1 //
   2 // Copyright (c) 2011, 2015, Oracle and/or its affiliates. All rights reserved.
   3 // Copyright 2012, 2015 SAP AG. All rights reserved.
   4 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5 //
   6 // This code is free software; you can redistribute it and/or modify it
   7 // under the terms of the GNU General Public License version 2 only, as
   8 // published by the Free Software Foundation.
   9 //
  10 // This code is distributed in the hope that it will be useful, but WITHOUT
  11 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12 // FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13 // version 2 for more details (a copy is included in the LICENSE file that
  14 // accompanied this code).
  15 //
  16 // You should have received a copy of the GNU General Public License version
  17 // 2 along with this work; if not, write to the Free Software Foundation,
  18 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19 //
  20 // Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21 // or visit www.oracle.com if you need additional information or have any
  22 // questions.
  23 //
  24 //
  25 
  26 //
  27 // PPC64 Architecture Description File
  28 //
  29 
  30 //----------REGISTER DEFINITION BLOCK------------------------------------------
  31 // This information is used by the matcher and the register allocator to
  32 // describe individual registers and classes of registers within the target
  33 // architecture.
  34 register %{
  35 //----------Architecture Description Register Definitions----------------------
  36 // General Registers
  37 // "reg_def"  name (register save type, C convention save type,
  38 //                  ideal register type, encoding);
  39 //
  40 // Register Save Types:
  41 //
  42 //   NS  = No-Save:     The register allocator assumes that these registers
  43 //                      can be used without saving upon entry to the method, &
  44 //                      that they do not need to be saved at call sites.
  45 //
  46 //   SOC = Save-On-Call: The register allocator assumes that these registers
  47 //                      can be used without saving upon entry to the method,
  48 //                      but that they must be saved at call sites.
  49 //                      These are called "volatiles" on ppc.
  50 //
  51 //   SOE = Save-On-Entry: The register allocator assumes that these registers
  52 //                      must be saved before using them upon entry to the
  53 //                      method, but they do not need to be saved at call
  54 //                      sites.
  55 //                      These are called "nonvolatiles" on ppc.
  56 //
  57 //   AS  = Always-Save:   The register allocator assumes that these registers
  58 //                      must be saved before using them upon entry to the
  59 //                      method, & that they must be saved at call sites.
  60 //
  61 // Ideal Register Type is used to determine how to save & restore a
  62 // register. Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
  63 // spilled with LoadP/StoreP. If the register supports both, use Op_RegI.
  64 //
  65 // The encoding number is the actual bit-pattern placed into the opcodes.
  66 //
  67 // PPC64 register definitions, based on the 64-bit PowerPC ELF ABI
  68 // Supplement Version 1.7 as of 2003-10-29.
  69 //
  70 // For each 64-bit register we must define two registers: the register
  71 // itself, e.g. R3, and a corresponding virtual other (32-bit-)'half',
  72 // e.g. R3_H, which is needed by the allocator, but is not used
  73 // for stores, loads, etc.
  74 
  75 // ----------------------------
  76 // Integer/Long Registers
  77 // ----------------------------
  78 
  79   // PPC64 has 32 64-bit integer registers.
  80 
  81   // types: v = volatile, nv = non-volatile, s = system
  82   reg_def R0   ( SOC, SOC, Op_RegI,  0, R0->as_VMReg()         );  // v   used in prologs
  83   reg_def R0_H ( SOC, SOC, Op_RegI, 99, R0->as_VMReg()->next() );
  84   reg_def R1   ( NS,  NS,  Op_RegI,  1, R1->as_VMReg()         );  // s   SP
  85   reg_def R1_H ( NS,  NS,  Op_RegI, 99, R1->as_VMReg()->next() );
  86   reg_def R2   ( SOC, SOC, Op_RegI,  2, R2->as_VMReg()         );  // v   TOC
  87   reg_def R2_H ( SOC, SOC, Op_RegI, 99, R2->as_VMReg()->next() );
  88   reg_def R3   ( SOC, SOC, Op_RegI,  3, R3->as_VMReg()         );  // v   iarg1 & iret
  89   reg_def R3_H ( SOC, SOC, Op_RegI, 99, R3->as_VMReg()->next() );
  90   reg_def R4   ( SOC, SOC, Op_RegI,  4, R4->as_VMReg()         );  //     iarg2
  91   reg_def R4_H ( SOC, SOC, Op_RegI, 99, R4->as_VMReg()->next() );
  92   reg_def R5   ( SOC, SOC, Op_RegI,  5, R5->as_VMReg()         );  // v   iarg3
  93   reg_def R5_H ( SOC, SOC, Op_RegI, 99, R5->as_VMReg()->next() );
  94   reg_def R6   ( SOC, SOC, Op_RegI,  6, R6->as_VMReg()         );  // v   iarg4
  95   reg_def R6_H ( SOC, SOC, Op_RegI, 99, R6->as_VMReg()->next() );
  96   reg_def R7   ( SOC, SOC, Op_RegI,  7, R7->as_VMReg()         );  // v   iarg5
  97   reg_def R7_H ( SOC, SOC, Op_RegI, 99, R7->as_VMReg()->next() );
  98   reg_def R8   ( SOC, SOC, Op_RegI,  8, R8->as_VMReg()         );  // v   iarg6
  99   reg_def R8_H ( SOC, SOC, Op_RegI, 99, R8->as_VMReg()->next() );
 100   reg_def R9   ( SOC, SOC, Op_RegI,  9, R9->as_VMReg()         );  // v   iarg7
 101   reg_def R9_H ( SOC, SOC, Op_RegI, 99, R9->as_VMReg()->next() );
 102   reg_def R10  ( SOC, SOC, Op_RegI, 10, R10->as_VMReg()        );  // v   iarg8
 103   reg_def R10_H( SOC, SOC, Op_RegI, 99, R10->as_VMReg()->next());
 104   reg_def R11  ( SOC, SOC, Op_RegI, 11, R11->as_VMReg()        );  // v   ENV / scratch
 105   reg_def R11_H( SOC, SOC, Op_RegI, 99, R11->as_VMReg()->next());
 106   reg_def R12  ( SOC, SOC, Op_RegI, 12, R12->as_VMReg()        );  // v   scratch
 107   reg_def R12_H( SOC, SOC, Op_RegI, 99, R12->as_VMReg()->next());
 108   reg_def R13  ( NS,  NS,  Op_RegI, 13, R13->as_VMReg()        );  // s   system thread id
 109   reg_def R13_H( NS,  NS,  Op_RegI, 99, R13->as_VMReg()->next());
 110   reg_def R14  ( SOC, SOE, Op_RegI, 14, R14->as_VMReg()        );  // nv
 111   reg_def R14_H( SOC, SOE, Op_RegI, 99, R14->as_VMReg()->next());
 112   reg_def R15  ( SOC, SOE, Op_RegI, 15, R15->as_VMReg()        );  // nv
 113   reg_def R15_H( SOC, SOE, Op_RegI, 99, R15->as_VMReg()->next());
 114   reg_def R16  ( SOC, SOE, Op_RegI, 16, R16->as_VMReg()        );  // nv
 115   reg_def R16_H( SOC, SOE, Op_RegI, 99, R16->as_VMReg()->next());
 116   reg_def R17  ( SOC, SOE, Op_RegI, 17, R17->as_VMReg()        );  // nv
 117   reg_def R17_H( SOC, SOE, Op_RegI, 99, R17->as_VMReg()->next());
 118   reg_def R18  ( SOC, SOE, Op_RegI, 18, R18->as_VMReg()        );  // nv
 119   reg_def R18_H( SOC, SOE, Op_RegI, 99, R18->as_VMReg()->next());
 120   reg_def R19  ( SOC, SOE, Op_RegI, 19, R19->as_VMReg()        );  // nv
 121   reg_def R19_H( SOC, SOE, Op_RegI, 99, R19->as_VMReg()->next());
 122   reg_def R20  ( SOC, SOE, Op_RegI, 20, R20->as_VMReg()        );  // nv
 123   reg_def R20_H( SOC, SOE, Op_RegI, 99, R20->as_VMReg()->next());
 124   reg_def R21  ( SOC, SOE, Op_RegI, 21, R21->as_VMReg()        );  // nv
 125   reg_def R21_H( SOC, SOE, Op_RegI, 99, R21->as_VMReg()->next());
 126   reg_def R22  ( SOC, SOE, Op_RegI, 22, R22->as_VMReg()        );  // nv
 127   reg_def R22_H( SOC, SOE, Op_RegI, 99, R22->as_VMReg()->next());
 128   reg_def R23  ( SOC, SOE, Op_RegI, 23, R23->as_VMReg()        );  // nv
 129   reg_def R23_H( SOC, SOE, Op_RegI, 99, R23->as_VMReg()->next());
 130   reg_def R24  ( SOC, SOE, Op_RegI, 24, R24->as_VMReg()        );  // nv
 131   reg_def R24_H( SOC, SOE, Op_RegI, 99, R24->as_VMReg()->next());
 132   reg_def R25  ( SOC, SOE, Op_RegI, 25, R25->as_VMReg()        );  // nv
 133   reg_def R25_H( SOC, SOE, Op_RegI, 99, R25->as_VMReg()->next());
 134   reg_def R26  ( SOC, SOE, Op_RegI, 26, R26->as_VMReg()        );  // nv
 135   reg_def R26_H( SOC, SOE, Op_RegI, 99, R26->as_VMReg()->next());
 136   reg_def R27  ( SOC, SOE, Op_RegI, 27, R27->as_VMReg()        );  // nv
 137   reg_def R27_H( SOC, SOE, Op_RegI, 99, R27->as_VMReg()->next());
 138   reg_def R28  ( SOC, SOE, Op_RegI, 28, R28->as_VMReg()        );  // nv
 139   reg_def R28_H( SOC, SOE, Op_RegI, 99, R28->as_VMReg()->next());
 140   reg_def R29  ( SOC, SOE, Op_RegI, 29, R29->as_VMReg()        );  // nv
 141   reg_def R29_H( SOC, SOE, Op_RegI, 99, R29->as_VMReg()->next());
 142   reg_def R30  ( SOC, SOE, Op_RegI, 30, R30->as_VMReg()        );  // nv
 143   reg_def R30_H( SOC, SOE, Op_RegI, 99, R30->as_VMReg()->next());
 144   reg_def R31  ( SOC, SOE, Op_RegI, 31, R31->as_VMReg()        );  // nv
 145   reg_def R31_H( SOC, SOE, Op_RegI, 99, R31->as_VMReg()->next());
 146 
 147 
 148 // ----------------------------
 149 // Float/Double Registers
 150 // ----------------------------
 151 
 152   // Double Registers
 153   // The rules of ADL require that double registers be defined in pairs.
 154   // Each pair must be two 32-bit values, but not necessarily a pair of
 155   // single float registers. In each pair, ADLC-assigned register numbers
 156   // must be adjacent, with the lower number even. Finally, when the
 157   // CPU stores such a register pair to memory, the word associated with
 158   // the lower ADLC-assigned number must be stored to the lower address.
 159 
 160   // PPC64 has 32 64-bit floating-point registers. Each can store a single
 161   // or double precision floating-point value.
 162 
 163   // types: v = volatile, nv = non-volatile, s = system
 164   reg_def F0   ( SOC, SOC, Op_RegF,  0, F0->as_VMReg()         );  // v   scratch
 165   reg_def F0_H ( SOC, SOC, Op_RegF, 99, F0->as_VMReg()->next() );
 166   reg_def F1   ( SOC, SOC, Op_RegF,  1, F1->as_VMReg()         );  // v   farg1 & fret
 167   reg_def F1_H ( SOC, SOC, Op_RegF, 99, F1->as_VMReg()->next() );
 168   reg_def F2   ( SOC, SOC, Op_RegF,  2, F2->as_VMReg()         );  // v   farg2
 169   reg_def F2_H ( SOC, SOC, Op_RegF, 99, F2->as_VMReg()->next() );
 170   reg_def F3   ( SOC, SOC, Op_RegF,  3, F3->as_VMReg()         );  // v   farg3
 171   reg_def F3_H ( SOC, SOC, Op_RegF, 99, F3->as_VMReg()->next() );
 172   reg_def F4   ( SOC, SOC, Op_RegF,  4, F4->as_VMReg()         );  // v   farg4
 173   reg_def F4_H ( SOC, SOC, Op_RegF, 99, F4->as_VMReg()->next() );
 174   reg_def F5   ( SOC, SOC, Op_RegF,  5, F5->as_VMReg()         );  // v   farg5
 175   reg_def F5_H ( SOC, SOC, Op_RegF, 99, F5->as_VMReg()->next() );
 176   reg_def F6   ( SOC, SOC, Op_RegF,  6, F6->as_VMReg()         );  // v   farg6
 177   reg_def F6_H ( SOC, SOC, Op_RegF, 99, F6->as_VMReg()->next() );
 178   reg_def F7   ( SOC, SOC, Op_RegF,  7, F7->as_VMReg()         );  // v   farg7
 179   reg_def F7_H ( SOC, SOC, Op_RegF, 99, F7->as_VMReg()->next() );
 180   reg_def F8   ( SOC, SOC, Op_RegF,  8, F8->as_VMReg()         );  // v   farg8
 181   reg_def F8_H ( SOC, SOC, Op_RegF, 99, F8->as_VMReg()->next() );
 182   reg_def F9   ( SOC, SOC, Op_RegF,  9, F9->as_VMReg()         );  // v   farg9
 183   reg_def F9_H ( SOC, SOC, Op_RegF, 99, F9->as_VMReg()->next() );
 184   reg_def F10  ( SOC, SOC, Op_RegF, 10, F10->as_VMReg()        );  // v   farg10
 185   reg_def F10_H( SOC, SOC, Op_RegF, 99, F10->as_VMReg()->next());
 186   reg_def F11  ( SOC, SOC, Op_RegF, 11, F11->as_VMReg()        );  // v   farg11
 187   reg_def F11_H( SOC, SOC, Op_RegF, 99, F11->as_VMReg()->next());
 188   reg_def F12  ( SOC, SOC, Op_RegF, 12, F12->as_VMReg()        );  // v   farg12
 189   reg_def F12_H( SOC, SOC, Op_RegF, 99, F12->as_VMReg()->next());
 190   reg_def F13  ( SOC, SOC, Op_RegF, 13, F13->as_VMReg()        );  // v   farg13
 191   reg_def F13_H( SOC, SOC, Op_RegF, 99, F13->as_VMReg()->next());
 192   reg_def F14  ( SOC, SOE, Op_RegF, 14, F14->as_VMReg()        );  // nv
 193   reg_def F14_H( SOC, SOE, Op_RegF, 99, F14->as_VMReg()->next());
 194   reg_def F15  ( SOC, SOE, Op_RegF, 15, F15->as_VMReg()        );  // nv
 195   reg_def F15_H( SOC, SOE, Op_RegF, 99, F15->as_VMReg()->next());
 196   reg_def F16  ( SOC, SOE, Op_RegF, 16, F16->as_VMReg()        );  // nv
 197   reg_def F16_H( SOC, SOE, Op_RegF, 99, F16->as_VMReg()->next());
 198   reg_def F17  ( SOC, SOE, Op_RegF, 17, F17->as_VMReg()        );  // nv
 199   reg_def F17_H( SOC, SOE, Op_RegF, 99, F17->as_VMReg()->next());
 200   reg_def F18  ( SOC, SOE, Op_RegF, 18, F18->as_VMReg()        );  // nv
 201   reg_def F18_H( SOC, SOE, Op_RegF, 99, F18->as_VMReg()->next());
 202   reg_def F19  ( SOC, SOE, Op_RegF, 19, F19->as_VMReg()        );  // nv
 203   reg_def F19_H( SOC, SOE, Op_RegF, 99, F19->as_VMReg()->next());
 204   reg_def F20  ( SOC, SOE, Op_RegF, 20, F20->as_VMReg()        );  // nv
 205   reg_def F20_H( SOC, SOE, Op_RegF, 99, F20->as_VMReg()->next());
 206   reg_def F21  ( SOC, SOE, Op_RegF, 21, F21->as_VMReg()        );  // nv
 207   reg_def F21_H( SOC, SOE, Op_RegF, 99, F21->as_VMReg()->next());
 208   reg_def F22  ( SOC, SOE, Op_RegF, 22, F22->as_VMReg()        );  // nv
 209   reg_def F22_H( SOC, SOE, Op_RegF, 99, F22->as_VMReg()->next());
 210   reg_def F23  ( SOC, SOE, Op_RegF, 23, F23->as_VMReg()        );  // nv
 211   reg_def F23_H( SOC, SOE, Op_RegF, 99, F23->as_VMReg()->next());
 212   reg_def F24  ( SOC, SOE, Op_RegF, 24, F24->as_VMReg()        );  // nv
 213   reg_def F24_H( SOC, SOE, Op_RegF, 99, F24->as_VMReg()->next());
 214   reg_def F25  ( SOC, SOE, Op_RegF, 25, F25->as_VMReg()        );  // nv
 215   reg_def F25_H( SOC, SOE, Op_RegF, 99, F25->as_VMReg()->next());
 216   reg_def F26  ( SOC, SOE, Op_RegF, 26, F26->as_VMReg()        );  // nv
 217   reg_def F26_H( SOC, SOE, Op_RegF, 99, F26->as_VMReg()->next());
 218   reg_def F27  ( SOC, SOE, Op_RegF, 27, F27->as_VMReg()        );  // nv
 219   reg_def F27_H( SOC, SOE, Op_RegF, 99, F27->as_VMReg()->next());
 220   reg_def F28  ( SOC, SOE, Op_RegF, 28, F28->as_VMReg()        );  // nv
 221   reg_def F28_H( SOC, SOE, Op_RegF, 99, F28->as_VMReg()->next());
 222   reg_def F29  ( SOC, SOE, Op_RegF, 29, F29->as_VMReg()        );  // nv
 223   reg_def F29_H( SOC, SOE, Op_RegF, 99, F29->as_VMReg()->next());
 224   reg_def F30  ( SOC, SOE, Op_RegF, 30, F30->as_VMReg()        );  // nv
 225   reg_def F30_H( SOC, SOE, Op_RegF, 99, F30->as_VMReg()->next());
 226   reg_def F31  ( SOC, SOE, Op_RegF, 31, F31->as_VMReg()        );  // nv
 227   reg_def F31_H( SOC, SOE, Op_RegF, 99, F31->as_VMReg()->next());
 228 
 229 // ----------------------------
 230 // Special Registers
 231 // ----------------------------
 232 
 233 // Condition Codes Flag Registers
 234 
 235   // PPC64 has 8 condition code "registers" which are all contained
 236   // in the CR register.
 237 
 238   // types: v = volatile, nv = non-volatile, s = system
 239   reg_def CCR0(SOC, SOC, Op_RegFlags, 0, CCR0->as_VMReg());  // v
 240   reg_def CCR1(SOC, SOC, Op_RegFlags, 1, CCR1->as_VMReg());  // v
 241   reg_def CCR2(SOC, SOC, Op_RegFlags, 2, CCR2->as_VMReg());  // nv
 242   reg_def CCR3(SOC, SOC, Op_RegFlags, 3, CCR3->as_VMReg());  // nv
 243   reg_def CCR4(SOC, SOC, Op_RegFlags, 4, CCR4->as_VMReg());  // nv
 244   reg_def CCR5(SOC, SOC, Op_RegFlags, 5, CCR5->as_VMReg());  // v
 245   reg_def CCR6(SOC, SOC, Op_RegFlags, 6, CCR6->as_VMReg());  // v
 246   reg_def CCR7(SOC, SOC, Op_RegFlags, 7, CCR7->as_VMReg());  // v
 247 
 248   // Special registers of PPC64
 249 
 250   reg_def SR_XER(    SOC, SOC, Op_RegP, 0, SR_XER->as_VMReg());     // v
 251   reg_def SR_LR(     SOC, SOC, Op_RegP, 1, SR_LR->as_VMReg());      // v
 252   reg_def SR_CTR(    SOC, SOC, Op_RegP, 2, SR_CTR->as_VMReg());     // v
 253   reg_def SR_VRSAVE( SOC, SOC, Op_RegP, 3, SR_VRSAVE->as_VMReg());  // v
 254   reg_def SR_SPEFSCR(SOC, SOC, Op_RegP, 4, SR_SPEFSCR->as_VMReg()); // v
 255   reg_def SR_PPR(    SOC, SOC, Op_RegP, 5, SR_PPR->as_VMReg());     // v
 256 
 257 
 258 // ----------------------------
 259 // Specify priority of register selection within phases of register
 260 // allocation. Highest priority is first. A useful heuristic is to
 261 // give registers a low priority when they are required by machine
 262 // instructions, like EAX and EDX on I486, and choose no-save registers
 263 // before save-on-call, & save-on-call before save-on-entry. Registers
 264 // which participate in fixed calling sequences should come last.
 265 // Registers which are used as pairs must fall on an even boundary.
 266 
 267 // It's worth about 1% on SPEC geomean to get this right.
 268 
 269 // Chunk0, chunk1, and chunk2 form the MachRegisterNumbers enumeration
 270 // in adGlobals_ppc.hpp which defines the <register>_num values, e.g.
 271 // R3_num. Therefore, R3_num may not be (and in reality is not)
 272 // the same as R3->encoding()! Furthermore, we cannot make any
 273 // assumptions on ordering, e.g. R3_num may be less than R2_num.
 274 // Additionally, the function
 275 //   static enum RC rc_class(OptoReg::Name reg )
 276 // maps a given <register>_num value to its chunk type (except for flags)
 277 // and its current implementation relies on chunk0 and chunk1 having a
 278 // size of 64 each.
 279 
 280 // If you change this allocation class, please have a look at the
 281 // default values for the parameters RoundRobinIntegerRegIntervalStart
 282 // and RoundRobinFloatRegIntervalStart
 283 
 284 alloc_class chunk0 (
 285   // Chunk0 contains *all* 64 integer registers halves.
 286 
 287   // "non-volatile" registers
 288   R14, R14_H,
 289   R15, R15_H,
 290   R17, R17_H,
 291   R18, R18_H,
 292   R19, R19_H,
 293   R20, R20_H,
 294   R21, R21_H,
 295   R22, R22_H,
 296   R23, R23_H,
 297   R24, R24_H,
 298   R25, R25_H,
 299   R26, R26_H,
 300   R27, R27_H,
 301   R28, R28_H,
 302   R29, R29_H,
 303   R30, R30_H,
 304   R31, R31_H,
 305 
 306   // scratch/special registers
 307   R11, R11_H,
 308   R12, R12_H,
 309 
 310   // argument registers
 311   R10, R10_H,
 312   R9,  R9_H,
 313   R8,  R8_H,
 314   R7,  R7_H,
 315   R6,  R6_H,
 316   R5,  R5_H,
 317   R4,  R4_H,
 318   R3,  R3_H,
 319 
 320   // special registers, not available for allocation
 321   R16, R16_H,     // R16_thread
 322   R13, R13_H,     // system thread id
 323   R2,  R2_H,      // may be used for TOC
 324   R1,  R1_H,      // SP
 325   R0,  R0_H       // R0 (scratch)
 326 );
 327 
 328 // If you change this allocation class, please have a look at the
 329 // default values for the parameters RoundRobinIntegerRegIntervalStart
 330 // and RoundRobinFloatRegIntervalStart
 331 
 332 alloc_class chunk1 (
 333   // Chunk1 contains *all* 64 floating-point registers halves.
 334 
 335   // scratch register
 336   F0,  F0_H,
 337 
 338   // argument registers
 339   F13, F13_H,
 340   F12, F12_H,
 341   F11, F11_H,
 342   F10, F10_H,
 343   F9,  F9_H,
 344   F8,  F8_H,
 345   F7,  F7_H,
 346   F6,  F6_H,
 347   F5,  F5_H,
 348   F4,  F4_H,
 349   F3,  F3_H,
 350   F2,  F2_H,
 351   F1,  F1_H,
 352 
 353   // non-volatile registers
 354   F14, F14_H,
 355   F15, F15_H,
 356   F16, F16_H,
 357   F17, F17_H,
 358   F18, F18_H,
 359   F19, F19_H,
 360   F20, F20_H,
 361   F21, F21_H,
 362   F22, F22_H,
 363   F23, F23_H,
 364   F24, F24_H,
 365   F25, F25_H,
 366   F26, F26_H,
 367   F27, F27_H,
 368   F28, F28_H,
 369   F29, F29_H,
 370   F30, F30_H,
 371   F31, F31_H
 372 );
 373 
 374 alloc_class chunk2 (
 375   // Chunk2 contains *all* 8 condition code registers.
 376 
 377   CCR0,
 378   CCR1,
 379   CCR2,
 380   CCR3,
 381   CCR4,
 382   CCR5,
 383   CCR6,
 384   CCR7
 385 );
 386 
 387 alloc_class chunk3 (
 388   // special registers
 389   // These registers are not allocated, but used for nodes generated by postalloc expand.
 390   SR_XER,
 391   SR_LR,
 392   SR_CTR,
 393   SR_VRSAVE,
 394   SR_SPEFSCR,
 395   SR_PPR
 396 );
 397 
 398 //-------Architecture Description Register Classes-----------------------
 399 
 400 // Several register classes are automatically defined based upon
 401 // information in this architecture description.
 402 
 403 // 1) reg_class inline_cache_reg           ( as defined in frame section )
 404 // 2) reg_class compiler_method_oop_reg    ( as defined in frame section )
 405 // 2) reg_class interpreter_method_oop_reg ( as defined in frame section )
 406 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
 407 //
 408 
 409 // ----------------------------
 410 // 32 Bit Register Classes
 411 // ----------------------------
 412 
 413 // We specify registers twice, once as read/write, and once read-only.
 414 // We use the read-only registers for source operands. With this, we
 415 // can include preset read only registers in this class, as a hard-coded
 416 // '0'-register. (We used to simulate this on ppc.)
 417 
 418 // 32 bit registers that can be read and written i.e. these registers
 419 // can be dest (or src) of normal instructions.
 420 reg_class bits32_reg_rw(
 421 /*R0*/              // R0
 422 /*R1*/              // SP
 423   R2,               // TOC
 424   R3,
 425   R4,
 426   R5,
 427   R6,
 428   R7,
 429   R8,
 430   R9,
 431   R10,
 432   R11,
 433   R12,
 434 /*R13*/             // system thread id
 435   R14,
 436   R15,
 437 /*R16*/             // R16_thread
 438   R17,
 439   R18,
 440   R19,
 441   R20,
 442   R21,
 443   R22,
 444   R23,
 445   R24,
 446   R25,
 447   R26,
 448   R27,
 449   R28,
 450 /*R29,*/             // global TOC
 451   R30,
 452   R31
 453 );
 454 
 455 // 32 bit registers that can only be read i.e. these registers can
 456 // only be src of all instructions.
 457 reg_class bits32_reg_ro(
 458 /*R0*/              // R0
 459 /*R1*/              // SP
 460   R2                // TOC
 461   R3,
 462   R4,
 463   R5,
 464   R6,
 465   R7,
 466   R8,
 467   R9,
 468   R10,
 469   R11,
 470   R12,
 471 /*R13*/             // system thread id
 472   R14,
 473   R15,
 474 /*R16*/             // R16_thread
 475   R17,
 476   R18,
 477   R19,
 478   R20,
 479   R21,
 480   R22,
 481   R23,
 482   R24,
 483   R25,
 484   R26,
 485   R27,
 486   R28,
 487 /*R29,*/
 488   R30,
 489   R31
 490 );
 491 
 492 reg_class rscratch1_bits32_reg(R11);
 493 reg_class rscratch2_bits32_reg(R12);
 494 reg_class rarg1_bits32_reg(R3);
 495 reg_class rarg2_bits32_reg(R4);
 496 reg_class rarg3_bits32_reg(R5);
 497 reg_class rarg4_bits32_reg(R6);
 498 
 499 // ----------------------------
 500 // 64 Bit Register Classes
 501 // ----------------------------
 502 // 64-bit build means 64-bit pointers means hi/lo pairs
 503 
 504 reg_class rscratch1_bits64_reg(R11_H, R11);
 505 reg_class rscratch2_bits64_reg(R12_H, R12);
 506 reg_class rarg1_bits64_reg(R3_H, R3);
 507 reg_class rarg2_bits64_reg(R4_H, R4);
 508 reg_class rarg3_bits64_reg(R5_H, R5);
 509 reg_class rarg4_bits64_reg(R6_H, R6);
 510 // Thread register, 'written' by tlsLoadP, see there.
 511 reg_class thread_bits64_reg(R16_H, R16);
 512 
 513 reg_class r19_bits64_reg(R19_H, R19);
 514 
 515 // 64 bit registers that can be read and written i.e. these registers
 516 // can be dest (or src) of normal instructions.
 517 reg_class bits64_reg_rw(
 518 /*R0_H,  R0*/     // R0
 519 /*R1_H,  R1*/     // SP
 520   R2_H,  R2,      // TOC
 521   R3_H,  R3,
 522   R4_H,  R4,
 523   R5_H,  R5,
 524   R6_H,  R6,
 525   R7_H,  R7,
 526   R8_H,  R8,
 527   R9_H,  R9,
 528   R10_H, R10,
 529   R11_H, R11,
 530   R12_H, R12,
 531 /*R13_H, R13*/   // system thread id
 532   R14_H, R14,
 533   R15_H, R15,
 534 /*R16_H, R16*/   // R16_thread
 535   R17_H, R17,
 536   R18_H, R18,
 537   R19_H, R19,
 538   R20_H, R20,
 539   R21_H, R21,
 540   R22_H, R22,
 541   R23_H, R23,
 542   R24_H, R24,
 543   R25_H, R25,
 544   R26_H, R26,
 545   R27_H, R27,
 546   R28_H, R28,
 547 /*R29_H, R29,*/
 548   R30_H, R30,
 549   R31_H, R31
 550 );
 551 
 552 // 64 bit registers used excluding r2, r11 and r12
 553 // Used to hold the TOC to avoid collisions with expanded LeafCall which uses
 554 // r2, r11 and r12 internally.
 555 reg_class bits64_reg_leaf_call(
 556 /*R0_H,  R0*/     // R0
 557 /*R1_H,  R1*/     // SP
 558 /*R2_H,  R2*/     // TOC
 559   R3_H,  R3,
 560   R4_H,  R4,
 561   R5_H,  R5,
 562   R6_H,  R6,
 563   R7_H,  R7,
 564   R8_H,  R8,
 565   R9_H,  R9,
 566   R10_H, R10,
 567 /*R11_H, R11*/
 568 /*R12_H, R12*/
 569 /*R13_H, R13*/   // system thread id
 570   R14_H, R14,
 571   R15_H, R15,
 572 /*R16_H, R16*/   // R16_thread
 573   R17_H, R17,
 574   R18_H, R18,
 575   R19_H, R19,
 576   R20_H, R20,
 577   R21_H, R21,
 578   R22_H, R22,
 579   R23_H, R23,
 580   R24_H, R24,
 581   R25_H, R25,
 582   R26_H, R26,
 583   R27_H, R27,
 584   R28_H, R28,
 585 /*R29_H, R29,*/
 586   R30_H, R30,
 587   R31_H, R31
 588 );
 589 
 590 // Used to hold the TOC to avoid collisions with expanded DynamicCall
 591 // which uses r19 as inline cache internally and expanded LeafCall which uses
 592 // r2, r11 and r12 internally.
 593 reg_class bits64_constant_table_base(
 594 /*R0_H,  R0*/     // R0
 595 /*R1_H,  R1*/     // SP
 596 /*R2_H,  R2*/     // TOC
 597   R3_H,  R3,
 598   R4_H,  R4,
 599   R5_H,  R5,
 600   R6_H,  R6,
 601   R7_H,  R7,
 602   R8_H,  R8,
 603   R9_H,  R9,
 604   R10_H, R10,
 605 /*R11_H, R11*/
 606 /*R12_H, R12*/
 607 /*R13_H, R13*/   // system thread id
 608   R14_H, R14,
 609   R15_H, R15,
 610 /*R16_H, R16*/   // R16_thread
 611   R17_H, R17,
 612   R18_H, R18,
 613 /*R19_H, R19*/
 614   R20_H, R20,
 615   R21_H, R21,
 616   R22_H, R22,
 617   R23_H, R23,
 618   R24_H, R24,
 619   R25_H, R25,
 620   R26_H, R26,
 621   R27_H, R27,
 622   R28_H, R28,
 623 /*R29_H, R29,*/
 624   R30_H, R30,
 625   R31_H, R31
 626 );
 627 
 628 // 64 bit registers that can only be read i.e. these registers can
 629 // only be src of all instructions.
 630 reg_class bits64_reg_ro(
 631 /*R0_H,  R0*/     // R0
 632   R1_H,  R1,
 633   R2_H,  R2,       // TOC
 634   R3_H,  R3,
 635   R4_H,  R4,
 636   R5_H,  R5,
 637   R6_H,  R6,
 638   R7_H,  R7,
 639   R8_H,  R8,
 640   R9_H,  R9,
 641   R10_H, R10,
 642   R11_H, R11,
 643   R12_H, R12,
 644 /*R13_H, R13*/   // system thread id
 645   R14_H, R14,
 646   R15_H, R15,
 647   R16_H, R16,    // R16_thread
 648   R17_H, R17,
 649   R18_H, R18,
 650   R19_H, R19,
 651   R20_H, R20,
 652   R21_H, R21,
 653   R22_H, R22,
 654   R23_H, R23,
 655   R24_H, R24,
 656   R25_H, R25,
 657   R26_H, R26,
 658   R27_H, R27,
 659   R28_H, R28,
 660 /*R29_H, R29,*/ // TODO: let allocator handle TOC!!
 661   R30_H, R30,
 662   R31_H, R31
 663 );
 664 
 665 
 666 // ----------------------------
 667 // Special Class for Condition Code Flags Register
 668 
 669 reg_class int_flags(
 670 /*CCR0*/             // scratch
 671 /*CCR1*/             // scratch
 672 /*CCR2*/             // nv!
 673 /*CCR3*/             // nv!
 674 /*CCR4*/             // nv!
 675   CCR5,
 676   CCR6,
 677   CCR7
 678 );
 679 
 680 reg_class int_flags_ro(
 681   CCR0,
 682   CCR1,
 683   CCR2,
 684   CCR3,
 685   CCR4,
 686   CCR5,
 687   CCR6,
 688   CCR7
 689 );
 690 
 691 reg_class int_flags_CR0(CCR0);
 692 reg_class int_flags_CR1(CCR1);
 693 reg_class int_flags_CR6(CCR6);
 694 reg_class ctr_reg(SR_CTR);
 695 
 696 // ----------------------------
 697 // Float Register Classes
 698 // ----------------------------
 699 
 700 reg_class flt_reg(
 701 /*F0*/              // scratch
 702   F1,
 703   F2,
 704   F3,
 705   F4,
 706   F5,
 707   F6,
 708   F7,
 709   F8,
 710   F9,
 711   F10,
 712   F11,
 713   F12,
 714   F13,
 715   F14,              // nv!
 716   F15,              // nv!
 717   F16,              // nv!
 718   F17,              // nv!
 719   F18,              // nv!
 720   F19,              // nv!
 721   F20,              // nv!
 722   F21,              // nv!
 723   F22,              // nv!
 724   F23,              // nv!
 725   F24,              // nv!
 726   F25,              // nv!
 727   F26,              // nv!
 728   F27,              // nv!
 729   F28,              // nv!
 730   F29,              // nv!
 731   F30,              // nv!
 732   F31               // nv!
 733 );
 734 
 735 // Double precision float registers have virtual `high halves' that
 736 // are needed by the allocator.
 737 reg_class dbl_reg(
 738 /*F0,  F0_H*/     // scratch
 739   F1,  F1_H,
 740   F2,  F2_H,
 741   F3,  F3_H,
 742   F4,  F4_H,
 743   F5,  F5_H,
 744   F6,  F6_H,
 745   F7,  F7_H,
 746   F8,  F8_H,
 747   F9,  F9_H,
 748   F10, F10_H,
 749   F11, F11_H,
 750   F12, F12_H,
 751   F13, F13_H,
 752   F14, F14_H,    // nv!
 753   F15, F15_H,    // nv!
 754   F16, F16_H,    // nv!
 755   F17, F17_H,    // nv!
 756   F18, F18_H,    // nv!
 757   F19, F19_H,    // nv!
 758   F20, F20_H,    // nv!
 759   F21, F21_H,    // nv!
 760   F22, F22_H,    // nv!
 761   F23, F23_H,    // nv!
 762   F24, F24_H,    // nv!
 763   F25, F25_H,    // nv!
 764   F26, F26_H,    // nv!
 765   F27, F27_H,    // nv!
 766   F28, F28_H,    // nv!
 767   F29, F29_H,    // nv!
 768   F30, F30_H,    // nv!
 769   F31, F31_H     // nv!
 770 );
 771 
 772  %}
 773 
 774 //----------DEFINITION BLOCK---------------------------------------------------
 775 // Define name --> value mappings to inform the ADLC of an integer valued name
 776 // Current support includes integer values in the range [0, 0x7FFFFFFF]
 777 // Format:
 778 //        int_def  <name>         ( <int_value>, <expression>);
 779 // Generated Code in ad_<arch>.hpp
 780 //        #define  <name>   (<expression>)
 781 //        // value == <int_value>
 782 // Generated code in ad_<arch>.cpp adlc_verification()
 783 //        assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
 784 //
 785 definitions %{
 786   // The default cost (of an ALU instruction).
 787   int_def DEFAULT_COST_LOW        (     30,      30);
 788   int_def DEFAULT_COST            (    100,     100);
 789   int_def HUGE_COST               (1000000, 1000000);
 790 
 791   // Memory refs
 792   int_def MEMORY_REF_COST_LOW     (    200, DEFAULT_COST * 2);
 793   int_def MEMORY_REF_COST         (    300, DEFAULT_COST * 3);
 794 
 795   // Branches are even more expensive.
 796   int_def BRANCH_COST             (    900, DEFAULT_COST * 9);
 797   int_def CALL_COST               (   1300, DEFAULT_COST * 13);
 798 %}
 799 
 800 
 801 //----------SOURCE BLOCK-------------------------------------------------------
 802 // This is a block of C++ code which provides values, functions, and
 803 // definitions necessary in the rest of the architecture description.
 804 source_hpp %{
 805   // Header information of the source block.
 806   // Method declarations/definitions which are used outside
 807   // the ad-scope can conveniently be defined here.
 808   //
 809   // To keep related declarations/definitions/uses close together,
 810   // we switch between source %{ }% and source_hpp %{ }% freely as needed.
 811 
 812   // Returns true if Node n is followed by a MemBar node that
 813   // will do an acquire. If so, this node must not do the acquire
 814   // operation.
 815   bool followed_by_acquire(const Node *n);
 816 %}
 817 
 818 source %{
 819 
 820 // Optimize load-acquire.
 821 //
 822 // Check if acquire is unnecessary due to following operation that does
 823 // acquire anyways.
 824 // Walk the pattern:
 825 //
 826 //      n: Load.acq
 827 //           |
 828 //      MemBarAcquire
 829 //       |         |
 830 //  Proj(ctrl)  Proj(mem)
 831 //       |         |
 832 //   MemBarRelease/Volatile
 833 //
 834 bool followed_by_acquire(const Node *load) {
 835   assert(load->is_Load(), "So far implemented only for loads.");
 836 
 837   // Find MemBarAcquire.
 838   const Node *mba = NULL;
 839   for (DUIterator_Fast imax, i = load->fast_outs(imax); i < imax; i++) {
 840     const Node *out = load->fast_out(i);
 841     if (out->Opcode() == Op_MemBarAcquire) {
 842       if (out->in(0) == load) continue; // Skip control edge, membar should be found via precedence edge.
 843       mba = out;
 844       break;
 845     }
 846   }
 847   if (!mba) return false;
 848 
 849   // Find following MemBar node.
 850   //
 851   // The following node must be reachable by control AND memory
 852   // edge to assure no other operations are in between the two nodes.
 853   //
 854   // So first get the Proj node, mem_proj, to use it to iterate forward.
 855   Node *mem_proj = NULL;
 856   for (DUIterator_Fast imax, i = mba->fast_outs(imax); i < imax; i++) {
 857     mem_proj = mba->fast_out(i);      // Throw out-of-bounds if proj not found
 858     assert(mem_proj->is_Proj(), "only projections here");
 859     ProjNode *proj = mem_proj->as_Proj();
 860     if (proj->_con == TypeFunc::Memory &&
 861         !Compile::current()->node_arena()->contains(mem_proj)) // Unmatched old-space only
 862       break;
 863   }
 864   assert(mem_proj->as_Proj()->_con == TypeFunc::Memory, "Graph broken");
 865 
 866   // Search MemBar behind Proj. If there are other memory operations
 867   // behind the Proj we lost.
 868   for (DUIterator_Fast jmax, j = mem_proj->fast_outs(jmax); j < jmax; j++) {
 869     Node *x = mem_proj->fast_out(j);
 870     // Proj might have an edge to a store or load node which precedes the membar.
 871     if (x->is_Mem()) return false;
 872 
 873     // On PPC64 release and volatile are implemented by an instruction
 874     // that also has acquire semantics. I.e. there is no need for an
 875     // acquire before these.
 876     int xop = x->Opcode();
 877     if (xop == Op_MemBarRelease || xop == Op_MemBarVolatile) {
 878       // Make sure we're not missing Call/Phi/MergeMem by checking
 879       // control edges. The control edge must directly lead back
 880       // to the MemBarAcquire
 881       Node *ctrl_proj = x->in(0);
 882       if (ctrl_proj->is_Proj() && ctrl_proj->in(0) == mba) {
 883         return true;
 884       }
 885     }
 886   }
 887 
 888   return false;
 889 }
 890 
 891 #define __ _masm.
 892 
 893 // Tertiary op of a LoadP or StoreP encoding.
 894 #define REGP_OP true
 895 
 896 // ****************************************************************************
 897 
 898 // REQUIRED FUNCTIONALITY
 899 
 900 // !!!!! Special hack to get all type of calls to specify the byte offset
 901 //       from the start of the call to the point where the return address
 902 //       will point.
 903 
 904 // PPC port: Removed use of lazy constant construct.
 905 
 906 int MachCallStaticJavaNode::ret_addr_offset() {
 907   // It's only a single branch-and-link instruction.
 908   return 4;
 909 }
 910 
 911 int MachCallDynamicJavaNode::ret_addr_offset() {
 912   // Offset is 4 with postalloc expanded calls (bl is one instruction). We use
 913   // postalloc expanded calls if we use inline caches and do not update method data.
 914   if (UseInlineCaches)
 915     return 4;
 916 
 917   int vtable_index = this->_vtable_index;
 918   if (vtable_index < 0) {
 919     // Must be invalid_vtable_index, not nonvirtual_vtable_index.
 920     assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
 921     return 12;
 922   } else {
 923     assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
 924     return 24;
 925   }
 926 }
 927 
 928 int MachCallRuntimeNode::ret_addr_offset() {
 929 #if defined(ABI_ELFv2)
 930   return 28;
 931 #else
 932   return 40;
 933 #endif
 934 }
 935 
 936 //=============================================================================
 937 
 938 // condition code conversions
 939 
 940 static int cc_to_boint(int cc) {
 941   return Assembler::bcondCRbiIs0 | (cc & 8);
 942 }
 943 
 944 static int cc_to_inverse_boint(int cc) {
 945   return Assembler::bcondCRbiIs0 | (8-(cc & 8));
 946 }
 947 
 948 static int cc_to_biint(int cc, int flags_reg) {
 949   return (flags_reg << 2) | (cc & 3);
 950 }
 951 
 952 //=============================================================================
 953 
 954 // Compute padding required for nodes which need alignment. The padding
 955 // is the number of bytes (not instructions) which will be inserted before
 956 // the instruction. The padding must match the size of a NOP instruction.
 957 
 958 int string_indexOf_imm1_charNode::compute_padding(int current_offset) const {
 959   return (3*4-current_offset)&31;
 960 }
 961 
 962 int string_indexOf_imm1Node::compute_padding(int current_offset) const {
 963   return (2*4-current_offset)&31;
 964 }
 965 
 966 int string_indexOf_immNode::compute_padding(int current_offset) const {
 967   return (3*4-current_offset)&31;
 968 }
 969 
 970 int string_indexOfNode::compute_padding(int current_offset) const {
 971   return (1*4-current_offset)&31;
 972 }
 973 
 974 int string_compareNode::compute_padding(int current_offset) const {
 975   return (4*4-current_offset)&31;
 976 }
 977 
 978 int string_equals_immNode::compute_padding(int current_offset) const {
 979   if (opnd_array(3)->constant() < 16) return 0; // Don't insert nops for short version (loop completely unrolled).
 980   return (2*4-current_offset)&31;
 981 }
 982 
 983 int string_equalsNode::compute_padding(int current_offset) const {
 984   return (7*4-current_offset)&31;
 985 }
 986 
 987 int inlineCallClearArrayNode::compute_padding(int current_offset) const {
 988   return (2*4-current_offset)&31;
 989 }
 990 
 991 //=============================================================================
 992 
 993 // Indicate if the safepoint node needs the polling page as an input.
 994 bool SafePointNode::needs_polling_address_input() {
 995   // The address is loaded from thread by a seperate node.
 996   return true;
 997 }
 998 
 999 //=============================================================================
1000 
1001 // Emit an interrupt that is caught by the debugger (for debugging compiler).
1002 void emit_break(CodeBuffer &cbuf) {
1003   MacroAssembler _masm(&cbuf);
1004   __ illtrap();
1005 }
1006 
1007 #ifndef PRODUCT
1008 void MachBreakpointNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1009   st->print("BREAKPOINT");
1010 }
1011 #endif
1012 
1013 void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1014   emit_break(cbuf);
1015 }
1016 
1017 uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
1018   return MachNode::size(ra_);
1019 }
1020 
1021 //=============================================================================
1022 
1023 void emit_nop(CodeBuffer &cbuf) {
1024   MacroAssembler _masm(&cbuf);
1025   __ nop();
1026 }
1027 
1028 static inline void emit_long(CodeBuffer &cbuf, int value) {
1029   *((int*)(cbuf.insts_end())) = value;
1030   cbuf.set_insts_end(cbuf.insts_end() + BytesPerInstWord);
1031 }
1032 
1033 //=============================================================================
1034 
1035 %} // interrupt source
1036 
1037 source_hpp %{ // Header information of the source block.
1038 
1039 //--------------------------------------------------------------
1040 //---<  Used for optimization in Compile::Shorten_branches  >---
1041 //--------------------------------------------------------------
1042 
1043 const uint trampoline_stub_size     =  6 * BytesPerInstWord;
1044 
1045 class CallStubImpl {
1046 
1047  public:
1048 
1049   // Emit call stub, compiled java to interpreter.
1050   static void emit_trampoline_stub(MacroAssembler &_masm, int destination_toc_offset, int insts_call_instruction_offset);
1051 
1052   // Size of call trampoline stub.
1053   // This doesn't need to be accurate to the byte, but it
1054   // must be larger than or equal to the real size of the stub.
1055   static uint size_call_trampoline() {
1056     return trampoline_stub_size;
1057   }
1058 
1059   // number of relocations needed by a call trampoline stub
1060   static uint reloc_call_trampoline() {
1061     return 5;
1062   }
1063 
1064 };
1065 
1066 %} // end source_hpp
1067 
1068 source %{
1069 
1070 // Emit a trampoline stub for a call to a target which is too far away.
1071 //
1072 // code sequences:
1073 //
1074 // call-site:
1075 //   branch-and-link to <destination> or <trampoline stub>
1076 //
1077 // Related trampoline stub for this call-site in the stub section:
1078 //   load the call target from the constant pool
1079 //   branch via CTR (LR/link still points to the call-site above)
1080 
1081 void CallStubImpl::emit_trampoline_stub(MacroAssembler &_masm, int destination_toc_offset, int insts_call_instruction_offset) {
1082   // Start the stub.
1083   address stub = __ start_a_stub(Compile::MAX_stubs_size/2);
1084   if (stub == NULL) {
1085     ciEnv::current()->record_failure("CodeCache is full");
1086     return;
1087   }
1088 
1089   // For java_to_interp stubs we use R11_scratch1 as scratch register
1090   // and in call trampoline stubs we use R12_scratch2. This way we
1091   // can distinguish them (see is_NativeCallTrampolineStub_at()).
1092   Register reg_scratch = R12_scratch2;
1093 
1094   // Create a trampoline stub relocation which relates this trampoline stub
1095   // with the call instruction at insts_call_instruction_offset in the
1096   // instructions code-section.
1097   __ relocate(trampoline_stub_Relocation::spec(__ code()->insts()->start() + insts_call_instruction_offset));
1098   const int stub_start_offset = __ offset();
1099 
1100   // Now, create the trampoline stub's code:
1101   // - load the TOC
1102   // - load the call target from the constant pool
1103   // - call
1104   __ calculate_address_from_global_toc(reg_scratch, __ method_toc());
1105   __ ld_largeoffset_unchecked(reg_scratch, destination_toc_offset, reg_scratch, false);
1106   __ mtctr(reg_scratch);
1107   __ bctr();
1108 
1109   const address stub_start_addr = __ addr_at(stub_start_offset);
1110 
1111   // FIXME: Assert that the trampoline stub can be identified and patched.
1112 
1113   // Assert that the encoded destination_toc_offset can be identified and that it is correct.
1114   assert(destination_toc_offset == NativeCallTrampolineStub_at(stub_start_addr)->destination_toc_offset(),
1115          "encoded offset into the constant pool must match");
1116   // Trampoline_stub_size should be good.
1117   assert((uint)(__ offset() - stub_start_offset) <= trampoline_stub_size, "should be good size");
1118   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
1119 
1120   // End the stub.
1121   __ end_a_stub();
1122 }
1123 
1124 //=============================================================================
1125 
1126 // Emit an inline branch-and-link call and a related trampoline stub.
1127 //
1128 // code sequences:
1129 //
1130 // call-site:
1131 //   branch-and-link to <destination> or <trampoline stub>
1132 //
1133 // Related trampoline stub for this call-site in the stub section:
1134 //   load the call target from the constant pool
1135 //   branch via CTR (LR/link still points to the call-site above)
1136 //
1137 
1138 typedef struct {
1139   int insts_call_instruction_offset;
1140   int ret_addr_offset;
1141 } EmitCallOffsets;
1142 
1143 // Emit a branch-and-link instruction that branches to a trampoline.
1144 // - Remember the offset of the branch-and-link instruction.
1145 // - Add a relocation at the branch-and-link instruction.
1146 // - Emit a branch-and-link.
1147 // - Remember the return pc offset.
1148 EmitCallOffsets emit_call_with_trampoline_stub(MacroAssembler &_masm, address entry_point, relocInfo::relocType rtype) {
1149   EmitCallOffsets offsets = { -1, -1 };
1150   const int start_offset = __ offset();
1151   offsets.insts_call_instruction_offset = __ offset();
1152 
1153   // No entry point given, use the current pc.
1154   if (entry_point == NULL) entry_point = __ pc();
1155 
1156   if (!Compile::current()->in_scratch_emit_size()) {
1157     // Put the entry point as a constant into the constant pool.
1158     const address entry_point_toc_addr   = __ address_constant(entry_point, RelocationHolder::none);
1159     const int     entry_point_toc_offset = __ offset_to_method_toc(entry_point_toc_addr);
1160 
1161     // Emit the trampoline stub which will be related to the branch-and-link below.
1162     CallStubImpl::emit_trampoline_stub(_masm, entry_point_toc_offset, offsets.insts_call_instruction_offset);
1163     if (ciEnv::current()->failing()) { return offsets; } // Code cache may be full.
1164     __ relocate(rtype);
1165   }
1166 
1167   // Note: At this point we do not have the address of the trampoline
1168   // stub, and the entry point might be too far away for bl, so __ pc()
1169   // serves as dummy and the bl will be patched later.
1170   __ bl((address) __ pc());
1171 
1172   offsets.ret_addr_offset = __ offset() - start_offset;
1173 
1174   return offsets;
1175 }
1176 
1177 //=============================================================================
1178 
1179 // Factory for creating loadConL* nodes for large/small constant pool.
1180 
1181 static inline jlong replicate_immF(float con) {
1182   // Replicate float con 2 times and pack into vector.
1183   int val = *((int*)&con);
1184   jlong lval = val;
1185   lval = (lval << 32) | (lval & 0xFFFFFFFFl);
1186   return lval;
1187 }
1188 
1189 //=============================================================================
1190 
1191 const RegMask& MachConstantBaseNode::_out_RegMask = BITS64_CONSTANT_TABLE_BASE_mask();
1192 int Compile::ConstantTable::calculate_table_base_offset() const {
1193   return 0;  // absolute addressing, no offset
1194 }
1195 
1196 bool MachConstantBaseNode::requires_postalloc_expand() const { return true; }
1197 void MachConstantBaseNode::postalloc_expand(GrowableArray <Node *> *nodes, PhaseRegAlloc *ra_) {
1198   iRegPdstOper *op_dst = new iRegPdstOper();
1199   MachNode *m1 = new loadToc_hiNode();
1200   MachNode *m2 = new loadToc_loNode();
1201 
1202   m1->add_req(NULL);
1203   m2->add_req(NULL, m1);
1204   m1->_opnds[0] = op_dst;
1205   m2->_opnds[0] = op_dst;
1206   m2->_opnds[1] = op_dst;
1207   ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
1208   ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
1209   nodes->push(m1);
1210   nodes->push(m2);
1211 }
1212 
1213 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
1214   // Is postalloc expanded.
1215   ShouldNotReachHere();
1216 }
1217 
1218 uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
1219   return 0;
1220 }
1221 
1222 #ifndef PRODUCT
1223 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
1224   st->print("-- \t// MachConstantBaseNode (empty encoding)");
1225 }
1226 #endif
1227 
1228 //=============================================================================
1229 
1230 #ifndef PRODUCT
1231 void MachPrologNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1232   Compile* C = ra_->C;
1233   const long framesize = C->frame_slots() << LogBytesPerInt;
1234 
1235   st->print("PROLOG\n\t");
1236   if (C->need_stack_bang(framesize)) {
1237     st->print("stack_overflow_check\n\t");
1238   }
1239 
1240   if (!false /* TODO: PPC port C->is_frameless_method()*/) {
1241     st->print("save return pc\n\t");
1242     st->print("push frame %ld\n\t", -framesize);
1243   }
1244 }
1245 #endif
1246 
1247 // Macro used instead of the common __ to emulate the pipes of PPC.
1248 // Instead of e.g. __ ld(...) one hase to write ___(ld) ld(...) This enables the
1249 // micro scheduler to cope with "hand written" assembler like in the prolog. Though
1250 // still no scheduling of this code is possible, the micro scheduler is aware of the
1251 // code and can update its internal data. The following mechanism is used to achieve this:
1252 // The micro scheduler calls size() of each compound node during scheduling. size() does a
1253 // dummy emit and only during this dummy emit C->hb_scheduling() is not NULL.
1254 #if 0 // TODO: PPC port
1255 #define ___(op) if (UsePower6SchedulerPPC64 && C->hb_scheduling())                    \
1256                   C->hb_scheduling()->_pdScheduling->PdEmulatePipe(ppc64Opcode_##op); \
1257                 _masm.
1258 #define ___stop if (UsePower6SchedulerPPC64 && C->hb_scheduling())                    \
1259                   C->hb_scheduling()->_pdScheduling->PdEmulatePipe(archOpcode_none)
1260 #define ___advance if (UsePower6SchedulerPPC64 && C->hb_scheduling())                 \
1261                   C->hb_scheduling()->_pdScheduling->advance_offset
1262 #else
1263 #define ___(op) if (UsePower6SchedulerPPC64)                                          \
1264                   Unimplemented();                                                    \
1265                 _masm.
1266 #define ___stop if (UsePower6SchedulerPPC64)                                          \
1267                   Unimplemented()
1268 #define ___advance if (UsePower6SchedulerPPC64)                                       \
1269                   Unimplemented()
1270 #endif
1271 
1272 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1273   Compile* C = ra_->C;
1274   MacroAssembler _masm(&cbuf);
1275 
1276   const long framesize = C->frame_size_in_bytes();
1277   assert(framesize % (2 * wordSize) == 0, "must preserve 2*wordSize alignment");
1278 
1279   const bool method_is_frameless      = false /* TODO: PPC port C->is_frameless_method()*/;
1280 
1281   const Register return_pc            = R20; // Must match return_addr() in frame section.
1282   const Register callers_sp           = R21;
1283   const Register push_frame_temp      = R22;
1284   const Register toc_temp             = R23;
1285   assert_different_registers(R11, return_pc, callers_sp, push_frame_temp, toc_temp);
1286 
1287   if (method_is_frameless) {
1288     // Add nop at beginning of all frameless methods to prevent any
1289     // oop instructions from getting overwritten by make_not_entrant
1290     // (patching attempt would fail).
1291     ___(nop) nop();
1292   } else {
1293     // Get return pc.
1294     ___(mflr) mflr(return_pc);
1295   }
1296 
1297   // Calls to C2R adapters often do not accept exceptional returns.
1298   // We require that their callers must bang for them. But be
1299   // careful, because some VM calls (such as call site linkage) can
1300   // use several kilobytes of stack. But the stack safety zone should
1301   // account for that. See bugs 4446381, 4468289, 4497237.
1302 
1303   int bangsize = C->bang_size_in_bytes();
1304   assert(bangsize >= framesize || bangsize <= 0, "stack bang size incorrect");
1305   if (C->need_stack_bang(bangsize) && UseStackBanging) {
1306     // Unfortunately we cannot use the function provided in
1307     // assembler.cpp as we have to emulate the pipes. So I had to
1308     // insert the code of generate_stack_overflow_check(), see
1309     // assembler.cpp for some illuminative comments.
1310     const int page_size = os::vm_page_size();
1311     int bang_end = StackShadowPages * page_size;
1312 
1313     // This is how far the previous frame's stack banging extended.
1314     const int bang_end_safe = bang_end;
1315 
1316     if (bangsize > page_size) {
1317       bang_end += bangsize;
1318     }
1319 
1320     int bang_offset = bang_end_safe;
1321 
1322     while (bang_offset <= bang_end) {
1323       // Need at least one stack bang at end of shadow zone.
1324 
1325       // Again I had to copy code, this time from assembler_ppc.cpp,
1326       // bang_stack_with_offset - see there for comments.
1327 
1328       // Stack grows down, caller passes positive offset.
1329       assert(bang_offset > 0, "must bang with positive offset");
1330 
1331       long stdoffset = -bang_offset;
1332 
1333       if (Assembler::is_simm(stdoffset, 16)) {
1334         // Signed 16 bit offset, a simple std is ok.
1335         if (UseLoadInstructionsForStackBangingPPC64) {
1336           ___(ld) ld(R0,  (int)(signed short)stdoffset, R1_SP);
1337         } else {
1338           ___(std) std(R0, (int)(signed short)stdoffset, R1_SP);
1339         }
1340       } else if (Assembler::is_simm(stdoffset, 31)) {
1341         // Use largeoffset calculations for addis & ld/std.
1342         const int hi = MacroAssembler::largeoffset_si16_si16_hi(stdoffset);
1343         const int lo = MacroAssembler::largeoffset_si16_si16_lo(stdoffset);
1344 
1345         Register tmp = R11;
1346         ___(addis) addis(tmp, R1_SP, hi);
1347         if (UseLoadInstructionsForStackBangingPPC64) {
1348           ___(ld) ld(R0, lo, tmp);
1349         } else {
1350           ___(std) std(R0, lo, tmp);
1351         }
1352       } else {
1353         ShouldNotReachHere();
1354       }
1355 
1356       bang_offset += page_size;
1357     }
1358     // R11 trashed
1359   } // C->need_stack_bang(framesize) && UseStackBanging
1360 
1361   unsigned int bytes = (unsigned int)framesize;
1362   long offset = Assembler::align_addr(bytes, frame::alignment_in_bytes);
1363   ciMethod *currMethod = C->method();
1364 
1365   // Optimized version for most common case.
1366   if (UsePower6SchedulerPPC64 &&
1367       !method_is_frameless && Assembler::is_simm((int)(-offset), 16) &&
1368       !(false /* ConstantsALot TODO: PPC port*/)) {
1369     ___(or) mr(callers_sp, R1_SP);
1370     ___(std) std(return_pc, _abi(lr), R1_SP);
1371     ___(stdu) stdu(R1_SP, -offset, R1_SP);
1372     return;
1373   }
1374 
1375   if (!method_is_frameless) {
1376     // Get callers sp.
1377     ___(or) mr(callers_sp, R1_SP);
1378 
1379     // Push method's frame, modifies SP.
1380     assert(Assembler::is_uimm(framesize, 32U), "wrong type");
1381     // The ABI is already accounted for in 'framesize' via the
1382     // 'out_preserve' area.
1383     Register tmp = push_frame_temp;
1384     // Had to insert code of push_frame((unsigned int)framesize, push_frame_temp).
1385     if (Assembler::is_simm(-offset, 16)) {
1386       ___(stdu) stdu(R1_SP, -offset, R1_SP);
1387     } else {
1388       long x = -offset;
1389       // Had to insert load_const(tmp, -offset).
1390       ___(addis)  lis( tmp, (int)((signed short)(((x >> 32) & 0xffff0000) >> 16)));
1391       ___(ori)    ori( tmp, tmp, ((x >> 32) & 0x0000ffff));
1392       ___(rldicr) sldi(tmp, tmp, 32);
1393       ___(oris)   oris(tmp, tmp, (x & 0xffff0000) >> 16);
1394       ___(ori)    ori( tmp, tmp, (x & 0x0000ffff));
1395 
1396       ___(stdux) stdux(R1_SP, R1_SP, tmp);
1397     }
1398   }
1399 #if 0 // TODO: PPC port
1400   // For testing large constant pools, emit a lot of constants to constant pool.
1401   // "Randomize" const_size.
1402   if (ConstantsALot) {
1403     const int num_consts = const_size();
1404     for (int i = 0; i < num_consts; i++) {
1405       __ long_constant(0xB0B5B00BBABE);
1406     }
1407   }
1408 #endif
1409   if (!method_is_frameless) {
1410     // Save return pc.
1411     ___(std) std(return_pc, _abi(lr), callers_sp);
1412   }
1413 }
1414 #undef ___
1415 #undef ___stop
1416 #undef ___advance
1417 
1418 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
1419   // Variable size. determine dynamically.
1420   return MachNode::size(ra_);
1421 }
1422 
1423 int MachPrologNode::reloc() const {
1424   // Return number of relocatable values contained in this instruction.
1425   return 1; // 1 reloc entry for load_const(toc).
1426 }
1427 
1428 //=============================================================================
1429 
1430 #ifndef PRODUCT
1431 void MachEpilogNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1432   Compile* C = ra_->C;
1433 
1434   st->print("EPILOG\n\t");
1435   st->print("restore return pc\n\t");
1436   st->print("pop frame\n\t");
1437 
1438   if (do_polling() && C->is_method_compilation()) {
1439     st->print("touch polling page\n\t");
1440   }
1441 }
1442 #endif
1443 
1444 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1445   Compile* C = ra_->C;
1446   MacroAssembler _masm(&cbuf);
1447 
1448   const long framesize = ((long)C->frame_slots()) << LogBytesPerInt;
1449   assert(framesize >= 0, "negative frame-size?");
1450 
1451   const bool method_needs_polling = do_polling() && C->is_method_compilation();
1452   const bool method_is_frameless  = false /* TODO: PPC port C->is_frameless_method()*/;
1453   const Register return_pc        = R11;
1454   const Register polling_page     = R12;
1455 
1456   if (!method_is_frameless) {
1457     // Restore return pc relative to callers' sp.
1458     __ ld(return_pc, ((int)framesize) + _abi(lr), R1_SP);
1459   }
1460 
1461   if (method_needs_polling) {
1462     if (LoadPollAddressFromThread) {
1463       // TODO: PPC port __ ld(polling_page, in_bytes(JavaThread::poll_address_offset()), R16_thread);
1464       Unimplemented();
1465     } else {
1466       __ load_const_optimized(polling_page, (long)(address) os::get_polling_page()); // TODO: PPC port: get_standard_polling_page()
1467     }
1468   }
1469 
1470   if (!method_is_frameless) {
1471     // Move return pc to LR.
1472     __ mtlr(return_pc);
1473     // Pop frame (fixed frame-size).
1474     __ addi(R1_SP, R1_SP, (int)framesize);
1475   }
1476 
1477   if (method_needs_polling) {
1478     // We need to mark the code position where the load from the safepoint
1479     // polling page was emitted as relocInfo::poll_return_type here.
1480     __ relocate(relocInfo::poll_return_type);
1481     __ load_from_polling_page(polling_page);
1482   }
1483 }
1484 
1485 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
1486   // Variable size. Determine dynamically.
1487   return MachNode::size(ra_);
1488 }
1489 
1490 int MachEpilogNode::reloc() const {
1491   // Return number of relocatable values contained in this instruction.
1492   return 1; // 1 for load_from_polling_page.
1493 }
1494 
1495 const Pipeline * MachEpilogNode::pipeline() const {
1496   return MachNode::pipeline_class();
1497 }
1498 
1499 // This method seems to be obsolete. It is declared in machnode.hpp
1500 // and defined in all *.ad files, but it is never called. Should we
1501 // get rid of it?
1502 int MachEpilogNode::safepoint_offset() const {
1503   assert(do_polling(), "no return for this epilog node");
1504   return 0;
1505 }
1506 
1507 #if 0 // TODO: PPC port
1508 void MachLoadPollAddrLateNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
1509   MacroAssembler _masm(&cbuf);
1510   if (LoadPollAddressFromThread) {
1511     _masm.ld(R11, in_bytes(JavaThread::poll_address_offset()), R16_thread);
1512   } else {
1513     _masm.nop();
1514   }
1515 }
1516 
1517 uint MachLoadPollAddrLateNode::size(PhaseRegAlloc* ra_) const {
1518   if (LoadPollAddressFromThread) {
1519     return 4;
1520   } else {
1521     return 4;
1522   }
1523 }
1524 
1525 #ifndef PRODUCT
1526 void MachLoadPollAddrLateNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
1527   st->print_cr(" LD R11, PollAddressOffset, R16_thread \t// LoadPollAddressFromThread");
1528 }
1529 #endif
1530 
1531 const RegMask &MachLoadPollAddrLateNode::out_RegMask() const {
1532   return RSCRATCH1_BITS64_REG_mask();
1533 }
1534 #endif // PPC port
1535 
1536 // =============================================================================
1537 
1538 // Figure out which register class each belongs in: rc_int, rc_float or
1539 // rc_stack.
1540 enum RC { rc_bad, rc_int, rc_float, rc_stack };
1541 
1542 static enum RC rc_class(OptoReg::Name reg) {
1543   // Return the register class for the given register. The given register
1544   // reg is a <register>_num value, which is an index into the MachRegisterNumbers
1545   // enumeration in adGlobals_ppc.hpp.
1546 
1547   if (reg == OptoReg::Bad) return rc_bad;
1548 
1549   // We have 64 integer register halves, starting at index 0.
1550   if (reg < 64) return rc_int;
1551 
1552   // We have 64 floating-point register halves, starting at index 64.
1553   if (reg < 64+64) return rc_float;
1554 
1555   // Between float regs & stack are the flags regs.
1556   assert(OptoReg::is_stack(reg), "blow up if spilling flags");
1557 
1558   return rc_stack;
1559 }
1560 
1561 static int ld_st_helper(CodeBuffer *cbuf, const char *op_str, uint opcode, int reg, int offset,
1562                         bool do_print, Compile* C, outputStream *st) {
1563 
1564   assert(opcode == Assembler::LD_OPCODE   ||
1565          opcode == Assembler::STD_OPCODE  ||
1566          opcode == Assembler::LWZ_OPCODE  ||
1567          opcode == Assembler::STW_OPCODE  ||
1568          opcode == Assembler::LFD_OPCODE  ||
1569          opcode == Assembler::STFD_OPCODE ||
1570          opcode == Assembler::LFS_OPCODE  ||
1571          opcode == Assembler::STFS_OPCODE,
1572          "opcode not supported");
1573 
1574   if (cbuf) {
1575     int d =
1576       (Assembler::LD_OPCODE == opcode || Assembler::STD_OPCODE == opcode) ?
1577         Assembler::ds(offset+0 /* TODO: PPC port C->frame_slots_sp_bias_in_bytes()*/)
1578       : Assembler::d1(offset+0 /* TODO: PPC port C->frame_slots_sp_bias_in_bytes()*/); // Makes no difference in opt build.
1579     emit_long(*cbuf, opcode | Assembler::rt(Matcher::_regEncode[reg]) | d | Assembler::ra(R1_SP));
1580   }
1581 #ifndef PRODUCT
1582   else if (do_print) {
1583     st->print("%-7s %s, [R1_SP + #%d+%d] \t// spill copy",
1584               op_str,
1585               Matcher::regName[reg],
1586               offset, 0 /* TODO: PPC port C->frame_slots_sp_bias_in_bytes()*/);
1587   }
1588 #endif
1589   return 4; // size
1590 }
1591 
1592 uint MachSpillCopyNode::implementation(CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream *st) const {
1593   Compile* C = ra_->C;
1594 
1595   // Get registers to move.
1596   OptoReg::Name src_hi = ra_->get_reg_second(in(1));
1597   OptoReg::Name src_lo = ra_->get_reg_first(in(1));
1598   OptoReg::Name dst_hi = ra_->get_reg_second(this);
1599   OptoReg::Name dst_lo = ra_->get_reg_first(this);
1600 
1601   enum RC src_hi_rc = rc_class(src_hi);
1602   enum RC src_lo_rc = rc_class(src_lo);
1603   enum RC dst_hi_rc = rc_class(dst_hi);
1604   enum RC dst_lo_rc = rc_class(dst_lo);
1605 
1606   assert(src_lo != OptoReg::Bad && dst_lo != OptoReg::Bad, "must move at least 1 register");
1607   if (src_hi != OptoReg::Bad)
1608     assert((src_lo&1)==0 && src_lo+1==src_hi &&
1609            (dst_lo&1)==0 && dst_lo+1==dst_hi,
1610            "expected aligned-adjacent pairs");
1611   // Generate spill code!
1612   int size = 0;
1613 
1614   if (src_lo == dst_lo && src_hi == dst_hi)
1615     return size;            // Self copy, no move.
1616 
1617   // --------------------------------------
1618   // Memory->Memory Spill. Use R0 to hold the value.
1619   if (src_lo_rc == rc_stack && dst_lo_rc == rc_stack) {
1620     int src_offset = ra_->reg2offset(src_lo);
1621     int dst_offset = ra_->reg2offset(dst_lo);
1622     if (src_hi != OptoReg::Bad) {
1623       assert(src_hi_rc==rc_stack && dst_hi_rc==rc_stack,
1624              "expected same type of move for high parts");
1625       size += ld_st_helper(cbuf, "LD  ", Assembler::LD_OPCODE,  R0_num, src_offset, !do_size, C, st);
1626       if (!cbuf && !do_size) st->print("\n\t");
1627       size += ld_st_helper(cbuf, "STD ", Assembler::STD_OPCODE, R0_num, dst_offset, !do_size, C, st);
1628     } else {
1629       size += ld_st_helper(cbuf, "LWZ ", Assembler::LWZ_OPCODE, R0_num, src_offset, !do_size, C, st);
1630       if (!cbuf && !do_size) st->print("\n\t");
1631       size += ld_st_helper(cbuf, "STW ", Assembler::STW_OPCODE, R0_num, dst_offset, !do_size, C, st);
1632     }
1633     return size;
1634   }
1635 
1636   // --------------------------------------
1637   // Check for float->int copy; requires a trip through memory.
1638   if (src_lo_rc == rc_float && dst_lo_rc == rc_int) {
1639     Unimplemented();
1640   }
1641 
1642   // --------------------------------------
1643   // Check for integer reg-reg copy.
1644   if (src_lo_rc == rc_int && dst_lo_rc == rc_int) {
1645       Register Rsrc = as_Register(Matcher::_regEncode[src_lo]);
1646       Register Rdst = as_Register(Matcher::_regEncode[dst_lo]);
1647       size = (Rsrc != Rdst) ? 4 : 0;
1648 
1649       if (cbuf) {
1650         MacroAssembler _masm(cbuf);
1651         if (size) {
1652           __ mr(Rdst, Rsrc);
1653         }
1654       }
1655 #ifndef PRODUCT
1656       else if (!do_size) {
1657         if (size) {
1658           st->print("%-7s %s, %s \t// spill copy", "MR", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
1659         } else {
1660           st->print("%-7s %s, %s \t// spill copy", "MR-NOP", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
1661         }
1662       }
1663 #endif
1664       return size;
1665   }
1666 
1667   // Check for integer store.
1668   if (src_lo_rc == rc_int && dst_lo_rc == rc_stack) {
1669     int dst_offset = ra_->reg2offset(dst_lo);
1670     if (src_hi != OptoReg::Bad) {
1671       assert(src_hi_rc==rc_int && dst_hi_rc==rc_stack,
1672              "expected same type of move for high parts");
1673       size += ld_st_helper(cbuf, "STD ", Assembler::STD_OPCODE, src_lo, dst_offset, !do_size, C, st);
1674     } else {
1675       size += ld_st_helper(cbuf, "STW ", Assembler::STW_OPCODE, src_lo, dst_offset, !do_size, C, st);
1676     }
1677     return size;
1678   }
1679 
1680   // Check for integer load.
1681   if (dst_lo_rc == rc_int && src_lo_rc == rc_stack) {
1682     int src_offset = ra_->reg2offset(src_lo);
1683     if (src_hi != OptoReg::Bad) {
1684       assert(dst_hi_rc==rc_int && src_hi_rc==rc_stack,
1685              "expected same type of move for high parts");
1686       size += ld_st_helper(cbuf, "LD  ", Assembler::LD_OPCODE, dst_lo, src_offset, !do_size, C, st);
1687     } else {
1688       size += ld_st_helper(cbuf, "LWZ ", Assembler::LWZ_OPCODE, dst_lo, src_offset, !do_size, C, st);
1689     }
1690     return size;
1691   }
1692 
1693   // Check for float reg-reg copy.
1694   if (src_lo_rc == rc_float && dst_lo_rc == rc_float) {
1695     if (cbuf) {
1696       MacroAssembler _masm(cbuf);
1697       FloatRegister Rsrc = as_FloatRegister(Matcher::_regEncode[src_lo]);
1698       FloatRegister Rdst = as_FloatRegister(Matcher::_regEncode[dst_lo]);
1699       __ fmr(Rdst, Rsrc);
1700     }
1701 #ifndef PRODUCT
1702     else if (!do_size) {
1703       st->print("%-7s %s, %s \t// spill copy", "FMR", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
1704     }
1705 #endif
1706     return 4;
1707   }
1708 
1709   // Check for float store.
1710   if (src_lo_rc == rc_float && dst_lo_rc == rc_stack) {
1711     int dst_offset = ra_->reg2offset(dst_lo);
1712     if (src_hi != OptoReg::Bad) {
1713       assert(src_hi_rc==rc_float && dst_hi_rc==rc_stack,
1714              "expected same type of move for high parts");
1715       size += ld_st_helper(cbuf, "STFD", Assembler::STFD_OPCODE, src_lo, dst_offset, !do_size, C, st);
1716     } else {
1717       size += ld_st_helper(cbuf, "STFS", Assembler::STFS_OPCODE, src_lo, dst_offset, !do_size, C, st);
1718     }
1719     return size;
1720   }
1721 
1722   // Check for float load.
1723   if (dst_lo_rc == rc_float && src_lo_rc == rc_stack) {
1724     int src_offset = ra_->reg2offset(src_lo);
1725     if (src_hi != OptoReg::Bad) {
1726       assert(dst_hi_rc==rc_float && src_hi_rc==rc_stack,
1727              "expected same type of move for high parts");
1728       size += ld_st_helper(cbuf, "LFD ", Assembler::LFD_OPCODE, dst_lo, src_offset, !do_size, C, st);
1729     } else {
1730       size += ld_st_helper(cbuf, "LFS ", Assembler::LFS_OPCODE, dst_lo, src_offset, !do_size, C, st);
1731     }
1732     return size;
1733   }
1734 
1735   // --------------------------------------------------------------------
1736   // Check for hi bits still needing moving. Only happens for misaligned
1737   // arguments to native calls.
1738   if (src_hi == dst_hi)
1739     return size;               // Self copy; no move.
1740 
1741   assert(src_hi_rc != rc_bad && dst_hi_rc != rc_bad, "src_hi & dst_hi cannot be Bad");
1742   ShouldNotReachHere(); // Unimplemented
1743   return 0;
1744 }
1745 
1746 #ifndef PRODUCT
1747 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1748   if (!ra_)
1749     st->print("N%d = SpillCopy(N%d)", _idx, in(1)->_idx);
1750   else
1751     implementation(NULL, ra_, false, st);
1752 }
1753 #endif
1754 
1755 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1756   implementation(&cbuf, ra_, false, NULL);
1757 }
1758 
1759 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1760   return implementation(NULL, ra_, true, NULL);
1761 }
1762 
1763 #if 0 // TODO: PPC port
1764 ArchOpcode MachSpillCopyNode_archOpcode(MachSpillCopyNode *n, PhaseRegAlloc *ra_) {
1765 #ifndef PRODUCT
1766   if (ra_->node_regs_max_index() == 0) return archOpcode_undefined;
1767 #endif
1768   assert(ra_->node_regs_max_index() != 0, "");
1769 
1770   // Get registers to move.
1771   OptoReg::Name src_hi = ra_->get_reg_second(n->in(1));
1772   OptoReg::Name src_lo = ra_->get_reg_first(n->in(1));
1773   OptoReg::Name dst_hi = ra_->get_reg_second(n);
1774   OptoReg::Name dst_lo = ra_->get_reg_first(n);
1775 
1776   enum RC src_lo_rc = rc_class(src_lo);
1777   enum RC dst_lo_rc = rc_class(dst_lo);
1778 
1779   if (src_lo == dst_lo && src_hi == dst_hi)
1780     return ppc64Opcode_none;            // Self copy, no move.
1781 
1782   // --------------------------------------
1783   // Memory->Memory Spill. Use R0 to hold the value.
1784   if (src_lo_rc == rc_stack && dst_lo_rc == rc_stack) {
1785     return ppc64Opcode_compound;
1786   }
1787 
1788   // --------------------------------------
1789   // Check for float->int copy; requires a trip through memory.
1790   if (src_lo_rc == rc_float && dst_lo_rc == rc_int) {
1791     Unimplemented();
1792   }
1793 
1794   // --------------------------------------
1795   // Check for integer reg-reg copy.
1796   if (src_lo_rc == rc_int && dst_lo_rc == rc_int) {
1797     Register Rsrc = as_Register(Matcher::_regEncode[src_lo]);
1798     Register Rdst = as_Register(Matcher::_regEncode[dst_lo]);
1799     if (Rsrc == Rdst) {
1800       return ppc64Opcode_none;
1801     } else {
1802       return ppc64Opcode_or;
1803     }
1804   }
1805 
1806   // Check for integer store.
1807   if (src_lo_rc == rc_int && dst_lo_rc == rc_stack) {
1808     if (src_hi != OptoReg::Bad) {
1809       return ppc64Opcode_std;
1810     } else {
1811       return ppc64Opcode_stw;
1812     }
1813   }
1814 
1815   // Check for integer load.
1816   if (dst_lo_rc == rc_int && src_lo_rc == rc_stack) {
1817     if (src_hi != OptoReg::Bad) {
1818       return ppc64Opcode_ld;
1819     } else {
1820       return ppc64Opcode_lwz;
1821     }
1822   }
1823 
1824   // Check for float reg-reg copy.
1825   if (src_lo_rc == rc_float && dst_lo_rc == rc_float) {
1826     return ppc64Opcode_fmr;
1827   }
1828 
1829   // Check for float store.
1830   if (src_lo_rc == rc_float && dst_lo_rc == rc_stack) {
1831     if (src_hi != OptoReg::Bad) {
1832       return ppc64Opcode_stfd;
1833     } else {
1834       return ppc64Opcode_stfs;
1835     }
1836   }
1837 
1838   // Check for float load.
1839   if (dst_lo_rc == rc_float && src_lo_rc == rc_stack) {
1840     if (src_hi != OptoReg::Bad) {
1841       return ppc64Opcode_lfd;
1842     } else {
1843       return ppc64Opcode_lfs;
1844     }
1845   }
1846 
1847   // --------------------------------------------------------------------
1848   // Check for hi bits still needing moving. Only happens for misaligned
1849   // arguments to native calls.
1850   if (src_hi == dst_hi) {
1851     return ppc64Opcode_none;               // Self copy; no move.
1852   }
1853 
1854   ShouldNotReachHere();
1855   return ppc64Opcode_undefined;
1856 }
1857 #endif // PPC port
1858 
1859 #ifndef PRODUCT
1860 void MachNopNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1861   st->print("NOP \t// %d nops to pad for loops.", _count);
1862 }
1863 #endif
1864 
1865 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *) const {
1866   MacroAssembler _masm(&cbuf);
1867   // _count contains the number of nops needed for padding.
1868   for (int i = 0; i < _count; i++) {
1869     __ nop();
1870   }
1871 }
1872 
1873 uint MachNopNode::size(PhaseRegAlloc *ra_) const {
1874   return _count * 4;
1875 }
1876 
1877 #ifndef PRODUCT
1878 void BoxLockNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1879   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1880   char reg_str[128];
1881   ra_->dump_register(this, reg_str);
1882   st->print("ADDI    %s, SP, %d \t// box node", reg_str, offset);
1883 }
1884 #endif
1885 
1886 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1887   MacroAssembler _masm(&cbuf);
1888 
1889   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1890   int reg    = ra_->get_encode(this);
1891 
1892   if (Assembler::is_simm(offset, 16)) {
1893     __ addi(as_Register(reg), R1, offset);
1894   } else {
1895     ShouldNotReachHere();
1896   }
1897 }
1898 
1899 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1900   // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_).
1901   return 4;
1902 }
1903 
1904 #ifndef PRODUCT
1905 void MachUEPNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
1906   st->print_cr("---- MachUEPNode ----");
1907   st->print_cr("...");
1908 }
1909 #endif
1910 
1911 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1912   // This is the unverified entry point.
1913   MacroAssembler _masm(&cbuf);
1914 
1915   // Inline_cache contains a klass.
1916   Register ic_klass       = as_Register(Matcher::inline_cache_reg_encode());
1917   Register receiver_klass = R12_scratch2;  // tmp
1918 
1919   assert_different_registers(ic_klass, receiver_klass, R11_scratch1, R3_ARG1);
1920   assert(R11_scratch1 == R11, "need prologue scratch register");
1921 
1922   // Check for NULL argument if we don't have implicit null checks.
1923   if (!ImplicitNullChecks || !os::zero_page_read_protected()) {
1924     if (TrapBasedNullChecks) {
1925       __ trap_null_check(R3_ARG1);
1926     } else {
1927       Label valid;
1928       __ cmpdi(CCR0, R3_ARG1, 0);
1929       __ bne_predict_taken(CCR0, valid);
1930       // We have a null argument, branch to ic_miss_stub.
1931       __ b64_patchable((address)SharedRuntime::get_ic_miss_stub(),
1932                            relocInfo::runtime_call_type);
1933       __ bind(valid);
1934     }
1935   }
1936   // Assume argument is not NULL, load klass from receiver.
1937   __ load_klass(receiver_klass, R3_ARG1);
1938 
1939   if (TrapBasedICMissChecks) {
1940     __ trap_ic_miss_check(receiver_klass, ic_klass);
1941   } else {
1942     Label valid;
1943     __ cmpd(CCR0, receiver_klass, ic_klass);
1944     __ beq_predict_taken(CCR0, valid);
1945     // We have an unexpected klass, branch to ic_miss_stub.
1946     __ b64_patchable((address)SharedRuntime::get_ic_miss_stub(),
1947                          relocInfo::runtime_call_type);
1948     __ bind(valid);
1949   }
1950 
1951   // Argument is valid and klass is as expected, continue.
1952 }
1953 
1954 #if 0 // TODO: PPC port
1955 // Optimize UEP code on z (save a load_const() call in main path).
1956 int MachUEPNode::ep_offset() {
1957   return 0;
1958 }
1959 #endif
1960 
1961 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1962   // Variable size. Determine dynamically.
1963   return MachNode::size(ra_);
1964 }
1965 
1966 //=============================================================================
1967 
1968 %} // interrupt source
1969 
1970 source_hpp %{ // Header information of the source block.
1971 
1972 class HandlerImpl {
1973 
1974  public:
1975 
1976   static int emit_exception_handler(CodeBuffer &cbuf);
1977   static int emit_deopt_handler(CodeBuffer& cbuf);
1978 
1979   static uint size_exception_handler() {
1980     // The exception_handler is a b64_patchable.
1981     return MacroAssembler::b64_patchable_size;
1982   }
1983 
1984   static uint size_deopt_handler() {
1985     // The deopt_handler is a bl64_patchable.
1986     return MacroAssembler::bl64_patchable_size;
1987   }
1988 
1989 };
1990 
1991 %} // end source_hpp
1992 
1993 source %{
1994 
1995 int HandlerImpl::emit_exception_handler(CodeBuffer &cbuf) {
1996   MacroAssembler _masm(&cbuf);
1997 
1998   address base = __ start_a_stub(size_exception_handler());
1999   if (base == NULL) return 0; // CodeBuffer::expand failed
2000 
2001   int offset = __ offset();
2002   __ b64_patchable((address)OptoRuntime::exception_blob()->content_begin(),
2003                        relocInfo::runtime_call_type);
2004   assert(__ offset() - offset == (int)size_exception_handler(), "must be fixed size");
2005   __ end_a_stub();
2006 
2007   return offset;
2008 }
2009 
2010 // The deopt_handler is like the exception handler, but it calls to
2011 // the deoptimization blob instead of jumping to the exception blob.
2012 int HandlerImpl::emit_deopt_handler(CodeBuffer& cbuf) {
2013   MacroAssembler _masm(&cbuf);
2014 
2015   address base = __ start_a_stub(size_deopt_handler());
2016   if (base == NULL) return 0; // CodeBuffer::expand failed
2017 
2018   int offset = __ offset();
2019   __ bl64_patchable((address)SharedRuntime::deopt_blob()->unpack(),
2020                         relocInfo::runtime_call_type);
2021   assert(__ offset() - offset == (int) size_deopt_handler(), "must be fixed size");
2022   __ end_a_stub();
2023 
2024   return offset;
2025 }
2026 
2027 //=============================================================================
2028 
2029 // Use a frame slots bias for frameless methods if accessing the stack.
2030 static int frame_slots_bias(int reg_enc, PhaseRegAlloc* ra_) {
2031   if (as_Register(reg_enc) == R1_SP) {
2032     return 0; // TODO: PPC port ra_->C->frame_slots_sp_bias_in_bytes();
2033   }
2034   return 0;
2035 }
2036 
2037 const bool Matcher::match_rule_supported(int opcode) {
2038   if (!has_match_rule(opcode))
2039     return false;
2040 
2041   switch (opcode) {
2042   case Op_SqrtD:
2043     return VM_Version::has_fsqrt();
2044   case Op_CountLeadingZerosI:
2045   case Op_CountLeadingZerosL:
2046   case Op_CountTrailingZerosI:
2047   case Op_CountTrailingZerosL:
2048     if (!UseCountLeadingZerosInstructionsPPC64)
2049       return false;
2050     break;
2051 
2052   case Op_PopCountI:
2053   case Op_PopCountL:
2054     return (UsePopCountInstruction && VM_Version::has_popcntw());
2055 
2056   case Op_StrComp:
2057     return SpecialStringCompareTo;
2058   case Op_StrEquals:
2059     return SpecialStringEquals;
2060   case Op_StrIndexOf:
2061     return SpecialStringIndexOf;
2062   }
2063 
2064   return true;  // Per default match rules are supported.
2065 }
2066 
2067 const int Matcher::float_pressure(int default_pressure_threshold) {
2068   return default_pressure_threshold;
2069 }
2070 
2071 int Matcher::regnum_to_fpu_offset(int regnum) {
2072   // No user for this method?
2073   Unimplemented();
2074   return 999;
2075 }
2076 
2077 const bool Matcher::convL2FSupported(void) {
2078   // fcfids can do the conversion (>= Power7).
2079   // fcfid + frsp showed rounding problem when result should be 0x3f800001.
2080   return VM_Version::has_fcfids(); // False means that conversion is done by runtime call.
2081 }
2082 
2083 // Vector width in bytes.
2084 const int Matcher::vector_width_in_bytes(BasicType bt) {
2085   assert(MaxVectorSize == 8, "");
2086   return 8;
2087 }
2088 
2089 // Vector ideal reg.
2090 const int Matcher::vector_ideal_reg(int size) {
2091   assert(MaxVectorSize == 8 && size == 8, "");
2092   return Op_RegL;
2093 }
2094 
2095 const int Matcher::vector_shift_count_ideal_reg(int size) {
2096   fatal("vector shift is not supported");
2097   return Node::NotAMachineReg;
2098 }
2099 
2100 // Limits on vector size (number of elements) loaded into vector.
2101 const int Matcher::max_vector_size(const BasicType bt) {
2102   assert(is_java_primitive(bt), "only primitive type vectors");
2103   return vector_width_in_bytes(bt)/type2aelembytes(bt);
2104 }
2105 
2106 const int Matcher::min_vector_size(const BasicType bt) {
2107   return max_vector_size(bt); // Same as max.
2108 }
2109 
2110 // PPC doesn't support misaligned vectors store/load.
2111 const bool Matcher::misaligned_vectors_ok() {
2112   return false;
2113 }
2114 
2115 // PPC AES support not yet implemented
2116 const bool Matcher::pass_original_key_for_aes() {
2117   return false;
2118 }
2119 
2120 // RETURNS: whether this branch offset is short enough that a short
2121 // branch can be used.
2122 //
2123 // If the platform does not provide any short branch variants, then
2124 // this method should return `false' for offset 0.
2125 //
2126 // `Compile::Fill_buffer' will decide on basis of this information
2127 // whether to do the pass `Compile::Shorten_branches' at all.
2128 //
2129 // And `Compile::Shorten_branches' will decide on basis of this
2130 // information whether to replace particular branch sites by short
2131 // ones.
2132 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
2133   // Is the offset within the range of a ppc64 pc relative branch?
2134   bool b;
2135 
2136   const int safety_zone = 3 * BytesPerInstWord;
2137   b = Assembler::is_simm((offset<0 ? offset-safety_zone : offset+safety_zone),
2138                          29 - 16 + 1 + 2);
2139   return b;
2140 }
2141 
2142 const bool Matcher::isSimpleConstant64(jlong value) {
2143   // Probably always true, even if a temp register is required.
2144   return true;
2145 }
2146 /* TODO: PPC port
2147 // Make a new machine dependent decode node (with its operands).
2148 MachTypeNode *Matcher::make_decode_node() {
2149   assert(Universe::narrow_oop_base() == NULL && Universe::narrow_oop_shift() == 0,
2150          "This method is only implemented for unscaled cOops mode so far");
2151   MachTypeNode *decode = new decodeN_unscaledNode();
2152   decode->set_opnd_array(0, new iRegPdstOper());
2153   decode->set_opnd_array(1, new iRegNsrcOper());
2154   return decode;
2155 }
2156 */
2157 // Threshold size for cleararray.
2158 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2159 
2160 // false => size gets scaled to BytesPerLong, ok.
2161 const bool Matcher::init_array_count_is_in_bytes = false;
2162 
2163 // Use conditional move (CMOVL) on Power7.
2164 const int Matcher::long_cmove_cost() { return 0; } // this only makes long cmoves more expensive than int cmoves
2165 
2166 // Suppress CMOVF. Conditional move available (sort of) on PPC64 only from P7 onwards. Not exploited yet.
2167 // fsel doesn't accept a condition register as input, so this would be slightly different.
2168 const int Matcher::float_cmove_cost() { return ConditionalMoveLimit; }
2169 
2170 // Power6 requires postalloc expand (see block.cpp for description of postalloc expand).
2171 const bool Matcher::require_postalloc_expand = true;
2172 
2173 // Should the Matcher clone shifts on addressing modes, expecting them to
2174 // be subsumed into complex addressing expressions or compute them into
2175 // registers? True for Intel but false for most RISCs.
2176 const bool Matcher::clone_shift_expressions = false;
2177 
2178 // Do we need to mask the count passed to shift instructions or does
2179 // the cpu only look at the lower 5/6 bits anyway?
2180 // PowerPC requires masked shift counts.
2181 const bool Matcher::need_masked_shift_count = true;
2182 
2183 // This affects two different things:
2184 //  - how Decode nodes are matched
2185 //  - how ImplicitNullCheck opportunities are recognized
2186 // If true, the matcher will try to remove all Decodes and match them
2187 // (as operands) into nodes. NullChecks are not prepared to deal with
2188 // Decodes by final_graph_reshaping().
2189 // If false, final_graph_reshaping() forces the decode behind the Cmp
2190 // for a NullCheck. The matcher matches the Decode node into a register.
2191 // Implicit_null_check optimization moves the Decode along with the
2192 // memory operation back up before the NullCheck.
2193 bool Matcher::narrow_oop_use_complex_address() {
2194   // TODO: PPC port if (MatchDecodeNodes) return true;
2195   return false;
2196 }
2197 
2198 bool Matcher::narrow_klass_use_complex_address() {
2199   NOT_LP64(ShouldNotCallThis());
2200   assert(UseCompressedClassPointers, "only for compressed klass code");
2201   // TODO: PPC port if (MatchDecodeNodes) return true;
2202   return false;
2203 }
2204 
2205 // Is it better to copy float constants, or load them directly from memory?
2206 // Intel can load a float constant from a direct address, requiring no
2207 // extra registers. Most RISCs will have to materialize an address into a
2208 // register first, so they would do better to copy the constant from stack.
2209 const bool Matcher::rematerialize_float_constants = false;
2210 
2211 // If CPU can load and store mis-aligned doubles directly then no fixup is
2212 // needed. Else we split the double into 2 integer pieces and move it
2213 // piece-by-piece. Only happens when passing doubles into C code as the
2214 // Java calling convention forces doubles to be aligned.
2215 const bool Matcher::misaligned_doubles_ok = true;
2216 
2217 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
2218  Unimplemented();
2219 }
2220 
2221 // Advertise here if the CPU requires explicit rounding operations
2222 // to implement the UseStrictFP mode.
2223 const bool Matcher::strict_fp_requires_explicit_rounding = false;
2224 
2225 // Do floats take an entire double register or just half?
2226 //
2227 // A float occupies a ppc64 double register. For the allocator, a
2228 // ppc64 double register appears as a pair of float registers.
2229 bool Matcher::float_in_double() { return true; }
2230 
2231 // Do ints take an entire long register or just half?
2232 // The relevant question is how the int is callee-saved:
2233 // the whole long is written but de-opt'ing will have to extract
2234 // the relevant 32 bits.
2235 const bool Matcher::int_in_long = true;
2236 
2237 // Constants for c2c and c calling conventions.
2238 
2239 const MachRegisterNumbers iarg_reg[8] = {
2240   R3_num, R4_num, R5_num, R6_num,
2241   R7_num, R8_num, R9_num, R10_num
2242 };
2243 
2244 const MachRegisterNumbers farg_reg[13] = {
2245   F1_num, F2_num, F3_num, F4_num,
2246   F5_num, F6_num, F7_num, F8_num,
2247   F9_num, F10_num, F11_num, F12_num,
2248   F13_num
2249 };
2250 
2251 const int num_iarg_registers = sizeof(iarg_reg) / sizeof(iarg_reg[0]);
2252 
2253 const int num_farg_registers = sizeof(farg_reg) / sizeof(farg_reg[0]);
2254 
2255 // Return whether or not this register is ever used as an argument. This
2256 // function is used on startup to build the trampoline stubs in generateOptoStub.
2257 // Registers not mentioned will be killed by the VM call in the trampoline, and
2258 // arguments in those registers not be available to the callee.
2259 bool Matcher::can_be_java_arg(int reg) {
2260   // We return true for all registers contained in iarg_reg[] and
2261   // farg_reg[] and their virtual halves.
2262   // We must include the virtual halves in order to get STDs and LDs
2263   // instead of STWs and LWs in the trampoline stubs.
2264 
2265   if (   reg == R3_num  || reg == R3_H_num
2266       || reg == R4_num  || reg == R4_H_num
2267       || reg == R5_num  || reg == R5_H_num
2268       || reg == R6_num  || reg == R6_H_num
2269       || reg == R7_num  || reg == R7_H_num
2270       || reg == R8_num  || reg == R8_H_num
2271       || reg == R9_num  || reg == R9_H_num
2272       || reg == R10_num || reg == R10_H_num)
2273     return true;
2274 
2275   if (   reg == F1_num  || reg == F1_H_num
2276       || reg == F2_num  || reg == F2_H_num
2277       || reg == F3_num  || reg == F3_H_num
2278       || reg == F4_num  || reg == F4_H_num
2279       || reg == F5_num  || reg == F5_H_num
2280       || reg == F6_num  || reg == F6_H_num
2281       || reg == F7_num  || reg == F7_H_num
2282       || reg == F8_num  || reg == F8_H_num
2283       || reg == F9_num  || reg == F9_H_num
2284       || reg == F10_num || reg == F10_H_num
2285       || reg == F11_num || reg == F11_H_num
2286       || reg == F12_num || reg == F12_H_num
2287       || reg == F13_num || reg == F13_H_num)
2288     return true;
2289 
2290   return false;
2291 }
2292 
2293 bool Matcher::is_spillable_arg(int reg) {
2294   return can_be_java_arg(reg);
2295 }
2296 
2297 bool Matcher::use_asm_for_ldiv_by_con(jlong divisor) {
2298   return false;
2299 }
2300 
2301 // Register for DIVI projection of divmodI.
2302 RegMask Matcher::divI_proj_mask() {
2303   ShouldNotReachHere();
2304   return RegMask();
2305 }
2306 
2307 // Register for MODI projection of divmodI.
2308 RegMask Matcher::modI_proj_mask() {
2309   ShouldNotReachHere();
2310   return RegMask();
2311 }
2312 
2313 // Register for DIVL projection of divmodL.
2314 RegMask Matcher::divL_proj_mask() {
2315   ShouldNotReachHere();
2316   return RegMask();
2317 }
2318 
2319 // Register for MODL projection of divmodL.
2320 RegMask Matcher::modL_proj_mask() {
2321   ShouldNotReachHere();
2322   return RegMask();
2323 }
2324 
2325 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2326   return RegMask();
2327 }
2328 
2329 %}
2330 
2331 //----------ENCODING BLOCK-----------------------------------------------------
2332 // This block specifies the encoding classes used by the compiler to output
2333 // byte streams. Encoding classes are parameterized macros used by
2334 // Machine Instruction Nodes in order to generate the bit encoding of the
2335 // instruction. Operands specify their base encoding interface with the
2336 // interface keyword. There are currently supported four interfaces,
2337 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER. REG_INTER causes an
2338 // operand to generate a function which returns its register number when
2339 // queried. CONST_INTER causes an operand to generate a function which
2340 // returns the value of the constant when queried. MEMORY_INTER causes an
2341 // operand to generate four functions which return the Base Register, the
2342 // Index Register, the Scale Value, and the Offset Value of the operand when
2343 // queried. COND_INTER causes an operand to generate six functions which
2344 // return the encoding code (ie - encoding bits for the instruction)
2345 // associated with each basic boolean condition for a conditional instruction.
2346 //
2347 // Instructions specify two basic values for encoding. Again, a function
2348 // is available to check if the constant displacement is an oop. They use the
2349 // ins_encode keyword to specify their encoding classes (which must be
2350 // a sequence of enc_class names, and their parameters, specified in
2351 // the encoding block), and they use the
2352 // opcode keyword to specify, in order, their primary, secondary, and
2353 // tertiary opcode. Only the opcode sections which a particular instruction
2354 // needs for encoding need to be specified.
2355 encode %{
2356   enc_class enc_unimplemented %{
2357     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2358     MacroAssembler _masm(&cbuf);
2359     __ unimplemented("Unimplemented mach node encoding in AD file.", 13);
2360   %}
2361 
2362   enc_class enc_untested %{
2363 #ifdef ASSERT
2364     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2365     MacroAssembler _masm(&cbuf);
2366     __ untested("Untested mach node encoding in AD file.");
2367 #else
2368     // TODO: PPC port $archOpcode(ppc64Opcode_none);
2369 #endif
2370   %}
2371 
2372   enc_class enc_lbz(iRegIdst dst, memory mem) %{
2373     // TODO: PPC port $archOpcode(ppc64Opcode_lbz);
2374     MacroAssembler _masm(&cbuf);
2375     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2376     __ lbz($dst$$Register, Idisp, $mem$$base$$Register);
2377   %}
2378 
2379   // Load acquire.
2380   enc_class enc_lbz_ac(iRegIdst dst, memory mem) %{
2381     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2382     MacroAssembler _masm(&cbuf);
2383     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2384     __ lbz($dst$$Register, Idisp, $mem$$base$$Register);
2385     __ twi_0($dst$$Register);
2386     __ isync();
2387   %}
2388 
2389   enc_class enc_lhz(iRegIdst dst, memory mem) %{
2390     // TODO: PPC port $archOpcode(ppc64Opcode_lhz);
2391 
2392     MacroAssembler _masm(&cbuf);
2393     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2394     __ lhz($dst$$Register, Idisp, $mem$$base$$Register);
2395   %}
2396 
2397   // Load acquire.
2398   enc_class enc_lhz_ac(iRegIdst dst, memory mem) %{
2399     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2400 
2401     MacroAssembler _masm(&cbuf);
2402     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2403     __ lhz($dst$$Register, Idisp, $mem$$base$$Register);
2404     __ twi_0($dst$$Register);
2405     __ isync();
2406   %}
2407 
2408   enc_class enc_lwz(iRegIdst dst, memory mem) %{
2409     // TODO: PPC port $archOpcode(ppc64Opcode_lwz);
2410 
2411     MacroAssembler _masm(&cbuf);
2412     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2413     __ lwz($dst$$Register, Idisp, $mem$$base$$Register);
2414   %}
2415 
2416   // Load acquire.
2417   enc_class enc_lwz_ac(iRegIdst dst, memory mem) %{
2418     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2419 
2420     MacroAssembler _masm(&cbuf);
2421     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2422     __ lwz($dst$$Register, Idisp, $mem$$base$$Register);
2423     __ twi_0($dst$$Register);
2424     __ isync();
2425   %}
2426 
2427   enc_class enc_ld(iRegLdst dst, memoryAlg4 mem) %{
2428     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
2429     MacroAssembler _masm(&cbuf);
2430     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2431     // Operand 'ds' requires 4-alignment.
2432     assert((Idisp & 0x3) == 0, "unaligned offset");
2433     __ ld($dst$$Register, Idisp, $mem$$base$$Register);
2434   %}
2435 
2436   // Load acquire.
2437   enc_class enc_ld_ac(iRegLdst dst, memoryAlg4 mem) %{
2438     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2439     MacroAssembler _masm(&cbuf);
2440     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2441     // Operand 'ds' requires 4-alignment.
2442     assert((Idisp & 0x3) == 0, "unaligned offset");
2443     __ ld($dst$$Register, Idisp, $mem$$base$$Register);
2444     __ twi_0($dst$$Register);
2445     __ isync();
2446   %}
2447 
2448   enc_class enc_lfd(RegF dst, memory mem) %{
2449     // TODO: PPC port $archOpcode(ppc64Opcode_lfd);
2450     MacroAssembler _masm(&cbuf);
2451     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2452     __ lfd($dst$$FloatRegister, Idisp, $mem$$base$$Register);
2453   %}
2454 
2455   enc_class enc_load_long_constL(iRegLdst dst, immL src, iRegLdst toc) %{
2456     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
2457 
2458     MacroAssembler _masm(&cbuf);
2459     int toc_offset = 0;
2460 
2461     if (!ra_->C->in_scratch_emit_size()) {
2462       address const_toc_addr;
2463       // Create a non-oop constant, no relocation needed.
2464       // If it is an IC, it has a virtual_call_Relocation.
2465       const_toc_addr = __ long_constant((jlong)$src$$constant);
2466 
2467       // Get the constant's TOC offset.
2468       toc_offset = __ offset_to_method_toc(const_toc_addr);
2469 
2470       // Keep the current instruction offset in mind.
2471       ((loadConLNode*)this)->_cbuf_insts_offset = __ offset();
2472     }
2473 
2474     __ ld($dst$$Register, toc_offset, $toc$$Register);
2475   %}
2476 
2477   enc_class enc_load_long_constL_hi(iRegLdst dst, iRegLdst toc, immL src) %{
2478     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
2479 
2480     MacroAssembler _masm(&cbuf);
2481 
2482     if (!ra_->C->in_scratch_emit_size()) {
2483       address const_toc_addr;
2484       // Create a non-oop constant, no relocation needed.
2485       // If it is an IC, it has a virtual_call_Relocation.
2486       const_toc_addr = __ long_constant((jlong)$src$$constant);
2487 
2488       // Get the constant's TOC offset.
2489       const int toc_offset = __ offset_to_method_toc(const_toc_addr);
2490       // Store the toc offset of the constant.
2491       ((loadConL_hiNode*)this)->_const_toc_offset = toc_offset;
2492 
2493       // Also keep the current instruction offset in mind.
2494       ((loadConL_hiNode*)this)->_cbuf_insts_offset = __ offset();
2495     }
2496 
2497     __ addis($dst$$Register, $toc$$Register, MacroAssembler::largeoffset_si16_si16_hi(_const_toc_offset));
2498   %}
2499 
2500 %} // encode
2501 
2502 source %{
2503 
2504 typedef struct {
2505   loadConL_hiNode *_large_hi;
2506   loadConL_loNode *_large_lo;
2507   loadConLNode    *_small;
2508   MachNode        *_last;
2509 } loadConLNodesTuple;
2510 
2511 loadConLNodesTuple loadConLNodesTuple_create(PhaseRegAlloc *ra_, Node *toc, immLOper *immSrc,
2512                                              OptoReg::Name reg_second, OptoReg::Name reg_first) {
2513   loadConLNodesTuple nodes;
2514 
2515   const bool large_constant_pool = true; // TODO: PPC port C->cfg()->_consts_size > 4000;
2516   if (large_constant_pool) {
2517     // Create new nodes.
2518     loadConL_hiNode *m1 = new loadConL_hiNode();
2519     loadConL_loNode *m2 = new loadConL_loNode();
2520 
2521     // inputs for new nodes
2522     m1->add_req(NULL, toc);
2523     m2->add_req(NULL, m1);
2524 
2525     // operands for new nodes
2526     m1->_opnds[0] = new iRegLdstOper(); // dst
2527     m1->_opnds[1] = immSrc;             // src
2528     m1->_opnds[2] = new iRegPdstOper(); // toc
2529     m2->_opnds[0] = new iRegLdstOper(); // dst
2530     m2->_opnds[1] = immSrc;             // src
2531     m2->_opnds[2] = new iRegLdstOper(); // base
2532 
2533     // Initialize ins_attrib TOC fields.
2534     m1->_const_toc_offset = -1;
2535     m2->_const_toc_offset_hi_node = m1;
2536 
2537     // Initialize ins_attrib instruction offset.
2538     m1->_cbuf_insts_offset = -1;
2539 
2540     // register allocation for new nodes
2541     ra_->set_pair(m1->_idx, reg_second, reg_first);
2542     ra_->set_pair(m2->_idx, reg_second, reg_first);
2543 
2544     // Create result.
2545     nodes._large_hi = m1;
2546     nodes._large_lo = m2;
2547     nodes._small = NULL;
2548     nodes._last = nodes._large_lo;
2549     assert(m2->bottom_type()->isa_long(), "must be long");
2550   } else {
2551     loadConLNode *m2 = new loadConLNode();
2552 
2553     // inputs for new nodes
2554     m2->add_req(NULL, toc);
2555 
2556     // operands for new nodes
2557     m2->_opnds[0] = new iRegLdstOper(); // dst
2558     m2->_opnds[1] = immSrc;             // src
2559     m2->_opnds[2] = new iRegPdstOper(); // toc
2560 
2561     // Initialize ins_attrib instruction offset.
2562     m2->_cbuf_insts_offset = -1;
2563 
2564     // register allocation for new nodes
2565     ra_->set_pair(m2->_idx, reg_second, reg_first);
2566 
2567     // Create result.
2568     nodes._large_hi = NULL;
2569     nodes._large_lo = NULL;
2570     nodes._small = m2;
2571     nodes._last = nodes._small;
2572     assert(m2->bottom_type()->isa_long(), "must be long");
2573   }
2574 
2575   return nodes;
2576 }
2577 
2578 %} // source
2579 
2580 encode %{
2581   // Postalloc expand emitter for loading a long constant from the method's TOC.
2582   // Enc_class needed as consttanttablebase is not supported by postalloc
2583   // expand.
2584   enc_class postalloc_expand_load_long_constant(iRegLdst dst, immL src, iRegLdst toc) %{
2585     // Create new nodes.
2586     loadConLNodesTuple loadConLNodes =
2587       loadConLNodesTuple_create(ra_, n_toc, op_src,
2588                                 ra_->get_reg_second(this), ra_->get_reg_first(this));
2589 
2590     // Push new nodes.
2591     if (loadConLNodes._large_hi) nodes->push(loadConLNodes._large_hi);
2592     if (loadConLNodes._last)     nodes->push(loadConLNodes._last);
2593 
2594     // some asserts
2595     assert(nodes->length() >= 1, "must have created at least 1 node");
2596     assert(loadConLNodes._last->bottom_type()->isa_long(), "must be long");
2597   %}
2598 
2599   enc_class enc_load_long_constP(iRegLdst dst, immP src, iRegLdst toc) %{
2600     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
2601 
2602     MacroAssembler _masm(&cbuf);
2603     int toc_offset = 0;
2604 
2605     if (!ra_->C->in_scratch_emit_size()) {
2606       intptr_t val = $src$$constant;
2607       relocInfo::relocType constant_reloc = $src->constant_reloc();  // src
2608       address const_toc_addr;
2609       if (constant_reloc == relocInfo::oop_type) {
2610         // Create an oop constant and a corresponding relocation.
2611         AddressLiteral a = __ allocate_oop_address((jobject)val);
2612         const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
2613         __ relocate(a.rspec());
2614       } else if (constant_reloc == relocInfo::metadata_type) {
2615         AddressLiteral a = __ constant_metadata_address((Metadata *)val);
2616         const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
2617         __ relocate(a.rspec());
2618       } else {
2619         // Create a non-oop constant, no relocation needed.
2620         const_toc_addr = __ long_constant((jlong)$src$$constant);
2621       }
2622 
2623       // Get the constant's TOC offset.
2624       toc_offset = __ offset_to_method_toc(const_toc_addr);
2625     }
2626 
2627     __ ld($dst$$Register, toc_offset, $toc$$Register);
2628   %}
2629 
2630   enc_class enc_load_long_constP_hi(iRegLdst dst, immP src, iRegLdst toc) %{
2631     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
2632 
2633     MacroAssembler _masm(&cbuf);
2634     if (!ra_->C->in_scratch_emit_size()) {
2635       intptr_t val = $src$$constant;
2636       relocInfo::relocType constant_reloc = $src->constant_reloc();  // src
2637       address const_toc_addr;
2638       if (constant_reloc == relocInfo::oop_type) {
2639         // Create an oop constant and a corresponding relocation.
2640         AddressLiteral a = __ allocate_oop_address((jobject)val);
2641         const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
2642         __ relocate(a.rspec());
2643       } else if (constant_reloc == relocInfo::metadata_type) {
2644         AddressLiteral a = __ constant_metadata_address((Metadata *)val);
2645         const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
2646         __ relocate(a.rspec());
2647       } else {  // non-oop pointers, e.g. card mark base, heap top
2648         // Create a non-oop constant, no relocation needed.
2649         const_toc_addr = __ long_constant((jlong)$src$$constant);
2650       }
2651 
2652       // Get the constant's TOC offset.
2653       const int toc_offset = __ offset_to_method_toc(const_toc_addr);
2654       // Store the toc offset of the constant.
2655       ((loadConP_hiNode*)this)->_const_toc_offset = toc_offset;
2656     }
2657 
2658     __ addis($dst$$Register, $toc$$Register, MacroAssembler::largeoffset_si16_si16_hi(_const_toc_offset));
2659   %}
2660 
2661   // Postalloc expand emitter for loading a ptr constant from the method's TOC.
2662   // Enc_class needed as consttanttablebase is not supported by postalloc
2663   // expand.
2664   enc_class postalloc_expand_load_ptr_constant(iRegPdst dst, immP src, iRegLdst toc) %{
2665     const bool large_constant_pool = true; // TODO: PPC port C->cfg()->_consts_size > 4000;
2666     if (large_constant_pool) {
2667       // Create new nodes.
2668       loadConP_hiNode *m1 = new loadConP_hiNode();
2669       loadConP_loNode *m2 = new loadConP_loNode();
2670 
2671       // inputs for new nodes
2672       m1->add_req(NULL, n_toc);
2673       m2->add_req(NULL, m1);
2674 
2675       // operands for new nodes
2676       m1->_opnds[0] = new iRegPdstOper(); // dst
2677       m1->_opnds[1] = op_src;             // src
2678       m1->_opnds[2] = new iRegPdstOper(); // toc
2679       m2->_opnds[0] = new iRegPdstOper(); // dst
2680       m2->_opnds[1] = op_src;             // src
2681       m2->_opnds[2] = new iRegLdstOper(); // base
2682 
2683       // Initialize ins_attrib TOC fields.
2684       m1->_const_toc_offset = -1;
2685       m2->_const_toc_offset_hi_node = m1;
2686 
2687       // Register allocation for new nodes.
2688       ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2689       ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2690 
2691       nodes->push(m1);
2692       nodes->push(m2);
2693       assert(m2->bottom_type()->isa_ptr(), "must be ptr");
2694     } else {
2695       loadConPNode *m2 = new loadConPNode();
2696 
2697       // inputs for new nodes
2698       m2->add_req(NULL, n_toc);
2699 
2700       // operands for new nodes
2701       m2->_opnds[0] = new iRegPdstOper(); // dst
2702       m2->_opnds[1] = op_src;             // src
2703       m2->_opnds[2] = new iRegPdstOper(); // toc
2704 
2705       // Register allocation for new nodes.
2706       ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2707 
2708       nodes->push(m2);
2709       assert(m2->bottom_type()->isa_ptr(), "must be ptr");
2710     }
2711   %}
2712 
2713   // Enc_class needed as consttanttablebase is not supported by postalloc
2714   // expand.
2715   enc_class postalloc_expand_load_float_constant(regF dst, immF src, iRegLdst toc) %{
2716     bool large_constant_pool = true; // TODO: PPC port C->cfg()->_consts_size > 4000;
2717 
2718     MachNode *m2;
2719     if (large_constant_pool) {
2720       m2 = new loadConFCompNode();
2721     } else {
2722       m2 = new loadConFNode();
2723     }
2724     // inputs for new nodes
2725     m2->add_req(NULL, n_toc);
2726 
2727     // operands for new nodes
2728     m2->_opnds[0] = op_dst;
2729     m2->_opnds[1] = op_src;
2730     m2->_opnds[2] = new iRegPdstOper(); // constanttablebase
2731 
2732     // register allocation for new nodes
2733     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2734     nodes->push(m2);
2735   %}
2736 
2737   // Enc_class needed as consttanttablebase is not supported by postalloc
2738   // expand.
2739   enc_class postalloc_expand_load_double_constant(regD dst, immD src, iRegLdst toc) %{
2740     bool large_constant_pool = true; // TODO: PPC port C->cfg()->_consts_size > 4000;
2741 
2742     MachNode *m2;
2743     if (large_constant_pool) {
2744       m2 = new loadConDCompNode();
2745     } else {
2746       m2 = new loadConDNode();
2747     }
2748     // inputs for new nodes
2749     m2->add_req(NULL, n_toc);
2750 
2751     // operands for new nodes
2752     m2->_opnds[0] = op_dst;
2753     m2->_opnds[1] = op_src;
2754     m2->_opnds[2] = new iRegPdstOper(); // constanttablebase
2755 
2756     // register allocation for new nodes
2757     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2758     nodes->push(m2);
2759   %}
2760 
2761   enc_class enc_stw(iRegIsrc src, memory mem) %{
2762     // TODO: PPC port $archOpcode(ppc64Opcode_stw);
2763     MacroAssembler _masm(&cbuf);
2764     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2765     __ stw($src$$Register, Idisp, $mem$$base$$Register);
2766   %}
2767 
2768   enc_class enc_std(iRegIsrc src, memoryAlg4 mem) %{
2769     // TODO: PPC port $archOpcode(ppc64Opcode_std);
2770     MacroAssembler _masm(&cbuf);
2771     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2772     // Operand 'ds' requires 4-alignment.
2773     assert((Idisp & 0x3) == 0, "unaligned offset");
2774     __ std($src$$Register, Idisp, $mem$$base$$Register);
2775   %}
2776 
2777   enc_class enc_stfs(RegF src, memory mem) %{
2778     // TODO: PPC port $archOpcode(ppc64Opcode_stfs);
2779     MacroAssembler _masm(&cbuf);
2780     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2781     __ stfs($src$$FloatRegister, Idisp, $mem$$base$$Register);
2782   %}
2783 
2784   enc_class enc_stfd(RegF src, memory mem) %{
2785     // TODO: PPC port $archOpcode(ppc64Opcode_stfd);
2786     MacroAssembler _masm(&cbuf);
2787     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
2788     __ stfd($src$$FloatRegister, Idisp, $mem$$base$$Register);
2789   %}
2790 
2791   // Use release_store for card-marking to ensure that previous
2792   // oop-stores are visible before the card-mark change.
2793   enc_class enc_cms_card_mark(memory mem, iRegLdst releaseFieldAddr, flagsReg crx) %{
2794     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
2795     // FIXME: Implement this as a cmove and use a fixed condition code
2796     // register which is written on every transition to compiled code,
2797     // e.g. in call-stub and when returning from runtime stubs.
2798     //
2799     // Proposed code sequence for the cmove implementation:
2800     //
2801     // Label skip_release;
2802     // __ beq(CCRfixed, skip_release);
2803     // __ release();
2804     // __ bind(skip_release);
2805     // __ stb(card mark);
2806 
2807     MacroAssembler _masm(&cbuf);
2808     Label skip_storestore;
2809 
2810 #if 0 // TODO: PPC port
2811     // Check CMSCollectorCardTableModRefBSExt::_requires_release and do the
2812     // StoreStore barrier conditionally.
2813     __ lwz(R0, 0, $releaseFieldAddr$$Register);
2814     __ cmpwi($crx$$CondRegister, R0, 0);
2815     __ beq_predict_taken($crx$$CondRegister, skip_storestore);
2816 #endif
2817     __ li(R0, 0);
2818     __ membar(Assembler::StoreStore);
2819 #if 0 // TODO: PPC port
2820     __ bind(skip_storestore);
2821 #endif
2822 
2823     // Do the store.
2824     if ($mem$$index == 0) {
2825       __ stb(R0, $mem$$disp, $mem$$base$$Register);
2826     } else {
2827       assert(0 == $mem$$disp, "no displacement possible with indexed load/stores on ppc");
2828       __ stbx(R0, $mem$$base$$Register, $mem$$index$$Register);
2829     }
2830   %}
2831 
2832   enc_class postalloc_expand_encode_oop(iRegNdst dst, iRegPdst src, flagsReg crx) %{
2833 
2834     if (VM_Version::has_isel()) {
2835       // use isel instruction with Power 7
2836       cmpP_reg_imm16Node *n_compare  = new cmpP_reg_imm16Node();
2837       encodeP_subNode    *n_sub_base = new encodeP_subNode();
2838       encodeP_shiftNode  *n_shift    = new encodeP_shiftNode();
2839       cond_set_0_oopNode *n_cond_set = new cond_set_0_oopNode();
2840 
2841       n_compare->add_req(n_region, n_src);
2842       n_compare->_opnds[0] = op_crx;
2843       n_compare->_opnds[1] = op_src;
2844       n_compare->_opnds[2] = new immL16Oper(0);
2845 
2846       n_sub_base->add_req(n_region, n_src);
2847       n_sub_base->_opnds[0] = op_dst;
2848       n_sub_base->_opnds[1] = op_src;
2849       n_sub_base->_bottom_type = _bottom_type;
2850 
2851       n_shift->add_req(n_region, n_sub_base);
2852       n_shift->_opnds[0] = op_dst;
2853       n_shift->_opnds[1] = op_dst;
2854       n_shift->_bottom_type = _bottom_type;
2855 
2856       n_cond_set->add_req(n_region, n_compare, n_shift);
2857       n_cond_set->_opnds[0] = op_dst;
2858       n_cond_set->_opnds[1] = op_crx;
2859       n_cond_set->_opnds[2] = op_dst;
2860       n_cond_set->_bottom_type = _bottom_type;
2861 
2862       ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
2863       ra_->set_pair(n_sub_base->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2864       ra_->set_pair(n_shift->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2865       ra_->set_pair(n_cond_set->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2866 
2867       nodes->push(n_compare);
2868       nodes->push(n_sub_base);
2869       nodes->push(n_shift);
2870       nodes->push(n_cond_set);
2871 
2872     } else {
2873       // before Power 7
2874       moveRegNode        *n_move     = new moveRegNode();
2875       cmpP_reg_imm16Node *n_compare  = new cmpP_reg_imm16Node();
2876       encodeP_shiftNode  *n_shift    = new encodeP_shiftNode();
2877       cond_sub_baseNode  *n_sub_base = new cond_sub_baseNode();
2878 
2879       n_move->add_req(n_region, n_src);
2880       n_move->_opnds[0] = op_dst;
2881       n_move->_opnds[1] = op_src;
2882       ra_->set_oop(n_move, true); // Until here, 'n_move' still produces an oop.
2883 
2884       n_compare->add_req(n_region, n_src);
2885       n_compare->add_prec(n_move);
2886 
2887       n_compare->_opnds[0] = op_crx;
2888       n_compare->_opnds[1] = op_src;
2889       n_compare->_opnds[2] = new immL16Oper(0);
2890 
2891       n_sub_base->add_req(n_region, n_compare, n_src);
2892       n_sub_base->_opnds[0] = op_dst;
2893       n_sub_base->_opnds[1] = op_crx;
2894       n_sub_base->_opnds[2] = op_src;
2895       n_sub_base->_bottom_type = _bottom_type;
2896 
2897       n_shift->add_req(n_region, n_sub_base);
2898       n_shift->_opnds[0] = op_dst;
2899       n_shift->_opnds[1] = op_dst;
2900       n_shift->_bottom_type = _bottom_type;
2901 
2902       ra_->set_pair(n_shift->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2903       ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
2904       ra_->set_pair(n_sub_base->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2905       ra_->set_pair(n_move->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2906 
2907       nodes->push(n_move);
2908       nodes->push(n_compare);
2909       nodes->push(n_sub_base);
2910       nodes->push(n_shift);
2911     }
2912 
2913     assert(!(ra_->is_oop(this)), "sanity"); // This is not supposed to be GC'ed.
2914   %}
2915 
2916   enc_class postalloc_expand_encode_oop_not_null(iRegNdst dst, iRegPdst src) %{
2917 
2918     encodeP_subNode *n1 = new encodeP_subNode();
2919     n1->add_req(n_region, n_src);
2920     n1->_opnds[0] = op_dst;
2921     n1->_opnds[1] = op_src;
2922     n1->_bottom_type = _bottom_type;
2923 
2924     encodeP_shiftNode *n2 = new encodeP_shiftNode();
2925     n2->add_req(n_region, n1);
2926     n2->_opnds[0] = op_dst;
2927     n2->_opnds[1] = op_dst;
2928     n2->_bottom_type = _bottom_type;
2929     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2930     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2931 
2932     nodes->push(n1);
2933     nodes->push(n2);
2934     assert(!(ra_->is_oop(this)), "sanity"); // This is not supposed to be GC'ed.
2935   %}
2936 
2937   enc_class postalloc_expand_decode_oop(iRegPdst dst, iRegNsrc src, flagsReg crx) %{
2938     decodeN_shiftNode *n_shift    = new decodeN_shiftNode();
2939     cmpN_reg_imm0Node *n_compare  = new cmpN_reg_imm0Node();
2940 
2941     n_compare->add_req(n_region, n_src);
2942     n_compare->_opnds[0] = op_crx;
2943     n_compare->_opnds[1] = op_src;
2944     n_compare->_opnds[2] = new immN_0Oper(TypeNarrowOop::NULL_PTR);
2945 
2946     n_shift->add_req(n_region, n_src);
2947     n_shift->_opnds[0] = op_dst;
2948     n_shift->_opnds[1] = op_src;
2949     n_shift->_bottom_type = _bottom_type;
2950 
2951     if (VM_Version::has_isel()) {
2952       // use isel instruction with Power 7
2953 
2954       decodeN_addNode *n_add_base = new decodeN_addNode();
2955       n_add_base->add_req(n_region, n_shift);
2956       n_add_base->_opnds[0] = op_dst;
2957       n_add_base->_opnds[1] = op_dst;
2958       n_add_base->_bottom_type = _bottom_type;
2959 
2960       cond_set_0_ptrNode *n_cond_set = new cond_set_0_ptrNode();
2961       n_cond_set->add_req(n_region, n_compare, n_add_base);
2962       n_cond_set->_opnds[0] = op_dst;
2963       n_cond_set->_opnds[1] = op_crx;
2964       n_cond_set->_opnds[2] = op_dst;
2965       n_cond_set->_bottom_type = _bottom_type;
2966 
2967       assert(ra_->is_oop(this) == true, "A decodeN node must produce an oop!");
2968       ra_->set_oop(n_cond_set, true);
2969 
2970       ra_->set_pair(n_shift->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2971       ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
2972       ra_->set_pair(n_add_base->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2973       ra_->set_pair(n_cond_set->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2974 
2975       nodes->push(n_compare);
2976       nodes->push(n_shift);
2977       nodes->push(n_add_base);
2978       nodes->push(n_cond_set);
2979 
2980     } else {
2981       // before Power 7
2982       cond_add_baseNode *n_add_base = new cond_add_baseNode();
2983 
2984       n_add_base->add_req(n_region, n_compare, n_shift);
2985       n_add_base->_opnds[0] = op_dst;
2986       n_add_base->_opnds[1] = op_crx;
2987       n_add_base->_opnds[2] = op_dst;
2988       n_add_base->_bottom_type = _bottom_type;
2989 
2990       assert(ra_->is_oop(this) == true, "A decodeN node must produce an oop!");
2991       ra_->set_oop(n_add_base, true);
2992 
2993       ra_->set_pair(n_shift->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2994       ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
2995       ra_->set_pair(n_add_base->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
2996 
2997       nodes->push(n_compare);
2998       nodes->push(n_shift);
2999       nodes->push(n_add_base);
3000     }
3001   %}
3002 
3003   enc_class postalloc_expand_decode_oop_not_null(iRegPdst dst, iRegNsrc src) %{
3004     decodeN_shiftNode *n1 = new decodeN_shiftNode();
3005     n1->add_req(n_region, n_src);
3006     n1->_opnds[0] = op_dst;
3007     n1->_opnds[1] = op_src;
3008     n1->_bottom_type = _bottom_type;
3009 
3010     decodeN_addNode *n2 = new decodeN_addNode();
3011     n2->add_req(n_region, n1);
3012     n2->_opnds[0] = op_dst;
3013     n2->_opnds[1] = op_dst;
3014     n2->_bottom_type = _bottom_type;
3015     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
3016     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
3017 
3018     assert(ra_->is_oop(this) == true, "A decodeN node must produce an oop!");
3019     ra_->set_oop(n2, true);
3020 
3021     nodes->push(n1);
3022     nodes->push(n2);
3023   %}
3024 
3025   enc_class enc_cmove_reg(iRegIdst dst, flagsRegSrc crx, iRegIsrc src, cmpOp cmp) %{
3026     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
3027 
3028     MacroAssembler _masm(&cbuf);
3029     int cc        = $cmp$$cmpcode;
3030     int flags_reg = $crx$$reg;
3031     Label done;
3032     assert((Assembler::bcondCRbiIs1 & ~Assembler::bcondCRbiIs0) == 8, "check encoding");
3033     // Branch if not (cmp crx).
3034     __ bc(cc_to_inverse_boint(cc), cc_to_biint(cc, flags_reg), done);
3035     __ mr($dst$$Register, $src$$Register);
3036     // TODO PPC port __ endgroup_if_needed(_size == 12);
3037     __ bind(done);
3038   %}
3039 
3040   enc_class enc_cmove_imm(iRegIdst dst, flagsRegSrc crx, immI16 src, cmpOp cmp) %{
3041     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
3042 
3043     MacroAssembler _masm(&cbuf);
3044     Label done;
3045     assert((Assembler::bcondCRbiIs1 & ~Assembler::bcondCRbiIs0) == 8, "check encoding");
3046     // Branch if not (cmp crx).
3047     __ bc(cc_to_inverse_boint($cmp$$cmpcode), cc_to_biint($cmp$$cmpcode, $crx$$reg), done);
3048     __ li($dst$$Register, $src$$constant);
3049     // TODO PPC port __ endgroup_if_needed(_size == 12);
3050     __ bind(done);
3051   %}
3052 
3053   // New atomics.
3054   enc_class enc_GetAndAddI(iRegIdst res, iRegPdst mem_ptr, iRegIsrc src) %{
3055     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3056 
3057     MacroAssembler _masm(&cbuf);
3058     Register Rtmp   = R0;
3059     Register Rres   = $res$$Register;
3060     Register Rsrc   = $src$$Register;
3061     Register Rptr   = $mem_ptr$$Register;
3062     bool RegCollision = (Rres == Rsrc) || (Rres == Rptr);
3063     Register Rold   = RegCollision ? Rtmp : Rres;
3064 
3065     Label Lretry;
3066     __ bind(Lretry);
3067     __ lwarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
3068     __ add(Rtmp, Rsrc, Rold);
3069     __ stwcx_(Rtmp, Rptr);
3070     if (UseStaticBranchPredictionInCompareAndSwapPPC64) {
3071       __ bne_predict_not_taken(CCR0, Lretry);
3072     } else {
3073       __ bne(                  CCR0, Lretry);
3074     }
3075     if (RegCollision) __ subf(Rres, Rsrc, Rtmp);
3076     __ fence();
3077   %}
3078 
3079   enc_class enc_GetAndAddL(iRegLdst res, iRegPdst mem_ptr, iRegLsrc src) %{
3080     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3081 
3082     MacroAssembler _masm(&cbuf);
3083     Register Rtmp   = R0;
3084     Register Rres   = $res$$Register;
3085     Register Rsrc   = $src$$Register;
3086     Register Rptr   = $mem_ptr$$Register;
3087     bool RegCollision = (Rres == Rsrc) || (Rres == Rptr);
3088     Register Rold   = RegCollision ? Rtmp : Rres;
3089 
3090     Label Lretry;
3091     __ bind(Lretry);
3092     __ ldarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
3093     __ add(Rtmp, Rsrc, Rold);
3094     __ stdcx_(Rtmp, Rptr);
3095     if (UseStaticBranchPredictionInCompareAndSwapPPC64) {
3096       __ bne_predict_not_taken(CCR0, Lretry);
3097     } else {
3098       __ bne(                  CCR0, Lretry);
3099     }
3100     if (RegCollision) __ subf(Rres, Rsrc, Rtmp);
3101     __ fence();
3102   %}
3103 
3104   enc_class enc_GetAndSetI(iRegIdst res, iRegPdst mem_ptr, iRegIsrc src) %{
3105     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3106 
3107     MacroAssembler _masm(&cbuf);
3108     Register Rtmp   = R0;
3109     Register Rres   = $res$$Register;
3110     Register Rsrc   = $src$$Register;
3111     Register Rptr   = $mem_ptr$$Register;
3112     bool RegCollision = (Rres == Rsrc) || (Rres == Rptr);
3113     Register Rold   = RegCollision ? Rtmp : Rres;
3114 
3115     Label Lretry;
3116     __ bind(Lretry);
3117     __ lwarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
3118     __ stwcx_(Rsrc, Rptr);
3119     if (UseStaticBranchPredictionInCompareAndSwapPPC64) {
3120       __ bne_predict_not_taken(CCR0, Lretry);
3121     } else {
3122       __ bne(                  CCR0, Lretry);
3123     }
3124     if (RegCollision) __ mr(Rres, Rtmp);
3125     __ fence();
3126   %}
3127 
3128   enc_class enc_GetAndSetL(iRegLdst res, iRegPdst mem_ptr, iRegLsrc src) %{
3129     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3130 
3131     MacroAssembler _masm(&cbuf);
3132     Register Rtmp   = R0;
3133     Register Rres   = $res$$Register;
3134     Register Rsrc   = $src$$Register;
3135     Register Rptr   = $mem_ptr$$Register;
3136     bool RegCollision = (Rres == Rsrc) || (Rres == Rptr);
3137     Register Rold   = RegCollision ? Rtmp : Rres;
3138 
3139     Label Lretry;
3140     __ bind(Lretry);
3141     __ ldarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
3142     __ stdcx_(Rsrc, Rptr);
3143     if (UseStaticBranchPredictionInCompareAndSwapPPC64) {
3144       __ bne_predict_not_taken(CCR0, Lretry);
3145     } else {
3146       __ bne(                  CCR0, Lretry);
3147     }
3148     if (RegCollision) __ mr(Rres, Rtmp);
3149     __ fence();
3150   %}
3151 
3152   // This enc_class is needed so that scheduler gets proper
3153   // input mapping for latency computation.
3154   enc_class enc_andc(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
3155     // TODO: PPC port $archOpcode(ppc64Opcode_andc);
3156     MacroAssembler _masm(&cbuf);
3157     __ andc($dst$$Register, $src1$$Register, $src2$$Register);
3158   %}
3159 
3160   enc_class enc_convI2B_regI__cmove(iRegIdst dst, iRegIsrc src, flagsReg crx, immI16 zero, immI16 notzero) %{
3161     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3162 
3163     MacroAssembler _masm(&cbuf);
3164 
3165     Label done;
3166     __ cmpwi($crx$$CondRegister, $src$$Register, 0);
3167     __ li($dst$$Register, $zero$$constant);
3168     __ beq($crx$$CondRegister, done);
3169     __ li($dst$$Register, $notzero$$constant);
3170     __ bind(done);
3171   %}
3172 
3173   enc_class enc_convP2B_regP__cmove(iRegIdst dst, iRegPsrc src, flagsReg crx, immI16 zero, immI16 notzero) %{
3174     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3175 
3176     MacroAssembler _masm(&cbuf);
3177 
3178     Label done;
3179     __ cmpdi($crx$$CondRegister, $src$$Register, 0);
3180     __ li($dst$$Register, $zero$$constant);
3181     __ beq($crx$$CondRegister, done);
3182     __ li($dst$$Register, $notzero$$constant);
3183     __ bind(done);
3184   %}
3185 
3186   enc_class enc_cmove_bso_stackSlotL(iRegLdst dst, flagsRegSrc crx, stackSlotL mem ) %{
3187     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
3188 
3189     MacroAssembler _masm(&cbuf);
3190     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
3191     Label done;
3192     __ bso($crx$$CondRegister, done);
3193     __ ld($dst$$Register, Idisp, $mem$$base$$Register);
3194     // TODO PPC port __ endgroup_if_needed(_size == 12);
3195     __ bind(done);
3196   %}
3197 
3198   enc_class enc_bc(flagsRegSrc crx, cmpOp cmp, Label lbl) %{
3199     // TODO: PPC port $archOpcode(ppc64Opcode_bc);
3200 
3201     MacroAssembler _masm(&cbuf);
3202     Label d;   // dummy
3203     __ bind(d);
3204     Label* p = ($lbl$$label);
3205     // `p' is `NULL' when this encoding class is used only to
3206     // determine the size of the encoded instruction.
3207     Label& l = (NULL == p)? d : *(p);
3208     int cc = $cmp$$cmpcode;
3209     int flags_reg = $crx$$reg;
3210     assert((Assembler::bcondCRbiIs1 & ~Assembler::bcondCRbiIs0) == 8, "check encoding");
3211     int bhint = Assembler::bhintNoHint;
3212 
3213     if (UseStaticBranchPredictionForUncommonPathsPPC64) {
3214       if (_prob <= PROB_NEVER) {
3215         bhint = Assembler::bhintIsNotTaken;
3216       } else if (_prob >= PROB_ALWAYS) {
3217         bhint = Assembler::bhintIsTaken;
3218       }
3219     }
3220 
3221     __ bc(Assembler::add_bhint_to_boint(bhint, cc_to_boint(cc)),
3222           cc_to_biint(cc, flags_reg),
3223           l);
3224   %}
3225 
3226   enc_class enc_bc_far(flagsRegSrc crx, cmpOp cmp, Label lbl) %{
3227     // The scheduler doesn't know about branch shortening, so we set the opcode
3228     // to ppc64Opcode_bc in order to hide this detail from the scheduler.
3229     // TODO: PPC port $archOpcode(ppc64Opcode_bc);
3230 
3231     MacroAssembler _masm(&cbuf);
3232     Label d;    // dummy
3233     __ bind(d);
3234     Label* p = ($lbl$$label);
3235     // `p' is `NULL' when this encoding class is used only to
3236     // determine the size of the encoded instruction.
3237     Label& l = (NULL == p)? d : *(p);
3238     int cc = $cmp$$cmpcode;
3239     int flags_reg = $crx$$reg;
3240     int bhint = Assembler::bhintNoHint;
3241 
3242     if (UseStaticBranchPredictionForUncommonPathsPPC64) {
3243       if (_prob <= PROB_NEVER) {
3244         bhint = Assembler::bhintIsNotTaken;
3245       } else if (_prob >= PROB_ALWAYS) {
3246         bhint = Assembler::bhintIsTaken;
3247       }
3248     }
3249 
3250     // Tell the conditional far branch to optimize itself when being relocated.
3251     __ bc_far(Assembler::add_bhint_to_boint(bhint, cc_to_boint(cc)),
3252                   cc_to_biint(cc, flags_reg),
3253                   l,
3254                   MacroAssembler::bc_far_optimize_on_relocate);
3255   %}
3256 
3257   // Branch used with Power6 scheduling (can be shortened without changing the node).
3258   enc_class enc_bc_short_far(flagsRegSrc crx, cmpOp cmp, Label lbl) %{
3259     // The scheduler doesn't know about branch shortening, so we set the opcode
3260     // to ppc64Opcode_bc in order to hide this detail from the scheduler.
3261     // TODO: PPC port $archOpcode(ppc64Opcode_bc);
3262 
3263     MacroAssembler _masm(&cbuf);
3264     Label d;   // dummy
3265     __ bind(d);
3266     Label* p = ($lbl$$label);
3267     // `p' is `NULL' when this encoding class is used only to
3268     // determine the size of the encoded instruction.
3269     Label& l = (NULL == p)? d : *(p);
3270     int cc = $cmp$$cmpcode;
3271     int flags_reg = $crx$$reg;
3272     int bhint = Assembler::bhintNoHint;
3273 
3274     if (UseStaticBranchPredictionForUncommonPathsPPC64) {
3275       if (_prob <= PROB_NEVER) {
3276         bhint = Assembler::bhintIsNotTaken;
3277       } else if (_prob >= PROB_ALWAYS) {
3278         bhint = Assembler::bhintIsTaken;
3279       }
3280     }
3281 
3282 #if 0 // TODO: PPC port
3283     if (_size == 8) {
3284       // Tell the conditional far branch to optimize itself when being relocated.
3285       __ bc_far(Assembler::add_bhint_to_boint(bhint, cc_to_boint(cc)),
3286                     cc_to_biint(cc, flags_reg),
3287                     l,
3288                     MacroAssembler::bc_far_optimize_on_relocate);
3289     } else {
3290       __ bc    (Assembler::add_bhint_to_boint(bhint, cc_to_boint(cc)),
3291                     cc_to_biint(cc, flags_reg),
3292                     l);
3293     }
3294 #endif
3295     Unimplemented();
3296   %}
3297 
3298   // Postalloc expand emitter for loading a replicatef float constant from
3299   // the method's TOC.
3300   // Enc_class needed as consttanttablebase is not supported by postalloc
3301   // expand.
3302   enc_class postalloc_expand_load_replF_constant(iRegLdst dst, immF src, iRegLdst toc) %{
3303     // Create new nodes.
3304 
3305     // Make an operand with the bit pattern to load as float.
3306     immLOper *op_repl = new immLOper((jlong)replicate_immF(op_src->constantF()));
3307 
3308     loadConLNodesTuple loadConLNodes =
3309       loadConLNodesTuple_create(ra_, n_toc, op_repl,
3310                                 ra_->get_reg_second(this), ra_->get_reg_first(this));
3311 
3312     // Push new nodes.
3313     if (loadConLNodes._large_hi) nodes->push(loadConLNodes._large_hi);
3314     if (loadConLNodes._last)     nodes->push(loadConLNodes._last);
3315 
3316     assert(nodes->length() >= 1, "must have created at least 1 node");
3317     assert(loadConLNodes._last->bottom_type()->isa_long(), "must be long");
3318   %}
3319 
3320   // This enc_class is needed so that scheduler gets proper
3321   // input mapping for latency computation.
3322   enc_class enc_poll(immI dst, iRegLdst poll) %{
3323     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
3324     // Fake operand dst needed for PPC scheduler.
3325     assert($dst$$constant == 0x0, "dst must be 0x0");
3326 
3327     MacroAssembler _masm(&cbuf);
3328     // Mark the code position where the load from the safepoint
3329     // polling page was emitted as relocInfo::poll_type.
3330     __ relocate(relocInfo::poll_type);
3331     __ load_from_polling_page($poll$$Register);
3332   %}
3333 
3334   // A Java static call or a runtime call.
3335   //
3336   // Branch-and-link relative to a trampoline.
3337   // The trampoline loads the target address and does a long branch to there.
3338   // In case we call java, the trampoline branches to a interpreter_stub
3339   // which loads the inline cache and the real call target from the constant pool.
3340   //
3341   // This basically looks like this:
3342   //
3343   // >>>> consts      -+  -+
3344   //                   |   |- offset1
3345   // [call target1]    | <-+
3346   // [IC cache]        |- offset2
3347   // [call target2] <--+
3348   //
3349   // <<<< consts
3350   // >>>> insts
3351   //
3352   // bl offset16               -+  -+             ??? // How many bits available?
3353   //                            |   |
3354   // <<<< insts                 |   |
3355   // >>>> stubs                 |   |
3356   //                            |   |- trampoline_stub_Reloc
3357   // trampoline stub:           | <-+
3358   //   r2 = toc                 |
3359   //   r2 = [r2 + offset1]      |       // Load call target1 from const section
3360   //   mtctr r2                 |
3361   //   bctr                     |- static_stub_Reloc
3362   // comp_to_interp_stub:   <---+
3363   //   r1 = toc
3364   //   ICreg = [r1 + IC_offset]         // Load IC from const section
3365   //   r1    = [r1 + offset2]           // Load call target2 from const section
3366   //   mtctr r1
3367   //   bctr
3368   //
3369   // <<<< stubs
3370   //
3371   // The call instruction in the code either
3372   // - Branches directly to a compiled method if the offset is encodable in instruction.
3373   // - Branches to the trampoline stub if the offset to the compiled method is not encodable.
3374   // - Branches to the compiled_to_interp stub if the target is interpreted.
3375   //
3376   // Further there are three relocations from the loads to the constants in
3377   // the constant section.
3378   //
3379   // Usage of r1 and r2 in the stubs allows to distinguish them.
3380   enc_class enc_java_static_call(method meth) %{
3381     // TODO: PPC port $archOpcode(ppc64Opcode_bl);
3382 
3383     MacroAssembler _masm(&cbuf);
3384     address entry_point = (address)$meth$$method;
3385 
3386     if (!_method) {
3387       // A call to a runtime wrapper, e.g. new, new_typeArray_Java, uncommon_trap.
3388       emit_call_with_trampoline_stub(_masm, entry_point, relocInfo::runtime_call_type);
3389     } else {
3390       // Remember the offset not the address.
3391       const int start_offset = __ offset();
3392       // The trampoline stub.
3393       if (!Compile::current()->in_scratch_emit_size()) {
3394         // No entry point given, use the current pc.
3395         // Make sure branch fits into
3396         if (entry_point == 0) entry_point = __ pc();
3397 
3398         // Put the entry point as a constant into the constant pool.
3399         const address entry_point_toc_addr   = __ address_constant(entry_point, RelocationHolder::none);
3400         const int     entry_point_toc_offset = __ offset_to_method_toc(entry_point_toc_addr);
3401 
3402 
3403         // Emit the trampoline stub which will be related to the branch-and-link below.
3404         CallStubImpl::emit_trampoline_stub(_masm, entry_point_toc_offset, start_offset);
3405         if (ciEnv::current()->failing()) { return; } // Code cache may be full.
3406         int method_index = resolved_method_index(cbuf);
3407         __ relocate(_optimized_virtual ? opt_virtual_call_Relocate::spec(method_index)
3408                                        : static_call_Relocate::spec(method_index));
3409       }
3410 
3411       // The real call.
3412       // Note: At this point we do not have the address of the trampoline
3413       // stub, and the entry point might be too far away for bl, so __ pc()
3414       // serves as dummy and the bl will be patched later.
3415       cbuf.set_insts_mark();
3416       __ bl(__ pc());  // Emits a relocation.
3417 
3418       // The stub for call to interpreter.
3419       address stub = CompiledStaticCall::emit_to_interp_stub(cbuf);
3420       if (stub == NULL) {
3421         ciEnv::current()->record_failure("CodeCache is full");
3422         return;
3423       }
3424     }
3425   %}
3426 
3427   // Emit a method handle call.
3428   //
3429   // Method handle calls from compiled to compiled are going thru a
3430   // c2i -> i2c adapter, extending the frame for their arguments. The
3431   // caller however, returns directly to the compiled callee, that has
3432   // to cope with the extended frame. We restore the original frame by
3433   // loading the callers sp and adding the calculated framesize.
3434   enc_class enc_java_handle_call(method meth) %{
3435     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3436 
3437     MacroAssembler _masm(&cbuf);
3438     address entry_point = (address)$meth$$method;
3439 
3440     // Remember the offset not the address.
3441     const int start_offset = __ offset();
3442     // The trampoline stub.
3443     if (!ra_->C->in_scratch_emit_size()) {
3444       // No entry point given, use the current pc.
3445       // Make sure branch fits into
3446       if (entry_point == 0) entry_point = __ pc();
3447 
3448       // Put the entry point as a constant into the constant pool.
3449       const address entry_point_toc_addr   = __ address_constant(entry_point, RelocationHolder::none);
3450       const int     entry_point_toc_offset = __ offset_to_method_toc(entry_point_toc_addr);
3451 
3452       assert(!_override_symbolic_info, "resolved method overriding not supported");
3453 
3454       // Emit the trampoline stub which will be related to the branch-and-link below.
3455       CallStubImpl::emit_trampoline_stub(_masm, entry_point_toc_offset, start_offset);
3456       if (ra_->C->env()->failing()) { return; } // Code cache may be full.
3457       assert(_optimized_virtual, "methodHandle call should be a virtual call");
3458       __ relocate(relocInfo::opt_virtual_call_type);
3459     }
3460 
3461     // The real call.
3462     // Note: At this point we do not have the address of the trampoline
3463     // stub, and the entry point might be too far away for bl, so __ pc()
3464     // serves as dummy and the bl will be patched later.
3465     cbuf.set_insts_mark();
3466     __ bl(__ pc());  // Emits a relocation.
3467 
3468     assert(_method, "execute next statement conditionally");
3469     // The stub for call to interpreter.
3470     address stub = CompiledStaticCall::emit_to_interp_stub(cbuf);
3471     if (stub == NULL) {
3472       ciEnv::current()->record_failure("CodeCache is full");
3473       return;
3474     }
3475 
3476     // Restore original sp.
3477     __ ld(R11_scratch1, 0, R1_SP); // Load caller sp.
3478     const long framesize = ra_->C->frame_slots() << LogBytesPerInt;
3479     unsigned int bytes = (unsigned int)framesize;
3480     long offset = Assembler::align_addr(bytes, frame::alignment_in_bytes);
3481     if (Assembler::is_simm(-offset, 16)) {
3482       __ addi(R1_SP, R11_scratch1, -offset);
3483     } else {
3484       __ load_const_optimized(R12_scratch2, -offset);
3485       __ add(R1_SP, R11_scratch1, R12_scratch2);
3486     }
3487 #ifdef ASSERT
3488   __ ld(R12_scratch2, 0, R1_SP); // Load from unextended_sp.
3489   __ cmpd(CCR0, R11_scratch1, R12_scratch2);
3490   __ asm_assert_eq("backlink changed", 0x8000);
3491 #endif
3492     // If fails should store backlink before unextending.
3493 
3494     if (ra_->C->env()->failing()) {
3495       return;
3496     }
3497   %}
3498 
3499   // Second node of expanded dynamic call - the call.
3500   enc_class enc_java_dynamic_call_sched(method meth) %{
3501     // TODO: PPC port $archOpcode(ppc64Opcode_bl);
3502 
3503     MacroAssembler _masm(&cbuf);
3504 
3505     if (!ra_->C->in_scratch_emit_size()) {
3506       // Create a call trampoline stub for the given method.
3507       const address entry_point = !($meth$$method) ? 0 : (address)$meth$$method;
3508       const address entry_point_const = __ address_constant(entry_point, RelocationHolder::none);
3509       const int entry_point_const_toc_offset = __ offset_to_method_toc(entry_point_const);
3510       CallStubImpl::emit_trampoline_stub(_masm, entry_point_const_toc_offset, __ offset());
3511       if (ra_->C->env()->failing()) { return; } // Code cache may be full.
3512 
3513       // Build relocation at call site with ic position as data.
3514       assert((_load_ic_hi_node != NULL && _load_ic_node == NULL) ||
3515              (_load_ic_hi_node == NULL && _load_ic_node != NULL),
3516              "must have one, but can't have both");
3517       assert((_load_ic_hi_node != NULL && _load_ic_hi_node->_cbuf_insts_offset != -1) ||
3518              (_load_ic_node != NULL    && _load_ic_node->_cbuf_insts_offset != -1),
3519              "must contain instruction offset");
3520       const int virtual_call_oop_addr_offset = _load_ic_hi_node != NULL
3521         ? _load_ic_hi_node->_cbuf_insts_offset
3522         : _load_ic_node->_cbuf_insts_offset;
3523       const address virtual_call_oop_addr = __ addr_at(virtual_call_oop_addr_offset);
3524       assert(MacroAssembler::is_load_const_from_method_toc_at(virtual_call_oop_addr),
3525              "should be load from TOC");
3526       int method_index = resolved_method_index(cbuf);
3527       __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr, method_index));
3528     }
3529 
3530     // At this point I do not have the address of the trampoline stub,
3531     // and the entry point might be too far away for bl. Pc() serves
3532     // as dummy and bl will be patched later.
3533     __ bl((address) __ pc());
3534   %}
3535 
3536   // postalloc expand emitter for virtual calls.
3537   enc_class postalloc_expand_java_dynamic_call_sched(method meth, iRegLdst toc) %{
3538 
3539     // Create the nodes for loading the IC from the TOC.
3540     loadConLNodesTuple loadConLNodes_IC =
3541       loadConLNodesTuple_create(ra_, n_toc, new immLOper((jlong)Universe::non_oop_word()),
3542                                 OptoReg::Name(R19_H_num), OptoReg::Name(R19_num));
3543 
3544     // Create the call node.
3545     CallDynamicJavaDirectSchedNode *call = new CallDynamicJavaDirectSchedNode();
3546     call->_method_handle_invoke = _method_handle_invoke;
3547     call->_vtable_index      = _vtable_index;
3548     call->_method            = _method;
3549     call->_bci               = _bci;
3550     call->_optimized_virtual = _optimized_virtual;
3551     call->_tf                = _tf;
3552     call->_entry_point       = _entry_point;
3553     call->_cnt               = _cnt;
3554     call->_argsize           = _argsize;
3555     call->_oop_map           = _oop_map;
3556     call->_jvms              = _jvms;
3557     call->_jvmadj            = _jvmadj;
3558     call->_in_rms            = _in_rms;
3559     call->_nesting           = _nesting;
3560 
3561     // New call needs all inputs of old call.
3562     // Req...
3563     for (uint i = 0; i < req(); ++i) {
3564       // The expanded node does not need toc any more.
3565       // Add the inline cache constant here instead. This expresses the
3566       // register of the inline cache must be live at the call.
3567       // Else we would have to adapt JVMState by -1.
3568       if (i == mach_constant_base_node_input()) {
3569         call->add_req(loadConLNodes_IC._last);
3570       } else {
3571         call->add_req(in(i));
3572       }
3573     }
3574     // ...as well as prec
3575     for (uint i = req(); i < len(); ++i) {
3576       call->add_prec(in(i));
3577     }
3578 
3579     // Remember nodes loading the inline cache into r19.
3580     call->_load_ic_hi_node = loadConLNodes_IC._large_hi;
3581     call->_load_ic_node    = loadConLNodes_IC._small;
3582 
3583     // Operands for new nodes.
3584     call->_opnds[0] = _opnds[0];
3585     call->_opnds[1] = _opnds[1];
3586 
3587     // Only the inline cache is associated with a register.
3588     assert(Matcher::inline_cache_reg() == OptoReg::Name(R19_num), "ic reg should be R19");
3589 
3590     // Push new nodes.
3591     if (loadConLNodes_IC._large_hi) nodes->push(loadConLNodes_IC._large_hi);
3592     if (loadConLNodes_IC._last)     nodes->push(loadConLNodes_IC._last);
3593     nodes->push(call);
3594   %}
3595 
3596   // Compound version of call dynamic
3597   // Toc is only passed so that it can be used in ins_encode statement.
3598   // In the code we have to use $constanttablebase.
3599   enc_class enc_java_dynamic_call(method meth, iRegLdst toc) %{
3600     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3601     MacroAssembler _masm(&cbuf);
3602     int start_offset = __ offset();
3603 
3604     Register Rtoc = (ra_) ? $constanttablebase : R2_TOC;
3605 #if 0
3606     int vtable_index = this->_vtable_index;
3607     if (_vtable_index < 0) {
3608       // Must be invalid_vtable_index, not nonvirtual_vtable_index.
3609       assert(_vtable_index == Method::invalid_vtable_index, "correct sentinel value");
3610       Register ic_reg = as_Register(Matcher::inline_cache_reg_encode());
3611 
3612       // Virtual call relocation will point to ic load.
3613       address virtual_call_meta_addr = __ pc();
3614       // Load a clear inline cache.
3615       AddressLiteral empty_ic((address) Universe::non_oop_word());
3616       __ load_const_from_method_toc(ic_reg, empty_ic, Rtoc);
3617       // CALL to fixup routine.  Fixup routine uses ScopeDesc info
3618       // to determine who we intended to call.
3619       __ relocate(virtual_call_Relocation::spec(virtual_call_meta_addr));
3620       emit_call_with_trampoline_stub(_masm, (address)$meth$$method, relocInfo::none);
3621       assert(((MachCallDynamicJavaNode*)this)->ret_addr_offset() == __ offset() - start_offset,
3622              "Fix constant in ret_addr_offset()");
3623     } else {
3624       assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
3625       // Go thru the vtable. Get receiver klass. Receiver already
3626       // checked for non-null. If we'll go thru a C2I adapter, the
3627       // interpreter expects method in R19_method.
3628 
3629       __ load_klass(R11_scratch1, R3);
3630 
3631       int entry_offset = InstanceKlass::vtable_start_offset() + _vtable_index * vtableEntry::size();
3632       int v_off = entry_offset * wordSize + vtableEntry::method_offset_in_bytes();
3633       __ li(R19_method, v_off);
3634       __ ldx(R19_method/*method oop*/, R19_method/*method offset*/, R11_scratch1/*class*/);
3635       // NOTE: for vtable dispatches, the vtable entry will never be
3636       // null. However it may very well end up in handle_wrong_method
3637       // if the method is abstract for the particular class.
3638       __ ld(R11_scratch1, in_bytes(Method::from_compiled_offset()), R19_method);
3639       // Call target. Either compiled code or C2I adapter.
3640       __ mtctr(R11_scratch1);
3641       __ bctrl();
3642       if (((MachCallDynamicJavaNode*)this)->ret_addr_offset() != __ offset() - start_offset) {
3643         tty->print(" %d, %d\n", ((MachCallDynamicJavaNode*)this)->ret_addr_offset(),__ offset() - start_offset);
3644       }
3645       assert(((MachCallDynamicJavaNode*)this)->ret_addr_offset() == __ offset() - start_offset,
3646              "Fix constant in ret_addr_offset()");
3647     }
3648 #endif
3649     Unimplemented();  // ret_addr_offset not yet fixed. Depends on compressed oops (load klass!).
3650   %}
3651 
3652   // a runtime call
3653   enc_class enc_java_to_runtime_call (method meth) %{
3654     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
3655 
3656     MacroAssembler _masm(&cbuf);
3657     const address start_pc = __ pc();
3658 
3659 #if defined(ABI_ELFv2)
3660     address entry= !($meth$$method) ? NULL : (address)$meth$$method;
3661     __ call_c(entry, relocInfo::runtime_call_type);
3662 #else
3663     // The function we're going to call.
3664     FunctionDescriptor fdtemp;
3665     const FunctionDescriptor* fd = !($meth$$method) ? &fdtemp : (FunctionDescriptor*)$meth$$method;
3666 
3667     Register Rtoc = R12_scratch2;
3668     // Calculate the method's TOC.
3669     __ calculate_address_from_global_toc(Rtoc, __ method_toc());
3670     // Put entry, env, toc into the constant pool, this needs up to 3 constant
3671     // pool entries; call_c_using_toc will optimize the call.
3672     __ call_c_using_toc(fd, relocInfo::runtime_call_type, Rtoc);
3673 #endif
3674 
3675     // Check the ret_addr_offset.
3676     assert(((MachCallRuntimeNode*)this)->ret_addr_offset() ==  __ last_calls_return_pc() - start_pc,
3677            "Fix constant in ret_addr_offset()");
3678   %}
3679 
3680   // Move to ctr for leaf call.
3681   // This enc_class is needed so that scheduler gets proper
3682   // input mapping for latency computation.
3683   enc_class enc_leaf_call_mtctr(iRegLsrc src) %{
3684     // TODO: PPC port $archOpcode(ppc64Opcode_mtctr);
3685     MacroAssembler _masm(&cbuf);
3686     __ mtctr($src$$Register);
3687   %}
3688 
3689   // Postalloc expand emitter for runtime leaf calls.
3690   enc_class postalloc_expand_java_to_runtime_call(method meth, iRegLdst toc) %{
3691     loadConLNodesTuple loadConLNodes_Entry;
3692 #if defined(ABI_ELFv2)
3693     jlong entry_address = (jlong) this->entry_point();
3694     assert(entry_address, "need address here");
3695     loadConLNodes_Entry = loadConLNodesTuple_create(ra_, n_toc, new immLOper(entry_address),
3696                                                     OptoReg::Name(R12_H_num), OptoReg::Name(R12_num));
3697 #else
3698     // Get the struct that describes the function we are about to call.
3699     FunctionDescriptor* fd = (FunctionDescriptor*) this->entry_point();
3700     assert(fd, "need fd here");
3701     jlong entry_address = (jlong) fd->entry();
3702     // new nodes
3703     loadConLNodesTuple loadConLNodes_Env;
3704     loadConLNodesTuple loadConLNodes_Toc;
3705 
3706     // Create nodes and operands for loading the entry point.
3707     loadConLNodes_Entry = loadConLNodesTuple_create(ra_, n_toc, new immLOper(entry_address),
3708                                                     OptoReg::Name(R12_H_num), OptoReg::Name(R12_num));
3709 
3710 
3711     // Create nodes and operands for loading the env pointer.
3712     if (fd->env() != NULL) {
3713       loadConLNodes_Env = loadConLNodesTuple_create(ra_, n_toc, new immLOper((jlong) fd->env()),
3714                                                     OptoReg::Name(R11_H_num), OptoReg::Name(R11_num));
3715     } else {
3716       loadConLNodes_Env._large_hi = NULL;
3717       loadConLNodes_Env._large_lo = NULL;
3718       loadConLNodes_Env._small    = NULL;
3719       loadConLNodes_Env._last = new loadConL16Node();
3720       loadConLNodes_Env._last->_opnds[0] = new iRegLdstOper();
3721       loadConLNodes_Env._last->_opnds[1] = new immL16Oper(0);
3722       ra_->set_pair(loadConLNodes_Env._last->_idx, OptoReg::Name(R11_H_num), OptoReg::Name(R11_num));
3723     }
3724 
3725     // Create nodes and operands for loading the Toc point.
3726     loadConLNodes_Toc = loadConLNodesTuple_create(ra_, n_toc, new immLOper((jlong) fd->toc()),
3727                                                   OptoReg::Name(R2_H_num), OptoReg::Name(R2_num));
3728 #endif // ABI_ELFv2
3729     // mtctr node
3730     MachNode *mtctr = new CallLeafDirect_mtctrNode();
3731 
3732     assert(loadConLNodes_Entry._last != NULL, "entry must exist");
3733     mtctr->add_req(0, loadConLNodes_Entry._last);
3734 
3735     mtctr->_opnds[0] = new iRegLdstOper();
3736     mtctr->_opnds[1] = new iRegLdstOper();
3737 
3738     // call node
3739     MachCallLeafNode *call = new CallLeafDirectNode();
3740 
3741     call->_opnds[0] = _opnds[0];
3742     call->_opnds[1] = new methodOper((intptr_t) entry_address); // May get set later.
3743 
3744     // Make the new call node look like the old one.
3745     call->_name        = _name;
3746     call->_tf          = _tf;
3747     call->_entry_point = _entry_point;
3748     call->_cnt         = _cnt;
3749     call->_argsize     = _argsize;
3750     call->_oop_map     = _oop_map;
3751     guarantee(!_jvms, "You must clone the jvms and adapt the offsets by fix_jvms().");
3752     call->_jvms        = NULL;
3753     call->_jvmadj      = _jvmadj;
3754     call->_in_rms      = _in_rms;
3755     call->_nesting     = _nesting;
3756 
3757 
3758     // New call needs all inputs of old call.
3759     // Req...
3760     for (uint i = 0; i < req(); ++i) {
3761       if (i != mach_constant_base_node_input()) {
3762         call->add_req(in(i));
3763       }
3764     }
3765 
3766     // These must be reqired edges, as the registers are live up to
3767     // the call. Else the constants are handled as kills.
3768     call->add_req(mtctr);
3769 #if !defined(ABI_ELFv2)
3770     call->add_req(loadConLNodes_Env._last);
3771     call->add_req(loadConLNodes_Toc._last);
3772 #endif
3773 
3774     // ...as well as prec
3775     for (uint i = req(); i < len(); ++i) {
3776       call->add_prec(in(i));
3777     }
3778 
3779     // registers
3780     ra_->set1(mtctr->_idx, OptoReg::Name(SR_CTR_num));
3781 
3782     // Insert the new nodes.
3783     if (loadConLNodes_Entry._large_hi) nodes->push(loadConLNodes_Entry._large_hi);
3784     if (loadConLNodes_Entry._last)     nodes->push(loadConLNodes_Entry._last);
3785 #if !defined(ABI_ELFv2)
3786     if (loadConLNodes_Env._large_hi)   nodes->push(loadConLNodes_Env._large_hi);
3787     if (loadConLNodes_Env._last)       nodes->push(loadConLNodes_Env._last);
3788     if (loadConLNodes_Toc._large_hi)   nodes->push(loadConLNodes_Toc._large_hi);
3789     if (loadConLNodes_Toc._last)       nodes->push(loadConLNodes_Toc._last);
3790 #endif
3791     nodes->push(mtctr);
3792     nodes->push(call);
3793   %}
3794 %}
3795 
3796 //----------FRAME--------------------------------------------------------------
3797 // Definition of frame structure and management information.
3798 
3799 frame %{
3800   // What direction does stack grow in (assumed to be same for native & Java).
3801   stack_direction(TOWARDS_LOW);
3802 
3803   // These two registers define part of the calling convention between
3804   // compiled code and the interpreter.
3805 
3806   // Inline Cache Register or method for I2C.
3807   inline_cache_reg(R19); // R19_method
3808 
3809   // Method Oop Register when calling interpreter.
3810   interpreter_method_oop_reg(R19); // R19_method
3811 
3812   // Optional: name the operand used by cisc-spilling to access
3813   // [stack_pointer + offset].
3814   cisc_spilling_operand_name(indOffset);
3815 
3816   // Number of stack slots consumed by a Monitor enter.
3817   sync_stack_slots((frame::jit_monitor_size / VMRegImpl::stack_slot_size));
3818 
3819   // Compiled code's Frame Pointer.
3820   frame_pointer(R1); // R1_SP
3821 
3822   // Interpreter stores its frame pointer in a register which is
3823   // stored to the stack by I2CAdaptors. I2CAdaptors convert from
3824   // interpreted java to compiled java.
3825   //
3826   // R14_state holds pointer to caller's cInterpreter.
3827   interpreter_frame_pointer(R14); // R14_state
3828 
3829   stack_alignment(frame::alignment_in_bytes);
3830 
3831   in_preserve_stack_slots((frame::jit_in_preserve_size / VMRegImpl::stack_slot_size));
3832 
3833   // Number of outgoing stack slots killed above the
3834   // out_preserve_stack_slots for calls to C. Supports the var-args
3835   // backing area for register parms.
3836   //
3837   varargs_C_out_slots_killed(((frame::abi_reg_args_size - frame::jit_out_preserve_size) / VMRegImpl::stack_slot_size));
3838 
3839   // The after-PROLOG location of the return address. Location of
3840   // return address specifies a type (REG or STACK) and a number
3841   // representing the register number (i.e. - use a register name) or
3842   // stack slot.
3843   //
3844   // A: Link register is stored in stack slot ...
3845   // M:  ... but it's in the caller's frame according to PPC-64 ABI.
3846   // J: Therefore, we make sure that the link register is also in R11_scratch1
3847   //    at the end of the prolog.
3848   // B: We use R20, now.
3849   //return_addr(REG R20);
3850 
3851   // G: After reading the comments made by all the luminaries on their
3852   //    failure to tell the compiler where the return address really is,
3853   //    I hardly dare to try myself.  However, I'm convinced it's in slot
3854   //    4 what apparently works and saves us some spills.
3855   return_addr(STACK 4);
3856 
3857   // This is the body of the function
3858   //
3859   // void Matcher::calling_convention(OptoRegPair* sig, // array of ideal regs
3860   //                                  uint length,      // length of array
3861   //                                  bool is_outgoing)
3862   //
3863   // The `sig' array is to be updated. sig[j] represents the location
3864   // of the j-th argument, either a register or a stack slot.
3865 
3866   // Comment taken from i486.ad:
3867   // Body of function which returns an integer array locating
3868   // arguments either in registers or in stack slots. Passed an array
3869   // of ideal registers called "sig" and a "length" count. Stack-slot
3870   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3871   // arguments for a CALLEE. Incoming stack arguments are
3872   // automatically biased by the preserve_stack_slots field above.
3873   calling_convention %{
3874     // No difference between ingoing/outgoing. Just pass false.
3875     SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
3876   %}
3877 
3878   // Comment taken from i486.ad:
3879   // Body of function which returns an integer array locating
3880   // arguments either in registers or in stack slots. Passed an array
3881   // of ideal registers called "sig" and a "length" count. Stack-slot
3882   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3883   // arguments for a CALLEE. Incoming stack arguments are
3884   // automatically biased by the preserve_stack_slots field above.
3885   c_calling_convention %{
3886     // This is obviously always outgoing.
3887     // C argument in register AND stack slot.
3888     (void) SharedRuntime::c_calling_convention(sig_bt, regs, /*regs2=*/NULL, length);
3889   %}
3890 
3891   // Location of native (C/C++) and interpreter return values. This
3892   // is specified to be the same as Java. In the 32-bit VM, long
3893   // values are actually returned from native calls in O0:O1 and
3894   // returned to the interpreter in I0:I1. The copying to and from
3895   // the register pairs is done by the appropriate call and epilog
3896   // opcodes. This simplifies the register allocator.
3897   c_return_value %{
3898     assert((ideal_reg >= Op_RegI && ideal_reg <= Op_RegL) ||
3899             (ideal_reg == Op_RegN && Universe::narrow_oop_base() == NULL && Universe::narrow_oop_shift() == 0),
3900             "only return normal values");
3901     // enum names from opcodes.hpp:    Op_Node Op_Set Op_RegN       Op_RegI       Op_RegP       Op_RegF       Op_RegD       Op_RegL
3902     static int typeToRegLo[Op_RegL+1] = { 0,   0,     R3_num,   R3_num,   R3_num,   F1_num,   F1_num,   R3_num };
3903     static int typeToRegHi[Op_RegL+1] = { 0,   0,     OptoReg::Bad, R3_H_num, R3_H_num, OptoReg::Bad, F1_H_num, R3_H_num };
3904     return OptoRegPair(typeToRegHi[ideal_reg], typeToRegLo[ideal_reg]);
3905   %}
3906 
3907   // Location of compiled Java return values.  Same as C
3908   return_value %{
3909     assert((ideal_reg >= Op_RegI && ideal_reg <= Op_RegL) ||
3910             (ideal_reg == Op_RegN && Universe::narrow_oop_base() == NULL && Universe::narrow_oop_shift() == 0),
3911             "only return normal values");
3912     // enum names from opcodes.hpp:    Op_Node Op_Set Op_RegN       Op_RegI       Op_RegP       Op_RegF       Op_RegD       Op_RegL
3913     static int typeToRegLo[Op_RegL+1] = { 0,   0,     R3_num,   R3_num,   R3_num,   F1_num,   F1_num,   R3_num };
3914     static int typeToRegHi[Op_RegL+1] = { 0,   0,     OptoReg::Bad, R3_H_num, R3_H_num, OptoReg::Bad, F1_H_num, R3_H_num };
3915     return OptoRegPair(typeToRegHi[ideal_reg], typeToRegLo[ideal_reg]);
3916   %}
3917 %}
3918 
3919 
3920 //----------ATTRIBUTES---------------------------------------------------------
3921 
3922 //----------Operand Attributes-------------------------------------------------
3923 op_attrib op_cost(1);          // Required cost attribute.
3924 
3925 //----------Instruction Attributes---------------------------------------------
3926 
3927 // Cost attribute. required.
3928 ins_attrib ins_cost(DEFAULT_COST);
3929 
3930 // Is this instruction a non-matching short branch variant of some
3931 // long branch? Not required.
3932 ins_attrib ins_short_branch(0);
3933 
3934 ins_attrib ins_is_TrapBasedCheckNode(true);
3935 
3936 // Number of constants.
3937 // This instruction uses the given number of constants
3938 // (optional attribute).
3939 // This is needed to determine in time whether the constant pool will
3940 // exceed 4000 entries. Before postalloc_expand the overall number of constants
3941 // is determined. It's also used to compute the constant pool size
3942 // in Output().
3943 ins_attrib ins_num_consts(0);
3944 
3945 // Required alignment attribute (must be a power of 2) specifies the
3946 // alignment that some part of the instruction (not necessarily the
3947 // start) requires. If > 1, a compute_padding() function must be
3948 // provided for the instruction.
3949 ins_attrib ins_alignment(1);
3950 
3951 // Enforce/prohibit rematerializations.
3952 // - If an instruction is attributed with 'ins_cannot_rematerialize(true)'
3953 //   then rematerialization of that instruction is prohibited and the
3954 //   instruction's value will be spilled if necessary.
3955 //   Causes that MachNode::rematerialize() returns false.
3956 // - If an instruction is attributed with 'ins_should_rematerialize(true)'
3957 //   then rematerialization should be enforced and a copy of the instruction
3958 //   should be inserted if possible; rematerialization is not guaranteed.
3959 //   Note: this may result in rematerializations in front of every use.
3960 //   Causes that MachNode::rematerialize() can return true.
3961 // (optional attribute)
3962 ins_attrib ins_cannot_rematerialize(false);
3963 ins_attrib ins_should_rematerialize(false);
3964 
3965 // Instruction has variable size depending on alignment.
3966 ins_attrib ins_variable_size_depending_on_alignment(false);
3967 
3968 // Instruction is a nop.
3969 ins_attrib ins_is_nop(false);
3970 
3971 // Instruction is mapped to a MachIfFastLock node (instead of MachFastLock).
3972 ins_attrib ins_use_mach_if_fast_lock_node(false);
3973 
3974 // Field for the toc offset of a constant.
3975 //
3976 // This is needed if the toc offset is not encodable as an immediate in
3977 // the PPC load instruction. If so, the upper (hi) bits of the offset are
3978 // added to the toc, and from this a load with immediate is performed.
3979 // With postalloc expand, we get two nodes that require the same offset
3980 // but which don't know about each other. The offset is only known
3981 // when the constant is added to the constant pool during emitting.
3982 // It is generated in the 'hi'-node adding the upper bits, and saved
3983 // in this node.  The 'lo'-node has a link to the 'hi'-node and reads
3984 // the offset from there when it gets encoded.
3985 ins_attrib ins_field_const_toc_offset(0);
3986 ins_attrib ins_field_const_toc_offset_hi_node(0);
3987 
3988 // A field that can hold the instructions offset in the code buffer.
3989 // Set in the nodes emitter.
3990 ins_attrib ins_field_cbuf_insts_offset(-1);
3991 
3992 // Fields for referencing a call's load-IC-node.
3993 // If the toc offset can not be encoded as an immediate in a load, we
3994 // use two nodes.
3995 ins_attrib ins_field_load_ic_hi_node(0);
3996 ins_attrib ins_field_load_ic_node(0);
3997 
3998 //----------OPERANDS-----------------------------------------------------------
3999 // Operand definitions must precede instruction definitions for correct
4000 // parsing in the ADLC because operands constitute user defined types
4001 // which are used in instruction definitions.
4002 //
4003 // Formats are generated automatically for constants and base registers.
4004 
4005 //----------Simple Operands----------------------------------------------------
4006 // Immediate Operands
4007 
4008 // Integer Immediate: 32-bit
4009 operand immI() %{
4010   match(ConI);
4011   op_cost(40);
4012   format %{ %}
4013   interface(CONST_INTER);
4014 %}
4015 
4016 operand immI8() %{
4017   predicate(Assembler::is_simm(n->get_int(), 8));
4018   op_cost(0);
4019   match(ConI);
4020   format %{ %}
4021   interface(CONST_INTER);
4022 %}
4023 
4024 // Integer Immediate: 16-bit
4025 operand immI16() %{
4026   predicate(Assembler::is_simm(n->get_int(), 16));
4027   op_cost(0);
4028   match(ConI);
4029   format %{ %}
4030   interface(CONST_INTER);
4031 %}
4032 
4033 // Integer Immediate: 32-bit, where lowest 16 bits are 0x0000.
4034 operand immIhi16() %{
4035   predicate(((n->get_int() & 0xffff0000) != 0) && ((n->get_int() & 0xffff) == 0));
4036   match(ConI);
4037   op_cost(0);
4038   format %{ %}
4039   interface(CONST_INTER);
4040 %}
4041 
4042 operand immInegpow2() %{
4043   predicate(is_power_of_2_long((jlong) (julong) (juint) (-(n->get_int()))));
4044   match(ConI);
4045   op_cost(0);
4046   format %{ %}
4047   interface(CONST_INTER);
4048 %}
4049 
4050 operand immIpow2minus1() %{
4051   predicate(is_power_of_2_long((((jlong) (n->get_int()))+1)));
4052   match(ConI);
4053   op_cost(0);
4054   format %{ %}
4055   interface(CONST_INTER);
4056 %}
4057 
4058 operand immIpowerOf2() %{
4059   predicate(is_power_of_2_long((((jlong) (julong) (juint) (n->get_int())))));
4060   match(ConI);
4061   op_cost(0);
4062   format %{ %}
4063   interface(CONST_INTER);
4064 %}
4065 
4066 // Unsigned Integer Immediate: the values 0-31
4067 operand uimmI5() %{
4068   predicate(Assembler::is_uimm(n->get_int(), 5));
4069   match(ConI);
4070   op_cost(0);
4071   format %{ %}
4072   interface(CONST_INTER);
4073 %}
4074 
4075 // Unsigned Integer Immediate: 6-bit
4076 operand uimmI6() %{
4077   predicate(Assembler::is_uimm(n->get_int(), 6));
4078   match(ConI);
4079   op_cost(0);
4080   format %{ %}
4081   interface(CONST_INTER);
4082 %}
4083 
4084 // Unsigned Integer Immediate:  6-bit int, greater than 32
4085 operand uimmI6_ge32() %{
4086   predicate(Assembler::is_uimm(n->get_int(), 6) && n->get_int() >= 32);
4087   match(ConI);
4088   op_cost(0);
4089   format %{ %}
4090   interface(CONST_INTER);
4091 %}
4092 
4093 // Unsigned Integer Immediate: 15-bit
4094 operand uimmI15() %{
4095   predicate(Assembler::is_uimm(n->get_int(), 15));
4096   match(ConI);
4097   op_cost(0);
4098   format %{ %}
4099   interface(CONST_INTER);
4100 %}
4101 
4102 // Unsigned Integer Immediate: 16-bit
4103 operand uimmI16() %{
4104   predicate(Assembler::is_uimm(n->get_int(), 16));
4105   match(ConI);
4106   op_cost(0);
4107   format %{ %}
4108   interface(CONST_INTER);
4109 %}
4110 
4111 // constant 'int 0'.
4112 operand immI_0() %{
4113   predicate(n->get_int() == 0);
4114   match(ConI);
4115   op_cost(0);
4116   format %{ %}
4117   interface(CONST_INTER);
4118 %}
4119 
4120 // constant 'int 1'.
4121 operand immI_1() %{
4122   predicate(n->get_int() == 1);
4123   match(ConI);
4124   op_cost(0);
4125   format %{ %}
4126   interface(CONST_INTER);
4127 %}
4128 
4129 // constant 'int -1'.
4130 operand immI_minus1() %{
4131   predicate(n->get_int() == -1);
4132   match(ConI);
4133   op_cost(0);
4134   format %{ %}
4135   interface(CONST_INTER);
4136 %}
4137 
4138 // int value 16.
4139 operand immI_16() %{
4140   predicate(n->get_int() == 16);
4141   match(ConI);
4142   op_cost(0);
4143   format %{ %}
4144   interface(CONST_INTER);
4145 %}
4146 
4147 // int value 24.
4148 operand immI_24() %{
4149   predicate(n->get_int() == 24);
4150   match(ConI);
4151   op_cost(0);
4152   format %{ %}
4153   interface(CONST_INTER);
4154 %}
4155 
4156 // Compressed oops constants
4157 // Pointer Immediate
4158 operand immN() %{
4159   match(ConN);
4160 
4161   op_cost(10);
4162   format %{ %}
4163   interface(CONST_INTER);
4164 %}
4165 
4166 // NULL Pointer Immediate
4167 operand immN_0() %{
4168   predicate(n->get_narrowcon() == 0);
4169   match(ConN);
4170 
4171   op_cost(0);
4172   format %{ %}
4173   interface(CONST_INTER);
4174 %}
4175 
4176 // Compressed klass constants
4177 operand immNKlass() %{
4178   match(ConNKlass);
4179 
4180   op_cost(0);
4181   format %{ %}
4182   interface(CONST_INTER);
4183 %}
4184 
4185 // This operand can be used to avoid matching of an instruct
4186 // with chain rule.
4187 operand immNKlass_NM() %{
4188   match(ConNKlass);
4189   predicate(false);
4190   op_cost(0);
4191   format %{ %}
4192   interface(CONST_INTER);
4193 %}
4194 
4195 // Pointer Immediate: 64-bit
4196 operand immP() %{
4197   match(ConP);
4198   op_cost(0);
4199   format %{ %}
4200   interface(CONST_INTER);
4201 %}
4202 
4203 // Operand to avoid match of loadConP.
4204 // This operand can be used to avoid matching of an instruct
4205 // with chain rule.
4206 operand immP_NM() %{
4207   match(ConP);
4208   predicate(false);
4209   op_cost(0);
4210   format %{ %}
4211   interface(CONST_INTER);
4212 %}
4213 
4214 // costant 'pointer 0'.
4215 operand immP_0() %{
4216   predicate(n->get_ptr() == 0);
4217   match(ConP);
4218   op_cost(0);
4219   format %{ %}
4220   interface(CONST_INTER);
4221 %}
4222 
4223 // pointer 0x0 or 0x1
4224 operand immP_0or1() %{
4225   predicate((n->get_ptr() == 0) || (n->get_ptr() == 1));
4226   match(ConP);
4227   op_cost(0);
4228   format %{ %}
4229   interface(CONST_INTER);
4230 %}
4231 
4232 operand immL() %{
4233   match(ConL);
4234   op_cost(40);
4235   format %{ %}
4236   interface(CONST_INTER);
4237 %}
4238 
4239 // Long Immediate: 16-bit
4240 operand immL16() %{
4241   predicate(Assembler::is_simm(n->get_long(), 16));
4242   match(ConL);
4243   op_cost(0);
4244   format %{ %}
4245   interface(CONST_INTER);
4246 %}
4247 
4248 // Long Immediate: 16-bit, 4-aligned
4249 operand immL16Alg4() %{
4250   predicate(Assembler::is_simm(n->get_long(), 16) && ((n->get_long() & 0x3) == 0));
4251   match(ConL);
4252   op_cost(0);
4253   format %{ %}
4254   interface(CONST_INTER);
4255 %}
4256 
4257 // Long Immediate: 32-bit, where lowest 16 bits are 0x0000.
4258 operand immL32hi16() %{
4259   predicate(Assembler::is_simm(n->get_long(), 32) && ((n->get_long() & 0xffffL) == 0L));
4260   match(ConL);
4261   op_cost(0);
4262   format %{ %}
4263   interface(CONST_INTER);
4264 %}
4265 
4266 // Long Immediate: 32-bit
4267 operand immL32() %{
4268   predicate(Assembler::is_simm(n->get_long(), 32));
4269   match(ConL);
4270   op_cost(0);
4271   format %{ %}
4272   interface(CONST_INTER);
4273 %}
4274 
4275 // Long Immediate: 64-bit, where highest 16 bits are not 0x0000.
4276 operand immLhighest16() %{
4277   predicate((n->get_long() & 0xffff000000000000L) != 0L && (n->get_long() & 0x0000ffffffffffffL) == 0L);
4278   match(ConL);
4279   op_cost(0);
4280   format %{ %}
4281   interface(CONST_INTER);
4282 %}
4283 
4284 operand immLnegpow2() %{
4285   predicate(is_power_of_2_long((jlong)-(n->get_long())));
4286   match(ConL);
4287   op_cost(0);
4288   format %{ %}
4289   interface(CONST_INTER);
4290 %}
4291 
4292 operand immLpow2minus1() %{
4293   predicate(is_power_of_2_long((((jlong) (n->get_long()))+1)) &&
4294             (n->get_long() != (jlong)0xffffffffffffffffL));
4295   match(ConL);
4296   op_cost(0);
4297   format %{ %}
4298   interface(CONST_INTER);
4299 %}
4300 
4301 // constant 'long 0'.
4302 operand immL_0() %{
4303   predicate(n->get_long() == 0L);
4304   match(ConL);
4305   op_cost(0);
4306   format %{ %}
4307   interface(CONST_INTER);
4308 %}
4309 
4310 // constat ' long -1'.
4311 operand immL_minus1() %{
4312   predicate(n->get_long() == -1L);
4313   match(ConL);
4314   op_cost(0);
4315   format %{ %}
4316   interface(CONST_INTER);
4317 %}
4318 
4319 // Long Immediate: low 32-bit mask
4320 operand immL_32bits() %{
4321   predicate(n->get_long() == 0xFFFFFFFFL);
4322   match(ConL);
4323   op_cost(0);
4324   format %{ %}
4325   interface(CONST_INTER);
4326 %}
4327 
4328 // Unsigned Long Immediate: 16-bit
4329 operand uimmL16() %{
4330   predicate(Assembler::is_uimm(n->get_long(), 16));
4331   match(ConL);
4332   op_cost(0);
4333   format %{ %}
4334   interface(CONST_INTER);
4335 %}
4336 
4337 // Float Immediate
4338 operand immF() %{
4339   match(ConF);
4340   op_cost(40);
4341   format %{ %}
4342   interface(CONST_INTER);
4343 %}
4344 
4345 // Float Immediate: +0.0f.
4346 operand immF_0() %{
4347   predicate(jint_cast(n->getf()) == 0);
4348   match(ConF);
4349 
4350   op_cost(0);
4351   format %{ %}
4352   interface(CONST_INTER);
4353 %}
4354 
4355 // Double Immediate
4356 operand immD() %{
4357   match(ConD);
4358   op_cost(40);
4359   format %{ %}
4360   interface(CONST_INTER);
4361 %}
4362 
4363 // Integer Register Operands
4364 // Integer Destination Register
4365 // See definition of reg_class bits32_reg_rw.
4366 operand iRegIdst() %{
4367   constraint(ALLOC_IN_RC(bits32_reg_rw));
4368   match(RegI);
4369   match(rscratch1RegI);
4370   match(rscratch2RegI);
4371   match(rarg1RegI);
4372   match(rarg2RegI);
4373   match(rarg3RegI);
4374   match(rarg4RegI);
4375   format %{ %}
4376   interface(REG_INTER);
4377 %}
4378 
4379 // Integer Source Register
4380 // See definition of reg_class bits32_reg_ro.
4381 operand iRegIsrc() %{
4382   constraint(ALLOC_IN_RC(bits32_reg_ro));
4383   match(RegI);
4384   match(rscratch1RegI);
4385   match(rscratch2RegI);
4386   match(rarg1RegI);
4387   match(rarg2RegI);
4388   match(rarg3RegI);
4389   match(rarg4RegI);
4390   format %{ %}
4391   interface(REG_INTER);
4392 %}
4393 
4394 operand rscratch1RegI() %{
4395   constraint(ALLOC_IN_RC(rscratch1_bits32_reg));
4396   match(iRegIdst);
4397   format %{ %}
4398   interface(REG_INTER);
4399 %}
4400 
4401 operand rscratch2RegI() %{
4402   constraint(ALLOC_IN_RC(rscratch2_bits32_reg));
4403   match(iRegIdst);
4404   format %{ %}
4405   interface(REG_INTER);
4406 %}
4407 
4408 operand rarg1RegI() %{
4409   constraint(ALLOC_IN_RC(rarg1_bits32_reg));
4410   match(iRegIdst);
4411   format %{ %}
4412   interface(REG_INTER);
4413 %}
4414 
4415 operand rarg2RegI() %{
4416   constraint(ALLOC_IN_RC(rarg2_bits32_reg));
4417   match(iRegIdst);
4418   format %{ %}
4419   interface(REG_INTER);
4420 %}
4421 
4422 operand rarg3RegI() %{
4423   constraint(ALLOC_IN_RC(rarg3_bits32_reg));
4424   match(iRegIdst);
4425   format %{ %}
4426   interface(REG_INTER);
4427 %}
4428 
4429 operand rarg4RegI() %{
4430   constraint(ALLOC_IN_RC(rarg4_bits32_reg));
4431   match(iRegIdst);
4432   format %{ %}
4433   interface(REG_INTER);
4434 %}
4435 
4436 operand rarg1RegL() %{
4437   constraint(ALLOC_IN_RC(rarg1_bits64_reg));
4438   match(iRegLdst);
4439   format %{ %}
4440   interface(REG_INTER);
4441 %}
4442 
4443 operand rarg2RegL() %{
4444   constraint(ALLOC_IN_RC(rarg2_bits64_reg));
4445   match(iRegLdst);
4446   format %{ %}
4447   interface(REG_INTER);
4448 %}
4449 
4450 operand rarg3RegL() %{
4451   constraint(ALLOC_IN_RC(rarg3_bits64_reg));
4452   match(iRegLdst);
4453   format %{ %}
4454   interface(REG_INTER);
4455 %}
4456 
4457 operand rarg4RegL() %{
4458   constraint(ALLOC_IN_RC(rarg4_bits64_reg));
4459   match(iRegLdst);
4460   format %{ %}
4461   interface(REG_INTER);
4462 %}
4463 
4464 // Pointer Destination Register
4465 // See definition of reg_class bits64_reg_rw.
4466 operand iRegPdst() %{
4467   constraint(ALLOC_IN_RC(bits64_reg_rw));
4468   match(RegP);
4469   match(rscratch1RegP);
4470   match(rscratch2RegP);
4471   match(rarg1RegP);
4472   match(rarg2RegP);
4473   match(rarg3RegP);
4474   match(rarg4RegP);
4475   format %{ %}
4476   interface(REG_INTER);
4477 %}
4478 
4479 // Pointer Destination Register
4480 // Operand not using r11 and r12 (killed in epilog).
4481 operand iRegPdstNoScratch() %{
4482   constraint(ALLOC_IN_RC(bits64_reg_leaf_call));
4483   match(RegP);
4484   match(rarg1RegP);
4485   match(rarg2RegP);
4486   match(rarg3RegP);
4487   match(rarg4RegP);
4488   format %{ %}
4489   interface(REG_INTER);
4490 %}
4491 
4492 // Pointer Source Register
4493 // See definition of reg_class bits64_reg_ro.
4494 operand iRegPsrc() %{
4495   constraint(ALLOC_IN_RC(bits64_reg_ro));
4496   match(RegP);
4497   match(iRegPdst);
4498   match(rscratch1RegP);
4499   match(rscratch2RegP);
4500   match(rarg1RegP);
4501   match(rarg2RegP);
4502   match(rarg3RegP);
4503   match(rarg4RegP);
4504   match(threadRegP);
4505   format %{ %}
4506   interface(REG_INTER);
4507 %}
4508 
4509 // Thread operand.
4510 operand threadRegP() %{
4511   constraint(ALLOC_IN_RC(thread_bits64_reg));
4512   match(iRegPdst);
4513   format %{ "R16" %}
4514   interface(REG_INTER);
4515 %}
4516 
4517 operand rscratch1RegP() %{
4518   constraint(ALLOC_IN_RC(rscratch1_bits64_reg));
4519   match(iRegPdst);
4520   format %{ "R11" %}
4521   interface(REG_INTER);
4522 %}
4523 
4524 operand rscratch2RegP() %{
4525   constraint(ALLOC_IN_RC(rscratch2_bits64_reg));
4526   match(iRegPdst);
4527   format %{ %}
4528   interface(REG_INTER);
4529 %}
4530 
4531 operand rarg1RegP() %{
4532   constraint(ALLOC_IN_RC(rarg1_bits64_reg));
4533   match(iRegPdst);
4534   format %{ %}
4535   interface(REG_INTER);
4536 %}
4537 
4538 operand rarg2RegP() %{
4539   constraint(ALLOC_IN_RC(rarg2_bits64_reg));
4540   match(iRegPdst);
4541   format %{ %}
4542   interface(REG_INTER);
4543 %}
4544 
4545 operand rarg3RegP() %{
4546   constraint(ALLOC_IN_RC(rarg3_bits64_reg));
4547   match(iRegPdst);
4548   format %{ %}
4549   interface(REG_INTER);
4550 %}
4551 
4552 operand rarg4RegP() %{
4553   constraint(ALLOC_IN_RC(rarg4_bits64_reg));
4554   match(iRegPdst);
4555   format %{ %}
4556   interface(REG_INTER);
4557 %}
4558 
4559 operand iRegNsrc() %{
4560   constraint(ALLOC_IN_RC(bits32_reg_ro));
4561   match(RegN);
4562   match(iRegNdst);
4563 
4564   format %{ %}
4565   interface(REG_INTER);
4566 %}
4567 
4568 operand iRegNdst() %{
4569   constraint(ALLOC_IN_RC(bits32_reg_rw));
4570   match(RegN);
4571 
4572   format %{ %}
4573   interface(REG_INTER);
4574 %}
4575 
4576 // Long Destination Register
4577 // See definition of reg_class bits64_reg_rw.
4578 operand iRegLdst() %{
4579   constraint(ALLOC_IN_RC(bits64_reg_rw));
4580   match(RegL);
4581   match(rscratch1RegL);
4582   match(rscratch2RegL);
4583   format %{ %}
4584   interface(REG_INTER);
4585 %}
4586 
4587 // Long Source Register
4588 // See definition of reg_class bits64_reg_ro.
4589 operand iRegLsrc() %{
4590   constraint(ALLOC_IN_RC(bits64_reg_ro));
4591   match(RegL);
4592   match(iRegLdst);
4593   match(rscratch1RegL);
4594   match(rscratch2RegL);
4595   format %{ %}
4596   interface(REG_INTER);
4597 %}
4598 
4599 // Special operand for ConvL2I.
4600 operand iRegL2Isrc(iRegLsrc reg) %{
4601   constraint(ALLOC_IN_RC(bits64_reg_ro));
4602   match(ConvL2I reg);
4603   format %{ "ConvL2I($reg)" %}
4604   interface(REG_INTER)
4605 %}
4606 
4607 operand rscratch1RegL() %{
4608   constraint(ALLOC_IN_RC(rscratch1_bits64_reg));
4609   match(RegL);
4610   format %{ %}
4611   interface(REG_INTER);
4612 %}
4613 
4614 operand rscratch2RegL() %{
4615   constraint(ALLOC_IN_RC(rscratch2_bits64_reg));
4616   match(RegL);
4617   format %{ %}
4618   interface(REG_INTER);
4619 %}
4620 
4621 // Condition Code Flag Registers
4622 operand flagsReg() %{
4623   constraint(ALLOC_IN_RC(int_flags));
4624   match(RegFlags);
4625   format %{ %}
4626   interface(REG_INTER);
4627 %}
4628 
4629 operand flagsRegSrc() %{
4630   constraint(ALLOC_IN_RC(int_flags_ro));
4631   match(RegFlags);
4632   match(flagsReg);
4633   match(flagsRegCR0);
4634   format %{ %}
4635   interface(REG_INTER);
4636 %}
4637 
4638 // Condition Code Flag Register CR0
4639 operand flagsRegCR0() %{
4640   constraint(ALLOC_IN_RC(int_flags_CR0));
4641   match(RegFlags);
4642   format %{ "CR0" %}
4643   interface(REG_INTER);
4644 %}
4645 
4646 operand flagsRegCR1() %{
4647   constraint(ALLOC_IN_RC(int_flags_CR1));
4648   match(RegFlags);
4649   format %{ "CR1" %}
4650   interface(REG_INTER);
4651 %}
4652 
4653 operand flagsRegCR6() %{
4654   constraint(ALLOC_IN_RC(int_flags_CR6));
4655   match(RegFlags);
4656   format %{ "CR6" %}
4657   interface(REG_INTER);
4658 %}
4659 
4660 operand regCTR() %{
4661   constraint(ALLOC_IN_RC(ctr_reg));
4662   // RegFlags should work. Introducing a RegSpecial type would cause a
4663   // lot of changes.
4664   match(RegFlags);
4665   format %{"SR_CTR" %}
4666   interface(REG_INTER);
4667 %}
4668 
4669 operand regD() %{
4670   constraint(ALLOC_IN_RC(dbl_reg));
4671   match(RegD);
4672   format %{ %}
4673   interface(REG_INTER);
4674 %}
4675 
4676 operand regF() %{
4677   constraint(ALLOC_IN_RC(flt_reg));
4678   match(RegF);
4679   format %{ %}
4680   interface(REG_INTER);
4681 %}
4682 
4683 // Special Registers
4684 
4685 // Method Register
4686 operand inline_cache_regP(iRegPdst reg) %{
4687   constraint(ALLOC_IN_RC(r19_bits64_reg)); // inline_cache_reg
4688   match(reg);
4689   format %{ %}
4690   interface(REG_INTER);
4691 %}
4692 
4693 operand compiler_method_oop_regP(iRegPdst reg) %{
4694   constraint(ALLOC_IN_RC(rscratch1_bits64_reg)); // compiler_method_oop_reg
4695   match(reg);
4696   format %{ %}
4697   interface(REG_INTER);
4698 %}
4699 
4700 operand interpreter_method_oop_regP(iRegPdst reg) %{
4701   constraint(ALLOC_IN_RC(r19_bits64_reg)); // interpreter_method_oop_reg
4702   match(reg);
4703   format %{ %}
4704   interface(REG_INTER);
4705 %}
4706 
4707 // Operands to remove register moves in unscaled mode.
4708 // Match read/write registers with an EncodeP node if neither shift nor add are required.
4709 operand iRegP2N(iRegPsrc reg) %{
4710   predicate(false /* TODO: PPC port MatchDecodeNodes*/&& Universe::narrow_oop_shift() == 0);
4711   constraint(ALLOC_IN_RC(bits64_reg_ro));
4712   match(EncodeP reg);
4713   format %{ "$reg" %}
4714   interface(REG_INTER)
4715 %}
4716 
4717 operand iRegN2P(iRegNsrc reg) %{
4718   predicate(false /* TODO: PPC port MatchDecodeNodes*/);
4719   constraint(ALLOC_IN_RC(bits32_reg_ro));
4720   match(DecodeN reg);
4721   format %{ "$reg" %}
4722   interface(REG_INTER)
4723 %}
4724 
4725 operand iRegN2P_klass(iRegNsrc reg) %{
4726   predicate(Universe::narrow_klass_base() == NULL && Universe::narrow_klass_shift() == 0);
4727   constraint(ALLOC_IN_RC(bits32_reg_ro));
4728   match(DecodeNKlass reg);
4729   format %{ "$reg" %}
4730   interface(REG_INTER)
4731 %}
4732 
4733 //----------Complex Operands---------------------------------------------------
4734 // Indirect Memory Reference
4735 operand indirect(iRegPsrc reg) %{
4736   constraint(ALLOC_IN_RC(bits64_reg_ro));
4737   match(reg);
4738   op_cost(100);
4739   format %{ "[$reg]" %}
4740   interface(MEMORY_INTER) %{
4741     base($reg);
4742     index(0x0);
4743     scale(0x0);
4744     disp(0x0);
4745   %}
4746 %}
4747 
4748 // Indirect with Offset
4749 operand indOffset16(iRegPsrc reg, immL16 offset) %{
4750   constraint(ALLOC_IN_RC(bits64_reg_ro));
4751   match(AddP reg offset);
4752   op_cost(100);
4753   format %{ "[$reg + $offset]" %}
4754   interface(MEMORY_INTER) %{
4755     base($reg);
4756     index(0x0);
4757     scale(0x0);
4758     disp($offset);
4759   %}
4760 %}
4761 
4762 // Indirect with 4-aligned Offset
4763 operand indOffset16Alg4(iRegPsrc reg, immL16Alg4 offset) %{
4764   constraint(ALLOC_IN_RC(bits64_reg_ro));
4765   match(AddP reg offset);
4766   op_cost(100);
4767   format %{ "[$reg + $offset]" %}
4768   interface(MEMORY_INTER) %{
4769     base($reg);
4770     index(0x0);
4771     scale(0x0);
4772     disp($offset);
4773   %}
4774 %}
4775 
4776 //----------Complex Operands for Compressed OOPs-------------------------------
4777 // Compressed OOPs with narrow_oop_shift == 0.
4778 
4779 // Indirect Memory Reference, compressed OOP
4780 operand indirectNarrow(iRegNsrc reg) %{
4781   predicate(false /* TODO: PPC port MatchDecodeNodes*/);
4782   constraint(ALLOC_IN_RC(bits64_reg_ro));
4783   match(DecodeN reg);
4784   op_cost(100);
4785   format %{ "[$reg]" %}
4786   interface(MEMORY_INTER) %{
4787     base($reg);
4788     index(0x0);
4789     scale(0x0);
4790     disp(0x0);
4791   %}
4792 %}
4793 
4794 operand indirectNarrow_klass(iRegNsrc reg) %{
4795   predicate(Universe::narrow_klass_base() == NULL && Universe::narrow_klass_shift() == 0);
4796   constraint(ALLOC_IN_RC(bits64_reg_ro));
4797   match(DecodeNKlass reg);
4798   op_cost(100);
4799   format %{ "[$reg]" %}
4800   interface(MEMORY_INTER) %{
4801     base($reg);
4802     index(0x0);
4803     scale(0x0);
4804     disp(0x0);
4805   %}
4806 %}
4807 
4808 // Indirect with Offset, compressed OOP
4809 operand indOffset16Narrow(iRegNsrc reg, immL16 offset) %{
4810   predicate(false /* TODO: PPC port MatchDecodeNodes*/);
4811   constraint(ALLOC_IN_RC(bits64_reg_ro));
4812   match(AddP (DecodeN reg) offset);
4813   op_cost(100);
4814   format %{ "[$reg + $offset]" %}
4815   interface(MEMORY_INTER) %{
4816     base($reg);
4817     index(0x0);
4818     scale(0x0);
4819     disp($offset);
4820   %}
4821 %}
4822 
4823 operand indOffset16Narrow_klass(iRegNsrc reg, immL16 offset) %{
4824   predicate(Universe::narrow_klass_base() == NULL && Universe::narrow_klass_shift() == 0);
4825   constraint(ALLOC_IN_RC(bits64_reg_ro));
4826   match(AddP (DecodeNKlass reg) offset);
4827   op_cost(100);
4828   format %{ "[$reg + $offset]" %}
4829   interface(MEMORY_INTER) %{
4830     base($reg);
4831     index(0x0);
4832     scale(0x0);
4833     disp($offset);
4834   %}
4835 %}
4836 
4837 // Indirect with 4-aligned Offset, compressed OOP
4838 operand indOffset16NarrowAlg4(iRegNsrc reg, immL16Alg4 offset) %{
4839   predicate(false /* TODO: PPC port MatchDecodeNodes*/);
4840   constraint(ALLOC_IN_RC(bits64_reg_ro));
4841   match(AddP (DecodeN reg) offset);
4842   op_cost(100);
4843   format %{ "[$reg + $offset]" %}
4844   interface(MEMORY_INTER) %{
4845     base($reg);
4846     index(0x0);
4847     scale(0x0);
4848     disp($offset);
4849   %}
4850 %}
4851 
4852 operand indOffset16NarrowAlg4_klass(iRegNsrc reg, immL16Alg4 offset) %{
4853   predicate(Universe::narrow_klass_base() == NULL && Universe::narrow_klass_shift() == 0);
4854   constraint(ALLOC_IN_RC(bits64_reg_ro));
4855   match(AddP (DecodeNKlass reg) offset);
4856   op_cost(100);
4857   format %{ "[$reg + $offset]" %}
4858   interface(MEMORY_INTER) %{
4859     base($reg);
4860     index(0x0);
4861     scale(0x0);
4862     disp($offset);
4863   %}
4864 %}
4865 
4866 //----------Special Memory Operands--------------------------------------------
4867 // Stack Slot Operand
4868 //
4869 // This operand is used for loading and storing temporary values on
4870 // the stack where a match requires a value to flow through memory.
4871 operand stackSlotI(sRegI reg) %{
4872   constraint(ALLOC_IN_RC(stack_slots));
4873   op_cost(100);
4874   //match(RegI);
4875   format %{ "[sp+$reg]" %}
4876   interface(MEMORY_INTER) %{
4877     base(0x1);   // R1_SP
4878     index(0x0);
4879     scale(0x0);
4880     disp($reg);  // Stack Offset
4881   %}
4882 %}
4883 
4884 operand stackSlotL(sRegL reg) %{
4885   constraint(ALLOC_IN_RC(stack_slots));
4886   op_cost(100);
4887   //match(RegL);
4888   format %{ "[sp+$reg]" %}
4889   interface(MEMORY_INTER) %{
4890     base(0x1);   // R1_SP
4891     index(0x0);
4892     scale(0x0);
4893     disp($reg);  // Stack Offset
4894   %}
4895 %}
4896 
4897 operand stackSlotP(sRegP reg) %{
4898   constraint(ALLOC_IN_RC(stack_slots));
4899   op_cost(100);
4900   //match(RegP);
4901   format %{ "[sp+$reg]" %}
4902   interface(MEMORY_INTER) %{
4903     base(0x1);   // R1_SP
4904     index(0x0);
4905     scale(0x0);
4906     disp($reg);  // Stack Offset
4907   %}
4908 %}
4909 
4910 operand stackSlotF(sRegF reg) %{
4911   constraint(ALLOC_IN_RC(stack_slots));
4912   op_cost(100);
4913   //match(RegF);
4914   format %{ "[sp+$reg]" %}
4915   interface(MEMORY_INTER) %{
4916     base(0x1);   // R1_SP
4917     index(0x0);
4918     scale(0x0);
4919     disp($reg);  // Stack Offset
4920   %}
4921 %}
4922 
4923 operand stackSlotD(sRegD reg) %{
4924   constraint(ALLOC_IN_RC(stack_slots));
4925   op_cost(100);
4926   //match(RegD);
4927   format %{ "[sp+$reg]" %}
4928   interface(MEMORY_INTER) %{
4929     base(0x1);   // R1_SP
4930     index(0x0);
4931     scale(0x0);
4932     disp($reg);  // Stack Offset
4933   %}
4934 %}
4935 
4936 // Operands for expressing Control Flow
4937 // NOTE: Label is a predefined operand which should not be redefined in
4938 //       the AD file. It is generically handled within the ADLC.
4939 
4940 //----------Conditional Branch Operands----------------------------------------
4941 // Comparison Op
4942 //
4943 // This is the operation of the comparison, and is limited to the
4944 // following set of codes: L (<), LE (<=), G (>), GE (>=), E (==), NE
4945 // (!=).
4946 //
4947 // Other attributes of the comparison, such as unsignedness, are specified
4948 // by the comparison instruction that sets a condition code flags register.
4949 // That result is represented by a flags operand whose subtype is appropriate
4950 // to the unsignedness (etc.) of the comparison.
4951 //
4952 // Later, the instruction which matches both the Comparison Op (a Bool) and
4953 // the flags (produced by the Cmp) specifies the coding of the comparison op
4954 // by matching a specific subtype of Bool operand below.
4955 
4956 // When used for floating point comparisons: unordered same as less.
4957 operand cmpOp() %{
4958   match(Bool);
4959   format %{ "" %}
4960   interface(COND_INTER) %{
4961                            // BO only encodes bit 4 of bcondCRbiIsX, as bits 1-3 are always '100'.
4962                            //           BO          &  BI
4963     equal(0xA);            // 10 10:   bcondCRbiIs1 & Condition::equal
4964     not_equal(0x2);        // 00 10:   bcondCRbiIs0 & Condition::equal
4965     less(0x8);             // 10 00:   bcondCRbiIs1 & Condition::less
4966     greater_equal(0x0);    // 00 00:   bcondCRbiIs0 & Condition::less
4967     less_equal(0x1);       // 00 01:   bcondCRbiIs0 & Condition::greater
4968     greater(0x9);          // 10 01:   bcondCRbiIs1 & Condition::greater
4969     overflow(0xB);         // 10 11:   bcondCRbiIs1 & Condition::summary_overflow
4970     no_overflow(0x3);      // 00 11:   bcondCRbiIs0 & Condition::summary_overflow
4971   %}
4972 %}
4973 
4974 //----------OPERAND CLASSES----------------------------------------------------
4975 // Operand Classes are groups of operands that are used to simplify
4976 // instruction definitions by not requiring the AD writer to specify
4977 // seperate instructions for every form of operand when the
4978 // instruction accepts multiple operand types with the same basic
4979 // encoding and format. The classic case of this is memory operands.
4980 // Indirect is not included since its use is limited to Compare & Swap.
4981 
4982 opclass memory(indirect, indOffset16 /*, indIndex, tlsReference*/, indirectNarrow, indirectNarrow_klass, indOffset16Narrow, indOffset16Narrow_klass);
4983 // Memory operand where offsets are 4-aligned. Required for ld, std.
4984 opclass memoryAlg4(indirect, indOffset16Alg4, indirectNarrow, indOffset16NarrowAlg4, indOffset16NarrowAlg4_klass);
4985 opclass indirectMemory(indirect, indirectNarrow);
4986 
4987 // Special opclass for I and ConvL2I.
4988 opclass iRegIsrc_iRegL2Isrc(iRegIsrc, iRegL2Isrc);
4989 
4990 // Operand classes to match encode and decode. iRegN_P2N is only used
4991 // for storeN. I have never seen an encode node elsewhere.
4992 opclass iRegN_P2N(iRegNsrc, iRegP2N);
4993 opclass iRegP_N2P(iRegPsrc, iRegN2P, iRegN2P_klass);
4994 
4995 //----------PIPELINE-----------------------------------------------------------
4996 
4997 pipeline %{
4998 
4999 // See J.M.Tendler et al. "Power4 system microarchitecture", IBM
5000 // J. Res. & Dev., No. 1, Jan. 2002.
5001 
5002 //----------ATTRIBUTES---------------------------------------------------------
5003 attributes %{
5004 
5005   // Power4 instructions are of fixed length.
5006   fixed_size_instructions;
5007 
5008   // TODO: if `bundle' means number of instructions fetched
5009   // per cycle, this is 8. If `bundle' means Power4 `group', that is
5010   // max instructions issued per cycle, this is 5.
5011   max_instructions_per_bundle = 8;
5012 
5013   // A Power4 instruction is 4 bytes long.
5014   instruction_unit_size = 4;
5015 
5016   // The Power4 processor fetches 64 bytes...
5017   instruction_fetch_unit_size = 64;
5018 
5019   // ...in one line
5020   instruction_fetch_units = 1
5021 
5022   // Unused, list one so that array generated by adlc is not empty.
5023   // Aix compiler chokes if _nop_count = 0.
5024   nops(fxNop);
5025 %}
5026 
5027 //----------RESOURCES----------------------------------------------------------
5028 // Resources are the functional units available to the machine
5029 resources(
5030    PPC_BR,         // branch unit
5031    PPC_CR,         // condition unit
5032    PPC_FX1,        // integer arithmetic unit 1
5033    PPC_FX2,        // integer arithmetic unit 2
5034    PPC_LDST1,      // load/store unit 1
5035    PPC_LDST2,      // load/store unit 2
5036    PPC_FP1,        // float arithmetic unit 1
5037    PPC_FP2,        // float arithmetic unit 2
5038    PPC_LDST = PPC_LDST1 | PPC_LDST2,
5039    PPC_FX = PPC_FX1 | PPC_FX2,
5040    PPC_FP = PPC_FP1 | PPC_FP2
5041  );
5042 
5043 //----------PIPELINE DESCRIPTION-----------------------------------------------
5044 // Pipeline Description specifies the stages in the machine's pipeline
5045 pipe_desc(
5046    // Power4 longest pipeline path
5047    PPC_IF,   // instruction fetch
5048    PPC_IC,
5049    //PPC_BP, // branch prediction
5050    PPC_D0,   // decode
5051    PPC_D1,   // decode
5052    PPC_D2,   // decode
5053    PPC_D3,   // decode
5054    PPC_Xfer1,
5055    PPC_GD,   // group definition
5056    PPC_MP,   // map
5057    PPC_ISS,  // issue
5058    PPC_RF,   // resource fetch
5059    PPC_EX1,  // execute (all units)
5060    PPC_EX2,  // execute (FP, LDST)
5061    PPC_EX3,  // execute (FP, LDST)
5062    PPC_EX4,  // execute (FP)
5063    PPC_EX5,  // execute (FP)
5064    PPC_EX6,  // execute (FP)
5065    PPC_WB,   // write back
5066    PPC_Xfer2,
5067    PPC_CP
5068  );
5069 
5070 //----------PIPELINE CLASSES---------------------------------------------------
5071 // Pipeline Classes describe the stages in which input and output are
5072 // referenced by the hardware pipeline.
5073 
5074 // Simple pipeline classes.
5075 
5076 // Default pipeline class.
5077 pipe_class pipe_class_default() %{
5078   single_instruction;
5079   fixed_latency(2);
5080 %}
5081 
5082 // Pipeline class for empty instructions.
5083 pipe_class pipe_class_empty() %{
5084   single_instruction;
5085   fixed_latency(0);
5086 %}
5087 
5088 // Pipeline class for compares.
5089 pipe_class pipe_class_compare() %{
5090   single_instruction;
5091   fixed_latency(16);
5092 %}
5093 
5094 // Pipeline class for traps.
5095 pipe_class pipe_class_trap() %{
5096   single_instruction;
5097   fixed_latency(100);
5098 %}
5099 
5100 // Pipeline class for memory operations.
5101 pipe_class pipe_class_memory() %{
5102   single_instruction;
5103   fixed_latency(16);
5104 %}
5105 
5106 // Pipeline class for call.
5107 pipe_class pipe_class_call() %{
5108   single_instruction;
5109   fixed_latency(100);
5110 %}
5111 
5112 // Define the class for the Nop node.
5113 define %{
5114    MachNop = pipe_class_default;
5115 %}
5116 
5117 %}
5118 
5119 //----------INSTRUCTIONS-------------------------------------------------------
5120 
5121 // Naming of instructions:
5122 //   opA_operB / opA_operB_operC:
5123 //     Operation 'op' with one or two source operands 'oper'. Result
5124 //     type is A, source operand types are B and C.
5125 //     Iff A == B == C, B and C are left out.
5126 //
5127 // The instructions are ordered according to the following scheme:
5128 //  - loads
5129 //  - load constants
5130 //  - prefetch
5131 //  - store
5132 //  - encode/decode
5133 //  - membar
5134 //  - conditional moves
5135 //  - compare & swap
5136 //  - arithmetic and logic operations
5137 //    * int: Add, Sub, Mul, Div, Mod
5138 //    * int: lShift, arShift, urShift, rot
5139 //    * float: Add, Sub, Mul, Div
5140 //    * and, or, xor ...
5141 //  - register moves: float <-> int, reg <-> stack, repl
5142 //  - cast (high level type cast, XtoP, castPP, castII, not_null etc.
5143 //  - conv (low level type cast requiring bit changes (sign extend etc)
5144 //  - compares, range & zero checks.
5145 //  - branches
5146 //  - complex operations, intrinsics, min, max, replicate
5147 //  - lock
5148 //  - Calls
5149 //
5150 // If there are similar instructions with different types they are sorted:
5151 // int before float
5152 // small before big
5153 // signed before unsigned
5154 // e.g., loadS before loadUS before loadI before loadF.
5155 
5156 
5157 //----------Load/Store Instructions--------------------------------------------
5158 
5159 //----------Load Instructions--------------------------------------------------
5160 
5161 // Converts byte to int.
5162 // As convB2I_reg, but without match rule.  The match rule of convB2I_reg
5163 // reuses the 'amount' operand, but adlc expects that operand specification
5164 // and operands in match rule are equivalent.
5165 instruct convB2I_reg_2(iRegIdst dst, iRegIsrc src) %{
5166   effect(DEF dst, USE src);
5167   format %{ "EXTSB   $dst, $src \t// byte->int" %}
5168   size(4);
5169   ins_encode %{
5170     // TODO: PPC port $archOpcode(ppc64Opcode_extsb);
5171     __ extsb($dst$$Register, $src$$Register);
5172   %}
5173   ins_pipe(pipe_class_default);
5174 %}
5175 
5176 instruct loadUB_indirect(iRegIdst dst, indirectMemory mem) %{
5177   // match-rule, false predicate
5178   match(Set dst (LoadB mem));
5179   predicate(false);
5180 
5181   format %{ "LBZ     $dst, $mem" %}
5182   size(4);
5183   ins_encode( enc_lbz(dst, mem) );
5184   ins_pipe(pipe_class_memory);
5185 %}
5186 
5187 instruct loadUB_indirect_ac(iRegIdst dst, indirectMemory mem) %{
5188   // match-rule, false predicate
5189   match(Set dst (LoadB mem));
5190   predicate(false);
5191 
5192   format %{ "LBZ     $dst, $mem\n\t"
5193             "TWI     $dst\n\t"
5194             "ISYNC" %}
5195   size(12);
5196   ins_encode( enc_lbz_ac(dst, mem) );
5197   ins_pipe(pipe_class_memory);
5198 %}
5199 
5200 // Load Byte (8bit signed). LoadB = LoadUB + ConvUB2B.
5201 instruct loadB_indirect_Ex(iRegIdst dst, indirectMemory mem) %{
5202   match(Set dst (LoadB mem));
5203   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5204   ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5205   expand %{
5206     iRegIdst tmp;
5207     loadUB_indirect(tmp, mem);
5208     convB2I_reg_2(dst, tmp);
5209   %}
5210 %}
5211 
5212 instruct loadB_indirect_ac_Ex(iRegIdst dst, indirectMemory mem) %{
5213   match(Set dst (LoadB mem));
5214   ins_cost(3*MEMORY_REF_COST + DEFAULT_COST);
5215   expand %{
5216     iRegIdst tmp;
5217     loadUB_indirect_ac(tmp, mem);
5218     convB2I_reg_2(dst, tmp);
5219   %}
5220 %}
5221 
5222 instruct loadUB_indOffset16(iRegIdst dst, indOffset16 mem) %{
5223   // match-rule, false predicate
5224   match(Set dst (LoadB mem));
5225   predicate(false);
5226 
5227   format %{ "LBZ     $dst, $mem" %}
5228   size(4);
5229   ins_encode( enc_lbz(dst, mem) );
5230   ins_pipe(pipe_class_memory);
5231 %}
5232 
5233 instruct loadUB_indOffset16_ac(iRegIdst dst, indOffset16 mem) %{
5234   // match-rule, false predicate
5235   match(Set dst (LoadB mem));
5236   predicate(false);
5237 
5238   format %{ "LBZ     $dst, $mem\n\t"
5239             "TWI     $dst\n\t"
5240             "ISYNC" %}
5241   size(12);
5242   ins_encode( enc_lbz_ac(dst, mem) );
5243   ins_pipe(pipe_class_memory);
5244 %}
5245 
5246 // Load Byte (8bit signed). LoadB = LoadUB + ConvUB2B.
5247 instruct loadB_indOffset16_Ex(iRegIdst dst, indOffset16 mem) %{
5248   match(Set dst (LoadB mem));
5249   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5250   ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5251 
5252   expand %{
5253     iRegIdst tmp;
5254     loadUB_indOffset16(tmp, mem);
5255     convB2I_reg_2(dst, tmp);
5256   %}
5257 %}
5258 
5259 instruct loadB_indOffset16_ac_Ex(iRegIdst dst, indOffset16 mem) %{
5260   match(Set dst (LoadB mem));
5261   ins_cost(3*MEMORY_REF_COST + DEFAULT_COST);
5262 
5263   expand %{
5264     iRegIdst tmp;
5265     loadUB_indOffset16_ac(tmp, mem);
5266     convB2I_reg_2(dst, tmp);
5267   %}
5268 %}
5269 
5270 // Load Unsigned Byte (8bit UNsigned) into an int reg.
5271 instruct loadUB(iRegIdst dst, memory mem) %{
5272   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5273   match(Set dst (LoadUB mem));
5274   ins_cost(MEMORY_REF_COST);
5275 
5276   format %{ "LBZ     $dst, $mem \t// byte, zero-extend to int" %}
5277   size(4);
5278   ins_encode( enc_lbz(dst, mem) );
5279   ins_pipe(pipe_class_memory);
5280 %}
5281 
5282 // Load  Unsigned Byte (8bit UNsigned) acquire.
5283 instruct loadUB_ac(iRegIdst dst, memory mem) %{
5284   match(Set dst (LoadUB mem));
5285   ins_cost(3*MEMORY_REF_COST);
5286 
5287   format %{ "LBZ     $dst, $mem \t// byte, zero-extend to int, acquire\n\t"
5288             "TWI     $dst\n\t"
5289             "ISYNC" %}
5290   size(12);
5291   ins_encode( enc_lbz_ac(dst, mem) );
5292   ins_pipe(pipe_class_memory);
5293 %}
5294 
5295 // Load Unsigned Byte (8bit UNsigned) into a Long Register.
5296 instruct loadUB2L(iRegLdst dst, memory mem) %{
5297   match(Set dst (ConvI2L (LoadUB mem)));
5298   predicate(_kids[0]->_leaf->as_Load()->is_unordered() || followed_by_acquire(_kids[0]->_leaf));
5299   ins_cost(MEMORY_REF_COST);
5300 
5301   format %{ "LBZ     $dst, $mem \t// byte, zero-extend to long" %}
5302   size(4);
5303   ins_encode( enc_lbz(dst, mem) );
5304   ins_pipe(pipe_class_memory);
5305 %}
5306 
5307 instruct loadUB2L_ac(iRegLdst dst, memory mem) %{
5308   match(Set dst (ConvI2L (LoadUB mem)));
5309   ins_cost(3*MEMORY_REF_COST);
5310 
5311   format %{ "LBZ     $dst, $mem \t// byte, zero-extend to long, acquire\n\t"
5312             "TWI     $dst\n\t"
5313             "ISYNC" %}
5314   size(12);
5315   ins_encode( enc_lbz_ac(dst, mem) );
5316   ins_pipe(pipe_class_memory);
5317 %}
5318 
5319 // Load Short (16bit signed)
5320 instruct loadS(iRegIdst dst, memory mem) %{
5321   match(Set dst (LoadS mem));
5322   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5323   ins_cost(MEMORY_REF_COST);
5324 
5325   format %{ "LHA     $dst, $mem" %}
5326   size(4);
5327   ins_encode %{
5328     // TODO: PPC port $archOpcode(ppc64Opcode_lha);
5329     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5330     __ lha($dst$$Register, Idisp, $mem$$base$$Register);
5331   %}
5332   ins_pipe(pipe_class_memory);
5333 %}
5334 
5335 // Load Short (16bit signed) acquire.
5336 instruct loadS_ac(iRegIdst dst, memory mem) %{
5337   match(Set dst (LoadS mem));
5338   ins_cost(3*MEMORY_REF_COST);
5339 
5340   format %{ "LHA     $dst, $mem\t acquire\n\t"
5341             "TWI     $dst\n\t"
5342             "ISYNC" %}
5343   size(12);
5344   ins_encode %{
5345     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
5346     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5347     __ lha($dst$$Register, Idisp, $mem$$base$$Register);
5348     __ twi_0($dst$$Register);
5349     __ isync();
5350   %}
5351   ins_pipe(pipe_class_memory);
5352 %}
5353 
5354 // Load Char (16bit unsigned)
5355 instruct loadUS(iRegIdst dst, memory mem) %{
5356   match(Set dst (LoadUS mem));
5357   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5358   ins_cost(MEMORY_REF_COST);
5359 
5360   format %{ "LHZ     $dst, $mem" %}
5361   size(4);
5362   ins_encode( enc_lhz(dst, mem) );
5363   ins_pipe(pipe_class_memory);
5364 %}
5365 
5366 // Load Char (16bit unsigned) acquire.
5367 instruct loadUS_ac(iRegIdst dst, memory mem) %{
5368   match(Set dst (LoadUS mem));
5369   ins_cost(3*MEMORY_REF_COST);
5370 
5371   format %{ "LHZ     $dst, $mem \t// acquire\n\t"
5372             "TWI     $dst\n\t"
5373             "ISYNC" %}
5374   size(12);
5375   ins_encode( enc_lhz_ac(dst, mem) );
5376   ins_pipe(pipe_class_memory);
5377 %}
5378 
5379 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register.
5380 instruct loadUS2L(iRegLdst dst, memory mem) %{
5381   match(Set dst (ConvI2L (LoadUS mem)));
5382   predicate(_kids[0]->_leaf->as_Load()->is_unordered() || followed_by_acquire(_kids[0]->_leaf));
5383   ins_cost(MEMORY_REF_COST);
5384 
5385   format %{ "LHZ     $dst, $mem \t// short, zero-extend to long" %}
5386   size(4);
5387   ins_encode( enc_lhz(dst, mem) );
5388   ins_pipe(pipe_class_memory);
5389 %}
5390 
5391 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register acquire.
5392 instruct loadUS2L_ac(iRegLdst dst, memory mem) %{
5393   match(Set dst (ConvI2L (LoadUS mem)));
5394   ins_cost(3*MEMORY_REF_COST);
5395 
5396   format %{ "LHZ     $dst, $mem \t// short, zero-extend to long, acquire\n\t"
5397             "TWI     $dst\n\t"
5398             "ISYNC" %}
5399   size(12);
5400   ins_encode( enc_lhz_ac(dst, mem) );
5401   ins_pipe(pipe_class_memory);
5402 %}
5403 
5404 // Load Integer.
5405 instruct loadI(iRegIdst dst, memory mem) %{
5406   match(Set dst (LoadI mem));
5407   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5408   ins_cost(MEMORY_REF_COST);
5409 
5410   format %{ "LWZ     $dst, $mem" %}
5411   size(4);
5412   ins_encode( enc_lwz(dst, mem) );
5413   ins_pipe(pipe_class_memory);
5414 %}
5415 
5416 // Load Integer acquire.
5417 instruct loadI_ac(iRegIdst dst, memory mem) %{
5418   match(Set dst (LoadI mem));
5419   ins_cost(3*MEMORY_REF_COST);
5420 
5421   format %{ "LWZ     $dst, $mem \t// load acquire\n\t"
5422             "TWI     $dst\n\t"
5423             "ISYNC" %}
5424   size(12);
5425   ins_encode( enc_lwz_ac(dst, mem) );
5426   ins_pipe(pipe_class_memory);
5427 %}
5428 
5429 // Match loading integer and casting it to unsigned int in
5430 // long register.
5431 // LoadI + ConvI2L + AndL 0xffffffff.
5432 instruct loadUI2L(iRegLdst dst, memory mem, immL_32bits mask) %{
5433   match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
5434   predicate(_kids[0]->_kids[0]->_leaf->as_Load()->is_unordered());
5435   ins_cost(MEMORY_REF_COST);
5436 
5437   format %{ "LWZ     $dst, $mem \t// zero-extend to long" %}
5438   size(4);
5439   ins_encode( enc_lwz(dst, mem) );
5440   ins_pipe(pipe_class_memory);
5441 %}
5442 
5443 // Match loading integer and casting it to long.
5444 instruct loadI2L(iRegLdst dst, memory mem) %{
5445   match(Set dst (ConvI2L (LoadI mem)));
5446   predicate(_kids[0]->_leaf->as_Load()->is_unordered());
5447   ins_cost(MEMORY_REF_COST);
5448 
5449   format %{ "LWA     $dst, $mem \t// loadI2L" %}
5450   size(4);
5451   ins_encode %{
5452     // TODO: PPC port $archOpcode(ppc64Opcode_lwa);
5453     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5454     __ lwa($dst$$Register, Idisp, $mem$$base$$Register);
5455   %}
5456   ins_pipe(pipe_class_memory);
5457 %}
5458 
5459 // Match loading integer and casting it to long - acquire.
5460 instruct loadI2L_ac(iRegLdst dst, memory mem) %{
5461   match(Set dst (ConvI2L (LoadI mem)));
5462   ins_cost(3*MEMORY_REF_COST);
5463 
5464   format %{ "LWA     $dst, $mem \t// loadI2L acquire"
5465             "TWI     $dst\n\t"
5466             "ISYNC" %}
5467   size(12);
5468   ins_encode %{
5469     // TODO: PPC port $archOpcode(ppc64Opcode_lwa);
5470     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5471     __ lwa($dst$$Register, Idisp, $mem$$base$$Register);
5472     __ twi_0($dst$$Register);
5473     __ isync();
5474   %}
5475   ins_pipe(pipe_class_memory);
5476 %}
5477 
5478 // Load Long - aligned
5479 instruct loadL(iRegLdst dst, memoryAlg4 mem) %{
5480   match(Set dst (LoadL mem));
5481   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5482   ins_cost(MEMORY_REF_COST);
5483 
5484   format %{ "LD      $dst, $mem \t// long" %}
5485   size(4);
5486   ins_encode( enc_ld(dst, mem) );
5487   ins_pipe(pipe_class_memory);
5488 %}
5489 
5490 // Load Long - aligned acquire.
5491 instruct loadL_ac(iRegLdst dst, memoryAlg4 mem) %{
5492   match(Set dst (LoadL mem));
5493   ins_cost(3*MEMORY_REF_COST);
5494 
5495   format %{ "LD      $dst, $mem \t// long acquire\n\t"
5496             "TWI     $dst\n\t"
5497             "ISYNC" %}
5498   size(12);
5499   ins_encode( enc_ld_ac(dst, mem) );
5500   ins_pipe(pipe_class_memory);
5501 %}
5502 
5503 // Load Long - UNaligned
5504 instruct loadL_unaligned(iRegLdst dst, memoryAlg4 mem) %{
5505   match(Set dst (LoadL_unaligned mem));
5506   // predicate(...) // Unaligned_ac is not needed (and wouldn't make sense).
5507   ins_cost(MEMORY_REF_COST);
5508 
5509   format %{ "LD      $dst, $mem \t// unaligned long" %}
5510   size(4);
5511   ins_encode( enc_ld(dst, mem) );
5512   ins_pipe(pipe_class_memory);
5513 %}
5514 
5515 // Load nodes for superwords
5516 
5517 // Load Aligned Packed Byte
5518 instruct loadV8(iRegLdst dst, memoryAlg4 mem) %{
5519   predicate(n->as_LoadVector()->memory_size() == 8);
5520   match(Set dst (LoadVector mem));
5521   ins_cost(MEMORY_REF_COST);
5522 
5523   format %{ "LD      $dst, $mem \t// load 8-byte Vector" %}
5524   size(4);
5525   ins_encode( enc_ld(dst, mem) );
5526   ins_pipe(pipe_class_memory);
5527 %}
5528 
5529 // Load Range, range = array length (=jint)
5530 instruct loadRange(iRegIdst dst, memory mem) %{
5531   match(Set dst (LoadRange mem));
5532   ins_cost(MEMORY_REF_COST);
5533 
5534   format %{ "LWZ     $dst, $mem \t// range" %}
5535   size(4);
5536   ins_encode( enc_lwz(dst, mem) );
5537   ins_pipe(pipe_class_memory);
5538 %}
5539 
5540 // Load Compressed Pointer
5541 instruct loadN(iRegNdst dst, memory mem) %{
5542   match(Set dst (LoadN mem));
5543   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5544   ins_cost(MEMORY_REF_COST);
5545 
5546   format %{ "LWZ     $dst, $mem \t// load compressed ptr" %}
5547   size(4);
5548   ins_encode( enc_lwz(dst, mem) );
5549   ins_pipe(pipe_class_memory);
5550 %}
5551 
5552 // Load Compressed Pointer acquire.
5553 instruct loadN_ac(iRegNdst dst, memory mem) %{
5554   match(Set dst (LoadN mem));
5555   ins_cost(3*MEMORY_REF_COST);
5556 
5557   format %{ "LWZ     $dst, $mem \t// load acquire compressed ptr\n\t"
5558             "TWI     $dst\n\t"
5559             "ISYNC" %}
5560   size(12);
5561   ins_encode( enc_lwz_ac(dst, mem) );
5562   ins_pipe(pipe_class_memory);
5563 %}
5564 
5565 // Load Compressed Pointer and decode it if narrow_oop_shift == 0.
5566 instruct loadN2P_unscaled(iRegPdst dst, memory mem) %{
5567   match(Set dst (DecodeN (LoadN mem)));
5568   predicate(_kids[0]->_leaf->as_Load()->is_unordered() && Universe::narrow_oop_shift() == 0);
5569   ins_cost(MEMORY_REF_COST);
5570 
5571   format %{ "LWZ     $dst, $mem \t// DecodeN (unscaled)" %}
5572   size(4);
5573   ins_encode( enc_lwz(dst, mem) );
5574   ins_pipe(pipe_class_memory);
5575 %}
5576 
5577 instruct loadN2P_klass_unscaled(iRegPdst dst, memory mem) %{
5578   match(Set dst (DecodeNKlass (LoadNKlass mem)));
5579   // SAPJVM GL 2014-05-21 Differs.
5580   predicate(Universe::narrow_klass_base() == NULL && Universe::narrow_klass_shift() == 0 &&
5581             _kids[0]->_leaf->as_Load()->is_unordered());
5582   ins_cost(MEMORY_REF_COST);
5583 
5584   format %{ "LWZ     $dst, $mem \t// DecodeN (unscaled)" %}
5585   size(4);
5586   ins_encode( enc_lwz(dst, mem) );
5587   ins_pipe(pipe_class_memory);
5588 %}
5589 
5590 // Load Pointer
5591 instruct loadP(iRegPdst dst, memoryAlg4 mem) %{
5592   match(Set dst (LoadP mem));
5593   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5594   ins_cost(MEMORY_REF_COST);
5595 
5596   format %{ "LD      $dst, $mem \t// ptr" %}
5597   size(4);
5598   ins_encode( enc_ld(dst, mem) );
5599   ins_pipe(pipe_class_memory);
5600 %}
5601 
5602 // Load Pointer acquire.
5603 instruct loadP_ac(iRegPdst dst, memoryAlg4 mem) %{
5604   match(Set dst (LoadP mem));
5605   ins_cost(3*MEMORY_REF_COST);
5606 
5607   format %{ "LD      $dst, $mem \t// ptr acquire\n\t"
5608             "TWI     $dst\n\t"
5609             "ISYNC" %}
5610   size(12);
5611   ins_encode( enc_ld_ac(dst, mem) );
5612   ins_pipe(pipe_class_memory);
5613 %}
5614 
5615 // LoadP + CastP2L
5616 instruct loadP2X(iRegLdst dst, memoryAlg4 mem) %{
5617   match(Set dst (CastP2X (LoadP mem)));
5618   predicate(_kids[0]->_leaf->as_Load()->is_unordered());
5619   ins_cost(MEMORY_REF_COST);
5620 
5621   format %{ "LD      $dst, $mem \t// ptr + p2x" %}
5622   size(4);
5623   ins_encode( enc_ld(dst, mem) );
5624   ins_pipe(pipe_class_memory);
5625 %}
5626 
5627 // Load compressed klass pointer.
5628 instruct loadNKlass(iRegNdst dst, memory mem) %{
5629   match(Set dst (LoadNKlass mem));
5630   ins_cost(MEMORY_REF_COST);
5631 
5632   format %{ "LWZ     $dst, $mem \t// compressed klass ptr" %}
5633   size(4);
5634   ins_encode( enc_lwz(dst, mem) );
5635   ins_pipe(pipe_class_memory);
5636 %}
5637 
5638 // Load Klass Pointer
5639 instruct loadKlass(iRegPdst dst, memoryAlg4 mem) %{
5640   match(Set dst (LoadKlass mem));
5641   ins_cost(MEMORY_REF_COST);
5642 
5643   format %{ "LD      $dst, $mem \t// klass ptr" %}
5644   size(4);
5645   ins_encode( enc_ld(dst, mem) );
5646   ins_pipe(pipe_class_memory);
5647 %}
5648 
5649 // Load Float
5650 instruct loadF(regF dst, memory mem) %{
5651   match(Set dst (LoadF mem));
5652   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5653   ins_cost(MEMORY_REF_COST);
5654 
5655   format %{ "LFS     $dst, $mem" %}
5656   size(4);
5657   ins_encode %{
5658     // TODO: PPC port $archOpcode(ppc64Opcode_lfs);
5659     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5660     __ lfs($dst$$FloatRegister, Idisp, $mem$$base$$Register);
5661   %}
5662   ins_pipe(pipe_class_memory);
5663 %}
5664 
5665 // Load Float acquire.
5666 instruct loadF_ac(regF dst, memory mem, flagsRegCR0 cr0) %{
5667   match(Set dst (LoadF mem));
5668   effect(TEMP cr0);
5669   ins_cost(3*MEMORY_REF_COST);
5670 
5671   format %{ "LFS     $dst, $mem \t// acquire\n\t"
5672             "FCMPU   cr0, $dst, $dst\n\t"
5673             "BNE     cr0, next\n"
5674             "next:\n\t"
5675             "ISYNC" %}
5676   size(16);
5677   ins_encode %{
5678     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
5679     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5680     Label next;
5681     __ lfs($dst$$FloatRegister, Idisp, $mem$$base$$Register);
5682     __ fcmpu(CCR0, $dst$$FloatRegister, $dst$$FloatRegister);
5683     __ bne(CCR0, next);
5684     __ bind(next);
5685     __ isync();
5686   %}
5687   ins_pipe(pipe_class_memory);
5688 %}
5689 
5690 // Load Double - aligned
5691 instruct loadD(regD dst, memory mem) %{
5692   match(Set dst (LoadD mem));
5693   predicate(n->as_Load()->is_unordered() || followed_by_acquire(n));
5694   ins_cost(MEMORY_REF_COST);
5695 
5696   format %{ "LFD     $dst, $mem" %}
5697   size(4);
5698   ins_encode( enc_lfd(dst, mem) );
5699   ins_pipe(pipe_class_memory);
5700 %}
5701 
5702 // Load Double - aligned acquire.
5703 instruct loadD_ac(regD dst, memory mem, flagsRegCR0 cr0) %{
5704   match(Set dst (LoadD mem));
5705   effect(TEMP cr0);
5706   ins_cost(3*MEMORY_REF_COST);
5707 
5708   format %{ "LFD     $dst, $mem \t// acquire\n\t"
5709             "FCMPU   cr0, $dst, $dst\n\t"
5710             "BNE     cr0, next\n"
5711             "next:\n\t"
5712             "ISYNC" %}
5713   size(16);
5714   ins_encode %{
5715     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
5716     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
5717     Label next;
5718     __ lfd($dst$$FloatRegister, Idisp, $mem$$base$$Register);
5719     __ fcmpu(CCR0, $dst$$FloatRegister, $dst$$FloatRegister);
5720     __ bne(CCR0, next);
5721     __ bind(next);
5722     __ isync();
5723   %}
5724   ins_pipe(pipe_class_memory);
5725 %}
5726 
5727 // Load Double - UNaligned
5728 instruct loadD_unaligned(regD dst, memory mem) %{
5729   match(Set dst (LoadD_unaligned mem));
5730   // predicate(...) // Unaligned_ac is not needed (and wouldn't make sense).
5731   ins_cost(MEMORY_REF_COST);
5732 
5733   format %{ "LFD     $dst, $mem" %}
5734   size(4);
5735   ins_encode( enc_lfd(dst, mem) );
5736   ins_pipe(pipe_class_memory);
5737 %}
5738 
5739 //----------Constants--------------------------------------------------------
5740 
5741 // Load MachConstantTableBase: add hi offset to global toc.
5742 // TODO: Handle hidden register r29 in bundler!
5743 instruct loadToc_hi(iRegLdst dst) %{
5744   effect(DEF dst);
5745   ins_cost(DEFAULT_COST);
5746 
5747   format %{ "ADDIS   $dst, R29, DISP.hi \t// load TOC hi" %}
5748   size(4);
5749   ins_encode %{
5750     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
5751     __ calculate_address_from_global_toc_hi16only($dst$$Register, __ method_toc());
5752   %}
5753   ins_pipe(pipe_class_default);
5754 %}
5755 
5756 // Load MachConstantTableBase: add lo offset to global toc.
5757 instruct loadToc_lo(iRegLdst dst, iRegLdst src) %{
5758   effect(DEF dst, USE src);
5759   ins_cost(DEFAULT_COST);
5760 
5761   format %{ "ADDI    $dst, $src, DISP.lo \t// load TOC lo" %}
5762   size(4);
5763   ins_encode %{
5764     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
5765     __ calculate_address_from_global_toc_lo16only($dst$$Register, __ method_toc());
5766   %}
5767   ins_pipe(pipe_class_default);
5768 %}
5769 
5770 // Load 16-bit integer constant 0xssss????
5771 instruct loadConI16(iRegIdst dst, immI16 src) %{
5772   match(Set dst src);
5773 
5774   format %{ "LI      $dst, $src" %}
5775   size(4);
5776   ins_encode %{
5777     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
5778     __ li($dst$$Register, (int)((short)($src$$constant & 0xFFFF)));
5779   %}
5780   ins_pipe(pipe_class_default);
5781 %}
5782 
5783 // Load integer constant 0x????0000
5784 instruct loadConIhi16(iRegIdst dst, immIhi16 src) %{
5785   match(Set dst src);
5786   ins_cost(DEFAULT_COST);
5787 
5788   format %{ "LIS     $dst, $src.hi" %}
5789   size(4);
5790   ins_encode %{
5791     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
5792     // Lis sign extends 16-bit src then shifts it 16 bit to the left.
5793     __ lis($dst$$Register, (int)((short)(($src$$constant & 0xFFFF0000) >> 16)));
5794   %}
5795   ins_pipe(pipe_class_default);
5796 %}
5797 
5798 // Part 2 of loading 32 bit constant: hi16 is is src1 (properly shifted
5799 // and sign extended), this adds the low 16 bits.
5800 instruct loadConI32_lo16(iRegIdst dst, iRegIsrc src1, immI16 src2) %{
5801   // no match-rule, false predicate
5802   effect(DEF dst, USE src1, USE src2);
5803   predicate(false);
5804 
5805   format %{ "ORI     $dst, $src1.hi, $src2.lo" %}
5806   size(4);
5807   ins_encode %{
5808     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
5809     __ ori($dst$$Register, $src1$$Register, ($src2$$constant) & 0xFFFF);
5810   %}
5811   ins_pipe(pipe_class_default);
5812 %}
5813 
5814 instruct loadConI_Ex(iRegIdst dst, immI src) %{
5815   match(Set dst src);
5816   ins_cost(DEFAULT_COST*2);
5817 
5818   expand %{
5819     // Would like to use $src$$constant.
5820     immI16 srcLo %{ _opnds[1]->constant() %}
5821     // srcHi can be 0000 if srcLo sign-extends to a negative number.
5822     immIhi16 srcHi %{ _opnds[1]->constant() %}
5823     iRegIdst tmpI;
5824     loadConIhi16(tmpI, srcHi);
5825     loadConI32_lo16(dst, tmpI, srcLo);
5826   %}
5827 %}
5828 
5829 // No constant pool entries required.
5830 instruct loadConL16(iRegLdst dst, immL16 src) %{
5831   match(Set dst src);
5832 
5833   format %{ "LI      $dst, $src \t// long" %}
5834   size(4);
5835   ins_encode %{
5836     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
5837     __ li($dst$$Register, (int)((short) ($src$$constant & 0xFFFF)));
5838   %}
5839   ins_pipe(pipe_class_default);
5840 %}
5841 
5842 // Load long constant 0xssssssss????0000
5843 instruct loadConL32hi16(iRegLdst dst, immL32hi16 src) %{
5844   match(Set dst src);
5845   ins_cost(DEFAULT_COST);
5846 
5847   format %{ "LIS     $dst, $src.hi \t// long" %}
5848   size(4);
5849   ins_encode %{
5850     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
5851     __ lis($dst$$Register, (int)((short)(($src$$constant & 0xFFFF0000) >> 16)));
5852   %}
5853   ins_pipe(pipe_class_default);
5854 %}
5855 
5856 // To load a 32 bit constant: merge lower 16 bits into already loaded
5857 // high 16 bits.
5858 instruct loadConL32_lo16(iRegLdst dst, iRegLsrc src1, immL16 src2) %{
5859   // no match-rule, false predicate
5860   effect(DEF dst, USE src1, USE src2);
5861   predicate(false);
5862 
5863   format %{ "ORI     $dst, $src1, $src2.lo" %}
5864   size(4);
5865   ins_encode %{
5866     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
5867     __ ori($dst$$Register, $src1$$Register, ($src2$$constant) & 0xFFFF);
5868   %}
5869   ins_pipe(pipe_class_default);
5870 %}
5871 
5872 // Load 32-bit long constant
5873 instruct loadConL32_Ex(iRegLdst dst, immL32 src) %{
5874   match(Set dst src);
5875   ins_cost(DEFAULT_COST*2);
5876 
5877   expand %{
5878     // Would like to use $src$$constant.
5879     immL16     srcLo %{ _opnds[1]->constant() /*& 0x0000FFFFL */%}
5880     // srcHi can be 0000 if srcLo sign-extends to a negative number.
5881     immL32hi16 srcHi %{ _opnds[1]->constant() /*& 0xFFFF0000L */%}
5882     iRegLdst tmpL;
5883     loadConL32hi16(tmpL, srcHi);
5884     loadConL32_lo16(dst, tmpL, srcLo);
5885   %}
5886 %}
5887 
5888 // Load long constant 0x????000000000000.
5889 instruct loadConLhighest16_Ex(iRegLdst dst, immLhighest16 src) %{
5890   match(Set dst src);
5891   ins_cost(DEFAULT_COST);
5892 
5893   expand %{
5894     immL32hi16 srcHi %{ _opnds[1]->constant() >> 32 /*& 0xFFFF0000L */%}
5895     immI shift32 %{ 32 %}
5896     iRegLdst tmpL;
5897     loadConL32hi16(tmpL, srcHi);
5898     lshiftL_regL_immI(dst, tmpL, shift32);
5899   %}
5900 %}
5901 
5902 // Expand node for constant pool load: small offset.
5903 instruct loadConL(iRegLdst dst, immL src, iRegLdst toc) %{
5904   effect(DEF dst, USE src, USE toc);
5905   ins_cost(MEMORY_REF_COST);
5906 
5907   ins_num_consts(1);
5908   // Needed so that CallDynamicJavaDirect can compute the address of this
5909   // instruction for relocation.
5910   ins_field_cbuf_insts_offset(int);
5911 
5912   format %{ "LD      $dst, offset, $toc \t// load long $src from TOC" %}
5913   size(4);
5914   ins_encode( enc_load_long_constL(dst, src, toc) );
5915   ins_pipe(pipe_class_memory);
5916 %}
5917 
5918 // Expand node for constant pool load: large offset.
5919 instruct loadConL_hi(iRegLdst dst, immL src, iRegLdst toc) %{
5920   effect(DEF dst, USE src, USE toc);
5921   predicate(false);
5922 
5923   ins_num_consts(1);
5924   ins_field_const_toc_offset(int);
5925   // Needed so that CallDynamicJavaDirect can compute the address of this
5926   // instruction for relocation.
5927   ins_field_cbuf_insts_offset(int);
5928 
5929   format %{ "ADDIS   $dst, $toc, offset \t// load long $src from TOC (hi)" %}
5930   size(4);
5931   ins_encode( enc_load_long_constL_hi(dst, toc, src) );
5932   ins_pipe(pipe_class_default);
5933 %}
5934 
5935 // Expand node for constant pool load: large offset.
5936 // No constant pool entries required.
5937 instruct loadConL_lo(iRegLdst dst, immL src, iRegLdst base) %{
5938   effect(DEF dst, USE src, USE base);
5939   predicate(false);
5940 
5941   ins_field_const_toc_offset_hi_node(loadConL_hiNode*);
5942 
5943   format %{ "LD      $dst, offset, $base \t// load long $src from TOC (lo)" %}
5944   size(4);
5945   ins_encode %{
5946     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
5947     int offset = ra_->C->in_scratch_emit_size() ? 0 : _const_toc_offset_hi_node->_const_toc_offset;
5948     __ ld($dst$$Register, MacroAssembler::largeoffset_si16_si16_lo(offset), $base$$Register);
5949   %}
5950   ins_pipe(pipe_class_memory);
5951 %}
5952 
5953 // Load long constant from constant table. Expand in case of
5954 // offset > 16 bit is needed.
5955 // Adlc adds toc node MachConstantTableBase.
5956 instruct loadConL_Ex(iRegLdst dst, immL src) %{
5957   match(Set dst src);
5958   ins_cost(MEMORY_REF_COST);
5959 
5960   format %{ "LD      $dst, offset, $constanttablebase\t// load long $src from table, postalloc expanded" %}
5961   // We can not inline the enc_class for the expand as that does not support constanttablebase.
5962   postalloc_expand( postalloc_expand_load_long_constant(dst, src, constanttablebase) );
5963 %}
5964 
5965 // Load NULL as compressed oop.
5966 instruct loadConN0(iRegNdst dst, immN_0 src) %{
5967   match(Set dst src);
5968   ins_cost(DEFAULT_COST);
5969 
5970   format %{ "LI      $dst, $src \t// compressed ptr" %}
5971   size(4);
5972   ins_encode %{
5973     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
5974     __ li($dst$$Register, 0);
5975   %}
5976   ins_pipe(pipe_class_default);
5977 %}
5978 
5979 // Load hi part of compressed oop constant.
5980 instruct loadConN_hi(iRegNdst dst, immN src) %{
5981   effect(DEF dst, USE src);
5982   ins_cost(DEFAULT_COST);
5983 
5984   format %{ "LIS     $dst, $src \t// narrow oop hi" %}
5985   size(4);
5986   ins_encode %{
5987     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
5988     __ lis($dst$$Register, (int)(short)(($src$$constant >> 16) & 0xffff));
5989   %}
5990   ins_pipe(pipe_class_default);
5991 %}
5992 
5993 // Add lo part of compressed oop constant to already loaded hi part.
5994 instruct loadConN_lo(iRegNdst dst, iRegNsrc src1, immN src2) %{
5995   effect(DEF dst, USE src1, USE src2);
5996   ins_cost(DEFAULT_COST);
5997 
5998   format %{ "ORI     $dst, $src1, $src2 \t// narrow oop lo" %}
5999   size(4);
6000   ins_encode %{
6001     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
6002     assert(__ oop_recorder() != NULL, "this assembler needs an OopRecorder");
6003     int oop_index = __ oop_recorder()->find_index((jobject)$src2$$constant);
6004     RelocationHolder rspec = oop_Relocation::spec(oop_index);
6005     __ relocate(rspec, 1);
6006     __ ori($dst$$Register, $src1$$Register, $src2$$constant & 0xffff);
6007   %}
6008   ins_pipe(pipe_class_default);
6009 %}
6010 
6011 // Needed to postalloc expand loadConN: ConN is loaded as ConI
6012 // leaving the upper 32 bits with sign-extension bits.
6013 // This clears these bits: dst = src & 0xFFFFFFFF.
6014 // TODO: Eventually call this maskN_regN_FFFFFFFF.
6015 instruct clearMs32b(iRegNdst dst, iRegNsrc src) %{
6016   effect(DEF dst, USE src);
6017   predicate(false);
6018 
6019   format %{ "MASK    $dst, $src, 0xFFFFFFFF" %} // mask
6020   size(4);
6021   ins_encode %{
6022     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
6023     __ clrldi($dst$$Register, $src$$Register, 0x20);
6024   %}
6025   ins_pipe(pipe_class_default);
6026 %}
6027 
6028 // Optimize DecodeN for disjoint base.
6029 // Load base of compressed oops into a register
6030 instruct loadBase(iRegLdst dst) %{
6031   effect(DEF dst);
6032 
6033   format %{ "LoadConst $dst, heapbase" %}
6034   ins_encode %{
6035     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6036     __ load_const_optimized($dst$$Register, Universe::narrow_oop_base(), R0);
6037   %}
6038   ins_pipe(pipe_class_default);
6039 %}
6040 
6041 // Loading ConN must be postalloc expanded so that edges between
6042 // the nodes are safe. They may not interfere with a safepoint.
6043 // GL TODO: This needs three instructions: better put this into the constant pool.
6044 instruct loadConN_Ex(iRegNdst dst, immN src) %{
6045   match(Set dst src);
6046   ins_cost(DEFAULT_COST*2);
6047 
6048   format %{ "LoadN   $dst, $src \t// postalloc expanded" %} // mask
6049   postalloc_expand %{
6050     MachNode *m1 = new loadConN_hiNode();
6051     MachNode *m2 = new loadConN_loNode();
6052     MachNode *m3 = new clearMs32bNode();
6053     m1->add_req(NULL);
6054     m2->add_req(NULL, m1);
6055     m3->add_req(NULL, m2);
6056     m1->_opnds[0] = op_dst;
6057     m1->_opnds[1] = op_src;
6058     m2->_opnds[0] = op_dst;
6059     m2->_opnds[1] = op_dst;
6060     m2->_opnds[2] = op_src;
6061     m3->_opnds[0] = op_dst;
6062     m3->_opnds[1] = op_dst;
6063     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6064     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6065     ra_->set_pair(m3->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6066     nodes->push(m1);
6067     nodes->push(m2);
6068     nodes->push(m3);
6069   %}
6070 %}
6071 
6072 // We have seen a safepoint between the hi and lo parts, and this node was handled
6073 // as an oop. Therefore this needs a match rule so that build_oop_map knows this is
6074 // not a narrow oop.
6075 instruct loadConNKlass_hi(iRegNdst dst, immNKlass_NM src) %{
6076   match(Set dst src);
6077   effect(DEF dst, USE src);
6078   ins_cost(DEFAULT_COST);
6079 
6080   format %{ "LIS     $dst, $src \t// narrow klass hi" %}
6081   size(4);
6082   ins_encode %{
6083     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
6084     intptr_t Csrc = Klass::encode_klass((Klass *)$src$$constant);
6085     __ lis($dst$$Register, (int)(short)((Csrc >> 16) & 0xffff));
6086   %}
6087   ins_pipe(pipe_class_default);
6088 %}
6089 
6090 // As loadConNKlass_hi this must be recognized as narrow klass, not oop!
6091 instruct loadConNKlass_mask(iRegNdst dst, immNKlass_NM src1, iRegNsrc src2) %{
6092   match(Set dst src1);
6093   effect(TEMP src2);
6094   ins_cost(DEFAULT_COST);
6095 
6096   format %{ "MASK    $dst, $src2, 0xFFFFFFFF" %} // mask
6097   size(4);
6098   ins_encode %{
6099     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
6100     __ clrldi($dst$$Register, $src2$$Register, 0x20);
6101   %}
6102   ins_pipe(pipe_class_default);
6103 %}
6104 
6105 // This needs a match rule so that build_oop_map knows this is
6106 // not a narrow oop.
6107 instruct loadConNKlass_lo(iRegNdst dst, immNKlass_NM src1, iRegNsrc src2) %{
6108   match(Set dst src1);
6109   effect(TEMP src2);
6110   ins_cost(DEFAULT_COST);
6111 
6112   format %{ "ORI     $dst, $src1, $src2 \t// narrow klass lo" %}
6113   size(4);
6114   ins_encode %{
6115     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
6116     intptr_t Csrc = Klass::encode_klass((Klass *)$src1$$constant);
6117     assert(__ oop_recorder() != NULL, "this assembler needs an OopRecorder");
6118     int klass_index = __ oop_recorder()->find_index((Klass *)$src1$$constant);
6119     RelocationHolder rspec = metadata_Relocation::spec(klass_index);
6120 
6121     __ relocate(rspec, 1);
6122     __ ori($dst$$Register, $src2$$Register, Csrc & 0xffff);
6123   %}
6124   ins_pipe(pipe_class_default);
6125 %}
6126 
6127 // Loading ConNKlass must be postalloc expanded so that edges between
6128 // the nodes are safe. They may not interfere with a safepoint.
6129 instruct loadConNKlass_Ex(iRegNdst dst, immNKlass src) %{
6130   match(Set dst src);
6131   ins_cost(DEFAULT_COST*2);
6132 
6133   format %{ "LoadN   $dst, $src \t// postalloc expanded" %} // mask
6134   postalloc_expand %{
6135     // Load high bits into register. Sign extended.
6136     MachNode *m1 = new loadConNKlass_hiNode();
6137     m1->add_req(NULL);
6138     m1->_opnds[0] = op_dst;
6139     m1->_opnds[1] = op_src;
6140     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6141     nodes->push(m1);
6142 
6143     MachNode *m2 = m1;
6144     if (!Assembler::is_uimm((jlong)Klass::encode_klass((Klass *)op_src->constant()), 31)) {
6145       // Value might be 1-extended. Mask out these bits.
6146       m2 = new loadConNKlass_maskNode();
6147       m2->add_req(NULL, m1);
6148       m2->_opnds[0] = op_dst;
6149       m2->_opnds[1] = op_src;
6150       m2->_opnds[2] = op_dst;
6151       ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6152       nodes->push(m2);
6153     }
6154 
6155     MachNode *m3 = new loadConNKlass_loNode();
6156     m3->add_req(NULL, m2);
6157     m3->_opnds[0] = op_dst;
6158     m3->_opnds[1] = op_src;
6159     m3->_opnds[2] = op_dst;
6160     ra_->set_pair(m3->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6161     nodes->push(m3);
6162   %}
6163 %}
6164 
6165 // 0x1 is used in object initialization (initial object header).
6166 // No constant pool entries required.
6167 instruct loadConP0or1(iRegPdst dst, immP_0or1 src) %{
6168   match(Set dst src);
6169 
6170   format %{ "LI      $dst, $src \t// ptr" %}
6171   size(4);
6172   ins_encode %{
6173     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
6174     __ li($dst$$Register, (int)((short)($src$$constant & 0xFFFF)));
6175   %}
6176   ins_pipe(pipe_class_default);
6177 %}
6178 
6179 // Expand node for constant pool load: small offset.
6180 // The match rule is needed to generate the correct bottom_type(),
6181 // however this node should never match. The use of predicate is not
6182 // possible since ADLC forbids predicates for chain rules. The higher
6183 // costs do not prevent matching in this case. For that reason the
6184 // operand immP_NM with predicate(false) is used.
6185 instruct loadConP(iRegPdst dst, immP_NM src, iRegLdst toc) %{
6186   match(Set dst src);
6187   effect(TEMP toc);
6188 
6189   ins_num_consts(1);
6190 
6191   format %{ "LD      $dst, offset, $toc \t// load ptr $src from TOC" %}
6192   size(4);
6193   ins_encode( enc_load_long_constP(dst, src, toc) );
6194   ins_pipe(pipe_class_memory);
6195 %}
6196 
6197 // Expand node for constant pool load: large offset.
6198 instruct loadConP_hi(iRegPdst dst, immP_NM src, iRegLdst toc) %{
6199   effect(DEF dst, USE src, USE toc);
6200   predicate(false);
6201 
6202   ins_num_consts(1);
6203   ins_field_const_toc_offset(int);
6204 
6205   format %{ "ADDIS   $dst, $toc, offset \t// load ptr $src from TOC (hi)" %}
6206   size(4);
6207   ins_encode( enc_load_long_constP_hi(dst, src, toc) );
6208   ins_pipe(pipe_class_default);
6209 %}
6210 
6211 // Expand node for constant pool load: large offset.
6212 instruct loadConP_lo(iRegPdst dst, immP_NM src, iRegLdst base) %{
6213   match(Set dst src);
6214   effect(TEMP base);
6215 
6216   ins_field_const_toc_offset_hi_node(loadConP_hiNode*);
6217 
6218   format %{ "LD      $dst, offset, $base \t// load ptr $src from TOC (lo)" %}
6219   size(4);
6220   ins_encode %{
6221     // TODO: PPC port $archOpcode(ppc64Opcode_ld);
6222     int offset = ra_->C->in_scratch_emit_size() ? 0 : _const_toc_offset_hi_node->_const_toc_offset;
6223     __ ld($dst$$Register, MacroAssembler::largeoffset_si16_si16_lo(offset), $base$$Register);
6224   %}
6225   ins_pipe(pipe_class_memory);
6226 %}
6227 
6228 // Load pointer constant from constant table. Expand in case an
6229 // offset > 16 bit is needed.
6230 // Adlc adds toc node MachConstantTableBase.
6231 instruct loadConP_Ex(iRegPdst dst, immP src) %{
6232   match(Set dst src);
6233   ins_cost(MEMORY_REF_COST);
6234 
6235   // This rule does not use "expand" because then
6236   // the result type is not known to be an Oop.  An ADLC
6237   // enhancement will be needed to make that work - not worth it!
6238 
6239   // If this instruction rematerializes, it prolongs the live range
6240   // of the toc node, causing illegal graphs.
6241   // assert(edge_from_to(_reg_node[reg_lo],def)) fails in verify_good_schedule().
6242   ins_cannot_rematerialize(true);
6243 
6244   format %{ "LD    $dst, offset, $constanttablebase \t//  load ptr $src from table, postalloc expanded" %}
6245   postalloc_expand( postalloc_expand_load_ptr_constant(dst, src, constanttablebase) );
6246 %}
6247 
6248 // Expand node for constant pool load: small offset.
6249 instruct loadConF(regF dst, immF src, iRegLdst toc) %{
6250   effect(DEF dst, USE src, USE toc);
6251   ins_cost(MEMORY_REF_COST);
6252 
6253   ins_num_consts(1);
6254 
6255   format %{ "LFS     $dst, offset, $toc \t// load float $src from TOC" %}
6256   size(4);
6257   ins_encode %{
6258     // TODO: PPC port $archOpcode(ppc64Opcode_lfs);
6259     address float_address = __ float_constant($src$$constant);
6260     __ lfs($dst$$FloatRegister, __ offset_to_method_toc(float_address), $toc$$Register);
6261   %}
6262   ins_pipe(pipe_class_memory);
6263 %}
6264 
6265 // Expand node for constant pool load: large offset.
6266 instruct loadConFComp(regF dst, immF src, iRegLdst toc) %{
6267   effect(DEF dst, USE src, USE toc);
6268   ins_cost(MEMORY_REF_COST);
6269 
6270   ins_num_consts(1);
6271 
6272   format %{ "ADDIS   $toc, $toc, offset_hi\n\t"
6273             "LFS     $dst, offset_lo, $toc \t// load float $src from TOC (hi/lo)\n\t"
6274             "ADDIS   $toc, $toc, -offset_hi"%}
6275   size(12);
6276   ins_encode %{
6277     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6278     FloatRegister Rdst    = $dst$$FloatRegister;
6279     Register Rtoc         = $toc$$Register;
6280     address float_address = __ float_constant($src$$constant);
6281     int offset            = __ offset_to_method_toc(float_address);
6282     int hi = (offset + (1<<15))>>16;
6283     int lo = offset - hi * (1<<16);
6284 
6285     __ addis(Rtoc, Rtoc, hi);
6286     __ lfs(Rdst, lo, Rtoc);
6287     __ addis(Rtoc, Rtoc, -hi);
6288   %}
6289   ins_pipe(pipe_class_memory);
6290 %}
6291 
6292 // Adlc adds toc node MachConstantTableBase.
6293 instruct loadConF_Ex(regF dst, immF src) %{
6294   match(Set dst src);
6295   ins_cost(MEMORY_REF_COST);
6296 
6297   // See loadConP.
6298   ins_cannot_rematerialize(true);
6299 
6300   format %{ "LFS     $dst, offset, $constanttablebase \t// load $src from table, postalloc expanded" %}
6301   postalloc_expand( postalloc_expand_load_float_constant(dst, src, constanttablebase) );
6302 %}
6303 
6304 // Expand node for constant pool load: small offset.
6305 instruct loadConD(regD dst, immD src, iRegLdst toc) %{
6306   effect(DEF dst, USE src, USE toc);
6307   ins_cost(MEMORY_REF_COST);
6308 
6309   ins_num_consts(1);
6310 
6311   format %{ "LFD     $dst, offset, $toc \t// load double $src from TOC" %}
6312   size(4);
6313   ins_encode %{
6314     // TODO: PPC port $archOpcode(ppc64Opcode_lfd);
6315     int offset =  __ offset_to_method_toc(__ double_constant($src$$constant));
6316     __ lfd($dst$$FloatRegister, offset, $toc$$Register);
6317   %}
6318   ins_pipe(pipe_class_memory);
6319 %}
6320 
6321 // Expand node for constant pool load: large offset.
6322 instruct loadConDComp(regD dst, immD src, iRegLdst toc) %{
6323   effect(DEF dst, USE src, USE toc);
6324   ins_cost(MEMORY_REF_COST);
6325 
6326   ins_num_consts(1);
6327 
6328   format %{ "ADDIS   $toc, $toc, offset_hi\n\t"
6329             "LFD     $dst, offset_lo, $toc \t// load double $src from TOC (hi/lo)\n\t"
6330             "ADDIS   $toc, $toc, -offset_hi" %}
6331   size(12);
6332   ins_encode %{
6333     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6334     FloatRegister Rdst    = $dst$$FloatRegister;
6335     Register      Rtoc    = $toc$$Register;
6336     address float_address = __ double_constant($src$$constant);
6337     int offset            = __ offset_to_method_toc(float_address);
6338     int hi = (offset + (1<<15))>>16;
6339     int lo = offset - hi * (1<<16);
6340 
6341     __ addis(Rtoc, Rtoc, hi);
6342     __ lfd(Rdst, lo, Rtoc);
6343     __ addis(Rtoc, Rtoc, -hi);
6344   %}
6345   ins_pipe(pipe_class_memory);
6346 %}
6347 
6348 // Adlc adds toc node MachConstantTableBase.
6349 instruct loadConD_Ex(regD dst, immD src) %{
6350   match(Set dst src);
6351   ins_cost(MEMORY_REF_COST);
6352 
6353   // See loadConP.
6354   ins_cannot_rematerialize(true);
6355 
6356   format %{ "ConD    $dst, offset, $constanttablebase \t// load $src from table, postalloc expanded" %}
6357   postalloc_expand( postalloc_expand_load_double_constant(dst, src, constanttablebase) );
6358 %}
6359 
6360 // Prefetch instructions.
6361 // Must be safe to execute with invalid address (cannot fault).
6362 
6363 // Special prefetch versions which use the dcbz instruction.
6364 instruct prefetch_alloc_zero(indirectMemory mem, iRegLsrc src) %{
6365   match(PrefetchAllocation (AddP mem src));
6366   predicate(AllocatePrefetchStyle == 3);
6367   ins_cost(MEMORY_REF_COST);
6368 
6369   format %{ "PREFETCH $mem, 2, $src \t// Prefetch write-many with zero" %}
6370   size(4);
6371   ins_encode %{
6372     // TODO: PPC port $archOpcode(ppc64Opcode_dcbtst);
6373     __ dcbz($src$$Register, $mem$$base$$Register);
6374   %}
6375   ins_pipe(pipe_class_memory);
6376 %}
6377 
6378 instruct prefetch_alloc_zero_no_offset(indirectMemory mem) %{
6379   match(PrefetchAllocation mem);
6380   predicate(AllocatePrefetchStyle == 3);
6381   ins_cost(MEMORY_REF_COST);
6382 
6383   format %{ "PREFETCH $mem, 2 \t// Prefetch write-many with zero" %}
6384   size(4);
6385   ins_encode %{
6386     // TODO: PPC port $archOpcode(ppc64Opcode_dcbtst);
6387     __ dcbz($mem$$base$$Register);
6388   %}
6389   ins_pipe(pipe_class_memory);
6390 %}
6391 
6392 instruct prefetch_alloc(indirectMemory mem, iRegLsrc src) %{
6393   match(PrefetchAllocation (AddP mem src));
6394   predicate(AllocatePrefetchStyle != 3);
6395   ins_cost(MEMORY_REF_COST);
6396 
6397   format %{ "PREFETCH $mem, 2, $src \t// Prefetch write-many" %}
6398   size(4);
6399   ins_encode %{
6400     // TODO: PPC port $archOpcode(ppc64Opcode_dcbtst);
6401     __ dcbtst($src$$Register, $mem$$base$$Register);
6402   %}
6403   ins_pipe(pipe_class_memory);
6404 %}
6405 
6406 instruct prefetch_alloc_no_offset(indirectMemory mem) %{
6407   match(PrefetchAllocation mem);
6408   predicate(AllocatePrefetchStyle != 3);
6409   ins_cost(MEMORY_REF_COST);
6410 
6411   format %{ "PREFETCH $mem, 2 \t// Prefetch write-many" %}
6412   size(4);
6413   ins_encode %{
6414     // TODO: PPC port $archOpcode(ppc64Opcode_dcbtst);
6415     __ dcbtst($mem$$base$$Register);
6416   %}
6417   ins_pipe(pipe_class_memory);
6418 %}
6419 
6420 //----------Store Instructions-------------------------------------------------
6421 
6422 // Store Byte
6423 instruct storeB(memory mem, iRegIsrc src) %{
6424   match(Set mem (StoreB mem src));
6425   ins_cost(MEMORY_REF_COST);
6426 
6427   format %{ "STB     $src, $mem \t// byte" %}
6428   size(4);
6429   ins_encode %{
6430     // TODO: PPC port $archOpcode(ppc64Opcode_stb);
6431     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
6432     __ stb($src$$Register, Idisp, $mem$$base$$Register);
6433   %}
6434   ins_pipe(pipe_class_memory);
6435 %}
6436 
6437 // Store Char/Short
6438 instruct storeC(memory mem, iRegIsrc src) %{
6439   match(Set mem (StoreC mem src));
6440   ins_cost(MEMORY_REF_COST);
6441 
6442   format %{ "STH     $src, $mem \t// short" %}
6443   size(4);
6444   ins_encode %{
6445     // TODO: PPC port $archOpcode(ppc64Opcode_sth);
6446     int Idisp = $mem$$disp + frame_slots_bias($mem$$base, ra_);
6447     __ sth($src$$Register, Idisp, $mem$$base$$Register);
6448   %}
6449   ins_pipe(pipe_class_memory);
6450 %}
6451 
6452 // Store Integer
6453 instruct storeI(memory mem, iRegIsrc src) %{
6454   match(Set mem (StoreI mem src));
6455   ins_cost(MEMORY_REF_COST);
6456 
6457   format %{ "STW     $src, $mem" %}
6458   size(4);
6459   ins_encode( enc_stw(src, mem) );
6460   ins_pipe(pipe_class_memory);
6461 %}
6462 
6463 // ConvL2I + StoreI.
6464 instruct storeI_convL2I(memory mem, iRegLsrc src) %{
6465   match(Set mem (StoreI mem (ConvL2I src)));
6466   ins_cost(MEMORY_REF_COST);
6467 
6468   format %{ "STW     l2i($src), $mem" %}
6469   size(4);
6470   ins_encode( enc_stw(src, mem) );
6471   ins_pipe(pipe_class_memory);
6472 %}
6473 
6474 // Store Long
6475 instruct storeL(memoryAlg4 mem, iRegLsrc src) %{
6476   match(Set mem (StoreL mem src));
6477   ins_cost(MEMORY_REF_COST);
6478 
6479   format %{ "STD     $src, $mem \t// long" %}
6480   size(4);
6481   ins_encode( enc_std(src, mem) );
6482   ins_pipe(pipe_class_memory);
6483 %}
6484 
6485 // Store super word nodes.
6486 
6487 // Store Aligned Packed Byte long register to memory
6488 instruct storeA8B(memoryAlg4 mem, iRegLsrc src) %{
6489   predicate(n->as_StoreVector()->memory_size() == 8);
6490   match(Set mem (StoreVector mem src));
6491   ins_cost(MEMORY_REF_COST);
6492 
6493   format %{ "STD     $mem, $src \t// packed8B" %}
6494   size(4);
6495   ins_encode( enc_std(src, mem) );
6496   ins_pipe(pipe_class_memory);
6497 %}
6498 
6499 // Store Compressed Oop
6500 instruct storeN(memory dst, iRegN_P2N src) %{
6501   match(Set dst (StoreN dst src));
6502   ins_cost(MEMORY_REF_COST);
6503 
6504   format %{ "STW     $src, $dst \t// compressed oop" %}
6505   size(4);
6506   ins_encode( enc_stw(src, dst) );
6507   ins_pipe(pipe_class_memory);
6508 %}
6509 
6510 // Store Compressed KLass
6511 instruct storeNKlass(memory dst, iRegN_P2N src) %{
6512   match(Set dst (StoreNKlass dst src));
6513   ins_cost(MEMORY_REF_COST);
6514 
6515   format %{ "STW     $src, $dst \t// compressed klass" %}
6516   size(4);
6517   ins_encode( enc_stw(src, dst) );
6518   ins_pipe(pipe_class_memory);
6519 %}
6520 
6521 // Store Pointer
6522 instruct storeP(memoryAlg4 dst, iRegPsrc src) %{
6523   match(Set dst (StoreP dst src));
6524   ins_cost(MEMORY_REF_COST);
6525 
6526   format %{ "STD     $src, $dst \t// ptr" %}
6527   size(4);
6528   ins_encode( enc_std(src, dst) );
6529   ins_pipe(pipe_class_memory);
6530 %}
6531 
6532 // Store Float
6533 instruct storeF(memory mem, regF src) %{
6534   match(Set mem (StoreF mem src));
6535   ins_cost(MEMORY_REF_COST);
6536 
6537   format %{ "STFS    $src, $mem" %}
6538   size(4);
6539   ins_encode( enc_stfs(src, mem) );
6540   ins_pipe(pipe_class_memory);
6541 %}
6542 
6543 // Store Double
6544 instruct storeD(memory mem, regD src) %{
6545   match(Set mem (StoreD mem src));
6546   ins_cost(MEMORY_REF_COST);
6547 
6548   format %{ "STFD    $src, $mem" %}
6549   size(4);
6550   ins_encode( enc_stfd(src, mem) );
6551   ins_pipe(pipe_class_memory);
6552 %}
6553 
6554 //----------Store Instructions With Zeros--------------------------------------
6555 
6556 // Card-mark for CMS garbage collection.
6557 // This cardmark does an optimization so that it must not always
6558 // do a releasing store. For this, it gets the address of
6559 // CMSCollectorCardTableModRefBSExt::_requires_release as input.
6560 // (Using releaseFieldAddr in the match rule is a hack.)
6561 instruct storeCM_CMS(memory mem, iRegLdst releaseFieldAddr, flagsReg crx) %{
6562   match(Set mem (StoreCM mem releaseFieldAddr));
6563   effect(TEMP crx);
6564   predicate(false);
6565   ins_cost(MEMORY_REF_COST);
6566 
6567   // See loadConP.
6568   ins_cannot_rematerialize(true);
6569 
6570   format %{ "STB     #0, $mem \t// CMS card-mark byte (must be 0!), checking requires_release in [$releaseFieldAddr]" %}
6571   ins_encode( enc_cms_card_mark(mem, releaseFieldAddr, crx) );
6572   ins_pipe(pipe_class_memory);
6573 %}
6574 
6575 // Card-mark for CMS garbage collection.
6576 // This cardmark does an optimization so that it must not always
6577 // do a releasing store. For this, it needs the constant address of
6578 // CMSCollectorCardTableModRefBSExt::_requires_release.
6579 // This constant address is split off here by expand so we can use
6580 // adlc / matcher functionality to load it from the constant section.
6581 instruct storeCM_CMS_ExEx(memory mem, immI_0 zero) %{
6582   match(Set mem (StoreCM mem zero));
6583   predicate(UseConcMarkSweepGC);
6584 
6585   expand %{
6586     immL baseImm %{ 0 /* TODO: PPC port (jlong)CMSCollectorCardTableModRefBSExt::requires_release_address() */ %}
6587     iRegLdst releaseFieldAddress;
6588     flagsReg crx;
6589     loadConL_Ex(releaseFieldAddress, baseImm);
6590     storeCM_CMS(mem, releaseFieldAddress, crx);
6591   %}
6592 %}
6593 
6594 instruct storeCM_G1(memory mem, immI_0 zero) %{
6595   match(Set mem (StoreCM mem zero));
6596   predicate(UseG1GC);
6597   ins_cost(MEMORY_REF_COST);
6598 
6599   ins_cannot_rematerialize(true);
6600 
6601   format %{ "STB     #0, $mem \t// CMS card-mark byte store (G1)" %}
6602   size(8);
6603   ins_encode %{
6604     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6605     __ li(R0, 0);
6606     //__ release(); // G1: oops are allowed to get visible after dirty marking
6607     guarantee($mem$$base$$Register != R1_SP, "use frame_slots_bias");
6608     __ stb(R0, $mem$$disp, $mem$$base$$Register);
6609   %}
6610   ins_pipe(pipe_class_memory);
6611 %}
6612 
6613 // Convert oop pointer into compressed form.
6614 
6615 // Nodes for postalloc expand.
6616 
6617 // Shift node for expand.
6618 instruct encodeP_shift(iRegNdst dst, iRegNsrc src) %{
6619   // The match rule is needed to make it a 'MachTypeNode'!
6620   match(Set dst (EncodeP src));
6621   predicate(false);
6622 
6623   format %{ "SRDI    $dst, $src, 3 \t// encode" %}
6624   size(4);
6625   ins_encode %{
6626     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
6627     __ srdi($dst$$Register, $src$$Register, Universe::narrow_oop_shift() & 0x3f);
6628   %}
6629   ins_pipe(pipe_class_default);
6630 %}
6631 
6632 // Add node for expand.
6633 instruct encodeP_sub(iRegPdst dst, iRegPdst src) %{
6634   // The match rule is needed to make it a 'MachTypeNode'!
6635   match(Set dst (EncodeP src));
6636   predicate(false);
6637 
6638   format %{ "SUB     $dst, $src, oop_base \t// encode" %}
6639   ins_encode %{
6640     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6641     __ sub_const_optimized($dst$$Register, $src$$Register, Universe::narrow_oop_base(), R0);
6642   %}
6643   ins_pipe(pipe_class_default);
6644 %}
6645 
6646 // Conditional sub base.
6647 instruct cond_sub_base(iRegNdst dst, flagsRegSrc crx, iRegPsrc src1) %{
6648   // The match rule is needed to make it a 'MachTypeNode'!
6649   match(Set dst (EncodeP (Binary crx src1)));
6650   predicate(false);
6651 
6652   format %{ "BEQ     $crx, done\n\t"
6653             "SUB     $dst, $src1, heapbase \t// encode: subtract base if != NULL\n"
6654             "done:" %}
6655   ins_encode %{
6656     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6657     Label done;
6658     __ beq($crx$$CondRegister, done);
6659     __ sub_const_optimized($dst$$Register, $src1$$Register, Universe::narrow_oop_base(), R0);
6660     __ bind(done);
6661   %}
6662   ins_pipe(pipe_class_default);
6663 %}
6664 
6665 // Power 7 can use isel instruction
6666 instruct cond_set_0_oop(iRegNdst dst, flagsRegSrc crx, iRegPsrc src1) %{
6667   // The match rule is needed to make it a 'MachTypeNode'!
6668   match(Set dst (EncodeP (Binary crx src1)));
6669   predicate(false);
6670 
6671   format %{ "CMOVE   $dst, $crx eq, 0, $src1 \t// encode: preserve 0" %}
6672   size(4);
6673   ins_encode %{
6674     // This is a Power7 instruction for which no machine description exists.
6675     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6676     __ isel_0($dst$$Register, $crx$$CondRegister, Assembler::equal, $src1$$Register);
6677   %}
6678   ins_pipe(pipe_class_default);
6679 %}
6680 
6681 // Disjoint narrow oop base.
6682 instruct encodeP_Disjoint(iRegNdst dst, iRegPsrc src) %{
6683   match(Set dst (EncodeP src));
6684   predicate(Universe::narrow_oop_base_disjoint());
6685 
6686   format %{ "EXTRDI  $dst, $src, #32, #3 \t// encode with disjoint base" %}
6687   size(4);
6688   ins_encode %{
6689     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
6690     __ rldicl($dst$$Register, $src$$Register, 64-Universe::narrow_oop_shift(), 32);
6691   %}
6692   ins_pipe(pipe_class_default);
6693 %}
6694 
6695 // shift != 0, base != 0
6696 instruct encodeP_Ex(iRegNdst dst, flagsReg crx, iRegPsrc src) %{
6697   match(Set dst (EncodeP src));
6698   effect(TEMP crx);
6699   predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull &&
6700             Universe::narrow_oop_shift() != 0 &&
6701             Universe::narrow_oop_base_overlaps());
6702 
6703   format %{ "EncodeP $dst, $crx, $src \t// postalloc expanded" %}
6704   postalloc_expand( postalloc_expand_encode_oop(dst, src, crx));
6705 %}
6706 
6707 // shift != 0, base != 0
6708 instruct encodeP_not_null_Ex(iRegNdst dst, iRegPsrc src) %{
6709   match(Set dst (EncodeP src));
6710   predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull &&
6711             Universe::narrow_oop_shift() != 0 &&
6712             Universe::narrow_oop_base_overlaps());
6713 
6714   format %{ "EncodeP $dst, $src\t// $src != Null, postalloc expanded" %}
6715   postalloc_expand( postalloc_expand_encode_oop_not_null(dst, src) );
6716 %}
6717 
6718 // shift != 0, base == 0
6719 // TODO: This is the same as encodeP_shift. Merge!
6720 instruct encodeP_not_null_base_null(iRegNdst dst, iRegPsrc src) %{
6721   match(Set dst (EncodeP src));
6722   predicate(Universe::narrow_oop_shift() != 0 &&
6723             Universe::narrow_oop_base() ==0);
6724 
6725   format %{ "SRDI    $dst, $src, #3 \t// encodeP, $src != NULL" %}
6726   size(4);
6727   ins_encode %{
6728     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
6729     __ srdi($dst$$Register, $src$$Register, Universe::narrow_oop_shift() & 0x3f);
6730   %}
6731   ins_pipe(pipe_class_default);
6732 %}
6733 
6734 // Compressed OOPs with narrow_oop_shift == 0.
6735 // shift == 0, base == 0
6736 instruct encodeP_narrow_oop_shift_0(iRegNdst dst, iRegPsrc src) %{
6737   match(Set dst (EncodeP src));
6738   predicate(Universe::narrow_oop_shift() == 0);
6739 
6740   format %{ "MR      $dst, $src \t// Ptr->Narrow" %}
6741   // variable size, 0 or 4.
6742   ins_encode %{
6743     // TODO: PPC port $archOpcode(ppc64Opcode_or);
6744     __ mr_if_needed($dst$$Register, $src$$Register);
6745   %}
6746   ins_pipe(pipe_class_default);
6747 %}
6748 
6749 // Decode nodes.
6750 
6751 // Shift node for expand.
6752 instruct decodeN_shift(iRegPdst dst, iRegPsrc src) %{
6753   // The match rule is needed to make it a 'MachTypeNode'!
6754   match(Set dst (DecodeN src));
6755   predicate(false);
6756 
6757   format %{ "SLDI    $dst, $src, #3 \t// DecodeN" %}
6758   size(4);
6759   ins_encode %{
6760     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
6761     __ sldi($dst$$Register, $src$$Register, Universe::narrow_oop_shift());
6762   %}
6763   ins_pipe(pipe_class_default);
6764 %}
6765 
6766 // Add node for expand.
6767 instruct decodeN_add(iRegPdst dst, iRegPdst src) %{
6768   // The match rule is needed to make it a 'MachTypeNode'!
6769   match(Set dst (DecodeN src));
6770   predicate(false);
6771 
6772   format %{ "ADD     $dst, $src, heapbase \t// DecodeN, add oop base" %}
6773   ins_encode %{
6774     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6775     __ add_const_optimized($dst$$Register, $src$$Register, Universe::narrow_oop_base(), R0);
6776   %}
6777   ins_pipe(pipe_class_default);
6778 %}
6779 
6780 // conditianal add base for expand
6781 instruct cond_add_base(iRegPdst dst, flagsRegSrc crx, iRegPsrc src) %{
6782   // The match rule is needed to make it a 'MachTypeNode'!
6783   // NOTICE that the rule is nonsense - we just have to make sure that:
6784   //  - _matrule->_rChild->_opType == "DecodeN" (see InstructForm::captures_bottom_type() in formssel.cpp)
6785   //  - we have to match 'crx' to avoid an "illegal USE of non-input: flagsReg crx" error in ADLC.
6786   match(Set dst (DecodeN (Binary crx src)));
6787   predicate(false);
6788 
6789   format %{ "BEQ     $crx, done\n\t"
6790             "ADD     $dst, $src, heapbase \t// DecodeN: add oop base if $src != NULL\n"
6791             "done:" %}
6792   ins_encode %{
6793     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6794     Label done;
6795     __ beq($crx$$CondRegister, done);
6796     __ add_const_optimized($dst$$Register, $src$$Register, Universe::narrow_oop_base(), R0);
6797     __ bind(done);
6798   %}
6799   ins_pipe(pipe_class_default);
6800 %}
6801 
6802 instruct cond_set_0_ptr(iRegPdst dst, flagsRegSrc crx, iRegPsrc src1) %{
6803   // The match rule is needed to make it a 'MachTypeNode'!
6804   // NOTICE that the rule is nonsense - we just have to make sure that:
6805   //  - _matrule->_rChild->_opType == "DecodeN" (see InstructForm::captures_bottom_type() in formssel.cpp)
6806   //  - we have to match 'crx' to avoid an "illegal USE of non-input: flagsReg crx" error in ADLC.
6807   match(Set dst (DecodeN (Binary crx src1)));
6808   predicate(false);
6809 
6810   format %{ "CMOVE   $dst, $crx eq, 0, $src1 \t// decode: preserve 0" %}
6811   size(4);
6812   ins_encode %{
6813     // This is a Power7 instruction for which no machine description exists.
6814     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
6815     __ isel_0($dst$$Register, $crx$$CondRegister, Assembler::equal, $src1$$Register);
6816   %}
6817   ins_pipe(pipe_class_default);
6818 %}
6819 
6820 //  shift != 0, base != 0
6821 instruct decodeN_Ex(iRegPdst dst, iRegNsrc src, flagsReg crx) %{
6822   match(Set dst (DecodeN src));
6823   predicate((n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
6824              n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant) &&
6825             Universe::narrow_oop_shift() != 0 &&
6826             Universe::narrow_oop_base() != 0);
6827   ins_cost(4 * DEFAULT_COST); // Should be more expensive than decodeN_Disjoint_isel_Ex.
6828   effect(TEMP crx);
6829 
6830   format %{ "DecodeN $dst, $src \t// Kills $crx, postalloc expanded" %}
6831   postalloc_expand( postalloc_expand_decode_oop(dst, src, crx) );
6832 %}
6833 
6834 // shift != 0, base == 0
6835 instruct decodeN_nullBase(iRegPdst dst, iRegNsrc src) %{
6836   match(Set dst (DecodeN src));
6837   predicate(Universe::narrow_oop_shift() != 0 &&
6838             Universe::narrow_oop_base() == 0);
6839 
6840   format %{ "SLDI    $dst, $src, #3 \t// DecodeN (zerobased)" %}
6841   size(4);
6842   ins_encode %{
6843     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
6844     __ sldi($dst$$Register, $src$$Register, Universe::narrow_oop_shift());
6845   %}
6846   ins_pipe(pipe_class_default);
6847 %}
6848 
6849 // Optimize DecodeN for disjoint base.
6850 // Shift narrow oop and or it into register that already contains the heap base.
6851 // Base == dst must hold, and is assured by construction in postaloc_expand.
6852 instruct decodeN_mergeDisjoint(iRegPdst dst, iRegNsrc src, iRegLsrc base) %{
6853   match(Set dst (DecodeN src));
6854   effect(TEMP base);
6855   predicate(false);
6856 
6857   format %{ "RLDIMI  $dst, $src, shift, 32-shift \t// DecodeN (disjoint base)" %}
6858   size(4);
6859   ins_encode %{
6860     // TODO: PPC port $archOpcode(ppc64Opcode_rldimi);
6861     __ rldimi($dst$$Register, $src$$Register, Universe::narrow_oop_shift(), 32-Universe::narrow_oop_shift());
6862   %}
6863   ins_pipe(pipe_class_default);
6864 %}
6865 
6866 // Optimize DecodeN for disjoint base.
6867 // This node requires only one cycle on the critical path.
6868 // We must postalloc_expand as we can not express use_def effects where
6869 // the used register is L and the def'ed register P.
6870 instruct decodeN_Disjoint_notNull_Ex(iRegPdst dst, iRegNsrc src) %{
6871   match(Set dst (DecodeN src));
6872   effect(TEMP_DEF dst);
6873   predicate((n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
6874              n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
6875             Universe::narrow_oop_base_disjoint());
6876   ins_cost(DEFAULT_COST);
6877 
6878   format %{ "MOV     $dst, heapbase \t\n"
6879             "RLDIMI  $dst, $src, shift, 32-shift \t// decode with disjoint base" %}
6880   postalloc_expand %{
6881     loadBaseNode *n1 = new loadBaseNode();
6882     n1->add_req(NULL);
6883     n1->_opnds[0] = op_dst;
6884 
6885     decodeN_mergeDisjointNode *n2 = new decodeN_mergeDisjointNode();
6886     n2->add_req(n_region, n_src, n1);
6887     n2->_opnds[0] = op_dst;
6888     n2->_opnds[1] = op_src;
6889     n2->_opnds[2] = op_dst;
6890     n2->_bottom_type = _bottom_type;
6891 
6892     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6893     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6894 
6895     nodes->push(n1);
6896     nodes->push(n2);
6897   %}
6898 %}
6899 
6900 instruct decodeN_Disjoint_isel_Ex(iRegPdst dst, iRegNsrc src, flagsReg crx) %{
6901   match(Set dst (DecodeN src));
6902   effect(TEMP_DEF dst, TEMP crx);
6903   predicate((n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
6904              n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant) &&
6905             Universe::narrow_oop_base_disjoint() && VM_Version::has_isel());
6906   ins_cost(3 * DEFAULT_COST);
6907 
6908   format %{ "DecodeN  $dst, $src \t// decode with disjoint base using isel" %}
6909   postalloc_expand %{
6910     loadBaseNode *n1 = new loadBaseNode();
6911     n1->add_req(NULL);
6912     n1->_opnds[0] = op_dst;
6913 
6914     cmpN_reg_imm0Node *n_compare  = new cmpN_reg_imm0Node();
6915     n_compare->add_req(n_region, n_src);
6916     n_compare->_opnds[0] = op_crx;
6917     n_compare->_opnds[1] = op_src;
6918     n_compare->_opnds[2] = new immN_0Oper(TypeNarrowOop::NULL_PTR);
6919 
6920     decodeN_mergeDisjointNode *n2 = new decodeN_mergeDisjointNode();
6921     n2->add_req(n_region, n_src, n1);
6922     n2->_opnds[0] = op_dst;
6923     n2->_opnds[1] = op_src;
6924     n2->_opnds[2] = op_dst;
6925     n2->_bottom_type = _bottom_type;
6926 
6927     cond_set_0_ptrNode *n_cond_set = new cond_set_0_ptrNode();
6928     n_cond_set->add_req(n_region, n_compare, n2);
6929     n_cond_set->_opnds[0] = op_dst;
6930     n_cond_set->_opnds[1] = op_crx;
6931     n_cond_set->_opnds[2] = op_dst;
6932     n_cond_set->_bottom_type = _bottom_type;
6933 
6934     assert(ra_->is_oop(this) == true, "A decodeN node must produce an oop!");
6935     ra_->set_oop(n_cond_set, true);
6936 
6937     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6938     ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
6939     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6940     ra_->set_pair(n_cond_set->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
6941 
6942     nodes->push(n1);
6943     nodes->push(n_compare);
6944     nodes->push(n2);
6945     nodes->push(n_cond_set);
6946   %}
6947 %}
6948 
6949 // src != 0, shift != 0, base != 0
6950 instruct decodeN_notNull_addBase_Ex(iRegPdst dst, iRegNsrc src) %{
6951   match(Set dst (DecodeN src));
6952   predicate((n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
6953              n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
6954             Universe::narrow_oop_shift() != 0 &&
6955             Universe::narrow_oop_base() != 0);
6956   ins_cost(2 * DEFAULT_COST);
6957 
6958   format %{ "DecodeN $dst, $src \t// $src != NULL, postalloc expanded" %}
6959   postalloc_expand( postalloc_expand_decode_oop_not_null(dst, src));
6960 %}
6961 
6962 // Compressed OOPs with narrow_oop_shift == 0.
6963 instruct decodeN_unscaled(iRegPdst dst, iRegNsrc src) %{
6964   match(Set dst (DecodeN src));
6965   predicate(Universe::narrow_oop_shift() == 0);
6966   ins_cost(DEFAULT_COST);
6967 
6968   format %{ "MR      $dst, $src \t// DecodeN (unscaled)" %}
6969   // variable size, 0 or 4.
6970   ins_encode %{
6971     // TODO: PPC port $archOpcode(ppc64Opcode_or);
6972     __ mr_if_needed($dst$$Register, $src$$Register);
6973   %}
6974   ins_pipe(pipe_class_default);
6975 %}
6976 
6977 // Convert compressed oop into int for vectors alignment masking.
6978 instruct decodeN2I_unscaled(iRegIdst dst, iRegNsrc src) %{
6979   match(Set dst (ConvL2I (CastP2X (DecodeN src))));
6980   predicate(Universe::narrow_oop_shift() == 0);
6981   ins_cost(DEFAULT_COST);
6982 
6983   format %{ "MR      $dst, $src \t// (int)DecodeN (unscaled)" %}
6984   // variable size, 0 or 4.
6985   ins_encode %{
6986     // TODO: PPC port $archOpcode(ppc64Opcode_or);
6987     __ mr_if_needed($dst$$Register, $src$$Register);
6988   %}
6989   ins_pipe(pipe_class_default);
6990 %}
6991 
6992 // Convert klass pointer into compressed form.
6993 
6994 // Nodes for postalloc expand.
6995 
6996 // Shift node for expand.
6997 instruct encodePKlass_shift(iRegNdst dst, iRegNsrc src) %{
6998   // The match rule is needed to make it a 'MachTypeNode'!
6999   match(Set dst (EncodePKlass src));
7000   predicate(false);
7001 
7002   format %{ "SRDI    $dst, $src, 3 \t// encode" %}
7003   size(4);
7004   ins_encode %{
7005     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
7006     __ srdi($dst$$Register, $src$$Register, Universe::narrow_klass_shift());
7007   %}
7008   ins_pipe(pipe_class_default);
7009 %}
7010 
7011 // Add node for expand.
7012 instruct encodePKlass_sub_base(iRegPdst dst, iRegLsrc base, iRegPdst src) %{
7013   // The match rule is needed to make it a 'MachTypeNode'!
7014   match(Set dst (EncodePKlass (Binary base src)));
7015   predicate(false);
7016 
7017   format %{ "SUB     $dst, $base, $src \t// encode" %}
7018   size(4);
7019   ins_encode %{
7020     // TODO: PPC port $archOpcode(ppc64Opcode_subf);
7021     __ subf($dst$$Register, $base$$Register, $src$$Register);
7022   %}
7023   ins_pipe(pipe_class_default);
7024 %}
7025 
7026 // Disjoint narrow oop base.
7027 instruct encodePKlass_Disjoint(iRegNdst dst, iRegPsrc src) %{
7028   match(Set dst (EncodePKlass src));
7029   predicate(false /* TODO: PPC port Universe::narrow_klass_base_disjoint()*/);
7030 
7031   format %{ "EXTRDI  $dst, $src, #32, #3 \t// encode with disjoint base" %}
7032   size(4);
7033   ins_encode %{
7034     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
7035     __ rldicl($dst$$Register, $src$$Register, 64-Universe::narrow_klass_shift(), 32);
7036   %}
7037   ins_pipe(pipe_class_default);
7038 %}
7039 
7040 // shift != 0, base != 0
7041 instruct encodePKlass_not_null_Ex(iRegNdst dst, iRegLsrc base, iRegPsrc src) %{
7042   match(Set dst (EncodePKlass (Binary base src)));
7043   predicate(false);
7044 
7045   format %{ "EncodePKlass $dst, $src\t// $src != Null, postalloc expanded" %}
7046   postalloc_expand %{
7047     encodePKlass_sub_baseNode *n1 = new encodePKlass_sub_baseNode();
7048     n1->add_req(n_region, n_base, n_src);
7049     n1->_opnds[0] = op_dst;
7050     n1->_opnds[1] = op_base;
7051     n1->_opnds[2] = op_src;
7052     n1->_bottom_type = _bottom_type;
7053 
7054     encodePKlass_shiftNode *n2 = new encodePKlass_shiftNode();
7055     n2->add_req(n_region, n1);
7056     n2->_opnds[0] = op_dst;
7057     n2->_opnds[1] = op_dst;
7058     n2->_bottom_type = _bottom_type;
7059     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
7060     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
7061 
7062     nodes->push(n1);
7063     nodes->push(n2);
7064   %}
7065 %}
7066 
7067 // shift != 0, base != 0
7068 instruct encodePKlass_not_null_ExEx(iRegNdst dst, iRegPsrc src) %{
7069   match(Set dst (EncodePKlass src));
7070   //predicate(Universe::narrow_klass_shift() != 0 &&
7071   //          true /* TODO: PPC port Universe::narrow_klass_base_overlaps()*/);
7072 
7073   //format %{ "EncodePKlass $dst, $src\t// $src != Null, postalloc expanded" %}
7074   ins_cost(DEFAULT_COST*2);  // Don't count constant.
7075   expand %{
7076     immL baseImm %{ (jlong)(intptr_t)Universe::narrow_klass_base() %}
7077     iRegLdst base;
7078     loadConL_Ex(base, baseImm);
7079     encodePKlass_not_null_Ex(dst, base, src);
7080   %}
7081 %}
7082 
7083 // Decode nodes.
7084 
7085 // Shift node for expand.
7086 instruct decodeNKlass_shift(iRegPdst dst, iRegPsrc src) %{
7087   // The match rule is needed to make it a 'MachTypeNode'!
7088   match(Set dst (DecodeNKlass src));
7089   predicate(false);
7090 
7091   format %{ "SLDI    $dst, $src, #3 \t// DecodeNKlass" %}
7092   size(4);
7093   ins_encode %{
7094     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
7095     __ sldi($dst$$Register, $src$$Register, Universe::narrow_klass_shift());
7096   %}
7097   ins_pipe(pipe_class_default);
7098 %}
7099 
7100 // Add node for expand.
7101 
7102 instruct decodeNKlass_add_base(iRegPdst dst, iRegLsrc base, iRegPdst src) %{
7103   // The match rule is needed to make it a 'MachTypeNode'!
7104   match(Set dst (DecodeNKlass (Binary base src)));
7105   predicate(false);
7106 
7107   format %{ "ADD     $dst, $base, $src \t// DecodeNKlass, add klass base" %}
7108   size(4);
7109   ins_encode %{
7110     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7111     __ add($dst$$Register, $base$$Register, $src$$Register);
7112   %}
7113   ins_pipe(pipe_class_default);
7114 %}
7115 
7116 // src != 0, shift != 0, base != 0
7117 instruct decodeNKlass_notNull_addBase_Ex(iRegPdst dst, iRegLsrc base, iRegNsrc src) %{
7118   match(Set dst (DecodeNKlass (Binary base src)));
7119   //effect(kill src); // We need a register for the immediate result after shifting.
7120   predicate(false);
7121 
7122   format %{ "DecodeNKlass $dst =  $base + ($src << 3) \t// $src != NULL, postalloc expanded" %}
7123   postalloc_expand %{
7124     decodeNKlass_add_baseNode *n1 = new decodeNKlass_add_baseNode();
7125     n1->add_req(n_region, n_base, n_src);
7126     n1->_opnds[0] = op_dst;
7127     n1->_opnds[1] = op_base;
7128     n1->_opnds[2] = op_src;
7129     n1->_bottom_type = _bottom_type;
7130 
7131     decodeNKlass_shiftNode *n2 = new decodeNKlass_shiftNode();
7132     n2->add_req(n_region, n1);
7133     n2->_opnds[0] = op_dst;
7134     n2->_opnds[1] = op_dst;
7135     n2->_bottom_type = _bottom_type;
7136 
7137     ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
7138     ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
7139 
7140     nodes->push(n1);
7141     nodes->push(n2);
7142   %}
7143 %}
7144 
7145 // src != 0, shift != 0, base != 0
7146 instruct decodeNKlass_notNull_addBase_ExEx(iRegPdst dst, iRegNsrc src) %{
7147   match(Set dst (DecodeNKlass src));
7148   // predicate(Universe::narrow_klass_shift() != 0 &&
7149   //           Universe::narrow_klass_base() != 0);
7150 
7151   //format %{ "DecodeNKlass $dst, $src \t// $src != NULL, expanded" %}
7152 
7153   ins_cost(DEFAULT_COST*2);  // Don't count constant.
7154   expand %{
7155     // We add first, then we shift. Like this, we can get along with one register less.
7156     // But we have to load the base pre-shifted.
7157     immL baseImm %{ (jlong)((intptr_t)Universe::narrow_klass_base() >> Universe::narrow_klass_shift()) %}
7158     iRegLdst base;
7159     loadConL_Ex(base, baseImm);
7160     decodeNKlass_notNull_addBase_Ex(dst, base, src);
7161   %}
7162 %}
7163 
7164 //----------MemBar Instructions-----------------------------------------------
7165 // Memory barrier flavors
7166 
7167 instruct membar_acquire() %{
7168   match(LoadFence);
7169   ins_cost(4*MEMORY_REF_COST);
7170 
7171   format %{ "MEMBAR-acquire" %}
7172   size(4);
7173   ins_encode %{
7174     // TODO: PPC port $archOpcode(ppc64Opcode_lwsync);
7175     __ acquire();
7176   %}
7177   ins_pipe(pipe_class_default);
7178 %}
7179 
7180 instruct unnecessary_membar_acquire() %{
7181   match(MemBarAcquire);
7182   ins_cost(0);
7183 
7184   format %{ " -- \t// redundant MEMBAR-acquire - empty" %}
7185   size(0);
7186   ins_encode( /*empty*/ );
7187   ins_pipe(pipe_class_default);
7188 %}
7189 
7190 instruct membar_acquire_lock() %{
7191   match(MemBarAcquireLock);
7192   ins_cost(0);
7193 
7194   format %{ " -- \t// redundant MEMBAR-acquire - empty (acquire as part of CAS in prior FastLock)" %}
7195   size(0);
7196   ins_encode( /*empty*/ );
7197   ins_pipe(pipe_class_default);
7198 %}
7199 
7200 instruct membar_release() %{
7201   match(MemBarRelease);
7202   match(StoreFence);
7203   ins_cost(4*MEMORY_REF_COST);
7204 
7205   format %{ "MEMBAR-release" %}
7206   size(4);
7207   ins_encode %{
7208     // TODO: PPC port $archOpcode(ppc64Opcode_lwsync);
7209     __ release();
7210   %}
7211   ins_pipe(pipe_class_default);
7212 %}
7213 
7214 instruct membar_storestore() %{
7215   match(MemBarStoreStore);
7216   ins_cost(4*MEMORY_REF_COST);
7217 
7218   format %{ "MEMBAR-store-store" %}
7219   size(4);
7220   ins_encode %{
7221     // TODO: PPC port $archOpcode(ppc64Opcode_lwsync);
7222     __ membar(Assembler::StoreStore);
7223   %}
7224   ins_pipe(pipe_class_default);
7225 %}
7226 
7227 instruct membar_release_lock() %{
7228   match(MemBarReleaseLock);
7229   ins_cost(0);
7230 
7231   format %{ " -- \t// redundant MEMBAR-release - empty (release in FastUnlock)" %}
7232   size(0);
7233   ins_encode( /*empty*/ );
7234   ins_pipe(pipe_class_default);
7235 %}
7236 
7237 instruct membar_volatile() %{
7238   match(MemBarVolatile);
7239   ins_cost(4*MEMORY_REF_COST);
7240 
7241   format %{ "MEMBAR-volatile" %}
7242   size(4);
7243   ins_encode %{
7244     // TODO: PPC port $archOpcode(ppc64Opcode_sync);
7245     __ fence();
7246   %}
7247   ins_pipe(pipe_class_default);
7248 %}
7249 
7250 // This optimization is wrong on PPC. The following pattern is not supported:
7251 //  MemBarVolatile
7252 //   ^        ^
7253 //   |        |
7254 //  CtrlProj MemProj
7255 //   ^        ^
7256 //   |        |
7257 //   |       Load
7258 //   |
7259 //  MemBarVolatile
7260 //
7261 //  The first MemBarVolatile could get optimized out! According to
7262 //  Vladimir, this pattern can not occur on Oracle platforms.
7263 //  However, it does occur on PPC64 (because of membars in
7264 //  inline_unsafe_load_store).
7265 //
7266 // Add this node again if we found a good solution for inline_unsafe_load_store().
7267 // Don't forget to look at the implementation of post_store_load_barrier again,
7268 // we did other fixes in that method.
7269 //instruct unnecessary_membar_volatile() %{
7270 //  match(MemBarVolatile);
7271 //  predicate(Matcher::post_store_load_barrier(n));
7272 //  ins_cost(0);
7273 //
7274 //  format %{ " -- \t// redundant MEMBAR-volatile - empty" %}
7275 //  size(0);
7276 //  ins_encode( /*empty*/ );
7277 //  ins_pipe(pipe_class_default);
7278 //%}
7279 
7280 instruct membar_CPUOrder() %{
7281   match(MemBarCPUOrder);
7282   ins_cost(0);
7283 
7284   format %{ " -- \t// MEMBAR-CPUOrder - empty: PPC64 processors are self-consistent." %}
7285   size(0);
7286   ins_encode( /*empty*/ );
7287   ins_pipe(pipe_class_default);
7288 %}
7289 
7290 //----------Conditional Move---------------------------------------------------
7291 
7292 // Cmove using isel.
7293 instruct cmovI_reg_isel(cmpOp cmp, flagsRegSrc crx, iRegIdst dst, iRegIsrc src) %{
7294   match(Set dst (CMoveI (Binary cmp crx) (Binary dst src)));
7295   predicate(VM_Version::has_isel());
7296   ins_cost(DEFAULT_COST);
7297 
7298   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7299   size(4);
7300   ins_encode %{
7301     // This is a Power7 instruction for which no machine description
7302     // exists. Anyways, the scheduler should be off on Power7.
7303     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7304     int cc        = $cmp$$cmpcode;
7305     __ isel($dst$$Register, $crx$$CondRegister,
7306             (Assembler::Condition)(cc & 3), /*invert*/((~cc) & 8), $src$$Register);
7307   %}
7308   ins_pipe(pipe_class_default);
7309 %}
7310 
7311 instruct cmovI_reg(cmpOp cmp, flagsRegSrc crx, iRegIdst dst, iRegIsrc src) %{
7312   match(Set dst (CMoveI (Binary cmp crx) (Binary dst src)));
7313   predicate(!VM_Version::has_isel());
7314   ins_cost(DEFAULT_COST+BRANCH_COST);
7315 
7316   ins_variable_size_depending_on_alignment(true);
7317 
7318   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7319   // Worst case is branch + move + stop, no stop without scheduler
7320   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7321   ins_encode( enc_cmove_reg(dst, crx, src, cmp) );
7322   ins_pipe(pipe_class_default);
7323 %}
7324 
7325 instruct cmovI_imm(cmpOp cmp, flagsRegSrc crx, iRegIdst dst, immI16 src) %{
7326   match(Set dst (CMoveI (Binary cmp crx) (Binary dst src)));
7327   ins_cost(DEFAULT_COST+BRANCH_COST);
7328 
7329   ins_variable_size_depending_on_alignment(true);
7330 
7331   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7332   // Worst case is branch + move + stop, no stop without scheduler
7333   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7334   ins_encode( enc_cmove_imm(dst, crx, src, cmp) );
7335   ins_pipe(pipe_class_default);
7336 %}
7337 
7338 // Cmove using isel.
7339 instruct cmovL_reg_isel(cmpOp cmp, flagsRegSrc crx, iRegLdst dst, iRegLsrc src) %{
7340   match(Set dst (CMoveL (Binary cmp crx) (Binary dst src)));
7341   predicate(VM_Version::has_isel());
7342   ins_cost(DEFAULT_COST);
7343 
7344   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7345   size(4);
7346   ins_encode %{
7347     // This is a Power7 instruction for which no machine description
7348     // exists. Anyways, the scheduler should be off on Power7.
7349     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7350     int cc        = $cmp$$cmpcode;
7351     __ isel($dst$$Register, $crx$$CondRegister,
7352             (Assembler::Condition)(cc & 3), /*invert*/((~cc) & 8), $src$$Register);
7353   %}
7354   ins_pipe(pipe_class_default);
7355 %}
7356 
7357 instruct cmovL_reg(cmpOp cmp, flagsRegSrc crx, iRegLdst dst, iRegLsrc src) %{
7358   match(Set dst (CMoveL (Binary cmp crx) (Binary dst src)));
7359   predicate(!VM_Version::has_isel());
7360   ins_cost(DEFAULT_COST+BRANCH_COST);
7361 
7362   ins_variable_size_depending_on_alignment(true);
7363 
7364   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7365   // Worst case is branch + move + stop, no stop without scheduler.
7366   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7367   ins_encode( enc_cmove_reg(dst, crx, src, cmp) );
7368   ins_pipe(pipe_class_default);
7369 %}
7370 
7371 instruct cmovL_imm(cmpOp cmp, flagsRegSrc crx, iRegLdst dst, immL16 src) %{
7372   match(Set dst (CMoveL (Binary cmp crx) (Binary dst src)));
7373   ins_cost(DEFAULT_COST+BRANCH_COST);
7374 
7375   ins_variable_size_depending_on_alignment(true);
7376 
7377   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7378   // Worst case is branch + move + stop, no stop without scheduler.
7379   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7380   ins_encode( enc_cmove_imm(dst, crx, src, cmp) );
7381   ins_pipe(pipe_class_default);
7382 %}
7383 
7384 // Cmove using isel.
7385 instruct cmovN_reg_isel(cmpOp cmp, flagsRegSrc crx, iRegNdst dst, iRegNsrc src) %{
7386   match(Set dst (CMoveN (Binary cmp crx) (Binary dst src)));
7387   predicate(VM_Version::has_isel());
7388   ins_cost(DEFAULT_COST);
7389 
7390   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7391   size(4);
7392   ins_encode %{
7393     // This is a Power7 instruction for which no machine description
7394     // exists. Anyways, the scheduler should be off on Power7.
7395     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7396     int cc        = $cmp$$cmpcode;
7397     __ isel($dst$$Register, $crx$$CondRegister,
7398             (Assembler::Condition)(cc & 3), /*invert*/((~cc) & 8), $src$$Register);
7399   %}
7400   ins_pipe(pipe_class_default);
7401 %}
7402 
7403 // Conditional move for RegN. Only cmov(reg, reg).
7404 instruct cmovN_reg(cmpOp cmp, flagsRegSrc crx, iRegNdst dst, iRegNsrc src) %{
7405   match(Set dst (CMoveN (Binary cmp crx) (Binary dst src)));
7406   predicate(!VM_Version::has_isel());
7407   ins_cost(DEFAULT_COST+BRANCH_COST);
7408 
7409   ins_variable_size_depending_on_alignment(true);
7410 
7411   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7412   // Worst case is branch + move + stop, no stop without scheduler.
7413   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7414   ins_encode( enc_cmove_reg(dst, crx, src, cmp) );
7415   ins_pipe(pipe_class_default);
7416 %}
7417 
7418 instruct cmovN_imm(cmpOp cmp, flagsRegSrc crx, iRegNdst dst, immN_0 src) %{
7419   match(Set dst (CMoveN (Binary cmp crx) (Binary dst src)));
7420   ins_cost(DEFAULT_COST+BRANCH_COST);
7421 
7422   ins_variable_size_depending_on_alignment(true);
7423 
7424   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7425   // Worst case is branch + move + stop, no stop without scheduler.
7426   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7427   ins_encode( enc_cmove_imm(dst, crx, src, cmp) );
7428   ins_pipe(pipe_class_default);
7429 %}
7430 
7431 // Cmove using isel.
7432 instruct cmovP_reg_isel(cmpOp cmp, flagsRegSrc crx, iRegPdst dst, iRegPsrc src) %{
7433   match(Set dst (CMoveP (Binary cmp crx) (Binary dst src)));
7434   predicate(VM_Version::has_isel());
7435   ins_cost(DEFAULT_COST);
7436 
7437   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7438   size(4);
7439   ins_encode %{
7440     // This is a Power7 instruction for which no machine description
7441     // exists. Anyways, the scheduler should be off on Power7.
7442     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7443     int cc        = $cmp$$cmpcode;
7444     __ isel($dst$$Register, $crx$$CondRegister,
7445             (Assembler::Condition)(cc & 3), /*invert*/((~cc) & 8), $src$$Register);
7446   %}
7447   ins_pipe(pipe_class_default);
7448 %}
7449 
7450 instruct cmovP_reg(cmpOp cmp, flagsRegSrc crx, iRegPdst dst, iRegP_N2P src) %{
7451   match(Set dst (CMoveP (Binary cmp crx) (Binary dst src)));
7452   predicate(!VM_Version::has_isel());
7453   ins_cost(DEFAULT_COST+BRANCH_COST);
7454 
7455   ins_variable_size_depending_on_alignment(true);
7456 
7457   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7458   // Worst case is branch + move + stop, no stop without scheduler.
7459   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7460   ins_encode( enc_cmove_reg(dst, crx, src, cmp) );
7461   ins_pipe(pipe_class_default);
7462 %}
7463 
7464 instruct cmovP_imm(cmpOp cmp, flagsRegSrc crx, iRegPdst dst, immP_0 src) %{
7465   match(Set dst (CMoveP (Binary cmp crx) (Binary dst src)));
7466   ins_cost(DEFAULT_COST+BRANCH_COST);
7467 
7468   ins_variable_size_depending_on_alignment(true);
7469 
7470   format %{ "CMOVE   $cmp, $crx, $dst, $src\n\t" %}
7471   // Worst case is branch + move + stop, no stop without scheduler.
7472   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
7473   ins_encode( enc_cmove_imm(dst, crx, src, cmp) );
7474   ins_pipe(pipe_class_default);
7475 %}
7476 
7477 instruct cmovF_reg(cmpOp cmp, flagsRegSrc crx, regF dst, regF src) %{
7478   match(Set dst (CMoveF (Binary cmp crx) (Binary dst src)));
7479   ins_cost(DEFAULT_COST+BRANCH_COST);
7480 
7481   ins_variable_size_depending_on_alignment(true);
7482 
7483   format %{ "CMOVEF  $cmp, $crx, $dst, $src\n\t" %}
7484   // Worst case is branch + move + stop, no stop without scheduler.
7485   size(false /* TODO: PPC PORT (InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 12 : 8);
7486   ins_encode %{
7487     // TODO: PPC port $archOpcode(ppc64Opcode_cmovef);
7488     Label done;
7489     assert((Assembler::bcondCRbiIs1 & ~Assembler::bcondCRbiIs0) == 8, "check encoding");
7490     // Branch if not (cmp crx).
7491     __ bc(cc_to_inverse_boint($cmp$$cmpcode), cc_to_biint($cmp$$cmpcode, $crx$$reg), done);
7492     __ fmr($dst$$FloatRegister, $src$$FloatRegister);
7493     // TODO PPC port __ endgroup_if_needed(_size == 12);
7494     __ bind(done);
7495   %}
7496   ins_pipe(pipe_class_default);
7497 %}
7498 
7499 instruct cmovD_reg(cmpOp cmp, flagsRegSrc crx, regD dst, regD src) %{
7500   match(Set dst (CMoveD (Binary cmp crx) (Binary dst src)));
7501   ins_cost(DEFAULT_COST+BRANCH_COST);
7502 
7503   ins_variable_size_depending_on_alignment(true);
7504 
7505   format %{ "CMOVEF  $cmp, $crx, $dst, $src\n\t" %}
7506   // Worst case is branch + move + stop, no stop without scheduler.
7507   size(false /* TODO: PPC PORT (InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 12 : 8);
7508   ins_encode %{
7509     // TODO: PPC port $archOpcode(ppc64Opcode_cmovef);
7510     Label done;
7511     assert((Assembler::bcondCRbiIs1 & ~Assembler::bcondCRbiIs0) == 8, "check encoding");
7512     // Branch if not (cmp crx).
7513     __ bc(cc_to_inverse_boint($cmp$$cmpcode), cc_to_biint($cmp$$cmpcode, $crx$$reg), done);
7514     __ fmr($dst$$FloatRegister, $src$$FloatRegister);
7515     // TODO PPC port __ endgroup_if_needed(_size == 12);
7516     __ bind(done);
7517   %}
7518   ins_pipe(pipe_class_default);
7519 %}
7520 
7521 //----------Conditional_store--------------------------------------------------
7522 // Conditional-store of the updated heap-top.
7523 // Used during allocation of the shared heap.
7524 // Sets flags (EQ) on success. Implemented with a CASA on Sparc.
7525 
7526 // As compareAndSwapL, but return flag register instead of boolean value in
7527 // int register.
7528 // Used by sun/misc/AtomicLongCSImpl.java.
7529 // Mem_ptr must be a memory operand, else this node does not get
7530 // Flag_needs_anti_dependence_check set by adlc. If this is not set this node
7531 // can be rematerialized which leads to errors.
7532 instruct storeLConditional_regP_regL_regL(flagsReg crx, indirect mem_ptr, iRegLsrc oldVal, iRegLsrc newVal, flagsRegCR0 cr0) %{
7533   match(Set crx (StoreLConditional mem_ptr (Binary oldVal newVal)));
7534   effect(TEMP cr0);
7535   format %{ "CMPXCHGD if ($crx = ($oldVal == *$mem_ptr)) *mem_ptr = $newVal; as bool" %}
7536   ins_encode %{
7537     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7538     __ cmpxchgd($crx$$CondRegister, R0, $oldVal$$Register, $newVal$$Register, $mem_ptr$$Register,
7539                 MacroAssembler::MemBarAcq, MacroAssembler::cmpxchgx_hint_atomic_update(),
7540                 noreg, NULL, true);
7541   %}
7542   ins_pipe(pipe_class_default);
7543 %}
7544 
7545 // As compareAndSwapP, but return flag register instead of boolean value in
7546 // int register.
7547 // This instruction is matched if UseTLAB is off.
7548 // Mem_ptr must be a memory operand, else this node does not get
7549 // Flag_needs_anti_dependence_check set by adlc. If this is not set this node
7550 // can be rematerialized which leads to errors.
7551 instruct storePConditional_regP_regP_regP(flagsRegCR0 cr0, indirect mem_ptr, iRegPsrc oldVal, iRegPsrc newVal) %{
7552   match(Set cr0 (StorePConditional mem_ptr (Binary oldVal newVal)));
7553   ins_cost(2*MEMORY_REF_COST);
7554 
7555   format %{ "STDCX_  if ($cr0 = ($oldVal == *$mem_ptr)) *mem_ptr = $newVal; as bool" %}
7556   ins_encode %{
7557     // TODO: PPC port $archOpcode(ppc64Opcode_stdcx_);
7558     __ stdcx_($newVal$$Register, $mem_ptr$$Register);
7559   %}
7560   ins_pipe(pipe_class_memory);
7561 %}
7562 
7563 // Implement LoadPLocked. Must be ordered against changes of the memory location
7564 // by storePConditional.
7565 // Don't know whether this is ever used.
7566 instruct loadPLocked(iRegPdst dst, memory mem) %{
7567   match(Set dst (LoadPLocked mem));
7568   ins_cost(2*MEMORY_REF_COST);
7569 
7570   format %{ "LDARX   $dst, $mem \t// loadPLocked\n\t" %}
7571   size(4);
7572   ins_encode %{
7573     // TODO: PPC port $archOpcode(ppc64Opcode_ldarx);
7574     __ ldarx($dst$$Register, $mem$$Register, MacroAssembler::cmpxchgx_hint_atomic_update());
7575   %}
7576   ins_pipe(pipe_class_memory);
7577 %}
7578 
7579 //----------Compare-And-Swap---------------------------------------------------
7580 
7581 // CompareAndSwap{P,I,L} have more than one output, therefore "CmpI
7582 // (CompareAndSwap ...)" or "If (CmpI (CompareAndSwap ..))"  cannot be
7583 // matched.
7584 
7585 instruct compareAndSwapI_regP_regI_regI(iRegIdst res, iRegPdst mem_ptr, iRegIsrc src1, iRegIsrc src2, flagsRegCR0 cr0) %{
7586   match(Set res (CompareAndSwapI mem_ptr (Binary src1 src2)));
7587   effect(TEMP cr0);
7588   format %{ "CMPXCHGW $res, $mem_ptr, $src1, $src2; as bool" %}
7589   // Variable size: instruction count smaller if regs are disjoint.
7590   ins_encode %{
7591     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7592     // CmpxchgX sets CCR0 to cmpX(src1, src2) and Rres to 'true'/'false'.
7593     __ cmpxchgw(CCR0, R0, $src1$$Register, $src2$$Register, $mem_ptr$$Register,
7594                 MacroAssembler::MemBarFenceAfter, MacroAssembler::cmpxchgx_hint_atomic_update(),
7595                 $res$$Register, true);
7596   %}
7597   ins_pipe(pipe_class_default);
7598 %}
7599 
7600 instruct compareAndSwapN_regP_regN_regN(iRegIdst res, iRegPdst mem_ptr, iRegNsrc src1, iRegNsrc src2, flagsRegCR0 cr0) %{
7601   match(Set res (CompareAndSwapN mem_ptr (Binary src1 src2)));
7602   effect(TEMP cr0);
7603   format %{ "CMPXCHGW $res, $mem_ptr, $src1, $src2; as bool" %}
7604   // Variable size: instruction count smaller if regs are disjoint.
7605   ins_encode %{
7606     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7607     // CmpxchgX sets CCR0 to cmpX(src1, src2) and Rres to 'true'/'false'.
7608     __ cmpxchgw(CCR0, R0, $src1$$Register, $src2$$Register, $mem_ptr$$Register,
7609                 MacroAssembler::MemBarFenceAfter, MacroAssembler::cmpxchgx_hint_atomic_update(),
7610                 $res$$Register, true);
7611   %}
7612   ins_pipe(pipe_class_default);
7613 %}
7614 
7615 instruct compareAndSwapL_regP_regL_regL(iRegIdst res, iRegPdst mem_ptr, iRegLsrc src1, iRegLsrc src2, flagsRegCR0 cr0) %{
7616   match(Set res (CompareAndSwapL mem_ptr (Binary src1 src2)));
7617   effect(TEMP cr0);
7618   format %{ "CMPXCHGD $res, $mem_ptr, $src1, $src2; as bool" %}
7619   // Variable size: instruction count smaller if regs are disjoint.
7620   ins_encode %{
7621     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7622     // CmpxchgX sets CCR0 to cmpX(src1, src2) and Rres to 'true'/'false'.
7623     __ cmpxchgd(CCR0, R0, $src1$$Register, $src2$$Register, $mem_ptr$$Register,
7624                 MacroAssembler::MemBarFenceAfter, MacroAssembler::cmpxchgx_hint_atomic_update(),
7625                 $res$$Register, NULL, true);
7626   %}
7627   ins_pipe(pipe_class_default);
7628 %}
7629 
7630 instruct compareAndSwapP_regP_regP_regP(iRegIdst res, iRegPdst mem_ptr, iRegPsrc src1, iRegPsrc src2, flagsRegCR0 cr0) %{
7631   match(Set res (CompareAndSwapP mem_ptr (Binary src1 src2)));
7632   effect(TEMP cr0);
7633   format %{ "CMPXCHGD $res, $mem_ptr, $src1, $src2; as bool; ptr" %}
7634   // Variable size: instruction count smaller if regs are disjoint.
7635   ins_encode %{
7636     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
7637     // CmpxchgX sets CCR0 to cmpX(src1, src2) and Rres to 'true'/'false'.
7638     __ cmpxchgd(CCR0, R0, $src1$$Register, $src2$$Register, $mem_ptr$$Register,
7639                 MacroAssembler::MemBarFenceAfter, MacroAssembler::cmpxchgx_hint_atomic_update(),
7640                 $res$$Register, NULL, true);
7641   %}
7642   ins_pipe(pipe_class_default);
7643 %}
7644 
7645 instruct getAndAddI(iRegIdst res, iRegPdst mem_ptr, iRegIsrc src, flagsRegCR0 cr0) %{
7646   match(Set res (GetAndAddI mem_ptr src));
7647   effect(TEMP cr0);
7648   format %{ "GetAndAddI $res, $mem_ptr, $src" %}
7649   // Variable size: instruction count smaller if regs are disjoint.
7650   ins_encode( enc_GetAndAddI(res, mem_ptr, src) );
7651   ins_pipe(pipe_class_default);
7652 %}
7653 
7654 instruct getAndAddL(iRegLdst res, iRegPdst mem_ptr, iRegLsrc src, flagsRegCR0 cr0) %{
7655   match(Set res (GetAndAddL mem_ptr src));
7656   effect(TEMP cr0);
7657   format %{ "GetAndAddL $res, $mem_ptr, $src" %}
7658   // Variable size: instruction count smaller if regs are disjoint.
7659   ins_encode( enc_GetAndAddL(res, mem_ptr, src) );
7660   ins_pipe(pipe_class_default);
7661 %}
7662 
7663 instruct getAndSetI(iRegIdst res, iRegPdst mem_ptr, iRegIsrc src, flagsRegCR0 cr0) %{
7664   match(Set res (GetAndSetI mem_ptr src));
7665   effect(TEMP cr0);
7666   format %{ "GetAndSetI $res, $mem_ptr, $src" %}
7667   // Variable size: instruction count smaller if regs are disjoint.
7668   ins_encode( enc_GetAndSetI(res, mem_ptr, src) );
7669   ins_pipe(pipe_class_default);
7670 %}
7671 
7672 instruct getAndSetL(iRegLdst res, iRegPdst mem_ptr, iRegLsrc src, flagsRegCR0 cr0) %{
7673   match(Set res (GetAndSetL mem_ptr src));
7674   effect(TEMP cr0);
7675   format %{ "GetAndSetL $res, $mem_ptr, $src" %}
7676   // Variable size: instruction count smaller if regs are disjoint.
7677   ins_encode( enc_GetAndSetL(res, mem_ptr, src) );
7678   ins_pipe(pipe_class_default);
7679 %}
7680 
7681 instruct getAndSetP(iRegPdst res, iRegPdst mem_ptr, iRegPsrc src, flagsRegCR0 cr0) %{
7682   match(Set res (GetAndSetP mem_ptr src));
7683   effect(TEMP cr0);
7684   format %{ "GetAndSetP $res, $mem_ptr, $src" %}
7685   // Variable size: instruction count smaller if regs are disjoint.
7686   ins_encode( enc_GetAndSetL(res, mem_ptr, src) );
7687   ins_pipe(pipe_class_default);
7688 %}
7689 
7690 instruct getAndSetN(iRegNdst res, iRegPdst mem_ptr, iRegNsrc src, flagsRegCR0 cr0) %{
7691   match(Set res (GetAndSetN mem_ptr src));
7692   effect(TEMP cr0);
7693   format %{ "GetAndSetN $res, $mem_ptr, $src" %}
7694   // Variable size: instruction count smaller if regs are disjoint.
7695   ins_encode( enc_GetAndSetI(res, mem_ptr, src) );
7696   ins_pipe(pipe_class_default);
7697 %}
7698 
7699 //----------Arithmetic Instructions--------------------------------------------
7700 // Addition Instructions
7701 
7702 // Register Addition
7703 instruct addI_reg_reg(iRegIdst dst, iRegIsrc_iRegL2Isrc src1, iRegIsrc_iRegL2Isrc src2) %{
7704   match(Set dst (AddI src1 src2));
7705   format %{ "ADD     $dst, $src1, $src2" %}
7706   size(4);
7707   ins_encode %{
7708     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7709     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7710   %}
7711   ins_pipe(pipe_class_default);
7712 %}
7713 
7714 // Expand does not work with above instruct. (??)
7715 instruct addI_reg_reg_2(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
7716   // no match-rule
7717   effect(DEF dst, USE src1, USE src2);
7718   format %{ "ADD     $dst, $src1, $src2" %}
7719   size(4);
7720   ins_encode %{
7721     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7722     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7723   %}
7724   ins_pipe(pipe_class_default);
7725 %}
7726 
7727 instruct tree_addI_addI_addI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2, iRegIsrc src3, iRegIsrc src4) %{
7728   match(Set dst (AddI (AddI (AddI src1 src2) src3) src4));
7729   ins_cost(DEFAULT_COST*3);
7730 
7731   expand %{
7732     // FIXME: we should do this in the ideal world.
7733     iRegIdst tmp1;
7734     iRegIdst tmp2;
7735     addI_reg_reg(tmp1, src1, src2);
7736     addI_reg_reg_2(tmp2, src3, src4); // Adlc complains about addI_reg_reg.
7737     addI_reg_reg(dst, tmp1, tmp2);
7738   %}
7739 %}
7740 
7741 // Immediate Addition
7742 instruct addI_reg_imm16(iRegIdst dst, iRegIsrc src1, immI16 src2) %{
7743   match(Set dst (AddI src1 src2));
7744   format %{ "ADDI    $dst, $src1, $src2" %}
7745   size(4);
7746   ins_encode %{
7747     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
7748     __ addi($dst$$Register, $src1$$Register, $src2$$constant);
7749   %}
7750   ins_pipe(pipe_class_default);
7751 %}
7752 
7753 // Immediate Addition with 16-bit shifted operand
7754 instruct addI_reg_immhi16(iRegIdst dst, iRegIsrc src1, immIhi16 src2) %{
7755   match(Set dst (AddI src1 src2));
7756   format %{ "ADDIS   $dst, $src1, $src2" %}
7757   size(4);
7758   ins_encode %{
7759     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
7760     __ addis($dst$$Register, $src1$$Register, ($src2$$constant)>>16);
7761   %}
7762   ins_pipe(pipe_class_default);
7763 %}
7764 
7765 // Long Addition
7766 instruct addL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
7767   match(Set dst (AddL src1 src2));
7768   format %{ "ADD     $dst, $src1, $src2 \t// long" %}
7769   size(4);
7770   ins_encode %{
7771     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7772     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7773   %}
7774   ins_pipe(pipe_class_default);
7775 %}
7776 
7777 // Expand does not work with above instruct. (??)
7778 instruct addL_reg_reg_2(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
7779   // no match-rule
7780   effect(DEF dst, USE src1, USE src2);
7781   format %{ "ADD     $dst, $src1, $src2 \t// long" %}
7782   size(4);
7783   ins_encode %{
7784     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7785     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7786   %}
7787   ins_pipe(pipe_class_default);
7788 %}
7789 
7790 instruct tree_addL_addL_addL_reg_reg_Ex(iRegLdst dst, iRegLsrc src1, iRegLsrc src2, iRegLsrc src3, iRegLsrc src4) %{
7791   match(Set dst (AddL (AddL (AddL src1 src2) src3) src4));
7792   ins_cost(DEFAULT_COST*3);
7793 
7794   expand %{
7795     // FIXME: we should do this in the ideal world.
7796     iRegLdst tmp1;
7797     iRegLdst tmp2;
7798     addL_reg_reg(tmp1, src1, src2);
7799     addL_reg_reg_2(tmp2, src3, src4); // Adlc complains about orI_reg_reg.
7800     addL_reg_reg(dst, tmp1, tmp2);
7801   %}
7802 %}
7803 
7804 // AddL + ConvL2I.
7805 instruct addI_regL_regL(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{
7806   match(Set dst (ConvL2I (AddL src1 src2)));
7807 
7808   format %{ "ADD     $dst, $src1, $src2 \t// long + l2i" %}
7809   size(4);
7810   ins_encode %{
7811     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7812     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7813   %}
7814   ins_pipe(pipe_class_default);
7815 %}
7816 
7817 // No constant pool entries required.
7818 instruct addL_reg_imm16(iRegLdst dst, iRegLsrc src1, immL16 src2) %{
7819   match(Set dst (AddL src1 src2));
7820 
7821   format %{ "ADDI    $dst, $src1, $src2" %}
7822   size(4);
7823   ins_encode %{
7824     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
7825     __ addi($dst$$Register, $src1$$Register, $src2$$constant);
7826   %}
7827   ins_pipe(pipe_class_default);
7828 %}
7829 
7830 // Long Immediate Addition with 16-bit shifted operand.
7831 // No constant pool entries required.
7832 instruct addL_reg_immhi16(iRegLdst dst, iRegLsrc src1, immL32hi16 src2) %{
7833   match(Set dst (AddL src1 src2));
7834 
7835   format %{ "ADDIS   $dst, $src1, $src2" %}
7836   size(4);
7837   ins_encode %{
7838     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
7839     __ addis($dst$$Register, $src1$$Register, ($src2$$constant)>>16);
7840   %}
7841   ins_pipe(pipe_class_default);
7842 %}
7843 
7844 // Pointer Register Addition
7845 instruct addP_reg_reg(iRegPdst dst, iRegP_N2P src1, iRegLsrc src2) %{
7846   match(Set dst (AddP src1 src2));
7847   format %{ "ADD     $dst, $src1, $src2" %}
7848   size(4);
7849   ins_encode %{
7850     // TODO: PPC port $archOpcode(ppc64Opcode_add);
7851     __ add($dst$$Register, $src1$$Register, $src2$$Register);
7852   %}
7853   ins_pipe(pipe_class_default);
7854 %}
7855 
7856 // Pointer Immediate Addition
7857 // No constant pool entries required.
7858 instruct addP_reg_imm16(iRegPdst dst, iRegP_N2P src1, immL16 src2) %{
7859   match(Set dst (AddP src1 src2));
7860 
7861   format %{ "ADDI    $dst, $src1, $src2" %}
7862   size(4);
7863   ins_encode %{
7864     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
7865     __ addi($dst$$Register, $src1$$Register, $src2$$constant);
7866   %}
7867   ins_pipe(pipe_class_default);
7868 %}
7869 
7870 // Pointer Immediate Addition with 16-bit shifted operand.
7871 // No constant pool entries required.
7872 instruct addP_reg_immhi16(iRegPdst dst, iRegP_N2P src1, immL32hi16 src2) %{
7873   match(Set dst (AddP src1 src2));
7874 
7875   format %{ "ADDIS   $dst, $src1, $src2" %}
7876   size(4);
7877   ins_encode %{
7878     // TODO: PPC port $archOpcode(ppc64Opcode_addis);
7879     __ addis($dst$$Register, $src1$$Register, ($src2$$constant)>>16);
7880   %}
7881   ins_pipe(pipe_class_default);
7882 %}
7883 
7884 //---------------------
7885 // Subtraction Instructions
7886 
7887 // Register Subtraction
7888 instruct subI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
7889   match(Set dst (SubI src1 src2));
7890   format %{ "SUBF    $dst, $src2, $src1" %}
7891   size(4);
7892   ins_encode %{
7893     // TODO: PPC port $archOpcode(ppc64Opcode_subf);
7894     __ subf($dst$$Register, $src2$$Register, $src1$$Register);
7895   %}
7896   ins_pipe(pipe_class_default);
7897 %}
7898 
7899 // Immediate Subtraction
7900 // Immediate Subtraction: The compiler converts "x-c0" into "x+ -c0" (see SubLNode::Ideal),
7901 // Don't try to use addi with - $src2$$constant since it can overflow when $src2$$constant == minI16.
7902 
7903 // SubI from constant (using subfic).
7904 instruct subI_imm16_reg(iRegIdst dst, immI16 src1, iRegIsrc src2) %{
7905   match(Set dst (SubI src1 src2));
7906   format %{ "SUBI    $dst, $src1, $src2" %}
7907 
7908   size(4);
7909   ins_encode %{
7910     // TODO: PPC port $archOpcode(ppc64Opcode_subfic);
7911     __ subfic($dst$$Register, $src2$$Register, $src1$$constant);
7912   %}
7913   ins_pipe(pipe_class_default);
7914 %}
7915 
7916 // Turn the sign-bit of an integer into a 32-bit mask, 0x0...0 for
7917 // positive integers and 0xF...F for negative ones.
7918 instruct signmask32I_regI(iRegIdst dst, iRegIsrc src) %{
7919   // no match-rule, false predicate
7920   effect(DEF dst, USE src);
7921   predicate(false);
7922 
7923   format %{ "SRAWI   $dst, $src, #31" %}
7924   size(4);
7925   ins_encode %{
7926     // TODO: PPC port $archOpcode(ppc64Opcode_srawi);
7927     __ srawi($dst$$Register, $src$$Register, 0x1f);
7928   %}
7929   ins_pipe(pipe_class_default);
7930 %}
7931 
7932 instruct absI_reg_Ex(iRegIdst dst, iRegIsrc src) %{
7933   match(Set dst (AbsI src));
7934   ins_cost(DEFAULT_COST*3);
7935 
7936   expand %{
7937     iRegIdst tmp1;
7938     iRegIdst tmp2;
7939     signmask32I_regI(tmp1, src);
7940     xorI_reg_reg(tmp2, tmp1, src);
7941     subI_reg_reg(dst, tmp2, tmp1);
7942   %}
7943 %}
7944 
7945 instruct negI_regI(iRegIdst dst, immI_0 zero, iRegIsrc src2) %{
7946   match(Set dst (SubI zero src2));
7947   format %{ "NEG     $dst, $src2" %}
7948   size(4);
7949   ins_encode %{
7950     // TODO: PPC port $archOpcode(ppc64Opcode_neg);
7951     __ neg($dst$$Register, $src2$$Register);
7952   %}
7953   ins_pipe(pipe_class_default);
7954 %}
7955 
7956 // Long subtraction
7957 instruct subL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
7958   match(Set dst (SubL src1 src2));
7959   format %{ "SUBF    $dst, $src2, $src1 \t// long" %}
7960   size(4);
7961   ins_encode %{
7962     // TODO: PPC port $archOpcode(ppc64Opcode_subf);
7963     __ subf($dst$$Register, $src2$$Register, $src1$$Register);
7964   %}
7965   ins_pipe(pipe_class_default);
7966 %}
7967 
7968 // SubL + convL2I.
7969 instruct subI_regL_regL(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{
7970   match(Set dst (ConvL2I (SubL src1 src2)));
7971 
7972   format %{ "SUBF    $dst, $src2, $src1 \t// long + l2i" %}
7973   size(4);
7974   ins_encode %{
7975     // TODO: PPC port $archOpcode(ppc64Opcode_subf);
7976     __ subf($dst$$Register, $src2$$Register, $src1$$Register);
7977   %}
7978   ins_pipe(pipe_class_default);
7979 %}
7980 
7981 // Turn the sign-bit of a long into a 64-bit mask, 0x0...0 for
7982 // positive longs and 0xF...F for negative ones.
7983 instruct signmask64I_regL(iRegIdst dst, iRegLsrc src) %{
7984   // no match-rule, false predicate
7985   effect(DEF dst, USE src);
7986   predicate(false);
7987 
7988   format %{ "SRADI   $dst, $src, #63" %}
7989   size(4);
7990   ins_encode %{
7991     // TODO: PPC port $archOpcode(ppc64Opcode_sradi);
7992     __ sradi($dst$$Register, $src$$Register, 0x3f);
7993   %}
7994   ins_pipe(pipe_class_default);
7995 %}
7996 
7997 // Turn the sign-bit of a long into a 64-bit mask, 0x0...0 for
7998 // positive longs and 0xF...F for negative ones.
7999 instruct signmask64L_regL(iRegLdst dst, iRegLsrc src) %{
8000   // no match-rule, false predicate
8001   effect(DEF dst, USE src);
8002   predicate(false);
8003 
8004   format %{ "SRADI   $dst, $src, #63" %}
8005   size(4);
8006   ins_encode %{
8007     // TODO: PPC port $archOpcode(ppc64Opcode_sradi);
8008     __ sradi($dst$$Register, $src$$Register, 0x3f);
8009   %}
8010   ins_pipe(pipe_class_default);
8011 %}
8012 
8013 // Long negation
8014 instruct negL_reg_reg(iRegLdst dst, immL_0 zero, iRegLsrc src2) %{
8015   match(Set dst (SubL zero src2));
8016   format %{ "NEG     $dst, $src2 \t// long" %}
8017   size(4);
8018   ins_encode %{
8019     // TODO: PPC port $archOpcode(ppc64Opcode_neg);
8020     __ neg($dst$$Register, $src2$$Register);
8021   %}
8022   ins_pipe(pipe_class_default);
8023 %}
8024 
8025 // NegL + ConvL2I.
8026 instruct negI_con0_regL(iRegIdst dst, immL_0 zero, iRegLsrc src2) %{
8027   match(Set dst (ConvL2I (SubL zero src2)));
8028 
8029   format %{ "NEG     $dst, $src2 \t// long + l2i" %}
8030   size(4);
8031   ins_encode %{
8032     // TODO: PPC port $archOpcode(ppc64Opcode_neg);
8033     __ neg($dst$$Register, $src2$$Register);
8034   %}
8035   ins_pipe(pipe_class_default);
8036 %}
8037 
8038 // Multiplication Instructions
8039 // Integer Multiplication
8040 
8041 // Register Multiplication
8042 instruct mulI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8043   match(Set dst (MulI src1 src2));
8044   ins_cost(DEFAULT_COST);
8045 
8046   format %{ "MULLW   $dst, $src1, $src2" %}
8047   size(4);
8048   ins_encode %{
8049     // TODO: PPC port $archOpcode(ppc64Opcode_mullw);
8050     __ mullw($dst$$Register, $src1$$Register, $src2$$Register);
8051   %}
8052   ins_pipe(pipe_class_default);
8053 %}
8054 
8055 // Immediate Multiplication
8056 instruct mulI_reg_imm16(iRegIdst dst, iRegIsrc src1, immI16 src2) %{
8057   match(Set dst (MulI src1 src2));
8058   ins_cost(DEFAULT_COST);
8059 
8060   format %{ "MULLI   $dst, $src1, $src2" %}
8061   size(4);
8062   ins_encode %{
8063     // TODO: PPC port $archOpcode(ppc64Opcode_mulli);
8064     __ mulli($dst$$Register, $src1$$Register, $src2$$constant);
8065   %}
8066   ins_pipe(pipe_class_default);
8067 %}
8068 
8069 instruct mulL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8070   match(Set dst (MulL src1 src2));
8071   ins_cost(DEFAULT_COST);
8072 
8073   format %{ "MULLD   $dst $src1, $src2 \t// long" %}
8074   size(4);
8075   ins_encode %{
8076     // TODO: PPC port $archOpcode(ppc64Opcode_mulld);
8077     __ mulld($dst$$Register, $src1$$Register, $src2$$Register);
8078   %}
8079   ins_pipe(pipe_class_default);
8080 %}
8081 
8082 // Multiply high for optimized long division by constant.
8083 instruct mulHighL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8084   match(Set dst (MulHiL src1 src2));
8085   ins_cost(DEFAULT_COST);
8086 
8087   format %{ "MULHD   $dst $src1, $src2 \t// long" %}
8088   size(4);
8089   ins_encode %{
8090     // TODO: PPC port $archOpcode(ppc64Opcode_mulhd);
8091     __ mulhd($dst$$Register, $src1$$Register, $src2$$Register);
8092   %}
8093   ins_pipe(pipe_class_default);
8094 %}
8095 
8096 // Immediate Multiplication
8097 instruct mulL_reg_imm16(iRegLdst dst, iRegLsrc src1, immL16 src2) %{
8098   match(Set dst (MulL src1 src2));
8099   ins_cost(DEFAULT_COST);
8100 
8101   format %{ "MULLI   $dst, $src1, $src2" %}
8102   size(4);
8103   ins_encode %{
8104     // TODO: PPC port $archOpcode(ppc64Opcode_mulli);
8105     __ mulli($dst$$Register, $src1$$Register, $src2$$constant);
8106   %}
8107   ins_pipe(pipe_class_default);
8108 %}
8109 
8110 // Integer Division with Immediate -1: Negate.
8111 instruct divI_reg_immIvalueMinus1(iRegIdst dst, iRegIsrc src1, immI_minus1 src2) %{
8112   match(Set dst (DivI src1 src2));
8113   ins_cost(DEFAULT_COST);
8114 
8115   format %{ "NEG     $dst, $src1 \t// /-1" %}
8116   size(4);
8117   ins_encode %{
8118     // TODO: PPC port $archOpcode(ppc64Opcode_neg);
8119     __ neg($dst$$Register, $src1$$Register);
8120   %}
8121   ins_pipe(pipe_class_default);
8122 %}
8123 
8124 // Integer Division with constant, but not -1.
8125 // We should be able to improve this by checking the type of src2.
8126 // It might well be that src2 is known to be positive.
8127 instruct divI_reg_regnotMinus1(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8128   match(Set dst (DivI src1 src2));
8129   predicate(n->in(2)->find_int_con(-1) != -1); // src2 is a constant, but not -1
8130   ins_cost(2*DEFAULT_COST);
8131 
8132   format %{ "DIVW    $dst, $src1, $src2 \t// /not-1" %}
8133   size(4);
8134   ins_encode %{
8135     // TODO: PPC port $archOpcode(ppc64Opcode_divw);
8136     __ divw($dst$$Register, $src1$$Register, $src2$$Register);
8137   %}
8138   ins_pipe(pipe_class_default);
8139 %}
8140 
8141 instruct cmovI_bne_negI_reg(iRegIdst dst, flagsRegSrc crx, iRegIsrc src1) %{
8142   effect(USE_DEF dst, USE src1, USE crx);
8143   predicate(false);
8144 
8145   ins_variable_size_depending_on_alignment(true);
8146 
8147   format %{ "CMOVE   $dst, neg($src1), $crx" %}
8148   // Worst case is branch + move + stop, no stop without scheduler.
8149   size(false /* TODO: PPC PORT (InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 12 : 8);
8150   ins_encode %{
8151     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
8152     Label done;
8153     __ bne($crx$$CondRegister, done);
8154     __ neg($dst$$Register, $src1$$Register);
8155     // TODO PPC port __ endgroup_if_needed(_size == 12);
8156     __ bind(done);
8157   %}
8158   ins_pipe(pipe_class_default);
8159 %}
8160 
8161 // Integer Division with Registers not containing constants.
8162 instruct divI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8163   match(Set dst (DivI src1 src2));
8164   ins_cost(10*DEFAULT_COST);
8165 
8166   expand %{
8167     immI16 imm %{ (int)-1 %}
8168     flagsReg tmp1;
8169     cmpI_reg_imm16(tmp1, src2, imm);          // check src2 == -1
8170     divI_reg_regnotMinus1(dst, src1, src2);   // dst = src1 / src2
8171     cmovI_bne_negI_reg(dst, tmp1, src1);      // cmove dst = neg(src1) if src2 == -1
8172   %}
8173 %}
8174 
8175 // Long Division with Immediate -1: Negate.
8176 instruct divL_reg_immLvalueMinus1(iRegLdst dst, iRegLsrc src1, immL_minus1 src2) %{
8177   match(Set dst (DivL src1 src2));
8178   ins_cost(DEFAULT_COST);
8179 
8180   format %{ "NEG     $dst, $src1 \t// /-1, long" %}
8181   size(4);
8182   ins_encode %{
8183     // TODO: PPC port $archOpcode(ppc64Opcode_neg);
8184     __ neg($dst$$Register, $src1$$Register);
8185   %}
8186   ins_pipe(pipe_class_default);
8187 %}
8188 
8189 // Long Division with constant, but not -1.
8190 instruct divL_reg_regnotMinus1(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8191   match(Set dst (DivL src1 src2));
8192   predicate(n->in(2)->find_long_con(-1L) != -1L); // Src2 is a constant, but not -1.
8193   ins_cost(2*DEFAULT_COST);
8194 
8195   format %{ "DIVD    $dst, $src1, $src2 \t// /not-1, long" %}
8196   size(4);
8197   ins_encode %{
8198     // TODO: PPC port $archOpcode(ppc64Opcode_divd);
8199     __ divd($dst$$Register, $src1$$Register, $src2$$Register);
8200   %}
8201   ins_pipe(pipe_class_default);
8202 %}
8203 
8204 instruct cmovL_bne_negL_reg(iRegLdst dst, flagsRegSrc crx, iRegLsrc src1) %{
8205   effect(USE_DEF dst, USE src1, USE crx);
8206   predicate(false);
8207 
8208   ins_variable_size_depending_on_alignment(true);
8209 
8210   format %{ "CMOVE   $dst, neg($src1), $crx" %}
8211   // Worst case is branch + move + stop, no stop without scheduler.
8212   size(false /* TODO: PPC PORT (InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 12 : 8);
8213   ins_encode %{
8214     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
8215     Label done;
8216     __ bne($crx$$CondRegister, done);
8217     __ neg($dst$$Register, $src1$$Register);
8218     // TODO PPC port __ endgroup_if_needed(_size == 12);
8219     __ bind(done);
8220   %}
8221   ins_pipe(pipe_class_default);
8222 %}
8223 
8224 // Long Division with Registers not containing constants.
8225 instruct divL_reg_reg_Ex(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8226   match(Set dst (DivL src1 src2));
8227   ins_cost(10*DEFAULT_COST);
8228 
8229   expand %{
8230     immL16 imm %{ (int)-1 %}
8231     flagsReg tmp1;
8232     cmpL_reg_imm16(tmp1, src2, imm);          // check src2 == -1
8233     divL_reg_regnotMinus1(dst, src1, src2);   // dst = src1 / src2
8234     cmovL_bne_negL_reg(dst, tmp1, src1);      // cmove dst = neg(src1) if src2 == -1
8235   %}
8236 %}
8237 
8238 // Integer Remainder with registers.
8239 instruct modI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8240   match(Set dst (ModI src1 src2));
8241   ins_cost(10*DEFAULT_COST);
8242 
8243   expand %{
8244     immI16 imm %{ (int)-1 %}
8245     flagsReg tmp1;
8246     iRegIdst tmp2;
8247     iRegIdst tmp3;
8248     cmpI_reg_imm16(tmp1, src2, imm);           // check src2 == -1
8249     divI_reg_regnotMinus1(tmp2, src1, src2);   // tmp2 = src1 / src2
8250     cmovI_bne_negI_reg(tmp2, tmp1, src1);      // cmove tmp2 = neg(src1) if src2 == -1
8251     mulI_reg_reg(tmp3, src2, tmp2);            // tmp3 = src2 * tmp2
8252     subI_reg_reg(dst, src1, tmp3);             // dst = src1 - tmp3
8253   %}
8254 %}
8255 
8256 // Long Remainder with registers
8257 instruct modL_reg_reg_Ex(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8258   match(Set dst (ModL src1 src2));
8259   ins_cost(10*DEFAULT_COST);
8260 
8261   expand %{
8262     immL16 imm %{ (int)-1 %}
8263     flagsReg tmp1;
8264     iRegLdst tmp2;
8265     iRegLdst tmp3;
8266     cmpL_reg_imm16(tmp1, src2, imm);             // check src2 == -1
8267     divL_reg_regnotMinus1(tmp2, src1, src2);     // tmp2 = src1 / src2
8268     cmovL_bne_negL_reg(tmp2, tmp1, src1);        // cmove tmp2 = neg(src1) if src2 == -1
8269     mulL_reg_reg(tmp3, src2, tmp2);              // tmp3 = src2 * tmp2
8270     subL_reg_reg(dst, src1, tmp3);               // dst = src1 - tmp3
8271   %}
8272 %}
8273 
8274 // Integer Shift Instructions
8275 
8276 // Register Shift Left
8277 
8278 // Clear all but the lowest #mask bits.
8279 // Used to normalize shift amounts in registers.
8280 instruct maskI_reg_imm(iRegIdst dst, iRegIsrc src, uimmI6 mask) %{
8281   // no match-rule, false predicate
8282   effect(DEF dst, USE src, USE mask);
8283   predicate(false);
8284 
8285   format %{ "MASK    $dst, $src, $mask \t// clear $mask upper bits" %}
8286   size(4);
8287   ins_encode %{
8288     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
8289     __ clrldi($dst$$Register, $src$$Register, $mask$$constant);
8290   %}
8291   ins_pipe(pipe_class_default);
8292 %}
8293 
8294 instruct lShiftI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8295   // no match-rule, false predicate
8296   effect(DEF dst, USE src1, USE src2);
8297   predicate(false);
8298 
8299   format %{ "SLW     $dst, $src1, $src2" %}
8300   size(4);
8301   ins_encode %{
8302     // TODO: PPC port $archOpcode(ppc64Opcode_slw);
8303     __ slw($dst$$Register, $src1$$Register, $src2$$Register);
8304   %}
8305   ins_pipe(pipe_class_default);
8306 %}
8307 
8308 instruct lShiftI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8309   match(Set dst (LShiftI src1 src2));
8310   ins_cost(DEFAULT_COST*2);
8311   expand %{
8312     uimmI6 mask %{ 0x3b /* clear 59 bits, keep 5 */ %}
8313     iRegIdst tmpI;
8314     maskI_reg_imm(tmpI, src2, mask);
8315     lShiftI_reg_reg(dst, src1, tmpI);
8316   %}
8317 %}
8318 
8319 // Register Shift Left Immediate
8320 instruct lShiftI_reg_imm(iRegIdst dst, iRegIsrc src1, immI src2) %{
8321   match(Set dst (LShiftI src1 src2));
8322 
8323   format %{ "SLWI    $dst, $src1, ($src2 & 0x1f)" %}
8324   size(4);
8325   ins_encode %{
8326     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
8327     __ slwi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x1f);
8328   %}
8329   ins_pipe(pipe_class_default);
8330 %}
8331 
8332 // AndI with negpow2-constant + LShiftI
8333 instruct lShiftI_andI_immInegpow2_imm5(iRegIdst dst, iRegIsrc src1, immInegpow2 src2, uimmI5 src3) %{
8334   match(Set dst (LShiftI (AndI src1 src2) src3));
8335   predicate(UseRotateAndMaskInstructionsPPC64);
8336 
8337   format %{ "RLWINM  $dst, lShiftI(AndI($src1, $src2), $src3)" %}
8338   size(4);
8339   ins_encode %{
8340     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm); // FIXME: assert that rlwinm is equal to addi
8341     long src2      = $src2$$constant;
8342     long src3      = $src3$$constant;
8343     long maskbits  = src3 + log2_long((jlong) (julong) (juint) -src2);
8344     if (maskbits >= 32) {
8345       __ li($dst$$Register, 0); // addi
8346     } else {
8347       __ rlwinm($dst$$Register, $src1$$Register, src3 & 0x1f, 0, (31-maskbits) & 0x1f);
8348     }
8349   %}
8350   ins_pipe(pipe_class_default);
8351 %}
8352 
8353 // RShiftI + AndI with negpow2-constant + LShiftI
8354 instruct lShiftI_andI_immInegpow2_rShiftI_imm5(iRegIdst dst, iRegIsrc src1, immInegpow2 src2, uimmI5 src3) %{
8355   match(Set dst (LShiftI (AndI (RShiftI src1 src3) src2) src3));
8356   predicate(UseRotateAndMaskInstructionsPPC64);
8357 
8358   format %{ "RLWINM  $dst, lShiftI(AndI(RShiftI($src1, $src3), $src2), $src3)" %}
8359   size(4);
8360   ins_encode %{
8361     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm); // FIXME: assert that rlwinm is equal to addi
8362     long src2      = $src2$$constant;
8363     long src3      = $src3$$constant;
8364     long maskbits  = src3 + log2_long((jlong) (julong) (juint) -src2);
8365     if (maskbits >= 32) {
8366       __ li($dst$$Register, 0); // addi
8367     } else {
8368       __ rlwinm($dst$$Register, $src1$$Register, 0, 0, (31-maskbits) & 0x1f);
8369     }
8370   %}
8371   ins_pipe(pipe_class_default);
8372 %}
8373 
8374 instruct lShiftL_regL_regI(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8375   // no match-rule, false predicate
8376   effect(DEF dst, USE src1, USE src2);
8377   predicate(false);
8378 
8379   format %{ "SLD     $dst, $src1, $src2" %}
8380   size(4);
8381   ins_encode %{
8382     // TODO: PPC port $archOpcode(ppc64Opcode_sld);
8383     __ sld($dst$$Register, $src1$$Register, $src2$$Register);
8384   %}
8385   ins_pipe(pipe_class_default);
8386 %}
8387 
8388 // Register Shift Left
8389 instruct lShiftL_regL_regI_Ex(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8390   match(Set dst (LShiftL src1 src2));
8391   ins_cost(DEFAULT_COST*2);
8392   expand %{
8393     uimmI6 mask %{ 0x3a /* clear 58 bits, keep 6 */ %}
8394     iRegIdst tmpI;
8395     maskI_reg_imm(tmpI, src2, mask);
8396     lShiftL_regL_regI(dst, src1, tmpI);
8397   %}
8398 %}
8399 
8400 // Register Shift Left Immediate
8401 instruct lshiftL_regL_immI(iRegLdst dst, iRegLsrc src1, immI src2) %{
8402   match(Set dst (LShiftL src1 src2));
8403   format %{ "SLDI    $dst, $src1, ($src2 & 0x3f)" %}
8404   size(4);
8405   ins_encode %{
8406     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
8407     __ sldi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8408   %}
8409   ins_pipe(pipe_class_default);
8410 %}
8411 
8412 // If we shift more than 32 bits, we need not convert I2L.
8413 instruct lShiftL_regI_immGE32(iRegLdst dst, iRegIsrc src1, uimmI6_ge32 src2) %{
8414   match(Set dst (LShiftL (ConvI2L src1) src2));
8415   ins_cost(DEFAULT_COST);
8416 
8417   size(4);
8418   format %{ "SLDI    $dst, i2l($src1), $src2" %}
8419   ins_encode %{
8420     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
8421     __ sldi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8422   %}
8423   ins_pipe(pipe_class_default);
8424 %}
8425 
8426 // Shift a postivie int to the left.
8427 // Clrlsldi clears the upper 32 bits and shifts.
8428 instruct scaledPositiveI2L_lShiftL_convI2L_reg_imm6(iRegLdst dst, iRegIsrc src1, uimmI6 src2) %{
8429   match(Set dst (LShiftL (ConvI2L src1) src2));
8430   predicate(((ConvI2LNode*)(_kids[0]->_leaf))->type()->is_long()->is_positive_int());
8431 
8432   format %{ "SLDI    $dst, i2l(positive_int($src1)), $src2" %}
8433   size(4);
8434   ins_encode %{
8435     // TODO: PPC port $archOpcode(ppc64Opcode_rldic);
8436     __ clrlsldi($dst$$Register, $src1$$Register, 0x20, $src2$$constant);
8437   %}
8438   ins_pipe(pipe_class_default);
8439 %}
8440 
8441 instruct arShiftI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8442   // no match-rule, false predicate
8443   effect(DEF dst, USE src1, USE src2);
8444   predicate(false);
8445 
8446   format %{ "SRAW    $dst, $src1, $src2" %}
8447   size(4);
8448   ins_encode %{
8449     // TODO: PPC port $archOpcode(ppc64Opcode_sraw);
8450     __ sraw($dst$$Register, $src1$$Register, $src2$$Register);
8451   %}
8452   ins_pipe(pipe_class_default);
8453 %}
8454 
8455 // Register Arithmetic Shift Right
8456 instruct arShiftI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8457   match(Set dst (RShiftI src1 src2));
8458   ins_cost(DEFAULT_COST*2);
8459   expand %{
8460     uimmI6 mask %{ 0x3b /* clear 59 bits, keep 5 */ %}
8461     iRegIdst tmpI;
8462     maskI_reg_imm(tmpI, src2, mask);
8463     arShiftI_reg_reg(dst, src1, tmpI);
8464   %}
8465 %}
8466 
8467 // Register Arithmetic Shift Right Immediate
8468 instruct arShiftI_reg_imm(iRegIdst dst, iRegIsrc src1, immI src2) %{
8469   match(Set dst (RShiftI src1 src2));
8470 
8471   format %{ "SRAWI   $dst, $src1, ($src2 & 0x1f)" %}
8472   size(4);
8473   ins_encode %{
8474     // TODO: PPC port $archOpcode(ppc64Opcode_srawi);
8475     __ srawi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x1f);
8476   %}
8477   ins_pipe(pipe_class_default);
8478 %}
8479 
8480 instruct arShiftL_regL_regI(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8481   // no match-rule, false predicate
8482   effect(DEF dst, USE src1, USE src2);
8483   predicate(false);
8484 
8485   format %{ "SRAD    $dst, $src1, $src2" %}
8486   size(4);
8487   ins_encode %{
8488     // TODO: PPC port $archOpcode(ppc64Opcode_srad);
8489     __ srad($dst$$Register, $src1$$Register, $src2$$Register);
8490   %}
8491   ins_pipe(pipe_class_default);
8492 %}
8493 
8494 // Register Shift Right Arithmetic Long
8495 instruct arShiftL_regL_regI_Ex(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8496   match(Set dst (RShiftL src1 src2));
8497   ins_cost(DEFAULT_COST*2);
8498 
8499   expand %{
8500     uimmI6 mask %{ 0x3a /* clear 58 bits, keep 6 */ %}
8501     iRegIdst tmpI;
8502     maskI_reg_imm(tmpI, src2, mask);
8503     arShiftL_regL_regI(dst, src1, tmpI);
8504   %}
8505 %}
8506 
8507 // Register Shift Right Immediate
8508 instruct arShiftL_regL_immI(iRegLdst dst, iRegLsrc src1, immI src2) %{
8509   match(Set dst (RShiftL src1 src2));
8510 
8511   format %{ "SRADI   $dst, $src1, ($src2 & 0x3f)" %}
8512   size(4);
8513   ins_encode %{
8514     // TODO: PPC port $archOpcode(ppc64Opcode_sradi);
8515     __ sradi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8516   %}
8517   ins_pipe(pipe_class_default);
8518 %}
8519 
8520 // RShiftL + ConvL2I
8521 instruct convL2I_arShiftL_regL_immI(iRegIdst dst, iRegLsrc src1, immI src2) %{
8522   match(Set dst (ConvL2I (RShiftL src1 src2)));
8523 
8524   format %{ "SRADI   $dst, $src1, ($src2 & 0x3f) \t// long + l2i" %}
8525   size(4);
8526   ins_encode %{
8527     // TODO: PPC port $archOpcode(ppc64Opcode_sradi);
8528     __ sradi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8529   %}
8530   ins_pipe(pipe_class_default);
8531 %}
8532 
8533 instruct urShiftI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8534   // no match-rule, false predicate
8535   effect(DEF dst, USE src1, USE src2);
8536   predicate(false);
8537 
8538   format %{ "SRW     $dst, $src1, $src2" %}
8539   size(4);
8540   ins_encode %{
8541     // TODO: PPC port $archOpcode(ppc64Opcode_srw);
8542     __ srw($dst$$Register, $src1$$Register, $src2$$Register);
8543   %}
8544   ins_pipe(pipe_class_default);
8545 %}
8546 
8547 // Register Shift Right
8548 instruct urShiftI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8549   match(Set dst (URShiftI src1 src2));
8550   ins_cost(DEFAULT_COST*2);
8551 
8552   expand %{
8553     uimmI6 mask %{ 0x3b /* clear 59 bits, keep 5 */ %}
8554     iRegIdst tmpI;
8555     maskI_reg_imm(tmpI, src2, mask);
8556     urShiftI_reg_reg(dst, src1, tmpI);
8557   %}
8558 %}
8559 
8560 // Register Shift Right Immediate
8561 instruct urShiftI_reg_imm(iRegIdst dst, iRegIsrc src1, immI src2) %{
8562   match(Set dst (URShiftI src1 src2));
8563 
8564   format %{ "SRWI    $dst, $src1, ($src2 & 0x1f)" %}
8565   size(4);
8566   ins_encode %{
8567     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
8568     __ srwi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x1f);
8569   %}
8570   ins_pipe(pipe_class_default);
8571 %}
8572 
8573 instruct urShiftL_regL_regI(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8574   // no match-rule, false predicate
8575   effect(DEF dst, USE src1, USE src2);
8576   predicate(false);
8577 
8578   format %{ "SRD     $dst, $src1, $src2" %}
8579   size(4);
8580   ins_encode %{
8581     // TODO: PPC port $archOpcode(ppc64Opcode_srd);
8582     __ srd($dst$$Register, $src1$$Register, $src2$$Register);
8583   %}
8584   ins_pipe(pipe_class_default);
8585 %}
8586 
8587 // Register Shift Right
8588 instruct urShiftL_regL_regI_Ex(iRegLdst dst, iRegLsrc src1, iRegIsrc src2) %{
8589   match(Set dst (URShiftL src1 src2));
8590   ins_cost(DEFAULT_COST*2);
8591 
8592   expand %{
8593     uimmI6 mask %{ 0x3a /* clear 58 bits, keep 6 */ %}
8594     iRegIdst tmpI;
8595     maskI_reg_imm(tmpI, src2, mask);
8596     urShiftL_regL_regI(dst, src1, tmpI);
8597   %}
8598 %}
8599 
8600 // Register Shift Right Immediate
8601 instruct urShiftL_regL_immI(iRegLdst dst, iRegLsrc src1, immI src2) %{
8602   match(Set dst (URShiftL src1 src2));
8603 
8604   format %{ "SRDI    $dst, $src1, ($src2 & 0x3f)" %}
8605   size(4);
8606   ins_encode %{
8607     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
8608     __ srdi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8609   %}
8610   ins_pipe(pipe_class_default);
8611 %}
8612 
8613 // URShiftL + ConvL2I.
8614 instruct convL2I_urShiftL_regL_immI(iRegIdst dst, iRegLsrc src1, immI src2) %{
8615   match(Set dst (ConvL2I (URShiftL src1 src2)));
8616 
8617   format %{ "SRDI    $dst, $src1, ($src2 & 0x3f) \t// long + l2i" %}
8618   size(4);
8619   ins_encode %{
8620     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
8621     __ srdi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8622   %}
8623   ins_pipe(pipe_class_default);
8624 %}
8625 
8626 // Register Shift Right Immediate with a CastP2X
8627 instruct shrP_convP2X_reg_imm6(iRegLdst dst, iRegP_N2P src1, uimmI6 src2) %{
8628   match(Set dst (URShiftL (CastP2X src1) src2));
8629 
8630   format %{ "SRDI    $dst, $src1, $src2 \t// Cast ptr $src1 to long and shift" %}
8631   size(4);
8632   ins_encode %{
8633     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
8634     __ srdi($dst$$Register, $src1$$Register, ($src2$$constant) & 0x3f);
8635   %}
8636   ins_pipe(pipe_class_default);
8637 %}
8638 
8639 instruct sxtI_reg(iRegIdst dst, iRegIsrc src) %{
8640   match(Set dst (ConvL2I (ConvI2L src)));
8641 
8642   format %{ "EXTSW   $dst, $src \t// int->int" %}
8643   size(4);
8644   ins_encode %{
8645     // TODO: PPC port $archOpcode(ppc64Opcode_extsw);
8646     __ extsw($dst$$Register, $src$$Register);
8647   %}
8648   ins_pipe(pipe_class_default);
8649 %}
8650 
8651 //----------Rotate Instructions------------------------------------------------
8652 
8653 // Rotate Left by 8-bit immediate
8654 instruct rotlI_reg_immi8(iRegIdst dst, iRegIsrc src, immI8 lshift, immI8 rshift) %{
8655   match(Set dst (OrI (LShiftI src lshift) (URShiftI src rshift)));
8656   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8657 
8658   format %{ "ROTLWI  $dst, $src, $lshift" %}
8659   size(4);
8660   ins_encode %{
8661     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
8662     __ rotlwi($dst$$Register, $src$$Register, $lshift$$constant);
8663   %}
8664   ins_pipe(pipe_class_default);
8665 %}
8666 
8667 // Rotate Right by 8-bit immediate
8668 instruct rotrI_reg_immi8(iRegIdst dst, iRegIsrc src, immI8 rshift, immI8 lshift) %{
8669   match(Set dst (OrI (URShiftI src rshift) (LShiftI src lshift)));
8670   predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8671 
8672   format %{ "ROTRWI  $dst, $rshift" %}
8673   size(4);
8674   ins_encode %{
8675     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
8676     __ rotrwi($dst$$Register, $src$$Register, $rshift$$constant);
8677   %}
8678   ins_pipe(pipe_class_default);
8679 %}
8680 
8681 //----------Floating Point Arithmetic Instructions-----------------------------
8682 
8683 // Add float single precision
8684 instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
8685   match(Set dst (AddF src1 src2));
8686 
8687   format %{ "FADDS   $dst, $src1, $src2" %}
8688   size(4);
8689   ins_encode %{
8690     // TODO: PPC port $archOpcode(ppc64Opcode_fadds);
8691     __ fadds($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8692   %}
8693   ins_pipe(pipe_class_default);
8694 %}
8695 
8696 // Add float double precision
8697 instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
8698   match(Set dst (AddD src1 src2));
8699 
8700   format %{ "FADD    $dst, $src1, $src2" %}
8701   size(4);
8702   ins_encode %{
8703     // TODO: PPC port $archOpcode(ppc64Opcode_fadd);
8704     __ fadd($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8705   %}
8706   ins_pipe(pipe_class_default);
8707 %}
8708 
8709 // Sub float single precision
8710 instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
8711   match(Set dst (SubF src1 src2));
8712 
8713   format %{ "FSUBS   $dst, $src1, $src2" %}
8714   size(4);
8715   ins_encode %{
8716     // TODO: PPC port $archOpcode(ppc64Opcode_fsubs);
8717     __ fsubs($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8718   %}
8719   ins_pipe(pipe_class_default);
8720 %}
8721 
8722 // Sub float double precision
8723 instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
8724   match(Set dst (SubD src1 src2));
8725   format %{ "FSUB    $dst, $src1, $src2" %}
8726   size(4);
8727   ins_encode %{
8728     // TODO: PPC port $archOpcode(ppc64Opcode_fsub);
8729     __ fsub($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8730   %}
8731   ins_pipe(pipe_class_default);
8732 %}
8733 
8734 // Mul float single precision
8735 instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
8736   match(Set dst (MulF src1 src2));
8737   format %{ "FMULS   $dst, $src1, $src2" %}
8738   size(4);
8739   ins_encode %{
8740     // TODO: PPC port $archOpcode(ppc64Opcode_fmuls);
8741     __ fmuls($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8742   %}
8743   ins_pipe(pipe_class_default);
8744 %}
8745 
8746 // Mul float double precision
8747 instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
8748   match(Set dst (MulD src1 src2));
8749   format %{ "FMUL    $dst, $src1, $src2" %}
8750   size(4);
8751   ins_encode %{
8752     // TODO: PPC port $archOpcode(ppc64Opcode_fmul);
8753     __ fmul($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8754   %}
8755   ins_pipe(pipe_class_default);
8756 %}
8757 
8758 // Div float single precision
8759 instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
8760   match(Set dst (DivF src1 src2));
8761   format %{ "FDIVS   $dst, $src1, $src2" %}
8762   size(4);
8763   ins_encode %{
8764     // TODO: PPC port $archOpcode(ppc64Opcode_fdivs);
8765     __ fdivs($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8766   %}
8767   ins_pipe(pipe_class_default);
8768 %}
8769 
8770 // Div float double precision
8771 instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
8772   match(Set dst (DivD src1 src2));
8773   format %{ "FDIV    $dst, $src1, $src2" %}
8774   size(4);
8775   ins_encode %{
8776     // TODO: PPC port $archOpcode(ppc64Opcode_fdiv);
8777     __ fdiv($dst$$FloatRegister, $src1$$FloatRegister, $src2$$FloatRegister);
8778   %}
8779   ins_pipe(pipe_class_default);
8780 %}
8781 
8782 // Absolute float single precision
8783 instruct absF_reg(regF dst, regF src) %{
8784   match(Set dst (AbsF src));
8785   format %{ "FABS    $dst, $src \t// float" %}
8786   size(4);
8787   ins_encode %{
8788     // TODO: PPC port $archOpcode(ppc64Opcode_fabs);
8789     __ fabs($dst$$FloatRegister, $src$$FloatRegister);
8790   %}
8791   ins_pipe(pipe_class_default);
8792 %}
8793 
8794 // Absolute float double precision
8795 instruct absD_reg(regD dst, regD src) %{
8796   match(Set dst (AbsD src));
8797   format %{ "FABS    $dst, $src \t// double" %}
8798   size(4);
8799   ins_encode %{
8800     // TODO: PPC port $archOpcode(ppc64Opcode_fabs);
8801     __ fabs($dst$$FloatRegister, $src$$FloatRegister);
8802   %}
8803   ins_pipe(pipe_class_default);
8804 %}
8805 
8806 instruct negF_reg(regF dst, regF src) %{
8807   match(Set dst (NegF src));
8808   format %{ "FNEG    $dst, $src \t// float" %}
8809   size(4);
8810   ins_encode %{
8811     // TODO: PPC port $archOpcode(ppc64Opcode_fneg);
8812     __ fneg($dst$$FloatRegister, $src$$FloatRegister);
8813   %}
8814   ins_pipe(pipe_class_default);
8815 %}
8816 
8817 instruct negD_reg(regD dst, regD src) %{
8818   match(Set dst (NegD src));
8819   format %{ "FNEG    $dst, $src \t// double" %}
8820   size(4);
8821   ins_encode %{
8822     // TODO: PPC port $archOpcode(ppc64Opcode_fneg);
8823     __ fneg($dst$$FloatRegister, $src$$FloatRegister);
8824   %}
8825   ins_pipe(pipe_class_default);
8826 %}
8827 
8828 // AbsF + NegF.
8829 instruct negF_absF_reg(regF dst, regF src) %{
8830   match(Set dst (NegF (AbsF src)));
8831   format %{ "FNABS   $dst, $src \t// float" %}
8832   size(4);
8833   ins_encode %{
8834     // TODO: PPC port $archOpcode(ppc64Opcode_fnabs);
8835     __ fnabs($dst$$FloatRegister, $src$$FloatRegister);
8836   %}
8837   ins_pipe(pipe_class_default);
8838 %}
8839 
8840 // AbsD + NegD.
8841 instruct negD_absD_reg(regD dst, regD src) %{
8842   match(Set dst (NegD (AbsD src)));
8843   format %{ "FNABS   $dst, $src \t// double" %}
8844   size(4);
8845   ins_encode %{
8846     // TODO: PPC port $archOpcode(ppc64Opcode_fnabs);
8847     __ fnabs($dst$$FloatRegister, $src$$FloatRegister);
8848   %}
8849   ins_pipe(pipe_class_default);
8850 %}
8851 
8852 // VM_Version::has_fsqrt() decides if this node will be used.
8853 // Sqrt float double precision
8854 instruct sqrtD_reg(regD dst, regD src) %{
8855   match(Set dst (SqrtD src));
8856   format %{ "FSQRT   $dst, $src" %}
8857   size(4);
8858   ins_encode %{
8859     // TODO: PPC port $archOpcode(ppc64Opcode_fsqrt);
8860     __ fsqrt($dst$$FloatRegister, $src$$FloatRegister);
8861   %}
8862   ins_pipe(pipe_class_default);
8863 %}
8864 
8865 // Single-precision sqrt.
8866 instruct sqrtF_reg(regF dst, regF src) %{
8867   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
8868   predicate(VM_Version::has_fsqrts());
8869   ins_cost(DEFAULT_COST);
8870 
8871   format %{ "FSQRTS  $dst, $src" %}
8872   size(4);
8873   ins_encode %{
8874     // TODO: PPC port $archOpcode(ppc64Opcode_fsqrts);
8875     __ fsqrts($dst$$FloatRegister, $src$$FloatRegister);
8876   %}
8877   ins_pipe(pipe_class_default);
8878 %}
8879 
8880 instruct roundDouble_nop(regD dst) %{
8881   match(Set dst (RoundDouble dst));
8882   ins_cost(0);
8883 
8884   format %{ " -- \t// RoundDouble not needed - empty" %}
8885   size(0);
8886   // PPC results are already "rounded" (i.e., normal-format IEEE).
8887   ins_encode( /*empty*/ );
8888   ins_pipe(pipe_class_default);
8889 %}
8890 
8891 instruct roundFloat_nop(regF dst) %{
8892   match(Set dst (RoundFloat dst));
8893   ins_cost(0);
8894 
8895   format %{ " -- \t// RoundFloat not needed - empty" %}
8896   size(0);
8897   // PPC results are already "rounded" (i.e., normal-format IEEE).
8898   ins_encode( /*empty*/ );
8899   ins_pipe(pipe_class_default);
8900 %}
8901 
8902 //----------Logical Instructions-----------------------------------------------
8903 
8904 // And Instructions
8905 
8906 // Register And
8907 instruct andI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
8908   match(Set dst (AndI src1 src2));
8909   format %{ "AND     $dst, $src1, $src2" %}
8910   size(4);
8911   ins_encode %{
8912     // TODO: PPC port $archOpcode(ppc64Opcode_and);
8913     __ andr($dst$$Register, $src1$$Register, $src2$$Register);
8914   %}
8915   ins_pipe(pipe_class_default);
8916 %}
8917 
8918 // Immediate And
8919 instruct andI_reg_uimm16(iRegIdst dst, iRegIsrc src1, uimmI16 src2, flagsRegCR0 cr0) %{
8920   match(Set dst (AndI src1 src2));
8921   effect(KILL cr0);
8922 
8923   format %{ "ANDI    $dst, $src1, $src2" %}
8924   size(4);
8925   ins_encode %{
8926     // TODO: PPC port $archOpcode(ppc64Opcode_andi_);
8927     // FIXME: avoid andi_ ?
8928     __ andi_($dst$$Register, $src1$$Register, $src2$$constant);
8929   %}
8930   ins_pipe(pipe_class_default);
8931 %}
8932 
8933 // Immediate And where the immediate is a negative power of 2.
8934 instruct andI_reg_immInegpow2(iRegIdst dst, iRegIsrc src1, immInegpow2 src2) %{
8935   match(Set dst (AndI src1 src2));
8936   format %{ "ANDWI   $dst, $src1, $src2" %}
8937   size(4);
8938   ins_encode %{
8939     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
8940     __ clrrdi($dst$$Register, $src1$$Register, log2_long((jlong)(julong)(juint)-($src2$$constant)));
8941   %}
8942   ins_pipe(pipe_class_default);
8943 %}
8944 
8945 instruct andI_reg_immIpow2minus1(iRegIdst dst, iRegIsrc src1, immIpow2minus1 src2) %{
8946   match(Set dst (AndI src1 src2));
8947   format %{ "ANDWI   $dst, $src1, $src2" %}
8948   size(4);
8949   ins_encode %{
8950     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
8951     __ clrldi($dst$$Register, $src1$$Register, 64-log2_long((((jlong) $src2$$constant)+1)));
8952   %}
8953   ins_pipe(pipe_class_default);
8954 %}
8955 
8956 instruct andI_reg_immIpowerOf2(iRegIdst dst, iRegIsrc src1, immIpowerOf2 src2) %{
8957   match(Set dst (AndI src1 src2));
8958   predicate(UseRotateAndMaskInstructionsPPC64);
8959   format %{ "ANDWI   $dst, $src1, $src2" %}
8960   size(4);
8961   ins_encode %{
8962     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
8963     __ rlwinm($dst$$Register, $src1$$Register, 0,
8964               (31-log2_long((jlong) $src2$$constant)) & 0x1f, (31-log2_long((jlong) $src2$$constant)) & 0x1f);
8965   %}
8966   ins_pipe(pipe_class_default);
8967 %}
8968 
8969 // Register And Long
8970 instruct andL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
8971   match(Set dst (AndL src1 src2));
8972   ins_cost(DEFAULT_COST);
8973 
8974   format %{ "AND     $dst, $src1, $src2 \t// long" %}
8975   size(4);
8976   ins_encode %{
8977     // TODO: PPC port $archOpcode(ppc64Opcode_and);
8978     __ andr($dst$$Register, $src1$$Register, $src2$$Register);
8979   %}
8980   ins_pipe(pipe_class_default);
8981 %}
8982 
8983 // Immediate And long
8984 instruct andL_reg_uimm16(iRegLdst dst, iRegLsrc src1, uimmL16 src2, flagsRegCR0 cr0) %{
8985   match(Set dst (AndL src1 src2));
8986   effect(KILL cr0);
8987 
8988   format %{ "ANDI    $dst, $src1, $src2 \t// long" %}
8989   size(4);
8990   ins_encode %{
8991     // TODO: PPC port $archOpcode(ppc64Opcode_andi_);
8992     // FIXME: avoid andi_ ?
8993     __ andi_($dst$$Register, $src1$$Register, $src2$$constant);
8994   %}
8995   ins_pipe(pipe_class_default);
8996 %}
8997 
8998 // Immediate And Long where the immediate is a negative power of 2.
8999 instruct andL_reg_immLnegpow2(iRegLdst dst, iRegLsrc src1, immLnegpow2 src2) %{
9000   match(Set dst (AndL src1 src2));
9001   format %{ "ANDDI   $dst, $src1, $src2" %}
9002   size(4);
9003   ins_encode %{
9004     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
9005     __ clrrdi($dst$$Register, $src1$$Register, log2_long((jlong)-$src2$$constant));
9006   %}
9007   ins_pipe(pipe_class_default);
9008 %}
9009 
9010 instruct andL_reg_immLpow2minus1(iRegLdst dst, iRegLsrc src1, immLpow2minus1 src2) %{
9011   match(Set dst (AndL src1 src2));
9012   format %{ "ANDDI   $dst, $src1, $src2" %}
9013   size(4);
9014   ins_encode %{
9015     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
9016     __ clrldi($dst$$Register, $src1$$Register, 64-log2_long((((jlong) $src2$$constant)+1)));
9017   %}
9018   ins_pipe(pipe_class_default);
9019 %}
9020 
9021 // AndL + ConvL2I.
9022 instruct convL2I_andL_reg_immLpow2minus1(iRegIdst dst, iRegLsrc src1, immLpow2minus1 src2) %{
9023   match(Set dst (ConvL2I (AndL src1 src2)));
9024   ins_cost(DEFAULT_COST);
9025 
9026   format %{ "ANDDI   $dst, $src1, $src2 \t// long + l2i" %}
9027   size(4);
9028   ins_encode %{
9029     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
9030     __ clrldi($dst$$Register, $src1$$Register, 64-log2_long((((jlong) $src2$$constant)+1)));
9031   %}
9032   ins_pipe(pipe_class_default);
9033 %}
9034 
9035 // Or Instructions
9036 
9037 // Register Or
9038 instruct orI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
9039   match(Set dst (OrI src1 src2));
9040   format %{ "OR      $dst, $src1, $src2" %}
9041   size(4);
9042   ins_encode %{
9043     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9044     __ or_unchecked($dst$$Register, $src1$$Register, $src2$$Register);
9045   %}
9046   ins_pipe(pipe_class_default);
9047 %}
9048 
9049 // Expand does not work with above instruct. (??)
9050 instruct orI_reg_reg_2(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
9051   // no match-rule
9052   effect(DEF dst, USE src1, USE src2);
9053   format %{ "OR      $dst, $src1, $src2" %}
9054   size(4);
9055   ins_encode %{
9056     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9057     __ or_unchecked($dst$$Register, $src1$$Register, $src2$$Register);
9058   %}
9059   ins_pipe(pipe_class_default);
9060 %}
9061 
9062 instruct tree_orI_orI_orI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2, iRegIsrc src3, iRegIsrc src4) %{
9063   match(Set dst (OrI (OrI (OrI src1 src2) src3) src4));
9064   ins_cost(DEFAULT_COST*3);
9065 
9066   expand %{
9067     // FIXME: we should do this in the ideal world.
9068     iRegIdst tmp1;
9069     iRegIdst tmp2;
9070     orI_reg_reg(tmp1, src1, src2);
9071     orI_reg_reg_2(tmp2, src3, src4); // Adlc complains about orI_reg_reg.
9072     orI_reg_reg(dst, tmp1, tmp2);
9073   %}
9074 %}
9075 
9076 // Immediate Or
9077 instruct orI_reg_uimm16(iRegIdst dst, iRegIsrc src1, uimmI16 src2) %{
9078   match(Set dst (OrI src1 src2));
9079   format %{ "ORI     $dst, $src1, $src2" %}
9080   size(4);
9081   ins_encode %{
9082     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
9083     __ ori($dst$$Register, $src1$$Register, ($src2$$constant) & 0xFFFF);
9084   %}
9085   ins_pipe(pipe_class_default);
9086 %}
9087 
9088 // Register Or Long
9089 instruct orL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
9090   match(Set dst (OrL src1 src2));
9091   ins_cost(DEFAULT_COST);
9092 
9093   size(4);
9094   format %{ "OR      $dst, $src1, $src2 \t// long" %}
9095   ins_encode %{
9096     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9097     __ or_unchecked($dst$$Register, $src1$$Register, $src2$$Register);
9098   %}
9099   ins_pipe(pipe_class_default);
9100 %}
9101 
9102 // OrL + ConvL2I.
9103 instruct orI_regL_regL(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{
9104   match(Set dst (ConvL2I (OrL src1 src2)));
9105   ins_cost(DEFAULT_COST);
9106 
9107   format %{ "OR      $dst, $src1, $src2 \t// long + l2i" %}
9108   size(4);
9109   ins_encode %{
9110     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9111     __ or_unchecked($dst$$Register, $src1$$Register, $src2$$Register);
9112   %}
9113   ins_pipe(pipe_class_default);
9114 %}
9115 
9116 // Immediate Or long
9117 instruct orL_reg_uimm16(iRegLdst dst, iRegLsrc src1, uimmL16 con) %{
9118   match(Set dst (OrL src1 con));
9119   ins_cost(DEFAULT_COST);
9120 
9121   format %{ "ORI     $dst, $src1, $con \t// long" %}
9122   size(4);
9123   ins_encode %{
9124     // TODO: PPC port $archOpcode(ppc64Opcode_ori);
9125     __ ori($dst$$Register, $src1$$Register, ($con$$constant) & 0xFFFF);
9126   %}
9127   ins_pipe(pipe_class_default);
9128 %}
9129 
9130 // Xor Instructions
9131 
9132 // Register Xor
9133 instruct xorI_reg_reg(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
9134   match(Set dst (XorI src1 src2));
9135   format %{ "XOR     $dst, $src1, $src2" %}
9136   size(4);
9137   ins_encode %{
9138     // TODO: PPC port $archOpcode(ppc64Opcode_xor);
9139     __ xorr($dst$$Register, $src1$$Register, $src2$$Register);
9140   %}
9141   ins_pipe(pipe_class_default);
9142 %}
9143 
9144 // Expand does not work with above instruct. (??)
9145 instruct xorI_reg_reg_2(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
9146   // no match-rule
9147   effect(DEF dst, USE src1, USE src2);
9148   format %{ "XOR     $dst, $src1, $src2" %}
9149   size(4);
9150   ins_encode %{
9151     // TODO: PPC port $archOpcode(ppc64Opcode_xor);
9152     __ xorr($dst$$Register, $src1$$Register, $src2$$Register);
9153   %}
9154   ins_pipe(pipe_class_default);
9155 %}
9156 
9157 instruct tree_xorI_xorI_xorI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2, iRegIsrc src3, iRegIsrc src4) %{
9158   match(Set dst (XorI (XorI (XorI src1 src2) src3) src4));
9159   ins_cost(DEFAULT_COST*3);
9160 
9161   expand %{
9162     // FIXME: we should do this in the ideal world.
9163     iRegIdst tmp1;
9164     iRegIdst tmp2;
9165     xorI_reg_reg(tmp1, src1, src2);
9166     xorI_reg_reg_2(tmp2, src3, src4); // Adlc complains about xorI_reg_reg.
9167     xorI_reg_reg(dst, tmp1, tmp2);
9168   %}
9169 %}
9170 
9171 // Immediate Xor
9172 instruct xorI_reg_uimm16(iRegIdst dst, iRegIsrc src1, uimmI16 src2) %{
9173   match(Set dst (XorI src1 src2));
9174   format %{ "XORI    $dst, $src1, $src2" %}
9175   size(4);
9176   ins_encode %{
9177     // TODO: PPC port $archOpcode(ppc64Opcode_xori);
9178     __ xori($dst$$Register, $src1$$Register, $src2$$constant);
9179   %}
9180   ins_pipe(pipe_class_default);
9181 %}
9182 
9183 // Register Xor Long
9184 instruct xorL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
9185   match(Set dst (XorL src1 src2));
9186   ins_cost(DEFAULT_COST);
9187 
9188   format %{ "XOR     $dst, $src1, $src2 \t// long" %}
9189   size(4);
9190   ins_encode %{
9191     // TODO: PPC port $archOpcode(ppc64Opcode_xor);
9192     __ xorr($dst$$Register, $src1$$Register, $src2$$Register);
9193   %}
9194   ins_pipe(pipe_class_default);
9195 %}
9196 
9197 // XorL + ConvL2I.
9198 instruct xorI_regL_regL(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{
9199   match(Set dst (ConvL2I (XorL src1 src2)));
9200   ins_cost(DEFAULT_COST);
9201 
9202   format %{ "XOR     $dst, $src1, $src2 \t// long + l2i" %}
9203   size(4);
9204   ins_encode %{
9205     // TODO: PPC port $archOpcode(ppc64Opcode_xor);
9206     __ xorr($dst$$Register, $src1$$Register, $src2$$Register);
9207   %}
9208   ins_pipe(pipe_class_default);
9209 %}
9210 
9211 // Immediate Xor Long
9212 instruct xorL_reg_uimm16(iRegLdst dst, iRegLsrc src1, uimmL16 src2) %{
9213   match(Set dst (XorL src1 src2));
9214   ins_cost(DEFAULT_COST);
9215 
9216   format %{ "XORI    $dst, $src1, $src2 \t// long" %}
9217   size(4);
9218   ins_encode %{
9219     // TODO: PPC port $archOpcode(ppc64Opcode_xori);
9220     __ xori($dst$$Register, $src1$$Register, $src2$$constant);
9221   %}
9222   ins_pipe(pipe_class_default);
9223 %}
9224 
9225 instruct notI_reg(iRegIdst dst, iRegIsrc src1, immI_minus1 src2) %{
9226   match(Set dst (XorI src1 src2));
9227   ins_cost(DEFAULT_COST);
9228 
9229   format %{ "NOT     $dst, $src1 ($src2)" %}
9230   size(4);
9231   ins_encode %{
9232     // TODO: PPC port $archOpcode(ppc64Opcode_nor);
9233     __ nor($dst$$Register, $src1$$Register, $src1$$Register);
9234   %}
9235   ins_pipe(pipe_class_default);
9236 %}
9237 
9238 instruct notL_reg(iRegLdst dst, iRegLsrc src1, immL_minus1 src2) %{
9239   match(Set dst (XorL src1 src2));
9240   ins_cost(DEFAULT_COST);
9241 
9242   format %{ "NOT     $dst, $src1 ($src2) \t// long" %}
9243   size(4);
9244   ins_encode %{
9245     // TODO: PPC port $archOpcode(ppc64Opcode_nor);
9246     __ nor($dst$$Register, $src1$$Register, $src1$$Register);
9247   %}
9248   ins_pipe(pipe_class_default);
9249 %}
9250 
9251 // And-complement
9252 instruct andcI_reg_reg(iRegIdst dst, iRegIsrc src1, immI_minus1 src2, iRegIsrc src3) %{
9253   match(Set dst (AndI (XorI src1 src2) src3));
9254   ins_cost(DEFAULT_COST);
9255 
9256   format %{ "ANDW    $dst, xori($src1, $src2), $src3" %}
9257   size(4);
9258   ins_encode( enc_andc(dst, src3, src1) );
9259   ins_pipe(pipe_class_default);
9260 %}
9261 
9262 // And-complement
9263 instruct andcL_reg_reg(iRegLdst dst, iRegLsrc src1, iRegLsrc src2) %{
9264   // no match-rule, false predicate
9265   effect(DEF dst, USE src1, USE src2);
9266   predicate(false);
9267 
9268   format %{ "ANDC    $dst, $src1, $src2" %}
9269   size(4);
9270   ins_encode %{
9271     // TODO: PPC port $archOpcode(ppc64Opcode_andc);
9272     __ andc($dst$$Register, $src1$$Register, $src2$$Register);
9273   %}
9274   ins_pipe(pipe_class_default);
9275 %}
9276 
9277 //----------Moves between int/long and float/double----------------------------
9278 //
9279 // The following rules move values from int/long registers/stack-locations
9280 // to float/double registers/stack-locations and vice versa, without doing any
9281 // conversions. These rules are used to implement the bit-conversion methods
9282 // of java.lang.Float etc., e.g.
9283 //   int   floatToIntBits(float value)
9284 //   float intBitsToFloat(int bits)
9285 //
9286 // Notes on the implementation on ppc64:
9287 // We only provide rules which move between a register and a stack-location,
9288 // because we always have to go through memory when moving between a float
9289 // register and an integer register.
9290 
9291 //---------- Chain stack slots between similar types --------
9292 
9293 // These are needed so that the rules below can match.
9294 
9295 // Load integer from stack slot
9296 instruct stkI_to_regI(iRegIdst dst, stackSlotI src) %{
9297   match(Set dst src);
9298   ins_cost(MEMORY_REF_COST);
9299 
9300   format %{ "LWZ     $dst, $src" %}
9301   size(4);
9302   ins_encode( enc_lwz(dst, src) );
9303   ins_pipe(pipe_class_memory);
9304 %}
9305 
9306 // Store integer to stack slot
9307 instruct regI_to_stkI(stackSlotI dst, iRegIsrc src) %{
9308   match(Set dst src);
9309   ins_cost(MEMORY_REF_COST);
9310 
9311   format %{ "STW     $src, $dst \t// stk" %}
9312   size(4);
9313   ins_encode( enc_stw(src, dst) ); // rs=rt
9314   ins_pipe(pipe_class_memory);
9315 %}
9316 
9317 // Load long from stack slot
9318 instruct stkL_to_regL(iRegLdst dst, stackSlotL src) %{
9319   match(Set dst src);
9320   ins_cost(MEMORY_REF_COST);
9321 
9322   format %{ "LD      $dst, $src \t// long" %}
9323   size(4);
9324   ins_encode( enc_ld(dst, src) );
9325   ins_pipe(pipe_class_memory);
9326 %}
9327 
9328 // Store long to stack slot
9329 instruct regL_to_stkL(stackSlotL dst, iRegLsrc src) %{
9330   match(Set dst src);
9331   ins_cost(MEMORY_REF_COST);
9332 
9333   format %{ "STD     $src, $dst \t// long" %}
9334   size(4);
9335   ins_encode( enc_std(src, dst) ); // rs=rt
9336   ins_pipe(pipe_class_memory);
9337 %}
9338 
9339 //----------Moves between int and float
9340 
9341 // Move float value from float stack-location to integer register.
9342 instruct moveF2I_stack_reg(iRegIdst dst, stackSlotF src) %{
9343   match(Set dst (MoveF2I src));
9344   ins_cost(MEMORY_REF_COST);
9345 
9346   format %{ "LWZ     $dst, $src \t// MoveF2I" %}
9347   size(4);
9348   ins_encode( enc_lwz(dst, src) );
9349   ins_pipe(pipe_class_memory);
9350 %}
9351 
9352 // Move float value from float register to integer stack-location.
9353 instruct moveF2I_reg_stack(stackSlotI dst, regF src) %{
9354   match(Set dst (MoveF2I src));
9355   ins_cost(MEMORY_REF_COST);
9356 
9357   format %{ "STFS    $src, $dst \t// MoveF2I" %}
9358   size(4);
9359   ins_encode( enc_stfs(src, dst) );
9360   ins_pipe(pipe_class_memory);
9361 %}
9362 
9363 // Move integer value from integer stack-location to float register.
9364 instruct moveI2F_stack_reg(regF dst, stackSlotI src) %{
9365   match(Set dst (MoveI2F src));
9366   ins_cost(MEMORY_REF_COST);
9367 
9368   format %{ "LFS     $dst, $src \t// MoveI2F" %}
9369   size(4);
9370   ins_encode %{
9371     // TODO: PPC port $archOpcode(ppc64Opcode_lfs);
9372     int Idisp = $src$$disp + frame_slots_bias($src$$base, ra_);
9373     __ lfs($dst$$FloatRegister, Idisp, $src$$base$$Register);
9374   %}
9375   ins_pipe(pipe_class_memory);
9376 %}
9377 
9378 // Move integer value from integer register to float stack-location.
9379 instruct moveI2F_reg_stack(stackSlotF dst, iRegIsrc src) %{
9380   match(Set dst (MoveI2F src));
9381   ins_cost(MEMORY_REF_COST);
9382 
9383   format %{ "STW     $src, $dst \t// MoveI2F" %}
9384   size(4);
9385   ins_encode( enc_stw(src, dst) );
9386   ins_pipe(pipe_class_memory);
9387 %}
9388 
9389 //----------Moves between long and float
9390 
9391 instruct moveF2L_reg_stack(stackSlotL dst, regF src) %{
9392   // no match-rule, false predicate
9393   effect(DEF dst, USE src);
9394   predicate(false);
9395 
9396   format %{ "storeD  $src, $dst \t// STACK" %}
9397   size(4);
9398   ins_encode( enc_stfd(src, dst) );
9399   ins_pipe(pipe_class_default);
9400 %}
9401 
9402 //----------Moves between long and double
9403 
9404 // Move double value from double stack-location to long register.
9405 instruct moveD2L_stack_reg(iRegLdst dst, stackSlotD src) %{
9406   match(Set dst (MoveD2L src));
9407   ins_cost(MEMORY_REF_COST);
9408   size(4);
9409   format %{ "LD      $dst, $src \t// MoveD2L" %}
9410   ins_encode( enc_ld(dst, src) );
9411   ins_pipe(pipe_class_memory);
9412 %}
9413 
9414 // Move double value from double register to long stack-location.
9415 instruct moveD2L_reg_stack(stackSlotL dst, regD src) %{
9416   match(Set dst (MoveD2L src));
9417   effect(DEF dst, USE src);
9418   ins_cost(MEMORY_REF_COST);
9419 
9420   format %{ "STFD    $src, $dst \t// MoveD2L" %}
9421   size(4);
9422   ins_encode( enc_stfd(src, dst) );
9423   ins_pipe(pipe_class_memory);
9424 %}
9425 
9426 // Move long value from long stack-location to double register.
9427 instruct moveL2D_stack_reg(regD dst, stackSlotL src) %{
9428   match(Set dst (MoveL2D src));
9429   ins_cost(MEMORY_REF_COST);
9430 
9431   format %{ "LFD     $dst, $src \t// MoveL2D" %}
9432   size(4);
9433   ins_encode( enc_lfd(dst, src) );
9434   ins_pipe(pipe_class_memory);
9435 %}
9436 
9437 // Move long value from long register to double stack-location.
9438 instruct moveL2D_reg_stack(stackSlotD dst, iRegLsrc src) %{
9439   match(Set dst (MoveL2D src));
9440   ins_cost(MEMORY_REF_COST);
9441 
9442   format %{ "STD     $src, $dst \t// MoveL2D" %}
9443   size(4);
9444   ins_encode( enc_std(src, dst) );
9445   ins_pipe(pipe_class_memory);
9446 %}
9447 
9448 //----------Register Move Instructions-----------------------------------------
9449 
9450 // Replicate for Superword
9451 
9452 instruct moveReg(iRegLdst dst, iRegIsrc src) %{
9453   predicate(false);
9454   effect(DEF dst, USE src);
9455 
9456   format %{ "MR      $dst, $src \t// replicate " %}
9457   // variable size, 0 or 4.
9458   ins_encode %{
9459     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9460     __ mr_if_needed($dst$$Register, $src$$Register);
9461   %}
9462   ins_pipe(pipe_class_default);
9463 %}
9464 
9465 //----------Cast instructions (Java-level type cast)---------------------------
9466 
9467 // Cast Long to Pointer for unsafe natives.
9468 instruct castX2P(iRegPdst dst, iRegLsrc src) %{
9469   match(Set dst (CastX2P src));
9470 
9471   format %{ "MR      $dst, $src \t// Long->Ptr" %}
9472   // variable size, 0 or 4.
9473   ins_encode %{
9474     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9475     __ mr_if_needed($dst$$Register, $src$$Register);
9476   %}
9477  ins_pipe(pipe_class_default);
9478 %}
9479 
9480 // Cast Pointer to Long for unsafe natives.
9481 instruct castP2X(iRegLdst dst, iRegP_N2P src) %{
9482   match(Set dst (CastP2X src));
9483 
9484   format %{ "MR      $dst, $src \t// Ptr->Long" %}
9485   // variable size, 0 or 4.
9486   ins_encode %{
9487     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9488     __ mr_if_needed($dst$$Register, $src$$Register);
9489   %}
9490   ins_pipe(pipe_class_default);
9491 %}
9492 
9493 instruct castPP(iRegPdst dst) %{
9494   match(Set dst (CastPP dst));
9495   format %{ " -- \t// castPP of $dst" %}
9496   size(0);
9497   ins_encode( /*empty*/ );
9498   ins_pipe(pipe_class_default);
9499 %}
9500 
9501 instruct castII(iRegIdst dst) %{
9502   match(Set dst (CastII dst));
9503   format %{ " -- \t// castII of $dst" %}
9504   size(0);
9505   ins_encode( /*empty*/ );
9506   ins_pipe(pipe_class_default);
9507 %}
9508 
9509 instruct checkCastPP(iRegPdst dst) %{
9510   match(Set dst (CheckCastPP dst));
9511   format %{ " -- \t// checkcastPP of $dst" %}
9512   size(0);
9513   ins_encode( /*empty*/ );
9514   ins_pipe(pipe_class_default);
9515 %}
9516 
9517 //----------Convert instructions-----------------------------------------------
9518 
9519 // Convert to boolean.
9520 
9521 // int_to_bool(src) : { 1   if src != 0
9522 //                    { 0   else
9523 //
9524 // strategy:
9525 // 1) Count leading zeros of 32 bit-value src,
9526 //    this returns 32 (0b10.0000) iff src == 0 and <32 otherwise.
9527 // 2) Shift 5 bits to the right, result is 0b1 iff src == 0, 0b0 otherwise.
9528 // 3) Xori the result to get 0b1 if src != 0 and 0b0 if src == 0.
9529 
9530 // convI2Bool
9531 instruct convI2Bool_reg__cntlz_Ex(iRegIdst dst, iRegIsrc src) %{
9532   match(Set dst (Conv2B src));
9533   predicate(UseCountLeadingZerosInstructionsPPC64);
9534   ins_cost(DEFAULT_COST);
9535 
9536   expand %{
9537     immI shiftAmount %{ 0x5 %}
9538     uimmI16 mask %{ 0x1 %}
9539     iRegIdst tmp1;
9540     iRegIdst tmp2;
9541     countLeadingZerosI(tmp1, src);
9542     urShiftI_reg_imm(tmp2, tmp1, shiftAmount);
9543     xorI_reg_uimm16(dst, tmp2, mask);
9544   %}
9545 %}
9546 
9547 instruct convI2Bool_reg__cmove(iRegIdst dst, iRegIsrc src, flagsReg crx) %{
9548   match(Set dst (Conv2B src));
9549   effect(TEMP crx);
9550   predicate(!UseCountLeadingZerosInstructionsPPC64);
9551   ins_cost(DEFAULT_COST);
9552 
9553   format %{ "CMPWI   $crx, $src, #0 \t// convI2B"
9554             "LI      $dst, #0\n\t"
9555             "BEQ     $crx, done\n\t"
9556             "LI      $dst, #1\n"
9557             "done:" %}
9558   size(16);
9559   ins_encode( enc_convI2B_regI__cmove(dst, src, crx, 0x0, 0x1) );
9560   ins_pipe(pipe_class_compare);
9561 %}
9562 
9563 // ConvI2B + XorI
9564 instruct xorI_convI2Bool_reg_immIvalue1__cntlz_Ex(iRegIdst dst, iRegIsrc src, immI_1 mask) %{
9565   match(Set dst (XorI (Conv2B src) mask));
9566   predicate(UseCountLeadingZerosInstructionsPPC64);
9567   ins_cost(DEFAULT_COST);
9568 
9569   expand %{
9570     immI shiftAmount %{ 0x5 %}
9571     iRegIdst tmp1;
9572     countLeadingZerosI(tmp1, src);
9573     urShiftI_reg_imm(dst, tmp1, shiftAmount);
9574   %}
9575 %}
9576 
9577 instruct xorI_convI2Bool_reg_immIvalue1__cmove(iRegIdst dst, iRegIsrc src, flagsReg crx, immI_1 mask) %{
9578   match(Set dst (XorI (Conv2B src) mask));
9579   effect(TEMP crx);
9580   predicate(!UseCountLeadingZerosInstructionsPPC64);
9581   ins_cost(DEFAULT_COST);
9582 
9583   format %{ "CMPWI   $crx, $src, #0 \t// Xor(convI2B($src), $mask)"
9584             "LI      $dst, #1\n\t"
9585             "BEQ     $crx, done\n\t"
9586             "LI      $dst, #0\n"
9587             "done:" %}
9588   size(16);
9589   ins_encode( enc_convI2B_regI__cmove(dst, src, crx, 0x1, 0x0) );
9590   ins_pipe(pipe_class_compare);
9591 %}
9592 
9593 // AndI 0b0..010..0 + ConvI2B
9594 instruct convI2Bool_andI_reg_immIpowerOf2(iRegIdst dst, iRegIsrc src, immIpowerOf2 mask) %{
9595   match(Set dst (Conv2B (AndI src mask)));
9596   predicate(UseRotateAndMaskInstructionsPPC64);
9597   ins_cost(DEFAULT_COST);
9598 
9599   format %{ "RLWINM  $dst, $src, $mask \t// convI2B(AndI($src, $mask))" %}
9600   size(4);
9601   ins_encode %{
9602     // TODO: PPC port $archOpcode(ppc64Opcode_rlwinm);
9603     __ rlwinm($dst$$Register, $src$$Register, (32-log2_long((jlong)$mask$$constant)) & 0x1f, 31, 31);
9604   %}
9605   ins_pipe(pipe_class_default);
9606 %}
9607 
9608 // Convert pointer to boolean.
9609 //
9610 // ptr_to_bool(src) : { 1   if src != 0
9611 //                    { 0   else
9612 //
9613 // strategy:
9614 // 1) Count leading zeros of 64 bit-value src,
9615 //    this returns 64 (0b100.0000) iff src == 0 and <64 otherwise.
9616 // 2) Shift 6 bits to the right, result is 0b1 iff src == 0, 0b0 otherwise.
9617 // 3) Xori the result to get 0b1 if src != 0 and 0b0 if src == 0.
9618 
9619 // ConvP2B
9620 instruct convP2Bool_reg__cntlz_Ex(iRegIdst dst, iRegP_N2P src) %{
9621   match(Set dst (Conv2B src));
9622   predicate(UseCountLeadingZerosInstructionsPPC64);
9623   ins_cost(DEFAULT_COST);
9624 
9625   expand %{
9626     immI shiftAmount %{ 0x6 %}
9627     uimmI16 mask %{ 0x1 %}
9628     iRegIdst tmp1;
9629     iRegIdst tmp2;
9630     countLeadingZerosP(tmp1, src);
9631     urShiftI_reg_imm(tmp2, tmp1, shiftAmount);
9632     xorI_reg_uimm16(dst, tmp2, mask);
9633   %}
9634 %}
9635 
9636 instruct convP2Bool_reg__cmove(iRegIdst dst, iRegP_N2P src, flagsReg crx) %{
9637   match(Set dst (Conv2B src));
9638   effect(TEMP crx);
9639   predicate(!UseCountLeadingZerosInstructionsPPC64);
9640   ins_cost(DEFAULT_COST);
9641 
9642   format %{ "CMPDI   $crx, $src, #0 \t// convP2B"
9643             "LI      $dst, #0\n\t"
9644             "BEQ     $crx, done\n\t"
9645             "LI      $dst, #1\n"
9646             "done:" %}
9647   size(16);
9648   ins_encode( enc_convP2B_regP__cmove(dst, src, crx, 0x0, 0x1) );
9649   ins_pipe(pipe_class_compare);
9650 %}
9651 
9652 // ConvP2B + XorI
9653 instruct xorI_convP2Bool_reg__cntlz_Ex(iRegIdst dst, iRegP_N2P src, immI_1 mask) %{
9654   match(Set dst (XorI (Conv2B src) mask));
9655   predicate(UseCountLeadingZerosInstructionsPPC64);
9656   ins_cost(DEFAULT_COST);
9657 
9658   expand %{
9659     immI shiftAmount %{ 0x6 %}
9660     iRegIdst tmp1;
9661     countLeadingZerosP(tmp1, src);
9662     urShiftI_reg_imm(dst, tmp1, shiftAmount);
9663   %}
9664 %}
9665 
9666 instruct xorI_convP2Bool_reg_immIvalue1__cmove(iRegIdst dst, iRegP_N2P src, flagsReg crx, immI_1 mask) %{
9667   match(Set dst (XorI (Conv2B src) mask));
9668   effect(TEMP crx);
9669   predicate(!UseCountLeadingZerosInstructionsPPC64);
9670   ins_cost(DEFAULT_COST);
9671 
9672   format %{ "CMPDI   $crx, $src, #0 \t// XorI(convP2B($src), $mask)"
9673             "LI      $dst, #1\n\t"
9674             "BEQ     $crx, done\n\t"
9675             "LI      $dst, #0\n"
9676             "done:" %}
9677   size(16);
9678   ins_encode( enc_convP2B_regP__cmove(dst, src, crx, 0x1, 0x0) );
9679   ins_pipe(pipe_class_compare);
9680 %}
9681 
9682 // if src1 < src2, return -1 else return 0
9683 instruct cmpLTMask_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
9684   match(Set dst (CmpLTMask src1 src2));
9685   ins_cost(DEFAULT_COST*4);
9686 
9687   expand %{
9688     iRegLdst src1s;
9689     iRegLdst src2s;
9690     iRegLdst diff;
9691     convI2L_reg(src1s, src1); // Ensure proper sign extension.
9692     convI2L_reg(src2s, src2); // Ensure proper sign extension.
9693     subL_reg_reg(diff, src1s, src2s);
9694     // Need to consider >=33 bit result, therefore we need signmaskL.
9695     signmask64I_regL(dst, diff);
9696   %}
9697 %}
9698 
9699 instruct cmpLTMask_reg_immI0(iRegIdst dst, iRegIsrc src1, immI_0 src2) %{
9700   match(Set dst (CmpLTMask src1 src2)); // if src1 < src2, return -1 else return 0
9701   format %{ "SRAWI   $dst, $src1, $src2 \t// CmpLTMask" %}
9702   size(4);
9703   ins_encode %{
9704     // TODO: PPC port $archOpcode(ppc64Opcode_srawi);
9705     __ srawi($dst$$Register, $src1$$Register, 0x1f);
9706   %}
9707   ins_pipe(pipe_class_default);
9708 %}
9709 
9710 //----------Arithmetic Conversion Instructions---------------------------------
9711 
9712 // Convert to Byte  -- nop
9713 // Convert to Short -- nop
9714 
9715 // Convert to Int
9716 
9717 instruct convB2I_reg(iRegIdst dst, iRegIsrc src, immI_24 amount) %{
9718   match(Set dst (RShiftI (LShiftI src amount) amount));
9719   format %{ "EXTSB   $dst, $src \t// byte->int" %}
9720   size(4);
9721   ins_encode %{
9722     // TODO: PPC port $archOpcode(ppc64Opcode_extsb);
9723     __ extsb($dst$$Register, $src$$Register);
9724   %}
9725   ins_pipe(pipe_class_default);
9726 %}
9727 
9728 // LShiftI 16 + RShiftI 16 converts short to int.
9729 instruct convS2I_reg(iRegIdst dst, iRegIsrc src, immI_16 amount) %{
9730   match(Set dst (RShiftI (LShiftI src amount) amount));
9731   format %{ "EXTSH   $dst, $src \t// short->int" %}
9732   size(4);
9733   ins_encode %{
9734     // TODO: PPC port $archOpcode(ppc64Opcode_extsh);
9735     __ extsh($dst$$Register, $src$$Register);
9736   %}
9737   ins_pipe(pipe_class_default);
9738 %}
9739 
9740 // ConvL2I + ConvI2L: Sign extend int in long register.
9741 instruct sxtI_L2L_reg(iRegLdst dst, iRegLsrc src) %{
9742   match(Set dst (ConvI2L (ConvL2I src)));
9743 
9744   format %{ "EXTSW   $dst, $src \t// long->long" %}
9745   size(4);
9746   ins_encode %{
9747     // TODO: PPC port $archOpcode(ppc64Opcode_extsw);
9748     __ extsw($dst$$Register, $src$$Register);
9749   %}
9750   ins_pipe(pipe_class_default);
9751 %}
9752 
9753 instruct convL2I_reg(iRegIdst dst, iRegLsrc src) %{
9754   match(Set dst (ConvL2I src));
9755   format %{ "MR      $dst, $src \t// long->int" %}
9756   // variable size, 0 or 4
9757   ins_encode %{
9758     // TODO: PPC port $archOpcode(ppc64Opcode_or);
9759     __ mr_if_needed($dst$$Register, $src$$Register);
9760   %}
9761   ins_pipe(pipe_class_default);
9762 %}
9763 
9764 instruct convD2IRaw_regD(regD dst, regD src) %{
9765   // no match-rule, false predicate
9766   effect(DEF dst, USE src);
9767   predicate(false);
9768 
9769   format %{ "FCTIWZ $dst, $src \t// convD2I, $src != NaN" %}
9770   size(4);
9771   ins_encode %{
9772     // TODO: PPC port $archOpcode(ppc64Opcode_fctiwz);;
9773     __ fctiwz($dst$$FloatRegister, $src$$FloatRegister);
9774   %}
9775   ins_pipe(pipe_class_default);
9776 %}
9777 
9778 instruct cmovI_bso_stackSlotL(iRegIdst dst, flagsRegSrc crx, stackSlotL src) %{
9779   // no match-rule, false predicate
9780   effect(DEF dst, USE crx, USE src);
9781   predicate(false);
9782 
9783   ins_variable_size_depending_on_alignment(true);
9784 
9785   format %{ "cmovI   $crx, $dst, $src" %}
9786   // Worst case is branch + move + stop, no stop without scheduler.
9787   size(false /* TODO: PPC PORT(InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 12 : 8);
9788   ins_encode( enc_cmove_bso_stackSlotL(dst, crx, src) );
9789   ins_pipe(pipe_class_default);
9790 %}
9791 
9792 instruct cmovI_bso_stackSlotL_conLvalue0_Ex(iRegIdst dst, flagsRegSrc crx, stackSlotL mem) %{
9793   // no match-rule, false predicate
9794   effect(DEF dst, USE crx, USE mem);
9795   predicate(false);
9796 
9797   format %{ "CmovI   $dst, $crx, $mem \t// postalloc expanded" %}
9798   postalloc_expand %{
9799     //
9800     // replaces
9801     //
9802     //   region  dst  crx  mem
9803     //    \       |    |   /
9804     //     dst=cmovI_bso_stackSlotL_conLvalue0
9805     //
9806     // with
9807     //
9808     //   region  dst
9809     //    \       /
9810     //     dst=loadConI16(0)
9811     //      |
9812     //      ^  region  dst  crx  mem
9813     //      |   \       |    |    /
9814     //      dst=cmovI_bso_stackSlotL
9815     //
9816 
9817     // Create new nodes.
9818     MachNode *m1 = new loadConI16Node();
9819     MachNode *m2 = new cmovI_bso_stackSlotLNode();
9820 
9821     // inputs for new nodes
9822     m1->add_req(n_region);
9823     m2->add_req(n_region, n_crx, n_mem);
9824 
9825     // precedences for new nodes
9826     m2->add_prec(m1);
9827 
9828     // operands for new nodes
9829     m1->_opnds[0] = op_dst;
9830     m1->_opnds[1] = new immI16Oper(0);
9831 
9832     m2->_opnds[0] = op_dst;
9833     m2->_opnds[1] = op_crx;
9834     m2->_opnds[2] = op_mem;
9835 
9836     // registers for new nodes
9837     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
9838     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
9839 
9840     // Insert new nodes.
9841     nodes->push(m1);
9842     nodes->push(m2);
9843   %}
9844 %}
9845 
9846 // Double to Int conversion, NaN is mapped to 0.
9847 instruct convD2I_reg_ExEx(iRegIdst dst, regD src) %{
9848   match(Set dst (ConvD2I src));
9849   ins_cost(DEFAULT_COST);
9850 
9851   expand %{
9852     regD tmpD;
9853     stackSlotL tmpS;
9854     flagsReg crx;
9855     cmpDUnordered_reg_reg(crx, src, src);               // Check whether src is NaN.
9856     convD2IRaw_regD(tmpD, src);                         // Convert float to int (speculated).
9857     moveD2L_reg_stack(tmpS, tmpD);                      // Store float to stack (speculated).
9858     cmovI_bso_stackSlotL_conLvalue0_Ex(dst, crx, tmpS); // Cmove based on NaN check.
9859   %}
9860 %}
9861 
9862 instruct convF2IRaw_regF(regF dst, regF src) %{
9863   // no match-rule, false predicate
9864   effect(DEF dst, USE src);
9865   predicate(false);
9866 
9867   format %{ "FCTIWZ $dst, $src \t// convF2I, $src != NaN" %}
9868   size(4);
9869   ins_encode %{
9870     // TODO: PPC port $archOpcode(ppc64Opcode_fctiwz);
9871     __ fctiwz($dst$$FloatRegister, $src$$FloatRegister);
9872   %}
9873   ins_pipe(pipe_class_default);
9874 %}
9875 
9876 // Float to Int conversion, NaN is mapped to 0.
9877 instruct convF2I_regF_ExEx(iRegIdst dst, regF src) %{
9878   match(Set dst (ConvF2I src));
9879   ins_cost(DEFAULT_COST);
9880 
9881   expand %{
9882     regF tmpF;
9883     stackSlotL tmpS;
9884     flagsReg crx;
9885     cmpFUnordered_reg_reg(crx, src, src);               // Check whether src is NaN.
9886     convF2IRaw_regF(tmpF, src);                         // Convert float to int (speculated).
9887     moveF2L_reg_stack(tmpS, tmpF);                      // Store float to stack (speculated).
9888     cmovI_bso_stackSlotL_conLvalue0_Ex(dst, crx, tmpS); // Cmove based on NaN check.
9889   %}
9890 %}
9891 
9892 // Convert to Long
9893 
9894 instruct convI2L_reg(iRegLdst dst, iRegIsrc src) %{
9895   match(Set dst (ConvI2L src));
9896   format %{ "EXTSW   $dst, $src \t// int->long" %}
9897   size(4);
9898   ins_encode %{
9899     // TODO: PPC port $archOpcode(ppc64Opcode_extsw);
9900     __ extsw($dst$$Register, $src$$Register);
9901   %}
9902   ins_pipe(pipe_class_default);
9903 %}
9904 
9905 // Zero-extend: convert unsigned int to long (convUI2L).
9906 instruct zeroExtendL_regI(iRegLdst dst, iRegIsrc src, immL_32bits mask) %{
9907   match(Set dst (AndL (ConvI2L src) mask));
9908   ins_cost(DEFAULT_COST);
9909 
9910   format %{ "CLRLDI  $dst, $src, #32 \t// zero-extend int to long" %}
9911   size(4);
9912   ins_encode %{
9913     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
9914     __ clrldi($dst$$Register, $src$$Register, 32);
9915   %}
9916   ins_pipe(pipe_class_default);
9917 %}
9918 
9919 // Zero-extend: convert unsigned int to long in long register.
9920 instruct zeroExtendL_regL(iRegLdst dst, iRegLsrc src, immL_32bits mask) %{
9921   match(Set dst (AndL src mask));
9922   ins_cost(DEFAULT_COST);
9923 
9924   format %{ "CLRLDI  $dst, $src, #32 \t// zero-extend int to long" %}
9925   size(4);
9926   ins_encode %{
9927     // TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
9928     __ clrldi($dst$$Register, $src$$Register, 32);
9929   %}
9930   ins_pipe(pipe_class_default);
9931 %}
9932 
9933 instruct convF2LRaw_regF(regF dst, regF src) %{
9934   // no match-rule, false predicate
9935   effect(DEF dst, USE src);
9936   predicate(false);
9937 
9938   format %{ "FCTIDZ $dst, $src \t// convF2L, $src != NaN" %}
9939   size(4);
9940   ins_encode %{
9941     // TODO: PPC port $archOpcode(ppc64Opcode_fctiwz);
9942     __ fctidz($dst$$FloatRegister, $src$$FloatRegister);
9943   %}
9944   ins_pipe(pipe_class_default);
9945 %}
9946 
9947 instruct cmovL_bso_stackSlotL(iRegLdst dst, flagsRegSrc crx, stackSlotL src) %{
9948   // no match-rule, false predicate
9949   effect(DEF dst, USE crx, USE src);
9950   predicate(false);
9951 
9952   ins_variable_size_depending_on_alignment(true);
9953 
9954   format %{ "cmovL   $crx, $dst, $src" %}
9955   // Worst case is branch + move + stop, no stop without scheduler.
9956   size(false /* TODO: PPC PORT Compile::current()->do_hb_scheduling()*/ ? 12 : 8);
9957   ins_encode( enc_cmove_bso_stackSlotL(dst, crx, src) );
9958   ins_pipe(pipe_class_default);
9959 %}
9960 
9961 instruct cmovL_bso_stackSlotL_conLvalue0_Ex(iRegLdst dst, flagsRegSrc crx, stackSlotL mem) %{
9962   // no match-rule, false predicate
9963   effect(DEF dst, USE crx, USE mem);
9964   predicate(false);
9965 
9966   format %{ "CmovL   $dst, $crx, $mem \t// postalloc expanded" %}
9967   postalloc_expand %{
9968     //
9969     // replaces
9970     //
9971     //   region  dst  crx  mem
9972     //    \       |    |   /
9973     //     dst=cmovL_bso_stackSlotL_conLvalue0
9974     //
9975     // with
9976     //
9977     //   region  dst
9978     //    \       /
9979     //     dst=loadConL16(0)
9980     //      |
9981     //      ^  region  dst  crx  mem
9982     //      |   \       |    |    /
9983     //      dst=cmovL_bso_stackSlotL
9984     //
9985 
9986     // Create new nodes.
9987     MachNode *m1 = new loadConL16Node();
9988     MachNode *m2 = new cmovL_bso_stackSlotLNode();
9989 
9990     // inputs for new nodes
9991     m1->add_req(n_region);
9992     m2->add_req(n_region, n_crx, n_mem);
9993     m2->add_prec(m1);
9994 
9995     // operands for new nodes
9996     m1->_opnds[0] = op_dst;
9997     m1->_opnds[1] = new immL16Oper(0);
9998     m2->_opnds[0] = op_dst;
9999     m2->_opnds[1] = op_crx;
10000     m2->_opnds[2] = op_mem;
10001 
10002     // registers for new nodes
10003     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
10004     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
10005 
10006     // Insert new nodes.
10007     nodes->push(m1);
10008     nodes->push(m2);
10009   %}
10010 %}
10011 
10012 // Float to Long conversion, NaN is mapped to 0.
10013 instruct convF2L_reg_ExEx(iRegLdst dst, regF src) %{
10014   match(Set dst (ConvF2L src));
10015   ins_cost(DEFAULT_COST);
10016 
10017   expand %{
10018     regF tmpF;
10019     stackSlotL tmpS;
10020     flagsReg crx;
10021     cmpFUnordered_reg_reg(crx, src, src);               // Check whether src is NaN.
10022     convF2LRaw_regF(tmpF, src);                         // Convert float to long (speculated).
10023     moveF2L_reg_stack(tmpS, tmpF);                      // Store float to stack (speculated).
10024     cmovL_bso_stackSlotL_conLvalue0_Ex(dst, crx, tmpS); // Cmove based on NaN check.
10025   %}
10026 %}
10027 
10028 instruct convD2LRaw_regD(regD dst, regD src) %{
10029   // no match-rule, false predicate
10030   effect(DEF dst, USE src);
10031   predicate(false);
10032 
10033   format %{ "FCTIDZ $dst, $src \t// convD2L $src != NaN" %}
10034   size(4);
10035   ins_encode %{
10036     // TODO: PPC port $archOpcode(ppc64Opcode_fctiwz);
10037     __ fctidz($dst$$FloatRegister, $src$$FloatRegister);
10038   %}
10039   ins_pipe(pipe_class_default);
10040 %}
10041 
10042 // Double to Long conversion, NaN is mapped to 0.
10043 instruct convD2L_reg_ExEx(iRegLdst dst, regD src) %{
10044   match(Set dst (ConvD2L src));
10045   ins_cost(DEFAULT_COST);
10046 
10047   expand %{
10048     regD tmpD;
10049     stackSlotL tmpS;
10050     flagsReg crx;
10051     cmpDUnordered_reg_reg(crx, src, src);               // Check whether src is NaN.
10052     convD2LRaw_regD(tmpD, src);                         // Convert float to long (speculated).
10053     moveD2L_reg_stack(tmpS, tmpD);                      // Store float to stack (speculated).
10054     cmovL_bso_stackSlotL_conLvalue0_Ex(dst, crx, tmpS); // Cmove based on NaN check.
10055   %}
10056 %}
10057 
10058 // Convert to Float
10059 
10060 // Placed here as needed in expand.
10061 instruct convL2DRaw_regD(regD dst, regD src) %{
10062   // no match-rule, false predicate
10063   effect(DEF dst, USE src);
10064   predicate(false);
10065 
10066   format %{ "FCFID $dst, $src \t// convL2D" %}
10067   size(4);
10068   ins_encode %{
10069     // TODO: PPC port $archOpcode(ppc64Opcode_fcfid);
10070     __ fcfid($dst$$FloatRegister, $src$$FloatRegister);
10071   %}
10072   ins_pipe(pipe_class_default);
10073 %}
10074 
10075 // Placed here as needed in expand.
10076 instruct convD2F_reg(regF dst, regD src) %{
10077   match(Set dst (ConvD2F src));
10078   format %{ "FRSP    $dst, $src \t// convD2F" %}
10079   size(4);
10080   ins_encode %{
10081     // TODO: PPC port $archOpcode(ppc64Opcode_frsp);
10082     __ frsp($dst$$FloatRegister, $src$$FloatRegister);
10083   %}
10084   ins_pipe(pipe_class_default);
10085 %}
10086 
10087 // Integer to Float conversion.
10088 instruct convI2F_ireg_Ex(regF dst, iRegIsrc src) %{
10089   match(Set dst (ConvI2F src));
10090   predicate(!VM_Version::has_fcfids());
10091   ins_cost(DEFAULT_COST);
10092 
10093   expand %{
10094     iRegLdst tmpL;
10095     stackSlotL tmpS;
10096     regD tmpD;
10097     regD tmpD2;
10098     convI2L_reg(tmpL, src);              // Sign-extension int to long.
10099     regL_to_stkL(tmpS, tmpL);            // Store long to stack.
10100     moveL2D_stack_reg(tmpD, tmpS);       // Load long into double register.
10101     convL2DRaw_regD(tmpD2, tmpD);        // Convert to double.
10102     convD2F_reg(dst, tmpD2);             // Convert double to float.
10103   %}
10104 %}
10105 
10106 instruct convL2FRaw_regF(regF dst, regD src) %{
10107   // no match-rule, false predicate
10108   effect(DEF dst, USE src);
10109   predicate(false);
10110 
10111   format %{ "FCFIDS $dst, $src \t// convL2F" %}
10112   size(4);
10113   ins_encode %{
10114     // TODO: PPC port $archOpcode(ppc64Opcode_fcfid);
10115     __ fcfids($dst$$FloatRegister, $src$$FloatRegister);
10116   %}
10117   ins_pipe(pipe_class_default);
10118 %}
10119 
10120 // Integer to Float conversion. Special version for Power7.
10121 instruct convI2F_ireg_fcfids_Ex(regF dst, iRegIsrc src) %{
10122   match(Set dst (ConvI2F src));
10123   predicate(VM_Version::has_fcfids());
10124   ins_cost(DEFAULT_COST);
10125 
10126   expand %{
10127     iRegLdst tmpL;
10128     stackSlotL tmpS;
10129     regD tmpD;
10130     convI2L_reg(tmpL, src);              // Sign-extension int to long.
10131     regL_to_stkL(tmpS, tmpL);            // Store long to stack.
10132     moveL2D_stack_reg(tmpD, tmpS);       // Load long into double register.
10133     convL2FRaw_regF(dst, tmpD);          // Convert to float.
10134   %}
10135 %}
10136 
10137 // L2F to avoid runtime call.
10138 instruct convL2F_ireg_fcfids_Ex(regF dst, iRegLsrc src) %{
10139   match(Set dst (ConvL2F src));
10140   predicate(VM_Version::has_fcfids());
10141   ins_cost(DEFAULT_COST);
10142 
10143   expand %{
10144     stackSlotL tmpS;
10145     regD tmpD;
10146     regL_to_stkL(tmpS, src);             // Store long to stack.
10147     moveL2D_stack_reg(tmpD, tmpS);       // Load long into double register.
10148     convL2FRaw_regF(dst, tmpD);          // Convert to float.
10149   %}
10150 %}
10151 
10152 // Moved up as used in expand.
10153 //instruct convD2F_reg(regF dst, regD src) %{%}
10154 
10155 // Convert to Double
10156 
10157 // Integer to Double conversion.
10158 instruct convI2D_reg_Ex(regD dst, iRegIsrc src) %{
10159   match(Set dst (ConvI2D src));
10160   ins_cost(DEFAULT_COST);
10161 
10162   expand %{
10163     iRegLdst tmpL;
10164     stackSlotL tmpS;
10165     regD tmpD;
10166     convI2L_reg(tmpL, src);              // Sign-extension int to long.
10167     regL_to_stkL(tmpS, tmpL);            // Store long to stack.
10168     moveL2D_stack_reg(tmpD, tmpS);       // Load long into double register.
10169     convL2DRaw_regD(dst, tmpD);          // Convert to double.
10170   %}
10171 %}
10172 
10173 // Long to Double conversion
10174 instruct convL2D_reg_Ex(regD dst, stackSlotL src) %{
10175   match(Set dst (ConvL2D src));
10176   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
10177 
10178   expand %{
10179     regD tmpD;
10180     moveL2D_stack_reg(tmpD, src);
10181     convL2DRaw_regD(dst, tmpD);
10182   %}
10183 %}
10184 
10185 instruct convF2D_reg(regD dst, regF src) %{
10186   match(Set dst (ConvF2D src));
10187   format %{ "FMR     $dst, $src \t// float->double" %}
10188   // variable size, 0 or 4
10189   ins_encode %{
10190     // TODO: PPC port $archOpcode(ppc64Opcode_fmr);
10191     __ fmr_if_needed($dst$$FloatRegister, $src$$FloatRegister);
10192   %}
10193   ins_pipe(pipe_class_default);
10194 %}
10195 
10196 //----------Control Flow Instructions------------------------------------------
10197 // Compare Instructions
10198 
10199 // Compare Integers
10200 instruct cmpI_reg_reg(flagsReg crx, iRegIsrc src1, iRegIsrc src2) %{
10201   match(Set crx (CmpI src1 src2));
10202   size(4);
10203   format %{ "CMPW    $crx, $src1, $src2" %}
10204   ins_encode %{
10205     // TODO: PPC port $archOpcode(ppc64Opcode_cmp);
10206     __ cmpw($crx$$CondRegister, $src1$$Register, $src2$$Register);
10207   %}
10208   ins_pipe(pipe_class_compare);
10209 %}
10210 
10211 instruct cmpI_reg_imm16(flagsReg crx, iRegIsrc src1, immI16 src2) %{
10212   match(Set crx (CmpI src1 src2));
10213   format %{ "CMPWI   $crx, $src1, $src2" %}
10214   size(4);
10215   ins_encode %{
10216     // TODO: PPC port $archOpcode(ppc64Opcode_cmpi);
10217     __ cmpwi($crx$$CondRegister, $src1$$Register, $src2$$constant);
10218   %}
10219   ins_pipe(pipe_class_compare);
10220 %}
10221 
10222 // (src1 & src2) == 0?
10223 instruct testI_reg_imm(flagsRegCR0 cr0, iRegIsrc src1, uimmI16 src2, immI_0 zero) %{
10224   match(Set cr0 (CmpI (AndI src1 src2) zero));
10225   // r0 is killed
10226   format %{ "ANDI    R0, $src1, $src2 \t// BTST int" %}
10227   size(4);
10228   ins_encode %{
10229     // TODO: PPC port $archOpcode(ppc64Opcode_andi_);
10230     __ andi_(R0, $src1$$Register, $src2$$constant);
10231   %}
10232   ins_pipe(pipe_class_compare);
10233 %}
10234 
10235 instruct cmpL_reg_reg(flagsReg crx, iRegLsrc src1, iRegLsrc src2) %{
10236   match(Set crx (CmpL src1 src2));
10237   format %{ "CMPD    $crx, $src1, $src2" %}
10238   size(4);
10239   ins_encode %{
10240     // TODO: PPC port $archOpcode(ppc64Opcode_cmp);
10241     __ cmpd($crx$$CondRegister, $src1$$Register, $src2$$Register);
10242   %}
10243   ins_pipe(pipe_class_compare);
10244 %}
10245 
10246 instruct cmpL_reg_imm16(flagsReg crx, iRegLsrc src1, immL16 src2) %{
10247   match(Set crx (CmpL src1 src2));
10248   format %{ "CMPDI   $crx, $src1, $src2" %}
10249   size(4);
10250   ins_encode %{
10251     // TODO: PPC port $archOpcode(ppc64Opcode_cmpi);
10252     __ cmpdi($crx$$CondRegister, $src1$$Register, $src2$$constant);
10253   %}
10254   ins_pipe(pipe_class_compare);
10255 %}
10256 
10257 instruct testL_reg_reg(flagsRegCR0 cr0, iRegLsrc src1, iRegLsrc src2, immL_0 zero) %{
10258   match(Set cr0 (CmpL (AndL src1 src2) zero));
10259   // r0 is killed
10260   format %{ "AND     R0, $src1, $src2 \t// BTST long" %}
10261   size(4);
10262   ins_encode %{
10263     // TODO: PPC port $archOpcode(ppc64Opcode_and_);
10264     __ and_(R0, $src1$$Register, $src2$$Register);
10265   %}
10266   ins_pipe(pipe_class_compare);
10267 %}
10268 
10269 instruct testL_reg_imm(flagsRegCR0 cr0, iRegLsrc src1, uimmL16 src2, immL_0 zero) %{
10270   match(Set cr0 (CmpL (AndL src1 src2) zero));
10271   // r0 is killed
10272   format %{ "ANDI    R0, $src1, $src2 \t// BTST long" %}
10273   size(4);
10274   ins_encode %{
10275     // TODO: PPC port $archOpcode(ppc64Opcode_andi_);
10276     __ andi_(R0, $src1$$Register, $src2$$constant);
10277   %}
10278   ins_pipe(pipe_class_compare);
10279 %}
10280 
10281 instruct cmovI_conIvalueMinus1_conIvalue1(iRegIdst dst, flagsRegSrc crx) %{
10282   // no match-rule, false predicate
10283   effect(DEF dst, USE crx);
10284   predicate(false);
10285 
10286   ins_variable_size_depending_on_alignment(true);
10287 
10288   format %{ "cmovI   $crx, $dst, -1, 0, +1" %}
10289   // Worst case is branch + move + branch + move + stop, no stop without scheduler.
10290   size(false /* TODO: PPC PORTInsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 20 : 16);
10291   ins_encode %{
10292     // TODO: PPC port $archOpcode(ppc64Opcode_cmove);
10293     Label done;
10294     // li(Rdst, 0);              // equal -> 0
10295     __ beq($crx$$CondRegister, done);
10296     __ li($dst$$Register, 1);    // greater -> +1
10297     __ bgt($crx$$CondRegister, done);
10298     __ li($dst$$Register, -1);   // unordered or less -> -1
10299     // TODO: PPC port__ endgroup_if_needed(_size == 20);
10300     __ bind(done);
10301   %}
10302   ins_pipe(pipe_class_compare);
10303 %}
10304 
10305 instruct cmovI_conIvalueMinus1_conIvalue0_conIvalue1_Ex(iRegIdst dst, flagsRegSrc crx) %{
10306   // no match-rule, false predicate
10307   effect(DEF dst, USE crx);
10308   predicate(false);
10309 
10310   format %{ "CmovI    $crx, $dst, -1, 0, +1 \t// postalloc expanded" %}
10311   postalloc_expand %{
10312     //
10313     // replaces
10314     //
10315     //   region  crx
10316     //    \       |
10317     //     dst=cmovI_conIvalueMinus1_conIvalue0_conIvalue1
10318     //
10319     // with
10320     //
10321     //   region
10322     //    \
10323     //     dst=loadConI16(0)
10324     //      |
10325     //      ^  region  crx
10326     //      |   \       |
10327     //      dst=cmovI_conIvalueMinus1_conIvalue1
10328     //
10329 
10330     // Create new nodes.
10331     MachNode *m1 = new loadConI16Node();
10332     MachNode *m2 = new cmovI_conIvalueMinus1_conIvalue1Node();
10333 
10334     // inputs for new nodes
10335     m1->add_req(n_region);
10336     m2->add_req(n_region, n_crx);
10337     m2->add_prec(m1);
10338 
10339     // operands for new nodes
10340     m1->_opnds[0] = op_dst;
10341     m1->_opnds[1] = new immI16Oper(0);
10342     m2->_opnds[0] = op_dst;
10343     m2->_opnds[1] = op_crx;
10344 
10345     // registers for new nodes
10346     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
10347     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // dst
10348 
10349     // Insert new nodes.
10350     nodes->push(m1);
10351     nodes->push(m2);
10352   %}
10353 %}
10354 
10355 // Manifest a CmpL3 result in an integer register. Very painful.
10356 // This is the test to avoid.
10357 // (src1 < src2) ? -1 : ((src1 > src2) ? 1 : 0)
10358 instruct cmpL3_reg_reg_ExEx(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{
10359   match(Set dst (CmpL3 src1 src2));
10360   ins_cost(DEFAULT_COST*5+BRANCH_COST);
10361 
10362   expand %{
10363     flagsReg tmp1;
10364     cmpL_reg_reg(tmp1, src1, src2);
10365     cmovI_conIvalueMinus1_conIvalue0_conIvalue1_Ex(dst, tmp1);
10366   %}
10367 %}
10368 
10369 // Implicit range checks.
10370 // A range check in the ideal world has one of the following shapes:
10371 //  - (If le (CmpU length index)), (IfTrue  throw exception)
10372 //  - (If lt (CmpU index length)), (IfFalse throw exception)
10373 //
10374 // Match range check 'If le (CmpU length index)'.
10375 instruct rangeCheck_iReg_uimm15(cmpOp cmp, iRegIsrc src_length, uimmI15 index, label labl) %{
10376   match(If cmp (CmpU src_length index));
10377   effect(USE labl);
10378   predicate(TrapBasedRangeChecks &&
10379             _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le &&
10380             PROB_UNLIKELY(_leaf->as_If()->_prob) >= PROB_ALWAYS &&
10381             (Matcher::branches_to_uncommon_trap(_leaf)));
10382 
10383   ins_is_TrapBasedCheckNode(true);
10384 
10385   format %{ "TWI     $index $cmp $src_length \t// RangeCheck => trap $labl" %}
10386   size(4);
10387   ins_encode %{
10388     // TODO: PPC port $archOpcode(ppc64Opcode_twi);
10389     if ($cmp$$cmpcode == 0x1 /* less_equal */) {
10390       __ trap_range_check_le($src_length$$Register, $index$$constant);
10391     } else {
10392       // Both successors are uncommon traps, probability is 0.
10393       // Node got flipped during fixup flow.
10394       assert($cmp$$cmpcode == 0x9, "must be greater");
10395       __ trap_range_check_g($src_length$$Register, $index$$constant);
10396     }
10397   %}
10398   ins_pipe(pipe_class_trap);
10399 %}
10400 
10401 // Match range check 'If lt (CmpU index length)'.
10402 instruct rangeCheck_iReg_iReg(cmpOp cmp, iRegIsrc src_index, iRegIsrc src_length, label labl) %{
10403   match(If cmp (CmpU src_index src_length));
10404   effect(USE labl);
10405   predicate(TrapBasedRangeChecks &&
10406             _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt &&
10407             _leaf->as_If()->_prob >= PROB_ALWAYS &&
10408             (Matcher::branches_to_uncommon_trap(_leaf)));
10409 
10410   ins_is_TrapBasedCheckNode(true);
10411 
10412   format %{ "TW      $src_index $cmp $src_length \t// RangeCheck => trap $labl" %}
10413   size(4);
10414   ins_encode %{
10415     // TODO: PPC port $archOpcode(ppc64Opcode_tw);
10416     if ($cmp$$cmpcode == 0x0 /* greater_equal */) {
10417       __ trap_range_check_ge($src_index$$Register, $src_length$$Register);
10418     } else {
10419       // Both successors are uncommon traps, probability is 0.
10420       // Node got flipped during fixup flow.
10421       assert($cmp$$cmpcode == 0x8, "must be less");
10422       __ trap_range_check_l($src_index$$Register, $src_length$$Register);
10423     }
10424   %}
10425   ins_pipe(pipe_class_trap);
10426 %}
10427 
10428 // Match range check 'If lt (CmpU index length)'.
10429 instruct rangeCheck_uimm15_iReg(cmpOp cmp, iRegIsrc src_index, uimmI15 length, label labl) %{
10430   match(If cmp (CmpU src_index length));
10431   effect(USE labl);
10432   predicate(TrapBasedRangeChecks &&
10433             _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt &&
10434             _leaf->as_If()->_prob >= PROB_ALWAYS &&
10435             (Matcher::branches_to_uncommon_trap(_leaf)));
10436 
10437   ins_is_TrapBasedCheckNode(true);
10438 
10439   format %{ "TWI     $src_index $cmp $length \t// RangeCheck => trap $labl" %}
10440   size(4);
10441   ins_encode %{
10442     // TODO: PPC port $archOpcode(ppc64Opcode_twi);
10443     if ($cmp$$cmpcode == 0x0 /* greater_equal */) {
10444       __ trap_range_check_ge($src_index$$Register, $length$$constant);
10445     } else {
10446       // Both successors are uncommon traps, probability is 0.
10447       // Node got flipped during fixup flow.
10448       assert($cmp$$cmpcode == 0x8, "must be less");
10449       __ trap_range_check_l($src_index$$Register, $length$$constant);
10450     }
10451   %}
10452   ins_pipe(pipe_class_trap);
10453 %}
10454 
10455 instruct compU_reg_reg(flagsReg crx, iRegIsrc src1, iRegIsrc src2) %{
10456   match(Set crx (CmpU src1 src2));
10457   format %{ "CMPLW   $crx, $src1, $src2 \t// unsigned" %}
10458   size(4);
10459   ins_encode %{
10460     // TODO: PPC port $archOpcode(ppc64Opcode_cmpl);
10461     __ cmplw($crx$$CondRegister, $src1$$Register, $src2$$Register);
10462   %}
10463   ins_pipe(pipe_class_compare);
10464 %}
10465 
10466 instruct compU_reg_uimm16(flagsReg crx, iRegIsrc src1, uimmI16 src2) %{
10467   match(Set crx (CmpU src1 src2));
10468   size(4);
10469   format %{ "CMPLWI  $crx, $src1, $src2" %}
10470   ins_encode %{
10471     // TODO: PPC port $archOpcode(ppc64Opcode_cmpli);
10472     __ cmplwi($crx$$CondRegister, $src1$$Register, $src2$$constant);
10473   %}
10474   ins_pipe(pipe_class_compare);
10475 %}
10476 
10477 // Implicit zero checks (more implicit null checks).
10478 // No constant pool entries required.
10479 instruct zeroCheckN_iReg_imm0(cmpOp cmp, iRegNsrc value, immN_0 zero, label labl) %{
10480   match(If cmp (CmpN value zero));
10481   effect(USE labl);
10482   predicate(TrapBasedNullChecks &&
10483             _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne &&
10484             _leaf->as_If()->_prob >= PROB_LIKELY_MAG(4) &&
10485             Matcher::branches_to_uncommon_trap(_leaf));
10486   ins_cost(1);
10487 
10488   ins_is_TrapBasedCheckNode(true);
10489 
10490   format %{ "TDI     $value $cmp $zero \t// ZeroCheckN => trap $labl" %}
10491   size(4);
10492   ins_encode %{
10493     // TODO: PPC port $archOpcode(ppc64Opcode_tdi);
10494     if ($cmp$$cmpcode == 0xA) {
10495       __ trap_null_check($value$$Register);
10496     } else {
10497       // Both successors are uncommon traps, probability is 0.
10498       // Node got flipped during fixup flow.
10499       assert($cmp$$cmpcode == 0x2 , "must be equal(0xA) or notEqual(0x2)");
10500       __ trap_null_check($value$$Register, Assembler::traptoGreaterThanUnsigned);
10501     }
10502   %}
10503   ins_pipe(pipe_class_trap);
10504 %}
10505 
10506 // Compare narrow oops.
10507 instruct cmpN_reg_reg(flagsReg crx, iRegNsrc src1, iRegNsrc src2) %{
10508   match(Set crx (CmpN src1 src2));
10509 
10510   size(4);
10511   ins_cost(2);
10512   format %{ "CMPLW   $crx, $src1, $src2 \t// compressed ptr" %}
10513   ins_encode %{
10514     // TODO: PPC port $archOpcode(ppc64Opcode_cmpl);
10515     __ cmplw($crx$$CondRegister, $src1$$Register, $src2$$Register);
10516   %}
10517   ins_pipe(pipe_class_compare);
10518 %}
10519 
10520 instruct cmpN_reg_imm0(flagsReg crx, iRegNsrc src1, immN_0 src2) %{
10521   match(Set crx (CmpN src1 src2));
10522   // Make this more expensive than zeroCheckN_iReg_imm0.
10523   ins_cost(2);
10524 
10525   format %{ "CMPLWI  $crx, $src1, $src2 \t// compressed ptr" %}
10526   size(4);
10527   ins_encode %{
10528     // TODO: PPC port $archOpcode(ppc64Opcode_cmpli);
10529     __ cmplwi($crx$$CondRegister, $src1$$Register, $src2$$constant);
10530   %}
10531   ins_pipe(pipe_class_compare);
10532 %}
10533 
10534 // Implicit zero checks (more implicit null checks).
10535 // No constant pool entries required.
10536 instruct zeroCheckP_reg_imm0(cmpOp cmp, iRegP_N2P value, immP_0 zero, label labl) %{
10537   match(If cmp (CmpP value zero));
10538   effect(USE labl);
10539   predicate(TrapBasedNullChecks &&
10540             _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne &&
10541             _leaf->as_If()->_prob >= PROB_LIKELY_MAG(4) &&
10542             Matcher::branches_to_uncommon_trap(_leaf));
10543   ins_cost(1); // Should not be cheaper than zeroCheckN.
10544 
10545   ins_is_TrapBasedCheckNode(true);
10546 
10547   format %{ "TDI     $value $cmp $zero \t// ZeroCheckP => trap $labl" %}
10548   size(4);
10549   ins_encode %{
10550     // TODO: PPC port $archOpcode(ppc64Opcode_tdi);
10551     if ($cmp$$cmpcode == 0xA) {
10552       __ trap_null_check($value$$Register);
10553     } else {
10554       // Both successors are uncommon traps, probability is 0.
10555       // Node got flipped during fixup flow.
10556       assert($cmp$$cmpcode == 0x2 , "must be equal(0xA) or notEqual(0x2)");
10557       __ trap_null_check($value$$Register, Assembler::traptoGreaterThanUnsigned);
10558     }
10559   %}
10560   ins_pipe(pipe_class_trap);
10561 %}
10562 
10563 // Compare Pointers
10564 instruct cmpP_reg_reg(flagsReg crx, iRegP_N2P src1, iRegP_N2P src2) %{
10565   match(Set crx (CmpP src1 src2));
10566   format %{ "CMPLD   $crx, $src1, $src2 \t// ptr" %}
10567   size(4);
10568   ins_encode %{
10569     // TODO: PPC port $archOpcode(ppc64Opcode_cmpl);
10570     __ cmpld($crx$$CondRegister, $src1$$Register, $src2$$Register);
10571   %}
10572   ins_pipe(pipe_class_compare);
10573 %}
10574 
10575 // Used in postalloc expand.
10576 instruct cmpP_reg_imm16(flagsReg crx, iRegPsrc src1, immL16 src2) %{
10577   // This match rule prevents reordering of node before a safepoint.
10578   // This only makes sense if this instructions is used exclusively
10579   // for the expansion of EncodeP!
10580   match(Set crx (CmpP src1 src2));
10581   predicate(false);
10582 
10583   format %{ "CMPDI   $crx, $src1, $src2" %}
10584   size(4);
10585   ins_encode %{
10586     // TODO: PPC port $archOpcode(ppc64Opcode_cmpi);
10587     __ cmpdi($crx$$CondRegister, $src1$$Register, $src2$$constant);
10588   %}
10589   ins_pipe(pipe_class_compare);
10590 %}
10591 
10592 //----------Float Compares----------------------------------------------------
10593 
10594 instruct cmpFUnordered_reg_reg(flagsReg crx, regF src1, regF src2) %{
10595   // Needs matchrule, see cmpDUnordered.
10596   match(Set crx (CmpF src1 src2));
10597   // no match-rule, false predicate
10598   predicate(false);
10599 
10600   format %{ "cmpFUrd $crx, $src1, $src2" %}
10601   size(4);
10602   ins_encode %{
10603     // TODO: PPC port $archOpcode(ppc64Opcode_fcmpu);
10604     __ fcmpu($crx$$CondRegister, $src1$$FloatRegister, $src2$$FloatRegister);
10605   %}
10606   ins_pipe(pipe_class_default);
10607 %}
10608 
10609 instruct cmov_bns_less(flagsReg crx) %{
10610   // no match-rule, false predicate
10611   effect(DEF crx);
10612   predicate(false);
10613 
10614   ins_variable_size_depending_on_alignment(true);
10615 
10616   format %{ "cmov    $crx" %}
10617   // Worst case is branch + move + stop, no stop without scheduler.
10618   size(false /* TODO: PPC PORT(InsertEndGroupPPC64 && Compile::current()->do_hb_scheduling())*/ ? 16 : 12);
10619   ins_encode %{
10620     // TODO: PPC port $archOpcode(ppc64Opcode_cmovecr);
10621     Label done;
10622     __ bns($crx$$CondRegister, done);        // not unordered -> keep crx
10623     __ li(R0, 0);
10624     __ cmpwi($crx$$CondRegister, R0, 1);     // unordered -> set crx to 'less'
10625     // TODO PPC port __ endgroup_if_needed(_size == 16);
10626     __ bind(done);
10627   %}
10628   ins_pipe(pipe_class_default);
10629 %}
10630 
10631 // Compare floating, generate condition code.
10632 instruct cmpF_reg_reg_Ex(flagsReg crx, regF src1, regF src2) %{
10633   // FIXME: should we match 'If cmp (CmpF src1 src2))' ??
10634   //
10635   // The following code sequence occurs a lot in mpegaudio:
10636   //
10637   // block BXX:
10638   // 0: instruct cmpFUnordered_reg_reg (cmpF_reg_reg-0):
10639   //    cmpFUrd CCR6, F11, F9
10640   // 4: instruct cmov_bns_less (cmpF_reg_reg-1):
10641   //    cmov CCR6
10642   // 8: instruct branchConSched:
10643   //    B_FARle CCR6, B56  P=0.500000 C=-1.000000
10644   match(Set crx (CmpF src1 src2));
10645   ins_cost(DEFAULT_COST+BRANCH_COST);
10646 
10647   format %{ "CmpF    $crx, $src1, $src2 \t// postalloc expanded" %}
10648   postalloc_expand %{
10649     //
10650     // replaces
10651     //
10652     //   region  src1  src2
10653     //    \       |     |
10654     //     crx=cmpF_reg_reg
10655     //
10656     // with
10657     //
10658     //   region  src1  src2
10659     //    \       |     |
10660     //     crx=cmpFUnordered_reg_reg
10661     //      |
10662     //      ^  region
10663     //      |   \
10664     //      crx=cmov_bns_less
10665     //
10666 
10667     // Create new nodes.
10668     MachNode *m1 = new cmpFUnordered_reg_regNode();
10669     MachNode *m2 = new cmov_bns_lessNode();
10670 
10671     // inputs for new nodes
10672     m1->add_req(n_region, n_src1, n_src2);
10673     m2->add_req(n_region);
10674     m2->add_prec(m1);
10675 
10676     // operands for new nodes
10677     m1->_opnds[0] = op_crx;
10678     m1->_opnds[1] = op_src1;
10679     m1->_opnds[2] = op_src2;
10680     m2->_opnds[0] = op_crx;
10681 
10682     // registers for new nodes
10683     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // crx
10684     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // crx
10685 
10686     // Insert new nodes.
10687     nodes->push(m1);
10688     nodes->push(m2);
10689   %}
10690 %}
10691 
10692 // Compare float, generate -1,0,1
10693 instruct cmpF3_reg_reg_ExEx(iRegIdst dst, regF src1, regF src2) %{
10694   match(Set dst (CmpF3 src1 src2));
10695   ins_cost(DEFAULT_COST*5+BRANCH_COST);
10696 
10697   expand %{
10698     flagsReg tmp1;
10699     cmpFUnordered_reg_reg(tmp1, src1, src2);
10700     cmovI_conIvalueMinus1_conIvalue0_conIvalue1_Ex(dst, tmp1);
10701   %}
10702 %}
10703 
10704 instruct cmpDUnordered_reg_reg(flagsReg crx, regD src1, regD src2) %{
10705   // Needs matchrule so that ideal opcode is Cmp. This causes that gcm places the
10706   // node right before the conditional move using it.
10707   // In jck test api/java_awt/geom/QuadCurve2DFloat/index.html#SetCurveTesttestCase7,
10708   // compilation of java.awt.geom.RectangularShape::getBounds()Ljava/awt/Rectangle
10709   // crashed in register allocation where the flags Reg between cmpDUnoredered and a
10710   // conditional move was supposed to be spilled.
10711   match(Set crx (CmpD src1 src2));
10712   // False predicate, shall not be matched.
10713   predicate(false);
10714 
10715   format %{ "cmpFUrd $crx, $src1, $src2" %}
10716   size(4);
10717   ins_encode %{
10718     // TODO: PPC port $archOpcode(ppc64Opcode_fcmpu);
10719     __ fcmpu($crx$$CondRegister, $src1$$FloatRegister, $src2$$FloatRegister);
10720   %}
10721   ins_pipe(pipe_class_default);
10722 %}
10723 
10724 instruct cmpD_reg_reg_Ex(flagsReg crx, regD src1, regD src2) %{
10725   match(Set crx (CmpD src1 src2));
10726   ins_cost(DEFAULT_COST+BRANCH_COST);
10727 
10728   format %{ "CmpD    $crx, $src1, $src2 \t// postalloc expanded" %}
10729   postalloc_expand %{
10730     //
10731     // replaces
10732     //
10733     //   region  src1  src2
10734     //    \       |     |
10735     //     crx=cmpD_reg_reg
10736     //
10737     // with
10738     //
10739     //   region  src1  src2
10740     //    \       |     |
10741     //     crx=cmpDUnordered_reg_reg
10742     //      |
10743     //      ^  region
10744     //      |   \
10745     //      crx=cmov_bns_less
10746     //
10747 
10748     // create new nodes
10749     MachNode *m1 = new cmpDUnordered_reg_regNode();
10750     MachNode *m2 = new cmov_bns_lessNode();
10751 
10752     // inputs for new nodes
10753     m1->add_req(n_region, n_src1, n_src2);
10754     m2->add_req(n_region);
10755     m2->add_prec(m1);
10756 
10757     // operands for new nodes
10758     m1->_opnds[0] = op_crx;
10759     m1->_opnds[1] = op_src1;
10760     m1->_opnds[2] = op_src2;
10761     m2->_opnds[0] = op_crx;
10762 
10763     // registers for new nodes
10764     ra_->set_pair(m1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // crx
10765     ra_->set_pair(m2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this)); // crx
10766 
10767     // Insert new nodes.
10768     nodes->push(m1);
10769     nodes->push(m2);
10770   %}
10771 %}
10772 
10773 // Compare double, generate -1,0,1
10774 instruct cmpD3_reg_reg_ExEx(iRegIdst dst, regD src1, regD src2) %{
10775   match(Set dst (CmpD3 src1 src2));
10776   ins_cost(DEFAULT_COST*5+BRANCH_COST);
10777 
10778   expand %{
10779     flagsReg tmp1;
10780     cmpDUnordered_reg_reg(tmp1, src1, src2);
10781     cmovI_conIvalueMinus1_conIvalue0_conIvalue1_Ex(dst, tmp1);
10782   %}
10783 %}
10784 
10785 //----------Branches---------------------------------------------------------
10786 // Jump
10787 
10788 // Direct Branch.
10789 instruct branch(label labl) %{
10790   match(Goto);
10791   effect(USE labl);
10792   ins_cost(BRANCH_COST);
10793 
10794   format %{ "B       $labl" %}
10795   size(4);
10796   ins_encode %{
10797     // TODO: PPC port $archOpcode(ppc64Opcode_b);
10798      Label d;    // dummy
10799      __ bind(d);
10800      Label* p = $labl$$label;
10801      // `p' is `NULL' when this encoding class is used only to
10802      // determine the size of the encoded instruction.
10803      Label& l = (NULL == p)? d : *(p);
10804      __ b(l);
10805   %}
10806   ins_pipe(pipe_class_default);
10807 %}
10808 
10809 // Conditional Near Branch
10810 instruct branchCon(cmpOp cmp, flagsRegSrc crx, label lbl) %{
10811   // Same match rule as `branchConFar'.
10812   match(If cmp crx);
10813   effect(USE lbl);
10814   ins_cost(BRANCH_COST);
10815 
10816   // If set to 1 this indicates that the current instruction is a
10817   // short variant of a long branch. This avoids using this
10818   // instruction in first-pass matching. It will then only be used in
10819   // the `Shorten_branches' pass.
10820   ins_short_branch(1);
10821 
10822   format %{ "B$cmp     $crx, $lbl" %}
10823   size(4);
10824   ins_encode( enc_bc(crx, cmp, lbl) );
10825   ins_pipe(pipe_class_default);
10826 %}
10827 
10828 // This is for cases when the ppc64 `bc' instruction does not
10829 // reach far enough. So we emit a far branch here, which is more
10830 // expensive.
10831 //
10832 // Conditional Far Branch
10833 instruct branchConFar(cmpOp cmp, flagsRegSrc crx, label lbl) %{
10834   // Same match rule as `branchCon'.
10835   match(If cmp crx);
10836   effect(USE crx, USE lbl);
10837   predicate(!false /* TODO: PPC port HB_Schedule*/);
10838   // Higher cost than `branchCon'.
10839   ins_cost(5*BRANCH_COST);
10840 
10841   // This is not a short variant of a branch, but the long variant.
10842   ins_short_branch(0);
10843 
10844   format %{ "B_FAR$cmp $crx, $lbl" %}
10845   size(8);
10846   ins_encode( enc_bc_far(crx, cmp, lbl) );
10847   ins_pipe(pipe_class_default);
10848 %}
10849 
10850 // Conditional Branch used with Power6 scheduler (can be far or short).
10851 instruct branchConSched(cmpOp cmp, flagsRegSrc crx, label lbl) %{
10852   // Same match rule as `branchCon'.
10853   match(If cmp crx);
10854   effect(USE crx, USE lbl);
10855   predicate(false /* TODO: PPC port HB_Schedule*/);
10856   // Higher cost than `branchCon'.
10857   ins_cost(5*BRANCH_COST);
10858 
10859   // Actually size doesn't depend on alignment but on shortening.
10860   ins_variable_size_depending_on_alignment(true);
10861   // long variant.
10862   ins_short_branch(0);
10863 
10864   format %{ "B_FAR$cmp $crx, $lbl" %}
10865   size(8); // worst case
10866   ins_encode( enc_bc_short_far(crx, cmp, lbl) );
10867   ins_pipe(pipe_class_default);
10868 %}
10869 
10870 instruct branchLoopEnd(cmpOp cmp, flagsRegSrc crx, label labl) %{
10871   match(CountedLoopEnd cmp crx);
10872   effect(USE labl);
10873   ins_cost(BRANCH_COST);
10874 
10875   // short variant.
10876   ins_short_branch(1);
10877 
10878   format %{ "B$cmp     $crx, $labl \t// counted loop end" %}
10879   size(4);
10880   ins_encode( enc_bc(crx, cmp, labl) );
10881   ins_pipe(pipe_class_default);
10882 %}
10883 
10884 instruct branchLoopEndFar(cmpOp cmp, flagsRegSrc crx, label labl) %{
10885   match(CountedLoopEnd cmp crx);
10886   effect(USE labl);
10887   predicate(!false /* TODO: PPC port HB_Schedule */);
10888   ins_cost(BRANCH_COST);
10889 
10890   // Long variant.
10891   ins_short_branch(0);
10892 
10893   format %{ "B_FAR$cmp $crx, $labl \t// counted loop end" %}
10894   size(8);
10895   ins_encode( enc_bc_far(crx, cmp, labl) );
10896   ins_pipe(pipe_class_default);
10897 %}
10898 
10899 // Conditional Branch used with Power6 scheduler (can be far or short).
10900 instruct branchLoopEndSched(cmpOp cmp, flagsRegSrc crx, label labl) %{
10901   match(CountedLoopEnd cmp crx);
10902   effect(USE labl);
10903   predicate(false /* TODO: PPC port HB_Schedule */);
10904   // Higher cost than `branchCon'.
10905   ins_cost(5*BRANCH_COST);
10906 
10907   // Actually size doesn't depend on alignment but on shortening.
10908   ins_variable_size_depending_on_alignment(true);
10909   // Long variant.
10910   ins_short_branch(0);
10911 
10912   format %{ "B_FAR$cmp $crx, $labl \t// counted loop end" %}
10913   size(8); // worst case
10914   ins_encode( enc_bc_short_far(crx, cmp, labl) );
10915   ins_pipe(pipe_class_default);
10916 %}
10917 
10918 // ============================================================================
10919 // Java runtime operations, intrinsics and other complex operations.
10920 
10921 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary superklass
10922 // array for an instance of the superklass. Set a hidden internal cache on a
10923 // hit (cache is checked with exposed code in gen_subtype_check()). Return
10924 // not zero for a miss or zero for a hit. The encoding ALSO sets flags.
10925 //
10926 // GL TODO: Improve this.
10927 // - result should not be a TEMP
10928 // - Add match rule as on sparc avoiding additional Cmp.
10929 instruct partialSubtypeCheck(iRegPdst result, iRegP_N2P subklass, iRegP_N2P superklass,
10930                              iRegPdst tmp_klass, iRegPdst tmp_arrayptr) %{
10931   match(Set result (PartialSubtypeCheck subklass superklass));
10932   effect(TEMP_DEF result, TEMP tmp_klass, TEMP tmp_arrayptr);
10933   ins_cost(DEFAULT_COST*10);
10934 
10935   format %{ "PartialSubtypeCheck $result = ($subklass instanceOf $superklass) tmp: $tmp_klass, $tmp_arrayptr" %}
10936   ins_encode %{
10937     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
10938     __ check_klass_subtype_slow_path($subklass$$Register, $superklass$$Register, $tmp_arrayptr$$Register,
10939                                      $tmp_klass$$Register, NULL, $result$$Register);
10940   %}
10941   ins_pipe(pipe_class_default);
10942 %}
10943 
10944 // inlined locking and unlocking
10945 
10946 instruct cmpFastLock(flagsReg crx, iRegPdst oop, iRegPdst box, iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3) %{
10947   match(Set crx (FastLock oop box));
10948   effect(TEMP tmp1, TEMP tmp2, TEMP tmp3);
10949   predicate(!Compile::current()->use_rtm());
10950 
10951   format %{ "FASTLOCK  $oop, $box, $tmp1, $tmp2, $tmp3" %}
10952   ins_encode %{
10953     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
10954     __ compiler_fast_lock_object($crx$$CondRegister, $oop$$Register, $box$$Register,
10955                                  $tmp3$$Register, $tmp1$$Register, $tmp2$$Register,
10956                                  UseBiasedLocking && !UseOptoBiasInlining); // SAPJVM MD 2014-11-06 UseOptoBiasInlining
10957     // If locking was successfull, crx should indicate 'EQ'.
10958     // The compiler generates a branch to the runtime call to
10959     // _complete_monitor_locking_Java for the case where crx is 'NE'.
10960   %}
10961   ins_pipe(pipe_class_compare);
10962 %}
10963 
10964 // Separate version for TM. Use bound register for box to enable USE_KILL.
10965 instruct cmpFastLock_tm(flagsReg crx, iRegPdst oop, rarg2RegP box, iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3) %{
10966   match(Set crx (FastLock oop box));
10967   effect(TEMP tmp1, TEMP tmp2, TEMP tmp3, USE_KILL box);
10968   predicate(Compile::current()->use_rtm());
10969 
10970   format %{ "FASTLOCK  $oop, $box, $tmp1, $tmp2, $tmp3 (TM)" %}
10971   ins_encode %{
10972     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
10973     __ compiler_fast_lock_object($crx$$CondRegister, $oop$$Register, $box$$Register,
10974                                  $tmp3$$Register, $tmp1$$Register, $tmp2$$Register,
10975                                  /*Biased Locking*/ false,
10976                                  _rtm_counters, _stack_rtm_counters,
10977                                  ((Method*)(ra_->C->method()->constant_encoding()))->method_data(),
10978                                  /*TM*/ true, ra_->C->profile_rtm());
10979     // If locking was successfull, crx should indicate 'EQ'.
10980     // The compiler generates a branch to the runtime call to
10981     // _complete_monitor_locking_Java for the case where crx is 'NE'.
10982   %}
10983   ins_pipe(pipe_class_compare);
10984 %}
10985 
10986 instruct cmpFastUnlock(flagsReg crx, iRegPdst oop, iRegPdst box, iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3) %{
10987   match(Set crx (FastUnlock oop box));
10988   effect(TEMP tmp1, TEMP tmp2, TEMP tmp3);
10989   predicate(!Compile::current()->use_rtm());
10990 
10991   format %{ "FASTUNLOCK  $oop, $box, $tmp1, $tmp2" %}
10992   ins_encode %{
10993     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
10994     __ compiler_fast_unlock_object($crx$$CondRegister, $oop$$Register, $box$$Register,
10995                                    $tmp3$$Register, $tmp1$$Register, $tmp2$$Register,
10996                                    UseBiasedLocking && !UseOptoBiasInlining,
10997                                    false);
10998     // If unlocking was successfull, crx should indicate 'EQ'.
10999     // The compiler generates a branch to the runtime call to
11000     // _complete_monitor_unlocking_Java for the case where crx is 'NE'.
11001   %}
11002   ins_pipe(pipe_class_compare);
11003 %}
11004 
11005 instruct cmpFastUnlock_tm(flagsReg crx, iRegPdst oop, iRegPdst box, iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3) %{
11006   match(Set crx (FastUnlock oop box));
11007   effect(TEMP tmp1, TEMP tmp2, TEMP tmp3);
11008   predicate(Compile::current()->use_rtm());
11009 
11010   format %{ "FASTUNLOCK  $oop, $box, $tmp1, $tmp2 (TM)" %}
11011   ins_encode %{
11012     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11013     __ compiler_fast_unlock_object($crx$$CondRegister, $oop$$Register, $box$$Register,
11014                                    $tmp3$$Register, $tmp1$$Register, $tmp2$$Register,
11015                                    /*Biased Locking*/ false, /*TM*/ true);
11016     // If unlocking was successfull, crx should indicate 'EQ'.
11017     // The compiler generates a branch to the runtime call to
11018     // _complete_monitor_unlocking_Java for the case where crx is 'NE'.
11019   %}
11020   ins_pipe(pipe_class_compare);
11021 %}
11022 
11023 // Align address.
11024 instruct align_addr(iRegPdst dst, iRegPsrc src, immLnegpow2 mask) %{
11025   match(Set dst (CastX2P (AndL (CastP2X src) mask)));
11026 
11027   format %{ "ANDDI   $dst, $src, $mask \t// next aligned address" %}
11028   size(4);
11029   ins_encode %{
11030     // TODO: PPC port $archOpcode(ppc64Opcode_rldicr);
11031     __ clrrdi($dst$$Register, $src$$Register, log2_long((jlong)-$mask$$constant));
11032   %}
11033   ins_pipe(pipe_class_default);
11034 %}
11035 
11036 // Array size computation.
11037 instruct array_size(iRegLdst dst, iRegPsrc end, iRegPsrc start) %{
11038   match(Set dst (SubL (CastP2X end) (CastP2X start)));
11039 
11040   format %{ "SUB     $dst, $end, $start \t// array size in bytes" %}
11041   size(4);
11042   ins_encode %{
11043     // TODO: PPC port $archOpcode(ppc64Opcode_subf);
11044     __ subf($dst$$Register, $start$$Register, $end$$Register);
11045   %}
11046   ins_pipe(pipe_class_default);
11047 %}
11048 
11049 // Clear-array with dynamic array-size.
11050 instruct inlineCallClearArray(rarg1RegL cnt, rarg2RegP base, Universe dummy, regCTR ctr) %{
11051   match(Set dummy (ClearArray cnt base));
11052   effect(USE_KILL cnt, USE_KILL base, KILL ctr);
11053   ins_cost(MEMORY_REF_COST);
11054 
11055   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11056 
11057   format %{ "ClearArray $cnt, $base" %}
11058   ins_encode %{
11059     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11060     __ clear_memory_doubleword($base$$Register, $cnt$$Register); // kills cnt, base, R0
11061   %}
11062   ins_pipe(pipe_class_default);
11063 %}
11064 
11065 // String_IndexOf for needle of length 1.
11066 //
11067 // Match needle into immediate operands: no loadConP node needed. Saves one
11068 // register and two instructions over string_indexOf_imm1Node.
11069 //
11070 // Assumes register result differs from all input registers.
11071 //
11072 // Preserves registers haystack, haycnt
11073 // Kills     registers tmp1, tmp2
11074 // Defines   registers result
11075 //
11076 // Use dst register classes if register gets killed, as it is the case for tmp registers!
11077 //
11078 // Unfortunately this does not match too often. In many situations the AddP is used
11079 // by several nodes, even several StrIndexOf nodes, breaking the match tree.
11080 instruct string_indexOf_imm1_char(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
11081                                   immP needleImm, immL offsetImm, immI_1 needlecntImm,
11082                                   iRegIdst tmp1, iRegIdst tmp2,
11083                                   flagsRegCR0 cr0, flagsRegCR1 cr1) %{
11084   predicate(SpecialStringIndexOf);  // type check implicit by parameter type, See Matcher::match_rule_supported
11085   match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
11086 
11087   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1);
11088 
11089   ins_cost(150);
11090   format %{ "String IndexOf CSCL1 $haystack[0..$haycnt], $needleImm+$offsetImm[0..$needlecntImm]"
11091             "-> $result \t// KILL $haycnt, $tmp1, $tmp2, $cr0, $cr1" %}
11092 
11093   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted
11094   ins_encode %{
11095     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11096     immPOper *needleOper = (immPOper *)$needleImm;
11097     const TypeOopPtr *t = needleOper->type()->isa_oopptr();
11098     ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
11099 
11100     __ string_indexof_1($result$$Register,
11101                         $haystack$$Register, $haycnt$$Register,
11102                         R0, needle_values->char_at(0),
11103                         $tmp1$$Register, $tmp2$$Register);
11104   %}
11105   ins_pipe(pipe_class_compare);
11106 %}
11107 
11108 // String_IndexOf for needle of length 1.
11109 //
11110 // Special case requires less registers and emits less instructions.
11111 //
11112 // Assumes register result differs from all input registers.
11113 //
11114 // Preserves registers haystack, haycnt
11115 // Kills     registers tmp1, tmp2, needle
11116 // Defines   registers result
11117 //
11118 // Use dst register classes if register gets killed, as it is the case for tmp registers!
11119 instruct string_indexOf_imm1(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
11120                              rscratch2RegP needle, immI_1 needlecntImm,
11121                              iRegIdst tmp1, iRegIdst tmp2,
11122                              flagsRegCR0 cr0, flagsRegCR1 cr1) %{
11123   match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
11124   effect(USE_KILL needle, /* TDEF needle, */ TEMP_DEF result,
11125          TEMP tmp1, TEMP tmp2);
11126   // Required for EA: check if it is still a type_array.
11127   predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
11128             n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
11129   ins_cost(180);
11130 
11131   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11132 
11133   format %{ "String IndexOf SCL1 $haystack[0..$haycnt], $needle[0..$needlecntImm]"
11134             " -> $result \t// KILL $haycnt, $needle, $tmp1, $tmp2, $cr0, $cr1" %}
11135   ins_encode %{
11136     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11137     Node *ndl = in(operand_index($needle));  // The node that defines needle.
11138     ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
11139     guarantee(needle_values, "sanity");
11140     if (needle_values != NULL) {
11141       __ string_indexof_1($result$$Register,
11142                           $haystack$$Register, $haycnt$$Register,
11143                           R0, needle_values->char_at(0),
11144                           $tmp1$$Register, $tmp2$$Register);
11145     } else {
11146       __ string_indexof_1($result$$Register,
11147                           $haystack$$Register, $haycnt$$Register,
11148                           $needle$$Register, 0,
11149                           $tmp1$$Register, $tmp2$$Register);
11150     }
11151   %}
11152   ins_pipe(pipe_class_compare);
11153 %}
11154 
11155 // String_IndexOf.
11156 //
11157 // Length of needle as immediate. This saves instruction loading constant needle
11158 // length.
11159 // @@@ TODO Specify rules for length < 8 or so, and roll out comparison of needle
11160 // completely or do it in vector instruction. This should save registers for
11161 // needlecnt and needle.
11162 //
11163 // Assumes register result differs from all input registers.
11164 // Overwrites haycnt, needlecnt.
11165 // Use dst register classes if register gets killed, as it is the case for tmp registers!
11166 instruct string_indexOf_imm(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt,
11167                             iRegPsrc needle, uimmI15 needlecntImm,
11168                             iRegIdst tmp1, iRegIdst tmp2, iRegIdst tmp3, iRegIdst tmp4, iRegIdst tmp5,
11169                             flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6) %{
11170   match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
11171   effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP_DEF result,
11172          TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, KILL cr0, KILL cr1, KILL cr6);
11173   // Required for EA: check if it is still a type_array.
11174   predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
11175             n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
11176   ins_cost(250);
11177 
11178   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11179 
11180   format %{ "String IndexOf SCL $haystack[0..$haycnt], $needle[0..$needlecntImm]"
11181             " -> $result \t// KILL $haycnt, $tmp1, $tmp2, $tmp3, $tmp4, $tmp5, $cr0, $cr1" %}
11182   ins_encode %{
11183     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11184     Node *ndl = in(operand_index($needle));  // The node that defines needle.
11185     ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
11186 
11187     __ string_indexof($result$$Register,
11188                       $haystack$$Register, $haycnt$$Register,
11189                       $needle$$Register, needle_values, $tmp5$$Register, $needlecntImm$$constant,
11190                       $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register);
11191   %}
11192   ins_pipe(pipe_class_compare);
11193 %}
11194 
11195 // StrIndexOf node.
11196 //
11197 // Assumes register result differs from all input registers.
11198 // Overwrites haycnt, needlecnt.
11199 // Use dst register classes if register gets killed, as it is the case for tmp registers!
11200 instruct string_indexOf(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt, iRegPsrc needle, rscratch2RegI needlecnt,
11201                         iRegLdst tmp1, iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4,
11202                         flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6) %{
11203   match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
11204   effect(USE_KILL haycnt, USE_KILL needlecnt, /*better: TDEF haycnt, TDEF needlecnt,*/
11205          TEMP_DEF result,
11206          TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr0, KILL cr1, KILL cr6);
11207   predicate(SpecialStringIndexOf);  // See Matcher::match_rule_supported.
11208   ins_cost(300);
11209 
11210   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11211 
11212   format %{ "String IndexOf $haystack[0..$haycnt], $needle[0..$needlecnt]"
11213              " -> $result \t// KILL $haycnt, $needlecnt, $tmp1, $tmp2, $tmp3, $tmp4, $cr0, $cr1" %}
11214   ins_encode %{
11215     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11216     __ string_indexof($result$$Register,
11217                       $haystack$$Register, $haycnt$$Register,
11218                       $needle$$Register, NULL, $needlecnt$$Register, 0,  // needlecnt not constant.
11219                       $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register);
11220   %}
11221   ins_pipe(pipe_class_compare);
11222 %}
11223 
11224 // String equals with immediate.
11225 instruct string_equals_imm(iRegPsrc str1, iRegPsrc str2, uimmI15 cntImm, iRegIdst result,
11226                            iRegPdst tmp1, iRegPdst tmp2,
11227                            flagsRegCR0 cr0, flagsRegCR6 cr6, regCTR ctr) %{
11228   match(Set result (StrEquals (Binary str1 str2) cntImm));
11229   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2,
11230          KILL cr0, KILL cr6, KILL ctr);
11231   predicate(SpecialStringEquals);  // See Matcher::match_rule_supported.
11232   ins_cost(250);
11233 
11234   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11235 
11236   format %{ "String Equals SCL [0..$cntImm]($str1),[0..$cntImm]($str2)"
11237             " -> $result \t// KILL $cr0, $cr6, $ctr, TEMP $result, $tmp1, $tmp2" %}
11238   ins_encode %{
11239     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11240     __ char_arrays_equalsImm($str1$$Register, $str2$$Register, $cntImm$$constant,
11241                              $result$$Register, $tmp1$$Register, $tmp2$$Register);
11242   %}
11243   ins_pipe(pipe_class_compare);
11244 %}
11245 
11246 // String equals.
11247 // Use dst register classes if register gets killed, as it is the case for TEMP operands!
11248 instruct string_equals(iRegPsrc str1, iRegPsrc str2, iRegIsrc cnt, iRegIdst result,
11249                        iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3, iRegPdst tmp4, iRegPdst tmp5,
11250                        flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
11251   match(Set result (StrEquals (Binary str1 str2) cnt));
11252   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
11253          KILL cr0, KILL cr1, KILL cr6, KILL ctr);
11254   predicate(SpecialStringEquals);  // See Matcher::match_rule_supported.
11255   ins_cost(300);
11256 
11257   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11258 
11259   format %{ "String Equals [0..$cnt]($str1),[0..$cnt]($str2) -> $result"
11260             " \t// KILL $cr0, $cr1, $cr6, $ctr, TEMP $result, $tmp1, $tmp2, $tmp3, $tmp4, $tmp5" %}
11261   ins_encode %{
11262     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11263     __ char_arrays_equals($str1$$Register, $str2$$Register, $cnt$$Register, $result$$Register,
11264                           $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, $tmp5$$Register);
11265   %}
11266   ins_pipe(pipe_class_compare);
11267 %}
11268 
11269 // String compare.
11270 // Char[] pointers are passed in.
11271 // Use dst register classes if register gets killed, as it is the case for TEMP operands!
11272 instruct string_compare(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
11273                         iRegPdst tmp, flagsRegCR0 cr0, regCTR ctr) %{
11274   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11275   effect(USE_KILL cnt1, USE_KILL cnt2, USE_KILL str1, USE_KILL str2, TEMP_DEF result, TEMP tmp, KILL cr0, KILL ctr);
11276   ins_cost(300);
11277 
11278   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
11279 
11280   format %{ "String Compare $str1[0..$cnt1], $str2[0..$cnt2] -> $result"
11281             " \t// TEMP $tmp, $result KILLs $str1, $cnt1, $str2, $cnt2, $cr0, $ctr" %}
11282   ins_encode %{
11283     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11284     __ string_compare($str1$$Register, $str2$$Register, $cnt1$$Register, $cnt2$$Register,
11285                       $result$$Register, $tmp$$Register);
11286   %}
11287   ins_pipe(pipe_class_compare);
11288 %}
11289 
11290 //---------- Min/Max Instructions ---------------------------------------------
11291 
11292 instruct minI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
11293   match(Set dst (MinI src1 src2));
11294   ins_cost(DEFAULT_COST*6);
11295 
11296   expand %{
11297     iRegLdst src1s;
11298     iRegLdst src2s;
11299     iRegLdst diff;
11300     iRegLdst sm;
11301     iRegLdst doz; // difference or zero
11302     convI2L_reg(src1s, src1); // Ensure proper sign extension.
11303     convI2L_reg(src2s, src2); // Ensure proper sign extension.
11304     subL_reg_reg(diff, src2s, src1s);
11305     // Need to consider >=33 bit result, therefore we need signmaskL.
11306     signmask64L_regL(sm, diff);
11307     andL_reg_reg(doz, diff, sm); // <=0
11308     addI_regL_regL(dst, doz, src1s);
11309   %}
11310 %}
11311 
11312 instruct maxI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
11313   match(Set dst (MaxI src1 src2));
11314   ins_cost(DEFAULT_COST*6);
11315 
11316   expand %{
11317     iRegLdst src1s;
11318     iRegLdst src2s;
11319     iRegLdst diff;
11320     iRegLdst sm;
11321     iRegLdst doz; // difference or zero
11322     convI2L_reg(src1s, src1); // Ensure proper sign extension.
11323     convI2L_reg(src2s, src2); // Ensure proper sign extension.
11324     subL_reg_reg(diff, src2s, src1s);
11325     // Need to consider >=33 bit result, therefore we need signmaskL.
11326     signmask64L_regL(sm, diff);
11327     andcL_reg_reg(doz, diff, sm); // >=0
11328     addI_regL_regL(dst, doz, src1s);
11329   %}
11330 %}
11331 
11332 //---------- Population Count Instructions ------------------------------------
11333 
11334 // Popcnt for Power7.
11335 instruct popCountI(iRegIdst dst, iRegIsrc src) %{
11336   match(Set dst (PopCountI src));
11337   predicate(UsePopCountInstruction && VM_Version::has_popcntw());
11338   ins_cost(DEFAULT_COST);
11339 
11340   format %{ "POPCNTW $dst, $src" %}
11341   size(4);
11342   ins_encode %{
11343     // TODO: PPC port $archOpcode(ppc64Opcode_popcntb);
11344     __ popcntw($dst$$Register, $src$$Register);
11345   %}
11346   ins_pipe(pipe_class_default);
11347 %}
11348 
11349 // Popcnt for Power7.
11350 instruct popCountL(iRegIdst dst, iRegLsrc src) %{
11351   predicate(UsePopCountInstruction && VM_Version::has_popcntw());
11352   match(Set dst (PopCountL src));
11353   ins_cost(DEFAULT_COST);
11354 
11355   format %{ "POPCNTD $dst, $src" %}
11356   size(4);
11357   ins_encode %{
11358     // TODO: PPC port $archOpcode(ppc64Opcode_popcntb);
11359     __ popcntd($dst$$Register, $src$$Register);
11360   %}
11361   ins_pipe(pipe_class_default);
11362 %}
11363 
11364 instruct countLeadingZerosI(iRegIdst dst, iRegIsrc src) %{
11365   match(Set dst (CountLeadingZerosI src));
11366   predicate(UseCountLeadingZerosInstructionsPPC64);  // See Matcher::match_rule_supported.
11367   ins_cost(DEFAULT_COST);
11368 
11369   format %{ "CNTLZW  $dst, $src" %}
11370   size(4);
11371   ins_encode %{
11372     // TODO: PPC port $archOpcode(ppc64Opcode_cntlzw);
11373     __ cntlzw($dst$$Register, $src$$Register);
11374   %}
11375   ins_pipe(pipe_class_default);
11376 %}
11377 
11378 instruct countLeadingZerosL(iRegIdst dst, iRegLsrc src) %{
11379   match(Set dst (CountLeadingZerosL src));
11380   predicate(UseCountLeadingZerosInstructionsPPC64);  // See Matcher::match_rule_supported.
11381   ins_cost(DEFAULT_COST);
11382 
11383   format %{ "CNTLZD  $dst, $src" %}
11384   size(4);
11385   ins_encode %{
11386     // TODO: PPC port $archOpcode(ppc64Opcode_cntlzd);
11387     __ cntlzd($dst$$Register, $src$$Register);
11388   %}
11389   ins_pipe(pipe_class_default);
11390 %}
11391 
11392 instruct countLeadingZerosP(iRegIdst dst, iRegPsrc src) %{
11393   // no match-rule, false predicate
11394   effect(DEF dst, USE src);
11395   predicate(false);
11396 
11397   format %{ "CNTLZD  $dst, $src" %}
11398   size(4);
11399   ins_encode %{
11400     // TODO: PPC port $archOpcode(ppc64Opcode_cntlzd);
11401     __ cntlzd($dst$$Register, $src$$Register);
11402   %}
11403   ins_pipe(pipe_class_default);
11404 %}
11405 
11406 instruct countTrailingZerosI_Ex(iRegIdst dst, iRegIsrc src) %{
11407   match(Set dst (CountTrailingZerosI src));
11408   predicate(UseCountLeadingZerosInstructionsPPC64);
11409   ins_cost(DEFAULT_COST);
11410 
11411   expand %{
11412     immI16 imm1 %{ (int)-1 %}
11413     immI16 imm2 %{ (int)32 %}
11414     immI_minus1 m1 %{ -1 %}
11415     iRegIdst tmpI1;
11416     iRegIdst tmpI2;
11417     iRegIdst tmpI3;
11418     addI_reg_imm16(tmpI1, src, imm1);
11419     andcI_reg_reg(tmpI2, src, m1, tmpI1);
11420     countLeadingZerosI(tmpI3, tmpI2);
11421     subI_imm16_reg(dst, imm2, tmpI3);
11422   %}
11423 %}
11424 
11425 instruct countTrailingZerosL_Ex(iRegIdst dst, iRegLsrc src) %{
11426   match(Set dst (CountTrailingZerosL src));
11427   predicate(UseCountLeadingZerosInstructionsPPC64);
11428   ins_cost(DEFAULT_COST);
11429 
11430   expand %{
11431     immL16 imm1 %{ (long)-1 %}
11432     immI16 imm2 %{ (int)64 %}
11433     iRegLdst tmpL1;
11434     iRegLdst tmpL2;
11435     iRegIdst tmpL3;
11436     addL_reg_imm16(tmpL1, src, imm1);
11437     andcL_reg_reg(tmpL2, tmpL1, src);
11438     countLeadingZerosL(tmpL3, tmpL2);
11439     subI_imm16_reg(dst, imm2, tmpL3);
11440  %}
11441 %}
11442 
11443 // Expand nodes for byte_reverse_int.
11444 instruct insrwi_a(iRegIdst dst, iRegIsrc src, immI16 pos, immI16 shift) %{
11445   effect(DEF dst, USE src, USE pos, USE shift);
11446   predicate(false);
11447 
11448   format %{ "INSRWI  $dst, $src, $pos, $shift" %}
11449   size(4);
11450   ins_encode %{
11451     // TODO: PPC port $archOpcode(ppc64Opcode_rlwimi);
11452     __ insrwi($dst$$Register, $src$$Register, $shift$$constant, $pos$$constant);
11453   %}
11454   ins_pipe(pipe_class_default);
11455 %}
11456 
11457 // As insrwi_a, but with USE_DEF.
11458 instruct insrwi(iRegIdst dst, iRegIsrc src, immI16 pos, immI16 shift) %{
11459   effect(USE_DEF dst, USE src, USE pos, USE shift);
11460   predicate(false);
11461 
11462   format %{ "INSRWI  $dst, $src, $pos, $shift" %}
11463   size(4);
11464   ins_encode %{
11465     // TODO: PPC port $archOpcode(ppc64Opcode_rlwimi);
11466     __ insrwi($dst$$Register, $src$$Register, $shift$$constant, $pos$$constant);
11467   %}
11468   ins_pipe(pipe_class_default);
11469 %}
11470 
11471 // Just slightly faster than java implementation.
11472 instruct bytes_reverse_int_Ex(iRegIdst dst, iRegIsrc src) %{
11473   match(Set dst (ReverseBytesI src));
11474   predicate(UseCountLeadingZerosInstructionsPPC64);
11475   ins_cost(DEFAULT_COST);
11476 
11477   expand %{
11478     immI16 imm24 %{ (int) 24 %}
11479     immI16 imm16 %{ (int) 16 %}
11480     immI16  imm8 %{ (int)  8 %}
11481     immI16  imm4 %{ (int)  4 %}
11482     immI16  imm0 %{ (int)  0 %}
11483     iRegLdst tmpI1;
11484     iRegLdst tmpI2;
11485     iRegLdst tmpI3;
11486 
11487     urShiftI_reg_imm(tmpI1, src, imm24);
11488     insrwi_a(dst, tmpI1, imm24, imm8);
11489     urShiftI_reg_imm(tmpI2, src, imm16);
11490     insrwi(dst, tmpI2, imm8, imm16);
11491     urShiftI_reg_imm(tmpI3, src, imm8);
11492     insrwi(dst, tmpI3, imm8, imm8);
11493     insrwi(dst, src, imm0, imm8);
11494   %}
11495 %}
11496 
11497 //---------- Replicate Vector Instructions ------------------------------------
11498 
11499 // Insrdi does replicate if src == dst.
11500 instruct repl32(iRegLdst dst) %{
11501   predicate(false);
11502   effect(USE_DEF dst);
11503 
11504   format %{ "INSRDI  $dst, #0, $dst, #32 \t// replicate" %}
11505   size(4);
11506   ins_encode %{
11507     // TODO: PPC port $archOpcode(ppc64Opcode_rldimi);
11508     __ insrdi($dst$$Register, $dst$$Register, 32, 0);
11509   %}
11510   ins_pipe(pipe_class_default);
11511 %}
11512 
11513 // Insrdi does replicate if src == dst.
11514 instruct repl48(iRegLdst dst) %{
11515   predicate(false);
11516   effect(USE_DEF dst);
11517 
11518   format %{ "INSRDI  $dst, #0, $dst, #48 \t// replicate" %}
11519   size(4);
11520   ins_encode %{
11521     // TODO: PPC port $archOpcode(ppc64Opcode_rldimi);
11522     __ insrdi($dst$$Register, $dst$$Register, 48, 0);
11523   %}
11524   ins_pipe(pipe_class_default);
11525 %}
11526 
11527 // Insrdi does replicate if src == dst.
11528 instruct repl56(iRegLdst dst) %{
11529   predicate(false);
11530   effect(USE_DEF dst);
11531 
11532   format %{ "INSRDI  $dst, #0, $dst, #56 \t// replicate" %}
11533   size(4);
11534   ins_encode %{
11535     // TODO: PPC port $archOpcode(ppc64Opcode_rldimi);
11536     __ insrdi($dst$$Register, $dst$$Register, 56, 0);
11537   %}
11538   ins_pipe(pipe_class_default);
11539 %}
11540 
11541 instruct repl8B_reg_Ex(iRegLdst dst, iRegIsrc src) %{
11542   match(Set dst (ReplicateB src));
11543   predicate(n->as_Vector()->length() == 8);
11544   expand %{
11545     moveReg(dst, src);
11546     repl56(dst);
11547     repl48(dst);
11548     repl32(dst);
11549   %}
11550 %}
11551 
11552 instruct repl8B_immI0(iRegLdst dst, immI_0 zero) %{
11553   match(Set dst (ReplicateB zero));
11554   predicate(n->as_Vector()->length() == 8);
11555   format %{ "LI      $dst, #0 \t// replicate8B" %}
11556   size(4);
11557   ins_encode %{
11558     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11559     __ li($dst$$Register, (int)((short)($zero$$constant & 0xFFFF)));
11560   %}
11561   ins_pipe(pipe_class_default);
11562 %}
11563 
11564 instruct repl8B_immIminus1(iRegLdst dst, immI_minus1 src) %{
11565   match(Set dst (ReplicateB src));
11566   predicate(n->as_Vector()->length() == 8);
11567   format %{ "LI      $dst, #-1 \t// replicate8B" %}
11568   size(4);
11569   ins_encode %{
11570     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11571     __ li($dst$$Register, (int)((short)($src$$constant & 0xFFFF)));
11572   %}
11573   ins_pipe(pipe_class_default);
11574 %}
11575 
11576 instruct repl4S_reg_Ex(iRegLdst dst, iRegIsrc src) %{
11577   match(Set dst (ReplicateS src));
11578   predicate(n->as_Vector()->length() == 4);
11579   expand %{
11580     moveReg(dst, src);
11581     repl48(dst);
11582     repl32(dst);
11583   %}
11584 %}
11585 
11586 instruct repl4S_immI0(iRegLdst dst, immI_0 zero) %{
11587   match(Set dst (ReplicateS zero));
11588   predicate(n->as_Vector()->length() == 4);
11589   format %{ "LI      $dst, #0 \t// replicate4C" %}
11590   size(4);
11591   ins_encode %{
11592     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11593     __ li($dst$$Register, (int)((short)($zero$$constant & 0xFFFF)));
11594   %}
11595   ins_pipe(pipe_class_default);
11596 %}
11597 
11598 instruct repl4S_immIminus1(iRegLdst dst, immI_minus1 src) %{
11599   match(Set dst (ReplicateS src));
11600   predicate(n->as_Vector()->length() == 4);
11601   format %{ "LI      $dst, -1 \t// replicate4C" %}
11602   size(4);
11603   ins_encode %{
11604     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11605     __ li($dst$$Register, (int)((short)($src$$constant & 0xFFFF)));
11606   %}
11607   ins_pipe(pipe_class_default);
11608 %}
11609 
11610 instruct repl2I_reg_Ex(iRegLdst dst, iRegIsrc src) %{
11611   match(Set dst (ReplicateI src));
11612   predicate(n->as_Vector()->length() == 2);
11613   ins_cost(2 * DEFAULT_COST);
11614   expand %{
11615     moveReg(dst, src);
11616     repl32(dst);
11617   %}
11618 %}
11619 
11620 instruct repl2I_immI0(iRegLdst dst, immI_0 zero) %{
11621   match(Set dst (ReplicateI zero));
11622   predicate(n->as_Vector()->length() == 2);
11623   format %{ "LI      $dst, #0 \t// replicate4C" %}
11624   size(4);
11625   ins_encode %{
11626     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11627     __ li($dst$$Register, (int)((short)($zero$$constant & 0xFFFF)));
11628   %}
11629   ins_pipe(pipe_class_default);
11630 %}
11631 
11632 instruct repl2I_immIminus1(iRegLdst dst, immI_minus1 src) %{
11633   match(Set dst (ReplicateI src));
11634   predicate(n->as_Vector()->length() == 2);
11635   format %{ "LI      $dst, -1 \t// replicate4C" %}
11636   size(4);
11637   ins_encode %{
11638     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11639     __ li($dst$$Register, (int)((short)($src$$constant & 0xFFFF)));
11640   %}
11641   ins_pipe(pipe_class_default);
11642 %}
11643 
11644 // Move float to int register via stack, replicate.
11645 instruct repl2F_reg_Ex(iRegLdst dst, regF src) %{
11646   match(Set dst (ReplicateF src));
11647   predicate(n->as_Vector()->length() == 2);
11648   ins_cost(2 * MEMORY_REF_COST + DEFAULT_COST);
11649   expand %{
11650     stackSlotL tmpS;
11651     iRegIdst tmpI;
11652     moveF2I_reg_stack(tmpS, src);   // Move float to stack.
11653     moveF2I_stack_reg(tmpI, tmpS);  // Move stack to int reg.
11654     moveReg(dst, tmpI);             // Move int to long reg.
11655     repl32(dst);                    // Replicate bitpattern.
11656   %}
11657 %}
11658 
11659 // Replicate scalar constant to packed float values in Double register
11660 instruct repl2F_immF_Ex(iRegLdst dst, immF src) %{
11661   match(Set dst (ReplicateF src));
11662   predicate(n->as_Vector()->length() == 2);
11663   ins_cost(5 * DEFAULT_COST);
11664 
11665   format %{ "LD      $dst, offset, $constanttablebase\t// load replicated float $src $src from table, postalloc expanded" %}
11666   postalloc_expand( postalloc_expand_load_replF_constant(dst, src, constanttablebase) );
11667 %}
11668 
11669 // Replicate scalar zero constant to packed float values in Double register
11670 instruct repl2F_immF0(iRegLdst dst, immF_0 zero) %{
11671   match(Set dst (ReplicateF zero));
11672   predicate(n->as_Vector()->length() == 2);
11673 
11674   format %{ "LI      $dst, #0 \t// replicate2F" %}
11675   ins_encode %{
11676     // TODO: PPC port $archOpcode(ppc64Opcode_addi);
11677     __ li($dst$$Register, 0x0);
11678   %}
11679   ins_pipe(pipe_class_default);
11680 %}
11681 
11682 
11683 //----------Overflow Math Instructions-----------------------------------------
11684 
11685 // Note that we have to make sure that XER.SO is reset before using overflow instructions.
11686 // Simple Overflow operations can be matched by very few instructions (e.g. addExact: xor, and_, bc).
11687 // Seems like only Long intrinsincs have an advantage. (The only expensive one is OverflowMulL.)
11688 
11689 instruct overflowAddL_reg_reg(flagsRegCR0 cr0, iRegLsrc op1, iRegLsrc op2) %{
11690   match(Set cr0 (OverflowAddL op1 op2));
11691 
11692   format %{ "add_    $op1, $op2\t# overflow check long" %}
11693   ins_encode %{
11694     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11695     __ li(R0, 0);
11696     __ mtxer(R0); // clear XER.SO
11697     __ addo_(R0, $op1$$Register, $op2$$Register);
11698   %}
11699   ins_pipe(pipe_class_default);
11700 %}
11701 
11702 instruct overflowSubL_reg_reg(flagsRegCR0 cr0, iRegLsrc op1, iRegLsrc op2) %{
11703   match(Set cr0 (OverflowSubL op1 op2));
11704 
11705   format %{ "subfo_  R0, $op2, $op1\t# overflow check long" %}
11706   ins_encode %{
11707     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11708     __ li(R0, 0);
11709     __ mtxer(R0); // clear XER.SO
11710     __ subfo_(R0, $op2$$Register, $op1$$Register);
11711   %}
11712   ins_pipe(pipe_class_default);
11713 %}
11714 
11715 instruct overflowNegL_reg(flagsRegCR0 cr0, immL_0 zero, iRegLsrc op2) %{
11716   match(Set cr0 (OverflowSubL zero op2));
11717 
11718   format %{ "nego_   R0, $op2\t# overflow check long" %}
11719   ins_encode %{
11720     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11721     __ li(R0, 0);
11722     __ mtxer(R0); // clear XER.SO
11723     __ nego_(R0, $op2$$Register);
11724   %}
11725   ins_pipe(pipe_class_default);
11726 %}
11727 
11728 instruct overflowMulL_reg_reg(flagsRegCR0 cr0, iRegLsrc op1, iRegLsrc op2) %{
11729   match(Set cr0 (OverflowMulL op1 op2));
11730 
11731   format %{ "mulldo_ R0, $op1, $op2\t# overflow check long" %}
11732   ins_encode %{
11733     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11734     __ li(R0, 0);
11735     __ mtxer(R0); // clear XER.SO
11736     __ mulldo_(R0, $op1$$Register, $op2$$Register);
11737   %}
11738   ins_pipe(pipe_class_default);
11739 %}
11740 
11741 
11742 // ============================================================================
11743 // Safepoint Instruction
11744 
11745 instruct safePoint_poll(iRegPdst poll) %{
11746   match(SafePoint poll);
11747   predicate(LoadPollAddressFromThread);
11748 
11749   // It caused problems to add the effect that r0 is killed, but this
11750   // effect no longer needs to be mentioned, since r0 is not contained
11751   // in a reg_class.
11752 
11753   format %{ "LD      R0, #0, $poll \t// Safepoint poll for GC" %}
11754   size(4);
11755   ins_encode( enc_poll(0x0, poll) );
11756   ins_pipe(pipe_class_default);
11757 %}
11758 
11759 // Safepoint without per-thread support. Load address of page to poll
11760 // as constant.
11761 // Rscratch2RegP is R12.
11762 // LoadConPollAddr node is added in pd_post_matching_hook(). It must be
11763 // a seperate node so that the oop map is at the right location.
11764 instruct safePoint_poll_conPollAddr(rscratch2RegP poll) %{
11765   match(SafePoint poll);
11766   predicate(!LoadPollAddressFromThread);
11767 
11768   // It caused problems to add the effect that r0 is killed, but this
11769   // effect no longer needs to be mentioned, since r0 is not contained
11770   // in a reg_class.
11771 
11772   format %{ "LD      R0, #0, R12 \t// Safepoint poll for GC" %}
11773   ins_encode( enc_poll(0x0, poll) );
11774   ins_pipe(pipe_class_default);
11775 %}
11776 
11777 // ============================================================================
11778 // Call Instructions
11779 
11780 // Call Java Static Instruction
11781 
11782 // Schedulable version of call static node.
11783 instruct CallStaticJavaDirect(method meth) %{
11784   match(CallStaticJava);
11785   effect(USE meth);
11786   predicate(!((CallStaticJavaNode*)n)->is_method_handle_invoke());
11787   ins_cost(CALL_COST);
11788 
11789   ins_num_consts(3 /* up to 3 patchable constants: inline cache, 2 call targets. */);
11790 
11791   format %{ "CALL,static $meth \t// ==> " %}
11792   size(4);
11793   ins_encode( enc_java_static_call(meth) );
11794   ins_pipe(pipe_class_call);
11795 %}
11796 
11797 // Schedulable version of call static node.
11798 instruct CallStaticJavaDirectHandle(method meth) %{
11799   match(CallStaticJava);
11800   effect(USE meth);
11801   predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
11802   ins_cost(CALL_COST);
11803 
11804   ins_num_consts(3 /* up to 3 patchable constants: inline cache, 2 call targets. */);
11805 
11806   format %{ "CALL,static $meth \t// ==> " %}
11807   ins_encode( enc_java_handle_call(meth) );
11808   ins_pipe(pipe_class_call);
11809 %}
11810 
11811 // Call Java Dynamic Instruction
11812 
11813 // Used by postalloc expand of CallDynamicJavaDirectSchedEx (actual call).
11814 // Loading of IC was postalloc expanded. The nodes loading the IC are reachable
11815 // via fields ins_field_load_ic_hi_node and ins_field_load_ic_node.
11816 // The call destination must still be placed in the constant pool.
11817 instruct CallDynamicJavaDirectSched(method meth) %{
11818   match(CallDynamicJava); // To get all the data fields we need ...
11819   effect(USE meth);
11820   predicate(false);       // ... but never match.
11821 
11822   ins_field_load_ic_hi_node(loadConL_hiNode*);
11823   ins_field_load_ic_node(loadConLNode*);
11824   ins_num_consts(1 /* 1 patchable constant: call destination */);
11825 
11826   format %{ "BL        \t// dynamic $meth ==> " %}
11827   size(4);
11828   ins_encode( enc_java_dynamic_call_sched(meth) );
11829   ins_pipe(pipe_class_call);
11830 %}
11831 
11832 // Schedulable (i.e. postalloc expanded) version of call dynamic java.
11833 // We use postalloc expanded calls if we use inline caches
11834 // and do not update method data.
11835 //
11836 // This instruction has two constants: inline cache (IC) and call destination.
11837 // Loading the inline cache will be postalloc expanded, thus leaving a call with
11838 // one constant.
11839 instruct CallDynamicJavaDirectSched_Ex(method meth) %{
11840   match(CallDynamicJava);
11841   effect(USE meth);
11842   predicate(UseInlineCaches);
11843   ins_cost(CALL_COST);
11844 
11845   ins_num_consts(2 /* 2 patchable constants: inline cache, call destination. */);
11846 
11847   format %{ "CALL,dynamic $meth \t// postalloc expanded" %}
11848   postalloc_expand( postalloc_expand_java_dynamic_call_sched(meth, constanttablebase) );
11849 %}
11850 
11851 // Compound version of call dynamic java
11852 // We use postalloc expanded calls if we use inline caches
11853 // and do not update method data.
11854 instruct CallDynamicJavaDirect(method meth) %{
11855   match(CallDynamicJava);
11856   effect(USE meth);
11857   predicate(!UseInlineCaches);
11858   ins_cost(CALL_COST);
11859 
11860   // Enc_java_to_runtime_call needs up to 4 constants (method data oop).
11861   ins_num_consts(4);
11862 
11863   format %{ "CALL,dynamic $meth \t// ==> " %}
11864   ins_encode( enc_java_dynamic_call(meth, constanttablebase) );
11865   ins_pipe(pipe_class_call);
11866 %}
11867 
11868 // Call Runtime Instruction
11869 
11870 instruct CallRuntimeDirect(method meth) %{
11871   match(CallRuntime);
11872   effect(USE meth);
11873   ins_cost(CALL_COST);
11874 
11875   // Enc_java_to_runtime_call needs up to 3 constants: call target,
11876   // env for callee, C-toc.
11877   ins_num_consts(3);
11878 
11879   format %{ "CALL,runtime" %}
11880   ins_encode( enc_java_to_runtime_call(meth) );
11881   ins_pipe(pipe_class_call);
11882 %}
11883 
11884 // Call Leaf
11885 
11886 // Used by postalloc expand of CallLeafDirect_Ex (mtctr).
11887 instruct CallLeafDirect_mtctr(iRegLdst dst, iRegLsrc src) %{
11888   effect(DEF dst, USE src);
11889 
11890   ins_num_consts(1);
11891 
11892   format %{ "MTCTR   $src" %}
11893   size(4);
11894   ins_encode( enc_leaf_call_mtctr(src) );
11895   ins_pipe(pipe_class_default);
11896 %}
11897 
11898 // Used by postalloc expand of CallLeafDirect_Ex (actual call).
11899 instruct CallLeafDirect(method meth) %{
11900   match(CallLeaf);   // To get the data all the data fields we need ...
11901   effect(USE meth);
11902   predicate(false);  // but never match.
11903 
11904   format %{ "BCTRL     \t// leaf call $meth ==> " %}
11905   size(4);
11906   ins_encode %{
11907     // TODO: PPC port $archOpcode(ppc64Opcode_bctrl);
11908     __ bctrl();
11909   %}
11910   ins_pipe(pipe_class_call);
11911 %}
11912 
11913 // postalloc expand of CallLeafDirect.
11914 // Load adress to call from TOC, then bl to it.
11915 instruct CallLeafDirect_Ex(method meth) %{
11916   match(CallLeaf);
11917   effect(USE meth);
11918   ins_cost(CALL_COST);
11919 
11920   // Postalloc_expand_java_to_runtime_call needs up to 3 constants: call target,
11921   // env for callee, C-toc.
11922   ins_num_consts(3);
11923 
11924   format %{ "CALL,runtime leaf $meth \t// postalloc expanded" %}
11925   postalloc_expand( postalloc_expand_java_to_runtime_call(meth, constanttablebase) );
11926 %}
11927 
11928 // Call runtime without safepoint - same as CallLeaf.
11929 // postalloc expand of CallLeafNoFPDirect.
11930 // Load adress to call from TOC, then bl to it.
11931 instruct CallLeafNoFPDirect_Ex(method meth) %{
11932   match(CallLeafNoFP);
11933   effect(USE meth);
11934   ins_cost(CALL_COST);
11935 
11936   // Enc_java_to_runtime_call needs up to 3 constants: call target,
11937   // env for callee, C-toc.
11938   ins_num_consts(3);
11939 
11940   format %{ "CALL,runtime leaf nofp $meth \t// postalloc expanded" %}
11941   postalloc_expand( postalloc_expand_java_to_runtime_call(meth, constanttablebase) );
11942 %}
11943 
11944 // Tail Call; Jump from runtime stub to Java code.
11945 // Also known as an 'interprocedural jump'.
11946 // Target of jump will eventually return to caller.
11947 // TailJump below removes the return address.
11948 instruct TailCalljmpInd(iRegPdstNoScratch jump_target, inline_cache_regP method_oop) %{
11949   match(TailCall jump_target method_oop);
11950   ins_cost(CALL_COST);
11951 
11952   format %{ "MTCTR   $jump_target \t// $method_oop holds method oop\n\t"
11953             "BCTR         \t// tail call" %}
11954   size(8);
11955   ins_encode %{
11956     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11957     __ mtctr($jump_target$$Register);
11958     __ bctr();
11959   %}
11960   ins_pipe(pipe_class_call);
11961 %}
11962 
11963 // Return Instruction
11964 instruct Ret() %{
11965   match(Return);
11966   format %{ "BLR      \t// branch to link register" %}
11967   size(4);
11968   ins_encode %{
11969     // TODO: PPC port $archOpcode(ppc64Opcode_blr);
11970     // LR is restored in MachEpilogNode. Just do the RET here.
11971     __ blr();
11972   %}
11973   ins_pipe(pipe_class_default);
11974 %}
11975 
11976 // Tail Jump; remove the return address; jump to target.
11977 // TailCall above leaves the return address around.
11978 // TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2).
11979 // ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a
11980 // "restore" before this instruction (in Epilogue), we need to materialize it
11981 // in %i0.
11982 instruct tailjmpInd(iRegPdstNoScratch jump_target, rarg1RegP ex_oop) %{
11983   match(TailJump jump_target ex_oop);
11984   ins_cost(CALL_COST);
11985 
11986   format %{ "LD      R4_ARG2 = LR\n\t"
11987             "MTCTR   $jump_target\n\t"
11988             "BCTR     \t// TailJump, exception oop: $ex_oop" %}
11989   size(12);
11990   ins_encode %{
11991     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
11992     __ ld(R4_ARG2/* issuing pc */, _abi(lr), R1_SP);
11993     __ mtctr($jump_target$$Register);
11994     __ bctr();
11995   %}
11996   ins_pipe(pipe_class_call);
11997 %}
11998 
11999 // Create exception oop: created by stack-crawling runtime code.
12000 // Created exception is now available to this handler, and is setup
12001 // just prior to jumping to this handler. No code emitted.
12002 instruct CreateException(rarg1RegP ex_oop) %{
12003   match(Set ex_oop (CreateEx));
12004   ins_cost(0);
12005 
12006   format %{ " -- \t// exception oop; no code emitted" %}
12007   size(0);
12008   ins_encode( /*empty*/ );
12009   ins_pipe(pipe_class_default);
12010 %}
12011 
12012 // Rethrow exception: The exception oop will come in the first
12013 // argument position. Then JUMP (not call) to the rethrow stub code.
12014 instruct RethrowException() %{
12015   match(Rethrow);
12016   ins_cost(CALL_COST);
12017 
12018   format %{ "Jmp     rethrow_stub" %}
12019   ins_encode %{
12020     // TODO: PPC port $archOpcode(ppc64Opcode_compound);
12021     cbuf.set_insts_mark();
12022     __ b64_patchable((address)OptoRuntime::rethrow_stub(), relocInfo::runtime_call_type);
12023   %}
12024   ins_pipe(pipe_class_call);
12025 %}
12026 
12027 // Die now.
12028 instruct ShouldNotReachHere() %{
12029   match(Halt);
12030   ins_cost(CALL_COST);
12031 
12032   format %{ "ShouldNotReachHere" %}
12033   size(4);
12034   ins_encode %{
12035     // TODO: PPC port $archOpcode(ppc64Opcode_tdi);
12036     __ trap_should_not_reach_here();
12037   %}
12038   ins_pipe(pipe_class_default);
12039 %}
12040 
12041 // This name is KNOWN by the ADLC and cannot be changed.  The ADLC
12042 // forces a 'TypeRawPtr::BOTTOM' output type for this guy.
12043 // Get a DEF on threadRegP, no costs, no encoding, use
12044 // 'ins_should_rematerialize(true)' to avoid spilling.
12045 instruct tlsLoadP(threadRegP dst) %{
12046   match(Set dst (ThreadLocal));
12047   ins_cost(0);
12048 
12049   ins_should_rematerialize(true);
12050 
12051   format %{ " -- \t// $dst=Thread::current(), empty" %}
12052   size(0);
12053   ins_encode( /*empty*/ );
12054   ins_pipe(pipe_class_empty);
12055 %}
12056 
12057 //---Some PPC specific nodes---------------------------------------------------
12058 
12059 // Stop a group.
12060 instruct endGroup() %{
12061   ins_cost(0);
12062 
12063   ins_is_nop(true);
12064 
12065   format %{ "End Bundle (ori r1, r1, 0)" %}
12066   size(4);
12067   ins_encode %{
12068     // TODO: PPC port $archOpcode(ppc64Opcode_endgroup);
12069     __ endgroup();
12070   %}
12071   ins_pipe(pipe_class_default);
12072 %}
12073 
12074 // Nop instructions
12075 
12076 instruct fxNop() %{
12077   ins_cost(0);
12078 
12079   ins_is_nop(true);
12080 
12081   format %{ "fxNop" %}
12082   size(4);
12083   ins_encode %{
12084     // TODO: PPC port $archOpcode(ppc64Opcode_fmr);
12085     __ nop();
12086   %}
12087   ins_pipe(pipe_class_default);
12088 %}
12089 
12090 instruct fpNop0() %{
12091   ins_cost(0);
12092 
12093   ins_is_nop(true);
12094 
12095   format %{ "fpNop0" %}
12096   size(4);
12097   ins_encode %{
12098     // TODO: PPC port $archOpcode(ppc64Opcode_fmr);
12099     __ fpnop0();
12100   %}
12101   ins_pipe(pipe_class_default);
12102 %}
12103 
12104 instruct fpNop1() %{
12105   ins_cost(0);
12106 
12107   ins_is_nop(true);
12108 
12109   format %{ "fpNop1" %}
12110   size(4);
12111   ins_encode %{
12112     // TODO: PPC port $archOpcode(ppc64Opcode_fmr);
12113     __ fpnop1();
12114   %}
12115   ins_pipe(pipe_class_default);
12116 %}
12117 
12118 instruct brNop0() %{
12119   ins_cost(0);
12120   size(4);
12121   format %{ "brNop0" %}
12122   ins_encode %{
12123     // TODO: PPC port $archOpcode(ppc64Opcode_mcrf);
12124     __ brnop0();
12125   %}
12126   ins_is_nop(true);
12127   ins_pipe(pipe_class_default);
12128 %}
12129 
12130 instruct brNop1() %{
12131   ins_cost(0);
12132 
12133   ins_is_nop(true);
12134 
12135   format %{ "brNop1" %}
12136   size(4);
12137   ins_encode %{
12138     // TODO: PPC port $archOpcode(ppc64Opcode_mcrf);
12139     __ brnop1();
12140   %}
12141   ins_pipe(pipe_class_default);
12142 %}
12143 
12144 instruct brNop2() %{
12145   ins_cost(0);
12146 
12147   ins_is_nop(true);
12148 
12149   format %{ "brNop2" %}
12150   size(4);
12151   ins_encode %{
12152     // TODO: PPC port $archOpcode(ppc64Opcode_mcrf);
12153     __ brnop2();
12154   %}
12155   ins_pipe(pipe_class_default);
12156 %}
12157 
12158 //----------PEEPHOLE RULES-----------------------------------------------------
12159 // These must follow all instruction definitions as they use the names
12160 // defined in the instructions definitions.
12161 //
12162 // peepmatch ( root_instr_name [preceeding_instruction]* );
12163 //
12164 // peepconstraint %{
12165 // (instruction_number.operand_name relational_op instruction_number.operand_name
12166 //  [, ...] );
12167 // // instruction numbers are zero-based using left to right order in peepmatch
12168 //
12169 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12170 // // provide an instruction_number.operand_name for each operand that appears
12171 // // in the replacement instruction's match rule
12172 //
12173 // ---------VM FLAGS---------------------------------------------------------
12174 //
12175 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12176 //
12177 // Each peephole rule is given an identifying number starting with zero and
12178 // increasing by one in the order seen by the parser. An individual peephole
12179 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12180 // on the command-line.
12181 //
12182 // ---------CURRENT LIMITATIONS----------------------------------------------
12183 //
12184 // Only match adjacent instructions in same basic block
12185 // Only equality constraints
12186 // Only constraints between operands, not (0.dest_reg == EAX_enc)
12187 // Only one replacement instruction
12188 //
12189 // ---------EXAMPLE----------------------------------------------------------
12190 //
12191 // // pertinent parts of existing instructions in architecture description
12192 // instruct movI(eRegI dst, eRegI src) %{
12193 //   match(Set dst (CopyI src));
12194 // %}
12195 //
12196 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
12197 //   match(Set dst (AddI dst src));
12198 //   effect(KILL cr);
12199 // %}
12200 //
12201 // // Change (inc mov) to lea
12202 // peephole %{
12203 //   // increment preceeded by register-register move
12204 //   peepmatch ( incI_eReg movI );
12205 //   // require that the destination register of the increment
12206 //   // match the destination register of the move
12207 //   peepconstraint ( 0.dst == 1.dst );
12208 //   // construct a replacement instruction that sets
12209 //   // the destination to ( move's source register + one )
12210 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
12211 // %}
12212 //
12213 // Implementation no longer uses movX instructions since
12214 // machine-independent system no longer uses CopyX nodes.
12215 //
12216 // peephole %{
12217 //   peepmatch ( incI_eReg movI );
12218 //   peepconstraint ( 0.dst == 1.dst );
12219 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
12220 // %}
12221 //
12222 // peephole %{
12223 //   peepmatch ( decI_eReg movI );
12224 //   peepconstraint ( 0.dst == 1.dst );
12225 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
12226 // %}
12227 //
12228 // peephole %{
12229 //   peepmatch ( addI_eReg_imm movI );
12230 //   peepconstraint ( 0.dst == 1.dst );
12231 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
12232 // %}
12233 //
12234 // peephole %{
12235 //   peepmatch ( addP_eReg_imm movP );
12236 //   peepconstraint ( 0.dst == 1.dst );
12237 //   peepreplace ( leaP_eReg_immI( 0.dst 1.src 0.src ) );
12238 // %}
12239 
12240 // // Change load of spilled value to only a spill
12241 // instruct storeI(memory mem, eRegI src) %{
12242 //   match(Set mem (StoreI mem src));
12243 // %}
12244 //
12245 // instruct loadI(eRegI dst, memory mem) %{
12246 //   match(Set dst (LoadI mem));
12247 // %}
12248 //
12249 peephole %{
12250   peepmatch ( loadI storeI );
12251   peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
12252   peepreplace ( storeI( 1.mem 1.mem 1.src ) );
12253 %}
12254 
12255 peephole %{
12256   peepmatch ( loadL storeL );
12257   peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
12258   peepreplace ( storeL( 1.mem 1.mem 1.src ) );
12259 %}
12260 
12261 peephole %{
12262   peepmatch ( loadP storeP );
12263   peepconstraint ( 1.src == 0.dst, 1.dst == 0.mem );
12264   peepreplace ( storeP( 1.dst 1.dst 1.src ) );
12265 %}
12266 
12267 //----------SMARTSPILL RULES---------------------------------------------------
12268 // These must follow all instruction definitions as they use the names
12269 // defined in the instructions definitions.