1 //
   2 // Copyright (c) 1997, 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 // X86 Architecture Description File
  26 
  27 //----------REGISTER DEFINITION BLOCK------------------------------------------
  28 // This information is used by the matcher and the register allocator to
  29 // describe individual registers and classes of registers within the target
  30 // archtecture.
  31 
  32 register %{
  33 //----------Architecture Description Register Definitions----------------------
  34 // General Registers
  35 // "reg_def"  name ( register save type, C convention save type,
  36 //                   ideal register type, encoding );
  37 // Register Save Types:
  38 //
  39 // NS  = No-Save:       The register allocator assumes that these registers
  40 //                      can be used without saving upon entry to the method, &
  41 //                      that they do not need to be saved at call sites.
  42 //
  43 // SOC = Save-On-Call:  The register allocator assumes that these registers
  44 //                      can be used without saving upon entry to the method,
  45 //                      but that they must be saved at call sites.
  46 //
  47 // SOE = Save-On-Entry: The register allocator assumes that these registers
  48 //                      must be saved before using them upon entry to the
  49 //                      method, but they do not need to be saved at call
  50 //                      sites.
  51 //
  52 // AS  = Always-Save:   The register allocator assumes that these registers
  53 //                      must be saved before using them upon entry to the
  54 //                      method, & that they must be saved at call sites.
  55 //
  56 // Ideal Register Type is used to determine how to save & restore a
  57 // register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
  58 // spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
  59 //
  60 // The encoding number is the actual bit-pattern placed into the opcodes.
  61 
  62 // General Registers
  63 // Previously set EBX, ESI, and EDI as save-on-entry for java code
  64 // Turn off SOE in java-code due to frequent use of uncommon-traps.
  65 // Now that allocator is better, turn on ESI and EDI as SOE registers.
  66 
  67 reg_def EBX(SOC, SOE, Op_RegI, 3, rbx->as_VMReg());
  68 reg_def ECX(SOC, SOC, Op_RegI, 1, rcx->as_VMReg());
  69 reg_def ESI(SOC, SOE, Op_RegI, 6, rsi->as_VMReg());
  70 reg_def EDI(SOC, SOE, Op_RegI, 7, rdi->as_VMReg());
  71 // now that adapter frames are gone EBP is always saved and restored by the prolog/epilog code
  72 reg_def EBP(NS, SOE, Op_RegI, 5, rbp->as_VMReg());
  73 reg_def EDX(SOC, SOC, Op_RegI, 2, rdx->as_VMReg());
  74 reg_def EAX(SOC, SOC, Op_RegI, 0, rax->as_VMReg());
  75 reg_def ESP( NS,  NS, Op_RegI, 4, rsp->as_VMReg());
  76 
  77 // Float registers.  We treat TOS/FPR0 special.  It is invisible to the
  78 // allocator, and only shows up in the encodings.
  79 reg_def FPR0L( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
  80 reg_def FPR0H( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
  81 // Ok so here's the trick FPR1 is really st(0) except in the midst
  82 // of emission of assembly for a machnode. During the emission the fpu stack
  83 // is pushed making FPR1 == st(1) temporarily. However at any safepoint
  84 // the stack will not have this element so FPR1 == st(0) from the
  85 // oopMap viewpoint. This same weirdness with numbering causes
  86 // instruction encoding to have to play games with the register
  87 // encode to correct for this 0/1 issue. See MachSpillCopyNode::implementation
  88 // where it does flt->flt moves to see an example
  89 //
  90 reg_def FPR1L( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg());
  91 reg_def FPR1H( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg()->next());
  92 reg_def FPR2L( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg());
  93 reg_def FPR2H( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg()->next());
  94 reg_def FPR3L( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg());
  95 reg_def FPR3H( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg()->next());
  96 reg_def FPR4L( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg());
  97 reg_def FPR4H( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg()->next());
  98 reg_def FPR5L( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg());
  99 reg_def FPR5H( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg()->next());
 100 reg_def FPR6L( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg());
 101 reg_def FPR6H( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg()->next());
 102 reg_def FPR7L( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg());
 103 reg_def FPR7H( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg()->next());
 104 
 105 // Specify priority of register selection within phases of register
 106 // allocation.  Highest priority is first.  A useful heuristic is to
 107 // give registers a low priority when they are required by machine
 108 // instructions, like EAX and EDX.  Registers which are used as
 109 // pairs must fall on an even boundary (witness the FPR#L's in this list).
 110 // For the Intel integer registers, the equivalent Long pairs are
 111 // EDX:EAX, EBX:ECX, and EDI:EBP.
 112 alloc_class chunk0( ECX,   EBX,   EBP,   EDI,   EAX,   EDX,   ESI, ESP,
 113                     FPR0L, FPR0H, FPR1L, FPR1H, FPR2L, FPR2H,
 114                     FPR3L, FPR3H, FPR4L, FPR4H, FPR5L, FPR5H,
 115                     FPR6L, FPR6H, FPR7L, FPR7H );
 116 
 117 
 118 //----------Architecture Description Register Classes--------------------------
 119 // Several register classes are automatically defined based upon information in
 120 // this architecture description.
 121 // 1) reg_class inline_cache_reg           ( /* as def'd in frame section */ )
 122 // 2) reg_class compiler_method_oop_reg    ( /* as def'd in frame section */ )
 123 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
 124 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
 125 //
 126 // Class for all registers
 127 reg_class any_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX, ESP);
 128 // Class for general registers
 129 reg_class int_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX);
 130 // Class for general registers which may be used for implicit null checks on win95
 131 // Also safe for use by tailjump. We don't want to allocate in rbp,
 132 reg_class int_reg_no_rbp(EAX, EDX, EDI, ESI, ECX, EBX);
 133 // Class of "X" registers
 134 reg_class int_x_reg(EBX, ECX, EDX, EAX);
 135 // Class of registers that can appear in an address with no offset.
 136 // EBP and ESP require an extra instruction byte for zero offset.
 137 // Used in fast-unlock
 138 reg_class p_reg(EDX, EDI, ESI, EBX);
 139 // Class for general registers not including ECX
 140 reg_class ncx_reg(EAX, EDX, EBP, EDI, ESI, EBX);
 141 // Class for general registers not including EAX
 142 reg_class nax_reg(EDX, EDI, ESI, ECX, EBX);
 143 // Class for general registers not including EAX or EBX.
 144 reg_class nabx_reg(EDX, EDI, ESI, ECX, EBP);
 145 // Class of EAX (for multiply and divide operations)
 146 reg_class eax_reg(EAX);
 147 // Class of EBX (for atomic add)
 148 reg_class ebx_reg(EBX);
 149 // Class of ECX (for shift and JCXZ operations and cmpLTMask)
 150 reg_class ecx_reg(ECX);
 151 // Class of EDX (for multiply and divide operations)
 152 reg_class edx_reg(EDX);
 153 // Class of EDI (for synchronization)
 154 reg_class edi_reg(EDI);
 155 // Class of ESI (for synchronization)
 156 reg_class esi_reg(ESI);
 157 // Singleton class for interpreter's stack pointer
 158 reg_class ebp_reg(EBP);
 159 // Singleton class for stack pointer
 160 reg_class sp_reg(ESP);
 161 // Singleton class for instruction pointer
 162 // reg_class ip_reg(EIP);
 163 // Class of integer register pairs
 164 reg_class long_reg( EAX,EDX, ECX,EBX, EBP,EDI );
 165 // Class of integer register pairs that aligns with calling convention
 166 reg_class eadx_reg( EAX,EDX );
 167 reg_class ebcx_reg( ECX,EBX );
 168 // Not AX or DX, used in divides
 169 reg_class nadx_reg( EBX,ECX,ESI,EDI,EBP );
 170 
 171 // Floating point registers.  Notice FPR0 is not a choice.
 172 // FPR0 is not ever allocated; we use clever encodings to fake
 173 // a 2-address instructions out of Intels FP stack.
 174 reg_class fp_flt_reg( FPR1L,FPR2L,FPR3L,FPR4L,FPR5L,FPR6L,FPR7L );
 175 
 176 reg_class fp_dbl_reg( FPR1L,FPR1H, FPR2L,FPR2H, FPR3L,FPR3H,
 177                       FPR4L,FPR4H, FPR5L,FPR5H, FPR6L,FPR6H,
 178                       FPR7L,FPR7H );
 179 
 180 reg_class fp_flt_reg0( FPR1L );
 181 reg_class fp_dbl_reg0( FPR1L,FPR1H );
 182 reg_class fp_dbl_reg1( FPR2L,FPR2H );
 183 reg_class fp_dbl_notreg0( FPR2L,FPR2H, FPR3L,FPR3H, FPR4L,FPR4H,
 184                           FPR5L,FPR5H, FPR6L,FPR6H, FPR7L,FPR7H );
 185 
 186 %}
 187 
 188 
 189 //----------SOURCE BLOCK-------------------------------------------------------
 190 // This is a block of C++ code which provides values, functions, and
 191 // definitions necessary in the rest of the architecture description
 192 source_hpp %{
 193 // Must be visible to the DFA in dfa_x86_32.cpp
 194 extern bool is_operand_hi32_zero(Node* n);
 195 %}
 196 
 197 source %{
 198 #define   RELOC_IMM32    Assembler::imm_operand
 199 #define   RELOC_DISP32   Assembler::disp32_operand
 200 
 201 #define __ _masm.
 202 
 203 // How to find the high register of a Long pair, given the low register
 204 #define   HIGH_FROM_LOW(x) ((x)+2)
 205 
 206 // These masks are used to provide 128-bit aligned bitmasks to the XMM
 207 // instructions, to allow sign-masking or sign-bit flipping.  They allow
 208 // fast versions of NegF/NegD and AbsF/AbsD.
 209 
 210 // Note: 'double' and 'long long' have 32-bits alignment on x86.
 211 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
 212   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
 213   // of 128-bits operands for SSE instructions.
 214   jlong *operand = (jlong*)(((uintptr_t)adr)&((uintptr_t)(~0xF)));
 215   // Store the value to a 128-bits operand.
 216   operand[0] = lo;
 217   operand[1] = hi;
 218   return operand;
 219 }
 220 
 221 // Buffer for 128-bits masks used by SSE instructions.
 222 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
 223 
 224 // Static initialization during VM startup.
 225 static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
 226 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
 227 static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
 228 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
 229 
 230 // Offset hacking within calls.
 231 static int pre_call_resets_size() {
 232   int size = 0;
 233   Compile* C = Compile::current();
 234   if (C->in_24_bit_fp_mode()) {
 235     size += 6; // fldcw
 236   }
 237   if (C->max_vector_size() > 16) {
 238     size += 3; // vzeroupper
 239   }
 240   return size;
 241 }
 242 
 243 static int preserve_SP_size() {
 244   return 2;  // op, rm(reg/reg)
 245 }
 246 
 247 // !!!!! Special hack to get all type of calls to specify the byte offset
 248 //       from the start of the call to the point where the return address
 249 //       will point.
 250 int MachCallStaticJavaNode::ret_addr_offset() {
 251   int offset = 5 + pre_call_resets_size();  // 5 bytes from start of call to where return address points
 252   if (_method_handle_invoke)
 253     offset += preserve_SP_size();
 254   return offset;
 255 }
 256 
 257 int MachCallDynamicJavaNode::ret_addr_offset() {
 258   return 10 + pre_call_resets_size();  // 10 bytes from start of call to where return address points
 259 }
 260 
 261 static int sizeof_FFree_Float_Stack_All = -1;
 262 
 263 int MachCallRuntimeNode::ret_addr_offset() {
 264   assert(sizeof_FFree_Float_Stack_All != -1, "must have been emitted already");
 265   return sizeof_FFree_Float_Stack_All + 5 + pre_call_resets_size();
 266 }
 267 
 268 // Indicate if the safepoint node needs the polling page as an input.
 269 // Since x86 does have absolute addressing, it doesn't.
 270 bool SafePointNode::needs_polling_address_input() {
 271   return false;
 272 }
 273 
 274 //
 275 // Compute padding required for nodes which need alignment
 276 //
 277 
 278 // The address of the call instruction needs to be 4-byte aligned to
 279 // ensure that it does not span a cache line so that it can be patched.
 280 int CallStaticJavaDirectNode::compute_padding(int current_offset) const {
 281   current_offset += pre_call_resets_size();  // skip fldcw, if any
 282   current_offset += 1;      // skip call opcode byte
 283   return round_to(current_offset, alignment_required()) - current_offset;
 284 }
 285 
 286 // The address of the call instruction needs to be 4-byte aligned to
 287 // ensure that it does not span a cache line so that it can be patched.
 288 int CallStaticJavaHandleNode::compute_padding(int current_offset) const {
 289   current_offset += pre_call_resets_size();  // skip fldcw, if any
 290   current_offset += preserve_SP_size();   // skip mov rbp, rsp
 291   current_offset += 1;      // skip call opcode byte
 292   return round_to(current_offset, alignment_required()) - current_offset;
 293 }
 294 
 295 // The address of the call instruction needs to be 4-byte aligned to
 296 // ensure that it does not span a cache line so that it can be patched.
 297 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const {
 298   current_offset += pre_call_resets_size();  // skip fldcw, if any
 299   current_offset += 5;      // skip MOV instruction
 300   current_offset += 1;      // skip call opcode byte
 301   return round_to(current_offset, alignment_required()) - current_offset;
 302 }
 303 
 304 // EMIT_RM()
 305 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) {
 306   unsigned char c = (unsigned char)((f1 << 6) | (f2 << 3) | f3);
 307   cbuf.insts()->emit_int8(c);
 308 }
 309 
 310 // EMIT_CC()
 311 void emit_cc(CodeBuffer &cbuf, int f1, int f2) {
 312   unsigned char c = (unsigned char)( f1 | f2 );
 313   cbuf.insts()->emit_int8(c);
 314 }
 315 
 316 // EMIT_OPCODE()
 317 void emit_opcode(CodeBuffer &cbuf, int code) {
 318   cbuf.insts()->emit_int8((unsigned char) code);
 319 }
 320 
 321 // EMIT_OPCODE() w/ relocation information
 322 void emit_opcode(CodeBuffer &cbuf, int code, relocInfo::relocType reloc, int offset = 0) {
 323   cbuf.relocate(cbuf.insts_mark() + offset, reloc);
 324   emit_opcode(cbuf, code);
 325 }
 326 
 327 // EMIT_D8()
 328 void emit_d8(CodeBuffer &cbuf, int d8) {
 329   cbuf.insts()->emit_int8((unsigned char) d8);
 330 }
 331 
 332 // EMIT_D16()
 333 void emit_d16(CodeBuffer &cbuf, int d16) {
 334   cbuf.insts()->emit_int16(d16);
 335 }
 336 
 337 // EMIT_D32()
 338 void emit_d32(CodeBuffer &cbuf, int d32) {
 339   cbuf.insts()->emit_int32(d32);
 340 }
 341 
 342 // emit 32 bit value and construct relocation entry from relocInfo::relocType
 343 void emit_d32_reloc(CodeBuffer &cbuf, int d32, relocInfo::relocType reloc,
 344         int format) {
 345   cbuf.relocate(cbuf.insts_mark(), reloc, format);
 346   cbuf.insts()->emit_int32(d32);
 347 }
 348 
 349 // emit 32 bit value and construct relocation entry from RelocationHolder
 350 void emit_d32_reloc(CodeBuffer &cbuf, int d32, RelocationHolder const& rspec,
 351         int format) {
 352 #ifdef ASSERT
 353   if (rspec.reloc()->type() == relocInfo::oop_type && d32 != 0 && d32 != (int)Universe::non_oop_word()) {
 354     assert(cast_to_oop(d32)->is_oop() && (ScavengeRootsInCode || !cast_to_oop(d32)->is_scavengable()), "cannot embed scavengable oops in code");
 355   }
 356 #endif
 357   cbuf.relocate(cbuf.insts_mark(), rspec, format);
 358   cbuf.insts()->emit_int32(d32);
 359 }
 360 
 361 // Access stack slot for load or store
 362 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp) {
 363   emit_opcode( cbuf, opcode );               // (e.g., FILD   [ESP+src])
 364   if( -128 <= disp && disp <= 127 ) {
 365     emit_rm( cbuf, 0x01, rm_field, ESP_enc );  // R/M byte
 366     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
 367     emit_d8 (cbuf, disp);     // Displacement  // R/M byte
 368   } else {
 369     emit_rm( cbuf, 0x02, rm_field, ESP_enc );  // R/M byte
 370     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
 371     emit_d32(cbuf, disp);     // Displacement  // R/M byte
 372   }
 373 }
 374 
 375    // rRegI ereg, memory mem) %{    // emit_reg_mem
 376 void encode_RegMem( CodeBuffer &cbuf, int reg_encoding, int base, int index, int scale, int displace, relocInfo::relocType disp_reloc ) {
 377   // There is no index & no scale, use form without SIB byte
 378   if ((index == 0x4) &&
 379       (scale == 0) && (base != ESP_enc)) {
 380     // If no displacement, mode is 0x0; unless base is [EBP]
 381     if ( (displace == 0) && (base != EBP_enc) ) {
 382       emit_rm(cbuf, 0x0, reg_encoding, base);
 383     }
 384     else {                    // If 8-bit displacement, mode 0x1
 385       if ((displace >= -128) && (displace <= 127)
 386           && (disp_reloc == relocInfo::none) ) {
 387         emit_rm(cbuf, 0x1, reg_encoding, base);
 388         emit_d8(cbuf, displace);
 389       }
 390       else {                  // If 32-bit displacement
 391         if (base == -1) { // Special flag for absolute address
 392           emit_rm(cbuf, 0x0, reg_encoding, 0x5);
 393           // (manual lies; no SIB needed here)
 394           if ( disp_reloc != relocInfo::none ) {
 395             emit_d32_reloc(cbuf, displace, disp_reloc, 1);
 396           } else {
 397             emit_d32      (cbuf, displace);
 398           }
 399         }
 400         else {                // Normal base + offset
 401           emit_rm(cbuf, 0x2, reg_encoding, base);
 402           if ( disp_reloc != relocInfo::none ) {
 403             emit_d32_reloc(cbuf, displace, disp_reloc, 1);
 404           } else {
 405             emit_d32      (cbuf, displace);
 406           }
 407         }
 408       }
 409     }
 410   }
 411   else {                      // Else, encode with the SIB byte
 412     // If no displacement, mode is 0x0; unless base is [EBP]
 413     if (displace == 0 && (base != EBP_enc)) {  // If no displacement
 414       emit_rm(cbuf, 0x0, reg_encoding, 0x4);
 415       emit_rm(cbuf, scale, index, base);
 416     }
 417     else {                    // If 8-bit displacement, mode 0x1
 418       if ((displace >= -128) && (displace <= 127)
 419           && (disp_reloc == relocInfo::none) ) {
 420         emit_rm(cbuf, 0x1, reg_encoding, 0x4);
 421         emit_rm(cbuf, scale, index, base);
 422         emit_d8(cbuf, displace);
 423       }
 424       else {                  // If 32-bit displacement
 425         if (base == 0x04 ) {
 426           emit_rm(cbuf, 0x2, reg_encoding, 0x4);
 427           emit_rm(cbuf, scale, index, 0x04);
 428         } else {
 429           emit_rm(cbuf, 0x2, reg_encoding, 0x4);
 430           emit_rm(cbuf, scale, index, base);
 431         }
 432         if ( disp_reloc != relocInfo::none ) {
 433           emit_d32_reloc(cbuf, displace, disp_reloc, 1);
 434         } else {
 435           emit_d32      (cbuf, displace);
 436         }
 437       }
 438     }
 439   }
 440 }
 441 
 442 
 443 void encode_Copy( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
 444   if( dst_encoding == src_encoding ) {
 445     // reg-reg copy, use an empty encoding
 446   } else {
 447     emit_opcode( cbuf, 0x8B );
 448     emit_rm(cbuf, 0x3, dst_encoding, src_encoding );
 449   }
 450 }
 451 
 452 void emit_cmpfp_fixup(MacroAssembler& _masm) {
 453   Label exit;
 454   __ jccb(Assembler::noParity, exit);
 455   __ pushf();
 456   //
 457   // comiss/ucomiss instructions set ZF,PF,CF flags and
 458   // zero OF,AF,SF for NaN values.
 459   // Fixup flags by zeroing ZF,PF so that compare of NaN
 460   // values returns 'less than' result (CF is set).
 461   // Leave the rest of flags unchanged.
 462   //
 463   //    7 6 5 4 3 2 1 0
 464   //   |S|Z|r|A|r|P|r|C|  (r - reserved bit)
 465   //    0 0 1 0 1 0 1 1   (0x2B)
 466   //
 467   __ andl(Address(rsp, 0), 0xffffff2b);
 468   __ popf();
 469   __ bind(exit);
 470 }
 471 
 472 void emit_cmpfp3(MacroAssembler& _masm, Register dst) {
 473   Label done;
 474   __ movl(dst, -1);
 475   __ jcc(Assembler::parity, done);
 476   __ jcc(Assembler::below, done);
 477   __ setb(Assembler::notEqual, dst);
 478   __ movzbl(dst, dst);
 479   __ bind(done);
 480 }
 481 
 482 
 483 //=============================================================================
 484 const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
 485 
 486 int Compile::ConstantTable::calculate_table_base_offset() const {
 487   return 0;  // absolute addressing, no offset
 488 }
 489 
 490 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
 491   // Empty encoding
 492 }
 493 
 494 uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
 495   return 0;
 496 }
 497 
 498 #ifndef PRODUCT
 499 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
 500   st->print("# MachConstantBaseNode (empty encoding)");
 501 }
 502 #endif
 503 
 504 
 505 //=============================================================================
 506 #ifndef PRODUCT
 507 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
 508   Compile* C = ra_->C;
 509 
 510   int framesize = C->frame_slots() << LogBytesPerInt;
 511   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 512   // Remove wordSize for return addr which is already pushed.
 513   framesize -= wordSize;
 514 
 515   if (C->need_stack_bang(framesize)) {
 516     framesize -= wordSize;
 517     st->print("# stack bang");
 518     st->print("\n\t");
 519     st->print("PUSH   EBP\t# Save EBP");
 520     if (framesize) {
 521       st->print("\n\t");
 522       st->print("SUB    ESP, #%d\t# Create frame",framesize);
 523     }
 524   } else {
 525     st->print("SUB    ESP, #%d\t# Create frame",framesize);
 526     st->print("\n\t");
 527     framesize -= wordSize;
 528     st->print("MOV    [ESP + #%d], EBP\t# Save EBP",framesize);
 529   }
 530 
 531   if (VerifyStackAtCalls) {
 532     st->print("\n\t");
 533     framesize -= wordSize;
 534     st->print("MOV    [ESP + #%d], 0xBADB100D\t# Majik cookie for stack depth check",framesize);
 535   }
 536 
 537   if( C->in_24_bit_fp_mode() ) {
 538     st->print("\n\t");
 539     st->print("FLDCW  \t# load 24 bit fpu control word");
 540   }
 541   if (UseSSE >= 2 && VerifyFPU) {
 542     st->print("\n\t");
 543     st->print("# verify FPU stack (must be clean on entry)");
 544   }
 545 
 546 #ifdef ASSERT
 547   if (VerifyStackAtCalls) {
 548     st->print("\n\t");
 549     st->print("# stack alignment check");
 550   }
 551 #endif
 552   st->cr();
 553 }
 554 #endif
 555 
 556 
 557 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
 558   Compile* C = ra_->C;
 559   MacroAssembler _masm(&cbuf);
 560 
 561   int framesize = C->frame_slots() << LogBytesPerInt;
 562 
 563   __ verified_entry(framesize, C->need_stack_bang(framesize), C->in_24_bit_fp_mode());
 564 
 565   C->set_frame_complete(cbuf.insts_size());
 566 
 567   if (C->has_mach_constant_base_node()) {
 568     // NOTE: We set the table base offset here because users might be
 569     // emitted before MachConstantBaseNode.
 570     Compile::ConstantTable& constant_table = C->constant_table();
 571     constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
 572   }
 573 }
 574 
 575 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
 576   return MachNode::size(ra_); // too many variables; just compute it the hard way
 577 }
 578 
 579 int MachPrologNode::reloc() const {
 580   return 0; // a large enough number
 581 }
 582 
 583 //=============================================================================
 584 #ifndef PRODUCT
 585 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
 586   Compile *C = ra_->C;
 587   int framesize = C->frame_slots() << LogBytesPerInt;
 588   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 589   // Remove two words for return addr and rbp,
 590   framesize -= 2*wordSize;
 591 
 592   if (C->max_vector_size() > 16) {
 593     st->print("VZEROUPPER");
 594     st->cr(); st->print("\t");
 595   }
 596   if (C->in_24_bit_fp_mode()) {
 597     st->print("FLDCW  standard control word");
 598     st->cr(); st->print("\t");
 599   }
 600   if (framesize) {
 601     st->print("ADD    ESP,%d\t# Destroy frame",framesize);
 602     st->cr(); st->print("\t");
 603   }
 604   st->print_cr("POPL   EBP"); st->print("\t");
 605   if (do_polling() && C->is_method_compilation()) {
 606     st->print("TEST   PollPage,EAX\t! Poll Safepoint");
 607     st->cr(); st->print("\t");
 608   }
 609 }
 610 #endif
 611 
 612 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
 613   Compile *C = ra_->C;
 614 
 615   if (C->max_vector_size() > 16) {
 616     // Clear upper bits of YMM registers when current compiled code uses
 617     // wide vectors to avoid AVX <-> SSE transition penalty during call.
 618     MacroAssembler masm(&cbuf);
 619     masm.vzeroupper();
 620   }
 621   // If method set FPU control word, restore to standard control word
 622   if (C->in_24_bit_fp_mode()) {
 623     MacroAssembler masm(&cbuf);
 624     masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 625   }
 626 
 627   int framesize = C->frame_slots() << LogBytesPerInt;
 628   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 629   // Remove two words for return addr and rbp,
 630   framesize -= 2*wordSize;
 631 
 632   // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
 633 
 634   if (framesize >= 128) {
 635     emit_opcode(cbuf, 0x81); // add  SP, #framesize
 636     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
 637     emit_d32(cbuf, framesize);
 638   } else if (framesize) {
 639     emit_opcode(cbuf, 0x83); // add  SP, #framesize
 640     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
 641     emit_d8(cbuf, framesize);
 642   }
 643 
 644   emit_opcode(cbuf, 0x58 | EBP_enc);
 645 
 646   if (do_polling() && C->is_method_compilation()) {
 647     cbuf.relocate(cbuf.insts_end(), relocInfo::poll_return_type, 0);
 648     emit_opcode(cbuf,0x85);
 649     emit_rm(cbuf, 0x0, EAX_enc, 0x5); // EAX
 650     emit_d32(cbuf, (intptr_t)os::get_polling_page());
 651   }
 652 }
 653 
 654 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
 655   Compile *C = ra_->C;
 656   // If method set FPU control word, restore to standard control word
 657   int size = C->in_24_bit_fp_mode() ? 6 : 0;
 658   if (C->max_vector_size() > 16) size += 3; // vzeroupper
 659   if (do_polling() && C->is_method_compilation()) size += 6;
 660 
 661   int framesize = C->frame_slots() << LogBytesPerInt;
 662   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 663   // Remove two words for return addr and rbp,
 664   framesize -= 2*wordSize;
 665 
 666   size++; // popl rbp,
 667 
 668   if (framesize >= 128) {
 669     size += 6;
 670   } else {
 671     size += framesize ? 3 : 0;
 672   }
 673   return size;
 674 }
 675 
 676 int MachEpilogNode::reloc() const {
 677   return 0; // a large enough number
 678 }
 679 
 680 const Pipeline * MachEpilogNode::pipeline() const {
 681   return MachNode::pipeline_class();
 682 }
 683 
 684 int MachEpilogNode::safepoint_offset() const { return 0; }
 685 
 686 //=============================================================================
 687 
 688 enum RC { rc_bad, rc_int, rc_float, rc_xmm, rc_stack };
 689 static enum RC rc_class( OptoReg::Name reg ) {
 690 
 691   if( !OptoReg::is_valid(reg)  ) return rc_bad;
 692   if (OptoReg::is_stack(reg)) return rc_stack;
 693 
 694   VMReg r = OptoReg::as_VMReg(reg);
 695   if (r->is_Register()) return rc_int;
 696   if (r->is_FloatRegister()) {
 697     assert(UseSSE < 2, "shouldn't be used in SSE2+ mode");
 698     return rc_float;
 699   }
 700   assert(r->is_XMMRegister(), "must be");
 701   return rc_xmm;
 702 }
 703 
 704 static int impl_helper( CodeBuffer *cbuf, bool do_size, bool is_load, int offset, int reg,
 705                         int opcode, const char *op_str, int size, outputStream* st ) {
 706   if( cbuf ) {
 707     emit_opcode  (*cbuf, opcode );
 708     encode_RegMem(*cbuf, Matcher::_regEncode[reg], ESP_enc, 0x4, 0, offset, relocInfo::none);
 709 #ifndef PRODUCT
 710   } else if( !do_size ) {
 711     if( size != 0 ) st->print("\n\t");
 712     if( opcode == 0x8B || opcode == 0x89 ) { // MOV
 713       if( is_load ) st->print("%s   %s,[ESP + #%d]",op_str,Matcher::regName[reg],offset);
 714       else          st->print("%s   [ESP + #%d],%s",op_str,offset,Matcher::regName[reg]);
 715     } else { // FLD, FST, PUSH, POP
 716       st->print("%s [ESP + #%d]",op_str,offset);
 717     }
 718 #endif
 719   }
 720   int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
 721   return size+3+offset_size;
 722 }
 723 
 724 // Helper for XMM registers.  Extra opcode bits, limited syntax.
 725 static int impl_x_helper( CodeBuffer *cbuf, bool do_size, bool is_load,
 726                          int offset, int reg_lo, int reg_hi, int size, outputStream* st ) {
 727   if (cbuf) {
 728     MacroAssembler _masm(cbuf);
 729     if (reg_lo+1 == reg_hi) { // double move?
 730       if (is_load) {
 731         __ movdbl(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
 732       } else {
 733         __ movdbl(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
 734       }
 735     } else {
 736       if (is_load) {
 737         __ movflt(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
 738       } else {
 739         __ movflt(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
 740       }
 741     }
 742 #ifndef PRODUCT
 743   } else if (!do_size) {
 744     if (size != 0) st->print("\n\t");
 745     if (reg_lo+1 == reg_hi) { // double move?
 746       if (is_load) st->print("%s %s,[ESP + #%d]",
 747                               UseXmmLoadAndClearUpper ? "MOVSD " : "MOVLPD",
 748                               Matcher::regName[reg_lo], offset);
 749       else         st->print("MOVSD  [ESP + #%d],%s",
 750                               offset, Matcher::regName[reg_lo]);
 751     } else {
 752       if (is_load) st->print("MOVSS  %s,[ESP + #%d]",
 753                               Matcher::regName[reg_lo], offset);
 754       else         st->print("MOVSS  [ESP + #%d],%s",
 755                               offset, Matcher::regName[reg_lo]);
 756     }
 757 #endif
 758   }
 759   int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
 760   // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes as SIMD prefix.
 761   return size+5+offset_size;
 762 }
 763 
 764 
 765 static int impl_movx_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 766                             int src_hi, int dst_hi, int size, outputStream* st ) {
 767   if (cbuf) {
 768     MacroAssembler _masm(cbuf);
 769     if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
 770       __ movdbl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 771                 as_XMMRegister(Matcher::_regEncode[src_lo]));
 772     } else {
 773       __ movflt(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 774                 as_XMMRegister(Matcher::_regEncode[src_lo]));
 775     }
 776 #ifndef PRODUCT
 777   } else if (!do_size) {
 778     if (size != 0) st->print("\n\t");
 779     if (UseXmmRegToRegMoveAll) {//Use movaps,movapd to move between xmm registers
 780       if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
 781         st->print("MOVAPD %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 782       } else {
 783         st->print("MOVAPS %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 784       }
 785     } else {
 786       if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double move?
 787         st->print("MOVSD  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 788       } else {
 789         st->print("MOVSS  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 790       }
 791     }
 792 #endif
 793   }
 794   // VEX_2bytes prefix is used if UseAVX > 0, and it takes the same 2 bytes as SIMD prefix.
 795   // Only MOVAPS SSE prefix uses 1 byte.
 796   int sz = 4;
 797   if (!(src_lo+1 == src_hi && dst_lo+1 == dst_hi) &&
 798       UseXmmRegToRegMoveAll && (UseAVX == 0)) sz = 3;
 799   return size + sz;
 800 }
 801 
 802 static int impl_movgpr2x_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 803                             int src_hi, int dst_hi, int size, outputStream* st ) {
 804   // 32-bit
 805   if (cbuf) {
 806     MacroAssembler _masm(cbuf);
 807     __ movdl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 808              as_Register(Matcher::_regEncode[src_lo]));
 809 #ifndef PRODUCT
 810   } else if (!do_size) {
 811     st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
 812 #endif
 813   }
 814   return 4;
 815 }
 816 
 817 
 818 static int impl_movx2gpr_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 819                                  int src_hi, int dst_hi, int size, outputStream* st ) {
 820   // 32-bit
 821   if (cbuf) {
 822     MacroAssembler _masm(cbuf);
 823     __ movdl(as_Register(Matcher::_regEncode[dst_lo]),
 824              as_XMMRegister(Matcher::_regEncode[src_lo]));
 825 #ifndef PRODUCT
 826   } else if (!do_size) {
 827     st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
 828 #endif
 829   }
 830   return 4;
 831 }
 832 
 833 static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int size, outputStream* st ) {
 834   if( cbuf ) {
 835     emit_opcode(*cbuf, 0x8B );
 836     emit_rm    (*cbuf, 0x3, Matcher::_regEncode[dst], Matcher::_regEncode[src] );
 837 #ifndef PRODUCT
 838   } else if( !do_size ) {
 839     if( size != 0 ) st->print("\n\t");
 840     st->print("MOV    %s,%s",Matcher::regName[dst],Matcher::regName[src]);
 841 #endif
 842   }
 843   return size+2;
 844 }
 845 
 846 static int impl_fp_store_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int src_hi, int dst_lo, int dst_hi,
 847                                  int offset, int size, outputStream* st ) {
 848   if( src_lo != FPR1L_num ) {      // Move value to top of FP stack, if not already there
 849     if( cbuf ) {
 850       emit_opcode( *cbuf, 0xD9 );  // FLD (i.e., push it)
 851       emit_d8( *cbuf, 0xC0-1+Matcher::_regEncode[src_lo] );
 852 #ifndef PRODUCT
 853     } else if( !do_size ) {
 854       if( size != 0 ) st->print("\n\t");
 855       st->print("FLD    %s",Matcher::regName[src_lo]);
 856 #endif
 857     }
 858     size += 2;
 859   }
 860 
 861   int st_op = (src_lo != FPR1L_num) ? EBX_num /*store & pop*/ : EDX_num /*store no pop*/;
 862   const char *op_str;
 863   int op;
 864   if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double store?
 865     op_str = (src_lo != FPR1L_num) ? "FSTP_D" : "FST_D ";
 866     op = 0xDD;
 867   } else {                   // 32-bit store
 868     op_str = (src_lo != FPR1L_num) ? "FSTP_S" : "FST_S ";
 869     op = 0xD9;
 870     assert( !OptoReg::is_valid(src_hi) && !OptoReg::is_valid(dst_hi), "no non-adjacent float-stores" );
 871   }
 872 
 873   return impl_helper(cbuf,do_size,false,offset,st_op,op,op_str,size, st);
 874 }
 875 
 876 // Next two methods are shared by 32- and 64-bit VM. They are defined in x86.ad.
 877 static int vec_mov_helper(CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 878                           int src_hi, int dst_hi, uint ireg, outputStream* st);
 879 
 880 static int vec_spill_helper(CodeBuffer *cbuf, bool do_size, bool is_load,
 881                             int stack_offset, int reg, uint ireg, outputStream* st);
 882 
 883 static int vec_stack_to_stack_helper(CodeBuffer *cbuf, bool do_size, int src_offset,
 884                                      int dst_offset, uint ireg, outputStream* st) {
 885   int calc_size = 0;
 886   int src_offset_size = (src_offset == 0) ? 0 : ((src_offset < 0x80) ? 1 : 4);
 887   int dst_offset_size = (dst_offset == 0) ? 0 : ((dst_offset < 0x80) ? 1 : 4);
 888   switch (ireg) {
 889   case Op_VecS:
 890     calc_size = 3+src_offset_size + 3+dst_offset_size;
 891     break;
 892   case Op_VecD:
 893     calc_size = 3+src_offset_size + 3+dst_offset_size;
 894     src_offset += 4;
 895     dst_offset += 4;
 896     src_offset_size = (src_offset == 0) ? 0 : ((src_offset < 0x80) ? 1 : 4);
 897     dst_offset_size = (dst_offset == 0) ? 0 : ((dst_offset < 0x80) ? 1 : 4);
 898     calc_size += 3+src_offset_size + 3+dst_offset_size;
 899     break;
 900   case Op_VecX:
 901     calc_size = 6 + 6 + 5+src_offset_size + 5+dst_offset_size;
 902     break;
 903   case Op_VecY:
 904     calc_size = 6 + 6 + 5+src_offset_size + 5+dst_offset_size;
 905     break;
 906   default:
 907     ShouldNotReachHere();
 908   }
 909   if (cbuf) {
 910     MacroAssembler _masm(cbuf);
 911     int offset = __ offset();
 912     switch (ireg) {
 913     case Op_VecS:
 914       __ pushl(Address(rsp, src_offset));
 915       __ popl (Address(rsp, dst_offset));
 916       break;
 917     case Op_VecD:
 918       __ pushl(Address(rsp, src_offset));
 919       __ popl (Address(rsp, dst_offset));
 920       __ pushl(Address(rsp, src_offset+4));
 921       __ popl (Address(rsp, dst_offset+4));
 922       break;
 923     case Op_VecX:
 924       __ movdqu(Address(rsp, -16), xmm0);
 925       __ movdqu(xmm0, Address(rsp, src_offset));
 926       __ movdqu(Address(rsp, dst_offset), xmm0);
 927       __ movdqu(xmm0, Address(rsp, -16));
 928       break;
 929     case Op_VecY:
 930       __ vmovdqu(Address(rsp, -32), xmm0);
 931       __ vmovdqu(xmm0, Address(rsp, src_offset));
 932       __ vmovdqu(Address(rsp, dst_offset), xmm0);
 933       __ vmovdqu(xmm0, Address(rsp, -32));
 934       break;
 935     default:
 936       ShouldNotReachHere();
 937     }
 938     int size = __ offset() - offset;
 939     assert(size == calc_size, "incorrect size calculattion");
 940     return size;
 941 #ifndef PRODUCT
 942   } else if (!do_size) {
 943     switch (ireg) {
 944     case Op_VecS:
 945       st->print("pushl   [rsp + #%d]\t# 32-bit mem-mem spill\n\t"
 946                 "popl    [rsp + #%d]",
 947                 src_offset, dst_offset);
 948       break;
 949     case Op_VecD:
 950       st->print("pushl   [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
 951                 "popq    [rsp + #%d]\n\t"
 952                 "pushl   [rsp + #%d]\n\t"
 953                 "popq    [rsp + #%d]",
 954                 src_offset, dst_offset, src_offset+4, dst_offset+4);
 955       break;
 956      case Op_VecX:
 957       st->print("movdqu  [rsp - #16], xmm0\t# 128-bit mem-mem spill\n\t"
 958                 "movdqu  xmm0, [rsp + #%d]\n\t"
 959                 "movdqu  [rsp + #%d], xmm0\n\t"
 960                 "movdqu  xmm0, [rsp - #16]",
 961                 src_offset, dst_offset);
 962       break;
 963     case Op_VecY:
 964       st->print("vmovdqu [rsp - #32], xmm0\t# 256-bit mem-mem spill\n\t"
 965                 "vmovdqu xmm0, [rsp + #%d]\n\t"
 966                 "vmovdqu [rsp + #%d], xmm0\n\t"
 967                 "vmovdqu xmm0, [rsp - #32]",
 968                 src_offset, dst_offset);
 969       break;
 970     default:
 971       ShouldNotReachHere();
 972     }
 973 #endif
 974   }
 975   return calc_size;
 976 }
 977 
 978 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream* st ) const {
 979   // Get registers to move
 980   OptoReg::Name src_second = ra_->get_reg_second(in(1));
 981   OptoReg::Name src_first = ra_->get_reg_first(in(1));
 982   OptoReg::Name dst_second = ra_->get_reg_second(this );
 983   OptoReg::Name dst_first = ra_->get_reg_first(this );
 984 
 985   enum RC src_second_rc = rc_class(src_second);
 986   enum RC src_first_rc = rc_class(src_first);
 987   enum RC dst_second_rc = rc_class(dst_second);
 988   enum RC dst_first_rc = rc_class(dst_first);
 989 
 990   assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );
 991 
 992   // Generate spill code!
 993   int size = 0;
 994 
 995   if( src_first == dst_first && src_second == dst_second )
 996     return size;            // Self copy, no move
 997 
 998   if (bottom_type()->isa_vect() != NULL) {
 999     uint ireg = ideal_reg();
1000     assert((src_first_rc != rc_int && dst_first_rc != rc_int), "sanity");
1001     assert((src_first_rc != rc_float && dst_first_rc != rc_float), "sanity");
1002     assert((ireg == Op_VecS || ireg == Op_VecD || ireg == Op_VecX || ireg == Op_VecY), "sanity");
1003     if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
1004       // mem -> mem
1005       int src_offset = ra_->reg2offset(src_first);
1006       int dst_offset = ra_->reg2offset(dst_first);
1007       return vec_stack_to_stack_helper(cbuf, do_size, src_offset, dst_offset, ireg, st);
1008     } else if (src_first_rc == rc_xmm && dst_first_rc == rc_xmm ) {
1009       return vec_mov_helper(cbuf, do_size, src_first, dst_first, src_second, dst_second, ireg, st);
1010     } else if (src_first_rc == rc_xmm && dst_first_rc == rc_stack ) {
1011       int stack_offset = ra_->reg2offset(dst_first);
1012       return vec_spill_helper(cbuf, do_size, false, stack_offset, src_first, ireg, st);
1013     } else if (src_first_rc == rc_stack && dst_first_rc == rc_xmm ) {
1014       int stack_offset = ra_->reg2offset(src_first);
1015       return vec_spill_helper(cbuf, do_size, true,  stack_offset, dst_first, ireg, st);
1016     } else {
1017       ShouldNotReachHere();
1018     }
1019   }
1020 
1021   // --------------------------------------
1022   // Check for mem-mem move.  push/pop to move.
1023   if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
1024     if( src_second == dst_first ) { // overlapping stack copy ranges
1025       assert( src_second_rc == rc_stack && dst_second_rc == rc_stack, "we only expect a stk-stk copy here" );
1026       size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
1027       size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
1028       src_second_rc = dst_second_rc = rc_bad;  // flag as already moved the second bits
1029     }
1030     // move low bits
1031     size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),ESI_num,0xFF,"PUSH  ",size, st);
1032     size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),EAX_num,0x8F,"POP   ",size, st);
1033     if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) { // mov second bits
1034       size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
1035       size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
1036     }
1037     return size;
1038   }
1039 
1040   // --------------------------------------
1041   // Check for integer reg-reg copy
1042   if( src_first_rc == rc_int && dst_first_rc == rc_int )
1043     size = impl_mov_helper(cbuf,do_size,src_first,dst_first,size, st);
1044 
1045   // Check for integer store
1046   if( src_first_rc == rc_int && dst_first_rc == rc_stack )
1047     size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first,0x89,"MOV ",size, st);
1048 
1049   // Check for integer load
1050   if( dst_first_rc == rc_int && src_first_rc == rc_stack )
1051     size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first,0x8B,"MOV ",size, st);
1052 
1053   // Check for integer reg-xmm reg copy
1054   if( src_first_rc == rc_int && dst_first_rc == rc_xmm ) {
1055     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
1056             "no 64 bit integer-float reg moves" );
1057     return impl_movgpr2x_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
1058   }
1059   // --------------------------------------
1060   // Check for float reg-reg copy
1061   if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
1062     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
1063             (src_first+1 == src_second && dst_first+1 == dst_second), "no non-adjacent float-moves" );
1064     if( cbuf ) {
1065 
1066       // Note the mucking with the register encode to compensate for the 0/1
1067       // indexing issue mentioned in a comment in the reg_def sections
1068       // for FPR registers many lines above here.
1069 
1070       if( src_first != FPR1L_num ) {
1071         emit_opcode  (*cbuf, 0xD9 );           // FLD    ST(i)
1072         emit_d8      (*cbuf, 0xC0+Matcher::_regEncode[src_first]-1 );
1073         emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
1074         emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
1075      } else {
1076         emit_opcode  (*cbuf, 0xDD );           // FST    ST(i)
1077         emit_d8      (*cbuf, 0xD0+Matcher::_regEncode[dst_first]-1 );
1078      }
1079 #ifndef PRODUCT
1080     } else if( !do_size ) {
1081       if( size != 0 ) st->print("\n\t");
1082       if( src_first != FPR1L_num ) st->print("FLD    %s\n\tFSTP   %s",Matcher::regName[src_first],Matcher::regName[dst_first]);
1083       else                      st->print(             "FST    %s",                            Matcher::regName[dst_first]);
1084 #endif
1085     }
1086     return size + ((src_first != FPR1L_num) ? 2+2 : 2);
1087   }
1088 
1089   // Check for float store
1090   if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
1091     return impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,ra_->reg2offset(dst_first),size, st);
1092   }
1093 
1094   // Check for float load
1095   if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
1096     int offset = ra_->reg2offset(src_first);
1097     const char *op_str;
1098     int op;
1099     if( src_first+1 == src_second && dst_first+1 == dst_second ) { // double load?
1100       op_str = "FLD_D";
1101       op = 0xDD;
1102     } else {                   // 32-bit load
1103       op_str = "FLD_S";
1104       op = 0xD9;
1105       assert( src_second_rc == rc_bad && dst_second_rc == rc_bad, "no non-adjacent float-loads" );
1106     }
1107     if( cbuf ) {
1108       emit_opcode  (*cbuf, op );
1109       encode_RegMem(*cbuf, 0x0, ESP_enc, 0x4, 0, offset, relocInfo::none);
1110       emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
1111       emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
1112 #ifndef PRODUCT
1113     } else if( !do_size ) {
1114       if( size != 0 ) st->print("\n\t");
1115       st->print("%s  ST,[ESP + #%d]\n\tFSTP   %s",op_str, offset,Matcher::regName[dst_first]);
1116 #endif
1117     }
1118     int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
1119     return size + 3+offset_size+2;
1120   }
1121 
1122   // Check for xmm reg-reg copy
1123   if( src_first_rc == rc_xmm && dst_first_rc == rc_xmm ) {
1124     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
1125             (src_first+1 == src_second && dst_first+1 == dst_second),
1126             "no non-adjacent float-moves" );
1127     return impl_movx_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
1128   }
1129 
1130   // Check for xmm reg-integer reg copy
1131   if( src_first_rc == rc_xmm && dst_first_rc == rc_int ) {
1132     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
1133             "no 64 bit float-integer reg moves" );
1134     return impl_movx2gpr_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
1135   }
1136 
1137   // Check for xmm store
1138   if( src_first_rc == rc_xmm && dst_first_rc == rc_stack ) {
1139     return impl_x_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first, src_second, size, st);
1140   }
1141 
1142   // Check for float xmm load
1143   if( dst_first_rc == rc_xmm && src_first_rc == rc_stack ) {
1144     return impl_x_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first, dst_second, size, st);
1145   }
1146 
1147   // Copy from float reg to xmm reg
1148   if( dst_first_rc == rc_xmm && src_first_rc == rc_float ) {
1149     // copy to the top of stack from floating point reg
1150     // and use LEA to preserve flags
1151     if( cbuf ) {
1152       emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP-8]
1153       emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
1154       emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
1155       emit_d8(*cbuf,0xF8);
1156 #ifndef PRODUCT
1157     } else if( !do_size ) {
1158       if( size != 0 ) st->print("\n\t");
1159       st->print("LEA    ESP,[ESP-8]");
1160 #endif
1161     }
1162     size += 4;
1163 
1164     size = impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,0,size, st);
1165 
1166     // Copy from the temp memory to the xmm reg.
1167     size = impl_x_helper(cbuf,do_size,true ,0,dst_first, dst_second, size, st);
1168 
1169     if( cbuf ) {
1170       emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP+8]
1171       emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
1172       emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
1173       emit_d8(*cbuf,0x08);
1174 #ifndef PRODUCT
1175     } else if( !do_size ) {
1176       if( size != 0 ) st->print("\n\t");
1177       st->print("LEA    ESP,[ESP+8]");
1178 #endif
1179     }
1180     size += 4;
1181     return size;
1182   }
1183 
1184   assert( size > 0, "missed a case" );
1185 
1186   // --------------------------------------------------------------------
1187   // Check for second bits still needing moving.
1188   if( src_second == dst_second )
1189     return size;               // Self copy; no move
1190   assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );
1191 
1192   // Check for second word int-int move
1193   if( src_second_rc == rc_int && dst_second_rc == rc_int )
1194     return impl_mov_helper(cbuf,do_size,src_second,dst_second,size, st);
1195 
1196   // Check for second word integer store
1197   if( src_second_rc == rc_int && dst_second_rc == rc_stack )
1198     return impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),src_second,0x89,"MOV ",size, st);
1199 
1200   // Check for second word integer load
1201   if( dst_second_rc == rc_int && src_second_rc == rc_stack )
1202     return impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),dst_second,0x8B,"MOV ",size, st);
1203 
1204 
1205   Unimplemented();
1206 }
1207 
1208 #ifndef PRODUCT
1209 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const {
1210   implementation( NULL, ra_, false, st );
1211 }
1212 #endif
1213 
1214 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1215   implementation( &cbuf, ra_, false, NULL );
1216 }
1217 
1218 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1219   return implementation( NULL, ra_, true, NULL );
1220 }
1221 
1222 
1223 //=============================================================================
1224 #ifndef PRODUCT
1225 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
1226   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1227   int reg = ra_->get_reg_first(this);
1228   st->print("LEA    %s,[ESP + #%d]",Matcher::regName[reg],offset);
1229 }
1230 #endif
1231 
1232 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1233   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1234   int reg = ra_->get_encode(this);
1235   if( offset >= 128 ) {
1236     emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
1237     emit_rm(cbuf, 0x2, reg, 0x04);
1238     emit_rm(cbuf, 0x0, 0x04, ESP_enc);
1239     emit_d32(cbuf, offset);
1240   }
1241   else {
1242     emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
1243     emit_rm(cbuf, 0x1, reg, 0x04);
1244     emit_rm(cbuf, 0x0, 0x04, ESP_enc);
1245     emit_d8(cbuf, offset);
1246   }
1247 }
1248 
1249 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1250   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1251   if( offset >= 128 ) {
1252     return 7;
1253   }
1254   else {
1255     return 4;
1256   }
1257 }
1258 
1259 //=============================================================================
1260 #ifndef PRODUCT
1261 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
1262   st->print_cr(  "CMP    EAX,[ECX+4]\t# Inline cache check");
1263   st->print_cr("\tJNE    SharedRuntime::handle_ic_miss_stub");
1264   st->print_cr("\tNOP");
1265   st->print_cr("\tNOP");
1266   if( !OptoBreakpoint )
1267     st->print_cr("\tNOP");
1268 }
1269 #endif
1270 
1271 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1272   MacroAssembler masm(&cbuf);
1273 #ifdef ASSERT
1274   uint insts_size = cbuf.insts_size();
1275 #endif
1276   masm.cmpptr(rax, Address(rcx, oopDesc::klass_offset_in_bytes()));
1277   masm.jump_cc(Assembler::notEqual,
1278                RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1279   /* WARNING these NOPs are critical so that verified entry point is properly
1280      aligned for patching by NativeJump::patch_verified_entry() */
1281   int nops_cnt = 2;
1282   if( !OptoBreakpoint ) // Leave space for int3
1283      nops_cnt += 1;
1284   masm.nop(nops_cnt);
1285 
1286   assert(cbuf.insts_size() - insts_size == size(ra_), "checking code size of inline cache node");
1287 }
1288 
1289 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1290   return OptoBreakpoint ? 11 : 12;
1291 }
1292 
1293 
1294 //=============================================================================
1295 uint size_exception_handler() {
1296   // NativeCall instruction size is the same as NativeJump.
1297   // exception handler starts out as jump and can be patched to
1298   // a call be deoptimization.  (4932387)
1299   // Note that this value is also credited (in output.cpp) to
1300   // the size of the code section.
1301   return NativeJump::instruction_size;
1302 }
1303 
1304 // Emit exception handler code.  Stuff framesize into a register
1305 // and call a VM stub routine.
1306 int emit_exception_handler(CodeBuffer& cbuf) {
1307 
1308   // Note that the code buffer's insts_mark is always relative to insts.
1309   // That's why we must use the macroassembler to generate a handler.
1310   MacroAssembler _masm(&cbuf);
1311   address base =
1312   __ start_a_stub(size_exception_handler());
1313   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1314   int offset = __ offset();
1315   __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
1316   assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1317   __ end_a_stub();
1318   return offset;
1319 }
1320 
1321 uint size_deopt_handler() {
1322   // NativeCall instruction size is the same as NativeJump.
1323   // exception handler starts out as jump and can be patched to
1324   // a call be deoptimization.  (4932387)
1325   // Note that this value is also credited (in output.cpp) to
1326   // the size of the code section.
1327   return 5 + NativeJump::instruction_size; // pushl(); jmp;
1328 }
1329 
1330 // Emit deopt handler code.
1331 int emit_deopt_handler(CodeBuffer& cbuf) {
1332 
1333   // Note that the code buffer's insts_mark is always relative to insts.
1334   // That's why we must use the macroassembler to generate a handler.
1335   MacroAssembler _masm(&cbuf);
1336   address base =
1337   __ start_a_stub(size_exception_handler());
1338   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1339   int offset = __ offset();
1340   InternalAddress here(__ pc());
1341   __ pushptr(here.addr());
1342 
1343   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1344   assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1345   __ end_a_stub();
1346   return offset;
1347 }
1348 
1349 int Matcher::regnum_to_fpu_offset(int regnum) {
1350   return regnum - 32; // The FP registers are in the second chunk
1351 }
1352 
1353 // This is UltraSparc specific, true just means we have fast l2f conversion
1354 const bool Matcher::convL2FSupported(void) {
1355   return true;
1356 }
1357 
1358 // Is this branch offset short enough that a short branch can be used?
1359 //
1360 // NOTE: If the platform does not provide any short branch variants, then
1361 //       this method should return false for offset 0.
1362 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
1363   // The passed offset is relative to address of the branch.
1364   // On 86 a branch displacement is calculated relative to address
1365   // of a next instruction.
1366   offset -= br_size;
1367 
1368   // the short version of jmpConUCF2 contains multiple branches,
1369   // making the reach slightly less
1370   if (rule == jmpConUCF2_rule)
1371     return (-126 <= offset && offset <= 125);
1372   return (-128 <= offset && offset <= 127);
1373 }
1374 
1375 const bool Matcher::isSimpleConstant64(jlong value) {
1376   // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1377   return false;
1378 }
1379 
1380 // The ecx parameter to rep stos for the ClearArray node is in dwords.
1381 const bool Matcher::init_array_count_is_in_bytes = false;
1382 
1383 // Threshold size for cleararray.
1384 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1385 
1386 // Needs 2 CMOV's for longs.
1387 const int Matcher::long_cmove_cost() { return 1; }
1388 
1389 // No CMOVF/CMOVD with SSE/SSE2
1390 const int Matcher::float_cmove_cost() { return (UseSSE>=1) ? ConditionalMoveLimit : 0; }
1391 
1392 // Should the Matcher clone shifts on addressing modes, expecting them to
1393 // be subsumed into complex addressing expressions or compute them into
1394 // registers?  True for Intel but false for most RISCs
1395 const bool Matcher::clone_shift_expressions = true;
1396 
1397 // Do we need to mask the count passed to shift instructions or does
1398 // the cpu only look at the lower 5/6 bits anyway?
1399 const bool Matcher::need_masked_shift_count = false;
1400 
1401 bool Matcher::narrow_oop_use_complex_address() {
1402   ShouldNotCallThis();
1403   return true;
1404 }
1405 
1406 bool Matcher::narrow_klass_use_complex_address() {
1407   ShouldNotCallThis();
1408   return true;
1409 }
1410 
1411 
1412 // Is it better to copy float constants, or load them directly from memory?
1413 // Intel can load a float constant from a direct address, requiring no
1414 // extra registers.  Most RISCs will have to materialize an address into a
1415 // register first, so they would do better to copy the constant from stack.
1416 const bool Matcher::rematerialize_float_constants = true;
1417 
1418 // If CPU can load and store mis-aligned doubles directly then no fixup is
1419 // needed.  Else we split the double into 2 integer pieces and move it
1420 // piece-by-piece.  Only happens when passing doubles into C code as the
1421 // Java calling convention forces doubles to be aligned.
1422 const bool Matcher::misaligned_doubles_ok = true;
1423 
1424 
1425 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
1426   // Get the memory operand from the node
1427   uint numopnds = node->num_opnds();        // Virtual call for number of operands
1428   uint skipped  = node->oper_input_base();  // Sum of leaves skipped so far
1429   assert( idx >= skipped, "idx too low in pd_implicit_null_fixup" );
1430   uint opcnt     = 1;                 // First operand
1431   uint num_edges = node->_opnds[1]->num_edges(); // leaves for first operand
1432   while( idx >= skipped+num_edges ) {
1433     skipped += num_edges;
1434     opcnt++;                          // Bump operand count
1435     assert( opcnt < numopnds, "Accessing non-existent operand" );
1436     num_edges = node->_opnds[opcnt]->num_edges(); // leaves for next operand
1437   }
1438 
1439   MachOper *memory = node->_opnds[opcnt];
1440   MachOper *new_memory = NULL;
1441   switch (memory->opcode()) {
1442   case DIRECT:
1443   case INDOFFSET32X:
1444     // No transformation necessary.
1445     return;
1446   case INDIRECT:
1447     new_memory = new (C) indirect_win95_safeOper( );
1448     break;
1449   case INDOFFSET8:
1450     new_memory = new (C) indOffset8_win95_safeOper(memory->disp(NULL, NULL, 0));
1451     break;
1452   case INDOFFSET32:
1453     new_memory = new (C) indOffset32_win95_safeOper(memory->disp(NULL, NULL, 0));
1454     break;
1455   case INDINDEXOFFSET:
1456     new_memory = new (C) indIndexOffset_win95_safeOper(memory->disp(NULL, NULL, 0));
1457     break;
1458   case INDINDEXSCALE:
1459     new_memory = new (C) indIndexScale_win95_safeOper(memory->scale());
1460     break;
1461   case INDINDEXSCALEOFFSET:
1462     new_memory = new (C) indIndexScaleOffset_win95_safeOper(memory->scale(), memory->disp(NULL, NULL, 0));
1463     break;
1464   case LOAD_LONG_INDIRECT:
1465   case LOAD_LONG_INDOFFSET32:
1466     // Does not use EBP as address register, use { EDX, EBX, EDI, ESI}
1467     return;
1468   default:
1469     assert(false, "unexpected memory operand in pd_implicit_null_fixup()");
1470     return;
1471   }
1472   node->_opnds[opcnt] = new_memory;
1473 }
1474 
1475 // Advertise here if the CPU requires explicit rounding operations
1476 // to implement the UseStrictFP mode.
1477 const bool Matcher::strict_fp_requires_explicit_rounding = true;
1478 
1479 // Are floats conerted to double when stored to stack during deoptimization?
1480 // On x32 it is stored with convertion only when FPU is used for floats.
1481 bool Matcher::float_in_double() { return (UseSSE == 0); }
1482 
1483 // Do ints take an entire long register or just half?
1484 const bool Matcher::int_in_long = false;
1485 
1486 // Return whether or not this register is ever used as an argument.  This
1487 // function is used on startup to build the trampoline stubs in generateOptoStub.
1488 // Registers not mentioned will be killed by the VM call in the trampoline, and
1489 // arguments in those registers not be available to the callee.
1490 bool Matcher::can_be_java_arg( int reg ) {
1491   if(  reg == ECX_num   || reg == EDX_num   ) return true;
1492   if( (reg == XMM0_num  || reg == XMM1_num ) && UseSSE>=1 ) return true;
1493   if( (reg == XMM0b_num || reg == XMM1b_num) && UseSSE>=2 ) return true;
1494   return false;
1495 }
1496 
1497 bool Matcher::is_spillable_arg( int reg ) {
1498   return can_be_java_arg(reg);
1499 }
1500 
1501 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
1502   // Use hardware integer DIV instruction when
1503   // it is faster than a code which use multiply.
1504   // Only when constant divisor fits into 32 bit
1505   // (min_jint is excluded to get only correct
1506   // positive 32 bit values from negative).
1507   return VM_Version::has_fast_idiv() &&
1508          (divisor == (int)divisor && divisor != min_jint);
1509 }
1510 
1511 // Register for DIVI projection of divmodI
1512 RegMask Matcher::divI_proj_mask() {
1513   return EAX_REG_mask();
1514 }
1515 
1516 // Register for MODI projection of divmodI
1517 RegMask Matcher::modI_proj_mask() {
1518   return EDX_REG_mask();
1519 }
1520 
1521 // Register for DIVL projection of divmodL
1522 RegMask Matcher::divL_proj_mask() {
1523   ShouldNotReachHere();
1524   return RegMask();
1525 }
1526 
1527 // Register for MODL projection of divmodL
1528 RegMask Matcher::modL_proj_mask() {
1529   ShouldNotReachHere();
1530   return RegMask();
1531 }
1532 
1533 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
1534   return EBP_REG_mask();
1535 }
1536 
1537 const RegMask Matcher::mathExactI_result_proj_mask() {
1538   return EAX_REG_mask();
1539 }
1540 
1541 const RegMask Matcher::mathExactI_flags_proj_mask() {
1542   return INT_FLAGS_mask();
1543 }
1544 
1545 // Returns true if the high 32 bits of the value is known to be zero.
1546 bool is_operand_hi32_zero(Node* n) {
1547   int opc = n->Opcode();
1548   if (opc == Op_AndL) {
1549     Node* o2 = n->in(2);
1550     if (o2->is_Con() && (o2->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
1551       return true;
1552     }
1553   }
1554   if (opc == Op_ConL && (n->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
1555     return true;
1556   }
1557   return false;
1558 }
1559 
1560 %}
1561 
1562 //----------ENCODING BLOCK-----------------------------------------------------
1563 // This block specifies the encoding classes used by the compiler to output
1564 // byte streams.  Encoding classes generate functions which are called by
1565 // Machine Instruction Nodes in order to generate the bit encoding of the
1566 // instruction.  Operands specify their base encoding interface with the
1567 // interface keyword.  There are currently supported four interfaces,
1568 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER.  REG_INTER causes an
1569 // operand to generate a function which returns its register number when
1570 // queried.   CONST_INTER causes an operand to generate a function which
1571 // returns the value of the constant when queried.  MEMORY_INTER causes an
1572 // operand to generate four functions which return the Base Register, the
1573 // Index Register, the Scale Value, and the Offset Value of the operand when
1574 // queried.  COND_INTER causes an operand to generate six functions which
1575 // return the encoding code (ie - encoding bits for the instruction)
1576 // associated with each basic boolean condition for a conditional instruction.
1577 // Instructions specify two basic values for encoding.  They use the
1578 // ins_encode keyword to specify their encoding class (which must be one of
1579 // the class names specified in the encoding block), and they use the
1580 // opcode keyword to specify, in order, their primary, secondary, and
1581 // tertiary opcode.  Only the opcode sections which a particular instruction
1582 // needs for encoding need to be specified.
1583 encode %{
1584   // Build emit functions for each basic byte or larger field in the intel
1585   // encoding scheme (opcode, rm, sib, immediate), and call them from C++
1586   // code in the enc_class source block.  Emit functions will live in the
1587   // main source block for now.  In future, we can generalize this by
1588   // adding a syntax that specifies the sizes of fields in an order,
1589   // so that the adlc can build the emit functions automagically
1590 
1591   // Emit primary opcode
1592   enc_class OpcP %{
1593     emit_opcode(cbuf, $primary);
1594   %}
1595 
1596   // Emit secondary opcode
1597   enc_class OpcS %{
1598     emit_opcode(cbuf, $secondary);
1599   %}
1600 
1601   // Emit opcode directly
1602   enc_class Opcode(immI d8) %{
1603     emit_opcode(cbuf, $d8$$constant);
1604   %}
1605 
1606   enc_class SizePrefix %{
1607     emit_opcode(cbuf,0x66);
1608   %}
1609 
1610   enc_class RegReg (rRegI dst, rRegI src) %{    // RegReg(Many)
1611     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1612   %}
1613 
1614   enc_class OpcRegReg (immI opcode, rRegI dst, rRegI src) %{    // OpcRegReg(Many)
1615     emit_opcode(cbuf,$opcode$$constant);
1616     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1617   %}
1618 
1619   enc_class mov_r32_imm0( rRegI dst ) %{
1620     emit_opcode( cbuf, 0xB8 + $dst$$reg ); // 0xB8+ rd   -- MOV r32  ,imm32
1621     emit_d32   ( cbuf, 0x0  );             //                         imm32==0x0
1622   %}
1623 
1624   enc_class cdq_enc %{
1625     // Full implementation of Java idiv and irem; checks for
1626     // special case as described in JVM spec., p.243 & p.271.
1627     //
1628     //         normal case                           special case
1629     //
1630     // input : rax,: dividend                         min_int
1631     //         reg: divisor                          -1
1632     //
1633     // output: rax,: quotient  (= rax, idiv reg)       min_int
1634     //         rdx: remainder (= rax, irem reg)       0
1635     //
1636     //  Code sequnce:
1637     //
1638     //  81 F8 00 00 00 80    cmp         rax,80000000h
1639     //  0F 85 0B 00 00 00    jne         normal_case
1640     //  33 D2                xor         rdx,edx
1641     //  83 F9 FF             cmp         rcx,0FFh
1642     //  0F 84 03 00 00 00    je          done
1643     //                  normal_case:
1644     //  99                   cdq
1645     //  F7 F9                idiv        rax,ecx
1646     //                  done:
1647     //
1648     emit_opcode(cbuf,0x81); emit_d8(cbuf,0xF8);
1649     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);
1650     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x80);                     // cmp rax,80000000h
1651     emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x85);
1652     emit_opcode(cbuf,0x0B); emit_d8(cbuf,0x00);
1653     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // jne normal_case
1654     emit_opcode(cbuf,0x33); emit_d8(cbuf,0xD2);                     // xor rdx,edx
1655     emit_opcode(cbuf,0x83); emit_d8(cbuf,0xF9); emit_d8(cbuf,0xFF); // cmp rcx,0FFh
1656     emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x84);
1657     emit_opcode(cbuf,0x03); emit_d8(cbuf,0x00);
1658     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // je done
1659     // normal_case:
1660     emit_opcode(cbuf,0x99);                                         // cdq
1661     // idiv (note: must be emitted by the user of this rule)
1662     // normal:
1663   %}
1664 
1665   // Dense encoding for older common ops
1666   enc_class Opc_plus(immI opcode, rRegI reg) %{
1667     emit_opcode(cbuf, $opcode$$constant + $reg$$reg);
1668   %}
1669 
1670 
1671   // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
1672   enc_class OpcSE (immI imm) %{ // Emit primary opcode and set sign-extend bit
1673     // Check for 8-bit immediate, and set sign extend bit in opcode
1674     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1675       emit_opcode(cbuf, $primary | 0x02);
1676     }
1677     else {                          // If 32-bit immediate
1678       emit_opcode(cbuf, $primary);
1679     }
1680   %}
1681 
1682   enc_class OpcSErm (rRegI dst, immI imm) %{    // OpcSEr/m
1683     // Emit primary opcode and set sign-extend bit
1684     // Check for 8-bit immediate, and set sign extend bit in opcode
1685     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1686       emit_opcode(cbuf, $primary | 0x02);    }
1687     else {                          // If 32-bit immediate
1688       emit_opcode(cbuf, $primary);
1689     }
1690     // Emit r/m byte with secondary opcode, after primary opcode.
1691     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1692   %}
1693 
1694   enc_class Con8or32 (immI imm) %{    // Con8or32(storeImmI), 8 or 32 bits
1695     // Check for 8-bit immediate, and set sign extend bit in opcode
1696     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1697       $$$emit8$imm$$constant;
1698     }
1699     else {                          // If 32-bit immediate
1700       // Output immediate
1701       $$$emit32$imm$$constant;
1702     }
1703   %}
1704 
1705   enc_class Long_OpcSErm_Lo(eRegL dst, immL imm) %{
1706     // Emit primary opcode and set sign-extend bit
1707     // Check for 8-bit immediate, and set sign extend bit in opcode
1708     int con = (int)$imm$$constant; // Throw away top bits
1709     emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
1710     // Emit r/m byte with secondary opcode, after primary opcode.
1711     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1712     if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
1713     else                               emit_d32(cbuf,con);
1714   %}
1715 
1716   enc_class Long_OpcSErm_Hi(eRegL dst, immL imm) %{
1717     // Emit primary opcode and set sign-extend bit
1718     // Check for 8-bit immediate, and set sign extend bit in opcode
1719     int con = (int)($imm$$constant >> 32); // Throw away bottom bits
1720     emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
1721     // Emit r/m byte with tertiary opcode, after primary opcode.
1722     emit_rm(cbuf, 0x3, $tertiary, HIGH_FROM_LOW($dst$$reg));
1723     if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
1724     else                               emit_d32(cbuf,con);
1725   %}
1726 
1727   enc_class OpcSReg (rRegI dst) %{    // BSWAP
1728     emit_cc(cbuf, $secondary, $dst$$reg );
1729   %}
1730 
1731   enc_class bswap_long_bytes(eRegL dst) %{ // BSWAP
1732     int destlo = $dst$$reg;
1733     int desthi = HIGH_FROM_LOW(destlo);
1734     // bswap lo
1735     emit_opcode(cbuf, 0x0F);
1736     emit_cc(cbuf, 0xC8, destlo);
1737     // bswap hi
1738     emit_opcode(cbuf, 0x0F);
1739     emit_cc(cbuf, 0xC8, desthi);
1740     // xchg lo and hi
1741     emit_opcode(cbuf, 0x87);
1742     emit_rm(cbuf, 0x3, destlo, desthi);
1743   %}
1744 
1745   enc_class RegOpc (rRegI div) %{    // IDIV, IMOD, JMP indirect, ...
1746     emit_rm(cbuf, 0x3, $secondary, $div$$reg );
1747   %}
1748 
1749   enc_class enc_cmov(cmpOp cop ) %{ // CMOV
1750     $$$emit8$primary;
1751     emit_cc(cbuf, $secondary, $cop$$cmpcode);
1752   %}
1753 
1754   enc_class enc_cmov_dpr(cmpOp cop, regDPR src ) %{ // CMOV
1755     int op = 0xDA00 + $cop$$cmpcode + ($src$$reg-1);
1756     emit_d8(cbuf, op >> 8 );
1757     emit_d8(cbuf, op & 255);
1758   %}
1759 
1760   // emulate a CMOV with a conditional branch around a MOV
1761   enc_class enc_cmov_branch( cmpOp cop, immI brOffs ) %{ // CMOV
1762     // Invert sense of branch from sense of CMOV
1763     emit_cc( cbuf, 0x70, ($cop$$cmpcode^1) );
1764     emit_d8( cbuf, $brOffs$$constant );
1765   %}
1766 
1767   enc_class enc_PartialSubtypeCheck( ) %{
1768     Register Redi = as_Register(EDI_enc); // result register
1769     Register Reax = as_Register(EAX_enc); // super class
1770     Register Recx = as_Register(ECX_enc); // killed
1771     Register Resi = as_Register(ESI_enc); // sub class
1772     Label miss;
1773 
1774     MacroAssembler _masm(&cbuf);
1775     __ check_klass_subtype_slow_path(Resi, Reax, Recx, Redi,
1776                                      NULL, &miss,
1777                                      /*set_cond_codes:*/ true);
1778     if ($primary) {
1779       __ xorptr(Redi, Redi);
1780     }
1781     __ bind(miss);
1782   %}
1783 
1784   enc_class FFree_Float_Stack_All %{    // Free_Float_Stack_All
1785     MacroAssembler masm(&cbuf);
1786     int start = masm.offset();
1787     if (UseSSE >= 2) {
1788       if (VerifyFPU) {
1789         masm.verify_FPU(0, "must be empty in SSE2+ mode");
1790       }
1791     } else {
1792       // External c_calling_convention expects the FPU stack to be 'clean'.
1793       // Compiled code leaves it dirty.  Do cleanup now.
1794       masm.empty_FPU_stack();
1795     }
1796     if (sizeof_FFree_Float_Stack_All == -1) {
1797       sizeof_FFree_Float_Stack_All = masm.offset() - start;
1798     } else {
1799       assert(masm.offset() - start == sizeof_FFree_Float_Stack_All, "wrong size");
1800     }
1801   %}
1802 
1803   enc_class Verify_FPU_For_Leaf %{
1804     if( VerifyFPU ) {
1805       MacroAssembler masm(&cbuf);
1806       masm.verify_FPU( -3, "Returning from Runtime Leaf call");
1807     }
1808   %}
1809 
1810   enc_class Java_To_Runtime (method meth) %{    // CALL Java_To_Runtime, Java_To_Runtime_Leaf
1811     // This is the instruction starting address for relocation info.
1812     cbuf.set_insts_mark();
1813     $$$emit8$primary;
1814     // CALL directly to the runtime
1815     emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1816                 runtime_call_Relocation::spec(), RELOC_IMM32 );
1817 
1818     if (UseSSE >= 2) {
1819       MacroAssembler _masm(&cbuf);
1820       BasicType rt = tf()->return_type();
1821 
1822       if ((rt == T_FLOAT || rt == T_DOUBLE) && !return_value_is_used()) {
1823         // A C runtime call where the return value is unused.  In SSE2+
1824         // mode the result needs to be removed from the FPU stack.  It's
1825         // likely that this function call could be removed by the
1826         // optimizer if the C function is a pure function.
1827         __ ffree(0);
1828       } else if (rt == T_FLOAT) {
1829         __ lea(rsp, Address(rsp, -4));
1830         __ fstp_s(Address(rsp, 0));
1831         __ movflt(xmm0, Address(rsp, 0));
1832         __ lea(rsp, Address(rsp,  4));
1833       } else if (rt == T_DOUBLE) {
1834         __ lea(rsp, Address(rsp, -8));
1835         __ fstp_d(Address(rsp, 0));
1836         __ movdbl(xmm0, Address(rsp, 0));
1837         __ lea(rsp, Address(rsp,  8));
1838       }
1839     }
1840   %}
1841 
1842 
1843   enc_class pre_call_resets %{
1844     // If method sets FPU control word restore it here
1845     debug_only(int off0 = cbuf.insts_size());
1846     if (ra_->C->in_24_bit_fp_mode()) {
1847       MacroAssembler _masm(&cbuf);
1848       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1849     }
1850     if (ra_->C->max_vector_size() > 16) {
1851       // Clear upper bits of YMM registers when current compiled code uses
1852       // wide vectors to avoid AVX <-> SSE transition penalty during call.
1853       MacroAssembler _masm(&cbuf);
1854       __ vzeroupper();
1855     }
1856     debug_only(int off1 = cbuf.insts_size());
1857     assert(off1 - off0 == pre_call_resets_size(), "correct size prediction");
1858   %}
1859 
1860   enc_class post_call_FPU %{
1861     // If method sets FPU control word do it here also
1862     if (Compile::current()->in_24_bit_fp_mode()) {
1863       MacroAssembler masm(&cbuf);
1864       masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
1865     }
1866   %}
1867 
1868   enc_class Java_Static_Call (method meth) %{    // JAVA STATIC CALL
1869     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
1870     // who we intended to call.
1871     cbuf.set_insts_mark();
1872     $$$emit8$primary;
1873     if (!_method) {
1874       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1875                      runtime_call_Relocation::spec(), RELOC_IMM32 );
1876     } else if (_optimized_virtual) {
1877       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1878                      opt_virtual_call_Relocation::spec(), RELOC_IMM32 );
1879     } else {
1880       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1881                      static_call_Relocation::spec(), RELOC_IMM32 );
1882     }
1883     if (_method) {  // Emit stub for static call.
1884       CompiledStaticCall::emit_to_interp_stub(cbuf);
1885     }
1886   %}
1887 
1888   enc_class Java_Dynamic_Call (method meth) %{    // JAVA DYNAMIC CALL
1889     MacroAssembler _masm(&cbuf);
1890     __ ic_call((address)$meth$$method);
1891   %}
1892 
1893   enc_class Java_Compiled_Call (method meth) %{    // JAVA COMPILED CALL
1894     int disp = in_bytes(Method::from_compiled_offset());
1895     assert( -128 <= disp && disp <= 127, "compiled_code_offset isn't small");
1896 
1897     // CALL *[EAX+in_bytes(Method::from_compiled_code_entry_point_offset())]
1898     cbuf.set_insts_mark();
1899     $$$emit8$primary;
1900     emit_rm(cbuf, 0x01, $secondary, EAX_enc );  // R/M byte
1901     emit_d8(cbuf, disp);             // Displacement
1902 
1903   %}
1904 
1905 //   Following encoding is no longer used, but may be restored if calling
1906 //   convention changes significantly.
1907 //   Became: Xor_Reg(EBP), Java_To_Runtime( labl )
1908 //
1909 //   enc_class Java_Interpreter_Call (label labl) %{    // JAVA INTERPRETER CALL
1910 //     // int ic_reg     = Matcher::inline_cache_reg();
1911 //     // int ic_encode  = Matcher::_regEncode[ic_reg];
1912 //     // int imo_reg    = Matcher::interpreter_method_oop_reg();
1913 //     // int imo_encode = Matcher::_regEncode[imo_reg];
1914 //
1915 //     // // Interpreter expects method_oop in EBX, currently a callee-saved register,
1916 //     // // so we load it immediately before the call
1917 //     // emit_opcode(cbuf, 0x8B);                     // MOV    imo_reg,ic_reg  # method_oop
1918 //     // emit_rm(cbuf, 0x03, imo_encode, ic_encode ); // R/M byte
1919 //
1920 //     // xor rbp,ebp
1921 //     emit_opcode(cbuf, 0x33);
1922 //     emit_rm(cbuf, 0x3, EBP_enc, EBP_enc);
1923 //
1924 //     // CALL to interpreter.
1925 //     cbuf.set_insts_mark();
1926 //     $$$emit8$primary;
1927 //     emit_d32_reloc(cbuf, ($labl$$label - (int)(cbuf.insts_end()) - 4),
1928 //                 runtime_call_Relocation::spec(), RELOC_IMM32 );
1929 //   %}
1930 
1931   enc_class RegOpcImm (rRegI dst, immI8 shift) %{    // SHL, SAR, SHR
1932     $$$emit8$primary;
1933     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1934     $$$emit8$shift$$constant;
1935   %}
1936 
1937   enc_class LdImmI (rRegI dst, immI src) %{    // Load Immediate
1938     // Load immediate does not have a zero or sign extended version
1939     // for 8-bit immediates
1940     emit_opcode(cbuf, 0xB8 + $dst$$reg);
1941     $$$emit32$src$$constant;
1942   %}
1943 
1944   enc_class LdImmP (rRegI dst, immI src) %{    // Load Immediate
1945     // Load immediate does not have a zero or sign extended version
1946     // for 8-bit immediates
1947     emit_opcode(cbuf, $primary + $dst$$reg);
1948     $$$emit32$src$$constant;
1949   %}
1950 
1951   enc_class LdImmL_Lo( eRegL dst, immL src) %{    // Load Immediate
1952     // Load immediate does not have a zero or sign extended version
1953     // for 8-bit immediates
1954     int dst_enc = $dst$$reg;
1955     int src_con = $src$$constant & 0x0FFFFFFFFL;
1956     if (src_con == 0) {
1957       // xor dst, dst
1958       emit_opcode(cbuf, 0x33);
1959       emit_rm(cbuf, 0x3, dst_enc, dst_enc);
1960     } else {
1961       emit_opcode(cbuf, $primary + dst_enc);
1962       emit_d32(cbuf, src_con);
1963     }
1964   %}
1965 
1966   enc_class LdImmL_Hi( eRegL dst, immL src) %{    // Load Immediate
1967     // Load immediate does not have a zero or sign extended version
1968     // for 8-bit immediates
1969     int dst_enc = $dst$$reg + 2;
1970     int src_con = ((julong)($src$$constant)) >> 32;
1971     if (src_con == 0) {
1972       // xor dst, dst
1973       emit_opcode(cbuf, 0x33);
1974       emit_rm(cbuf, 0x3, dst_enc, dst_enc);
1975     } else {
1976       emit_opcode(cbuf, $primary + dst_enc);
1977       emit_d32(cbuf, src_con);
1978     }
1979   %}
1980 
1981 
1982   // Encode a reg-reg copy.  If it is useless, then empty encoding.
1983   enc_class enc_Copy( rRegI dst, rRegI src ) %{
1984     encode_Copy( cbuf, $dst$$reg, $src$$reg );
1985   %}
1986 
1987   enc_class enc_CopyL_Lo( rRegI dst, eRegL src ) %{
1988     encode_Copy( cbuf, $dst$$reg, $src$$reg );
1989   %}
1990 
1991   enc_class RegReg (rRegI dst, rRegI src) %{    // RegReg(Many)
1992     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1993   %}
1994 
1995   enc_class RegReg_Lo(eRegL dst, eRegL src) %{    // RegReg(Many)
1996     $$$emit8$primary;
1997     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1998   %}
1999 
2000   enc_class RegReg_Hi(eRegL dst, eRegL src) %{    // RegReg(Many)
2001     $$$emit8$secondary;
2002     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
2003   %}
2004 
2005   enc_class RegReg_Lo2(eRegL dst, eRegL src) %{    // RegReg(Many)
2006     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2007   %}
2008 
2009   enc_class RegReg_Hi2(eRegL dst, eRegL src) %{    // RegReg(Many)
2010     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
2011   %}
2012 
2013   enc_class RegReg_HiLo( eRegL src, rRegI dst ) %{
2014     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($src$$reg));
2015   %}
2016 
2017   enc_class Con32 (immI src) %{    // Con32(storeImmI)
2018     // Output immediate
2019     $$$emit32$src$$constant;
2020   %}
2021 
2022   enc_class Con32FPR_as_bits(immFPR src) %{        // storeF_imm
2023     // Output Float immediate bits
2024     jfloat jf = $src$$constant;
2025     int    jf_as_bits = jint_cast( jf );
2026     emit_d32(cbuf, jf_as_bits);
2027   %}
2028 
2029   enc_class Con32F_as_bits(immF src) %{      // storeX_imm
2030     // Output Float immediate bits
2031     jfloat jf = $src$$constant;
2032     int    jf_as_bits = jint_cast( jf );
2033     emit_d32(cbuf, jf_as_bits);
2034   %}
2035 
2036   enc_class Con16 (immI src) %{    // Con16(storeImmI)
2037     // Output immediate
2038     $$$emit16$src$$constant;
2039   %}
2040 
2041   enc_class Con_d32(immI src) %{
2042     emit_d32(cbuf,$src$$constant);
2043   %}
2044 
2045   enc_class conmemref (eRegP t1) %{    // Con32(storeImmI)
2046     // Output immediate memory reference
2047     emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2048     emit_d32(cbuf, 0x00);
2049   %}
2050 
2051   enc_class lock_prefix( ) %{
2052     if( os::is_MP() )
2053       emit_opcode(cbuf,0xF0);         // [Lock]
2054   %}
2055 
2056   // Cmp-xchg long value.
2057   // Note: we need to swap rbx, and rcx before and after the
2058   //       cmpxchg8 instruction because the instruction uses
2059   //       rcx as the high order word of the new value to store but
2060   //       our register encoding uses rbx,.
2061   enc_class enc_cmpxchg8(eSIRegP mem_ptr) %{
2062 
2063     // XCHG  rbx,ecx
2064     emit_opcode(cbuf,0x87);
2065     emit_opcode(cbuf,0xD9);
2066     // [Lock]
2067     if( os::is_MP() )
2068       emit_opcode(cbuf,0xF0);
2069     // CMPXCHG8 [Eptr]
2070     emit_opcode(cbuf,0x0F);
2071     emit_opcode(cbuf,0xC7);
2072     emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
2073     // XCHG  rbx,ecx
2074     emit_opcode(cbuf,0x87);
2075     emit_opcode(cbuf,0xD9);
2076   %}
2077 
2078   enc_class enc_cmpxchg(eSIRegP mem_ptr) %{
2079     // [Lock]
2080     if( os::is_MP() )
2081       emit_opcode(cbuf,0xF0);
2082 
2083     // CMPXCHG [Eptr]
2084     emit_opcode(cbuf,0x0F);
2085     emit_opcode(cbuf,0xB1);
2086     emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
2087   %}
2088 
2089   enc_class enc_flags_ne_to_boolean( iRegI res ) %{
2090     int res_encoding = $res$$reg;
2091 
2092     // MOV  res,0
2093     emit_opcode( cbuf, 0xB8 + res_encoding);
2094     emit_d32( cbuf, 0 );
2095     // JNE,s  fail
2096     emit_opcode(cbuf,0x75);
2097     emit_d8(cbuf, 5 );
2098     // MOV  res,1
2099     emit_opcode( cbuf, 0xB8 + res_encoding);
2100     emit_d32( cbuf, 1 );
2101     // fail:
2102   %}
2103 
2104   enc_class set_instruction_start( ) %{
2105     cbuf.set_insts_mark();            // Mark start of opcode for reloc info in mem operand
2106   %}
2107 
2108   enc_class RegMem (rRegI ereg, memory mem) %{    // emit_reg_mem
2109     int reg_encoding = $ereg$$reg;
2110     int base  = $mem$$base;
2111     int index = $mem$$index;
2112     int scale = $mem$$scale;
2113     int displace = $mem$$disp;
2114     relocInfo::relocType disp_reloc = $mem->disp_reloc();
2115     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
2116   %}
2117 
2118   enc_class RegMem_Hi(eRegL ereg, memory mem) %{    // emit_reg_mem
2119     int reg_encoding = HIGH_FROM_LOW($ereg$$reg);  // Hi register of pair, computed from lo
2120     int base  = $mem$$base;
2121     int index = $mem$$index;
2122     int scale = $mem$$scale;
2123     int displace = $mem$$disp + 4;      // Offset is 4 further in memory
2124     assert( $mem->disp_reloc() == relocInfo::none, "Cannot add 4 to oop" );
2125     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, relocInfo::none);
2126   %}
2127 
2128   enc_class move_long_small_shift( eRegL dst, immI_1_31 cnt ) %{
2129     int r1, r2;
2130     if( $tertiary == 0xA4 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
2131     else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }
2132     emit_opcode(cbuf,0x0F);
2133     emit_opcode(cbuf,$tertiary);
2134     emit_rm(cbuf, 0x3, r1, r2);
2135     emit_d8(cbuf,$cnt$$constant);
2136     emit_d8(cbuf,$primary);
2137     emit_rm(cbuf, 0x3, $secondary, r1);
2138     emit_d8(cbuf,$cnt$$constant);
2139   %}
2140 
2141   enc_class move_long_big_shift_sign( eRegL dst, immI_32_63 cnt ) %{
2142     emit_opcode( cbuf, 0x8B ); // Move
2143     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
2144     if( $cnt$$constant > 32 ) { // Shift, if not by zero
2145       emit_d8(cbuf,$primary);
2146       emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
2147       emit_d8(cbuf,$cnt$$constant-32);
2148     }
2149     emit_d8(cbuf,$primary);
2150     emit_rm(cbuf, 0x3, $secondary, HIGH_FROM_LOW($dst$$reg));
2151     emit_d8(cbuf,31);
2152   %}
2153 
2154   enc_class move_long_big_shift_clr( eRegL dst, immI_32_63 cnt ) %{
2155     int r1, r2;
2156     if( $secondary == 0x5 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
2157     else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }
2158 
2159     emit_opcode( cbuf, 0x8B ); // Move r1,r2
2160     emit_rm(cbuf, 0x3, r1, r2);
2161     if( $cnt$$constant > 32 ) { // Shift, if not by zero
2162       emit_opcode(cbuf,$primary);
2163       emit_rm(cbuf, 0x3, $secondary, r1);
2164       emit_d8(cbuf,$cnt$$constant-32);
2165     }
2166     emit_opcode(cbuf,0x33);  // XOR r2,r2
2167     emit_rm(cbuf, 0x3, r2, r2);
2168   %}
2169 
2170   // Clone of RegMem but accepts an extra parameter to access each
2171   // half of a double in memory; it never needs relocation info.
2172   enc_class Mov_MemD_half_to_Reg (immI opcode, memory mem, immI disp_for_half, rRegI rm_reg) %{
2173     emit_opcode(cbuf,$opcode$$constant);
2174     int reg_encoding = $rm_reg$$reg;
2175     int base     = $mem$$base;
2176     int index    = $mem$$index;
2177     int scale    = $mem$$scale;
2178     int displace = $mem$$disp + $disp_for_half$$constant;
2179     relocInfo::relocType disp_reloc = relocInfo::none;
2180     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
2181   %}
2182 
2183   // !!!!! Special Custom Code used by MemMove, and stack access instructions !!!!!
2184   //
2185   // Clone of RegMem except the RM-byte's reg/opcode field is an ADLC-time constant
2186   // and it never needs relocation information.
2187   // Frequently used to move data between FPU's Stack Top and memory.
2188   enc_class RMopc_Mem_no_oop (immI rm_opcode, memory mem) %{
2189     int rm_byte_opcode = $rm_opcode$$constant;
2190     int base     = $mem$$base;
2191     int index    = $mem$$index;
2192     int scale    = $mem$$scale;
2193     int displace = $mem$$disp;
2194     assert( $mem->disp_reloc() == relocInfo::none, "No oops here because no reloc info allowed" );
2195     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, relocInfo::none);
2196   %}
2197 
2198   enc_class RMopc_Mem (immI rm_opcode, memory mem) %{
2199     int rm_byte_opcode = $rm_opcode$$constant;
2200     int base     = $mem$$base;
2201     int index    = $mem$$index;
2202     int scale    = $mem$$scale;
2203     int displace = $mem$$disp;
2204     relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
2205     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
2206   %}
2207 
2208   enc_class RegLea (rRegI dst, rRegI src0, immI src1 ) %{    // emit_reg_lea
2209     int reg_encoding = $dst$$reg;
2210     int base         = $src0$$reg;      // 0xFFFFFFFF indicates no base
2211     int index        = 0x04;            // 0x04 indicates no index
2212     int scale        = 0x00;            // 0x00 indicates no scale
2213     int displace     = $src1$$constant; // 0x00 indicates no displacement
2214     relocInfo::relocType disp_reloc = relocInfo::none;
2215     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
2216   %}
2217 
2218   enc_class min_enc (rRegI dst, rRegI src) %{    // MIN
2219     // Compare dst,src
2220     emit_opcode(cbuf,0x3B);
2221     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2222     // jmp dst < src around move
2223     emit_opcode(cbuf,0x7C);
2224     emit_d8(cbuf,2);
2225     // move dst,src
2226     emit_opcode(cbuf,0x8B);
2227     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2228   %}
2229 
2230   enc_class max_enc (rRegI dst, rRegI src) %{    // MAX
2231     // Compare dst,src
2232     emit_opcode(cbuf,0x3B);
2233     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2234     // jmp dst > src around move
2235     emit_opcode(cbuf,0x7F);
2236     emit_d8(cbuf,2);
2237     // move dst,src
2238     emit_opcode(cbuf,0x8B);
2239     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2240   %}
2241 
2242   enc_class enc_FPR_store(memory mem, regDPR src) %{
2243     // If src is FPR1, we can just FST to store it.
2244     // Else we need to FLD it to FPR1, then FSTP to store/pop it.
2245     int reg_encoding = 0x2; // Just store
2246     int base  = $mem$$base;
2247     int index = $mem$$index;
2248     int scale = $mem$$scale;
2249     int displace = $mem$$disp;
2250     relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
2251     if( $src$$reg != FPR1L_enc ) {
2252       reg_encoding = 0x3;  // Store & pop
2253       emit_opcode( cbuf, 0xD9 ); // FLD (i.e., push it)
2254       emit_d8( cbuf, 0xC0-1+$src$$reg );
2255     }
2256     cbuf.set_insts_mark();       // Mark start of opcode for reloc info in mem operand
2257     emit_opcode(cbuf,$primary);
2258     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_reloc);
2259   %}
2260 
2261   enc_class neg_reg(rRegI dst) %{
2262     // NEG $dst
2263     emit_opcode(cbuf,0xF7);
2264     emit_rm(cbuf, 0x3, 0x03, $dst$$reg );
2265   %}
2266 
2267   enc_class setLT_reg(eCXRegI dst) %{
2268     // SETLT $dst
2269     emit_opcode(cbuf,0x0F);
2270     emit_opcode(cbuf,0x9C);
2271     emit_rm( cbuf, 0x3, 0x4, $dst$$reg );
2272   %}
2273 
2274   enc_class enc_cmpLTP(ncxRegI p, ncxRegI q, ncxRegI y, eCXRegI tmp) %{    // cadd_cmpLT
2275     int tmpReg = $tmp$$reg;
2276 
2277     // SUB $p,$q
2278     emit_opcode(cbuf,0x2B);
2279     emit_rm(cbuf, 0x3, $p$$reg, $q$$reg);
2280     // SBB $tmp,$tmp
2281     emit_opcode(cbuf,0x1B);
2282     emit_rm(cbuf, 0x3, tmpReg, tmpReg);
2283     // AND $tmp,$y
2284     emit_opcode(cbuf,0x23);
2285     emit_rm(cbuf, 0x3, tmpReg, $y$$reg);
2286     // ADD $p,$tmp
2287     emit_opcode(cbuf,0x03);
2288     emit_rm(cbuf, 0x3, $p$$reg, tmpReg);
2289   %}
2290 
2291   enc_class shift_left_long( eRegL dst, eCXRegI shift ) %{
2292     // TEST shift,32
2293     emit_opcode(cbuf,0xF7);
2294     emit_rm(cbuf, 0x3, 0, ECX_enc);
2295     emit_d32(cbuf,0x20);
2296     // JEQ,s small
2297     emit_opcode(cbuf, 0x74);
2298     emit_d8(cbuf, 0x04);
2299     // MOV    $dst.hi,$dst.lo
2300     emit_opcode( cbuf, 0x8B );
2301     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
2302     // CLR    $dst.lo
2303     emit_opcode(cbuf, 0x33);
2304     emit_rm(cbuf, 0x3, $dst$$reg, $dst$$reg);
2305 // small:
2306     // SHLD   $dst.hi,$dst.lo,$shift
2307     emit_opcode(cbuf,0x0F);
2308     emit_opcode(cbuf,0xA5);
2309     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
2310     // SHL    $dst.lo,$shift"
2311     emit_opcode(cbuf,0xD3);
2312     emit_rm(cbuf, 0x3, 0x4, $dst$$reg );
2313   %}
2314 
2315   enc_class shift_right_long( eRegL dst, eCXRegI shift ) %{
2316     // TEST shift,32
2317     emit_opcode(cbuf,0xF7);
2318     emit_rm(cbuf, 0x3, 0, ECX_enc);
2319     emit_d32(cbuf,0x20);
2320     // JEQ,s small
2321     emit_opcode(cbuf, 0x74);
2322     emit_d8(cbuf, 0x04);
2323     // MOV    $dst.lo,$dst.hi
2324     emit_opcode( cbuf, 0x8B );
2325     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
2326     // CLR    $dst.hi
2327     emit_opcode(cbuf, 0x33);
2328     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($dst$$reg));
2329 // small:
2330     // SHRD   $dst.lo,$dst.hi,$shift
2331     emit_opcode(cbuf,0x0F);
2332     emit_opcode(cbuf,0xAD);
2333     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
2334     // SHR    $dst.hi,$shift"
2335     emit_opcode(cbuf,0xD3);
2336     emit_rm(cbuf, 0x3, 0x5, HIGH_FROM_LOW($dst$$reg) );
2337   %}
2338 
2339   enc_class shift_right_arith_long( eRegL dst, eCXRegI shift ) %{
2340     // TEST shift,32
2341     emit_opcode(cbuf,0xF7);
2342     emit_rm(cbuf, 0x3, 0, ECX_enc);
2343     emit_d32(cbuf,0x20);
2344     // JEQ,s small
2345     emit_opcode(cbuf, 0x74);
2346     emit_d8(cbuf, 0x05);
2347     // MOV    $dst.lo,$dst.hi
2348     emit_opcode( cbuf, 0x8B );
2349     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
2350     // SAR    $dst.hi,31
2351     emit_opcode(cbuf, 0xC1);
2352     emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW($dst$$reg) );
2353     emit_d8(cbuf, 0x1F );
2354 // small:
2355     // SHRD   $dst.lo,$dst.hi,$shift
2356     emit_opcode(cbuf,0x0F);
2357     emit_opcode(cbuf,0xAD);
2358     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
2359     // SAR    $dst.hi,$shift"
2360     emit_opcode(cbuf,0xD3);
2361     emit_rm(cbuf, 0x3, 0x7, HIGH_FROM_LOW($dst$$reg) );
2362   %}
2363 
2364 
2365   // ----------------- Encodings for floating point unit -----------------
2366   // May leave result in FPU-TOS or FPU reg depending on opcodes
2367   enc_class OpcReg_FPR(regFPR src) %{    // FMUL, FDIV
2368     $$$emit8$primary;
2369     emit_rm(cbuf, 0x3, $secondary, $src$$reg );
2370   %}
2371 
2372   // Pop argument in FPR0 with FSTP ST(0)
2373   enc_class PopFPU() %{
2374     emit_opcode( cbuf, 0xDD );
2375     emit_d8( cbuf, 0xD8 );
2376   %}
2377 
2378   // !!!!! equivalent to Pop_Reg_F
2379   enc_class Pop_Reg_DPR( regDPR dst ) %{
2380     emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
2381     emit_d8( cbuf, 0xD8+$dst$$reg );
2382   %}
2383 
2384   enc_class Push_Reg_DPR( regDPR dst ) %{
2385     emit_opcode( cbuf, 0xD9 );
2386     emit_d8( cbuf, 0xC0-1+$dst$$reg );   // FLD ST(i-1)
2387   %}
2388 
2389   enc_class strictfp_bias1( regDPR dst ) %{
2390     emit_opcode( cbuf, 0xDB );           // FLD m80real
2391     emit_opcode( cbuf, 0x2D );
2392     emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias1() );
2393     emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
2394     emit_opcode( cbuf, 0xC8+$dst$$reg );
2395   %}
2396 
2397   enc_class strictfp_bias2( regDPR dst ) %{
2398     emit_opcode( cbuf, 0xDB );           // FLD m80real
2399     emit_opcode( cbuf, 0x2D );
2400     emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias2() );
2401     emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
2402     emit_opcode( cbuf, 0xC8+$dst$$reg );
2403   %}
2404 
2405   // Special case for moving an integer register to a stack slot.
2406   enc_class OpcPRegSS( stackSlotI dst, rRegI src ) %{ // RegSS
2407     store_to_stackslot( cbuf, $primary, $src$$reg, $dst$$disp );
2408   %}
2409 
2410   // Special case for moving a register to a stack slot.
2411   enc_class RegSS( stackSlotI dst, rRegI src ) %{ // RegSS
2412     // Opcode already emitted
2413     emit_rm( cbuf, 0x02, $src$$reg, ESP_enc );   // R/M byte
2414     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);          // SIB byte
2415     emit_d32(cbuf, $dst$$disp);   // Displacement
2416   %}
2417 
2418   // Push the integer in stackSlot 'src' onto FP-stack
2419   enc_class Push_Mem_I( memory src ) %{    // FILD   [ESP+src]
2420     store_to_stackslot( cbuf, $primary, $secondary, $src$$disp );
2421   %}
2422 
2423   // Push FPU's TOS float to a stack-slot, and pop FPU-stack
2424   enc_class Pop_Mem_FPR( stackSlotF dst ) %{ // FSTP_S [ESP+dst]
2425     store_to_stackslot( cbuf, 0xD9, 0x03, $dst$$disp );
2426   %}
2427 
2428   // Same as Pop_Mem_F except for opcode
2429   // Push FPU's TOS double to a stack-slot, and pop FPU-stack
2430   enc_class Pop_Mem_DPR( stackSlotD dst ) %{ // FSTP_D [ESP+dst]
2431     store_to_stackslot( cbuf, 0xDD, 0x03, $dst$$disp );
2432   %}
2433 
2434   enc_class Pop_Reg_FPR( regFPR dst ) %{
2435     emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
2436     emit_d8( cbuf, 0xD8+$dst$$reg );
2437   %}
2438 
2439   enc_class Push_Reg_FPR( regFPR dst ) %{
2440     emit_opcode( cbuf, 0xD9 );           // FLD    ST(i-1)
2441     emit_d8( cbuf, 0xC0-1+$dst$$reg );
2442   %}
2443 
2444   // Push FPU's float to a stack-slot, and pop FPU-stack
2445   enc_class Pop_Mem_Reg_FPR( stackSlotF dst, regFPR src ) %{
2446     int pop = 0x02;
2447     if ($src$$reg != FPR1L_enc) {
2448       emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
2449       emit_d8( cbuf, 0xC0-1+$src$$reg );
2450       pop = 0x03;
2451     }
2452     store_to_stackslot( cbuf, 0xD9, pop, $dst$$disp ); // FST<P>_S  [ESP+dst]
2453   %}
2454 
2455   // Push FPU's double to a stack-slot, and pop FPU-stack
2456   enc_class Pop_Mem_Reg_DPR( stackSlotD dst, regDPR src ) %{
2457     int pop = 0x02;
2458     if ($src$$reg != FPR1L_enc) {
2459       emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
2460       emit_d8( cbuf, 0xC0-1+$src$$reg );
2461       pop = 0x03;
2462     }
2463     store_to_stackslot( cbuf, 0xDD, pop, $dst$$disp ); // FST<P>_D  [ESP+dst]
2464   %}
2465 
2466   // Push FPU's double to a FPU-stack-slot, and pop FPU-stack
2467   enc_class Pop_Reg_Reg_DPR( regDPR dst, regFPR src ) %{
2468     int pop = 0xD0 - 1; // -1 since we skip FLD
2469     if ($src$$reg != FPR1L_enc) {
2470       emit_opcode( cbuf, 0xD9 );         // FLD    ST(src-1)
2471       emit_d8( cbuf, 0xC0-1+$src$$reg );
2472       pop = 0xD8;
2473     }
2474     emit_opcode( cbuf, 0xDD );
2475     emit_d8( cbuf, pop+$dst$$reg );      // FST<P> ST(i)
2476   %}
2477 
2478 
2479   enc_class Push_Reg_Mod_DPR( regDPR dst, regDPR src) %{
2480     // load dst in FPR0
2481     emit_opcode( cbuf, 0xD9 );
2482     emit_d8( cbuf, 0xC0-1+$dst$$reg );
2483     if ($src$$reg != FPR1L_enc) {
2484       // fincstp
2485       emit_opcode (cbuf, 0xD9);
2486       emit_opcode (cbuf, 0xF7);
2487       // swap src with FPR1:
2488       // FXCH FPR1 with src
2489       emit_opcode(cbuf, 0xD9);
2490       emit_d8(cbuf, 0xC8-1+$src$$reg );
2491       // fdecstp
2492       emit_opcode (cbuf, 0xD9);
2493       emit_opcode (cbuf, 0xF6);
2494     }
2495   %}
2496 
2497   enc_class Push_ModD_encoding(regD src0, regD src1) %{
2498     MacroAssembler _masm(&cbuf);
2499     __ subptr(rsp, 8);
2500     __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
2501     __ fld_d(Address(rsp, 0));
2502     __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
2503     __ fld_d(Address(rsp, 0));
2504   %}
2505 
2506   enc_class Push_ModF_encoding(regF src0, regF src1) %{
2507     MacroAssembler _masm(&cbuf);
2508     __ subptr(rsp, 4);
2509     __ movflt(Address(rsp, 0), $src1$$XMMRegister);
2510     __ fld_s(Address(rsp, 0));
2511     __ movflt(Address(rsp, 0), $src0$$XMMRegister);
2512     __ fld_s(Address(rsp, 0));
2513   %}
2514 
2515   enc_class Push_ResultD(regD dst) %{
2516     MacroAssembler _masm(&cbuf);
2517     __ fstp_d(Address(rsp, 0));
2518     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
2519     __ addptr(rsp, 8);
2520   %}
2521 
2522   enc_class Push_ResultF(regF dst, immI d8) %{
2523     MacroAssembler _masm(&cbuf);
2524     __ fstp_s(Address(rsp, 0));
2525     __ movflt($dst$$XMMRegister, Address(rsp, 0));
2526     __ addptr(rsp, $d8$$constant);
2527   %}
2528 
2529   enc_class Push_SrcD(regD src) %{
2530     MacroAssembler _masm(&cbuf);
2531     __ subptr(rsp, 8);
2532     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
2533     __ fld_d(Address(rsp, 0));
2534   %}
2535 
2536   enc_class push_stack_temp_qword() %{
2537     MacroAssembler _masm(&cbuf);
2538     __ subptr(rsp, 8);
2539   %}
2540 
2541   enc_class pop_stack_temp_qword() %{
2542     MacroAssembler _masm(&cbuf);
2543     __ addptr(rsp, 8);
2544   %}
2545 
2546   enc_class push_xmm_to_fpr1(regD src) %{
2547     MacroAssembler _masm(&cbuf);
2548     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
2549     __ fld_d(Address(rsp, 0));
2550   %}
2551 
2552   enc_class Push_Result_Mod_DPR( regDPR src) %{
2553     if ($src$$reg != FPR1L_enc) {
2554       // fincstp
2555       emit_opcode (cbuf, 0xD9);
2556       emit_opcode (cbuf, 0xF7);
2557       // FXCH FPR1 with src
2558       emit_opcode(cbuf, 0xD9);
2559       emit_d8(cbuf, 0xC8-1+$src$$reg );
2560       // fdecstp
2561       emit_opcode (cbuf, 0xD9);
2562       emit_opcode (cbuf, 0xF6);
2563     }
2564     // // following asm replaced with Pop_Reg_F or Pop_Mem_F
2565     // // FSTP   FPR$dst$$reg
2566     // emit_opcode( cbuf, 0xDD );
2567     // emit_d8( cbuf, 0xD8+$dst$$reg );
2568   %}
2569 
2570   enc_class fnstsw_sahf_skip_parity() %{
2571     // fnstsw ax
2572     emit_opcode( cbuf, 0xDF );
2573     emit_opcode( cbuf, 0xE0 );
2574     // sahf
2575     emit_opcode( cbuf, 0x9E );
2576     // jnp  ::skip
2577     emit_opcode( cbuf, 0x7B );
2578     emit_opcode( cbuf, 0x05 );
2579   %}
2580 
2581   enc_class emitModDPR() %{
2582     // fprem must be iterative
2583     // :: loop
2584     // fprem
2585     emit_opcode( cbuf, 0xD9 );
2586     emit_opcode( cbuf, 0xF8 );
2587     // wait
2588     emit_opcode( cbuf, 0x9b );
2589     // fnstsw ax
2590     emit_opcode( cbuf, 0xDF );
2591     emit_opcode( cbuf, 0xE0 );
2592     // sahf
2593     emit_opcode( cbuf, 0x9E );
2594     // jp  ::loop
2595     emit_opcode( cbuf, 0x0F );
2596     emit_opcode( cbuf, 0x8A );
2597     emit_opcode( cbuf, 0xF4 );
2598     emit_opcode( cbuf, 0xFF );
2599     emit_opcode( cbuf, 0xFF );
2600     emit_opcode( cbuf, 0xFF );
2601   %}
2602 
2603   enc_class fpu_flags() %{
2604     // fnstsw_ax
2605     emit_opcode( cbuf, 0xDF);
2606     emit_opcode( cbuf, 0xE0);
2607     // test ax,0x0400
2608     emit_opcode( cbuf, 0x66 );   // operand-size prefix for 16-bit immediate
2609     emit_opcode( cbuf, 0xA9 );
2610     emit_d16   ( cbuf, 0x0400 );
2611     // // // This sequence works, but stalls for 12-16 cycles on PPro
2612     // // test rax,0x0400
2613     // emit_opcode( cbuf, 0xA9 );
2614     // emit_d32   ( cbuf, 0x00000400 );
2615     //
2616     // jz exit (no unordered comparison)
2617     emit_opcode( cbuf, 0x74 );
2618     emit_d8    ( cbuf, 0x02 );
2619     // mov ah,1 - treat as LT case (set carry flag)
2620     emit_opcode( cbuf, 0xB4 );
2621     emit_d8    ( cbuf, 0x01 );
2622     // sahf
2623     emit_opcode( cbuf, 0x9E);
2624   %}
2625 
2626   enc_class cmpF_P6_fixup() %{
2627     // Fixup the integer flags in case comparison involved a NaN
2628     //
2629     // JNP exit (no unordered comparison, P-flag is set by NaN)
2630     emit_opcode( cbuf, 0x7B );
2631     emit_d8    ( cbuf, 0x03 );
2632     // MOV AH,1 - treat as LT case (set carry flag)
2633     emit_opcode( cbuf, 0xB4 );
2634     emit_d8    ( cbuf, 0x01 );
2635     // SAHF
2636     emit_opcode( cbuf, 0x9E);
2637     // NOP     // target for branch to avoid branch to branch
2638     emit_opcode( cbuf, 0x90);
2639   %}
2640 
2641 //     fnstsw_ax();
2642 //     sahf();
2643 //     movl(dst, nan_result);
2644 //     jcc(Assembler::parity, exit);
2645 //     movl(dst, less_result);
2646 //     jcc(Assembler::below, exit);
2647 //     movl(dst, equal_result);
2648 //     jcc(Assembler::equal, exit);
2649 //     movl(dst, greater_result);
2650 
2651 // less_result     =  1;
2652 // greater_result  = -1;
2653 // equal_result    = 0;
2654 // nan_result      = -1;
2655 
2656   enc_class CmpF_Result(rRegI dst) %{
2657     // fnstsw_ax();
2658     emit_opcode( cbuf, 0xDF);
2659     emit_opcode( cbuf, 0xE0);
2660     // sahf
2661     emit_opcode( cbuf, 0x9E);
2662     // movl(dst, nan_result);
2663     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2664     emit_d32( cbuf, -1 );
2665     // jcc(Assembler::parity, exit);
2666     emit_opcode( cbuf, 0x7A );
2667     emit_d8    ( cbuf, 0x13 );
2668     // movl(dst, less_result);
2669     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2670     emit_d32( cbuf, -1 );
2671     // jcc(Assembler::below, exit);
2672     emit_opcode( cbuf, 0x72 );
2673     emit_d8    ( cbuf, 0x0C );
2674     // movl(dst, equal_result);
2675     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2676     emit_d32( cbuf, 0 );
2677     // jcc(Assembler::equal, exit);
2678     emit_opcode( cbuf, 0x74 );
2679     emit_d8    ( cbuf, 0x05 );
2680     // movl(dst, greater_result);
2681     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2682     emit_d32( cbuf, 1 );
2683   %}
2684 
2685 
2686   // Compare the longs and set flags
2687   // BROKEN!  Do Not use as-is
2688   enc_class cmpl_test( eRegL src1, eRegL src2 ) %{
2689     // CMP    $src1.hi,$src2.hi
2690     emit_opcode( cbuf, 0x3B );
2691     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
2692     // JNE,s  done
2693     emit_opcode(cbuf,0x75);
2694     emit_d8(cbuf, 2 );
2695     // CMP    $src1.lo,$src2.lo
2696     emit_opcode( cbuf, 0x3B );
2697     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2698 // done:
2699   %}
2700 
2701   enc_class convert_int_long( regL dst, rRegI src ) %{
2702     // mov $dst.lo,$src
2703     int dst_encoding = $dst$$reg;
2704     int src_encoding = $src$$reg;
2705     encode_Copy( cbuf, dst_encoding  , src_encoding );
2706     // mov $dst.hi,$src
2707     encode_Copy( cbuf, HIGH_FROM_LOW(dst_encoding), src_encoding );
2708     // sar $dst.hi,31
2709     emit_opcode( cbuf, 0xC1 );
2710     emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW(dst_encoding) );
2711     emit_d8(cbuf, 0x1F );
2712   %}
2713 
2714   enc_class convert_long_double( eRegL src ) %{
2715     // push $src.hi
2716     emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
2717     // push $src.lo
2718     emit_opcode(cbuf, 0x50+$src$$reg  );
2719     // fild 64-bits at [SP]
2720     emit_opcode(cbuf,0xdf);
2721     emit_d8(cbuf, 0x6C);
2722     emit_d8(cbuf, 0x24);
2723     emit_d8(cbuf, 0x00);
2724     // pop stack
2725     emit_opcode(cbuf, 0x83); // add  SP, #8
2726     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2727     emit_d8(cbuf, 0x8);
2728   %}
2729 
2730   enc_class multiply_con_and_shift_high( eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr ) %{
2731     // IMUL   EDX:EAX,$src1
2732     emit_opcode( cbuf, 0xF7 );
2733     emit_rm( cbuf, 0x3, 0x5, $src1$$reg );
2734     // SAR    EDX,$cnt-32
2735     int shift_count = ((int)$cnt$$constant) - 32;
2736     if (shift_count > 0) {
2737       emit_opcode(cbuf, 0xC1);
2738       emit_rm(cbuf, 0x3, 7, $dst$$reg );
2739       emit_d8(cbuf, shift_count);
2740     }
2741   %}
2742 
2743   // this version doesn't have add sp, 8
2744   enc_class convert_long_double2( eRegL src ) %{
2745     // push $src.hi
2746     emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
2747     // push $src.lo
2748     emit_opcode(cbuf, 0x50+$src$$reg  );
2749     // fild 64-bits at [SP]
2750     emit_opcode(cbuf,0xdf);
2751     emit_d8(cbuf, 0x6C);
2752     emit_d8(cbuf, 0x24);
2753     emit_d8(cbuf, 0x00);
2754   %}
2755 
2756   enc_class long_int_multiply( eADXRegL dst, nadxRegI src) %{
2757     // Basic idea: long = (long)int * (long)int
2758     // IMUL EDX:EAX, src
2759     emit_opcode( cbuf, 0xF7 );
2760     emit_rm( cbuf, 0x3, 0x5, $src$$reg);
2761   %}
2762 
2763   enc_class long_uint_multiply( eADXRegL dst, nadxRegI src) %{
2764     // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
2765     // MUL EDX:EAX, src
2766     emit_opcode( cbuf, 0xF7 );
2767     emit_rm( cbuf, 0x3, 0x4, $src$$reg);
2768   %}
2769 
2770   enc_class long_multiply( eADXRegL dst, eRegL src, rRegI tmp ) %{
2771     // Basic idea: lo(result) = lo(x_lo * y_lo)
2772     //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
2773     // MOV    $tmp,$src.lo
2774     encode_Copy( cbuf, $tmp$$reg, $src$$reg );
2775     // IMUL   $tmp,EDX
2776     emit_opcode( cbuf, 0x0F );
2777     emit_opcode( cbuf, 0xAF );
2778     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2779     // MOV    EDX,$src.hi
2780     encode_Copy( cbuf, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg) );
2781     // IMUL   EDX,EAX
2782     emit_opcode( cbuf, 0x0F );
2783     emit_opcode( cbuf, 0xAF );
2784     emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
2785     // ADD    $tmp,EDX
2786     emit_opcode( cbuf, 0x03 );
2787     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2788     // MUL   EDX:EAX,$src.lo
2789     emit_opcode( cbuf, 0xF7 );
2790     emit_rm( cbuf, 0x3, 0x4, $src$$reg );
2791     // ADD    EDX,ESI
2792     emit_opcode( cbuf, 0x03 );
2793     emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $tmp$$reg );
2794   %}
2795 
2796   enc_class long_multiply_con( eADXRegL dst, immL_127 src, rRegI tmp ) %{
2797     // Basic idea: lo(result) = lo(src * y_lo)
2798     //             hi(result) = hi(src * y_lo) + lo(src * y_hi)
2799     // IMUL   $tmp,EDX,$src
2800     emit_opcode( cbuf, 0x6B );
2801     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2802     emit_d8( cbuf, (int)$src$$constant );
2803     // MOV    EDX,$src
2804     emit_opcode(cbuf, 0xB8 + EDX_enc);
2805     emit_d32( cbuf, (int)$src$$constant );
2806     // MUL   EDX:EAX,EDX
2807     emit_opcode( cbuf, 0xF7 );
2808     emit_rm( cbuf, 0x3, 0x4, EDX_enc );
2809     // ADD    EDX,ESI
2810     emit_opcode( cbuf, 0x03 );
2811     emit_rm( cbuf, 0x3, EDX_enc, $tmp$$reg );
2812   %}
2813 
2814   enc_class long_div( eRegL src1, eRegL src2 ) %{
2815     // PUSH src1.hi
2816     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
2817     // PUSH src1.lo
2818     emit_opcode(cbuf,               0x50+$src1$$reg  );
2819     // PUSH src2.hi
2820     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
2821     // PUSH src2.lo
2822     emit_opcode(cbuf,               0x50+$src2$$reg  );
2823     // CALL directly to the runtime
2824     cbuf.set_insts_mark();
2825     emit_opcode(cbuf,0xE8);       // Call into runtime
2826     emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::ldiv) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
2827     // Restore stack
2828     emit_opcode(cbuf, 0x83); // add  SP, #framesize
2829     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2830     emit_d8(cbuf, 4*4);
2831   %}
2832 
2833   enc_class long_mod( eRegL src1, eRegL src2 ) %{
2834     // PUSH src1.hi
2835     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
2836     // PUSH src1.lo
2837     emit_opcode(cbuf,               0x50+$src1$$reg  );
2838     // PUSH src2.hi
2839     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
2840     // PUSH src2.lo
2841     emit_opcode(cbuf,               0x50+$src2$$reg  );
2842     // CALL directly to the runtime
2843     cbuf.set_insts_mark();
2844     emit_opcode(cbuf,0xE8);       // Call into runtime
2845     emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::lrem ) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
2846     // Restore stack
2847     emit_opcode(cbuf, 0x83); // add  SP, #framesize
2848     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2849     emit_d8(cbuf, 4*4);
2850   %}
2851 
2852   enc_class long_cmp_flags0( eRegL src, rRegI tmp ) %{
2853     // MOV   $tmp,$src.lo
2854     emit_opcode(cbuf, 0x8B);
2855     emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg);
2856     // OR    $tmp,$src.hi
2857     emit_opcode(cbuf, 0x0B);
2858     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg));
2859   %}
2860 
2861   enc_class long_cmp_flags1( eRegL src1, eRegL src2 ) %{
2862     // CMP    $src1.lo,$src2.lo
2863     emit_opcode( cbuf, 0x3B );
2864     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2865     // JNE,s  skip
2866     emit_cc(cbuf, 0x70, 0x5);
2867     emit_d8(cbuf,2);
2868     // CMP    $src1.hi,$src2.hi
2869     emit_opcode( cbuf, 0x3B );
2870     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
2871   %}
2872 
2873   enc_class long_cmp_flags2( eRegL src1, eRegL src2, rRegI tmp ) %{
2874     // CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits
2875     emit_opcode( cbuf, 0x3B );
2876     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2877     // MOV    $tmp,$src1.hi
2878     emit_opcode( cbuf, 0x8B );
2879     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src1$$reg) );
2880     // SBB   $tmp,$src2.hi\t! Compute flags for long compare
2881     emit_opcode( cbuf, 0x1B );
2882     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src2$$reg) );
2883   %}
2884 
2885   enc_class long_cmp_flags3( eRegL src, rRegI tmp ) %{
2886     // XOR    $tmp,$tmp
2887     emit_opcode(cbuf,0x33);  // XOR
2888     emit_rm(cbuf,0x3, $tmp$$reg, $tmp$$reg);
2889     // CMP    $tmp,$src.lo
2890     emit_opcode( cbuf, 0x3B );
2891     emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg );
2892     // SBB    $tmp,$src.hi
2893     emit_opcode( cbuf, 0x1B );
2894     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg) );
2895   %}
2896 
2897  // Sniff, sniff... smells like Gnu Superoptimizer
2898   enc_class neg_long( eRegL dst ) %{
2899     emit_opcode(cbuf,0xF7);    // NEG hi
2900     emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
2901     emit_opcode(cbuf,0xF7);    // NEG lo
2902     emit_rm    (cbuf,0x3, 0x3,               $dst$$reg );
2903     emit_opcode(cbuf,0x83);    // SBB hi,0
2904     emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
2905     emit_d8    (cbuf,0 );
2906   %}
2907 
2908 
2909   // Because the transitions from emitted code to the runtime
2910   // monitorenter/exit helper stubs are so slow it's critical that
2911   // we inline both the stack-locking fast-path and the inflated fast path.
2912   //
2913   // See also: cmpFastLock and cmpFastUnlock.
2914   //
2915   // What follows is a specialized inline transliteration of the code
2916   // in slow_enter() and slow_exit().  If we're concerned about I$ bloat
2917   // another option would be to emit TrySlowEnter and TrySlowExit methods
2918   // at startup-time.  These methods would accept arguments as
2919   // (rax,=Obj, rbx=Self, rcx=box, rdx=Scratch) and return success-failure
2920   // indications in the icc.ZFlag.  Fast_Lock and Fast_Unlock would simply
2921   // marshal the arguments and emit calls to TrySlowEnter and TrySlowExit.
2922   // In practice, however, the # of lock sites is bounded and is usually small.
2923   // Besides the call overhead, TrySlowEnter and TrySlowExit might suffer
2924   // if the processor uses simple bimodal branch predictors keyed by EIP
2925   // Since the helper routines would be called from multiple synchronization
2926   // sites.
2927   //
2928   // An even better approach would be write "MonitorEnter()" and "MonitorExit()"
2929   // in java - using j.u.c and unsafe - and just bind the lock and unlock sites
2930   // to those specialized methods.  That'd give us a mostly platform-independent
2931   // implementation that the JITs could optimize and inline at their pleasure.
2932   // Done correctly, the only time we'd need to cross to native could would be
2933   // to park() or unpark() threads.  We'd also need a few more unsafe operators
2934   // to (a) prevent compiler-JIT reordering of non-volatile accesses, and
2935   // (b) explicit barriers or fence operations.
2936   //
2937   // TODO:
2938   //
2939   // *  Arrange for C2 to pass "Self" into Fast_Lock and Fast_Unlock in one of the registers (scr).
2940   //    This avoids manifesting the Self pointer in the Fast_Lock and Fast_Unlock terminals.
2941   //    Given TLAB allocation, Self is usually manifested in a register, so passing it into
2942   //    the lock operators would typically be faster than reifying Self.
2943   //
2944   // *  Ideally I'd define the primitives as:
2945   //       fast_lock   (nax Obj, nax box, EAX tmp, nax scr) where box, tmp and scr are KILLED.
2946   //       fast_unlock (nax Obj, EAX box, nax tmp) where box and tmp are KILLED
2947   //    Unfortunately ADLC bugs prevent us from expressing the ideal form.
2948   //    Instead, we're stuck with a rather awkward and brittle register assignments below.
2949   //    Furthermore the register assignments are overconstrained, possibly resulting in
2950   //    sub-optimal code near the synchronization site.
2951   //
2952   // *  Eliminate the sp-proximity tests and just use "== Self" tests instead.
2953   //    Alternately, use a better sp-proximity test.
2954   //
2955   // *  Currently ObjectMonitor._Owner can hold either an sp value or a (THREAD *) value.
2956   //    Either one is sufficient to uniquely identify a thread.
2957   //    TODO: eliminate use of sp in _owner and use get_thread(tr) instead.
2958   //
2959   // *  Intrinsify notify() and notifyAll() for the common cases where the
2960   //    object is locked by the calling thread but the waitlist is empty.
2961   //    avoid the expensive JNI call to JVM_Notify() and JVM_NotifyAll().
2962   //
2963   // *  use jccb and jmpb instead of jcc and jmp to improve code density.
2964   //    But beware of excessive branch density on AMD Opterons.
2965   //
2966   // *  Both Fast_Lock and Fast_Unlock set the ICC.ZF to indicate success
2967   //    or failure of the fast-path.  If the fast-path fails then we pass
2968   //    control to the slow-path, typically in C.  In Fast_Lock and
2969   //    Fast_Unlock we often branch to DONE_LABEL, just to find that C2
2970   //    will emit a conditional branch immediately after the node.
2971   //    So we have branches to branches and lots of ICC.ZF games.
2972   //    Instead, it might be better to have C2 pass a "FailureLabel"
2973   //    into Fast_Lock and Fast_Unlock.  In the case of success, control
2974   //    will drop through the node.  ICC.ZF is undefined at exit.
2975   //    In the case of failure, the node will branch directly to the
2976   //    FailureLabel
2977 
2978 
2979   // obj: object to lock
2980   // box: on-stack box address (displaced header location) - KILLED
2981   // rax,: tmp -- KILLED
2982   // scr: tmp -- KILLED
2983   enc_class Fast_Lock( eRegP obj, eRegP box, eAXRegI tmp, eRegP scr ) %{
2984 
2985     Register objReg = as_Register($obj$$reg);
2986     Register boxReg = as_Register($box$$reg);
2987     Register tmpReg = as_Register($tmp$$reg);
2988     Register scrReg = as_Register($scr$$reg);
2989 
2990     // Ensure the register assignents are disjoint
2991     guarantee (objReg != boxReg, "") ;
2992     guarantee (objReg != tmpReg, "") ;
2993     guarantee (objReg != scrReg, "") ;
2994     guarantee (boxReg != tmpReg, "") ;
2995     guarantee (boxReg != scrReg, "") ;
2996     guarantee (tmpReg == as_Register(EAX_enc), "") ;
2997 
2998     MacroAssembler masm(&cbuf);
2999 
3000     if (_counters != NULL) {
3001       masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3002     }
3003     if (EmitSync & 1) {
3004         // set box->dhw = unused_mark (3)
3005         // Force all sync thru slow-path: slow_enter() and slow_exit() 
3006         masm.movptr (Address(boxReg, 0), int32_t(markOopDesc::unused_mark())) ;             
3007         masm.cmpptr (rsp, (int32_t)0) ;                        
3008     } else 
3009     if (EmitSync & 2) { 
3010         Label DONE_LABEL ;           
3011         if (UseBiasedLocking) {
3012            // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3013            masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3014         }
3015 
3016         masm.movptr(tmpReg, Address(objReg, 0)) ;          // fetch markword 
3017         masm.orptr (tmpReg, 0x1);
3018         masm.movptr(Address(boxReg, 0), tmpReg);           // Anticipate successful CAS 
3019         if (os::is_MP()) { masm.lock();  }
3020         masm.cmpxchgptr(boxReg, Address(objReg, 0));          // Updates tmpReg
3021         masm.jcc(Assembler::equal, DONE_LABEL);
3022         // Recursive locking
3023         masm.subptr(tmpReg, rsp);
3024         masm.andptr(tmpReg, (int32_t) 0xFFFFF003 );
3025         masm.movptr(Address(boxReg, 0), tmpReg);
3026         masm.bind(DONE_LABEL) ; 
3027     } else {  
3028       // Possible cases that we'll encounter in fast_lock 
3029       // ------------------------------------------------
3030       // * Inflated
3031       //    -- unlocked
3032       //    -- Locked
3033       //       = by self
3034       //       = by other
3035       // * biased
3036       //    -- by Self
3037       //    -- by other
3038       // * neutral
3039       // * stack-locked
3040       //    -- by self
3041       //       = sp-proximity test hits
3042       //       = sp-proximity test generates false-negative
3043       //    -- by other
3044       //
3045 
3046       Label IsInflated, DONE_LABEL, PopDone ;
3047 
3048       // TODO: optimize away redundant LDs of obj->mark and improve the markword triage
3049       // order to reduce the number of conditional branches in the most common cases.
3050       // Beware -- there's a subtle invariant that fetch of the markword
3051       // at [FETCH], below, will never observe a biased encoding (*101b).
3052       // If this invariant is not held we risk exclusion (safety) failure.
3053       if (UseBiasedLocking && !UseOptoBiasInlining) {
3054         masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3055       }
3056 
3057       masm.movptr(tmpReg, Address(objReg, 0)) ;         // [FETCH]
3058       masm.testptr(tmpReg, 0x02) ;                      // Inflated v (Stack-locked or neutral)
3059       masm.jccb  (Assembler::notZero, IsInflated) ;
3060 
3061       // Attempt stack-locking ...
3062       masm.orptr (tmpReg, 0x1);
3063       masm.movptr(Address(boxReg, 0), tmpReg);          // Anticipate successful CAS
3064       if (os::is_MP()) { masm.lock();  }
3065       masm.cmpxchgptr(boxReg, Address(objReg, 0));           // Updates tmpReg
3066       if (_counters != NULL) {
3067         masm.cond_inc32(Assembler::equal,
3068                         ExternalAddress((address)_counters->fast_path_entry_count_addr()));
3069       }
3070       masm.jccb (Assembler::equal, DONE_LABEL);
3071 
3072       // Recursive locking
3073       masm.subptr(tmpReg, rsp);
3074       masm.andptr(tmpReg, 0xFFFFF003 );
3075       masm.movptr(Address(boxReg, 0), tmpReg);
3076       if (_counters != NULL) {
3077         masm.cond_inc32(Assembler::equal,
3078                         ExternalAddress((address)_counters->fast_path_entry_count_addr()));
3079       }
3080       masm.jmp  (DONE_LABEL) ;
3081 
3082       masm.bind (IsInflated) ;
3083 
3084       // The object is inflated.
3085       //
3086       // TODO-FIXME: eliminate the ugly use of manifest constants:
3087       //   Use markOopDesc::monitor_value instead of "2".
3088       //   use markOop::unused_mark() instead of "3".
3089       // The tmpReg value is an objectMonitor reference ORed with
3090       // markOopDesc::monitor_value (2).   We can either convert tmpReg to an
3091       // objectmonitor pointer by masking off the "2" bit or we can just
3092       // use tmpReg as an objectmonitor pointer but bias the objectmonitor
3093       // field offsets with "-2" to compensate for and annul the low-order tag bit.
3094       //
3095       // I use the latter as it avoids AGI stalls.
3096       // As such, we write "mov r, [tmpReg+OFFSETOF(Owner)-2]"
3097       // instead of "mov r, [tmpReg+OFFSETOF(Owner)]".
3098       //
3099       #define OFFSET_SKEWED(f) ((ObjectMonitor::f ## _offset_in_bytes())-2)
3100 
3101       // boxReg refers to the on-stack BasicLock in the current frame.
3102       // We'd like to write:
3103       //   set box->_displaced_header = markOop::unused_mark().  Any non-0 value suffices.
3104       // This is convenient but results a ST-before-CAS penalty.  The following CAS suffers
3105       // additional latency as we have another ST in the store buffer that must drain.
3106 
3107       if (EmitSync & 8192) { 
3108          masm.movptr(Address(boxReg, 0), 3) ;            // results in ST-before-CAS penalty
3109          masm.get_thread (scrReg) ; 
3110          masm.movptr(boxReg, tmpReg);                    // consider: LEA box, [tmp-2] 
3111          masm.movptr(tmpReg, NULL_WORD);                 // consider: xor vs mov
3112          if (os::is_MP()) { masm.lock(); } 
3113          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3114       } else 
3115       if ((EmitSync & 128) == 0) {                      // avoid ST-before-CAS
3116          masm.movptr(scrReg, boxReg) ; 
3117          masm.movptr(boxReg, tmpReg);                   // consider: LEA box, [tmp-2] 
3118 
3119          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
3120          if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3121             // prefetchw [eax + Offset(_owner)-2]
3122             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
3123          }
3124 
3125          if ((EmitSync & 64) == 0) {
3126            // Optimistic form: consider XORL tmpReg,tmpReg
3127            masm.movptr(tmpReg, NULL_WORD) ; 
3128          } else { 
3129            // Can suffer RTS->RTO upgrades on shared or cold $ lines
3130            // Test-And-CAS instead of CAS
3131            masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
3132            masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
3133            masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
3134          }
3135 
3136          // Appears unlocked - try to swing _owner from null to non-null.
3137          // Ideally, I'd manifest "Self" with get_thread and then attempt
3138          // to CAS the register containing Self into m->Owner.
3139          // But we don't have enough registers, so instead we can either try to CAS
3140          // rsp or the address of the box (in scr) into &m->owner.  If the CAS succeeds
3141          // we later store "Self" into m->Owner.  Transiently storing a stack address
3142          // (rsp or the address of the box) into  m->owner is harmless.
3143          // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
3144          if (os::is_MP()) { masm.lock();  }
3145          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3146          masm.movptr(Address(scrReg, 0), 3) ;          // box->_displaced_header = 3
3147          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3148          masm.get_thread (scrReg) ;                    // beware: clobbers ICCs
3149          masm.movptr(Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2), scrReg) ; 
3150          masm.xorptr(boxReg, boxReg) ;                 // set icc.ZFlag = 1 to indicate success
3151                        
3152          // If the CAS fails we can either retry or pass control to the slow-path.  
3153          // We use the latter tactic.  
3154          // Pass the CAS result in the icc.ZFlag into DONE_LABEL
3155          // If the CAS was successful ...
3156          //   Self has acquired the lock
3157          //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
3158          // Intentional fall-through into DONE_LABEL ...
3159       } else {
3160          masm.movptr(Address(boxReg, 0), 3) ;       // results in ST-before-CAS penalty
3161          masm.movptr(boxReg, tmpReg) ; 
3162 
3163          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
3164          if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3165             // prefetchw [eax + Offset(_owner)-2]
3166             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
3167          }
3168 
3169          if ((EmitSync & 64) == 0) {
3170            // Optimistic form
3171            masm.xorptr  (tmpReg, tmpReg) ; 
3172          } else { 
3173            // Can suffer RTS->RTO upgrades on shared or cold $ lines
3174            masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
3175            masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
3176            masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
3177          }
3178 
3179          // Appears unlocked - try to swing _owner from null to non-null.
3180          // Use either "Self" (in scr) or rsp as thread identity in _owner.
3181          // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
3182          masm.get_thread (scrReg) ;
3183          if (os::is_MP()) { masm.lock(); }
3184          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3185 
3186          // If the CAS fails we can either retry or pass control to the slow-path.
3187          // We use the latter tactic.
3188          // Pass the CAS result in the icc.ZFlag into DONE_LABEL
3189          // If the CAS was successful ...
3190          //   Self has acquired the lock
3191          //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
3192          // Intentional fall-through into DONE_LABEL ...
3193       }
3194 
3195       // DONE_LABEL is a hot target - we'd really like to place it at the
3196       // start of cache line by padding with NOPs.
3197       // See the AMD and Intel software optimization manuals for the
3198       // most efficient "long" NOP encodings.
3199       // Unfortunately none of our alignment mechanisms suffice.
3200       masm.bind(DONE_LABEL);
3201 
3202       // Avoid branch-to-branch on AMD processors
3203       // This appears to be superstition.
3204       if (EmitSync & 32) masm.nop() ;
3205 
3206 
3207       // At DONE_LABEL the icc ZFlag is set as follows ...
3208       // Fast_Unlock uses the same protocol.
3209       // ZFlag == 1 -> Success
3210       // ZFlag == 0 -> Failure - force control through the slow-path
3211     }
3212   %}
3213 
3214   // obj: object to unlock
3215   // box: box address (displaced header location), killed.  Must be EAX.
3216   // rbx,: killed tmp; cannot be obj nor box.
3217   //
3218   // Some commentary on balanced locking:
3219   //
3220   // Fast_Lock and Fast_Unlock are emitted only for provably balanced lock sites.
3221   // Methods that don't have provably balanced locking are forced to run in the
3222   // interpreter - such methods won't be compiled to use fast_lock and fast_unlock.
3223   // The interpreter provides two properties:
3224   // I1:  At return-time the interpreter automatically and quietly unlocks any
3225   //      objects acquired the current activation (frame).  Recall that the
3226   //      interpreter maintains an on-stack list of locks currently held by
3227   //      a frame.
3228   // I2:  If a method attempts to unlock an object that is not held by the
3229   //      the frame the interpreter throws IMSX.
3230   //
3231   // Lets say A(), which has provably balanced locking, acquires O and then calls B().
3232   // B() doesn't have provably balanced locking so it runs in the interpreter.
3233   // Control returns to A() and A() unlocks O.  By I1 and I2, above, we know that O
3234   // is still locked by A().
3235   //
3236   // The only other source of unbalanced locking would be JNI.  The "Java Native Interface:
3237   // Programmer's Guide and Specification" claims that an object locked by jni_monitorenter
3238   // should not be unlocked by "normal" java-level locking and vice-versa.  The specification
3239   // doesn't specify what will occur if a program engages in such mixed-mode locking, however.
3240 
3241   enc_class Fast_Unlock( nabxRegP obj, eAXRegP box, eRegP tmp) %{
3242 
3243     Register objReg = as_Register($obj$$reg);
3244     Register boxReg = as_Register($box$$reg);
3245     Register tmpReg = as_Register($tmp$$reg);
3246 
3247     guarantee (objReg != boxReg, "") ;
3248     guarantee (objReg != tmpReg, "") ;
3249     guarantee (boxReg != tmpReg, "") ;
3250     guarantee (boxReg == as_Register(EAX_enc), "") ;
3251     MacroAssembler masm(&cbuf);
3252 
3253     if (EmitSync & 4) {
3254       // Disable - inhibit all inlining.  Force control through the slow-path
3255       masm.cmpptr (rsp, 0) ; 
3256     } else 
3257     if (EmitSync & 8) {
3258       Label DONE_LABEL ;
3259       if (UseBiasedLocking) {
3260          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3261       }
3262       // classic stack-locking code ...
3263       masm.movptr(tmpReg, Address(boxReg, 0)) ;
3264       masm.testptr(tmpReg, tmpReg) ;
3265       masm.jcc   (Assembler::zero, DONE_LABEL) ;
3266       if (os::is_MP()) { masm.lock(); }
3267       masm.cmpxchgptr(tmpReg, Address(objReg, 0));          // Uses EAX which is box
3268       masm.bind(DONE_LABEL);
3269     } else {
3270       Label DONE_LABEL, Stacked, CheckSucc, Inflated ;
3271 
3272       // Critically, the biased locking test must have precedence over
3273       // and appear before the (box->dhw == 0) recursive stack-lock test.
3274       if (UseBiasedLocking && !UseOptoBiasInlining) {
3275          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3276       }
3277       
3278       masm.cmpptr(Address(boxReg, 0), 0) ;            // Examine the displaced header
3279       masm.movptr(tmpReg, Address(objReg, 0)) ;       // Examine the object's markword
3280       masm.jccb  (Assembler::zero, DONE_LABEL) ;      // 0 indicates recursive stack-lock
3281 
3282       masm.testptr(tmpReg, 0x02) ;                     // Inflated? 
3283       masm.jccb  (Assembler::zero, Stacked) ;
3284 
3285       masm.bind  (Inflated) ;
3286       // It's inflated.
3287       // Despite our balanced locking property we still check that m->_owner == Self
3288       // as java routines or native JNI code called by this thread might
3289       // have released the lock.
3290       // Refer to the comments in synchronizer.cpp for how we might encode extra
3291       // state in _succ so we can avoid fetching EntryList|cxq.
3292       //
3293       // I'd like to add more cases in fast_lock() and fast_unlock() --
3294       // such as recursive enter and exit -- but we have to be wary of
3295       // I$ bloat, T$ effects and BP$ effects.
3296       //
3297       // If there's no contention try a 1-0 exit.  That is, exit without
3298       // a costly MEMBAR or CAS.  See synchronizer.cpp for details on how
3299       // we detect and recover from the race that the 1-0 exit admits.
3300       //
3301       // Conceptually Fast_Unlock() must execute a STST|LDST "release" barrier
3302       // before it STs null into _owner, releasing the lock.  Updates
3303       // to data protected by the critical section must be visible before
3304       // we drop the lock (and thus before any other thread could acquire
3305       // the lock and observe the fields protected by the lock).
3306       // IA32's memory-model is SPO, so STs are ordered with respect to
3307       // each other and there's no need for an explicit barrier (fence).
3308       // See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
3309 
3310       masm.get_thread (boxReg) ;
3311       if ((EmitSync & 4096) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3312         // prefetchw [ebx + Offset(_owner)-2]
3313         masm.prefetchw(Address(rbx, ObjectMonitor::owner_offset_in_bytes()-2));
3314       }
3315 
3316       // Note that we could employ various encoding schemes to reduce
3317       // the number of loads below (currently 4) to just 2 or 3.
3318       // Refer to the comments in synchronizer.cpp.
3319       // In practice the chain of fetches doesn't seem to impact performance, however.
3320       if ((EmitSync & 65536) == 0 && (EmitSync & 256)) {
3321          // Attempt to reduce branch density - AMD's branch predictor.
3322          masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
3323          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3324          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
3325          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
3326          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3327          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3328          masm.jmpb  (DONE_LABEL) ; 
3329       } else { 
3330          masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
3331          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3332          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3333          masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
3334          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
3335          masm.jccb  (Assembler::notZero, CheckSucc) ; 
3336          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3337          masm.jmpb  (DONE_LABEL) ; 
3338       }
3339 
3340       // The Following code fragment (EmitSync & 65536) improves the performance of
3341       // contended applications and contended synchronization microbenchmarks.
3342       // Unfortunately the emission of the code - even though not executed - causes regressions
3343       // in scimark and jetstream, evidently because of $ effects.  Replacing the code
3344       // with an equal number of never-executed NOPs results in the same regression.
3345       // We leave it off by default.
3346 
3347       if ((EmitSync & 65536) != 0) {
3348          Label LSuccess, LGoSlowPath ;
3349 
3350          masm.bind  (CheckSucc) ;
3351 
3352          // Optional pre-test ... it's safe to elide this
3353          if ((EmitSync & 16) == 0) { 
3354             masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
3355             masm.jccb  (Assembler::zero, LGoSlowPath) ; 
3356          }
3357 
3358          // We have a classic Dekker-style idiom:
3359          //    ST m->_owner = 0 ; MEMBAR; LD m->_succ
3360          // There are a number of ways to implement the barrier:
3361          // (1) lock:andl &m->_owner, 0
3362          //     is fast, but mask doesn't currently support the "ANDL M,IMM32" form.
3363          //     LOCK: ANDL [ebx+Offset(_Owner)-2], 0
3364          //     Encodes as 81 31 OFF32 IMM32 or 83 63 OFF8 IMM8
3365          // (2) If supported, an explicit MFENCE is appealing.
3366          //     In older IA32 processors MFENCE is slower than lock:add or xchg
3367          //     particularly if the write-buffer is full as might be the case if
3368          //     if stores closely precede the fence or fence-equivalent instruction.
3369          //     In more modern implementations MFENCE appears faster, however.
3370          // (3) In lieu of an explicit fence, use lock:addl to the top-of-stack
3371          //     The $lines underlying the top-of-stack should be in M-state.
3372          //     The locked add instruction is serializing, of course.
3373          // (4) Use xchg, which is serializing
3374          //     mov boxReg, 0; xchgl boxReg, [tmpReg + Offset(_owner)-2] also works
3375          // (5) ST m->_owner = 0 and then execute lock:orl &m->_succ, 0.
3376          //     The integer condition codes will tell us if succ was 0.
3377          //     Since _succ and _owner should reside in the same $line and
3378          //     we just stored into _owner, it's likely that the $line
3379          //     remains in M-state for the lock:orl.
3380          //
3381          // We currently use (3), although it's likely that switching to (2)
3382          // is correct for the future.
3383             
3384          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3385          if (os::is_MP()) { 
3386             if (VM_Version::supports_sse2() && 1 == FenceInstruction) { 
3387               masm.mfence();
3388             } else { 
3389               masm.lock () ; masm.addptr(Address(rsp, 0), 0) ; 
3390             }
3391          }
3392          // Ratify _succ remains non-null
3393          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
3394          masm.jccb  (Assembler::notZero, LSuccess) ; 
3395 
3396          masm.xorptr(boxReg, boxReg) ;                  // box is really EAX
3397          if (os::is_MP()) { masm.lock(); }
3398          masm.cmpxchgptr(rsp, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3399          masm.jccb  (Assembler::notEqual, LSuccess) ;
3400          // Since we're low on registers we installed rsp as a placeholding in _owner.
3401          // Now install Self over rsp.  This is safe as we're transitioning from
3402          // non-null to non=null
3403          masm.get_thread (boxReg) ;
3404          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), boxReg) ;
3405          // Intentional fall-through into LGoSlowPath ...
3406 
3407          masm.bind  (LGoSlowPath) ; 
3408          masm.orptr(boxReg, 1) ;                      // set ICC.ZF=0 to indicate failure
3409          masm.jmpb  (DONE_LABEL) ; 
3410 
3411          masm.bind  (LSuccess) ; 
3412          masm.xorptr(boxReg, boxReg) ;                 // set ICC.ZF=1 to indicate success
3413          masm.jmpb  (DONE_LABEL) ; 
3414       }
3415 
3416       masm.bind (Stacked) ;
3417       // It's not inflated and it's not recursively stack-locked and it's not biased.
3418       // It must be stack-locked.
3419       // Try to reset the header to displaced header.
3420       // The "box" value on the stack is stable, so we can reload
3421       // and be assured we observe the same value as above.
3422       masm.movptr(tmpReg, Address(boxReg, 0)) ;
3423       if (os::is_MP()) {   masm.lock();    }
3424       masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses EAX which is box
3425       // Intention fall-thru into DONE_LABEL
3426 
3427 
3428       // DONE_LABEL is a hot target - we'd really like to place it at the
3429       // start of cache line by padding with NOPs.
3430       // See the AMD and Intel software optimization manuals for the
3431       // most efficient "long" NOP encodings.
3432       // Unfortunately none of our alignment mechanisms suffice.
3433       if ((EmitSync & 65536) == 0) {
3434          masm.bind (CheckSucc) ;
3435       }
3436       masm.bind(DONE_LABEL);
3437 
3438       // Avoid branch to branch on AMD processors
3439       if (EmitSync & 32768) { masm.nop() ; }
3440     }
3441   %}
3442 
3443 
3444   enc_class enc_pop_rdx() %{
3445     emit_opcode(cbuf,0x5A);
3446   %}
3447 
3448   enc_class enc_rethrow() %{
3449     cbuf.set_insts_mark();
3450     emit_opcode(cbuf, 0xE9);        // jmp    entry
3451     emit_d32_reloc(cbuf, (int)OptoRuntime::rethrow_stub() - ((int)cbuf.insts_end())-4,
3452                    runtime_call_Relocation::spec(), RELOC_IMM32 );
3453   %}
3454 
3455 
3456   // Convert a double to an int.  Java semantics require we do complex
3457   // manglelations in the corner cases.  So we set the rounding mode to
3458   // 'zero', store the darned double down as an int, and reset the
3459   // rounding mode to 'nearest'.  The hardware throws an exception which
3460   // patches up the correct value directly to the stack.
3461   enc_class DPR2I_encoding( regDPR src ) %{
3462     // Flip to round-to-zero mode.  We attempted to allow invalid-op
3463     // exceptions here, so that a NAN or other corner-case value will
3464     // thrown an exception (but normal values get converted at full speed).
3465     // However, I2C adapters and other float-stack manglers leave pending
3466     // invalid-op exceptions hanging.  We would have to clear them before
3467     // enabling them and that is more expensive than just testing for the
3468     // invalid value Intel stores down in the corner cases.
3469     emit_opcode(cbuf,0xD9);            // FLDCW  trunc
3470     emit_opcode(cbuf,0x2D);
3471     emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
3472     // Allocate a word
3473     emit_opcode(cbuf,0x83);            // SUB ESP,4
3474     emit_opcode(cbuf,0xEC);
3475     emit_d8(cbuf,0x04);
3476     // Encoding assumes a double has been pushed into FPR0.
3477     // Store down the double as an int, popping the FPU stack
3478     emit_opcode(cbuf,0xDB);            // FISTP [ESP]
3479     emit_opcode(cbuf,0x1C);
3480     emit_d8(cbuf,0x24);
3481     // Restore the rounding mode; mask the exception
3482     emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
3483     emit_opcode(cbuf,0x2D);
3484     emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
3485         ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
3486         : (int)StubRoutines::addr_fpu_cntrl_wrd_std());
3487 
3488     // Load the converted int; adjust CPU stack
3489     emit_opcode(cbuf,0x58);       // POP EAX
3490     emit_opcode(cbuf,0x3D);       // CMP EAX,imm
3491     emit_d32   (cbuf,0x80000000); //         0x80000000
3492     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3493     emit_d8    (cbuf,0x07);       // Size of slow_call
3494     // Push src onto stack slow-path
3495     emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
3496     emit_d8    (cbuf,0xC0-1+$src$$reg );
3497     // CALL directly to the runtime
3498     cbuf.set_insts_mark();
3499     emit_opcode(cbuf,0xE8);       // Call into runtime
3500     emit_d32_reloc(cbuf, (StubRoutines::d2i_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
3501     // Carry on here...
3502   %}
3503 
3504   enc_class DPR2L_encoding( regDPR src ) %{
3505     emit_opcode(cbuf,0xD9);            // FLDCW  trunc
3506     emit_opcode(cbuf,0x2D);
3507     emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
3508     // Allocate a word
3509     emit_opcode(cbuf,0x83);            // SUB ESP,8
3510     emit_opcode(cbuf,0xEC);
3511     emit_d8(cbuf,0x08);
3512     // Encoding assumes a double has been pushed into FPR0.
3513     // Store down the double as a long, popping the FPU stack
3514     emit_opcode(cbuf,0xDF);            // FISTP [ESP]
3515     emit_opcode(cbuf,0x3C);
3516     emit_d8(cbuf,0x24);
3517     // Restore the rounding mode; mask the exception
3518     emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
3519     emit_opcode(cbuf,0x2D);
3520     emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
3521         ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
3522         : (int)StubRoutines::addr_fpu_cntrl_wrd_std());
3523 
3524     // Load the converted int; adjust CPU stack
3525     emit_opcode(cbuf,0x58);       // POP EAX
3526     emit_opcode(cbuf,0x5A);       // POP EDX
3527     emit_opcode(cbuf,0x81);       // CMP EDX,imm
3528     emit_d8    (cbuf,0xFA);       // rdx
3529     emit_d32   (cbuf,0x80000000); //         0x80000000
3530     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3531     emit_d8    (cbuf,0x07+4);     // Size of slow_call
3532     emit_opcode(cbuf,0x85);       // TEST EAX,EAX
3533     emit_opcode(cbuf,0xC0);       // 2/rax,/rax,
3534     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3535     emit_d8    (cbuf,0x07);       // Size of slow_call
3536     // Push src onto stack slow-path
3537     emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
3538     emit_d8    (cbuf,0xC0-1+$src$$reg );
3539     // CALL directly to the runtime
3540     cbuf.set_insts_mark();
3541     emit_opcode(cbuf,0xE8);       // Call into runtime
3542     emit_d32_reloc(cbuf, (StubRoutines::d2l_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
3543     // Carry on here...
3544   %}
3545 
3546   enc_class FMul_ST_reg( eRegFPR src1 ) %{
3547     // Operand was loaded from memory into fp ST (stack top)
3548     // FMUL   ST,$src  /* D8 C8+i */
3549     emit_opcode(cbuf, 0xD8);
3550     emit_opcode(cbuf, 0xC8 + $src1$$reg);
3551   %}
3552 
3553   enc_class FAdd_ST_reg( eRegFPR src2 ) %{
3554     // FADDP  ST,src2  /* D8 C0+i */
3555     emit_opcode(cbuf, 0xD8);
3556     emit_opcode(cbuf, 0xC0 + $src2$$reg);
3557     //could use FADDP  src2,fpST  /* DE C0+i */
3558   %}
3559 
3560   enc_class FAddP_reg_ST( eRegFPR src2 ) %{
3561     // FADDP  src2,ST  /* DE C0+i */
3562     emit_opcode(cbuf, 0xDE);
3563     emit_opcode(cbuf, 0xC0 + $src2$$reg);
3564   %}
3565 
3566   enc_class subFPR_divFPR_encode( eRegFPR src1, eRegFPR src2) %{
3567     // Operand has been loaded into fp ST (stack top)
3568       // FSUB   ST,$src1
3569       emit_opcode(cbuf, 0xD8);
3570       emit_opcode(cbuf, 0xE0 + $src1$$reg);
3571 
3572       // FDIV
3573       emit_opcode(cbuf, 0xD8);
3574       emit_opcode(cbuf, 0xF0 + $src2$$reg);
3575   %}
3576 
3577   enc_class MulFAddF (eRegFPR src1, eRegFPR src2) %{
3578     // Operand was loaded from memory into fp ST (stack top)
3579     // FADD   ST,$src  /* D8 C0+i */
3580     emit_opcode(cbuf, 0xD8);
3581     emit_opcode(cbuf, 0xC0 + $src1$$reg);
3582 
3583     // FMUL  ST,src2  /* D8 C*+i */
3584     emit_opcode(cbuf, 0xD8);
3585     emit_opcode(cbuf, 0xC8 + $src2$$reg);
3586   %}
3587 
3588 
3589   enc_class MulFAddFreverse (eRegFPR src1, eRegFPR src2) %{
3590     // Operand was loaded from memory into fp ST (stack top)
3591     // FADD   ST,$src  /* D8 C0+i */
3592     emit_opcode(cbuf, 0xD8);
3593     emit_opcode(cbuf, 0xC0 + $src1$$reg);
3594 
3595     // FMULP  src2,ST  /* DE C8+i */
3596     emit_opcode(cbuf, 0xDE);
3597     emit_opcode(cbuf, 0xC8 + $src2$$reg);
3598   %}
3599 
3600   // Atomically load the volatile long
3601   enc_class enc_loadL_volatile( memory mem, stackSlotL dst ) %{
3602     emit_opcode(cbuf,0xDF);
3603     int rm_byte_opcode = 0x05;
3604     int base     = $mem$$base;
3605     int index    = $mem$$index;
3606     int scale    = $mem$$scale;
3607     int displace = $mem$$disp;
3608     relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
3609     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
3610     store_to_stackslot( cbuf, 0x0DF, 0x07, $dst$$disp );
3611   %}
3612 
3613   // Volatile Store Long.  Must be atomic, so move it into
3614   // the FP TOS and then do a 64-bit FIST.  Has to probe the
3615   // target address before the store (for null-ptr checks)
3616   // so the memory operand is used twice in the encoding.
3617   enc_class enc_storeL_volatile( memory mem, stackSlotL src ) %{
3618     store_to_stackslot( cbuf, 0x0DF, 0x05, $src$$disp );
3619     cbuf.set_insts_mark();            // Mark start of FIST in case $mem has an oop
3620     emit_opcode(cbuf,0xDF);
3621     int rm_byte_opcode = 0x07;
3622     int base     = $mem$$base;
3623     int index    = $mem$$index;
3624     int scale    = $mem$$scale;
3625     int displace = $mem$$disp;
3626     relocInfo::relocType disp_reloc = $mem->disp_reloc(); // disp-as-oop when working with static globals
3627     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_reloc);
3628   %}
3629 
3630   // Safepoint Poll.  This polls the safepoint page, and causes an
3631   // exception if it is not readable. Unfortunately, it kills the condition code
3632   // in the process
3633   // We current use TESTL [spp],EDI
3634   // A better choice might be TESTB [spp + pagesize() - CacheLineSize()],0
3635 
3636   enc_class Safepoint_Poll() %{
3637     cbuf.relocate(cbuf.insts_mark(), relocInfo::poll_type, 0);
3638     emit_opcode(cbuf,0x85);
3639     emit_rm (cbuf, 0x0, 0x7, 0x5);
3640     emit_d32(cbuf, (intptr_t)os::get_polling_page());
3641   %}
3642 %}
3643 
3644 
3645 //----------FRAME--------------------------------------------------------------
3646 // Definition of frame structure and management information.
3647 //
3648 //  S T A C K   L A Y O U T    Allocators stack-slot number
3649 //                             |   (to get allocators register number
3650 //  G  Owned by    |        |  v    add OptoReg::stack0())
3651 //  r   CALLER     |        |
3652 //  o     |        +--------+      pad to even-align allocators stack-slot
3653 //  w     V        |  pad0  |        numbers; owned by CALLER
3654 //  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
3655 //  h     ^        |   in   |  5
3656 //        |        |  args  |  4   Holes in incoming args owned by SELF
3657 //  |     |        |        |  3
3658 //  |     |        +--------+
3659 //  V     |        | old out|      Empty on Intel, window on Sparc
3660 //        |    old |preserve|      Must be even aligned.
3661 //        |     SP-+--------+----> Matcher::_old_SP, even aligned
3662 //        |        |   in   |  3   area for Intel ret address
3663 //     Owned by    |preserve|      Empty on Sparc.
3664 //       SELF      +--------+
3665 //        |        |  pad2  |  2   pad to align old SP
3666 //        |        +--------+  1
3667 //        |        | locks  |  0
3668 //        |        +--------+----> OptoReg::stack0(), even aligned
3669 //        |        |  pad1  | 11   pad to align new SP
3670 //        |        +--------+
3671 //        |        |        | 10
3672 //        |        | spills |  9   spills
3673 //        V        |        |  8   (pad0 slot for callee)
3674 //      -----------+--------+----> Matcher::_out_arg_limit, unaligned
3675 //        ^        |  out   |  7
3676 //        |        |  args  |  6   Holes in outgoing args owned by CALLEE
3677 //     Owned by    +--------+
3678 //      CALLEE     | new out|  6   Empty on Intel, window on Sparc
3679 //        |    new |preserve|      Must be even-aligned.
3680 //        |     SP-+--------+----> Matcher::_new_SP, even aligned
3681 //        |        |        |
3682 //
3683 // Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
3684 //         known from SELF's arguments and the Java calling convention.
3685 //         Region 6-7 is determined per call site.
3686 // Note 2: If the calling convention leaves holes in the incoming argument
3687 //         area, those holes are owned by SELF.  Holes in the outgoing area
3688 //         are owned by the CALLEE.  Holes should not be nessecary in the
3689 //         incoming area, as the Java calling convention is completely under
3690 //         the control of the AD file.  Doubles can be sorted and packed to
3691 //         avoid holes.  Holes in the outgoing arguments may be nessecary for
3692 //         varargs C calling conventions.
3693 // Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
3694 //         even aligned with pad0 as needed.
3695 //         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
3696 //         region 6-11 is even aligned; it may be padded out more so that
3697 //         the region from SP to FP meets the minimum stack alignment.
3698 
3699 frame %{
3700   // What direction does stack grow in (assumed to be same for C & Java)
3701   stack_direction(TOWARDS_LOW);
3702 
3703   // These three registers define part of the calling convention
3704   // between compiled code and the interpreter.
3705   inline_cache_reg(EAX);                // Inline Cache Register
3706   interpreter_method_oop_reg(EBX);      // Method Oop Register when calling interpreter
3707 
3708   // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
3709   cisc_spilling_operand_name(indOffset32);
3710 
3711   // Number of stack slots consumed by locking an object
3712   sync_stack_slots(1);
3713 
3714   // Compiled code's Frame Pointer
3715   frame_pointer(ESP);
3716   // Interpreter stores its frame pointer in a register which is
3717   // stored to the stack by I2CAdaptors.
3718   // I2CAdaptors convert from interpreted java to compiled java.
3719   interpreter_frame_pointer(EBP);
3720 
3721   // Stack alignment requirement
3722   // Alignment size in bytes (128-bit -> 16 bytes)
3723   stack_alignment(StackAlignmentInBytes);
3724 
3725   // Number of stack slots between incoming argument block and the start of
3726   // a new frame.  The PROLOG must add this many slots to the stack.  The
3727   // EPILOG must remove this many slots.  Intel needs one slot for
3728   // return address and one for rbp, (must save rbp)
3729   in_preserve_stack_slots(2+VerifyStackAtCalls);
3730 
3731   // Number of outgoing stack slots killed above the out_preserve_stack_slots
3732   // for calls to C.  Supports the var-args backing area for register parms.
3733   varargs_C_out_slots_killed(0);
3734 
3735   // The after-PROLOG location of the return address.  Location of
3736   // return address specifies a type (REG or STACK) and a number
3737   // representing the register number (i.e. - use a register name) or
3738   // stack slot.
3739   // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
3740   // Otherwise, it is above the locks and verification slot and alignment word
3741   return_addr(STACK - 1 +
3742               round_to((Compile::current()->in_preserve_stack_slots() +
3743                         Compile::current()->fixed_slots()),
3744                        stack_alignment_in_slots()));
3745 
3746   // Body of function which returns an integer array locating
3747   // arguments either in registers or in stack slots.  Passed an array
3748   // of ideal registers called "sig" and a "length" count.  Stack-slot
3749   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3750   // arguments for a CALLEE.  Incoming stack arguments are
3751   // automatically biased by the preserve_stack_slots field above.
3752   calling_convention %{
3753     // No difference between ingoing/outgoing just pass false
3754     SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
3755   %}
3756 
3757 
3758   // Body of function which returns an integer array locating
3759   // arguments either in registers or in stack slots.  Passed an array
3760   // of ideal registers called "sig" and a "length" count.  Stack-slot
3761   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3762   // arguments for a CALLEE.  Incoming stack arguments are
3763   // automatically biased by the preserve_stack_slots field above.
3764   c_calling_convention %{
3765     // This is obviously always outgoing
3766     (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
3767   %}
3768 
3769   // Location of C & interpreter return values
3770   c_return_value %{
3771     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3772     static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
3773     static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };
3774 
3775     // in SSE2+ mode we want to keep the FPU stack clean so pretend
3776     // that C functions return float and double results in XMM0.
3777     if( ideal_reg == Op_RegD && UseSSE>=2 )
3778       return OptoRegPair(XMM0b_num,XMM0_num);
3779     if( ideal_reg == Op_RegF && UseSSE>=2 )
3780       return OptoRegPair(OptoReg::Bad,XMM0_num);
3781 
3782     return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
3783   %}
3784 
3785   // Location of return values
3786   return_value %{
3787     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3788     static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
3789     static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };
3790     if( ideal_reg == Op_RegD && UseSSE>=2 )
3791       return OptoRegPair(XMM0b_num,XMM0_num);
3792     if( ideal_reg == Op_RegF && UseSSE>=1 )
3793       return OptoRegPair(OptoReg::Bad,XMM0_num);
3794     return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
3795   %}
3796 
3797 %}
3798 
3799 //----------ATTRIBUTES---------------------------------------------------------
3800 //----------Operand Attributes-------------------------------------------------
3801 op_attrib op_cost(0);        // Required cost attribute
3802 
3803 //----------Instruction Attributes---------------------------------------------
3804 ins_attrib ins_cost(100);       // Required cost attribute
3805 ins_attrib ins_size(8);         // Required size attribute (in bits)
3806 ins_attrib ins_short_branch(0); // Required flag: is this instruction a
3807                                 // non-matching short branch variant of some
3808                                                             // long branch?
3809 ins_attrib ins_alignment(1);    // Required alignment attribute (must be a power of 2)
3810                                 // specifies the alignment that some part of the instruction (not
3811                                 // necessarily the start) requires.  If > 1, a compute_padding()
3812                                 // function must be provided for the instruction
3813 
3814 //----------OPERANDS-----------------------------------------------------------
3815 // Operand definitions must precede instruction definitions for correct parsing
3816 // in the ADLC because operands constitute user defined types which are used in
3817 // instruction definitions.
3818 
3819 //----------Simple Operands----------------------------------------------------
3820 // Immediate Operands
3821 // Integer Immediate
3822 operand immI() %{
3823   match(ConI);
3824 
3825   op_cost(10);
3826   format %{ %}
3827   interface(CONST_INTER);
3828 %}
3829 
3830 // Constant for test vs zero
3831 operand immI0() %{
3832   predicate(n->get_int() == 0);
3833   match(ConI);
3834 
3835   op_cost(0);
3836   format %{ %}
3837   interface(CONST_INTER);
3838 %}
3839 
3840 // Constant for increment
3841 operand immI1() %{
3842   predicate(n->get_int() == 1);
3843   match(ConI);
3844 
3845   op_cost(0);
3846   format %{ %}
3847   interface(CONST_INTER);
3848 %}
3849 
3850 // Constant for decrement
3851 operand immI_M1() %{
3852   predicate(n->get_int() == -1);
3853   match(ConI);
3854 
3855   op_cost(0);
3856   format %{ %}
3857   interface(CONST_INTER);
3858 %}
3859 
3860 // Valid scale values for addressing modes
3861 operand immI2() %{
3862   predicate(0 <= n->get_int() && (n->get_int() <= 3));
3863   match(ConI);
3864 
3865   format %{ %}
3866   interface(CONST_INTER);
3867 %}
3868 
3869 operand immI8() %{
3870   predicate((-128 <= n->get_int()) && (n->get_int() <= 127));
3871   match(ConI);
3872 
3873   op_cost(5);
3874   format %{ %}
3875   interface(CONST_INTER);
3876 %}
3877 
3878 operand immI16() %{
3879   predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
3880   match(ConI);
3881 
3882   op_cost(10);
3883   format %{ %}
3884   interface(CONST_INTER);
3885 %}
3886 
3887 // Constant for long shifts
3888 operand immI_32() %{
3889   predicate( n->get_int() == 32 );
3890   match(ConI);
3891 
3892   op_cost(0);
3893   format %{ %}
3894   interface(CONST_INTER);
3895 %}
3896 
3897 operand immI_1_31() %{
3898   predicate( n->get_int() >= 1 && n->get_int() <= 31 );
3899   match(ConI);
3900 
3901   op_cost(0);
3902   format %{ %}
3903   interface(CONST_INTER);
3904 %}
3905 
3906 operand immI_32_63() %{
3907   predicate( n->get_int() >= 32 && n->get_int() <= 63 );
3908   match(ConI);
3909   op_cost(0);
3910 
3911   format %{ %}
3912   interface(CONST_INTER);
3913 %}
3914 
3915 operand immI_1() %{
3916   predicate( n->get_int() == 1 );
3917   match(ConI);
3918 
3919   op_cost(0);
3920   format %{ %}
3921   interface(CONST_INTER);
3922 %}
3923 
3924 operand immI_2() %{
3925   predicate( n->get_int() == 2 );
3926   match(ConI);
3927 
3928   op_cost(0);
3929   format %{ %}
3930   interface(CONST_INTER);
3931 %}
3932 
3933 operand immI_3() %{
3934   predicate( n->get_int() == 3 );
3935   match(ConI);
3936 
3937   op_cost(0);
3938   format %{ %}
3939   interface(CONST_INTER);
3940 %}
3941 
3942 // Pointer Immediate
3943 operand immP() %{
3944   match(ConP);
3945 
3946   op_cost(10);
3947   format %{ %}
3948   interface(CONST_INTER);
3949 %}
3950 
3951 // NULL Pointer Immediate
3952 operand immP0() %{
3953   predicate( n->get_ptr() == 0 );
3954   match(ConP);
3955   op_cost(0);
3956 
3957   format %{ %}
3958   interface(CONST_INTER);
3959 %}
3960 
3961 // Long Immediate
3962 operand immL() %{
3963   match(ConL);
3964 
3965   op_cost(20);
3966   format %{ %}
3967   interface(CONST_INTER);
3968 %}
3969 
3970 // Long Immediate zero
3971 operand immL0() %{
3972   predicate( n->get_long() == 0L );
3973   match(ConL);
3974   op_cost(0);
3975 
3976   format %{ %}
3977   interface(CONST_INTER);
3978 %}
3979 
3980 // Long Immediate zero
3981 operand immL_M1() %{
3982   predicate( n->get_long() == -1L );
3983   match(ConL);
3984   op_cost(0);
3985 
3986   format %{ %}
3987   interface(CONST_INTER);
3988 %}
3989 
3990 // Long immediate from 0 to 127.
3991 // Used for a shorter form of long mul by 10.
3992 operand immL_127() %{
3993   predicate((0 <= n->get_long()) && (n->get_long() <= 127));
3994   match(ConL);
3995   op_cost(0);
3996 
3997   format %{ %}
3998   interface(CONST_INTER);
3999 %}
4000 
4001 // Long Immediate: low 32-bit mask
4002 operand immL_32bits() %{
4003   predicate(n->get_long() == 0xFFFFFFFFL);
4004   match(ConL);
4005   op_cost(0);
4006 
4007   format %{ %}
4008   interface(CONST_INTER);
4009 %}
4010 
4011 // Long Immediate: low 32-bit mask
4012 operand immL32() %{
4013   predicate(n->get_long() == (int)(n->get_long()));
4014   match(ConL);
4015   op_cost(20);
4016 
4017   format %{ %}
4018   interface(CONST_INTER);
4019 %}
4020 
4021 //Double Immediate zero
4022 operand immDPR0() %{
4023   // Do additional (and counter-intuitive) test against NaN to work around VC++
4024   // bug that generates code such that NaNs compare equal to 0.0
4025   predicate( UseSSE<=1 && n->getd() == 0.0 && !g_isnan(n->getd()) );
4026   match(ConD);
4027 
4028   op_cost(5);
4029   format %{ %}
4030   interface(CONST_INTER);
4031 %}
4032 
4033 // Double Immediate one
4034 operand immDPR1() %{
4035   predicate( UseSSE<=1 && n->getd() == 1.0 );
4036   match(ConD);
4037 
4038   op_cost(5);
4039   format %{ %}
4040   interface(CONST_INTER);
4041 %}
4042 
4043 // Double Immediate
4044 operand immDPR() %{
4045   predicate(UseSSE<=1);
4046   match(ConD);
4047 
4048   op_cost(5);
4049   format %{ %}
4050   interface(CONST_INTER);
4051 %}
4052 
4053 operand immD() %{
4054   predicate(UseSSE>=2);
4055   match(ConD);
4056 
4057   op_cost(5);
4058   format %{ %}
4059   interface(CONST_INTER);
4060 %}
4061 
4062 // Double Immediate zero
4063 operand immD0() %{
4064   // Do additional (and counter-intuitive) test against NaN to work around VC++
4065   // bug that generates code such that NaNs compare equal to 0.0 AND do not
4066   // compare equal to -0.0.
4067   predicate( UseSSE>=2 && jlong_cast(n->getd()) == 0 );
4068   match(ConD);
4069 
4070   format %{ %}
4071   interface(CONST_INTER);
4072 %}
4073 
4074 // Float Immediate zero
4075 operand immFPR0() %{
4076   predicate(UseSSE == 0 && n->getf() == 0.0F);
4077   match(ConF);
4078 
4079   op_cost(5);
4080   format %{ %}
4081   interface(CONST_INTER);
4082 %}
4083 
4084 // Float Immediate one
4085 operand immFPR1() %{
4086   predicate(UseSSE == 0 && n->getf() == 1.0F);
4087   match(ConF);
4088 
4089   op_cost(5);
4090   format %{ %}
4091   interface(CONST_INTER);
4092 %}
4093 
4094 // Float Immediate
4095 operand immFPR() %{
4096   predicate( UseSSE == 0 );
4097   match(ConF);
4098 
4099   op_cost(5);
4100   format %{ %}
4101   interface(CONST_INTER);
4102 %}
4103 
4104 // Float Immediate
4105 operand immF() %{
4106   predicate(UseSSE >= 1);
4107   match(ConF);
4108 
4109   op_cost(5);
4110   format %{ %}
4111   interface(CONST_INTER);
4112 %}
4113 
4114 // Float Immediate zero.  Zero and not -0.0
4115 operand immF0() %{
4116   predicate( UseSSE >= 1 && jint_cast(n->getf()) == 0 );
4117   match(ConF);
4118 
4119   op_cost(5);
4120   format %{ %}
4121   interface(CONST_INTER);
4122 %}
4123 
4124 // Immediates for special shifts (sign extend)
4125 
4126 // Constants for increment
4127 operand immI_16() %{
4128   predicate( n->get_int() == 16 );
4129   match(ConI);
4130 
4131   format %{ %}
4132   interface(CONST_INTER);
4133 %}
4134 
4135 operand immI_24() %{
4136   predicate( n->get_int() == 24 );
4137   match(ConI);
4138 
4139   format %{ %}
4140   interface(CONST_INTER);
4141 %}
4142 
4143 // Constant for byte-wide masking
4144 operand immI_255() %{
4145   predicate( n->get_int() == 255 );
4146   match(ConI);
4147 
4148   format %{ %}
4149   interface(CONST_INTER);
4150 %}
4151 
4152 // Constant for short-wide masking
4153 operand immI_65535() %{
4154   predicate(n->get_int() == 65535);
4155   match(ConI);
4156 
4157   format %{ %}
4158   interface(CONST_INTER);
4159 %}
4160 
4161 // Register Operands
4162 // Integer Register
4163 operand rRegI() %{
4164   constraint(ALLOC_IN_RC(int_reg));
4165   match(RegI);
4166   match(xRegI);
4167   match(eAXRegI);
4168   match(eBXRegI);
4169   match(eCXRegI);
4170   match(eDXRegI);
4171   match(eDIRegI);
4172   match(eSIRegI);
4173 
4174   format %{ %}
4175   interface(REG_INTER);
4176 %}
4177 
4178 // Subset of Integer Register
4179 operand xRegI(rRegI reg) %{
4180   constraint(ALLOC_IN_RC(int_x_reg));
4181   match(reg);
4182   match(eAXRegI);
4183   match(eBXRegI);
4184   match(eCXRegI);
4185   match(eDXRegI);
4186 
4187   format %{ %}
4188   interface(REG_INTER);
4189 %}
4190 
4191 // Special Registers
4192 operand eAXRegI(xRegI reg) %{
4193   constraint(ALLOC_IN_RC(eax_reg));
4194   match(reg);
4195   match(rRegI);
4196 
4197   format %{ "EAX" %}
4198   interface(REG_INTER);
4199 %}
4200 
4201 // Special Registers
4202 operand eBXRegI(xRegI reg) %{
4203   constraint(ALLOC_IN_RC(ebx_reg));
4204   match(reg);
4205   match(rRegI);
4206 
4207   format %{ "EBX" %}
4208   interface(REG_INTER);
4209 %}
4210 
4211 operand eCXRegI(xRegI reg) %{
4212   constraint(ALLOC_IN_RC(ecx_reg));
4213   match(reg);
4214   match(rRegI);
4215 
4216   format %{ "ECX" %}
4217   interface(REG_INTER);
4218 %}
4219 
4220 operand eDXRegI(xRegI reg) %{
4221   constraint(ALLOC_IN_RC(edx_reg));
4222   match(reg);
4223   match(rRegI);
4224 
4225   format %{ "EDX" %}
4226   interface(REG_INTER);
4227 %}
4228 
4229 operand eDIRegI(xRegI reg) %{
4230   constraint(ALLOC_IN_RC(edi_reg));
4231   match(reg);
4232   match(rRegI);
4233 
4234   format %{ "EDI" %}
4235   interface(REG_INTER);
4236 %}
4237 
4238 operand naxRegI() %{
4239   constraint(ALLOC_IN_RC(nax_reg));
4240   match(RegI);
4241   match(eCXRegI);
4242   match(eDXRegI);
4243   match(eSIRegI);
4244   match(eDIRegI);
4245 
4246   format %{ %}
4247   interface(REG_INTER);
4248 %}
4249 
4250 operand nadxRegI() %{
4251   constraint(ALLOC_IN_RC(nadx_reg));
4252   match(RegI);
4253   match(eBXRegI);
4254   match(eCXRegI);
4255   match(eSIRegI);
4256   match(eDIRegI);
4257 
4258   format %{ %}
4259   interface(REG_INTER);
4260 %}
4261 
4262 operand ncxRegI() %{
4263   constraint(ALLOC_IN_RC(ncx_reg));
4264   match(RegI);
4265   match(eAXRegI);
4266   match(eDXRegI);
4267   match(eSIRegI);
4268   match(eDIRegI);
4269 
4270   format %{ %}
4271   interface(REG_INTER);
4272 %}
4273 
4274 // // This operand was used by cmpFastUnlock, but conflicted with 'object' reg
4275 // //
4276 operand eSIRegI(xRegI reg) %{
4277    constraint(ALLOC_IN_RC(esi_reg));
4278    match(reg);
4279    match(rRegI);
4280 
4281    format %{ "ESI" %}
4282    interface(REG_INTER);
4283 %}
4284 
4285 // Pointer Register
4286 operand anyRegP() %{
4287   constraint(ALLOC_IN_RC(any_reg));
4288   match(RegP);
4289   match(eAXRegP);
4290   match(eBXRegP);
4291   match(eCXRegP);
4292   match(eDIRegP);
4293   match(eRegP);
4294 
4295   format %{ %}
4296   interface(REG_INTER);
4297 %}
4298 
4299 operand eRegP() %{
4300   constraint(ALLOC_IN_RC(int_reg));
4301   match(RegP);
4302   match(eAXRegP);
4303   match(eBXRegP);
4304   match(eCXRegP);
4305   match(eDIRegP);
4306 
4307   format %{ %}
4308   interface(REG_INTER);
4309 %}
4310 
4311 // On windows95, EBP is not safe to use for implicit null tests.
4312 operand eRegP_no_EBP() %{
4313   constraint(ALLOC_IN_RC(int_reg_no_rbp));
4314   match(RegP);
4315   match(eAXRegP);
4316   match(eBXRegP);
4317   match(eCXRegP);
4318   match(eDIRegP);
4319 
4320   op_cost(100);
4321   format %{ %}
4322   interface(REG_INTER);
4323 %}
4324 
4325 operand naxRegP() %{
4326   constraint(ALLOC_IN_RC(nax_reg));
4327   match(RegP);
4328   match(eBXRegP);
4329   match(eDXRegP);
4330   match(eCXRegP);
4331   match(eSIRegP);
4332   match(eDIRegP);
4333 
4334   format %{ %}
4335   interface(REG_INTER);
4336 %}
4337 
4338 operand nabxRegP() %{
4339   constraint(ALLOC_IN_RC(nabx_reg));
4340   match(RegP);
4341   match(eCXRegP);
4342   match(eDXRegP);
4343   match(eSIRegP);
4344   match(eDIRegP);
4345 
4346   format %{ %}
4347   interface(REG_INTER);
4348 %}
4349 
4350 operand pRegP() %{
4351   constraint(ALLOC_IN_RC(p_reg));
4352   match(RegP);
4353   match(eBXRegP);
4354   match(eDXRegP);
4355   match(eSIRegP);
4356   match(eDIRegP);
4357 
4358   format %{ %}
4359   interface(REG_INTER);
4360 %}
4361 
4362 // Special Registers
4363 // Return a pointer value
4364 operand eAXRegP(eRegP reg) %{
4365   constraint(ALLOC_IN_RC(eax_reg));
4366   match(reg);
4367   format %{ "EAX" %}
4368   interface(REG_INTER);
4369 %}
4370 
4371 // Used in AtomicAdd
4372 operand eBXRegP(eRegP reg) %{
4373   constraint(ALLOC_IN_RC(ebx_reg));
4374   match(reg);
4375   format %{ "EBX" %}
4376   interface(REG_INTER);
4377 %}
4378 
4379 // Tail-call (interprocedural jump) to interpreter
4380 operand eCXRegP(eRegP reg) %{
4381   constraint(ALLOC_IN_RC(ecx_reg));
4382   match(reg);
4383   format %{ "ECX" %}
4384   interface(REG_INTER);
4385 %}
4386 
4387 operand eSIRegP(eRegP reg) %{
4388   constraint(ALLOC_IN_RC(esi_reg));
4389   match(reg);
4390   format %{ "ESI" %}
4391   interface(REG_INTER);
4392 %}
4393 
4394 // Used in rep stosw
4395 operand eDIRegP(eRegP reg) %{
4396   constraint(ALLOC_IN_RC(edi_reg));
4397   match(reg);
4398   format %{ "EDI" %}
4399   interface(REG_INTER);
4400 %}
4401 
4402 operand eBPRegP() %{
4403   constraint(ALLOC_IN_RC(ebp_reg));
4404   match(RegP);
4405   format %{ "EBP" %}
4406   interface(REG_INTER);
4407 %}
4408 
4409 operand eRegL() %{
4410   constraint(ALLOC_IN_RC(long_reg));
4411   match(RegL);
4412   match(eADXRegL);
4413 
4414   format %{ %}
4415   interface(REG_INTER);
4416 %}
4417 
4418 operand eADXRegL( eRegL reg ) %{
4419   constraint(ALLOC_IN_RC(eadx_reg));
4420   match(reg);
4421 
4422   format %{ "EDX:EAX" %}
4423   interface(REG_INTER);
4424 %}
4425 
4426 operand eBCXRegL( eRegL reg ) %{
4427   constraint(ALLOC_IN_RC(ebcx_reg));
4428   match(reg);
4429 
4430   format %{ "EBX:ECX" %}
4431   interface(REG_INTER);
4432 %}
4433 
4434 // Special case for integer high multiply
4435 operand eADXRegL_low_only() %{
4436   constraint(ALLOC_IN_RC(eadx_reg));
4437   match(RegL);
4438 
4439   format %{ "EAX" %}
4440   interface(REG_INTER);
4441 %}
4442 
4443 // Flags register, used as output of compare instructions
4444 operand eFlagsReg() %{
4445   constraint(ALLOC_IN_RC(int_flags));
4446   match(RegFlags);
4447 
4448   format %{ "EFLAGS" %}
4449   interface(REG_INTER);
4450 %}
4451 
4452 // Flags register, used as output of FLOATING POINT compare instructions
4453 operand eFlagsRegU() %{
4454   constraint(ALLOC_IN_RC(int_flags));
4455   match(RegFlags);
4456 
4457   format %{ "EFLAGS_U" %}
4458   interface(REG_INTER);
4459 %}
4460 
4461 operand eFlagsRegUCF() %{
4462   constraint(ALLOC_IN_RC(int_flags));
4463   match(RegFlags);
4464   predicate(false);
4465 
4466   format %{ "EFLAGS_U_CF" %}
4467   interface(REG_INTER);
4468 %}
4469 
4470 // Condition Code Register used by long compare
4471 operand flagsReg_long_LTGE() %{
4472   constraint(ALLOC_IN_RC(int_flags));
4473   match(RegFlags);
4474   format %{ "FLAGS_LTGE" %}
4475   interface(REG_INTER);
4476 %}
4477 operand flagsReg_long_EQNE() %{
4478   constraint(ALLOC_IN_RC(int_flags));
4479   match(RegFlags);
4480   format %{ "FLAGS_EQNE" %}
4481   interface(REG_INTER);
4482 %}
4483 operand flagsReg_long_LEGT() %{
4484   constraint(ALLOC_IN_RC(int_flags));
4485   match(RegFlags);
4486   format %{ "FLAGS_LEGT" %}
4487   interface(REG_INTER);
4488 %}
4489 
4490 // Float register operands
4491 operand regDPR() %{
4492   predicate( UseSSE < 2 );
4493   constraint(ALLOC_IN_RC(fp_dbl_reg));
4494   match(RegD);
4495   match(regDPR1);
4496   match(regDPR2);
4497   format %{ %}
4498   interface(REG_INTER);
4499 %}
4500 
4501 operand regDPR1(regDPR reg) %{
4502   predicate( UseSSE < 2 );
4503   constraint(ALLOC_IN_RC(fp_dbl_reg0));
4504   match(reg);
4505   format %{ "FPR1" %}
4506   interface(REG_INTER);
4507 %}
4508 
4509 operand regDPR2(regDPR reg) %{
4510   predicate( UseSSE < 2 );
4511   constraint(ALLOC_IN_RC(fp_dbl_reg1));
4512   match(reg);
4513   format %{ "FPR2" %}
4514   interface(REG_INTER);
4515 %}
4516 
4517 operand regnotDPR1(regDPR reg) %{
4518   predicate( UseSSE < 2 );
4519   constraint(ALLOC_IN_RC(fp_dbl_notreg0));
4520   match(reg);
4521   format %{ %}
4522   interface(REG_INTER);
4523 %}
4524 
4525 // Float register operands
4526 operand regFPR() %{
4527   predicate( UseSSE < 2 );
4528   constraint(ALLOC_IN_RC(fp_flt_reg));
4529   match(RegF);
4530   match(regFPR1);
4531   format %{ %}
4532   interface(REG_INTER);
4533 %}
4534 
4535 // Float register operands
4536 operand regFPR1(regFPR reg) %{
4537   predicate( UseSSE < 2 );
4538   constraint(ALLOC_IN_RC(fp_flt_reg0));
4539   match(reg);
4540   format %{ "FPR1" %}
4541   interface(REG_INTER);
4542 %}
4543 
4544 // XMM Float register operands
4545 operand regF() %{
4546   predicate( UseSSE>=1 );
4547   constraint(ALLOC_IN_RC(float_reg));
4548   match(RegF);
4549   format %{ %}
4550   interface(REG_INTER);
4551 %}
4552 
4553 // XMM Double register operands
4554 operand regD() %{
4555   predicate( UseSSE>=2 );
4556   constraint(ALLOC_IN_RC(double_reg));
4557   match(RegD);
4558   format %{ %}
4559   interface(REG_INTER);
4560 %}
4561 
4562 
4563 //----------Memory Operands----------------------------------------------------
4564 // Direct Memory Operand
4565 operand direct(immP addr) %{
4566   match(addr);
4567 
4568   format %{ "[$addr]" %}
4569   interface(MEMORY_INTER) %{
4570     base(0xFFFFFFFF);
4571     index(0x4);
4572     scale(0x0);
4573     disp($addr);
4574   %}
4575 %}
4576 
4577 // Indirect Memory Operand
4578 operand indirect(eRegP reg) %{
4579   constraint(ALLOC_IN_RC(int_reg));
4580   match(reg);
4581 
4582   format %{ "[$reg]" %}
4583   interface(MEMORY_INTER) %{
4584     base($reg);
4585     index(0x4);
4586     scale(0x0);
4587     disp(0x0);
4588   %}
4589 %}
4590 
4591 // Indirect Memory Plus Short Offset Operand
4592 operand indOffset8(eRegP reg, immI8 off) %{
4593   match(AddP reg off);
4594 
4595   format %{ "[$reg + $off]" %}
4596   interface(MEMORY_INTER) %{
4597     base($reg);
4598     index(0x4);
4599     scale(0x0);
4600     disp($off);
4601   %}
4602 %}
4603 
4604 // Indirect Memory Plus Long Offset Operand
4605 operand indOffset32(eRegP reg, immI off) %{
4606   match(AddP reg off);
4607 
4608   format %{ "[$reg + $off]" %}
4609   interface(MEMORY_INTER) %{
4610     base($reg);
4611     index(0x4);
4612     scale(0x0);
4613     disp($off);
4614   %}
4615 %}
4616 
4617 // Indirect Memory Plus Long Offset Operand
4618 operand indOffset32X(rRegI reg, immP off) %{
4619   match(AddP off reg);
4620 
4621   format %{ "[$reg + $off]" %}
4622   interface(MEMORY_INTER) %{
4623     base($reg);
4624     index(0x4);
4625     scale(0x0);
4626     disp($off);
4627   %}
4628 %}
4629 
4630 // Indirect Memory Plus Index Register Plus Offset Operand
4631 operand indIndexOffset(eRegP reg, rRegI ireg, immI off) %{
4632   match(AddP (AddP reg ireg) off);
4633 
4634   op_cost(10);
4635   format %{"[$reg + $off + $ireg]" %}
4636   interface(MEMORY_INTER) %{
4637     base($reg);
4638     index($ireg);
4639     scale(0x0);
4640     disp($off);
4641   %}
4642 %}
4643 
4644 // Indirect Memory Plus Index Register Plus Offset Operand
4645 operand indIndex(eRegP reg, rRegI ireg) %{
4646   match(AddP reg ireg);
4647 
4648   op_cost(10);
4649   format %{"[$reg + $ireg]" %}
4650   interface(MEMORY_INTER) %{
4651     base($reg);
4652     index($ireg);
4653     scale(0x0);
4654     disp(0x0);
4655   %}
4656 %}
4657 
4658 // // -------------------------------------------------------------------------
4659 // // 486 architecture doesn't support "scale * index + offset" with out a base
4660 // // -------------------------------------------------------------------------
4661 // // Scaled Memory Operands
4662 // // Indirect Memory Times Scale Plus Offset Operand
4663 // operand indScaleOffset(immP off, rRegI ireg, immI2 scale) %{
4664 //   match(AddP off (LShiftI ireg scale));
4665 //
4666 //   op_cost(10);
4667 //   format %{"[$off + $ireg << $scale]" %}
4668 //   interface(MEMORY_INTER) %{
4669 //     base(0x4);
4670 //     index($ireg);
4671 //     scale($scale);
4672 //     disp($off);
4673 //   %}
4674 // %}
4675 
4676 // Indirect Memory Times Scale Plus Index Register
4677 operand indIndexScale(eRegP reg, rRegI ireg, immI2 scale) %{
4678   match(AddP reg (LShiftI ireg scale));
4679 
4680   op_cost(10);
4681   format %{"[$reg + $ireg << $scale]" %}
4682   interface(MEMORY_INTER) %{
4683     base($reg);
4684     index($ireg);
4685     scale($scale);
4686     disp(0x0);
4687   %}
4688 %}
4689 
4690 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4691 operand indIndexScaleOffset(eRegP reg, immI off, rRegI ireg, immI2 scale) %{
4692   match(AddP (AddP reg (LShiftI ireg scale)) off);
4693 
4694   op_cost(10);
4695   format %{"[$reg + $off + $ireg << $scale]" %}
4696   interface(MEMORY_INTER) %{
4697     base($reg);
4698     index($ireg);
4699     scale($scale);
4700     disp($off);
4701   %}
4702 %}
4703 
4704 //----------Load Long Memory Operands------------------------------------------
4705 // The load-long idiom will use it's address expression again after loading
4706 // the first word of the long.  If the load-long destination overlaps with
4707 // registers used in the addressing expression, the 2nd half will be loaded
4708 // from a clobbered address.  Fix this by requiring that load-long use
4709 // address registers that do not overlap with the load-long target.
4710 
4711 // load-long support
4712 operand load_long_RegP() %{
4713   constraint(ALLOC_IN_RC(esi_reg));
4714   match(RegP);
4715   match(eSIRegP);
4716   op_cost(100);
4717   format %{  %}
4718   interface(REG_INTER);
4719 %}
4720 
4721 // Indirect Memory Operand Long
4722 operand load_long_indirect(load_long_RegP reg) %{
4723   constraint(ALLOC_IN_RC(esi_reg));
4724   match(reg);
4725 
4726   format %{ "[$reg]" %}
4727   interface(MEMORY_INTER) %{
4728     base($reg);
4729     index(0x4);
4730     scale(0x0);
4731     disp(0x0);
4732   %}
4733 %}
4734 
4735 // Indirect Memory Plus Long Offset Operand
4736 operand load_long_indOffset32(load_long_RegP reg, immI off) %{
4737   match(AddP reg off);
4738 
4739   format %{ "[$reg + $off]" %}
4740   interface(MEMORY_INTER) %{
4741     base($reg);
4742     index(0x4);
4743     scale(0x0);
4744     disp($off);
4745   %}
4746 %}
4747 
4748 opclass load_long_memory(load_long_indirect, load_long_indOffset32);
4749 
4750 
4751 //----------Special Memory Operands--------------------------------------------
4752 // Stack Slot Operand - This operand is used for loading and storing temporary
4753 //                      values on the stack where a match requires a value to
4754 //                      flow through memory.
4755 operand stackSlotP(sRegP reg) %{
4756   constraint(ALLOC_IN_RC(stack_slots));
4757   // No match rule because this operand is only generated in matching
4758   format %{ "[$reg]" %}
4759   interface(MEMORY_INTER) %{
4760     base(0x4);   // ESP
4761     index(0x4);  // No Index
4762     scale(0x0);  // No Scale
4763     disp($reg);  // Stack Offset
4764   %}
4765 %}
4766 
4767 operand stackSlotI(sRegI reg) %{
4768   constraint(ALLOC_IN_RC(stack_slots));
4769   // No match rule because this operand is only generated in matching
4770   format %{ "[$reg]" %}
4771   interface(MEMORY_INTER) %{
4772     base(0x4);   // ESP
4773     index(0x4);  // No Index
4774     scale(0x0);  // No Scale
4775     disp($reg);  // Stack Offset
4776   %}
4777 %}
4778 
4779 operand stackSlotF(sRegF reg) %{
4780   constraint(ALLOC_IN_RC(stack_slots));
4781   // No match rule because this operand is only generated in matching
4782   format %{ "[$reg]" %}
4783   interface(MEMORY_INTER) %{
4784     base(0x4);   // ESP
4785     index(0x4);  // No Index
4786     scale(0x0);  // No Scale
4787     disp($reg);  // Stack Offset
4788   %}
4789 %}
4790 
4791 operand stackSlotD(sRegD reg) %{
4792   constraint(ALLOC_IN_RC(stack_slots));
4793   // No match rule because this operand is only generated in matching
4794   format %{ "[$reg]" %}
4795   interface(MEMORY_INTER) %{
4796     base(0x4);   // ESP
4797     index(0x4);  // No Index
4798     scale(0x0);  // No Scale
4799     disp($reg);  // Stack Offset
4800   %}
4801 %}
4802 
4803 operand stackSlotL(sRegL reg) %{
4804   constraint(ALLOC_IN_RC(stack_slots));
4805   // No match rule because this operand is only generated in matching
4806   format %{ "[$reg]" %}
4807   interface(MEMORY_INTER) %{
4808     base(0x4);   // ESP
4809     index(0x4);  // No Index
4810     scale(0x0);  // No Scale
4811     disp($reg);  // Stack Offset
4812   %}
4813 %}
4814 
4815 //----------Memory Operands - Win95 Implicit Null Variants----------------
4816 // Indirect Memory Operand
4817 operand indirect_win95_safe(eRegP_no_EBP reg)
4818 %{
4819   constraint(ALLOC_IN_RC(int_reg));
4820   match(reg);
4821 
4822   op_cost(100);
4823   format %{ "[$reg]" %}
4824   interface(MEMORY_INTER) %{
4825     base($reg);
4826     index(0x4);
4827     scale(0x0);
4828     disp(0x0);
4829   %}
4830 %}
4831 
4832 // Indirect Memory Plus Short Offset Operand
4833 operand indOffset8_win95_safe(eRegP_no_EBP reg, immI8 off)
4834 %{
4835   match(AddP reg off);
4836 
4837   op_cost(100);
4838   format %{ "[$reg + $off]" %}
4839   interface(MEMORY_INTER) %{
4840     base($reg);
4841     index(0x4);
4842     scale(0x0);
4843     disp($off);
4844   %}
4845 %}
4846 
4847 // Indirect Memory Plus Long Offset Operand
4848 operand indOffset32_win95_safe(eRegP_no_EBP reg, immI off)
4849 %{
4850   match(AddP reg off);
4851 
4852   op_cost(100);
4853   format %{ "[$reg + $off]" %}
4854   interface(MEMORY_INTER) %{
4855     base($reg);
4856     index(0x4);
4857     scale(0x0);
4858     disp($off);
4859   %}
4860 %}
4861 
4862 // Indirect Memory Plus Index Register Plus Offset Operand
4863 operand indIndexOffset_win95_safe(eRegP_no_EBP reg, rRegI ireg, immI off)
4864 %{
4865   match(AddP (AddP reg ireg) off);
4866 
4867   op_cost(100);
4868   format %{"[$reg + $off + $ireg]" %}
4869   interface(MEMORY_INTER) %{
4870     base($reg);
4871     index($ireg);
4872     scale(0x0);
4873     disp($off);
4874   %}
4875 %}
4876 
4877 // Indirect Memory Times Scale Plus Index Register
4878 operand indIndexScale_win95_safe(eRegP_no_EBP reg, rRegI ireg, immI2 scale)
4879 %{
4880   match(AddP reg (LShiftI ireg scale));
4881 
4882   op_cost(100);
4883   format %{"[$reg + $ireg << $scale]" %}
4884   interface(MEMORY_INTER) %{
4885     base($reg);
4886     index($ireg);
4887     scale($scale);
4888     disp(0x0);
4889   %}
4890 %}
4891 
4892 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4893 operand indIndexScaleOffset_win95_safe(eRegP_no_EBP reg, immI off, rRegI ireg, immI2 scale)
4894 %{
4895   match(AddP (AddP reg (LShiftI ireg scale)) off);
4896 
4897   op_cost(100);
4898   format %{"[$reg + $off + $ireg << $scale]" %}
4899   interface(MEMORY_INTER) %{
4900     base($reg);
4901     index($ireg);
4902     scale($scale);
4903     disp($off);
4904   %}
4905 %}
4906 
4907 //----------Conditional Branch Operands----------------------------------------
4908 // Comparison Op  - This is the operation of the comparison, and is limited to
4909 //                  the following set of codes:
4910 //                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
4911 //
4912 // Other attributes of the comparison, such as unsignedness, are specified
4913 // by the comparison instruction that sets a condition code flags register.
4914 // That result is represented by a flags operand whose subtype is appropriate
4915 // to the unsignedness (etc.) of the comparison.
4916 //
4917 // Later, the instruction which matches both the Comparison Op (a Bool) and
4918 // the flags (produced by the Cmp) specifies the coding of the comparison op
4919 // by matching a specific subtype of Bool operand below, such as cmpOpU.
4920 
4921 // Comparision Code
4922 operand cmpOp() %{
4923   match(Bool);
4924 
4925   format %{ "" %}
4926   interface(COND_INTER) %{
4927     equal(0x4, "e");
4928     not_equal(0x5, "ne");
4929     less(0xC, "l");
4930     greater_equal(0xD, "ge");
4931     less_equal(0xE, "le");
4932     greater(0xF, "g");
4933     overflow(0x0, "o");
4934     no_overflow(0x1, "no");
4935   %}
4936 %}
4937 
4938 // Comparison Code, unsigned compare.  Used by FP also, with
4939 // C2 (unordered) turned into GT or LT already.  The other bits
4940 // C0 and C3 are turned into Carry & Zero flags.
4941 operand cmpOpU() %{
4942   match(Bool);
4943 
4944   format %{ "" %}
4945   interface(COND_INTER) %{
4946     equal(0x4, "e");
4947     not_equal(0x5, "ne");
4948     less(0x2, "b");
4949     greater_equal(0x3, "nb");
4950     less_equal(0x6, "be");
4951     greater(0x7, "nbe");
4952     overflow(0x0, "o");
4953     no_overflow(0x1, "no");
4954   %}
4955 %}
4956 
4957 // Floating comparisons that don't require any fixup for the unordered case
4958 operand cmpOpUCF() %{
4959   match(Bool);
4960   predicate(n->as_Bool()->_test._test == BoolTest::lt ||
4961             n->as_Bool()->_test._test == BoolTest::ge ||
4962             n->as_Bool()->_test._test == BoolTest::le ||
4963             n->as_Bool()->_test._test == BoolTest::gt);
4964   format %{ "" %}
4965   interface(COND_INTER) %{
4966     equal(0x4, "e");
4967     not_equal(0x5, "ne");
4968     less(0x2, "b");
4969     greater_equal(0x3, "nb");
4970     less_equal(0x6, "be");
4971     greater(0x7, "nbe");
4972     overflow(0x0, "o");
4973     no_overflow(0x1, "no");
4974   %}
4975 %}
4976 
4977 
4978 // Floating comparisons that can be fixed up with extra conditional jumps
4979 operand cmpOpUCF2() %{
4980   match(Bool);
4981   predicate(n->as_Bool()->_test._test == BoolTest::ne ||
4982             n->as_Bool()->_test._test == BoolTest::eq);
4983   format %{ "" %}
4984   interface(COND_INTER) %{
4985     equal(0x4, "e");
4986     not_equal(0x5, "ne");
4987     less(0x2, "b");
4988     greater_equal(0x3, "nb");
4989     less_equal(0x6, "be");
4990     greater(0x7, "nbe");
4991     overflow(0x0, "o");
4992     no_overflow(0x1, "no");
4993   %}
4994 %}
4995 
4996 // Comparison Code for FP conditional move
4997 operand cmpOp_fcmov() %{
4998   match(Bool);
4999 
5000   predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
5001             n->as_Bool()->_test._test != BoolTest::no_overflow);
5002   format %{ "" %}
5003   interface(COND_INTER) %{
5004     equal        (0x0C8);
5005     not_equal    (0x1C8);
5006     less         (0x0C0);
5007     greater_equal(0x1C0);
5008     less_equal   (0x0D0);
5009     greater      (0x1D0);
5010     overflow(0x0, "o"); // not really supported by the instruction
5011     no_overflow(0x1, "no"); // not really supported by the instruction
5012   %}
5013 %}
5014 
5015 // Comparision Code used in long compares
5016 operand cmpOp_commute() %{
5017   match(Bool);
5018 
5019   format %{ "" %}
5020   interface(COND_INTER) %{
5021     equal(0x4, "e");
5022     not_equal(0x5, "ne");
5023     less(0xF, "g");
5024     greater_equal(0xE, "le");
5025     less_equal(0xD, "ge");
5026     greater(0xC, "l");
5027     overflow(0x0, "o");
5028     no_overflow(0x1, "no");
5029   %}
5030 %}
5031 
5032 //----------OPERAND CLASSES----------------------------------------------------
5033 // Operand Classes are groups of operands that are used as to simplify
5034 // instruction definitions by not requiring the AD writer to specify separate
5035 // instructions for every form of operand when the instruction accepts
5036 // multiple operand types with the same basic encoding and format.  The classic
5037 // case of this is memory operands.
5038 
5039 opclass memory(direct, indirect, indOffset8, indOffset32, indOffset32X, indIndexOffset,
5040                indIndex, indIndexScale, indIndexScaleOffset);
5041 
5042 // Long memory operations are encoded in 2 instructions and a +4 offset.
5043 // This means some kind of offset is always required and you cannot use
5044 // an oop as the offset (done when working on static globals).
5045 opclass long_memory(direct, indirect, indOffset8, indOffset32, indIndexOffset,
5046                     indIndex, indIndexScale, indIndexScaleOffset);
5047 
5048 
5049 //----------PIPELINE-----------------------------------------------------------
5050 // Rules which define the behavior of the target architectures pipeline.
5051 pipeline %{
5052 
5053 //----------ATTRIBUTES---------------------------------------------------------
5054 attributes %{
5055   variable_size_instructions;        // Fixed size instructions
5056   max_instructions_per_bundle = 3;   // Up to 3 instructions per bundle
5057   instruction_unit_size = 1;         // An instruction is 1 bytes long
5058   instruction_fetch_unit_size = 16;  // The processor fetches one line
5059   instruction_fetch_units = 1;       // of 16 bytes
5060 
5061   // List of nop instructions
5062   nops( MachNop );
5063 %}
5064 
5065 //----------RESOURCES----------------------------------------------------------
5066 // Resources are the functional units available to the machine
5067 
5068 // Generic P2/P3 pipeline
5069 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5070 // 3 instructions decoded per cycle.
5071 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5072 // 2 ALU op, only ALU0 handles mul/div instructions.
5073 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5074            MS0, MS1, MEM = MS0 | MS1,
5075            BR, FPU,
5076            ALU0, ALU1, ALU = ALU0 | ALU1 );
5077 
5078 //----------PIPELINE DESCRIPTION-----------------------------------------------
5079 // Pipeline Description specifies the stages in the machine's pipeline
5080 
5081 // Generic P2/P3 pipeline
5082 pipe_desc(S0, S1, S2, S3, S4, S5);
5083 
5084 //----------PIPELINE CLASSES---------------------------------------------------
5085 // Pipeline Classes describe the stages in which input and output are
5086 // referenced by the hardware pipeline.
5087 
5088 // Naming convention: ialu or fpu
5089 // Then: _reg
5090 // Then: _reg if there is a 2nd register
5091 // Then: _long if it's a pair of instructions implementing a long
5092 // Then: _fat if it requires the big decoder
5093 //   Or: _mem if it requires the big decoder and a memory unit.
5094 
5095 // Integer ALU reg operation
5096 pipe_class ialu_reg(rRegI dst) %{
5097     single_instruction;
5098     dst    : S4(write);
5099     dst    : S3(read);
5100     DECODE : S0;        // any decoder
5101     ALU    : S3;        // any alu
5102 %}
5103 
5104 // Long ALU reg operation
5105 pipe_class ialu_reg_long(eRegL dst) %{
5106     instruction_count(2);
5107     dst    : S4(write);
5108     dst    : S3(read);
5109     DECODE : S0(2);     // any 2 decoders
5110     ALU    : S3(2);     // both alus
5111 %}
5112 
5113 // Integer ALU reg operation using big decoder
5114 pipe_class ialu_reg_fat(rRegI dst) %{
5115     single_instruction;
5116     dst    : S4(write);
5117     dst    : S3(read);
5118     D0     : S0;        // big decoder only
5119     ALU    : S3;        // any alu
5120 %}
5121 
5122 // Long ALU reg operation using big decoder
5123 pipe_class ialu_reg_long_fat(eRegL dst) %{
5124     instruction_count(2);
5125     dst    : S4(write);
5126     dst    : S3(read);
5127     D0     : S0(2);     // big decoder only; twice
5128     ALU    : S3(2);     // any 2 alus
5129 %}
5130 
5131 // Integer ALU reg-reg operation
5132 pipe_class ialu_reg_reg(rRegI dst, rRegI src) %{
5133     single_instruction;
5134     dst    : S4(write);
5135     src    : S3(read);
5136     DECODE : S0;        // any decoder
5137     ALU    : S3;        // any alu
5138 %}
5139 
5140 // Long ALU reg-reg operation
5141 pipe_class ialu_reg_reg_long(eRegL dst, eRegL src) %{
5142     instruction_count(2);
5143     dst    : S4(write);
5144     src    : S3(read);
5145     DECODE : S0(2);     // any 2 decoders
5146     ALU    : S3(2);     // both alus
5147 %}
5148 
5149 // Integer ALU reg-reg operation
5150 pipe_class ialu_reg_reg_fat(rRegI dst, memory src) %{
5151     single_instruction;
5152     dst    : S4(write);
5153     src    : S3(read);
5154     D0     : S0;        // big decoder only
5155     ALU    : S3;        // any alu
5156 %}
5157 
5158 // Long ALU reg-reg operation
5159 pipe_class ialu_reg_reg_long_fat(eRegL dst, eRegL src) %{
5160     instruction_count(2);
5161     dst    : S4(write);
5162     src    : S3(read);
5163     D0     : S0(2);     // big decoder only; twice
5164     ALU    : S3(2);     // both alus
5165 %}
5166 
5167 // Integer ALU reg-mem operation
5168 pipe_class ialu_reg_mem(rRegI dst, memory mem) %{
5169     single_instruction;
5170     dst    : S5(write);
5171     mem    : S3(read);
5172     D0     : S0;        // big decoder only
5173     ALU    : S4;        // any alu
5174     MEM    : S3;        // any mem
5175 %}
5176 
5177 // Long ALU reg-mem operation
5178 pipe_class ialu_reg_long_mem(eRegL dst, load_long_memory mem) %{
5179     instruction_count(2);
5180     dst    : S5(write);
5181     mem    : S3(read);
5182     D0     : S0(2);     // big decoder only; twice
5183     ALU    : S4(2);     // any 2 alus
5184     MEM    : S3(2);     // both mems
5185 %}
5186 
5187 // Integer mem operation (prefetch)
5188 pipe_class ialu_mem(memory mem)
5189 %{
5190     single_instruction;
5191     mem    : S3(read);
5192     D0     : S0;        // big decoder only
5193     MEM    : S3;        // any mem
5194 %}
5195 
5196 // Integer Store to Memory
5197 pipe_class ialu_mem_reg(memory mem, rRegI src) %{
5198     single_instruction;
5199     mem    : S3(read);
5200     src    : S5(read);
5201     D0     : S0;        // big decoder only
5202     ALU    : S4;        // any alu
5203     MEM    : S3;
5204 %}
5205 
5206 // Long Store to Memory
5207 pipe_class ialu_mem_long_reg(memory mem, eRegL src) %{
5208     instruction_count(2);
5209     mem    : S3(read);
5210     src    : S5(read);
5211     D0     : S0(2);     // big decoder only; twice
5212     ALU    : S4(2);     // any 2 alus
5213     MEM    : S3(2);     // Both mems
5214 %}
5215 
5216 // Integer Store to Memory
5217 pipe_class ialu_mem_imm(memory mem) %{
5218     single_instruction;
5219     mem    : S3(read);
5220     D0     : S0;        // big decoder only
5221     ALU    : S4;        // any alu
5222     MEM    : S3;
5223 %}
5224 
5225 // Integer ALU0 reg-reg operation
5226 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src) %{
5227     single_instruction;
5228     dst    : S4(write);
5229     src    : S3(read);
5230     D0     : S0;        // Big decoder only
5231     ALU0   : S3;        // only alu0
5232 %}
5233 
5234 // Integer ALU0 reg-mem operation
5235 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem) %{
5236     single_instruction;
5237     dst    : S5(write);
5238     mem    : S3(read);
5239     D0     : S0;        // big decoder only
5240     ALU0   : S4;        // ALU0 only
5241     MEM    : S3;        // any mem
5242 %}
5243 
5244 // Integer ALU reg-reg operation
5245 pipe_class ialu_cr_reg_reg(eFlagsReg cr, rRegI src1, rRegI src2) %{
5246     single_instruction;
5247     cr     : S4(write);
5248     src1   : S3(read);
5249     src2   : S3(read);
5250     DECODE : S0;        // any decoder
5251     ALU    : S3;        // any alu
5252 %}
5253 
5254 // Integer ALU reg-imm operation
5255 pipe_class ialu_cr_reg_imm(eFlagsReg cr, rRegI src1) %{
5256     single_instruction;
5257     cr     : S4(write);
5258     src1   : S3(read);
5259     DECODE : S0;        // any decoder
5260     ALU    : S3;        // any alu
5261 %}
5262 
5263 // Integer ALU reg-mem operation
5264 pipe_class ialu_cr_reg_mem(eFlagsReg cr, rRegI src1, memory src2) %{
5265     single_instruction;
5266     cr     : S4(write);
5267     src1   : S3(read);
5268     src2   : S3(read);
5269     D0     : S0;        // big decoder only
5270     ALU    : S4;        // any alu
5271     MEM    : S3;
5272 %}
5273 
5274 // Conditional move reg-reg
5275 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y ) %{
5276     instruction_count(4);
5277     y      : S4(read);
5278     q      : S3(read);
5279     p      : S3(read);
5280     DECODE : S0(4);     // any decoder
5281 %}
5282 
5283 // Conditional move reg-reg
5284 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, eFlagsReg cr ) %{
5285     single_instruction;
5286     dst    : S4(write);
5287     src    : S3(read);
5288     cr     : S3(read);
5289     DECODE : S0;        // any decoder
5290 %}
5291 
5292 // Conditional move reg-mem
5293 pipe_class pipe_cmov_mem( eFlagsReg cr, rRegI dst, memory src) %{
5294     single_instruction;
5295     dst    : S4(write);
5296     src    : S3(read);
5297     cr     : S3(read);
5298     DECODE : S0;        // any decoder
5299     MEM    : S3;
5300 %}
5301 
5302 // Conditional move reg-reg long
5303 pipe_class pipe_cmov_reg_long( eFlagsReg cr, eRegL dst, eRegL src) %{
5304     single_instruction;
5305     dst    : S4(write);
5306     src    : S3(read);
5307     cr     : S3(read);
5308     DECODE : S0(2);     // any 2 decoders
5309 %}
5310 
5311 // Conditional move double reg-reg
5312 pipe_class pipe_cmovDPR_reg( eFlagsReg cr, regDPR1 dst, regDPR src) %{
5313     single_instruction;
5314     dst    : S4(write);
5315     src    : S3(read);
5316     cr     : S3(read);
5317     DECODE : S0;        // any decoder
5318 %}
5319 
5320 // Float reg-reg operation
5321 pipe_class fpu_reg(regDPR dst) %{
5322     instruction_count(2);
5323     dst    : S3(read);
5324     DECODE : S0(2);     // any 2 decoders
5325     FPU    : S3;
5326 %}
5327 
5328 // Float reg-reg operation
5329 pipe_class fpu_reg_reg(regDPR dst, regDPR src) %{
5330     instruction_count(2);
5331     dst    : S4(write);
5332     src    : S3(read);
5333     DECODE : S0(2);     // any 2 decoders
5334     FPU    : S3;
5335 %}
5336 
5337 // Float reg-reg operation
5338 pipe_class fpu_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2) %{
5339     instruction_count(3);
5340     dst    : S4(write);
5341     src1   : S3(read);
5342     src2   : S3(read);
5343     DECODE : S0(3);     // any 3 decoders
5344     FPU    : S3(2);
5345 %}
5346 
5347 // Float reg-reg operation
5348 pipe_class fpu_reg_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2, regDPR src3) %{
5349     instruction_count(4);
5350     dst    : S4(write);
5351     src1   : S3(read);
5352     src2   : S3(read);
5353     src3   : S3(read);
5354     DECODE : S0(4);     // any 3 decoders
5355     FPU    : S3(2);
5356 %}
5357 
5358 // Float reg-reg operation
5359 pipe_class fpu_reg_mem_reg_reg(regDPR dst, memory src1, regDPR src2, regDPR src3) %{
5360     instruction_count(4);
5361     dst    : S4(write);
5362     src1   : S3(read);
5363     src2   : S3(read);
5364     src3   : S3(read);
5365     DECODE : S1(3);     // any 3 decoders
5366     D0     : S0;        // Big decoder only
5367     FPU    : S3(2);
5368     MEM    : S3;
5369 %}
5370 
5371 // Float reg-mem operation
5372 pipe_class fpu_reg_mem(regDPR dst, memory mem) %{
5373     instruction_count(2);
5374     dst    : S5(write);
5375     mem    : S3(read);
5376     D0     : S0;        // big decoder only
5377     DECODE : S1;        // any decoder for FPU POP
5378     FPU    : S4;
5379     MEM    : S3;        // any mem
5380 %}
5381 
5382 // Float reg-mem operation
5383 pipe_class fpu_reg_reg_mem(regDPR dst, regDPR src1, memory mem) %{
5384     instruction_count(3);
5385     dst    : S5(write);
5386     src1   : S3(read);
5387     mem    : S3(read);
5388     D0     : S0;        // big decoder only
5389     DECODE : S1(2);     // any decoder for FPU POP
5390     FPU    : S4;
5391     MEM    : S3;        // any mem
5392 %}
5393 
5394 // Float mem-reg operation
5395 pipe_class fpu_mem_reg(memory mem, regDPR src) %{
5396     instruction_count(2);
5397     src    : S5(read);
5398     mem    : S3(read);
5399     DECODE : S0;        // any decoder for FPU PUSH
5400     D0     : S1;        // big decoder only
5401     FPU    : S4;
5402     MEM    : S3;        // any mem
5403 %}
5404 
5405 pipe_class fpu_mem_reg_reg(memory mem, regDPR src1, regDPR src2) %{
5406     instruction_count(3);
5407     src1   : S3(read);
5408     src2   : S3(read);
5409     mem    : S3(read);
5410     DECODE : S0(2);     // any decoder for FPU PUSH
5411     D0     : S1;        // big decoder only
5412     FPU    : S4;
5413     MEM    : S3;        // any mem
5414 %}
5415 
5416 pipe_class fpu_mem_reg_mem(memory mem, regDPR src1, memory src2) %{
5417     instruction_count(3);
5418     src1   : S3(read);
5419     src2   : S3(read);
5420     mem    : S4(read);
5421     DECODE : S0;        // any decoder for FPU PUSH
5422     D0     : S0(2);     // big decoder only
5423     FPU    : S4;
5424     MEM    : S3(2);     // any mem
5425 %}
5426 
5427 pipe_class fpu_mem_mem(memory dst, memory src1) %{
5428     instruction_count(2);
5429     src1   : S3(read);
5430     dst    : S4(read);
5431     D0     : S0(2);     // big decoder only
5432     MEM    : S3(2);     // any mem
5433 %}
5434 
5435 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2) %{
5436     instruction_count(3);
5437     src1   : S3(read);
5438     src2   : S3(read);
5439     dst    : S4(read);
5440     D0     : S0(3);     // big decoder only
5441     FPU    : S4;
5442     MEM    : S3(3);     // any mem
5443 %}
5444 
5445 pipe_class fpu_mem_reg_con(memory mem, regDPR src1) %{
5446     instruction_count(3);
5447     src1   : S4(read);
5448     mem    : S4(read);
5449     DECODE : S0;        // any decoder for FPU PUSH
5450     D0     : S0(2);     // big decoder only
5451     FPU    : S4;
5452     MEM    : S3(2);     // any mem
5453 %}
5454 
5455 // Float load constant
5456 pipe_class fpu_reg_con(regDPR dst) %{
5457     instruction_count(2);
5458     dst    : S5(write);
5459     D0     : S0;        // big decoder only for the load
5460     DECODE : S1;        // any decoder for FPU POP
5461     FPU    : S4;
5462     MEM    : S3;        // any mem
5463 %}
5464 
5465 // Float load constant
5466 pipe_class fpu_reg_reg_con(regDPR dst, regDPR src) %{
5467     instruction_count(3);
5468     dst    : S5(write);
5469     src    : S3(read);
5470     D0     : S0;        // big decoder only for the load
5471     DECODE : S1(2);     // any decoder for FPU POP
5472     FPU    : S4;
5473     MEM    : S3;        // any mem
5474 %}
5475 
5476 // UnConditional branch
5477 pipe_class pipe_jmp( label labl ) %{
5478     single_instruction;
5479     BR   : S3;
5480 %}
5481 
5482 // Conditional branch
5483 pipe_class pipe_jcc( cmpOp cmp, eFlagsReg cr, label labl ) %{
5484     single_instruction;
5485     cr    : S1(read);
5486     BR    : S3;
5487 %}
5488 
5489 // Allocation idiom
5490 pipe_class pipe_cmpxchg( eRegP dst, eRegP heap_ptr ) %{
5491     instruction_count(1); force_serialization;
5492     fixed_latency(6);
5493     heap_ptr : S3(read);
5494     DECODE   : S0(3);
5495     D0       : S2;
5496     MEM      : S3;
5497     ALU      : S3(2);
5498     dst      : S5(write);
5499     BR       : S5;
5500 %}
5501 
5502 // Generic big/slow expanded idiom
5503 pipe_class pipe_slow(  ) %{
5504     instruction_count(10); multiple_bundles; force_serialization;
5505     fixed_latency(100);
5506     D0  : S0(2);
5507     MEM : S3(2);
5508 %}
5509 
5510 // The real do-nothing guy
5511 pipe_class empty( ) %{
5512     instruction_count(0);
5513 %}
5514 
5515 // Define the class for the Nop node
5516 define %{
5517    MachNop = empty;
5518 %}
5519 
5520 %}
5521 
5522 //----------INSTRUCTIONS-------------------------------------------------------
5523 //
5524 // match      -- States which machine-independent subtree may be replaced
5525 //               by this instruction.
5526 // ins_cost   -- The estimated cost of this instruction is used by instruction
5527 //               selection to identify a minimum cost tree of machine
5528 //               instructions that matches a tree of machine-independent
5529 //               instructions.
5530 // format     -- A string providing the disassembly for this instruction.
5531 //               The value of an instruction's operand may be inserted
5532 //               by referring to it with a '$' prefix.
5533 // opcode     -- Three instruction opcodes may be provided.  These are referred
5534 //               to within an encode class as $primary, $secondary, and $tertiary
5535 //               respectively.  The primary opcode is commonly used to
5536 //               indicate the type of machine instruction, while secondary
5537 //               and tertiary are often used for prefix options or addressing
5538 //               modes.
5539 // ins_encode -- A list of encode classes with parameters. The encode class
5540 //               name must have been defined in an 'enc_class' specification
5541 //               in the encode section of the architecture description.
5542 
5543 //----------BSWAP-Instruction--------------------------------------------------
5544 instruct bytes_reverse_int(rRegI dst) %{
5545   match(Set dst (ReverseBytesI dst));
5546 
5547   format %{ "BSWAP  $dst" %}
5548   opcode(0x0F, 0xC8);
5549   ins_encode( OpcP, OpcSReg(dst) );
5550   ins_pipe( ialu_reg );
5551 %}
5552 
5553 instruct bytes_reverse_long(eRegL dst) %{
5554   match(Set dst (ReverseBytesL dst));
5555 
5556   format %{ "BSWAP  $dst.lo\n\t"
5557             "BSWAP  $dst.hi\n\t"
5558             "XCHG   $dst.lo $dst.hi" %}
5559 
5560   ins_cost(125);
5561   ins_encode( bswap_long_bytes(dst) );
5562   ins_pipe( ialu_reg_reg);
5563 %}
5564 
5565 instruct bytes_reverse_unsigned_short(rRegI dst, eFlagsReg cr) %{
5566   match(Set dst (ReverseBytesUS dst));
5567   effect(KILL cr);
5568 
5569   format %{ "BSWAP  $dst\n\t" 
5570             "SHR    $dst,16\n\t" %}
5571   ins_encode %{
5572     __ bswapl($dst$$Register);
5573     __ shrl($dst$$Register, 16); 
5574   %}
5575   ins_pipe( ialu_reg );
5576 %}
5577 
5578 instruct bytes_reverse_short(rRegI dst, eFlagsReg cr) %{
5579   match(Set dst (ReverseBytesS dst));
5580   effect(KILL cr);
5581 
5582   format %{ "BSWAP  $dst\n\t" 
5583             "SAR    $dst,16\n\t" %}
5584   ins_encode %{
5585     __ bswapl($dst$$Register);
5586     __ sarl($dst$$Register, 16); 
5587   %}
5588   ins_pipe( ialu_reg );
5589 %}
5590 
5591 
5592 //---------- Zeros Count Instructions ------------------------------------------
5593 
5594 instruct countLeadingZerosI(rRegI dst, rRegI src, eFlagsReg cr) %{
5595   predicate(UseCountLeadingZerosInstruction);
5596   match(Set dst (CountLeadingZerosI src));
5597   effect(KILL cr);
5598 
5599   format %{ "LZCNT  $dst, $src\t# count leading zeros (int)" %}
5600   ins_encode %{
5601     __ lzcntl($dst$$Register, $src$$Register);
5602   %}
5603   ins_pipe(ialu_reg);
5604 %}
5605 
5606 instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, eFlagsReg cr) %{
5607   predicate(!UseCountLeadingZerosInstruction);
5608   match(Set dst (CountLeadingZerosI src));
5609   effect(KILL cr);
5610 
5611   format %{ "BSR    $dst, $src\t# count leading zeros (int)\n\t"
5612             "JNZ    skip\n\t"
5613             "MOV    $dst, -1\n"
5614       "skip:\n\t"
5615             "NEG    $dst\n\t"
5616             "ADD    $dst, 31" %}
5617   ins_encode %{
5618     Register Rdst = $dst$$Register;
5619     Register Rsrc = $src$$Register;
5620     Label skip;
5621     __ bsrl(Rdst, Rsrc);
5622     __ jccb(Assembler::notZero, skip);
5623     __ movl(Rdst, -1);
5624     __ bind(skip);
5625     __ negl(Rdst);
5626     __ addl(Rdst, BitsPerInt - 1);
5627   %}
5628   ins_pipe(ialu_reg);
5629 %}
5630 
5631 instruct countLeadingZerosL(rRegI dst, eRegL src, eFlagsReg cr) %{
5632   predicate(UseCountLeadingZerosInstruction);
5633   match(Set dst (CountLeadingZerosL src));
5634   effect(TEMP dst, KILL cr);
5635 
5636   format %{ "LZCNT  $dst, $src.hi\t# count leading zeros (long)\n\t"
5637             "JNC    done\n\t"
5638             "LZCNT  $dst, $src.lo\n\t"
5639             "ADD    $dst, 32\n"
5640       "done:" %}
5641   ins_encode %{
5642     Register Rdst = $dst$$Register;
5643     Register Rsrc = $src$$Register;
5644     Label done;
5645     __ lzcntl(Rdst, HIGH_FROM_LOW(Rsrc));
5646     __ jccb(Assembler::carryClear, done);
5647     __ lzcntl(Rdst, Rsrc);
5648     __ addl(Rdst, BitsPerInt);
5649     __ bind(done);
5650   %}
5651   ins_pipe(ialu_reg);
5652 %}
5653 
5654 instruct countLeadingZerosL_bsr(rRegI dst, eRegL src, eFlagsReg cr) %{
5655   predicate(!UseCountLeadingZerosInstruction);
5656   match(Set dst (CountLeadingZerosL src));
5657   effect(TEMP dst, KILL cr);
5658 
5659   format %{ "BSR    $dst, $src.hi\t# count leading zeros (long)\n\t"
5660             "JZ     msw_is_zero\n\t"
5661             "ADD    $dst, 32\n\t"
5662             "JMP    not_zero\n"
5663       "msw_is_zero:\n\t"
5664             "BSR    $dst, $src.lo\n\t"
5665             "JNZ    not_zero\n\t"
5666             "MOV    $dst, -1\n"
5667       "not_zero:\n\t"
5668             "NEG    $dst\n\t"
5669             "ADD    $dst, 63\n" %}
5670  ins_encode %{
5671     Register Rdst = $dst$$Register;
5672     Register Rsrc = $src$$Register;
5673     Label msw_is_zero;
5674     Label not_zero;
5675     __ bsrl(Rdst, HIGH_FROM_LOW(Rsrc));
5676     __ jccb(Assembler::zero, msw_is_zero);
5677     __ addl(Rdst, BitsPerInt);
5678     __ jmpb(not_zero);
5679     __ bind(msw_is_zero);
5680     __ bsrl(Rdst, Rsrc);
5681     __ jccb(Assembler::notZero, not_zero);
5682     __ movl(Rdst, -1);
5683     __ bind(not_zero);
5684     __ negl(Rdst);
5685     __ addl(Rdst, BitsPerLong - 1);
5686   %}
5687   ins_pipe(ialu_reg);
5688 %}
5689 
5690 instruct countTrailingZerosI(rRegI dst, rRegI src, eFlagsReg cr) %{
5691   match(Set dst (CountTrailingZerosI src));
5692   effect(KILL cr);
5693 
5694   format %{ "BSF    $dst, $src\t# count trailing zeros (int)\n\t"
5695             "JNZ    done\n\t"
5696             "MOV    $dst, 32\n"
5697       "done:" %}
5698   ins_encode %{
5699     Register Rdst = $dst$$Register;
5700     Label done;
5701     __ bsfl(Rdst, $src$$Register);
5702     __ jccb(Assembler::notZero, done);
5703     __ movl(Rdst, BitsPerInt);
5704     __ bind(done);
5705   %}
5706   ins_pipe(ialu_reg);
5707 %}
5708 
5709 instruct countTrailingZerosL(rRegI dst, eRegL src, eFlagsReg cr) %{
5710   match(Set dst (CountTrailingZerosL src));
5711   effect(TEMP dst, KILL cr);
5712 
5713   format %{ "BSF    $dst, $src.lo\t# count trailing zeros (long)\n\t"
5714             "JNZ    done\n\t"
5715             "BSF    $dst, $src.hi\n\t"
5716             "JNZ    msw_not_zero\n\t"
5717             "MOV    $dst, 32\n"
5718       "msw_not_zero:\n\t"
5719             "ADD    $dst, 32\n"
5720       "done:" %}
5721   ins_encode %{
5722     Register Rdst = $dst$$Register;
5723     Register Rsrc = $src$$Register;
5724     Label msw_not_zero;
5725     Label done;
5726     __ bsfl(Rdst, Rsrc);
5727     __ jccb(Assembler::notZero, done);
5728     __ bsfl(Rdst, HIGH_FROM_LOW(Rsrc));
5729     __ jccb(Assembler::notZero, msw_not_zero);
5730     __ movl(Rdst, BitsPerInt);
5731     __ bind(msw_not_zero);
5732     __ addl(Rdst, BitsPerInt);
5733     __ bind(done);
5734   %}
5735   ins_pipe(ialu_reg);
5736 %}
5737 
5738 
5739 //---------- Population Count Instructions -------------------------------------
5740 
5741 instruct popCountI(rRegI dst, rRegI src, eFlagsReg cr) %{
5742   predicate(UsePopCountInstruction);
5743   match(Set dst (PopCountI src));
5744   effect(KILL cr);
5745 
5746   format %{ "POPCNT $dst, $src" %}
5747   ins_encode %{
5748     __ popcntl($dst$$Register, $src$$Register);
5749   %}
5750   ins_pipe(ialu_reg);
5751 %}
5752 
5753 instruct popCountI_mem(rRegI dst, memory mem, eFlagsReg cr) %{
5754   predicate(UsePopCountInstruction);
5755   match(Set dst (PopCountI (LoadI mem)));
5756   effect(KILL cr);
5757 
5758   format %{ "POPCNT $dst, $mem" %}
5759   ins_encode %{
5760     __ popcntl($dst$$Register, $mem$$Address);
5761   %}
5762   ins_pipe(ialu_reg);
5763 %}
5764 
5765 // Note: Long.bitCount(long) returns an int.
5766 instruct popCountL(rRegI dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
5767   predicate(UsePopCountInstruction);
5768   match(Set dst (PopCountL src));
5769   effect(KILL cr, TEMP tmp, TEMP dst);
5770 
5771   format %{ "POPCNT $dst, $src.lo\n\t"
5772             "POPCNT $tmp, $src.hi\n\t"
5773             "ADD    $dst, $tmp" %}
5774   ins_encode %{
5775     __ popcntl($dst$$Register, $src$$Register);
5776     __ popcntl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
5777     __ addl($dst$$Register, $tmp$$Register);
5778   %}
5779   ins_pipe(ialu_reg);
5780 %}
5781 
5782 // Note: Long.bitCount(long) returns an int.
5783 instruct popCountL_mem(rRegI dst, memory mem, rRegI tmp, eFlagsReg cr) %{
5784   predicate(UsePopCountInstruction);
5785   match(Set dst (PopCountL (LoadL mem)));
5786   effect(KILL cr, TEMP tmp, TEMP dst);
5787 
5788   format %{ "POPCNT $dst, $mem\n\t"
5789             "POPCNT $tmp, $mem+4\n\t"
5790             "ADD    $dst, $tmp" %}
5791   ins_encode %{
5792     //__ popcntl($dst$$Register, $mem$$Address$$first);
5793     //__ popcntl($tmp$$Register, $mem$$Address$$second);
5794     __ popcntl($dst$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, relocInfo::none));
5795     __ popcntl($tmp$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, relocInfo::none));
5796     __ addl($dst$$Register, $tmp$$Register);
5797   %}
5798   ins_pipe(ialu_reg);
5799 %}
5800 
5801 
5802 //----------Load/Store/Move Instructions---------------------------------------
5803 //----------Load Instructions--------------------------------------------------
5804 // Load Byte (8bit signed)
5805 instruct loadB(xRegI dst, memory mem) %{
5806   match(Set dst (LoadB mem));
5807 
5808   ins_cost(125);
5809   format %{ "MOVSX8 $dst,$mem\t# byte" %}
5810 
5811   ins_encode %{
5812     __ movsbl($dst$$Register, $mem$$Address);
5813   %}
5814 
5815   ins_pipe(ialu_reg_mem);
5816 %}
5817 
5818 // Load Byte (8bit signed) into Long Register
5819 instruct loadB2L(eRegL dst, memory mem, eFlagsReg cr) %{
5820   match(Set dst (ConvI2L (LoadB mem)));
5821   effect(KILL cr);
5822 
5823   ins_cost(375);
5824   format %{ "MOVSX8 $dst.lo,$mem\t# byte -> long\n\t"
5825             "MOV    $dst.hi,$dst.lo\n\t"
5826             "SAR    $dst.hi,7" %}
5827 
5828   ins_encode %{
5829     __ movsbl($dst$$Register, $mem$$Address);
5830     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
5831     __ sarl(HIGH_FROM_LOW($dst$$Register), 7); // 24+1 MSB are already signed extended.
5832   %}
5833 
5834   ins_pipe(ialu_reg_mem);
5835 %}
5836 
5837 // Load Unsigned Byte (8bit UNsigned)
5838 instruct loadUB(xRegI dst, memory mem) %{
5839   match(Set dst (LoadUB mem));
5840 
5841   ins_cost(125);
5842   format %{ "MOVZX8 $dst,$mem\t# ubyte -> int" %}
5843 
5844   ins_encode %{
5845     __ movzbl($dst$$Register, $mem$$Address);
5846   %}
5847 
5848   ins_pipe(ialu_reg_mem);
5849 %}
5850 
5851 // Load Unsigned Byte (8 bit UNsigned) into Long Register
5852 instruct loadUB2L(eRegL dst, memory mem, eFlagsReg cr) %{
5853   match(Set dst (ConvI2L (LoadUB mem)));
5854   effect(KILL cr);
5855 
5856   ins_cost(250);
5857   format %{ "MOVZX8 $dst.lo,$mem\t# ubyte -> long\n\t"
5858             "XOR    $dst.hi,$dst.hi" %}
5859 
5860   ins_encode %{
5861     Register Rdst = $dst$$Register;
5862     __ movzbl(Rdst, $mem$$Address);
5863     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5864   %}
5865 
5866   ins_pipe(ialu_reg_mem);
5867 %}
5868 
5869 // Load Unsigned Byte (8 bit UNsigned) with mask into Long Register
5870 instruct loadUB2L_immI8(eRegL dst, memory mem, immI8 mask, eFlagsReg cr) %{
5871   match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5872   effect(KILL cr);
5873 
5874   format %{ "MOVZX8 $dst.lo,$mem\t# ubyte & 8-bit mask -> long\n\t"
5875             "XOR    $dst.hi,$dst.hi\n\t"
5876             "AND    $dst.lo,$mask" %}
5877   ins_encode %{
5878     Register Rdst = $dst$$Register;
5879     __ movzbl(Rdst, $mem$$Address);
5880     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5881     __ andl(Rdst, $mask$$constant);
5882   %}
5883   ins_pipe(ialu_reg_mem);
5884 %}
5885 
5886 // Load Short (16bit signed)
5887 instruct loadS(rRegI dst, memory mem) %{
5888   match(Set dst (LoadS mem));
5889 
5890   ins_cost(125);
5891   format %{ "MOVSX  $dst,$mem\t# short" %}
5892 
5893   ins_encode %{
5894     __ movswl($dst$$Register, $mem$$Address);
5895   %}
5896 
5897   ins_pipe(ialu_reg_mem);
5898 %}
5899 
5900 // Load Short (16 bit signed) to Byte (8 bit signed)
5901 instruct loadS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5902   match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5903 
5904   ins_cost(125);
5905   format %{ "MOVSX  $dst, $mem\t# short -> byte" %}
5906   ins_encode %{
5907     __ movsbl($dst$$Register, $mem$$Address);
5908   %}
5909   ins_pipe(ialu_reg_mem);
5910 %}
5911 
5912 // Load Short (16bit signed) into Long Register
5913 instruct loadS2L(eRegL dst, memory mem, eFlagsReg cr) %{
5914   match(Set dst (ConvI2L (LoadS mem)));
5915   effect(KILL cr);
5916 
5917   ins_cost(375);
5918   format %{ "MOVSX  $dst.lo,$mem\t# short -> long\n\t"
5919             "MOV    $dst.hi,$dst.lo\n\t"
5920             "SAR    $dst.hi,15" %}
5921 
5922   ins_encode %{
5923     __ movswl($dst$$Register, $mem$$Address);
5924     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
5925     __ sarl(HIGH_FROM_LOW($dst$$Register), 15); // 16+1 MSB are already signed extended.
5926   %}
5927 
5928   ins_pipe(ialu_reg_mem);
5929 %}
5930 
5931 // Load Unsigned Short/Char (16bit unsigned)
5932 instruct loadUS(rRegI dst, memory mem) %{
5933   match(Set dst (LoadUS mem));
5934 
5935   ins_cost(125);
5936   format %{ "MOVZX  $dst,$mem\t# ushort/char -> int" %}
5937 
5938   ins_encode %{
5939     __ movzwl($dst$$Register, $mem$$Address);
5940   %}
5941 
5942   ins_pipe(ialu_reg_mem);
5943 %}
5944 
5945 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5946 instruct loadUS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5947   match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5948 
5949   ins_cost(125);
5950   format %{ "MOVSX  $dst, $mem\t# ushort -> byte" %}
5951   ins_encode %{
5952     __ movsbl($dst$$Register, $mem$$Address);
5953   %}
5954   ins_pipe(ialu_reg_mem);
5955 %}
5956 
5957 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
5958 instruct loadUS2L(eRegL dst, memory mem, eFlagsReg cr) %{
5959   match(Set dst (ConvI2L (LoadUS mem)));
5960   effect(KILL cr);
5961 
5962   ins_cost(250);
5963   format %{ "MOVZX  $dst.lo,$mem\t# ushort/char -> long\n\t"
5964             "XOR    $dst.hi,$dst.hi" %}
5965 
5966   ins_encode %{
5967     __ movzwl($dst$$Register, $mem$$Address);
5968     __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
5969   %}
5970 
5971   ins_pipe(ialu_reg_mem);
5972 %}
5973 
5974 // Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
5975 instruct loadUS2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
5976   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5977   effect(KILL cr);
5978 
5979   format %{ "MOVZX8 $dst.lo,$mem\t# ushort/char & 0xFF -> long\n\t"
5980             "XOR    $dst.hi,$dst.hi" %}
5981   ins_encode %{
5982     Register Rdst = $dst$$Register;
5983     __ movzbl(Rdst, $mem$$Address);
5984     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5985   %}
5986   ins_pipe(ialu_reg_mem);
5987 %}
5988 
5989 // Load Unsigned Short/Char (16 bit UNsigned) with a 16-bit mask into Long Register
5990 instruct loadUS2L_immI16(eRegL dst, memory mem, immI16 mask, eFlagsReg cr) %{
5991   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5992   effect(KILL cr);
5993 
5994   format %{ "MOVZX  $dst.lo, $mem\t# ushort/char & 16-bit mask -> long\n\t"
5995             "XOR    $dst.hi,$dst.hi\n\t"
5996             "AND    $dst.lo,$mask" %}
5997   ins_encode %{
5998     Register Rdst = $dst$$Register;
5999     __ movzwl(Rdst, $mem$$Address);
6000     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6001     __ andl(Rdst, $mask$$constant);
6002   %}
6003   ins_pipe(ialu_reg_mem);
6004 %}
6005 
6006 // Load Integer
6007 instruct loadI(rRegI dst, memory mem) %{
6008   match(Set dst (LoadI mem));
6009 
6010   ins_cost(125);
6011   format %{ "MOV    $dst,$mem\t# int" %}
6012 
6013   ins_encode %{
6014     __ movl($dst$$Register, $mem$$Address);
6015   %}
6016 
6017   ins_pipe(ialu_reg_mem);
6018 %}
6019 
6020 // Load Integer (32 bit signed) to Byte (8 bit signed)
6021 instruct loadI2B(rRegI dst, memory mem, immI_24 twentyfour) %{
6022   match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
6023 
6024   ins_cost(125);
6025   format %{ "MOVSX  $dst, $mem\t# int -> byte" %}
6026   ins_encode %{
6027     __ movsbl($dst$$Register, $mem$$Address);
6028   %}
6029   ins_pipe(ialu_reg_mem);
6030 %}
6031 
6032 // Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
6033 instruct loadI2UB(rRegI dst, memory mem, immI_255 mask) %{
6034   match(Set dst (AndI (LoadI mem) mask));
6035 
6036   ins_cost(125);
6037   format %{ "MOVZX  $dst, $mem\t# int -> ubyte" %}
6038   ins_encode %{
6039     __ movzbl($dst$$Register, $mem$$Address);
6040   %}
6041   ins_pipe(ialu_reg_mem);
6042 %}
6043 
6044 // Load Integer (32 bit signed) to Short (16 bit signed)
6045 instruct loadI2S(rRegI dst, memory mem, immI_16 sixteen) %{
6046   match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
6047 
6048   ins_cost(125);
6049   format %{ "MOVSX  $dst, $mem\t# int -> short" %}
6050   ins_encode %{
6051     __ movswl($dst$$Register, $mem$$Address);
6052   %}
6053   ins_pipe(ialu_reg_mem);
6054 %}
6055 
6056 // Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
6057 instruct loadI2US(rRegI dst, memory mem, immI_65535 mask) %{
6058   match(Set dst (AndI (LoadI mem) mask));
6059 
6060   ins_cost(125);
6061   format %{ "MOVZX  $dst, $mem\t# int -> ushort/char" %}
6062   ins_encode %{
6063     __ movzwl($dst$$Register, $mem$$Address);
6064   %}
6065   ins_pipe(ialu_reg_mem);
6066 %}
6067 
6068 // Load Integer into Long Register
6069 instruct loadI2L(eRegL dst, memory mem, eFlagsReg cr) %{
6070   match(Set dst (ConvI2L (LoadI mem)));
6071   effect(KILL cr);
6072 
6073   ins_cost(375);
6074   format %{ "MOV    $dst.lo,$mem\t# int -> long\n\t"
6075             "MOV    $dst.hi,$dst.lo\n\t"
6076             "SAR    $dst.hi,31" %}
6077 
6078   ins_encode %{
6079     __ movl($dst$$Register, $mem$$Address);
6080     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
6081     __ sarl(HIGH_FROM_LOW($dst$$Register), 31);
6082   %}
6083 
6084   ins_pipe(ialu_reg_mem);
6085 %}
6086 
6087 // Load Integer with mask 0xFF into Long Register
6088 instruct loadI2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
6089   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6090   effect(KILL cr);
6091 
6092   format %{ "MOVZX8 $dst.lo,$mem\t# int & 0xFF -> long\n\t"
6093             "XOR    $dst.hi,$dst.hi" %}
6094   ins_encode %{
6095     Register Rdst = $dst$$Register;
6096     __ movzbl(Rdst, $mem$$Address);
6097     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6098   %}
6099   ins_pipe(ialu_reg_mem);
6100 %}
6101 
6102 // Load Integer with mask 0xFFFF into Long Register
6103 instruct loadI2L_immI_65535(eRegL dst, memory mem, immI_65535 mask, eFlagsReg cr) %{
6104   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6105   effect(KILL cr);
6106 
6107   format %{ "MOVZX  $dst.lo,$mem\t# int & 0xFFFF -> long\n\t"
6108             "XOR    $dst.hi,$dst.hi" %}
6109   ins_encode %{
6110     Register Rdst = $dst$$Register;
6111     __ movzwl(Rdst, $mem$$Address);
6112     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6113   %}
6114   ins_pipe(ialu_reg_mem);
6115 %}
6116 
6117 // Load Integer with 32-bit mask into Long Register
6118 instruct loadI2L_immI(eRegL dst, memory mem, immI mask, eFlagsReg cr) %{
6119   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6120   effect(KILL cr);
6121 
6122   format %{ "MOV    $dst.lo,$mem\t# int & 32-bit mask -> long\n\t"
6123             "XOR    $dst.hi,$dst.hi\n\t"
6124             "AND    $dst.lo,$mask" %}
6125   ins_encode %{
6126     Register Rdst = $dst$$Register;
6127     __ movl(Rdst, $mem$$Address);
6128     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6129     __ andl(Rdst, $mask$$constant);
6130   %}
6131   ins_pipe(ialu_reg_mem);
6132 %}
6133 
6134 // Load Unsigned Integer into Long Register
6135 instruct loadUI2L(eRegL dst, memory mem, immL_32bits mask, eFlagsReg cr) %{
6136   match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
6137   effect(KILL cr);
6138 
6139   ins_cost(250);
6140   format %{ "MOV    $dst.lo,$mem\t# uint -> long\n\t"
6141             "XOR    $dst.hi,$dst.hi" %}
6142 
6143   ins_encode %{
6144     __ movl($dst$$Register, $mem$$Address);
6145     __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
6146   %}
6147 
6148   ins_pipe(ialu_reg_mem);
6149 %}
6150 
6151 // Load Long.  Cannot clobber address while loading, so restrict address
6152 // register to ESI
6153 instruct loadL(eRegL dst, load_long_memory mem) %{
6154   predicate(!((LoadLNode*)n)->require_atomic_access());
6155   match(Set dst (LoadL mem));
6156 
6157   ins_cost(250);
6158   format %{ "MOV    $dst.lo,$mem\t# long\n\t"
6159             "MOV    $dst.hi,$mem+4" %}
6160 
6161   ins_encode %{
6162     Address Amemlo = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, relocInfo::none);
6163     Address Amemhi = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, relocInfo::none);
6164     __ movl($dst$$Register, Amemlo);
6165     __ movl(HIGH_FROM_LOW($dst$$Register), Amemhi);
6166   %}
6167 
6168   ins_pipe(ialu_reg_long_mem);
6169 %}
6170 
6171 // Volatile Load Long.  Must be atomic, so do 64-bit FILD
6172 // then store it down to the stack and reload on the int
6173 // side.
6174 instruct loadL_volatile(stackSlotL dst, memory mem) %{
6175   predicate(UseSSE<=1 && ((LoadLNode*)n)->require_atomic_access());
6176   match(Set dst (LoadL mem));
6177 
6178   ins_cost(200);
6179   format %{ "FILD   $mem\t# Atomic volatile long load\n\t"
6180             "FISTp  $dst" %}
6181   ins_encode(enc_loadL_volatile(mem,dst));
6182   ins_pipe( fpu_reg_mem );
6183 %}
6184 
6185 instruct loadLX_volatile(stackSlotL dst, memory mem, regD tmp) %{
6186   predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
6187   match(Set dst (LoadL mem));
6188   effect(TEMP tmp);
6189   ins_cost(180);
6190   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
6191             "MOVSD  $dst,$tmp" %}
6192   ins_encode %{
6193     __ movdbl($tmp$$XMMRegister, $mem$$Address);
6194     __ movdbl(Address(rsp, $dst$$disp), $tmp$$XMMRegister);
6195   %}
6196   ins_pipe( pipe_slow );
6197 %}
6198 
6199 instruct loadLX_reg_volatile(eRegL dst, memory mem, regD tmp) %{
6200   predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
6201   match(Set dst (LoadL mem));
6202   effect(TEMP tmp);
6203   ins_cost(160);
6204   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
6205             "MOVD   $dst.lo,$tmp\n\t"
6206             "PSRLQ  $tmp,32\n\t"
6207             "MOVD   $dst.hi,$tmp" %}
6208   ins_encode %{
6209     __ movdbl($tmp$$XMMRegister, $mem$$Address);
6210     __ movdl($dst$$Register, $tmp$$XMMRegister);
6211     __ psrlq($tmp$$XMMRegister, 32);
6212     __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
6213   %}
6214   ins_pipe( pipe_slow );
6215 %}
6216 
6217 // Load Range
6218 instruct loadRange(rRegI dst, memory mem) %{
6219   match(Set dst (LoadRange mem));
6220 
6221   ins_cost(125);
6222   format %{ "MOV    $dst,$mem" %}
6223   opcode(0x8B);
6224   ins_encode( OpcP, RegMem(dst,mem));
6225   ins_pipe( ialu_reg_mem );
6226 %}
6227 
6228 
6229 // Load Pointer
6230 instruct loadP(eRegP dst, memory mem) %{
6231   match(Set dst (LoadP mem));
6232 
6233   ins_cost(125);
6234   format %{ "MOV    $dst,$mem" %}
6235   opcode(0x8B);
6236   ins_encode( OpcP, RegMem(dst,mem));
6237   ins_pipe( ialu_reg_mem );
6238 %}
6239 
6240 // Load Klass Pointer
6241 instruct loadKlass(eRegP dst, memory mem) %{
6242   match(Set dst (LoadKlass mem));
6243 
6244   ins_cost(125);
6245   format %{ "MOV    $dst,$mem" %}
6246   opcode(0x8B);
6247   ins_encode( OpcP, RegMem(dst,mem));
6248   ins_pipe( ialu_reg_mem );
6249 %}
6250 
6251 // Load Double
6252 instruct loadDPR(regDPR dst, memory mem) %{
6253   predicate(UseSSE<=1);
6254   match(Set dst (LoadD mem));
6255 
6256   ins_cost(150);
6257   format %{ "FLD_D  ST,$mem\n\t"
6258             "FSTP   $dst" %}
6259   opcode(0xDD);               /* DD /0 */
6260   ins_encode( OpcP, RMopc_Mem(0x00,mem),
6261               Pop_Reg_DPR(dst) );
6262   ins_pipe( fpu_reg_mem );
6263 %}
6264 
6265 // Load Double to XMM
6266 instruct loadD(regD dst, memory mem) %{
6267   predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
6268   match(Set dst (LoadD mem));
6269   ins_cost(145);
6270   format %{ "MOVSD  $dst,$mem" %}
6271   ins_encode %{
6272     __ movdbl ($dst$$XMMRegister, $mem$$Address);
6273   %}
6274   ins_pipe( pipe_slow );
6275 %}
6276 
6277 instruct loadD_partial(regD dst, memory mem) %{
6278   predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
6279   match(Set dst (LoadD mem));
6280   ins_cost(145);
6281   format %{ "MOVLPD $dst,$mem" %}
6282   ins_encode %{
6283     __ movdbl ($dst$$XMMRegister, $mem$$Address);
6284   %}
6285   ins_pipe( pipe_slow );
6286 %}
6287 
6288 // Load to XMM register (single-precision floating point)
6289 // MOVSS instruction
6290 instruct loadF(regF dst, memory mem) %{
6291   predicate(UseSSE>=1);
6292   match(Set dst (LoadF mem));
6293   ins_cost(145);
6294   format %{ "MOVSS  $dst,$mem" %}
6295   ins_encode %{
6296     __ movflt ($dst$$XMMRegister, $mem$$Address);
6297   %}
6298   ins_pipe( pipe_slow );
6299 %}
6300 
6301 // Load Float
6302 instruct loadFPR(regFPR dst, memory mem) %{
6303   predicate(UseSSE==0);
6304   match(Set dst (LoadF mem));
6305 
6306   ins_cost(150);
6307   format %{ "FLD_S  ST,$mem\n\t"
6308             "FSTP   $dst" %}
6309   opcode(0xD9);               /* D9 /0 */
6310   ins_encode( OpcP, RMopc_Mem(0x00,mem),
6311               Pop_Reg_FPR(dst) );
6312   ins_pipe( fpu_reg_mem );
6313 %}
6314 
6315 // Load Effective Address
6316 instruct leaP8(eRegP dst, indOffset8 mem) %{
6317   match(Set dst mem);
6318 
6319   ins_cost(110);
6320   format %{ "LEA    $dst,$mem" %}
6321   opcode(0x8D);
6322   ins_encode( OpcP, RegMem(dst,mem));
6323   ins_pipe( ialu_reg_reg_fat );
6324 %}
6325 
6326 instruct leaP32(eRegP dst, indOffset32 mem) %{
6327   match(Set dst mem);
6328 
6329   ins_cost(110);
6330   format %{ "LEA    $dst,$mem" %}
6331   opcode(0x8D);
6332   ins_encode( OpcP, RegMem(dst,mem));
6333   ins_pipe( ialu_reg_reg_fat );
6334 %}
6335 
6336 instruct leaPIdxOff(eRegP dst, indIndexOffset mem) %{
6337   match(Set dst mem);
6338 
6339   ins_cost(110);
6340   format %{ "LEA    $dst,$mem" %}
6341   opcode(0x8D);
6342   ins_encode( OpcP, RegMem(dst,mem));
6343   ins_pipe( ialu_reg_reg_fat );
6344 %}
6345 
6346 instruct leaPIdxScale(eRegP dst, indIndexScale mem) %{
6347   match(Set dst mem);
6348 
6349   ins_cost(110);
6350   format %{ "LEA    $dst,$mem" %}
6351   opcode(0x8D);
6352   ins_encode( OpcP, RegMem(dst,mem));
6353   ins_pipe( ialu_reg_reg_fat );
6354 %}
6355 
6356 instruct leaPIdxScaleOff(eRegP dst, indIndexScaleOffset mem) %{
6357   match(Set dst mem);
6358 
6359   ins_cost(110);
6360   format %{ "LEA    $dst,$mem" %}
6361   opcode(0x8D);
6362   ins_encode( OpcP, RegMem(dst,mem));
6363   ins_pipe( ialu_reg_reg_fat );
6364 %}
6365 
6366 // Load Constant
6367 instruct loadConI(rRegI dst, immI src) %{
6368   match(Set dst src);
6369 
6370   format %{ "MOV    $dst,$src" %}
6371   ins_encode( LdImmI(dst, src) );
6372   ins_pipe( ialu_reg_fat );
6373 %}
6374 
6375 // Load Constant zero
6376 instruct loadConI0(rRegI dst, immI0 src, eFlagsReg cr) %{
6377   match(Set dst src);
6378   effect(KILL cr);
6379 
6380   ins_cost(50);
6381   format %{ "XOR    $dst,$dst" %}
6382   opcode(0x33);  /* + rd */
6383   ins_encode( OpcP, RegReg( dst, dst ) );
6384   ins_pipe( ialu_reg );
6385 %}
6386 
6387 instruct loadConP(eRegP dst, immP src) %{
6388   match(Set dst src);
6389 
6390   format %{ "MOV    $dst,$src" %}
6391   opcode(0xB8);  /* + rd */
6392   ins_encode( LdImmP(dst, src) );
6393   ins_pipe( ialu_reg_fat );
6394 %}
6395 
6396 instruct loadConL(eRegL dst, immL src, eFlagsReg cr) %{
6397   match(Set dst src);
6398   effect(KILL cr);
6399   ins_cost(200);
6400   format %{ "MOV    $dst.lo,$src.lo\n\t"
6401             "MOV    $dst.hi,$src.hi" %}
6402   opcode(0xB8);
6403   ins_encode( LdImmL_Lo(dst, src), LdImmL_Hi(dst, src) );
6404   ins_pipe( ialu_reg_long_fat );
6405 %}
6406 
6407 instruct loadConL0(eRegL dst, immL0 src, eFlagsReg cr) %{
6408   match(Set dst src);
6409   effect(KILL cr);
6410   ins_cost(150);
6411   format %{ "XOR    $dst.lo,$dst.lo\n\t"
6412             "XOR    $dst.hi,$dst.hi" %}
6413   opcode(0x33,0x33);
6414   ins_encode( RegReg_Lo(dst,dst), RegReg_Hi(dst, dst) );
6415   ins_pipe( ialu_reg_long );
6416 %}
6417 
6418 // The instruction usage is guarded by predicate in operand immFPR().
6419 instruct loadConFPR(regFPR dst, immFPR con) %{
6420   match(Set dst con);
6421   ins_cost(125);
6422   format %{ "FLD_S  ST,[$constantaddress]\t# load from constant table: float=$con\n\t"
6423             "FSTP   $dst" %}
6424   ins_encode %{
6425     __ fld_s($constantaddress($con));
6426     __ fstp_d($dst$$reg);
6427   %}
6428   ins_pipe(fpu_reg_con);
6429 %}
6430 
6431 // The instruction usage is guarded by predicate in operand immFPR0().
6432 instruct loadConFPR0(regFPR dst, immFPR0 con) %{
6433   match(Set dst con);
6434   ins_cost(125);
6435   format %{ "FLDZ   ST\n\t"
6436             "FSTP   $dst" %}
6437   ins_encode %{
6438     __ fldz();
6439     __ fstp_d($dst$$reg);
6440   %}
6441   ins_pipe(fpu_reg_con);
6442 %}
6443 
6444 // The instruction usage is guarded by predicate in operand immFPR1().
6445 instruct loadConFPR1(regFPR dst, immFPR1 con) %{
6446   match(Set dst con);
6447   ins_cost(125);
6448   format %{ "FLD1   ST\n\t"
6449             "FSTP   $dst" %}
6450   ins_encode %{
6451     __ fld1();
6452     __ fstp_d($dst$$reg);
6453   %}
6454   ins_pipe(fpu_reg_con);
6455 %}
6456 
6457 // The instruction usage is guarded by predicate in operand immF().
6458 instruct loadConF(regF dst, immF con) %{
6459   match(Set dst con);
6460   ins_cost(125);
6461   format %{ "MOVSS  $dst,[$constantaddress]\t# load from constant table: float=$con" %}
6462   ins_encode %{
6463     __ movflt($dst$$XMMRegister, $constantaddress($con));
6464   %}
6465   ins_pipe(pipe_slow);
6466 %}
6467 
6468 // The instruction usage is guarded by predicate in operand immF0().
6469 instruct loadConF0(regF dst, immF0 src) %{
6470   match(Set dst src);
6471   ins_cost(100);
6472   format %{ "XORPS  $dst,$dst\t# float 0.0" %}
6473   ins_encode %{
6474     __ xorps($dst$$XMMRegister, $dst$$XMMRegister);
6475   %}
6476   ins_pipe(pipe_slow);
6477 %}
6478 
6479 // The instruction usage is guarded by predicate in operand immDPR().
6480 instruct loadConDPR(regDPR dst, immDPR con) %{
6481   match(Set dst con);
6482   ins_cost(125);
6483 
6484   format %{ "FLD_D  ST,[$constantaddress]\t# load from constant table: double=$con\n\t"
6485             "FSTP   $dst" %}
6486   ins_encode %{
6487     __ fld_d($constantaddress($con));
6488     __ fstp_d($dst$$reg);
6489   %}
6490   ins_pipe(fpu_reg_con);
6491 %}
6492 
6493 // The instruction usage is guarded by predicate in operand immDPR0().
6494 instruct loadConDPR0(regDPR dst, immDPR0 con) %{
6495   match(Set dst con);
6496   ins_cost(125);
6497 
6498   format %{ "FLDZ   ST\n\t"
6499             "FSTP   $dst" %}
6500   ins_encode %{
6501     __ fldz();
6502     __ fstp_d($dst$$reg);
6503   %}
6504   ins_pipe(fpu_reg_con);
6505 %}
6506 
6507 // The instruction usage is guarded by predicate in operand immDPR1().
6508 instruct loadConDPR1(regDPR dst, immDPR1 con) %{
6509   match(Set dst con);
6510   ins_cost(125);
6511 
6512   format %{ "FLD1   ST\n\t"
6513             "FSTP   $dst" %}
6514   ins_encode %{
6515     __ fld1();
6516     __ fstp_d($dst$$reg);
6517   %}
6518   ins_pipe(fpu_reg_con);
6519 %}
6520 
6521 // The instruction usage is guarded by predicate in operand immD().
6522 instruct loadConD(regD dst, immD con) %{
6523   match(Set dst con);
6524   ins_cost(125);
6525   format %{ "MOVSD  $dst,[$constantaddress]\t# load from constant table: double=$con" %}
6526   ins_encode %{
6527     __ movdbl($dst$$XMMRegister, $constantaddress($con));
6528   %}
6529   ins_pipe(pipe_slow);
6530 %}
6531 
6532 // The instruction usage is guarded by predicate in operand immD0().
6533 instruct loadConD0(regD dst, immD0 src) %{
6534   match(Set dst src);
6535   ins_cost(100);
6536   format %{ "XORPD  $dst,$dst\t# double 0.0" %}
6537   ins_encode %{
6538     __ xorpd ($dst$$XMMRegister, $dst$$XMMRegister);
6539   %}
6540   ins_pipe( pipe_slow );
6541 %}
6542 
6543 // Load Stack Slot
6544 instruct loadSSI(rRegI dst, stackSlotI src) %{
6545   match(Set dst src);
6546   ins_cost(125);
6547 
6548   format %{ "MOV    $dst,$src" %}
6549   opcode(0x8B);
6550   ins_encode( OpcP, RegMem(dst,src));
6551   ins_pipe( ialu_reg_mem );
6552 %}
6553 
6554 instruct loadSSL(eRegL dst, stackSlotL src) %{
6555   match(Set dst src);
6556 
6557   ins_cost(200);
6558   format %{ "MOV    $dst,$src.lo\n\t"
6559             "MOV    $dst+4,$src.hi" %}
6560   opcode(0x8B, 0x8B);
6561   ins_encode( OpcP, RegMem( dst, src ), OpcS, RegMem_Hi( dst, src ) );
6562   ins_pipe( ialu_mem_long_reg );
6563 %}
6564 
6565 // Load Stack Slot
6566 instruct loadSSP(eRegP dst, stackSlotP src) %{
6567   match(Set dst src);
6568   ins_cost(125);
6569 
6570   format %{ "MOV    $dst,$src" %}
6571   opcode(0x8B);
6572   ins_encode( OpcP, RegMem(dst,src));
6573   ins_pipe( ialu_reg_mem );
6574 %}
6575 
6576 // Load Stack Slot
6577 instruct loadSSF(regFPR dst, stackSlotF src) %{
6578   match(Set dst src);
6579   ins_cost(125);
6580 
6581   format %{ "FLD_S  $src\n\t"
6582             "FSTP   $dst" %}
6583   opcode(0xD9);               /* D9 /0, FLD m32real */
6584   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
6585               Pop_Reg_FPR(dst) );
6586   ins_pipe( fpu_reg_mem );
6587 %}
6588 
6589 // Load Stack Slot
6590 instruct loadSSD(regDPR dst, stackSlotD src) %{
6591   match(Set dst src);
6592   ins_cost(125);
6593 
6594   format %{ "FLD_D  $src\n\t"
6595             "FSTP   $dst" %}
6596   opcode(0xDD);               /* DD /0, FLD m64real */
6597   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
6598               Pop_Reg_DPR(dst) );
6599   ins_pipe( fpu_reg_mem );
6600 %}
6601 
6602 // Prefetch instructions.
6603 // Must be safe to execute with invalid address (cannot fault).
6604 
6605 instruct prefetchr0( memory mem ) %{
6606   predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
6607   match(PrefetchRead mem);
6608   ins_cost(0);
6609   size(0);
6610   format %{ "PREFETCHR (non-SSE is empty encoding)" %}
6611   ins_encode();
6612   ins_pipe(empty);
6613 %}
6614 
6615 instruct prefetchr( memory mem ) %{
6616   predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch() || ReadPrefetchInstr==3);
6617   match(PrefetchRead mem);
6618   ins_cost(100);
6619 
6620   format %{ "PREFETCHR $mem\t! Prefetch into level 1 cache for read" %}
6621   ins_encode %{
6622     __ prefetchr($mem$$Address);
6623   %}
6624   ins_pipe(ialu_mem);
6625 %}
6626 
6627 instruct prefetchrNTA( memory mem ) %{
6628   predicate(UseSSE>=1 && ReadPrefetchInstr==0);
6629   match(PrefetchRead mem);
6630   ins_cost(100);
6631 
6632   format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for read" %}
6633   ins_encode %{
6634     __ prefetchnta($mem$$Address);
6635   %}
6636   ins_pipe(ialu_mem);
6637 %}
6638 
6639 instruct prefetchrT0( memory mem ) %{
6640   predicate(UseSSE>=1 && ReadPrefetchInstr==1);
6641   match(PrefetchRead mem);
6642   ins_cost(100);
6643 
6644   format %{ "PREFETCHT0 $mem\t! Prefetch into L1 and L2 caches for read" %}
6645   ins_encode %{
6646     __ prefetcht0($mem$$Address);
6647   %}
6648   ins_pipe(ialu_mem);
6649 %}
6650 
6651 instruct prefetchrT2( memory mem ) %{
6652   predicate(UseSSE>=1 && ReadPrefetchInstr==2);
6653   match(PrefetchRead mem);
6654   ins_cost(100);
6655 
6656   format %{ "PREFETCHT2 $mem\t! Prefetch into L2 cache for read" %}
6657   ins_encode %{
6658     __ prefetcht2($mem$$Address);
6659   %}
6660   ins_pipe(ialu_mem);
6661 %}
6662 
6663 instruct prefetchw0( memory mem ) %{
6664   predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
6665   match(PrefetchWrite mem);
6666   ins_cost(0);
6667   size(0);
6668   format %{ "Prefetch (non-SSE is empty encoding)" %}
6669   ins_encode();
6670   ins_pipe(empty);
6671 %}
6672 
6673 instruct prefetchw( memory mem ) %{
6674   predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch());
6675   match( PrefetchWrite mem );
6676   ins_cost(100);
6677 
6678   format %{ "PREFETCHW $mem\t! Prefetch into L1 cache and mark modified" %}
6679   ins_encode %{
6680     __ prefetchw($mem$$Address);
6681   %}
6682   ins_pipe(ialu_mem);
6683 %}
6684 
6685 instruct prefetchwNTA( memory mem ) %{
6686   predicate(UseSSE>=1);
6687   match(PrefetchWrite mem);
6688   ins_cost(100);
6689 
6690   format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for write" %}
6691   ins_encode %{
6692     __ prefetchnta($mem$$Address);
6693   %}
6694   ins_pipe(ialu_mem);
6695 %}
6696 
6697 // Prefetch instructions for allocation.
6698 
6699 instruct prefetchAlloc0( memory mem ) %{
6700   predicate(UseSSE==0 && AllocatePrefetchInstr!=3);
6701   match(PrefetchAllocation mem);
6702   ins_cost(0);
6703   size(0);
6704   format %{ "Prefetch allocation (non-SSE is empty encoding)" %}
6705   ins_encode();
6706   ins_pipe(empty);
6707 %}
6708 
6709 instruct prefetchAlloc( memory mem ) %{
6710   predicate(AllocatePrefetchInstr==3);
6711   match( PrefetchAllocation mem );
6712   ins_cost(100);
6713 
6714   format %{ "PREFETCHW $mem\t! Prefetch allocation into L1 cache and mark modified" %}
6715   ins_encode %{
6716     __ prefetchw($mem$$Address);
6717   %}
6718   ins_pipe(ialu_mem);
6719 %}
6720 
6721 instruct prefetchAllocNTA( memory mem ) %{
6722   predicate(UseSSE>=1 && AllocatePrefetchInstr==0);
6723   match(PrefetchAllocation mem);
6724   ins_cost(100);
6725 
6726   format %{ "PREFETCHNTA $mem\t! Prefetch allocation into non-temporal cache for write" %}
6727   ins_encode %{
6728     __ prefetchnta($mem$$Address);
6729   %}
6730   ins_pipe(ialu_mem);
6731 %}
6732 
6733 instruct prefetchAllocT0( memory mem ) %{
6734   predicate(UseSSE>=1 && AllocatePrefetchInstr==1);
6735   match(PrefetchAllocation mem);
6736   ins_cost(100);
6737 
6738   format %{ "PREFETCHT0 $mem\t! Prefetch allocation into L1 and L2 caches for write" %}
6739   ins_encode %{
6740     __ prefetcht0($mem$$Address);
6741   %}
6742   ins_pipe(ialu_mem);
6743 %}
6744 
6745 instruct prefetchAllocT2( memory mem ) %{
6746   predicate(UseSSE>=1 && AllocatePrefetchInstr==2);
6747   match(PrefetchAllocation mem);
6748   ins_cost(100);
6749 
6750   format %{ "PREFETCHT2 $mem\t! Prefetch allocation into L2 cache for write" %}
6751   ins_encode %{
6752     __ prefetcht2($mem$$Address);
6753   %}
6754   ins_pipe(ialu_mem);
6755 %}
6756 
6757 //----------Store Instructions-------------------------------------------------
6758 
6759 // Store Byte
6760 instruct storeB(memory mem, xRegI src) %{
6761   match(Set mem (StoreB mem src));
6762 
6763   ins_cost(125);
6764   format %{ "MOV8   $mem,$src" %}
6765   opcode(0x88);
6766   ins_encode( OpcP, RegMem( src, mem ) );
6767   ins_pipe( ialu_mem_reg );
6768 %}
6769 
6770 // Store Char/Short
6771 instruct storeC(memory mem, rRegI src) %{
6772   match(Set mem (StoreC mem src));
6773 
6774   ins_cost(125);
6775   format %{ "MOV16  $mem,$src" %}
6776   opcode(0x89, 0x66);
6777   ins_encode( OpcS, OpcP, RegMem( src, mem ) );
6778   ins_pipe( ialu_mem_reg );
6779 %}
6780 
6781 // Store Integer
6782 instruct storeI(memory mem, rRegI src) %{
6783   match(Set mem (StoreI mem src));
6784 
6785   ins_cost(125);
6786   format %{ "MOV    $mem,$src" %}
6787   opcode(0x89);
6788   ins_encode( OpcP, RegMem( src, mem ) );
6789   ins_pipe( ialu_mem_reg );
6790 %}
6791 
6792 // Store Long
6793 instruct storeL(long_memory mem, eRegL src) %{
6794   predicate(!((StoreLNode*)n)->require_atomic_access());
6795   match(Set mem (StoreL mem src));
6796 
6797   ins_cost(200);
6798   format %{ "MOV    $mem,$src.lo\n\t"
6799             "MOV    $mem+4,$src.hi" %}
6800   opcode(0x89, 0x89);
6801   ins_encode( OpcP, RegMem( src, mem ), OpcS, RegMem_Hi( src, mem ) );
6802   ins_pipe( ialu_mem_long_reg );
6803 %}
6804 
6805 // Store Long to Integer
6806 instruct storeL2I(memory mem, eRegL src) %{
6807   match(Set mem (StoreI mem (ConvL2I src)));
6808 
6809   format %{ "MOV    $mem,$src.lo\t# long -> int" %}
6810   ins_encode %{
6811     __ movl($mem$$Address, $src$$Register);
6812   %}
6813   ins_pipe(ialu_mem_reg);
6814 %}
6815 
6816 // Volatile Store Long.  Must be atomic, so move it into
6817 // the FP TOS and then do a 64-bit FIST.  Has to probe the
6818 // target address before the store (for null-ptr checks)
6819 // so the memory operand is used twice in the encoding.
6820 instruct storeL_volatile(memory mem, stackSlotL src, eFlagsReg cr ) %{
6821   predicate(UseSSE<=1 && ((StoreLNode*)n)->require_atomic_access());
6822   match(Set mem (StoreL mem src));
6823   effect( KILL cr );
6824   ins_cost(400);
6825   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6826             "FILD   $src\n\t"
6827             "FISTp  $mem\t # 64-bit atomic volatile long store" %}
6828   opcode(0x3B);
6829   ins_encode( OpcP, RegMem( EAX, mem ), enc_storeL_volatile(mem,src));
6830   ins_pipe( fpu_reg_mem );
6831 %}
6832 
6833 instruct storeLX_volatile(memory mem, stackSlotL src, regD tmp, eFlagsReg cr) %{
6834   predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
6835   match(Set mem (StoreL mem src));
6836   effect( TEMP tmp, KILL cr );
6837   ins_cost(380);
6838   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6839             "MOVSD  $tmp,$src\n\t"
6840             "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
6841   ins_encode %{
6842     __ cmpl(rax, $mem$$Address);
6843     __ movdbl($tmp$$XMMRegister, Address(rsp, $src$$disp));
6844     __ movdbl($mem$$Address, $tmp$$XMMRegister);
6845   %}
6846   ins_pipe( pipe_slow );
6847 %}
6848 
6849 instruct storeLX_reg_volatile(memory mem, eRegL src, regD tmp2, regD tmp, eFlagsReg cr) %{
6850   predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
6851   match(Set mem (StoreL mem src));
6852   effect( TEMP tmp2 , TEMP tmp, KILL cr );
6853   ins_cost(360);
6854   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6855             "MOVD   $tmp,$src.lo\n\t"
6856             "MOVD   $tmp2,$src.hi\n\t"
6857             "PUNPCKLDQ $tmp,$tmp2\n\t"
6858             "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
6859   ins_encode %{
6860     __ cmpl(rax, $mem$$Address);
6861     __ movdl($tmp$$XMMRegister, $src$$Register);
6862     __ movdl($tmp2$$XMMRegister, HIGH_FROM_LOW($src$$Register));
6863     __ punpckldq($tmp$$XMMRegister, $tmp2$$XMMRegister);
6864     __ movdbl($mem$$Address, $tmp$$XMMRegister);
6865   %}
6866   ins_pipe( pipe_slow );
6867 %}
6868 
6869 // Store Pointer; for storing unknown oops and raw pointers
6870 instruct storeP(memory mem, anyRegP src) %{
6871   match(Set mem (StoreP mem src));
6872 
6873   ins_cost(125);
6874   format %{ "MOV    $mem,$src" %}
6875   opcode(0x89);
6876   ins_encode( OpcP, RegMem( src, mem ) );
6877   ins_pipe( ialu_mem_reg );
6878 %}
6879 
6880 // Store Integer Immediate
6881 instruct storeImmI(memory mem, immI src) %{
6882   match(Set mem (StoreI mem src));
6883 
6884   ins_cost(150);
6885   format %{ "MOV    $mem,$src" %}
6886   opcode(0xC7);               /* C7 /0 */
6887   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
6888   ins_pipe( ialu_mem_imm );
6889 %}
6890 
6891 // Store Short/Char Immediate
6892 instruct storeImmI16(memory mem, immI16 src) %{
6893   predicate(UseStoreImmI16);
6894   match(Set mem (StoreC mem src));
6895 
6896   ins_cost(150);
6897   format %{ "MOV16  $mem,$src" %}
6898   opcode(0xC7);     /* C7 /0 Same as 32 store immediate with prefix */
6899   ins_encode( SizePrefix, OpcP, RMopc_Mem(0x00,mem),  Con16( src ));
6900   ins_pipe( ialu_mem_imm );
6901 %}
6902 
6903 // Store Pointer Immediate; null pointers or constant oops that do not
6904 // need card-mark barriers.
6905 instruct storeImmP(memory mem, immP src) %{
6906   match(Set mem (StoreP mem src));
6907 
6908   ins_cost(150);
6909   format %{ "MOV    $mem,$src" %}
6910   opcode(0xC7);               /* C7 /0 */
6911   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
6912   ins_pipe( ialu_mem_imm );
6913 %}
6914 
6915 // Store Byte Immediate
6916 instruct storeImmB(memory mem, immI8 src) %{
6917   match(Set mem (StoreB mem src));
6918 
6919   ins_cost(150);
6920   format %{ "MOV8   $mem,$src" %}
6921   opcode(0xC6);               /* C6 /0 */
6922   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
6923   ins_pipe( ialu_mem_imm );
6924 %}
6925 
6926 // Store CMS card-mark Immediate
6927 instruct storeImmCM(memory mem, immI8 src) %{
6928   match(Set mem (StoreCM mem src));
6929 
6930   ins_cost(150);
6931   format %{ "MOV8   $mem,$src\t! CMS card-mark imm0" %}
6932   opcode(0xC6);               /* C6 /0 */
6933   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
6934   ins_pipe( ialu_mem_imm );
6935 %}
6936 
6937 // Store Double
6938 instruct storeDPR( memory mem, regDPR1 src) %{
6939   predicate(UseSSE<=1);
6940   match(Set mem (StoreD mem src));
6941 
6942   ins_cost(100);
6943   format %{ "FST_D  $mem,$src" %}
6944   opcode(0xDD);       /* DD /2 */
6945   ins_encode( enc_FPR_store(mem,src) );
6946   ins_pipe( fpu_mem_reg );
6947 %}
6948 
6949 // Store double does rounding on x86
6950 instruct storeDPR_rounded( memory mem, regDPR1 src) %{
6951   predicate(UseSSE<=1);
6952   match(Set mem (StoreD mem (RoundDouble src)));
6953 
6954   ins_cost(100);
6955   format %{ "FST_D  $mem,$src\t# round" %}
6956   opcode(0xDD);       /* DD /2 */
6957   ins_encode( enc_FPR_store(mem,src) );
6958   ins_pipe( fpu_mem_reg );
6959 %}
6960 
6961 // Store XMM register to memory (double-precision floating points)
6962 // MOVSD instruction
6963 instruct storeD(memory mem, regD src) %{
6964   predicate(UseSSE>=2);
6965   match(Set mem (StoreD mem src));
6966   ins_cost(95);
6967   format %{ "MOVSD  $mem,$src" %}
6968   ins_encode %{
6969     __ movdbl($mem$$Address, $src$$XMMRegister);
6970   %}
6971   ins_pipe( pipe_slow );
6972 %}
6973 
6974 // Store XMM register to memory (single-precision floating point)
6975 // MOVSS instruction
6976 instruct storeF(memory mem, regF src) %{
6977   predicate(UseSSE>=1);
6978   match(Set mem (StoreF mem src));
6979   ins_cost(95);
6980   format %{ "MOVSS  $mem,$src" %}
6981   ins_encode %{
6982     __ movflt($mem$$Address, $src$$XMMRegister);
6983   %}
6984   ins_pipe( pipe_slow );
6985 %}
6986 
6987 // Store Float
6988 instruct storeFPR( memory mem, regFPR1 src) %{
6989   predicate(UseSSE==0);
6990   match(Set mem (StoreF mem src));
6991 
6992   ins_cost(100);
6993   format %{ "FST_S  $mem,$src" %}
6994   opcode(0xD9);       /* D9 /2 */
6995   ins_encode( enc_FPR_store(mem,src) );
6996   ins_pipe( fpu_mem_reg );
6997 %}
6998 
6999 // Store Float does rounding on x86
7000 instruct storeFPR_rounded( memory mem, regFPR1 src) %{
7001   predicate(UseSSE==0);
7002   match(Set mem (StoreF mem (RoundFloat src)));
7003 
7004   ins_cost(100);
7005   format %{ "FST_S  $mem,$src\t# round" %}
7006   opcode(0xD9);       /* D9 /2 */
7007   ins_encode( enc_FPR_store(mem,src) );
7008   ins_pipe( fpu_mem_reg );
7009 %}
7010 
7011 // Store Float does rounding on x86
7012 instruct storeFPR_Drounded( memory mem, regDPR1 src) %{
7013   predicate(UseSSE<=1);
7014   match(Set mem (StoreF mem (ConvD2F src)));
7015 
7016   ins_cost(100);
7017   format %{ "FST_S  $mem,$src\t# D-round" %}
7018   opcode(0xD9);       /* D9 /2 */
7019   ins_encode( enc_FPR_store(mem,src) );
7020   ins_pipe( fpu_mem_reg );
7021 %}
7022 
7023 // Store immediate Float value (it is faster than store from FPU register)
7024 // The instruction usage is guarded by predicate in operand immFPR().
7025 instruct storeFPR_imm( memory mem, immFPR src) %{
7026   match(Set mem (StoreF mem src));
7027 
7028   ins_cost(50);
7029   format %{ "MOV    $mem,$src\t# store float" %}
7030   opcode(0xC7);               /* C7 /0 */
7031   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32FPR_as_bits( src ));
7032   ins_pipe( ialu_mem_imm );
7033 %}
7034 
7035 // Store immediate Float value (it is faster than store from XMM register)
7036 // The instruction usage is guarded by predicate in operand immF().
7037 instruct storeF_imm( memory mem, immF src) %{
7038   match(Set mem (StoreF mem src));
7039 
7040   ins_cost(50);
7041   format %{ "MOV    $mem,$src\t# store float" %}
7042   opcode(0xC7);               /* C7 /0 */
7043   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32F_as_bits( src ));
7044   ins_pipe( ialu_mem_imm );
7045 %}
7046 
7047 // Store Integer to stack slot
7048 instruct storeSSI(stackSlotI dst, rRegI src) %{
7049   match(Set dst src);
7050 
7051   ins_cost(100);
7052   format %{ "MOV    $dst,$src" %}
7053   opcode(0x89);
7054   ins_encode( OpcPRegSS( dst, src ) );
7055   ins_pipe( ialu_mem_reg );
7056 %}
7057 
7058 // Store Integer to stack slot
7059 instruct storeSSP(stackSlotP dst, eRegP src) %{
7060   match(Set dst src);
7061 
7062   ins_cost(100);
7063   format %{ "MOV    $dst,$src" %}
7064   opcode(0x89);
7065   ins_encode( OpcPRegSS( dst, src ) );
7066   ins_pipe( ialu_mem_reg );
7067 %}
7068 
7069 // Store Long to stack slot
7070 instruct storeSSL(stackSlotL dst, eRegL src) %{
7071   match(Set dst src);
7072 
7073   ins_cost(200);
7074   format %{ "MOV    $dst,$src.lo\n\t"
7075             "MOV    $dst+4,$src.hi" %}
7076   opcode(0x89, 0x89);
7077   ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
7078   ins_pipe( ialu_mem_long_reg );
7079 %}
7080 
7081 //----------MemBar Instructions-----------------------------------------------
7082 // Memory barrier flavors
7083 
7084 instruct membar_acquire() %{
7085   match(MemBarAcquire);
7086   ins_cost(400);
7087 
7088   size(0);
7089   format %{ "MEMBAR-acquire ! (empty encoding)" %}
7090   ins_encode();
7091   ins_pipe(empty);
7092 %}
7093 
7094 instruct membar_acquire_lock() %{
7095   match(MemBarAcquireLock);
7096   ins_cost(0);
7097 
7098   size(0);
7099   format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7100   ins_encode( );
7101   ins_pipe(empty);
7102 %}
7103 
7104 instruct membar_release() %{
7105   match(MemBarRelease);
7106   ins_cost(400);
7107 
7108   size(0);
7109   format %{ "MEMBAR-release ! (empty encoding)" %}
7110   ins_encode( );
7111   ins_pipe(empty);
7112 %}
7113 
7114 instruct membar_release_lock() %{
7115   match(MemBarReleaseLock);
7116   ins_cost(0);
7117 
7118   size(0);
7119   format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7120   ins_encode( );
7121   ins_pipe(empty);
7122 %}
7123 
7124 instruct membar_volatile(eFlagsReg cr) %{
7125   match(MemBarVolatile);
7126   effect(KILL cr);
7127   ins_cost(400);
7128 
7129   format %{ 
7130     $$template
7131     if (os::is_MP()) {
7132       $$emit$$"LOCK ADDL [ESP + #0], 0\t! membar_volatile"
7133     } else {
7134       $$emit$$"MEMBAR-volatile ! (empty encoding)"
7135     }
7136   %}
7137   ins_encode %{
7138     __ membar(Assembler::StoreLoad);
7139   %}
7140   ins_pipe(pipe_slow);
7141 %}
7142 
7143 instruct unnecessary_membar_volatile() %{
7144   match(MemBarVolatile);
7145   predicate(Matcher::post_store_load_barrier(n));
7146   ins_cost(0);
7147 
7148   size(0);
7149   format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7150   ins_encode( );
7151   ins_pipe(empty);
7152 %}
7153 
7154 instruct membar_storestore() %{
7155   match(MemBarStoreStore);
7156   ins_cost(0);
7157 
7158   size(0);
7159   format %{ "MEMBAR-storestore (empty encoding)" %}
7160   ins_encode( );
7161   ins_pipe(empty);
7162 %}
7163 
7164 //----------Move Instructions--------------------------------------------------
7165 instruct castX2P(eAXRegP dst, eAXRegI src) %{
7166   match(Set dst (CastX2P src));
7167   format %{ "# X2P  $dst, $src" %}
7168   ins_encode( /*empty encoding*/ );
7169   ins_cost(0);
7170   ins_pipe(empty);
7171 %}
7172 
7173 instruct castP2X(rRegI dst, eRegP src ) %{
7174   match(Set dst (CastP2X src));
7175   ins_cost(50);
7176   format %{ "MOV    $dst, $src\t# CastP2X" %}
7177   ins_encode( enc_Copy( dst, src) );
7178   ins_pipe( ialu_reg_reg );
7179 %}
7180 
7181 //----------Conditional Move---------------------------------------------------
7182 // Conditional move
7183 instruct jmovI_reg(cmpOp cop, eFlagsReg cr, rRegI dst, rRegI src) %{
7184   predicate(!VM_Version::supports_cmov() );
7185   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7186   ins_cost(200);
7187   format %{ "J$cop,us skip\t# signed cmove\n\t"
7188             "MOV    $dst,$src\n"
7189       "skip:" %}
7190   ins_encode %{
7191     Label Lskip;
7192     // Invert sense of branch from sense of CMOV
7193     __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
7194     __ movl($dst$$Register, $src$$Register);
7195     __ bind(Lskip);
7196   %}
7197   ins_pipe( pipe_cmov_reg );
7198 %}
7199 
7200 instruct jmovI_regU(cmpOpU cop, eFlagsRegU cr, rRegI dst, rRegI src) %{
7201   predicate(!VM_Version::supports_cmov() );
7202   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7203   ins_cost(200);
7204   format %{ "J$cop,us skip\t# unsigned cmove\n\t"
7205             "MOV    $dst,$src\n"
7206       "skip:" %}
7207   ins_encode %{
7208     Label Lskip;
7209     // Invert sense of branch from sense of CMOV
7210     __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
7211     __ movl($dst$$Register, $src$$Register);
7212     __ bind(Lskip);
7213   %}
7214   ins_pipe( pipe_cmov_reg );
7215 %}
7216 
7217 instruct cmovI_reg(rRegI dst, rRegI src, eFlagsReg cr, cmpOp cop ) %{
7218   predicate(VM_Version::supports_cmov() );
7219   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7220   ins_cost(200);
7221   format %{ "CMOV$cop $dst,$src" %}
7222   opcode(0x0F,0x40);
7223   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7224   ins_pipe( pipe_cmov_reg );
7225 %}
7226 
7227 instruct cmovI_regU( cmpOpU cop, eFlagsRegU cr, rRegI dst, rRegI src ) %{
7228   predicate(VM_Version::supports_cmov() );
7229   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7230   ins_cost(200);
7231   format %{ "CMOV$cop $dst,$src" %}
7232   opcode(0x0F,0x40);
7233   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7234   ins_pipe( pipe_cmov_reg );
7235 %}
7236 
7237 instruct cmovI_regUCF( cmpOpUCF cop, eFlagsRegUCF cr, rRegI dst, rRegI src ) %{
7238   predicate(VM_Version::supports_cmov() );
7239   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7240   ins_cost(200);
7241   expand %{
7242     cmovI_regU(cop, cr, dst, src);
7243   %}
7244 %}
7245 
7246 // Conditional move
7247 instruct cmovI_mem(cmpOp cop, eFlagsReg cr, rRegI dst, memory src) %{
7248   predicate(VM_Version::supports_cmov() );
7249   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7250   ins_cost(250);
7251   format %{ "CMOV$cop $dst,$src" %}
7252   opcode(0x0F,0x40);
7253   ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7254   ins_pipe( pipe_cmov_mem );
7255 %}
7256 
7257 // Conditional move
7258 instruct cmovI_memU(cmpOpU cop, eFlagsRegU cr, rRegI dst, memory src) %{
7259   predicate(VM_Version::supports_cmov() );
7260   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7261   ins_cost(250);
7262   format %{ "CMOV$cop $dst,$src" %}
7263   opcode(0x0F,0x40);
7264   ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7265   ins_pipe( pipe_cmov_mem );
7266 %}
7267 
7268 instruct cmovI_memUCF(cmpOpUCF cop, eFlagsRegUCF cr, rRegI dst, memory src) %{
7269   predicate(VM_Version::supports_cmov() );
7270   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7271   ins_cost(250);
7272   expand %{
7273     cmovI_memU(cop, cr, dst, src);
7274   %}
7275 %}
7276 
7277 // Conditional move
7278 instruct cmovP_reg(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
7279   predicate(VM_Version::supports_cmov() );
7280   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7281   ins_cost(200);
7282   format %{ "CMOV$cop $dst,$src\t# ptr" %}
7283   opcode(0x0F,0x40);
7284   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7285   ins_pipe( pipe_cmov_reg );
7286 %}
7287 
7288 // Conditional move (non-P6 version)
7289 // Note:  a CMoveP is generated for  stubs and native wrappers
7290 //        regardless of whether we are on a P6, so we
7291 //        emulate a cmov here
7292 instruct cmovP_reg_nonP6(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
7293   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7294   ins_cost(300);
7295   format %{ "Jn$cop   skip\n\t"
7296           "MOV    $dst,$src\t# pointer\n"
7297       "skip:" %}
7298   opcode(0x8b);
7299   ins_encode( enc_cmov_branch(cop, 0x2), OpcP, RegReg(dst, src));
7300   ins_pipe( pipe_cmov_reg );
7301 %}
7302 
7303 // Conditional move
7304 instruct cmovP_regU(cmpOpU cop, eFlagsRegU cr, eRegP dst, eRegP src ) %{
7305   predicate(VM_Version::supports_cmov() );
7306   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7307   ins_cost(200);
7308   format %{ "CMOV$cop $dst,$src\t# ptr" %}
7309   opcode(0x0F,0x40);
7310   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7311   ins_pipe( pipe_cmov_reg );
7312 %}
7313 
7314 instruct cmovP_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegP dst, eRegP src ) %{
7315   predicate(VM_Version::supports_cmov() );
7316   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7317   ins_cost(200);
7318   expand %{
7319     cmovP_regU(cop, cr, dst, src);
7320   %}
7321 %}
7322 
7323 // DISABLED: Requires the ADLC to emit a bottom_type call that
7324 // correctly meets the two pointer arguments; one is an incoming
7325 // register but the other is a memory operand.  ALSO appears to
7326 // be buggy with implicit null checks.
7327 //
7328 //// Conditional move
7329 //instruct cmovP_mem(cmpOp cop, eFlagsReg cr, eRegP dst, memory src) %{
7330 //  predicate(VM_Version::supports_cmov() );
7331 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7332 //  ins_cost(250);
7333 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7334 //  opcode(0x0F,0x40);
7335 //  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7336 //  ins_pipe( pipe_cmov_mem );
7337 //%}
7338 //
7339 //// Conditional move
7340 //instruct cmovP_memU(cmpOpU cop, eFlagsRegU cr, eRegP dst, memory src) %{
7341 //  predicate(VM_Version::supports_cmov() );
7342 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7343 //  ins_cost(250);
7344 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7345 //  opcode(0x0F,0x40);
7346 //  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7347 //  ins_pipe( pipe_cmov_mem );
7348 //%}
7349 
7350 // Conditional move
7351 instruct fcmovDPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regDPR1 dst, regDPR src) %{
7352   predicate(UseSSE<=1);
7353   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7354   ins_cost(200);
7355   format %{ "FCMOV$cop $dst,$src\t# double" %}
7356   opcode(0xDA);
7357   ins_encode( enc_cmov_dpr(cop,src) );
7358   ins_pipe( pipe_cmovDPR_reg );
7359 %}
7360 
7361 // Conditional move
7362 instruct fcmovFPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regFPR1 dst, regFPR src) %{
7363   predicate(UseSSE==0);
7364   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7365   ins_cost(200);
7366   format %{ "FCMOV$cop $dst,$src\t# float" %}
7367   opcode(0xDA);
7368   ins_encode( enc_cmov_dpr(cop,src) );
7369   ins_pipe( pipe_cmovDPR_reg );
7370 %}
7371 
7372 // Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
7373 instruct fcmovDPR_regS(cmpOp cop, eFlagsReg cr, regDPR dst, regDPR src) %{
7374   predicate(UseSSE<=1);
7375   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7376   ins_cost(200);
7377   format %{ "Jn$cop   skip\n\t"
7378             "MOV    $dst,$src\t# double\n"
7379       "skip:" %}
7380   opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
7381   ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_DPR(src), OpcP, RegOpc(dst) );
7382   ins_pipe( pipe_cmovDPR_reg );
7383 %}
7384 
7385 // Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
7386 instruct fcmovFPR_regS(cmpOp cop, eFlagsReg cr, regFPR dst, regFPR src) %{
7387   predicate(UseSSE==0);
7388   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7389   ins_cost(200);
7390   format %{ "Jn$cop    skip\n\t"
7391             "MOV    $dst,$src\t# float\n"
7392       "skip:" %}
7393   opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
7394   ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_FPR(src), OpcP, RegOpc(dst) );
7395   ins_pipe( pipe_cmovDPR_reg );
7396 %}
7397 
7398 // No CMOVE with SSE/SSE2
7399 instruct fcmovF_regS(cmpOp cop, eFlagsReg cr, regF dst, regF src) %{
7400   predicate (UseSSE>=1);
7401   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7402   ins_cost(200);
7403   format %{ "Jn$cop   skip\n\t"
7404             "MOVSS  $dst,$src\t# float\n"
7405       "skip:" %}
7406   ins_encode %{
7407     Label skip;
7408     // Invert sense of branch from sense of CMOV
7409     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7410     __ movflt($dst$$XMMRegister, $src$$XMMRegister);
7411     __ bind(skip);
7412   %}
7413   ins_pipe( pipe_slow );
7414 %}
7415 
7416 // No CMOVE with SSE/SSE2
7417 instruct fcmovD_regS(cmpOp cop, eFlagsReg cr, regD dst, regD src) %{
7418   predicate (UseSSE>=2);
7419   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7420   ins_cost(200);
7421   format %{ "Jn$cop   skip\n\t"
7422             "MOVSD  $dst,$src\t# float\n"
7423       "skip:" %}
7424   ins_encode %{
7425     Label skip;
7426     // Invert sense of branch from sense of CMOV
7427     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7428     __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
7429     __ bind(skip);
7430   %}
7431   ins_pipe( pipe_slow );
7432 %}
7433 
7434 // unsigned version
7435 instruct fcmovF_regU(cmpOpU cop, eFlagsRegU cr, regF dst, regF src) %{
7436   predicate (UseSSE>=1);
7437   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7438   ins_cost(200);
7439   format %{ "Jn$cop   skip\n\t"
7440             "MOVSS  $dst,$src\t# float\n"
7441       "skip:" %}
7442   ins_encode %{
7443     Label skip;
7444     // Invert sense of branch from sense of CMOV
7445     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7446     __ movflt($dst$$XMMRegister, $src$$XMMRegister);
7447     __ bind(skip);
7448   %}
7449   ins_pipe( pipe_slow );
7450 %}
7451 
7452 instruct fcmovF_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regF dst, regF src) %{
7453   predicate (UseSSE>=1);
7454   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7455   ins_cost(200);
7456   expand %{
7457     fcmovF_regU(cop, cr, dst, src);
7458   %}
7459 %}
7460 
7461 // unsigned version
7462 instruct fcmovD_regU(cmpOpU cop, eFlagsRegU cr, regD dst, regD src) %{
7463   predicate (UseSSE>=2);
7464   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7465   ins_cost(200);
7466   format %{ "Jn$cop   skip\n\t"
7467             "MOVSD  $dst,$src\t# float\n"
7468       "skip:" %}
7469   ins_encode %{
7470     Label skip;
7471     // Invert sense of branch from sense of CMOV
7472     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7473     __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
7474     __ bind(skip);
7475   %}
7476   ins_pipe( pipe_slow );
7477 %}
7478 
7479 instruct fcmovD_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regD dst, regD src) %{
7480   predicate (UseSSE>=2);
7481   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7482   ins_cost(200);
7483   expand %{
7484     fcmovD_regU(cop, cr, dst, src);
7485   %}
7486 %}
7487 
7488 instruct cmovL_reg(cmpOp cop, eFlagsReg cr, eRegL dst, eRegL src) %{
7489   predicate(VM_Version::supports_cmov() );
7490   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7491   ins_cost(200);
7492   format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
7493             "CMOV$cop $dst.hi,$src.hi" %}
7494   opcode(0x0F,0x40);
7495   ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
7496   ins_pipe( pipe_cmov_reg_long );
7497 %}
7498 
7499 instruct cmovL_regU(cmpOpU cop, eFlagsRegU cr, eRegL dst, eRegL src) %{
7500   predicate(VM_Version::supports_cmov() );
7501   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7502   ins_cost(200);
7503   format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
7504             "CMOV$cop $dst.hi,$src.hi" %}
7505   opcode(0x0F,0x40);
7506   ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
7507   ins_pipe( pipe_cmov_reg_long );
7508 %}
7509 
7510 instruct cmovL_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegL dst, eRegL src) %{
7511   predicate(VM_Version::supports_cmov() );
7512   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7513   ins_cost(200);
7514   expand %{
7515     cmovL_regU(cop, cr, dst, src);
7516   %}
7517 %}
7518 
7519 //----------Arithmetic Instructions--------------------------------------------
7520 //----------Addition Instructions----------------------------------------------
7521 
7522 instruct addExactI_rReg(eAXRegI dst, rRegI src, eFlagsReg cr)
7523 %{
7524   match(AddExactI dst src);
7525   effect(DEF cr);
7526 
7527   format %{ "ADD    $dst, $src\t# addExact int" %}
7528   ins_encode %{
7529     __ addl($dst$$Register, $src$$Register);
7530   %}
7531   ins_pipe(ialu_reg_reg);
7532 %}
7533 
7534 instruct addExactI_rReg_imm(eAXRegI dst, immI src, eFlagsReg cr)
7535 %{
7536   match(AddExactI dst src);
7537   effect(DEF cr);
7538 
7539   format %{ "ADD    $dst, $src\t# addExact int" %}
7540   ins_encode %{
7541     __ addl($dst$$Register, $src$$constant);
7542   %}
7543   ins_pipe(ialu_reg_reg);
7544 %}
7545 
7546 // Integer Addition Instructions
7547 instruct addI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
7548   match(Set dst (AddI dst src));
7549   effect(KILL cr);
7550 
7551   size(2);
7552   format %{ "ADD    $dst,$src" %}
7553   opcode(0x03);
7554   ins_encode( OpcP, RegReg( dst, src) );
7555   ins_pipe( ialu_reg_reg );
7556 %}
7557 
7558 instruct addI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
7559   match(Set dst (AddI dst src));
7560   effect(KILL cr);
7561 
7562   format %{ "ADD    $dst,$src" %}
7563   opcode(0x81, 0x00); /* /0 id */
7564   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7565   ins_pipe( ialu_reg );
7566 %}
7567 
7568 instruct incI_eReg(rRegI dst, immI1 src, eFlagsReg cr) %{
7569   predicate(UseIncDec);
7570   match(Set dst (AddI dst src));
7571   effect(KILL cr);
7572 
7573   size(1);
7574   format %{ "INC    $dst" %}
7575   opcode(0x40); /*  */
7576   ins_encode( Opc_plus( primary, dst ) );
7577   ins_pipe( ialu_reg );
7578 %}
7579 
7580 instruct leaI_eReg_immI(rRegI dst, rRegI src0, immI src1) %{
7581   match(Set dst (AddI src0 src1));
7582   ins_cost(110);
7583 
7584   format %{ "LEA    $dst,[$src0 + $src1]" %}
7585   opcode(0x8D); /* 0x8D /r */
7586   ins_encode( OpcP, RegLea( dst, src0, src1 ) );
7587   ins_pipe( ialu_reg_reg );
7588 %}
7589 
7590 instruct leaP_eReg_immI(eRegP dst, eRegP src0, immI src1) %{
7591   match(Set dst (AddP src0 src1));
7592   ins_cost(110);
7593 
7594   format %{ "LEA    $dst,[$src0 + $src1]\t# ptr" %}
7595   opcode(0x8D); /* 0x8D /r */
7596   ins_encode( OpcP, RegLea( dst, src0, src1 ) );
7597   ins_pipe( ialu_reg_reg );
7598 %}
7599 
7600 instruct decI_eReg(rRegI dst, immI_M1 src, eFlagsReg cr) %{
7601   predicate(UseIncDec);
7602   match(Set dst (AddI dst src));
7603   effect(KILL cr);
7604 
7605   size(1);
7606   format %{ "DEC    $dst" %}
7607   opcode(0x48); /*  */
7608   ins_encode( Opc_plus( primary, dst ) );
7609   ins_pipe( ialu_reg );
7610 %}
7611 
7612 instruct addP_eReg(eRegP dst, rRegI src, eFlagsReg cr) %{
7613   match(Set dst (AddP dst src));
7614   effect(KILL cr);
7615 
7616   size(2);
7617   format %{ "ADD    $dst,$src" %}
7618   opcode(0x03);
7619   ins_encode( OpcP, RegReg( dst, src) );
7620   ins_pipe( ialu_reg_reg );
7621 %}
7622 
7623 instruct addP_eReg_imm(eRegP dst, immI src, eFlagsReg cr) %{
7624   match(Set dst (AddP dst src));
7625   effect(KILL cr);
7626 
7627   format %{ "ADD    $dst,$src" %}
7628   opcode(0x81,0x00); /* Opcode 81 /0 id */
7629   // ins_encode( RegImm( dst, src) );
7630   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7631   ins_pipe( ialu_reg );
7632 %}
7633 
7634 instruct addI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
7635   match(Set dst (AddI dst (LoadI src)));
7636   effect(KILL cr);
7637 
7638   ins_cost(125);
7639   format %{ "ADD    $dst,$src" %}
7640   opcode(0x03);
7641   ins_encode( OpcP, RegMem( dst, src) );
7642   ins_pipe( ialu_reg_mem );
7643 %}
7644 
7645 instruct addI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
7646   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7647   effect(KILL cr);
7648 
7649   ins_cost(150);
7650   format %{ "ADD    $dst,$src" %}
7651   opcode(0x01);  /* Opcode 01 /r */
7652   ins_encode( OpcP, RegMem( src, dst ) );
7653   ins_pipe( ialu_mem_reg );
7654 %}
7655 
7656 // Add Memory with Immediate
7657 instruct addI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
7658   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7659   effect(KILL cr);
7660 
7661   ins_cost(125);
7662   format %{ "ADD    $dst,$src" %}
7663   opcode(0x81);               /* Opcode 81 /0 id */
7664   ins_encode( OpcSE( src ), RMopc_Mem(0x00,dst), Con8or32( src ) );
7665   ins_pipe( ialu_mem_imm );
7666 %}
7667 
7668 instruct incI_mem(memory dst, immI1 src, eFlagsReg cr) %{
7669   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7670   effect(KILL cr);
7671 
7672   ins_cost(125);
7673   format %{ "INC    $dst" %}
7674   opcode(0xFF);               /* Opcode FF /0 */
7675   ins_encode( OpcP, RMopc_Mem(0x00,dst));
7676   ins_pipe( ialu_mem_imm );
7677 %}
7678 
7679 instruct decI_mem(memory dst, immI_M1 src, eFlagsReg cr) %{
7680   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7681   effect(KILL cr);
7682 
7683   ins_cost(125);
7684   format %{ "DEC    $dst" %}
7685   opcode(0xFF);               /* Opcode FF /1 */
7686   ins_encode( OpcP, RMopc_Mem(0x01,dst));
7687   ins_pipe( ialu_mem_imm );
7688 %}
7689 
7690 
7691 instruct checkCastPP( eRegP dst ) %{
7692   match(Set dst (CheckCastPP dst));
7693 
7694   size(0);
7695   format %{ "#checkcastPP of $dst" %}
7696   ins_encode( /*empty encoding*/ );
7697   ins_pipe( empty );
7698 %}
7699 
7700 instruct castPP( eRegP dst ) %{
7701   match(Set dst (CastPP dst));
7702   format %{ "#castPP of $dst" %}
7703   ins_encode( /*empty encoding*/ );
7704   ins_pipe( empty );
7705 %}
7706 
7707 instruct castII( rRegI dst ) %{
7708   match(Set dst (CastII dst));
7709   format %{ "#castII of $dst" %}
7710   ins_encode( /*empty encoding*/ );
7711   ins_cost(0);
7712   ins_pipe( empty );
7713 %}
7714 
7715 
7716 // Load-locked - same as a regular pointer load when used with compare-swap
7717 instruct loadPLocked(eRegP dst, memory mem) %{
7718   match(Set dst (LoadPLocked mem));
7719 
7720   ins_cost(125);
7721   format %{ "MOV    $dst,$mem\t# Load ptr. locked" %}
7722   opcode(0x8B);
7723   ins_encode( OpcP, RegMem(dst,mem));
7724   ins_pipe( ialu_reg_mem );
7725 %}
7726 
7727 // Conditional-store of the updated heap-top.
7728 // Used during allocation of the shared heap.
7729 // Sets flags (EQ) on success.  Implemented with a CMPXCHG on Intel.
7730 instruct storePConditional( memory heap_top_ptr, eAXRegP oldval, eRegP newval, eFlagsReg cr ) %{
7731   match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
7732   // EAX is killed if there is contention, but then it's also unused.
7733   // In the common case of no contention, EAX holds the new oop address.
7734   format %{ "CMPXCHG $heap_top_ptr,$newval\t# If EAX==$heap_top_ptr Then store $newval into $heap_top_ptr" %}
7735   ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval,heap_top_ptr) );
7736   ins_pipe( pipe_cmpxchg );
7737 %}
7738 
7739 // Conditional-store of an int value.
7740 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG on Intel.
7741 instruct storeIConditional( memory mem, eAXRegI oldval, rRegI newval, eFlagsReg cr ) %{
7742   match(Set cr (StoreIConditional mem (Binary oldval newval)));
7743   effect(KILL oldval);
7744   format %{ "CMPXCHG $mem,$newval\t# If EAX==$mem Then store $newval into $mem" %}
7745   ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval, mem) );
7746   ins_pipe( pipe_cmpxchg );
7747 %}
7748 
7749 // Conditional-store of a long value.
7750 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG8 on Intel.
7751 instruct storeLConditional( memory mem, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
7752   match(Set cr (StoreLConditional mem (Binary oldval newval)));
7753   effect(KILL oldval);
7754   format %{ "XCHG   EBX,ECX\t# correct order for CMPXCHG8 instruction\n\t"
7755             "CMPXCHG8 $mem,ECX:EBX\t# If EDX:EAX==$mem Then store ECX:EBX into $mem\n\t"
7756             "XCHG   EBX,ECX"
7757   %}
7758   ins_encode %{
7759     // Note: we need to swap rbx, and rcx before and after the
7760     //       cmpxchg8 instruction because the instruction uses
7761     //       rcx as the high order word of the new value to store but
7762     //       our register encoding uses rbx.
7763     __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
7764     if( os::is_MP() )
7765       __ lock();
7766     __ cmpxchg8($mem$$Address);
7767     __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
7768   %}
7769   ins_pipe( pipe_cmpxchg );
7770 %}
7771 
7772 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7773 
7774 instruct compareAndSwapL( rRegI res, eSIRegP mem_ptr, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
7775   predicate(VM_Version::supports_cx8());
7776   match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7777   effect(KILL cr, KILL oldval);
7778   format %{ "CMPXCHG8 [$mem_ptr],$newval\t# If EDX:EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7779             "MOV    $res,0\n\t"
7780             "JNE,s  fail\n\t"
7781             "MOV    $res,1\n"
7782           "fail:" %}
7783   ins_encode( enc_cmpxchg8(mem_ptr),
7784               enc_flags_ne_to_boolean(res) );
7785   ins_pipe( pipe_cmpxchg );
7786 %}
7787 
7788 instruct compareAndSwapP( rRegI res,  pRegP mem_ptr, eAXRegP oldval, eCXRegP newval, eFlagsReg cr) %{
7789   match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7790   effect(KILL cr, KILL oldval);
7791   format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7792             "MOV    $res,0\n\t"
7793             "JNE,s  fail\n\t"
7794             "MOV    $res,1\n"
7795           "fail:" %}
7796   ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
7797   ins_pipe( pipe_cmpxchg );
7798 %}
7799 
7800 instruct compareAndSwapI( rRegI res, pRegP mem_ptr, eAXRegI oldval, eCXRegI newval, eFlagsReg cr) %{
7801   match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7802   effect(KILL cr, KILL oldval);
7803   format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7804             "MOV    $res,0\n\t"
7805             "JNE,s  fail\n\t"
7806             "MOV    $res,1\n"
7807           "fail:" %}
7808   ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
7809   ins_pipe( pipe_cmpxchg );
7810 %}
7811 
7812 instruct xaddI_no_res( memory mem, Universe dummy, immI add, eFlagsReg cr) %{
7813   predicate(n->as_LoadStore()->result_not_used());
7814   match(Set dummy (GetAndAddI mem add));
7815   effect(KILL cr);
7816   format %{ "ADDL  [$mem],$add" %}
7817   ins_encode %{
7818     if (os::is_MP()) { __ lock(); }
7819     __ addl($mem$$Address, $add$$constant);
7820   %}
7821   ins_pipe( pipe_cmpxchg );
7822 %}
7823 
7824 instruct xaddI( memory mem, rRegI newval, eFlagsReg cr) %{
7825   match(Set newval (GetAndAddI mem newval));
7826   effect(KILL cr);
7827   format %{ "XADDL  [$mem],$newval" %}
7828   ins_encode %{
7829     if (os::is_MP()) { __ lock(); }
7830     __ xaddl($mem$$Address, $newval$$Register);
7831   %}
7832   ins_pipe( pipe_cmpxchg );
7833 %}
7834 
7835 instruct xchgI( memory mem, rRegI newval) %{
7836   match(Set newval (GetAndSetI mem newval));
7837   format %{ "XCHGL  $newval,[$mem]" %}
7838   ins_encode %{
7839     __ xchgl($newval$$Register, $mem$$Address);
7840   %}
7841   ins_pipe( pipe_cmpxchg );
7842 %}
7843 
7844 instruct xchgP( memory mem, pRegP newval) %{
7845   match(Set newval (GetAndSetP mem newval));
7846   format %{ "XCHGL  $newval,[$mem]" %}
7847   ins_encode %{
7848     __ xchgl($newval$$Register, $mem$$Address);
7849   %}
7850   ins_pipe( pipe_cmpxchg );
7851 %}
7852 
7853 //----------Subtraction Instructions-------------------------------------------
7854 // Integer Subtraction Instructions
7855 instruct subI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
7856   match(Set dst (SubI dst src));
7857   effect(KILL cr);
7858 
7859   size(2);
7860   format %{ "SUB    $dst,$src" %}
7861   opcode(0x2B);
7862   ins_encode( OpcP, RegReg( dst, src) );
7863   ins_pipe( ialu_reg_reg );
7864 %}
7865 
7866 instruct subI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
7867   match(Set dst (SubI dst src));
7868   effect(KILL cr);
7869 
7870   format %{ "SUB    $dst,$src" %}
7871   opcode(0x81,0x05);  /* Opcode 81 /5 */
7872   // ins_encode( RegImm( dst, src) );
7873   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7874   ins_pipe( ialu_reg );
7875 %}
7876 
7877 instruct subI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
7878   match(Set dst (SubI dst (LoadI src)));
7879   effect(KILL cr);
7880 
7881   ins_cost(125);
7882   format %{ "SUB    $dst,$src" %}
7883   opcode(0x2B);
7884   ins_encode( OpcP, RegMem( dst, src) );
7885   ins_pipe( ialu_reg_mem );
7886 %}
7887 
7888 instruct subI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
7889   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
7890   effect(KILL cr);
7891 
7892   ins_cost(150);
7893   format %{ "SUB    $dst,$src" %}
7894   opcode(0x29);  /* Opcode 29 /r */
7895   ins_encode( OpcP, RegMem( src, dst ) );
7896   ins_pipe( ialu_mem_reg );
7897 %}
7898 
7899 // Subtract from a pointer
7900 instruct subP_eReg(eRegP dst, rRegI src, immI0 zero, eFlagsReg cr) %{
7901   match(Set dst (AddP dst (SubI zero src)));
7902   effect(KILL cr);
7903 
7904   size(2);
7905   format %{ "SUB    $dst,$src" %}
7906   opcode(0x2B);
7907   ins_encode( OpcP, RegReg( dst, src) );
7908   ins_pipe( ialu_reg_reg );
7909 %}
7910 
7911 instruct negI_eReg(rRegI dst, immI0 zero, eFlagsReg cr) %{
7912   match(Set dst (SubI zero dst));
7913   effect(KILL cr);
7914 
7915   size(2);
7916   format %{ "NEG    $dst" %}
7917   opcode(0xF7,0x03);  // Opcode F7 /3
7918   ins_encode( OpcP, RegOpc( dst ) );
7919   ins_pipe( ialu_reg );
7920 %}
7921 
7922 
7923 //----------Multiplication/Division Instructions-------------------------------
7924 // Integer Multiplication Instructions
7925 // Multiply Register
7926 instruct mulI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
7927   match(Set dst (MulI dst src));
7928   effect(KILL cr);
7929 
7930   size(3);
7931   ins_cost(300);
7932   format %{ "IMUL   $dst,$src" %}
7933   opcode(0xAF, 0x0F);
7934   ins_encode( OpcS, OpcP, RegReg( dst, src) );
7935   ins_pipe( ialu_reg_reg_alu0 );
7936 %}
7937 
7938 // Multiply 32-bit Immediate
7939 instruct mulI_eReg_imm(rRegI dst, rRegI src, immI imm, eFlagsReg cr) %{
7940   match(Set dst (MulI src imm));
7941   effect(KILL cr);
7942 
7943   ins_cost(300);
7944   format %{ "IMUL   $dst,$src,$imm" %}
7945   opcode(0x69);  /* 69 /r id */
7946   ins_encode( OpcSE(imm), RegReg( dst, src ), Con8or32( imm ) );
7947   ins_pipe( ialu_reg_reg_alu0 );
7948 %}
7949 
7950 instruct loadConL_low_only(eADXRegL_low_only dst, immL32 src, eFlagsReg cr) %{
7951   match(Set dst src);
7952   effect(KILL cr);
7953 
7954   // Note that this is artificially increased to make it more expensive than loadConL
7955   ins_cost(250);
7956   format %{ "MOV    EAX,$src\t// low word only" %}
7957   opcode(0xB8);
7958   ins_encode( LdImmL_Lo(dst, src) );
7959   ins_pipe( ialu_reg_fat );
7960 %}
7961 
7962 // Multiply by 32-bit Immediate, taking the shifted high order results
7963 //  (special case for shift by 32)
7964 instruct mulI_imm_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32 cnt, eFlagsReg cr) %{
7965   match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
7966   predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
7967              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
7968              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
7969   effect(USE src1, KILL cr);
7970 
7971   // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
7972   ins_cost(0*100 + 1*400 - 150);
7973   format %{ "IMUL   EDX:EAX,$src1" %}
7974   ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
7975   ins_pipe( pipe_slow );
7976 %}
7977 
7978 // Multiply by 32-bit Immediate, taking the shifted high order results
7979 instruct mulI_imm_RShift_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr) %{
7980   match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
7981   predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
7982              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
7983              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
7984   effect(USE src1, KILL cr);
7985 
7986   // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
7987   ins_cost(1*100 + 1*400 - 150);
7988   format %{ "IMUL   EDX:EAX,$src1\n\t"
7989             "SAR    EDX,$cnt-32" %}
7990   ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
7991   ins_pipe( pipe_slow );
7992 %}
7993 
7994 // Multiply Memory 32-bit Immediate
7995 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, eFlagsReg cr) %{
7996   match(Set dst (MulI (LoadI src) imm));
7997   effect(KILL cr);
7998 
7999   ins_cost(300);
8000   format %{ "IMUL   $dst,$src,$imm" %}
8001   opcode(0x69);  /* 69 /r id */
8002   ins_encode( OpcSE(imm), RegMem( dst, src ), Con8or32( imm ) );
8003   ins_pipe( ialu_reg_mem_alu0 );
8004 %}
8005 
8006 // Multiply Memory
8007 instruct mulI(rRegI dst, memory src, eFlagsReg cr) %{
8008   match(Set dst (MulI dst (LoadI src)));
8009   effect(KILL cr);
8010 
8011   ins_cost(350);
8012   format %{ "IMUL   $dst,$src" %}
8013   opcode(0xAF, 0x0F);
8014   ins_encode( OpcS, OpcP, RegMem( dst, src) );
8015   ins_pipe( ialu_reg_mem_alu0 );
8016 %}
8017 
8018 // Multiply Register Int to Long
8019 instruct mulI2L(eADXRegL dst, eAXRegI src, nadxRegI src1, eFlagsReg flags) %{
8020   // Basic Idea: long = (long)int * (long)int
8021   match(Set dst (MulL (ConvI2L src) (ConvI2L src1)));
8022   effect(DEF dst, USE src, USE src1, KILL flags);
8023 
8024   ins_cost(300);
8025   format %{ "IMUL   $dst,$src1" %}
8026 
8027   ins_encode( long_int_multiply( dst, src1 ) );
8028   ins_pipe( ialu_reg_reg_alu0 );
8029 %}
8030 
8031 instruct mulIS_eReg(eADXRegL dst, immL_32bits mask, eFlagsReg flags, eAXRegI src, nadxRegI src1) %{
8032   // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
8033   match(Set dst (MulL (AndL (ConvI2L src) mask) (AndL (ConvI2L src1) mask)));
8034   effect(KILL flags);
8035 
8036   ins_cost(300);
8037   format %{ "MUL    $dst,$src1" %}
8038 
8039   ins_encode( long_uint_multiply(dst, src1) );
8040   ins_pipe( ialu_reg_reg_alu0 );
8041 %}
8042 
8043 // Multiply Register Long
8044 instruct mulL_eReg(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
8045   match(Set dst (MulL dst src));
8046   effect(KILL cr, TEMP tmp);
8047   ins_cost(4*100+3*400);
8048 // Basic idea: lo(result) = lo(x_lo * y_lo)
8049 //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
8050   format %{ "MOV    $tmp,$src.lo\n\t"
8051             "IMUL   $tmp,EDX\n\t"
8052             "MOV    EDX,$src.hi\n\t"
8053             "IMUL   EDX,EAX\n\t"
8054             "ADD    $tmp,EDX\n\t"
8055             "MUL    EDX:EAX,$src.lo\n\t"
8056             "ADD    EDX,$tmp" %}
8057   ins_encode( long_multiply( dst, src, tmp ) );
8058   ins_pipe( pipe_slow );
8059 %}
8060 
8061 // Multiply Register Long where the left operand's high 32 bits are zero
8062 instruct mulL_eReg_lhi0(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
8063   predicate(is_operand_hi32_zero(n->in(1)));
8064   match(Set dst (MulL dst src));
8065   effect(KILL cr, TEMP tmp);
8066   ins_cost(2*100+2*400);
8067 // Basic idea: lo(result) = lo(x_lo * y_lo)
8068 //             hi(result) = hi(x_lo * y_lo) + lo(x_lo * y_hi) where lo(x_hi * y_lo) = 0 because x_hi = 0
8069   format %{ "MOV    $tmp,$src.hi\n\t"
8070             "IMUL   $tmp,EAX\n\t"
8071             "MUL    EDX:EAX,$src.lo\n\t"
8072             "ADD    EDX,$tmp" %}
8073   ins_encode %{
8074     __ movl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
8075     __ imull($tmp$$Register, rax);
8076     __ mull($src$$Register);
8077     __ addl(rdx, $tmp$$Register);
8078   %}
8079   ins_pipe( pipe_slow );
8080 %}
8081 
8082 // Multiply Register Long where the right operand's high 32 bits are zero
8083 instruct mulL_eReg_rhi0(eADXRegL dst, eRegL src, rRegI tmp, eFlagsReg cr) %{
8084   predicate(is_operand_hi32_zero(n->in(2)));
8085   match(Set dst (MulL dst src));
8086   effect(KILL cr, TEMP tmp);
8087   ins_cost(2*100+2*400);
8088 // Basic idea: lo(result) = lo(x_lo * y_lo)
8089 //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) where lo(x_lo * y_hi) = 0 because y_hi = 0
8090   format %{ "MOV    $tmp,$src.lo\n\t"
8091             "IMUL   $tmp,EDX\n\t"
8092             "MUL    EDX:EAX,$src.lo\n\t"
8093             "ADD    EDX,$tmp" %}
8094   ins_encode %{
8095     __ movl($tmp$$Register, $src$$Register);
8096     __ imull($tmp$$Register, rdx);
8097     __ mull($src$$Register);
8098     __ addl(rdx, $tmp$$Register);
8099   %}
8100   ins_pipe( pipe_slow );
8101 %}
8102 
8103 // Multiply Register Long where the left and the right operands' high 32 bits are zero
8104 instruct mulL_eReg_hi0(eADXRegL dst, eRegL src, eFlagsReg cr) %{
8105   predicate(is_operand_hi32_zero(n->in(1)) && is_operand_hi32_zero(n->in(2)));
8106   match(Set dst (MulL dst src));
8107   effect(KILL cr);
8108   ins_cost(1*400);
8109 // Basic idea: lo(result) = lo(x_lo * y_lo)
8110 //             hi(result) = hi(x_lo * y_lo) where lo(x_hi * y_lo) = 0 and lo(x_lo * y_hi) = 0 because x_hi = 0 and y_hi = 0
8111   format %{ "MUL    EDX:EAX,$src.lo\n\t" %}
8112   ins_encode %{
8113     __ mull($src$$Register);
8114   %}
8115   ins_pipe( pipe_slow );
8116 %}
8117 
8118 // Multiply Register Long by small constant
8119 instruct mulL_eReg_con(eADXRegL dst, immL_127 src, rRegI tmp, eFlagsReg cr) %{
8120   match(Set dst (MulL dst src));
8121   effect(KILL cr, TEMP tmp);
8122   ins_cost(2*100+2*400);
8123   size(12);
8124 // Basic idea: lo(result) = lo(src * EAX)
8125 //             hi(result) = hi(src * EAX) + lo(src * EDX)
8126   format %{ "IMUL   $tmp,EDX,$src\n\t"
8127             "MOV    EDX,$src\n\t"
8128             "MUL    EDX\t# EDX*EAX -> EDX:EAX\n\t"
8129             "ADD    EDX,$tmp" %}
8130   ins_encode( long_multiply_con( dst, src, tmp ) );
8131   ins_pipe( pipe_slow );
8132 %}
8133 
8134 // Integer DIV with Register
8135 instruct divI_eReg(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
8136   match(Set rax (DivI rax div));
8137   effect(KILL rdx, KILL cr);
8138   size(26);
8139   ins_cost(30*100+10*100);
8140   format %{ "CMP    EAX,0x80000000\n\t"
8141             "JNE,s  normal\n\t"
8142             "XOR    EDX,EDX\n\t"
8143             "CMP    ECX,-1\n\t"
8144             "JE,s   done\n"
8145     "normal: CDQ\n\t"
8146             "IDIV   $div\n\t"
8147     "done:"        %}
8148   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8149   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8150   ins_pipe( ialu_reg_reg_alu0 );
8151 %}
8152 
8153 // Divide Register Long
8154 instruct divL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
8155   match(Set dst (DivL src1 src2));
8156   effect( KILL cr, KILL cx, KILL bx );
8157   ins_cost(10000);
8158   format %{ "PUSH   $src1.hi\n\t"
8159             "PUSH   $src1.lo\n\t"
8160             "PUSH   $src2.hi\n\t"
8161             "PUSH   $src2.lo\n\t"
8162             "CALL   SharedRuntime::ldiv\n\t"
8163             "ADD    ESP,16" %}
8164   ins_encode( long_div(src1,src2) );
8165   ins_pipe( pipe_slow );
8166 %}
8167 
8168 // Integer DIVMOD with Register, both quotient and mod results
8169 instruct divModI_eReg_divmod(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
8170   match(DivModI rax div);
8171   effect(KILL cr);
8172   size(26);
8173   ins_cost(30*100+10*100);
8174   format %{ "CMP    EAX,0x80000000\n\t"
8175             "JNE,s  normal\n\t"
8176             "XOR    EDX,EDX\n\t"
8177             "CMP    ECX,-1\n\t"
8178             "JE,s   done\n"
8179     "normal: CDQ\n\t"
8180             "IDIV   $div\n\t"
8181     "done:"        %}
8182   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8183   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8184   ins_pipe( pipe_slow );
8185 %}
8186 
8187 // Integer MOD with Register
8188 instruct modI_eReg(eDXRegI rdx, eAXRegI rax, eCXRegI div, eFlagsReg cr) %{
8189   match(Set rdx (ModI rax div));
8190   effect(KILL rax, KILL cr);
8191 
8192   size(26);
8193   ins_cost(300);
8194   format %{ "CDQ\n\t"
8195             "IDIV   $div" %}
8196   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8197   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8198   ins_pipe( ialu_reg_reg_alu0 );
8199 %}
8200 
8201 // Remainder Register Long
8202 instruct modL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
8203   match(Set dst (ModL src1 src2));
8204   effect( KILL cr, KILL cx, KILL bx );
8205   ins_cost(10000);
8206   format %{ "PUSH   $src1.hi\n\t"
8207             "PUSH   $src1.lo\n\t"
8208             "PUSH   $src2.hi\n\t"
8209             "PUSH   $src2.lo\n\t"
8210             "CALL   SharedRuntime::lrem\n\t"
8211             "ADD    ESP,16" %}
8212   ins_encode( long_mod(src1,src2) );
8213   ins_pipe( pipe_slow );
8214 %}
8215 
8216 // Divide Register Long (no special case since divisor != -1)
8217 instruct divL_eReg_imm32( eADXRegL dst, immL32 imm, rRegI tmp, rRegI tmp2, eFlagsReg cr ) %{
8218   match(Set dst (DivL dst imm));
8219   effect( TEMP tmp, TEMP tmp2, KILL cr );
8220   ins_cost(1000);
8221   format %{ "MOV    $tmp,abs($imm) # ldiv EDX:EAX,$imm\n\t"
8222             "XOR    $tmp2,$tmp2\n\t"
8223             "CMP    $tmp,EDX\n\t"
8224             "JA,s   fast\n\t"
8225             "MOV    $tmp2,EAX\n\t"
8226             "MOV    EAX,EDX\n\t"
8227             "MOV    EDX,0\n\t"
8228             "JLE,s  pos\n\t"
8229             "LNEG   EAX : $tmp2\n\t"
8230             "DIV    $tmp # unsigned division\n\t"
8231             "XCHG   EAX,$tmp2\n\t"
8232             "DIV    $tmp\n\t"
8233             "LNEG   $tmp2 : EAX\n\t"
8234             "JMP,s  done\n"
8235     "pos:\n\t"
8236             "DIV    $tmp\n\t"
8237             "XCHG   EAX,$tmp2\n"
8238     "fast:\n\t"
8239             "DIV    $tmp\n"
8240     "done:\n\t"
8241             "MOV    EDX,$tmp2\n\t"
8242             "NEG    EDX:EAX # if $imm < 0" %}
8243   ins_encode %{
8244     int con = (int)$imm$$constant;
8245     assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
8246     int pcon = (con > 0) ? con : -con;
8247     Label Lfast, Lpos, Ldone;
8248 
8249     __ movl($tmp$$Register, pcon);
8250     __ xorl($tmp2$$Register,$tmp2$$Register);
8251     __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
8252     __ jccb(Assembler::above, Lfast); // result fits into 32 bit
8253 
8254     __ movl($tmp2$$Register, $dst$$Register); // save
8255     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8256     __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
8257     __ jccb(Assembler::lessEqual, Lpos); // result is positive
8258 
8259     // Negative dividend.
8260     // convert value to positive to use unsigned division
8261     __ lneg($dst$$Register, $tmp2$$Register);
8262     __ divl($tmp$$Register);
8263     __ xchgl($dst$$Register, $tmp2$$Register);
8264     __ divl($tmp$$Register);
8265     // revert result back to negative
8266     __ lneg($tmp2$$Register, $dst$$Register);
8267     __ jmpb(Ldone);
8268 
8269     __ bind(Lpos);
8270     __ divl($tmp$$Register); // Use unsigned division
8271     __ xchgl($dst$$Register, $tmp2$$Register);
8272     // Fallthrow for final divide, tmp2 has 32 bit hi result
8273 
8274     __ bind(Lfast);
8275     // fast path: src is positive
8276     __ divl($tmp$$Register); // Use unsigned division
8277 
8278     __ bind(Ldone);
8279     __ movl(HIGH_FROM_LOW($dst$$Register),$tmp2$$Register);
8280     if (con < 0) {
8281       __ lneg(HIGH_FROM_LOW($dst$$Register), $dst$$Register);
8282     }
8283   %}
8284   ins_pipe( pipe_slow );
8285 %}
8286 
8287 // Remainder Register Long (remainder fit into 32 bits)
8288 instruct modL_eReg_imm32( eADXRegL dst, immL32 imm, rRegI tmp, rRegI tmp2, eFlagsReg cr ) %{
8289   match(Set dst (ModL dst imm));
8290   effect( TEMP tmp, TEMP tmp2, KILL cr );
8291   ins_cost(1000);
8292   format %{ "MOV    $tmp,abs($imm) # lrem EDX:EAX,$imm\n\t"
8293             "CMP    $tmp,EDX\n\t"
8294             "JA,s   fast\n\t"
8295             "MOV    $tmp2,EAX\n\t"
8296             "MOV    EAX,EDX\n\t"
8297             "MOV    EDX,0\n\t"
8298             "JLE,s  pos\n\t"
8299             "LNEG   EAX : $tmp2\n\t"
8300             "DIV    $tmp # unsigned division\n\t"
8301             "MOV    EAX,$tmp2\n\t"
8302             "DIV    $tmp\n\t"
8303             "NEG    EDX\n\t"
8304             "JMP,s  done\n"
8305     "pos:\n\t"
8306             "DIV    $tmp\n\t"
8307             "MOV    EAX,$tmp2\n"
8308     "fast:\n\t"
8309             "DIV    $tmp\n"
8310     "done:\n\t"
8311             "MOV    EAX,EDX\n\t"
8312             "SAR    EDX,31\n\t" %}
8313   ins_encode %{
8314     int con = (int)$imm$$constant;
8315     assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
8316     int pcon = (con > 0) ? con : -con;
8317     Label  Lfast, Lpos, Ldone;
8318 
8319     __ movl($tmp$$Register, pcon);
8320     __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
8321     __ jccb(Assembler::above, Lfast); // src is positive and result fits into 32 bit
8322 
8323     __ movl($tmp2$$Register, $dst$$Register); // save
8324     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8325     __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
8326     __ jccb(Assembler::lessEqual, Lpos); // result is positive
8327 
8328     // Negative dividend.
8329     // convert value to positive to use unsigned division
8330     __ lneg($dst$$Register, $tmp2$$Register);
8331     __ divl($tmp$$Register);
8332     __ movl($dst$$Register, $tmp2$$Register);
8333     __ divl($tmp$$Register);
8334     // revert remainder back to negative
8335     __ negl(HIGH_FROM_LOW($dst$$Register));
8336     __ jmpb(Ldone);
8337 
8338     __ bind(Lpos);
8339     __ divl($tmp$$Register);
8340     __ movl($dst$$Register, $tmp2$$Register);
8341 
8342     __ bind(Lfast);
8343     // fast path: src is positive
8344     __ divl($tmp$$Register);
8345 
8346     __ bind(Ldone);
8347     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8348     __ sarl(HIGH_FROM_LOW($dst$$Register), 31); // result sign
8349 
8350   %}
8351   ins_pipe( pipe_slow );
8352 %}
8353 
8354 // Integer Shift Instructions
8355 // Shift Left by one
8356 instruct shlI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
8357   match(Set dst (LShiftI dst shift));
8358   effect(KILL cr);
8359 
8360   size(2);
8361   format %{ "SHL    $dst,$shift" %}
8362   opcode(0xD1, 0x4);  /* D1 /4 */
8363   ins_encode( OpcP, RegOpc( dst ) );
8364   ins_pipe( ialu_reg );
8365 %}
8366 
8367 // Shift Left by 8-bit immediate
8368 instruct salI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
8369   match(Set dst (LShiftI dst shift));
8370   effect(KILL cr);
8371 
8372   size(3);
8373   format %{ "SHL    $dst,$shift" %}
8374   opcode(0xC1, 0x4);  /* C1 /4 ib */
8375   ins_encode( RegOpcImm( dst, shift) );
8376   ins_pipe( ialu_reg );
8377 %}
8378 
8379 // Shift Left by variable
8380 instruct salI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
8381   match(Set dst (LShiftI dst shift));
8382   effect(KILL cr);
8383 
8384   size(2);
8385   format %{ "SHL    $dst,$shift" %}
8386   opcode(0xD3, 0x4);  /* D3 /4 */
8387   ins_encode( OpcP, RegOpc( dst ) );
8388   ins_pipe( ialu_reg_reg );
8389 %}
8390 
8391 // Arithmetic shift right by one
8392 instruct sarI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
8393   match(Set dst (RShiftI dst shift));
8394   effect(KILL cr);
8395 
8396   size(2);
8397   format %{ "SAR    $dst,$shift" %}
8398   opcode(0xD1, 0x7);  /* D1 /7 */
8399   ins_encode( OpcP, RegOpc( dst ) );
8400   ins_pipe( ialu_reg );
8401 %}
8402 
8403 // Arithmetic shift right by one
8404 instruct sarI_mem_1(memory dst, immI1 shift, eFlagsReg cr) %{
8405   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8406   effect(KILL cr);
8407   format %{ "SAR    $dst,$shift" %}
8408   opcode(0xD1, 0x7);  /* D1 /7 */
8409   ins_encode( OpcP, RMopc_Mem(secondary,dst) );
8410   ins_pipe( ialu_mem_imm );
8411 %}
8412 
8413 // Arithmetic Shift Right by 8-bit immediate
8414 instruct sarI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
8415   match(Set dst (RShiftI dst shift));
8416   effect(KILL cr);
8417 
8418   size(3);
8419   format %{ "SAR    $dst,$shift" %}
8420   opcode(0xC1, 0x7);  /* C1 /7 ib */
8421   ins_encode( RegOpcImm( dst, shift ) );
8422   ins_pipe( ialu_mem_imm );
8423 %}
8424 
8425 // Arithmetic Shift Right by 8-bit immediate
8426 instruct sarI_mem_imm(memory dst, immI8 shift, eFlagsReg cr) %{
8427   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8428   effect(KILL cr);
8429 
8430   format %{ "SAR    $dst,$shift" %}
8431   opcode(0xC1, 0x7);  /* C1 /7 ib */
8432   ins_encode( OpcP, RMopc_Mem(secondary, dst ), Con8or32( shift ) );
8433   ins_pipe( ialu_mem_imm );
8434 %}
8435 
8436 // Arithmetic Shift Right by variable
8437 instruct sarI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
8438   match(Set dst (RShiftI dst shift));
8439   effect(KILL cr);
8440 
8441   size(2);
8442   format %{ "SAR    $dst,$shift" %}
8443   opcode(0xD3, 0x7);  /* D3 /7 */
8444   ins_encode( OpcP, RegOpc( dst ) );
8445   ins_pipe( ialu_reg_reg );
8446 %}
8447 
8448 // Logical shift right by one
8449 instruct shrI_eReg_1(rRegI dst, immI1 shift, eFlagsReg cr) %{
8450   match(Set dst (URShiftI dst shift));
8451   effect(KILL cr);
8452 
8453   size(2);
8454   format %{ "SHR    $dst,$shift" %}
8455   opcode(0xD1, 0x5);  /* D1 /5 */
8456   ins_encode( OpcP, RegOpc( dst ) );
8457   ins_pipe( ialu_reg );
8458 %}
8459 
8460 // Logical Shift Right by 8-bit immediate
8461 instruct shrI_eReg_imm(rRegI dst, immI8 shift, eFlagsReg cr) %{
8462   match(Set dst (URShiftI dst shift));
8463   effect(KILL cr);
8464 
8465   size(3);
8466   format %{ "SHR    $dst,$shift" %}
8467   opcode(0xC1, 0x5);  /* C1 /5 ib */
8468   ins_encode( RegOpcImm( dst, shift) );
8469   ins_pipe( ialu_reg );
8470 %}
8471 
8472 
8473 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
8474 // This idiom is used by the compiler for the i2b bytecode.
8475 instruct i2b(rRegI dst, xRegI src, immI_24 twentyfour) %{
8476   match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
8477 
8478   size(3);
8479   format %{ "MOVSX  $dst,$src :8" %}
8480   ins_encode %{
8481     __ movsbl($dst$$Register, $src$$Register);
8482   %}
8483   ins_pipe(ialu_reg_reg);
8484 %}
8485 
8486 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
8487 // This idiom is used by the compiler the i2s bytecode.
8488 instruct i2s(rRegI dst, xRegI src, immI_16 sixteen) %{
8489   match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
8490 
8491   size(3);
8492   format %{ "MOVSX  $dst,$src :16" %}
8493   ins_encode %{
8494     __ movswl($dst$$Register, $src$$Register);
8495   %}
8496   ins_pipe(ialu_reg_reg);
8497 %}
8498 
8499 
8500 // Logical Shift Right by variable
8501 instruct shrI_eReg_CL(rRegI dst, eCXRegI shift, eFlagsReg cr) %{
8502   match(Set dst (URShiftI dst shift));
8503   effect(KILL cr);
8504 
8505   size(2);
8506   format %{ "SHR    $dst,$shift" %}
8507   opcode(0xD3, 0x5);  /* D3 /5 */
8508   ins_encode( OpcP, RegOpc( dst ) );
8509   ins_pipe( ialu_reg_reg );
8510 %}
8511 
8512 
8513 //----------Logical Instructions-----------------------------------------------
8514 //----------Integer Logical Instructions---------------------------------------
8515 // And Instructions
8516 // And Register with Register
8517 instruct andI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
8518   match(Set dst (AndI dst src));
8519   effect(KILL cr);
8520 
8521   size(2);
8522   format %{ "AND    $dst,$src" %}
8523   opcode(0x23);
8524   ins_encode( OpcP, RegReg( dst, src) );
8525   ins_pipe( ialu_reg_reg );
8526 %}
8527 
8528 // And Register with Immediate
8529 instruct andI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
8530   match(Set dst (AndI dst src));
8531   effect(KILL cr);
8532 
8533   format %{ "AND    $dst,$src" %}
8534   opcode(0x81,0x04);  /* Opcode 81 /4 */
8535   // ins_encode( RegImm( dst, src) );
8536   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8537   ins_pipe( ialu_reg );
8538 %}
8539 
8540 // And Register with Memory
8541 instruct andI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
8542   match(Set dst (AndI dst (LoadI src)));
8543   effect(KILL cr);
8544 
8545   ins_cost(125);
8546   format %{ "AND    $dst,$src" %}
8547   opcode(0x23);
8548   ins_encode( OpcP, RegMem( dst, src) );
8549   ins_pipe( ialu_reg_mem );
8550 %}
8551 
8552 // And Memory with Register
8553 instruct andI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
8554   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
8555   effect(KILL cr);
8556 
8557   ins_cost(150);
8558   format %{ "AND    $dst,$src" %}
8559   opcode(0x21);  /* Opcode 21 /r */
8560   ins_encode( OpcP, RegMem( src, dst ) );
8561   ins_pipe( ialu_mem_reg );
8562 %}
8563 
8564 // And Memory with Immediate
8565 instruct andI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8566   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
8567   effect(KILL cr);
8568 
8569   ins_cost(125);
8570   format %{ "AND    $dst,$src" %}
8571   opcode(0x81, 0x4);  /* Opcode 81 /4 id */
8572   // ins_encode( MemImm( dst, src) );
8573   ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
8574   ins_pipe( ialu_mem_imm );
8575 %}
8576 
8577 // Or Instructions
8578 // Or Register with Register
8579 instruct orI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
8580   match(Set dst (OrI dst src));
8581   effect(KILL cr);
8582 
8583   size(2);
8584   format %{ "OR     $dst,$src" %}
8585   opcode(0x0B);
8586   ins_encode( OpcP, RegReg( dst, src) );
8587   ins_pipe( ialu_reg_reg );
8588 %}
8589 
8590 instruct orI_eReg_castP2X(rRegI dst, eRegP src, eFlagsReg cr) %{
8591   match(Set dst (OrI dst (CastP2X src)));
8592   effect(KILL cr);
8593 
8594   size(2);
8595   format %{ "OR     $dst,$src" %}
8596   opcode(0x0B);
8597   ins_encode( OpcP, RegReg( dst, src) );
8598   ins_pipe( ialu_reg_reg );
8599 %}
8600 
8601 
8602 // Or Register with Immediate
8603 instruct orI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
8604   match(Set dst (OrI dst src));
8605   effect(KILL cr);
8606 
8607   format %{ "OR     $dst,$src" %}
8608   opcode(0x81,0x01);  /* Opcode 81 /1 id */
8609   // ins_encode( RegImm( dst, src) );
8610   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8611   ins_pipe( ialu_reg );
8612 %}
8613 
8614 // Or Register with Memory
8615 instruct orI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
8616   match(Set dst (OrI dst (LoadI src)));
8617   effect(KILL cr);
8618 
8619   ins_cost(125);
8620   format %{ "OR     $dst,$src" %}
8621   opcode(0x0B);
8622   ins_encode( OpcP, RegMem( dst, src) );
8623   ins_pipe( ialu_reg_mem );
8624 %}
8625 
8626 // Or Memory with Register
8627 instruct orI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
8628   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
8629   effect(KILL cr);
8630 
8631   ins_cost(150);
8632   format %{ "OR     $dst,$src" %}
8633   opcode(0x09);  /* Opcode 09 /r */
8634   ins_encode( OpcP, RegMem( src, dst ) );
8635   ins_pipe( ialu_mem_reg );
8636 %}
8637 
8638 // Or Memory with Immediate
8639 instruct orI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8640   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
8641   effect(KILL cr);
8642 
8643   ins_cost(125);
8644   format %{ "OR     $dst,$src" %}
8645   opcode(0x81,0x1);  /* Opcode 81 /1 id */
8646   // ins_encode( MemImm( dst, src) );
8647   ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
8648   ins_pipe( ialu_mem_imm );
8649 %}
8650 
8651 // ROL/ROR
8652 // ROL expand
8653 instruct rolI_eReg_imm1(rRegI dst, immI1 shift, eFlagsReg cr) %{
8654   effect(USE_DEF dst, USE shift, KILL cr);
8655 
8656   format %{ "ROL    $dst, $shift" %}
8657   opcode(0xD1, 0x0); /* Opcode D1 /0 */
8658   ins_encode( OpcP, RegOpc( dst ));
8659   ins_pipe( ialu_reg );
8660 %}
8661 
8662 instruct rolI_eReg_imm8(rRegI dst, immI8 shift, eFlagsReg cr) %{
8663   effect(USE_DEF dst, USE shift, KILL cr);
8664 
8665   format %{ "ROL    $dst, $shift" %}
8666   opcode(0xC1, 0x0); /*Opcode /C1  /0  */
8667   ins_encode( RegOpcImm(dst, shift) );
8668   ins_pipe(ialu_reg);
8669 %}
8670 
8671 instruct rolI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr) %{
8672   effect(USE_DEF dst, USE shift, KILL cr);
8673 
8674   format %{ "ROL    $dst, $shift" %}
8675   opcode(0xD3, 0x0);    /* Opcode D3 /0 */
8676   ins_encode(OpcP, RegOpc(dst));
8677   ins_pipe( ialu_reg_reg );
8678 %}
8679 // end of ROL expand
8680 
8681 // ROL 32bit by one once
8682 instruct rolI_eReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, eFlagsReg cr) %{
8683   match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8684 
8685   expand %{
8686     rolI_eReg_imm1(dst, lshift, cr);
8687   %}
8688 %}
8689 
8690 // ROL 32bit var by imm8 once
8691 instruct rolI_eReg_i8(rRegI dst, immI8 lshift, immI8 rshift, eFlagsReg cr) %{
8692   predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8693   match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8694 
8695   expand %{
8696     rolI_eReg_imm8(dst, lshift, cr);
8697   %}
8698 %}
8699 
8700 // ROL 32bit var by var once
8701 instruct rolI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
8702   match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
8703 
8704   expand %{
8705     rolI_eReg_CL(dst, shift, cr);
8706   %}
8707 %}
8708 
8709 // ROL 32bit var by var once
8710 instruct rolI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
8711   match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
8712 
8713   expand %{
8714     rolI_eReg_CL(dst, shift, cr);
8715   %}
8716 %}
8717 
8718 // ROR expand
8719 instruct rorI_eReg_imm1(rRegI dst, immI1 shift, eFlagsReg cr) %{
8720   effect(USE_DEF dst, USE shift, KILL cr);
8721 
8722   format %{ "ROR    $dst, $shift" %}
8723   opcode(0xD1,0x1);  /* Opcode D1 /1 */
8724   ins_encode( OpcP, RegOpc( dst ) );
8725   ins_pipe( ialu_reg );
8726 %}
8727 
8728 instruct rorI_eReg_imm8(rRegI dst, immI8 shift, eFlagsReg cr) %{
8729   effect (USE_DEF dst, USE shift, KILL cr);
8730 
8731   format %{ "ROR    $dst, $shift" %}
8732   opcode(0xC1, 0x1); /* Opcode /C1 /1 ib */
8733   ins_encode( RegOpcImm(dst, shift) );
8734   ins_pipe( ialu_reg );
8735 %}
8736 
8737 instruct rorI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr)%{
8738   effect(USE_DEF dst, USE shift, KILL cr);
8739 
8740   format %{ "ROR    $dst, $shift" %}
8741   opcode(0xD3, 0x1);    /* Opcode D3 /1 */
8742   ins_encode(OpcP, RegOpc(dst));
8743   ins_pipe( ialu_reg_reg );
8744 %}
8745 // end of ROR expand
8746 
8747 // ROR right once
8748 instruct rorI_eReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, eFlagsReg cr) %{
8749   match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8750 
8751   expand %{
8752     rorI_eReg_imm1(dst, rshift, cr);
8753   %}
8754 %}
8755 
8756 // ROR 32bit by immI8 once
8757 instruct rorI_eReg_i8(rRegI dst, immI8 rshift, immI8 lshift, eFlagsReg cr) %{
8758   predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8759   match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8760 
8761   expand %{
8762     rorI_eReg_imm8(dst, rshift, cr);
8763   %}
8764 %}
8765 
8766 // ROR 32bit var by var once
8767 instruct rorI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
8768   match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
8769 
8770   expand %{
8771     rorI_eReg_CL(dst, shift, cr);
8772   %}
8773 %}
8774 
8775 // ROR 32bit var by var once
8776 instruct rorI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
8777   match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
8778 
8779   expand %{
8780     rorI_eReg_CL(dst, shift, cr);
8781   %}
8782 %}
8783 
8784 // Xor Instructions
8785 // Xor Register with Register
8786 instruct xorI_eReg(rRegI dst, rRegI src, eFlagsReg cr) %{
8787   match(Set dst (XorI dst src));
8788   effect(KILL cr);
8789 
8790   size(2);
8791   format %{ "XOR    $dst,$src" %}
8792   opcode(0x33);
8793   ins_encode( OpcP, RegReg( dst, src) );
8794   ins_pipe( ialu_reg_reg );
8795 %}
8796 
8797 // Xor Register with Immediate -1
8798 instruct xorI_eReg_im1(rRegI dst, immI_M1 imm) %{
8799   match(Set dst (XorI dst imm));  
8800 
8801   size(2);
8802   format %{ "NOT    $dst" %}  
8803   ins_encode %{
8804      __ notl($dst$$Register);
8805   %}
8806   ins_pipe( ialu_reg );
8807 %}
8808 
8809 // Xor Register with Immediate
8810 instruct xorI_eReg_imm(rRegI dst, immI src, eFlagsReg cr) %{
8811   match(Set dst (XorI dst src));
8812   effect(KILL cr);
8813 
8814   format %{ "XOR    $dst,$src" %}
8815   opcode(0x81,0x06);  /* Opcode 81 /6 id */
8816   // ins_encode( RegImm( dst, src) );
8817   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8818   ins_pipe( ialu_reg );
8819 %}
8820 
8821 // Xor Register with Memory
8822 instruct xorI_eReg_mem(rRegI dst, memory src, eFlagsReg cr) %{
8823   match(Set dst (XorI dst (LoadI src)));
8824   effect(KILL cr);
8825 
8826   ins_cost(125);
8827   format %{ "XOR    $dst,$src" %}
8828   opcode(0x33);
8829   ins_encode( OpcP, RegMem(dst, src) );
8830   ins_pipe( ialu_reg_mem );
8831 %}
8832 
8833 // Xor Memory with Register
8834 instruct xorI_mem_eReg(memory dst, rRegI src, eFlagsReg cr) %{
8835   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
8836   effect(KILL cr);
8837 
8838   ins_cost(150);
8839   format %{ "XOR    $dst,$src" %}
8840   opcode(0x31);  /* Opcode 31 /r */
8841   ins_encode( OpcP, RegMem( src, dst ) );
8842   ins_pipe( ialu_mem_reg );
8843 %}
8844 
8845 // Xor Memory with Immediate
8846 instruct xorI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8847   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
8848   effect(KILL cr);
8849 
8850   ins_cost(125);
8851   format %{ "XOR    $dst,$src" %}
8852   opcode(0x81,0x6);  /* Opcode 81 /6 id */
8853   ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
8854   ins_pipe( ialu_mem_imm );
8855 %}
8856 
8857 //----------Convert Int to Boolean---------------------------------------------
8858 
8859 instruct movI_nocopy(rRegI dst, rRegI src) %{
8860   effect( DEF dst, USE src );
8861   format %{ "MOV    $dst,$src" %}
8862   ins_encode( enc_Copy( dst, src) );
8863   ins_pipe( ialu_reg_reg );
8864 %}
8865 
8866 instruct ci2b( rRegI dst, rRegI src, eFlagsReg cr ) %{
8867   effect( USE_DEF dst, USE src, KILL cr );
8868 
8869   size(4);
8870   format %{ "NEG    $dst\n\t"
8871             "ADC    $dst,$src" %}
8872   ins_encode( neg_reg(dst),
8873               OpcRegReg(0x13,dst,src) );
8874   ins_pipe( ialu_reg_reg_long );
8875 %}
8876 
8877 instruct convI2B( rRegI dst, rRegI src, eFlagsReg cr ) %{
8878   match(Set dst (Conv2B src));
8879 
8880   expand %{
8881     movI_nocopy(dst,src);
8882     ci2b(dst,src,cr);
8883   %}
8884 %}
8885 
8886 instruct movP_nocopy(rRegI dst, eRegP src) %{
8887   effect( DEF dst, USE src );
8888   format %{ "MOV    $dst,$src" %}
8889   ins_encode( enc_Copy( dst, src) );
8890   ins_pipe( ialu_reg_reg );
8891 %}
8892 
8893 instruct cp2b( rRegI dst, eRegP src, eFlagsReg cr ) %{
8894   effect( USE_DEF dst, USE src, KILL cr );
8895   format %{ "NEG    $dst\n\t"
8896             "ADC    $dst,$src" %}
8897   ins_encode( neg_reg(dst),
8898               OpcRegReg(0x13,dst,src) );
8899   ins_pipe( ialu_reg_reg_long );
8900 %}
8901 
8902 instruct convP2B( rRegI dst, eRegP src, eFlagsReg cr ) %{
8903   match(Set dst (Conv2B src));
8904 
8905   expand %{
8906     movP_nocopy(dst,src);
8907     cp2b(dst,src,cr);
8908   %}
8909 %}
8910 
8911 instruct cmpLTMask(eCXRegI dst, ncxRegI p, ncxRegI q, eFlagsReg cr) %{
8912   match(Set dst (CmpLTMask p q));
8913   effect(KILL cr);
8914   ins_cost(400);
8915 
8916   // SETlt can only use low byte of EAX,EBX, ECX, or EDX as destination
8917   format %{ "XOR    $dst,$dst\n\t"
8918             "CMP    $p,$q\n\t"
8919             "SETlt  $dst\n\t"
8920             "NEG    $dst" %}
8921   ins_encode %{
8922     Register Rp = $p$$Register;
8923     Register Rq = $q$$Register;
8924     Register Rd = $dst$$Register;
8925     Label done;
8926     __ xorl(Rd, Rd);
8927     __ cmpl(Rp, Rq);
8928     __ setb(Assembler::less, Rd);
8929     __ negl(Rd);
8930   %}
8931 
8932   ins_pipe(pipe_slow);
8933 %}
8934 
8935 instruct cmpLTMask0(rRegI dst, immI0 zero, eFlagsReg cr) %{
8936   match(Set dst (CmpLTMask dst zero));
8937   effect(DEF dst, KILL cr);
8938   ins_cost(100);
8939 
8940   format %{ "SAR    $dst,31\t# cmpLTMask0" %}
8941   ins_encode %{
8942   __ sarl($dst$$Register, 31);
8943   %}
8944   ins_pipe(ialu_reg);
8945 %}
8946 
8947 /* better to save a register than avoid a branch */
8948 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y, eFlagsReg cr) %{
8949   match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
8950   effect(KILL cr);
8951   ins_cost(400);
8952   format %{ "SUB    $p,$q\t# cadd_cmpLTMask\n\t"
8953             "JGE    done\n\t"
8954             "ADD    $p,$y\n"
8955             "done:  " %}
8956   ins_encode %{
8957     Register Rp = $p$$Register;
8958     Register Rq = $q$$Register;
8959     Register Ry = $y$$Register;
8960     Label done;
8961     __ subl(Rp, Rq);
8962     __ jccb(Assembler::greaterEqual, done);
8963     __ addl(Rp, Ry);
8964     __ bind(done);
8965   %}
8966 
8967   ins_pipe(pipe_cmplt);
8968 %}
8969 
8970 /* better to save a register than avoid a branch */
8971 instruct and_cmpLTMask(rRegI p, rRegI q, rRegI y, eFlagsReg cr) %{
8972   match(Set y (AndI (CmpLTMask p q) y));
8973   effect(KILL cr);
8974 
8975   ins_cost(300);
8976 
8977   format %{ "CMPL     $p, $q\t# and_cmpLTMask\n\t"
8978             "JLT      done\n\t"
8979             "XORL     $y, $y\n"
8980             "done:  " %}
8981   ins_encode %{
8982     Register Rp = $p$$Register;
8983     Register Rq = $q$$Register;
8984     Register Ry = $y$$Register;
8985     Label done;
8986     __ cmpl(Rp, Rq);
8987     __ jccb(Assembler::less, done);
8988     __ xorl(Ry, Ry);
8989     __ bind(done);
8990   %}
8991 
8992   ins_pipe(pipe_cmplt);
8993 %}
8994 
8995 /* If I enable this, I encourage spilling in the inner loop of compress.
8996 instruct cadd_cmpLTMask_mem(ncxRegI p, ncxRegI q, memory y, eCXRegI tmp, eFlagsReg cr) %{
8997   match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
8998 */
8999 
9000 //----------Long Instructions------------------------------------------------
9001 // Add Long Register with Register
9002 instruct addL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9003   match(Set dst (AddL dst src));
9004   effect(KILL cr);
9005   ins_cost(200);
9006   format %{ "ADD    $dst.lo,$src.lo\n\t"
9007             "ADC    $dst.hi,$src.hi" %}
9008   opcode(0x03, 0x13);
9009   ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
9010   ins_pipe( ialu_reg_reg_long );
9011 %}
9012 
9013 // Add Long Register with Immediate
9014 instruct addL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9015   match(Set dst (AddL dst src));
9016   effect(KILL cr);
9017   format %{ "ADD    $dst.lo,$src.lo\n\t"
9018             "ADC    $dst.hi,$src.hi" %}
9019   opcode(0x81,0x00,0x02);  /* Opcode 81 /0, 81 /2 */
9020   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9021   ins_pipe( ialu_reg_long );
9022 %}
9023 
9024 // Add Long Register with Memory
9025 instruct addL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9026   match(Set dst (AddL dst (LoadL mem)));
9027   effect(KILL cr);
9028   ins_cost(125);
9029   format %{ "ADD    $dst.lo,$mem\n\t"
9030             "ADC    $dst.hi,$mem+4" %}
9031   opcode(0x03, 0x13);
9032   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9033   ins_pipe( ialu_reg_long_mem );
9034 %}
9035 
9036 // Subtract Long Register with Register.
9037 instruct subL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9038   match(Set dst (SubL dst src));
9039   effect(KILL cr);
9040   ins_cost(200);
9041   format %{ "SUB    $dst.lo,$src.lo\n\t"
9042             "SBB    $dst.hi,$src.hi" %}
9043   opcode(0x2B, 0x1B);
9044   ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
9045   ins_pipe( ialu_reg_reg_long );
9046 %}
9047 
9048 // Subtract Long Register with Immediate
9049 instruct subL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9050   match(Set dst (SubL dst src));
9051   effect(KILL cr);
9052   format %{ "SUB    $dst.lo,$src.lo\n\t"
9053             "SBB    $dst.hi,$src.hi" %}
9054   opcode(0x81,0x05,0x03);  /* Opcode 81 /5, 81 /3 */
9055   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9056   ins_pipe( ialu_reg_long );
9057 %}
9058 
9059 // Subtract Long Register with Memory
9060 instruct subL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9061   match(Set dst (SubL dst (LoadL mem)));
9062   effect(KILL cr);
9063   ins_cost(125);
9064   format %{ "SUB    $dst.lo,$mem\n\t"
9065             "SBB    $dst.hi,$mem+4" %}
9066   opcode(0x2B, 0x1B);
9067   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9068   ins_pipe( ialu_reg_long_mem );
9069 %}
9070 
9071 instruct negL_eReg(eRegL dst, immL0 zero, eFlagsReg cr) %{
9072   match(Set dst (SubL zero dst));
9073   effect(KILL cr);
9074   ins_cost(300);
9075   format %{ "NEG    $dst.hi\n\tNEG    $dst.lo\n\tSBB    $dst.hi,0" %}
9076   ins_encode( neg_long(dst) );
9077   ins_pipe( ialu_reg_reg_long );
9078 %}
9079 
9080 // And Long Register with Register
9081 instruct andL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9082   match(Set dst (AndL dst src));
9083   effect(KILL cr);
9084   format %{ "AND    $dst.lo,$src.lo\n\t"
9085             "AND    $dst.hi,$src.hi" %}
9086   opcode(0x23,0x23);
9087   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9088   ins_pipe( ialu_reg_reg_long );
9089 %}
9090 
9091 // And Long Register with Immediate
9092 instruct andL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9093   match(Set dst (AndL dst src));
9094   effect(KILL cr);
9095   format %{ "AND    $dst.lo,$src.lo\n\t"
9096             "AND    $dst.hi,$src.hi" %}
9097   opcode(0x81,0x04,0x04);  /* Opcode 81 /4, 81 /4 */
9098   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9099   ins_pipe( ialu_reg_long );
9100 %}
9101 
9102 // And Long Register with Memory
9103 instruct andL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9104   match(Set dst (AndL dst (LoadL mem)));
9105   effect(KILL cr);
9106   ins_cost(125);
9107   format %{ "AND    $dst.lo,$mem\n\t"
9108             "AND    $dst.hi,$mem+4" %}
9109   opcode(0x23, 0x23);
9110   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9111   ins_pipe( ialu_reg_long_mem );
9112 %}
9113 
9114 // Or Long Register with Register
9115 instruct orl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9116   match(Set dst (OrL dst src));
9117   effect(KILL cr);
9118   format %{ "OR     $dst.lo,$src.lo\n\t"
9119             "OR     $dst.hi,$src.hi" %}
9120   opcode(0x0B,0x0B);
9121   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9122   ins_pipe( ialu_reg_reg_long );
9123 %}
9124 
9125 // Or Long Register with Immediate
9126 instruct orl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9127   match(Set dst (OrL dst src));
9128   effect(KILL cr);
9129   format %{ "OR     $dst.lo,$src.lo\n\t"
9130             "OR     $dst.hi,$src.hi" %}
9131   opcode(0x81,0x01,0x01);  /* Opcode 81 /1, 81 /1 */
9132   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9133   ins_pipe( ialu_reg_long );
9134 %}
9135 
9136 // Or Long Register with Memory
9137 instruct orl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9138   match(Set dst (OrL dst (LoadL mem)));
9139   effect(KILL cr);
9140   ins_cost(125);
9141   format %{ "OR     $dst.lo,$mem\n\t"
9142             "OR     $dst.hi,$mem+4" %}
9143   opcode(0x0B,0x0B);
9144   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9145   ins_pipe( ialu_reg_long_mem );
9146 %}
9147 
9148 // Xor Long Register with Register
9149 instruct xorl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9150   match(Set dst (XorL dst src));
9151   effect(KILL cr);
9152   format %{ "XOR    $dst.lo,$src.lo\n\t"
9153             "XOR    $dst.hi,$src.hi" %}
9154   opcode(0x33,0x33);
9155   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9156   ins_pipe( ialu_reg_reg_long );
9157 %}
9158 
9159 // Xor Long Register with Immediate -1
9160 instruct xorl_eReg_im1(eRegL dst, immL_M1 imm) %{
9161   match(Set dst (XorL dst imm));  
9162   format %{ "NOT    $dst.lo\n\t"
9163             "NOT    $dst.hi" %}
9164   ins_encode %{
9165      __ notl($dst$$Register);
9166      __ notl(HIGH_FROM_LOW($dst$$Register));
9167   %}
9168   ins_pipe( ialu_reg_long );
9169 %}
9170 
9171 // Xor Long Register with Immediate
9172 instruct xorl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9173   match(Set dst (XorL dst src));
9174   effect(KILL cr);
9175   format %{ "XOR    $dst.lo,$src.lo\n\t"
9176             "XOR    $dst.hi,$src.hi" %}
9177   opcode(0x81,0x06,0x06);  /* Opcode 81 /6, 81 /6 */
9178   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9179   ins_pipe( ialu_reg_long );
9180 %}
9181 
9182 // Xor Long Register with Memory
9183 instruct xorl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9184   match(Set dst (XorL dst (LoadL mem)));
9185   effect(KILL cr);
9186   ins_cost(125);
9187   format %{ "XOR    $dst.lo,$mem\n\t"
9188             "XOR    $dst.hi,$mem+4" %}
9189   opcode(0x33,0x33);
9190   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9191   ins_pipe( ialu_reg_long_mem );
9192 %}
9193 
9194 // Shift Left Long by 1
9195 instruct shlL_eReg_1(eRegL dst, immI_1 cnt, eFlagsReg cr) %{
9196   predicate(UseNewLongLShift);
9197   match(Set dst (LShiftL dst cnt));
9198   effect(KILL cr);
9199   ins_cost(100);
9200   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9201             "ADC    $dst.hi,$dst.hi" %}
9202   ins_encode %{
9203     __ addl($dst$$Register,$dst$$Register);
9204     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9205   %}
9206   ins_pipe( ialu_reg_long );
9207 %}
9208 
9209 // Shift Left Long by 2
9210 instruct shlL_eReg_2(eRegL dst, immI_2 cnt, eFlagsReg cr) %{
9211   predicate(UseNewLongLShift);
9212   match(Set dst (LShiftL dst cnt));
9213   effect(KILL cr);
9214   ins_cost(100);
9215   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9216             "ADC    $dst.hi,$dst.hi\n\t" 
9217             "ADD    $dst.lo,$dst.lo\n\t"
9218             "ADC    $dst.hi,$dst.hi" %}
9219   ins_encode %{
9220     __ addl($dst$$Register,$dst$$Register);
9221     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9222     __ addl($dst$$Register,$dst$$Register);
9223     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9224   %}
9225   ins_pipe( ialu_reg_long );
9226 %}
9227 
9228 // Shift Left Long by 3
9229 instruct shlL_eReg_3(eRegL dst, immI_3 cnt, eFlagsReg cr) %{
9230   predicate(UseNewLongLShift);
9231   match(Set dst (LShiftL dst cnt));
9232   effect(KILL cr);
9233   ins_cost(100);
9234   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9235             "ADC    $dst.hi,$dst.hi\n\t" 
9236             "ADD    $dst.lo,$dst.lo\n\t"
9237             "ADC    $dst.hi,$dst.hi\n\t" 
9238             "ADD    $dst.lo,$dst.lo\n\t"
9239             "ADC    $dst.hi,$dst.hi" %}
9240   ins_encode %{
9241     __ addl($dst$$Register,$dst$$Register);
9242     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9243     __ addl($dst$$Register,$dst$$Register);
9244     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9245     __ addl($dst$$Register,$dst$$Register);
9246     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9247   %}
9248   ins_pipe( ialu_reg_long );
9249 %}
9250 
9251 // Shift Left Long by 1-31
9252 instruct shlL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9253   match(Set dst (LShiftL dst cnt));
9254   effect(KILL cr);
9255   ins_cost(200);
9256   format %{ "SHLD   $dst.hi,$dst.lo,$cnt\n\t"
9257             "SHL    $dst.lo,$cnt" %}
9258   opcode(0xC1, 0x4, 0xA4);  /* 0F/A4, then C1 /4 ib */
9259   ins_encode( move_long_small_shift(dst,cnt) );
9260   ins_pipe( ialu_reg_long );
9261 %}
9262 
9263 // Shift Left Long by 32-63
9264 instruct shlL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9265   match(Set dst (LShiftL dst cnt));
9266   effect(KILL cr);
9267   ins_cost(300);
9268   format %{ "MOV    $dst.hi,$dst.lo\n"
9269           "\tSHL    $dst.hi,$cnt-32\n"
9270           "\tXOR    $dst.lo,$dst.lo" %}
9271   opcode(0xC1, 0x4);  /* C1 /4 ib */
9272   ins_encode( move_long_big_shift_clr(dst,cnt) );
9273   ins_pipe( ialu_reg_long );
9274 %}
9275 
9276 // Shift Left Long by variable
9277 instruct salL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9278   match(Set dst (LShiftL dst shift));
9279   effect(KILL cr);
9280   ins_cost(500+200);
9281   size(17);
9282   format %{ "TEST   $shift,32\n\t"
9283             "JEQ,s  small\n\t"
9284             "MOV    $dst.hi,$dst.lo\n\t"
9285             "XOR    $dst.lo,$dst.lo\n"
9286     "small:\tSHLD   $dst.hi,$dst.lo,$shift\n\t"
9287             "SHL    $dst.lo,$shift" %}
9288   ins_encode( shift_left_long( dst, shift ) );
9289   ins_pipe( pipe_slow );
9290 %}
9291 
9292 // Shift Right Long by 1-31
9293 instruct shrL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9294   match(Set dst (URShiftL dst cnt));
9295   effect(KILL cr);
9296   ins_cost(200);
9297   format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
9298             "SHR    $dst.hi,$cnt" %}
9299   opcode(0xC1, 0x5, 0xAC);  /* 0F/AC, then C1 /5 ib */
9300   ins_encode( move_long_small_shift(dst,cnt) );
9301   ins_pipe( ialu_reg_long );
9302 %}
9303 
9304 // Shift Right Long by 32-63
9305 instruct shrL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9306   match(Set dst (URShiftL dst cnt));
9307   effect(KILL cr);
9308   ins_cost(300);
9309   format %{ "MOV    $dst.lo,$dst.hi\n"
9310           "\tSHR    $dst.lo,$cnt-32\n"
9311           "\tXOR    $dst.hi,$dst.hi" %}
9312   opcode(0xC1, 0x5);  /* C1 /5 ib */
9313   ins_encode( move_long_big_shift_clr(dst,cnt) );
9314   ins_pipe( ialu_reg_long );
9315 %}
9316 
9317 // Shift Right Long by variable
9318 instruct shrL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9319   match(Set dst (URShiftL dst shift));
9320   effect(KILL cr);
9321   ins_cost(600);
9322   size(17);
9323   format %{ "TEST   $shift,32\n\t"
9324             "JEQ,s  small\n\t"
9325             "MOV    $dst.lo,$dst.hi\n\t"
9326             "XOR    $dst.hi,$dst.hi\n"
9327     "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
9328             "SHR    $dst.hi,$shift" %}
9329   ins_encode( shift_right_long( dst, shift ) );
9330   ins_pipe( pipe_slow );
9331 %}
9332 
9333 // Shift Right Long by 1-31
9334 instruct sarL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9335   match(Set dst (RShiftL dst cnt));
9336   effect(KILL cr);
9337   ins_cost(200);
9338   format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
9339             "SAR    $dst.hi,$cnt" %}
9340   opcode(0xC1, 0x7, 0xAC);  /* 0F/AC, then C1 /7 ib */
9341   ins_encode( move_long_small_shift(dst,cnt) );
9342   ins_pipe( ialu_reg_long );
9343 %}
9344 
9345 // Shift Right Long by 32-63
9346 instruct sarL_eReg_32_63( eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9347   match(Set dst (RShiftL dst cnt));
9348   effect(KILL cr);
9349   ins_cost(300);
9350   format %{ "MOV    $dst.lo,$dst.hi\n"
9351           "\tSAR    $dst.lo,$cnt-32\n"
9352           "\tSAR    $dst.hi,31" %}
9353   opcode(0xC1, 0x7);  /* C1 /7 ib */
9354   ins_encode( move_long_big_shift_sign(dst,cnt) );
9355   ins_pipe( ialu_reg_long );
9356 %}
9357 
9358 // Shift Right arithmetic Long by variable
9359 instruct sarL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9360   match(Set dst (RShiftL dst shift));
9361   effect(KILL cr);
9362   ins_cost(600);
9363   size(18);
9364   format %{ "TEST   $shift,32\n\t"
9365             "JEQ,s  small\n\t"
9366             "MOV    $dst.lo,$dst.hi\n\t"
9367             "SAR    $dst.hi,31\n"
9368     "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
9369             "SAR    $dst.hi,$shift" %}
9370   ins_encode( shift_right_arith_long( dst, shift ) );
9371   ins_pipe( pipe_slow );
9372 %}
9373 
9374 
9375 //----------Double Instructions------------------------------------------------
9376 // Double Math
9377 
9378 // Compare & branch
9379 
9380 // P6 version of float compare, sets condition codes in EFLAGS
9381 instruct cmpDPR_cc_P6(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
9382   predicate(VM_Version::supports_cmov() && UseSSE <=1);
9383   match(Set cr (CmpD src1 src2));
9384   effect(KILL rax);
9385   ins_cost(150);
9386   format %{ "FLD    $src1\n\t"
9387             "FUCOMIP ST,$src2  // P6 instruction\n\t"
9388             "JNP    exit\n\t"
9389             "MOV    ah,1       // saw a NaN, set CF\n\t"
9390             "SAHF\n"
9391      "exit:\tNOP               // avoid branch to branch" %}
9392   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
9393   ins_encode( Push_Reg_DPR(src1),
9394               OpcP, RegOpc(src2),
9395               cmpF_P6_fixup );
9396   ins_pipe( pipe_slow );
9397 %}
9398 
9399 instruct cmpDPR_cc_P6CF(eFlagsRegUCF cr, regDPR src1, regDPR src2) %{
9400   predicate(VM_Version::supports_cmov() && UseSSE <=1);
9401   match(Set cr (CmpD src1 src2));
9402   ins_cost(150);
9403   format %{ "FLD    $src1\n\t"
9404             "FUCOMIP ST,$src2  // P6 instruction" %}
9405   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
9406   ins_encode( Push_Reg_DPR(src1),
9407               OpcP, RegOpc(src2));
9408   ins_pipe( pipe_slow );
9409 %}
9410 
9411 // Compare & branch
9412 instruct cmpDPR_cc(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
9413   predicate(UseSSE<=1);
9414   match(Set cr (CmpD src1 src2));
9415   effect(KILL rax);
9416   ins_cost(200);
9417   format %{ "FLD    $src1\n\t"
9418             "FCOMp  $src2\n\t"
9419             "FNSTSW AX\n\t"
9420             "TEST   AX,0x400\n\t"
9421             "JZ,s   flags\n\t"
9422             "MOV    AH,1\t# unordered treat as LT\n"
9423     "flags:\tSAHF" %}
9424   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
9425   ins_encode( Push_Reg_DPR(src1),
9426               OpcP, RegOpc(src2),
9427               fpu_flags);
9428   ins_pipe( pipe_slow );
9429 %}
9430 
9431 // Compare vs zero into -1,0,1
9432 instruct cmpDPR_0(rRegI dst, regDPR src1, immDPR0 zero, eAXRegI rax, eFlagsReg cr) %{
9433   predicate(UseSSE<=1);
9434   match(Set dst (CmpD3 src1 zero));
9435   effect(KILL cr, KILL rax);
9436   ins_cost(280);
9437   format %{ "FTSTD  $dst,$src1" %}
9438   opcode(0xE4, 0xD9);
9439   ins_encode( Push_Reg_DPR(src1),
9440               OpcS, OpcP, PopFPU,
9441               CmpF_Result(dst));
9442   ins_pipe( pipe_slow );
9443 %}
9444 
9445 // Compare into -1,0,1
9446 instruct cmpDPR_reg(rRegI dst, regDPR src1, regDPR src2, eAXRegI rax, eFlagsReg cr) %{
9447   predicate(UseSSE<=1);
9448   match(Set dst (CmpD3 src1 src2));
9449   effect(KILL cr, KILL rax);
9450   ins_cost(300);
9451   format %{ "FCMPD  $dst,$src1,$src2" %}
9452   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
9453   ins_encode( Push_Reg_DPR(src1),
9454               OpcP, RegOpc(src2),
9455               CmpF_Result(dst));
9456   ins_pipe( pipe_slow );
9457 %}
9458 
9459 // float compare and set condition codes in EFLAGS by XMM regs
9460 instruct cmpD_cc(eFlagsRegU cr, regD src1, regD src2) %{
9461   predicate(UseSSE>=2);
9462   match(Set cr (CmpD src1 src2));
9463   ins_cost(145);
9464   format %{ "UCOMISD $src1,$src2\n\t"
9465             "JNP,s   exit\n\t"
9466             "PUSHF\t# saw NaN, set CF\n\t"
9467             "AND     [rsp], #0xffffff2b\n\t"
9468             "POPF\n"
9469     "exit:" %}
9470   ins_encode %{
9471     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9472     emit_cmpfp_fixup(_masm);
9473   %}
9474   ins_pipe( pipe_slow );
9475 %}
9476 
9477 instruct cmpD_ccCF(eFlagsRegUCF cr, regD src1, regD src2) %{
9478   predicate(UseSSE>=2);
9479   match(Set cr (CmpD src1 src2));
9480   ins_cost(100);
9481   format %{ "UCOMISD $src1,$src2" %}
9482   ins_encode %{
9483     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9484   %}
9485   ins_pipe( pipe_slow );
9486 %}
9487 
9488 // float compare and set condition codes in EFLAGS by XMM regs
9489 instruct cmpD_ccmem(eFlagsRegU cr, regD src1, memory src2) %{
9490   predicate(UseSSE>=2);
9491   match(Set cr (CmpD src1 (LoadD src2)));
9492   ins_cost(145);
9493   format %{ "UCOMISD $src1,$src2\n\t"
9494             "JNP,s   exit\n\t"
9495             "PUSHF\t# saw NaN, set CF\n\t"
9496             "AND     [rsp], #0xffffff2b\n\t"
9497             "POPF\n"
9498     "exit:" %}
9499   ins_encode %{
9500     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9501     emit_cmpfp_fixup(_masm);
9502   %}
9503   ins_pipe( pipe_slow );
9504 %}
9505 
9506 instruct cmpD_ccmemCF(eFlagsRegUCF cr, regD src1, memory src2) %{
9507   predicate(UseSSE>=2);
9508   match(Set cr (CmpD src1 (LoadD src2)));
9509   ins_cost(100);
9510   format %{ "UCOMISD $src1,$src2" %}
9511   ins_encode %{
9512     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9513   %}
9514   ins_pipe( pipe_slow );
9515 %}
9516 
9517 // Compare into -1,0,1 in XMM
9518 instruct cmpD_reg(xRegI dst, regD src1, regD src2, eFlagsReg cr) %{
9519   predicate(UseSSE>=2);
9520   match(Set dst (CmpD3 src1 src2));
9521   effect(KILL cr);
9522   ins_cost(255);
9523   format %{ "UCOMISD $src1, $src2\n\t"
9524             "MOV     $dst, #-1\n\t"
9525             "JP,s    done\n\t"
9526             "JB,s    done\n\t"
9527             "SETNE   $dst\n\t"
9528             "MOVZB   $dst, $dst\n"
9529     "done:" %}
9530   ins_encode %{
9531     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9532     emit_cmpfp3(_masm, $dst$$Register);
9533   %}
9534   ins_pipe( pipe_slow );
9535 %}
9536 
9537 // Compare into -1,0,1 in XMM and memory
9538 instruct cmpD_regmem(xRegI dst, regD src1, memory src2, eFlagsReg cr) %{
9539   predicate(UseSSE>=2);
9540   match(Set dst (CmpD3 src1 (LoadD src2)));
9541   effect(KILL cr);
9542   ins_cost(275);
9543   format %{ "UCOMISD $src1, $src2\n\t"
9544             "MOV     $dst, #-1\n\t"
9545             "JP,s    done\n\t"
9546             "JB,s    done\n\t"
9547             "SETNE   $dst\n\t"
9548             "MOVZB   $dst, $dst\n"
9549     "done:" %}
9550   ins_encode %{
9551     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9552     emit_cmpfp3(_masm, $dst$$Register);
9553   %}
9554   ins_pipe( pipe_slow );
9555 %}
9556 
9557 
9558 instruct subDPR_reg(regDPR dst, regDPR src) %{
9559   predicate (UseSSE <=1);
9560   match(Set dst (SubD dst src));
9561 
9562   format %{ "FLD    $src\n\t"
9563             "DSUBp  $dst,ST" %}
9564   opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
9565   ins_cost(150);
9566   ins_encode( Push_Reg_DPR(src),
9567               OpcP, RegOpc(dst) );
9568   ins_pipe( fpu_reg_reg );
9569 %}
9570 
9571 instruct subDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9572   predicate (UseSSE <=1);
9573   match(Set dst (RoundDouble (SubD src1 src2)));
9574   ins_cost(250);
9575 
9576   format %{ "FLD    $src2\n\t"
9577             "DSUB   ST,$src1\n\t"
9578             "FSTP_D $dst\t# D-round" %}
9579   opcode(0xD8, 0x5);
9580   ins_encode( Push_Reg_DPR(src2),
9581               OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
9582   ins_pipe( fpu_mem_reg_reg );
9583 %}
9584 
9585 
9586 instruct subDPR_reg_mem(regDPR dst, memory src) %{
9587   predicate (UseSSE <=1);
9588   match(Set dst (SubD dst (LoadD src)));
9589   ins_cost(150);
9590 
9591   format %{ "FLD    $src\n\t"
9592             "DSUBp  $dst,ST" %}
9593   opcode(0xDE, 0x5, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
9594   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9595               OpcP, RegOpc(dst) );
9596   ins_pipe( fpu_reg_mem );
9597 %}
9598 
9599 instruct absDPR_reg(regDPR1 dst, regDPR1 src) %{
9600   predicate (UseSSE<=1);
9601   match(Set dst (AbsD src));
9602   ins_cost(100);
9603   format %{ "FABS" %}
9604   opcode(0xE1, 0xD9);
9605   ins_encode( OpcS, OpcP );
9606   ins_pipe( fpu_reg_reg );
9607 %}
9608 
9609 instruct negDPR_reg(regDPR1 dst, regDPR1 src) %{
9610   predicate(UseSSE<=1);
9611   match(Set dst (NegD src));
9612   ins_cost(100);
9613   format %{ "FCHS" %}
9614   opcode(0xE0, 0xD9);
9615   ins_encode( OpcS, OpcP );
9616   ins_pipe( fpu_reg_reg );
9617 %}
9618 
9619 instruct addDPR_reg(regDPR dst, regDPR src) %{
9620   predicate(UseSSE<=1);
9621   match(Set dst (AddD dst src));
9622   format %{ "FLD    $src\n\t"
9623             "DADD   $dst,ST" %}
9624   size(4);
9625   ins_cost(150);
9626   opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
9627   ins_encode( Push_Reg_DPR(src),
9628               OpcP, RegOpc(dst) );
9629   ins_pipe( fpu_reg_reg );
9630 %}
9631 
9632 
9633 instruct addDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9634   predicate(UseSSE<=1);
9635   match(Set dst (RoundDouble (AddD src1 src2)));
9636   ins_cost(250);
9637 
9638   format %{ "FLD    $src2\n\t"
9639             "DADD   ST,$src1\n\t"
9640             "FSTP_D $dst\t# D-round" %}
9641   opcode(0xD8, 0x0); /* D8 C0+i or D8 /0*/
9642   ins_encode( Push_Reg_DPR(src2),
9643               OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
9644   ins_pipe( fpu_mem_reg_reg );
9645 %}
9646 
9647 
9648 instruct addDPR_reg_mem(regDPR dst, memory src) %{
9649   predicate(UseSSE<=1);
9650   match(Set dst (AddD dst (LoadD src)));
9651   ins_cost(150);
9652 
9653   format %{ "FLD    $src\n\t"
9654             "DADDp  $dst,ST" %}
9655   opcode(0xDE, 0x0, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
9656   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9657               OpcP, RegOpc(dst) );
9658   ins_pipe( fpu_reg_mem );
9659 %}
9660 
9661 // add-to-memory
9662 instruct addDPR_mem_reg(memory dst, regDPR src) %{
9663   predicate(UseSSE<=1);
9664   match(Set dst (StoreD dst (RoundDouble (AddD (LoadD dst) src))));
9665   ins_cost(150);
9666 
9667   format %{ "FLD_D  $dst\n\t"
9668             "DADD   ST,$src\n\t"
9669             "FST_D  $dst" %}
9670   opcode(0xDD, 0x0);
9671   ins_encode( Opcode(0xDD), RMopc_Mem(0x00,dst),
9672               Opcode(0xD8), RegOpc(src),
9673               set_instruction_start,
9674               Opcode(0xDD), RMopc_Mem(0x03,dst) );
9675   ins_pipe( fpu_reg_mem );
9676 %}
9677 
9678 instruct addDPR_reg_imm1(regDPR dst, immDPR1 con) %{
9679   predicate(UseSSE<=1);
9680   match(Set dst (AddD dst con));
9681   ins_cost(125);
9682   format %{ "FLD1\n\t"
9683             "DADDp  $dst,ST" %}
9684   ins_encode %{
9685     __ fld1();
9686     __ faddp($dst$$reg);
9687   %}
9688   ins_pipe(fpu_reg);
9689 %}
9690 
9691 instruct addDPR_reg_imm(regDPR dst, immDPR con) %{
9692   predicate(UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
9693   match(Set dst (AddD dst con));
9694   ins_cost(200);
9695   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9696             "DADDp  $dst,ST" %}
9697   ins_encode %{
9698     __ fld_d($constantaddress($con));
9699     __ faddp($dst$$reg);
9700   %}
9701   ins_pipe(fpu_reg_mem);
9702 %}
9703 
9704 instruct addDPR_reg_imm_round(stackSlotD dst, regDPR src, immDPR con) %{
9705   predicate(UseSSE<=1 && _kids[0]->_kids[1]->_leaf->getd() != 0.0 && _kids[0]->_kids[1]->_leaf->getd() != 1.0 );
9706   match(Set dst (RoundDouble (AddD src con)));
9707   ins_cost(200);
9708   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9709             "DADD   ST,$src\n\t"
9710             "FSTP_D $dst\t# D-round" %}
9711   ins_encode %{
9712     __ fld_d($constantaddress($con));
9713     __ fadd($src$$reg);
9714     __ fstp_d(Address(rsp, $dst$$disp));
9715   %}
9716   ins_pipe(fpu_mem_reg_con);
9717 %}
9718 
9719 instruct mulDPR_reg(regDPR dst, regDPR src) %{
9720   predicate(UseSSE<=1);
9721   match(Set dst (MulD dst src));
9722   format %{ "FLD    $src\n\t"
9723             "DMULp  $dst,ST" %}
9724   opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
9725   ins_cost(150);
9726   ins_encode( Push_Reg_DPR(src),
9727               OpcP, RegOpc(dst) );
9728   ins_pipe( fpu_reg_reg );
9729 %}
9730 
9731 // Strict FP instruction biases argument before multiply then
9732 // biases result to avoid double rounding of subnormals.
9733 //
9734 // scale arg1 by multiplying arg1 by 2^(-15360)
9735 // load arg2
9736 // multiply scaled arg1 by arg2
9737 // rescale product by 2^(15360)
9738 //
9739 instruct strictfp_mulDPR_reg(regDPR1 dst, regnotDPR1 src) %{
9740   predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
9741   match(Set dst (MulD dst src));
9742   ins_cost(1);   // Select this instruction for all strict FP double multiplies
9743 
9744   format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
9745             "DMULp  $dst,ST\n\t"
9746             "FLD    $src\n\t"
9747             "DMULp  $dst,ST\n\t"
9748             "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
9749             "DMULp  $dst,ST\n\t" %}
9750   opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
9751   ins_encode( strictfp_bias1(dst),
9752               Push_Reg_DPR(src),
9753               OpcP, RegOpc(dst),
9754               strictfp_bias2(dst) );
9755   ins_pipe( fpu_reg_reg );
9756 %}
9757 
9758 instruct mulDPR_reg_imm(regDPR dst, immDPR con) %{
9759   predicate( UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
9760   match(Set dst (MulD dst con));
9761   ins_cost(200);
9762   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9763             "DMULp  $dst,ST" %}
9764   ins_encode %{
9765     __ fld_d($constantaddress($con));
9766     __ fmulp($dst$$reg);
9767   %}
9768   ins_pipe(fpu_reg_mem);
9769 %}
9770 
9771 
9772 instruct mulDPR_reg_mem(regDPR dst, memory src) %{
9773   predicate( UseSSE<=1 );
9774   match(Set dst (MulD dst (LoadD src)));
9775   ins_cost(200);
9776   format %{ "FLD_D  $src\n\t"
9777             "DMULp  $dst,ST" %}
9778   opcode(0xDE, 0x1, 0xDD); /* DE C8+i or DE /1*/  /* LoadD  DD /0 */
9779   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9780               OpcP, RegOpc(dst) );
9781   ins_pipe( fpu_reg_mem );
9782 %}
9783 
9784 //
9785 // Cisc-alternate to reg-reg multiply
9786 instruct mulDPR_reg_mem_cisc(regDPR dst, regDPR src, memory mem) %{
9787   predicate( UseSSE<=1 );
9788   match(Set dst (MulD src (LoadD mem)));
9789   ins_cost(250);
9790   format %{ "FLD_D  $mem\n\t"
9791             "DMUL   ST,$src\n\t"
9792             "FSTP_D $dst" %}
9793   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadD D9 /0 */
9794   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem),
9795               OpcReg_FPR(src),
9796               Pop_Reg_DPR(dst) );
9797   ins_pipe( fpu_reg_reg_mem );
9798 %}
9799 
9800 
9801 // MACRO3 -- addDPR a mulDPR
9802 // This instruction is a '2-address' instruction in that the result goes
9803 // back to src2.  This eliminates a move from the macro; possibly the
9804 // register allocator will have to add it back (and maybe not).
9805 instruct addDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
9806   predicate( UseSSE<=1 );
9807   match(Set src2 (AddD (MulD src0 src1) src2));
9808   format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
9809             "DMUL   ST,$src1\n\t"
9810             "DADDp  $src2,ST" %}
9811   ins_cost(250);
9812   opcode(0xDD); /* LoadD DD /0 */
9813   ins_encode( Push_Reg_FPR(src0),
9814               FMul_ST_reg(src1),
9815               FAddP_reg_ST(src2) );
9816   ins_pipe( fpu_reg_reg_reg );
9817 %}
9818 
9819 
9820 // MACRO3 -- subDPR a mulDPR
9821 instruct subDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
9822   predicate( UseSSE<=1 );
9823   match(Set src2 (SubD (MulD src0 src1) src2));
9824   format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
9825             "DMUL   ST,$src1\n\t"
9826             "DSUBRp $src2,ST" %}
9827   ins_cost(250);
9828   ins_encode( Push_Reg_FPR(src0),
9829               FMul_ST_reg(src1),
9830               Opcode(0xDE), Opc_plus(0xE0,src2));
9831   ins_pipe( fpu_reg_reg_reg );
9832 %}
9833 
9834 
9835 instruct divDPR_reg(regDPR dst, regDPR src) %{
9836   predicate( UseSSE<=1 );
9837   match(Set dst (DivD dst src));
9838 
9839   format %{ "FLD    $src\n\t"
9840             "FDIVp  $dst,ST" %}
9841   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
9842   ins_cost(150);
9843   ins_encode( Push_Reg_DPR(src),
9844               OpcP, RegOpc(dst) );
9845   ins_pipe( fpu_reg_reg );
9846 %}
9847 
9848 // Strict FP instruction biases argument before division then
9849 // biases result, to avoid double rounding of subnormals.
9850 //
9851 // scale dividend by multiplying dividend by 2^(-15360)
9852 // load divisor
9853 // divide scaled dividend by divisor
9854 // rescale quotient by 2^(15360)
9855 //
9856 instruct strictfp_divDPR_reg(regDPR1 dst, regnotDPR1 src) %{
9857   predicate (UseSSE<=1);
9858   match(Set dst (DivD dst src));
9859   predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
9860   ins_cost(01);
9861 
9862   format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
9863             "DMULp  $dst,ST\n\t"
9864             "FLD    $src\n\t"
9865             "FDIVp  $dst,ST\n\t"
9866             "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
9867             "DMULp  $dst,ST\n\t" %}
9868   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
9869   ins_encode( strictfp_bias1(dst),
9870               Push_Reg_DPR(src),
9871               OpcP, RegOpc(dst),
9872               strictfp_bias2(dst) );
9873   ins_pipe( fpu_reg_reg );
9874 %}
9875 
9876 instruct divDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9877   predicate( UseSSE<=1 && !(Compile::current()->has_method() && Compile::current()->method()->is_strict()) );
9878   match(Set dst (RoundDouble (DivD src1 src2)));
9879 
9880   format %{ "FLD    $src1\n\t"
9881             "FDIV   ST,$src2\n\t"
9882             "FSTP_D $dst\t# D-round" %}
9883   opcode(0xD8, 0x6); /* D8 F0+i or D8 /6 */
9884   ins_encode( Push_Reg_DPR(src1),
9885               OpcP, RegOpc(src2), Pop_Mem_DPR(dst) );
9886   ins_pipe( fpu_mem_reg_reg );
9887 %}
9888 
9889 
9890 instruct modDPR_reg(regDPR dst, regDPR src, eAXRegI rax, eFlagsReg cr) %{
9891   predicate(UseSSE<=1);
9892   match(Set dst (ModD dst src));
9893   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
9894 
9895   format %{ "DMOD   $dst,$src" %}
9896   ins_cost(250);
9897   ins_encode(Push_Reg_Mod_DPR(dst, src),
9898               emitModDPR(),
9899               Push_Result_Mod_DPR(src),
9900               Pop_Reg_DPR(dst));
9901   ins_pipe( pipe_slow );
9902 %}
9903 
9904 instruct modD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eFlagsReg cr) %{
9905   predicate(UseSSE>=2);
9906   match(Set dst (ModD src0 src1));
9907   effect(KILL rax, KILL cr);
9908 
9909   format %{ "SUB    ESP,8\t # DMOD\n"
9910           "\tMOVSD  [ESP+0],$src1\n"
9911           "\tFLD_D  [ESP+0]\n"
9912           "\tMOVSD  [ESP+0],$src0\n"
9913           "\tFLD_D  [ESP+0]\n"
9914      "loop:\tFPREM\n"
9915           "\tFWAIT\n"
9916           "\tFNSTSW AX\n"
9917           "\tSAHF\n"
9918           "\tJP     loop\n"
9919           "\tFSTP_D [ESP+0]\n"
9920           "\tMOVSD  $dst,[ESP+0]\n"
9921           "\tADD    ESP,8\n"
9922           "\tFSTP   ST0\t # Restore FPU Stack"
9923     %}
9924   ins_cost(250);
9925   ins_encode( Push_ModD_encoding(src0, src1), emitModDPR(), Push_ResultD(dst), PopFPU);
9926   ins_pipe( pipe_slow );
9927 %}
9928 
9929 instruct sinDPR_reg(regDPR1 dst, regDPR1 src) %{
9930   predicate (UseSSE<=1);
9931   match(Set dst (SinD src));
9932   ins_cost(1800);
9933   format %{ "DSIN   $dst" %}
9934   opcode(0xD9, 0xFE);
9935   ins_encode( OpcP, OpcS );
9936   ins_pipe( pipe_slow );
9937 %}
9938 
9939 instruct sinD_reg(regD dst, eFlagsReg cr) %{
9940   predicate (UseSSE>=2);
9941   match(Set dst (SinD dst));
9942   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
9943   ins_cost(1800);
9944   format %{ "DSIN   $dst" %}
9945   opcode(0xD9, 0xFE);
9946   ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
9947   ins_pipe( pipe_slow );
9948 %}
9949 
9950 instruct cosDPR_reg(regDPR1 dst, regDPR1 src) %{
9951   predicate (UseSSE<=1);
9952   match(Set dst (CosD src));
9953   ins_cost(1800);
9954   format %{ "DCOS   $dst" %}
9955   opcode(0xD9, 0xFF);
9956   ins_encode( OpcP, OpcS );
9957   ins_pipe( pipe_slow );
9958 %}
9959 
9960 instruct cosD_reg(regD dst, eFlagsReg cr) %{
9961   predicate (UseSSE>=2);
9962   match(Set dst (CosD dst));
9963   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
9964   ins_cost(1800);
9965   format %{ "DCOS   $dst" %}
9966   opcode(0xD9, 0xFF);
9967   ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
9968   ins_pipe( pipe_slow );
9969 %}
9970 
9971 instruct tanDPR_reg(regDPR1 dst, regDPR1 src) %{
9972   predicate (UseSSE<=1);
9973   match(Set dst(TanD src));
9974   format %{ "DTAN   $dst" %}
9975   ins_encode( Opcode(0xD9), Opcode(0xF2),    // fptan
9976               Opcode(0xDD), Opcode(0xD8));   // fstp st
9977   ins_pipe( pipe_slow );
9978 %}
9979 
9980 instruct tanD_reg(regD dst, eFlagsReg cr) %{
9981   predicate (UseSSE>=2);
9982   match(Set dst(TanD dst));
9983   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
9984   format %{ "DTAN   $dst" %}
9985   ins_encode( Push_SrcD(dst),
9986               Opcode(0xD9), Opcode(0xF2),    // fptan
9987               Opcode(0xDD), Opcode(0xD8),   // fstp st
9988               Push_ResultD(dst) );
9989   ins_pipe( pipe_slow );
9990 %}
9991 
9992 instruct atanDPR_reg(regDPR dst, regDPR src) %{
9993   predicate (UseSSE<=1);
9994   match(Set dst(AtanD dst src));
9995   format %{ "DATA   $dst,$src" %}
9996   opcode(0xD9, 0xF3);
9997   ins_encode( Push_Reg_DPR(src),
9998               OpcP, OpcS, RegOpc(dst) );
9999   ins_pipe( pipe_slow );
10000 %}
10001 
10002 instruct atanD_reg(regD dst, regD src, eFlagsReg cr) %{
10003   predicate (UseSSE>=2);
10004   match(Set dst(AtanD dst src));
10005   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
10006   format %{ "DATA   $dst,$src" %}
10007   opcode(0xD9, 0xF3);
10008   ins_encode( Push_SrcD(src),
10009               OpcP, OpcS, Push_ResultD(dst) );
10010   ins_pipe( pipe_slow );
10011 %}
10012 
10013 instruct sqrtDPR_reg(regDPR dst, regDPR src) %{
10014   predicate (UseSSE<=1);
10015   match(Set dst (SqrtD src));
10016   format %{ "DSQRT  $dst,$src" %}
10017   opcode(0xFA, 0xD9);
10018   ins_encode( Push_Reg_DPR(src),
10019               OpcS, OpcP, Pop_Reg_DPR(dst) );
10020   ins_pipe( pipe_slow );
10021 %}
10022 
10023 instruct powDPR_reg(regDPR X, regDPR1 Y, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10024   predicate (UseSSE<=1);
10025   match(Set Y (PowD X Y));  // Raise X to the Yth power
10026   effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
10027   format %{ "fast_pow $X $Y -> $Y  // KILL $rax, $rcx, $rdx" %}
10028   ins_encode %{
10029     __ subptr(rsp, 8);
10030     __ fld_s($X$$reg - 1);
10031     __ fast_pow();
10032     __ addptr(rsp, 8);
10033   %}
10034   ins_pipe( pipe_slow );
10035 %}
10036 
10037 instruct powD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10038   predicate (UseSSE>=2);
10039   match(Set dst (PowD src0 src1));  // Raise src0 to the src1'th power
10040   effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
10041   format %{ "fast_pow $src0 $src1 -> $dst  // KILL $rax, $rcx, $rdx" %}
10042   ins_encode %{
10043     __ subptr(rsp, 8);
10044     __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
10045     __ fld_d(Address(rsp, 0));
10046     __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
10047     __ fld_d(Address(rsp, 0));
10048     __ fast_pow();
10049     __ fstp_d(Address(rsp, 0));
10050     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
10051     __ addptr(rsp, 8);
10052   %}
10053   ins_pipe( pipe_slow );
10054 %}
10055 
10056 
10057 instruct expDPR_reg(regDPR1 dpr1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10058   predicate (UseSSE<=1);
10059   match(Set dpr1 (ExpD dpr1));
10060   effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
10061   format %{ "fast_exp $dpr1 -> $dpr1  // KILL $rax, $rcx, $rdx" %}
10062   ins_encode %{
10063     __ fast_exp();
10064   %}
10065   ins_pipe( pipe_slow );
10066 %}
10067 
10068 instruct expD_reg(regD dst, regD src, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10069   predicate (UseSSE>=2);
10070   match(Set dst (ExpD src));
10071   effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
10072   format %{ "fast_exp $dst -> $src  // KILL $rax, $rcx, $rdx" %}
10073   ins_encode %{
10074     __ subptr(rsp, 8);
10075     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
10076     __ fld_d(Address(rsp, 0));
10077     __ fast_exp();
10078     __ fstp_d(Address(rsp, 0));
10079     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
10080     __ addptr(rsp, 8);
10081   %}
10082   ins_pipe( pipe_slow );
10083 %}
10084 
10085 instruct log10DPR_reg(regDPR1 dst, regDPR1 src) %{
10086   predicate (UseSSE<=1);
10087   // The source Double operand on FPU stack
10088   match(Set dst (Log10D src));
10089   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
10090   // fxch         ; swap ST(0) with ST(1)
10091   // fyl2x        ; compute log_10(2) * log_2(x)
10092   format %{ "FLDLG2 \t\t\t#Log10\n\t"
10093             "FXCH   \n\t"
10094             "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
10095          %}
10096   ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
10097               Opcode(0xD9), Opcode(0xC9),   // fxch
10098               Opcode(0xD9), Opcode(0xF1));  // fyl2x
10099 
10100   ins_pipe( pipe_slow );
10101 %}
10102 
10103 instruct log10D_reg(regD dst, regD src, eFlagsReg cr) %{
10104   predicate (UseSSE>=2);
10105   effect(KILL cr);
10106   match(Set dst (Log10D src));
10107   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
10108   // fyl2x        ; compute log_10(2) * log_2(x)
10109   format %{ "FLDLG2 \t\t\t#Log10\n\t"
10110             "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
10111          %}
10112   ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
10113               Push_SrcD(src),
10114               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10115               Push_ResultD(dst));
10116 
10117   ins_pipe( pipe_slow );
10118 %}
10119 
10120 instruct logDPR_reg(regDPR1 dst, regDPR1 src) %{
10121   predicate (UseSSE<=1);
10122   // The source Double operand on FPU stack
10123   match(Set dst (LogD src));
10124   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
10125   // fxch         ; swap ST(0) with ST(1)
10126   // fyl2x        ; compute log_e(2) * log_2(x)
10127   format %{ "FLDLN2 \t\t\t#Log_e\n\t"
10128             "FXCH   \n\t"
10129             "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
10130          %}
10131   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
10132               Opcode(0xD9), Opcode(0xC9),   // fxch
10133               Opcode(0xD9), Opcode(0xF1));  // fyl2x
10134 
10135   ins_pipe( pipe_slow );
10136 %}
10137 
10138 instruct logD_reg(regD dst, regD src, eFlagsReg cr) %{
10139   predicate (UseSSE>=2);
10140   effect(KILL cr);
10141   // The source and result Double operands in XMM registers
10142   match(Set dst (LogD src));
10143   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
10144   // fyl2x        ; compute log_e(2) * log_2(x)
10145   format %{ "FLDLN2 \t\t\t#Log_e\n\t"
10146             "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
10147          %}
10148   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
10149               Push_SrcD(src),
10150               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10151               Push_ResultD(dst));
10152   ins_pipe( pipe_slow );
10153 %}
10154 
10155 //-------------Float Instructions-------------------------------
10156 // Float Math
10157 
10158 // Code for float compare:
10159 //     fcompp();
10160 //     fwait(); fnstsw_ax();
10161 //     sahf();
10162 //     movl(dst, unordered_result);
10163 //     jcc(Assembler::parity, exit);
10164 //     movl(dst, less_result);
10165 //     jcc(Assembler::below, exit);
10166 //     movl(dst, equal_result);
10167 //     jcc(Assembler::equal, exit);
10168 //     movl(dst, greater_result);
10169 //   exit:
10170 
10171 // P6 version of float compare, sets condition codes in EFLAGS
10172 instruct cmpFPR_cc_P6(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
10173   predicate(VM_Version::supports_cmov() && UseSSE == 0);
10174   match(Set cr (CmpF src1 src2));
10175   effect(KILL rax);
10176   ins_cost(150);
10177   format %{ "FLD    $src1\n\t"
10178             "FUCOMIP ST,$src2  // P6 instruction\n\t"
10179             "JNP    exit\n\t"
10180             "MOV    ah,1       // saw a NaN, set CF (treat as LT)\n\t"
10181             "SAHF\n"
10182      "exit:\tNOP               // avoid branch to branch" %}
10183   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
10184   ins_encode( Push_Reg_DPR(src1),
10185               OpcP, RegOpc(src2),
10186               cmpF_P6_fixup );
10187   ins_pipe( pipe_slow );
10188 %}
10189 
10190 instruct cmpFPR_cc_P6CF(eFlagsRegUCF cr, regFPR src1, regFPR src2) %{
10191   predicate(VM_Version::supports_cmov() && UseSSE == 0);
10192   match(Set cr (CmpF src1 src2));
10193   ins_cost(100);
10194   format %{ "FLD    $src1\n\t"
10195             "FUCOMIP ST,$src2  // P6 instruction" %}
10196   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
10197   ins_encode( Push_Reg_DPR(src1),
10198               OpcP, RegOpc(src2));
10199   ins_pipe( pipe_slow );
10200 %}
10201 
10202 
10203 // Compare & branch
10204 instruct cmpFPR_cc(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
10205   predicate(UseSSE == 0);
10206   match(Set cr (CmpF src1 src2));
10207   effect(KILL rax);
10208   ins_cost(200);
10209   format %{ "FLD    $src1\n\t"
10210             "FCOMp  $src2\n\t"
10211             "FNSTSW AX\n\t"
10212             "TEST   AX,0x400\n\t"
10213             "JZ,s   flags\n\t"
10214             "MOV    AH,1\t# unordered treat as LT\n"
10215     "flags:\tSAHF" %}
10216   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
10217   ins_encode( Push_Reg_DPR(src1),
10218               OpcP, RegOpc(src2),
10219               fpu_flags);
10220   ins_pipe( pipe_slow );
10221 %}
10222 
10223 // Compare vs zero into -1,0,1
10224 instruct cmpFPR_0(rRegI dst, regFPR src1, immFPR0 zero, eAXRegI rax, eFlagsReg cr) %{
10225   predicate(UseSSE == 0);
10226   match(Set dst (CmpF3 src1 zero));
10227   effect(KILL cr, KILL rax);
10228   ins_cost(280);
10229   format %{ "FTSTF  $dst,$src1" %}
10230   opcode(0xE4, 0xD9);
10231   ins_encode( Push_Reg_DPR(src1),
10232               OpcS, OpcP, PopFPU,
10233               CmpF_Result(dst));
10234   ins_pipe( pipe_slow );
10235 %}
10236 
10237 // Compare into -1,0,1
10238 instruct cmpFPR_reg(rRegI dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
10239   predicate(UseSSE == 0);
10240   match(Set dst (CmpF3 src1 src2));
10241   effect(KILL cr, KILL rax);
10242   ins_cost(300);
10243   format %{ "FCMPF  $dst,$src1,$src2" %}
10244   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
10245   ins_encode( Push_Reg_DPR(src1),
10246               OpcP, RegOpc(src2),
10247               CmpF_Result(dst));
10248   ins_pipe( pipe_slow );
10249 %}
10250 
10251 // float compare and set condition codes in EFLAGS by XMM regs
10252 instruct cmpF_cc(eFlagsRegU cr, regF src1, regF src2) %{
10253   predicate(UseSSE>=1);
10254   match(Set cr (CmpF src1 src2));
10255   ins_cost(145);
10256   format %{ "UCOMISS $src1,$src2\n\t"
10257             "JNP,s   exit\n\t"
10258             "PUSHF\t# saw NaN, set CF\n\t"
10259             "AND     [rsp], #0xffffff2b\n\t"
10260             "POPF\n"
10261     "exit:" %}
10262   ins_encode %{
10263     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10264     emit_cmpfp_fixup(_masm);
10265   %}
10266   ins_pipe( pipe_slow );
10267 %}
10268 
10269 instruct cmpF_ccCF(eFlagsRegUCF cr, regF src1, regF src2) %{
10270   predicate(UseSSE>=1);
10271   match(Set cr (CmpF src1 src2));
10272   ins_cost(100);
10273   format %{ "UCOMISS $src1,$src2" %}
10274   ins_encode %{
10275     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10276   %}
10277   ins_pipe( pipe_slow );
10278 %}
10279 
10280 // float compare and set condition codes in EFLAGS by XMM regs
10281 instruct cmpF_ccmem(eFlagsRegU cr, regF src1, memory src2) %{
10282   predicate(UseSSE>=1);
10283   match(Set cr (CmpF src1 (LoadF src2)));
10284   ins_cost(165);
10285   format %{ "UCOMISS $src1,$src2\n\t"
10286             "JNP,s   exit\n\t"
10287             "PUSHF\t# saw NaN, set CF\n\t"
10288             "AND     [rsp], #0xffffff2b\n\t"
10289             "POPF\n"
10290     "exit:" %}
10291   ins_encode %{
10292     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10293     emit_cmpfp_fixup(_masm);
10294   %}
10295   ins_pipe( pipe_slow );
10296 %}
10297 
10298 instruct cmpF_ccmemCF(eFlagsRegUCF cr, regF src1, memory src2) %{
10299   predicate(UseSSE>=1);
10300   match(Set cr (CmpF src1 (LoadF src2)));
10301   ins_cost(100);
10302   format %{ "UCOMISS $src1,$src2" %}
10303   ins_encode %{
10304     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10305   %}
10306   ins_pipe( pipe_slow );
10307 %}
10308 
10309 // Compare into -1,0,1 in XMM
10310 instruct cmpF_reg(xRegI dst, regF src1, regF src2, eFlagsReg cr) %{
10311   predicate(UseSSE>=1);
10312   match(Set dst (CmpF3 src1 src2));
10313   effect(KILL cr);
10314   ins_cost(255);
10315   format %{ "UCOMISS $src1, $src2\n\t"
10316             "MOV     $dst, #-1\n\t"
10317             "JP,s    done\n\t"
10318             "JB,s    done\n\t"
10319             "SETNE   $dst\n\t"
10320             "MOVZB   $dst, $dst\n"
10321     "done:" %}
10322   ins_encode %{
10323     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10324     emit_cmpfp3(_masm, $dst$$Register);
10325   %}
10326   ins_pipe( pipe_slow );
10327 %}
10328 
10329 // Compare into -1,0,1 in XMM and memory
10330 instruct cmpF_regmem(xRegI dst, regF src1, memory src2, eFlagsReg cr) %{
10331   predicate(UseSSE>=1);
10332   match(Set dst (CmpF3 src1 (LoadF src2)));
10333   effect(KILL cr);
10334   ins_cost(275);
10335   format %{ "UCOMISS $src1, $src2\n\t"
10336             "MOV     $dst, #-1\n\t"
10337             "JP,s    done\n\t"
10338             "JB,s    done\n\t"
10339             "SETNE   $dst\n\t"
10340             "MOVZB   $dst, $dst\n"
10341     "done:" %}
10342   ins_encode %{
10343     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10344     emit_cmpfp3(_masm, $dst$$Register);
10345   %}
10346   ins_pipe( pipe_slow );
10347 %}
10348 
10349 // Spill to obtain 24-bit precision
10350 instruct subFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10351   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10352   match(Set dst (SubF src1 src2));
10353 
10354   format %{ "FSUB   $dst,$src1 - $src2" %}
10355   opcode(0xD8, 0x4); /* D8 E0+i or D8 /4 mod==0x3 ;; result in TOS */
10356   ins_encode( Push_Reg_FPR(src1),
10357               OpcReg_FPR(src2),
10358               Pop_Mem_FPR(dst) );
10359   ins_pipe( fpu_mem_reg_reg );
10360 %}
10361 //
10362 // This instruction does not round to 24-bits
10363 instruct subFPR_reg(regFPR dst, regFPR src) %{
10364   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10365   match(Set dst (SubF dst src));
10366 
10367   format %{ "FSUB   $dst,$src" %}
10368   opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
10369   ins_encode( Push_Reg_FPR(src),
10370               OpcP, RegOpc(dst) );
10371   ins_pipe( fpu_reg_reg );
10372 %}
10373 
10374 // Spill to obtain 24-bit precision
10375 instruct addFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10376   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10377   match(Set dst (AddF src1 src2));
10378 
10379   format %{ "FADD   $dst,$src1,$src2" %}
10380   opcode(0xD8, 0x0); /* D8 C0+i */
10381   ins_encode( Push_Reg_FPR(src2),
10382               OpcReg_FPR(src1),
10383               Pop_Mem_FPR(dst) );
10384   ins_pipe( fpu_mem_reg_reg );
10385 %}
10386 //
10387 // This instruction does not round to 24-bits
10388 instruct addFPR_reg(regFPR dst, regFPR src) %{
10389   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10390   match(Set dst (AddF dst src));
10391 
10392   format %{ "FLD    $src\n\t"
10393             "FADDp  $dst,ST" %}
10394   opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
10395   ins_encode( Push_Reg_FPR(src),
10396               OpcP, RegOpc(dst) );
10397   ins_pipe( fpu_reg_reg );
10398 %}
10399 
10400 instruct absFPR_reg(regFPR1 dst, regFPR1 src) %{
10401   predicate(UseSSE==0);
10402   match(Set dst (AbsF src));
10403   ins_cost(100);
10404   format %{ "FABS" %}
10405   opcode(0xE1, 0xD9);
10406   ins_encode( OpcS, OpcP );
10407   ins_pipe( fpu_reg_reg );
10408 %}
10409 
10410 instruct negFPR_reg(regFPR1 dst, regFPR1 src) %{
10411   predicate(UseSSE==0);
10412   match(Set dst (NegF src));
10413   ins_cost(100);
10414   format %{ "FCHS" %}
10415   opcode(0xE0, 0xD9);
10416   ins_encode( OpcS, OpcP );
10417   ins_pipe( fpu_reg_reg );
10418 %}
10419 
10420 // Cisc-alternate to addFPR_reg
10421 // Spill to obtain 24-bit precision
10422 instruct addFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
10423   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10424   match(Set dst (AddF src1 (LoadF src2)));
10425 
10426   format %{ "FLD    $src2\n\t"
10427             "FADD   ST,$src1\n\t"
10428             "FSTP_S $dst" %}
10429   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10430   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10431               OpcReg_FPR(src1),
10432               Pop_Mem_FPR(dst) );
10433   ins_pipe( fpu_mem_reg_mem );
10434 %}
10435 //
10436 // Cisc-alternate to addFPR_reg
10437 // This instruction does not round to 24-bits
10438 instruct addFPR_reg_mem(regFPR dst, memory src) %{
10439   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10440   match(Set dst (AddF dst (LoadF src)));
10441 
10442   format %{ "FADD   $dst,$src" %}
10443   opcode(0xDE, 0x0, 0xD9); /* DE C0+i or DE /0*/  /* LoadF  D9 /0 */
10444   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
10445               OpcP, RegOpc(dst) );
10446   ins_pipe( fpu_reg_mem );
10447 %}
10448 
10449 // // Following two instructions for _222_mpegaudio
10450 // Spill to obtain 24-bit precision
10451 instruct addFPR24_mem_reg(stackSlotF dst, regFPR src2, memory src1 ) %{
10452   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10453   match(Set dst (AddF src1 src2));
10454 
10455   format %{ "FADD   $dst,$src1,$src2" %}
10456   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10457   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src1),
10458               OpcReg_FPR(src2),
10459               Pop_Mem_FPR(dst) );
10460   ins_pipe( fpu_mem_reg_mem );
10461 %}
10462 
10463 // Cisc-spill variant
10464 // Spill to obtain 24-bit precision
10465 instruct addFPR24_mem_cisc(stackSlotF dst, memory src1, memory src2) %{
10466   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10467   match(Set dst (AddF src1 (LoadF src2)));
10468 
10469   format %{ "FADD   $dst,$src1,$src2 cisc" %}
10470   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10471   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10472               set_instruction_start,
10473               OpcP, RMopc_Mem(secondary,src1),
10474               Pop_Mem_FPR(dst) );
10475   ins_pipe( fpu_mem_mem_mem );
10476 %}
10477 
10478 // Spill to obtain 24-bit precision
10479 instruct addFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
10480   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10481   match(Set dst (AddF src1 src2));
10482 
10483   format %{ "FADD   $dst,$src1,$src2" %}
10484   opcode(0xD8, 0x0, 0xD9); /* D8 /0 */  /* LoadF  D9 /0 */
10485   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10486               set_instruction_start,
10487               OpcP, RMopc_Mem(secondary,src1),
10488               Pop_Mem_FPR(dst) );
10489   ins_pipe( fpu_mem_mem_mem );
10490 %}
10491 
10492 
10493 // Spill to obtain 24-bit precision
10494 instruct addFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
10495   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10496   match(Set dst (AddF src con));
10497   format %{ "FLD    $src\n\t"
10498             "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10499             "FSTP_S $dst"  %}
10500   ins_encode %{
10501     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10502     __ fadd_s($constantaddress($con));
10503     __ fstp_s(Address(rsp, $dst$$disp));
10504   %}
10505   ins_pipe(fpu_mem_reg_con);
10506 %}
10507 //
10508 // This instruction does not round to 24-bits
10509 instruct addFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
10510   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10511   match(Set dst (AddF src con));
10512   format %{ "FLD    $src\n\t"
10513             "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10514             "FSTP   $dst"  %}
10515   ins_encode %{
10516     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10517     __ fadd_s($constantaddress($con));
10518     __ fstp_d($dst$$reg);
10519   %}
10520   ins_pipe(fpu_reg_reg_con);
10521 %}
10522 
10523 // Spill to obtain 24-bit precision
10524 instruct mulFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10525   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10526   match(Set dst (MulF src1 src2));
10527 
10528   format %{ "FLD    $src1\n\t"
10529             "FMUL   $src2\n\t"
10530             "FSTP_S $dst"  %}
10531   opcode(0xD8, 0x1); /* D8 C8+i or D8 /1 ;; result in TOS */
10532   ins_encode( Push_Reg_FPR(src1),
10533               OpcReg_FPR(src2),
10534               Pop_Mem_FPR(dst) );
10535   ins_pipe( fpu_mem_reg_reg );
10536 %}
10537 //
10538 // This instruction does not round to 24-bits
10539 instruct mulFPR_reg(regFPR dst, regFPR src1, regFPR src2) %{
10540   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10541   match(Set dst (MulF src1 src2));
10542 
10543   format %{ "FLD    $src1\n\t"
10544             "FMUL   $src2\n\t"
10545             "FSTP_S $dst"  %}
10546   opcode(0xD8, 0x1); /* D8 C8+i */
10547   ins_encode( Push_Reg_FPR(src2),
10548               OpcReg_FPR(src1),
10549               Pop_Reg_FPR(dst) );
10550   ins_pipe( fpu_reg_reg_reg );
10551 %}
10552 
10553 
10554 // Spill to obtain 24-bit precision
10555 // Cisc-alternate to reg-reg multiply
10556 instruct mulFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
10557   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10558   match(Set dst (MulF src1 (LoadF src2)));
10559 
10560   format %{ "FLD_S  $src2\n\t"
10561             "FMUL   $src1\n\t"
10562             "FSTP_S $dst"  %}
10563   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or DE /1*/  /* LoadF D9 /0 */
10564   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10565               OpcReg_FPR(src1),
10566               Pop_Mem_FPR(dst) );
10567   ins_pipe( fpu_mem_reg_mem );
10568 %}
10569 //
10570 // This instruction does not round to 24-bits
10571 // Cisc-alternate to reg-reg multiply
10572 instruct mulFPR_reg_mem(regFPR dst, regFPR src1, memory src2) %{
10573   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10574   match(Set dst (MulF src1 (LoadF src2)));
10575 
10576   format %{ "FMUL   $dst,$src1,$src2" %}
10577   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadF D9 /0 */
10578   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10579               OpcReg_FPR(src1),
10580               Pop_Reg_FPR(dst) );
10581   ins_pipe( fpu_reg_reg_mem );
10582 %}
10583 
10584 // Spill to obtain 24-bit precision
10585 instruct mulFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
10586   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10587   match(Set dst (MulF src1 src2));
10588 
10589   format %{ "FMUL   $dst,$src1,$src2" %}
10590   opcode(0xD8, 0x1, 0xD9); /* D8 /1 */  /* LoadF D9 /0 */
10591   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10592               set_instruction_start,
10593               OpcP, RMopc_Mem(secondary,src1),
10594               Pop_Mem_FPR(dst) );
10595   ins_pipe( fpu_mem_mem_mem );
10596 %}
10597 
10598 // Spill to obtain 24-bit precision
10599 instruct mulFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
10600   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10601   match(Set dst (MulF src con));
10602 
10603   format %{ "FLD    $src\n\t"
10604             "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10605             "FSTP_S $dst"  %}
10606   ins_encode %{
10607     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10608     __ fmul_s($constantaddress($con));
10609     __ fstp_s(Address(rsp, $dst$$disp));
10610   %}
10611   ins_pipe(fpu_mem_reg_con);
10612 %}
10613 //
10614 // This instruction does not round to 24-bits
10615 instruct mulFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
10616   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10617   match(Set dst (MulF src con));
10618 
10619   format %{ "FLD    $src\n\t"
10620             "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10621             "FSTP   $dst"  %}
10622   ins_encode %{
10623     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10624     __ fmul_s($constantaddress($con));
10625     __ fstp_d($dst$$reg);
10626   %}
10627   ins_pipe(fpu_reg_reg_con);
10628 %}
10629 
10630 
10631 //
10632 // MACRO1 -- subsume unshared load into mulFPR
10633 // This instruction does not round to 24-bits
10634 instruct mulFPR_reg_load1(regFPR dst, regFPR src, memory mem1 ) %{
10635   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10636   match(Set dst (MulF (LoadF mem1) src));
10637 
10638   format %{ "FLD    $mem1    ===MACRO1===\n\t"
10639             "FMUL   ST,$src\n\t"
10640             "FSTP   $dst" %}
10641   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or D8 /1 */  /* LoadF D9 /0 */
10642   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem1),
10643               OpcReg_FPR(src),
10644               Pop_Reg_FPR(dst) );
10645   ins_pipe( fpu_reg_reg_mem );
10646 %}
10647 //
10648 // MACRO2 -- addFPR a mulFPR which subsumed an unshared load
10649 // This instruction does not round to 24-bits
10650 instruct addFPR_mulFPR_reg_load1(regFPR dst, memory mem1, regFPR src1, regFPR src2) %{
10651   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10652   match(Set dst (AddF (MulF (LoadF mem1) src1) src2));
10653   ins_cost(95);
10654 
10655   format %{ "FLD    $mem1     ===MACRO2===\n\t"
10656             "FMUL   ST,$src1  subsume mulFPR left load\n\t"
10657             "FADD   ST,$src2\n\t"
10658             "FSTP   $dst" %}
10659   opcode(0xD9); /* LoadF D9 /0 */
10660   ins_encode( OpcP, RMopc_Mem(0x00,mem1),
10661               FMul_ST_reg(src1),
10662               FAdd_ST_reg(src2),
10663               Pop_Reg_FPR(dst) );
10664   ins_pipe( fpu_reg_mem_reg_reg );
10665 %}
10666 
10667 // MACRO3 -- addFPR a mulFPR
10668 // This instruction does not round to 24-bits.  It is a '2-address'
10669 // instruction in that the result goes back to src2.  This eliminates
10670 // a move from the macro; possibly the register allocator will have
10671 // to add it back (and maybe not).
10672 instruct addFPR_mulFPR_reg(regFPR src2, regFPR src1, regFPR src0) %{
10673   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10674   match(Set src2 (AddF (MulF src0 src1) src2));
10675 
10676   format %{ "FLD    $src0     ===MACRO3===\n\t"
10677             "FMUL   ST,$src1\n\t"
10678             "FADDP  $src2,ST" %}
10679   opcode(0xD9); /* LoadF D9 /0 */
10680   ins_encode( Push_Reg_FPR(src0),
10681               FMul_ST_reg(src1),
10682               FAddP_reg_ST(src2) );
10683   ins_pipe( fpu_reg_reg_reg );
10684 %}
10685 
10686 // MACRO4 -- divFPR subFPR
10687 // This instruction does not round to 24-bits
10688 instruct subFPR_divFPR_reg(regFPR dst, regFPR src1, regFPR src2, regFPR src3) %{
10689   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10690   match(Set dst (DivF (SubF src2 src1) src3));
10691 
10692   format %{ "FLD    $src2   ===MACRO4===\n\t"
10693             "FSUB   ST,$src1\n\t"
10694             "FDIV   ST,$src3\n\t"
10695             "FSTP  $dst" %}
10696   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
10697   ins_encode( Push_Reg_FPR(src2),
10698               subFPR_divFPR_encode(src1,src3),
10699               Pop_Reg_FPR(dst) );
10700   ins_pipe( fpu_reg_reg_reg_reg );
10701 %}
10702 
10703 // Spill to obtain 24-bit precision
10704 instruct divFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10705   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10706   match(Set dst (DivF src1 src2));
10707 
10708   format %{ "FDIV   $dst,$src1,$src2" %}
10709   opcode(0xD8, 0x6); /* D8 F0+i or DE /6*/
10710   ins_encode( Push_Reg_FPR(src1),
10711               OpcReg_FPR(src2),
10712               Pop_Mem_FPR(dst) );
10713   ins_pipe( fpu_mem_reg_reg );
10714 %}
10715 //
10716 // This instruction does not round to 24-bits
10717 instruct divFPR_reg(regFPR dst, regFPR src) %{
10718   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10719   match(Set dst (DivF dst src));
10720 
10721   format %{ "FDIV   $dst,$src" %}
10722   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
10723   ins_encode( Push_Reg_FPR(src),
10724               OpcP, RegOpc(dst) );
10725   ins_pipe( fpu_reg_reg );
10726 %}
10727 
10728 
10729 // Spill to obtain 24-bit precision
10730 instruct modFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
10731   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
10732   match(Set dst (ModF src1 src2));
10733   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
10734 
10735   format %{ "FMOD   $dst,$src1,$src2" %}
10736   ins_encode( Push_Reg_Mod_DPR(src1, src2),
10737               emitModDPR(),
10738               Push_Result_Mod_DPR(src2),
10739               Pop_Mem_FPR(dst));
10740   ins_pipe( pipe_slow );
10741 %}
10742 //
10743 // This instruction does not round to 24-bits
10744 instruct modFPR_reg(regFPR dst, regFPR src, eAXRegI rax, eFlagsReg cr) %{
10745   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
10746   match(Set dst (ModF dst src));
10747   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
10748 
10749   format %{ "FMOD   $dst,$src" %}
10750   ins_encode(Push_Reg_Mod_DPR(dst, src),
10751               emitModDPR(),
10752               Push_Result_Mod_DPR(src),
10753               Pop_Reg_FPR(dst));
10754   ins_pipe( pipe_slow );
10755 %}
10756 
10757 instruct modF_reg(regF dst, regF src0, regF src1, eAXRegI rax, eFlagsReg cr) %{
10758   predicate(UseSSE>=1);
10759   match(Set dst (ModF src0 src1));
10760   effect(KILL rax, KILL cr);
10761   format %{ "SUB    ESP,4\t # FMOD\n"
10762           "\tMOVSS  [ESP+0],$src1\n"
10763           "\tFLD_S  [ESP+0]\n"
10764           "\tMOVSS  [ESP+0],$src0\n"
10765           "\tFLD_S  [ESP+0]\n"
10766      "loop:\tFPREM\n"
10767           "\tFWAIT\n"
10768           "\tFNSTSW AX\n"
10769           "\tSAHF\n"
10770           "\tJP     loop\n"
10771           "\tFSTP_S [ESP+0]\n"
10772           "\tMOVSS  $dst,[ESP+0]\n"
10773           "\tADD    ESP,4\n"
10774           "\tFSTP   ST0\t # Restore FPU Stack"
10775     %}
10776   ins_cost(250);
10777   ins_encode( Push_ModF_encoding(src0, src1), emitModDPR(), Push_ResultF(dst,0x4), PopFPU);
10778   ins_pipe( pipe_slow );
10779 %}
10780 
10781 
10782 //----------Arithmetic Conversion Instructions---------------------------------
10783 // The conversions operations are all Alpha sorted.  Please keep it that way!
10784 
10785 instruct roundFloat_mem_reg(stackSlotF dst, regFPR src) %{
10786   predicate(UseSSE==0);
10787   match(Set dst (RoundFloat src));
10788   ins_cost(125);
10789   format %{ "FST_S  $dst,$src\t# F-round" %}
10790   ins_encode( Pop_Mem_Reg_FPR(dst, src) );
10791   ins_pipe( fpu_mem_reg );
10792 %}
10793 
10794 instruct roundDouble_mem_reg(stackSlotD dst, regDPR src) %{
10795   predicate(UseSSE<=1);
10796   match(Set dst (RoundDouble src));
10797   ins_cost(125);
10798   format %{ "FST_D  $dst,$src\t# D-round" %}
10799   ins_encode( Pop_Mem_Reg_DPR(dst, src) );
10800   ins_pipe( fpu_mem_reg );
10801 %}
10802 
10803 // Force rounding to 24-bit precision and 6-bit exponent
10804 instruct convDPR2FPR_reg(stackSlotF dst, regDPR src) %{
10805   predicate(UseSSE==0);
10806   match(Set dst (ConvD2F src));
10807   format %{ "FST_S  $dst,$src\t# F-round" %}
10808   expand %{
10809     roundFloat_mem_reg(dst,src);
10810   %}
10811 %}
10812 
10813 // Force rounding to 24-bit precision and 6-bit exponent
10814 instruct convDPR2F_reg(regF dst, regDPR src, eFlagsReg cr) %{
10815   predicate(UseSSE==1);
10816   match(Set dst (ConvD2F src));
10817   effect( KILL cr );
10818   format %{ "SUB    ESP,4\n\t"
10819             "FST_S  [ESP],$src\t# F-round\n\t"
10820             "MOVSS  $dst,[ESP]\n\t"
10821             "ADD ESP,4" %}
10822   ins_encode %{
10823     __ subptr(rsp, 4);
10824     if ($src$$reg != FPR1L_enc) {
10825       __ fld_s($src$$reg-1);
10826       __ fstp_s(Address(rsp, 0));
10827     } else {
10828       __ fst_s(Address(rsp, 0));
10829     }
10830     __ movflt($dst$$XMMRegister, Address(rsp, 0));
10831     __ addptr(rsp, 4);
10832   %}
10833   ins_pipe( pipe_slow );
10834 %}
10835 
10836 // Force rounding double precision to single precision
10837 instruct convD2F_reg(regF dst, regD src) %{
10838   predicate(UseSSE>=2);
10839   match(Set dst (ConvD2F src));
10840   format %{ "CVTSD2SS $dst,$src\t# F-round" %}
10841   ins_encode %{
10842     __ cvtsd2ss ($dst$$XMMRegister, $src$$XMMRegister);
10843   %}
10844   ins_pipe( pipe_slow );
10845 %}
10846 
10847 instruct convFPR2DPR_reg_reg(regDPR dst, regFPR src) %{
10848   predicate(UseSSE==0);
10849   match(Set dst (ConvF2D src));
10850   format %{ "FST_S  $dst,$src\t# D-round" %}
10851   ins_encode( Pop_Reg_Reg_DPR(dst, src));
10852   ins_pipe( fpu_reg_reg );
10853 %}
10854 
10855 instruct convFPR2D_reg(stackSlotD dst, regFPR src) %{
10856   predicate(UseSSE==1);
10857   match(Set dst (ConvF2D src));
10858   format %{ "FST_D  $dst,$src\t# D-round" %}
10859   expand %{
10860     roundDouble_mem_reg(dst,src);
10861   %}
10862 %}
10863 
10864 instruct convF2DPR_reg(regDPR dst, regF src, eFlagsReg cr) %{
10865   predicate(UseSSE==1);
10866   match(Set dst (ConvF2D src));
10867   effect( KILL cr );
10868   format %{ "SUB    ESP,4\n\t"
10869             "MOVSS  [ESP] $src\n\t"
10870             "FLD_S  [ESP]\n\t"
10871             "ADD    ESP,4\n\t"
10872             "FSTP   $dst\t# D-round" %}
10873   ins_encode %{
10874     __ subptr(rsp, 4);
10875     __ movflt(Address(rsp, 0), $src$$XMMRegister);
10876     __ fld_s(Address(rsp, 0));
10877     __ addptr(rsp, 4);
10878     __ fstp_d($dst$$reg);
10879   %}
10880   ins_pipe( pipe_slow );
10881 %}
10882 
10883 instruct convF2D_reg(regD dst, regF src) %{
10884   predicate(UseSSE>=2);
10885   match(Set dst (ConvF2D src));
10886   format %{ "CVTSS2SD $dst,$src\t# D-round" %}
10887   ins_encode %{
10888     __ cvtss2sd ($dst$$XMMRegister, $src$$XMMRegister);
10889   %}
10890   ins_pipe( pipe_slow );
10891 %}
10892 
10893 // Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
10894 instruct convDPR2I_reg_reg( eAXRegI dst, eDXRegI tmp, regDPR src, eFlagsReg cr ) %{
10895   predicate(UseSSE<=1);
10896   match(Set dst (ConvD2I src));
10897   effect( KILL tmp, KILL cr );
10898   format %{ "FLD    $src\t# Convert double to int \n\t"
10899             "FLDCW  trunc mode\n\t"
10900             "SUB    ESP,4\n\t"
10901             "FISTp  [ESP + #0]\n\t"
10902             "FLDCW  std/24-bit mode\n\t"
10903             "POP    EAX\n\t"
10904             "CMP    EAX,0x80000000\n\t"
10905             "JNE,s  fast\n\t"
10906             "FLD_D  $src\n\t"
10907             "CALL   d2i_wrapper\n"
10908       "fast:" %}
10909   ins_encode( Push_Reg_DPR(src), DPR2I_encoding(src) );
10910   ins_pipe( pipe_slow );
10911 %}
10912 
10913 // Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
10914 instruct convD2I_reg_reg( eAXRegI dst, eDXRegI tmp, regD src, eFlagsReg cr ) %{
10915   predicate(UseSSE>=2);
10916   match(Set dst (ConvD2I src));
10917   effect( KILL tmp, KILL cr );
10918   format %{ "CVTTSD2SI $dst, $src\n\t"
10919             "CMP    $dst,0x80000000\n\t"
10920             "JNE,s  fast\n\t"
10921             "SUB    ESP, 8\n\t"
10922             "MOVSD  [ESP], $src\n\t"
10923             "FLD_D  [ESP]\n\t"
10924             "ADD    ESP, 8\n\t"
10925             "CALL   d2i_wrapper\n"
10926       "fast:" %}
10927   ins_encode %{
10928     Label fast;
10929     __ cvttsd2sil($dst$$Register, $src$$XMMRegister);
10930     __ cmpl($dst$$Register, 0x80000000);
10931     __ jccb(Assembler::notEqual, fast);
10932     __ subptr(rsp, 8);
10933     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
10934     __ fld_d(Address(rsp, 0));
10935     __ addptr(rsp, 8);
10936     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
10937     __ bind(fast);
10938   %}
10939   ins_pipe( pipe_slow );
10940 %}
10941 
10942 instruct convDPR2L_reg_reg( eADXRegL dst, regDPR src, eFlagsReg cr ) %{
10943   predicate(UseSSE<=1);
10944   match(Set dst (ConvD2L src));
10945   effect( KILL cr );
10946   format %{ "FLD    $src\t# Convert double to long\n\t"
10947             "FLDCW  trunc mode\n\t"
10948             "SUB    ESP,8\n\t"
10949             "FISTp  [ESP + #0]\n\t"
10950             "FLDCW  std/24-bit mode\n\t"
10951             "POP    EAX\n\t"
10952             "POP    EDX\n\t"
10953             "CMP    EDX,0x80000000\n\t"
10954             "JNE,s  fast\n\t"
10955             "TEST   EAX,EAX\n\t"
10956             "JNE,s  fast\n\t"
10957             "FLD    $src\n\t"
10958             "CALL   d2l_wrapper\n"
10959       "fast:" %}
10960   ins_encode( Push_Reg_DPR(src),  DPR2L_encoding(src) );
10961   ins_pipe( pipe_slow );
10962 %}
10963 
10964 // XMM lacks a float/double->long conversion, so use the old FPU stack.
10965 instruct convD2L_reg_reg( eADXRegL dst, regD src, eFlagsReg cr ) %{
10966   predicate (UseSSE>=2);
10967   match(Set dst (ConvD2L src));
10968   effect( KILL cr );
10969   format %{ "SUB    ESP,8\t# Convert double to long\n\t"
10970             "MOVSD  [ESP],$src\n\t"
10971             "FLD_D  [ESP]\n\t"
10972             "FLDCW  trunc mode\n\t"
10973             "FISTp  [ESP + #0]\n\t"
10974             "FLDCW  std/24-bit mode\n\t"
10975             "POP    EAX\n\t"
10976             "POP    EDX\n\t"
10977             "CMP    EDX,0x80000000\n\t"
10978             "JNE,s  fast\n\t"
10979             "TEST   EAX,EAX\n\t"
10980             "JNE,s  fast\n\t"
10981             "SUB    ESP,8\n\t"
10982             "MOVSD  [ESP],$src\n\t"
10983             "FLD_D  [ESP]\n\t"
10984             "ADD    ESP,8\n\t"
10985             "CALL   d2l_wrapper\n"
10986       "fast:" %}
10987   ins_encode %{
10988     Label fast;
10989     __ subptr(rsp, 8);
10990     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
10991     __ fld_d(Address(rsp, 0));
10992     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
10993     __ fistp_d(Address(rsp, 0));
10994     // Restore the rounding mode, mask the exception
10995     if (Compile::current()->in_24_bit_fp_mode()) {
10996       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
10997     } else {
10998       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
10999     }
11000     // Load the converted long, adjust CPU stack
11001     __ pop(rax);
11002     __ pop(rdx);
11003     __ cmpl(rdx, 0x80000000);
11004     __ jccb(Assembler::notEqual, fast);
11005     __ testl(rax, rax);
11006     __ jccb(Assembler::notEqual, fast);
11007     __ subptr(rsp, 8);
11008     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
11009     __ fld_d(Address(rsp, 0));
11010     __ addptr(rsp, 8);
11011     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
11012     __ bind(fast);
11013   %}
11014   ins_pipe( pipe_slow );
11015 %}
11016 
11017 // Convert a double to an int.  Java semantics require we do complex
11018 // manglations in the corner cases.  So we set the rounding mode to
11019 // 'zero', store the darned double down as an int, and reset the
11020 // rounding mode to 'nearest'.  The hardware stores a flag value down
11021 // if we would overflow or converted a NAN; we check for this and
11022 // and go the slow path if needed.
11023 instruct convFPR2I_reg_reg(eAXRegI dst, eDXRegI tmp, regFPR src, eFlagsReg cr ) %{
11024   predicate(UseSSE==0);
11025   match(Set dst (ConvF2I src));
11026   effect( KILL tmp, KILL cr );
11027   format %{ "FLD    $src\t# Convert float to int \n\t"
11028             "FLDCW  trunc mode\n\t"
11029             "SUB    ESP,4\n\t"
11030             "FISTp  [ESP + #0]\n\t"
11031             "FLDCW  std/24-bit mode\n\t"
11032             "POP    EAX\n\t"
11033             "CMP    EAX,0x80000000\n\t"
11034             "JNE,s  fast\n\t"
11035             "FLD    $src\n\t"
11036             "CALL   d2i_wrapper\n"
11037       "fast:" %}
11038   // DPR2I_encoding works for FPR2I
11039   ins_encode( Push_Reg_FPR(src), DPR2I_encoding(src) );
11040   ins_pipe( pipe_slow );
11041 %}
11042 
11043 // Convert a float in xmm to an int reg.
11044 instruct convF2I_reg(eAXRegI dst, eDXRegI tmp, regF src, eFlagsReg cr ) %{
11045   predicate(UseSSE>=1);
11046   match(Set dst (ConvF2I src));
11047   effect( KILL tmp, KILL cr );
11048   format %{ "CVTTSS2SI $dst, $src\n\t"
11049             "CMP    $dst,0x80000000\n\t"
11050             "JNE,s  fast\n\t"
11051             "SUB    ESP, 4\n\t"
11052             "MOVSS  [ESP], $src\n\t"
11053             "FLD    [ESP]\n\t"
11054             "ADD    ESP, 4\n\t"
11055             "CALL   d2i_wrapper\n"
11056       "fast:" %}
11057   ins_encode %{
11058     Label fast;
11059     __ cvttss2sil($dst$$Register, $src$$XMMRegister);
11060     __ cmpl($dst$$Register, 0x80000000);
11061     __ jccb(Assembler::notEqual, fast);
11062     __ subptr(rsp, 4);
11063     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11064     __ fld_s(Address(rsp, 0));
11065     __ addptr(rsp, 4);
11066     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
11067     __ bind(fast);
11068   %}
11069   ins_pipe( pipe_slow );
11070 %}
11071 
11072 instruct convFPR2L_reg_reg( eADXRegL dst, regFPR src, eFlagsReg cr ) %{
11073   predicate(UseSSE==0);
11074   match(Set dst (ConvF2L src));
11075   effect( KILL cr );
11076   format %{ "FLD    $src\t# Convert float to long\n\t"
11077             "FLDCW  trunc mode\n\t"
11078             "SUB    ESP,8\n\t"
11079             "FISTp  [ESP + #0]\n\t"
11080             "FLDCW  std/24-bit mode\n\t"
11081             "POP    EAX\n\t"
11082             "POP    EDX\n\t"
11083             "CMP    EDX,0x80000000\n\t"
11084             "JNE,s  fast\n\t"
11085             "TEST   EAX,EAX\n\t"
11086             "JNE,s  fast\n\t"
11087             "FLD    $src\n\t"
11088             "CALL   d2l_wrapper\n"
11089       "fast:" %}
11090   // DPR2L_encoding works for FPR2L
11091   ins_encode( Push_Reg_FPR(src), DPR2L_encoding(src) );
11092   ins_pipe( pipe_slow );
11093 %}
11094 
11095 // XMM lacks a float/double->long conversion, so use the old FPU stack.
11096 instruct convF2L_reg_reg( eADXRegL dst, regF src, eFlagsReg cr ) %{
11097   predicate (UseSSE>=1);
11098   match(Set dst (ConvF2L src));
11099   effect( KILL cr );
11100   format %{ "SUB    ESP,8\t# Convert float to long\n\t"
11101             "MOVSS  [ESP],$src\n\t"
11102             "FLD_S  [ESP]\n\t"
11103             "FLDCW  trunc mode\n\t"
11104             "FISTp  [ESP + #0]\n\t"
11105             "FLDCW  std/24-bit mode\n\t"
11106             "POP    EAX\n\t"
11107             "POP    EDX\n\t"
11108             "CMP    EDX,0x80000000\n\t"
11109             "JNE,s  fast\n\t"
11110             "TEST   EAX,EAX\n\t"
11111             "JNE,s  fast\n\t"
11112             "SUB    ESP,4\t# Convert float to long\n\t"
11113             "MOVSS  [ESP],$src\n\t"
11114             "FLD_S  [ESP]\n\t"
11115             "ADD    ESP,4\n\t"
11116             "CALL   d2l_wrapper\n"
11117       "fast:" %}
11118   ins_encode %{
11119     Label fast;
11120     __ subptr(rsp, 8);
11121     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11122     __ fld_s(Address(rsp, 0));
11123     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
11124     __ fistp_d(Address(rsp, 0));
11125     // Restore the rounding mode, mask the exception
11126     if (Compile::current()->in_24_bit_fp_mode()) {
11127       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
11128     } else {
11129       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
11130     }
11131     // Load the converted long, adjust CPU stack
11132     __ pop(rax);
11133     __ pop(rdx);
11134     __ cmpl(rdx, 0x80000000);
11135     __ jccb(Assembler::notEqual, fast);
11136     __ testl(rax, rax);
11137     __ jccb(Assembler::notEqual, fast);
11138     __ subptr(rsp, 4);
11139     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11140     __ fld_s(Address(rsp, 0));
11141     __ addptr(rsp, 4);
11142     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
11143     __ bind(fast);
11144   %}
11145   ins_pipe( pipe_slow );
11146 %}
11147 
11148 instruct convI2DPR_reg(regDPR dst, stackSlotI src) %{
11149   predicate( UseSSE<=1 );
11150   match(Set dst (ConvI2D src));
11151   format %{ "FILD   $src\n\t"
11152             "FSTP   $dst" %}
11153   opcode(0xDB, 0x0);  /* DB /0 */
11154   ins_encode(Push_Mem_I(src), Pop_Reg_DPR(dst));
11155   ins_pipe( fpu_reg_mem );
11156 %}
11157 
11158 instruct convI2D_reg(regD dst, rRegI src) %{
11159   predicate( UseSSE>=2 && !UseXmmI2D );
11160   match(Set dst (ConvI2D src));
11161   format %{ "CVTSI2SD $dst,$src" %}
11162   ins_encode %{
11163     __ cvtsi2sdl ($dst$$XMMRegister, $src$$Register);
11164   %}
11165   ins_pipe( pipe_slow );
11166 %}
11167 
11168 instruct convI2D_mem(regD dst, memory mem) %{
11169   predicate( UseSSE>=2 );
11170   match(Set dst (ConvI2D (LoadI mem)));
11171   format %{ "CVTSI2SD $dst,$mem" %}
11172   ins_encode %{
11173     __ cvtsi2sdl ($dst$$XMMRegister, $mem$$Address);
11174   %}
11175   ins_pipe( pipe_slow );
11176 %}
11177 
11178 instruct convXI2D_reg(regD dst, rRegI src)
11179 %{
11180   predicate( UseSSE>=2 && UseXmmI2D );
11181   match(Set dst (ConvI2D src));
11182 
11183   format %{ "MOVD  $dst,$src\n\t"
11184             "CVTDQ2PD $dst,$dst\t# i2d" %}
11185   ins_encode %{
11186     __ movdl($dst$$XMMRegister, $src$$Register);
11187     __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11188   %}
11189   ins_pipe(pipe_slow); // XXX
11190 %}
11191 
11192 instruct convI2DPR_mem(regDPR dst, memory mem) %{
11193   predicate( UseSSE<=1 && !Compile::current()->select_24_bit_instr());
11194   match(Set dst (ConvI2D (LoadI mem)));
11195   format %{ "FILD   $mem\n\t"
11196             "FSTP   $dst" %}
11197   opcode(0xDB);      /* DB /0 */
11198   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11199               Pop_Reg_DPR(dst));
11200   ins_pipe( fpu_reg_mem );
11201 %}
11202 
11203 // Convert a byte to a float; no rounding step needed.
11204 instruct conv24I2FPR_reg(regFPR dst, stackSlotI src) %{
11205   predicate( UseSSE==0 && n->in(1)->Opcode() == Op_AndI && n->in(1)->in(2)->is_Con() && n->in(1)->in(2)->get_int() == 255 );
11206   match(Set dst (ConvI2F src));
11207   format %{ "FILD   $src\n\t"
11208             "FSTP   $dst" %}
11209 
11210   opcode(0xDB, 0x0);  /* DB /0 */
11211   ins_encode(Push_Mem_I(src), Pop_Reg_FPR(dst));
11212   ins_pipe( fpu_reg_mem );
11213 %}
11214 
11215 // In 24-bit mode, force exponent rounding by storing back out
11216 instruct convI2FPR_SSF(stackSlotF dst, stackSlotI src) %{
11217   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
11218   match(Set dst (ConvI2F src));
11219   ins_cost(200);
11220   format %{ "FILD   $src\n\t"
11221             "FSTP_S $dst" %}
11222   opcode(0xDB, 0x0);  /* DB /0 */
11223   ins_encode( Push_Mem_I(src),
11224               Pop_Mem_FPR(dst));
11225   ins_pipe( fpu_mem_mem );
11226 %}
11227 
11228 // In 24-bit mode, force exponent rounding by storing back out
11229 instruct convI2FPR_SSF_mem(stackSlotF dst, memory mem) %{
11230   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
11231   match(Set dst (ConvI2F (LoadI mem)));
11232   ins_cost(200);
11233   format %{ "FILD   $mem\n\t"
11234             "FSTP_S $dst" %}
11235   opcode(0xDB);  /* DB /0 */
11236   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11237               Pop_Mem_FPR(dst));
11238   ins_pipe( fpu_mem_mem );
11239 %}
11240 
11241 // This instruction does not round to 24-bits
11242 instruct convI2FPR_reg(regFPR dst, stackSlotI src) %{
11243   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
11244   match(Set dst (ConvI2F src));
11245   format %{ "FILD   $src\n\t"
11246             "FSTP   $dst" %}
11247   opcode(0xDB, 0x0);  /* DB /0 */
11248   ins_encode( Push_Mem_I(src),
11249               Pop_Reg_FPR(dst));
11250   ins_pipe( fpu_reg_mem );
11251 %}
11252 
11253 // This instruction does not round to 24-bits
11254 instruct convI2FPR_mem(regFPR dst, memory mem) %{
11255   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
11256   match(Set dst (ConvI2F (LoadI mem)));
11257   format %{ "FILD   $mem\n\t"
11258             "FSTP   $dst" %}
11259   opcode(0xDB);      /* DB /0 */
11260   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11261               Pop_Reg_FPR(dst));
11262   ins_pipe( fpu_reg_mem );
11263 %}
11264 
11265 // Convert an int to a float in xmm; no rounding step needed.
11266 instruct convI2F_reg(regF dst, rRegI src) %{
11267   predicate( UseSSE==1 || UseSSE>=2 && !UseXmmI2F );
11268   match(Set dst (ConvI2F src));
11269   format %{ "CVTSI2SS $dst, $src" %}
11270   ins_encode %{
11271     __ cvtsi2ssl ($dst$$XMMRegister, $src$$Register);
11272   %}
11273   ins_pipe( pipe_slow );
11274 %}
11275 
11276  instruct convXI2F_reg(regF dst, rRegI src)
11277 %{
11278   predicate( UseSSE>=2 && UseXmmI2F );
11279   match(Set dst (ConvI2F src));
11280 
11281   format %{ "MOVD  $dst,$src\n\t"
11282             "CVTDQ2PS $dst,$dst\t# i2f" %}
11283   ins_encode %{
11284     __ movdl($dst$$XMMRegister, $src$$Register);
11285     __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11286   %}
11287   ins_pipe(pipe_slow); // XXX
11288 %}
11289 
11290 instruct convI2L_reg( eRegL dst, rRegI src, eFlagsReg cr) %{
11291   match(Set dst (ConvI2L src));
11292   effect(KILL cr);
11293   ins_cost(375);
11294   format %{ "MOV    $dst.lo,$src\n\t"
11295             "MOV    $dst.hi,$src\n\t"
11296             "SAR    $dst.hi,31" %}
11297   ins_encode(convert_int_long(dst,src));
11298   ins_pipe( ialu_reg_reg_long );
11299 %}
11300 
11301 // Zero-extend convert int to long
11302 instruct convI2L_reg_zex(eRegL dst, rRegI src, immL_32bits mask, eFlagsReg flags ) %{
11303   match(Set dst (AndL (ConvI2L src) mask) );
11304   effect( KILL flags );
11305   ins_cost(250);
11306   format %{ "MOV    $dst.lo,$src\n\t"
11307             "XOR    $dst.hi,$dst.hi" %}
11308   opcode(0x33); // XOR
11309   ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
11310   ins_pipe( ialu_reg_reg_long );
11311 %}
11312 
11313 // Zero-extend long
11314 instruct zerox_long(eRegL dst, eRegL src, immL_32bits mask, eFlagsReg flags ) %{
11315   match(Set dst (AndL src mask) );
11316   effect( KILL flags );
11317   ins_cost(250);
11318   format %{ "MOV    $dst.lo,$src.lo\n\t"
11319             "XOR    $dst.hi,$dst.hi\n\t" %}
11320   opcode(0x33); // XOR
11321   ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
11322   ins_pipe( ialu_reg_reg_long );
11323 %}
11324 
11325 instruct convL2DPR_reg( stackSlotD dst, eRegL src, eFlagsReg cr) %{
11326   predicate (UseSSE<=1);
11327   match(Set dst (ConvL2D src));
11328   effect( KILL cr );
11329   format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
11330             "PUSH   $src.lo\n\t"
11331             "FILD   ST,[ESP + #0]\n\t"
11332             "ADD    ESP,8\n\t"
11333             "FSTP_D $dst\t# D-round" %}
11334   opcode(0xDF, 0x5);  /* DF /5 */
11335   ins_encode(convert_long_double(src), Pop_Mem_DPR(dst));
11336   ins_pipe( pipe_slow );
11337 %}
11338 
11339 instruct convL2D_reg( regD dst, eRegL src, eFlagsReg cr) %{
11340   predicate (UseSSE>=2);
11341   match(Set dst (ConvL2D src));
11342   effect( KILL cr );
11343   format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
11344             "PUSH   $src.lo\n\t"
11345             "FILD_D [ESP]\n\t"
11346             "FSTP_D [ESP]\n\t"
11347             "MOVSD  $dst,[ESP]\n\t"
11348             "ADD    ESP,8" %}
11349   opcode(0xDF, 0x5);  /* DF /5 */
11350   ins_encode(convert_long_double2(src), Push_ResultD(dst));
11351   ins_pipe( pipe_slow );
11352 %}
11353 
11354 instruct convL2F_reg( regF dst, eRegL src, eFlagsReg cr) %{
11355   predicate (UseSSE>=1);
11356   match(Set dst (ConvL2F src));
11357   effect( KILL cr );
11358   format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
11359             "PUSH   $src.lo\n\t"
11360             "FILD_D [ESP]\n\t"
11361             "FSTP_S [ESP]\n\t"
11362             "MOVSS  $dst,[ESP]\n\t"
11363             "ADD    ESP,8" %}
11364   opcode(0xDF, 0x5);  /* DF /5 */
11365   ins_encode(convert_long_double2(src), Push_ResultF(dst,0x8));
11366   ins_pipe( pipe_slow );
11367 %}
11368 
11369 instruct convL2FPR_reg( stackSlotF dst, eRegL src, eFlagsReg cr) %{
11370   match(Set dst (ConvL2F src));
11371   effect( KILL cr );
11372   format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
11373             "PUSH   $src.lo\n\t"
11374             "FILD   ST,[ESP + #0]\n\t"
11375             "ADD    ESP,8\n\t"
11376             "FSTP_S $dst\t# F-round" %}
11377   opcode(0xDF, 0x5);  /* DF /5 */
11378   ins_encode(convert_long_double(src), Pop_Mem_FPR(dst));
11379   ins_pipe( pipe_slow );
11380 %}
11381 
11382 instruct convL2I_reg( rRegI dst, eRegL src ) %{
11383   match(Set dst (ConvL2I src));
11384   effect( DEF dst, USE src );
11385   format %{ "MOV    $dst,$src.lo" %}
11386   ins_encode(enc_CopyL_Lo(dst,src));
11387   ins_pipe( ialu_reg_reg );
11388 %}
11389 
11390 
11391 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11392   match(Set dst (MoveF2I src));
11393   effect( DEF dst, USE src );
11394   ins_cost(100);
11395   format %{ "MOV    $dst,$src\t# MoveF2I_stack_reg" %}
11396   ins_encode %{
11397     __ movl($dst$$Register, Address(rsp, $src$$disp));
11398   %}
11399   ins_pipe( ialu_reg_mem );
11400 %}
11401 
11402 instruct MoveFPR2I_reg_stack(stackSlotI dst, regFPR src) %{
11403   predicate(UseSSE==0);
11404   match(Set dst (MoveF2I src));
11405   effect( DEF dst, USE src );
11406 
11407   ins_cost(125);
11408   format %{ "FST_S  $dst,$src\t# MoveF2I_reg_stack" %}
11409   ins_encode( Pop_Mem_Reg_FPR(dst, src) );
11410   ins_pipe( fpu_mem_reg );
11411 %}
11412 
11413 instruct MoveF2I_reg_stack_sse(stackSlotI dst, regF src) %{
11414   predicate(UseSSE>=1);
11415   match(Set dst (MoveF2I src));
11416   effect( DEF dst, USE src );
11417 
11418   ins_cost(95);
11419   format %{ "MOVSS  $dst,$src\t# MoveF2I_reg_stack_sse" %}
11420   ins_encode %{
11421     __ movflt(Address(rsp, $dst$$disp), $src$$XMMRegister);
11422   %}
11423   ins_pipe( pipe_slow );
11424 %}
11425 
11426 instruct MoveF2I_reg_reg_sse(rRegI dst, regF src) %{
11427   predicate(UseSSE>=2);
11428   match(Set dst (MoveF2I src));
11429   effect( DEF dst, USE src );
11430   ins_cost(85);
11431   format %{ "MOVD   $dst,$src\t# MoveF2I_reg_reg_sse" %}
11432   ins_encode %{
11433     __ movdl($dst$$Register, $src$$XMMRegister);
11434   %}
11435   ins_pipe( pipe_slow );
11436 %}
11437 
11438 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11439   match(Set dst (MoveI2F src));
11440   effect( DEF dst, USE src );
11441 
11442   ins_cost(100);
11443   format %{ "MOV    $dst,$src\t# MoveI2F_reg_stack" %}
11444   ins_encode %{
11445     __ movl(Address(rsp, $dst$$disp), $src$$Register);
11446   %}
11447   ins_pipe( ialu_mem_reg );
11448 %}
11449 
11450 
11451 instruct MoveI2FPR_stack_reg(regFPR dst, stackSlotI src) %{
11452   predicate(UseSSE==0);
11453   match(Set dst (MoveI2F src));
11454   effect(DEF dst, USE src);
11455 
11456   ins_cost(125);
11457   format %{ "FLD_S  $src\n\t"
11458             "FSTP   $dst\t# MoveI2F_stack_reg" %}
11459   opcode(0xD9);               /* D9 /0, FLD m32real */
11460   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
11461               Pop_Reg_FPR(dst) );
11462   ins_pipe( fpu_reg_mem );
11463 %}
11464 
11465 instruct MoveI2F_stack_reg_sse(regF dst, stackSlotI src) %{
11466   predicate(UseSSE>=1);
11467   match(Set dst (MoveI2F src));
11468   effect( DEF dst, USE src );
11469 
11470   ins_cost(95);
11471   format %{ "MOVSS  $dst,$src\t# MoveI2F_stack_reg_sse" %}
11472   ins_encode %{
11473     __ movflt($dst$$XMMRegister, Address(rsp, $src$$disp));
11474   %}
11475   ins_pipe( pipe_slow );
11476 %}
11477 
11478 instruct MoveI2F_reg_reg_sse(regF dst, rRegI src) %{
11479   predicate(UseSSE>=2);
11480   match(Set dst (MoveI2F src));
11481   effect( DEF dst, USE src );
11482 
11483   ins_cost(85);
11484   format %{ "MOVD   $dst,$src\t# MoveI2F_reg_reg_sse" %}
11485   ins_encode %{
11486     __ movdl($dst$$XMMRegister, $src$$Register);
11487   %}
11488   ins_pipe( pipe_slow );
11489 %}
11490 
11491 instruct MoveD2L_stack_reg(eRegL dst, stackSlotD src) %{
11492   match(Set dst (MoveD2L src));
11493   effect(DEF dst, USE src);
11494 
11495   ins_cost(250);
11496   format %{ "MOV    $dst.lo,$src\n\t"
11497             "MOV    $dst.hi,$src+4\t# MoveD2L_stack_reg" %}
11498   opcode(0x8B, 0x8B);
11499   ins_encode( OpcP, RegMem(dst,src), OpcS, RegMem_Hi(dst,src));
11500   ins_pipe( ialu_mem_long_reg );
11501 %}
11502 
11503 instruct MoveDPR2L_reg_stack(stackSlotL dst, regDPR src) %{
11504   predicate(UseSSE<=1);
11505   match(Set dst (MoveD2L src));
11506   effect(DEF dst, USE src);
11507 
11508   ins_cost(125);
11509   format %{ "FST_D  $dst,$src\t# MoveD2L_reg_stack" %}
11510   ins_encode( Pop_Mem_Reg_DPR(dst, src) );
11511   ins_pipe( fpu_mem_reg );
11512 %}
11513 
11514 instruct MoveD2L_reg_stack_sse(stackSlotL dst, regD src) %{
11515   predicate(UseSSE>=2);
11516   match(Set dst (MoveD2L src));
11517   effect(DEF dst, USE src);
11518   ins_cost(95);
11519   format %{ "MOVSD  $dst,$src\t# MoveD2L_reg_stack_sse" %}
11520   ins_encode %{
11521     __ movdbl(Address(rsp, $dst$$disp), $src$$XMMRegister);
11522   %}
11523   ins_pipe( pipe_slow );
11524 %}
11525 
11526 instruct MoveD2L_reg_reg_sse(eRegL dst, regD src, regD tmp) %{
11527   predicate(UseSSE>=2);
11528   match(Set dst (MoveD2L src));
11529   effect(DEF dst, USE src, TEMP tmp);
11530   ins_cost(85);
11531   format %{ "MOVD   $dst.lo,$src\n\t"
11532             "PSHUFLW $tmp,$src,0x4E\n\t"
11533             "MOVD   $dst.hi,$tmp\t# MoveD2L_reg_reg_sse" %}
11534   ins_encode %{
11535     __ movdl($dst$$Register, $src$$XMMRegister);
11536     __ pshuflw($tmp$$XMMRegister, $src$$XMMRegister, 0x4e);
11537     __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
11538   %}
11539   ins_pipe( pipe_slow );
11540 %}
11541 
11542 instruct MoveL2D_reg_stack(stackSlotD dst, eRegL src) %{
11543   match(Set dst (MoveL2D src));
11544   effect(DEF dst, USE src);
11545 
11546   ins_cost(200);
11547   format %{ "MOV    $dst,$src.lo\n\t"
11548             "MOV    $dst+4,$src.hi\t# MoveL2D_reg_stack" %}
11549   opcode(0x89, 0x89);
11550   ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
11551   ins_pipe( ialu_mem_long_reg );
11552 %}
11553 
11554 
11555 instruct MoveL2DPR_stack_reg(regDPR dst, stackSlotL src) %{
11556   predicate(UseSSE<=1);
11557   match(Set dst (MoveL2D src));
11558   effect(DEF dst, USE src);
11559   ins_cost(125);
11560 
11561   format %{ "FLD_D  $src\n\t"
11562             "FSTP   $dst\t# MoveL2D_stack_reg" %}
11563   opcode(0xDD);               /* DD /0, FLD m64real */
11564   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
11565               Pop_Reg_DPR(dst) );
11566   ins_pipe( fpu_reg_mem );
11567 %}
11568 
11569 
11570 instruct MoveL2D_stack_reg_sse(regD dst, stackSlotL src) %{
11571   predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
11572   match(Set dst (MoveL2D src));
11573   effect(DEF dst, USE src);
11574 
11575   ins_cost(95);
11576   format %{ "MOVSD  $dst,$src\t# MoveL2D_stack_reg_sse" %}
11577   ins_encode %{
11578     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
11579   %}
11580   ins_pipe( pipe_slow );
11581 %}
11582 
11583 instruct MoveL2D_stack_reg_sse_partial(regD dst, stackSlotL src) %{
11584   predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
11585   match(Set dst (MoveL2D src));
11586   effect(DEF dst, USE src);
11587 
11588   ins_cost(95);
11589   format %{ "MOVLPD $dst,$src\t# MoveL2D_stack_reg_sse" %}
11590   ins_encode %{
11591     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
11592   %}
11593   ins_pipe( pipe_slow );
11594 %}
11595 
11596 instruct MoveL2D_reg_reg_sse(regD dst, eRegL src, regD tmp) %{
11597   predicate(UseSSE>=2);
11598   match(Set dst (MoveL2D src));
11599   effect(TEMP dst, USE src, TEMP tmp);
11600   ins_cost(85);
11601   format %{ "MOVD   $dst,$src.lo\n\t"
11602             "MOVD   $tmp,$src.hi\n\t"
11603             "PUNPCKLDQ $dst,$tmp\t# MoveL2D_reg_reg_sse" %}
11604   ins_encode %{
11605     __ movdl($dst$$XMMRegister, $src$$Register);
11606     __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
11607     __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
11608   %}
11609   ins_pipe( pipe_slow );
11610 %}
11611 
11612 
11613 // =======================================================================
11614 // fast clearing of an array
11615 instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
11616   predicate(!UseFastStosb);
11617   match(Set dummy (ClearArray cnt base));
11618   effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11619   format %{ "XOR    EAX,EAX\t# ClearArray:\n\t"
11620             "SHL    ECX,1\t# Convert doublewords to words\n\t"
11621             "REP STOS\t# store EAX into [EDI++] while ECX--" %}
11622   ins_encode %{ 
11623     __ clear_mem($base$$Register, $cnt$$Register, $zero$$Register);
11624   %}
11625   ins_pipe( pipe_slow );
11626 %}
11627 
11628 instruct rep_fast_stosb(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
11629   predicate(UseFastStosb);
11630   match(Set dummy (ClearArray cnt base));
11631   effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11632   format %{ "XOR    EAX,EAX\t# ClearArray:\n\t"
11633             "SHL    ECX,3\t# Convert doublewords to bytes\n\t"
11634             "REP STOSB\t# store EAX into [EDI++] while ECX--" %}
11635   ins_encode %{ 
11636     __ clear_mem($base$$Register, $cnt$$Register, $zero$$Register);
11637   %}
11638   ins_pipe( pipe_slow );
11639 %}
11640 
11641 instruct string_compare(eDIRegP str1, eCXRegI cnt1, eSIRegP str2, eDXRegI cnt2,
11642                         eAXRegI result, regD tmp1, eFlagsReg cr) %{
11643   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11644   effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
11645 
11646   format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result   // KILL $tmp1" %}
11647   ins_encode %{
11648     __ string_compare($str1$$Register, $str2$$Register,
11649                       $cnt1$$Register, $cnt2$$Register, $result$$Register,
11650                       $tmp1$$XMMRegister);
11651   %}
11652   ins_pipe( pipe_slow );
11653 %}
11654 
11655 // fast string equals
11656 instruct string_equals(eDIRegP str1, eSIRegP str2, eCXRegI cnt, eAXRegI result,
11657                        regD tmp1, regD tmp2, eBXRegI tmp3, eFlagsReg cr) %{
11658   match(Set result (StrEquals (Binary str1 str2) cnt));
11659   effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);
11660 
11661   format %{ "String Equals $str1,$str2,$cnt -> $result    // KILL $tmp1, $tmp2, $tmp3" %}
11662   ins_encode %{
11663     __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
11664                           $cnt$$Register, $result$$Register, $tmp3$$Register,
11665                           $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11666   %}
11667   ins_pipe( pipe_slow );
11668 %}
11669 
11670 // fast search of substring with known size.
11671 instruct string_indexof_con(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, immI int_cnt2,
11672                             eBXRegI result, regD vec, eAXRegI cnt2, eCXRegI tmp, eFlagsReg cr) %{
11673   predicate(UseSSE42Intrinsics);
11674   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
11675   effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, KILL cnt2, KILL tmp, KILL cr);
11676 
11677   format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result   // KILL $vec, $cnt1, $cnt2, $tmp" %}
11678   ins_encode %{
11679     int icnt2 = (int)$int_cnt2$$constant;
11680     if (icnt2 >= 8) {
11681       // IndexOf for constant substrings with size >= 8 elements
11682       // which don't need to be loaded through stack.
11683       __ string_indexofC8($str1$$Register, $str2$$Register,
11684                           $cnt1$$Register, $cnt2$$Register,
11685                           icnt2, $result$$Register,
11686                           $vec$$XMMRegister, $tmp$$Register);
11687     } else {
11688       // Small strings are loaded through stack if they cross page boundary.
11689       __ string_indexof($str1$$Register, $str2$$Register,
11690                         $cnt1$$Register, $cnt2$$Register,
11691                         icnt2, $result$$Register,
11692                         $vec$$XMMRegister, $tmp$$Register);
11693     }
11694   %}
11695   ins_pipe( pipe_slow );
11696 %}
11697 
11698 instruct string_indexof(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, eAXRegI cnt2,
11699                         eBXRegI result, regD vec, eCXRegI tmp, eFlagsReg cr) %{
11700   predicate(UseSSE42Intrinsics);
11701   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
11702   effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp, KILL cr);
11703 
11704   format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result   // KILL all" %}
11705   ins_encode %{
11706     __ string_indexof($str1$$Register, $str2$$Register,
11707                       $cnt1$$Register, $cnt2$$Register,
11708                       (-1), $result$$Register,
11709                       $vec$$XMMRegister, $tmp$$Register);
11710   %}
11711   ins_pipe( pipe_slow );
11712 %}
11713 
11714 // fast array equals
11715 instruct array_equals(eDIRegP ary1, eSIRegP ary2, eAXRegI result,
11716                       regD tmp1, regD tmp2, eCXRegI tmp3, eBXRegI tmp4, eFlagsReg cr)
11717 %{
11718   match(Set result (AryEq ary1 ary2));
11719   effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
11720   //ins_cost(300);
11721 
11722   format %{ "Array Equals $ary1,$ary2 -> $result   // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
11723   ins_encode %{
11724     __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
11725                           $tmp3$$Register, $result$$Register, $tmp4$$Register,
11726                           $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11727   %}
11728   ins_pipe( pipe_slow );
11729 %}
11730 
11731 // encode char[] to byte[] in ISO_8859_1
11732 instruct encode_iso_array(eSIRegP src, eDIRegP dst, eDXRegI len,
11733                           regD tmp1, regD tmp2, regD tmp3, regD tmp4,
11734                           eCXRegI tmp5, eAXRegI result, eFlagsReg cr) %{
11735   match(Set result (EncodeISOArray src (Binary dst len)));
11736   effect(TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, USE_KILL src, USE_KILL dst, USE_KILL len, KILL tmp5, KILL cr);
11737 
11738   format %{ "Encode array $src,$dst,$len -> $result    // KILL ECX, EDX, $tmp1, $tmp2, $tmp3, $tmp4, ESI, EDI " %}
11739   ins_encode %{
11740     __ encode_iso_array($src$$Register, $dst$$Register, $len$$Register,
11741                         $tmp1$$XMMRegister, $tmp2$$XMMRegister, $tmp3$$XMMRegister,
11742                         $tmp4$$XMMRegister, $tmp5$$Register, $result$$Register);
11743   %}
11744   ins_pipe( pipe_slow );
11745 %}
11746 
11747 
11748 //----------Control Flow Instructions------------------------------------------
11749 // Signed compare Instructions
11750 instruct compI_eReg(eFlagsReg cr, rRegI op1, rRegI op2) %{
11751   match(Set cr (CmpI op1 op2));
11752   effect( DEF cr, USE op1, USE op2 );
11753   format %{ "CMP    $op1,$op2" %}
11754   opcode(0x3B);  /* Opcode 3B /r */
11755   ins_encode( OpcP, RegReg( op1, op2) );
11756   ins_pipe( ialu_cr_reg_reg );
11757 %}
11758 
11759 instruct compI_eReg_imm(eFlagsReg cr, rRegI op1, immI op2) %{
11760   match(Set cr (CmpI op1 op2));
11761   effect( DEF cr, USE op1 );
11762   format %{ "CMP    $op1,$op2" %}
11763   opcode(0x81,0x07);  /* Opcode 81 /7 */
11764   // ins_encode( RegImm( op1, op2) );  /* Was CmpImm */
11765   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
11766   ins_pipe( ialu_cr_reg_imm );
11767 %}
11768 
11769 // Cisc-spilled version of cmpI_eReg
11770 instruct compI_eReg_mem(eFlagsReg cr, rRegI op1, memory op2) %{
11771   match(Set cr (CmpI op1 (LoadI op2)));
11772 
11773   format %{ "CMP    $op1,$op2" %}
11774   ins_cost(500);
11775   opcode(0x3B);  /* Opcode 3B /r */
11776   ins_encode( OpcP, RegMem( op1, op2) );
11777   ins_pipe( ialu_cr_reg_mem );
11778 %}
11779 
11780 instruct testI_reg( eFlagsReg cr, rRegI src, immI0 zero ) %{
11781   match(Set cr (CmpI src zero));
11782   effect( DEF cr, USE src );
11783 
11784   format %{ "TEST   $src,$src" %}
11785   opcode(0x85);
11786   ins_encode( OpcP, RegReg( src, src ) );
11787   ins_pipe( ialu_cr_reg_imm );
11788 %}
11789 
11790 instruct testI_reg_imm( eFlagsReg cr, rRegI src, immI con, immI0 zero ) %{
11791   match(Set cr (CmpI (AndI src con) zero));
11792 
11793   format %{ "TEST   $src,$con" %}
11794   opcode(0xF7,0x00);
11795   ins_encode( OpcP, RegOpc(src), Con32(con) );
11796   ins_pipe( ialu_cr_reg_imm );
11797 %}
11798 
11799 instruct testI_reg_mem( eFlagsReg cr, rRegI src, memory mem, immI0 zero ) %{
11800   match(Set cr (CmpI (AndI src mem) zero));
11801 
11802   format %{ "TEST   $src,$mem" %}
11803   opcode(0x85);
11804   ins_encode( OpcP, RegMem( src, mem ) );
11805   ins_pipe( ialu_cr_reg_mem );
11806 %}
11807 
11808 // Unsigned compare Instructions; really, same as signed except they
11809 // produce an eFlagsRegU instead of eFlagsReg.
11810 instruct compU_eReg(eFlagsRegU cr, rRegI op1, rRegI op2) %{
11811   match(Set cr (CmpU op1 op2));
11812 
11813   format %{ "CMPu   $op1,$op2" %}
11814   opcode(0x3B);  /* Opcode 3B /r */
11815   ins_encode( OpcP, RegReg( op1, op2) );
11816   ins_pipe( ialu_cr_reg_reg );
11817 %}
11818 
11819 instruct compU_eReg_imm(eFlagsRegU cr, rRegI op1, immI op2) %{
11820   match(Set cr (CmpU op1 op2));
11821 
11822   format %{ "CMPu   $op1,$op2" %}
11823   opcode(0x81,0x07);  /* Opcode 81 /7 */
11824   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
11825   ins_pipe( ialu_cr_reg_imm );
11826 %}
11827 
11828 // // Cisc-spilled version of cmpU_eReg
11829 instruct compU_eReg_mem(eFlagsRegU cr, rRegI op1, memory op2) %{
11830   match(Set cr (CmpU op1 (LoadI op2)));
11831 
11832   format %{ "CMPu   $op1,$op2" %}
11833   ins_cost(500);
11834   opcode(0x3B);  /* Opcode 3B /r */
11835   ins_encode( OpcP, RegMem( op1, op2) );
11836   ins_pipe( ialu_cr_reg_mem );
11837 %}
11838 
11839 // // Cisc-spilled version of cmpU_eReg
11840 //instruct compU_mem_eReg(eFlagsRegU cr, memory op1, rRegI op2) %{
11841 //  match(Set cr (CmpU (LoadI op1) op2));
11842 //
11843 //  format %{ "CMPu   $op1,$op2" %}
11844 //  ins_cost(500);
11845 //  opcode(0x39);  /* Opcode 39 /r */
11846 //  ins_encode( OpcP, RegMem( op1, op2) );
11847 //%}
11848 
11849 instruct testU_reg( eFlagsRegU cr, rRegI src, immI0 zero ) %{
11850   match(Set cr (CmpU src zero));
11851 
11852   format %{ "TESTu  $src,$src" %}
11853   opcode(0x85);
11854   ins_encode( OpcP, RegReg( src, src ) );
11855   ins_pipe( ialu_cr_reg_imm );
11856 %}
11857 
11858 // Unsigned pointer compare Instructions
11859 instruct compP_eReg(eFlagsRegU cr, eRegP op1, eRegP op2) %{
11860   match(Set cr (CmpP op1 op2));
11861 
11862   format %{ "CMPu   $op1,$op2" %}
11863   opcode(0x3B);  /* Opcode 3B /r */
11864   ins_encode( OpcP, RegReg( op1, op2) );
11865   ins_pipe( ialu_cr_reg_reg );
11866 %}
11867 
11868 instruct compP_eReg_imm(eFlagsRegU cr, eRegP op1, immP op2) %{
11869   match(Set cr (CmpP op1 op2));
11870 
11871   format %{ "CMPu   $op1,$op2" %}
11872   opcode(0x81,0x07);  /* Opcode 81 /7 */
11873   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
11874   ins_pipe( ialu_cr_reg_imm );
11875 %}
11876 
11877 // // Cisc-spilled version of cmpP_eReg
11878 instruct compP_eReg_mem(eFlagsRegU cr, eRegP op1, memory op2) %{
11879   match(Set cr (CmpP op1 (LoadP op2)));
11880 
11881   format %{ "CMPu   $op1,$op2" %}
11882   ins_cost(500);
11883   opcode(0x3B);  /* Opcode 3B /r */
11884   ins_encode( OpcP, RegMem( op1, op2) );
11885   ins_pipe( ialu_cr_reg_mem );
11886 %}
11887 
11888 // // Cisc-spilled version of cmpP_eReg
11889 //instruct compP_mem_eReg(eFlagsRegU cr, memory op1, eRegP op2) %{
11890 //  match(Set cr (CmpP (LoadP op1) op2));
11891 //
11892 //  format %{ "CMPu   $op1,$op2" %}
11893 //  ins_cost(500);
11894 //  opcode(0x39);  /* Opcode 39 /r */
11895 //  ins_encode( OpcP, RegMem( op1, op2) );
11896 //%}
11897 
11898 // Compare raw pointer (used in out-of-heap check).
11899 // Only works because non-oop pointers must be raw pointers
11900 // and raw pointers have no anti-dependencies.
11901 instruct compP_mem_eReg( eFlagsRegU cr, eRegP op1, memory op2 ) %{
11902   predicate( n->in(2)->in(2)->bottom_type()->reloc() == relocInfo::none );
11903   match(Set cr (CmpP op1 (LoadP op2)));
11904 
11905   format %{ "CMPu   $op1,$op2" %}
11906   opcode(0x3B);  /* Opcode 3B /r */
11907   ins_encode( OpcP, RegMem( op1, op2) );
11908   ins_pipe( ialu_cr_reg_mem );
11909 %}
11910 
11911 //
11912 // This will generate a signed flags result. This should be ok
11913 // since any compare to a zero should be eq/neq.
11914 instruct testP_reg( eFlagsReg cr, eRegP src, immP0 zero ) %{
11915   match(Set cr (CmpP src zero));
11916 
11917   format %{ "TEST   $src,$src" %}
11918   opcode(0x85);
11919   ins_encode( OpcP, RegReg( src, src ) );
11920   ins_pipe( ialu_cr_reg_imm );
11921 %}
11922 
11923 // Cisc-spilled version of testP_reg
11924 // This will generate a signed flags result. This should be ok
11925 // since any compare to a zero should be eq/neq.
11926 instruct testP_Reg_mem( eFlagsReg cr, memory op, immI0 zero ) %{
11927   match(Set cr (CmpP (LoadP op) zero));
11928 
11929   format %{ "TEST   $op,0xFFFFFFFF" %}
11930   ins_cost(500);
11931   opcode(0xF7);               /* Opcode F7 /0 */
11932   ins_encode( OpcP, RMopc_Mem(0x00,op), Con_d32(0xFFFFFFFF) );
11933   ins_pipe( ialu_cr_reg_imm );
11934 %}
11935 
11936 // Yanked all unsigned pointer compare operations.
11937 // Pointer compares are done with CmpP which is already unsigned.
11938 
11939 //----------Max and Min--------------------------------------------------------
11940 // Min Instructions
11941 ////
11942 //   *** Min and Max using the conditional move are slower than the
11943 //   *** branch version on a Pentium III.
11944 // // Conditional move for min
11945 //instruct cmovI_reg_lt( rRegI op2, rRegI op1, eFlagsReg cr ) %{
11946 //  effect( USE_DEF op2, USE op1, USE cr );
11947 //  format %{ "CMOVlt $op2,$op1\t! min" %}
11948 //  opcode(0x4C,0x0F);
11949 //  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
11950 //  ins_pipe( pipe_cmov_reg );
11951 //%}
11952 //
11953 //// Min Register with Register (P6 version)
11954 //instruct minI_eReg_p6( rRegI op1, rRegI op2 ) %{
11955 //  predicate(VM_Version::supports_cmov() );
11956 //  match(Set op2 (MinI op1 op2));
11957 //  ins_cost(200);
11958 //  expand %{
11959 //    eFlagsReg cr;
11960 //    compI_eReg(cr,op1,op2);
11961 //    cmovI_reg_lt(op2,op1,cr);
11962 //  %}
11963 //%}
11964 
11965 // Min Register with Register (generic version)
11966 instruct minI_eReg(rRegI dst, rRegI src, eFlagsReg flags) %{
11967   match(Set dst (MinI dst src));
11968   effect(KILL flags);
11969   ins_cost(300);
11970 
11971   format %{ "MIN    $dst,$src" %}
11972   opcode(0xCC);
11973   ins_encode( min_enc(dst,src) );
11974   ins_pipe( pipe_slow );
11975 %}
11976 
11977 // Max Register with Register
11978 //   *** Min and Max using the conditional move are slower than the
11979 //   *** branch version on a Pentium III.
11980 // // Conditional move for max
11981 //instruct cmovI_reg_gt( rRegI op2, rRegI op1, eFlagsReg cr ) %{
11982 //  effect( USE_DEF op2, USE op1, USE cr );
11983 //  format %{ "CMOVgt $op2,$op1\t! max" %}
11984 //  opcode(0x4F,0x0F);
11985 //  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
11986 //  ins_pipe( pipe_cmov_reg );
11987 //%}
11988 //
11989 // // Max Register with Register (P6 version)
11990 //instruct maxI_eReg_p6( rRegI op1, rRegI op2 ) %{
11991 //  predicate(VM_Version::supports_cmov() );
11992 //  match(Set op2 (MaxI op1 op2));
11993 //  ins_cost(200);
11994 //  expand %{
11995 //    eFlagsReg cr;
11996 //    compI_eReg(cr,op1,op2);
11997 //    cmovI_reg_gt(op2,op1,cr);
11998 //  %}
11999 //%}
12000 
12001 // Max Register with Register (generic version)
12002 instruct maxI_eReg(rRegI dst, rRegI src, eFlagsReg flags) %{
12003   match(Set dst (MaxI dst src));
12004   effect(KILL flags);
12005   ins_cost(300);
12006 
12007   format %{ "MAX    $dst,$src" %}
12008   opcode(0xCC);
12009   ins_encode( max_enc(dst,src) );
12010   ins_pipe( pipe_slow );
12011 %}
12012 
12013 // ============================================================================
12014 // Counted Loop limit node which represents exact final iterator value.
12015 // Note: the resulting value should fit into integer range since
12016 // counted loops have limit check on overflow.
12017 instruct loopLimit_eReg(eAXRegI limit, nadxRegI init, immI stride, eDXRegI limit_hi, nadxRegI tmp, eFlagsReg flags) %{
12018   match(Set limit (LoopLimit (Binary init limit) stride));
12019   effect(TEMP limit_hi, TEMP tmp, KILL flags);
12020   ins_cost(300);
12021 
12022   format %{ "loopLimit $init,$limit,$stride  # $limit = $init + $stride *( $limit - $init + $stride -1)/ $stride, kills $limit_hi" %}
12023   ins_encode %{
12024     int strd = (int)$stride$$constant;
12025     assert(strd != 1 && strd != -1, "sanity");
12026     int m1 = (strd > 0) ? 1 : -1;
12027     // Convert limit to long (EAX:EDX)
12028     __ cdql();
12029     // Convert init to long (init:tmp)
12030     __ movl($tmp$$Register, $init$$Register);
12031     __ sarl($tmp$$Register, 31);
12032     // $limit - $init
12033     __ subl($limit$$Register, $init$$Register);
12034     __ sbbl($limit_hi$$Register, $tmp$$Register);
12035     // + ($stride - 1)
12036     if (strd > 0) {
12037       __ addl($limit$$Register, (strd - 1));
12038       __ adcl($limit_hi$$Register, 0);
12039       __ movl($tmp$$Register, strd);
12040     } else {
12041       __ addl($limit$$Register, (strd + 1));
12042       __ adcl($limit_hi$$Register, -1);
12043       __ lneg($limit_hi$$Register, $limit$$Register);
12044       __ movl($tmp$$Register, -strd);
12045     }
12046     // signed devision: (EAX:EDX) / pos_stride
12047     __ idivl($tmp$$Register);
12048     if (strd < 0) {
12049       // restore sign
12050       __ negl($tmp$$Register);
12051     }
12052     // (EAX) * stride
12053     __ mull($tmp$$Register);
12054     // + init (ignore upper bits)
12055     __ addl($limit$$Register, $init$$Register);
12056   %}
12057   ins_pipe( pipe_slow );
12058 %}
12059 
12060 // ============================================================================
12061 // Branch Instructions
12062 // Jump Table
12063 instruct jumpXtnd(rRegI switch_val) %{
12064   match(Jump switch_val);
12065   ins_cost(350);
12066   format %{  "JMP    [$constantaddress](,$switch_val,1)\n\t" %}
12067   ins_encode %{
12068     // Jump to Address(table_base + switch_reg)
12069     Address index(noreg, $switch_val$$Register, Address::times_1);
12070     __ jump(ArrayAddress($constantaddress, index));
12071   %}
12072   ins_pipe(pipe_jmp);
12073 %}
12074 
12075 // Jump Direct - Label defines a relative address from JMP+1
12076 instruct jmpDir(label labl) %{
12077   match(Goto);
12078   effect(USE labl);
12079 
12080   ins_cost(300);
12081   format %{ "JMP    $labl" %}
12082   size(5);
12083   ins_encode %{
12084     Label* L = $labl$$label;
12085     __ jmp(*L, false); // Always long jump
12086   %}
12087   ins_pipe( pipe_jmp );
12088 %}
12089 
12090 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12091 instruct jmpCon(cmpOp cop, eFlagsReg cr, label labl) %{
12092   match(If cop cr);
12093   effect(USE labl);
12094 
12095   ins_cost(300);
12096   format %{ "J$cop    $labl" %}
12097   size(6);
12098   ins_encode %{
12099     Label* L = $labl$$label;
12100     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12101   %}
12102   ins_pipe( pipe_jcc );
12103 %}
12104 
12105 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12106 instruct jmpLoopEnd(cmpOp cop, eFlagsReg cr, label labl) %{
12107   match(CountedLoopEnd cop cr);
12108   effect(USE labl);
12109 
12110   ins_cost(300);
12111   format %{ "J$cop    $labl\t# Loop end" %}
12112   size(6);
12113   ins_encode %{
12114     Label* L = $labl$$label;
12115     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12116   %}
12117   ins_pipe( pipe_jcc );
12118 %}
12119 
12120 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12121 instruct jmpLoopEndU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12122   match(CountedLoopEnd cop cmp);
12123   effect(USE labl);
12124 
12125   ins_cost(300);
12126   format %{ "J$cop,u  $labl\t# Loop end" %}
12127   size(6);
12128   ins_encode %{
12129     Label* L = $labl$$label;
12130     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12131   %}
12132   ins_pipe( pipe_jcc );
12133 %}
12134 
12135 instruct jmpLoopEndUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12136   match(CountedLoopEnd cop cmp);
12137   effect(USE labl);
12138 
12139   ins_cost(200);
12140   format %{ "J$cop,u  $labl\t# Loop end" %}
12141   size(6);
12142   ins_encode %{
12143     Label* L = $labl$$label;
12144     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12145   %}
12146   ins_pipe( pipe_jcc );
12147 %}
12148 
12149 // Jump Direct Conditional - using unsigned comparison
12150 instruct jmpConU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12151   match(If cop cmp);
12152   effect(USE labl);
12153 
12154   ins_cost(300);
12155   format %{ "J$cop,u  $labl" %}
12156   size(6);
12157   ins_encode %{
12158     Label* L = $labl$$label;
12159     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12160   %}
12161   ins_pipe(pipe_jcc);
12162 %}
12163 
12164 instruct jmpConUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12165   match(If cop cmp);
12166   effect(USE labl);
12167 
12168   ins_cost(200);
12169   format %{ "J$cop,u  $labl" %}
12170   size(6);
12171   ins_encode %{
12172     Label* L = $labl$$label;
12173     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12174   %}
12175   ins_pipe(pipe_jcc);
12176 %}
12177 
12178 instruct jmpConUCF2(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
12179   match(If cop cmp);
12180   effect(USE labl);
12181 
12182   ins_cost(200);
12183   format %{ $$template
12184     if ($cop$$cmpcode == Assembler::notEqual) {
12185       $$emit$$"JP,u   $labl\n\t"
12186       $$emit$$"J$cop,u   $labl"
12187     } else {
12188       $$emit$$"JP,u   done\n\t"
12189       $$emit$$"J$cop,u   $labl\n\t"
12190       $$emit$$"done:"
12191     }
12192   %}
12193   ins_encode %{
12194     Label* l = $labl$$label;
12195     if ($cop$$cmpcode == Assembler::notEqual) {
12196       __ jcc(Assembler::parity, *l, false);
12197       __ jcc(Assembler::notEqual, *l, false);
12198     } else if ($cop$$cmpcode == Assembler::equal) {
12199       Label done;
12200       __ jccb(Assembler::parity, done);
12201       __ jcc(Assembler::equal, *l, false);
12202       __ bind(done);
12203     } else {
12204        ShouldNotReachHere();
12205     }
12206   %}
12207   ins_pipe(pipe_jcc);
12208 %}
12209 
12210 // ============================================================================
12211 // The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary superklass
12212 // array for an instance of the superklass.  Set a hidden internal cache on a
12213 // hit (cache is checked with exposed code in gen_subtype_check()).  Return
12214 // NZ for a miss or zero for a hit.  The encoding ALSO sets flags.
12215 instruct partialSubtypeCheck( eDIRegP result, eSIRegP sub, eAXRegP super, eCXRegI rcx, eFlagsReg cr ) %{
12216   match(Set result (PartialSubtypeCheck sub super));
12217   effect( KILL rcx, KILL cr );
12218 
12219   ins_cost(1100);  // slightly larger than the next version
12220   format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
12221             "MOV    ECX,[EDI+ArrayKlass::length]\t# length to scan\n\t"
12222             "ADD    EDI,ArrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
12223             "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
12224             "JNE,s  miss\t\t# Missed: EDI not-zero\n\t"
12225             "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache\n\t"
12226             "XOR    $result,$result\t\t Hit: EDI zero\n\t"
12227      "miss:\t" %}
12228 
12229   opcode(0x1); // Force a XOR of EDI
12230   ins_encode( enc_PartialSubtypeCheck() );
12231   ins_pipe( pipe_slow );
12232 %}
12233 
12234 instruct partialSubtypeCheck_vs_Zero( eFlagsReg cr, eSIRegP sub, eAXRegP super, eCXRegI rcx, eDIRegP result, immP0 zero ) %{
12235   match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12236   effect( KILL rcx, KILL result );
12237 
12238   ins_cost(1000);
12239   format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
12240             "MOV    ECX,[EDI+ArrayKlass::length]\t# length to scan\n\t"
12241             "ADD    EDI,ArrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
12242             "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
12243             "JNE,s  miss\t\t# Missed: flags NZ\n\t"
12244             "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache, flags Z\n\t"
12245      "miss:\t" %}
12246 
12247   opcode(0x0);  // No need to XOR EDI
12248   ins_encode( enc_PartialSubtypeCheck() );
12249   ins_pipe( pipe_slow );
12250 %}
12251 
12252 // ============================================================================
12253 // Branch Instructions -- short offset versions
12254 //
12255 // These instructions are used to replace jumps of a long offset (the default
12256 // match) with jumps of a shorter offset.  These instructions are all tagged
12257 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12258 // match rules in general matching.  Instead, the ADLC generates a conversion
12259 // method in the MachNode which can be used to do in-place replacement of the
12260 // long variant with the shorter variant.  The compiler will determine if a
12261 // branch can be taken by the is_short_branch_offset() predicate in the machine
12262 // specific code section of the file.
12263 
12264 // Jump Direct - Label defines a relative address from JMP+1
12265 instruct jmpDir_short(label labl) %{
12266   match(Goto);
12267   effect(USE labl);
12268 
12269   ins_cost(300);
12270   format %{ "JMP,s  $labl" %}
12271   size(2);
12272   ins_encode %{
12273     Label* L = $labl$$label;
12274     __ jmpb(*L);
12275   %}
12276   ins_pipe( pipe_jmp );
12277   ins_short_branch(1);
12278 %}
12279 
12280 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12281 instruct jmpCon_short(cmpOp cop, eFlagsReg cr, label labl) %{
12282   match(If cop cr);
12283   effect(USE labl);
12284 
12285   ins_cost(300);
12286   format %{ "J$cop,s  $labl" %}
12287   size(2);
12288   ins_encode %{
12289     Label* L = $labl$$label;
12290     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12291   %}
12292   ins_pipe( pipe_jcc );
12293   ins_short_branch(1);
12294 %}
12295 
12296 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12297 instruct jmpLoopEnd_short(cmpOp cop, eFlagsReg cr, label labl) %{
12298   match(CountedLoopEnd cop cr);
12299   effect(USE labl);
12300 
12301   ins_cost(300);
12302   format %{ "J$cop,s  $labl\t# Loop end" %}
12303   size(2);
12304   ins_encode %{
12305     Label* L = $labl$$label;
12306     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12307   %}
12308   ins_pipe( pipe_jcc );
12309   ins_short_branch(1);
12310 %}
12311 
12312 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12313 instruct jmpLoopEndU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12314   match(CountedLoopEnd cop cmp);
12315   effect(USE labl);
12316 
12317   ins_cost(300);
12318   format %{ "J$cop,us $labl\t# Loop end" %}
12319   size(2);
12320   ins_encode %{
12321     Label* L = $labl$$label;
12322     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12323   %}
12324   ins_pipe( pipe_jcc );
12325   ins_short_branch(1);
12326 %}
12327 
12328 instruct jmpLoopEndUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12329   match(CountedLoopEnd cop cmp);
12330   effect(USE labl);
12331 
12332   ins_cost(300);
12333   format %{ "J$cop,us $labl\t# Loop end" %}
12334   size(2);
12335   ins_encode %{
12336     Label* L = $labl$$label;
12337     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12338   %}
12339   ins_pipe( pipe_jcc );
12340   ins_short_branch(1);
12341 %}
12342 
12343 // Jump Direct Conditional - using unsigned comparison
12344 instruct jmpConU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12345   match(If cop cmp);
12346   effect(USE labl);
12347 
12348   ins_cost(300);
12349   format %{ "J$cop,us $labl" %}
12350   size(2);
12351   ins_encode %{
12352     Label* L = $labl$$label;
12353     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12354   %}
12355   ins_pipe( pipe_jcc );
12356   ins_short_branch(1);
12357 %}
12358 
12359 instruct jmpConUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12360   match(If cop cmp);
12361   effect(USE labl);
12362 
12363   ins_cost(300);
12364   format %{ "J$cop,us $labl" %}
12365   size(2);
12366   ins_encode %{
12367     Label* L = $labl$$label;
12368     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12369   %}
12370   ins_pipe( pipe_jcc );
12371   ins_short_branch(1);
12372 %}
12373 
12374 instruct jmpConUCF2_short(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
12375   match(If cop cmp);
12376   effect(USE labl);
12377 
12378   ins_cost(300);
12379   format %{ $$template
12380     if ($cop$$cmpcode == Assembler::notEqual) {
12381       $$emit$$"JP,u,s   $labl\n\t"
12382       $$emit$$"J$cop,u,s   $labl"
12383     } else {
12384       $$emit$$"JP,u,s   done\n\t"
12385       $$emit$$"J$cop,u,s  $labl\n\t"
12386       $$emit$$"done:"
12387     }
12388   %}
12389   size(4);
12390   ins_encode %{
12391     Label* l = $labl$$label;
12392     if ($cop$$cmpcode == Assembler::notEqual) {
12393       __ jccb(Assembler::parity, *l);
12394       __ jccb(Assembler::notEqual, *l);
12395     } else if ($cop$$cmpcode == Assembler::equal) {
12396       Label done;
12397       __ jccb(Assembler::parity, done);
12398       __ jccb(Assembler::equal, *l);
12399       __ bind(done);
12400     } else {
12401        ShouldNotReachHere();
12402     }
12403   %}
12404   ins_pipe(pipe_jcc);
12405   ins_short_branch(1);
12406 %}
12407 
12408 // ============================================================================
12409 // Long Compare
12410 //
12411 // Currently we hold longs in 2 registers.  Comparing such values efficiently
12412 // is tricky.  The flavor of compare used depends on whether we are testing
12413 // for LT, LE, or EQ.  For a simple LT test we can check just the sign bit.
12414 // The GE test is the negated LT test.  The LE test can be had by commuting
12415 // the operands (yielding a GE test) and then negating; negate again for the
12416 // GT test.  The EQ test is done by ORcc'ing the high and low halves, and the
12417 // NE test is negated from that.
12418 
12419 // Due to a shortcoming in the ADLC, it mixes up expressions like:
12420 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)).  Note the
12421 // difference between 'Y' and '0L'.  The tree-matches for the CmpI sections
12422 // are collapsed internally in the ADLC's dfa-gen code.  The match for
12423 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
12424 // foo match ends up with the wrong leaf.  One fix is to not match both
12425 // reg-reg and reg-zero forms of long-compare.  This is unfortunate because
12426 // both forms beat the trinary form of long-compare and both are very useful
12427 // on Intel which has so few registers.
12428 
12429 // Manifest a CmpL result in an integer register.  Very painful.
12430 // This is the test to avoid.
12431 instruct cmpL3_reg_reg(eSIRegI dst, eRegL src1, eRegL src2, eFlagsReg flags ) %{
12432   match(Set dst (CmpL3 src1 src2));
12433   effect( KILL flags );
12434   ins_cost(1000);
12435   format %{ "XOR    $dst,$dst\n\t"
12436             "CMP    $src1.hi,$src2.hi\n\t"
12437             "JLT,s  m_one\n\t"
12438             "JGT,s  p_one\n\t"
12439             "CMP    $src1.lo,$src2.lo\n\t"
12440             "JB,s   m_one\n\t"
12441             "JEQ,s  done\n"
12442     "p_one:\tINC    $dst\n\t"
12443             "JMP,s  done\n"
12444     "m_one:\tDEC    $dst\n"
12445      "done:" %}
12446   ins_encode %{
12447     Label p_one, m_one, done;
12448     __ xorptr($dst$$Register, $dst$$Register);
12449     __ cmpl(HIGH_FROM_LOW($src1$$Register), HIGH_FROM_LOW($src2$$Register));
12450     __ jccb(Assembler::less,    m_one);
12451     __ jccb(Assembler::greater, p_one);
12452     __ cmpl($src1$$Register, $src2$$Register);
12453     __ jccb(Assembler::below,   m_one);
12454     __ jccb(Assembler::equal,   done);
12455     __ bind(p_one);
12456     __ incrementl($dst$$Register);
12457     __ jmpb(done);
12458     __ bind(m_one);
12459     __ decrementl($dst$$Register);
12460     __ bind(done);
12461   %}
12462   ins_pipe( pipe_slow );
12463 %}
12464 
12465 //======
12466 // Manifest a CmpL result in the normal flags.  Only good for LT or GE
12467 // compares.  Can be used for LE or GT compares by reversing arguments.
12468 // NOT GOOD FOR EQ/NE tests.
12469 instruct cmpL_zero_flags_LTGE( flagsReg_long_LTGE flags, eRegL src, immL0 zero ) %{
12470   match( Set flags (CmpL src zero ));
12471   ins_cost(100);
12472   format %{ "TEST   $src.hi,$src.hi" %}
12473   opcode(0x85);
12474   ins_encode( OpcP, RegReg_Hi2( src, src ) );
12475   ins_pipe( ialu_cr_reg_reg );
12476 %}
12477 
12478 // Manifest a CmpL result in the normal flags.  Only good for LT or GE
12479 // compares.  Can be used for LE or GT compares by reversing arguments.
12480 // NOT GOOD FOR EQ/NE tests.
12481 instruct cmpL_reg_flags_LTGE( flagsReg_long_LTGE flags, eRegL src1, eRegL src2, rRegI tmp ) %{
12482   match( Set flags (CmpL src1 src2 ));
12483   effect( TEMP tmp );
12484   ins_cost(300);
12485   format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
12486             "MOV    $tmp,$src1.hi\n\t"
12487             "SBB    $tmp,$src2.hi\t! Compute flags for long compare" %}
12488   ins_encode( long_cmp_flags2( src1, src2, tmp ) );
12489   ins_pipe( ialu_cr_reg_reg );
12490 %}
12491 
12492 // Long compares reg < zero/req OR reg >= zero/req.
12493 // Just a wrapper for a normal branch, plus the predicate test.
12494 instruct cmpL_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, label labl) %{
12495   match(If cmp flags);
12496   effect(USE labl);
12497   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12498   expand %{
12499     jmpCon(cmp,flags,labl);    // JLT or JGE...
12500   %}
12501 %}
12502 
12503 // Compare 2 longs and CMOVE longs.
12504 instruct cmovLL_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, eRegL src) %{
12505   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12506   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
12507   ins_cost(400);
12508   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12509             "CMOV$cmp $dst.hi,$src.hi" %}
12510   opcode(0x0F,0x40);
12511   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12512   ins_pipe( pipe_cmov_reg_long );
12513 %}
12514 
12515 instruct cmovLL_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, load_long_memory src) %{
12516   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12517   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
12518   ins_cost(500);
12519   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12520             "CMOV$cmp $dst.hi,$src.hi" %}
12521   opcode(0x0F,0x40);
12522   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12523   ins_pipe( pipe_cmov_reg_long );
12524 %}
12525 
12526 // Compare 2 longs and CMOVE ints.
12527 instruct cmovII_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, rRegI dst, rRegI src) %{
12528   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
12529   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12530   ins_cost(200);
12531   format %{ "CMOV$cmp $dst,$src" %}
12532   opcode(0x0F,0x40);
12533   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12534   ins_pipe( pipe_cmov_reg );
12535 %}
12536 
12537 instruct cmovII_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, rRegI dst, memory src) %{
12538   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
12539   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12540   ins_cost(250);
12541   format %{ "CMOV$cmp $dst,$src" %}
12542   opcode(0x0F,0x40);
12543   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12544   ins_pipe( pipe_cmov_mem );
12545 %}
12546 
12547 // Compare 2 longs and CMOVE ints.
12548 instruct cmovPP_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegP dst, eRegP src) %{
12549   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge ));
12550   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
12551   ins_cost(200);
12552   format %{ "CMOV$cmp $dst,$src" %}
12553   opcode(0x0F,0x40);
12554   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12555   ins_pipe( pipe_cmov_reg );
12556 %}
12557 
12558 // Compare 2 longs and CMOVE doubles
12559 instruct cmovDDPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regDPR dst, regDPR src) %{
12560   predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12561   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12562   ins_cost(200);
12563   expand %{
12564     fcmovDPR_regS(cmp,flags,dst,src);
12565   %}
12566 %}
12567 
12568 // Compare 2 longs and CMOVE doubles
12569 instruct cmovDD_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regD dst, regD src) %{
12570   predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12571   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12572   ins_cost(200);
12573   expand %{
12574     fcmovD_regS(cmp,flags,dst,src);
12575   %}
12576 %}
12577 
12578 instruct cmovFFPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regFPR dst, regFPR src) %{
12579   predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12580   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12581   ins_cost(200);
12582   expand %{
12583     fcmovFPR_regS(cmp,flags,dst,src);
12584   %}
12585 %}
12586 
12587 instruct cmovFF_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regF dst, regF src) %{
12588   predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12589   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12590   ins_cost(200);
12591   expand %{
12592     fcmovF_regS(cmp,flags,dst,src);
12593   %}
12594 %}
12595 
12596 //======
12597 // Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
12598 instruct cmpL_zero_flags_EQNE( flagsReg_long_EQNE flags, eRegL src, immL0 zero, rRegI tmp ) %{
12599   match( Set flags (CmpL src zero ));
12600   effect(TEMP tmp);
12601   ins_cost(200);
12602   format %{ "MOV    $tmp,$src.lo\n\t"
12603             "OR     $tmp,$src.hi\t! Long is EQ/NE 0?" %}
12604   ins_encode( long_cmp_flags0( src, tmp ) );
12605   ins_pipe( ialu_reg_reg_long );
12606 %}
12607 
12608 // Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
12609 instruct cmpL_reg_flags_EQNE( flagsReg_long_EQNE flags, eRegL src1, eRegL src2 ) %{
12610   match( Set flags (CmpL src1 src2 ));
12611   ins_cost(200+300);
12612   format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
12613             "JNE,s  skip\n\t"
12614             "CMP    $src1.hi,$src2.hi\n\t"
12615      "skip:\t" %}
12616   ins_encode( long_cmp_flags1( src1, src2 ) );
12617   ins_pipe( ialu_cr_reg_reg );
12618 %}
12619 
12620 // Long compare reg == zero/reg OR reg != zero/reg
12621 // Just a wrapper for a normal branch, plus the predicate test.
12622 instruct cmpL_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, label labl) %{
12623   match(If cmp flags);
12624   effect(USE labl);
12625   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12626   expand %{
12627     jmpCon(cmp,flags,labl);    // JEQ or JNE...
12628   %}
12629 %}
12630 
12631 // Compare 2 longs and CMOVE longs.
12632 instruct cmovLL_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, eRegL src) %{
12633   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12634   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
12635   ins_cost(400);
12636   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12637             "CMOV$cmp $dst.hi,$src.hi" %}
12638   opcode(0x0F,0x40);
12639   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12640   ins_pipe( pipe_cmov_reg_long );
12641 %}
12642 
12643 instruct cmovLL_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, load_long_memory src) %{
12644   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12645   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
12646   ins_cost(500);
12647   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12648             "CMOV$cmp $dst.hi,$src.hi" %}
12649   opcode(0x0F,0x40);
12650   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12651   ins_pipe( pipe_cmov_reg_long );
12652 %}
12653 
12654 // Compare 2 longs and CMOVE ints.
12655 instruct cmovII_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, rRegI dst, rRegI src) %{
12656   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
12657   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12658   ins_cost(200);
12659   format %{ "CMOV$cmp $dst,$src" %}
12660   opcode(0x0F,0x40);
12661   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12662   ins_pipe( pipe_cmov_reg );
12663 %}
12664 
12665 instruct cmovII_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, rRegI dst, memory src) %{
12666   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
12667   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12668   ins_cost(250);
12669   format %{ "CMOV$cmp $dst,$src" %}
12670   opcode(0x0F,0x40);
12671   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12672   ins_pipe( pipe_cmov_mem );
12673 %}
12674 
12675 // Compare 2 longs and CMOVE ints.
12676 instruct cmovPP_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegP dst, eRegP src) %{
12677   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne ));
12678   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
12679   ins_cost(200);
12680   format %{ "CMOV$cmp $dst,$src" %}
12681   opcode(0x0F,0x40);
12682   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12683   ins_pipe( pipe_cmov_reg );
12684 %}
12685 
12686 // Compare 2 longs and CMOVE doubles
12687 instruct cmovDDPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regDPR dst, regDPR src) %{
12688   predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12689   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12690   ins_cost(200);
12691   expand %{
12692     fcmovDPR_regS(cmp,flags,dst,src);
12693   %}
12694 %}
12695 
12696 // Compare 2 longs and CMOVE doubles
12697 instruct cmovDD_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regD dst, regD src) %{
12698   predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12699   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12700   ins_cost(200);
12701   expand %{
12702     fcmovD_regS(cmp,flags,dst,src);
12703   %}
12704 %}
12705 
12706 instruct cmovFFPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regFPR dst, regFPR src) %{
12707   predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12708   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12709   ins_cost(200);
12710   expand %{
12711     fcmovFPR_regS(cmp,flags,dst,src);
12712   %}
12713 %}
12714 
12715 instruct cmovFF_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regF dst, regF src) %{
12716   predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12717   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12718   ins_cost(200);
12719   expand %{
12720     fcmovF_regS(cmp,flags,dst,src);
12721   %}
12722 %}
12723 
12724 //======
12725 // Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
12726 // Same as cmpL_reg_flags_LEGT except must negate src
12727 instruct cmpL_zero_flags_LEGT( flagsReg_long_LEGT flags, eRegL src, immL0 zero, rRegI tmp ) %{
12728   match( Set flags (CmpL src zero ));
12729   effect( TEMP tmp );
12730   ins_cost(300);
12731   format %{ "XOR    $tmp,$tmp\t# Long compare for -$src < 0, use commuted test\n\t"
12732             "CMP    $tmp,$src.lo\n\t"
12733             "SBB    $tmp,$src.hi\n\t" %}
12734   ins_encode( long_cmp_flags3(src, tmp) );
12735   ins_pipe( ialu_reg_reg_long );
12736 %}
12737 
12738 // Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
12739 // Same as cmpL_reg_flags_LTGE except operands swapped.  Swapping operands
12740 // requires a commuted test to get the same result.
12741 instruct cmpL_reg_flags_LEGT( flagsReg_long_LEGT flags, eRegL src1, eRegL src2, rRegI tmp ) %{
12742   match( Set flags (CmpL src1 src2 ));
12743   effect( TEMP tmp );
12744   ins_cost(300);
12745   format %{ "CMP    $src2.lo,$src1.lo\t! Long compare, swapped operands, use with commuted test\n\t"
12746             "MOV    $tmp,$src2.hi\n\t"
12747             "SBB    $tmp,$src1.hi\t! Compute flags for long compare" %}
12748   ins_encode( long_cmp_flags2( src2, src1, tmp ) );
12749   ins_pipe( ialu_cr_reg_reg );
12750 %}
12751 
12752 // Long compares reg < zero/req OR reg >= zero/req.
12753 // Just a wrapper for a normal branch, plus the predicate test
12754 instruct cmpL_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, label labl) %{
12755   match(If cmp flags);
12756   effect(USE labl);
12757   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le );
12758   ins_cost(300);
12759   expand %{
12760     jmpCon(cmp,flags,labl);    // JGT or JLE...
12761   %}
12762 %}
12763 
12764 // Compare 2 longs and CMOVE longs.
12765 instruct cmovLL_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, eRegL src) %{
12766   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12767   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
12768   ins_cost(400);
12769   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12770             "CMOV$cmp $dst.hi,$src.hi" %}
12771   opcode(0x0F,0x40);
12772   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12773   ins_pipe( pipe_cmov_reg_long );
12774 %}
12775 
12776 instruct cmovLL_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, load_long_memory src) %{
12777   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12778   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
12779   ins_cost(500);
12780   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12781             "CMOV$cmp $dst.hi,$src.hi+4" %}
12782   opcode(0x0F,0x40);
12783   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12784   ins_pipe( pipe_cmov_reg_long );
12785 %}
12786 
12787 // Compare 2 longs and CMOVE ints.
12788 instruct cmovII_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, rRegI dst, rRegI src) %{
12789   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
12790   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12791   ins_cost(200);
12792   format %{ "CMOV$cmp $dst,$src" %}
12793   opcode(0x0F,0x40);
12794   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12795   ins_pipe( pipe_cmov_reg );
12796 %}
12797 
12798 instruct cmovII_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, rRegI dst, memory src) %{
12799   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
12800   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12801   ins_cost(250);
12802   format %{ "CMOV$cmp $dst,$src" %}
12803   opcode(0x0F,0x40);
12804   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12805   ins_pipe( pipe_cmov_mem );
12806 %}
12807 
12808 // Compare 2 longs and CMOVE ptrs.
12809 instruct cmovPP_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegP dst, eRegP src) %{
12810   predicate(VM_Version::supports_cmov() && ( _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt ));
12811   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
12812   ins_cost(200);
12813   format %{ "CMOV$cmp $dst,$src" %}
12814   opcode(0x0F,0x40);
12815   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12816   ins_pipe( pipe_cmov_reg );
12817 %}
12818 
12819 // Compare 2 longs and CMOVE doubles
12820 instruct cmovDDPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regDPR dst, regDPR src) %{
12821   predicate( UseSSE<=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
12822   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12823   ins_cost(200);
12824   expand %{
12825     fcmovDPR_regS(cmp,flags,dst,src);
12826   %}
12827 %}
12828 
12829 // Compare 2 longs and CMOVE doubles
12830 instruct cmovDD_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regD dst, regD src) %{
12831   predicate( UseSSE>=2 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
12832   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12833   ins_cost(200);
12834   expand %{
12835     fcmovD_regS(cmp,flags,dst,src);
12836   %}
12837 %}
12838 
12839 instruct cmovFFPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regFPR dst, regFPR src) %{
12840   predicate( UseSSE==0 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
12841   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12842   ins_cost(200);
12843   expand %{
12844     fcmovFPR_regS(cmp,flags,dst,src);
12845   %}
12846 %}
12847 
12848 
12849 instruct cmovFF_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regF dst, regF src) %{
12850   predicate( UseSSE>=1 && _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le || _kids[0]->_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt );
12851   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12852   ins_cost(200);
12853   expand %{
12854     fcmovF_regS(cmp,flags,dst,src);
12855   %}
12856 %}
12857 
12858 
12859 // ============================================================================
12860 // Procedure Call/Return Instructions
12861 // Call Java Static Instruction
12862 // Note: If this code changes, the corresponding ret_addr_offset() and
12863 //       compute_padding() functions will have to be adjusted.
12864 instruct CallStaticJavaDirect(method meth) %{
12865   match(CallStaticJava);
12866   predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
12867   effect(USE meth);
12868 
12869   ins_cost(300);
12870   format %{ "CALL,static " %}
12871   opcode(0xE8); /* E8 cd */
12872   ins_encode( pre_call_resets,
12873               Java_Static_Call( meth ),
12874               call_epilog,
12875               post_call_FPU );
12876   ins_pipe( pipe_slow );
12877   ins_alignment(4);
12878 %}
12879 
12880 // Call Java Static Instruction (method handle version)
12881 // Note: If this code changes, the corresponding ret_addr_offset() and
12882 //       compute_padding() functions will have to be adjusted.
12883 instruct CallStaticJavaHandle(method meth, eBPRegP ebp_mh_SP_save) %{
12884   match(CallStaticJava);
12885   predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
12886   effect(USE meth);
12887   // EBP is saved by all callees (for interpreter stack correction).
12888   // We use it here for a similar purpose, in {preserve,restore}_SP.
12889 
12890   ins_cost(300);
12891   format %{ "CALL,static/MethodHandle " %}
12892   opcode(0xE8); /* E8 cd */
12893   ins_encode( pre_call_resets,
12894               preserve_SP,
12895               Java_Static_Call( meth ),
12896               restore_SP,
12897               call_epilog,
12898               post_call_FPU );
12899   ins_pipe( pipe_slow );
12900   ins_alignment(4);
12901 %}
12902 
12903 // Call Java Dynamic Instruction
12904 // Note: If this code changes, the corresponding ret_addr_offset() and
12905 //       compute_padding() functions will have to be adjusted.
12906 instruct CallDynamicJavaDirect(method meth) %{
12907   match(CallDynamicJava);
12908   effect(USE meth);
12909 
12910   ins_cost(300);
12911   format %{ "MOV    EAX,(oop)-1\n\t"
12912             "CALL,dynamic" %}
12913   opcode(0xE8); /* E8 cd */
12914   ins_encode( pre_call_resets,
12915               Java_Dynamic_Call( meth ),
12916               call_epilog,
12917               post_call_FPU );
12918   ins_pipe( pipe_slow );
12919   ins_alignment(4);
12920 %}
12921 
12922 // Call Runtime Instruction
12923 instruct CallRuntimeDirect(method meth) %{
12924   match(CallRuntime );
12925   effect(USE meth);
12926 
12927   ins_cost(300);
12928   format %{ "CALL,runtime " %}
12929   opcode(0xE8); /* E8 cd */
12930   // Use FFREEs to clear entries in float stack
12931   ins_encode( pre_call_resets,
12932               FFree_Float_Stack_All,
12933               Java_To_Runtime( meth ),
12934               post_call_FPU );
12935   ins_pipe( pipe_slow );
12936 %}
12937 
12938 // Call runtime without safepoint
12939 instruct CallLeafDirect(method meth) %{
12940   match(CallLeaf);
12941   effect(USE meth);
12942 
12943   ins_cost(300);
12944   format %{ "CALL_LEAF,runtime " %}
12945   opcode(0xE8); /* E8 cd */
12946   ins_encode( pre_call_resets,
12947               FFree_Float_Stack_All,
12948               Java_To_Runtime( meth ),
12949               Verify_FPU_For_Leaf, post_call_FPU );
12950   ins_pipe( pipe_slow );
12951 %}
12952 
12953 instruct CallLeafNoFPDirect(method meth) %{
12954   match(CallLeafNoFP);
12955   effect(USE meth);
12956 
12957   ins_cost(300);
12958   format %{ "CALL_LEAF_NOFP,runtime " %}
12959   opcode(0xE8); /* E8 cd */
12960   ins_encode(Java_To_Runtime(meth));
12961   ins_pipe( pipe_slow );
12962 %}
12963 
12964 
12965 // Return Instruction
12966 // Remove the return address & jump to it.
12967 instruct Ret() %{
12968   match(Return);
12969   format %{ "RET" %}
12970   opcode(0xC3);
12971   ins_encode(OpcP);
12972   ins_pipe( pipe_jmp );
12973 %}
12974 
12975 // Tail Call; Jump from runtime stub to Java code.
12976 // Also known as an 'interprocedural jump'.
12977 // Target of jump will eventually return to caller.
12978 // TailJump below removes the return address.
12979 instruct TailCalljmpInd(eRegP_no_EBP jump_target, eBXRegP method_oop) %{
12980   match(TailCall jump_target method_oop );
12981   ins_cost(300);
12982   format %{ "JMP    $jump_target \t# EBX holds method oop" %}
12983   opcode(0xFF, 0x4);  /* Opcode FF /4 */
12984   ins_encode( OpcP, RegOpc(jump_target) );
12985   ins_pipe( pipe_jmp );
12986 %}
12987 
12988 
12989 // Tail Jump; remove the return address; jump to target.
12990 // TailCall above leaves the return address around.
12991 instruct tailjmpInd(eRegP_no_EBP jump_target, eAXRegP ex_oop) %{
12992   match( TailJump jump_target ex_oop );
12993   ins_cost(300);
12994   format %{ "POP    EDX\t# pop return address into dummy\n\t"
12995             "JMP    $jump_target " %}
12996   opcode(0xFF, 0x4);  /* Opcode FF /4 */
12997   ins_encode( enc_pop_rdx,
12998               OpcP, RegOpc(jump_target) );
12999   ins_pipe( pipe_jmp );
13000 %}
13001 
13002 // Create exception oop: created by stack-crawling runtime code.
13003 // Created exception is now available to this handler, and is setup
13004 // just prior to jumping to this handler.  No code emitted.
13005 instruct CreateException( eAXRegP ex_oop )
13006 %{
13007   match(Set ex_oop (CreateEx));
13008 
13009   size(0);
13010   // use the following format syntax
13011   format %{ "# exception oop is in EAX; no code emitted" %}
13012   ins_encode();
13013   ins_pipe( empty );
13014 %}
13015 
13016 
13017 // Rethrow exception:
13018 // The exception oop will come in the first argument position.
13019 // Then JUMP (not call) to the rethrow stub code.
13020 instruct RethrowException()
13021 %{
13022   match(Rethrow);
13023 
13024   // use the following format syntax
13025   format %{ "JMP    rethrow_stub" %}
13026   ins_encode(enc_rethrow);
13027   ins_pipe( pipe_jmp );
13028 %}
13029 
13030 // inlined locking and unlocking
13031 
13032 
13033 instruct cmpFastLock( eFlagsReg cr, eRegP object, eBXRegP box, eAXRegI tmp, eRegP scr) %{
13034   match( Set cr (FastLock object box) );
13035   effect( TEMP tmp, TEMP scr, USE_KILL box );
13036   ins_cost(300);
13037   format %{ "FASTLOCK $object,$box\t! kills $box,$tmp,$scr" %}
13038   ins_encode( Fast_Lock(object,box,tmp,scr) );
13039   ins_pipe( pipe_slow );
13040 %}
13041 
13042 instruct cmpFastUnlock( eFlagsReg cr, eRegP object, eAXRegP box, eRegP tmp ) %{
13043   match( Set cr (FastUnlock object box) );
13044   effect( TEMP tmp, USE_KILL box );
13045   ins_cost(300);
13046   format %{ "FASTUNLOCK $object,$box\t! kills $box,$tmp" %}
13047   ins_encode( Fast_Unlock(object,box,tmp) );
13048   ins_pipe( pipe_slow );
13049 %}
13050 
13051 
13052 
13053 // ============================================================================
13054 // Safepoint Instruction
13055 instruct safePoint_poll(eFlagsReg cr) %{
13056   match(SafePoint);
13057   effect(KILL cr);
13058 
13059   // TODO-FIXME: we currently poll at offset 0 of the safepoint polling page.
13060   // On SPARC that might be acceptable as we can generate the address with
13061   // just a sethi, saving an or.  By polling at offset 0 we can end up
13062   // putting additional pressure on the index-0 in the D$.  Because of
13063   // alignment (just like the situation at hand) the lower indices tend
13064   // to see more traffic.  It'd be better to change the polling address
13065   // to offset 0 of the last $line in the polling page.
13066 
13067   format %{ "TSTL   #polladdr,EAX\t! Safepoint: poll for GC" %}
13068   ins_cost(125);
13069   size(6) ;
13070   ins_encode( Safepoint_Poll() );
13071   ins_pipe( ialu_reg_mem );
13072 %}
13073 
13074 
13075 // ============================================================================
13076 // This name is KNOWN by the ADLC and cannot be changed.
13077 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type
13078 // for this guy.
13079 instruct tlsLoadP(eRegP dst, eFlagsReg cr) %{
13080   match(Set dst (ThreadLocal));
13081   effect(DEF dst, KILL cr);
13082 
13083   format %{ "MOV    $dst, Thread::current()" %}
13084   ins_encode %{
13085     Register dstReg = as_Register($dst$$reg);
13086     __ get_thread(dstReg);
13087   %}
13088   ins_pipe( ialu_reg_fat );
13089 %}
13090 
13091 
13092 
13093 //----------PEEPHOLE RULES-----------------------------------------------------
13094 // These must follow all instruction definitions as they use the names
13095 // defined in the instructions definitions.
13096 //
13097 // peepmatch ( root_instr_name [preceding_instruction]* );
13098 //
13099 // peepconstraint %{
13100 // (instruction_number.operand_name relational_op instruction_number.operand_name
13101 //  [, ...] );
13102 // // instruction numbers are zero-based using left to right order in peepmatch
13103 //
13104 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
13105 // // provide an instruction_number.operand_name for each operand that appears
13106 // // in the replacement instruction's match rule
13107 //
13108 // ---------VM FLAGS---------------------------------------------------------
13109 //
13110 // All peephole optimizations can be turned off using -XX:-OptoPeephole
13111 //
13112 // Each peephole rule is given an identifying number starting with zero and
13113 // increasing by one in the order seen by the parser.  An individual peephole
13114 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
13115 // on the command-line.
13116 //
13117 // ---------CURRENT LIMITATIONS----------------------------------------------
13118 //
13119 // Only match adjacent instructions in same basic block
13120 // Only equality constraints
13121 // Only constraints between operands, not (0.dest_reg == EAX_enc)
13122 // Only one replacement instruction
13123 //
13124 // ---------EXAMPLE----------------------------------------------------------
13125 //
13126 // // pertinent parts of existing instructions in architecture description
13127 // instruct movI(rRegI dst, rRegI src) %{
13128 //   match(Set dst (CopyI src));
13129 // %}
13130 //
13131 // instruct incI_eReg(rRegI dst, immI1 src, eFlagsReg cr) %{
13132 //   match(Set dst (AddI dst src));
13133 //   effect(KILL cr);
13134 // %}
13135 //
13136 // // Change (inc mov) to lea
13137 // peephole %{
13138 //   // increment preceeded by register-register move
13139 //   peepmatch ( incI_eReg movI );
13140 //   // require that the destination register of the increment
13141 //   // match the destination register of the move
13142 //   peepconstraint ( 0.dst == 1.dst );
13143 //   // construct a replacement instruction that sets
13144 //   // the destination to ( move's source register + one )
13145 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13146 // %}
13147 //
13148 // Implementation no longer uses movX instructions since
13149 // machine-independent system no longer uses CopyX nodes.
13150 //
13151 // peephole %{
13152 //   peepmatch ( incI_eReg movI );
13153 //   peepconstraint ( 0.dst == 1.dst );
13154 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13155 // %}
13156 //
13157 // peephole %{
13158 //   peepmatch ( decI_eReg movI );
13159 //   peepconstraint ( 0.dst == 1.dst );
13160 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13161 // %}
13162 //
13163 // peephole %{
13164 //   peepmatch ( addI_eReg_imm movI );
13165 //   peepconstraint ( 0.dst == 1.dst );
13166 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13167 // %}
13168 //
13169 // peephole %{
13170 //   peepmatch ( addP_eReg_imm movP );
13171 //   peepconstraint ( 0.dst == 1.dst );
13172 //   peepreplace ( leaP_eReg_immI( 0.dst 1.src 0.src ) );
13173 // %}
13174 
13175 // // Change load of spilled value to only a spill
13176 // instruct storeI(memory mem, rRegI src) %{
13177 //   match(Set mem (StoreI mem src));
13178 // %}
13179 //
13180 // instruct loadI(rRegI dst, memory mem) %{
13181 //   match(Set dst (LoadI mem));
13182 // %}
13183 //
13184 peephole %{
13185   peepmatch ( loadI storeI );
13186   peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
13187   peepreplace ( storeI( 1.mem 1.mem 1.src ) );
13188 %}
13189 
13190 //----------SMARTSPILL RULES---------------------------------------------------
13191 // These must follow all instruction definitions as they use the names
13192 // defined in the instructions definitions.