1 //
   2 // Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved.
   3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4 //
   5 // This code is free software; you can redistribute it and/or modify it
   6 // under the terms of the GNU General Public License version 2 only, as
   7 // published by the Free Software Foundation.
   8 //
   9 // This code is distributed in the hope that it will be useful, but WITHOUT
  10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11 // FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12 // version 2 for more details (a copy is included in the LICENSE file that
  13 // accompanied this code).
  14 //
  15 // You should have received a copy of the GNU General Public License version
  16 // 2 along with this work; if not, write to the Free Software Foundation,
  17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18 //
  19 // Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20 // or visit www.oracle.com if you need additional information or have any
  21 // questions.
  22 //
  23 //
  24 
  25 // SPARC Architecture Description File
  26 
  27 //----------REGISTER DEFINITION BLOCK------------------------------------------
  28 // This information is used by the matcher and the register allocator to
  29 // describe individual registers and classes of registers within the target
  30 // archtecture.
  31 register %{
  32 //----------Architecture Description Register Definitions----------------------
  33 // General Registers
  34 // "reg_def"  name ( register save type, C convention save type,
  35 //                   ideal register type, encoding, vm name );
  36 // Register Save Types:
  37 //
  38 // NS  = No-Save:       The register allocator assumes that these registers
  39 //                      can be used without saving upon entry to the method, &
  40 //                      that they do not need to be saved at call sites.
  41 //
  42 // SOC = Save-On-Call:  The register allocator assumes that these registers
  43 //                      can be used without saving upon entry to the method,
  44 //                      but that they must be saved at call sites.
  45 //
  46 // SOE = Save-On-Entry: The register allocator assumes that these registers
  47 //                      must be saved before using them upon entry to the
  48 //                      method, but they do not need to be saved at call
  49 //                      sites.
  50 //
  51 // AS  = Always-Save:   The register allocator assumes that these registers
  52 //                      must be saved before using them upon entry to the
  53 //                      method, & that they must be saved at call sites.
  54 //
  55 // Ideal Register Type is used to determine how to save & restore a
  56 // register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
  57 // spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
  58 //
  59 // The encoding number is the actual bit-pattern placed into the opcodes.
  60 
  61 
  62 // ----------------------------
  63 // Integer/Long Registers
  64 // ----------------------------
  65 
  66 // Need to expose the hi/lo aspect of 64-bit registers
  67 // This register set is used for both the 64-bit build and
  68 // the 32-bit build with 1-register longs.
  69 
  70 // Global Registers 0-7
  71 reg_def R_G0H( NS,  NS, Op_RegI,128, G0->as_VMReg()->next());
  72 reg_def R_G0 ( NS,  NS, Op_RegI,  0, G0->as_VMReg());
  73 reg_def R_G1H(SOC, SOC, Op_RegI,129, G1->as_VMReg()->next());
  74 reg_def R_G1 (SOC, SOC, Op_RegI,  1, G1->as_VMReg());
  75 reg_def R_G2H( NS,  NS, Op_RegI,130, G2->as_VMReg()->next());
  76 reg_def R_G2 ( NS,  NS, Op_RegI,  2, G2->as_VMReg());
  77 reg_def R_G3H(SOC, SOC, Op_RegI,131, G3->as_VMReg()->next());
  78 reg_def R_G3 (SOC, SOC, Op_RegI,  3, G3->as_VMReg());
  79 reg_def R_G4H(SOC, SOC, Op_RegI,132, G4->as_VMReg()->next());
  80 reg_def R_G4 (SOC, SOC, Op_RegI,  4, G4->as_VMReg());
  81 reg_def R_G5H(SOC, SOC, Op_RegI,133, G5->as_VMReg()->next());
  82 reg_def R_G5 (SOC, SOC, Op_RegI,  5, G5->as_VMReg());
  83 reg_def R_G6H( NS,  NS, Op_RegI,134, G6->as_VMReg()->next());
  84 reg_def R_G6 ( NS,  NS, Op_RegI,  6, G6->as_VMReg());
  85 reg_def R_G7H( NS,  NS, Op_RegI,135, G7->as_VMReg()->next());
  86 reg_def R_G7 ( NS,  NS, Op_RegI,  7, G7->as_VMReg());
  87 
  88 // Output Registers 0-7
  89 reg_def R_O0H(SOC, SOC, Op_RegI,136, O0->as_VMReg()->next());
  90 reg_def R_O0 (SOC, SOC, Op_RegI,  8, O0->as_VMReg());
  91 reg_def R_O1H(SOC, SOC, Op_RegI,137, O1->as_VMReg()->next());
  92 reg_def R_O1 (SOC, SOC, Op_RegI,  9, O1->as_VMReg());
  93 reg_def R_O2H(SOC, SOC, Op_RegI,138, O2->as_VMReg()->next());
  94 reg_def R_O2 (SOC, SOC, Op_RegI, 10, O2->as_VMReg());
  95 reg_def R_O3H(SOC, SOC, Op_RegI,139, O3->as_VMReg()->next());
  96 reg_def R_O3 (SOC, SOC, Op_RegI, 11, O3->as_VMReg());
  97 reg_def R_O4H(SOC, SOC, Op_RegI,140, O4->as_VMReg()->next());
  98 reg_def R_O4 (SOC, SOC, Op_RegI, 12, O4->as_VMReg());
  99 reg_def R_O5H(SOC, SOC, Op_RegI,141, O5->as_VMReg()->next());
 100 reg_def R_O5 (SOC, SOC, Op_RegI, 13, O5->as_VMReg());
 101 reg_def R_SPH( NS,  NS, Op_RegI,142, SP->as_VMReg()->next());
 102 reg_def R_SP ( NS,  NS, Op_RegI, 14, SP->as_VMReg());
 103 reg_def R_O7H(SOC, SOC, Op_RegI,143, O7->as_VMReg()->next());
 104 reg_def R_O7 (SOC, SOC, Op_RegI, 15, O7->as_VMReg());
 105 
 106 // Local Registers 0-7
 107 reg_def R_L0H( NS,  NS, Op_RegI,144, L0->as_VMReg()->next());
 108 reg_def R_L0 ( NS,  NS, Op_RegI, 16, L0->as_VMReg());
 109 reg_def R_L1H( NS,  NS, Op_RegI,145, L1->as_VMReg()->next());
 110 reg_def R_L1 ( NS,  NS, Op_RegI, 17, L1->as_VMReg());
 111 reg_def R_L2H( NS,  NS, Op_RegI,146, L2->as_VMReg()->next());
 112 reg_def R_L2 ( NS,  NS, Op_RegI, 18, L2->as_VMReg());
 113 reg_def R_L3H( NS,  NS, Op_RegI,147, L3->as_VMReg()->next());
 114 reg_def R_L3 ( NS,  NS, Op_RegI, 19, L3->as_VMReg());
 115 reg_def R_L4H( NS,  NS, Op_RegI,148, L4->as_VMReg()->next());
 116 reg_def R_L4 ( NS,  NS, Op_RegI, 20, L4->as_VMReg());
 117 reg_def R_L5H( NS,  NS, Op_RegI,149, L5->as_VMReg()->next());
 118 reg_def R_L5 ( NS,  NS, Op_RegI, 21, L5->as_VMReg());
 119 reg_def R_L6H( NS,  NS, Op_RegI,150, L6->as_VMReg()->next());
 120 reg_def R_L6 ( NS,  NS, Op_RegI, 22, L6->as_VMReg());
 121 reg_def R_L7H( NS,  NS, Op_RegI,151, L7->as_VMReg()->next());
 122 reg_def R_L7 ( NS,  NS, Op_RegI, 23, L7->as_VMReg());
 123 
 124 // Input Registers 0-7
 125 reg_def R_I0H( NS,  NS, Op_RegI,152, I0->as_VMReg()->next());
 126 reg_def R_I0 ( NS,  NS, Op_RegI, 24, I0->as_VMReg());
 127 reg_def R_I1H( NS,  NS, Op_RegI,153, I1->as_VMReg()->next());
 128 reg_def R_I1 ( NS,  NS, Op_RegI, 25, I1->as_VMReg());
 129 reg_def R_I2H( NS,  NS, Op_RegI,154, I2->as_VMReg()->next());
 130 reg_def R_I2 ( NS,  NS, Op_RegI, 26, I2->as_VMReg());
 131 reg_def R_I3H( NS,  NS, Op_RegI,155, I3->as_VMReg()->next());
 132 reg_def R_I3 ( NS,  NS, Op_RegI, 27, I3->as_VMReg());
 133 reg_def R_I4H( NS,  NS, Op_RegI,156, I4->as_VMReg()->next());
 134 reg_def R_I4 ( NS,  NS, Op_RegI, 28, I4->as_VMReg());
 135 reg_def R_I5H( NS,  NS, Op_RegI,157, I5->as_VMReg()->next());
 136 reg_def R_I5 ( NS,  NS, Op_RegI, 29, I5->as_VMReg());
 137 reg_def R_FPH( NS,  NS, Op_RegI,158, FP->as_VMReg()->next());
 138 reg_def R_FP ( NS,  NS, Op_RegI, 30, FP->as_VMReg());
 139 reg_def R_I7H( NS,  NS, Op_RegI,159, I7->as_VMReg()->next());
 140 reg_def R_I7 ( NS,  NS, Op_RegI, 31, I7->as_VMReg());
 141 
 142 // ----------------------------
 143 // Float/Double Registers
 144 // ----------------------------
 145 
 146 // Float Registers
 147 reg_def R_F0 ( SOC, SOC, Op_RegF,  0, F0->as_VMReg());
 148 reg_def R_F1 ( SOC, SOC, Op_RegF,  1, F1->as_VMReg());
 149 reg_def R_F2 ( SOC, SOC, Op_RegF,  2, F2->as_VMReg());
 150 reg_def R_F3 ( SOC, SOC, Op_RegF,  3, F3->as_VMReg());
 151 reg_def R_F4 ( SOC, SOC, Op_RegF,  4, F4->as_VMReg());
 152 reg_def R_F5 ( SOC, SOC, Op_RegF,  5, F5->as_VMReg());
 153 reg_def R_F6 ( SOC, SOC, Op_RegF,  6, F6->as_VMReg());
 154 reg_def R_F7 ( SOC, SOC, Op_RegF,  7, F7->as_VMReg());
 155 reg_def R_F8 ( SOC, SOC, Op_RegF,  8, F8->as_VMReg());
 156 reg_def R_F9 ( SOC, SOC, Op_RegF,  9, F9->as_VMReg());
 157 reg_def R_F10( SOC, SOC, Op_RegF, 10, F10->as_VMReg());
 158 reg_def R_F11( SOC, SOC, Op_RegF, 11, F11->as_VMReg());
 159 reg_def R_F12( SOC, SOC, Op_RegF, 12, F12->as_VMReg());
 160 reg_def R_F13( SOC, SOC, Op_RegF, 13, F13->as_VMReg());
 161 reg_def R_F14( SOC, SOC, Op_RegF, 14, F14->as_VMReg());
 162 reg_def R_F15( SOC, SOC, Op_RegF, 15, F15->as_VMReg());
 163 reg_def R_F16( SOC, SOC, Op_RegF, 16, F16->as_VMReg());
 164 reg_def R_F17( SOC, SOC, Op_RegF, 17, F17->as_VMReg());
 165 reg_def R_F18( SOC, SOC, Op_RegF, 18, F18->as_VMReg());
 166 reg_def R_F19( SOC, SOC, Op_RegF, 19, F19->as_VMReg());
 167 reg_def R_F20( SOC, SOC, Op_RegF, 20, F20->as_VMReg());
 168 reg_def R_F21( SOC, SOC, Op_RegF, 21, F21->as_VMReg());
 169 reg_def R_F22( SOC, SOC, Op_RegF, 22, F22->as_VMReg());
 170 reg_def R_F23( SOC, SOC, Op_RegF, 23, F23->as_VMReg());
 171 reg_def R_F24( SOC, SOC, Op_RegF, 24, F24->as_VMReg());
 172 reg_def R_F25( SOC, SOC, Op_RegF, 25, F25->as_VMReg());
 173 reg_def R_F26( SOC, SOC, Op_RegF, 26, F26->as_VMReg());
 174 reg_def R_F27( SOC, SOC, Op_RegF, 27, F27->as_VMReg());
 175 reg_def R_F28( SOC, SOC, Op_RegF, 28, F28->as_VMReg());
 176 reg_def R_F29( SOC, SOC, Op_RegF, 29, F29->as_VMReg());
 177 reg_def R_F30( SOC, SOC, Op_RegF, 30, F30->as_VMReg());
 178 reg_def R_F31( SOC, SOC, Op_RegF, 31, F31->as_VMReg());
 179 
 180 // Double Registers
 181 // The rules of ADL require that double registers be defined in pairs.
 182 // Each pair must be two 32-bit values, but not necessarily a pair of
 183 // single float registers.  In each pair, ADLC-assigned register numbers
 184 // must be adjacent, with the lower number even.  Finally, when the
 185 // CPU stores such a register pair to memory, the word associated with
 186 // the lower ADLC-assigned number must be stored to the lower address.
 187 
 188 // These definitions specify the actual bit encodings of the sparc
 189 // double fp register numbers.  FloatRegisterImpl in register_sparc.hpp
 190 // wants 0-63, so we have to convert every time we want to use fp regs
 191 // with the macroassembler, using reg_to_DoubleFloatRegister_object().
 192 // 255 is a flag meaning "don't go here".
 193 // I believe we can't handle callee-save doubles D32 and up until
 194 // the place in the sparc stack crawler that asserts on the 255 is
 195 // fixed up.
 196 reg_def R_D32 (SOC, SOC, Op_RegD,  1, F32->as_VMReg());
 197 reg_def R_D32x(SOC, SOC, Op_RegD,255, F32->as_VMReg()->next());
 198 reg_def R_D34 (SOC, SOC, Op_RegD,  3, F34->as_VMReg());
 199 reg_def R_D34x(SOC, SOC, Op_RegD,255, F34->as_VMReg()->next());
 200 reg_def R_D36 (SOC, SOC, Op_RegD,  5, F36->as_VMReg());
 201 reg_def R_D36x(SOC, SOC, Op_RegD,255, F36->as_VMReg()->next());
 202 reg_def R_D38 (SOC, SOC, Op_RegD,  7, F38->as_VMReg());
 203 reg_def R_D38x(SOC, SOC, Op_RegD,255, F38->as_VMReg()->next());
 204 reg_def R_D40 (SOC, SOC, Op_RegD,  9, F40->as_VMReg());
 205 reg_def R_D40x(SOC, SOC, Op_RegD,255, F40->as_VMReg()->next());
 206 reg_def R_D42 (SOC, SOC, Op_RegD, 11, F42->as_VMReg());
 207 reg_def R_D42x(SOC, SOC, Op_RegD,255, F42->as_VMReg()->next());
 208 reg_def R_D44 (SOC, SOC, Op_RegD, 13, F44->as_VMReg());
 209 reg_def R_D44x(SOC, SOC, Op_RegD,255, F44->as_VMReg()->next());
 210 reg_def R_D46 (SOC, SOC, Op_RegD, 15, F46->as_VMReg());
 211 reg_def R_D46x(SOC, SOC, Op_RegD,255, F46->as_VMReg()->next());
 212 reg_def R_D48 (SOC, SOC, Op_RegD, 17, F48->as_VMReg());
 213 reg_def R_D48x(SOC, SOC, Op_RegD,255, F48->as_VMReg()->next());
 214 reg_def R_D50 (SOC, SOC, Op_RegD, 19, F50->as_VMReg());
 215 reg_def R_D50x(SOC, SOC, Op_RegD,255, F50->as_VMReg()->next());
 216 reg_def R_D52 (SOC, SOC, Op_RegD, 21, F52->as_VMReg());
 217 reg_def R_D52x(SOC, SOC, Op_RegD,255, F52->as_VMReg()->next());
 218 reg_def R_D54 (SOC, SOC, Op_RegD, 23, F54->as_VMReg());
 219 reg_def R_D54x(SOC, SOC, Op_RegD,255, F54->as_VMReg()->next());
 220 reg_def R_D56 (SOC, SOC, Op_RegD, 25, F56->as_VMReg());
 221 reg_def R_D56x(SOC, SOC, Op_RegD,255, F56->as_VMReg()->next());
 222 reg_def R_D58 (SOC, SOC, Op_RegD, 27, F58->as_VMReg());
 223 reg_def R_D58x(SOC, SOC, Op_RegD,255, F58->as_VMReg()->next());
 224 reg_def R_D60 (SOC, SOC, Op_RegD, 29, F60->as_VMReg());
 225 reg_def R_D60x(SOC, SOC, Op_RegD,255, F60->as_VMReg()->next());
 226 reg_def R_D62 (SOC, SOC, Op_RegD, 31, F62->as_VMReg());
 227 reg_def R_D62x(SOC, SOC, Op_RegD,255, F62->as_VMReg()->next());
 228 
 229 
 230 // ----------------------------
 231 // Special Registers
 232 // Condition Codes Flag Registers
 233 // I tried to break out ICC and XCC but it's not very pretty.
 234 // Every Sparc instruction which defs/kills one also kills the other.
 235 // Hence every compare instruction which defs one kind of flags ends
 236 // up needing a kill of the other.
 237 reg_def CCR (SOC, SOC,  Op_RegFlags, 0, VMRegImpl::Bad());
 238 
 239 reg_def FCC0(SOC, SOC,  Op_RegFlags, 0, VMRegImpl::Bad());
 240 reg_def FCC1(SOC, SOC,  Op_RegFlags, 1, VMRegImpl::Bad());
 241 reg_def FCC2(SOC, SOC,  Op_RegFlags, 2, VMRegImpl::Bad());
 242 reg_def FCC3(SOC, SOC,  Op_RegFlags, 3, VMRegImpl::Bad());
 243 
 244 // ----------------------------
 245 // Specify the enum values for the registers.  These enums are only used by the
 246 // OptoReg "class". We can convert these enum values at will to VMReg when needed
 247 // for visibility to the rest of the vm. The order of this enum influences the
 248 // register allocator so having the freedom to set this order and not be stuck
 249 // with the order that is natural for the rest of the vm is worth it.
 250 alloc_class chunk0(
 251   R_L0,R_L0H, R_L1,R_L1H, R_L2,R_L2H, R_L3,R_L3H, R_L4,R_L4H, R_L5,R_L5H, R_L6,R_L6H, R_L7,R_L7H,
 252   R_G0,R_G0H, R_G1,R_G1H, R_G2,R_G2H, R_G3,R_G3H, R_G4,R_G4H, R_G5,R_G5H, R_G6,R_G6H, R_G7,R_G7H,
 253   R_O7,R_O7H, R_SP,R_SPH, R_O0,R_O0H, R_O1,R_O1H, R_O2,R_O2H, R_O3,R_O3H, R_O4,R_O4H, R_O5,R_O5H,
 254   R_I0,R_I0H, R_I1,R_I1H, R_I2,R_I2H, R_I3,R_I3H, R_I4,R_I4H, R_I5,R_I5H, R_FP,R_FPH, R_I7,R_I7H);
 255 
 256 // Note that a register is not allocatable unless it is also mentioned
 257 // in a widely-used reg_class below.  Thus, R_G7 and R_G0 are outside i_reg.
 258 
 259 alloc_class chunk1(
 260   // The first registers listed here are those most likely to be used
 261   // as temporaries.  We move F0..F7 away from the front of the list,
 262   // to reduce the likelihood of interferences with parameters and
 263   // return values.  Likewise, we avoid using F0/F1 for parameters,
 264   // since they are used for return values.
 265   // This FPU fine-tuning is worth about 1% on the SPEC geomean.
 266   R_F8 ,R_F9 ,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
 267   R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,
 268   R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,R_F30,R_F31,
 269   R_F0 ,R_F1 ,R_F2 ,R_F3 ,R_F4 ,R_F5 ,R_F6 ,R_F7 , // used for arguments and return values
 270   R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,
 271   R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
 272   R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,
 273   R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x);
 274 
 275 alloc_class chunk2(CCR, FCC0, FCC1, FCC2, FCC3);
 276 
 277 //----------Architecture Description Register Classes--------------------------
 278 // Several register classes are automatically defined based upon information in
 279 // this architecture description.
 280 // 1) reg_class inline_cache_reg           ( as defined in frame section )
 281 // 2) reg_class interpreter_method_oop_reg ( as defined in frame section )
 282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
 283 //
 284 
 285 // G0 is not included in integer class since it has special meaning.
 286 reg_class g0_reg(R_G0);
 287 
 288 // ----------------------------
 289 // Integer Register Classes
 290 // ----------------------------
 291 // Exclusions from i_reg:
 292 // R_G0: hardwired zero
 293 // R_G2: reserved by HotSpot to the TLS register (invariant within Java)
 294 // R_G6: reserved by Solaris ABI to tools
 295 // R_G7: reserved by Solaris ABI to libthread
 296 // R_O7: Used as a temp in many encodings
 297 reg_class int_reg(R_G1,R_G3,R_G4,R_G5,R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
 298 
 299 // Class for all integer registers, except the G registers.  This is used for
 300 // encodings which use G registers as temps.  The regular inputs to such
 301 // instructions use a "notemp_" prefix, as a hack to ensure that the allocator
 302 // will not put an input into a temp register.
 303 reg_class notemp_int_reg(R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
 304 
 305 reg_class g1_regI(R_G1);
 306 reg_class g3_regI(R_G3);
 307 reg_class g4_regI(R_G4);
 308 reg_class o0_regI(R_O0);
 309 reg_class o7_regI(R_O7);
 310 
 311 // ----------------------------
 312 // Pointer Register Classes
 313 // ----------------------------
 314 #ifdef _LP64
 315 // 64-bit build means 64-bit pointers means hi/lo pairs
 316 reg_class ptr_reg(            R_G1H,R_G1,             R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
 317                   R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
 318                   R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
 319                   R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
 320 // Lock encodings use G3 and G4 internally
 321 reg_class lock_ptr_reg(       R_G1H,R_G1,                                     R_G5H,R_G5,
 322                   R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
 323                   R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
 324                   R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
 325 // Special class for storeP instructions, which can store SP or RPC to TLS.
 326 // It is also used for memory addressing, allowing direct TLS addressing.
 327 reg_class sp_ptr_reg(         R_G1H,R_G1, R_G2H,R_G2, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
 328                   R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5, R_SPH,R_SP,
 329                   R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
 330                   R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5, R_FPH,R_FP );
 331 // R_L7 is the lowest-priority callee-save (i.e., NS) register
 332 // We use it to save R_G2 across calls out of Java.
 333 reg_class l7_regP(R_L7H,R_L7);
 334 
 335 // Other special pointer regs
 336 reg_class g1_regP(R_G1H,R_G1);
 337 reg_class g2_regP(R_G2H,R_G2);
 338 reg_class g3_regP(R_G3H,R_G3);
 339 reg_class g4_regP(R_G4H,R_G4);
 340 reg_class g5_regP(R_G5H,R_G5);
 341 reg_class i0_regP(R_I0H,R_I0);
 342 reg_class o0_regP(R_O0H,R_O0);
 343 reg_class o1_regP(R_O1H,R_O1);
 344 reg_class o2_regP(R_O2H,R_O2);
 345 reg_class o7_regP(R_O7H,R_O7);
 346 
 347 #else // _LP64
 348 // 32-bit build means 32-bit pointers means 1 register.
 349 reg_class ptr_reg(     R_G1,     R_G3,R_G4,R_G5,
 350                   R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
 351                   R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
 352                   R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
 353 // Lock encodings use G3 and G4 internally
 354 reg_class lock_ptr_reg(R_G1,               R_G5,
 355                   R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
 356                   R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
 357                   R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
 358 // Special class for storeP instructions, which can store SP or RPC to TLS.
 359 // It is also used for memory addressing, allowing direct TLS addressing.
 360 reg_class sp_ptr_reg(  R_G1,R_G2,R_G3,R_G4,R_G5,
 361                   R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_SP,
 362                   R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
 363                   R_I0,R_I1,R_I2,R_I3,R_I4,R_I5,R_FP);
 364 // R_L7 is the lowest-priority callee-save (i.e., NS) register
 365 // We use it to save R_G2 across calls out of Java.
 366 reg_class l7_regP(R_L7);
 367 
 368 // Other special pointer regs
 369 reg_class g1_regP(R_G1);
 370 reg_class g2_regP(R_G2);
 371 reg_class g3_regP(R_G3);
 372 reg_class g4_regP(R_G4);
 373 reg_class g5_regP(R_G5);
 374 reg_class i0_regP(R_I0);
 375 reg_class o0_regP(R_O0);
 376 reg_class o1_regP(R_O1);
 377 reg_class o2_regP(R_O2);
 378 reg_class o7_regP(R_O7);
 379 #endif // _LP64
 380 
 381 
 382 // ----------------------------
 383 // Long Register Classes
 384 // ----------------------------
 385 // Longs in 1 register.  Aligned adjacent hi/lo pairs.
 386 // Note:  O7 is never in this class; it is sometimes used as an encoding temp.
 387 reg_class long_reg(             R_G1H,R_G1,             R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5
 388                    ,R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5
 389 #ifdef _LP64
 390 // 64-bit, longs in 1 register: use all 64-bit integer registers
 391 // 32-bit, longs in 1 register: cannot use I's and L's.  Restrict to O's and G's.
 392                    ,R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7
 393                    ,R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5
 394 #endif // _LP64
 395                   );
 396 
 397 reg_class g1_regL(R_G1H,R_G1);
 398 reg_class g3_regL(R_G3H,R_G3);
 399 reg_class o2_regL(R_O2H,R_O2);
 400 reg_class o7_regL(R_O7H,R_O7);
 401 
 402 // ----------------------------
 403 // Special Class for Condition Code Flags Register
 404 reg_class int_flags(CCR);
 405 reg_class float_flags(FCC0,FCC1,FCC2,FCC3);
 406 reg_class float_flag0(FCC0);
 407 
 408 
 409 // ----------------------------
 410 // Float Point Register Classes
 411 // ----------------------------
 412 // Skip F30/F31, they are reserved for mem-mem copies
 413 reg_class sflt_reg(R_F0,R_F1,R_F2,R_F3,R_F4,R_F5,R_F6,R_F7,R_F8,R_F9,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
 414 
 415 // Paired floating point registers--they show up in the same order as the floats,
 416 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
 417 reg_class dflt_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
 418                    R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,
 419                    /* Use extra V9 double registers; this AD file does not support V8 */
 420                    R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
 421                    R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x
 422                    );
 423 
 424 // Paired floating point registers--they show up in the same order as the floats,
 425 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
 426 // This class is usable for mis-aligned loads as happen in I2C adapters.
 427 reg_class dflt_low_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
 428                    R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
 429 %}
 430 
 431 //----------DEFINITION BLOCK---------------------------------------------------
 432 // Define name --> value mappings to inform the ADLC of an integer valued name
 433 // Current support includes integer values in the range [0, 0x7FFFFFFF]
 434 // Format:
 435 //        int_def  <name>         ( <int_value>, <expression>);
 436 // Generated Code in ad_<arch>.hpp
 437 //        #define  <name>   (<expression>)
 438 //        // value == <int_value>
 439 // Generated code in ad_<arch>.cpp adlc_verification()
 440 //        assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
 441 //
 442 definitions %{
 443 // The default cost (of an ALU instruction).
 444   int_def DEFAULT_COST      (    100,     100);
 445   int_def HUGE_COST         (1000000, 1000000);
 446 
 447 // Memory refs are twice as expensive as run-of-the-mill.
 448   int_def MEMORY_REF_COST   (    200, DEFAULT_COST * 2);
 449 
 450 // Branches are even more expensive.
 451   int_def BRANCH_COST       (    300, DEFAULT_COST * 3);
 452   int_def CALL_COST         (    300, DEFAULT_COST * 3);
 453 %}
 454 
 455 
 456 //----------SOURCE BLOCK-------------------------------------------------------
 457 // This is a block of C++ code which provides values, functions, and
 458 // definitions necessary in the rest of the architecture description
 459 source_hpp %{
 460 // Must be visible to the DFA in dfa_sparc.cpp
 461 extern bool can_branch_register( Node *bol, Node *cmp );
 462 
 463 extern bool use_block_zeroing(Node* count);
 464 
 465 // Macros to extract hi & lo halves from a long pair.
 466 // G0 is not part of any long pair, so assert on that.
 467 // Prevents accidentally using G1 instead of G0.
 468 #define LONG_HI_REG(x) (x)
 469 #define LONG_LO_REG(x) (x)
 470 
 471 %}
 472 
 473 source %{
 474 #define __ _masm.
 475 
 476 // tertiary op of a LoadP or StoreP encoding
 477 #define REGP_OP true
 478 
 479 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding);
 480 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding);
 481 static Register reg_to_register_object(int register_encoding);
 482 
 483 // Used by the DFA in dfa_sparc.cpp.
 484 // Check for being able to use a V9 branch-on-register.  Requires a
 485 // compare-vs-zero, equal/not-equal, of a value which was zero- or sign-
 486 // extended.  Doesn't work following an integer ADD, for example, because of
 487 // overflow (-1 incremented yields 0 plus a carry in the high-order word).  On
 488 // 32-bit V9 systems, interrupts currently blow away the high-order 32 bits and
 489 // replace them with zero, which could become sign-extension in a different OS
 490 // release.  There's no obvious reason why an interrupt will ever fill these
 491 // bits with non-zero junk (the registers are reloaded with standard LD
 492 // instructions which either zero-fill or sign-fill).
 493 bool can_branch_register( Node *bol, Node *cmp ) {
 494   if( !BranchOnRegister ) return false;
 495 #ifdef _LP64
 496   if( cmp->Opcode() == Op_CmpP )
 497     return true;  // No problems with pointer compares
 498 #endif
 499   if( cmp->Opcode() == Op_CmpL )
 500     return true;  // No problems with long compares
 501 
 502   if( !SparcV9RegsHiBitsZero ) return false;
 503   if( bol->as_Bool()->_test._test != BoolTest::ne &&
 504       bol->as_Bool()->_test._test != BoolTest::eq )
 505      return false;
 506 
 507   // Check for comparing against a 'safe' value.  Any operation which
 508   // clears out the high word is safe.  Thus, loads and certain shifts
 509   // are safe, as are non-negative constants.  Any operation which
 510   // preserves zero bits in the high word is safe as long as each of its
 511   // inputs are safe.  Thus, phis and bitwise booleans are safe if their
 512   // inputs are safe.  At present, the only important case to recognize
 513   // seems to be loads.  Constants should fold away, and shifts &
 514   // logicals can use the 'cc' forms.
 515   Node *x = cmp->in(1);
 516   if( x->is_Load() ) return true;
 517   if( x->is_Phi() ) {
 518     for( uint i = 1; i < x->req(); i++ )
 519       if( !x->in(i)->is_Load() )
 520         return false;
 521     return true;
 522   }
 523   return false;
 524 }
 525 
 526 bool use_block_zeroing(Node* count) {
 527   // Use BIS for zeroing if count is not constant
 528   // or it is >= BlockZeroingLowLimit.
 529   return UseBlockZeroing && (count->find_intptr_t_con(BlockZeroingLowLimit) >= BlockZeroingLowLimit);
 530 }
 531 
 532 // ****************************************************************************
 533 
 534 // REQUIRED FUNCTIONALITY
 535 
 536 // !!!!! Special hack to get all type of calls to specify the byte offset
 537 //       from the start of the call to the point where the return address
 538 //       will point.
 539 //       The "return address" is the address of the call instruction, plus 8.
 540 
 541 int MachCallStaticJavaNode::ret_addr_offset() {
 542   int offset = NativeCall::instruction_size;  // call; delay slot
 543   if (_method_handle_invoke)
 544     offset += 4;  // restore SP
 545   return offset;
 546 }
 547 
 548 int MachCallDynamicJavaNode::ret_addr_offset() {
 549   int vtable_index = this->_vtable_index;
 550   if (vtable_index < 0) {
 551     // must be invalid_vtable_index, not nonvirtual_vtable_index
 552     assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
 553     return (NativeMovConstReg::instruction_size +
 554            NativeCall::instruction_size);  // sethi; setlo; call; delay slot
 555   } else {
 556     assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
 557     int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
 558     int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
 559     int klass_load_size;
 560     if (UseCompressedKlassPointers) {
 561       assert(Universe::heap() != NULL, "java heap should be initialized");
 562       klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
 563     } else {
 564       klass_load_size = 1*BytesPerInstWord;
 565     }
 566     if (Assembler::is_simm13(v_off)) {
 567       return klass_load_size +
 568              (2*BytesPerInstWord +           // ld_ptr, ld_ptr
 569              NativeCall::instruction_size);  // call; delay slot
 570     } else {
 571       return klass_load_size +
 572              (4*BytesPerInstWord +           // set_hi, set, ld_ptr, ld_ptr
 573              NativeCall::instruction_size);  // call; delay slot
 574     }
 575   }
 576 }
 577 
 578 int MachCallRuntimeNode::ret_addr_offset() {
 579 #ifdef _LP64
 580   if (MacroAssembler::is_far_target(entry_point())) {
 581     return NativeFarCall::instruction_size;
 582   } else {
 583     return NativeCall::instruction_size;
 584   }
 585 #else
 586   return NativeCall::instruction_size;  // call; delay slot
 587 #endif
 588 }
 589 
 590 // Indicate if the safepoint node needs the polling page as an input.
 591 // Since Sparc does not have absolute addressing, it does.
 592 bool SafePointNode::needs_polling_address_input() {
 593   return true;
 594 }
 595 
 596 // emit an interrupt that is caught by the debugger (for debugging compiler)
 597 void emit_break(CodeBuffer &cbuf) {
 598   MacroAssembler _masm(&cbuf);
 599   __ breakpoint_trap();
 600 }
 601 
 602 #ifndef PRODUCT
 603 void MachBreakpointNode::format( PhaseRegAlloc *, outputStream *st ) const {
 604   st->print("TA");
 605 }
 606 #endif
 607 
 608 void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
 609   emit_break(cbuf);
 610 }
 611 
 612 uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
 613   return MachNode::size(ra_);
 614 }
 615 
 616 // Traceable jump
 617 void  emit_jmpl(CodeBuffer &cbuf, int jump_target) {
 618   MacroAssembler _masm(&cbuf);
 619   Register rdest = reg_to_register_object(jump_target);
 620   __ JMP(rdest, 0);
 621   __ delayed()->nop();
 622 }
 623 
 624 // Traceable jump and set exception pc
 625 void  emit_jmpl_set_exception_pc(CodeBuffer &cbuf, int jump_target) {
 626   MacroAssembler _masm(&cbuf);
 627   Register rdest = reg_to_register_object(jump_target);
 628   __ JMP(rdest, 0);
 629   __ delayed()->add(O7, frame::pc_return_offset, Oissuing_pc );
 630 }
 631 
 632 void emit_nop(CodeBuffer &cbuf) {
 633   MacroAssembler _masm(&cbuf);
 634   __ nop();
 635 }
 636 
 637 void emit_illtrap(CodeBuffer &cbuf) {
 638   MacroAssembler _masm(&cbuf);
 639   __ illtrap(0);
 640 }
 641 
 642 
 643 intptr_t get_offset_from_base(const MachNode* n, const TypePtr* atype, int disp32) {
 644   assert(n->rule() != loadUB_rule, "");
 645 
 646   intptr_t offset = 0;
 647   const TypePtr *adr_type = TYPE_PTR_SENTINAL;  // Check for base==RegI, disp==immP
 648   const Node* addr = n->get_base_and_disp(offset, adr_type);
 649   assert(adr_type == (const TypePtr*)-1, "VerifyOops: no support for sparc operands with base==RegI, disp==immP");
 650   assert(addr != NULL && addr != (Node*)-1, "invalid addr");
 651   assert(addr->bottom_type()->isa_oopptr() == atype, "");
 652   atype = atype->add_offset(offset);
 653   assert(disp32 == offset, "wrong disp32");
 654   return atype->_offset;
 655 }
 656 
 657 
 658 intptr_t get_offset_from_base_2(const MachNode* n, const TypePtr* atype, int disp32) {
 659   assert(n->rule() != loadUB_rule, "");
 660 
 661   intptr_t offset = 0;
 662   Node* addr = n->in(2);
 663   assert(addr->bottom_type()->isa_oopptr() == atype, "");
 664   if (addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP) {
 665     Node* a = addr->in(2/*AddPNode::Address*/);
 666     Node* o = addr->in(3/*AddPNode::Offset*/);
 667     offset = o->is_Con() ? o->bottom_type()->is_intptr_t()->get_con() : Type::OffsetBot;
 668     atype = a->bottom_type()->is_ptr()->add_offset(offset);
 669     assert(atype->isa_oop_ptr(), "still an oop");
 670   }
 671   offset = atype->is_ptr()->_offset;
 672   if (offset != Type::OffsetBot)  offset += disp32;
 673   return offset;
 674 }
 675 
 676 static inline jdouble replicate_immI(int con, int count, int width) {
 677   // Load a constant replicated "count" times with width "width"
 678   assert(count*width == 8 && width <= 4, "sanity");
 679   int bit_width = width * 8;
 680   jlong val = con;
 681   val &= (((jlong) 1) << bit_width) - 1;  // mask off sign bits
 682   for (int i = 0; i < count - 1; i++) {
 683     val |= (val << bit_width);
 684   }
 685   jdouble dval = *((jdouble*) &val);  // coerce to double type
 686   return dval;
 687 }
 688 
 689 static inline jdouble replicate_immF(float con) {
 690   // Replicate float con 2 times and pack into vector.
 691   int val = *((int*)&con);
 692   jlong lval = val;
 693   lval = (lval << 32) | (lval & 0xFFFFFFFFl);
 694   jdouble dval = *((jdouble*) &lval);  // coerce to double type
 695   return dval;
 696 }
 697 
 698 // Standard Sparc opcode form2 field breakdown
 699 static inline void emit2_19(CodeBuffer &cbuf, int f30, int f29, int f25, int f22, int f20, int f19, int f0 ) {
 700   f0 &= (1<<19)-1;     // Mask displacement to 19 bits
 701   int op = (f30 << 30) |
 702            (f29 << 29) |
 703            (f25 << 25) |
 704            (f22 << 22) |
 705            (f20 << 20) |
 706            (f19 << 19) |
 707            (f0  <<  0);
 708   cbuf.insts()->emit_int32(op);
 709 }
 710 
 711 // Standard Sparc opcode form2 field breakdown
 712 static inline void emit2_22(CodeBuffer &cbuf, int f30, int f25, int f22, int f0 ) {
 713   f0 >>= 10;           // Drop 10 bits
 714   f0 &= (1<<22)-1;     // Mask displacement to 22 bits
 715   int op = (f30 << 30) |
 716            (f25 << 25) |
 717            (f22 << 22) |
 718            (f0  <<  0);
 719   cbuf.insts()->emit_int32(op);
 720 }
 721 
 722 // Standard Sparc opcode form3 field breakdown
 723 static inline void emit3(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int f5, int f0 ) {
 724   int op = (f30 << 30) |
 725            (f25 << 25) |
 726            (f19 << 19) |
 727            (f14 << 14) |
 728            (f5  <<  5) |
 729            (f0  <<  0);
 730   cbuf.insts()->emit_int32(op);
 731 }
 732 
 733 // Standard Sparc opcode form3 field breakdown
 734 static inline void emit3_simm13(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm13 ) {
 735   simm13 &= (1<<13)-1; // Mask to 13 bits
 736   int op = (f30 << 30) |
 737            (f25 << 25) |
 738            (f19 << 19) |
 739            (f14 << 14) |
 740            (1   << 13) | // bit to indicate immediate-mode
 741            (simm13<<0);
 742   cbuf.insts()->emit_int32(op);
 743 }
 744 
 745 static inline void emit3_simm10(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm10 ) {
 746   simm10 &= (1<<10)-1; // Mask to 10 bits
 747   emit3_simm13(cbuf,f30,f25,f19,f14,simm10);
 748 }
 749 
 750 #ifdef ASSERT
 751 // Helper function for VerifyOops in emit_form3_mem_reg
 752 void verify_oops_warning(const MachNode *n, int ideal_op, int mem_op) {
 753   warning("VerifyOops encountered unexpected instruction:");
 754   n->dump(2);
 755   warning("Instruction has ideal_Opcode==Op_%s and op_ld==Op_%s \n", NodeClassNames[ideal_op], NodeClassNames[mem_op]);
 756 }
 757 #endif
 758 
 759 
 760 void emit_form3_mem_reg(CodeBuffer &cbuf, const MachNode* n, int primary, int tertiary,
 761                         int src1_enc, int disp32, int src2_enc, int dst_enc) {
 762 
 763 #ifdef ASSERT
 764   // The following code implements the +VerifyOops feature.
 765   // It verifies oop values which are loaded into or stored out of
 766   // the current method activation.  +VerifyOops complements techniques
 767   // like ScavengeALot, because it eagerly inspects oops in transit,
 768   // as they enter or leave the stack, as opposed to ScavengeALot,
 769   // which inspects oops "at rest", in the stack or heap, at safepoints.
 770   // For this reason, +VerifyOops can sometimes detect bugs very close
 771   // to their point of creation.  It can also serve as a cross-check
 772   // on the validity of oop maps, when used toegether with ScavengeALot.
 773 
 774   // It would be good to verify oops at other points, especially
 775   // when an oop is used as a base pointer for a load or store.
 776   // This is presently difficult, because it is hard to know when
 777   // a base address is biased or not.  (If we had such information,
 778   // it would be easy and useful to make a two-argument version of
 779   // verify_oop which unbiases the base, and performs verification.)
 780 
 781   assert((uint)tertiary == 0xFFFFFFFF || tertiary == REGP_OP, "valid tertiary");
 782   bool is_verified_oop_base  = false;
 783   bool is_verified_oop_load  = false;
 784   bool is_verified_oop_store = false;
 785   int tmp_enc = -1;
 786   if (VerifyOops && src1_enc != R_SP_enc) {
 787     // classify the op, mainly for an assert check
 788     int st_op = 0, ld_op = 0;
 789     switch (primary) {
 790     case Assembler::stb_op3:  st_op = Op_StoreB; break;
 791     case Assembler::sth_op3:  st_op = Op_StoreC; break;
 792     case Assembler::stx_op3:  // may become StoreP or stay StoreI or StoreD0
 793     case Assembler::stw_op3:  st_op = Op_StoreI; break;
 794     case Assembler::std_op3:  st_op = Op_StoreL; break;
 795     case Assembler::stf_op3:  st_op = Op_StoreF; break;
 796     case Assembler::stdf_op3: st_op = Op_StoreD; break;
 797 
 798     case Assembler::ldsb_op3: ld_op = Op_LoadB; break;
 799     case Assembler::ldub_op3: ld_op = Op_LoadUB; break;
 800     case Assembler::lduh_op3: ld_op = Op_LoadUS; break;
 801     case Assembler::ldsh_op3: ld_op = Op_LoadS; break;
 802     case Assembler::ldx_op3:  // may become LoadP or stay LoadI
 803     case Assembler::ldsw_op3: // may become LoadP or stay LoadI
 804     case Assembler::lduw_op3: ld_op = Op_LoadI; break;
 805     case Assembler::ldd_op3:  ld_op = Op_LoadL; break;
 806     case Assembler::ldf_op3:  ld_op = Op_LoadF; break;
 807     case Assembler::lddf_op3: ld_op = Op_LoadD; break;
 808     case Assembler::prefetch_op3: ld_op = Op_LoadI; break;
 809 
 810     default: ShouldNotReachHere();
 811     }
 812     if (tertiary == REGP_OP) {
 813       if      (st_op == Op_StoreI)  st_op = Op_StoreP;
 814       else if (ld_op == Op_LoadI)   ld_op = Op_LoadP;
 815       else                          ShouldNotReachHere();
 816       if (st_op) {
 817         // a store
 818         // inputs are (0:control, 1:memory, 2:address, 3:value)
 819         Node* n2 = n->in(3);
 820         if (n2 != NULL) {
 821           const Type* t = n2->bottom_type();
 822           is_verified_oop_store = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
 823         }
 824       } else {
 825         // a load
 826         const Type* t = n->bottom_type();
 827         is_verified_oop_load = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
 828       }
 829     }
 830 
 831     if (ld_op) {
 832       // a Load
 833       // inputs are (0:control, 1:memory, 2:address)
 834       if (!(n->ideal_Opcode()==ld_op)       && // Following are special cases
 835           !(n->ideal_Opcode()==Op_LoadPLocked && ld_op==Op_LoadP) &&
 836           !(n->ideal_Opcode()==Op_LoadI     && ld_op==Op_LoadF) &&
 837           !(n->ideal_Opcode()==Op_LoadF     && ld_op==Op_LoadI) &&
 838           !(n->ideal_Opcode()==Op_LoadRange && ld_op==Op_LoadI) &&
 839           !(n->ideal_Opcode()==Op_LoadKlass && ld_op==Op_LoadP) &&
 840           !(n->ideal_Opcode()==Op_LoadL     && ld_op==Op_LoadI) &&
 841           !(n->ideal_Opcode()==Op_LoadL_unaligned && ld_op==Op_LoadI) &&
 842           !(n->ideal_Opcode()==Op_LoadD_unaligned && ld_op==Op_LoadF) &&
 843           !(n->ideal_Opcode()==Op_ConvI2F   && ld_op==Op_LoadF) &&
 844           !(n->ideal_Opcode()==Op_ConvI2D   && ld_op==Op_LoadF) &&
 845           !(n->ideal_Opcode()==Op_PrefetchRead  && ld_op==Op_LoadI) &&
 846           !(n->ideal_Opcode()==Op_PrefetchWrite && ld_op==Op_LoadI) &&
 847           !(n->ideal_Opcode()==Op_PrefetchAllocation && ld_op==Op_LoadI) &&
 848           !(n->ideal_Opcode()==Op_LoadVector && ld_op==Op_LoadD) &&
 849           !(n->rule() == loadUB_rule)) {
 850         verify_oops_warning(n, n->ideal_Opcode(), ld_op);
 851       }
 852     } else if (st_op) {
 853       // a Store
 854       // inputs are (0:control, 1:memory, 2:address, 3:value)
 855       if (!(n->ideal_Opcode()==st_op)    && // Following are special cases
 856           !(n->ideal_Opcode()==Op_StoreCM && st_op==Op_StoreB) &&
 857           !(n->ideal_Opcode()==Op_StoreI && st_op==Op_StoreF) &&
 858           !(n->ideal_Opcode()==Op_StoreF && st_op==Op_StoreI) &&
 859           !(n->ideal_Opcode()==Op_StoreL && st_op==Op_StoreI) &&
 860           !(n->ideal_Opcode()==Op_StoreVector && st_op==Op_StoreD) &&
 861           !(n->ideal_Opcode()==Op_StoreD && st_op==Op_StoreI && n->rule() == storeD0_rule)) {
 862         verify_oops_warning(n, n->ideal_Opcode(), st_op);
 863       }
 864     }
 865 
 866     if (src2_enc == R_G0_enc && n->rule() != loadUB_rule && n->ideal_Opcode() != Op_StoreCM ) {
 867       Node* addr = n->in(2);
 868       if (!(addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP)) {
 869         const TypeOopPtr* atype = addr->bottom_type()->isa_instptr();  // %%% oopptr?
 870         if (atype != NULL) {
 871           intptr_t offset = get_offset_from_base(n, atype, disp32);
 872           intptr_t offset_2 = get_offset_from_base_2(n, atype, disp32);
 873           if (offset != offset_2) {
 874             get_offset_from_base(n, atype, disp32);
 875             get_offset_from_base_2(n, atype, disp32);
 876           }
 877           assert(offset == offset_2, "different offsets");
 878           if (offset == disp32) {
 879             // we now know that src1 is a true oop pointer
 880             is_verified_oop_base = true;
 881             if (ld_op && src1_enc == dst_enc && ld_op != Op_LoadF && ld_op != Op_LoadD) {
 882               if( primary == Assembler::ldd_op3 ) {
 883                 is_verified_oop_base = false; // Cannot 'ldd' into O7
 884               } else {
 885                 tmp_enc = dst_enc;
 886                 dst_enc = R_O7_enc; // Load into O7; preserve source oop
 887                 assert(src1_enc != dst_enc, "");
 888               }
 889             }
 890           }
 891           if (st_op && (( offset == oopDesc::klass_offset_in_bytes())
 892                        || offset == oopDesc::mark_offset_in_bytes())) {
 893                       // loading the mark should not be allowed either, but
 894                       // we don't check this since it conflicts with InlineObjectHash
 895                       // usage of LoadINode to get the mark. We could keep the
 896                       // check if we create a new LoadMarkNode
 897             // but do not verify the object before its header is initialized
 898             ShouldNotReachHere();
 899           }
 900         }
 901       }
 902     }
 903   }
 904 #endif
 905 
 906   uint instr;
 907   instr = (Assembler::ldst_op << 30)
 908         | (dst_enc        << 25)
 909         | (primary        << 19)
 910         | (src1_enc       << 14);
 911 
 912   uint index = src2_enc;
 913   int disp = disp32;
 914 
 915   if (src1_enc == R_SP_enc || src1_enc == R_FP_enc)
 916     disp += STACK_BIAS;
 917 
 918   // We should have a compiler bailout here rather than a guarantee.
 919   // Better yet would be some mechanism to handle variable-size matches correctly.
 920   guarantee(Assembler::is_simm13(disp), "Do not match large constant offsets" );
 921 
 922   if( disp == 0 ) {
 923     // use reg-reg form
 924     // bit 13 is already zero
 925     instr |= index;
 926   } else {
 927     // use reg-imm form
 928     instr |= 0x00002000;          // set bit 13 to one
 929     instr |= disp & 0x1FFF;
 930   }
 931 
 932   cbuf.insts()->emit_int32(instr);
 933 
 934 #ifdef ASSERT
 935   {
 936     MacroAssembler _masm(&cbuf);
 937     if (is_verified_oop_base) {
 938       __ verify_oop(reg_to_register_object(src1_enc));
 939     }
 940     if (is_verified_oop_store) {
 941       __ verify_oop(reg_to_register_object(dst_enc));
 942     }
 943     if (tmp_enc != -1) {
 944       __ mov(O7, reg_to_register_object(tmp_enc));
 945     }
 946     if (is_verified_oop_load) {
 947       __ verify_oop(reg_to_register_object(dst_enc));
 948     }
 949   }
 950 #endif
 951 }
 952 
 953 void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false) {
 954   // The method which records debug information at every safepoint
 955   // expects the call to be the first instruction in the snippet as
 956   // it creates a PcDesc structure which tracks the offset of a call
 957   // from the start of the codeBlob. This offset is computed as
 958   // code_end() - code_begin() of the code which has been emitted
 959   // so far.
 960   // In this particular case we have skirted around the problem by
 961   // putting the "mov" instruction in the delay slot but the problem
 962   // may bite us again at some other point and a cleaner/generic
 963   // solution using relocations would be needed.
 964   MacroAssembler _masm(&cbuf);
 965   __ set_inst_mark();
 966 
 967   // We flush the current window just so that there is a valid stack copy
 968   // the fact that the current window becomes active again instantly is
 969   // not a problem there is nothing live in it.
 970 
 971 #ifdef ASSERT
 972   int startpos = __ offset();
 973 #endif /* ASSERT */
 974 
 975   __ call((address)entry_point, rtype);
 976 
 977   if (preserve_g2)   __ delayed()->mov(G2, L7);
 978   else __ delayed()->nop();
 979 
 980   if (preserve_g2)   __ mov(L7, G2);
 981 
 982 #ifdef ASSERT
 983   if (preserve_g2 && (VerifyCompiledCode || VerifyOops)) {
 984 #ifdef _LP64
 985     // Trash argument dump slots.
 986     __ set(0xb0b8ac0db0b8ac0d, G1);
 987     __ mov(G1, G5);
 988     __ stx(G1, SP, STACK_BIAS + 0x80);
 989     __ stx(G1, SP, STACK_BIAS + 0x88);
 990     __ stx(G1, SP, STACK_BIAS + 0x90);
 991     __ stx(G1, SP, STACK_BIAS + 0x98);
 992     __ stx(G1, SP, STACK_BIAS + 0xA0);
 993     __ stx(G1, SP, STACK_BIAS + 0xA8);
 994 #else // _LP64
 995     // this is also a native call, so smash the first 7 stack locations,
 996     // and the various registers
 997 
 998     // Note:  [SP+0x40] is sp[callee_aggregate_return_pointer_sp_offset],
 999     // while [SP+0x44..0x58] are the argument dump slots.
1000     __ set((intptr_t)0xbaadf00d, G1);
1001     __ mov(G1, G5);
1002     __ sllx(G1, 32, G1);
1003     __ or3(G1, G5, G1);
1004     __ mov(G1, G5);
1005     __ stx(G1, SP, 0x40);
1006     __ stx(G1, SP, 0x48);
1007     __ stx(G1, SP, 0x50);
1008     __ stw(G1, SP, 0x58); // Do not trash [SP+0x5C] which is a usable spill slot
1009 #endif // _LP64
1010   }
1011 #endif /*ASSERT*/
1012 }
1013 
1014 //=============================================================================
1015 // REQUIRED FUNCTIONALITY for encoding
1016 void emit_lo(CodeBuffer &cbuf, int val) {  }
1017 void emit_hi(CodeBuffer &cbuf, int val) {  }
1018 
1019 
1020 //=============================================================================
1021 const RegMask& MachConstantBaseNode::_out_RegMask = PTR_REG_mask();
1022 
1023 int Compile::ConstantTable::calculate_table_base_offset() const {
1024   if (UseRDPCForConstantTableBase) {
1025     // The table base offset might be less but then it fits into
1026     // simm13 anyway and we are good (cf. MachConstantBaseNode::emit).
1027     return Assembler::min_simm13();
1028   } else {
1029     int offset = -(size() / 2);
1030     if (!Assembler::is_simm13(offset)) {
1031       offset = Assembler::min_simm13();
1032     }
1033     return offset;
1034   }
1035 }
1036 
1037 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
1038   Compile* C = ra_->C;
1039   Compile::ConstantTable& constant_table = C->constant_table();
1040   MacroAssembler _masm(&cbuf);
1041 
1042   Register r = as_Register(ra_->get_encode(this));
1043   CodeSection* consts_section = __ code()->consts();
1044   int consts_size = consts_section->align_at_start(consts_section->size());
1045   assert(constant_table.size() == consts_size, err_msg("must be: %d == %d", constant_table.size(), consts_size));
1046 
1047   if (UseRDPCForConstantTableBase) {
1048     // For the following RDPC logic to work correctly the consts
1049     // section must be allocated right before the insts section.  This
1050     // assert checks for that.  The layout and the SECT_* constants
1051     // are defined in src/share/vm/asm/codeBuffer.hpp.
1052     assert(CodeBuffer::SECT_CONSTS + 1 == CodeBuffer::SECT_INSTS, "must be");
1053     int insts_offset = __ offset();
1054 
1055     // Layout:
1056     //
1057     // |----------- consts section ------------|----------- insts section -----------...
1058     // |------ constant table -----|- padding -|------------------x----
1059     //                                                            \ current PC (RDPC instruction)
1060     // |<------------- consts_size ----------->|<- insts_offset ->|
1061     //                                                            \ table base
1062     // The table base offset is later added to the load displacement
1063     // so it has to be negative.
1064     int table_base_offset = -(consts_size + insts_offset);
1065     int disp;
1066 
1067     // If the displacement from the current PC to the constant table
1068     // base fits into simm13 we set the constant table base to the
1069     // current PC.
1070     if (Assembler::is_simm13(table_base_offset)) {
1071       constant_table.set_table_base_offset(table_base_offset);
1072       disp = 0;
1073     } else {
1074       // Otherwise we set the constant table base offset to the
1075       // maximum negative displacement of load instructions to keep
1076       // the disp as small as possible:
1077       //
1078       // |<------------- consts_size ----------->|<- insts_offset ->|
1079       // |<--------- min_simm13 --------->|<-------- disp --------->|
1080       //                                  \ table base
1081       table_base_offset = Assembler::min_simm13();
1082       constant_table.set_table_base_offset(table_base_offset);
1083       disp = (consts_size + insts_offset) + table_base_offset;
1084     }
1085 
1086     __ rdpc(r);
1087 
1088     if (disp != 0) {
1089       assert(r != O7, "need temporary");
1090       __ sub(r, __ ensure_simm13_or_reg(disp, O7), r);
1091     }
1092   }
1093   else {
1094     // Materialize the constant table base.
1095     address baseaddr = consts_section->start() + -(constant_table.table_base_offset());
1096     RelocationHolder rspec = internal_word_Relocation::spec(baseaddr);
1097     AddressLiteral base(baseaddr, rspec);
1098     __ set(base, r);
1099   }
1100 }
1101 
1102 uint MachConstantBaseNode::size(PhaseRegAlloc*) const {
1103   if (UseRDPCForConstantTableBase) {
1104     // This is really the worst case but generally it's only 1 instruction.
1105     return (1 /*rdpc*/ + 1 /*sub*/ + MacroAssembler::worst_case_insts_for_set()) * BytesPerInstWord;
1106   } else {
1107     return MacroAssembler::worst_case_insts_for_set() * BytesPerInstWord;
1108   }
1109 }
1110 
1111 #ifndef PRODUCT
1112 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
1113   char reg[128];
1114   ra_->dump_register(this, reg);
1115   if (UseRDPCForConstantTableBase) {
1116     st->print("RDPC   %s\t! constant table base", reg);
1117   } else {
1118     st->print("SET    &constanttable,%s\t! constant table base", reg);
1119   }
1120 }
1121 #endif
1122 
1123 
1124 //=============================================================================
1125 
1126 #ifndef PRODUCT
1127 void MachPrologNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1128   Compile* C = ra_->C;
1129 
1130   for (int i = 0; i < OptoPrologueNops; i++) {
1131     st->print_cr("NOP"); st->print("\t");
1132   }
1133 
1134   if( VerifyThread ) {
1135     st->print_cr("Verify_Thread"); st->print("\t");
1136   }
1137 
1138   size_t framesize = C->frame_slots() << LogBytesPerInt;
1139 
1140   // Calls to C2R adapters often do not accept exceptional returns.
1141   // We require that their callers must bang for them.  But be careful, because
1142   // some VM calls (such as call site linkage) can use several kilobytes of
1143   // stack.  But the stack safety zone should account for that.
1144   // See bugs 4446381, 4468289, 4497237.
1145   if (C->need_stack_bang(framesize)) {
1146     st->print_cr("! stack bang"); st->print("\t");
1147   }
1148 
1149   if (Assembler::is_simm13(-framesize)) {
1150     st->print   ("SAVE   R_SP,-%d,R_SP",framesize);
1151   } else {
1152     st->print_cr("SETHI  R_SP,hi%%(-%d),R_G3",framesize); st->print("\t");
1153     st->print_cr("ADD    R_G3,lo%%(-%d),R_G3",framesize); st->print("\t");
1154     st->print   ("SAVE   R_SP,R_G3,R_SP");
1155   }
1156 
1157 }
1158 #endif
1159 
1160 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1161   Compile* C = ra_->C;
1162   MacroAssembler _masm(&cbuf);
1163 
1164   for (int i = 0; i < OptoPrologueNops; i++) {
1165     __ nop();
1166   }
1167 
1168   __ verify_thread();
1169 
1170   size_t framesize = C->frame_slots() << LogBytesPerInt;
1171   assert(framesize >= 16*wordSize, "must have room for reg. save area");
1172   assert(framesize%(2*wordSize) == 0, "must preserve 2*wordSize alignment");
1173 
1174   // Calls to C2R adapters often do not accept exceptional returns.
1175   // We require that their callers must bang for them.  But be careful, because
1176   // some VM calls (such as call site linkage) can use several kilobytes of
1177   // stack.  But the stack safety zone should account for that.
1178   // See bugs 4446381, 4468289, 4497237.
1179   if (C->need_stack_bang(framesize)) {
1180     __ generate_stack_overflow_check(framesize);
1181   }
1182 
1183   if (Assembler::is_simm13(-framesize)) {
1184     __ save(SP, -framesize, SP);
1185   } else {
1186     __ sethi(-framesize & ~0x3ff, G3);
1187     __ add(G3, -framesize & 0x3ff, G3);
1188     __ save(SP, G3, SP);
1189   }
1190   C->set_frame_complete( __ offset() );
1191 
1192   if (!UseRDPCForConstantTableBase && C->has_mach_constant_base_node()) {
1193     // NOTE: We set the table base offset here because users might be
1194     // emitted before MachConstantBaseNode.
1195     Compile::ConstantTable& constant_table = C->constant_table();
1196     constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
1197   }
1198 }
1199 
1200 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
1201   return MachNode::size(ra_);
1202 }
1203 
1204 int MachPrologNode::reloc() const {
1205   return 10; // a large enough number
1206 }
1207 
1208 //=============================================================================
1209 #ifndef PRODUCT
1210 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1211   Compile* C = ra_->C;
1212 
1213   if( do_polling() && ra_->C->is_method_compilation() ) {
1214     st->print("SETHI  #PollAddr,L0\t! Load Polling address\n\t");
1215 #ifdef _LP64
1216     st->print("LDX    [L0],G0\t!Poll for Safepointing\n\t");
1217 #else
1218     st->print("LDUW   [L0],G0\t!Poll for Safepointing\n\t");
1219 #endif
1220   }
1221 
1222   if( do_polling() )
1223     st->print("RET\n\t");
1224 
1225   st->print("RESTORE");
1226 }
1227 #endif
1228 
1229 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1230   MacroAssembler _masm(&cbuf);
1231   Compile* C = ra_->C;
1232 
1233   __ verify_thread();
1234 
1235   // If this does safepoint polling, then do it here
1236   if( do_polling() && ra_->C->is_method_compilation() ) {
1237     AddressLiteral polling_page(os::get_polling_page());
1238     __ sethi(polling_page, L0);
1239     __ relocate(relocInfo::poll_return_type);
1240     __ ld_ptr( L0, 0, G0 );
1241   }
1242 
1243   // If this is a return, then stuff the restore in the delay slot
1244   if( do_polling() ) {
1245     __ ret();
1246     __ delayed()->restore();
1247   } else {
1248     __ restore();
1249   }
1250 }
1251 
1252 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
1253   return MachNode::size(ra_);
1254 }
1255 
1256 int MachEpilogNode::reloc() const {
1257   return 16; // a large enough number
1258 }
1259 
1260 const Pipeline * MachEpilogNode::pipeline() const {
1261   return MachNode::pipeline_class();
1262 }
1263 
1264 int MachEpilogNode::safepoint_offset() const {
1265   assert( do_polling(), "no return for this epilog node");
1266   return MacroAssembler::insts_for_sethi(os::get_polling_page()) * BytesPerInstWord;
1267 }
1268 
1269 //=============================================================================
1270 
1271 // Figure out which register class each belongs in: rc_int, rc_float, rc_stack
1272 enum RC { rc_bad, rc_int, rc_float, rc_stack };
1273 static enum RC rc_class( OptoReg::Name reg ) {
1274   if( !OptoReg::is_valid(reg)  ) return rc_bad;
1275   if (OptoReg::is_stack(reg)) return rc_stack;
1276   VMReg r = OptoReg::as_VMReg(reg);
1277   if (r->is_Register()) return rc_int;
1278   assert(r->is_FloatRegister(), "must be");
1279   return rc_float;
1280 }
1281 
1282 static int impl_helper( const MachNode *mach, CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, bool is_load, int offset, int reg, int opcode, const char *op_str, int size, outputStream* st ) {
1283   if( cbuf ) {
1284     // Better yet would be some mechanism to handle variable-size matches correctly
1285     if (!Assembler::is_simm13(offset + STACK_BIAS)) {
1286       ra_->C->record_method_not_compilable("unable to handle large constant offsets");
1287     } else {
1288       emit_form3_mem_reg(*cbuf, mach, opcode, -1, R_SP_enc, offset, 0, Matcher::_regEncode[reg]);
1289     }
1290   }
1291 #ifndef PRODUCT
1292   else if( !do_size ) {
1293     if( size != 0 ) st->print("\n\t");
1294     if( is_load ) st->print("%s   [R_SP + #%d],R_%s\t! spill",op_str,offset,OptoReg::regname(reg));
1295     else          st->print("%s   R_%s,[R_SP + #%d]\t! spill",op_str,OptoReg::regname(reg),offset);
1296   }
1297 #endif
1298   return size+4;
1299 }
1300 
1301 static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int op1, int op2, const char *op_str, int size, outputStream* st ) {
1302   if( cbuf ) emit3( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst], op1, 0, op2, Matcher::_regEncode[src] );
1303 #ifndef PRODUCT
1304   else if( !do_size ) {
1305     if( size != 0 ) st->print("\n\t");
1306     st->print("%s  R_%s,R_%s\t! spill",op_str,OptoReg::regname(src),OptoReg::regname(dst));
1307   }
1308 #endif
1309   return size+4;
1310 }
1311 
1312 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf,
1313                                         PhaseRegAlloc *ra_,
1314                                         bool do_size,
1315                                         outputStream* st ) const {
1316   // Get registers to move
1317   OptoReg::Name src_second = ra_->get_reg_second(in(1));
1318   OptoReg::Name src_first = ra_->get_reg_first(in(1));
1319   OptoReg::Name dst_second = ra_->get_reg_second(this );
1320   OptoReg::Name dst_first = ra_->get_reg_first(this );
1321 
1322   enum RC src_second_rc = rc_class(src_second);
1323   enum RC src_first_rc = rc_class(src_first);
1324   enum RC dst_second_rc = rc_class(dst_second);
1325   enum RC dst_first_rc = rc_class(dst_first);
1326 
1327   assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );
1328 
1329   // Generate spill code!
1330   int size = 0;
1331 
1332   if( src_first == dst_first && src_second == dst_second )
1333     return size;            // Self copy, no move
1334 
1335   // --------------------------------------
1336   // Check for mem-mem move.  Load into unused float registers and fall into
1337   // the float-store case.
1338   if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
1339     int offset = ra_->reg2offset(src_first);
1340     // Further check for aligned-adjacent pair, so we can use a double load
1341     if( (src_first&1)==0 && src_first+1 == src_second ) {
1342       src_second    = OptoReg::Name(R_F31_num);
1343       src_second_rc = rc_float;
1344       size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::lddf_op3,"LDDF",size, st);
1345     } else {
1346       size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::ldf_op3 ,"LDF ",size, st);
1347     }
1348     src_first    = OptoReg::Name(R_F30_num);
1349     src_first_rc = rc_float;
1350   }
1351 
1352   if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) {
1353     int offset = ra_->reg2offset(src_second);
1354     size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F31_num,Assembler::ldf_op3,"LDF ",size, st);
1355     src_second    = OptoReg::Name(R_F31_num);
1356     src_second_rc = rc_float;
1357   }
1358 
1359   // --------------------------------------
1360   // Check for float->int copy; requires a trip through memory
1361   if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS < 3) {
1362     int offset = frame::register_save_words*wordSize;
1363     if (cbuf) {
1364       emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::sub_op3, R_SP_enc, 16 );
1365       impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1366       impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1367       emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::add_op3, R_SP_enc, 16 );
1368     }
1369 #ifndef PRODUCT
1370     else if (!do_size) {
1371       if (size != 0) st->print("\n\t");
1372       st->print(  "SUB    R_SP,16,R_SP\n");
1373       impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1374       impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1375       st->print("\tADD    R_SP,16,R_SP\n");
1376     }
1377 #endif
1378     size += 16;
1379   }
1380 
1381   // Check for float->int copy on T4
1382   if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS >= 3) {
1383     // Further check for aligned-adjacent pair, so we can use a double move
1384     if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1385       return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mdtox_opf,"MOVDTOX",size, st);
1386     size  =  impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mstouw_opf,"MOVSTOUW",size, st);
1387   }
1388   // Check for int->float copy on T4
1389   if (src_first_rc == rc_int && dst_first_rc == rc_float && UseVIS >= 3) {
1390     // Further check for aligned-adjacent pair, so we can use a double move
1391     if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1392       return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mxtod_opf,"MOVXTOD",size, st);
1393     size  =  impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mwtos_opf,"MOVWTOS",size, st);
1394   }
1395 
1396   // --------------------------------------
1397   // In the 32-bit 1-reg-longs build ONLY, I see mis-aligned long destinations.
1398   // In such cases, I have to do the big-endian swap.  For aligned targets, the
1399   // hardware does the flop for me.  Doubles are always aligned, so no problem
1400   // there.  Misaligned sources only come from native-long-returns (handled
1401   // special below).
1402 #ifndef _LP64
1403   if( src_first_rc == rc_int &&     // source is already big-endian
1404       src_second_rc != rc_bad &&    // 64-bit move
1405       ((dst_first&1)!=0 || dst_second != dst_first+1) ) { // misaligned dst
1406     assert( (src_first&1)==0 && src_second == src_first+1, "source must be aligned" );
1407     // Do the big-endian flop.
1408     OptoReg::Name tmp    = dst_first   ; dst_first    = dst_second   ; dst_second    = tmp   ;
1409     enum RC       tmp_rc = dst_first_rc; dst_first_rc = dst_second_rc; dst_second_rc = tmp_rc;
1410   }
1411 #endif
1412 
1413   // --------------------------------------
1414   // Check for integer reg-reg copy
1415   if( src_first_rc == rc_int && dst_first_rc == rc_int ) {
1416 #ifndef _LP64
1417     if( src_first == R_O0_num && src_second == R_O1_num ) {  // Check for the evil O0/O1 native long-return case
1418       // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1419       //       as stored in memory.  On a big-endian machine like SPARC, this means that the _second
1420       //       operand contains the least significant word of the 64-bit value and vice versa.
1421       OptoReg::Name tmp = OptoReg::Name(R_O7_num);
1422       assert( (dst_first&1)==0 && dst_second == dst_first+1, "return a native O0/O1 long to an aligned-adjacent 64-bit reg" );
1423       // Shift O0 left in-place, zero-extend O1, then OR them into the dst
1424       if( cbuf ) {
1425         emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tmp], Assembler::sllx_op3, Matcher::_regEncode[src_first], 0x1020 );
1426         emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[src_second], Assembler::srl_op3, Matcher::_regEncode[src_second], 0x0000 );
1427         emit3       ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler:: or_op3, Matcher::_regEncode[tmp], 0, Matcher::_regEncode[src_second] );
1428 #ifndef PRODUCT
1429       } else if( !do_size ) {
1430         if( size != 0 ) st->print("\n\t");
1431         st->print("SLLX   R_%s,32,R_%s\t! Move O0-first to O7-high\n\t", OptoReg::regname(src_first), OptoReg::regname(tmp));
1432         st->print("SRL    R_%s, 0,R_%s\t! Zero-extend O1\n\t", OptoReg::regname(src_second), OptoReg::regname(src_second));
1433         st->print("OR     R_%s,R_%s,R_%s\t! spill",OptoReg::regname(tmp), OptoReg::regname(src_second), OptoReg::regname(dst_first));
1434 #endif
1435       }
1436       return size+12;
1437     }
1438     else if( dst_first == R_I0_num && dst_second == R_I1_num ) {
1439       // returning a long value in I0/I1
1440       // a SpillCopy must be able to target a return instruction's reg_class
1441       // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1442       //       as stored in memory.  On a big-endian machine like SPARC, this means that the _second
1443       //       operand contains the least significant word of the 64-bit value and vice versa.
1444       OptoReg::Name tdest = dst_first;
1445 
1446       if (src_first == dst_first) {
1447         tdest = OptoReg::Name(R_O7_num);
1448         size += 4;
1449       }
1450 
1451       if( cbuf ) {
1452         assert( (src_first&1) == 0 && (src_first+1) == src_second, "return value was in an aligned-adjacent 64-bit reg");
1453         // Shift value in upper 32-bits of src to lower 32-bits of I0; move lower 32-bits to I1
1454         // ShrL_reg_imm6
1455         emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tdest], Assembler::srlx_op3, Matcher::_regEncode[src_second], 32 | 0x1000 );
1456         // ShrR_reg_imm6  src, 0, dst
1457         emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srl_op3, Matcher::_regEncode[src_first], 0x0000 );
1458         if (tdest != dst_first) {
1459           emit3     ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler::or_op3, 0/*G0*/, 0/*op2*/, Matcher::_regEncode[tdest] );
1460         }
1461       }
1462 #ifndef PRODUCT
1463       else if( !do_size ) {
1464         if( size != 0 ) st->print("\n\t");  // %%%%% !!!!!
1465         st->print("SRLX   R_%s,32,R_%s\t! Extract MSW\n\t",OptoReg::regname(src_second),OptoReg::regname(tdest));
1466         st->print("SRL    R_%s, 0,R_%s\t! Extract LSW\n\t",OptoReg::regname(src_first),OptoReg::regname(dst_second));
1467         if (tdest != dst_first) {
1468           st->print("MOV    R_%s,R_%s\t! spill\n\t", OptoReg::regname(tdest), OptoReg::regname(dst_first));
1469         }
1470       }
1471 #endif // PRODUCT
1472       return size+8;
1473     }
1474 #endif // !_LP64
1475     // Else normal reg-reg copy
1476     assert( src_second != dst_first, "smashed second before evacuating it" );
1477     size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::or_op3,0,"MOV  ",size, st);
1478     assert( (src_first&1) == 0 && (dst_first&1) == 0, "never move second-halves of int registers" );
1479     // This moves an aligned adjacent pair.
1480     // See if we are done.
1481     if( src_first+1 == src_second && dst_first+1 == dst_second )
1482       return size;
1483   }
1484 
1485   // Check for integer store
1486   if( src_first_rc == rc_int && dst_first_rc == rc_stack ) {
1487     int offset = ra_->reg2offset(dst_first);
1488     // Further check for aligned-adjacent pair, so we can use a double store
1489     if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1490       return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stx_op3,"STX ",size, st);
1491     size  =  impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stw_op3,"STW ",size, st);
1492   }
1493 
1494   // Check for integer load
1495   if( dst_first_rc == rc_int && src_first_rc == rc_stack ) {
1496     int offset = ra_->reg2offset(src_first);
1497     // Further check for aligned-adjacent pair, so we can use a double load
1498     if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1499       return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldx_op3 ,"LDX ",size, st);
1500     size  =  impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1501   }
1502 
1503   // Check for float reg-reg copy
1504   if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
1505     // Further check for aligned-adjacent pair, so we can use a double move
1506     if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1507       return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovd_opf,"FMOVD",size, st);
1508     size  =  impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovs_opf,"FMOVS",size, st);
1509   }
1510 
1511   // Check for float store
1512   if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
1513     int offset = ra_->reg2offset(dst_first);
1514     // Further check for aligned-adjacent pair, so we can use a double store
1515     if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1516       return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stdf_op3,"STDF",size, st);
1517     size  =  impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1518   }
1519 
1520   // Check for float load
1521   if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
1522     int offset = ra_->reg2offset(src_first);
1523     // Further check for aligned-adjacent pair, so we can use a double load
1524     if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1525       return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lddf_op3,"LDDF",size, st);
1526     size  =  impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldf_op3 ,"LDF ",size, st);
1527   }
1528 
1529   // --------------------------------------------------------------------
1530   // Check for hi bits still needing moving.  Only happens for misaligned
1531   // arguments to native calls.
1532   if( src_second == dst_second )
1533     return size;               // Self copy; no move
1534   assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );
1535 
1536 #ifndef _LP64
1537   // In the LP64 build, all registers can be moved as aligned/adjacent
1538   // pairs, so there's never any need to move the high bits separately.
1539   // The 32-bit builds have to deal with the 32-bit ABI which can force
1540   // all sorts of silly alignment problems.
1541 
1542   // Check for integer reg-reg copy.  Hi bits are stuck up in the top
1543   // 32-bits of a 64-bit register, but are needed in low bits of another
1544   // register (else it's a hi-bits-to-hi-bits copy which should have
1545   // happened already as part of a 64-bit move)
1546   if( src_second_rc == rc_int && dst_second_rc == rc_int ) {
1547     assert( (src_second&1)==1, "its the evil O0/O1 native return case" );
1548     assert( (dst_second&1)==0, "should have moved with 1 64-bit move" );
1549     // Shift src_second down to dst_second's low bits.
1550     if( cbuf ) {
1551       emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1552 #ifndef PRODUCT
1553     } else if( !do_size ) {
1554       if( size != 0 ) st->print("\n\t");
1555       st->print("SRLX   R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(dst_second));
1556 #endif
1557     }
1558     return size+4;
1559   }
1560 
1561   // Check for high word integer store.  Must down-shift the hi bits
1562   // into a temp register, then fall into the case of storing int bits.
1563   if( src_second_rc == rc_int && dst_second_rc == rc_stack && (src_second&1)==1 ) {
1564     // Shift src_second down to dst_second's low bits.
1565     if( cbuf ) {
1566       emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[R_O7_num], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1567 #ifndef PRODUCT
1568     } else if( !do_size ) {
1569       if( size != 0 ) st->print("\n\t");
1570       st->print("SRLX   R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(R_O7_num));
1571 #endif
1572     }
1573     size+=4;
1574     src_second = OptoReg::Name(R_O7_num); // Not R_O7H_num!
1575   }
1576 
1577   // Check for high word integer load
1578   if( dst_second_rc == rc_int && src_second_rc == rc_stack )
1579     return impl_helper(this,cbuf,ra_,do_size,true ,ra_->reg2offset(src_second),dst_second,Assembler::lduw_op3,"LDUW",size, st);
1580 
1581   // Check for high word integer store
1582   if( src_second_rc == rc_int && dst_second_rc == rc_stack )
1583     return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stw_op3 ,"STW ",size, st);
1584 
1585   // Check for high word float store
1586   if( src_second_rc == rc_float && dst_second_rc == rc_stack )
1587     return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stf_op3 ,"STF ",size, st);
1588 
1589 #endif // !_LP64
1590 
1591   Unimplemented();
1592 }
1593 
1594 #ifndef PRODUCT
1595 void MachSpillCopyNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1596   implementation( NULL, ra_, false, st );
1597 }
1598 #endif
1599 
1600 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1601   implementation( &cbuf, ra_, false, NULL );
1602 }
1603 
1604 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1605   return implementation( NULL, ra_, true, NULL );
1606 }
1607 
1608 //=============================================================================
1609 #ifndef PRODUCT
1610 void MachNopNode::format( PhaseRegAlloc *, outputStream *st ) const {
1611   st->print("NOP \t# %d bytes pad for loops and calls", 4 * _count);
1612 }
1613 #endif
1614 
1615 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ) const {
1616   MacroAssembler _masm(&cbuf);
1617   for(int i = 0; i < _count; i += 1) {
1618     __ nop();
1619   }
1620 }
1621 
1622 uint MachNopNode::size(PhaseRegAlloc *ra_) const {
1623   return 4 * _count;
1624 }
1625 
1626 
1627 //=============================================================================
1628 #ifndef PRODUCT
1629 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1630   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1631   int reg = ra_->get_reg_first(this);
1632   st->print("LEA    [R_SP+#%d+BIAS],%s",offset,Matcher::regName[reg]);
1633 }
1634 #endif
1635 
1636 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1637   MacroAssembler _masm(&cbuf);
1638   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()) + STACK_BIAS;
1639   int reg = ra_->get_encode(this);
1640 
1641   if (Assembler::is_simm13(offset)) {
1642      __ add(SP, offset, reg_to_register_object(reg));
1643   } else {
1644      __ set(offset, O7);
1645      __ add(SP, O7, reg_to_register_object(reg));
1646   }
1647 }
1648 
1649 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1650   // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_)
1651   assert(ra_ == ra_->C->regalloc(), "sanity");
1652   return ra_->C->scratch_emit_size(this);
1653 }
1654 
1655 //=============================================================================
1656 #ifndef PRODUCT
1657 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1658   st->print_cr("\nUEP:");
1659 #ifdef    _LP64
1660   if (UseCompressedKlassPointers) {
1661     assert(Universe::heap() != NULL, "java heap should be initialized");
1662     st->print_cr("\tLDUW   [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass");
1663     st->print_cr("\tSET    Universe::narrow_klass_base,R_G6_heap_base");
1664     if (Universe::narrow_klass_shift() != 0) {
1665       st->print_cr("\tSLL    R_G5,3,R_G5");
1666     }
1667     st->print_cr("\tADD    R_G5,R_G6_heap_base,R_G5");
1668     st->print_cr("\tSET    Universe::narrow_ptrs_base,R_G6_heap_base");
1669   } else {
1670     st->print_cr("\tLDX    [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1671   }
1672   st->print_cr("\tCMP    R_G5,R_G3" );
1673   st->print   ("\tTne    xcc,R_G0+ST_RESERVED_FOR_USER_0+2");
1674 #else  // _LP64
1675   st->print_cr("\tLDUW   [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1676   st->print_cr("\tCMP    R_G5,R_G3" );
1677   st->print   ("\tTne    icc,R_G0+ST_RESERVED_FOR_USER_0+2");
1678 #endif // _LP64
1679 }
1680 #endif
1681 
1682 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1683   MacroAssembler _masm(&cbuf);
1684   Register G5_ic_reg  = reg_to_register_object(Matcher::inline_cache_reg_encode());
1685   Register temp_reg   = G3;
1686   assert( G5_ic_reg != temp_reg, "conflicting registers" );
1687 
1688   // Load klass from receiver
1689   __ load_klass(O0, temp_reg);
1690   // Compare against expected klass
1691   __ cmp(temp_reg, G5_ic_reg);
1692   // Branch to miss code, checks xcc or icc depending
1693   __ trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2);
1694 }
1695 
1696 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1697   return MachNode::size(ra_);
1698 }
1699 
1700 
1701 //=============================================================================
1702 
1703 uint size_exception_handler() {
1704   if (TraceJumps) {
1705     return (400); // just a guess
1706   }
1707   return ( NativeJump::instruction_size ); // sethi;jmp;nop
1708 }
1709 
1710 uint size_deopt_handler() {
1711   if (TraceJumps) {
1712     return (400); // just a guess
1713   }
1714   return ( 4+  NativeJump::instruction_size ); // save;sethi;jmp;restore
1715 }
1716 
1717 // Emit exception handler code.
1718 int emit_exception_handler(CodeBuffer& cbuf) {
1719   Register temp_reg = G3;
1720   AddressLiteral exception_blob(OptoRuntime::exception_blob()->entry_point());
1721   MacroAssembler _masm(&cbuf);
1722 
1723   address base =
1724   __ start_a_stub(size_exception_handler());
1725   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1726 
1727   int offset = __ offset();
1728 
1729   __ JUMP(exception_blob, temp_reg, 0); // sethi;jmp
1730   __ delayed()->nop();
1731 
1732   assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1733 
1734   __ end_a_stub();
1735 
1736   return offset;
1737 }
1738 
1739 int emit_deopt_handler(CodeBuffer& cbuf) {
1740   // Can't use any of the current frame's registers as we may have deopted
1741   // at a poll and everything (including G3) can be live.
1742   Register temp_reg = L0;
1743   AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
1744   MacroAssembler _masm(&cbuf);
1745 
1746   address base =
1747   __ start_a_stub(size_deopt_handler());
1748   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1749 
1750   int offset = __ offset();
1751   __ save_frame(0);
1752   __ JUMP(deopt_blob, temp_reg, 0); // sethi;jmp
1753   __ delayed()->restore();
1754 
1755   assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1756 
1757   __ end_a_stub();
1758   return offset;
1759 
1760 }
1761 
1762 // Given a register encoding, produce a Integer Register object
1763 static Register reg_to_register_object(int register_encoding) {
1764   assert(L5->encoding() == R_L5_enc && G1->encoding() == R_G1_enc, "right coding");
1765   return as_Register(register_encoding);
1766 }
1767 
1768 // Given a register encoding, produce a single-precision Float Register object
1769 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding) {
1770   assert(F5->encoding(FloatRegisterImpl::S) == R_F5_enc && F12->encoding(FloatRegisterImpl::S) == R_F12_enc, "right coding");
1771   return as_SingleFloatRegister(register_encoding);
1772 }
1773 
1774 // Given a register encoding, produce a double-precision Float Register object
1775 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding) {
1776   assert(F4->encoding(FloatRegisterImpl::D) == R_F4_enc, "right coding");
1777   assert(F32->encoding(FloatRegisterImpl::D) == R_D32_enc, "right coding");
1778   return as_DoubleFloatRegister(register_encoding);
1779 }
1780 
1781 const bool Matcher::match_rule_supported(int opcode) {
1782   if (!has_match_rule(opcode))
1783     return false;
1784 
1785   switch (opcode) {
1786   case Op_CountLeadingZerosI:
1787   case Op_CountLeadingZerosL:
1788   case Op_CountTrailingZerosI:
1789   case Op_CountTrailingZerosL:
1790   case Op_PopCountI:
1791   case Op_PopCountL:
1792     if (!UsePopCountInstruction)
1793       return false;
1794   case Op_CompareAndSwapL:
1795 #ifdef _LP64
1796   case Op_CompareAndSwapP:
1797 #endif
1798     if (!VM_Version::supports_cx8())
1799       return false;
1800     break;
1801   }
1802 
1803   return true;  // Per default match rules are supported.
1804 }
1805 
1806 int Matcher::regnum_to_fpu_offset(int regnum) {
1807   return regnum - 32; // The FP registers are in the second chunk
1808 }
1809 
1810 #ifdef ASSERT
1811 address last_rethrow = NULL;  // debugging aid for Rethrow encoding
1812 #endif
1813 
1814 // Vector width in bytes
1815 const int Matcher::vector_width_in_bytes(BasicType bt) {
1816   assert(MaxVectorSize == 8, "");
1817   return 8;
1818 }
1819 
1820 // Vector ideal reg
1821 const int Matcher::vector_ideal_reg(int size) {
1822   assert(MaxVectorSize == 8, "");
1823   return Op_RegD;
1824 }
1825 
1826 const int Matcher::vector_shift_count_ideal_reg(int size) {
1827   fatal("vector shift is not supported");
1828   return Node::NotAMachineReg;
1829 }
1830 
1831 // Limits on vector size (number of elements) loaded into vector.
1832 const int Matcher::max_vector_size(const BasicType bt) {
1833   assert(is_java_primitive(bt), "only primitive type vectors");
1834   return vector_width_in_bytes(bt)/type2aelembytes(bt);
1835 }
1836 
1837 const int Matcher::min_vector_size(const BasicType bt) {
1838   return max_vector_size(bt); // Same as max.
1839 }
1840 
1841 // SPARC doesn't support misaligned vectors store/load.
1842 const bool Matcher::misaligned_vectors_ok() {
1843   return false;
1844 }
1845 
1846 // USII supports fxtof through the whole range of number, USIII doesn't
1847 const bool Matcher::convL2FSupported(void) {
1848   return VM_Version::has_fast_fxtof();
1849 }
1850 
1851 // Is this branch offset short enough that a short branch can be used?
1852 //
1853 // NOTE: If the platform does not provide any short branch variants, then
1854 //       this method should return false for offset 0.
1855 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
1856   // The passed offset is relative to address of the branch.
1857   // Don't need to adjust the offset.
1858   return UseCBCond && Assembler::is_simm12(offset);
1859 }
1860 
1861 const bool Matcher::isSimpleConstant64(jlong value) {
1862   // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1863   // Depends on optimizations in MacroAssembler::setx.
1864   int hi = (int)(value >> 32);
1865   int lo = (int)(value & ~0);
1866   return (hi == 0) || (hi == -1) || (lo == 0);
1867 }
1868 
1869 // No scaling for the parameter the ClearArray node.
1870 const bool Matcher::init_array_count_is_in_bytes = true;
1871 
1872 // Threshold size for cleararray.
1873 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1874 
1875 // No additional cost for CMOVL.
1876 const int Matcher::long_cmove_cost() { return 0; }
1877 
1878 // CMOVF/CMOVD are expensive on T4 and on SPARC64.
1879 const int Matcher::float_cmove_cost() {
1880   return (VM_Version::is_T4() || VM_Version::is_sparc64()) ? ConditionalMoveLimit : 0;
1881 }
1882 
1883 // Should the Matcher clone shifts on addressing modes, expecting them to
1884 // be subsumed into complex addressing expressions or compute them into
1885 // registers?  True for Intel but false for most RISCs
1886 const bool Matcher::clone_shift_expressions = false;
1887 
1888 // Do we need to mask the count passed to shift instructions or does
1889 // the cpu only look at the lower 5/6 bits anyway?
1890 const bool Matcher::need_masked_shift_count = false;
1891 
1892 bool Matcher::narrow_oop_use_complex_address() {
1893   NOT_LP64(ShouldNotCallThis());
1894   assert(UseCompressedOops, "only for compressed oops code");
1895   return false;
1896 }
1897 
1898 bool Matcher::narrow_klass_use_complex_address() {
1899   NOT_LP64(ShouldNotCallThis());
1900   assert(UseCompressedKlassPointers, "only for compressed klass code");
1901   return false;
1902 }
1903 
1904 // Is it better to copy float constants, or load them directly from memory?
1905 // Intel can load a float constant from a direct address, requiring no
1906 // extra registers.  Most RISCs will have to materialize an address into a
1907 // register first, so they would do better to copy the constant from stack.
1908 const bool Matcher::rematerialize_float_constants = false;
1909 
1910 // If CPU can load and store mis-aligned doubles directly then no fixup is
1911 // needed.  Else we split the double into 2 integer pieces and move it
1912 // piece-by-piece.  Only happens when passing doubles into C code as the
1913 // Java calling convention forces doubles to be aligned.
1914 #ifdef _LP64
1915 const bool Matcher::misaligned_doubles_ok = true;
1916 #else
1917 const bool Matcher::misaligned_doubles_ok = false;
1918 #endif
1919 
1920 // No-op on SPARC.
1921 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
1922 }
1923 
1924 // Advertise here if the CPU requires explicit rounding operations
1925 // to implement the UseStrictFP mode.
1926 const bool Matcher::strict_fp_requires_explicit_rounding = false;
1927 
1928 // Are floats conerted to double when stored to stack during deoptimization?
1929 // Sparc does not handle callee-save floats.
1930 bool Matcher::float_in_double() { return false; }
1931 
1932 // Do ints take an entire long register or just half?
1933 // Note that we if-def off of _LP64.
1934 // The relevant question is how the int is callee-saved.  In _LP64
1935 // the whole long is written but de-opt'ing will have to extract
1936 // the relevant 32 bits, in not-_LP64 only the low 32 bits is written.
1937 #ifdef _LP64
1938 const bool Matcher::int_in_long = true;
1939 #else
1940 const bool Matcher::int_in_long = false;
1941 #endif
1942 
1943 // Return whether or not this register is ever used as an argument.  This
1944 // function is used on startup to build the trampoline stubs in generateOptoStub.
1945 // Registers not mentioned will be killed by the VM call in the trampoline, and
1946 // arguments in those registers not be available to the callee.
1947 bool Matcher::can_be_java_arg( int reg ) {
1948   // Standard sparc 6 args in registers
1949   if( reg == R_I0_num ||
1950       reg == R_I1_num ||
1951       reg == R_I2_num ||
1952       reg == R_I3_num ||
1953       reg == R_I4_num ||
1954       reg == R_I5_num ) return true;
1955 #ifdef _LP64
1956   // 64-bit builds can pass 64-bit pointers and longs in
1957   // the high I registers
1958   if( reg == R_I0H_num ||
1959       reg == R_I1H_num ||
1960       reg == R_I2H_num ||
1961       reg == R_I3H_num ||
1962       reg == R_I4H_num ||
1963       reg == R_I5H_num ) return true;
1964 
1965   if ((UseCompressedOops) && (reg == R_G6_num || reg == R_G6H_num)) {
1966     return true;
1967   }
1968 
1969 #else
1970   // 32-bit builds with longs-in-one-entry pass longs in G1 & G4.
1971   // Longs cannot be passed in O regs, because O regs become I regs
1972   // after a 'save' and I regs get their high bits chopped off on
1973   // interrupt.
1974   if( reg == R_G1H_num || reg == R_G1_num ) return true;
1975   if( reg == R_G4H_num || reg == R_G4_num ) return true;
1976 #endif
1977   // A few float args in registers
1978   if( reg >= R_F0_num && reg <= R_F7_num ) return true;
1979 
1980   return false;
1981 }
1982 
1983 bool Matcher::is_spillable_arg( int reg ) {
1984   return can_be_java_arg(reg);
1985 }
1986 
1987 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
1988   // Use hardware SDIVX instruction when it is
1989   // faster than a code which use multiply.
1990   return VM_Version::has_fast_idiv();
1991 }
1992 
1993 // Register for DIVI projection of divmodI
1994 RegMask Matcher::divI_proj_mask() {
1995   ShouldNotReachHere();
1996   return RegMask();
1997 }
1998 
1999 // Register for MODI projection of divmodI
2000 RegMask Matcher::modI_proj_mask() {
2001   ShouldNotReachHere();
2002   return RegMask();
2003 }
2004 
2005 // Register for DIVL projection of divmodL
2006 RegMask Matcher::divL_proj_mask() {
2007   ShouldNotReachHere();
2008   return RegMask();
2009 }
2010 
2011 // Register for MODL projection of divmodL
2012 RegMask Matcher::modL_proj_mask() {
2013   ShouldNotReachHere();
2014   return RegMask();
2015 }
2016 
2017 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2018   return L7_REGP_mask();
2019 }
2020 
2021 %}
2022 
2023 
2024 // The intptr_t operand types, defined by textual substitution.
2025 // (Cf. opto/type.hpp.  This lets us avoid many, many other ifdefs.)
2026 #ifdef _LP64
2027 #define immX      immL
2028 #define immX13    immL13
2029 #define immX13m7  immL13m7
2030 #define iRegX     iRegL
2031 #define g1RegX    g1RegL
2032 #else
2033 #define immX      immI
2034 #define immX13    immI13
2035 #define immX13m7  immI13m7
2036 #define iRegX     iRegI
2037 #define g1RegX    g1RegI
2038 #endif
2039 
2040 //----------ENCODING BLOCK-----------------------------------------------------
2041 // This block specifies the encoding classes used by the compiler to output
2042 // byte streams.  Encoding classes are parameterized macros used by
2043 // Machine Instruction Nodes in order to generate the bit encoding of the
2044 // instruction.  Operands specify their base encoding interface with the
2045 // interface keyword.  There are currently supported four interfaces,
2046 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER.  REG_INTER causes an
2047 // operand to generate a function which returns its register number when
2048 // queried.   CONST_INTER causes an operand to generate a function which
2049 // returns the value of the constant when queried.  MEMORY_INTER causes an
2050 // operand to generate four functions which return the Base Register, the
2051 // Index Register, the Scale Value, and the Offset Value of the operand when
2052 // queried.  COND_INTER causes an operand to generate six functions which
2053 // return the encoding code (ie - encoding bits for the instruction)
2054 // associated with each basic boolean condition for a conditional instruction.
2055 //
2056 // Instructions specify two basic values for encoding.  Again, a function
2057 // is available to check if the constant displacement is an oop. They use the
2058 // ins_encode keyword to specify their encoding classes (which must be
2059 // a sequence of enc_class names, and their parameters, specified in
2060 // the encoding block), and they use the
2061 // opcode keyword to specify, in order, their primary, secondary, and
2062 // tertiary opcode.  Only the opcode sections which a particular instruction
2063 // needs for encoding need to be specified.
2064 encode %{
2065   enc_class enc_untested %{
2066 #ifdef ASSERT
2067     MacroAssembler _masm(&cbuf);
2068     __ untested("encoding");
2069 #endif
2070   %}
2071 
2072   enc_class form3_mem_reg( memory mem, iRegI dst ) %{
2073     emit_form3_mem_reg(cbuf, this, $primary, $tertiary,
2074                        $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2075   %}
2076 
2077   enc_class simple_form3_mem_reg( memory mem, iRegI dst ) %{
2078     emit_form3_mem_reg(cbuf, this, $primary, -1,
2079                        $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2080   %}
2081 
2082   enc_class form3_mem_prefetch_read( memory mem ) %{
2083     emit_form3_mem_reg(cbuf, this, $primary, -1,
2084                        $mem$$base, $mem$$disp, $mem$$index, 0/*prefetch function many-reads*/);
2085   %}
2086 
2087   enc_class form3_mem_prefetch_write( memory mem ) %{
2088     emit_form3_mem_reg(cbuf, this, $primary, -1,
2089                        $mem$$base, $mem$$disp, $mem$$index, 2/*prefetch function many-writes*/);
2090   %}
2091 
2092   enc_class form3_mem_reg_long_unaligned_marshal( memory mem, iRegL reg ) %{
2093     assert(Assembler::is_simm13($mem$$disp  ), "need disp and disp+4");
2094     assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2095     guarantee($mem$$index == R_G0_enc, "double index?");
2096     emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, R_O7_enc );
2097     emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp,   R_G0_enc, $reg$$reg );
2098     emit3_simm13( cbuf, Assembler::arith_op, $reg$$reg, Assembler::sllx_op3, $reg$$reg, 0x1020 );
2099     emit3( cbuf, Assembler::arith_op, $reg$$reg, Assembler::or_op3, $reg$$reg, 0, R_O7_enc );
2100   %}
2101 
2102   enc_class form3_mem_reg_double_unaligned( memory mem, RegD_low reg ) %{
2103     assert(Assembler::is_simm13($mem$$disp  ), "need disp and disp+4");
2104     assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2105     guarantee($mem$$index == R_G0_enc, "double index?");
2106     // Load long with 2 instructions
2107     emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp,   R_G0_enc, $reg$$reg+0 );
2108     emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, $reg$$reg+1 );
2109   %}
2110 
2111   //%%% form3_mem_plus_4_reg is a hack--get rid of it
2112   enc_class form3_mem_plus_4_reg( memory mem, iRegI dst ) %{
2113     guarantee($mem$$disp, "cannot offset a reg-reg operand by 4");
2114     emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp + 4, $mem$$index, $dst$$reg);
2115   %}
2116 
2117   enc_class form3_g0_rs2_rd_move( iRegI rs2, iRegI rd ) %{
2118     // Encode a reg-reg copy.  If it is useless, then empty encoding.
2119     if( $rs2$$reg != $rd$$reg )
2120       emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, $rs2$$reg );
2121   %}
2122 
2123   // Target lo half of long
2124   enc_class form3_g0_rs2_rd_move_lo( iRegI rs2, iRegL rd ) %{
2125     // Encode a reg-reg copy.  If it is useless, then empty encoding.
2126     if( $rs2$$reg != LONG_LO_REG($rd$$reg) )
2127       emit3( cbuf, Assembler::arith_op, LONG_LO_REG($rd$$reg), Assembler::or_op3, 0, 0, $rs2$$reg );
2128   %}
2129 
2130   // Source lo half of long
2131   enc_class form3_g0_rs2_rd_move_lo2( iRegL rs2, iRegI rd ) %{
2132     // Encode a reg-reg copy.  If it is useless, then empty encoding.
2133     if( LONG_LO_REG($rs2$$reg) != $rd$$reg )
2134       emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_LO_REG($rs2$$reg) );
2135   %}
2136 
2137   // Target hi half of long
2138   enc_class form3_rs1_rd_copysign_hi( iRegI rs1, iRegL rd ) %{
2139     emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 31 );
2140   %}
2141 
2142   // Source lo half of long, and leave it sign extended.
2143   enc_class form3_rs1_rd_signextend_lo1( iRegL rs1, iRegI rd ) %{
2144     // Sign extend low half
2145     emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 0, 0 );
2146   %}
2147 
2148   // Source hi half of long, and leave it sign extended.
2149   enc_class form3_rs1_rd_copy_hi1( iRegL rs1, iRegI rd ) %{
2150     // Shift high half to low half
2151     emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::srlx_op3, $rs1$$reg, 32 );
2152   %}
2153 
2154   // Source hi half of long
2155   enc_class form3_g0_rs2_rd_move_hi2( iRegL rs2, iRegI rd ) %{
2156     // Encode a reg-reg copy.  If it is useless, then empty encoding.
2157     if( LONG_HI_REG($rs2$$reg) != $rd$$reg )
2158       emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_HI_REG($rs2$$reg) );
2159   %}
2160 
2161   enc_class form3_rs1_rs2_rd( iRegI rs1, iRegI rs2, iRegI rd ) %{
2162     emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0, $rs2$$reg );
2163   %}
2164 
2165   enc_class enc_to_bool( iRegI src, iRegI dst ) %{
2166     emit3       ( cbuf, Assembler::arith_op,         0, Assembler::subcc_op3, 0, 0, $src$$reg );
2167     emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::addc_op3 , 0, 0 );
2168   %}
2169 
2170   enc_class enc_ltmask( iRegI p, iRegI q, iRegI dst ) %{
2171     emit3       ( cbuf, Assembler::arith_op,         0, Assembler::subcc_op3, $p$$reg, 0, $q$$reg );
2172     // clear if nothing else is happening
2173     emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0,  0 );
2174     // blt,a,pn done
2175     emit2_19    ( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less, Assembler::bp_op2, Assembler::icc, 0/*predict not taken*/, 2 );
2176     // mov dst,-1 in delay slot
2177     emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2178   %}
2179 
2180   enc_class form3_rs1_imm5_rd( iRegI rs1, immU5 imm5, iRegI rd ) %{
2181     emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $imm5$$constant & 0x1F );
2182   %}
2183 
2184   enc_class form3_sd_rs1_imm6_rd( iRegL rs1, immU6 imm6, iRegL rd ) %{
2185     emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, ($imm6$$constant & 0x3F) | 0x1000 );
2186   %}
2187 
2188   enc_class form3_sd_rs1_rs2_rd( iRegL rs1, iRegI rs2, iRegL rd ) %{
2189     emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0x80, $rs2$$reg );
2190   %}
2191 
2192   enc_class form3_rs1_simm13_rd( iRegI rs1, immI13 simm13, iRegI rd ) %{
2193     emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $simm13$$constant );
2194   %}
2195 
2196   enc_class move_return_pc_to_o1() %{
2197     emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::add_op3, R_O7_enc, frame::pc_return_offset );
2198   %}
2199 
2200 #ifdef _LP64
2201   /* %%% merge with enc_to_bool */
2202   enc_class enc_convP2B( iRegI dst, iRegP src ) %{
2203     MacroAssembler _masm(&cbuf);
2204 
2205     Register   src_reg = reg_to_register_object($src$$reg);
2206     Register   dst_reg = reg_to_register_object($dst$$reg);
2207     __ movr(Assembler::rc_nz, src_reg, 1, dst_reg);
2208   %}
2209 #endif
2210 
2211   enc_class enc_cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp ) %{
2212     // (Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)))
2213     MacroAssembler _masm(&cbuf);
2214 
2215     Register   p_reg = reg_to_register_object($p$$reg);
2216     Register   q_reg = reg_to_register_object($q$$reg);
2217     Register   y_reg = reg_to_register_object($y$$reg);
2218     Register tmp_reg = reg_to_register_object($tmp$$reg);
2219 
2220     __ subcc( p_reg, q_reg,   p_reg );
2221     __ add  ( p_reg, y_reg, tmp_reg );
2222     __ movcc( Assembler::less, false, Assembler::icc, tmp_reg, p_reg );
2223   %}
2224 
2225   enc_class form_d2i_helper(regD src, regF dst) %{
2226     // fcmp %fcc0,$src,$src
2227     emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2228     // branch %fcc0 not-nan, predict taken
2229     emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2230     // fdtoi $src,$dst
2231     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fdtoi_opf, $src$$reg );
2232     // fitos $dst,$dst (if nan)
2233     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fitos_opf, $dst$$reg );
2234     // clear $dst (if nan)
2235     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2236     // carry on here...
2237   %}
2238 
2239   enc_class form_d2l_helper(regD src, regD dst) %{
2240     // fcmp %fcc0,$src,$src  check for NAN
2241     emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2242     // branch %fcc0 not-nan, predict taken
2243     emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2244     // fdtox $src,$dst   convert in delay slot
2245     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fdtox_opf, $src$$reg );
2246     // fxtod $dst,$dst  (if nan)
2247     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fxtod_opf, $dst$$reg );
2248     // clear $dst (if nan)
2249     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2250     // carry on here...
2251   %}
2252 
2253   enc_class form_f2i_helper(regF src, regF dst) %{
2254     // fcmps %fcc0,$src,$src
2255     emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2256     // branch %fcc0 not-nan, predict taken
2257     emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2258     // fstoi $src,$dst
2259     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fstoi_opf, $src$$reg );
2260     // fitos $dst,$dst (if nan)
2261     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fitos_opf, $dst$$reg );
2262     // clear $dst (if nan)
2263     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2264     // carry on here...
2265   %}
2266 
2267   enc_class form_f2l_helper(regF src, regD dst) %{
2268     // fcmps %fcc0,$src,$src
2269     emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2270     // branch %fcc0 not-nan, predict taken
2271     emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2272     // fstox $src,$dst
2273     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fstox_opf, $src$$reg );
2274     // fxtod $dst,$dst (if nan)
2275     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3,         0, Assembler::fxtod_opf, $dst$$reg );
2276     // clear $dst (if nan)
2277     emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2278     // carry on here...
2279   %}
2280 
2281   enc_class form3_opf_rs2F_rdF(regF rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2282   enc_class form3_opf_rs2F_rdD(regF rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2283   enc_class form3_opf_rs2D_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2284   enc_class form3_opf_rs2D_rdD(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2285 
2286   enc_class form3_opf_rs2D_lo_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg+1); %}
2287 
2288   enc_class form3_opf_rs2D_hi_rdD_hi(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2289   enc_class form3_opf_rs2D_lo_rdD_lo(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg+1,$primary,0,$tertiary,$rs2$$reg+1); %}
2290 
2291   enc_class form3_opf_rs1F_rs2F_rdF( regF rs1, regF rs2, regF rd ) %{
2292     emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2293   %}
2294 
2295   enc_class form3_opf_rs1D_rs2D_rdD( regD rs1, regD rs2, regD rd ) %{
2296     emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2297   %}
2298 
2299   enc_class form3_opf_rs1F_rs2F_fcc( regF rs1, regF rs2, flagsRegF fcc ) %{
2300     emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2301   %}
2302 
2303   enc_class form3_opf_rs1D_rs2D_fcc( regD rs1, regD rs2, flagsRegF fcc ) %{
2304     emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2305   %}
2306 
2307   enc_class form3_convI2F(regF rs2, regF rd) %{
2308     emit3(cbuf,Assembler::arith_op,$rd$$reg,Assembler::fpop1_op3,0,$secondary,$rs2$$reg);
2309   %}
2310 
2311   // Encloding class for traceable jumps
2312   enc_class form_jmpl(g3RegP dest) %{
2313     emit_jmpl(cbuf, $dest$$reg);
2314   %}
2315 
2316   enc_class form_jmpl_set_exception_pc(g1RegP dest) %{
2317     emit_jmpl_set_exception_pc(cbuf, $dest$$reg);
2318   %}
2319 
2320   enc_class form2_nop() %{
2321     emit_nop(cbuf);
2322   %}
2323 
2324   enc_class form2_illtrap() %{
2325     emit_illtrap(cbuf);
2326   %}
2327 
2328 
2329   // Compare longs and convert into -1, 0, 1.
2330   enc_class cmpl_flag( iRegL src1, iRegL src2, iRegI dst ) %{
2331     // CMP $src1,$src2
2332     emit3( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $src1$$reg, 0, $src2$$reg );
2333     // blt,a,pn done
2334     emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less   , Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 5 );
2335     // mov dst,-1 in delay slot
2336     emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2337     // bgt,a,pn done
2338     emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::greater, Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 3 );
2339     // mov dst,1 in delay slot
2340     emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0,  1 );
2341     // CLR    $dst
2342     emit3( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3 , 0, 0, 0 );
2343   %}
2344 
2345   enc_class enc_PartialSubtypeCheck() %{
2346     MacroAssembler _masm(&cbuf);
2347     __ call(StubRoutines::Sparc::partial_subtype_check(), relocInfo::runtime_call_type);
2348     __ delayed()->nop();
2349   %}
2350 
2351   enc_class enc_bp( label labl, cmpOp cmp, flagsReg cc ) %{
2352     MacroAssembler _masm(&cbuf);
2353     Label* L = $labl$$label;
2354     Assembler::Predict predict_taken =
2355       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2356 
2357     __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
2358     __ delayed()->nop();
2359   %}
2360 
2361   enc_class enc_bpr( label labl, cmpOp_reg cmp, iRegI op1 ) %{
2362     MacroAssembler _masm(&cbuf);
2363     Label* L = $labl$$label;
2364     Assembler::Predict predict_taken =
2365       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2366 
2367     __ bpr( (Assembler::RCondition)($cmp$$cmpcode), false, predict_taken, as_Register($op1$$reg), *L);
2368     __ delayed()->nop();
2369   %}
2370 
2371   enc_class enc_cmov_reg( cmpOp cmp, iRegI dst, iRegI src, immI pcc) %{
2372     int op = (Assembler::arith_op << 30) |
2373              ($dst$$reg << 25) |
2374              (Assembler::movcc_op3 << 19) |
2375              (1 << 18) |                    // cc2 bit for 'icc'
2376              ($cmp$$cmpcode << 14) |
2377              (0 << 13) |                    // select register move
2378              ($pcc$$constant << 11) |       // cc1, cc0 bits for 'icc' or 'xcc'
2379              ($src$$reg << 0);
2380     cbuf.insts()->emit_int32(op);
2381   %}
2382 
2383   enc_class enc_cmov_imm( cmpOp cmp, iRegI dst, immI11 src, immI pcc ) %{
2384     int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2385     int op = (Assembler::arith_op << 30) |
2386              ($dst$$reg << 25) |
2387              (Assembler::movcc_op3 << 19) |
2388              (1 << 18) |                    // cc2 bit for 'icc'
2389              ($cmp$$cmpcode << 14) |
2390              (1 << 13) |                    // select immediate move
2391              ($pcc$$constant << 11) |       // cc1, cc0 bits for 'icc'
2392              (simm11 << 0);
2393     cbuf.insts()->emit_int32(op);
2394   %}
2395 
2396   enc_class enc_cmov_reg_f( cmpOpF cmp, iRegI dst, iRegI src, flagsRegF fcc ) %{
2397     int op = (Assembler::arith_op << 30) |
2398              ($dst$$reg << 25) |
2399              (Assembler::movcc_op3 << 19) |
2400              (0 << 18) |                    // cc2 bit for 'fccX'
2401              ($cmp$$cmpcode << 14) |
2402              (0 << 13) |                    // select register move
2403              ($fcc$$reg << 11) |            // cc1, cc0 bits for fcc0-fcc3
2404              ($src$$reg << 0);
2405     cbuf.insts()->emit_int32(op);
2406   %}
2407 
2408   enc_class enc_cmov_imm_f( cmpOp cmp, iRegI dst, immI11 src, flagsRegF fcc ) %{
2409     int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2410     int op = (Assembler::arith_op << 30) |
2411              ($dst$$reg << 25) |
2412              (Assembler::movcc_op3 << 19) |
2413              (0 << 18) |                    // cc2 bit for 'fccX'
2414              ($cmp$$cmpcode << 14) |
2415              (1 << 13) |                    // select immediate move
2416              ($fcc$$reg << 11) |            // cc1, cc0 bits for fcc0-fcc3
2417              (simm11 << 0);
2418     cbuf.insts()->emit_int32(op);
2419   %}
2420 
2421   enc_class enc_cmovf_reg( cmpOp cmp, regD dst, regD src, immI pcc ) %{
2422     int op = (Assembler::arith_op << 30) |
2423              ($dst$$reg << 25) |
2424              (Assembler::fpop2_op3 << 19) |
2425              (0 << 18) |
2426              ($cmp$$cmpcode << 14) |
2427              (1 << 13) |                    // select register move
2428              ($pcc$$constant << 11) |       // cc1-cc0 bits for 'icc' or 'xcc'
2429              ($primary << 5) |              // select single, double or quad
2430              ($src$$reg << 0);
2431     cbuf.insts()->emit_int32(op);
2432   %}
2433 
2434   enc_class enc_cmovff_reg( cmpOpF cmp, flagsRegF fcc, regD dst, regD src ) %{
2435     int op = (Assembler::arith_op << 30) |
2436              ($dst$$reg << 25) |
2437              (Assembler::fpop2_op3 << 19) |
2438              (0 << 18) |
2439              ($cmp$$cmpcode << 14) |
2440              ($fcc$$reg << 11) |            // cc2-cc0 bits for 'fccX'
2441              ($primary << 5) |              // select single, double or quad
2442              ($src$$reg << 0);
2443     cbuf.insts()->emit_int32(op);
2444   %}
2445 
2446   // Used by the MIN/MAX encodings.  Same as a CMOV, but
2447   // the condition comes from opcode-field instead of an argument.
2448   enc_class enc_cmov_reg_minmax( iRegI dst, iRegI src ) %{
2449     int op = (Assembler::arith_op << 30) |
2450              ($dst$$reg << 25) |
2451              (Assembler::movcc_op3 << 19) |
2452              (1 << 18) |                    // cc2 bit for 'icc'
2453              ($primary << 14) |
2454              (0 << 13) |                    // select register move
2455              (0 << 11) |                    // cc1, cc0 bits for 'icc'
2456              ($src$$reg << 0);
2457     cbuf.insts()->emit_int32(op);
2458   %}
2459 
2460   enc_class enc_cmov_reg_minmax_long( iRegL dst, iRegL src ) %{
2461     int op = (Assembler::arith_op << 30) |
2462              ($dst$$reg << 25) |
2463              (Assembler::movcc_op3 << 19) |
2464              (6 << 16) |                    // cc2 bit for 'xcc'
2465              ($primary << 14) |
2466              (0 << 13) |                    // select register move
2467              (0 << 11) |                    // cc1, cc0 bits for 'icc'
2468              ($src$$reg << 0);
2469     cbuf.insts()->emit_int32(op);
2470   %}
2471 
2472   enc_class Set13( immI13 src, iRegI rd ) %{
2473     emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, $src$$constant );
2474   %}
2475 
2476   enc_class SetHi22( immI src, iRegI rd ) %{
2477     emit2_22( cbuf, Assembler::branch_op, $rd$$reg, Assembler::sethi_op2, $src$$constant );
2478   %}
2479 
2480   enc_class Set32( immI src, iRegI rd ) %{
2481     MacroAssembler _masm(&cbuf);
2482     __ set($src$$constant, reg_to_register_object($rd$$reg));
2483   %}
2484 
2485   enc_class call_epilog %{
2486     if( VerifyStackAtCalls ) {
2487       MacroAssembler _masm(&cbuf);
2488       int framesize = ra_->C->frame_slots() << LogBytesPerInt;
2489       Register temp_reg = G3;
2490       __ add(SP, framesize, temp_reg);
2491       __ cmp(temp_reg, FP);
2492       __ breakpoint_trap(Assembler::notEqual, Assembler::ptr_cc);
2493     }
2494   %}
2495 
2496   // Long values come back from native calls in O0:O1 in the 32-bit VM, copy the value
2497   // to G1 so the register allocator will not have to deal with the misaligned register
2498   // pair.
2499   enc_class adjust_long_from_native_call %{
2500 #ifndef _LP64
2501     if (returns_long()) {
2502       //    sllx  O0,32,O0
2503       emit3_simm13( cbuf, Assembler::arith_op, R_O0_enc, Assembler::sllx_op3, R_O0_enc, 0x1020 );
2504       //    srl   O1,0,O1
2505       emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::srl_op3, R_O1_enc, 0x0000 );
2506       //    or    O0,O1,G1
2507       emit3       ( cbuf, Assembler::arith_op, R_G1_enc, Assembler:: or_op3, R_O0_enc, 0, R_O1_enc );
2508     }
2509 #endif
2510   %}
2511 
2512   enc_class Java_To_Runtime (method meth) %{    // CALL Java_To_Runtime
2513     // CALL directly to the runtime
2514     // The user of this is responsible for ensuring that R_L7 is empty (killed).
2515     emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type,
2516                     /*preserve_g2=*/true);
2517   %}
2518 
2519   enc_class preserve_SP %{
2520     MacroAssembler _masm(&cbuf);
2521     __ mov(SP, L7_mh_SP_save);
2522   %}
2523 
2524   enc_class restore_SP %{
2525     MacroAssembler _masm(&cbuf);
2526     __ mov(L7_mh_SP_save, SP);
2527   %}
2528 
2529   enc_class Java_Static_Call (method meth) %{    // JAVA STATIC CALL
2530     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
2531     // who we intended to call.
2532     if (!_method) {
2533       emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type);
2534     } else if (_optimized_virtual) {
2535       emit_call_reloc(cbuf, $meth$$method, relocInfo::opt_virtual_call_type);
2536     } else {
2537       emit_call_reloc(cbuf, $meth$$method, relocInfo::static_call_type);
2538     }
2539     if (_method) {  // Emit stub for static call.
2540       CompiledStaticCall::emit_to_interp_stub(cbuf);
2541     }
2542   %}
2543 
2544   enc_class Java_Dynamic_Call (method meth) %{    // JAVA DYNAMIC CALL
2545     MacroAssembler _masm(&cbuf);
2546     __ set_inst_mark();
2547     int vtable_index = this->_vtable_index;
2548     // MachCallDynamicJavaNode::ret_addr_offset uses this same test
2549     if (vtable_index < 0) {
2550       // must be invalid_vtable_index, not nonvirtual_vtable_index
2551       assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
2552       Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2553       assert(G5_ic_reg == G5_inline_cache_reg, "G5_inline_cache_reg used in assemble_ic_buffer_code()");
2554       assert(G5_ic_reg == G5_megamorphic_method, "G5_megamorphic_method used in megamorphic call stub");
2555       __ ic_call((address)$meth$$method);
2556     } else {
2557       assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
2558       // Just go thru the vtable
2559       // get receiver klass (receiver already checked for non-null)
2560       // If we end up going thru a c2i adapter interpreter expects method in G5
2561       int off = __ offset();
2562       __ load_klass(O0, G3_scratch);
2563       int klass_load_size;
2564       if (UseCompressedKlassPointers) {
2565         assert(Universe::heap() != NULL, "java heap should be initialized");
2566         klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
2567       } else {
2568         klass_load_size = 1*BytesPerInstWord;
2569       }
2570       int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
2571       int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
2572       if (Assembler::is_simm13(v_off)) {
2573         __ ld_ptr(G3, v_off, G5_method);
2574       } else {
2575         // Generate 2 instructions
2576         __ Assembler::sethi(v_off & ~0x3ff, G5_method);
2577         __ or3(G5_method, v_off & 0x3ff, G5_method);
2578         // ld_ptr, set_hi, set
2579         assert(__ offset() - off == klass_load_size + 2*BytesPerInstWord,
2580                "Unexpected instruction size(s)");
2581         __ ld_ptr(G3, G5_method, G5_method);
2582       }
2583       // NOTE: for vtable dispatches, the vtable entry will never be null.
2584       // However it may very well end up in handle_wrong_method if the
2585       // method is abstract for the particular class.
2586       __ ld_ptr(G5_method, in_bytes(Method::from_compiled_offset()), G3_scratch);
2587       // jump to target (either compiled code or c2iadapter)
2588       __ jmpl(G3_scratch, G0, O7);
2589       __ delayed()->nop();
2590     }
2591   %}
2592 
2593   enc_class Java_Compiled_Call (method meth) %{    // JAVA COMPILED CALL
2594     MacroAssembler _masm(&cbuf);
2595 
2596     Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2597     Register temp_reg = G3;   // caller must kill G3!  We cannot reuse G5_ic_reg here because
2598                               // we might be calling a C2I adapter which needs it.
2599 
2600     assert(temp_reg != G5_ic_reg, "conflicting registers");
2601     // Load nmethod
2602     __ ld_ptr(G5_ic_reg, in_bytes(Method::from_compiled_offset()), temp_reg);
2603 
2604     // CALL to compiled java, indirect the contents of G3
2605     __ set_inst_mark();
2606     __ callr(temp_reg, G0);
2607     __ delayed()->nop();
2608   %}
2609 
2610 enc_class idiv_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst) %{
2611     MacroAssembler _masm(&cbuf);
2612     Register Rdividend = reg_to_register_object($src1$$reg);
2613     Register Rdivisor = reg_to_register_object($src2$$reg);
2614     Register Rresult = reg_to_register_object($dst$$reg);
2615 
2616     __ sra(Rdivisor, 0, Rdivisor);
2617     __ sra(Rdividend, 0, Rdividend);
2618     __ sdivx(Rdividend, Rdivisor, Rresult);
2619 %}
2620 
2621 enc_class idiv_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst) %{
2622     MacroAssembler _masm(&cbuf);
2623 
2624     Register Rdividend = reg_to_register_object($src1$$reg);
2625     int divisor = $imm$$constant;
2626     Register Rresult = reg_to_register_object($dst$$reg);
2627 
2628     __ sra(Rdividend, 0, Rdividend);
2629     __ sdivx(Rdividend, divisor, Rresult);
2630 %}
2631 
2632 enc_class enc_mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2) %{
2633     MacroAssembler _masm(&cbuf);
2634     Register Rsrc1 = reg_to_register_object($src1$$reg);
2635     Register Rsrc2 = reg_to_register_object($src2$$reg);
2636     Register Rdst  = reg_to_register_object($dst$$reg);
2637 
2638     __ sra( Rsrc1, 0, Rsrc1 );
2639     __ sra( Rsrc2, 0, Rsrc2 );
2640     __ mulx( Rsrc1, Rsrc2, Rdst );
2641     __ srlx( Rdst, 32, Rdst );
2642 %}
2643 
2644 enc_class irem_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst, o7RegL scratch) %{
2645     MacroAssembler _masm(&cbuf);
2646     Register Rdividend = reg_to_register_object($src1$$reg);
2647     Register Rdivisor = reg_to_register_object($src2$$reg);
2648     Register Rresult = reg_to_register_object($dst$$reg);
2649     Register Rscratch = reg_to_register_object($scratch$$reg);
2650 
2651     assert(Rdividend != Rscratch, "");
2652     assert(Rdivisor  != Rscratch, "");
2653 
2654     __ sra(Rdividend, 0, Rdividend);
2655     __ sra(Rdivisor, 0, Rdivisor);
2656     __ sdivx(Rdividend, Rdivisor, Rscratch);
2657     __ mulx(Rscratch, Rdivisor, Rscratch);
2658     __ sub(Rdividend, Rscratch, Rresult);
2659 %}
2660 
2661 enc_class irem_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst, o7RegL scratch) %{
2662     MacroAssembler _masm(&cbuf);
2663 
2664     Register Rdividend = reg_to_register_object($src1$$reg);
2665     int divisor = $imm$$constant;
2666     Register Rresult = reg_to_register_object($dst$$reg);
2667     Register Rscratch = reg_to_register_object($scratch$$reg);
2668 
2669     assert(Rdividend != Rscratch, "");
2670 
2671     __ sra(Rdividend, 0, Rdividend);
2672     __ sdivx(Rdividend, divisor, Rscratch);
2673     __ mulx(Rscratch, divisor, Rscratch);
2674     __ sub(Rdividend, Rscratch, Rresult);
2675 %}
2676 
2677 enc_class fabss (sflt_reg dst, sflt_reg src) %{
2678     MacroAssembler _masm(&cbuf);
2679 
2680     FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2681     FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2682 
2683     __ fabs(FloatRegisterImpl::S, Fsrc, Fdst);
2684 %}
2685 
2686 enc_class fabsd (dflt_reg dst, dflt_reg src) %{
2687     MacroAssembler _masm(&cbuf);
2688 
2689     FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2690     FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2691 
2692     __ fabs(FloatRegisterImpl::D, Fsrc, Fdst);
2693 %}
2694 
2695 enc_class fnegd (dflt_reg dst, dflt_reg src) %{
2696     MacroAssembler _masm(&cbuf);
2697 
2698     FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2699     FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2700 
2701     __ fneg(FloatRegisterImpl::D, Fsrc, Fdst);
2702 %}
2703 
2704 enc_class fsqrts (sflt_reg dst, sflt_reg src) %{
2705     MacroAssembler _masm(&cbuf);
2706 
2707     FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2708     FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2709 
2710     __ fsqrt(FloatRegisterImpl::S, Fsrc, Fdst);
2711 %}
2712 
2713 enc_class fsqrtd (dflt_reg dst, dflt_reg src) %{
2714     MacroAssembler _masm(&cbuf);
2715 
2716     FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2717     FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2718 
2719     __ fsqrt(FloatRegisterImpl::D, Fsrc, Fdst);
2720 %}
2721 
2722 enc_class fmovs (dflt_reg dst, dflt_reg src) %{
2723     MacroAssembler _masm(&cbuf);
2724 
2725     FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2726     FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2727 
2728     __ fmov(FloatRegisterImpl::S, Fsrc, Fdst);
2729 %}
2730 
2731 enc_class fmovd (dflt_reg dst, dflt_reg src) %{
2732     MacroAssembler _masm(&cbuf);
2733 
2734     FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2735     FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2736 
2737     __ fmov(FloatRegisterImpl::D, Fsrc, Fdst);
2738 %}
2739 
2740 enc_class Fast_Lock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2741     MacroAssembler _masm(&cbuf);
2742 
2743     Register Roop  = reg_to_register_object($oop$$reg);
2744     Register Rbox  = reg_to_register_object($box$$reg);
2745     Register Rscratch = reg_to_register_object($scratch$$reg);
2746     Register Rmark =    reg_to_register_object($scratch2$$reg);
2747 
2748     assert(Roop  != Rscratch, "");
2749     assert(Roop  != Rmark, "");
2750     assert(Rbox  != Rscratch, "");
2751     assert(Rbox  != Rmark, "");
2752 
2753     __ compiler_lock_object(Roop, Rmark, Rbox, Rscratch, _counters, UseBiasedLocking && !UseOptoBiasInlining);
2754 %}
2755 
2756 enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2757     MacroAssembler _masm(&cbuf);
2758 
2759     Register Roop  = reg_to_register_object($oop$$reg);
2760     Register Rbox  = reg_to_register_object($box$$reg);
2761     Register Rscratch = reg_to_register_object($scratch$$reg);
2762     Register Rmark =    reg_to_register_object($scratch2$$reg);
2763 
2764     assert(Roop  != Rscratch, "");
2765     assert(Roop  != Rmark, "");
2766     assert(Rbox  != Rscratch, "");
2767     assert(Rbox  != Rmark, "");
2768 
2769     __ compiler_unlock_object(Roop, Rmark, Rbox, Rscratch, UseBiasedLocking && !UseOptoBiasInlining);
2770   %}
2771 
2772   enc_class enc_cas( iRegP mem, iRegP old, iRegP new ) %{
2773     MacroAssembler _masm(&cbuf);
2774     Register Rmem = reg_to_register_object($mem$$reg);
2775     Register Rold = reg_to_register_object($old$$reg);
2776     Register Rnew = reg_to_register_object($new$$reg);
2777 
2778     __ cas_ptr(Rmem, Rold, Rnew); // Swap(*Rmem,Rnew) if *Rmem == Rold
2779     __ cmp( Rold, Rnew );
2780   %}
2781 
2782   enc_class enc_casx( iRegP mem, iRegL old, iRegL new) %{
2783     Register Rmem = reg_to_register_object($mem$$reg);
2784     Register Rold = reg_to_register_object($old$$reg);
2785     Register Rnew = reg_to_register_object($new$$reg);
2786 
2787     MacroAssembler _masm(&cbuf);
2788     __ mov(Rnew, O7);
2789     __ casx(Rmem, Rold, O7);
2790     __ cmp( Rold, O7 );
2791   %}
2792 
2793   // raw int cas, used for compareAndSwap
2794   enc_class enc_casi( iRegP mem, iRegL old, iRegL new) %{
2795     Register Rmem = reg_to_register_object($mem$$reg);
2796     Register Rold = reg_to_register_object($old$$reg);
2797     Register Rnew = reg_to_register_object($new$$reg);
2798 
2799     MacroAssembler _masm(&cbuf);
2800     __ mov(Rnew, O7);
2801     __ cas(Rmem, Rold, O7);
2802     __ cmp( Rold, O7 );
2803   %}
2804 
2805   enc_class enc_lflags_ne_to_boolean( iRegI res ) %{
2806     Register Rres = reg_to_register_object($res$$reg);
2807 
2808     MacroAssembler _masm(&cbuf);
2809     __ mov(1, Rres);
2810     __ movcc( Assembler::notEqual, false, Assembler::xcc, G0, Rres );
2811   %}
2812 
2813   enc_class enc_iflags_ne_to_boolean( iRegI res ) %{
2814     Register Rres = reg_to_register_object($res$$reg);
2815 
2816     MacroAssembler _masm(&cbuf);
2817     __ mov(1, Rres);
2818     __ movcc( Assembler::notEqual, false, Assembler::icc, G0, Rres );
2819   %}
2820 
2821   enc_class floating_cmp ( iRegP dst, regF src1, regF src2 ) %{
2822     MacroAssembler _masm(&cbuf);
2823     Register Rdst = reg_to_register_object($dst$$reg);
2824     FloatRegister Fsrc1 = $primary ? reg_to_SingleFloatRegister_object($src1$$reg)
2825                                      : reg_to_DoubleFloatRegister_object($src1$$reg);
2826     FloatRegister Fsrc2 = $primary ? reg_to_SingleFloatRegister_object($src2$$reg)
2827                                      : reg_to_DoubleFloatRegister_object($src2$$reg);
2828 
2829     // Convert condition code fcc0 into -1,0,1; unordered reports less-than (-1)
2830     __ float_cmp( $primary, -1, Fsrc1, Fsrc2, Rdst);
2831   %}
2832 
2833 
2834   enc_class enc_String_Compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result) %{
2835     Label Ldone, Lloop;
2836     MacroAssembler _masm(&cbuf);
2837 
2838     Register   str1_reg = reg_to_register_object($str1$$reg);
2839     Register   str2_reg = reg_to_register_object($str2$$reg);
2840     Register   cnt1_reg = reg_to_register_object($cnt1$$reg);
2841     Register   cnt2_reg = reg_to_register_object($cnt2$$reg);
2842     Register result_reg = reg_to_register_object($result$$reg);
2843 
2844     assert(result_reg != str1_reg &&
2845            result_reg != str2_reg &&
2846            result_reg != cnt1_reg &&
2847            result_reg != cnt2_reg ,
2848            "need different registers");
2849 
2850     // Compute the minimum of the string lengths(str1_reg) and the
2851     // difference of the string lengths (stack)
2852 
2853     // See if the lengths are different, and calculate min in str1_reg.
2854     // Stash diff in O7 in case we need it for a tie-breaker.
2855     Label Lskip;
2856     __ subcc(cnt1_reg, cnt2_reg, O7);
2857     __ sll(cnt1_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2858     __ br(Assembler::greater, true, Assembler::pt, Lskip);
2859     // cnt2 is shorter, so use its count:
2860     __ delayed()->sll(cnt2_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2861     __ bind(Lskip);
2862 
2863     // reallocate cnt1_reg, cnt2_reg, result_reg
2864     // Note:  limit_reg holds the string length pre-scaled by 2
2865     Register limit_reg =   cnt1_reg;
2866     Register  chr2_reg =   cnt2_reg;
2867     Register  chr1_reg = result_reg;
2868     // str{12} are the base pointers
2869 
2870     // Is the minimum length zero?
2871     __ cmp(limit_reg, (int)(0 * sizeof(jchar))); // use cast to resolve overloading ambiguity
2872     __ br(Assembler::equal, true, Assembler::pn, Ldone);
2873     __ delayed()->mov(O7, result_reg);  // result is difference in lengths
2874 
2875     // Load first characters
2876     __ lduh(str1_reg, 0, chr1_reg);
2877     __ lduh(str2_reg, 0, chr2_reg);
2878 
2879     // Compare first characters
2880     __ subcc(chr1_reg, chr2_reg, chr1_reg);
2881     __ br(Assembler::notZero, false, Assembler::pt,  Ldone);
2882     assert(chr1_reg == result_reg, "result must be pre-placed");
2883     __ delayed()->nop();
2884 
2885     {
2886       // Check after comparing first character to see if strings are equivalent
2887       Label LSkip2;
2888       // Check if the strings start at same location
2889       __ cmp(str1_reg, str2_reg);
2890       __ brx(Assembler::notEqual, true, Assembler::pt, LSkip2);
2891       __ delayed()->nop();
2892 
2893       // Check if the length difference is zero (in O7)
2894       __ cmp(G0, O7);
2895       __ br(Assembler::equal, true, Assembler::pn, Ldone);
2896       __ delayed()->mov(G0, result_reg);  // result is zero
2897 
2898       // Strings might not be equal
2899       __ bind(LSkip2);
2900     }
2901 
2902     __ subcc(limit_reg, 1 * sizeof(jchar), chr1_reg);
2903     __ br(Assembler::equal, true, Assembler::pn, Ldone);
2904     __ delayed()->mov(O7, result_reg);  // result is difference in lengths
2905 
2906     // Shift str1_reg and str2_reg to the end of the arrays, negate limit
2907     __ add(str1_reg, limit_reg, str1_reg);
2908     __ add(str2_reg, limit_reg, str2_reg);
2909     __ neg(chr1_reg, limit_reg);  // limit = -(limit-2)
2910 
2911     // Compare the rest of the characters
2912     __ lduh(str1_reg, limit_reg, chr1_reg);
2913     __ bind(Lloop);
2914     // __ lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2915     __ lduh(str2_reg, limit_reg, chr2_reg);
2916     __ subcc(chr1_reg, chr2_reg, chr1_reg);
2917     __ br(Assembler::notZero, false, Assembler::pt, Ldone);
2918     assert(chr1_reg == result_reg, "result must be pre-placed");
2919     __ delayed()->inccc(limit_reg, sizeof(jchar));
2920     // annul LDUH if branch is not taken to prevent access past end of string
2921     __ br(Assembler::notZero, true, Assembler::pt, Lloop);
2922     __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2923 
2924     // If strings are equal up to min length, return the length difference.
2925     __ mov(O7, result_reg);
2926 
2927     // Otherwise, return the difference between the first mismatched chars.
2928     __ bind(Ldone);
2929   %}
2930 
2931 enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result) %{
2932     Label Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone;
2933     MacroAssembler _masm(&cbuf);
2934 
2935     Register   str1_reg = reg_to_register_object($str1$$reg);
2936     Register   str2_reg = reg_to_register_object($str2$$reg);
2937     Register    cnt_reg = reg_to_register_object($cnt$$reg);
2938     Register   tmp1_reg = O7;
2939     Register result_reg = reg_to_register_object($result$$reg);
2940 
2941     assert(result_reg != str1_reg &&
2942            result_reg != str2_reg &&
2943            result_reg !=  cnt_reg &&
2944            result_reg != tmp1_reg ,
2945            "need different registers");
2946 
2947     __ cmp(str1_reg, str2_reg); //same char[] ?
2948     __ brx(Assembler::equal, true, Assembler::pn, Ldone);
2949     __ delayed()->add(G0, 1, result_reg);
2950 
2951     __ cmp_zero_and_br(Assembler::zero, cnt_reg, Ldone, true, Assembler::pn);
2952     __ delayed()->add(G0, 1, result_reg); // count == 0
2953 
2954     //rename registers
2955     Register limit_reg =    cnt_reg;
2956     Register  chr1_reg = result_reg;
2957     Register  chr2_reg =   tmp1_reg;
2958 
2959     //check for alignment and position the pointers to the ends
2960     __ or3(str1_reg, str2_reg, chr1_reg);
2961     __ andcc(chr1_reg, 0x3, chr1_reg);
2962     // notZero means at least one not 4-byte aligned.
2963     // We could optimize the case when both arrays are not aligned
2964     // but it is not frequent case and it requires additional checks.
2965     __ br(Assembler::notZero, false, Assembler::pn, Lchar); // char by char compare
2966     __ delayed()->sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); // set byte count
2967 
2968     // Compare char[] arrays aligned to 4 bytes.
2969     __ char_arrays_equals(str1_reg, str2_reg, limit_reg, result_reg,
2970                           chr1_reg, chr2_reg, Ldone);
2971     __ ba(Ldone);
2972     __ delayed()->add(G0, 1, result_reg);
2973 
2974     // char by char compare
2975     __ bind(Lchar);
2976     __ add(str1_reg, limit_reg, str1_reg);
2977     __ add(str2_reg, limit_reg, str2_reg);
2978     __ neg(limit_reg); //negate count
2979 
2980     __ lduh(str1_reg, limit_reg, chr1_reg);
2981     // Lchar_loop
2982     __ bind(Lchar_loop);
2983     __ lduh(str2_reg, limit_reg, chr2_reg);
2984     __ cmp(chr1_reg, chr2_reg);
2985     __ br(Assembler::notEqual, true, Assembler::pt, Ldone);
2986     __ delayed()->mov(G0, result_reg); //not equal
2987     __ inccc(limit_reg, sizeof(jchar));
2988     // annul LDUH if branch is not taken to prevent access past end of string
2989     __ br(Assembler::notZero, true, Assembler::pt, Lchar_loop);
2990     __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2991 
2992     __ add(G0, 1, result_reg);  //equal
2993 
2994     __ bind(Ldone);
2995   %}
2996 
2997 enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, notemp_iRegI result) %{
2998     Label Lvector, Ldone, Lloop;
2999     MacroAssembler _masm(&cbuf);
3000 
3001     Register   ary1_reg = reg_to_register_object($ary1$$reg);
3002     Register   ary2_reg = reg_to_register_object($ary2$$reg);
3003     Register   tmp1_reg = reg_to_register_object($tmp1$$reg);
3004     Register   tmp2_reg = O7;
3005     Register result_reg = reg_to_register_object($result$$reg);
3006 
3007     int length_offset  = arrayOopDesc::length_offset_in_bytes();
3008     int base_offset    = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3009 
3010     // return true if the same array
3011     __ cmp(ary1_reg, ary2_reg);
3012     __ brx(Assembler::equal, true, Assembler::pn, Ldone);
3013     __ delayed()->add(G0, 1, result_reg); // equal
3014 
3015     __ br_null(ary1_reg, true, Assembler::pn, Ldone);
3016     __ delayed()->mov(G0, result_reg);    // not equal
3017 
3018     __ br_null(ary2_reg, true, Assembler::pn, Ldone);
3019     __ delayed()->mov(G0, result_reg);    // not equal
3020 
3021     //load the lengths of arrays
3022     __ ld(Address(ary1_reg, length_offset), tmp1_reg);
3023     __ ld(Address(ary2_reg, length_offset), tmp2_reg);
3024 
3025     // return false if the two arrays are not equal length
3026     __ cmp(tmp1_reg, tmp2_reg);
3027     __ br(Assembler::notEqual, true, Assembler::pn, Ldone);
3028     __ delayed()->mov(G0, result_reg);     // not equal
3029 
3030     __ cmp_zero_and_br(Assembler::zero, tmp1_reg, Ldone, true, Assembler::pn);
3031     __ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal
3032 
3033     // load array addresses
3034     __ add(ary1_reg, base_offset, ary1_reg);
3035     __ add(ary2_reg, base_offset, ary2_reg);
3036 
3037     // renaming registers
3038     Register chr1_reg  =  result_reg; // for characters in ary1
3039     Register chr2_reg  =  tmp2_reg;   // for characters in ary2
3040     Register limit_reg =  tmp1_reg;   // length
3041 
3042     // set byte count
3043     __ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg);
3044 
3045     // Compare char[] arrays aligned to 4 bytes.
3046     __ char_arrays_equals(ary1_reg, ary2_reg, limit_reg, result_reg,
3047                           chr1_reg, chr2_reg, Ldone);
3048     __ add(G0, 1, result_reg); // equals
3049 
3050     __ bind(Ldone);
3051   %}
3052 
3053   enc_class enc_rethrow() %{
3054     cbuf.set_insts_mark();
3055     Register temp_reg = G3;
3056     AddressLiteral rethrow_stub(OptoRuntime::rethrow_stub());
3057     assert(temp_reg != reg_to_register_object(R_I0_num), "temp must not break oop_reg");
3058     MacroAssembler _masm(&cbuf);
3059 #ifdef ASSERT
3060     __ save_frame(0);
3061     AddressLiteral last_rethrow_addrlit(&last_rethrow);
3062     __ sethi(last_rethrow_addrlit, L1);
3063     Address addr(L1, last_rethrow_addrlit.low10());
3064     __ rdpc(L2);
3065     __ inc(L2, 3 * BytesPerInstWord);  // skip this & 2 more insns to point at jump_to
3066     __ st_ptr(L2, addr);
3067     __ restore();
3068 #endif
3069     __ JUMP(rethrow_stub, temp_reg, 0); // sethi;jmp
3070     __ delayed()->nop();
3071   %}
3072 
3073   enc_class emit_mem_nop() %{
3074     // Generates the instruction LDUXA [o6,g0],#0x82,g0
3075     cbuf.insts()->emit_int32((unsigned int) 0xc0839040);
3076   %}
3077 
3078   enc_class emit_fadd_nop() %{
3079     // Generates the instruction FMOVS f31,f31
3080     cbuf.insts()->emit_int32((unsigned int) 0xbfa0003f);
3081   %}
3082 
3083   enc_class emit_br_nop() %{
3084     // Generates the instruction BPN,PN .
3085     cbuf.insts()->emit_int32((unsigned int) 0x00400000);
3086   %}
3087 
3088   enc_class enc_membar_acquire %{
3089     MacroAssembler _masm(&cbuf);
3090     __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::LoadLoad) );
3091   %}
3092 
3093   enc_class enc_membar_release %{
3094     MacroAssembler _masm(&cbuf);
3095     __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore) );
3096   %}
3097 
3098   enc_class enc_membar_volatile %{
3099     MacroAssembler _masm(&cbuf);
3100     __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3101   %}
3102 
3103 %}
3104 
3105 //----------FRAME--------------------------------------------------------------
3106 // Definition of frame structure and management information.
3107 //
3108 //  S T A C K   L A Y O U T    Allocators stack-slot number
3109 //                             |   (to get allocators register number
3110 //  G  Owned by    |        |  v    add VMRegImpl::stack0)
3111 //  r   CALLER     |        |
3112 //  o     |        +--------+      pad to even-align allocators stack-slot
3113 //  w     V        |  pad0  |        numbers; owned by CALLER
3114 //  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
3115 //  h     ^        |   in   |  5
3116 //        |        |  args  |  4   Holes in incoming args owned by SELF
3117 //  |     |        |        |  3
3118 //  |     |        +--------+
3119 //  V     |        | old out|      Empty on Intel, window on Sparc
3120 //        |    old |preserve|      Must be even aligned.
3121 //        |     SP-+--------+----> Matcher::_old_SP, 8 (or 16 in LP64)-byte aligned
3122 //        |        |   in   |  3   area for Intel ret address
3123 //     Owned by    |preserve|      Empty on Sparc.
3124 //       SELF      +--------+
3125 //        |        |  pad2  |  2   pad to align old SP
3126 //        |        +--------+  1
3127 //        |        | locks  |  0
3128 //        |        +--------+----> VMRegImpl::stack0, 8 (or 16 in LP64)-byte aligned
3129 //        |        |  pad1  | 11   pad to align new SP
3130 //        |        +--------+
3131 //        |        |        | 10
3132 //        |        | spills |  9   spills
3133 //        V        |        |  8   (pad0 slot for callee)
3134 //      -----------+--------+----> Matcher::_out_arg_limit, unaligned
3135 //        ^        |  out   |  7
3136 //        |        |  args  |  6   Holes in outgoing args owned by CALLEE
3137 //     Owned by    +--------+
3138 //      CALLEE     | new out|  6   Empty on Intel, window on Sparc
3139 //        |    new |preserve|      Must be even-aligned.
3140 //        |     SP-+--------+----> Matcher::_new_SP, even aligned
3141 //        |        |        |
3142 //
3143 // Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
3144 //         known from SELF's arguments and the Java calling convention.
3145 //         Region 6-7 is determined per call site.
3146 // Note 2: If the calling convention leaves holes in the incoming argument
3147 //         area, those holes are owned by SELF.  Holes in the outgoing area
3148 //         are owned by the CALLEE.  Holes should not be nessecary in the
3149 //         incoming area, as the Java calling convention is completely under
3150 //         the control of the AD file.  Doubles can be sorted and packed to
3151 //         avoid holes.  Holes in the outgoing arguments may be nessecary for
3152 //         varargs C calling conventions.
3153 // Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
3154 //         even aligned with pad0 as needed.
3155 //         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
3156 //         region 6-11 is even aligned; it may be padded out more so that
3157 //         the region from SP to FP meets the minimum stack alignment.
3158 
3159 frame %{
3160   // What direction does stack grow in (assumed to be same for native & Java)
3161   stack_direction(TOWARDS_LOW);
3162 
3163   // These two registers define part of the calling convention
3164   // between compiled code and the interpreter.
3165   inline_cache_reg(R_G5);                // Inline Cache Register or Method* for I2C
3166   interpreter_method_oop_reg(R_G5);      // Method Oop Register when calling interpreter
3167 
3168   // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
3169   cisc_spilling_operand_name(indOffset);
3170 
3171   // Number of stack slots consumed by a Monitor enter
3172 #ifdef _LP64
3173   sync_stack_slots(2);
3174 #else
3175   sync_stack_slots(1);
3176 #endif
3177 
3178   // Compiled code's Frame Pointer
3179   frame_pointer(R_SP);
3180 
3181   // Stack alignment requirement
3182   stack_alignment(StackAlignmentInBytes);
3183   //  LP64: Alignment size in bytes (128-bit -> 16 bytes)
3184   // !LP64: Alignment size in bytes (64-bit  ->  8 bytes)
3185 
3186   // Number of stack slots between incoming argument block and the start of
3187   // a new frame.  The PROLOG must add this many slots to the stack.  The
3188   // EPILOG must remove this many slots.
3189   in_preserve_stack_slots(0);
3190 
3191   // Number of outgoing stack slots killed above the out_preserve_stack_slots
3192   // for calls to C.  Supports the var-args backing area for register parms.
3193   // ADLC doesn't support parsing expressions, so I folded the math by hand.
3194 #ifdef _LP64
3195   // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (0)) * 2-stack-slots-per-word
3196   varargs_C_out_slots_killed(12);
3197 #else
3198   // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (1)) * 1-stack-slots-per-word
3199   varargs_C_out_slots_killed( 7);
3200 #endif
3201 
3202   // The after-PROLOG location of the return address.  Location of
3203   // return address specifies a type (REG or STACK) and a number
3204   // representing the register number (i.e. - use a register name) or
3205   // stack slot.
3206   return_addr(REG R_I7);          // Ret Addr is in register I7
3207 
3208   // Body of function which returns an OptoRegs array locating
3209   // arguments either in registers or in stack slots for calling
3210   // java
3211   calling_convention %{
3212     (void) SharedRuntime::java_calling_convention(sig_bt, regs, length, is_outgoing);
3213 
3214   %}
3215 
3216   // Body of function which returns an OptoRegs array locating
3217   // arguments either in registers or in stack slots for callin
3218   // C.
3219   c_calling_convention %{
3220     // This is obviously always outgoing
3221     (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
3222   %}
3223 
3224   // Location of native (C/C++) and interpreter return values.  This is specified to
3225   // be the  same as Java.  In the 32-bit VM, long values are actually returned from
3226   // native calls in O0:O1 and returned to the interpreter in I0:I1.  The copying
3227   // to and from the register pairs is done by the appropriate call and epilog
3228   // opcodes.  This simplifies the register allocator.
3229   c_return_value %{
3230     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3231 #ifdef     _LP64
3232     static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_O0_num };
3233     static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num,    OptoReg::Bad, R_F1_num, R_O0H_num};
3234     static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_I0_num };
3235     static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num,    OptoReg::Bad, R_F1_num, R_I0H_num};
3236 #else  // !_LP64
3237     static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_G1_num };
3238     static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3239     static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_G1_num };
3240     static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3241 #endif
3242     return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3243                         (is_outgoing?lo_out:lo_in)[ideal_reg] );
3244   %}
3245 
3246   // Location of compiled Java return values.  Same as C
3247   return_value %{
3248     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3249 #ifdef     _LP64
3250     static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_O0_num };
3251     static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num,    OptoReg::Bad, R_F1_num, R_O0H_num};
3252     static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_I0_num };
3253     static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num,    OptoReg::Bad, R_F1_num, R_I0H_num};
3254 #else  // !_LP64
3255     static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num,     R_O0_num,     R_O0_num,     R_F0_num,     R_F0_num, R_G1_num };
3256     static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3257     static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num,     R_I0_num,     R_I0_num,     R_F0_num,     R_F0_num, R_G1_num };
3258     static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3259 #endif
3260     return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3261                         (is_outgoing?lo_out:lo_in)[ideal_reg] );
3262   %}
3263 
3264 %}
3265 
3266 
3267 //----------ATTRIBUTES---------------------------------------------------------
3268 //----------Operand Attributes-------------------------------------------------
3269 op_attrib op_cost(1);          // Required cost attribute
3270 
3271 //----------Instruction Attributes---------------------------------------------
3272 ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute
3273 ins_attrib ins_size(32);           // Required size attribute (in bits)
3274 ins_attrib ins_avoid_back_to_back(0); // instruction should not be generated back to back
3275 ins_attrib ins_short_branch(0);    // Required flag: is this instruction a
3276                                    // non-matching short branch variant of some
3277                                                             // long branch?
3278 
3279 //----------OPERANDS-----------------------------------------------------------
3280 // Operand definitions must precede instruction definitions for correct parsing
3281 // in the ADLC because operands constitute user defined types which are used in
3282 // instruction definitions.
3283 
3284 //----------Simple Operands----------------------------------------------------
3285 // Immediate Operands
3286 // Integer Immediate: 32-bit
3287 operand immI() %{
3288   match(ConI);
3289 
3290   op_cost(0);
3291   // formats are generated automatically for constants and base registers
3292   format %{ %}
3293   interface(CONST_INTER);
3294 %}
3295 
3296 // Integer Immediate: 8-bit
3297 operand immI8() %{
3298   predicate(Assembler::is_simm8(n->get_int()));
3299   match(ConI);
3300   op_cost(0);
3301   format %{ %}
3302   interface(CONST_INTER);
3303 %}
3304 
3305 // Integer Immediate: 13-bit
3306 operand immI13() %{
3307   predicate(Assembler::is_simm13(n->get_int()));
3308   match(ConI);
3309   op_cost(0);
3310 
3311   format %{ %}
3312   interface(CONST_INTER);
3313 %}
3314 
3315 // Integer Immediate: 13-bit minus 7
3316 operand immI13m7() %{
3317   predicate((-4096 < n->get_int()) && ((n->get_int() + 7) <= 4095));
3318   match(ConI);
3319   op_cost(0);
3320 
3321   format %{ %}
3322   interface(CONST_INTER);
3323 %}
3324 
3325 // Integer Immediate: 16-bit
3326 operand immI16() %{
3327   predicate(Assembler::is_simm16(n->get_int()));
3328   match(ConI);
3329   op_cost(0);
3330   format %{ %}
3331   interface(CONST_INTER);
3332 %}
3333 
3334 // Unsigned (positive) Integer Immediate: 13-bit
3335 operand immU13() %{
3336   predicate((0 <= n->get_int()) && Assembler::is_simm13(n->get_int()));
3337   match(ConI);
3338   op_cost(0);
3339 
3340   format %{ %}
3341   interface(CONST_INTER);
3342 %}
3343 
3344 // Integer Immediate: 6-bit
3345 operand immU6() %{
3346   predicate(n->get_int() >= 0 && n->get_int() <= 63);
3347   match(ConI);
3348   op_cost(0);
3349   format %{ %}
3350   interface(CONST_INTER);
3351 %}
3352 
3353 // Integer Immediate: 11-bit
3354 operand immI11() %{
3355   predicate(Assembler::is_simm11(n->get_int()));
3356   match(ConI);
3357   op_cost(0);
3358   format %{ %}
3359   interface(CONST_INTER);
3360 %}
3361 
3362 // Integer Immediate: 5-bit
3363 operand immI5() %{
3364   predicate(Assembler::is_simm5(n->get_int()));
3365   match(ConI);
3366   op_cost(0);
3367   format %{ %}
3368   interface(CONST_INTER);
3369 %}
3370 
3371 // Integer Immediate: 0-bit
3372 operand immI0() %{
3373   predicate(n->get_int() == 0);
3374   match(ConI);
3375   op_cost(0);
3376 
3377   format %{ %}
3378   interface(CONST_INTER);
3379 %}
3380 
3381 // Integer Immediate: the value 10
3382 operand immI10() %{
3383   predicate(n->get_int() == 10);
3384   match(ConI);
3385   op_cost(0);
3386 
3387   format %{ %}
3388   interface(CONST_INTER);
3389 %}
3390 
3391 // Integer Immediate: the values 0-31
3392 operand immU5() %{
3393   predicate(n->get_int() >= 0 && n->get_int() <= 31);
3394   match(ConI);
3395   op_cost(0);
3396 
3397   format %{ %}
3398   interface(CONST_INTER);
3399 %}
3400 
3401 // Integer Immediate: the values 1-31
3402 operand immI_1_31() %{
3403   predicate(n->get_int() >= 1 && n->get_int() <= 31);
3404   match(ConI);
3405   op_cost(0);
3406 
3407   format %{ %}
3408   interface(CONST_INTER);
3409 %}
3410 
3411 // Integer Immediate: the values 32-63
3412 operand immI_32_63() %{
3413   predicate(n->get_int() >= 32 && n->get_int() <= 63);
3414   match(ConI);
3415   op_cost(0);
3416 
3417   format %{ %}
3418   interface(CONST_INTER);
3419 %}
3420 
3421 // Immediates for special shifts (sign extend)
3422 
3423 // Integer Immediate: the value 16
3424 operand immI_16() %{
3425   predicate(n->get_int() == 16);
3426   match(ConI);
3427   op_cost(0);
3428 
3429   format %{ %}
3430   interface(CONST_INTER);
3431 %}
3432 
3433 // Integer Immediate: the value 24
3434 operand immI_24() %{
3435   predicate(n->get_int() == 24);
3436   match(ConI);
3437   op_cost(0);
3438 
3439   format %{ %}
3440   interface(CONST_INTER);
3441 %}
3442 
3443 // Integer Immediate: the value 255
3444 operand immI_255() %{
3445   predicate( n->get_int() == 255 );
3446   match(ConI);
3447   op_cost(0);
3448 
3449   format %{ %}
3450   interface(CONST_INTER);
3451 %}
3452 
3453 // Integer Immediate: the value 65535
3454 operand immI_65535() %{
3455   predicate(n->get_int() == 65535);
3456   match(ConI);
3457   op_cost(0);
3458 
3459   format %{ %}
3460   interface(CONST_INTER);
3461 %}
3462 
3463 // Long Immediate: the value FF
3464 operand immL_FF() %{
3465   predicate( n->get_long() == 0xFFL );
3466   match(ConL);
3467   op_cost(0);
3468 
3469   format %{ %}
3470   interface(CONST_INTER);
3471 %}
3472 
3473 // Long Immediate: the value FFFF
3474 operand immL_FFFF() %{
3475   predicate( n->get_long() == 0xFFFFL );
3476   match(ConL);
3477   op_cost(0);
3478 
3479   format %{ %}
3480   interface(CONST_INTER);
3481 %}
3482 
3483 // Pointer Immediate: 32 or 64-bit
3484 operand immP() %{
3485   match(ConP);
3486 
3487   op_cost(5);
3488   // formats are generated automatically for constants and base registers
3489   format %{ %}
3490   interface(CONST_INTER);
3491 %}
3492 
3493 #ifdef _LP64
3494 // Pointer Immediate: 64-bit
3495 operand immP_set() %{
3496   predicate(!VM_Version::is_niagara_plus());
3497   match(ConP);
3498 
3499   op_cost(5);
3500   // formats are generated automatically for constants and base registers
3501   format %{ %}
3502   interface(CONST_INTER);
3503 %}
3504 
3505 // Pointer Immediate: 64-bit
3506 // From Niagara2 processors on a load should be better than materializing.
3507 operand immP_load() %{
3508   predicate(VM_Version::is_niagara_plus() && (n->bottom_type()->isa_oop_ptr() || (MacroAssembler::insts_for_set(n->get_ptr()) > 3)));
3509   match(ConP);
3510 
3511   op_cost(5);
3512   // formats are generated automatically for constants and base registers
3513   format %{ %}
3514   interface(CONST_INTER);
3515 %}
3516 
3517 // Pointer Immediate: 64-bit
3518 operand immP_no_oop_cheap() %{
3519   predicate(VM_Version::is_niagara_plus() && !n->bottom_type()->isa_oop_ptr() && (MacroAssembler::insts_for_set(n->get_ptr()) <= 3));
3520   match(ConP);
3521 
3522   op_cost(5);
3523   // formats are generated automatically for constants and base registers
3524   format %{ %}
3525   interface(CONST_INTER);
3526 %}
3527 #endif
3528 
3529 operand immP13() %{
3530   predicate((-4096 < n->get_ptr()) && (n->get_ptr() <= 4095));
3531   match(ConP);
3532   op_cost(0);
3533 
3534   format %{ %}
3535   interface(CONST_INTER);
3536 %}
3537 
3538 operand immP0() %{
3539   predicate(n->get_ptr() == 0);
3540   match(ConP);
3541   op_cost(0);
3542 
3543   format %{ %}
3544   interface(CONST_INTER);
3545 %}
3546 
3547 operand immP_poll() %{
3548   predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
3549   match(ConP);
3550 
3551   // formats are generated automatically for constants and base registers
3552   format %{ %}
3553   interface(CONST_INTER);
3554 %}
3555 
3556 // Pointer Immediate
3557 operand immN()
3558 %{
3559   match(ConN);
3560 
3561   op_cost(10);
3562   format %{ %}
3563   interface(CONST_INTER);
3564 %}
3565 
3566 operand immNKlass()
3567 %{
3568   match(ConNKlass);
3569 
3570   op_cost(10);
3571   format %{ %}
3572   interface(CONST_INTER);
3573 %}
3574 
3575 // NULL Pointer Immediate
3576 operand immN0()
3577 %{
3578   predicate(n->get_narrowcon() == 0);
3579   match(ConN);
3580 
3581   op_cost(0);
3582   format %{ %}
3583   interface(CONST_INTER);
3584 %}
3585 
3586 operand immL() %{
3587   match(ConL);
3588   op_cost(40);
3589   // formats are generated automatically for constants and base registers
3590   format %{ %}
3591   interface(CONST_INTER);
3592 %}
3593 
3594 operand immL0() %{
3595   predicate(n->get_long() == 0L);
3596   match(ConL);
3597   op_cost(0);
3598   // formats are generated automatically for constants and base registers
3599   format %{ %}
3600   interface(CONST_INTER);
3601 %}
3602 
3603 // Integer Immediate: 5-bit
3604 operand immL5() %{
3605   predicate(n->get_long() == (int)n->get_long() && Assembler::is_simm5((int)n->get_long()));
3606   match(ConL);
3607   op_cost(0);
3608   format %{ %}
3609   interface(CONST_INTER);
3610 %}
3611 
3612 // Long Immediate: 13-bit
3613 operand immL13() %{
3614   predicate((-4096L < n->get_long()) && (n->get_long() <= 4095L));
3615   match(ConL);
3616   op_cost(0);
3617 
3618   format %{ %}
3619   interface(CONST_INTER);
3620 %}
3621 
3622 // Long Immediate: 13-bit minus 7
3623 operand immL13m7() %{
3624   predicate((-4096L < n->get_long()) && ((n->get_long() + 7L) <= 4095L));
3625   match(ConL);
3626   op_cost(0);
3627 
3628   format %{ %}
3629   interface(CONST_INTER);
3630 %}
3631 
3632 // Long Immediate: low 32-bit mask
3633 operand immL_32bits() %{
3634   predicate(n->get_long() == 0xFFFFFFFFL);
3635   match(ConL);
3636   op_cost(0);
3637 
3638   format %{ %}
3639   interface(CONST_INTER);
3640 %}
3641 
3642 // Long Immediate: cheap (materialize in <= 3 instructions)
3643 operand immL_cheap() %{
3644   predicate(!VM_Version::is_niagara_plus() || MacroAssembler::insts_for_set64(n->get_long()) <= 3);
3645   match(ConL);
3646   op_cost(0);
3647 
3648   format %{ %}
3649   interface(CONST_INTER);
3650 %}
3651 
3652 // Long Immediate: expensive (materialize in > 3 instructions)
3653 operand immL_expensive() %{
3654   predicate(VM_Version::is_niagara_plus() && MacroAssembler::insts_for_set64(n->get_long()) > 3);
3655   match(ConL);
3656   op_cost(0);
3657 
3658   format %{ %}
3659   interface(CONST_INTER);
3660 %}
3661 
3662 // Double Immediate
3663 operand immD() %{
3664   match(ConD);
3665 
3666   op_cost(40);
3667   format %{ %}
3668   interface(CONST_INTER);
3669 %}
3670 
3671 operand immD0() %{
3672 #ifdef _LP64
3673   // on 64-bit architectures this comparision is faster
3674   predicate(jlong_cast(n->getd()) == 0);
3675 #else
3676   predicate((n->getd() == 0) && (fpclass(n->getd()) == FP_PZERO));
3677 #endif
3678   match(ConD);
3679 
3680   op_cost(0);
3681   format %{ %}
3682   interface(CONST_INTER);
3683 %}
3684 
3685 // Float Immediate
3686 operand immF() %{
3687   match(ConF);
3688 
3689   op_cost(20);
3690   format %{ %}
3691   interface(CONST_INTER);
3692 %}
3693 
3694 // Float Immediate: 0
3695 operand immF0() %{
3696   predicate((n->getf() == 0) && (fpclass(n->getf()) == FP_PZERO));
3697   match(ConF);
3698 
3699   op_cost(0);
3700   format %{ %}
3701   interface(CONST_INTER);
3702 %}
3703 
3704 // Integer Register Operands
3705 // Integer Register
3706 operand iRegI() %{
3707   constraint(ALLOC_IN_RC(int_reg));
3708   match(RegI);
3709 
3710   match(notemp_iRegI);
3711   match(g1RegI);
3712   match(o0RegI);
3713   match(iRegIsafe);
3714 
3715   format %{ %}
3716   interface(REG_INTER);
3717 %}
3718 
3719 operand notemp_iRegI() %{
3720   constraint(ALLOC_IN_RC(notemp_int_reg));
3721   match(RegI);
3722 
3723   match(o0RegI);
3724 
3725   format %{ %}
3726   interface(REG_INTER);
3727 %}
3728 
3729 operand o0RegI() %{
3730   constraint(ALLOC_IN_RC(o0_regI));
3731   match(iRegI);
3732 
3733   format %{ %}
3734   interface(REG_INTER);
3735 %}
3736 
3737 // Pointer Register
3738 operand iRegP() %{
3739   constraint(ALLOC_IN_RC(ptr_reg));
3740   match(RegP);
3741 
3742   match(lock_ptr_RegP);
3743   match(g1RegP);
3744   match(g2RegP);
3745   match(g3RegP);
3746   match(g4RegP);
3747   match(i0RegP);
3748   match(o0RegP);
3749   match(o1RegP);
3750   match(l7RegP);
3751 
3752   format %{ %}
3753   interface(REG_INTER);
3754 %}
3755 
3756 operand sp_ptr_RegP() %{
3757   constraint(ALLOC_IN_RC(sp_ptr_reg));
3758   match(RegP);
3759   match(iRegP);
3760 
3761   format %{ %}
3762   interface(REG_INTER);
3763 %}
3764 
3765 operand lock_ptr_RegP() %{
3766   constraint(ALLOC_IN_RC(lock_ptr_reg));
3767   match(RegP);
3768   match(i0RegP);
3769   match(o0RegP);
3770   match(o1RegP);
3771   match(l7RegP);
3772 
3773   format %{ %}
3774   interface(REG_INTER);
3775 %}
3776 
3777 operand g1RegP() %{
3778   constraint(ALLOC_IN_RC(g1_regP));
3779   match(iRegP);
3780 
3781   format %{ %}
3782   interface(REG_INTER);
3783 %}
3784 
3785 operand g2RegP() %{
3786   constraint(ALLOC_IN_RC(g2_regP));
3787   match(iRegP);
3788 
3789   format %{ %}
3790   interface(REG_INTER);
3791 %}
3792 
3793 operand g3RegP() %{
3794   constraint(ALLOC_IN_RC(g3_regP));
3795   match(iRegP);
3796 
3797   format %{ %}
3798   interface(REG_INTER);
3799 %}
3800 
3801 operand g1RegI() %{
3802   constraint(ALLOC_IN_RC(g1_regI));
3803   match(iRegI);
3804 
3805   format %{ %}
3806   interface(REG_INTER);
3807 %}
3808 
3809 operand g3RegI() %{
3810   constraint(ALLOC_IN_RC(g3_regI));
3811   match(iRegI);
3812 
3813   format %{ %}
3814   interface(REG_INTER);
3815 %}
3816 
3817 operand g4RegI() %{
3818   constraint(ALLOC_IN_RC(g4_regI));
3819   match(iRegI);
3820 
3821   format %{ %}
3822   interface(REG_INTER);
3823 %}
3824 
3825 operand g4RegP() %{
3826   constraint(ALLOC_IN_RC(g4_regP));
3827   match(iRegP);
3828 
3829   format %{ %}
3830   interface(REG_INTER);
3831 %}
3832 
3833 operand i0RegP() %{
3834   constraint(ALLOC_IN_RC(i0_regP));
3835   match(iRegP);
3836 
3837   format %{ %}
3838   interface(REG_INTER);
3839 %}
3840 
3841 operand o0RegP() %{
3842   constraint(ALLOC_IN_RC(o0_regP));
3843   match(iRegP);
3844 
3845   format %{ %}
3846   interface(REG_INTER);
3847 %}
3848 
3849 operand o1RegP() %{
3850   constraint(ALLOC_IN_RC(o1_regP));
3851   match(iRegP);
3852 
3853   format %{ %}
3854   interface(REG_INTER);
3855 %}
3856 
3857 operand o2RegP() %{
3858   constraint(ALLOC_IN_RC(o2_regP));
3859   match(iRegP);
3860 
3861   format %{ %}
3862   interface(REG_INTER);
3863 %}
3864 
3865 operand o7RegP() %{
3866   constraint(ALLOC_IN_RC(o7_regP));
3867   match(iRegP);
3868 
3869   format %{ %}
3870   interface(REG_INTER);
3871 %}
3872 
3873 operand l7RegP() %{
3874   constraint(ALLOC_IN_RC(l7_regP));
3875   match(iRegP);
3876 
3877   format %{ %}
3878   interface(REG_INTER);
3879 %}
3880 
3881 operand o7RegI() %{
3882   constraint(ALLOC_IN_RC(o7_regI));
3883   match(iRegI);
3884 
3885   format %{ %}
3886   interface(REG_INTER);
3887 %}
3888 
3889 operand iRegN() %{
3890   constraint(ALLOC_IN_RC(int_reg));
3891   match(RegN);
3892 
3893   format %{ %}
3894   interface(REG_INTER);
3895 %}
3896 
3897 // Long Register
3898 operand iRegL() %{
3899   constraint(ALLOC_IN_RC(long_reg));
3900   match(RegL);
3901 
3902   format %{ %}
3903   interface(REG_INTER);
3904 %}
3905 
3906 operand o2RegL() %{
3907   constraint(ALLOC_IN_RC(o2_regL));
3908   match(iRegL);
3909 
3910   format %{ %}
3911   interface(REG_INTER);
3912 %}
3913 
3914 operand o7RegL() %{
3915   constraint(ALLOC_IN_RC(o7_regL));
3916   match(iRegL);
3917 
3918   format %{ %}
3919   interface(REG_INTER);
3920 %}
3921 
3922 operand g1RegL() %{
3923   constraint(ALLOC_IN_RC(g1_regL));
3924   match(iRegL);
3925 
3926   format %{ %}
3927   interface(REG_INTER);
3928 %}
3929 
3930 operand g3RegL() %{
3931   constraint(ALLOC_IN_RC(g3_regL));
3932   match(iRegL);
3933 
3934   format %{ %}
3935   interface(REG_INTER);
3936 %}
3937 
3938 // Int Register safe
3939 // This is 64bit safe
3940 operand iRegIsafe() %{
3941   constraint(ALLOC_IN_RC(long_reg));
3942 
3943   match(iRegI);
3944 
3945   format %{ %}
3946   interface(REG_INTER);
3947 %}
3948 
3949 // Condition Code Flag Register
3950 operand flagsReg() %{
3951   constraint(ALLOC_IN_RC(int_flags));
3952   match(RegFlags);
3953 
3954   format %{ "ccr" %} // both ICC and XCC
3955   interface(REG_INTER);
3956 %}
3957 
3958 // Condition Code Register, unsigned comparisons.
3959 operand flagsRegU() %{
3960   constraint(ALLOC_IN_RC(int_flags));
3961   match(RegFlags);
3962 
3963   format %{ "icc_U" %}
3964   interface(REG_INTER);
3965 %}
3966 
3967 // Condition Code Register, pointer comparisons.
3968 operand flagsRegP() %{
3969   constraint(ALLOC_IN_RC(int_flags));
3970   match(RegFlags);
3971 
3972 #ifdef _LP64
3973   format %{ "xcc_P" %}
3974 #else
3975   format %{ "icc_P" %}
3976 #endif
3977   interface(REG_INTER);
3978 %}
3979 
3980 // Condition Code Register, long comparisons.
3981 operand flagsRegL() %{
3982   constraint(ALLOC_IN_RC(int_flags));
3983   match(RegFlags);
3984 
3985   format %{ "xcc_L" %}
3986   interface(REG_INTER);
3987 %}
3988 
3989 // Condition Code Register, floating comparisons, unordered same as "less".
3990 operand flagsRegF() %{
3991   constraint(ALLOC_IN_RC(float_flags));
3992   match(RegFlags);
3993   match(flagsRegF0);
3994 
3995   format %{ %}
3996   interface(REG_INTER);
3997 %}
3998 
3999 operand flagsRegF0() %{
4000   constraint(ALLOC_IN_RC(float_flag0));
4001   match(RegFlags);
4002 
4003   format %{ %}
4004   interface(REG_INTER);
4005 %}
4006 
4007 
4008 // Condition Code Flag Register used by long compare
4009 operand flagsReg_long_LTGE() %{
4010   constraint(ALLOC_IN_RC(int_flags));
4011   match(RegFlags);
4012   format %{ "icc_LTGE" %}
4013   interface(REG_INTER);
4014 %}
4015 operand flagsReg_long_EQNE() %{
4016   constraint(ALLOC_IN_RC(int_flags));
4017   match(RegFlags);
4018   format %{ "icc_EQNE" %}
4019   interface(REG_INTER);
4020 %}
4021 operand flagsReg_long_LEGT() %{
4022   constraint(ALLOC_IN_RC(int_flags));
4023   match(RegFlags);
4024   format %{ "icc_LEGT" %}
4025   interface(REG_INTER);
4026 %}
4027 
4028 
4029 operand regD() %{
4030   constraint(ALLOC_IN_RC(dflt_reg));
4031   match(RegD);
4032 
4033   match(regD_low);
4034 
4035   format %{ %}
4036   interface(REG_INTER);
4037 %}
4038 
4039 operand regF() %{
4040   constraint(ALLOC_IN_RC(sflt_reg));
4041   match(RegF);
4042 
4043   format %{ %}
4044   interface(REG_INTER);
4045 %}
4046 
4047 operand regD_low() %{
4048   constraint(ALLOC_IN_RC(dflt_low_reg));
4049   match(regD);
4050 
4051   format %{ %}
4052   interface(REG_INTER);
4053 %}
4054 
4055 // Special Registers
4056 
4057 // Method Register
4058 operand inline_cache_regP(iRegP reg) %{
4059   constraint(ALLOC_IN_RC(g5_regP)); // G5=inline_cache_reg but uses 2 bits instead of 1
4060   match(reg);
4061   format %{ %}
4062   interface(REG_INTER);
4063 %}
4064 
4065 operand interpreter_method_oop_regP(iRegP reg) %{
4066   constraint(ALLOC_IN_RC(g5_regP)); // G5=interpreter_method_oop_reg but uses 2 bits instead of 1
4067   match(reg);
4068   format %{ %}
4069   interface(REG_INTER);
4070 %}
4071 
4072 
4073 //----------Complex Operands---------------------------------------------------
4074 // Indirect Memory Reference
4075 operand indirect(sp_ptr_RegP reg) %{
4076   constraint(ALLOC_IN_RC(sp_ptr_reg));
4077   match(reg);
4078 
4079   op_cost(100);
4080   format %{ "[$reg]" %}
4081   interface(MEMORY_INTER) %{
4082     base($reg);
4083     index(0x0);
4084     scale(0x0);
4085     disp(0x0);
4086   %}
4087 %}
4088 
4089 // Indirect with simm13 Offset
4090 operand indOffset13(sp_ptr_RegP reg, immX13 offset) %{
4091   constraint(ALLOC_IN_RC(sp_ptr_reg));
4092   match(AddP reg offset);
4093 
4094   op_cost(100);
4095   format %{ "[$reg + $offset]" %}
4096   interface(MEMORY_INTER) %{
4097     base($reg);
4098     index(0x0);
4099     scale(0x0);
4100     disp($offset);
4101   %}
4102 %}
4103 
4104 // Indirect with simm13 Offset minus 7
4105 operand indOffset13m7(sp_ptr_RegP reg, immX13m7 offset) %{
4106   constraint(ALLOC_IN_RC(sp_ptr_reg));
4107   match(AddP reg offset);
4108 
4109   op_cost(100);
4110   format %{ "[$reg + $offset]" %}
4111   interface(MEMORY_INTER) %{
4112     base($reg);
4113     index(0x0);
4114     scale(0x0);
4115     disp($offset);
4116   %}
4117 %}
4118 
4119 // Note:  Intel has a swapped version also, like this:
4120 //operand indOffsetX(iRegI reg, immP offset) %{
4121 //  constraint(ALLOC_IN_RC(int_reg));
4122 //  match(AddP offset reg);
4123 //
4124 //  op_cost(100);
4125 //  format %{ "[$reg + $offset]" %}
4126 //  interface(MEMORY_INTER) %{
4127 //    base($reg);
4128 //    index(0x0);
4129 //    scale(0x0);
4130 //    disp($offset);
4131 //  %}
4132 //%}
4133 //// However, it doesn't make sense for SPARC, since
4134 // we have no particularly good way to embed oops in
4135 // single instructions.
4136 
4137 // Indirect with Register Index
4138 operand indIndex(iRegP addr, iRegX index) %{
4139   constraint(ALLOC_IN_RC(ptr_reg));
4140   match(AddP addr index);
4141 
4142   op_cost(100);
4143   format %{ "[$addr + $index]" %}
4144   interface(MEMORY_INTER) %{
4145     base($addr);
4146     index($index);
4147     scale(0x0);
4148     disp(0x0);
4149   %}
4150 %}
4151 
4152 //----------Special Memory Operands--------------------------------------------
4153 // Stack Slot Operand - This operand is used for loading and storing temporary
4154 //                      values on the stack where a match requires a value to
4155 //                      flow through memory.
4156 operand stackSlotI(sRegI reg) %{
4157   constraint(ALLOC_IN_RC(stack_slots));
4158   op_cost(100);
4159   //match(RegI);
4160   format %{ "[$reg]" %}
4161   interface(MEMORY_INTER) %{
4162     base(0xE);   // R_SP
4163     index(0x0);
4164     scale(0x0);
4165     disp($reg);  // Stack Offset
4166   %}
4167 %}
4168 
4169 operand stackSlotP(sRegP reg) %{
4170   constraint(ALLOC_IN_RC(stack_slots));
4171   op_cost(100);
4172   //match(RegP);
4173   format %{ "[$reg]" %}
4174   interface(MEMORY_INTER) %{
4175     base(0xE);   // R_SP
4176     index(0x0);
4177     scale(0x0);
4178     disp($reg);  // Stack Offset
4179   %}
4180 %}
4181 
4182 operand stackSlotF(sRegF reg) %{
4183   constraint(ALLOC_IN_RC(stack_slots));
4184   op_cost(100);
4185   //match(RegF);
4186   format %{ "[$reg]" %}
4187   interface(MEMORY_INTER) %{
4188     base(0xE);   // R_SP
4189     index(0x0);
4190     scale(0x0);
4191     disp($reg);  // Stack Offset
4192   %}
4193 %}
4194 operand stackSlotD(sRegD reg) %{
4195   constraint(ALLOC_IN_RC(stack_slots));
4196   op_cost(100);
4197   //match(RegD);
4198   format %{ "[$reg]" %}
4199   interface(MEMORY_INTER) %{
4200     base(0xE);   // R_SP
4201     index(0x0);
4202     scale(0x0);
4203     disp($reg);  // Stack Offset
4204   %}
4205 %}
4206 operand stackSlotL(sRegL reg) %{
4207   constraint(ALLOC_IN_RC(stack_slots));
4208   op_cost(100);
4209   //match(RegL);
4210   format %{ "[$reg]" %}
4211   interface(MEMORY_INTER) %{
4212     base(0xE);   // R_SP
4213     index(0x0);
4214     scale(0x0);
4215     disp($reg);  // Stack Offset
4216   %}
4217 %}
4218 
4219 // Operands for expressing Control Flow
4220 // NOTE:  Label is a predefined operand which should not be redefined in
4221 //        the AD file.  It is generically handled within the ADLC.
4222 
4223 //----------Conditional Branch Operands----------------------------------------
4224 // Comparison Op  - This is the operation of the comparison, and is limited to
4225 //                  the following set of codes:
4226 //                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
4227 //
4228 // Other attributes of the comparison, such as unsignedness, are specified
4229 // by the comparison instruction that sets a condition code flags register.
4230 // That result is represented by a flags operand whose subtype is appropriate
4231 // to the unsignedness (etc.) of the comparison.
4232 //
4233 // Later, the instruction which matches both the Comparison Op (a Bool) and
4234 // the flags (produced by the Cmp) specifies the coding of the comparison op
4235 // by matching a specific subtype of Bool operand below, such as cmpOpU.
4236 
4237 operand cmpOp() %{
4238   match(Bool);
4239 
4240   format %{ "" %}
4241   interface(COND_INTER) %{
4242     equal(0x1);
4243     not_equal(0x9);
4244     less(0x3);
4245     greater_equal(0xB);
4246     less_equal(0x2);
4247     greater(0xA);
4248   %}
4249 %}
4250 
4251 // Comparison Op, unsigned
4252 operand cmpOpU() %{
4253   match(Bool);
4254 
4255   format %{ "u" %}
4256   interface(COND_INTER) %{
4257     equal(0x1);
4258     not_equal(0x9);
4259     less(0x5);
4260     greater_equal(0xD);
4261     less_equal(0x4);
4262     greater(0xC);
4263   %}
4264 %}
4265 
4266 // Comparison Op, pointer (same as unsigned)
4267 operand cmpOpP() %{
4268   match(Bool);
4269 
4270   format %{ "p" %}
4271   interface(COND_INTER) %{
4272     equal(0x1);
4273     not_equal(0x9);
4274     less(0x5);
4275     greater_equal(0xD);
4276     less_equal(0x4);
4277     greater(0xC);
4278   %}
4279 %}
4280 
4281 // Comparison Op, branch-register encoding
4282 operand cmpOp_reg() %{
4283   match(Bool);
4284 
4285   format %{ "" %}
4286   interface(COND_INTER) %{
4287     equal        (0x1);
4288     not_equal    (0x5);
4289     less         (0x3);
4290     greater_equal(0x7);
4291     less_equal   (0x2);
4292     greater      (0x6);
4293   %}
4294 %}
4295 
4296 // Comparison Code, floating, unordered same as less
4297 operand cmpOpF() %{
4298   match(Bool);
4299 
4300   format %{ "fl" %}
4301   interface(COND_INTER) %{
4302     equal(0x9);
4303     not_equal(0x1);
4304     less(0x3);
4305     greater_equal(0xB);
4306     less_equal(0xE);
4307     greater(0x6);
4308   %}
4309 %}
4310 
4311 // Used by long compare
4312 operand cmpOp_commute() %{
4313   match(Bool);
4314 
4315   format %{ "" %}
4316   interface(COND_INTER) %{
4317     equal(0x1);
4318     not_equal(0x9);
4319     less(0xA);
4320     greater_equal(0x2);
4321     less_equal(0xB);
4322     greater(0x3);
4323   %}
4324 %}
4325 
4326 //----------OPERAND CLASSES----------------------------------------------------
4327 // Operand Classes are groups of operands that are used to simplify
4328 // instruction definitions by not requiring the AD writer to specify separate
4329 // instructions for every form of operand when the instruction accepts
4330 // multiple operand types with the same basic encoding and format.  The classic
4331 // case of this is memory operands.
4332 opclass memory( indirect, indOffset13, indIndex );
4333 opclass indIndexMemory( indIndex );
4334 
4335 //----------PIPELINE-----------------------------------------------------------
4336 pipeline %{
4337 
4338 //----------ATTRIBUTES---------------------------------------------------------
4339 attributes %{
4340   fixed_size_instructions;           // Fixed size instructions
4341   branch_has_delay_slot;             // Branch has delay slot following
4342   max_instructions_per_bundle = 4;   // Up to 4 instructions per bundle
4343   instruction_unit_size = 4;         // An instruction is 4 bytes long
4344   instruction_fetch_unit_size = 16;  // The processor fetches one line
4345   instruction_fetch_units = 1;       // of 16 bytes
4346 
4347   // List of nop instructions
4348   nops( Nop_A0, Nop_A1, Nop_MS, Nop_FA, Nop_BR );
4349 %}
4350 
4351 //----------RESOURCES----------------------------------------------------------
4352 // Resources are the functional units available to the machine
4353 resources(A0, A1, MS, BR, FA, FM, IDIV, FDIV, IALU = A0 | A1);
4354 
4355 //----------PIPELINE DESCRIPTION-----------------------------------------------
4356 // Pipeline Description specifies the stages in the machine's pipeline
4357 
4358 pipe_desc(A, P, F, B, I, J, S, R, E, C, M, W, X, T, D);
4359 
4360 //----------PIPELINE CLASSES---------------------------------------------------
4361 // Pipeline Classes describe the stages in which input and output are
4362 // referenced by the hardware pipeline.
4363 
4364 // Integer ALU reg-reg operation
4365 pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4366     single_instruction;
4367     dst   : E(write);
4368     src1  : R(read);
4369     src2  : R(read);
4370     IALU  : R;
4371 %}
4372 
4373 // Integer ALU reg-reg long operation
4374 pipe_class ialu_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
4375     instruction_count(2);
4376     dst   : E(write);
4377     src1  : R(read);
4378     src2  : R(read);
4379     IALU  : R;
4380     IALU  : R;
4381 %}
4382 
4383 // Integer ALU reg-reg long dependent operation
4384 pipe_class ialu_reg_reg_2_dep(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
4385     instruction_count(1); multiple_bundles;
4386     dst   : E(write);
4387     src1  : R(read);
4388     src2  : R(read);
4389     cr    : E(write);
4390     IALU  : R(2);
4391 %}
4392 
4393 // Integer ALU reg-imm operaion
4394 pipe_class ialu_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4395     single_instruction;
4396     dst   : E(write);
4397     src1  : R(read);
4398     IALU  : R;
4399 %}
4400 
4401 // Integer ALU reg-reg operation with condition code
4402 pipe_class ialu_cc_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
4403     single_instruction;
4404     dst   : E(write);
4405     cr    : E(write);
4406     src1  : R(read);
4407     src2  : R(read);
4408     IALU  : R;
4409 %}
4410 
4411 // Integer ALU reg-imm operation with condition code
4412 pipe_class ialu_cc_reg_imm(iRegI dst, iRegI src1, immI13 src2, flagsReg cr) %{
4413     single_instruction;
4414     dst   : E(write);
4415     cr    : E(write);
4416     src1  : R(read);
4417     IALU  : R;
4418 %}
4419 
4420 // Integer ALU zero-reg operation
4421 pipe_class ialu_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
4422     single_instruction;
4423     dst   : E(write);
4424     src2  : R(read);
4425     IALU  : R;
4426 %}
4427 
4428 // Integer ALU zero-reg operation with condition code only
4429 pipe_class ialu_cconly_zero_reg(flagsReg cr, iRegI src) %{
4430     single_instruction;
4431     cr    : E(write);
4432     src   : R(read);
4433     IALU  : R;
4434 %}
4435 
4436 // Integer ALU reg-reg operation with condition code only
4437 pipe_class ialu_cconly_reg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4438     single_instruction;
4439     cr    : E(write);
4440     src1  : R(read);
4441     src2  : R(read);
4442     IALU  : R;
4443 %}
4444 
4445 // Integer ALU reg-imm operation with condition code only
4446 pipe_class ialu_cconly_reg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4447     single_instruction;
4448     cr    : E(write);
4449     src1  : R(read);
4450     IALU  : R;
4451 %}
4452 
4453 // Integer ALU reg-reg-zero operation with condition code only
4454 pipe_class ialu_cconly_reg_reg_zero(flagsReg cr, iRegI src1, iRegI src2, immI0 zero) %{
4455     single_instruction;
4456     cr    : E(write);
4457     src1  : R(read);
4458     src2  : R(read);
4459     IALU  : R;
4460 %}
4461 
4462 // Integer ALU reg-imm-zero operation with condition code only
4463 pipe_class ialu_cconly_reg_imm_zero(flagsReg cr, iRegI src1, immI13 src2, immI0 zero) %{
4464     single_instruction;
4465     cr    : E(write);
4466     src1  : R(read);
4467     IALU  : R;
4468 %}
4469 
4470 // Integer ALU reg-reg operation with condition code, src1 modified
4471 pipe_class ialu_cc_rwreg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4472     single_instruction;
4473     cr    : E(write);
4474     src1  : E(write);
4475     src1  : R(read);
4476     src2  : R(read);
4477     IALU  : R;
4478 %}
4479 
4480 // Integer ALU reg-imm operation with condition code, src1 modified
4481 pipe_class ialu_cc_rwreg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4482     single_instruction;
4483     cr    : E(write);
4484     src1  : E(write);
4485     src1  : R(read);
4486     IALU  : R;
4487 %}
4488 
4489 pipe_class cmpL_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg cr ) %{
4490     multiple_bundles;
4491     dst   : E(write)+4;
4492     cr    : E(write);
4493     src1  : R(read);
4494     src2  : R(read);
4495     IALU  : R(3);
4496     BR    : R(2);
4497 %}
4498 
4499 // Integer ALU operation
4500 pipe_class ialu_none(iRegI dst) %{
4501     single_instruction;
4502     dst   : E(write);
4503     IALU  : R;
4504 %}
4505 
4506 // Integer ALU reg operation
4507 pipe_class ialu_reg(iRegI dst, iRegI src) %{
4508     single_instruction; may_have_no_code;
4509     dst   : E(write);
4510     src   : R(read);
4511     IALU  : R;
4512 %}
4513 
4514 // Integer ALU reg conditional operation
4515 // This instruction has a 1 cycle stall, and cannot execute
4516 // in the same cycle as the instruction setting the condition
4517 // code. We kludge this by pretending to read the condition code
4518 // 1 cycle earlier, and by marking the functional units as busy
4519 // for 2 cycles with the result available 1 cycle later than
4520 // is really the case.
4521 pipe_class ialu_reg_flags( iRegI op2_out, iRegI op2_in, iRegI op1, flagsReg cr ) %{
4522     single_instruction;
4523     op2_out : C(write);
4524     op1     : R(read);
4525     cr      : R(read);       // This is really E, with a 1 cycle stall
4526     BR      : R(2);
4527     MS      : R(2);
4528 %}
4529 
4530 #ifdef _LP64
4531 pipe_class ialu_clr_and_mover( iRegI dst, iRegP src ) %{
4532     instruction_count(1); multiple_bundles;
4533     dst     : C(write)+1;
4534     src     : R(read)+1;
4535     IALU    : R(1);
4536     BR      : E(2);
4537     MS      : E(2);
4538 %}
4539 #endif
4540 
4541 // Integer ALU reg operation
4542 pipe_class ialu_move_reg_L_to_I(iRegI dst, iRegL src) %{
4543     single_instruction; may_have_no_code;
4544     dst   : E(write);
4545     src   : R(read);
4546     IALU  : R;
4547 %}
4548 pipe_class ialu_move_reg_I_to_L(iRegL dst, iRegI src) %{
4549     single_instruction; may_have_no_code;
4550     dst   : E(write);
4551     src   : R(read);
4552     IALU  : R;
4553 %}
4554 
4555 // Two integer ALU reg operations
4556 pipe_class ialu_reg_2(iRegL dst, iRegL src) %{
4557     instruction_count(2);
4558     dst   : E(write);
4559     src   : R(read);
4560     A0    : R;
4561     A1    : R;
4562 %}
4563 
4564 // Two integer ALU reg operations
4565 pipe_class ialu_move_reg_L_to_L(iRegL dst, iRegL src) %{
4566     instruction_count(2); may_have_no_code;
4567     dst   : E(write);
4568     src   : R(read);
4569     A0    : R;
4570     A1    : R;
4571 %}
4572 
4573 // Integer ALU imm operation
4574 pipe_class ialu_imm(iRegI dst, immI13 src) %{
4575     single_instruction;
4576     dst   : E(write);
4577     IALU  : R;
4578 %}
4579 
4580 // Integer ALU reg-reg with carry operation
4581 pipe_class ialu_reg_reg_cy(iRegI dst, iRegI src1, iRegI src2, iRegI cy) %{
4582     single_instruction;
4583     dst   : E(write);
4584     src1  : R(read);
4585     src2  : R(read);
4586     IALU  : R;
4587 %}
4588 
4589 // Integer ALU cc operation
4590 pipe_class ialu_cc(iRegI dst, flagsReg cc) %{
4591     single_instruction;
4592     dst   : E(write);
4593     cc    : R(read);
4594     IALU  : R;
4595 %}
4596 
4597 // Integer ALU cc / second IALU operation
4598 pipe_class ialu_reg_ialu( iRegI dst, iRegI src ) %{
4599     instruction_count(1); multiple_bundles;
4600     dst   : E(write)+1;
4601     src   : R(read);
4602     IALU  : R;
4603 %}
4604 
4605 // Integer ALU cc / second IALU operation
4606 pipe_class ialu_reg_reg_ialu( iRegI dst, iRegI p, iRegI q ) %{
4607     instruction_count(1); multiple_bundles;
4608     dst   : E(write)+1;
4609     p     : R(read);
4610     q     : R(read);
4611     IALU  : R;
4612 %}
4613 
4614 // Integer ALU hi-lo-reg operation
4615 pipe_class ialu_hi_lo_reg(iRegI dst, immI src) %{
4616     instruction_count(1); multiple_bundles;
4617     dst   : E(write)+1;
4618     IALU  : R(2);
4619 %}
4620 
4621 // Float ALU hi-lo-reg operation (with temp)
4622 pipe_class ialu_hi_lo_reg_temp(regF dst, immF src, g3RegP tmp) %{
4623     instruction_count(1); multiple_bundles;
4624     dst   : E(write)+1;
4625     IALU  : R(2);
4626 %}
4627 
4628 // Long Constant
4629 pipe_class loadConL( iRegL dst, immL src ) %{
4630     instruction_count(2); multiple_bundles;
4631     dst   : E(write)+1;
4632     IALU  : R(2);
4633     IALU  : R(2);
4634 %}
4635 
4636 // Pointer Constant
4637 pipe_class loadConP( iRegP dst, immP src ) %{
4638     instruction_count(0); multiple_bundles;
4639     fixed_latency(6);
4640 %}
4641 
4642 // Polling Address
4643 pipe_class loadConP_poll( iRegP dst, immP_poll src ) %{
4644 #ifdef _LP64
4645     instruction_count(0); multiple_bundles;
4646     fixed_latency(6);
4647 #else
4648     dst   : E(write);
4649     IALU  : R;
4650 #endif
4651 %}
4652 
4653 // Long Constant small
4654 pipe_class loadConLlo( iRegL dst, immL src ) %{
4655     instruction_count(2);
4656     dst   : E(write);
4657     IALU  : R;
4658     IALU  : R;
4659 %}
4660 
4661 // [PHH] This is wrong for 64-bit.  See LdImmF/D.
4662 pipe_class loadConFD(regF dst, immF src, g3RegP tmp) %{
4663     instruction_count(1); multiple_bundles;
4664     src   : R(read);
4665     dst   : M(write)+1;
4666     IALU  : R;
4667     MS    : E;
4668 %}
4669 
4670 // Integer ALU nop operation
4671 pipe_class ialu_nop() %{
4672     single_instruction;
4673     IALU  : R;
4674 %}
4675 
4676 // Integer ALU nop operation
4677 pipe_class ialu_nop_A0() %{
4678     single_instruction;
4679     A0    : R;
4680 %}
4681 
4682 // Integer ALU nop operation
4683 pipe_class ialu_nop_A1() %{
4684     single_instruction;
4685     A1    : R;
4686 %}
4687 
4688 // Integer Multiply reg-reg operation
4689 pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4690     single_instruction;
4691     dst   : E(write);
4692     src1  : R(read);
4693     src2  : R(read);
4694     MS    : R(5);
4695 %}
4696 
4697 // Integer Multiply reg-imm operation
4698 pipe_class imul_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4699     single_instruction;
4700     dst   : E(write);
4701     src1  : R(read);
4702     MS    : R(5);
4703 %}
4704 
4705 pipe_class mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4706     single_instruction;
4707     dst   : E(write)+4;
4708     src1  : R(read);
4709     src2  : R(read);
4710     MS    : R(6);
4711 %}
4712 
4713 pipe_class mulL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4714     single_instruction;
4715     dst   : E(write)+4;
4716     src1  : R(read);
4717     MS    : R(6);
4718 %}
4719 
4720 // Integer Divide reg-reg
4721 pipe_class sdiv_reg_reg(iRegI dst, iRegI src1, iRegI src2, iRegI temp, flagsReg cr) %{
4722     instruction_count(1); multiple_bundles;
4723     dst   : E(write);
4724     temp  : E(write);
4725     src1  : R(read);
4726     src2  : R(read);
4727     temp  : R(read);
4728     MS    : R(38);
4729 %}
4730 
4731 // Integer Divide reg-imm
4732 pipe_class sdiv_reg_imm(iRegI dst, iRegI src1, immI13 src2, iRegI temp, flagsReg cr) %{
4733     instruction_count(1); multiple_bundles;
4734     dst   : E(write);
4735     temp  : E(write);
4736     src1  : R(read);
4737     temp  : R(read);
4738     MS    : R(38);
4739 %}
4740 
4741 // Long Divide
4742 pipe_class divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4743     dst  : E(write)+71;
4744     src1 : R(read);
4745     src2 : R(read)+1;
4746     MS   : R(70);
4747 %}
4748 
4749 pipe_class divL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4750     dst  : E(write)+71;
4751     src1 : R(read);
4752     MS   : R(70);
4753 %}
4754 
4755 // Floating Point Add Float
4756 pipe_class faddF_reg_reg(regF dst, regF src1, regF src2) %{
4757     single_instruction;
4758     dst   : X(write);
4759     src1  : E(read);
4760     src2  : E(read);
4761     FA    : R;
4762 %}
4763 
4764 // Floating Point Add Double
4765 pipe_class faddD_reg_reg(regD dst, regD src1, regD src2) %{
4766     single_instruction;
4767     dst   : X(write);
4768     src1  : E(read);
4769     src2  : E(read);
4770     FA    : R;
4771 %}
4772 
4773 // Floating Point Conditional Move based on integer flags
4774 pipe_class int_conditional_float_move (cmpOp cmp, flagsReg cr, regF dst, regF src) %{
4775     single_instruction;
4776     dst   : X(write);
4777     src   : E(read);
4778     cr    : R(read);
4779     FA    : R(2);
4780     BR    : R(2);
4781 %}
4782 
4783 // Floating Point Conditional Move based on integer flags
4784 pipe_class int_conditional_double_move (cmpOp cmp, flagsReg cr, regD dst, regD src) %{
4785     single_instruction;
4786     dst   : X(write);
4787     src   : E(read);
4788     cr    : R(read);
4789     FA    : R(2);
4790     BR    : R(2);
4791 %}
4792 
4793 // Floating Point Multiply Float
4794 pipe_class fmulF_reg_reg(regF dst, regF src1, regF src2) %{
4795     single_instruction;
4796     dst   : X(write);
4797     src1  : E(read);
4798     src2  : E(read);
4799     FM    : R;
4800 %}
4801 
4802 // Floating Point Multiply Double
4803 pipe_class fmulD_reg_reg(regD dst, regD src1, regD src2) %{
4804     single_instruction;
4805     dst   : X(write);
4806     src1  : E(read);
4807     src2  : E(read);
4808     FM    : R;
4809 %}
4810 
4811 // Floating Point Divide Float
4812 pipe_class fdivF_reg_reg(regF dst, regF src1, regF src2) %{
4813     single_instruction;
4814     dst   : X(write);
4815     src1  : E(read);
4816     src2  : E(read);
4817     FM    : R;
4818     FDIV  : C(14);
4819 %}
4820 
4821 // Floating Point Divide Double
4822 pipe_class fdivD_reg_reg(regD dst, regD src1, regD src2) %{
4823     single_instruction;
4824     dst   : X(write);
4825     src1  : E(read);
4826     src2  : E(read);
4827     FM    : R;
4828     FDIV  : C(17);
4829 %}
4830 
4831 // Floating Point Move/Negate/Abs Float
4832 pipe_class faddF_reg(regF dst, regF src) %{
4833     single_instruction;
4834     dst   : W(write);
4835     src   : E(read);
4836     FA    : R(1);
4837 %}
4838 
4839 // Floating Point Move/Negate/Abs Double
4840 pipe_class faddD_reg(regD dst, regD src) %{
4841     single_instruction;
4842     dst   : W(write);
4843     src   : E(read);
4844     FA    : R;
4845 %}
4846 
4847 // Floating Point Convert F->D
4848 pipe_class fcvtF2D(regD dst, regF src) %{
4849     single_instruction;
4850     dst   : X(write);
4851     src   : E(read);
4852     FA    : R;
4853 %}
4854 
4855 // Floating Point Convert I->D
4856 pipe_class fcvtI2D(regD dst, regF src) %{
4857     single_instruction;
4858     dst   : X(write);
4859     src   : E(read);
4860     FA    : R;
4861 %}
4862 
4863 // Floating Point Convert LHi->D
4864 pipe_class fcvtLHi2D(regD dst, regD src) %{
4865     single_instruction;
4866     dst   : X(write);
4867     src   : E(read);
4868     FA    : R;
4869 %}
4870 
4871 // Floating Point Convert L->D
4872 pipe_class fcvtL2D(regD dst, regF src) %{
4873     single_instruction;
4874     dst   : X(write);
4875     src   : E(read);
4876     FA    : R;
4877 %}
4878 
4879 // Floating Point Convert L->F
4880 pipe_class fcvtL2F(regD dst, regF src) %{
4881     single_instruction;
4882     dst   : X(write);
4883     src   : E(read);
4884     FA    : R;
4885 %}
4886 
4887 // Floating Point Convert D->F
4888 pipe_class fcvtD2F(regD dst, regF src) %{
4889     single_instruction;
4890     dst   : X(write);
4891     src   : E(read);
4892     FA    : R;
4893 %}
4894 
4895 // Floating Point Convert I->L
4896 pipe_class fcvtI2L(regD dst, regF src) %{
4897     single_instruction;
4898     dst   : X(write);
4899     src   : E(read);
4900     FA    : R;
4901 %}
4902 
4903 // Floating Point Convert D->F
4904 pipe_class fcvtD2I(regF dst, regD src, flagsReg cr) %{
4905     instruction_count(1); multiple_bundles;
4906     dst   : X(write)+6;
4907     src   : E(read);
4908     FA    : R;
4909 %}
4910 
4911 // Floating Point Convert D->L
4912 pipe_class fcvtD2L(regD dst, regD src, flagsReg cr) %{
4913     instruction_count(1); multiple_bundles;
4914     dst   : X(write)+6;
4915     src   : E(read);
4916     FA    : R;
4917 %}
4918 
4919 // Floating Point Convert F->I
4920 pipe_class fcvtF2I(regF dst, regF src, flagsReg cr) %{
4921     instruction_count(1); multiple_bundles;
4922     dst   : X(write)+6;
4923     src   : E(read);
4924     FA    : R;
4925 %}
4926 
4927 // Floating Point Convert F->L
4928 pipe_class fcvtF2L(regD dst, regF src, flagsReg cr) %{
4929     instruction_count(1); multiple_bundles;
4930     dst   : X(write)+6;
4931     src   : E(read);
4932     FA    : R;
4933 %}
4934 
4935 // Floating Point Convert I->F
4936 pipe_class fcvtI2F(regF dst, regF src) %{
4937     single_instruction;
4938     dst   : X(write);
4939     src   : E(read);
4940     FA    : R;
4941 %}
4942 
4943 // Floating Point Compare
4944 pipe_class faddF_fcc_reg_reg_zero(flagsRegF cr, regF src1, regF src2, immI0 zero) %{
4945     single_instruction;
4946     cr    : X(write);
4947     src1  : E(read);
4948     src2  : E(read);
4949     FA    : R;
4950 %}
4951 
4952 // Floating Point Compare
4953 pipe_class faddD_fcc_reg_reg_zero(flagsRegF cr, regD src1, regD src2, immI0 zero) %{
4954     single_instruction;
4955     cr    : X(write);
4956     src1  : E(read);
4957     src2  : E(read);
4958     FA    : R;
4959 %}
4960 
4961 // Floating Add Nop
4962 pipe_class fadd_nop() %{
4963     single_instruction;
4964     FA  : R;
4965 %}
4966 
4967 // Integer Store to Memory
4968 pipe_class istore_mem_reg(memory mem, iRegI src) %{
4969     single_instruction;
4970     mem   : R(read);
4971     src   : C(read);
4972     MS    : R;
4973 %}
4974 
4975 // Integer Store to Memory
4976 pipe_class istore_mem_spORreg(memory mem, sp_ptr_RegP src) %{
4977     single_instruction;
4978     mem   : R(read);
4979     src   : C(read);
4980     MS    : R;
4981 %}
4982 
4983 // Integer Store Zero to Memory
4984 pipe_class istore_mem_zero(memory mem, immI0 src) %{
4985     single_instruction;
4986     mem   : R(read);
4987     MS    : R;
4988 %}
4989 
4990 // Special Stack Slot Store
4991 pipe_class istore_stk_reg(stackSlotI stkSlot, iRegI src) %{
4992     single_instruction;
4993     stkSlot : R(read);
4994     src     : C(read);
4995     MS      : R;
4996 %}
4997 
4998 // Special Stack Slot Store
4999 pipe_class lstoreI_stk_reg(stackSlotL stkSlot, iRegI src) %{
5000     instruction_count(2); multiple_bundles;
5001     stkSlot : R(read);
5002     src     : C(read);
5003     MS      : R(2);
5004 %}
5005 
5006 // Float Store
5007 pipe_class fstoreF_mem_reg(memory mem, RegF src) %{
5008     single_instruction;
5009     mem : R(read);
5010     src : C(read);
5011     MS  : R;
5012 %}
5013 
5014 // Float Store
5015 pipe_class fstoreF_mem_zero(memory mem, immF0 src) %{
5016     single_instruction;
5017     mem : R(read);
5018     MS  : R;
5019 %}
5020 
5021 // Double Store
5022 pipe_class fstoreD_mem_reg(memory mem, RegD src) %{
5023     instruction_count(1);
5024     mem : R(read);
5025     src : C(read);
5026     MS  : R;
5027 %}
5028 
5029 // Double Store
5030 pipe_class fstoreD_mem_zero(memory mem, immD0 src) %{
5031     single_instruction;
5032     mem : R(read);
5033     MS  : R;
5034 %}
5035 
5036 // Special Stack Slot Float Store
5037 pipe_class fstoreF_stk_reg(stackSlotI stkSlot, RegF src) %{
5038     single_instruction;
5039     stkSlot : R(read);
5040     src     : C(read);
5041     MS      : R;
5042 %}
5043 
5044 // Special Stack Slot Double Store
5045 pipe_class fstoreD_stk_reg(stackSlotI stkSlot, RegD src) %{
5046     single_instruction;
5047     stkSlot : R(read);
5048     src     : C(read);
5049     MS      : R;
5050 %}
5051 
5052 // Integer Load (when sign bit propagation not needed)
5053 pipe_class iload_mem(iRegI dst, memory mem) %{
5054     single_instruction;
5055     mem : R(read);
5056     dst : C(write);
5057     MS  : R;
5058 %}
5059 
5060 // Integer Load from stack operand
5061 pipe_class iload_stkD(iRegI dst, stackSlotD mem ) %{
5062     single_instruction;
5063     mem : R(read);
5064     dst : C(write);
5065     MS  : R;
5066 %}
5067 
5068 // Integer Load (when sign bit propagation or masking is needed)
5069 pipe_class iload_mask_mem(iRegI dst, memory mem) %{
5070     single_instruction;
5071     mem : R(read);
5072     dst : M(write);
5073     MS  : R;
5074 %}
5075 
5076 // Float Load
5077 pipe_class floadF_mem(regF dst, memory mem) %{
5078     single_instruction;
5079     mem : R(read);
5080     dst : M(write);
5081     MS  : R;
5082 %}
5083 
5084 // Float Load
5085 pipe_class floadD_mem(regD dst, memory mem) %{
5086     instruction_count(1); multiple_bundles; // Again, unaligned argument is only multiple case
5087     mem : R(read);
5088     dst : M(write);
5089     MS  : R;
5090 %}
5091 
5092 // Float Load
5093 pipe_class floadF_stk(regF dst, stackSlotI stkSlot) %{
5094     single_instruction;
5095     stkSlot : R(read);
5096     dst : M(write);
5097     MS  : R;
5098 %}
5099 
5100 // Float Load
5101 pipe_class floadD_stk(regD dst, stackSlotI stkSlot) %{
5102     single_instruction;
5103     stkSlot : R(read);
5104     dst : M(write);
5105     MS  : R;
5106 %}
5107 
5108 // Memory Nop
5109 pipe_class mem_nop() %{
5110     single_instruction;
5111     MS  : R;
5112 %}
5113 
5114 pipe_class sethi(iRegP dst, immI src) %{
5115     single_instruction;
5116     dst  : E(write);
5117     IALU : R;
5118 %}
5119 
5120 pipe_class loadPollP(iRegP poll) %{
5121     single_instruction;
5122     poll : R(read);
5123     MS   : R;
5124 %}
5125 
5126 pipe_class br(Universe br, label labl) %{
5127     single_instruction_with_delay_slot;
5128     BR  : R;
5129 %}
5130 
5131 pipe_class br_cc(Universe br, cmpOp cmp, flagsReg cr, label labl) %{
5132     single_instruction_with_delay_slot;
5133     cr    : E(read);
5134     BR    : R;
5135 %}
5136 
5137 pipe_class br_reg(Universe br, cmpOp cmp, iRegI op1, label labl) %{
5138     single_instruction_with_delay_slot;
5139     op1 : E(read);
5140     BR  : R;
5141     MS  : R;
5142 %}
5143 
5144 // Compare and branch
5145 pipe_class cmp_br_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
5146     instruction_count(2); has_delay_slot;
5147     cr    : E(write);
5148     src1  : R(read);
5149     src2  : R(read);
5150     IALU  : R;
5151     BR    : R;
5152 %}
5153 
5154 // Compare and branch
5155 pipe_class cmp_br_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI13 src2, label labl, flagsReg cr) %{
5156     instruction_count(2); has_delay_slot;
5157     cr    : E(write);
5158     src1  : R(read);
5159     IALU  : R;
5160     BR    : R;
5161 %}
5162 
5163 // Compare and branch using cbcond
5164 pipe_class cbcond_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl) %{
5165     single_instruction;
5166     src1  : E(read);
5167     src2  : E(read);
5168     IALU  : R;
5169     BR    : R;
5170 %}
5171 
5172 // Compare and branch using cbcond
5173 pipe_class cbcond_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI5 src2, label labl) %{
5174     single_instruction;
5175     src1  : E(read);
5176     IALU  : R;
5177     BR    : R;
5178 %}
5179 
5180 pipe_class br_fcc(Universe br, cmpOpF cc, flagsReg cr, label labl) %{
5181     single_instruction_with_delay_slot;
5182     cr    : E(read);
5183     BR    : R;
5184 %}
5185 
5186 pipe_class br_nop() %{
5187     single_instruction;
5188     BR  : R;
5189 %}
5190 
5191 pipe_class simple_call(method meth) %{
5192     instruction_count(2); multiple_bundles; force_serialization;
5193     fixed_latency(100);
5194     BR  : R(1);
5195     MS  : R(1);
5196     A0  : R(1);
5197 %}
5198 
5199 pipe_class compiled_call(method meth) %{
5200     instruction_count(1); multiple_bundles; force_serialization;
5201     fixed_latency(100);
5202     MS  : R(1);
5203 %}
5204 
5205 pipe_class call(method meth) %{
5206     instruction_count(0); multiple_bundles; force_serialization;
5207     fixed_latency(100);
5208 %}
5209 
5210 pipe_class tail_call(Universe ignore, label labl) %{
5211     single_instruction; has_delay_slot;
5212     fixed_latency(100);
5213     BR  : R(1);
5214     MS  : R(1);
5215 %}
5216 
5217 pipe_class ret(Universe ignore) %{
5218     single_instruction; has_delay_slot;
5219     BR  : R(1);
5220     MS  : R(1);
5221 %}
5222 
5223 pipe_class ret_poll(g3RegP poll) %{
5224     instruction_count(3); has_delay_slot;
5225     poll : E(read);
5226     MS   : R;
5227 %}
5228 
5229 // The real do-nothing guy
5230 pipe_class empty( ) %{
5231     instruction_count(0);
5232 %}
5233 
5234 pipe_class long_memory_op() %{
5235     instruction_count(0); multiple_bundles; force_serialization;
5236     fixed_latency(25);
5237     MS  : R(1);
5238 %}
5239 
5240 // Check-cast
5241 pipe_class partial_subtype_check_pipe(Universe ignore, iRegP array, iRegP match ) %{
5242     array : R(read);
5243     match  : R(read);
5244     IALU   : R(2);
5245     BR     : R(2);
5246     MS     : R;
5247 %}
5248 
5249 // Convert FPU flags into +1,0,-1
5250 pipe_class floating_cmp( iRegI dst, regF src1, regF src2 ) %{
5251     src1  : E(read);
5252     src2  : E(read);
5253     dst   : E(write);
5254     FA    : R;
5255     MS    : R(2);
5256     BR    : R(2);
5257 %}
5258 
5259 // Compare for p < q, and conditionally add y
5260 pipe_class cadd_cmpltmask( iRegI p, iRegI q, iRegI y ) %{
5261     p     : E(read);
5262     q     : E(read);
5263     y     : E(read);
5264     IALU  : R(3)
5265 %}
5266 
5267 // Perform a compare, then move conditionally in a branch delay slot.
5268 pipe_class min_max( iRegI src2, iRegI srcdst ) %{
5269     src2   : E(read);
5270     srcdst : E(read);
5271     IALU   : R;
5272     BR     : R;
5273 %}
5274 
5275 // Define the class for the Nop node
5276 define %{
5277    MachNop = ialu_nop;
5278 %}
5279 
5280 %}
5281 
5282 //----------INSTRUCTIONS-------------------------------------------------------
5283 
5284 //------------Special Stack Slot instructions - no match rules-----------------
5285 instruct stkI_to_regF(regF dst, stackSlotI src) %{
5286   // No match rule to avoid chain rule match.
5287   effect(DEF dst, USE src);
5288   ins_cost(MEMORY_REF_COST);
5289   size(4);
5290   format %{ "LDF    $src,$dst\t! stkI to regF" %}
5291   opcode(Assembler::ldf_op3);
5292   ins_encode(simple_form3_mem_reg(src, dst));
5293   ins_pipe(floadF_stk);
5294 %}
5295 
5296 instruct stkL_to_regD(regD dst, stackSlotL src) %{
5297   // No match rule to avoid chain rule match.
5298   effect(DEF dst, USE src);
5299   ins_cost(MEMORY_REF_COST);
5300   size(4);
5301   format %{ "LDDF   $src,$dst\t! stkL to regD" %}
5302   opcode(Assembler::lddf_op3);
5303   ins_encode(simple_form3_mem_reg(src, dst));
5304   ins_pipe(floadD_stk);
5305 %}
5306 
5307 instruct regF_to_stkI(stackSlotI dst, regF src) %{
5308   // No match rule to avoid chain rule match.
5309   effect(DEF dst, USE src);
5310   ins_cost(MEMORY_REF_COST);
5311   size(4);
5312   format %{ "STF    $src,$dst\t! regF to stkI" %}
5313   opcode(Assembler::stf_op3);
5314   ins_encode(simple_form3_mem_reg(dst, src));
5315   ins_pipe(fstoreF_stk_reg);
5316 %}
5317 
5318 instruct regD_to_stkL(stackSlotL dst, regD src) %{
5319   // No match rule to avoid chain rule match.
5320   effect(DEF dst, USE src);
5321   ins_cost(MEMORY_REF_COST);
5322   size(4);
5323   format %{ "STDF   $src,$dst\t! regD to stkL" %}
5324   opcode(Assembler::stdf_op3);
5325   ins_encode(simple_form3_mem_reg(dst, src));
5326   ins_pipe(fstoreD_stk_reg);
5327 %}
5328 
5329 instruct regI_to_stkLHi(stackSlotL dst, iRegI src) %{
5330   effect(DEF dst, USE src);
5331   ins_cost(MEMORY_REF_COST*2);
5332   size(8);
5333   format %{ "STW    $src,$dst.hi\t! long\n\t"
5334             "STW    R_G0,$dst.lo" %}
5335   opcode(Assembler::stw_op3);
5336   ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, R_G0));
5337   ins_pipe(lstoreI_stk_reg);
5338 %}
5339 
5340 instruct regL_to_stkD(stackSlotD dst, iRegL src) %{
5341   // No match rule to avoid chain rule match.
5342   effect(DEF dst, USE src);
5343   ins_cost(MEMORY_REF_COST);
5344   size(4);
5345   format %{ "STX    $src,$dst\t! regL to stkD" %}
5346   opcode(Assembler::stx_op3);
5347   ins_encode(simple_form3_mem_reg( dst, src ) );
5348   ins_pipe(istore_stk_reg);
5349 %}
5350 
5351 //---------- Chain stack slots between similar types --------
5352 
5353 // Load integer from stack slot
5354 instruct stkI_to_regI( iRegI dst, stackSlotI src ) %{
5355   match(Set dst src);
5356   ins_cost(MEMORY_REF_COST);
5357 
5358   size(4);
5359   format %{ "LDUW   $src,$dst\t!stk" %}
5360   opcode(Assembler::lduw_op3);
5361   ins_encode(simple_form3_mem_reg( src, dst ) );
5362   ins_pipe(iload_mem);
5363 %}
5364 
5365 // Store integer to stack slot
5366 instruct regI_to_stkI( stackSlotI dst, iRegI src ) %{
5367   match(Set dst src);
5368   ins_cost(MEMORY_REF_COST);
5369 
5370   size(4);
5371   format %{ "STW    $src,$dst\t!stk" %}
5372   opcode(Assembler::stw_op3);
5373   ins_encode(simple_form3_mem_reg( dst, src ) );
5374   ins_pipe(istore_mem_reg);
5375 %}
5376 
5377 // Load long from stack slot
5378 instruct stkL_to_regL( iRegL dst, stackSlotL src ) %{
5379   match(Set dst src);
5380 
5381   ins_cost(MEMORY_REF_COST);
5382   size(4);
5383   format %{ "LDX    $src,$dst\t! long" %}
5384   opcode(Assembler::ldx_op3);
5385   ins_encode(simple_form3_mem_reg( src, dst ) );
5386   ins_pipe(iload_mem);
5387 %}
5388 
5389 // Store long to stack slot
5390 instruct regL_to_stkL(stackSlotL dst, iRegL src) %{
5391   match(Set dst src);
5392 
5393   ins_cost(MEMORY_REF_COST);
5394   size(4);
5395   format %{ "STX    $src,$dst\t! long" %}
5396   opcode(Assembler::stx_op3);
5397   ins_encode(simple_form3_mem_reg( dst, src ) );
5398   ins_pipe(istore_mem_reg);
5399 %}
5400 
5401 #ifdef _LP64
5402 // Load pointer from stack slot, 64-bit encoding
5403 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5404   match(Set dst src);
5405   ins_cost(MEMORY_REF_COST);
5406   size(4);
5407   format %{ "LDX    $src,$dst\t!ptr" %}
5408   opcode(Assembler::ldx_op3);
5409   ins_encode(simple_form3_mem_reg( src, dst ) );
5410   ins_pipe(iload_mem);
5411 %}
5412 
5413 // Store pointer to stack slot
5414 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5415   match(Set dst src);
5416   ins_cost(MEMORY_REF_COST);
5417   size(4);
5418   format %{ "STX    $src,$dst\t!ptr" %}
5419   opcode(Assembler::stx_op3);
5420   ins_encode(simple_form3_mem_reg( dst, src ) );
5421   ins_pipe(istore_mem_reg);
5422 %}
5423 #else // _LP64
5424 // Load pointer from stack slot, 32-bit encoding
5425 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5426   match(Set dst src);
5427   ins_cost(MEMORY_REF_COST);
5428   format %{ "LDUW   $src,$dst\t!ptr" %}
5429   opcode(Assembler::lduw_op3, Assembler::ldst_op);
5430   ins_encode(simple_form3_mem_reg( src, dst ) );
5431   ins_pipe(iload_mem);
5432 %}
5433 
5434 // Store pointer to stack slot
5435 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5436   match(Set dst src);
5437   ins_cost(MEMORY_REF_COST);
5438   format %{ "STW    $src,$dst\t!ptr" %}
5439   opcode(Assembler::stw_op3, Assembler::ldst_op);
5440   ins_encode(simple_form3_mem_reg( dst, src ) );
5441   ins_pipe(istore_mem_reg);
5442 %}
5443 #endif // _LP64
5444 
5445 //------------Special Nop instructions for bundling - no match rules-----------
5446 // Nop using the A0 functional unit
5447 instruct Nop_A0() %{
5448   ins_cost(0);
5449 
5450   format %{ "NOP    ! Alu Pipeline" %}
5451   opcode(Assembler::or_op3, Assembler::arith_op);
5452   ins_encode( form2_nop() );
5453   ins_pipe(ialu_nop_A0);
5454 %}
5455 
5456 // Nop using the A1 functional unit
5457 instruct Nop_A1( ) %{
5458   ins_cost(0);
5459 
5460   format %{ "NOP    ! Alu Pipeline" %}
5461   opcode(Assembler::or_op3, Assembler::arith_op);
5462   ins_encode( form2_nop() );
5463   ins_pipe(ialu_nop_A1);
5464 %}
5465 
5466 // Nop using the memory functional unit
5467 instruct Nop_MS( ) %{
5468   ins_cost(0);
5469 
5470   format %{ "NOP    ! Memory Pipeline" %}
5471   ins_encode( emit_mem_nop );
5472   ins_pipe(mem_nop);
5473 %}
5474 
5475 // Nop using the floating add functional unit
5476 instruct Nop_FA( ) %{
5477   ins_cost(0);
5478 
5479   format %{ "NOP    ! Floating Add Pipeline" %}
5480   ins_encode( emit_fadd_nop );
5481   ins_pipe(fadd_nop);
5482 %}
5483 
5484 // Nop using the branch functional unit
5485 instruct Nop_BR( ) %{
5486   ins_cost(0);
5487 
5488   format %{ "NOP    ! Branch Pipeline" %}
5489   ins_encode( emit_br_nop );
5490   ins_pipe(br_nop);
5491 %}
5492 
5493 //----------Load/Store/Move Instructions---------------------------------------
5494 //----------Load Instructions--------------------------------------------------
5495 // Load Byte (8bit signed)
5496 instruct loadB(iRegI dst, memory mem) %{
5497   match(Set dst (LoadB mem));
5498   ins_cost(MEMORY_REF_COST);
5499 
5500   size(4);
5501   format %{ "LDSB   $mem,$dst\t! byte" %}
5502   ins_encode %{
5503     __ ldsb($mem$$Address, $dst$$Register);
5504   %}
5505   ins_pipe(iload_mask_mem);
5506 %}
5507 
5508 // Load Byte (8bit signed) into a Long Register
5509 instruct loadB2L(iRegL dst, memory mem) %{
5510   match(Set dst (ConvI2L (LoadB mem)));
5511   ins_cost(MEMORY_REF_COST);
5512 
5513   size(4);
5514   format %{ "LDSB   $mem,$dst\t! byte -> long" %}
5515   ins_encode %{
5516     __ ldsb($mem$$Address, $dst$$Register);
5517   %}
5518   ins_pipe(iload_mask_mem);
5519 %}
5520 
5521 // Load Unsigned Byte (8bit UNsigned) into an int reg
5522 instruct loadUB(iRegI dst, memory mem) %{
5523   match(Set dst (LoadUB mem));
5524   ins_cost(MEMORY_REF_COST);
5525 
5526   size(4);
5527   format %{ "LDUB   $mem,$dst\t! ubyte" %}
5528   ins_encode %{
5529     __ ldub($mem$$Address, $dst$$Register);
5530   %}
5531   ins_pipe(iload_mem);
5532 %}
5533 
5534 // Load Unsigned Byte (8bit UNsigned) into a Long Register
5535 instruct loadUB2L(iRegL dst, memory mem) %{
5536   match(Set dst (ConvI2L (LoadUB mem)));
5537   ins_cost(MEMORY_REF_COST);
5538 
5539   size(4);
5540   format %{ "LDUB   $mem,$dst\t! ubyte -> long" %}
5541   ins_encode %{
5542     __ ldub($mem$$Address, $dst$$Register);
5543   %}
5544   ins_pipe(iload_mem);
5545 %}
5546 
5547 // Load Unsigned Byte (8 bit UNsigned) with 8-bit mask into Long Register
5548 instruct loadUB2L_immI8(iRegL dst, memory mem, immI8 mask) %{
5549   match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5550   ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5551 
5552   size(2*4);
5553   format %{ "LDUB   $mem,$dst\t# ubyte & 8-bit mask -> long\n\t"
5554             "AND    $dst,$mask,$dst" %}
5555   ins_encode %{
5556     __ ldub($mem$$Address, $dst$$Register);
5557     __ and3($dst$$Register, $mask$$constant, $dst$$Register);
5558   %}
5559   ins_pipe(iload_mem);
5560 %}
5561 
5562 // Load Short (16bit signed)
5563 instruct loadS(iRegI dst, memory mem) %{
5564   match(Set dst (LoadS mem));
5565   ins_cost(MEMORY_REF_COST);
5566 
5567   size(4);
5568   format %{ "LDSH   $mem,$dst\t! short" %}
5569   ins_encode %{
5570     __ ldsh($mem$$Address, $dst$$Register);
5571   %}
5572   ins_pipe(iload_mask_mem);
5573 %}
5574 
5575 // Load Short (16 bit signed) to Byte (8 bit signed)
5576 instruct loadS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5577   match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5578   ins_cost(MEMORY_REF_COST);
5579 
5580   size(4);
5581 
5582   format %{ "LDSB   $mem+1,$dst\t! short -> byte" %}
5583   ins_encode %{
5584     __ ldsb($mem$$Address, $dst$$Register, 1);
5585   %}
5586   ins_pipe(iload_mask_mem);
5587 %}
5588 
5589 // Load Short (16bit signed) into a Long Register
5590 instruct loadS2L(iRegL dst, memory mem) %{
5591   match(Set dst (ConvI2L (LoadS mem)));
5592   ins_cost(MEMORY_REF_COST);
5593 
5594   size(4);
5595   format %{ "LDSH   $mem,$dst\t! short -> long" %}
5596   ins_encode %{
5597     __ ldsh($mem$$Address, $dst$$Register);
5598   %}
5599   ins_pipe(iload_mask_mem);
5600 %}
5601 
5602 // Load Unsigned Short/Char (16bit UNsigned)
5603 instruct loadUS(iRegI dst, memory mem) %{
5604   match(Set dst (LoadUS mem));
5605   ins_cost(MEMORY_REF_COST);
5606 
5607   size(4);
5608   format %{ "LDUH   $mem,$dst\t! ushort/char" %}
5609   ins_encode %{
5610     __ lduh($mem$$Address, $dst$$Register);
5611   %}
5612   ins_pipe(iload_mem);
5613 %}
5614 
5615 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5616 instruct loadUS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5617   match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5618   ins_cost(MEMORY_REF_COST);
5619 
5620   size(4);
5621   format %{ "LDSB   $mem+1,$dst\t! ushort -> byte" %}
5622   ins_encode %{
5623     __ ldsb($mem$$Address, $dst$$Register, 1);
5624   %}
5625   ins_pipe(iload_mask_mem);
5626 %}
5627 
5628 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register
5629 instruct loadUS2L(iRegL dst, memory mem) %{
5630   match(Set dst (ConvI2L (LoadUS mem)));
5631   ins_cost(MEMORY_REF_COST);
5632 
5633   size(4);
5634   format %{ "LDUH   $mem,$dst\t! ushort/char -> long" %}
5635   ins_encode %{
5636     __ lduh($mem$$Address, $dst$$Register);
5637   %}
5638   ins_pipe(iload_mem);
5639 %}
5640 
5641 // Load Unsigned Short/Char (16bit UNsigned) with mask 0xFF into a Long Register
5642 instruct loadUS2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5643   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5644   ins_cost(MEMORY_REF_COST);
5645 
5646   size(4);
5647   format %{ "LDUB   $mem+1,$dst\t! ushort/char & 0xFF -> long" %}
5648   ins_encode %{
5649     __ ldub($mem$$Address, $dst$$Register, 1);  // LSB is index+1 on BE
5650   %}
5651   ins_pipe(iload_mem);
5652 %}
5653 
5654 // Load Unsigned Short/Char (16bit UNsigned) with a 13-bit mask into a Long Register
5655 instruct loadUS2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5656   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5657   ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5658 
5659   size(2*4);
5660   format %{ "LDUH   $mem,$dst\t! ushort/char & 13-bit mask -> long\n\t"
5661             "AND    $dst,$mask,$dst" %}
5662   ins_encode %{
5663     Register Rdst = $dst$$Register;
5664     __ lduh($mem$$Address, Rdst);
5665     __ and3(Rdst, $mask$$constant, Rdst);
5666   %}
5667   ins_pipe(iload_mem);
5668 %}
5669 
5670 // Load Unsigned Short/Char (16bit UNsigned) with a 16-bit mask into a Long Register
5671 instruct loadUS2L_immI16(iRegL dst, memory mem, immI16 mask, iRegL tmp) %{
5672   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5673   effect(TEMP dst, TEMP tmp);
5674   ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5675 
5676   size((3+1)*4);  // set may use two instructions.
5677   format %{ "LDUH   $mem,$dst\t! ushort/char & 16-bit mask -> long\n\t"
5678             "SET    $mask,$tmp\n\t"
5679             "AND    $dst,$tmp,$dst" %}
5680   ins_encode %{
5681     Register Rdst = $dst$$Register;
5682     Register Rtmp = $tmp$$Register;
5683     __ lduh($mem$$Address, Rdst);
5684     __ set($mask$$constant, Rtmp);
5685     __ and3(Rdst, Rtmp, Rdst);
5686   %}
5687   ins_pipe(iload_mem);
5688 %}
5689 
5690 // Load Integer
5691 instruct loadI(iRegI dst, memory mem) %{
5692   match(Set dst (LoadI mem));
5693   ins_cost(MEMORY_REF_COST);
5694 
5695   size(4);
5696   format %{ "LDUW   $mem,$dst\t! int" %}
5697   ins_encode %{
5698     __ lduw($mem$$Address, $dst$$Register);
5699   %}
5700   ins_pipe(iload_mem);
5701 %}
5702 
5703 // Load Integer to Byte (8 bit signed)
5704 instruct loadI2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5705   match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
5706   ins_cost(MEMORY_REF_COST);
5707 
5708   size(4);
5709 
5710   format %{ "LDSB   $mem+3,$dst\t! int -> byte" %}
5711   ins_encode %{
5712     __ ldsb($mem$$Address, $dst$$Register, 3);
5713   %}
5714   ins_pipe(iload_mask_mem);
5715 %}
5716 
5717 // Load Integer to Unsigned Byte (8 bit UNsigned)
5718 instruct loadI2UB(iRegI dst, indOffset13m7 mem, immI_255 mask) %{
5719   match(Set dst (AndI (LoadI mem) mask));
5720   ins_cost(MEMORY_REF_COST);
5721 
5722   size(4);
5723 
5724   format %{ "LDUB   $mem+3,$dst\t! int -> ubyte" %}
5725   ins_encode %{
5726     __ ldub($mem$$Address, $dst$$Register, 3);
5727   %}
5728   ins_pipe(iload_mask_mem);
5729 %}
5730 
5731 // Load Integer to Short (16 bit signed)
5732 instruct loadI2S(iRegI dst, indOffset13m7 mem, immI_16 sixteen) %{
5733   match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
5734   ins_cost(MEMORY_REF_COST);
5735 
5736   size(4);
5737 
5738   format %{ "LDSH   $mem+2,$dst\t! int -> short" %}
5739   ins_encode %{
5740     __ ldsh($mem$$Address, $dst$$Register, 2);
5741   %}
5742   ins_pipe(iload_mask_mem);
5743 %}
5744 
5745 // Load Integer to Unsigned Short (16 bit UNsigned)
5746 instruct loadI2US(iRegI dst, indOffset13m7 mem, immI_65535 mask) %{
5747   match(Set dst (AndI (LoadI mem) mask));
5748   ins_cost(MEMORY_REF_COST);
5749 
5750   size(4);
5751 
5752   format %{ "LDUH   $mem+2,$dst\t! int -> ushort/char" %}
5753   ins_encode %{
5754     __ lduh($mem$$Address, $dst$$Register, 2);
5755   %}
5756   ins_pipe(iload_mask_mem);
5757 %}
5758 
5759 // Load Integer into a Long Register
5760 instruct loadI2L(iRegL dst, memory mem) %{
5761   match(Set dst (ConvI2L (LoadI mem)));
5762   ins_cost(MEMORY_REF_COST);
5763 
5764   size(4);
5765   format %{ "LDSW   $mem,$dst\t! int -> long" %}
5766   ins_encode %{
5767     __ ldsw($mem$$Address, $dst$$Register);
5768   %}
5769   ins_pipe(iload_mask_mem);
5770 %}
5771 
5772 // Load Integer with mask 0xFF into a Long Register
5773 instruct loadI2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5774   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5775   ins_cost(MEMORY_REF_COST);
5776 
5777   size(4);
5778   format %{ "LDUB   $mem+3,$dst\t! int & 0xFF -> long" %}
5779   ins_encode %{
5780     __ ldub($mem$$Address, $dst$$Register, 3);  // LSB is index+3 on BE
5781   %}
5782   ins_pipe(iload_mem);
5783 %}
5784 
5785 // Load Integer with mask 0xFFFF into a Long Register
5786 instruct loadI2L_immI_65535(iRegL dst, indOffset13m7 mem, immI_65535 mask) %{
5787   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5788   ins_cost(MEMORY_REF_COST);
5789 
5790   size(4);
5791   format %{ "LDUH   $mem+2,$dst\t! int & 0xFFFF -> long" %}
5792   ins_encode %{
5793     __ lduh($mem$$Address, $dst$$Register, 2);  // LSW is index+2 on BE
5794   %}
5795   ins_pipe(iload_mem);
5796 %}
5797 
5798 // Load Integer with a 13-bit mask into a Long Register
5799 instruct loadI2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5800   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5801   ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5802 
5803   size(2*4);
5804   format %{ "LDUW   $mem,$dst\t! int & 13-bit mask -> long\n\t"
5805             "AND    $dst,$mask,$dst" %}
5806   ins_encode %{
5807     Register Rdst = $dst$$Register;
5808     __ lduw($mem$$Address, Rdst);
5809     __ and3(Rdst, $mask$$constant, Rdst);
5810   %}
5811   ins_pipe(iload_mem);
5812 %}
5813 
5814 // Load Integer with a 32-bit mask into a Long Register
5815 instruct loadI2L_immI(iRegL dst, memory mem, immI mask, iRegL tmp) %{
5816   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5817   effect(TEMP dst, TEMP tmp);
5818   ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5819 
5820   size((3+1)*4);  // set may use two instructions.
5821   format %{ "LDUW   $mem,$dst\t! int & 32-bit mask -> long\n\t"
5822             "SET    $mask,$tmp\n\t"
5823             "AND    $dst,$tmp,$dst" %}
5824   ins_encode %{
5825     Register Rdst = $dst$$Register;
5826     Register Rtmp = $tmp$$Register;
5827     __ lduw($mem$$Address, Rdst);
5828     __ set($mask$$constant, Rtmp);
5829     __ and3(Rdst, Rtmp, Rdst);
5830   %}
5831   ins_pipe(iload_mem);
5832 %}
5833 
5834 // Load Unsigned Integer into a Long Register
5835 instruct loadUI2L(iRegL dst, memory mem, immL_32bits mask) %{
5836   match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
5837   ins_cost(MEMORY_REF_COST);
5838 
5839   size(4);
5840   format %{ "LDUW   $mem,$dst\t! uint -> long" %}
5841   ins_encode %{
5842     __ lduw($mem$$Address, $dst$$Register);
5843   %}
5844   ins_pipe(iload_mem);
5845 %}
5846 
5847 // Load Long - aligned
5848 instruct loadL(iRegL dst, memory mem ) %{
5849   match(Set dst (LoadL mem));
5850   ins_cost(MEMORY_REF_COST);
5851 
5852   size(4);
5853   format %{ "LDX    $mem,$dst\t! long" %}
5854   ins_encode %{
5855     __ ldx($mem$$Address, $dst$$Register);
5856   %}
5857   ins_pipe(iload_mem);
5858 %}
5859 
5860 // Load Long - UNaligned
5861 instruct loadL_unaligned(iRegL dst, memory mem, o7RegI tmp) %{
5862   match(Set dst (LoadL_unaligned mem));
5863   effect(KILL tmp);
5864   ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
5865   size(16);
5866   format %{ "LDUW   $mem+4,R_O7\t! misaligned long\n"
5867           "\tLDUW   $mem  ,$dst\n"
5868           "\tSLLX   #32, $dst, $dst\n"
5869           "\tOR     $dst, R_O7, $dst" %}
5870   opcode(Assembler::lduw_op3);
5871   ins_encode(form3_mem_reg_long_unaligned_marshal( mem, dst ));
5872   ins_pipe(iload_mem);
5873 %}
5874 
5875 // Load Range
5876 instruct loadRange(iRegI dst, memory mem) %{
5877   match(Set dst (LoadRange mem));
5878   ins_cost(MEMORY_REF_COST);
5879 
5880   size(4);
5881   format %{ "LDUW   $mem,$dst\t! range" %}
5882   opcode(Assembler::lduw_op3);
5883   ins_encode(simple_form3_mem_reg( mem, dst ) );
5884   ins_pipe(iload_mem);
5885 %}
5886 
5887 // Load Integer into %f register (for fitos/fitod)
5888 instruct loadI_freg(regF dst, memory mem) %{
5889   match(Set dst (LoadI mem));
5890   ins_cost(MEMORY_REF_COST);
5891   size(4);
5892 
5893   format %{ "LDF    $mem,$dst\t! for fitos/fitod" %}
5894   opcode(Assembler::ldf_op3);
5895   ins_encode(simple_form3_mem_reg( mem, dst ) );
5896   ins_pipe(floadF_mem);
5897 %}
5898 
5899 // Load Pointer
5900 instruct loadP(iRegP dst, memory mem) %{
5901   match(Set dst (LoadP mem));
5902   ins_cost(MEMORY_REF_COST);
5903   size(4);
5904 
5905 #ifndef _LP64
5906   format %{ "LDUW   $mem,$dst\t! ptr" %}
5907   ins_encode %{
5908     __ lduw($mem$$Address, $dst$$Register);
5909   %}
5910 #else
5911   format %{ "LDX    $mem,$dst\t! ptr" %}
5912   ins_encode %{
5913     __ ldx($mem$$Address, $dst$$Register);
5914   %}
5915 #endif
5916   ins_pipe(iload_mem);
5917 %}
5918 
5919 // Load Compressed Pointer
5920 instruct loadN(iRegN dst, memory mem) %{
5921   match(Set dst (LoadN mem));
5922   ins_cost(MEMORY_REF_COST);
5923   size(4);
5924 
5925   format %{ "LDUW   $mem,$dst\t! compressed ptr" %}
5926   ins_encode %{
5927     __ lduw($mem$$Address, $dst$$Register);
5928   %}
5929   ins_pipe(iload_mem);
5930 %}
5931 
5932 // Load Klass Pointer
5933 instruct loadKlass(iRegP dst, memory mem) %{
5934   match(Set dst (LoadKlass mem));
5935   ins_cost(MEMORY_REF_COST);
5936   size(4);
5937 
5938 #ifndef _LP64
5939   format %{ "LDUW   $mem,$dst\t! klass ptr" %}
5940   ins_encode %{
5941     __ lduw($mem$$Address, $dst$$Register);
5942   %}
5943 #else
5944   format %{ "LDX    $mem,$dst\t! klass ptr" %}
5945   ins_encode %{
5946     __ ldx($mem$$Address, $dst$$Register);
5947   %}
5948 #endif
5949   ins_pipe(iload_mem);
5950 %}
5951 
5952 // Load narrow Klass Pointer
5953 instruct loadNKlass(iRegN dst, memory mem) %{
5954   match(Set dst (LoadNKlass mem));
5955   ins_cost(MEMORY_REF_COST);
5956   size(4);
5957 
5958   format %{ "LDUW   $mem,$dst\t! compressed klass ptr" %}
5959   ins_encode %{
5960     __ lduw($mem$$Address, $dst$$Register);
5961   %}
5962   ins_pipe(iload_mem);
5963 %}
5964 
5965 // Load Double
5966 instruct loadD(regD dst, memory mem) %{
5967   match(Set dst (LoadD mem));
5968   ins_cost(MEMORY_REF_COST);
5969 
5970   size(4);
5971   format %{ "LDDF   $mem,$dst" %}
5972   opcode(Assembler::lddf_op3);
5973   ins_encode(simple_form3_mem_reg( mem, dst ) );
5974   ins_pipe(floadD_mem);
5975 %}
5976 
5977 // Load Double - UNaligned
5978 instruct loadD_unaligned(regD_low dst, memory mem ) %{
5979   match(Set dst (LoadD_unaligned mem));
5980   ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
5981   size(8);
5982   format %{ "LDF    $mem  ,$dst.hi\t! misaligned double\n"
5983           "\tLDF    $mem+4,$dst.lo\t!" %}
5984   opcode(Assembler::ldf_op3);
5985   ins_encode( form3_mem_reg_double_unaligned( mem, dst ));
5986   ins_pipe(iload_mem);
5987 %}
5988 
5989 // Load Float
5990 instruct loadF(regF dst, memory mem) %{
5991   match(Set dst (LoadF mem));
5992   ins_cost(MEMORY_REF_COST);
5993 
5994   size(4);
5995   format %{ "LDF    $mem,$dst" %}
5996   opcode(Assembler::ldf_op3);
5997   ins_encode(simple_form3_mem_reg( mem, dst ) );
5998   ins_pipe(floadF_mem);
5999 %}
6000 
6001 // Load Constant
6002 instruct loadConI( iRegI dst, immI src ) %{
6003   match(Set dst src);
6004   ins_cost(DEFAULT_COST * 3/2);
6005   format %{ "SET    $src,$dst" %}
6006   ins_encode( Set32(src, dst) );
6007   ins_pipe(ialu_hi_lo_reg);
6008 %}
6009 
6010 instruct loadConI13( iRegI dst, immI13 src ) %{
6011   match(Set dst src);
6012 
6013   size(4);
6014   format %{ "MOV    $src,$dst" %}
6015   ins_encode( Set13( src, dst ) );
6016   ins_pipe(ialu_imm);
6017 %}
6018 
6019 #ifndef _LP64
6020 instruct loadConP(iRegP dst, immP con) %{
6021   match(Set dst con);
6022   ins_cost(DEFAULT_COST * 3/2);
6023   format %{ "SET    $con,$dst\t!ptr" %}
6024   ins_encode %{
6025     relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6026       intptr_t val = $con$$constant;
6027     if (constant_reloc == relocInfo::oop_type) {
6028       __ set_oop_constant((jobject) val, $dst$$Register);
6029     } else if (constant_reloc == relocInfo::metadata_type) {
6030       __ set_metadata_constant((Metadata*)val, $dst$$Register);
6031     } else {          // non-oop pointers, e.g. card mark base, heap top
6032       assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6033       __ set(val, $dst$$Register);
6034     }
6035   %}
6036   ins_pipe(loadConP);
6037 %}
6038 #else
6039 instruct loadConP_set(iRegP dst, immP_set con) %{
6040   match(Set dst con);
6041   ins_cost(DEFAULT_COST * 3/2);
6042   format %{ "SET    $con,$dst\t! ptr" %}
6043   ins_encode %{
6044     relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6045       intptr_t val = $con$$constant;
6046     if (constant_reloc == relocInfo::oop_type) {
6047       __ set_oop_constant((jobject) val, $dst$$Register);
6048     } else if (constant_reloc == relocInfo::metadata_type) {
6049       __ set_metadata_constant((Metadata*)val, $dst$$Register);
6050     } else {          // non-oop pointers, e.g. card mark base, heap top
6051       assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6052       __ set(val, $dst$$Register);
6053     }
6054   %}
6055   ins_pipe(loadConP);
6056 %}
6057 
6058 instruct loadConP_load(iRegP dst, immP_load con) %{
6059   match(Set dst con);
6060   ins_cost(MEMORY_REF_COST);
6061   format %{ "LD     [$constanttablebase + $constantoffset],$dst\t! load from constant table: ptr=$con" %}
6062   ins_encode %{
6063     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6064     __ ld_ptr($constanttablebase, con_offset, $dst$$Register);
6065   %}
6066   ins_pipe(loadConP);
6067 %}
6068 
6069 instruct loadConP_no_oop_cheap(iRegP dst, immP_no_oop_cheap con) %{
6070   match(Set dst con);
6071   ins_cost(DEFAULT_COST * 3/2);
6072   format %{ "SET    $con,$dst\t! non-oop ptr" %}
6073   ins_encode %{
6074     __ set($con$$constant, $dst$$Register);
6075   %}
6076   ins_pipe(loadConP);
6077 %}
6078 #endif // _LP64
6079 
6080 instruct loadConP0(iRegP dst, immP0 src) %{
6081   match(Set dst src);
6082 
6083   size(4);
6084   format %{ "CLR    $dst\t!ptr" %}
6085   ins_encode %{
6086     __ clr($dst$$Register);
6087   %}
6088   ins_pipe(ialu_imm);
6089 %}
6090 
6091 instruct loadConP_poll(iRegP dst, immP_poll src) %{
6092   match(Set dst src);
6093   ins_cost(DEFAULT_COST);
6094   format %{ "SET    $src,$dst\t!ptr" %}
6095   ins_encode %{
6096     AddressLiteral polling_page(os::get_polling_page());
6097     __ sethi(polling_page, reg_to_register_object($dst$$reg));
6098   %}
6099   ins_pipe(loadConP_poll);
6100 %}
6101 
6102 instruct loadConN0(iRegN dst, immN0 src) %{
6103   match(Set dst src);
6104 
6105   size(4);
6106   format %{ "CLR    $dst\t! compressed NULL ptr" %}
6107   ins_encode %{
6108     __ clr($dst$$Register);
6109   %}
6110   ins_pipe(ialu_imm);
6111 %}
6112 
6113 instruct loadConN(iRegN dst, immN src) %{
6114   match(Set dst src);
6115   ins_cost(DEFAULT_COST * 3/2);
6116   format %{ "SET    $src,$dst\t! compressed ptr" %}
6117   ins_encode %{
6118     Register dst = $dst$$Register;
6119     __ set_narrow_oop((jobject)$src$$constant, dst);
6120   %}
6121   ins_pipe(ialu_hi_lo_reg);
6122 %}
6123 
6124 instruct loadConNKlass(iRegN dst, immNKlass src) %{
6125   match(Set dst src);
6126   ins_cost(DEFAULT_COST * 3/2);
6127   format %{ "SET    $src,$dst\t! compressed klass ptr" %}
6128   ins_encode %{
6129     Register dst = $dst$$Register;
6130     __ set_narrow_klass((Klass*)$src$$constant, dst);
6131   %}
6132   ins_pipe(ialu_hi_lo_reg);
6133 %}
6134 
6135 // Materialize long value (predicated by immL_cheap).
6136 instruct loadConL_set64(iRegL dst, immL_cheap con, o7RegL tmp) %{
6137   match(Set dst con);
6138   effect(KILL tmp);
6139   ins_cost(DEFAULT_COST * 3);
6140   format %{ "SET64   $con,$dst KILL $tmp\t! cheap long" %}
6141   ins_encode %{
6142     __ set64($con$$constant, $dst$$Register, $tmp$$Register);
6143   %}
6144   ins_pipe(loadConL);
6145 %}
6146 
6147 // Load long value from constant table (predicated by immL_expensive).
6148 instruct loadConL_ldx(iRegL dst, immL_expensive con) %{
6149   match(Set dst con);
6150   ins_cost(MEMORY_REF_COST);
6151   format %{ "LDX     [$constanttablebase + $constantoffset],$dst\t! load from constant table: long=$con" %}
6152   ins_encode %{
6153       RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6154     __ ldx($constanttablebase, con_offset, $dst$$Register);
6155   %}
6156   ins_pipe(loadConL);
6157 %}
6158 
6159 instruct loadConL0( iRegL dst, immL0 src ) %{
6160   match(Set dst src);
6161   ins_cost(DEFAULT_COST);
6162   size(4);
6163   format %{ "CLR    $dst\t! long" %}
6164   ins_encode( Set13( src, dst ) );
6165   ins_pipe(ialu_imm);
6166 %}
6167 
6168 instruct loadConL13( iRegL dst, immL13 src ) %{
6169   match(Set dst src);
6170   ins_cost(DEFAULT_COST * 2);
6171 
6172   size(4);
6173   format %{ "MOV    $src,$dst\t! long" %}
6174   ins_encode( Set13( src, dst ) );
6175   ins_pipe(ialu_imm);
6176 %}
6177 
6178 instruct loadConF(regF dst, immF con, o7RegI tmp) %{
6179   match(Set dst con);
6180   effect(KILL tmp);
6181   format %{ "LDF    [$constanttablebase + $constantoffset],$dst\t! load from constant table: float=$con" %}
6182   ins_encode %{
6183       RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6184     __ ldf(FloatRegisterImpl::S, $constanttablebase, con_offset, $dst$$FloatRegister);
6185   %}
6186   ins_pipe(loadConFD);
6187 %}
6188 
6189 instruct loadConD(regD dst, immD con, o7RegI tmp) %{
6190   match(Set dst con);
6191   effect(KILL tmp);
6192   format %{ "LDDF   [$constanttablebase + $constantoffset],$dst\t! load from constant table: double=$con" %}
6193   ins_encode %{
6194     // XXX This is a quick fix for 6833573.
6195     //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset($con), $dst$$FloatRegister);
6196     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6197     __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
6198   %}
6199   ins_pipe(loadConFD);
6200 %}
6201 
6202 // Prefetch instructions.
6203 // Must be safe to execute with invalid address (cannot fault).
6204 
6205 instruct prefetchr( memory mem ) %{
6206   match( PrefetchRead mem );
6207   ins_cost(MEMORY_REF_COST);
6208   size(4);
6209 
6210   format %{ "PREFETCH $mem,0\t! Prefetch read-many" %}
6211   opcode(Assembler::prefetch_op3);
6212   ins_encode( form3_mem_prefetch_read( mem ) );
6213   ins_pipe(iload_mem);
6214 %}
6215 
6216 instruct prefetchw( memory mem ) %{
6217   match( PrefetchWrite mem );
6218   ins_cost(MEMORY_REF_COST);
6219   size(4);
6220 
6221   format %{ "PREFETCH $mem,2\t! Prefetch write-many (and read)" %}
6222   opcode(Assembler::prefetch_op3);
6223   ins_encode( form3_mem_prefetch_write( mem ) );
6224   ins_pipe(iload_mem);
6225 %}
6226 
6227 // Prefetch instructions for allocation.
6228 
6229 instruct prefetchAlloc( memory mem ) %{
6230   predicate(AllocatePrefetchInstr == 0);
6231   match( PrefetchAllocation mem );
6232   ins_cost(MEMORY_REF_COST);
6233   size(4);
6234 
6235   format %{ "PREFETCH $mem,2\t! Prefetch allocation" %}
6236   opcode(Assembler::prefetch_op3);
6237   ins_encode( form3_mem_prefetch_write( mem ) );
6238   ins_pipe(iload_mem);
6239 %}
6240 
6241 // Use BIS instruction to prefetch for allocation.
6242 // Could fault, need space at the end of TLAB.
6243 instruct prefetchAlloc_bis( iRegP dst ) %{
6244   predicate(AllocatePrefetchInstr == 1);
6245   match( PrefetchAllocation dst );
6246   ins_cost(MEMORY_REF_COST);
6247   size(4);
6248 
6249   format %{ "STXA   [$dst]\t! // Prefetch allocation using BIS" %}
6250   ins_encode %{
6251     __ stxa(G0, $dst$$Register, G0, Assembler::ASI_ST_BLKINIT_PRIMARY);
6252   %}
6253   ins_pipe(istore_mem_reg);
6254 %}
6255 
6256 // Next code is used for finding next cache line address to prefetch.
6257 #ifndef _LP64
6258 instruct cacheLineAdr( iRegP dst, iRegP src, immI13 mask ) %{
6259   match(Set dst (CastX2P (AndI (CastP2X src) mask)));
6260   ins_cost(DEFAULT_COST);
6261   size(4);
6262 
6263   format %{ "AND    $src,$mask,$dst\t! next cache line address" %}
6264   ins_encode %{
6265     __ and3($src$$Register, $mask$$constant, $dst$$Register);
6266   %}
6267   ins_pipe(ialu_reg_imm);
6268 %}
6269 #else
6270 instruct cacheLineAdr( iRegP dst, iRegP src, immL13 mask ) %{
6271   match(Set dst (CastX2P (AndL (CastP2X src) mask)));
6272   ins_cost(DEFAULT_COST);
6273   size(4);
6274 
6275   format %{ "AND    $src,$mask,$dst\t! next cache line address" %}
6276   ins_encode %{
6277     __ and3($src$$Register, $mask$$constant, $dst$$Register);
6278   %}
6279   ins_pipe(ialu_reg_imm);
6280 %}
6281 #endif
6282 
6283 //----------Store Instructions-------------------------------------------------
6284 // Store Byte
6285 instruct storeB(memory mem, iRegI src) %{
6286   match(Set mem (StoreB mem src));
6287   ins_cost(MEMORY_REF_COST);
6288 
6289   size(4);
6290   format %{ "STB    $src,$mem\t! byte" %}
6291   opcode(Assembler::stb_op3);
6292   ins_encode(simple_form3_mem_reg( mem, src ) );
6293   ins_pipe(istore_mem_reg);
6294 %}
6295 
6296 instruct storeB0(memory mem, immI0 src) %{
6297   match(Set mem (StoreB mem src));
6298   ins_cost(MEMORY_REF_COST);
6299 
6300   size(4);
6301   format %{ "STB    $src,$mem\t! byte" %}
6302   opcode(Assembler::stb_op3);
6303   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6304   ins_pipe(istore_mem_zero);
6305 %}
6306 
6307 instruct storeCM0(memory mem, immI0 src) %{
6308   match(Set mem (StoreCM mem src));
6309   ins_cost(MEMORY_REF_COST);
6310 
6311   size(4);
6312   format %{ "STB    $src,$mem\t! CMS card-mark byte 0" %}
6313   opcode(Assembler::stb_op3);
6314   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6315   ins_pipe(istore_mem_zero);
6316 %}
6317 
6318 // Store Char/Short
6319 instruct storeC(memory mem, iRegI src) %{
6320   match(Set mem (StoreC mem src));
6321   ins_cost(MEMORY_REF_COST);
6322 
6323   size(4);
6324   format %{ "STH    $src,$mem\t! short" %}
6325   opcode(Assembler::sth_op3);
6326   ins_encode(simple_form3_mem_reg( mem, src ) );
6327   ins_pipe(istore_mem_reg);
6328 %}
6329 
6330 instruct storeC0(memory mem, immI0 src) %{
6331   match(Set mem (StoreC mem src));
6332   ins_cost(MEMORY_REF_COST);
6333 
6334   size(4);
6335   format %{ "STH    $src,$mem\t! short" %}
6336   opcode(Assembler::sth_op3);
6337   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6338   ins_pipe(istore_mem_zero);
6339 %}
6340 
6341 // Store Integer
6342 instruct storeI(memory mem, iRegI src) %{
6343   match(Set mem (StoreI mem src));
6344   ins_cost(MEMORY_REF_COST);
6345 
6346   size(4);
6347   format %{ "STW    $src,$mem" %}
6348   opcode(Assembler::stw_op3);
6349   ins_encode(simple_form3_mem_reg( mem, src ) );
6350   ins_pipe(istore_mem_reg);
6351 %}
6352 
6353 // Store Long
6354 instruct storeL(memory mem, iRegL src) %{
6355   match(Set mem (StoreL mem src));
6356   ins_cost(MEMORY_REF_COST);
6357   size(4);
6358   format %{ "STX    $src,$mem\t! long" %}
6359   opcode(Assembler::stx_op3);
6360   ins_encode(simple_form3_mem_reg( mem, src ) );
6361   ins_pipe(istore_mem_reg);
6362 %}
6363 
6364 instruct storeI0(memory mem, immI0 src) %{
6365   match(Set mem (StoreI mem src));
6366   ins_cost(MEMORY_REF_COST);
6367 
6368   size(4);
6369   format %{ "STW    $src,$mem" %}
6370   opcode(Assembler::stw_op3);
6371   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6372   ins_pipe(istore_mem_zero);
6373 %}
6374 
6375 instruct storeL0(memory mem, immL0 src) %{
6376   match(Set mem (StoreL mem src));
6377   ins_cost(MEMORY_REF_COST);
6378 
6379   size(4);
6380   format %{ "STX    $src,$mem" %}
6381   opcode(Assembler::stx_op3);
6382   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6383   ins_pipe(istore_mem_zero);
6384 %}
6385 
6386 // Store Integer from float register (used after fstoi)
6387 instruct storeI_Freg(memory mem, regF src) %{
6388   match(Set mem (StoreI mem src));
6389   ins_cost(MEMORY_REF_COST);
6390 
6391   size(4);
6392   format %{ "STF    $src,$mem\t! after fstoi/fdtoi" %}
6393   opcode(Assembler::stf_op3);
6394   ins_encode(simple_form3_mem_reg( mem, src ) );
6395   ins_pipe(fstoreF_mem_reg);
6396 %}
6397 
6398 // Store Pointer
6399 instruct storeP(memory dst, sp_ptr_RegP src) %{
6400   match(Set dst (StoreP dst src));
6401   ins_cost(MEMORY_REF_COST);
6402   size(4);
6403 
6404 #ifndef _LP64
6405   format %{ "STW    $src,$dst\t! ptr" %}
6406   opcode(Assembler::stw_op3, 0, REGP_OP);
6407 #else
6408   format %{ "STX    $src,$dst\t! ptr" %}
6409   opcode(Assembler::stx_op3, 0, REGP_OP);
6410 #endif
6411   ins_encode( form3_mem_reg( dst, src ) );
6412   ins_pipe(istore_mem_spORreg);
6413 %}
6414 
6415 instruct storeP0(memory dst, immP0 src) %{
6416   match(Set dst (StoreP dst src));
6417   ins_cost(MEMORY_REF_COST);
6418   size(4);
6419 
6420 #ifndef _LP64
6421   format %{ "STW    $src,$dst\t! ptr" %}
6422   opcode(Assembler::stw_op3, 0, REGP_OP);
6423 #else
6424   format %{ "STX    $src,$dst\t! ptr" %}
6425   opcode(Assembler::stx_op3, 0, REGP_OP);
6426 #endif
6427   ins_encode( form3_mem_reg( dst, R_G0 ) );
6428   ins_pipe(istore_mem_zero);
6429 %}
6430 
6431 // Store Compressed Pointer
6432 instruct storeN(memory dst, iRegN src) %{
6433    match(Set dst (StoreN dst src));
6434    ins_cost(MEMORY_REF_COST);
6435    size(4);
6436 
6437    format %{ "STW    $src,$dst\t! compressed ptr" %}
6438    ins_encode %{
6439      Register base = as_Register($dst$$base);
6440      Register index = as_Register($dst$$index);
6441      Register src = $src$$Register;
6442      if (index != G0) {
6443        __ stw(src, base, index);
6444      } else {
6445        __ stw(src, base, $dst$$disp);
6446      }
6447    %}
6448    ins_pipe(istore_mem_spORreg);
6449 %}
6450 
6451 instruct storeNKlass(memory dst, iRegN src) %{
6452    match(Set dst (StoreNKlass dst src));
6453    ins_cost(MEMORY_REF_COST);
6454    size(4);
6455 
6456    format %{ "STW    $src,$dst\t! compressed klass ptr" %}
6457    ins_encode %{
6458      Register base = as_Register($dst$$base);
6459      Register index = as_Register($dst$$index);
6460      Register src = $src$$Register;
6461      if (index != G0) {
6462        __ stw(src, base, index);
6463      } else {
6464        __ stw(src, base, $dst$$disp);
6465      }
6466    %}
6467    ins_pipe(istore_mem_spORreg);
6468 %}
6469 
6470 instruct storeN0(memory dst, immN0 src) %{
6471    match(Set dst (StoreN dst src));
6472    ins_cost(MEMORY_REF_COST);
6473    size(4);
6474 
6475    format %{ "STW    $src,$dst\t! compressed ptr" %}
6476    ins_encode %{
6477      Register base = as_Register($dst$$base);
6478      Register index = as_Register($dst$$index);
6479      if (index != G0) {
6480        __ stw(0, base, index);
6481      } else {
6482        __ stw(0, base, $dst$$disp);
6483      }
6484    %}
6485    ins_pipe(istore_mem_zero);
6486 %}
6487 
6488 // Store Double
6489 instruct storeD( memory mem, regD src) %{
6490   match(Set mem (StoreD mem src));
6491   ins_cost(MEMORY_REF_COST);
6492 
6493   size(4);
6494   format %{ "STDF   $src,$mem" %}
6495   opcode(Assembler::stdf_op3);
6496   ins_encode(simple_form3_mem_reg( mem, src ) );
6497   ins_pipe(fstoreD_mem_reg);
6498 %}
6499 
6500 instruct storeD0( memory mem, immD0 src) %{
6501   match(Set mem (StoreD mem src));
6502   ins_cost(MEMORY_REF_COST);
6503 
6504   size(4);
6505   format %{ "STX    $src,$mem" %}
6506   opcode(Assembler::stx_op3);
6507   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6508   ins_pipe(fstoreD_mem_zero);
6509 %}
6510 
6511 // Store Float
6512 instruct storeF( memory mem, regF src) %{
6513   match(Set mem (StoreF mem src));
6514   ins_cost(MEMORY_REF_COST);
6515 
6516   size(4);
6517   format %{ "STF    $src,$mem" %}
6518   opcode(Assembler::stf_op3);
6519   ins_encode(simple_form3_mem_reg( mem, src ) );
6520   ins_pipe(fstoreF_mem_reg);
6521 %}
6522 
6523 instruct storeF0( memory mem, immF0 src) %{
6524   match(Set mem (StoreF mem src));
6525   ins_cost(MEMORY_REF_COST);
6526 
6527   size(4);
6528   format %{ "STW    $src,$mem\t! storeF0" %}
6529   opcode(Assembler::stw_op3);
6530   ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6531   ins_pipe(fstoreF_mem_zero);
6532 %}
6533 
6534 // Convert oop pointer into compressed form
6535 instruct encodeHeapOop(iRegN dst, iRegP src) %{
6536   predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
6537   match(Set dst (EncodeP src));
6538   format %{ "encode_heap_oop $src, $dst" %}
6539   ins_encode %{
6540     __ encode_heap_oop($src$$Register, $dst$$Register);
6541   %}
6542   ins_pipe(ialu_reg);
6543 %}
6544 
6545 instruct encodeHeapOop_not_null(iRegN dst, iRegP src) %{
6546   predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
6547   match(Set dst (EncodeP src));
6548   format %{ "encode_heap_oop_not_null $src, $dst" %}
6549   ins_encode %{
6550     __ encode_heap_oop_not_null($src$$Register, $dst$$Register);
6551   %}
6552   ins_pipe(ialu_reg);
6553 %}
6554 
6555 instruct decodeHeapOop(iRegP dst, iRegN src) %{
6556   predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
6557             n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
6558   match(Set dst (DecodeN src));
6559   format %{ "decode_heap_oop $src, $dst" %}
6560   ins_encode %{
6561     __ decode_heap_oop($src$$Register, $dst$$Register);
6562   %}
6563   ins_pipe(ialu_reg);
6564 %}
6565 
6566 instruct decodeHeapOop_not_null(iRegP dst, iRegN src) %{
6567   predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
6568             n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
6569   match(Set dst (DecodeN src));
6570   format %{ "decode_heap_oop_not_null $src, $dst" %}
6571   ins_encode %{
6572     __ decode_heap_oop_not_null($src$$Register, $dst$$Register);
6573   %}
6574   ins_pipe(ialu_reg);
6575 %}
6576 
6577 instruct encodeKlass_not_null(iRegN dst, iRegP src) %{
6578   match(Set dst (EncodePKlass src));
6579   format %{ "encode_klass_not_null $src, $dst" %}
6580   ins_encode %{
6581     __ encode_klass_not_null($src$$Register, $dst$$Register);
6582   %}
6583   ins_pipe(ialu_reg);
6584 %}
6585 
6586 instruct decodeKlass_not_null(iRegP dst, iRegN src) %{
6587   match(Set dst (DecodeNKlass src));
6588   format %{ "decode_klass_not_null $src, $dst" %}
6589   ins_encode %{
6590     __ decode_klass_not_null($src$$Register, $dst$$Register);
6591   %}
6592   ins_pipe(ialu_reg);
6593 %}
6594 
6595 //----------MemBar Instructions-----------------------------------------------
6596 // Memory barrier flavors
6597 
6598 instruct membar_acquire() %{
6599   match(MemBarAcquire);
6600   ins_cost(4*MEMORY_REF_COST);
6601 
6602   size(0);
6603   format %{ "MEMBAR-acquire" %}
6604   ins_encode( enc_membar_acquire );
6605   ins_pipe(long_memory_op);
6606 %}
6607 
6608 instruct membar_acquire_lock() %{
6609   match(MemBarAcquireLock);
6610   ins_cost(0);
6611 
6612   size(0);
6613   format %{ "!MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %}
6614   ins_encode( );
6615   ins_pipe(empty);
6616 %}
6617 
6618 instruct membar_release() %{
6619   match(MemBarRelease);
6620   ins_cost(4*MEMORY_REF_COST);
6621 
6622   size(0);
6623   format %{ "MEMBAR-release" %}
6624   ins_encode( enc_membar_release );
6625   ins_pipe(long_memory_op);
6626 %}
6627 
6628 instruct membar_release_lock() %{
6629   match(MemBarReleaseLock);
6630   ins_cost(0);
6631 
6632   size(0);
6633   format %{ "!MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %}
6634   ins_encode( );
6635   ins_pipe(empty);
6636 %}
6637 
6638 instruct membar_volatile() %{
6639   match(MemBarVolatile);
6640   ins_cost(4*MEMORY_REF_COST);
6641 
6642   size(4);
6643   format %{ "MEMBAR-volatile" %}
6644   ins_encode( enc_membar_volatile );
6645   ins_pipe(long_memory_op);
6646 %}
6647 
6648 instruct unnecessary_membar_volatile() %{
6649   match(MemBarVolatile);
6650   predicate(Matcher::post_store_load_barrier(n));
6651   ins_cost(0);
6652 
6653   size(0);
6654   format %{ "!MEMBAR-volatile (unnecessary so empty encoding)" %}
6655   ins_encode( );
6656   ins_pipe(empty);
6657 %}
6658 
6659 instruct membar_storestore() %{
6660   match(MemBarStoreStore);
6661   ins_cost(0);
6662 
6663   size(0);
6664   format %{ "!MEMBAR-storestore (empty encoding)" %}
6665   ins_encode( );
6666   ins_pipe(empty);
6667 %}
6668 
6669 //----------Register Move Instructions-----------------------------------------
6670 instruct roundDouble_nop(regD dst) %{
6671   match(Set dst (RoundDouble dst));
6672   ins_cost(0);
6673   // SPARC results are already "rounded" (i.e., normal-format IEEE)
6674   ins_encode( );
6675   ins_pipe(empty);
6676 %}
6677 
6678 
6679 instruct roundFloat_nop(regF dst) %{
6680   match(Set dst (RoundFloat dst));
6681   ins_cost(0);
6682   // SPARC results are already "rounded" (i.e., normal-format IEEE)
6683   ins_encode( );
6684   ins_pipe(empty);
6685 %}
6686 
6687 
6688 // Cast Index to Pointer for unsafe natives
6689 instruct castX2P(iRegX src, iRegP dst) %{
6690   match(Set dst (CastX2P src));
6691 
6692   format %{ "MOV    $src,$dst\t! IntX->Ptr" %}
6693   ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6694   ins_pipe(ialu_reg);
6695 %}
6696 
6697 // Cast Pointer to Index for unsafe natives
6698 instruct castP2X(iRegP src, iRegX dst) %{
6699   match(Set dst (CastP2X src));
6700 
6701   format %{ "MOV    $src,$dst\t! Ptr->IntX" %}
6702   ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6703   ins_pipe(ialu_reg);
6704 %}
6705 
6706 instruct stfSSD(stackSlotD stkSlot, regD src) %{
6707   // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6708   match(Set stkSlot src);   // chain rule
6709   ins_cost(MEMORY_REF_COST);
6710   format %{ "STDF   $src,$stkSlot\t!stk" %}
6711   opcode(Assembler::stdf_op3);
6712   ins_encode(simple_form3_mem_reg(stkSlot, src));
6713   ins_pipe(fstoreD_stk_reg);
6714 %}
6715 
6716 instruct ldfSSD(regD dst, stackSlotD stkSlot) %{
6717   // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6718   match(Set dst stkSlot);   // chain rule
6719   ins_cost(MEMORY_REF_COST);
6720   format %{ "LDDF   $stkSlot,$dst\t!stk" %}
6721   opcode(Assembler::lddf_op3);
6722   ins_encode(simple_form3_mem_reg(stkSlot, dst));
6723   ins_pipe(floadD_stk);
6724 %}
6725 
6726 instruct stfSSF(stackSlotF stkSlot, regF src) %{
6727   // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6728   match(Set stkSlot src);   // chain rule
6729   ins_cost(MEMORY_REF_COST);
6730   format %{ "STF   $src,$stkSlot\t!stk" %}
6731   opcode(Assembler::stf_op3);
6732   ins_encode(simple_form3_mem_reg(stkSlot, src));
6733   ins_pipe(fstoreF_stk_reg);
6734 %}
6735 
6736 //----------Conditional Move---------------------------------------------------
6737 // Conditional move
6738 instruct cmovIP_reg(cmpOpP cmp, flagsRegP pcc, iRegI dst, iRegI src) %{
6739   match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6740   ins_cost(150);
6741   format %{ "MOV$cmp $pcc,$src,$dst" %}
6742   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6743   ins_pipe(ialu_reg);
6744 %}
6745 
6746 instruct cmovIP_imm(cmpOpP cmp, flagsRegP pcc, iRegI dst, immI11 src) %{
6747   match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6748   ins_cost(140);
6749   format %{ "MOV$cmp $pcc,$src,$dst" %}
6750   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6751   ins_pipe(ialu_imm);
6752 %}
6753 
6754 instruct cmovII_reg(cmpOp cmp, flagsReg icc, iRegI dst, iRegI src) %{
6755   match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6756   ins_cost(150);
6757   size(4);
6758   format %{ "MOV$cmp  $icc,$src,$dst" %}
6759   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6760   ins_pipe(ialu_reg);
6761 %}
6762 
6763 instruct cmovII_imm(cmpOp cmp, flagsReg icc, iRegI dst, immI11 src) %{
6764   match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6765   ins_cost(140);
6766   size(4);
6767   format %{ "MOV$cmp  $icc,$src,$dst" %}
6768   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6769   ins_pipe(ialu_imm);
6770 %}
6771 
6772 instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{
6773   match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6774   ins_cost(150);
6775   size(4);
6776   format %{ "MOV$cmp  $icc,$src,$dst" %}
6777   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6778   ins_pipe(ialu_reg);
6779 %}
6780 
6781 instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{
6782   match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6783   ins_cost(140);
6784   size(4);
6785   format %{ "MOV$cmp  $icc,$src,$dst" %}
6786   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6787   ins_pipe(ialu_imm);
6788 %}
6789 
6790 instruct cmovIF_reg(cmpOpF cmp, flagsRegF fcc, iRegI dst, iRegI src) %{
6791   match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6792   ins_cost(150);
6793   size(4);
6794   format %{ "MOV$cmp $fcc,$src,$dst" %}
6795   ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6796   ins_pipe(ialu_reg);
6797 %}
6798 
6799 instruct cmovIF_imm(cmpOpF cmp, flagsRegF fcc, iRegI dst, immI11 src) %{
6800   match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6801   ins_cost(140);
6802   size(4);
6803   format %{ "MOV$cmp $fcc,$src,$dst" %}
6804   ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6805   ins_pipe(ialu_imm);
6806 %}
6807 
6808 // Conditional move for RegN. Only cmov(reg,reg).
6809 instruct cmovNP_reg(cmpOpP cmp, flagsRegP pcc, iRegN dst, iRegN src) %{
6810   match(Set dst (CMoveN (Binary cmp pcc) (Binary dst src)));
6811   ins_cost(150);
6812   format %{ "MOV$cmp $pcc,$src,$dst" %}
6813   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6814   ins_pipe(ialu_reg);
6815 %}
6816 
6817 // This instruction also works with CmpN so we don't need cmovNN_reg.
6818 instruct cmovNI_reg(cmpOp cmp, flagsReg icc, iRegN dst, iRegN src) %{
6819   match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6820   ins_cost(150);
6821   size(4);
6822   format %{ "MOV$cmp  $icc,$src,$dst" %}
6823   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6824   ins_pipe(ialu_reg);
6825 %}
6826 
6827 // This instruction also works with CmpN so we don't need cmovNN_reg.
6828 instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{
6829   match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6830   ins_cost(150);
6831   size(4);
6832   format %{ "MOV$cmp  $icc,$src,$dst" %}
6833   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6834   ins_pipe(ialu_reg);
6835 %}
6836 
6837 instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{
6838   match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src)));
6839   ins_cost(150);
6840   size(4);
6841   format %{ "MOV$cmp $fcc,$src,$dst" %}
6842   ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6843   ins_pipe(ialu_reg);
6844 %}
6845 
6846 // Conditional move
6847 instruct cmovPP_reg(cmpOpP cmp, flagsRegP pcc, iRegP dst, iRegP src) %{
6848   match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6849   ins_cost(150);
6850   format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6851   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6852   ins_pipe(ialu_reg);
6853 %}
6854 
6855 instruct cmovPP_imm(cmpOpP cmp, flagsRegP pcc, iRegP dst, immP0 src) %{
6856   match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6857   ins_cost(140);
6858   format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6859   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6860   ins_pipe(ialu_imm);
6861 %}
6862 
6863 // This instruction also works with CmpN so we don't need cmovPN_reg.
6864 instruct cmovPI_reg(cmpOp cmp, flagsReg icc, iRegP dst, iRegP src) %{
6865   match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6866   ins_cost(150);
6867 
6868   size(4);
6869   format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
6870   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6871   ins_pipe(ialu_reg);
6872 %}
6873 
6874 instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{
6875   match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6876   ins_cost(150);
6877 
6878   size(4);
6879   format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
6880   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6881   ins_pipe(ialu_reg);
6882 %}
6883 
6884 instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{
6885   match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6886   ins_cost(140);
6887 
6888   size(4);
6889   format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
6890   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6891   ins_pipe(ialu_imm);
6892 %}
6893 
6894 instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{
6895   match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6896   ins_cost(140);
6897 
6898   size(4);
6899   format %{ "MOV$cmp  $icc,$src,$dst\t! ptr" %}
6900   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6901   ins_pipe(ialu_imm);
6902 %}
6903 
6904 instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{
6905   match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6906   ins_cost(150);
6907   size(4);
6908   format %{ "MOV$cmp $fcc,$src,$dst" %}
6909   ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6910   ins_pipe(ialu_imm);
6911 %}
6912 
6913 instruct cmovPF_imm(cmpOpF cmp, flagsRegF fcc, iRegP dst, immP0 src) %{
6914   match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6915   ins_cost(140);
6916   size(4);
6917   format %{ "MOV$cmp $fcc,$src,$dst" %}
6918   ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6919   ins_pipe(ialu_imm);
6920 %}
6921 
6922 // Conditional move
6923 instruct cmovFP_reg(cmpOpP cmp, flagsRegP pcc, regF dst, regF src) %{
6924   match(Set dst (CMoveF (Binary cmp pcc) (Binary dst src)));
6925   ins_cost(150);
6926   opcode(0x101);
6927   format %{ "FMOVD$cmp $pcc,$src,$dst" %}
6928   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6929   ins_pipe(int_conditional_float_move);
6930 %}
6931 
6932 instruct cmovFI_reg(cmpOp cmp, flagsReg icc, regF dst, regF src) %{
6933   match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6934   ins_cost(150);
6935 
6936   size(4);
6937   format %{ "FMOVS$cmp $icc,$src,$dst" %}
6938   opcode(0x101);
6939   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6940   ins_pipe(int_conditional_float_move);
6941 %}
6942 
6943 instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{
6944   match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6945   ins_cost(150);
6946 
6947   size(4);
6948   format %{ "FMOVS$cmp $icc,$src,$dst" %}
6949   opcode(0x101);
6950   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6951   ins_pipe(int_conditional_float_move);
6952 %}
6953 
6954 // Conditional move,
6955 instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{
6956   match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src)));
6957   ins_cost(150);
6958   size(4);
6959   format %{ "FMOVF$cmp $fcc,$src,$dst" %}
6960   opcode(0x1);
6961   ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
6962   ins_pipe(int_conditional_double_move);
6963 %}
6964 
6965 // Conditional move
6966 instruct cmovDP_reg(cmpOpP cmp, flagsRegP pcc, regD dst, regD src) %{
6967   match(Set dst (CMoveD (Binary cmp pcc) (Binary dst src)));
6968   ins_cost(150);
6969   size(4);
6970   opcode(0x102);
6971   format %{ "FMOVD$cmp $pcc,$src,$dst" %}
6972   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6973   ins_pipe(int_conditional_double_move);
6974 %}
6975 
6976 instruct cmovDI_reg(cmpOp cmp, flagsReg icc, regD dst, regD src) %{
6977   match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
6978   ins_cost(150);
6979 
6980   size(4);
6981   format %{ "FMOVD$cmp $icc,$src,$dst" %}
6982   opcode(0x102);
6983   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6984   ins_pipe(int_conditional_double_move);
6985 %}
6986 
6987 instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{
6988   match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
6989   ins_cost(150);
6990 
6991   size(4);
6992   format %{ "FMOVD$cmp $icc,$src,$dst" %}
6993   opcode(0x102);
6994   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6995   ins_pipe(int_conditional_double_move);
6996 %}
6997 
6998 // Conditional move,
6999 instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{
7000   match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src)));
7001   ins_cost(150);
7002   size(4);
7003   format %{ "FMOVD$cmp $fcc,$src,$dst" %}
7004   opcode(0x2);
7005   ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
7006   ins_pipe(int_conditional_double_move);
7007 %}
7008 
7009 // Conditional move
7010 instruct cmovLP_reg(cmpOpP cmp, flagsRegP pcc, iRegL dst, iRegL src) %{
7011   match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7012   ins_cost(150);
7013   format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7014   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
7015   ins_pipe(ialu_reg);
7016 %}
7017 
7018 instruct cmovLP_imm(cmpOpP cmp, flagsRegP pcc, iRegL dst, immI11 src) %{
7019   match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7020   ins_cost(140);
7021   format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7022   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
7023   ins_pipe(ialu_imm);
7024 %}
7025 
7026 instruct cmovLI_reg(cmpOp cmp, flagsReg icc, iRegL dst, iRegL src) %{
7027   match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7028   ins_cost(150);
7029 
7030   size(4);
7031   format %{ "MOV$cmp  $icc,$src,$dst\t! long" %}
7032   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7033   ins_pipe(ialu_reg);
7034 %}
7035 
7036 
7037 instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{
7038   match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7039   ins_cost(150);
7040 
7041   size(4);
7042   format %{ "MOV$cmp  $icc,$src,$dst\t! long" %}
7043   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7044   ins_pipe(ialu_reg);
7045 %}
7046 
7047 
7048 instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{
7049   match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src)));
7050   ins_cost(150);
7051 
7052   size(4);
7053   format %{ "MOV$cmp  $fcc,$src,$dst\t! long" %}
7054   ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
7055   ins_pipe(ialu_reg);
7056 %}
7057 
7058 
7059 
7060 //----------OS and Locking Instructions----------------------------------------
7061 
7062 // This name is KNOWN by the ADLC and cannot be changed.
7063 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type
7064 // for this guy.
7065 instruct tlsLoadP(g2RegP dst) %{
7066   match(Set dst (ThreadLocal));
7067 
7068   size(0);
7069   ins_cost(0);
7070   format %{ "# TLS is in G2" %}
7071   ins_encode( /*empty encoding*/ );
7072   ins_pipe(ialu_none);
7073 %}
7074 
7075 instruct checkCastPP( iRegP dst ) %{
7076   match(Set dst (CheckCastPP dst));
7077 
7078   size(0);
7079   format %{ "# checkcastPP of $dst" %}
7080   ins_encode( /*empty encoding*/ );
7081   ins_pipe(empty);
7082 %}
7083 
7084 
7085 instruct castPP( iRegP dst ) %{
7086   match(Set dst (CastPP dst));
7087   format %{ "# castPP of $dst" %}
7088   ins_encode( /*empty encoding*/ );
7089   ins_pipe(empty);
7090 %}
7091 
7092 instruct castII( iRegI dst ) %{
7093   match(Set dst (CastII dst));
7094   format %{ "# castII of $dst" %}
7095   ins_encode( /*empty encoding*/ );
7096   ins_cost(0);
7097   ins_pipe(empty);
7098 %}
7099 
7100 //----------Arithmetic Instructions--------------------------------------------
7101 // Addition Instructions
7102 // Register Addition
7103 instruct addI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7104   match(Set dst (AddI src1 src2));
7105 
7106   size(4);
7107   format %{ "ADD    $src1,$src2,$dst" %}
7108   ins_encode %{
7109     __ add($src1$$Register, $src2$$Register, $dst$$Register);
7110   %}
7111   ins_pipe(ialu_reg_reg);
7112 %}
7113 
7114 // Immediate Addition
7115 instruct addI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7116   match(Set dst (AddI src1 src2));
7117 
7118   size(4);
7119   format %{ "ADD    $src1,$src2,$dst" %}
7120   opcode(Assembler::add_op3, Assembler::arith_op);
7121   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7122   ins_pipe(ialu_reg_imm);
7123 %}
7124 
7125 // Pointer Register Addition
7126 instruct addP_reg_reg(iRegP dst, iRegP src1, iRegX src2) %{
7127   match(Set dst (AddP src1 src2));
7128 
7129   size(4);
7130   format %{ "ADD    $src1,$src2,$dst" %}
7131   opcode(Assembler::add_op3, Assembler::arith_op);
7132   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7133   ins_pipe(ialu_reg_reg);
7134 %}
7135 
7136 // Pointer Immediate Addition
7137 instruct addP_reg_imm13(iRegP dst, iRegP src1, immX13 src2) %{
7138   match(Set dst (AddP src1 src2));
7139 
7140   size(4);
7141   format %{ "ADD    $src1,$src2,$dst" %}
7142   opcode(Assembler::add_op3, Assembler::arith_op);
7143   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7144   ins_pipe(ialu_reg_imm);
7145 %}
7146 
7147 // Long Addition
7148 instruct addL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7149   match(Set dst (AddL src1 src2));
7150 
7151   size(4);
7152   format %{ "ADD    $src1,$src2,$dst\t! long" %}
7153   opcode(Assembler::add_op3, Assembler::arith_op);
7154   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7155   ins_pipe(ialu_reg_reg);
7156 %}
7157 
7158 instruct addL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7159   match(Set dst (AddL src1 con));
7160 
7161   size(4);
7162   format %{ "ADD    $src1,$con,$dst" %}
7163   opcode(Assembler::add_op3, Assembler::arith_op);
7164   ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7165   ins_pipe(ialu_reg_imm);
7166 %}
7167 
7168 //----------Conditional_store--------------------------------------------------
7169 // Conditional-store of the updated heap-top.
7170 // Used during allocation of the shared heap.
7171 // Sets flags (EQ) on success.  Implemented with a CASA on Sparc.
7172 
7173 // LoadP-locked.  Same as a regular pointer load when used with a compare-swap
7174 instruct loadPLocked(iRegP dst, memory mem) %{
7175   match(Set dst (LoadPLocked mem));
7176   ins_cost(MEMORY_REF_COST);
7177 
7178 #ifndef _LP64
7179   size(4);
7180   format %{ "LDUW   $mem,$dst\t! ptr" %}
7181   opcode(Assembler::lduw_op3, 0, REGP_OP);
7182 #else
7183   format %{ "LDX    $mem,$dst\t! ptr" %}
7184   opcode(Assembler::ldx_op3, 0, REGP_OP);
7185 #endif
7186   ins_encode( form3_mem_reg( mem, dst ) );
7187   ins_pipe(iload_mem);
7188 %}
7189 
7190 instruct storePConditional( iRegP heap_top_ptr, iRegP oldval, g3RegP newval, flagsRegP pcc ) %{
7191   match(Set pcc (StorePConditional heap_top_ptr (Binary oldval newval)));
7192   effect( KILL newval );
7193   format %{ "CASA   [$heap_top_ptr],$oldval,R_G3\t! If $oldval==[$heap_top_ptr] Then store R_G3 into [$heap_top_ptr], set R_G3=[$heap_top_ptr] in any case\n\t"
7194             "CMP    R_G3,$oldval\t\t! See if we made progress"  %}
7195   ins_encode( enc_cas(heap_top_ptr,oldval,newval) );
7196   ins_pipe( long_memory_op );
7197 %}
7198 
7199 // Conditional-store of an int value.
7200 instruct storeIConditional( iRegP mem_ptr, iRegI oldval, g3RegI newval, flagsReg icc ) %{
7201   match(Set icc (StoreIConditional mem_ptr (Binary oldval newval)));
7202   effect( KILL newval );
7203   format %{ "CASA   [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7204             "CMP    $oldval,$newval\t\t! See if we made progress"  %}
7205   ins_encode( enc_cas(mem_ptr,oldval,newval) );
7206   ins_pipe( long_memory_op );
7207 %}
7208 
7209 // Conditional-store of a long value.
7210 instruct storeLConditional( iRegP mem_ptr, iRegL oldval, g3RegL newval, flagsRegL xcc ) %{
7211   match(Set xcc (StoreLConditional mem_ptr (Binary oldval newval)));
7212   effect( KILL newval );
7213   format %{ "CASXA  [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7214             "CMP    $oldval,$newval\t\t! See if we made progress"  %}
7215   ins_encode( enc_cas(mem_ptr,oldval,newval) );
7216   ins_pipe( long_memory_op );
7217 %}
7218 
7219 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7220 
7221 instruct compareAndSwapL_bool(iRegP mem_ptr, iRegL oldval, iRegL newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7222   predicate(VM_Version::supports_cx8());
7223   match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7224   effect( USE mem_ptr, KILL ccr, KILL tmp1);
7225   format %{
7226             "MOV    $newval,O7\n\t"
7227             "CASXA  [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7228             "CMP    $oldval,O7\t\t! See if we made progress\n\t"
7229             "MOV    1,$res\n\t"
7230             "MOVne  xcc,R_G0,$res"
7231   %}
7232   ins_encode( enc_casx(mem_ptr, oldval, newval),
7233               enc_lflags_ne_to_boolean(res) );
7234   ins_pipe( long_memory_op );
7235 %}
7236 
7237 
7238 instruct compareAndSwapI_bool(iRegP mem_ptr, iRegI oldval, iRegI newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7239   match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7240   effect( USE mem_ptr, KILL ccr, KILL tmp1);
7241   format %{
7242             "MOV    $newval,O7\n\t"
7243             "CASA   [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7244             "CMP    $oldval,O7\t\t! See if we made progress\n\t"
7245             "MOV    1,$res\n\t"
7246             "MOVne  icc,R_G0,$res"
7247   %}
7248   ins_encode( enc_casi(mem_ptr, oldval, newval),
7249               enc_iflags_ne_to_boolean(res) );
7250   ins_pipe( long_memory_op );
7251 %}
7252 
7253 instruct compareAndSwapP_bool(iRegP mem_ptr, iRegP oldval, iRegP newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7254 #ifdef _LP64
7255   predicate(VM_Version::supports_cx8());
7256 #endif
7257   match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7258   effect( USE mem_ptr, KILL ccr, KILL tmp1);
7259   format %{
7260             "MOV    $newval,O7\n\t"
7261             "CASA_PTR  [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7262             "CMP    $oldval,O7\t\t! See if we made progress\n\t"
7263             "MOV    1,$res\n\t"
7264             "MOVne  xcc,R_G0,$res"
7265   %}
7266 #ifdef _LP64
7267   ins_encode( enc_casx(mem_ptr, oldval, newval),
7268               enc_lflags_ne_to_boolean(res) );
7269 #else
7270   ins_encode( enc_casi(mem_ptr, oldval, newval),
7271               enc_iflags_ne_to_boolean(res) );
7272 #endif
7273   ins_pipe( long_memory_op );
7274 %}
7275 
7276 instruct compareAndSwapN_bool(iRegP mem_ptr, iRegN oldval, iRegN newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7277   match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
7278   effect( USE mem_ptr, KILL ccr, KILL tmp1);
7279   format %{
7280             "MOV    $newval,O7\n\t"
7281             "CASA   [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7282             "CMP    $oldval,O7\t\t! See if we made progress\n\t"
7283             "MOV    1,$res\n\t"
7284             "MOVne  icc,R_G0,$res"
7285   %}
7286   ins_encode( enc_casi(mem_ptr, oldval, newval),
7287               enc_iflags_ne_to_boolean(res) );
7288   ins_pipe( long_memory_op );
7289 %}
7290 
7291 instruct xchgI( memory mem, iRegI newval) %{
7292   match(Set newval (GetAndSetI mem newval));
7293   format %{ "SWAP  [$mem],$newval" %}
7294   size(4);
7295   ins_encode %{
7296     __ swap($mem$$Address, $newval$$Register);
7297   %}
7298   ins_pipe( long_memory_op );
7299 %}
7300 
7301 #ifndef _LP64
7302 instruct xchgP( memory mem, iRegP newval) %{
7303   match(Set newval (GetAndSetP mem newval));
7304   format %{ "SWAP  [$mem],$newval" %}
7305   size(4);
7306   ins_encode %{
7307     __ swap($mem$$Address, $newval$$Register);
7308   %}
7309   ins_pipe( long_memory_op );
7310 %}
7311 #endif
7312 
7313 instruct xchgN( memory mem, iRegN newval) %{
7314   match(Set newval (GetAndSetN mem newval));
7315   format %{ "SWAP  [$mem],$newval" %}
7316   size(4);
7317   ins_encode %{
7318     __ swap($mem$$Address, $newval$$Register);
7319   %}
7320   ins_pipe( long_memory_op );
7321 %}
7322 
7323 //---------------------
7324 // Subtraction Instructions
7325 // Register Subtraction
7326 instruct subI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7327   match(Set dst (SubI src1 src2));
7328 
7329   size(4);
7330   format %{ "SUB    $src1,$src2,$dst" %}
7331   opcode(Assembler::sub_op3, Assembler::arith_op);
7332   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7333   ins_pipe(ialu_reg_reg);
7334 %}
7335 
7336 // Immediate Subtraction
7337 instruct subI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7338   match(Set dst (SubI src1 src2));
7339 
7340   size(4);
7341   format %{ "SUB    $src1,$src2,$dst" %}
7342   opcode(Assembler::sub_op3, Assembler::arith_op);
7343   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7344   ins_pipe(ialu_reg_imm);
7345 %}
7346 
7347 instruct subI_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
7348   match(Set dst (SubI zero src2));
7349 
7350   size(4);
7351   format %{ "NEG    $src2,$dst" %}
7352   opcode(Assembler::sub_op3, Assembler::arith_op);
7353   ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7354   ins_pipe(ialu_zero_reg);
7355 %}
7356 
7357 // Long subtraction
7358 instruct subL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7359   match(Set dst (SubL src1 src2));
7360 
7361   size(4);
7362   format %{ "SUB    $src1,$src2,$dst\t! long" %}
7363   opcode(Assembler::sub_op3, Assembler::arith_op);
7364   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7365   ins_pipe(ialu_reg_reg);
7366 %}
7367 
7368 // Immediate Subtraction
7369 instruct subL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7370   match(Set dst (SubL src1 con));
7371 
7372   size(4);
7373   format %{ "SUB    $src1,$con,$dst\t! long" %}
7374   opcode(Assembler::sub_op3, Assembler::arith_op);
7375   ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7376   ins_pipe(ialu_reg_imm);
7377 %}
7378 
7379 // Long negation
7380 instruct negL_reg_reg(iRegL dst, immL0 zero, iRegL src2) %{
7381   match(Set dst (SubL zero src2));
7382 
7383   size(4);
7384   format %{ "NEG    $src2,$dst\t! long" %}
7385   opcode(Assembler::sub_op3, Assembler::arith_op);
7386   ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7387   ins_pipe(ialu_zero_reg);
7388 %}
7389 
7390 // Multiplication Instructions
7391 // Integer Multiplication
7392 // Register Multiplication
7393 instruct mulI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7394   match(Set dst (MulI src1 src2));
7395 
7396   size(4);
7397   format %{ "MULX   $src1,$src2,$dst" %}
7398   opcode(Assembler::mulx_op3, Assembler::arith_op);
7399   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7400   ins_pipe(imul_reg_reg);
7401 %}
7402 
7403 // Immediate Multiplication
7404 instruct mulI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7405   match(Set dst (MulI src1 src2));
7406 
7407   size(4);
7408   format %{ "MULX   $src1,$src2,$dst" %}
7409   opcode(Assembler::mulx_op3, Assembler::arith_op);
7410   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7411   ins_pipe(imul_reg_imm);
7412 %}
7413 
7414 instruct mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7415   match(Set dst (MulL src1 src2));
7416   ins_cost(DEFAULT_COST * 5);
7417   size(4);
7418   format %{ "MULX   $src1,$src2,$dst\t! long" %}
7419   opcode(Assembler::mulx_op3, Assembler::arith_op);
7420   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7421   ins_pipe(mulL_reg_reg);
7422 %}
7423 
7424 // Immediate Multiplication
7425 instruct mulL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7426   match(Set dst (MulL src1 src2));
7427   ins_cost(DEFAULT_COST * 5);
7428   size(4);
7429   format %{ "MULX   $src1,$src2,$dst" %}
7430   opcode(Assembler::mulx_op3, Assembler::arith_op);
7431   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7432   ins_pipe(mulL_reg_imm);
7433 %}
7434 
7435 // Integer Division
7436 // Register Division
7437 instruct divI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2) %{
7438   match(Set dst (DivI src1 src2));
7439   ins_cost((2+71)*DEFAULT_COST);
7440 
7441   format %{ "SRA     $src2,0,$src2\n\t"
7442             "SRA     $src1,0,$src1\n\t"
7443             "SDIVX   $src1,$src2,$dst" %}
7444   ins_encode( idiv_reg( src1, src2, dst ) );
7445   ins_pipe(sdiv_reg_reg);
7446 %}
7447 
7448 // Immediate Division
7449 instruct divI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2) %{
7450   match(Set dst (DivI src1 src2));
7451   ins_cost((2+71)*DEFAULT_COST);
7452 
7453   format %{ "SRA     $src1,0,$src1\n\t"
7454             "SDIVX   $src1,$src2,$dst" %}
7455   ins_encode( idiv_imm( src1, src2, dst ) );
7456   ins_pipe(sdiv_reg_imm);
7457 %}
7458 
7459 //----------Div-By-10-Expansion------------------------------------------------
7460 // Extract hi bits of a 32x32->64 bit multiply.
7461 // Expand rule only, not matched
7462 instruct mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2 ) %{
7463   effect( DEF dst, USE src1, USE src2 );
7464   format %{ "MULX   $src1,$src2,$dst\t! Used in div-by-10\n\t"
7465             "SRLX   $dst,#32,$dst\t\t! Extract only hi word of result" %}
7466   ins_encode( enc_mul_hi(dst,src1,src2));
7467   ins_pipe(sdiv_reg_reg);
7468 %}
7469 
7470 // Magic constant, reciprocal of 10
7471 instruct loadConI_x66666667(iRegIsafe dst) %{
7472   effect( DEF dst );
7473 
7474   size(8);
7475   format %{ "SET    0x66666667,$dst\t! Used in div-by-10" %}
7476   ins_encode( Set32(0x66666667, dst) );
7477   ins_pipe(ialu_hi_lo_reg);
7478 %}
7479 
7480 // Register Shift Right Arithmetic Long by 32-63
7481 instruct sra_31( iRegI dst, iRegI src ) %{
7482   effect( DEF dst, USE src );
7483   format %{ "SRA    $src,31,$dst\t! Used in div-by-10" %}
7484   ins_encode( form3_rs1_rd_copysign_hi(src,dst) );
7485   ins_pipe(ialu_reg_reg);
7486 %}
7487 
7488 // Arithmetic Shift Right by 8-bit immediate
7489 instruct sra_reg_2( iRegI dst, iRegI src ) %{
7490   effect( DEF dst, USE src );
7491   format %{ "SRA    $src,2,$dst\t! Used in div-by-10" %}
7492   opcode(Assembler::sra_op3, Assembler::arith_op);
7493   ins_encode( form3_rs1_simm13_rd( src, 0x2, dst ) );
7494   ins_pipe(ialu_reg_imm);
7495 %}
7496 
7497 // Integer DIV with 10
7498 instruct divI_10( iRegI dst, iRegIsafe src, immI10 div ) %{
7499   match(Set dst (DivI src div));
7500   ins_cost((6+6)*DEFAULT_COST);
7501   expand %{
7502     iRegIsafe tmp1;               // Killed temps;
7503     iRegIsafe tmp2;               // Killed temps;
7504     iRegI tmp3;                   // Killed temps;
7505     iRegI tmp4;                   // Killed temps;
7506     loadConI_x66666667( tmp1 );   // SET  0x66666667 -> tmp1
7507     mul_hi( tmp2, src, tmp1 );    // MUL  hibits(src * tmp1) -> tmp2
7508     sra_31( tmp3, src );          // SRA  src,31 -> tmp3
7509     sra_reg_2( tmp4, tmp2 );      // SRA  tmp2,2 -> tmp4
7510     subI_reg_reg( dst,tmp4,tmp3); // SUB  tmp4 - tmp3 -> dst
7511   %}
7512 %}
7513 
7514 // Register Long Division
7515 instruct divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7516   match(Set dst (DivL src1 src2));
7517   ins_cost(DEFAULT_COST*71);
7518   size(4);
7519   format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
7520   opcode(Assembler::sdivx_op3, Assembler::arith_op);
7521   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7522   ins_pipe(divL_reg_reg);
7523 %}
7524 
7525 // Register Long Division
7526 instruct divL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7527   match(Set dst (DivL src1 src2));
7528   ins_cost(DEFAULT_COST*71);
7529   size(4);
7530   format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
7531   opcode(Assembler::sdivx_op3, Assembler::arith_op);
7532   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7533   ins_pipe(divL_reg_imm);
7534 %}
7535 
7536 // Integer Remainder
7537 // Register Remainder
7538 instruct modI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2, o7RegP temp, flagsReg ccr ) %{
7539   match(Set dst (ModI src1 src2));
7540   effect( KILL ccr, KILL temp);
7541 
7542   format %{ "SREM   $src1,$src2,$dst" %}
7543   ins_encode( irem_reg(src1, src2, dst, temp) );
7544   ins_pipe(sdiv_reg_reg);
7545 %}
7546 
7547 // Immediate Remainder
7548 instruct modI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2, o7RegP temp, flagsReg ccr ) %{
7549   match(Set dst (ModI src1 src2));
7550   effect( KILL ccr, KILL temp);
7551 
7552   format %{ "SREM   $src1,$src2,$dst" %}
7553   ins_encode( irem_imm(src1, src2, dst, temp) );
7554   ins_pipe(sdiv_reg_imm);
7555 %}
7556 
7557 // Register Long Remainder
7558 instruct divL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7559   effect(DEF dst, USE src1, USE src2);
7560   size(4);
7561   format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
7562   opcode(Assembler::sdivx_op3, Assembler::arith_op);
7563   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7564   ins_pipe(divL_reg_reg);
7565 %}
7566 
7567 // Register Long Division
7568 instruct divL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7569   effect(DEF dst, USE src1, USE src2);
7570   size(4);
7571   format %{ "SDIVX  $src1,$src2,$dst\t! long" %}
7572   opcode(Assembler::sdivx_op3, Assembler::arith_op);
7573   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7574   ins_pipe(divL_reg_imm);
7575 %}
7576 
7577 instruct mulL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7578   effect(DEF dst, USE src1, USE src2);
7579   size(4);
7580   format %{ "MULX   $src1,$src2,$dst\t! long" %}
7581   opcode(Assembler::mulx_op3, Assembler::arith_op);
7582   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7583   ins_pipe(mulL_reg_reg);
7584 %}
7585 
7586 // Immediate Multiplication
7587 instruct mulL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7588   effect(DEF dst, USE src1, USE src2);
7589   size(4);
7590   format %{ "MULX   $src1,$src2,$dst" %}
7591   opcode(Assembler::mulx_op3, Assembler::arith_op);
7592   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7593   ins_pipe(mulL_reg_imm);
7594 %}
7595 
7596 instruct subL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7597   effect(DEF dst, USE src1, USE src2);
7598   size(4);
7599   format %{ "SUB    $src1,$src2,$dst\t! long" %}
7600   opcode(Assembler::sub_op3, Assembler::arith_op);
7601   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7602   ins_pipe(ialu_reg_reg);
7603 %}
7604 
7605 instruct subL_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
7606   effect(DEF dst, USE src1, USE src2);
7607   size(4);
7608   format %{ "SUB    $src1,$src2,$dst\t! long" %}
7609   opcode(Assembler::sub_op3, Assembler::arith_op);
7610   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7611   ins_pipe(ialu_reg_reg);
7612 %}
7613 
7614 // Register Long Remainder
7615 instruct modL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7616   match(Set dst (ModL src1 src2));
7617   ins_cost(DEFAULT_COST*(71 + 6 + 1));
7618   expand %{
7619     iRegL tmp1;
7620     iRegL tmp2;
7621     divL_reg_reg_1(tmp1, src1, src2);
7622     mulL_reg_reg_1(tmp2, tmp1, src2);
7623     subL_reg_reg_1(dst,  src1, tmp2);
7624   %}
7625 %}
7626 
7627 // Register Long Remainder
7628 instruct modL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7629   match(Set dst (ModL src1 src2));
7630   ins_cost(DEFAULT_COST*(71 + 6 + 1));
7631   expand %{
7632     iRegL tmp1;
7633     iRegL tmp2;
7634     divL_reg_imm13_1(tmp1, src1, src2);
7635     mulL_reg_imm13_1(tmp2, tmp1, src2);
7636     subL_reg_reg_2  (dst,  src1, tmp2);
7637   %}
7638 %}
7639 
7640 // Integer Shift Instructions
7641 // Register Shift Left
7642 instruct shlI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7643   match(Set dst (LShiftI src1 src2));
7644 
7645   size(4);
7646   format %{ "SLL    $src1,$src2,$dst" %}
7647   opcode(Assembler::sll_op3, Assembler::arith_op);
7648   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7649   ins_pipe(ialu_reg_reg);
7650 %}
7651 
7652 // Register Shift Left Immediate
7653 instruct shlI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7654   match(Set dst (LShiftI src1 src2));
7655 
7656   size(4);
7657   format %{ "SLL    $src1,$src2,$dst" %}
7658   opcode(Assembler::sll_op3, Assembler::arith_op);
7659   ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7660   ins_pipe(ialu_reg_imm);
7661 %}
7662 
7663 // Register Shift Left
7664 instruct shlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7665   match(Set dst (LShiftL src1 src2));
7666 
7667   size(4);
7668   format %{ "SLLX   $src1,$src2,$dst" %}
7669   opcode(Assembler::sllx_op3, Assembler::arith_op);
7670   ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7671   ins_pipe(ialu_reg_reg);
7672 %}
7673 
7674 // Register Shift Left Immediate
7675 instruct shlL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7676   match(Set dst (LShiftL src1 src2));
7677 
7678   size(4);
7679   format %{ "SLLX   $src1,$src2,$dst" %}
7680   opcode(Assembler::sllx_op3, Assembler::arith_op);
7681   ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7682   ins_pipe(ialu_reg_imm);
7683 %}
7684 
7685 // Register Arithmetic Shift Right
7686 instruct sarI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7687   match(Set dst (RShiftI src1 src2));
7688   size(4);
7689   format %{ "SRA    $src1,$src2,$dst" %}
7690   opcode(Assembler::sra_op3, Assembler::arith_op);
7691   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7692   ins_pipe(ialu_reg_reg);
7693 %}
7694 
7695 // Register Arithmetic Shift Right Immediate
7696 instruct sarI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7697   match(Set dst (RShiftI src1 src2));
7698 
7699   size(4);
7700   format %{ "SRA    $src1,$src2,$dst" %}
7701   opcode(Assembler::sra_op3, Assembler::arith_op);
7702   ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7703   ins_pipe(ialu_reg_imm);
7704 %}
7705 
7706 // Register Shift Right Arithmatic Long
7707 instruct sarL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7708   match(Set dst (RShiftL src1 src2));
7709 
7710   size(4);
7711   format %{ "SRAX   $src1,$src2,$dst" %}
7712   opcode(Assembler::srax_op3, Assembler::arith_op);
7713   ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7714   ins_pipe(ialu_reg_reg);
7715 %}
7716 
7717 // Register Shift Left Immediate
7718 instruct sarL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7719   match(Set dst (RShiftL src1 src2));
7720 
7721   size(4);
7722   format %{ "SRAX   $src1,$src2,$dst" %}
7723   opcode(Assembler::srax_op3, Assembler::arith_op);
7724   ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7725   ins_pipe(ialu_reg_imm);
7726 %}
7727 
7728 // Register Shift Right
7729 instruct shrI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7730   match(Set dst (URShiftI src1 src2));
7731 
7732   size(4);
7733   format %{ "SRL    $src1,$src2,$dst" %}
7734   opcode(Assembler::srl_op3, Assembler::arith_op);
7735   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7736   ins_pipe(ialu_reg_reg);
7737 %}
7738 
7739 // Register Shift Right Immediate
7740 instruct shrI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7741   match(Set dst (URShiftI src1 src2));
7742 
7743   size(4);
7744   format %{ "SRL    $src1,$src2,$dst" %}
7745   opcode(Assembler::srl_op3, Assembler::arith_op);
7746   ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7747   ins_pipe(ialu_reg_imm);
7748 %}
7749 
7750 // Register Shift Right
7751 instruct shrL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7752   match(Set dst (URShiftL src1 src2));
7753 
7754   size(4);
7755   format %{ "SRLX   $src1,$src2,$dst" %}
7756   opcode(Assembler::srlx_op3, Assembler::arith_op);
7757   ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7758   ins_pipe(ialu_reg_reg);
7759 %}
7760 
7761 // Register Shift Right Immediate
7762 instruct shrL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7763   match(Set dst (URShiftL src1 src2));
7764 
7765   size(4);
7766   format %{ "SRLX   $src1,$src2,$dst" %}
7767   opcode(Assembler::srlx_op3, Assembler::arith_op);
7768   ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7769   ins_pipe(ialu_reg_imm);
7770 %}
7771 
7772 // Register Shift Right Immediate with a CastP2X
7773 #ifdef _LP64
7774 instruct shrP_reg_imm6(iRegL dst, iRegP src1, immU6 src2) %{
7775   match(Set dst (URShiftL (CastP2X src1) src2));
7776   size(4);
7777   format %{ "SRLX   $src1,$src2,$dst\t! Cast ptr $src1 to long and shift" %}
7778   opcode(Assembler::srlx_op3, Assembler::arith_op);
7779   ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7780   ins_pipe(ialu_reg_imm);
7781 %}
7782 #else
7783 instruct shrP_reg_imm5(iRegI dst, iRegP src1, immU5 src2) %{
7784   match(Set dst (URShiftI (CastP2X src1) src2));
7785   size(4);
7786   format %{ "SRL    $src1,$src2,$dst\t! Cast ptr $src1 to int and shift" %}
7787   opcode(Assembler::srl_op3, Assembler::arith_op);
7788   ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7789   ins_pipe(ialu_reg_imm);
7790 %}
7791 #endif
7792 
7793 
7794 //----------Floating Point Arithmetic Instructions-----------------------------
7795 
7796 //  Add float single precision
7797 instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
7798   match(Set dst (AddF src1 src2));
7799 
7800   size(4);
7801   format %{ "FADDS  $src1,$src2,$dst" %}
7802   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fadds_opf);
7803   ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7804   ins_pipe(faddF_reg_reg);
7805 %}
7806 
7807 //  Add float double precision
7808 instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
7809   match(Set dst (AddD src1 src2));
7810 
7811   size(4);
7812   format %{ "FADDD  $src1,$src2,$dst" %}
7813   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
7814   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7815   ins_pipe(faddD_reg_reg);
7816 %}
7817 
7818 //  Sub float single precision
7819 instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
7820   match(Set dst (SubF src1 src2));
7821 
7822   size(4);
7823   format %{ "FSUBS  $src1,$src2,$dst" %}
7824   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubs_opf);
7825   ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7826   ins_pipe(faddF_reg_reg);
7827 %}
7828 
7829 //  Sub float double precision
7830 instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
7831   match(Set dst (SubD src1 src2));
7832 
7833   size(4);
7834   format %{ "FSUBD  $src1,$src2,$dst" %}
7835   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
7836   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7837   ins_pipe(faddD_reg_reg);
7838 %}
7839 
7840 //  Mul float single precision
7841 instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
7842   match(Set dst (MulF src1 src2));
7843 
7844   size(4);
7845   format %{ "FMULS  $src1,$src2,$dst" %}
7846   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuls_opf);
7847   ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7848   ins_pipe(fmulF_reg_reg);
7849 %}
7850 
7851 //  Mul float double precision
7852 instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
7853   match(Set dst (MulD src1 src2));
7854 
7855   size(4);
7856   format %{ "FMULD  $src1,$src2,$dst" %}
7857   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
7858   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7859   ins_pipe(fmulD_reg_reg);
7860 %}
7861 
7862 //  Div float single precision
7863 instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
7864   match(Set dst (DivF src1 src2));
7865 
7866   size(4);
7867   format %{ "FDIVS  $src1,$src2,$dst" %}
7868   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivs_opf);
7869   ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7870   ins_pipe(fdivF_reg_reg);
7871 %}
7872 
7873 //  Div float double precision
7874 instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
7875   match(Set dst (DivD src1 src2));
7876 
7877   size(4);
7878   format %{ "FDIVD  $src1,$src2,$dst" %}
7879   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivd_opf);
7880   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7881   ins_pipe(fdivD_reg_reg);
7882 %}
7883 
7884 //  Absolute float double precision
7885 instruct absD_reg(regD dst, regD src) %{
7886   match(Set dst (AbsD src));
7887 
7888   format %{ "FABSd  $src,$dst" %}
7889   ins_encode(fabsd(dst, src));
7890   ins_pipe(faddD_reg);
7891 %}
7892 
7893 //  Absolute float single precision
7894 instruct absF_reg(regF dst, regF src) %{
7895   match(Set dst (AbsF src));
7896 
7897   format %{ "FABSs  $src,$dst" %}
7898   ins_encode(fabss(dst, src));
7899   ins_pipe(faddF_reg);
7900 %}
7901 
7902 instruct negF_reg(regF dst, regF src) %{
7903   match(Set dst (NegF src));
7904 
7905   size(4);
7906   format %{ "FNEGs  $src,$dst" %}
7907   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fnegs_opf);
7908   ins_encode(form3_opf_rs2F_rdF(src, dst));
7909   ins_pipe(faddF_reg);
7910 %}
7911 
7912 instruct negD_reg(regD dst, regD src) %{
7913   match(Set dst (NegD src));
7914 
7915   format %{ "FNEGd  $src,$dst" %}
7916   ins_encode(fnegd(dst, src));
7917   ins_pipe(faddD_reg);
7918 %}
7919 
7920 //  Sqrt float double precision
7921 instruct sqrtF_reg_reg(regF dst, regF src) %{
7922   match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
7923 
7924   size(4);
7925   format %{ "FSQRTS $src,$dst" %}
7926   ins_encode(fsqrts(dst, src));
7927   ins_pipe(fdivF_reg_reg);
7928 %}
7929 
7930 //  Sqrt float double precision
7931 instruct sqrtD_reg_reg(regD dst, regD src) %{
7932   match(Set dst (SqrtD src));
7933 
7934   size(4);
7935   format %{ "FSQRTD $src,$dst" %}
7936   ins_encode(fsqrtd(dst, src));
7937   ins_pipe(fdivD_reg_reg);
7938 %}
7939 
7940 //----------Logical Instructions-----------------------------------------------
7941 // And Instructions
7942 // Register And
7943 instruct andI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7944   match(Set dst (AndI src1 src2));
7945 
7946   size(4);
7947   format %{ "AND    $src1,$src2,$dst" %}
7948   opcode(Assembler::and_op3, Assembler::arith_op);
7949   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7950   ins_pipe(ialu_reg_reg);
7951 %}
7952 
7953 // Immediate And
7954 instruct andI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7955   match(Set dst (AndI src1 src2));
7956 
7957   size(4);
7958   format %{ "AND    $src1,$src2,$dst" %}
7959   opcode(Assembler::and_op3, Assembler::arith_op);
7960   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7961   ins_pipe(ialu_reg_imm);
7962 %}
7963 
7964 // Register And Long
7965 instruct andL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7966   match(Set dst (AndL src1 src2));
7967 
7968   ins_cost(DEFAULT_COST);
7969   size(4);
7970   format %{ "AND    $src1,$src2,$dst\t! long" %}
7971   opcode(Assembler::and_op3, Assembler::arith_op);
7972   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7973   ins_pipe(ialu_reg_reg);
7974 %}
7975 
7976 instruct andL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7977   match(Set dst (AndL src1 con));
7978 
7979   ins_cost(DEFAULT_COST);
7980   size(4);
7981   format %{ "AND    $src1,$con,$dst\t! long" %}
7982   opcode(Assembler::and_op3, Assembler::arith_op);
7983   ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7984   ins_pipe(ialu_reg_imm);
7985 %}
7986 
7987 // Or Instructions
7988 // Register Or
7989 instruct orI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7990   match(Set dst (OrI src1 src2));
7991 
7992   size(4);
7993   format %{ "OR     $src1,$src2,$dst" %}
7994   opcode(Assembler::or_op3, Assembler::arith_op);
7995   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7996   ins_pipe(ialu_reg_reg);
7997 %}
7998 
7999 // Immediate Or
8000 instruct orI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8001   match(Set dst (OrI src1 src2));
8002 
8003   size(4);
8004   format %{ "OR     $src1,$src2,$dst" %}
8005   opcode(Assembler::or_op3, Assembler::arith_op);
8006   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8007   ins_pipe(ialu_reg_imm);
8008 %}
8009 
8010 // Register Or Long
8011 instruct orL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8012   match(Set dst (OrL src1 src2));
8013 
8014   ins_cost(DEFAULT_COST);
8015   size(4);
8016   format %{ "OR     $src1,$src2,$dst\t! long" %}
8017   opcode(Assembler::or_op3, Assembler::arith_op);
8018   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8019   ins_pipe(ialu_reg_reg);
8020 %}
8021 
8022 instruct orL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8023   match(Set dst (OrL src1 con));
8024   ins_cost(DEFAULT_COST*2);
8025 
8026   ins_cost(DEFAULT_COST);
8027   size(4);
8028   format %{ "OR     $src1,$con,$dst\t! long" %}
8029   opcode(Assembler::or_op3, Assembler::arith_op);
8030   ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8031   ins_pipe(ialu_reg_imm);
8032 %}
8033 
8034 #ifndef _LP64
8035 
8036 // Use sp_ptr_RegP to match G2 (TLS register) without spilling.
8037 instruct orI_reg_castP2X(iRegI dst, iRegI src1, sp_ptr_RegP src2) %{
8038   match(Set dst (OrI src1 (CastP2X src2)));
8039 
8040   size(4);
8041   format %{ "OR     $src1,$src2,$dst" %}
8042   opcode(Assembler::or_op3, Assembler::arith_op);
8043   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8044   ins_pipe(ialu_reg_reg);
8045 %}
8046 
8047 #else
8048 
8049 instruct orL_reg_castP2X(iRegL dst, iRegL src1, sp_ptr_RegP src2) %{
8050   match(Set dst (OrL src1 (CastP2X src2)));
8051 
8052   ins_cost(DEFAULT_COST);
8053   size(4);
8054   format %{ "OR     $src1,$src2,$dst\t! long" %}
8055   opcode(Assembler::or_op3, Assembler::arith_op);
8056   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8057   ins_pipe(ialu_reg_reg);
8058 %}
8059 
8060 #endif
8061 
8062 // Xor Instructions
8063 // Register Xor
8064 instruct xorI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
8065   match(Set dst (XorI src1 src2));
8066 
8067   size(4);
8068   format %{ "XOR    $src1,$src2,$dst" %}
8069   opcode(Assembler::xor_op3, Assembler::arith_op);
8070   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8071   ins_pipe(ialu_reg_reg);
8072 %}
8073 
8074 // Immediate Xor
8075 instruct xorI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8076   match(Set dst (XorI src1 src2));
8077 
8078   size(4);
8079   format %{ "XOR    $src1,$src2,$dst" %}
8080   opcode(Assembler::xor_op3, Assembler::arith_op);
8081   ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8082   ins_pipe(ialu_reg_imm);
8083 %}
8084 
8085 // Register Xor Long
8086 instruct xorL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8087   match(Set dst (XorL src1 src2));
8088 
8089   ins_cost(DEFAULT_COST);
8090   size(4);
8091   format %{ "XOR    $src1,$src2,$dst\t! long" %}
8092   opcode(Assembler::xor_op3, Assembler::arith_op);
8093   ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8094   ins_pipe(ialu_reg_reg);
8095 %}
8096 
8097 instruct xorL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8098   match(Set dst (XorL src1 con));
8099 
8100   ins_cost(DEFAULT_COST);
8101   size(4);
8102   format %{ "XOR    $src1,$con,$dst\t! long" %}
8103   opcode(Assembler::xor_op3, Assembler::arith_op);
8104   ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8105   ins_pipe(ialu_reg_imm);
8106 %}
8107 
8108 //----------Convert to Boolean-------------------------------------------------
8109 // Nice hack for 32-bit tests but doesn't work for
8110 // 64-bit pointers.
8111 instruct convI2B( iRegI dst, iRegI src, flagsReg ccr ) %{
8112   match(Set dst (Conv2B src));
8113   effect( KILL ccr );
8114   ins_cost(DEFAULT_COST*2);
8115   format %{ "CMP    R_G0,$src\n\t"
8116             "ADDX   R_G0,0,$dst" %}
8117   ins_encode( enc_to_bool( src, dst ) );
8118   ins_pipe(ialu_reg_ialu);
8119 %}
8120 
8121 #ifndef _LP64
8122 instruct convP2B( iRegI dst, iRegP src, flagsReg ccr ) %{
8123   match(Set dst (Conv2B src));
8124   effect( KILL ccr );
8125   ins_cost(DEFAULT_COST*2);
8126   format %{ "CMP    R_G0,$src\n\t"
8127             "ADDX   R_G0,0,$dst" %}
8128   ins_encode( enc_to_bool( src, dst ) );
8129   ins_pipe(ialu_reg_ialu);
8130 %}
8131 #else
8132 instruct convP2B( iRegI dst, iRegP src ) %{
8133   match(Set dst (Conv2B src));
8134   ins_cost(DEFAULT_COST*2);
8135   format %{ "MOV    $src,$dst\n\t"
8136             "MOVRNZ $src,1,$dst" %}
8137   ins_encode( form3_g0_rs2_rd_move( src, dst ), enc_convP2B( dst, src ) );
8138   ins_pipe(ialu_clr_and_mover);
8139 %}
8140 #endif
8141 
8142 instruct cmpLTMask0( iRegI dst, iRegI src, immI0 zero, flagsReg ccr ) %{
8143   match(Set dst (CmpLTMask src zero));
8144   effect(KILL ccr);
8145   size(4);
8146   format %{ "SRA    $src,#31,$dst\t# cmpLTMask0" %}
8147   ins_encode %{
8148     __ sra($src$$Register, 31, $dst$$Register);
8149   %}
8150   ins_pipe(ialu_reg_imm);
8151 %}
8152 
8153 instruct cmpLTMask_reg_reg( iRegI dst, iRegI p, iRegI q, flagsReg ccr ) %{
8154   match(Set dst (CmpLTMask p q));
8155   effect( KILL ccr );
8156   ins_cost(DEFAULT_COST*4);
8157   format %{ "CMP    $p,$q\n\t"
8158             "MOV    #0,$dst\n\t"
8159             "BLT,a  .+8\n\t"
8160             "MOV    #-1,$dst" %}
8161   ins_encode( enc_ltmask(p,q,dst) );
8162   ins_pipe(ialu_reg_reg_ialu);
8163 %}
8164 
8165 instruct cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{
8166   match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
8167   effect(KILL ccr, TEMP tmp);
8168   ins_cost(DEFAULT_COST*3);
8169 
8170   format %{ "SUBcc  $p,$q,$p\t! p' = p-q\n\t"
8171             "ADD    $p,$y,$tmp\t! g3=p-q+y\n\t"
8172             "MOVlt  $tmp,$p\t! p' < 0 ? p'+y : p'" %}
8173   ins_encode(enc_cadd_cmpLTMask(p, q, y, tmp));
8174   ins_pipe(cadd_cmpltmask);
8175 %}
8176 
8177 instruct and_cmpLTMask(iRegI p, iRegI q, iRegI y, flagsReg ccr) %{
8178   match(Set p (AndI (CmpLTMask p q) y));
8179   effect(KILL ccr);
8180   ins_cost(DEFAULT_COST*3);
8181 
8182   format %{ "CMP  $p,$q\n\t"
8183             "MOV  $y,$p\n\t"
8184             "MOVge G0,$p" %}
8185   ins_encode %{
8186     __ cmp($p$$Register, $q$$Register);
8187     __ mov($y$$Register, $p$$Register);
8188     __ movcc(Assembler::greaterEqual, false, Assembler::icc, G0, $p$$Register);
8189   %}
8190   ins_pipe(ialu_reg_reg_ialu);
8191 %}
8192 
8193 //-----------------------------------------------------------------
8194 // Direct raw moves between float and general registers using VIS3.
8195 
8196 //  ins_pipe(faddF_reg);
8197 instruct MoveF2I_reg_reg(iRegI dst, regF src) %{
8198   predicate(UseVIS >= 3);
8199   match(Set dst (MoveF2I src));
8200 
8201   format %{ "MOVSTOUW $src,$dst\t! MoveF2I" %}
8202   ins_encode %{
8203     __ movstouw($src$$FloatRegister, $dst$$Register);
8204   %}
8205   ins_pipe(ialu_reg_reg);
8206 %}
8207 
8208 instruct MoveI2F_reg_reg(regF dst, iRegI src) %{
8209   predicate(UseVIS >= 3);
8210   match(Set dst (MoveI2F src));
8211 
8212   format %{ "MOVWTOS $src,$dst\t! MoveI2F" %}
8213   ins_encode %{
8214     __ movwtos($src$$Register, $dst$$FloatRegister);
8215   %}
8216   ins_pipe(ialu_reg_reg);
8217 %}
8218 
8219 instruct MoveD2L_reg_reg(iRegL dst, regD src) %{
8220   predicate(UseVIS >= 3);
8221   match(Set dst (MoveD2L src));
8222 
8223   format %{ "MOVDTOX $src,$dst\t! MoveD2L" %}
8224   ins_encode %{
8225     __ movdtox(as_DoubleFloatRegister($src$$reg), $dst$$Register);
8226   %}
8227   ins_pipe(ialu_reg_reg);
8228 %}
8229 
8230 instruct MoveL2D_reg_reg(regD dst, iRegL src) %{
8231   predicate(UseVIS >= 3);
8232   match(Set dst (MoveL2D src));
8233 
8234   format %{ "MOVXTOD $src,$dst\t! MoveL2D" %}
8235   ins_encode %{
8236     __ movxtod($src$$Register, as_DoubleFloatRegister($dst$$reg));
8237   %}
8238   ins_pipe(ialu_reg_reg);
8239 %}
8240 
8241 
8242 // Raw moves between float and general registers using stack.
8243 
8244 instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{
8245   match(Set dst (MoveF2I src));
8246   effect(DEF dst, USE src);
8247   ins_cost(MEMORY_REF_COST);
8248 
8249   size(4);
8250   format %{ "LDUW   $src,$dst\t! MoveF2I" %}
8251   opcode(Assembler::lduw_op3);
8252   ins_encode(simple_form3_mem_reg( src, dst ) );
8253   ins_pipe(iload_mem);
8254 %}
8255 
8256 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
8257   match(Set dst (MoveI2F src));
8258   effect(DEF dst, USE src);
8259   ins_cost(MEMORY_REF_COST);
8260 
8261   size(4);
8262   format %{ "LDF    $src,$dst\t! MoveI2F" %}
8263   opcode(Assembler::ldf_op3);
8264   ins_encode(simple_form3_mem_reg(src, dst));
8265   ins_pipe(floadF_stk);
8266 %}
8267 
8268 instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{
8269   match(Set dst (MoveD2L src));
8270   effect(DEF dst, USE src);
8271   ins_cost(MEMORY_REF_COST);
8272 
8273   size(4);
8274   format %{ "LDX    $src,$dst\t! MoveD2L" %}
8275   opcode(Assembler::ldx_op3);
8276   ins_encode(simple_form3_mem_reg( src, dst ) );
8277   ins_pipe(iload_mem);
8278 %}
8279 
8280 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
8281   match(Set dst (MoveL2D src));
8282   effect(DEF dst, USE src);
8283   ins_cost(MEMORY_REF_COST);
8284 
8285   size(4);
8286   format %{ "LDDF   $src,$dst\t! MoveL2D" %}
8287   opcode(Assembler::lddf_op3);
8288   ins_encode(simple_form3_mem_reg(src, dst));
8289   ins_pipe(floadD_stk);
8290 %}
8291 
8292 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
8293   match(Set dst (MoveF2I src));
8294   effect(DEF dst, USE src);
8295   ins_cost(MEMORY_REF_COST);
8296 
8297   size(4);
8298   format %{ "STF   $src,$dst\t! MoveF2I" %}
8299   opcode(Assembler::stf_op3);
8300   ins_encode(simple_form3_mem_reg(dst, src));
8301   ins_pipe(fstoreF_stk_reg);
8302 %}
8303 
8304 instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
8305   match(Set dst (MoveI2F src));
8306   effect(DEF dst, USE src);
8307   ins_cost(MEMORY_REF_COST);
8308 
8309   size(4);
8310   format %{ "STW    $src,$dst\t! MoveI2F" %}
8311   opcode(Assembler::stw_op3);
8312   ins_encode(simple_form3_mem_reg( dst, src ) );
8313   ins_pipe(istore_mem_reg);
8314 %}
8315 
8316 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
8317   match(Set dst (MoveD2L src));
8318   effect(DEF dst, USE src);
8319   ins_cost(MEMORY_REF_COST);
8320 
8321   size(4);
8322   format %{ "STDF   $src,$dst\t! MoveD2L" %}
8323   opcode(Assembler::stdf_op3);
8324   ins_encode(simple_form3_mem_reg(dst, src));
8325   ins_pipe(fstoreD_stk_reg);
8326 %}
8327 
8328 instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
8329   match(Set dst (MoveL2D src));
8330   effect(DEF dst, USE src);
8331   ins_cost(MEMORY_REF_COST);
8332 
8333   size(4);
8334   format %{ "STX    $src,$dst\t! MoveL2D" %}
8335   opcode(Assembler::stx_op3);
8336   ins_encode(simple_form3_mem_reg( dst, src ) );
8337   ins_pipe(istore_mem_reg);
8338 %}
8339 
8340 
8341 //----------Arithmetic Conversion Instructions---------------------------------
8342 // The conversions operations are all Alpha sorted.  Please keep it that way!
8343 
8344 instruct convD2F_reg(regF dst, regD src) %{
8345   match(Set dst (ConvD2F src));
8346   size(4);
8347   format %{ "FDTOS  $src,$dst" %}
8348   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdtos_opf);
8349   ins_encode(form3_opf_rs2D_rdF(src, dst));
8350   ins_pipe(fcvtD2F);
8351 %}
8352 
8353 
8354 // Convert a double to an int in a float register.
8355 // If the double is a NAN, stuff a zero in instead.
8356 instruct convD2I_helper(regF dst, regD src, flagsRegF0 fcc0) %{
8357   effect(DEF dst, USE src, KILL fcc0);
8358   format %{ "FCMPd  fcc0,$src,$src\t! check for NAN\n\t"
8359             "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8360             "FDTOI  $src,$dst\t! convert in delay slot\n\t"
8361             "FITOS  $dst,$dst\t! change NaN/max-int to valid float\n\t"
8362             "FSUBs  $dst,$dst,$dst\t! cleared only if nan\n"
8363       "skip:" %}
8364   ins_encode(form_d2i_helper(src,dst));
8365   ins_pipe(fcvtD2I);
8366 %}
8367 
8368 instruct convD2I_stk(stackSlotI dst, regD src) %{
8369   match(Set dst (ConvD2I src));
8370   ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8371   expand %{
8372     regF tmp;
8373     convD2I_helper(tmp, src);
8374     regF_to_stkI(dst, tmp);
8375   %}
8376 %}
8377 
8378 instruct convD2I_reg(iRegI dst, regD src) %{
8379   predicate(UseVIS >= 3);
8380   match(Set dst (ConvD2I src));
8381   ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8382   expand %{
8383     regF tmp;
8384     convD2I_helper(tmp, src);
8385     MoveF2I_reg_reg(dst, tmp);
8386   %}
8387 %}
8388 
8389 
8390 // Convert a double to a long in a double register.
8391 // If the double is a NAN, stuff a zero in instead.
8392 instruct convD2L_helper(regD dst, regD src, flagsRegF0 fcc0) %{
8393   effect(DEF dst, USE src, KILL fcc0);
8394   format %{ "FCMPd  fcc0,$src,$src\t! check for NAN\n\t"
8395             "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8396             "FDTOX  $src,$dst\t! convert in delay slot\n\t"
8397             "FXTOD  $dst,$dst\t! change NaN/max-long to valid double\n\t"
8398             "FSUBd  $dst,$dst,$dst\t! cleared only if nan\n"
8399       "skip:" %}
8400   ins_encode(form_d2l_helper(src,dst));
8401   ins_pipe(fcvtD2L);
8402 %}
8403 
8404 instruct convD2L_stk(stackSlotL dst, regD src) %{
8405   match(Set dst (ConvD2L src));
8406   ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8407   expand %{
8408     regD tmp;
8409     convD2L_helper(tmp, src);
8410     regD_to_stkL(dst, tmp);
8411   %}
8412 %}
8413 
8414 instruct convD2L_reg(iRegL dst, regD src) %{
8415   predicate(UseVIS >= 3);
8416   match(Set dst (ConvD2L src));
8417   ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8418   expand %{
8419     regD tmp;
8420     convD2L_helper(tmp, src);
8421     MoveD2L_reg_reg(dst, tmp);
8422   %}
8423 %}
8424 
8425 
8426 instruct convF2D_reg(regD dst, regF src) %{
8427   match(Set dst (ConvF2D src));
8428   format %{ "FSTOD  $src,$dst" %}
8429   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fstod_opf);
8430   ins_encode(form3_opf_rs2F_rdD(src, dst));
8431   ins_pipe(fcvtF2D);
8432 %}
8433 
8434 
8435 // Convert a float to an int in a float register.
8436 // If the float is a NAN, stuff a zero in instead.
8437 instruct convF2I_helper(regF dst, regF src, flagsRegF0 fcc0) %{
8438   effect(DEF dst, USE src, KILL fcc0);
8439   format %{ "FCMPs  fcc0,$src,$src\t! check for NAN\n\t"
8440             "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8441             "FSTOI  $src,$dst\t! convert in delay slot\n\t"
8442             "FITOS  $dst,$dst\t! change NaN/max-int to valid float\n\t"
8443             "FSUBs  $dst,$dst,$dst\t! cleared only if nan\n"
8444       "skip:" %}
8445   ins_encode(form_f2i_helper(src,dst));
8446   ins_pipe(fcvtF2I);
8447 %}
8448 
8449 instruct convF2I_stk(stackSlotI dst, regF src) %{
8450   match(Set dst (ConvF2I src));
8451   ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8452   expand %{
8453     regF tmp;
8454     convF2I_helper(tmp, src);
8455     regF_to_stkI(dst, tmp);
8456   %}
8457 %}
8458 
8459 instruct convF2I_reg(iRegI dst, regF src) %{
8460   predicate(UseVIS >= 3);
8461   match(Set dst (ConvF2I src));
8462   ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8463   expand %{
8464     regF tmp;
8465     convF2I_helper(tmp, src);
8466     MoveF2I_reg_reg(dst, tmp);
8467   %}
8468 %}
8469 
8470 
8471 // Convert a float to a long in a float register.
8472 // If the float is a NAN, stuff a zero in instead.
8473 instruct convF2L_helper(regD dst, regF src, flagsRegF0 fcc0) %{
8474   effect(DEF dst, USE src, KILL fcc0);
8475   format %{ "FCMPs  fcc0,$src,$src\t! check for NAN\n\t"
8476             "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8477             "FSTOX  $src,$dst\t! convert in delay slot\n\t"
8478             "FXTOD  $dst,$dst\t! change NaN/max-long to valid double\n\t"
8479             "FSUBd  $dst,$dst,$dst\t! cleared only if nan\n"
8480       "skip:" %}
8481   ins_encode(form_f2l_helper(src,dst));
8482   ins_pipe(fcvtF2L);
8483 %}
8484 
8485 instruct convF2L_stk(stackSlotL dst, regF src) %{
8486   match(Set dst (ConvF2L src));
8487   ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8488   expand %{
8489     regD tmp;
8490     convF2L_helper(tmp, src);
8491     regD_to_stkL(dst, tmp);
8492   %}
8493 %}
8494 
8495 instruct convF2L_reg(iRegL dst, regF src) %{
8496   predicate(UseVIS >= 3);
8497   match(Set dst (ConvF2L src));
8498   ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8499   expand %{
8500     regD tmp;
8501     convF2L_helper(tmp, src);
8502     MoveD2L_reg_reg(dst, tmp);
8503   %}
8504 %}
8505 
8506 
8507 instruct convI2D_helper(regD dst, regF tmp) %{
8508   effect(USE tmp, DEF dst);
8509   format %{ "FITOD  $tmp,$dst" %}
8510   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8511   ins_encode(form3_opf_rs2F_rdD(tmp, dst));
8512   ins_pipe(fcvtI2D);
8513 %}
8514 
8515 instruct convI2D_stk(stackSlotI src, regD dst) %{
8516   match(Set dst (ConvI2D src));
8517   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8518   expand %{
8519     regF tmp;
8520     stkI_to_regF(tmp, src);
8521     convI2D_helper(dst, tmp);
8522   %}
8523 %}
8524 
8525 instruct convI2D_reg(regD_low dst, iRegI src) %{
8526   predicate(UseVIS >= 3);
8527   match(Set dst (ConvI2D src));
8528   expand %{
8529     regF tmp;
8530     MoveI2F_reg_reg(tmp, src);
8531     convI2D_helper(dst, tmp);
8532   %}
8533 %}
8534 
8535 instruct convI2D_mem(regD_low dst, memory mem) %{
8536   match(Set dst (ConvI2D (LoadI mem)));
8537   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8538   size(8);
8539   format %{ "LDF    $mem,$dst\n\t"
8540             "FITOD  $dst,$dst" %}
8541   opcode(Assembler::ldf_op3, Assembler::fitod_opf);
8542   ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8543   ins_pipe(floadF_mem);
8544 %}
8545 
8546 
8547 instruct convI2F_helper(regF dst, regF tmp) %{
8548   effect(DEF dst, USE tmp);
8549   format %{ "FITOS  $tmp,$dst" %}
8550   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitos_opf);
8551   ins_encode(form3_opf_rs2F_rdF(tmp, dst));
8552   ins_pipe(fcvtI2F);
8553 %}
8554 
8555 instruct convI2F_stk(regF dst, stackSlotI src) %{
8556   match(Set dst (ConvI2F src));
8557   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8558   expand %{
8559     regF tmp;
8560     stkI_to_regF(tmp,src);
8561     convI2F_helper(dst, tmp);
8562   %}
8563 %}
8564 
8565 instruct convI2F_reg(regF dst, iRegI src) %{
8566   predicate(UseVIS >= 3);
8567   match(Set dst (ConvI2F src));
8568   ins_cost(DEFAULT_COST);
8569   expand %{
8570     regF tmp;
8571     MoveI2F_reg_reg(tmp, src);
8572     convI2F_helper(dst, tmp);
8573   %}
8574 %}
8575 
8576 instruct convI2F_mem( regF dst, memory mem ) %{
8577   match(Set dst (ConvI2F (LoadI mem)));
8578   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8579   size(8);
8580   format %{ "LDF    $mem,$dst\n\t"
8581             "FITOS  $dst,$dst" %}
8582   opcode(Assembler::ldf_op3, Assembler::fitos_opf);
8583   ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8584   ins_pipe(floadF_mem);
8585 %}
8586 
8587 
8588 instruct convI2L_reg(iRegL dst, iRegI src) %{
8589   match(Set dst (ConvI2L src));
8590   size(4);
8591   format %{ "SRA    $src,0,$dst\t! int->long" %}
8592   opcode(Assembler::sra_op3, Assembler::arith_op);
8593   ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8594   ins_pipe(ialu_reg_reg);
8595 %}
8596 
8597 // Zero-extend convert int to long
8598 instruct convI2L_reg_zex(iRegL dst, iRegI src, immL_32bits mask ) %{
8599   match(Set dst (AndL (ConvI2L src) mask) );
8600   size(4);
8601   format %{ "SRL    $src,0,$dst\t! zero-extend int to long" %}
8602   opcode(Assembler::srl_op3, Assembler::arith_op);
8603   ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8604   ins_pipe(ialu_reg_reg);
8605 %}
8606 
8607 // Zero-extend long
8608 instruct zerox_long(iRegL dst, iRegL src, immL_32bits mask ) %{
8609   match(Set dst (AndL src mask) );
8610   size(4);
8611   format %{ "SRL    $src,0,$dst\t! zero-extend long" %}
8612   opcode(Assembler::srl_op3, Assembler::arith_op);
8613   ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8614   ins_pipe(ialu_reg_reg);
8615 %}
8616 
8617 
8618 //-----------
8619 // Long to Double conversion using V8 opcodes.
8620 // Still useful because cheetah traps and becomes
8621 // amazingly slow for some common numbers.
8622 
8623 // Magic constant, 0x43300000
8624 instruct loadConI_x43300000(iRegI dst) %{
8625   effect(DEF dst);
8626   size(4);
8627   format %{ "SETHI  HI(0x43300000),$dst\t! 2^52" %}
8628   ins_encode(SetHi22(0x43300000, dst));
8629   ins_pipe(ialu_none);
8630 %}
8631 
8632 // Magic constant, 0x41f00000
8633 instruct loadConI_x41f00000(iRegI dst) %{
8634   effect(DEF dst);
8635   size(4);
8636   format %{ "SETHI  HI(0x41f00000),$dst\t! 2^32" %}
8637   ins_encode(SetHi22(0x41f00000, dst));
8638   ins_pipe(ialu_none);
8639 %}
8640 
8641 // Construct a double from two float halves
8642 instruct regDHi_regDLo_to_regD(regD_low dst, regD_low src1, regD_low src2) %{
8643   effect(DEF dst, USE src1, USE src2);
8644   size(8);
8645   format %{ "FMOVS  $src1.hi,$dst.hi\n\t"
8646             "FMOVS  $src2.lo,$dst.lo" %}
8647   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmovs_opf);
8648   ins_encode(form3_opf_rs2D_hi_rdD_hi(src1, dst), form3_opf_rs2D_lo_rdD_lo(src2, dst));
8649   ins_pipe(faddD_reg_reg);
8650 %}
8651 
8652 // Convert integer in high half of a double register (in the lower half of
8653 // the double register file) to double
8654 instruct convI2D_regDHi_regD(regD dst, regD_low src) %{
8655   effect(DEF dst, USE src);
8656   size(4);
8657   format %{ "FITOD  $src,$dst" %}
8658   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8659   ins_encode(form3_opf_rs2D_rdD(src, dst));
8660   ins_pipe(fcvtLHi2D);
8661 %}
8662 
8663 // Add float double precision
8664 instruct addD_regD_regD(regD dst, regD src1, regD src2) %{
8665   effect(DEF dst, USE src1, USE src2);
8666   size(4);
8667   format %{ "FADDD  $src1,$src2,$dst" %}
8668   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
8669   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8670   ins_pipe(faddD_reg_reg);
8671 %}
8672 
8673 // Sub float double precision
8674 instruct subD_regD_regD(regD dst, regD src1, regD src2) %{
8675   effect(DEF dst, USE src1, USE src2);
8676   size(4);
8677   format %{ "FSUBD  $src1,$src2,$dst" %}
8678   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
8679   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8680   ins_pipe(faddD_reg_reg);
8681 %}
8682 
8683 // Mul float double precision
8684 instruct mulD_regD_regD(regD dst, regD src1, regD src2) %{
8685   effect(DEF dst, USE src1, USE src2);
8686   size(4);
8687   format %{ "FMULD  $src1,$src2,$dst" %}
8688   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
8689   ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8690   ins_pipe(fmulD_reg_reg);
8691 %}
8692 
8693 instruct convL2D_reg_slow_fxtof(regD dst, stackSlotL src) %{
8694   match(Set dst (ConvL2D src));
8695   ins_cost(DEFAULT_COST*8 + MEMORY_REF_COST*6);
8696 
8697   expand %{
8698     regD_low   tmpsrc;
8699     iRegI      ix43300000;
8700     iRegI      ix41f00000;
8701     stackSlotL lx43300000;
8702     stackSlotL lx41f00000;
8703     regD_low   dx43300000;
8704     regD       dx41f00000;
8705     regD       tmp1;
8706     regD_low   tmp2;
8707     regD       tmp3;
8708     regD       tmp4;
8709 
8710     stkL_to_regD(tmpsrc, src);
8711 
8712     loadConI_x43300000(ix43300000);
8713     loadConI_x41f00000(ix41f00000);
8714     regI_to_stkLHi(lx43300000, ix43300000);
8715     regI_to_stkLHi(lx41f00000, ix41f00000);
8716     stkL_to_regD(dx43300000, lx43300000);
8717     stkL_to_regD(dx41f00000, lx41f00000);
8718 
8719     convI2D_regDHi_regD(tmp1, tmpsrc);
8720     regDHi_regDLo_to_regD(tmp2, dx43300000, tmpsrc);
8721     subD_regD_regD(tmp3, tmp2, dx43300000);
8722     mulD_regD_regD(tmp4, tmp1, dx41f00000);
8723     addD_regD_regD(dst, tmp3, tmp4);
8724   %}
8725 %}
8726 
8727 // Long to Double conversion using fast fxtof
8728 instruct convL2D_helper(regD dst, regD tmp) %{
8729   effect(DEF dst, USE tmp);
8730   size(4);
8731   format %{ "FXTOD  $tmp,$dst" %}
8732   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtod_opf);
8733   ins_encode(form3_opf_rs2D_rdD(tmp, dst));
8734   ins_pipe(fcvtL2D);
8735 %}
8736 
8737 instruct convL2D_stk_fast_fxtof(regD dst, stackSlotL src) %{
8738   predicate(VM_Version::has_fast_fxtof());
8739   match(Set dst (ConvL2D src));
8740   ins_cost(DEFAULT_COST + 3 * MEMORY_REF_COST);
8741   expand %{
8742     regD tmp;
8743     stkL_to_regD(tmp, src);
8744     convL2D_helper(dst, tmp);
8745   %}
8746 %}
8747 
8748 instruct convL2D_reg(regD dst, iRegL src) %{
8749   predicate(UseVIS >= 3);
8750   match(Set dst (ConvL2D src));
8751   expand %{
8752     regD tmp;
8753     MoveL2D_reg_reg(tmp, src);
8754     convL2D_helper(dst, tmp);
8755   %}
8756 %}
8757 
8758 // Long to Float conversion using fast fxtof
8759 instruct convL2F_helper(regF dst, regD tmp) %{
8760   effect(DEF dst, USE tmp);
8761   size(4);
8762   format %{ "FXTOS  $tmp,$dst" %}
8763   opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtos_opf);
8764   ins_encode(form3_opf_rs2D_rdF(tmp, dst));
8765   ins_pipe(fcvtL2F);
8766 %}
8767 
8768 instruct convL2F_stk_fast_fxtof(regF dst, stackSlotL src) %{
8769   match(Set dst (ConvL2F src));
8770   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8771   expand %{
8772     regD tmp;
8773     stkL_to_regD(tmp, src);
8774     convL2F_helper(dst, tmp);
8775   %}
8776 %}
8777 
8778 instruct convL2F_reg(regF dst, iRegL src) %{
8779   predicate(UseVIS >= 3);
8780   match(Set dst (ConvL2F src));
8781   ins_cost(DEFAULT_COST);
8782   expand %{
8783     regD tmp;
8784     MoveL2D_reg_reg(tmp, src);
8785     convL2F_helper(dst, tmp);
8786   %}
8787 %}
8788 
8789 //-----------
8790 
8791 instruct convL2I_reg(iRegI dst, iRegL src) %{
8792   match(Set dst (ConvL2I src));
8793 #ifndef _LP64
8794   format %{ "MOV    $src.lo,$dst\t! long->int" %}
8795   ins_encode( form3_g0_rs2_rd_move_lo2( src, dst ) );
8796   ins_pipe(ialu_move_reg_I_to_L);
8797 #else
8798   size(4);
8799   format %{ "SRA    $src,R_G0,$dst\t! long->int" %}
8800   ins_encode( form3_rs1_rd_signextend_lo1( src, dst ) );
8801   ins_pipe(ialu_reg);
8802 #endif
8803 %}
8804 
8805 // Register Shift Right Immediate
8806 instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt) %{
8807   match(Set dst (ConvL2I (RShiftL src cnt)));
8808 
8809   size(4);
8810   format %{ "SRAX   $src,$cnt,$dst" %}
8811   opcode(Assembler::srax_op3, Assembler::arith_op);
8812   ins_encode( form3_sd_rs1_imm6_rd( src, cnt, dst ) );
8813   ins_pipe(ialu_reg_imm);
8814 %}
8815 
8816 //----------Control Flow Instructions------------------------------------------
8817 // Compare Instructions
8818 // Compare Integers
8819 instruct compI_iReg(flagsReg icc, iRegI op1, iRegI op2) %{
8820   match(Set icc (CmpI op1 op2));
8821   effect( DEF icc, USE op1, USE op2 );
8822 
8823   size(4);
8824   format %{ "CMP    $op1,$op2" %}
8825   opcode(Assembler::subcc_op3, Assembler::arith_op);
8826   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8827   ins_pipe(ialu_cconly_reg_reg);
8828 %}
8829 
8830 instruct compU_iReg(flagsRegU icc, iRegI op1, iRegI op2) %{
8831   match(Set icc (CmpU op1 op2));
8832 
8833   size(4);
8834   format %{ "CMP    $op1,$op2\t! unsigned" %}
8835   opcode(Assembler::subcc_op3, Assembler::arith_op);
8836   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8837   ins_pipe(ialu_cconly_reg_reg);
8838 %}
8839 
8840 instruct compI_iReg_imm13(flagsReg icc, iRegI op1, immI13 op2) %{
8841   match(Set icc (CmpI op1 op2));
8842   effect( DEF icc, USE op1 );
8843 
8844   size(4);
8845   format %{ "CMP    $op1,$op2" %}
8846   opcode(Assembler::subcc_op3, Assembler::arith_op);
8847   ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8848   ins_pipe(ialu_cconly_reg_imm);
8849 %}
8850 
8851 instruct testI_reg_reg( flagsReg icc, iRegI op1, iRegI op2, immI0 zero ) %{
8852   match(Set icc (CmpI (AndI op1 op2) zero));
8853 
8854   size(4);
8855   format %{ "BTST   $op2,$op1" %}
8856   opcode(Assembler::andcc_op3, Assembler::arith_op);
8857   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8858   ins_pipe(ialu_cconly_reg_reg_zero);
8859 %}
8860 
8861 instruct testI_reg_imm( flagsReg icc, iRegI op1, immI13 op2, immI0 zero ) %{
8862   match(Set icc (CmpI (AndI op1 op2) zero));
8863 
8864   size(4);
8865   format %{ "BTST   $op2,$op1" %}
8866   opcode(Assembler::andcc_op3, Assembler::arith_op);
8867   ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8868   ins_pipe(ialu_cconly_reg_imm_zero);
8869 %}
8870 
8871 instruct compL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2 ) %{
8872   match(Set xcc (CmpL op1 op2));
8873   effect( DEF xcc, USE op1, USE op2 );
8874 
8875   size(4);
8876   format %{ "CMP    $op1,$op2\t\t! long" %}
8877   opcode(Assembler::subcc_op3, Assembler::arith_op);
8878   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8879   ins_pipe(ialu_cconly_reg_reg);
8880 %}
8881 
8882 instruct compL_reg_con(flagsRegL xcc, iRegL op1, immL13 con) %{
8883   match(Set xcc (CmpL op1 con));
8884   effect( DEF xcc, USE op1, USE con );
8885 
8886   size(4);
8887   format %{ "CMP    $op1,$con\t\t! long" %}
8888   opcode(Assembler::subcc_op3, Assembler::arith_op);
8889   ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8890   ins_pipe(ialu_cconly_reg_reg);
8891 %}
8892 
8893 instruct testL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2, immL0 zero) %{
8894   match(Set xcc (CmpL (AndL op1 op2) zero));
8895   effect( DEF xcc, USE op1, USE op2 );
8896 
8897   size(4);
8898   format %{ "BTST   $op1,$op2\t\t! long" %}
8899   opcode(Assembler::andcc_op3, Assembler::arith_op);
8900   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8901   ins_pipe(ialu_cconly_reg_reg);
8902 %}
8903 
8904 // useful for checking the alignment of a pointer:
8905 instruct testL_reg_con(flagsRegL xcc, iRegL op1, immL13 con, immL0 zero) %{
8906   match(Set xcc (CmpL (AndL op1 con) zero));
8907   effect( DEF xcc, USE op1, USE con );
8908 
8909   size(4);
8910   format %{ "BTST   $op1,$con\t\t! long" %}
8911   opcode(Assembler::andcc_op3, Assembler::arith_op);
8912   ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8913   ins_pipe(ialu_cconly_reg_reg);
8914 %}
8915 
8916 instruct compU_iReg_imm13(flagsRegU icc, iRegI op1, immU13 op2 ) %{
8917   match(Set icc (CmpU op1 op2));
8918 
8919   size(4);
8920   format %{ "CMP    $op1,$op2\t! unsigned" %}
8921   opcode(Assembler::subcc_op3, Assembler::arith_op);
8922   ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8923   ins_pipe(ialu_cconly_reg_imm);
8924 %}
8925 
8926 // Compare Pointers
8927 instruct compP_iRegP(flagsRegP pcc, iRegP op1, iRegP op2 ) %{
8928   match(Set pcc (CmpP op1 op2));
8929 
8930   size(4);
8931   format %{ "CMP    $op1,$op2\t! ptr" %}
8932   opcode(Assembler::subcc_op3, Assembler::arith_op);
8933   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8934   ins_pipe(ialu_cconly_reg_reg);
8935 %}
8936 
8937 instruct compP_iRegP_imm13(flagsRegP pcc, iRegP op1, immP13 op2 ) %{
8938   match(Set pcc (CmpP op1 op2));
8939 
8940   size(4);
8941   format %{ "CMP    $op1,$op2\t! ptr" %}
8942   opcode(Assembler::subcc_op3, Assembler::arith_op);
8943   ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8944   ins_pipe(ialu_cconly_reg_imm);
8945 %}
8946 
8947 // Compare Narrow oops
8948 instruct compN_iRegN(flagsReg icc, iRegN op1, iRegN op2 ) %{
8949   match(Set icc (CmpN op1 op2));
8950 
8951   size(4);
8952   format %{ "CMP    $op1,$op2\t! compressed ptr" %}
8953   opcode(Assembler::subcc_op3, Assembler::arith_op);
8954   ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8955   ins_pipe(ialu_cconly_reg_reg);
8956 %}
8957 
8958 instruct compN_iRegN_immN0(flagsReg icc, iRegN op1, immN0 op2 ) %{
8959   match(Set icc (CmpN op1 op2));
8960 
8961   size(4);
8962   format %{ "CMP    $op1,$op2\t! compressed ptr" %}
8963   opcode(Assembler::subcc_op3, Assembler::arith_op);
8964   ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8965   ins_pipe(ialu_cconly_reg_imm);
8966 %}
8967 
8968 //----------Max and Min--------------------------------------------------------
8969 // Min Instructions
8970 // Conditional move for min
8971 instruct cmovI_reg_lt( iRegI op2, iRegI op1, flagsReg icc ) %{
8972   effect( USE_DEF op2, USE op1, USE icc );
8973 
8974   size(4);
8975   format %{ "MOVlt  icc,$op1,$op2\t! min" %}
8976   opcode(Assembler::less);
8977   ins_encode( enc_cmov_reg_minmax(op2,op1) );
8978   ins_pipe(ialu_reg_flags);
8979 %}
8980 
8981 // Min Register with Register.
8982 instruct minI_eReg(iRegI op1, iRegI op2) %{
8983   match(Set op2 (MinI op1 op2));
8984   ins_cost(DEFAULT_COST*2);
8985   expand %{
8986     flagsReg icc;
8987     compI_iReg(icc,op1,op2);
8988     cmovI_reg_lt(op2,op1,icc);
8989   %}
8990 %}
8991 
8992 // Max Instructions
8993 // Conditional move for max
8994 instruct cmovI_reg_gt( iRegI op2, iRegI op1, flagsReg icc ) %{
8995   effect( USE_DEF op2, USE op1, USE icc );
8996   format %{ "MOVgt  icc,$op1,$op2\t! max" %}
8997   opcode(Assembler::greater);
8998   ins_encode( enc_cmov_reg_minmax(op2,op1) );
8999   ins_pipe(ialu_reg_flags);
9000 %}
9001 
9002 // Max Register with Register
9003 instruct maxI_eReg(iRegI op1, iRegI op2) %{
9004   match(Set op2 (MaxI op1 op2));
9005   ins_cost(DEFAULT_COST*2);
9006   expand %{
9007     flagsReg icc;
9008     compI_iReg(icc,op1,op2);
9009     cmovI_reg_gt(op2,op1,icc);
9010   %}
9011 %}
9012 
9013 
9014 //----------Float Compares----------------------------------------------------
9015 // Compare floating, generate condition code
9016 instruct cmpF_cc(flagsRegF fcc, regF src1, regF src2) %{
9017   match(Set fcc (CmpF src1 src2));
9018 
9019   size(4);
9020   format %{ "FCMPs  $fcc,$src1,$src2" %}
9021   opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmps_opf);
9022   ins_encode( form3_opf_rs1F_rs2F_fcc( src1, src2, fcc ) );
9023   ins_pipe(faddF_fcc_reg_reg_zero);
9024 %}
9025 
9026 instruct cmpD_cc(flagsRegF fcc, regD src1, regD src2) %{
9027   match(Set fcc (CmpD src1 src2));
9028 
9029   size(4);
9030   format %{ "FCMPd  $fcc,$src1,$src2" %}
9031   opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmpd_opf);
9032   ins_encode( form3_opf_rs1D_rs2D_fcc( src1, src2, fcc ) );
9033   ins_pipe(faddD_fcc_reg_reg_zero);
9034 %}
9035 
9036 
9037 // Compare floating, generate -1,0,1
9038 instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsRegF0 fcc0) %{
9039   match(Set dst (CmpF3 src1 src2));
9040   effect(KILL fcc0);
9041   ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9042   format %{ "fcmpl  $dst,$src1,$src2" %}
9043   // Primary = float
9044   opcode( true );
9045   ins_encode( floating_cmp( dst, src1, src2 ) );
9046   ins_pipe( floating_cmp );
9047 %}
9048 
9049 instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsRegF0 fcc0) %{
9050   match(Set dst (CmpD3 src1 src2));
9051   effect(KILL fcc0);
9052   ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9053   format %{ "dcmpl  $dst,$src1,$src2" %}
9054   // Primary = double (not float)
9055   opcode( false );
9056   ins_encode( floating_cmp( dst, src1, src2 ) );
9057   ins_pipe( floating_cmp );
9058 %}
9059 
9060 //----------Branches---------------------------------------------------------
9061 // Jump
9062 // (compare 'operand indIndex' and 'instruct addP_reg_reg' above)
9063 instruct jumpXtnd(iRegX switch_val, o7RegI table) %{
9064   match(Jump switch_val);
9065   effect(TEMP table);
9066 
9067   ins_cost(350);
9068 
9069   format %{  "ADD    $constanttablebase, $constantoffset, O7\n\t"
9070              "LD     [O7 + $switch_val], O7\n\t"
9071              "JUMP   O7" %}
9072   ins_encode %{
9073     // Calculate table address into a register.
9074     Register table_reg;
9075     Register label_reg = O7;
9076     // If we are calculating the size of this instruction don't trust
9077     // zero offsets because they might change when
9078     // MachConstantBaseNode decides to optimize the constant table
9079     // base.
9080     if ((constant_offset() == 0) && !Compile::current()->in_scratch_emit_size()) {
9081       table_reg = $constanttablebase;
9082     } else {
9083       table_reg = O7;
9084       RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset, O7);
9085       __ add($constanttablebase, con_offset, table_reg);
9086     }
9087 
9088     // Jump to base address + switch value
9089     __ ld_ptr(table_reg, $switch_val$$Register, label_reg);
9090     __ jmp(label_reg, G0);
9091     __ delayed()->nop();
9092   %}
9093   ins_pipe(ialu_reg_reg);
9094 %}
9095 
9096 // Direct Branch.  Use V8 version with longer range.
9097 instruct branch(label labl) %{
9098   match(Goto);
9099   effect(USE labl);
9100 
9101   size(8);
9102   ins_cost(BRANCH_COST);
9103   format %{ "BA     $labl" %}
9104   ins_encode %{
9105     Label* L = $labl$$label;
9106     __ ba(*L);
9107     __ delayed()->nop();
9108   %}
9109   ins_pipe(br);
9110 %}
9111 
9112 // Direct Branch, short with no delay slot
9113 instruct branch_short(label labl) %{
9114   match(Goto);
9115   predicate(UseCBCond);
9116   effect(USE labl);
9117 
9118   size(4);
9119   ins_cost(BRANCH_COST);
9120   format %{ "BA     $labl\t! short branch" %}
9121   ins_encode %{ 
9122     Label* L = $labl$$label;
9123     assert(__ use_cbcond(*L), "back to back cbcond");
9124     __ ba_short(*L);
9125   %}
9126   ins_short_branch(1);
9127   ins_avoid_back_to_back(1);
9128   ins_pipe(cbcond_reg_imm);
9129 %}
9130 
9131 // Conditional Direct Branch
9132 instruct branchCon(cmpOp cmp, flagsReg icc, label labl) %{
9133   match(If cmp icc);
9134   effect(USE labl);
9135 
9136   size(8);
9137   ins_cost(BRANCH_COST);
9138   format %{ "BP$cmp   $icc,$labl" %}
9139   // Prim = bits 24-22, Secnd = bits 31-30
9140   ins_encode( enc_bp( labl, cmp, icc ) );
9141   ins_pipe(br_cc);
9142 %}
9143 
9144 instruct branchConU(cmpOpU cmp, flagsRegU icc, label labl) %{
9145   match(If cmp icc);
9146   effect(USE labl);
9147 
9148   ins_cost(BRANCH_COST);
9149   format %{ "BP$cmp  $icc,$labl" %}
9150   // Prim = bits 24-22, Secnd = bits 31-30
9151   ins_encode( enc_bp( labl, cmp, icc ) );
9152   ins_pipe(br_cc);
9153 %}
9154 
9155 instruct branchConP(cmpOpP cmp, flagsRegP pcc, label labl) %{
9156   match(If cmp pcc);
9157   effect(USE labl);
9158 
9159   size(8);
9160   ins_cost(BRANCH_COST);
9161   format %{ "BP$cmp  $pcc,$labl" %}
9162   ins_encode %{
9163     Label* L = $labl$$label;
9164     Assembler::Predict predict_taken =
9165       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9166 
9167     __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9168     __ delayed()->nop();
9169   %}
9170   ins_pipe(br_cc);
9171 %}
9172 
9173 instruct branchConF(cmpOpF cmp, flagsRegF fcc, label labl) %{
9174   match(If cmp fcc);
9175   effect(USE labl);
9176 
9177   size(8);
9178   ins_cost(BRANCH_COST);
9179   format %{ "FBP$cmp $fcc,$labl" %}
9180   ins_encode %{
9181     Label* L = $labl$$label;
9182     Assembler::Predict predict_taken =
9183       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9184 
9185     __ fbp( (Assembler::Condition)($cmp$$cmpcode), false, (Assembler::CC)($fcc$$reg), predict_taken, *L);
9186     __ delayed()->nop();
9187   %}
9188   ins_pipe(br_fcc);
9189 %}
9190 
9191 instruct branchLoopEnd(cmpOp cmp, flagsReg icc, label labl) %{
9192   match(CountedLoopEnd cmp icc);
9193   effect(USE labl);
9194 
9195   size(8);
9196   ins_cost(BRANCH_COST);
9197   format %{ "BP$cmp   $icc,$labl\t! Loop end" %}
9198   // Prim = bits 24-22, Secnd = bits 31-30
9199   ins_encode( enc_bp( labl, cmp, icc ) );
9200   ins_pipe(br_cc);
9201 %}
9202 
9203 instruct branchLoopEndU(cmpOpU cmp, flagsRegU icc, label labl) %{
9204   match(CountedLoopEnd cmp icc);
9205   effect(USE labl);
9206 
9207   size(8);
9208   ins_cost(BRANCH_COST);
9209   format %{ "BP$cmp  $icc,$labl\t! Loop end" %}
9210   // Prim = bits 24-22, Secnd = bits 31-30
9211   ins_encode( enc_bp( labl, cmp, icc ) );
9212   ins_pipe(br_cc);
9213 %}
9214 
9215 // Compare and branch instructions
9216 instruct cmpI_reg_branch(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9217   match(If cmp (CmpI op1 op2));
9218   effect(USE labl, KILL icc);
9219 
9220   size(12);
9221   ins_cost(BRANCH_COST);
9222   format %{ "CMP    $op1,$op2\t! int\n\t"
9223             "BP$cmp   $labl" %}
9224   ins_encode %{
9225     Label* L = $labl$$label;
9226     Assembler::Predict predict_taken =
9227       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9228     __ cmp($op1$$Register, $op2$$Register);
9229     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9230     __ delayed()->nop();
9231   %}
9232   ins_pipe(cmp_br_reg_reg);
9233 %}
9234 
9235 instruct cmpI_imm_branch(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9236   match(If cmp (CmpI op1 op2));
9237   effect(USE labl, KILL icc);
9238 
9239   size(12);
9240   ins_cost(BRANCH_COST);
9241   format %{ "CMP    $op1,$op2\t! int\n\t"
9242             "BP$cmp   $labl" %}
9243   ins_encode %{
9244     Label* L = $labl$$label;
9245     Assembler::Predict predict_taken =
9246       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9247     __ cmp($op1$$Register, $op2$$constant);
9248     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9249     __ delayed()->nop();
9250   %}
9251   ins_pipe(cmp_br_reg_imm);
9252 %}
9253 
9254 instruct cmpU_reg_branch(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9255   match(If cmp (CmpU op1 op2));
9256   effect(USE labl, KILL icc);
9257 
9258   size(12);
9259   ins_cost(BRANCH_COST);
9260   format %{ "CMP    $op1,$op2\t! unsigned\n\t"
9261             "BP$cmp  $labl" %}
9262   ins_encode %{
9263     Label* L = $labl$$label;
9264     Assembler::Predict predict_taken =
9265       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9266     __ cmp($op1$$Register, $op2$$Register);
9267     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9268     __ delayed()->nop();
9269   %}
9270   ins_pipe(cmp_br_reg_reg);
9271 %}
9272 
9273 instruct cmpU_imm_branch(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9274   match(If cmp (CmpU op1 op2));
9275   effect(USE labl, KILL icc);
9276 
9277   size(12);
9278   ins_cost(BRANCH_COST);
9279   format %{ "CMP    $op1,$op2\t! unsigned\n\t"
9280             "BP$cmp  $labl" %}
9281   ins_encode %{
9282     Label* L = $labl$$label;
9283     Assembler::Predict predict_taken =
9284       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9285     __ cmp($op1$$Register, $op2$$constant);
9286     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9287     __ delayed()->nop();
9288   %}
9289   ins_pipe(cmp_br_reg_imm);
9290 %}
9291 
9292 instruct cmpL_reg_branch(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9293   match(If cmp (CmpL op1 op2));
9294   effect(USE labl, KILL xcc);
9295 
9296   size(12);
9297   ins_cost(BRANCH_COST);
9298   format %{ "CMP    $op1,$op2\t! long\n\t"
9299             "BP$cmp   $labl" %}
9300   ins_encode %{
9301     Label* L = $labl$$label;
9302     Assembler::Predict predict_taken =
9303       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9304     __ cmp($op1$$Register, $op2$$Register);
9305     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9306     __ delayed()->nop();
9307   %}
9308   ins_pipe(cmp_br_reg_reg);
9309 %}
9310 
9311 instruct cmpL_imm_branch(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9312   match(If cmp (CmpL op1 op2));
9313   effect(USE labl, KILL xcc);
9314 
9315   size(12);
9316   ins_cost(BRANCH_COST);
9317   format %{ "CMP    $op1,$op2\t! long\n\t"
9318             "BP$cmp   $labl" %}
9319   ins_encode %{
9320     Label* L = $labl$$label;
9321     Assembler::Predict predict_taken =
9322       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9323     __ cmp($op1$$Register, $op2$$constant);
9324     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9325     __ delayed()->nop();
9326   %}
9327   ins_pipe(cmp_br_reg_imm);
9328 %}
9329 
9330 // Compare Pointers and branch
9331 instruct cmpP_reg_branch(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9332   match(If cmp (CmpP op1 op2));
9333   effect(USE labl, KILL pcc);
9334 
9335   size(12);
9336   ins_cost(BRANCH_COST);
9337   format %{ "CMP    $op1,$op2\t! ptr\n\t"
9338             "B$cmp   $labl" %}
9339   ins_encode %{
9340     Label* L = $labl$$label;
9341     Assembler::Predict predict_taken =
9342       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9343     __ cmp($op1$$Register, $op2$$Register);
9344     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9345     __ delayed()->nop();
9346   %}
9347   ins_pipe(cmp_br_reg_reg);
9348 %}
9349 
9350 instruct cmpP_null_branch(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9351   match(If cmp (CmpP op1 null));
9352   effect(USE labl, KILL pcc);
9353 
9354   size(12);
9355   ins_cost(BRANCH_COST);
9356   format %{ "CMP    $op1,0\t! ptr\n\t"
9357             "B$cmp   $labl" %}
9358   ins_encode %{
9359     Label* L = $labl$$label;
9360     Assembler::Predict predict_taken =
9361       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9362     __ cmp($op1$$Register, G0);
9363     // bpr() is not used here since it has shorter distance.
9364     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9365     __ delayed()->nop();
9366   %}
9367   ins_pipe(cmp_br_reg_reg);
9368 %}
9369 
9370 instruct cmpN_reg_branch(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9371   match(If cmp (CmpN op1 op2));
9372   effect(USE labl, KILL icc);
9373 
9374   size(12);
9375   ins_cost(BRANCH_COST);
9376   format %{ "CMP    $op1,$op2\t! compressed ptr\n\t"
9377             "BP$cmp   $labl" %}
9378   ins_encode %{
9379     Label* L = $labl$$label;
9380     Assembler::Predict predict_taken =
9381       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9382     __ cmp($op1$$Register, $op2$$Register);
9383     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9384     __ delayed()->nop();
9385   %}
9386   ins_pipe(cmp_br_reg_reg);
9387 %}
9388 
9389 instruct cmpN_null_branch(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9390   match(If cmp (CmpN op1 null));
9391   effect(USE labl, KILL icc);
9392 
9393   size(12);
9394   ins_cost(BRANCH_COST);
9395   format %{ "CMP    $op1,0\t! compressed ptr\n\t"
9396             "BP$cmp   $labl" %}
9397   ins_encode %{
9398     Label* L = $labl$$label;
9399     Assembler::Predict predict_taken =
9400       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9401     __ cmp($op1$$Register, G0);
9402     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9403     __ delayed()->nop();
9404   %}
9405   ins_pipe(cmp_br_reg_reg);
9406 %}
9407 
9408 // Loop back branch
9409 instruct cmpI_reg_branchLoopEnd(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9410   match(CountedLoopEnd cmp (CmpI op1 op2));
9411   effect(USE labl, KILL icc);
9412 
9413   size(12);
9414   ins_cost(BRANCH_COST);
9415   format %{ "CMP    $op1,$op2\t! int\n\t"
9416             "BP$cmp   $labl\t! Loop end" %}
9417   ins_encode %{
9418     Label* L = $labl$$label;
9419     Assembler::Predict predict_taken =
9420       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9421     __ cmp($op1$$Register, $op2$$Register);
9422     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9423     __ delayed()->nop();
9424   %}
9425   ins_pipe(cmp_br_reg_reg);
9426 %}
9427 
9428 instruct cmpI_imm_branchLoopEnd(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9429   match(CountedLoopEnd cmp (CmpI op1 op2));
9430   effect(USE labl, KILL icc);
9431 
9432   size(12);
9433   ins_cost(BRANCH_COST);
9434   format %{ "CMP    $op1,$op2\t! int\n\t"
9435             "BP$cmp   $labl\t! Loop end" %}
9436   ins_encode %{
9437     Label* L = $labl$$label;
9438     Assembler::Predict predict_taken =
9439       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9440     __ cmp($op1$$Register, $op2$$constant);
9441     __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9442     __ delayed()->nop();
9443   %}
9444   ins_pipe(cmp_br_reg_imm);
9445 %}
9446 
9447 // Short compare and branch instructions
9448 instruct cmpI_reg_branch_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9449   match(If cmp (CmpI op1 op2));
9450   predicate(UseCBCond);
9451   effect(USE labl, KILL icc);
9452 
9453   size(4);
9454   ins_cost(BRANCH_COST);
9455   format %{ "CWB$cmp  $op1,$op2,$labl\t! int" %}
9456   ins_encode %{
9457     Label* L = $labl$$label;
9458     assert(__ use_cbcond(*L), "back to back cbcond");
9459     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9460   %}
9461   ins_short_branch(1);
9462   ins_avoid_back_to_back(1);
9463   ins_pipe(cbcond_reg_reg);
9464 %}
9465 
9466 instruct cmpI_imm_branch_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9467   match(If cmp (CmpI op1 op2));
9468   predicate(UseCBCond);
9469   effect(USE labl, KILL icc);
9470 
9471   size(4);
9472   ins_cost(BRANCH_COST);
9473   format %{ "CWB$cmp  $op1,$op2,$labl\t! int" %}
9474   ins_encode %{
9475     Label* L = $labl$$label;
9476     assert(__ use_cbcond(*L), "back to back cbcond");
9477     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9478   %}
9479   ins_short_branch(1);
9480   ins_avoid_back_to_back(1);
9481   ins_pipe(cbcond_reg_imm);
9482 %}
9483 
9484 instruct cmpU_reg_branch_short(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9485   match(If cmp (CmpU op1 op2));
9486   predicate(UseCBCond);
9487   effect(USE labl, KILL icc);
9488 
9489   size(4);
9490   ins_cost(BRANCH_COST);
9491   format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9492   ins_encode %{
9493     Label* L = $labl$$label;
9494     assert(__ use_cbcond(*L), "back to back cbcond");
9495     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9496   %}
9497   ins_short_branch(1);
9498   ins_avoid_back_to_back(1);
9499   ins_pipe(cbcond_reg_reg);
9500 %}
9501 
9502 instruct cmpU_imm_branch_short(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9503   match(If cmp (CmpU op1 op2));
9504   predicate(UseCBCond);
9505   effect(USE labl, KILL icc);
9506 
9507   size(4);
9508   ins_cost(BRANCH_COST);
9509   format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9510   ins_encode %{
9511     Label* L = $labl$$label;
9512     assert(__ use_cbcond(*L), "back to back cbcond");
9513     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9514   %}
9515   ins_short_branch(1);
9516   ins_avoid_back_to_back(1);
9517   ins_pipe(cbcond_reg_imm);
9518 %}
9519 
9520 instruct cmpL_reg_branch_short(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9521   match(If cmp (CmpL op1 op2));
9522   predicate(UseCBCond);
9523   effect(USE labl, KILL xcc);
9524 
9525   size(4);
9526   ins_cost(BRANCH_COST);
9527   format %{ "CXB$cmp  $op1,$op2,$labl\t! long" %}
9528   ins_encode %{
9529     Label* L = $labl$$label;
9530     assert(__ use_cbcond(*L), "back to back cbcond");
9531     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$Register, *L);
9532   %}
9533   ins_short_branch(1);
9534   ins_avoid_back_to_back(1);
9535   ins_pipe(cbcond_reg_reg);
9536 %}
9537 
9538 instruct cmpL_imm_branch_short(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9539   match(If cmp (CmpL op1 op2));
9540   predicate(UseCBCond);
9541   effect(USE labl, KILL xcc);
9542 
9543   size(4);
9544   ins_cost(BRANCH_COST);
9545   format %{ "CXB$cmp  $op1,$op2,$labl\t! long" %}
9546   ins_encode %{
9547     Label* L = $labl$$label;
9548     assert(__ use_cbcond(*L), "back to back cbcond");
9549     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$constant, *L);
9550   %}
9551   ins_short_branch(1);
9552   ins_avoid_back_to_back(1);
9553   ins_pipe(cbcond_reg_imm);
9554 %}
9555 
9556 // Compare Pointers and branch
9557 instruct cmpP_reg_branch_short(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9558   match(If cmp (CmpP op1 op2));
9559   predicate(UseCBCond);
9560   effect(USE labl, KILL pcc);
9561 
9562   size(4);
9563   ins_cost(BRANCH_COST);
9564 #ifdef _LP64
9565   format %{ "CXB$cmp $op1,$op2,$labl\t! ptr" %}
9566 #else
9567   format %{ "CWB$cmp $op1,$op2,$labl\t! ptr" %}
9568 #endif
9569   ins_encode %{
9570     Label* L = $labl$$label;
9571     assert(__ use_cbcond(*L), "back to back cbcond");
9572     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, $op2$$Register, *L);
9573   %}
9574   ins_short_branch(1);
9575   ins_avoid_back_to_back(1);
9576   ins_pipe(cbcond_reg_reg);
9577 %}
9578 
9579 instruct cmpP_null_branch_short(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9580   match(If cmp (CmpP op1 null));
9581   predicate(UseCBCond);
9582   effect(USE labl, KILL pcc);
9583 
9584   size(4);
9585   ins_cost(BRANCH_COST);
9586 #ifdef _LP64
9587   format %{ "CXB$cmp $op1,0,$labl\t! ptr" %}
9588 #else
9589   format %{ "CWB$cmp $op1,0,$labl\t! ptr" %}
9590 #endif
9591   ins_encode %{
9592     Label* L = $labl$$label;
9593     assert(__ use_cbcond(*L), "back to back cbcond");
9594     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, G0, *L);
9595   %}
9596   ins_short_branch(1);
9597   ins_avoid_back_to_back(1);
9598   ins_pipe(cbcond_reg_reg);
9599 %}
9600 
9601 instruct cmpN_reg_branch_short(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9602   match(If cmp (CmpN op1 op2));
9603   predicate(UseCBCond);
9604   effect(USE labl, KILL icc);
9605 
9606   size(4);
9607   ins_cost(BRANCH_COST);
9608   format %{ "CWB$cmp  $op1,op2,$labl\t! compressed ptr" %}
9609   ins_encode %{
9610     Label* L = $labl$$label;
9611     assert(__ use_cbcond(*L), "back to back cbcond");
9612     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9613   %}
9614   ins_short_branch(1);
9615   ins_avoid_back_to_back(1);
9616   ins_pipe(cbcond_reg_reg);
9617 %}
9618 
9619 instruct cmpN_null_branch_short(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9620   match(If cmp (CmpN op1 null));
9621   predicate(UseCBCond);
9622   effect(USE labl, KILL icc);
9623 
9624   size(4);
9625   ins_cost(BRANCH_COST);
9626   format %{ "CWB$cmp  $op1,0,$labl\t! compressed ptr" %}
9627   ins_encode %{
9628     Label* L = $labl$$label;
9629     assert(__ use_cbcond(*L), "back to back cbcond");
9630     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, G0, *L);
9631   %}
9632   ins_short_branch(1);
9633   ins_avoid_back_to_back(1);
9634   ins_pipe(cbcond_reg_reg);
9635 %}
9636 
9637 // Loop back branch
9638 instruct cmpI_reg_branchLoopEnd_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9639   match(CountedLoopEnd cmp (CmpI op1 op2));
9640   predicate(UseCBCond);
9641   effect(USE labl, KILL icc);
9642 
9643   size(4);
9644   ins_cost(BRANCH_COST);
9645   format %{ "CWB$cmp  $op1,$op2,$labl\t! Loop end" %}
9646   ins_encode %{
9647     Label* L = $labl$$label;
9648     assert(__ use_cbcond(*L), "back to back cbcond");
9649     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9650   %}
9651   ins_short_branch(1);
9652   ins_avoid_back_to_back(1);
9653   ins_pipe(cbcond_reg_reg);
9654 %}
9655 
9656 instruct cmpI_imm_branchLoopEnd_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9657   match(CountedLoopEnd cmp (CmpI op1 op2));
9658   predicate(UseCBCond);
9659   effect(USE labl, KILL icc);
9660 
9661   size(4);
9662   ins_cost(BRANCH_COST);
9663   format %{ "CWB$cmp  $op1,$op2,$labl\t! Loop end" %}
9664   ins_encode %{
9665     Label* L = $labl$$label;
9666     assert(__ use_cbcond(*L), "back to back cbcond");
9667     __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9668   %}
9669   ins_short_branch(1);
9670   ins_avoid_back_to_back(1);
9671   ins_pipe(cbcond_reg_imm);
9672 %}
9673 
9674 // Branch-on-register tests all 64 bits.  We assume that values
9675 // in 64-bit registers always remains zero or sign extended
9676 // unless our code munges the high bits.  Interrupts can chop
9677 // the high order bits to zero or sign at any time.
9678 instruct branchCon_regI(cmpOp_reg cmp, iRegI op1, immI0 zero, label labl) %{
9679   match(If cmp (CmpI op1 zero));
9680   predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9681   effect(USE labl);
9682 
9683   size(8);
9684   ins_cost(BRANCH_COST);
9685   format %{ "BR$cmp   $op1,$labl" %}
9686   ins_encode( enc_bpr( labl, cmp, op1 ) );
9687   ins_pipe(br_reg);
9688 %}
9689 
9690 instruct branchCon_regP(cmpOp_reg cmp, iRegP op1, immP0 null, label labl) %{
9691   match(If cmp (CmpP op1 null));
9692   predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9693   effect(USE labl);
9694 
9695   size(8);
9696   ins_cost(BRANCH_COST);
9697   format %{ "BR$cmp   $op1,$labl" %}
9698   ins_encode( enc_bpr( labl, cmp, op1 ) );
9699   ins_pipe(br_reg);
9700 %}
9701 
9702 instruct branchCon_regL(cmpOp_reg cmp, iRegL op1, immL0 zero, label labl) %{
9703   match(If cmp (CmpL op1 zero));
9704   predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9705   effect(USE labl);
9706 
9707   size(8);
9708   ins_cost(BRANCH_COST);
9709   format %{ "BR$cmp   $op1,$labl" %}
9710   ins_encode( enc_bpr( labl, cmp, op1 ) );
9711   ins_pipe(br_reg);
9712 %}
9713 
9714 
9715 // ============================================================================
9716 // Long Compare
9717 //
9718 // Currently we hold longs in 2 registers.  Comparing such values efficiently
9719 // is tricky.  The flavor of compare used depends on whether we are testing
9720 // for LT, LE, or EQ.  For a simple LT test we can check just the sign bit.
9721 // The GE test is the negated LT test.  The LE test can be had by commuting
9722 // the operands (yielding a GE test) and then negating; negate again for the
9723 // GT test.  The EQ test is done by ORcc'ing the high and low halves, and the
9724 // NE test is negated from that.
9725 
9726 // Due to a shortcoming in the ADLC, it mixes up expressions like:
9727 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)).  Note the
9728 // difference between 'Y' and '0L'.  The tree-matches for the CmpI sections
9729 // are collapsed internally in the ADLC's dfa-gen code.  The match for
9730 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
9731 // foo match ends up with the wrong leaf.  One fix is to not match both
9732 // reg-reg and reg-zero forms of long-compare.  This is unfortunate because
9733 // both forms beat the trinary form of long-compare and both are very useful
9734 // on Intel which has so few registers.
9735 
9736 instruct branchCon_long(cmpOp cmp, flagsRegL xcc, label labl) %{
9737   match(If cmp xcc);
9738   effect(USE labl);
9739 
9740   size(8);
9741   ins_cost(BRANCH_COST);
9742   format %{ "BP$cmp   $xcc,$labl" %}
9743   ins_encode %{
9744     Label* L = $labl$$label;
9745     Assembler::Predict predict_taken =
9746       cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9747 
9748     __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9749     __ delayed()->nop();
9750   %}
9751   ins_pipe(br_cc);
9752 %}
9753 
9754 // Manifest a CmpL3 result in an integer register.  Very painful.
9755 // This is the test to avoid.
9756 instruct cmpL3_reg_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg ccr ) %{
9757   match(Set dst (CmpL3 src1 src2) );
9758   effect( KILL ccr );
9759   ins_cost(6*DEFAULT_COST);
9760   size(24);
9761   format %{ "CMP    $src1,$src2\t\t! long\n"
9762           "\tBLT,a,pn done\n"
9763           "\tMOV    -1,$dst\t! delay slot\n"
9764           "\tBGT,a,pn done\n"
9765           "\tMOV    1,$dst\t! delay slot\n"
9766           "\tCLR    $dst\n"
9767     "done:"     %}
9768   ins_encode( cmpl_flag(src1,src2,dst) );
9769   ins_pipe(cmpL_reg);
9770 %}
9771 
9772 // Conditional move
9773 instruct cmovLL_reg(cmpOp cmp, flagsRegL xcc, iRegL dst, iRegL src) %{
9774   match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9775   ins_cost(150);
9776   format %{ "MOV$cmp  $xcc,$src,$dst\t! long" %}
9777   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9778   ins_pipe(ialu_reg);
9779 %}
9780 
9781 instruct cmovLL_imm(cmpOp cmp, flagsRegL xcc, iRegL dst, immL0 src) %{
9782   match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9783   ins_cost(140);
9784   format %{ "MOV$cmp  $xcc,$src,$dst\t! long" %}
9785   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9786   ins_pipe(ialu_imm);
9787 %}
9788 
9789 instruct cmovIL_reg(cmpOp cmp, flagsRegL xcc, iRegI dst, iRegI src) %{
9790   match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9791   ins_cost(150);
9792   format %{ "MOV$cmp  $xcc,$src,$dst" %}
9793   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9794   ins_pipe(ialu_reg);
9795 %}
9796 
9797 instruct cmovIL_imm(cmpOp cmp, flagsRegL xcc, iRegI dst, immI11 src) %{
9798   match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9799   ins_cost(140);
9800   format %{ "MOV$cmp  $xcc,$src,$dst" %}
9801   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9802   ins_pipe(ialu_imm);
9803 %}
9804 
9805 instruct cmovNL_reg(cmpOp cmp, flagsRegL xcc, iRegN dst, iRegN src) %{
9806   match(Set dst (CMoveN (Binary cmp xcc) (Binary dst src)));
9807   ins_cost(150);
9808   format %{ "MOV$cmp  $xcc,$src,$dst" %}
9809   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9810   ins_pipe(ialu_reg);
9811 %}
9812 
9813 instruct cmovPL_reg(cmpOp cmp, flagsRegL xcc, iRegP dst, iRegP src) %{
9814   match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9815   ins_cost(150);
9816   format %{ "MOV$cmp  $xcc,$src,$dst" %}
9817   ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9818   ins_pipe(ialu_reg);
9819 %}
9820 
9821 instruct cmovPL_imm(cmpOp cmp, flagsRegL xcc, iRegP dst, immP0 src) %{
9822   match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9823   ins_cost(140);
9824   format %{ "MOV$cmp  $xcc,$src,$dst" %}
9825   ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9826   ins_pipe(ialu_imm);
9827 %}
9828 
9829 instruct cmovFL_reg(cmpOp cmp, flagsRegL xcc, regF dst, regF src) %{
9830   match(Set dst (CMoveF (Binary cmp xcc) (Binary dst src)));
9831   ins_cost(150);
9832   opcode(0x101);
9833   format %{ "FMOVS$cmp $xcc,$src,$dst" %}
9834   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9835   ins_pipe(int_conditional_float_move);
9836 %}
9837 
9838 instruct cmovDL_reg(cmpOp cmp, flagsRegL xcc, regD dst, regD src) %{
9839   match(Set dst (CMoveD (Binary cmp xcc) (Binary dst src)));
9840   ins_cost(150);
9841   opcode(0x102);
9842   format %{ "FMOVD$cmp $xcc,$src,$dst" %}
9843   ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9844   ins_pipe(int_conditional_float_move);
9845 %}
9846 
9847 // ============================================================================
9848 // Safepoint Instruction
9849 instruct safePoint_poll(iRegP poll) %{
9850   match(SafePoint poll);
9851   effect(USE poll);
9852 
9853   size(4);
9854 #ifdef _LP64
9855   format %{ "LDX    [$poll],R_G0\t! Safepoint: poll for GC" %}
9856 #else
9857   format %{ "LDUW   [$poll],R_G0\t! Safepoint: poll for GC" %}
9858 #endif
9859   ins_encode %{
9860     __ relocate(relocInfo::poll_type);
9861     __ ld_ptr($poll$$Register, 0, G0);
9862   %}
9863   ins_pipe(loadPollP);
9864 %}
9865 
9866 // ============================================================================
9867 // Call Instructions
9868 // Call Java Static Instruction
9869 instruct CallStaticJavaDirect( method meth ) %{
9870   match(CallStaticJava);
9871   predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
9872   effect(USE meth);
9873 
9874   size(8);
9875   ins_cost(CALL_COST);
9876   format %{ "CALL,static  ; NOP ==> " %}
9877   ins_encode( Java_Static_Call( meth ), call_epilog );
9878   ins_pipe(simple_call);
9879 %}
9880 
9881 // Call Java Static Instruction (method handle version)
9882 instruct CallStaticJavaHandle(method meth, l7RegP l7_mh_SP_save) %{
9883   match(CallStaticJava);
9884   predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
9885   effect(USE meth, KILL l7_mh_SP_save);
9886 
9887   size(16);
9888   ins_cost(CALL_COST);
9889   format %{ "CALL,static/MethodHandle" %}
9890   ins_encode(preserve_SP, Java_Static_Call(meth), restore_SP, call_epilog);
9891   ins_pipe(simple_call);
9892 %}
9893 
9894 // Call Java Dynamic Instruction
9895 instruct CallDynamicJavaDirect( method meth ) %{
9896   match(CallDynamicJava);
9897   effect(USE meth);
9898 
9899   ins_cost(CALL_COST);
9900   format %{ "SET    (empty),R_G5\n\t"
9901             "CALL,dynamic  ; NOP ==> " %}
9902   ins_encode( Java_Dynamic_Call( meth ), call_epilog );
9903   ins_pipe(call);
9904 %}
9905 
9906 // Call Runtime Instruction
9907 instruct CallRuntimeDirect(method meth, l7RegP l7) %{
9908   match(CallRuntime);
9909   effect(USE meth, KILL l7);
9910   ins_cost(CALL_COST);
9911   format %{ "CALL,runtime" %}
9912   ins_encode( Java_To_Runtime( meth ),
9913               call_epilog, adjust_long_from_native_call );
9914   ins_pipe(simple_call);
9915 %}
9916 
9917 // Call runtime without safepoint - same as CallRuntime
9918 instruct CallLeafDirect(method meth, l7RegP l7) %{
9919   match(CallLeaf);
9920   effect(USE meth, KILL l7);
9921   ins_cost(CALL_COST);
9922   format %{ "CALL,runtime leaf" %}
9923   ins_encode( Java_To_Runtime( meth ),
9924               call_epilog,
9925               adjust_long_from_native_call );
9926   ins_pipe(simple_call);
9927 %}
9928 
9929 // Call runtime without safepoint - same as CallLeaf
9930 instruct CallLeafNoFPDirect(method meth, l7RegP l7) %{
9931   match(CallLeafNoFP);
9932   effect(USE meth, KILL l7);
9933   ins_cost(CALL_COST);
9934   format %{ "CALL,runtime leaf nofp" %}
9935   ins_encode( Java_To_Runtime( meth ),
9936               call_epilog,
9937               adjust_long_from_native_call );
9938   ins_pipe(simple_call);
9939 %}
9940 
9941 // Tail Call; Jump from runtime stub to Java code.
9942 // Also known as an 'interprocedural jump'.
9943 // Target of jump will eventually return to caller.
9944 // TailJump below removes the return address.
9945 instruct TailCalljmpInd(g3RegP jump_target, inline_cache_regP method_oop) %{
9946   match(TailCall jump_target method_oop );
9947 
9948   ins_cost(CALL_COST);
9949   format %{ "Jmp     $jump_target  ; NOP \t! $method_oop holds method oop" %}
9950   ins_encode(form_jmpl(jump_target));
9951   ins_pipe(tail_call);
9952 %}
9953 
9954 
9955 // Return Instruction
9956 instruct Ret() %{
9957   match(Return);
9958 
9959   // The epilogue node did the ret already.
9960   size(0);
9961   format %{ "! return" %}
9962   ins_encode();
9963   ins_pipe(empty);
9964 %}
9965 
9966 
9967 // Tail Jump; remove the return address; jump to target.
9968 // TailCall above leaves the return address around.
9969 // TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2).
9970 // ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a
9971 // "restore" before this instruction (in Epilogue), we need to materialize it
9972 // in %i0.
9973 instruct tailjmpInd(g1RegP jump_target, i0RegP ex_oop) %{
9974   match( TailJump jump_target ex_oop );
9975   ins_cost(CALL_COST);
9976   format %{ "! discard R_O7\n\t"
9977             "Jmp     $jump_target  ; ADD O7,8,O1 \t! $ex_oop holds exc. oop" %}
9978   ins_encode(form_jmpl_set_exception_pc(jump_target));
9979   // opcode(Assembler::jmpl_op3, Assembler::arith_op);
9980   // The hack duplicates the exception oop into G3, so that CreateEx can use it there.
9981   // ins_encode( form3_rs1_simm13_rd( jump_target, 0x00, R_G0 ), move_return_pc_to_o1() );
9982   ins_pipe(tail_call);
9983 %}
9984 
9985 // Create exception oop: created by stack-crawling runtime code.
9986 // Created exception is now available to this handler, and is setup
9987 // just prior to jumping to this handler.  No code emitted.
9988 instruct CreateException( o0RegP ex_oop )
9989 %{
9990   match(Set ex_oop (CreateEx));
9991   ins_cost(0);
9992 
9993   size(0);
9994   // use the following format syntax
9995   format %{ "! exception oop is in R_O0; no code emitted" %}
9996   ins_encode();
9997   ins_pipe(empty);
9998 %}
9999 
10000 
10001 // Rethrow exception:
10002 // The exception oop will come in the first argument position.
10003 // Then JUMP (not call) to the rethrow stub code.
10004 instruct RethrowException()
10005 %{
10006   match(Rethrow);
10007   ins_cost(CALL_COST);
10008 
10009   // use the following format syntax
10010   format %{ "Jmp    rethrow_stub" %}
10011   ins_encode(enc_rethrow);
10012   ins_pipe(tail_call);
10013 %}
10014 
10015 
10016 // Die now
10017 instruct ShouldNotReachHere( )
10018 %{
10019   match(Halt);
10020   ins_cost(CALL_COST);
10021 
10022   size(4);
10023   // Use the following format syntax
10024   format %{ "ILLTRAP   ; ShouldNotReachHere" %}
10025   ins_encode( form2_illtrap() );
10026   ins_pipe(tail_call);
10027 %}
10028 
10029 // ============================================================================
10030 // The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary superklass
10031 // array for an instance of the superklass.  Set a hidden internal cache on a
10032 // hit (cache is checked with exposed code in gen_subtype_check()).  Return
10033 // not zero for a miss or zero for a hit.  The encoding ALSO sets flags.
10034 instruct partialSubtypeCheck( o0RegP index, o1RegP sub, o2RegP super, flagsRegP pcc, o7RegP o7 ) %{
10035   match(Set index (PartialSubtypeCheck sub super));
10036   effect( KILL pcc, KILL o7 );
10037   ins_cost(DEFAULT_COST*10);
10038   format %{ "CALL   PartialSubtypeCheck\n\tNOP" %}
10039   ins_encode( enc_PartialSubtypeCheck() );
10040   ins_pipe(partial_subtype_check_pipe);
10041 %}
10042 
10043 instruct partialSubtypeCheck_vs_zero( flagsRegP pcc, o1RegP sub, o2RegP super, immP0 zero, o0RegP idx, o7RegP o7 ) %{
10044   match(Set pcc (CmpP (PartialSubtypeCheck sub super) zero));
10045   effect( KILL idx, KILL o7 );
10046   ins_cost(DEFAULT_COST*10);
10047   format %{ "CALL   PartialSubtypeCheck\n\tNOP\t# (sets condition codes)" %}
10048   ins_encode( enc_PartialSubtypeCheck() );
10049   ins_pipe(partial_subtype_check_pipe);
10050 %}
10051 
10052 
10053 // ============================================================================
10054 // inlined locking and unlocking
10055 
10056 instruct cmpFastLock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10057   match(Set pcc (FastLock object box));
10058 
10059   effect(TEMP scratch2, USE_KILL box, KILL scratch);
10060   ins_cost(100);
10061 
10062   format %{ "FASTLOCK  $object,$box\t! kills $box,$scratch,$scratch2" %}
10063   ins_encode( Fast_Lock(object, box, scratch, scratch2) );
10064   ins_pipe(long_memory_op);
10065 %}
10066 
10067 
10068 instruct cmpFastUnlock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10069   match(Set pcc (FastUnlock object box));
10070   effect(TEMP scratch2, USE_KILL box, KILL scratch);
10071   ins_cost(100);
10072 
10073   format %{ "FASTUNLOCK  $object,$box\t! kills $box,$scratch,$scratch2" %}
10074   ins_encode( Fast_Unlock(object, box, scratch, scratch2) );
10075   ins_pipe(long_memory_op);
10076 %}
10077 
10078 // The encodings are generic.
10079 instruct clear_array(iRegX cnt, iRegP base, iRegX temp, Universe dummy, flagsReg ccr) %{
10080   predicate(!use_block_zeroing(n->in(2)) );
10081   match(Set dummy (ClearArray cnt base));
10082   effect(TEMP temp, KILL ccr);
10083   ins_cost(300);
10084   format %{ "MOV    $cnt,$temp\n"
10085     "loop:   SUBcc  $temp,8,$temp\t! Count down a dword of bytes\n"
10086     "        BRge   loop\t\t! Clearing loop\n"
10087     "        STX    G0,[$base+$temp]\t! delay slot" %}
10088 
10089   ins_encode %{
10090     // Compiler ensures base is doubleword aligned and cnt is count of doublewords
10091     Register nof_bytes_arg    = $cnt$$Register;
10092     Register nof_bytes_tmp    = $temp$$Register;
10093     Register base_pointer_arg = $base$$Register;
10094 
10095     Label loop;
10096     __ mov(nof_bytes_arg, nof_bytes_tmp);
10097 
10098     // Loop and clear, walking backwards through the array.
10099     // nof_bytes_tmp (if >0) is always the number of bytes to zero
10100     __ bind(loop);
10101     __ deccc(nof_bytes_tmp, 8);
10102     __ br(Assembler::greaterEqual, true, Assembler::pt, loop);
10103     __ delayed()-> stx(G0, base_pointer_arg, nof_bytes_tmp);
10104     // %%%% this mini-loop must not cross a cache boundary!
10105   %}
10106   ins_pipe(long_memory_op);
10107 %}
10108 
10109 instruct clear_array_bis(g1RegX cnt, o0RegP base, Universe dummy, flagsReg ccr) %{
10110   predicate(use_block_zeroing(n->in(2)));
10111   match(Set dummy (ClearArray cnt base));
10112   effect(USE_KILL cnt, USE_KILL base, KILL ccr);
10113   ins_cost(300);
10114   format %{ "CLEAR  [$base, $cnt]\t! ClearArray" %}
10115 
10116   ins_encode %{
10117 
10118     assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10119     Register to    = $base$$Register;
10120     Register count = $cnt$$Register;
10121 
10122     Label Ldone;
10123     __ nop(); // Separate short branches
10124     // Use BIS for zeroing (temp is not used).
10125     __ bis_zeroing(to, count, G0, Ldone);
10126     __ bind(Ldone);
10127 
10128   %}
10129   ins_pipe(long_memory_op);
10130 %}
10131 
10132 instruct clear_array_bis_2(g1RegX cnt, o0RegP base, iRegX tmp, Universe dummy, flagsReg ccr) %{
10133   predicate(use_block_zeroing(n->in(2)) && !Assembler::is_simm13((int)BlockZeroingLowLimit));
10134   match(Set dummy (ClearArray cnt base));
10135   effect(TEMP tmp, USE_KILL cnt, USE_KILL base, KILL ccr);
10136   ins_cost(300);
10137   format %{ "CLEAR  [$base, $cnt]\t! ClearArray" %}
10138 
10139   ins_encode %{
10140 
10141     assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10142     Register to    = $base$$Register;
10143     Register count = $cnt$$Register;
10144     Register temp  = $tmp$$Register;
10145 
10146     Label Ldone;
10147     __ nop(); // Separate short branches
10148     // Use BIS for zeroing
10149     __ bis_zeroing(to, count, temp, Ldone);
10150     __ bind(Ldone);
10151 
10152   %}
10153   ins_pipe(long_memory_op);
10154 %}
10155 
10156 instruct string_compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result,
10157                         o7RegI tmp, flagsReg ccr) %{
10158   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
10159   effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ccr, KILL tmp);
10160   ins_cost(300);
10161   format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result   // KILL $tmp" %}
10162   ins_encode( enc_String_Compare(str1, str2, cnt1, cnt2, result) );
10163   ins_pipe(long_memory_op);
10164 %}
10165 
10166 instruct string_equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result,
10167                        o7RegI tmp, flagsReg ccr) %{
10168   match(Set result (StrEquals (Binary str1 str2) cnt));
10169   effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp, KILL ccr);
10170   ins_cost(300);
10171   format %{ "String Equals $str1,$str2,$cnt -> $result   // KILL $tmp" %}
10172   ins_encode( enc_String_Equals(str1, str2, cnt, result) );
10173   ins_pipe(long_memory_op);
10174 %}
10175 
10176 instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegI tmp1, notemp_iRegI result,
10177                       o7RegI tmp2, flagsReg ccr) %{
10178   match(Set result (AryEq ary1 ary2));
10179   effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr);
10180   ins_cost(300);
10181   format %{ "Array Equals $ary1,$ary2 -> $result   // KILL $tmp1,$tmp2" %}
10182   ins_encode( enc_Array_Equals(ary1, ary2, tmp1, result));
10183   ins_pipe(long_memory_op);
10184 %}
10185 
10186 
10187 //---------- Zeros Count Instructions ------------------------------------------
10188 
10189 instruct countLeadingZerosI(iRegIsafe dst, iRegI src, iRegI tmp, flagsReg cr) %{
10190   predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
10191   match(Set dst (CountLeadingZerosI src));
10192   effect(TEMP dst, TEMP tmp, KILL cr);
10193 
10194   // x |= (x >> 1);
10195   // x |= (x >> 2);
10196   // x |= (x >> 4);
10197   // x |= (x >> 8);
10198   // x |= (x >> 16);
10199   // return (WORDBITS - popc(x));
10200   format %{ "SRL     $src,1,$tmp\t! count leading zeros (int)\n\t"
10201             "SRL     $src,0,$dst\t! 32-bit zero extend\n\t"
10202             "OR      $dst,$tmp,$dst\n\t"
10203             "SRL     $dst,2,$tmp\n\t"
10204             "OR      $dst,$tmp,$dst\n\t"
10205             "SRL     $dst,4,$tmp\n\t"
10206             "OR      $dst,$tmp,$dst\n\t"
10207             "SRL     $dst,8,$tmp\n\t"
10208             "OR      $dst,$tmp,$dst\n\t"
10209             "SRL     $dst,16,$tmp\n\t"
10210             "OR      $dst,$tmp,$dst\n\t"
10211             "POPC    $dst,$dst\n\t"
10212             "MOV     32,$tmp\n\t"
10213             "SUB     $tmp,$dst,$dst" %}
10214   ins_encode %{
10215     Register Rdst = $dst$$Register;
10216     Register Rsrc = $src$$Register;
10217     Register Rtmp = $tmp$$Register;
10218     __ srl(Rsrc, 1,    Rtmp);
10219     __ srl(Rsrc, 0,    Rdst);
10220     __ or3(Rdst, Rtmp, Rdst);
10221     __ srl(Rdst, 2,    Rtmp);
10222     __ or3(Rdst, Rtmp, Rdst);
10223     __ srl(Rdst, 4,    Rtmp);
10224     __ or3(Rdst, Rtmp, Rdst);
10225     __ srl(Rdst, 8,    Rtmp);
10226     __ or3(Rdst, Rtmp, Rdst);
10227     __ srl(Rdst, 16,   Rtmp);
10228     __ or3(Rdst, Rtmp, Rdst);
10229     __ popc(Rdst, Rdst);
10230     __ mov(BitsPerInt, Rtmp);
10231     __ sub(Rtmp, Rdst, Rdst);
10232   %}
10233   ins_pipe(ialu_reg);
10234 %}
10235 
10236 instruct countLeadingZerosL(iRegIsafe dst, iRegL src, iRegL tmp, flagsReg cr) %{
10237   predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
10238   match(Set dst (CountLeadingZerosL src));
10239   effect(TEMP dst, TEMP tmp, KILL cr);
10240 
10241   // x |= (x >> 1);
10242   // x |= (x >> 2);
10243   // x |= (x >> 4);
10244   // x |= (x >> 8);
10245   // x |= (x >> 16);
10246   // x |= (x >> 32);
10247   // return (WORDBITS - popc(x));
10248   format %{ "SRLX    $src,1,$tmp\t! count leading zeros (long)\n\t"
10249             "OR      $src,$tmp,$dst\n\t"
10250             "SRLX    $dst,2,$tmp\n\t"
10251             "OR      $dst,$tmp,$dst\n\t"
10252             "SRLX    $dst,4,$tmp\n\t"
10253             "OR      $dst,$tmp,$dst\n\t"
10254             "SRLX    $dst,8,$tmp\n\t"
10255             "OR      $dst,$tmp,$dst\n\t"
10256             "SRLX    $dst,16,$tmp\n\t"
10257             "OR      $dst,$tmp,$dst\n\t"
10258             "SRLX    $dst,32,$tmp\n\t"
10259             "OR      $dst,$tmp,$dst\n\t"
10260             "POPC    $dst,$dst\n\t"
10261             "MOV     64,$tmp\n\t"
10262             "SUB     $tmp,$dst,$dst" %}
10263   ins_encode %{
10264     Register Rdst = $dst$$Register;
10265     Register Rsrc = $src$$Register;
10266     Register Rtmp = $tmp$$Register;
10267     __ srlx(Rsrc, 1,    Rtmp);
10268     __ or3( Rsrc, Rtmp, Rdst);
10269     __ srlx(Rdst, 2,    Rtmp);
10270     __ or3( Rdst, Rtmp, Rdst);
10271     __ srlx(Rdst, 4,    Rtmp);
10272     __ or3( Rdst, Rtmp, Rdst);
10273     __ srlx(Rdst, 8,    Rtmp);
10274     __ or3( Rdst, Rtmp, Rdst);
10275     __ srlx(Rdst, 16,   Rtmp);
10276     __ or3( Rdst, Rtmp, Rdst);
10277     __ srlx(Rdst, 32,   Rtmp);
10278     __ or3( Rdst, Rtmp, Rdst);
10279     __ popc(Rdst, Rdst);
10280     __ mov(BitsPerLong, Rtmp);
10281     __ sub(Rtmp, Rdst, Rdst);
10282   %}
10283   ins_pipe(ialu_reg);
10284 %}
10285 
10286 instruct countTrailingZerosI(iRegIsafe dst, iRegI src, flagsReg cr) %{
10287   predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
10288   match(Set dst (CountTrailingZerosI src));
10289   effect(TEMP dst, KILL cr);
10290 
10291   // return popc(~x & (x - 1));
10292   format %{ "SUB     $src,1,$dst\t! count trailing zeros (int)\n\t"
10293             "ANDN    $dst,$src,$dst\n\t"
10294             "SRL     $dst,R_G0,$dst\n\t"
10295             "POPC    $dst,$dst" %}
10296   ins_encode %{
10297     Register Rdst = $dst$$Register;
10298     Register Rsrc = $src$$Register;
10299     __ sub(Rsrc, 1, Rdst);
10300     __ andn(Rdst, Rsrc, Rdst);
10301     __ srl(Rdst, G0, Rdst);
10302     __ popc(Rdst, Rdst);
10303   %}
10304   ins_pipe(ialu_reg);
10305 %}
10306 
10307 instruct countTrailingZerosL(iRegIsafe dst, iRegL src, flagsReg cr) %{
10308   predicate(UsePopCountInstruction);  // See Matcher::match_rule_supported
10309   match(Set dst (CountTrailingZerosL src));
10310   effect(TEMP dst, KILL cr);
10311 
10312   // return popc(~x & (x - 1));
10313   format %{ "SUB     $src,1,$dst\t! count trailing zeros (long)\n\t"
10314             "ANDN    $dst,$src,$dst\n\t"
10315             "POPC    $dst,$dst" %}
10316   ins_encode %{
10317     Register Rdst = $dst$$Register;
10318     Register Rsrc = $src$$Register;
10319     __ sub(Rsrc, 1, Rdst);
10320     __ andn(Rdst, Rsrc, Rdst);
10321     __ popc(Rdst, Rdst);
10322   %}
10323   ins_pipe(ialu_reg);
10324 %}
10325 
10326 
10327 //---------- Population Count Instructions -------------------------------------
10328 
10329 instruct popCountI(iRegIsafe dst, iRegI src) %{
10330   predicate(UsePopCountInstruction);
10331   match(Set dst (PopCountI src));
10332 
10333   format %{ "SRL    $src, G0, $dst\t! clear upper word for 64 bit POPC\n\t"
10334             "POPC   $dst, $dst" %}
10335   ins_encode %{
10336     __ srl($src$$Register, G0, $dst$$Register);
10337     __ popc($dst$$Register, $dst$$Register);
10338   %}
10339   ins_pipe(ialu_reg);
10340 %}
10341 
10342 // Note: Long.bitCount(long) returns an int.
10343 instruct popCountL(iRegIsafe dst, iRegL src) %{
10344   predicate(UsePopCountInstruction);
10345   match(Set dst (PopCountL src));
10346 
10347   format %{ "POPC   $src, $dst" %}
10348   ins_encode %{
10349     __ popc($src$$Register, $dst$$Register);
10350   %}
10351   ins_pipe(ialu_reg);
10352 %}
10353 
10354 
10355 // ============================================================================
10356 //------------Bytes reverse--------------------------------------------------
10357 
10358 instruct bytes_reverse_int(iRegI dst, stackSlotI src) %{
10359   match(Set dst (ReverseBytesI src));
10360 
10361   // Op cost is artificially doubled to make sure that load or store
10362   // instructions are preferred over this one which requires a spill
10363   // onto a stack slot.
10364   ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10365   format %{ "LDUWA  $src, $dst\t!asi=primary_little" %}
10366 
10367   ins_encode %{
10368     __ set($src$$disp + STACK_BIAS, O7);
10369     __ lduwa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10370   %}
10371   ins_pipe( iload_mem );
10372 %}
10373 
10374 instruct bytes_reverse_long(iRegL dst, stackSlotL src) %{
10375   match(Set dst (ReverseBytesL src));
10376 
10377   // Op cost is artificially doubled to make sure that load or store
10378   // instructions are preferred over this one which requires a spill
10379   // onto a stack slot.
10380   ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10381   format %{ "LDXA   $src, $dst\t!asi=primary_little" %}
10382 
10383   ins_encode %{
10384     __ set($src$$disp + STACK_BIAS, O7);
10385     __ ldxa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10386   %}
10387   ins_pipe( iload_mem );
10388 %}
10389 
10390 instruct bytes_reverse_unsigned_short(iRegI dst, stackSlotI src) %{
10391   match(Set dst (ReverseBytesUS src));
10392 
10393   // Op cost is artificially doubled to make sure that load or store
10394   // instructions are preferred over this one which requires a spill
10395   // onto a stack slot.
10396   ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10397   format %{ "LDUHA  $src, $dst\t!asi=primary_little\n\t" %}
10398 
10399   ins_encode %{
10400     // the value was spilled as an int so bias the load
10401     __ set($src$$disp + STACK_BIAS + 2, O7);
10402     __ lduha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10403   %}
10404   ins_pipe( iload_mem );
10405 %}
10406 
10407 instruct bytes_reverse_short(iRegI dst, stackSlotI src) %{
10408   match(Set dst (ReverseBytesS src));
10409 
10410   // Op cost is artificially doubled to make sure that load or store
10411   // instructions are preferred over this one which requires a spill
10412   // onto a stack slot.
10413   ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10414   format %{ "LDSHA  $src, $dst\t!asi=primary_little\n\t" %}
10415 
10416   ins_encode %{
10417     // the value was spilled as an int so bias the load
10418     __ set($src$$disp + STACK_BIAS + 2, O7);
10419     __ ldsha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10420   %}
10421   ins_pipe( iload_mem );
10422 %}
10423 
10424 // Load Integer reversed byte order
10425 instruct loadI_reversed(iRegI dst, indIndexMemory src) %{
10426   match(Set dst (ReverseBytesI (LoadI src)));
10427 
10428   ins_cost(DEFAULT_COST + MEMORY_REF_COST);
10429   size(4);
10430   format %{ "LDUWA  $src, $dst\t!asi=primary_little" %}
10431 
10432   ins_encode %{
10433     __ lduwa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10434   %}
10435   ins_pipe(iload_mem);
10436 %}
10437 
10438 // Load Long - aligned and reversed
10439 instruct loadL_reversed(iRegL dst, indIndexMemory src) %{
10440   match(Set dst (ReverseBytesL (LoadL src)));
10441 
10442   ins_cost(MEMORY_REF_COST);
10443   size(4);
10444   format %{ "LDXA   $src, $dst\t!asi=primary_little" %}
10445 
10446   ins_encode %{
10447     __ ldxa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10448   %}
10449   ins_pipe(iload_mem);
10450 %}
10451 
10452 // Load unsigned short / char reversed byte order
10453 instruct loadUS_reversed(iRegI dst, indIndexMemory src) %{
10454   match(Set dst (ReverseBytesUS (LoadUS src)));
10455 
10456   ins_cost(MEMORY_REF_COST);
10457   size(4);
10458   format %{ "LDUHA  $src, $dst\t!asi=primary_little" %}
10459 
10460   ins_encode %{
10461     __ lduha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10462   %}
10463   ins_pipe(iload_mem);
10464 %}
10465 
10466 // Load short reversed byte order
10467 instruct loadS_reversed(iRegI dst, indIndexMemory src) %{
10468   match(Set dst (ReverseBytesS (LoadS src)));
10469 
10470   ins_cost(MEMORY_REF_COST);
10471   size(4);
10472   format %{ "LDSHA  $src, $dst\t!asi=primary_little" %}
10473 
10474   ins_encode %{
10475     __ ldsha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10476   %}
10477   ins_pipe(iload_mem);
10478 %}
10479 
10480 // Store Integer reversed byte order
10481 instruct storeI_reversed(indIndexMemory dst, iRegI src) %{
10482   match(Set dst (StoreI dst (ReverseBytesI src)));
10483 
10484   ins_cost(MEMORY_REF_COST);
10485   size(4);
10486   format %{ "STWA   $src, $dst\t!asi=primary_little" %}
10487 
10488   ins_encode %{
10489     __ stwa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10490   %}
10491   ins_pipe(istore_mem_reg);
10492 %}
10493 
10494 // Store Long reversed byte order
10495 instruct storeL_reversed(indIndexMemory dst, iRegL src) %{
10496   match(Set dst (StoreL dst (ReverseBytesL src)));
10497 
10498   ins_cost(MEMORY_REF_COST);
10499   size(4);
10500   format %{ "STXA   $src, $dst\t!asi=primary_little" %}
10501 
10502   ins_encode %{
10503     __ stxa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10504   %}
10505   ins_pipe(istore_mem_reg);
10506 %}
10507 
10508 // Store unsighed short/char reversed byte order
10509 instruct storeUS_reversed(indIndexMemory dst, iRegI src) %{
10510   match(Set dst (StoreC dst (ReverseBytesUS src)));
10511 
10512   ins_cost(MEMORY_REF_COST);
10513   size(4);
10514   format %{ "STHA   $src, $dst\t!asi=primary_little" %}
10515 
10516   ins_encode %{
10517     __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10518   %}
10519   ins_pipe(istore_mem_reg);
10520 %}
10521 
10522 // Store short reversed byte order
10523 instruct storeS_reversed(indIndexMemory dst, iRegI src) %{
10524   match(Set dst (StoreC dst (ReverseBytesS src)));
10525 
10526   ins_cost(MEMORY_REF_COST);
10527   size(4);
10528   format %{ "STHA   $src, $dst\t!asi=primary_little" %}
10529 
10530   ins_encode %{
10531     __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10532   %}
10533   ins_pipe(istore_mem_reg);
10534 %}
10535 
10536 // ====================VECTOR INSTRUCTIONS=====================================
10537 
10538 // Load Aligned Packed values into a Double Register
10539 instruct loadV8(regD dst, memory mem) %{
10540   predicate(n->as_LoadVector()->memory_size() == 8);
10541   match(Set dst (LoadVector mem));
10542   ins_cost(MEMORY_REF_COST);
10543   size(4);
10544   format %{ "LDDF   $mem,$dst\t! load vector (8 bytes)" %}
10545   ins_encode %{
10546     __ ldf(FloatRegisterImpl::D, $mem$$Address, as_DoubleFloatRegister($dst$$reg));
10547   %}
10548   ins_pipe(floadD_mem);
10549 %}
10550 
10551 // Store Vector in Double register to memory
10552 instruct storeV8(memory mem, regD src) %{
10553   predicate(n->as_StoreVector()->memory_size() == 8);
10554   match(Set mem (StoreVector mem src));
10555   ins_cost(MEMORY_REF_COST);
10556   size(4);
10557   format %{ "STDF   $src,$mem\t! store vector (8 bytes)" %}
10558   ins_encode %{
10559     __ stf(FloatRegisterImpl::D, as_DoubleFloatRegister($src$$reg), $mem$$Address);
10560   %}
10561   ins_pipe(fstoreD_mem_reg);
10562 %}
10563 
10564 // Store Zero into vector in memory
10565 instruct storeV8B_zero(memory mem, immI0 zero) %{
10566   predicate(n->as_StoreVector()->memory_size() == 8);
10567   match(Set mem (StoreVector mem (ReplicateB zero)));
10568   ins_cost(MEMORY_REF_COST);
10569   size(4);
10570   format %{ "STX    $zero,$mem\t! store zero vector (8 bytes)" %}
10571   ins_encode %{
10572     __ stx(G0, $mem$$Address);
10573   %}
10574   ins_pipe(fstoreD_mem_zero);
10575 %}
10576 
10577 instruct storeV4S_zero(memory mem, immI0 zero) %{
10578   predicate(n->as_StoreVector()->memory_size() == 8);
10579   match(Set mem (StoreVector mem (ReplicateS zero)));
10580   ins_cost(MEMORY_REF_COST);
10581   size(4);
10582   format %{ "STX    $zero,$mem\t! store zero vector (4 shorts)" %}
10583   ins_encode %{
10584     __ stx(G0, $mem$$Address);
10585   %}
10586   ins_pipe(fstoreD_mem_zero);
10587 %}
10588 
10589 instruct storeV2I_zero(memory mem, immI0 zero) %{
10590   predicate(n->as_StoreVector()->memory_size() == 8);
10591   match(Set mem (StoreVector mem (ReplicateI zero)));
10592   ins_cost(MEMORY_REF_COST);
10593   size(4);
10594   format %{ "STX    $zero,$mem\t! store zero vector (2 ints)" %}
10595   ins_encode %{
10596     __ stx(G0, $mem$$Address);
10597   %}
10598   ins_pipe(fstoreD_mem_zero);
10599 %}
10600 
10601 instruct storeV2F_zero(memory mem, immF0 zero) %{
10602   predicate(n->as_StoreVector()->memory_size() == 8);
10603   match(Set mem (StoreVector mem (ReplicateF zero)));
10604   ins_cost(MEMORY_REF_COST);
10605   size(4);
10606   format %{ "STX    $zero,$mem\t! store zero vector (2 floats)" %}
10607   ins_encode %{
10608     __ stx(G0, $mem$$Address);
10609   %}
10610   ins_pipe(fstoreD_mem_zero);
10611 %}
10612 
10613 // Replicate scalar to packed byte values into Double register
10614 instruct Repl8B_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10615   predicate(n->as_Vector()->length() == 8 && UseVIS >= 3);
10616   match(Set dst (ReplicateB src));
10617   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10618   format %{ "SLLX  $src,56,$tmp\n\t"
10619             "SRLX  $tmp, 8,$tmp2\n\t"
10620             "OR    $tmp,$tmp2,$tmp\n\t"
10621             "SRLX  $tmp,16,$tmp2\n\t"
10622             "OR    $tmp,$tmp2,$tmp\n\t"
10623             "SRLX  $tmp,32,$tmp2\n\t"
10624             "OR    $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10625             "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10626   ins_encode %{
10627     Register Rsrc = $src$$Register;
10628     Register Rtmp = $tmp$$Register;
10629     Register Rtmp2 = $tmp2$$Register;
10630     __ sllx(Rsrc,    56, Rtmp);
10631     __ srlx(Rtmp,     8, Rtmp2);
10632     __ or3 (Rtmp, Rtmp2, Rtmp);
10633     __ srlx(Rtmp,    16, Rtmp2);
10634     __ or3 (Rtmp, Rtmp2, Rtmp);
10635     __ srlx(Rtmp,    32, Rtmp2);
10636     __ or3 (Rtmp, Rtmp2, Rtmp);
10637     __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10638   %}
10639   ins_pipe(ialu_reg);
10640 %}
10641 
10642 // Replicate scalar to packed byte values into Double stack
10643 instruct Repl8B_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10644   predicate(n->as_Vector()->length() == 8 && UseVIS < 3);
10645   match(Set dst (ReplicateB src));
10646   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10647   format %{ "SLLX  $src,56,$tmp\n\t"
10648             "SRLX  $tmp, 8,$tmp2\n\t"
10649             "OR    $tmp,$tmp2,$tmp\n\t"
10650             "SRLX  $tmp,16,$tmp2\n\t"
10651             "OR    $tmp,$tmp2,$tmp\n\t"
10652             "SRLX  $tmp,32,$tmp2\n\t"
10653             "OR    $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10654             "STX   $tmp,$dst\t! regL to stkD" %}
10655   ins_encode %{
10656     Register Rsrc = $src$$Register;
10657     Register Rtmp = $tmp$$Register;
10658     Register Rtmp2 = $tmp2$$Register;
10659     __ sllx(Rsrc,    56, Rtmp);
10660     __ srlx(Rtmp,     8, Rtmp2);
10661     __ or3 (Rtmp, Rtmp2, Rtmp);
10662     __ srlx(Rtmp,    16, Rtmp2);
10663     __ or3 (Rtmp, Rtmp2, Rtmp);
10664     __ srlx(Rtmp,    32, Rtmp2);
10665     __ or3 (Rtmp, Rtmp2, Rtmp);
10666     __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10667     __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10668   %}
10669   ins_pipe(ialu_reg);
10670 %}
10671 
10672 // Replicate scalar constant to packed byte values in Double register
10673 instruct Repl8B_immI(regD dst, immI13 con, o7RegI tmp) %{
10674   predicate(n->as_Vector()->length() == 8);
10675   match(Set dst (ReplicateB con));
10676   effect(KILL tmp);
10677   format %{ "LDDF   [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl8B($con)" %}
10678   ins_encode %{
10679     // XXX This is a quick fix for 6833573.
10680     //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 8, 1)), $dst$$FloatRegister);
10681     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 8, 1)), $tmp$$Register);
10682     __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10683   %}
10684   ins_pipe(loadConFD);
10685 %}
10686 
10687 // Replicate scalar to packed char/short values into Double register
10688 instruct Repl4S_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10689   predicate(n->as_Vector()->length() == 4 && UseVIS >= 3);
10690   match(Set dst (ReplicateS src));
10691   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10692   format %{ "SLLX  $src,48,$tmp\n\t"
10693             "SRLX  $tmp,16,$tmp2\n\t"
10694             "OR    $tmp,$tmp2,$tmp\n\t"
10695             "SRLX  $tmp,32,$tmp2\n\t"
10696             "OR    $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10697             "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10698   ins_encode %{
10699     Register Rsrc = $src$$Register;
10700     Register Rtmp = $tmp$$Register;
10701     Register Rtmp2 = $tmp2$$Register;
10702     __ sllx(Rsrc,    48, Rtmp);
10703     __ srlx(Rtmp,    16, Rtmp2);
10704     __ or3 (Rtmp, Rtmp2, Rtmp);
10705     __ srlx(Rtmp,    32, Rtmp2);
10706     __ or3 (Rtmp, Rtmp2, Rtmp);
10707     __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10708   %}
10709   ins_pipe(ialu_reg);
10710 %}
10711 
10712 // Replicate scalar to packed char/short values into Double stack
10713 instruct Repl4S_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10714   predicate(n->as_Vector()->length() == 4 && UseVIS < 3);
10715   match(Set dst (ReplicateS src));
10716   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10717   format %{ "SLLX  $src,48,$tmp\n\t"
10718             "SRLX  $tmp,16,$tmp2\n\t"
10719             "OR    $tmp,$tmp2,$tmp\n\t"
10720             "SRLX  $tmp,32,$tmp2\n\t"
10721             "OR    $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10722             "STX   $tmp,$dst\t! regL to stkD" %}
10723   ins_encode %{
10724     Register Rsrc = $src$$Register;
10725     Register Rtmp = $tmp$$Register;
10726     Register Rtmp2 = $tmp2$$Register;
10727     __ sllx(Rsrc,    48, Rtmp);
10728     __ srlx(Rtmp,    16, Rtmp2);
10729     __ or3 (Rtmp, Rtmp2, Rtmp);
10730     __ srlx(Rtmp,    32, Rtmp2);
10731     __ or3 (Rtmp, Rtmp2, Rtmp);
10732     __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10733     __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10734   %}
10735   ins_pipe(ialu_reg);
10736 %}
10737 
10738 // Replicate scalar constant to packed char/short values in Double register
10739 instruct Repl4S_immI(regD dst, immI con, o7RegI tmp) %{
10740   predicate(n->as_Vector()->length() == 4);
10741   match(Set dst (ReplicateS con));
10742   effect(KILL tmp);
10743   format %{ "LDDF   [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl4S($con)" %}
10744   ins_encode %{
10745     // XXX This is a quick fix for 6833573.
10746     //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 4, 2)), $dst$$FloatRegister);
10747     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 4, 2)), $tmp$$Register);
10748     __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10749   %}
10750   ins_pipe(loadConFD);
10751 %}
10752 
10753 // Replicate scalar to packed int values into Double register
10754 instruct Repl2I_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10755   predicate(n->as_Vector()->length() == 2 && UseVIS >= 3);
10756   match(Set dst (ReplicateI src));
10757   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10758   format %{ "SLLX  $src,32,$tmp\n\t"
10759             "SRLX  $tmp,32,$tmp2\n\t"
10760             "OR    $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10761             "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10762   ins_encode %{
10763     Register Rsrc = $src$$Register;
10764     Register Rtmp = $tmp$$Register;
10765     Register Rtmp2 = $tmp2$$Register;
10766     __ sllx(Rsrc,    32, Rtmp);
10767     __ srlx(Rtmp,    32, Rtmp2);
10768     __ or3 (Rtmp, Rtmp2, Rtmp);
10769     __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10770   %}
10771   ins_pipe(ialu_reg);
10772 %}
10773 
10774 // Replicate scalar to packed int values into Double stack
10775 instruct Repl2I_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10776   predicate(n->as_Vector()->length() == 2 && UseVIS < 3);
10777   match(Set dst (ReplicateI src));
10778   effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10779   format %{ "SLLX  $src,32,$tmp\n\t"
10780             "SRLX  $tmp,32,$tmp2\n\t"
10781             "OR    $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10782             "STX   $tmp,$dst\t! regL to stkD" %}
10783   ins_encode %{
10784     Register Rsrc = $src$$Register;
10785     Register Rtmp = $tmp$$Register;
10786     Register Rtmp2 = $tmp2$$Register;
10787     __ sllx(Rsrc,    32, Rtmp);
10788     __ srlx(Rtmp,    32, Rtmp2);
10789     __ or3 (Rtmp, Rtmp2, Rtmp);
10790     __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10791     __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10792   %}
10793   ins_pipe(ialu_reg);
10794 %}
10795 
10796 // Replicate scalar zero constant to packed int values in Double register
10797 instruct Repl2I_immI(regD dst, immI con, o7RegI tmp) %{
10798   predicate(n->as_Vector()->length() == 2);
10799   match(Set dst (ReplicateI con));
10800   effect(KILL tmp);
10801   format %{ "LDDF   [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2I($con)" %}
10802   ins_encode %{
10803     // XXX This is a quick fix for 6833573.
10804     //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 2, 4)), $dst$$FloatRegister);
10805     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 2, 4)), $tmp$$Register);
10806     __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10807   %}
10808   ins_pipe(loadConFD);
10809 %}
10810 
10811 // Replicate scalar to packed float values into Double stack
10812 instruct Repl2F_stk(stackSlotD dst, regF src) %{
10813   predicate(n->as_Vector()->length() == 2);
10814   match(Set dst (ReplicateF src));
10815   ins_cost(MEMORY_REF_COST*2);
10816   format %{ "STF    $src,$dst.hi\t! packed2F\n\t"
10817             "STF    $src,$dst.lo" %}
10818   opcode(Assembler::stf_op3);
10819   ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, src));
10820   ins_pipe(fstoreF_stk_reg);
10821 %}
10822 
10823 // Replicate scalar zero constant to packed float values in Double register
10824 instruct Repl2F_immF(regD dst, immF con, o7RegI tmp) %{
10825   predicate(n->as_Vector()->length() == 2);
10826   match(Set dst (ReplicateF con));
10827   effect(KILL tmp);
10828   format %{ "LDDF   [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2F($con)" %}
10829   ins_encode %{
10830     // XXX This is a quick fix for 6833573.
10831     //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immF($con$$constant)), $dst$$FloatRegister);
10832     RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immF($con$$constant)), $tmp$$Register);
10833     __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10834   %}
10835   ins_pipe(loadConFD);
10836 %}
10837 
10838 //----------PEEPHOLE RULES-----------------------------------------------------
10839 // These must follow all instruction definitions as they use the names
10840 // defined in the instructions definitions.
10841 //
10842 // peepmatch ( root_instr_name [preceding_instruction]* );
10843 //
10844 // peepconstraint %{
10845 // (instruction_number.operand_name relational_op instruction_number.operand_name
10846 //  [, ...] );
10847 // // instruction numbers are zero-based using left to right order in peepmatch
10848 //
10849 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
10850 // // provide an instruction_number.operand_name for each operand that appears
10851 // // in the replacement instruction's match rule
10852 //
10853 // ---------VM FLAGS---------------------------------------------------------
10854 //
10855 // All peephole optimizations can be turned off using -XX:-OptoPeephole
10856 //
10857 // Each peephole rule is given an identifying number starting with zero and
10858 // increasing by one in the order seen by the parser.  An individual peephole
10859 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
10860 // on the command-line.
10861 //
10862 // ---------CURRENT LIMITATIONS----------------------------------------------
10863 //
10864 // Only match adjacent instructions in same basic block
10865 // Only equality constraints
10866 // Only constraints between operands, not (0.dest_reg == EAX_enc)
10867 // Only one replacement instruction
10868 //
10869 // ---------EXAMPLE----------------------------------------------------------
10870 //
10871 // // pertinent parts of existing instructions in architecture description
10872 // instruct movI(eRegI dst, eRegI src) %{
10873 //   match(Set dst (CopyI src));
10874 // %}
10875 //
10876 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
10877 //   match(Set dst (AddI dst src));
10878 //   effect(KILL cr);
10879 // %}
10880 //
10881 // // Change (inc mov) to lea
10882 // peephole %{
10883 //   // increment preceeded by register-register move
10884 //   peepmatch ( incI_eReg movI );
10885 //   // require that the destination register of the increment
10886 //   // match the destination register of the move
10887 //   peepconstraint ( 0.dst == 1.dst );
10888 //   // construct a replacement instruction that sets
10889 //   // the destination to ( move's source register + one )
10890 //   peepreplace ( incI_eReg_immI1( 0.dst 1.src 0.src ) );
10891 // %}
10892 //
10893 
10894 // // Change load of spilled value to only a spill
10895 // instruct storeI(memory mem, eRegI src) %{
10896 //   match(Set mem (StoreI mem src));
10897 // %}
10898 //
10899 // instruct loadI(eRegI dst, memory mem) %{
10900 //   match(Set dst (LoadI mem));
10901 // %}
10902 //
10903 // peephole %{
10904 //   peepmatch ( loadI storeI );
10905 //   peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
10906 //   peepreplace ( storeI( 1.mem 1.mem 1.src ) );
10907 // %}
10908 
10909 //----------SMARTSPILL RULES---------------------------------------------------
10910 // These must follow all instruction definitions as they use the names
10911 // defined in the instructions definitions.
10912 //
10913 // SPARC will probably not have any of these rules due to RISC instruction set.
10914 
10915 //----------PIPELINE-----------------------------------------------------------
10916 // Rules which define the behavior of the target architectures pipeline.