe_slow);
 11257 %}
 11259 instruct absD_reg(regD dst)
 11260 %{
 11261   match(Set dst (AbsD dst));
 11263   format %{ "andpd   $dst, [0x7fffffffffffffff]\t"
 11264             "# abs double by sign masking" %}
 11265   ins_encode(absD_encoding(dst));
 11266   ins_pipe(pipe_slow);
 11267 %}
 11269 instruct negF_reg(regF dst)
 11270 %{
 11271   match(Set dst (NegF dst));
 11273   format %{ "xorps   $dst, [0x80000000]\t# neg float by sign flipping" %}
 11274   ins_encode(negF_encoding(dst));
 11275   ins_pipe(pipe_slow);
 11276 %}
 11278 instruct negD_reg(regD dst)
 11279 %{
 11280   match(Set dst (NegD dst));
 11282   format %{ "xorpd   $dst, [0x8000000000000000]\t"
 11283             "# neg double by sign flipping" %}
 11284   ins_encode(negD_encoding(dst));
 11285   ins_pipe(pipe_slow);
 11286 %}
 11288 // -----------Trig and Trancendental Instructions------------------------------
 11289 instruct cosD_reg(regD dst) %{
 11290   match(Set dst (CosD dst));
 11292   format %{ "dcos   $dst\n\t" %}
 11293   opcode(0xD9, 0xFF);
 11294   ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
 11295   ins_pipe( pipe_slow );
 11296 %}
 11298 instruct sinD_reg(regD dst) %{
 11299   match(Set dst (SinD dst));
 11301   format %{ "dsin   $dst\n\t" %}
 11302   opcode(0xD9, 0xFE);
 11303   ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
 11304   ins_pipe( pipe_slow );
 11305 %}
 11307 instruct tanD_reg(regD dst) %{
 11308   match(Set dst (TanD dst));
 11310   format %{ "dtan   $dst\n\t" %}
 11311   ins_encode( Push_SrcXD(dst),
 11312               Opcode(0xD9), Opcode(0xF2),   //fptan
 11313               Opcode(0xDD), Opcode(0xD8),   //fstp st
 11314               Push_ResultXD(dst) );
 11315   ins_pipe( pipe_slow );
 11316 %}
 11318 instruct log10D_reg(regD dst) %{
 11319   // The source and result Double operands in XMM registers
 11320   match(Set dst (Log10D dst));
 11321   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
 11322   // fyl2x        ; compute log_10(2) * log_2(x)
 11323   format %{ "fldlg2\t\t\t#Log10\n\t"
 11324             "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
 11325          %}
 11326    ins_encode(Opcode(0xD9), Opcode(0xEC),   // fldlg2
 11327               Push_SrcXD(dst),
 11328               Opcode(0xD9), Opcode(0xF1),   // fyl2x
 11329               Push_ResultXD(dst));
 11331   ins_pipe( pipe_slow );
 11332 %}
 11334 instruct logD_reg(regD dst) %{
 11335   // The source and result Double operands in XMM registers
 11336   match(Set dst (LogD dst));
 11337   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
 11338   // fyl2x        ; compute log_e(2) * log_2(x)
 11339   format %{ "fldln2\t\t\t#Log_e\n\t"
 11340             "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
 11341          %}
 11342   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
 11343               Push_SrcXD(dst),
 11344               Opcode(0xD9), Opcode(0xF1),   // fyl2x
 11345               Push_ResultXD(dst));
 11346   ins_pipe( pipe_slow );
 11347 %}
 11351 //----------Arithmetic Conversion Instructions---------------------------------
 11353 instruct roundFloat_nop(regF dst)
 11354 %{
 11355   match(Set dst (RoundFloat dst));
 11357   ins_cost(0);
 11358   ins_encode();
 11359   ins_pipe(empty);
 11360 %}
 11362 instruct roundDouble_nop(regD dst)
 11363 %{
 11364   match(Set dst (RoundDouble dst));
 11366   ins_cost(0);
 11367   ins_encode();
 11368   ins_pipe(empty);
 11369 %}
 11371 instruct convF2D_reg_reg(regD dst, regF src)
 11372 %{
 11373   match(Set dst (ConvF2D src));
 11375   format %{ "cvtss2sd $dst, $src" %}
 11376   opcode(0xF3, 0x0F, 0x5A);
 11377   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11378   ins_pipe(pipe_slow); // XXX
 11379 %}
 11381 instruct convF2D_reg_mem(regD dst, memory src)
 11382 %{
 11383   match(Set dst (ConvF2D (LoadF src)));
 11385   format %{ "cvtss2sd $dst, $src" %}
 11386   opcode(0xF3, 0x0F, 0x5A);
 11387   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11388   ins_pipe(pipe_slow); // XXX
 11389 %}
 11391 instruct convD2F_reg_reg(regF dst, regD src)
 11392 %{
 11393   match(Set dst (ConvD2F src));
 11395   format %{ "cvtsd2ss $dst, $src" %}
 11396   opcode(0xF2, 0x0F, 0x5A);
 11397   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11398   ins_pipe(pipe_slow); // XXX
 11399 %}
 11401 instruct convD2F_reg_mem(regF dst, memory src)
 11402 %{
 11403   match(Set dst (ConvD2F (LoadD src)));
 11405   format %{ "cvtsd2ss $dst, $src" %}
 11406   opcode(0xF2, 0x0F, 0x5A);
 11407   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11408   ins_pipe(pipe_slow); // XXX
 11409 %}
 11411 // XXX do mem variants
 11412 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
 11413 %{
 11414   match(Set dst (ConvF2I src));
 11415   effect(KILL cr);
 11417   format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
 11418             "cmpl    $dst, #0x80000000\n\t"
 11419             "jne,s   done\n\t"
 11420             "subq    rsp, #8\n\t"
 11421             "movss   [rsp], $src\n\t"
 11422             "call    f2i_fixup\n\t"
 11423             "popq    $dst\n"
 11424     "done:   "%}
 11425   opcode(0xF3, 0x0F, 0x2C);
 11426   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
 11427              f2i_fixup(dst, src));
 11428   ins_pipe(pipe_slow);
 11429 %}
 11431 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
 11432 %{
 11433   match(Set dst (ConvF2L src));
 11434   effect(KILL cr);
 11436   format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
 11437             "cmpq    $dst, [0x8000000000000000]\n\t"
 11438             "jne,s   done\n\t"
 11439             "subq    rsp, #8\n\t"
 11440             "movss   [rsp], $src\n\t"
 11441             "call    f2l_fixup\n\t"
 11442             "popq    $dst\n"
 11443     "done:   "%}
 11444   opcode(0xF3, 0x0F, 0x2C);
 11445   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
 11446              f2l_fixup(dst, src));
 11447   ins_pipe(pipe_slow);
 11448 %}
 11450 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
 11451 %{
 11452   match(Set dst (ConvD2I src));
 11453   effect(KILL cr);
 11455   format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
 11456             "cmpl    $dst, #0x80000000\n\t"
 11457             "jne,s   done\n\t"
 11458             "subq    rsp, #8\n\t"
 11459             "movsd   [rsp], $src\n\t"
 11460             "call    d2i_fixup\n\t"
 11461             "popq    $dst\n"
 11462     "done:   "%}
 11463   opcode(0xF2, 0x0F, 0x2C);
 11464   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
 11465              d2i_fixup(dst, src));
 11466   ins_pipe(pipe_slow);
 11467 %}
 11469 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
 11470 %{
 11471   match(Set dst (ConvD2L src));
 11472   effect(KILL cr);
 11474   format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
 11475             "cmpq    $dst, [0x8000000000000000]\n\t"
 11476             "jne,s   done\n\t"
 11477             "subq    rsp, #8\n\t"
 11478             "movsd   [rsp], $src\n\t"
 11479             "call    d2l_fixup\n\t"
 11480             "popq    $dst\n"
 11481     "done:   "%}
 11482   opcode(0xF2, 0x0F, 0x2C);
 11483   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
 11484              d2l_fixup(dst, src));
 11485   ins_pipe(pipe_slow);
 11486 %}
 11488 instruct convI2F_reg_reg(regF dst, rRegI src)
 11489 %{
 11490   predicate(!UseXmmI2F);
 11491   match(Set dst (ConvI2F src));
 11493   format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
 11494   opcode(0xF3, 0x0F, 0x2A);
 11495   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11496   ins_pipe(pipe_slow); // XXX
 11497 %}
 11499 instruct convI2F_reg_mem(regF dst, memory src)
 11500 %{
 11501   match(Set dst (ConvI2F (LoadI src)));
 11503   format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
 11504   opcode(0xF3, 0x0F, 0x2A);
 11505   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11506   ins_pipe(pipe_slow); // XXX
 11507 %}
 11509 instruct convI2D_reg_reg(regD dst, rRegI src)
 11510 %{
 11511   predicate(!UseXmmI2D);
 11512   match(Set dst (ConvI2D src));
 11514   format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
 11515   opcode(0xF2, 0x0F, 0x2A);
 11516   ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11517   ins_pipe(pipe_slow); // XXX
 11518 %}
 11520 instruct convI2D_reg_mem(regD dst, memory src)
 11521 %{
 11522   match(Set dst (ConvI2D (LoadI src)));
 11524   format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
 11525   opcode(0xF2, 0x0F, 0x2A);
 11526   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11527   ins_pipe(pipe_slow); // XXX
 11528 %}
 11530 instruct convXI2F_reg(regF dst, rRegI src)
 11531 %{
 11532   predicate(UseXmmI2F);
 11533   match(Set dst (ConvI2F src));
 11535   format %{ "movdl $dst, $src\n\t"
 11536             "cvtdq2psl $dst, $dst\t# i2f" %}
 11537   ins_encode %{
 11538     __ movdl($dst$$XMMRegister, $src$$Register);
 11539     __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
 11540   %}
 11541   ins_pipe(pipe_slow); // XXX
 11542 %}
 11544 instruct convXI2D_reg(regD dst, rRegI src)
 11545 %{
 11546   predicate(UseXmmI2D);
 11547   match(Set dst (ConvI2D src));
 11549   format %{ "movdl $dst, $src\n\t"
 11550             "cvtdq2pdl $dst, $dst\t# i2d" %}
 11551   ins_encode %{
 11552     __ movdl($dst$$XMMRegister, $src$$Register);
 11553     __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
 11554   %}
 11555   ins_pipe(pipe_slow); // XXX
 11556 %}
 11558 instruct convL2F_reg_reg(regF dst, rRegL src)
 11559 %{
 11560   match(Set dst (ConvL2F src));
 11562   format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
 11563   opcode(0xF3, 0x0F, 0x2A);
 11564   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11565   ins_pipe(pipe_slow); // XXX
 11566 %}
 11568 instruct convL2F_reg_mem(regF dst, memory src)
 11569 %{
 11570   match(Set dst (ConvL2F (LoadL src)));
 11572   format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
 11573   opcode(0xF3, 0x0F, 0x2A);
 11574   ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11575   ins_pipe(pipe_slow); // XXX
 11576 %}
 11578 instruct convL2D_reg_reg(regD dst, rRegL src)
 11579 %{
 11580   match(Set dst (ConvL2D src));
 11582   format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
 11583   opcode(0xF2, 0x0F, 0x2A);
 11584   ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
 11585   ins_pipe(pipe_slow); // XXX
 11586 %}
 11588 instruct convL2D_reg_mem(regD dst, memory src)
 11589 %{
 11590   match(Set dst (ConvL2D (LoadL src)));
 11592   format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
 11593   opcode(0xF2, 0x0F, 0x2A);
 11594   ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11595   ins_pipe(pipe_slow); // XXX
 11596 %}
 11598 instruct convI2L_reg_reg(rRegL dst, rRegI src)
 11599 %{
 11600   match(Set dst (ConvI2L src));
 11602   ins_cost(125);
 11603   format %{ "movslq  $dst, $src\t# i2l" %}
 11604   opcode(0x63); // needs REX.W
 11605   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
 11606   ins_pipe(ialu_reg_reg);
 11607 %}
 11609 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
 11610 // %{
 11611 //   match(Set dst (ConvI2L src));
 11612 // //   predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
 11613 // //             _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
 11614 //   predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
 11615 //             (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
 11616 //             ((const TypeNode*) n)->type()->is_long()->_lo ==
 11617 //             (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
 11619 //   format %{ "movl    $dst, $src\t# unsigned i2l" %}
 11620 //   ins_encode(enc_copy(dst, src));
 11621 // //   opcode(0x63); // needs REX.W
 11622 // //   ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
 11623 //   ins_pipe(ialu_reg_reg);
 11624 // %}
 11626 // Zero-extend convert int to long
 11627 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
 11628 %{
 11629   match(Set dst (AndL (ConvI2L src) mask));
 11631   format %{ "movl    $dst, $src\t# i2l zero-extend\n\t" %}
 11632   ins_encode(enc_copy(dst, src));
 11633   ins_pipe(ialu_reg_reg);
 11634 %}
 11636 // Zero-extend convert int to long
 11637 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
 11638 %{
 11639   match(Set dst (AndL (ConvI2L (LoadI src)) mask));
 11641   format %{ "movl    $dst, $src\t# i2l zero-extend\n\t" %}
 11642   opcode(0x8B);
 11643   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
 11644   ins_pipe(ialu_reg_mem);
 11645 %}
 11647 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
 11648 %{
 11649   match(Set dst (AndL src mask));
 11651   format %{ "movl    $dst, $src\t# zero-extend long" %}
 11652   ins_encode(enc_copy_always(dst, src));
 11653   ins_pipe(ialu_reg_reg);
 11654 %}
 11656 instruct convL2I_reg_reg(rRegI dst, rRegL src)
 11657 %{
 11658   match(Set dst (ConvL2I src));
 11660   format %{ "movl    $dst, $src\t# l2i" %}
 11661   ins_encode(enc_copy_always(dst, src));
 11662   ins_pipe(ialu_reg_reg);
 11663 %}
 11666 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
 11667   match(Set dst (MoveF2I src));
 11668   effect(DEF dst, USE src);
 11670   ins_cost(125);
 11671   format %{ "movl    $dst, $src\t# MoveF2I_stack_reg" %}
 11672   opcode(0x8B);
 11673   ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
 11674   ins_pipe(ialu_reg_mem);
 11675 %}
 11677 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
 11678   match(Set dst (MoveI2F src));
 11679   effect(DEF dst, USE src);
 11681   ins_cost(125);
 11682   format %{ "movss   $dst, $src\t# MoveI2F_stack_reg" %}
 11683   opcode(0xF3, 0x0F, 0x10);
 11684   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11685   ins_pipe(pipe_slow);
 11686 %}
 11688 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
 11689   match(Set dst (MoveD2L src));
 11690   effect(DEF dst, USE src);
 11692   ins_cost(125);
 11693   format %{ "movq    $dst, $src\t# MoveD2L_stack_reg" %}
 11694   opcode(0x8B);
 11695   ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
 11696   ins_pipe(ialu_reg_mem);
 11697 %}
 11699 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
 11700   predicate(!UseXmmLoadAndClearUpper);
 11701   match(Set dst (MoveL2D src));
 11702   effect(DEF dst, USE src);
 11704   ins_cost(125);
 11705   format %{ "movlpd  $dst, $src\t# MoveL2D_stack_reg" %}
 11706   opcode(0x66, 0x0F, 0x12);
 11707   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11708   ins_pipe(pipe_slow);
 11709 %}
 11711 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
 11712   predicate(UseXmmLoadAndClearUpper);
 11713   match(Set dst (MoveL2D src));
 11714   effect(DEF dst, USE src);
 11716   ins_cost(125);
 11717   format %{ "movsd   $dst, $src\t# MoveL2D_stack_reg" %}
 11718   opcode(0xF2, 0x0F, 0x10);
 11719   ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
 11720   ins_pipe(pipe_slow);
 11721 %}
 11724 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
 11725   match(Set dst (MoveF2I src));
 11726   effect(DEF dst, USE src);
 11728   ins_cost(95); // XXX
 11729   format %{ "movss   $dst, $src\t# MoveF2I_reg_stack" %}
 11730   opcode(0xF3, 0x0F, 0x11);
 11731   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
 11732   ins_pipe(pipe_slow);
 11733 %}
 11735 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
 11736   match(Set dst (MoveI2F src));
 11737   effect(DEF dst, USE src);
 11739   ins_cost(100);
 11740   format %{ "movl    $dst, $src\t# MoveI2F_reg_stack" %}
 11741   opcode(0x89);
 11742   ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
 11743   ins_pipe( ialu_mem_reg );
 11744 %}
 11746 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
 11747   match(Set dst (MoveD2L src));
 11748   effect(DEF dst, USE src);
 11750   ins_cost(95); // XXX
 11751   format %{ "movsd   $dst, $src\t# MoveL2D_reg_stack" %}
 11752   opcode(0xF2, 0x0F, 0x11);
 11753   ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
 11754   ins_pipe(pipe_slow);
 11755 %}
 11757 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
 11758   match(Set dst (MoveL2D src));
 11759   effect(DEF dst, USE src);
 11761   ins_cost(100);
 11762   format %{ "movq    $dst, $src\t# MoveL2D_reg_stack" %}
 11763   opcode(0x89);
 11764   ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
 11765   ins_pipe(ialu_mem_reg);
 11766 %}
 11768 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
 11769   match(Set dst (MoveF2I src));
 11770   effect(DEF dst, USE src);
 11771   ins_cost(85);
 11772   format %{ "movd    $dst,$src\t# MoveF2I" %}
 11773   ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
 11774   ins_pipe( pipe_slow );
 11775 %}
 11777 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
 11778   match(Set dst (MoveD2L src));
 11779   effect(DEF dst, USE src);
 11780   ins_cost(85);
 11781   format %{ "movd    $dst,$src\t# MoveD2L" %}
 11782   ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
 11783   ins_pipe( pipe_slow );
 11784 %}
 11786 // The next instructions have long latency and use Int unit. Set high cost.
 11787 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
 11788   match(Set dst (MoveI2F src));
 11789   effect(DEF dst, USE src);
 11790   ins_cost(300);
 11791   format %{ "movd    $dst,$src\t# MoveI2F" %}
 11792   ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
 11793   ins_pipe( pipe_slow );
 11794 %}
 11796 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
 11797   match(Set dst (MoveL2D src));
 11798   effect(DEF dst, USE src);
 11799   ins_cost(300);
 11800   format %{ "movd    $dst,$src\t# MoveL2D" %}
 11801   ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
 11802   ins_pipe( pipe_slow );
 11803 %}
 11805 // Replicate scalar to packed byte (1 byte) values in xmm
 11806 instruct Repl8B_reg(regD dst, regD src) %{
 11807   match(Set dst (Replicate8B src));
 11808   format %{ "MOVDQA  $dst,$src\n\t"
 11809             "PUNPCKLBW $dst,$dst\n\t"
 11810             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
 11811   ins_encode( pshufd_8x8(dst, src));
 11812   ins_pipe( pipe_slow );
 11813 %}
 11815 // Replicate scalar to packed byte (1 byte) values in xmm
 11816 instruct Repl8B_rRegI(regD dst, rRegI src) %{
 11817   match(Set dst (Replicate8B src));
 11818   format %{ "MOVD    $dst,$src\n\t"
 11819             "PUNPCKLBW $dst,$dst\n\t"
 11820             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
 11821   ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
 11822   ins_pipe( pipe_slow );
 11823 %}
 11825 // Replicate scalar zero to packed byte (1 byte) values in xmm
 11826 instruct Repl8B_immI0(regD dst, immI0 zero) %{
 11827   match(Set dst (Replicate8B zero));
 11828   format %{ "PXOR  $dst,$dst\t! replicate8B" %}
 11829   ins_encode( pxor(dst, dst));
 11830   ins_pipe( fpu_reg_reg );
 11831 %}
 11833 // Replicate scalar to packed shore (2 byte) values in xmm
 11834 instruct Repl4S_reg(regD dst, regD src) %{
 11835   match(Set dst (Replicate4S src));
 11836   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
 11837   ins_encode( pshufd_4x16(dst, src));
 11838   ins_pipe( fpu_reg_reg );
 11839 %}
 11841 // Replicate scalar to packed shore (2 byte) values in xmm
 11842 instruct Repl4S_rRegI(regD dst, rRegI src) %{
 11843   match(Set dst (Replicate4S src));
 11844   format %{ "MOVD    $dst,$src\n\t"
 11845             "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
 11846   ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
 11847   ins_pipe( fpu_reg_reg );
 11848 %}
 11850 // Replicate scalar zero to packed short (2 byte) values in xmm
 11851 instruct Repl4S_immI0(regD dst, immI0 zero) %{
 11852   match(Set dst (Replicate4S zero));
 11853   format %{ "PXOR  $dst,$dst\t! replicate4S" %}
 11854   ins_encode( pxor(dst, dst));
 11855   ins_pipe( fpu_reg_reg );
 11856 %}
 11858 // Replicate scalar to packed char (2 byte) values in xmm
 11859 instruct Repl4C_reg(regD dst, regD src) %{
 11860   match(Set dst (Replicate4C src));
 11861   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
 11862   ins_encode( pshufd_4x16(dst, src));
 11863   ins_pipe( fpu_reg_reg );
 11864 %}
 11866 // Replicate scalar to packed char (2 byte) values in xmm
 11867 instruct Repl4C_rRegI(regD dst, rRegI src) %{
 11868   match(Set dst (Replicate4C src));
 11869   format %{ "MOVD    $dst,$src\n\t"
 11870             "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
 11871   ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
 11872   ins_pipe( fpu_reg_reg );
 11873 %}
 11875 // Replicate scalar zero to packed char (2 byte) values in xmm
 11876 instruct Repl4C_immI0(regD dst, immI0 zero) %{
 11877   match(Set dst (Replicate4C zero));
 11878   format %{ "PXOR  $dst,$dst\t! replicate4C" %}
 11879   ins_encode( pxor(dst, dst));
 11880   ins_pipe( fpu_reg_reg );
 11881 %}
 11883 // Replicate scalar to packed integer (4 byte) values in xmm
 11884 instruct Repl2I_reg(regD dst, regD src) %{
 11885   match(Set dst (Replicate2I src));
 11886   format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
 11887   ins_encode( pshufd(dst, src, 0x00));
 11888   ins_pipe( fpu_reg_reg );
 11889 %}
 11891 // Replicate scalar to packed integer (4 byte) values in xmm
 11892 instruct Repl2I_rRegI(regD dst, rRegI src) %{
 11893   match(Set dst (Replicate2I src));
 11894   format %{ "MOVD   $dst,$src\n\t"
 11895             "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
 11896   ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
 11897   ins_pipe( fpu_reg_reg );
 11898 %}
 11900 // Replicate scalar zero to packed integer (2 byte) values in xmm
 11901 instruct Repl2I_immI0(regD dst, immI0 zero) %{
 11902   match(Set dst (Replicate2I zero));
 11903   format %{ "PXOR  $dst,$dst\t! replicate2I" %}
 11904   ins_encode( pxor(dst, dst));
 11905   ins_pipe( fpu_reg_reg );
 11906 %}
 11908 // Replicate scalar to packed single precision floating point values in xmm
 11909 instruct Repl2F_reg(regD dst, regD src) %{
 11910   match(Set dst (Replicate2F src));
 11911   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
 11912   ins_encode( pshufd(dst, src, 0xe0));
 11913   ins_pipe( fpu_reg_reg );
 11914 %}
 11916 // Replicate scalar to packed single precision floating point values in xmm
 11917 instruct Repl2F_regF(regD dst, regF src) %{
 11918   match(Set dst (Replicate2F src));
 11919   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
 11920   ins_encode( pshufd(dst, src, 0xe0));
 11921   ins_pipe( fpu_reg_reg );
 11922 %}
 11924 // Replicate scalar to packed single precision floating point values in xmm
 11925 instruct Repl2F_immF0(regD dst, immF0 zero) %{
 11926   match(Set dst (Replicate2F zero));
 11927   format %{ "PXOR  $dst,$dst\t! replicate2F" %}
 11928   ins_encode( pxor(dst, dst));
 11929   ins_pipe( fpu_reg_reg );
 11930 %}
 11933 // =======================================================================
 11934 // fast clearing of an array
 11935 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
 11936                   rFlagsReg cr)
 11937 %{
 11938   match(Set dummy (ClearArray cnt base));
 11939   effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
 11941   format %{ "xorl    rax, rax\t# ClearArray:\n\t"
 11942             "rep stosq\t# Store rax to *rdi++ while rcx--" %}
 11943   ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
 11944              Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
 11945   ins_pipe(pipe_slow);
 11946 %}
 11948 instruct string_compare(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
 11949                         rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result, rFlagsReg cr)
 11950 %{
 11951   match(Set result (StrComp str1 str2));
 11952   effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
 11953   //ins_cost(300);
 11955   format %{ "String Compare $str1, $str2 -> $result    // XXX KILL RAX, RBX" %}
 11956   ins_encode( enc_String_Compare(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
 11957   ins_pipe( pipe_slow );
 11958 %}
 11960 instruct string_indexof(rsi_RegP str1, rdi_RegP str2, regD tmp1, rax_RegI tmp2,
 11961                         rcx_RegI tmp3, rdx_RegI tmp4, rbx_RegI result, rFlagsReg cr)
 11962 %{
 11963   predicate(UseSSE42Intrinsics);
 11964   match(Set result (StrIndexOf str1 str2));
 11965   effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
 11967   format %{ "String IndexOf $str1,$str2 -> $result   // KILL RAX, RCX, RDX" %}
 11968   ins_encode( enc_String_IndexOf(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
 11969   ins_pipe( pipe_slow );
 11970 %}
 11972 // fast string equals
 11973 instruct string_equals(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2, rbx_RegI tmp3,
 11974                        rcx_RegI tmp4, rax_RegI result, rFlagsReg cr)
 11975 %{
 11976   match(Set result (StrEquals str1 str2));
 11977   effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
 11979   format %{ "String Equals $str1,$str2 -> $result    // KILL RBX, RCX" %}
 11980   ins_encode( enc_String_Equals(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
 11981   ins_pipe( pipe_slow );
 11982 %}
 11984 // fast array equals
 11985 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, regD tmp1, regD tmp2, rax_RegI tmp3,
 11986                       rbx_RegI tmp4, rcx_RegI result, rFlagsReg cr)
 11987 %{
 11988   match(Set result (AryEq ary1 ary2));
 11989   effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
 11990   //ins_cost(300);
 11992   format %{ "Array Equals $ary1,$ary2 -> $result   // KILL RAX, RBX" %}
 11993   ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, tmp3, tmp4, result) );
 11994   ins_pipe( pipe_slow );
 11995 %}
 11997 //----------Control Flow Instructions------------------------------------------
 11998 // Signed compare Instructions
 12000 // XXX more variants!!
 12001 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
 12002 %{
 12003   match(Set cr (CmpI op1 op2));
 12004   effect(DEF cr, USE op1, USE op2);
 12006   format %{ "cmpl    $op1, $op2" %}
 12007   opcode(0x3B);  /* Opcode 3B /r */
 12008   ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
 12009   ins_pipe(ialu_cr_reg_reg);
 12010 %}
 12012 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
 12013 %{
 12014   match(Set cr (CmpI op1 op2));
 12016   format %{ "cmpl    $op1, $op2" %}
 12017   opcode(0x81, 0x07); /* Opcode 81 /7 */
 12018   ins_encode(OpcSErm(op1, op2), Con8or32(op2));
 12019   ins_pipe(ialu_cr_reg_imm);
 12020 %}
 12022 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
 12023 %{
 12024   match(Set cr (CmpI op1 (LoadI op2)));
 12026   ins_cost(500); // XXX
 12027   format %{ "cmpl    $op1, $op2" %}
 12028   opcode(0x3B); /* Opcode 3B /r */
 12029   ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
 12030   ins_pipe(ialu_cr_reg_mem);
 12031 %}
 12033 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
 12034 %{
 12035   match(Set cr (CmpI src zero));
 12037   format %{ "testl   $src, $src" %}
 12038   opcode(0x85);
 12039   ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
 12040   ins_pipe(ialu_cr_reg_imm);
 12041 %}
 12043 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
 12044 %{
 12045   match(Set cr (CmpI (AndI src con) zero));
 12047   format %{ "testl   $src, $con" %}
 12048   opcode(0xF7, 0x00);
 12049   ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
 12050   ins_pipe(ialu_cr_reg_imm);
 12051 %}
 12053 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
 12054 %{
 12055   match(Set cr (CmpI (AndI src (LoadI mem)) zero));
 12057   format %{ "testl   $src, $mem" %}
 12058   opcode(0x85);
 12059   ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
 12060   ins_pipe(ialu_cr_reg_mem);
 12061 %}
 12063 // Unsigned compare Instructions; really, same as signed except they
 12064 // produce an rFlagsRegU instead of rFlagsReg.
 12065 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
 12066 %{
 12067   match(Set cr (CmpU op1 op2));
 12069   format %{ "cmpl    $op1, $op2\t# unsigned" %}
 12070   opcode(0x3B); /* Opcode 3B /r */
 12071   ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
 12072   ins_pipe(ialu_cr_reg_reg);
 12073 %}
 12075 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
 12076 %{
 12077   match(Set cr (CmpU op1 op2));
 12079   format %{ "cmpl    $op1, $op2\t# unsigned" %}
 12080   opcode(0x81,0x07); /* Opcode 81 /7 */
 12081   ins_encode(OpcSErm(op1, op2), Con8or32(op2));
 12082   ins_pipe(ialu_cr_reg_imm);
 12083 %}
 12085 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
 12086 %{
 12087   match(Set cr (CmpU op1 (LoadI op2)));
 12089   ins_cost(500); // XXX
 12090   format %{ "cmpl    $op1, $op2\t# unsigned" %}
 12091   opcode(0x3B); /* Opcode 3B /r */
 12092   ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
 12093   ins_pipe(ialu_cr_reg_mem);
 12094 %}
 12096 // // // Cisc-spilled version of cmpU_rReg
 12097 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
 12098 // //%{
 12099 // //  match(Set cr (CmpU (LoadI op1) op2));
 12100 // //
 12101 // //  format %{ "CMPu   $op1,$op2" %}
 12102 // //  ins_cost(500);
 12103 // //  opcode(0x39);  /* Opcode 39 /r */
 12104 // //  ins_encode( OpcP, reg_mem( op1, op2) );
 12105 // //%}
 12107 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
 12108 %{
 12109   match(Set cr (CmpU src zero));
 12111   format %{ "testl  $src, $src\t# unsigned" %}
 12112   opcode(0x85);
 12113   ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
 12114   ins_pipe(ialu_cr_reg_imm);
 12115 %}
 12117 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
 12118 %{
 12119   match(Set cr (CmpP op1 op2));
 12121   format %{ "cmpq    $op1, $op2\t# ptr" %}
 12122   opcode(0x3B); /* Opcode 3B /r */
 12123   ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
 12124   ins_pipe(ialu_cr_reg_reg);
 12125 %}
 12127 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
 12128 %{
 12129   match(Set cr (CmpP op1 (LoadP op2)));
 12131   ins_cost(500); // XXX
 12132   format %{ "cmpq    $op1, $op2\t# ptr" %}
 12133   opcode(0x3B); /* Opcode 3B /r */
 12134   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
 12135   ins_pipe(ialu_cr_reg_mem);
 12136 %}
 12138 // // // Cisc-spilled version of cmpP_rReg
 12139 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
 12140 // //%{
 12141 // //  match(Set cr (CmpP (LoadP op1) op2));
 12142 // //
 12143 // //  format %{ "CMPu   $op1,$op2" %}
 12144 // //  ins_cost(500);
 12145 // //  opcode(0x39);  /* Opcode 39 /r */
 12146 // //  ins_encode( OpcP, reg_mem( op1, op2) );
 12147 // //%}
 12149 // XXX this is generalized by compP_rReg_mem???
 12150 // Compare raw pointer (used in out-of-heap check).
 12151 // Only works because non-oop pointers must be raw pointers
 12152 // and raw pointers have no anti-dependencies.
 12153 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
 12154 %{
 12155   predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
 12156   match(Set cr (CmpP op1 (LoadP op2)));
 12158   format %{ "cmpq    $op1, $op2\t# raw ptr" %}
 12159   opcode(0x3B); /* Opcode 3B /r */
 12160   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
 12161   ins_pipe(ialu_cr_reg_mem);
 12162 %}
 12164 // This will generate a signed flags result. This should be OK since
 12165 // any compare to a zero should be eq/neq.
 12166 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
 12167 %{
 12168   match(Set cr (CmpP src zero));
 12170   format %{ "testq   $src, $src\t# ptr" %}
 12171   opcode(0x85);
 12172   ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
 12173   ins_pipe(ialu_cr_reg_imm);
 12174 %}
 12176 // This will generate a signed flags result. This should be OK since
 12177 // any compare to a zero should be eq/neq.
 12178 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero)
 12179 %{
 12180   predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
 12181   match(Set cr (CmpP (LoadP op) zero));
 12183   ins_cost(500); // XXX
 12184   format %{ "testq   $op, 0xffffffffffffffff\t# ptr" %}
 12185   opcode(0xF7); /* Opcode F7 /0 */
 12186   ins_encode(REX_mem_wide(op),
 12187              OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
 12188   ins_pipe(ialu_cr_reg_imm);
 12189 %}
 12191 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero)
 12192 %{
 12193   predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
 12194   match(Set cr (CmpP (LoadP mem) zero));
 12196   format %{ "cmpq    R12, $mem\t# ptr (R12_heapbase==0)" %}
 12197   ins_encode %{
 12198     __ cmpq(r12, $mem$$Address);
 12199   %}
 12200   ins_pipe(ialu_cr_reg_mem);
 12201 %}
 12203 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
 12204 %{
 12205   match(Set cr (CmpN op1 op2));
 12207   format %{ "cmpl    $op1, $op2\t# compressed ptr" %}
 12208   ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %}
 12209   ins_pipe(ialu_cr_reg_reg);
 12210 %}
 12212 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
 12213 %{
 12214   match(Set cr (CmpN src (LoadN mem)));
 12216   format %{ "cmpl    $src, $mem\t# compressed ptr" %}
 12217   ins_encode %{
 12218     __ cmpl($src$$Register, $mem$$Address);
 12219   %}
 12220   ins_pipe(ialu_cr_reg_mem);
 12221 %}
 12223 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{
 12224   match(Set cr (CmpN op1 op2));
 12226   format %{ "cmpl    $op1, $op2\t# compressed ptr" %}
 12227   ins_encode %{
 12228     __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant);
 12229   %}
 12230   ins_pipe(ialu_cr_reg_imm);
 12231 %}
 12233 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src)
 12234 %{
 12235   match(Set cr (CmpN src (LoadN mem)));
 12237   format %{ "cmpl    $mem, $src\t# compressed ptr" %}
 12238   ins_encode %{
 12239     __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant);
 12240   %}
 12241   ins_pipe(ialu_cr_reg_mem);
 12242 %}
 12244 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
 12245   match(Set cr (CmpN src zero));
 12247   format %{ "testl   $src, $src\t# compressed ptr" %}
 12248   ins_encode %{ __ testl($src$$Register, $src$$Register); %}
 12249   ins_pipe(ialu_cr_reg_imm);
 12250 %}
 12252 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero)
 12253 %{
 12254   predicate(Universe::narrow_oop_base() != NULL);
 12255   match(Set cr (CmpN (LoadN mem) zero));
 12257   ins_cost(500); // XXX
 12258   format %{ "testl   $mem, 0xffffffff\t# compressed ptr" %}
 12259   ins_encode %{
 12260     __ cmpl($mem$$Address, (int)0xFFFFFFFF);
 12261   %}
 12262   ins_pipe(ialu_cr_reg_mem);
 12263 %}
 12265 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero)
 12266 %{
 12267   predicate(Universe::narrow_oop_base() == NULL);
 12268   match(Set cr (CmpN (LoadN mem) zero));
 12270   format %{ "cmpl    R12, $mem\t# compressed ptr (R12_heapbase==0)" %}
 12271   ins_encode %{
 12272     __ cmpl(r12, $mem$$Address);
 12273   %}
 12274   ins_pipe(ialu_cr_reg_mem);
 12275 %}
 12277 // Yanked all unsigned pointer compare operations.
 12278 // Pointer compares are done with CmpP which is already unsigned.
 12280 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
 12281 %{
 12282   match(Set cr (CmpL op1 op2));
 12284   format %{ "cmpq    $op1, $op2" %}
 12285   opcode(0x3B);  /* Opcode 3B /r */
 12286   ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
 12287   ins_pipe(ialu_cr_reg_reg);
 12288 %}
 12290 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
 12291 %{
 12292   match(Set cr (CmpL op1 op2));
 12294   format %{ "cmpq    $op1, $op2" %}
 12295   opcode(0x81, 0x07); /* Opcode 81 /7 */
 12296   ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
 12297   ins_pipe(ialu_cr_reg_imm);
 12298 %}
 12300 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
 12301 %{
 12302   match(Set cr (CmpL op1 (LoadL op2)));
 12304   format %{ "cmpq    $op1, $op2" %}
 12305   opcode(0x3B); /* Opcode 3B /r */
 12306   ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
 12307   ins_pipe(ialu_cr_reg_mem);
 12308 %}
 12310 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
 12311 %{
 12312   match(Set cr (CmpL src zero));
 12314   format %{ "testq   $src, $src" %}
 12315   opcode(0x85);
 12316   ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
 12317   ins_pipe(ialu_cr_reg_imm);
 12318 %}
 12320 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
 12321 %{
 12322   match(Set cr (CmpL (AndL src con) zero));
 12324   format %{ "testq   $src, $con\t# long" %}
 12325   opcode(0xF7, 0x00);
 12326   ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
 12327   ins_pipe(ialu_cr_reg_imm);
 12328 %}
 12330 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
 12331 %{
 12332   match(Set cr (CmpL (AndL src (LoadL mem)) zero));
 12334   format %{ "testq   $src, $mem" %}
 12335   opcode(0x85);
 12336   ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
 12337   ins_pipe(ialu_cr_reg_mem);
 12338 %}
 12340 // Manifest a CmpL result in an integer register.  Very painful.
 12341 // This is the test to avoid.
 12342 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
 12343 %{
 12344   match(Set dst (CmpL3 src1 src2));
 12345   effect(KILL flags);
 12347   ins_cost(275); // XXX
 12348   format %{ "cmpq    $src1, $src2\t# CmpL3\n\t"
 12349             "movl    $dst, -1\n\t"
 12350             "jl,s    done\n\t"
 12351             "setne   $dst\n\t"
 12352             "movzbl  $dst, $dst\n\t"
 12353     "done:" %}
 12354   ins_encode(cmpl3_flag(src1, src2, dst));
 12355   ins_pipe(pipe_slow);
 12356 %}
 12358 //----------Max and Min--------------------------------------------------------
 12359 // Min Instructions
 12361 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
 12362 %{
 12363   effect(USE_DEF dst, USE src, USE cr);
 12365   format %{ "cmovlgt $dst, $src\t# min" %}
 12366   opcode(0x0F, 0x4F);
 12367   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
 12368   ins_pipe(pipe_cmov_reg);
 12369 %}
 12372 instruct minI_rReg(rRegI dst, rRegI src)
 12373 %{
 12374   match(Set dst (MinI dst src));
 12376   ins_cost(200);
 12377   expand %{
 12378     rFlagsReg cr;
 12379     compI_rReg(cr, dst, src);
 12380     cmovI_reg_g(dst, src, cr);
 12381   %}
 12382 %}
 12384 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
 12385 %{
 12386   effect(USE_DEF dst, USE src, USE cr);
 12388   format %{ "cmovllt $dst, $src\t# max" %}
 12389   opcode(0x0F, 0x4C);
 12390   ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
 12391   ins_pipe(pipe_cmov_reg);
 12392 %}
 12395 instruct maxI_rReg(rRegI dst, rRegI src)
 12396 %{
 12397   match(Set dst (MaxI dst src));
 12399   ins_cost(200);
 12400   expand %{
 12401     rFlagsReg cr;
 12402     compI_rReg(cr, dst, src);
 12403     cmovI_reg_l(dst, src, cr);
 12404   %}
 12405 %}
 12407 // ============================================================================
 12408 // Branch Instructions
 12410 // Jump Direct - Label defines a relative address from JMP+1
 12411 instruct jmpDir(label labl)
 12412 %{
 12413   match(Goto);
 12414   effect(USE labl);
 12416   ins_cost(300);
 12417   format %{ "jmp     $labl" %}
 12418   size(5);
 12419   opcode(0xE9);
 12420   ins_encode(OpcP, Lbl(labl));
 12421   ins_pipe(pipe_jmp);
 12422   ins_pc_relative(1);
 12423 %}
 12425 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12426 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
 12427 %{
 12428   match(If cop cr);
 12429   effect(USE labl);
 12431   ins_cost(300);
 12432   format %{ "j$cop     $labl" %}
 12433   size(6);
 12434   opcode(0x0F, 0x80);
 12435   ins_encode(Jcc(cop, labl));
 12436   ins_pipe(pipe_jcc);
 12437   ins_pc_relative(1);
 12438 %}
 12440 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12441 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
 12442 %{
 12443   match(CountedLoopEnd cop cr);
 12444   effect(USE labl);
 12446   ins_cost(300);
 12447   format %{ "j$cop     $labl\t# loop end" %}
 12448   size(6);
 12449   opcode(0x0F, 0x80);
 12450   ins_encode(Jcc(cop, labl));
 12451   ins_pipe(pipe_jcc);
 12452   ins_pc_relative(1);
 12453 %}
 12455 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12456 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
 12457   match(CountedLoopEnd cop cmp);
 12458   effect(USE labl);
 12460   ins_cost(300);
 12461   format %{ "j$cop,u   $labl\t# loop end" %}
 12462   size(6);
 12463   opcode(0x0F, 0x80);
 12464   ins_encode(Jcc(cop, labl));
 12465   ins_pipe(pipe_jcc);
 12466   ins_pc_relative(1);
 12467 %}
 12469 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
 12470   match(CountedLoopEnd cop cmp);
 12471   effect(USE labl);
 12473   ins_cost(200);
 12474   format %{ "j$cop,u   $labl\t# loop end" %}
 12475   size(6);
 12476   opcode(0x0F, 0x80);
 12477   ins_encode(Jcc(cop, labl));
 12478   ins_pipe(pipe_jcc);
 12479   ins_pc_relative(1);
 12480 %}
 12482 // Jump Direct Conditional - using unsigned comparison
 12483 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
 12484   match(If cop cmp);
 12485   effect(USE labl);
 12487   ins_cost(300);
 12488   format %{ "j$cop,u  $labl" %}
 12489   size(6);
 12490   opcode(0x0F, 0x80);
 12491   ins_encode(Jcc(cop, labl));
 12492   ins_pipe(pipe_jcc);
 12493   ins_pc_relative(1);
 12494 %}
 12496 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
 12497   match(If cop cmp);
 12498   effect(USE labl);
 12500   ins_cost(200);
 12501   format %{ "j$cop,u  $labl" %}
 12502   size(6);
 12503   opcode(0x0F, 0x80);
 12504   ins_encode(Jcc(cop, labl));
 12505   ins_pipe(pipe_jcc);
 12506   ins_pc_relative(1);
 12507 %}
 12509 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
 12510   match(If cop cmp);
 12511   effect(USE labl);
 12513   ins_cost(200);
 12514   format %{ $$template
 12515     if ($cop$$cmpcode == Assembler::notEqual) {
 12516       $$emit$$"jp,u   $labl\n\t"
 12517       $$emit$$"j$cop,u   $labl"
 12518     } else {
 12519       $$emit$$"jp,u   done\n\t"
 12520       $$emit$$"j$cop,u   $labl\n\t"
 12521       $$emit$$"done:"
 12523   %}
 12524   size(12);
 12525   opcode(0x0F, 0x80);
 12526   ins_encode %{
 12527     Label* l = $labl$$label;
 12528     $$$emit8$primary;
 12529     emit_cc(cbuf, $secondary, Assembler::parity);
 12530     int parity_disp = -1;
 12531     if ($cop$$cmpcode == Assembler::notEqual) {
 12532        // the two jumps 6 bytes apart so the jump distances are too
 12533        parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
 12534     } else if ($cop$$cmpcode == Assembler::equal) {
 12535        parity_disp = 6;
 12536     } else {
 12537        ShouldNotReachHere();
 12539     emit_d32(cbuf, parity_disp);
 12540     $$$emit8$primary;
 12541     emit_cc(cbuf, $secondary, $cop$$cmpcode);
 12542     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
 12543     emit_d32(cbuf, disp);
 12544   %}
 12545   ins_pipe(pipe_jcc);
 12546   ins_pc_relative(1);
 12547 %}
 12549 // ============================================================================
 12550 // The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary
 12551 // superklass array for an instance of the superklass.  Set a hidden
 12552 // internal cache on a hit (cache is checked with exposed code in
 12553 // gen_subtype_check()).  Return NZ for a miss or zero for a hit.  The
 12554 // encoding ALSO sets flags.
 12556 instruct partialSubtypeCheck(rdi_RegP result,
 12557                              rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
 12558                              rFlagsReg cr)
 12559 %{
 12560   match(Set result (PartialSubtypeCheck sub super));
 12561   effect(KILL rcx, KILL cr);
 12563   ins_cost(1100);  // slightly larger than the next version
 12564   format %{ "movq    rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
 12565             "movl    rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
 12566             "addq    rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
 12567             "repne   scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
 12568             "jne,s   miss\t\t# Missed: rdi not-zero\n\t"
 12569             "movq    [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
 12570             "xorq    $result, $result\t\t Hit: rdi zero\n\t"
 12571     "miss:\t" %}
 12573   opcode(0x1); // Force a XOR of RDI
 12574   ins_encode(enc_PartialSubtypeCheck());
 12575   ins_pipe(pipe_slow);
 12576 %}
 12578 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
 12579                                      rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
 12580                                      immP0 zero,
 12581                                      rdi_RegP result)
 12582 %{
 12583   match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
 12584   effect(KILL rcx, KILL result);
 12586   ins_cost(1000);
 12587   format %{ "movq    rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
 12588             "movl    rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
 12589             "addq    rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
 12590             "repne   scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
 12591             "jne,s   miss\t\t# Missed: flags nz\n\t"
 12592             "movq    [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
 12593     "miss:\t" %}
 12595   opcode(0x0); // No need to XOR RDI
 12596   ins_encode(enc_PartialSubtypeCheck());
 12597   ins_pipe(pipe_slow);
 12598 %}
 12600 // ============================================================================
 12601 // Branch Instructions -- short offset versions
 12602 //
 12603 // These instructions are used to replace jumps of a long offset (the default
 12604 // match) with jumps of a shorter offset.  These instructions are all tagged
 12605 // with the ins_short_branch attribute, which causes the ADLC to suppress the
 12606 // match rules in general matching.  Instead, the ADLC generates a conversion
 12607 // method in the MachNode which can be used to do in-place replacement of the
 12608 // long variant with the shorter variant.  The compiler will determine if a
 12609 // branch can be taken by the is_short_branch_offset() predicate in the machine
 12610 // specific code section of the file.
 12612 // Jump Direct - Label defines a relative address from JMP+1
 12613 instruct jmpDir_short(label labl) %{
 12614   match(Goto);
 12615   effect(USE labl);
 12617   ins_cost(300);
 12618   format %{ "jmp,s   $labl" %}
 12619   size(2);
 12620   opcode(0xEB);
 12621   ins_encode(OpcP, LblShort(labl));
 12622   ins_pipe(pipe_jmp);
 12623   ins_pc_relative(1);
 12624   ins_short_branch(1);
 12625 %}
 12627 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12628 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
 12629   match(If cop cr);
 12630   effect(USE labl);
 12632   ins_cost(300);
 12633   format %{ "j$cop,s   $labl" %}
 12634   size(2);
 12635   opcode(0x70);
 12636   ins_encode(JccShort(cop, labl));
 12637   ins_pipe(pipe_jcc);
 12638   ins_pc_relative(1);
 12639   ins_short_branch(1);
 12640 %}
 12642 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12643 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
 12644   match(CountedLoopEnd cop cr);
 12645   effect(USE labl);
 12647   ins_cost(300);
 12648   format %{ "j$cop,s   $labl\t# loop end" %}
 12649   size(2);
 12650   opcode(0x70);
 12651   ins_encode(JccShort(cop, labl));
 12652   ins_pipe(pipe_jcc);
 12653   ins_pc_relative(1);
 12654   ins_short_branch(1);
 12655 %}
 12657 // Jump Direct Conditional - Label defines a relative address from Jcc+1
 12658 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
 12659   match(CountedLoopEnd cop cmp);
 12660   effect(USE labl);
 12662   ins_cost(300);
 12663   format %{ "j$cop,us  $labl\t# loop end" %}
 12664   size(2);
 12665   opcode(0x70);
 12666   ins_encode(JccShort(cop, labl));
 12667   ins_pipe(pipe_jcc);
 12668   ins_pc_relative(1);
 12669   ins_short_branch(1);
 12670 %}
 12672 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
 12673   match(CountedLoopEnd cop cmp);
 12674   effect(USE labl);
 12676   ins_cost(300);
 12677   format %{ "j$cop,us  $labl\t# loop end" %}
 12678   size(2);
 12679   opcode(0x70);
 12680   ins_encode(JccShort(cop, labl));
 12681   ins_pipe(pipe_jcc);
 12682   ins_pc_relative(1);
 12683   ins_short_branch(1);
 12684 %}
 12686 // Jump Direct Conditional - using unsigned comparison
 12687 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
 12688   match(If cop cmp);
 12689   effect(USE labl);
 12691   ins_cost(300);
 12692   format %{ "j$cop,us  $labl" %}
 12693   size(2);
 12694   opcode(0x70);
 12695   ins_encode(JccShort(cop, labl));
 12696   ins_pipe(pipe_jcc);
 12697   ins_pc_relative(1);
 12698   ins_short_branch(1);
 12699 %}
 12701 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
 12702   match(If cop cmp);
 12703   effect(USE labl);
 12705   ins_cost(300);
 12706   format %{ "j$cop,us  $labl" %}
 12707   size(2);
 12708   opcode(0x70);
 12709   ins_encode(JccShort(cop, labl));
 12710   ins_pipe(pipe_jcc);
 12711   ins_pc_relative(1);
 12712   ins_short_branch(1);
 12713 %}
 12715 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
 12716   match(If cop cmp);
 12717   effect(USE labl);
 12719   ins_cost(300);
 12720   format %{ $$template
 12721     if ($cop$$cmpcode == Assembler::notEqual) {
 12722       $$emit$$"jp,u,s   $labl\n\t"
 12723       $$emit$$"j$cop,u,s   $labl"
 12724     } else {
 12725       $$emit$$"jp,u,s   done\n\t"
 12726       $$emit$$"j$cop,u,s  $labl\n\t"
 12727       $$emit$$"done:"
 12729   %}
 12730   size(4);
 12731   opcode(0x70);
 12732   ins_encode %{
 12733     Label* l = $labl$$label;
 12734     emit_cc(cbuf, $primary, Assembler::parity);
 12735     int parity_disp = -1;
 12736     if ($cop$$cmpcode == Assembler::notEqual) {
 12737       parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
 12738     } else if ($cop$$cmpcode == Assembler::equal) {
 12739       parity_disp = 2;
 12740     } else {
 12741       ShouldNotReachHere();
 12743     emit_d8(cbuf, parity_disp);
 12744     emit_cc(cbuf, $primary, $cop$$cmpcode);
 12745     int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
 12746     emit_d8(cbuf, disp);
 12747     assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
 12748     assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
 12749   %}
 12750   ins_pipe(pipe_jcc);
 12751   ins_pc_relative(1);
 12752   ins_short_branch(1);
 12753 %}
 12755 // ============================================================================
 12756 // inlined locking and unlocking
 12758 instruct cmpFastLock(rFlagsReg cr,
 12759                      rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
 12760 %{
 12761   match(Set cr (FastLock object box));
 12762   effect(TEMP tmp, TEMP scr);
 12764   ins_cost(300);
 12765   format %{ "fastlock $object,$box,$tmp,$scr" %}
 12766   ins_encode(Fast_Lock(object, box, tmp, scr));
 12767   ins_pipe(pipe_slow);
 12768   ins_pc_relative(1);
 12769 %}
 12771 instruct cmpFastUnlock(rFlagsReg cr,
 12772                        rRegP object, rax_RegP box, rRegP tmp)
 12773 %{
 12774   match(Set cr (FastUnlock object box));
 12775   effect(TEMP tmp);
 12777   ins_cost(300);
 12778   format %{ "fastunlock $object, $box, $tmp" %}
 12779   ins_encode(Fast_Unlock(object, box, tmp));
 12780   ins_pipe(pipe_slow);
 12781   ins_pc_relative(1);
 12782 %}
 12785 // ============================================================================
 12786 // Safepoint Instructions
 12787 instruct safePoint_poll(rFlagsReg cr)
 12788 %{
 12789   match(SafePoint);
 12790   effect(KILL cr);
 12792   format %{ "testl   rax, [rip + #offset_to_poll_page]\t"
 12793             "# Safepoint: poll for GC" %}
 12794   size(6); // Opcode + ModRM + Disp32 == 6 bytes
 12795   ins_cost(125);
 12796   ins_encode(enc_safepoint_poll);
 12797   ins_pipe(ialu_reg_mem);
 12798 %}
 12800 // ============================================================================
 12801 // Procedure Call/Return Instructions
 12802 // Call Java Static Instruction
 12803 // Note: If this code changes, the corresponding ret_addr_offset() and
 12804 //       compute_padding() functions will have to be adjusted.
 12805 instruct CallStaticJavaDirect(method meth)
 12806 %{
 12807   match(CallStaticJava);
 12808   effect(USE meth);
 12810   ins_cost(300);
 12811   format %{ "call,static " %}
 12812   opcode(0xE8); /* E8 cd */
 12813   ins_encode(Java_Static_Call(meth), call_epilog);
 12814   ins_pipe(pipe_slow);
 12815   ins_pc_relative(1);
 12816   ins_alignment(4);
 12817 %}
 12819 // Call Java Dynamic Instruction
 12820 // Note: If this code changes, the corresponding ret_addr_offset() and
 12821 //       compute_padding() functions will have to be adjusted.
 12822 instruct CallDynamicJavaDirect(method meth)
 12823 %{
 12824   match(CallDynamicJava);
 12825   effect(USE meth);
 12827   ins_cost(300);
 12828   format %{ "movq    rax, #Universe::non_oop_word()\n\t"
 12829             "call,dynamic " %}
 12830   opcode(0xE8); /* E8 cd */
 12831   ins_encode(Java_Dynamic_Call(meth), call_epilog);
 12832   ins_pipe(pipe_slow);
 12833   ins_pc_relative(1);
 12834   ins_alignment(4);
 12835 %}
 12837 // Call Runtime Instruction
 12838 instruct CallRuntimeDirect(method meth)
 12839 %{
 12840   match(CallRuntime);
 12841   effect(USE meth);
 12843   ins_cost(300);
 12844   format %{ "call,runtime " %}
 12845   opcode(0xE8); /* E8 cd */
 12846   ins_encode(Java_To_Runtime(meth));
 12847   ins_pipe(pipe_slow);
 12848   ins_pc_relative(1);
 12849 %}
 12851 // Call runtime without safepoint
 12852 instruct CallLeafDirect(method meth)
 12853 %{
 12854   match(CallLeaf);
 12855   effect(USE meth);
 12857   ins_cost(300);
 12858   format %{ "call_leaf,runtime " %}
 12859   opcode(0xE8); /* E8 cd */
 12860   ins_encode(Java_To_Runtime(meth));
 12861   ins_pipe(pipe_slow);
 12862   ins_pc_relative(1);
 12863 %}
 12865 // Call runtime without safepoint
 12866 instruct CallLeafNoFPDirect(method meth)
 12867 %{
 12868   match(CallLeafNoFP);
 12869   effect(USE meth);
 12871   ins_cost(300);
 12872   format %{ "call_leaf_nofp,runtime " %}
 12873   opcode(0xE8); /* E8 cd */
 12874   ins_encode(Java_To_Runtime(meth));
 12875   ins_pipe(pipe_slow);
 12876   ins_pc_relative(1);
 12877 %}
 12879 // Return Instruction
 12880 // Remove the return address & jump to it.
 12881 // Notice: We always emit a nop after a ret to make sure there is room
 12882 // for safepoint patching
 12883 instruct Ret()
 12884 %{
 12885   match(Return);
 12887   format %{ "ret" %}
 12888   opcode(0xC3);
 12889   ins_encode(OpcP);
 12890   ins_pipe(pipe_jmp);
 12891 %}
 12893 // Tail Call; Jump from runtime stub to Java code.
 12894 // Also known as an 'interprocedural jump'.
 12895 // Target of jump will eventually return to caller.
 12896 // TailJump below removes the return address.
 12897 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
 12898 %{
 12899   match(TailCall jump_target method_oop);
 12901   ins_cost(300);
 12902   format %{ "jmp     $jump_target\t# rbx holds method oop" %}
 12903   opcode(0xFF, 0x4); /* Opcode FF /4 */
 12904   ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
 12905   ins_pipe(pipe_jmp);
 12906 %}
 12908 // Tail Jump; remove the return address; jump to target.
 12909 // TailCall above leaves the return address around.
 12910 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
 12911 %{
 12912   match(TailJump jump_target ex_oop);
 12914   ins_cost(300);
 12915   format %{ "popq    rdx\t# pop return address\n\t"
 12916             "jmp     $jump_target" %}
 12917   opcode(0xFF, 0x4); /* Opcode FF /4 */
 12918   ins_encode(Opcode(0x5a), // popq rdx
 12919              REX_reg(jump_target), OpcP, reg_opc(jump_target));
 12920   ins_pipe(pipe_jmp);
 12921 %}
 12923 // Create exception oop: created by stack-crawling runtime code.
 12924 // Created exception is now available to this handler, and is setup
 12925 // just prior to jumping to this handler.  No code emitted.
 12926 instruct CreateException(rax_RegP ex_oop)
 12927 %{
 12928   match(Set ex_oop (CreateEx));
 12930   size(0);
 12931   // use the following format syntax
 12932   format %{ "# exception oop is in rax; no code emitted" %}
 12933   ins_encode();
 12934   ins_pipe(empty);
 12935 %}
 12937 // Rethrow exception:
 12938 // The exception oop will come in the first argument position.
 12939 // Then JUMP (not call) to the rethrow stub code.
 12940 instruct RethrowException()
 12941 %{
 12942   match(Rethrow);
 12944   // use the following format syntax
 12945   format %{ "jmp     rethrow_stub" %}
 12946   ins_encode(enc_rethrow);
 12947   ins_pipe(pipe_jmp);
 12948 %}
 12951 //----------PEEPHOLE RULES-----------------------------------------------------
 12952 // These must follow all instruction definitions as they use the names
 12953 // defined in the instructions definitions.
 12954 //
 12955 // peepmatch ( root_instr_name [preceding_instruction]* );
 12956 //
 12957 // peepconstraint %{
 12958 // (instruction_number.operand_name relational_op instruction_number.operand_name
 12959 //  [, ...] );
 12960 // // instruction numbers are zero-based using left to right order in peepmatch
 12961 //
 12962 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
 12963 // // provide an instruction_number.operand_name for each operand that appears
 12964 // // in the replacement instruction's match rule
 12965 //
 12966 // ---------VM FLAGS---------------------------------------------------------
 12967 //
 12968 // All peephole optimizations can be turned off using -XX:-OptoPeephole
 12969 //
 12970 // Each peephole rule is given an identifying number starting with zero and
 12971 // increasing by one in the order seen by the parser.  An individual peephole
 12972 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
 12973 // on the command-line.
 12974 //
 12975 // ---------CURRENT LIMITATIONS----------------------------------------------
 12976 //
 12977 // Only match adjacent instructions in same basic block
 12978 // Only equality constraints
 12979 // Only constraints between operands, not (0.dest_reg == RAX_enc)
 12980 // Only one replacement instruction
 12981 //
 12982 // ---------EXAMPLE----------------------------------------------------------
 12983 //
 12984 // // pertinent parts of existing instructions in architecture description
 12985 // instruct movI(rRegI dst, rRegI src)
 12986 // %{
 12987 //   match(Set dst (CopyI src));
 12988 // %}
 12989 //
 12990 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
 12991 // %{
 12992 //   match(Set dst (AddI dst src));
 12993 //   effect(KILL cr);
 12994 // %}
 12995 //
 12996 // // Change (inc mov) to lea
 12997 // peephole %{
 12998 //   // increment preceeded by register-register move
 12999 //   peepmatch ( incI_rReg movI );
 13000 //   // require that the destination register of the increment
 13001 //   // match the destination register of the move
 13002 //   peepconstraint ( 0.dst == 1.dst );
 13003 //   // construct a replacement instruction that sets
 13004 //   // the destination to ( move's source register + one )
 13005 //   peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
 13006 // %}
 13007 //
 13009 // Implementation no longer uses movX instructions since
 13010 // machine-independent system no longer uses CopyX nodes.
 13011 //
 13012 // peephole
 13013 // %{
 13014 //   peepmatch (incI_rReg movI);
 13015 //   peepconstraint (0.dst == 1.dst);
 13016 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
 13017 // %}
 13019 // peephole
 13020 // %{
 13021 //   peepmatch (decI_rReg movI);
 13022 //   peepconstraint (0.dst == 1.dst);
 13023 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
 13024 // %}
 13026 // peephole
 13027 // %{
 13028 //   peepmatch (addI_rReg_imm movI);
 13029 //   peepconstraint (0.dst == 1.dst);
 13030 //   peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
 13031 // %}
 13033 // peephole
 13034 // %{
 13035 //   peepmatch (incL_rReg movL);
 13036 //   peepconstraint (0.dst == 1.dst);
 13037 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
 13038 // %}
 13040 // peephole
 13041 // %{
 13042 //   peepmatch (decL_rReg movL);
 13043 //   peepconstraint (0.dst == 1.dst);
 13044 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
 13045 // %}
 13047 // peephole
 13048 // %{
 13049 //   peepmatch (addL_rReg_imm movL);
 13050 //   peepconstraint (0.dst == 1.dst);
 13051 //   peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
 13052 // %}
 13054 // peephole
 13055 // %{
 13056 //   peepmatch (addP_rReg_imm movP);
 13057 //   peepconstraint (0.dst == 1.dst);
 13058 //   peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
 13059 // %}
 13061 // // Change load of spilled value to only a spill
 13062 // instruct storeI(memory mem, rRegI src)
 13063 // %{
 13064 //   match(Set mem (StoreI mem src));
 13065 // %}
 13066 //
 13067 // instruct loadI(rRegI dst, memory mem)
 13068 // %{
 13069 //   match(Set dst (LoadI mem));
 13070 // %}
 13071 //
 13073 peephole
 13074 %{
 13075   peepmatch (loadI storeI);
 13076   peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
 13077   peepreplace (storeI(1.mem 1.mem 1.src));
 13078 %}
 13080 peephole
 13081 %{
 13082   peepmatch (loadL storeL);
 13083   peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
 13084   peepreplace (storeL(1.mem 1.mem 1.src));
 13085 %}
 13087 //----------SMARTSPILL RULES---------------------------------------------------
 13088 // These must follow all instruction definitions as they use the names
 13089 // defined in the instructions definitions.

mercurial