1 //
   2 // Copyright (c) 1997, 2012, 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 // Special Registers
  78 reg_def EFLAGS(SOC, SOC, 0, 8, VMRegImpl::Bad());
  79 
  80 // Float registers.  We treat TOS/FPR0 special.  It is invisible to the
  81 // allocator, and only shows up in the encodings.
  82 reg_def FPR0L( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
  83 reg_def FPR0H( SOC, SOC, Op_RegF, 0, VMRegImpl::Bad());
  84 // Ok so here's the trick FPR1 is really st(0) except in the midst
  85 // of emission of assembly for a machnode. During the emission the fpu stack
  86 // is pushed making FPR1 == st(1) temporarily. However at any safepoint
  87 // the stack will not have this element so FPR1 == st(0) from the
  88 // oopMap viewpoint. This same weirdness with numbering causes
  89 // instruction encoding to have to play games with the register
  90 // encode to correct for this 0/1 issue. See MachSpillCopyNode::implementation
  91 // where it does flt->flt moves to see an example
  92 //
  93 reg_def FPR1L( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg());
  94 reg_def FPR1H( SOC, SOC, Op_RegF, 1, as_FloatRegister(0)->as_VMReg()->next());
  95 reg_def FPR2L( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg());
  96 reg_def FPR2H( SOC, SOC, Op_RegF, 2, as_FloatRegister(1)->as_VMReg()->next());
  97 reg_def FPR3L( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg());
  98 reg_def FPR3H( SOC, SOC, Op_RegF, 3, as_FloatRegister(2)->as_VMReg()->next());
  99 reg_def FPR4L( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg());
 100 reg_def FPR4H( SOC, SOC, Op_RegF, 4, as_FloatRegister(3)->as_VMReg()->next());
 101 reg_def FPR5L( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg());
 102 reg_def FPR5H( SOC, SOC, Op_RegF, 5, as_FloatRegister(4)->as_VMReg()->next());
 103 reg_def FPR6L( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg());
 104 reg_def FPR6H( SOC, SOC, Op_RegF, 6, as_FloatRegister(5)->as_VMReg()->next());
 105 reg_def FPR7L( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg());
 106 reg_def FPR7H( SOC, SOC, Op_RegF, 7, as_FloatRegister(6)->as_VMReg()->next());
 107 
 108 // XMM registers.  128-bit registers or 4 words each, labeled a-d.
 109 // Word a in each register holds a Float, words ab hold a Double.
 110 // We currently do not use the SIMD capabilities, so registers cd
 111 // are unused at the moment.
 112 reg_def XMM0a( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg());
 113 reg_def XMM0b( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next());
 114 reg_def XMM1a( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg());
 115 reg_def XMM1b( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next());
 116 reg_def XMM2a( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg());
 117 reg_def XMM2b( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next());
 118 reg_def XMM3a( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg());
 119 reg_def XMM3b( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next());
 120 reg_def XMM4a( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg());
 121 reg_def XMM4b( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next());
 122 reg_def XMM5a( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg());
 123 reg_def XMM5b( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next());
 124 reg_def XMM6a( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg());
 125 reg_def XMM6b( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next());
 126 reg_def XMM7a( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg());
 127 reg_def XMM7b( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next());
 128 
 129 // Specify priority of register selection within phases of register
 130 // allocation.  Highest priority is first.  A useful heuristic is to
 131 // give registers a low priority when they are required by machine
 132 // instructions, like EAX and EDX.  Registers which are used as
 133 // pairs must fall on an even boundary (witness the FPR#L's in this list).
 134 // For the Intel integer registers, the equivalent Long pairs are
 135 // EDX:EAX, EBX:ECX, and EDI:EBP.
 136 alloc_class chunk0( ECX,   EBX,   EBP,   EDI,   EAX,   EDX,   ESI, ESP,
 137                     FPR0L, FPR0H, FPR1L, FPR1H, FPR2L, FPR2H,
 138                     FPR3L, FPR3H, FPR4L, FPR4H, FPR5L, FPR5H,
 139                     FPR6L, FPR6H, FPR7L, FPR7H );
 140 
 141 alloc_class chunk1( XMM0a, XMM0b,
 142                     XMM1a, XMM1b,
 143                     XMM2a, XMM2b,
 144                     XMM3a, XMM3b,
 145                     XMM4a, XMM4b,
 146                     XMM5a, XMM5b,
 147                     XMM6a, XMM6b,
 148                     XMM7a, XMM7b, EFLAGS);
 149 
 150 
 151 //----------Architecture Description Register Classes--------------------------
 152 // Several register classes are automatically defined based upon information in
 153 // this architecture description.
 154 // 1) reg_class inline_cache_reg           ( /* as def'd in frame section */ )
 155 // 2) reg_class compiler_method_oop_reg    ( /* as def'd in frame section */ )
 156 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
 157 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
 158 //
 159 // Class for all registers
 160 reg_class any_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX, ESP);
 161 // Class for general registers
 162 reg_class e_reg(EAX, EDX, EBP, EDI, ESI, ECX, EBX);
 163 // Class for general registers which may be used for implicit null checks on win95
 164 // Also safe for use by tailjump. We don't want to allocate in rbp,
 165 reg_class e_reg_no_rbp(EAX, EDX, EDI, ESI, ECX, EBX);
 166 // Class of "X" registers
 167 reg_class x_reg(EBX, ECX, EDX, EAX);
 168 // Class of registers that can appear in an address with no offset.
 169 // EBP and ESP require an extra instruction byte for zero offset.
 170 // Used in fast-unlock
 171 reg_class p_reg(EDX, EDI, ESI, EBX);
 172 // Class for general registers not including ECX
 173 reg_class ncx_reg(EAX, EDX, EBP, EDI, ESI, EBX);
 174 // Class for general registers not including EAX
 175 reg_class nax_reg(EDX, EDI, ESI, ECX, EBX);
 176 // Class for general registers not including EAX or EBX.
 177 reg_class nabx_reg(EDX, EDI, ESI, ECX, EBP);
 178 // Class of EAX (for multiply and divide operations)
 179 reg_class eax_reg(EAX);
 180 // Class of EBX (for atomic add)
 181 reg_class ebx_reg(EBX);
 182 // Class of ECX (for shift and JCXZ operations and cmpLTMask)
 183 reg_class ecx_reg(ECX);
 184 // Class of EDX (for multiply and divide operations)
 185 reg_class edx_reg(EDX);
 186 // Class of EDI (for synchronization)
 187 reg_class edi_reg(EDI);
 188 // Class of ESI (for synchronization)
 189 reg_class esi_reg(ESI);
 190 // Singleton class for interpreter's stack pointer
 191 reg_class ebp_reg(EBP);
 192 // Singleton class for stack pointer
 193 reg_class sp_reg(ESP);
 194 // Singleton class for instruction pointer
 195 // reg_class ip_reg(EIP);
 196 // Singleton class for condition codes
 197 reg_class int_flags(EFLAGS);
 198 // Class of integer register pairs
 199 reg_class long_reg( EAX,EDX, ECX,EBX, EBP,EDI );
 200 // Class of integer register pairs that aligns with calling convention
 201 reg_class eadx_reg( EAX,EDX );
 202 reg_class ebcx_reg( ECX,EBX );
 203 // Not AX or DX, used in divides
 204 reg_class nadx_reg( EBX,ECX,ESI,EDI,EBP );
 205 
 206 // Floating point registers.  Notice FPR0 is not a choice.
 207 // FPR0 is not ever allocated; we use clever encodings to fake
 208 // a 2-address instructions out of Intels FP stack.
 209 reg_class flt_reg( FPR1L,FPR2L,FPR3L,FPR4L,FPR5L,FPR6L,FPR7L );
 210 
 211 // make a register class for SSE registers
 212 reg_class xmm_reg(XMM0a, XMM1a, XMM2a, XMM3a, XMM4a, XMM5a, XMM6a, XMM7a);
 213 
 214 // make a double register class for SSE2 registers
 215 reg_class xdb_reg(XMM0a,XMM0b, XMM1a,XMM1b, XMM2a,XMM2b, XMM3a,XMM3b,
 216                   XMM4a,XMM4b, XMM5a,XMM5b, XMM6a,XMM6b, XMM7a,XMM7b );
 217 
 218 reg_class dbl_reg( FPR1L,FPR1H, FPR2L,FPR2H, FPR3L,FPR3H,
 219                    FPR4L,FPR4H, FPR5L,FPR5H, FPR6L,FPR6H,
 220                    FPR7L,FPR7H );
 221 
 222 reg_class flt_reg0( FPR1L );
 223 reg_class dbl_reg0( FPR1L,FPR1H );
 224 reg_class dbl_reg1( FPR2L,FPR2H );
 225 reg_class dbl_notreg0( FPR2L,FPR2H, FPR3L,FPR3H, FPR4L,FPR4H,
 226                        FPR5L,FPR5H, FPR6L,FPR6H, FPR7L,FPR7H );
 227 
 228 // XMM6 and XMM7 could be used as temporary registers for long, float and
 229 // double values for SSE2.
 230 reg_class xdb_reg6( XMM6a,XMM6b );
 231 reg_class xdb_reg7( XMM7a,XMM7b );
 232 %}
 233 
 234 
 235 //----------SOURCE BLOCK-------------------------------------------------------
 236 // This is a block of C++ code which provides values, functions, and
 237 // definitions necessary in the rest of the architecture description
 238 source_hpp %{
 239 // Must be visible to the DFA in dfa_x86_32.cpp
 240 extern bool is_operand_hi32_zero(Node* n);
 241 %}
 242 
 243 source %{
 244 #define   RELOC_IMM32    Assembler::imm_operand
 245 #define   RELOC_DISP32   Assembler::disp32_operand
 246 
 247 #define __ _masm.
 248 
 249 // How to find the high register of a Long pair, given the low register
 250 #define   HIGH_FROM_LOW(x) ((x)+2)
 251 
 252 // These masks are used to provide 128-bit aligned bitmasks to the XMM
 253 // instructions, to allow sign-masking or sign-bit flipping.  They allow
 254 // fast versions of NegF/NegD and AbsF/AbsD.
 255 
 256 // Note: 'double' and 'long long' have 32-bits alignment on x86.
 257 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
 258   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
 259   // of 128-bits operands for SSE instructions.
 260   jlong *operand = (jlong*)(((uintptr_t)adr)&((uintptr_t)(~0xF)));
 261   // Store the value to a 128-bits operand.
 262   operand[0] = lo;
 263   operand[1] = hi;
 264   return operand;
 265 }
 266 
 267 // Buffer for 128-bits masks used by SSE instructions.
 268 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
 269 
 270 // Static initialization during VM startup.
 271 static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
 272 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
 273 static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
 274 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
 275 
 276 // Offset hacking within calls.
 277 static int pre_call_FPU_size() {
 278   if (Compile::current()->in_24_bit_fp_mode())
 279     return 6; // fldcw
 280   return 0;
 281 }
 282 
 283 static int preserve_SP_size() {
 284   return 2;  // op, rm(reg/reg)
 285 }
 286 
 287 // !!!!! Special hack to get all type of calls to specify the byte offset
 288 //       from the start of the call to the point where the return address
 289 //       will point.
 290 int MachCallStaticJavaNode::ret_addr_offset() {
 291   int offset = 5 + pre_call_FPU_size();  // 5 bytes from start of call to where return address points
 292   if (_method_handle_invoke)
 293     offset += preserve_SP_size();
 294   return offset;
 295 }
 296 
 297 int MachCallDynamicJavaNode::ret_addr_offset() {
 298   return 10 + pre_call_FPU_size();  // 10 bytes from start of call to where return address points
 299 }
 300 
 301 static int sizeof_FFree_Float_Stack_All = -1;
 302 
 303 int MachCallRuntimeNode::ret_addr_offset() {
 304   assert(sizeof_FFree_Float_Stack_All != -1, "must have been emitted already");
 305   return sizeof_FFree_Float_Stack_All + 5 + pre_call_FPU_size();
 306 }
 307 
 308 // Indicate if the safepoint node needs the polling page as an input.
 309 // Since x86 does have absolute addressing, it doesn't.
 310 bool SafePointNode::needs_polling_address_input() {
 311   return false;
 312 }
 313 
 314 //
 315 // Compute padding required for nodes which need alignment
 316 //
 317 
 318 // The address of the call instruction needs to be 4-byte aligned to
 319 // ensure that it does not span a cache line so that it can be patched.
 320 int CallStaticJavaDirectNode::compute_padding(int current_offset) const {
 321   current_offset += pre_call_FPU_size();  // skip fldcw, if any
 322   current_offset += 1;      // skip call opcode byte
 323   return round_to(current_offset, alignment_required()) - current_offset;
 324 }
 325 
 326 // The address of the call instruction needs to be 4-byte aligned to
 327 // ensure that it does not span a cache line so that it can be patched.
 328 int CallStaticJavaHandleNode::compute_padding(int current_offset) const {
 329   current_offset += pre_call_FPU_size();  // skip fldcw, if any
 330   current_offset += preserve_SP_size();   // skip mov rbp, rsp
 331   current_offset += 1;      // skip call opcode byte
 332   return round_to(current_offset, alignment_required()) - current_offset;
 333 }
 334 
 335 // The address of the call instruction needs to be 4-byte aligned to
 336 // ensure that it does not span a cache line so that it can be patched.
 337 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const {
 338   current_offset += pre_call_FPU_size();  // skip fldcw, if any
 339   current_offset += 5;      // skip MOV instruction
 340   current_offset += 1;      // skip call opcode byte
 341   return round_to(current_offset, alignment_required()) - current_offset;
 342 }
 343 
 344 // EMIT_RM()
 345 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) {
 346   unsigned char c = (unsigned char)((f1 << 6) | (f2 << 3) | f3);
 347   cbuf.insts()->emit_int8(c);
 348 }
 349 
 350 // EMIT_CC()
 351 void emit_cc(CodeBuffer &cbuf, int f1, int f2) {
 352   unsigned char c = (unsigned char)( f1 | f2 );
 353   cbuf.insts()->emit_int8(c);
 354 }
 355 
 356 // EMIT_OPCODE()
 357 void emit_opcode(CodeBuffer &cbuf, int code) {
 358   cbuf.insts()->emit_int8((unsigned char) code);
 359 }
 360 
 361 // EMIT_OPCODE() w/ relocation information
 362 void emit_opcode(CodeBuffer &cbuf, int code, relocInfo::relocType reloc, int offset = 0) {
 363   cbuf.relocate(cbuf.insts_mark() + offset, reloc);
 364   emit_opcode(cbuf, code);
 365 }
 366 
 367 // EMIT_D8()
 368 void emit_d8(CodeBuffer &cbuf, int d8) {
 369   cbuf.insts()->emit_int8((unsigned char) d8);
 370 }
 371 
 372 // EMIT_D16()
 373 void emit_d16(CodeBuffer &cbuf, int d16) {
 374   cbuf.insts()->emit_int16(d16);
 375 }
 376 
 377 // EMIT_D32()
 378 void emit_d32(CodeBuffer &cbuf, int d32) {
 379   cbuf.insts()->emit_int32(d32);
 380 }
 381 
 382 // emit 32 bit value and construct relocation entry from relocInfo::relocType
 383 void emit_d32_reloc(CodeBuffer &cbuf, int d32, relocInfo::relocType reloc,
 384         int format) {
 385   cbuf.relocate(cbuf.insts_mark(), reloc, format);
 386   cbuf.insts()->emit_int32(d32);
 387 }
 388 
 389 // emit 32 bit value and construct relocation entry from RelocationHolder
 390 void emit_d32_reloc(CodeBuffer &cbuf, int d32, RelocationHolder const& rspec,
 391         int format) {
 392 #ifdef ASSERT
 393   if (rspec.reloc()->type() == relocInfo::oop_type && d32 != 0 && d32 != (int)Universe::non_oop_word()) {
 394     assert(oop(d32)->is_oop() && (ScavengeRootsInCode || !oop(d32)->is_scavengable()), "cannot embed scavengable oops in code");
 395   }
 396 #endif
 397   cbuf.relocate(cbuf.insts_mark(), rspec, format);
 398   cbuf.insts()->emit_int32(d32);
 399 }
 400 
 401 // Access stack slot for load or store
 402 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp) {
 403   emit_opcode( cbuf, opcode );               // (e.g., FILD   [ESP+src])
 404   if( -128 <= disp && disp <= 127 ) {
 405     emit_rm( cbuf, 0x01, rm_field, ESP_enc );  // R/M byte
 406     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
 407     emit_d8 (cbuf, disp);     // Displacement  // R/M byte
 408   } else {
 409     emit_rm( cbuf, 0x02, rm_field, ESP_enc );  // R/M byte
 410     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);    // SIB byte
 411     emit_d32(cbuf, disp);     // Displacement  // R/M byte
 412   }
 413 }
 414 
 415    // eRegI ereg, memory mem) %{    // emit_reg_mem
 416 void encode_RegMem( CodeBuffer &cbuf, int reg_encoding, int base, int index, int scale, int displace, bool displace_is_oop ) {
 417   // There is no index & no scale, use form without SIB byte
 418   if ((index == 0x4) &&
 419       (scale == 0) && (base != ESP_enc)) {
 420     // If no displacement, mode is 0x0; unless base is [EBP]
 421     if ( (displace == 0) && (base != EBP_enc) ) {
 422       emit_rm(cbuf, 0x0, reg_encoding, base);
 423     }
 424     else {                    // If 8-bit displacement, mode 0x1
 425       if ((displace >= -128) && (displace <= 127)
 426           && !(displace_is_oop) ) {
 427         emit_rm(cbuf, 0x1, reg_encoding, base);
 428         emit_d8(cbuf, displace);
 429       }
 430       else {                  // If 32-bit displacement
 431         if (base == -1) { // Special flag for absolute address
 432           emit_rm(cbuf, 0x0, reg_encoding, 0x5);
 433           // (manual lies; no SIB needed here)
 434           if ( displace_is_oop ) {
 435             emit_d32_reloc(cbuf, displace, relocInfo::oop_type, 1);
 436           } else {
 437             emit_d32      (cbuf, displace);
 438           }
 439         }
 440         else {                // Normal base + offset
 441           emit_rm(cbuf, 0x2, reg_encoding, base);
 442           if ( displace_is_oop ) {
 443             emit_d32_reloc(cbuf, displace, relocInfo::oop_type, 1);
 444           } else {
 445             emit_d32      (cbuf, displace);
 446           }
 447         }
 448       }
 449     }
 450   }
 451   else {                      // Else, encode with the SIB byte
 452     // If no displacement, mode is 0x0; unless base is [EBP]
 453     if (displace == 0 && (base != EBP_enc)) {  // If no displacement
 454       emit_rm(cbuf, 0x0, reg_encoding, 0x4);
 455       emit_rm(cbuf, scale, index, base);
 456     }
 457     else {                    // If 8-bit displacement, mode 0x1
 458       if ((displace >= -128) && (displace <= 127)
 459           && !(displace_is_oop) ) {
 460         emit_rm(cbuf, 0x1, reg_encoding, 0x4);
 461         emit_rm(cbuf, scale, index, base);
 462         emit_d8(cbuf, displace);
 463       }
 464       else {                  // If 32-bit displacement
 465         if (base == 0x04 ) {
 466           emit_rm(cbuf, 0x2, reg_encoding, 0x4);
 467           emit_rm(cbuf, scale, index, 0x04);
 468         } else {
 469           emit_rm(cbuf, 0x2, reg_encoding, 0x4);
 470           emit_rm(cbuf, scale, index, base);
 471         }
 472         if ( displace_is_oop ) {
 473           emit_d32_reloc(cbuf, displace, relocInfo::oop_type, 1);
 474         } else {
 475           emit_d32      (cbuf, displace);
 476         }
 477       }
 478     }
 479   }
 480 }
 481 
 482 
 483 void encode_Copy( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
 484   if( dst_encoding == src_encoding ) {
 485     // reg-reg copy, use an empty encoding
 486   } else {
 487     emit_opcode( cbuf, 0x8B );
 488     emit_rm(cbuf, 0x3, dst_encoding, src_encoding );
 489   }
 490 }
 491 
 492 void emit_cmpfp_fixup(MacroAssembler& _masm) {
 493   Label exit;
 494   __ jccb(Assembler::noParity, exit);
 495   __ pushf();
 496   //
 497   // comiss/ucomiss instructions set ZF,PF,CF flags and
 498   // zero OF,AF,SF for NaN values.
 499   // Fixup flags by zeroing ZF,PF so that compare of NaN
 500   // values returns 'less than' result (CF is set).
 501   // Leave the rest of flags unchanged.
 502   //
 503   //    7 6 5 4 3 2 1 0
 504   //   |S|Z|r|A|r|P|r|C|  (r - reserved bit)
 505   //    0 0 1 0 1 0 1 1   (0x2B)
 506   //
 507   __ andl(Address(rsp, 0), 0xffffff2b);
 508   __ popf();
 509   __ bind(exit);
 510 }
 511 
 512 void emit_cmpfp3(MacroAssembler& _masm, Register dst) {
 513   Label done;
 514   __ movl(dst, -1);
 515   __ jcc(Assembler::parity, done);
 516   __ jcc(Assembler::below, done);
 517   __ setb(Assembler::notEqual, dst);
 518   __ movzbl(dst, dst);
 519   __ bind(done);
 520 }
 521 
 522 
 523 //=============================================================================
 524 const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
 525 
 526 int Compile::ConstantTable::calculate_table_base_offset() const {
 527   return 0;  // absolute addressing, no offset
 528 }
 529 
 530 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
 531   // Empty encoding
 532 }
 533 
 534 uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
 535   return 0;
 536 }
 537 
 538 #ifndef PRODUCT
 539 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
 540   st->print("# MachConstantBaseNode (empty encoding)");
 541 }
 542 #endif
 543 
 544 
 545 //=============================================================================
 546 #ifndef PRODUCT
 547 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
 548   Compile* C = ra_->C;
 549 
 550   int framesize = C->frame_slots() << LogBytesPerInt;
 551   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 552   // Remove wordSize for return addr which is already pushed.
 553   framesize -= wordSize;
 554 
 555   if (C->need_stack_bang(framesize)) {
 556     framesize -= wordSize;
 557     st->print("# stack bang");
 558     st->print("\n\t");
 559     st->print("PUSH   EBP\t# Save EBP");
 560     if (framesize) {
 561       st->print("\n\t");
 562       st->print("SUB    ESP, #%d\t# Create frame",framesize);
 563     }
 564   } else {
 565     st->print("SUB    ESP, #%d\t# Create frame",framesize);
 566     st->print("\n\t");
 567     framesize -= wordSize;
 568     st->print("MOV    [ESP + #%d], EBP\t# Save EBP",framesize);
 569   }
 570 
 571   if (VerifyStackAtCalls) {
 572     st->print("\n\t");
 573     framesize -= wordSize;
 574     st->print("MOV    [ESP + #%d], 0xBADB100D\t# Majik cookie for stack depth check",framesize);
 575   }
 576 
 577   if( C->in_24_bit_fp_mode() ) {
 578     st->print("\n\t");
 579     st->print("FLDCW  \t# load 24 bit fpu control word");
 580   }
 581   if (UseSSE >= 2 && VerifyFPU) {
 582     st->print("\n\t");
 583     st->print("# verify FPU stack (must be clean on entry)");
 584   }
 585 
 586 #ifdef ASSERT
 587   if (VerifyStackAtCalls) {
 588     st->print("\n\t");
 589     st->print("# stack alignment check");
 590   }
 591 #endif
 592   st->cr();
 593 }
 594 #endif
 595 
 596 
 597 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
 598   Compile* C = ra_->C;
 599   MacroAssembler _masm(&cbuf);
 600 
 601   int framesize = C->frame_slots() << LogBytesPerInt;
 602 
 603   __ verified_entry(framesize, C->need_stack_bang(framesize), C->in_24_bit_fp_mode());
 604 
 605   C->set_frame_complete(cbuf.insts_size());
 606 
 607   if (C->has_mach_constant_base_node()) {
 608     // NOTE: We set the table base offset here because users might be
 609     // emitted before MachConstantBaseNode.
 610     Compile::ConstantTable& constant_table = C->constant_table();
 611     constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
 612   }
 613 }
 614 
 615 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
 616   return MachNode::size(ra_); // too many variables; just compute it the hard way
 617 }
 618 
 619 int MachPrologNode::reloc() const {
 620   return 0; // a large enough number
 621 }
 622 
 623 //=============================================================================
 624 #ifndef PRODUCT
 625 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
 626   Compile *C = ra_->C;
 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   if( C->in_24_bit_fp_mode() ) {
 633     st->print("FLDCW  standard control word");
 634     st->cr(); st->print("\t");
 635   }
 636   if( framesize ) {
 637     st->print("ADD    ESP,%d\t# Destroy frame",framesize);
 638     st->cr(); st->print("\t");
 639   }
 640   st->print_cr("POPL   EBP"); st->print("\t");
 641   if( do_polling() && C->is_method_compilation() ) {
 642     st->print("TEST   PollPage,EAX\t! Poll Safepoint");
 643     st->cr(); st->print("\t");
 644   }
 645 }
 646 #endif
 647 
 648 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
 649   Compile *C = ra_->C;
 650 
 651   // If method set FPU control word, restore to standard control word
 652   if( C->in_24_bit_fp_mode() ) {
 653     MacroAssembler masm(&cbuf);
 654     masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 655   }
 656 
 657   int framesize = C->frame_slots() << LogBytesPerInt;
 658   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 659   // Remove two words for return addr and rbp,
 660   framesize -= 2*wordSize;
 661 
 662   // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
 663 
 664   if( framesize >= 128 ) {
 665     emit_opcode(cbuf, 0x81); // add  SP, #framesize
 666     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
 667     emit_d32(cbuf, framesize);
 668   }
 669   else if( framesize ) {
 670     emit_opcode(cbuf, 0x83); // add  SP, #framesize
 671     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
 672     emit_d8(cbuf, framesize);
 673   }
 674 
 675   emit_opcode(cbuf, 0x58 | EBP_enc);
 676 
 677   if( do_polling() && C->is_method_compilation() ) {
 678     cbuf.relocate(cbuf.insts_end(), relocInfo::poll_return_type, 0);
 679     emit_opcode(cbuf,0x85);
 680     emit_rm(cbuf, 0x0, EAX_enc, 0x5); // EAX
 681     emit_d32(cbuf, (intptr_t)os::get_polling_page());
 682   }
 683 }
 684 
 685 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
 686   Compile *C = ra_->C;
 687   // If method set FPU control word, restore to standard control word
 688   int size = C->in_24_bit_fp_mode() ? 6 : 0;
 689   if( do_polling() && C->is_method_compilation() ) size += 6;
 690 
 691   int framesize = C->frame_slots() << LogBytesPerInt;
 692   assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 693   // Remove two words for return addr and rbp,
 694   framesize -= 2*wordSize;
 695 
 696   size++; // popl rbp,
 697 
 698   if( framesize >= 128 ) {
 699     size += 6;
 700   } else {
 701     size += framesize ? 3 : 0;
 702   }
 703   return size;
 704 }
 705 
 706 int MachEpilogNode::reloc() const {
 707   return 0; // a large enough number
 708 }
 709 
 710 const Pipeline * MachEpilogNode::pipeline() const {
 711   return MachNode::pipeline_class();
 712 }
 713 
 714 int MachEpilogNode::safepoint_offset() const { return 0; }
 715 
 716 //=============================================================================
 717 
 718 enum RC { rc_bad, rc_int, rc_float, rc_xmm, rc_stack };
 719 static enum RC rc_class( OptoReg::Name reg ) {
 720 
 721   if( !OptoReg::is_valid(reg)  ) return rc_bad;
 722   if (OptoReg::is_stack(reg)) return rc_stack;
 723 
 724   VMReg r = OptoReg::as_VMReg(reg);
 725   if (r->is_Register()) return rc_int;
 726   if (r->is_FloatRegister()) {
 727     assert(UseSSE < 2, "shouldn't be used in SSE2+ mode");
 728     return rc_float;
 729   }
 730   assert(r->is_XMMRegister(), "must be");
 731   return rc_xmm;
 732 }
 733 
 734 static int impl_helper( CodeBuffer *cbuf, bool do_size, bool is_load, int offset, int reg,
 735                         int opcode, const char *op_str, int size, outputStream* st ) {
 736   if( cbuf ) {
 737     emit_opcode  (*cbuf, opcode );
 738     encode_RegMem(*cbuf, Matcher::_regEncode[reg], ESP_enc, 0x4, 0, offset, false);
 739 #ifndef PRODUCT
 740   } else if( !do_size ) {
 741     if( size != 0 ) st->print("\n\t");
 742     if( opcode == 0x8B || opcode == 0x89 ) { // MOV
 743       if( is_load ) st->print("%s   %s,[ESP + #%d]",op_str,Matcher::regName[reg],offset);
 744       else          st->print("%s   [ESP + #%d],%s",op_str,offset,Matcher::regName[reg]);
 745     } else { // FLD, FST, PUSH, POP
 746       st->print("%s [ESP + #%d]",op_str,offset);
 747     }
 748 #endif
 749   }
 750   int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
 751   return size+3+offset_size;
 752 }
 753 
 754 // Helper for XMM registers.  Extra opcode bits, limited syntax.
 755 static int impl_x_helper( CodeBuffer *cbuf, bool do_size, bool is_load,
 756                          int offset, int reg_lo, int reg_hi, int size, outputStream* st ) {
 757   if (cbuf) {
 758     MacroAssembler _masm(cbuf);
 759     if (reg_lo+1 == reg_hi) { // double move?
 760       if (is_load) {
 761         __ movdbl(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
 762       } else {
 763         __ movdbl(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
 764       }
 765     } else {
 766       if (is_load) {
 767         __ movflt(as_XMMRegister(Matcher::_regEncode[reg_lo]), Address(rsp, offset));
 768       } else {
 769         __ movflt(Address(rsp, offset), as_XMMRegister(Matcher::_regEncode[reg_lo]));
 770       }
 771     }
 772 #ifndef PRODUCT
 773   } else if (!do_size) {
 774     if (size != 0) st->print("\n\t");
 775     if (reg_lo+1 == reg_hi) { // double move?
 776       if (is_load) st->print("%s %s,[ESP + #%d]",
 777                               UseXmmLoadAndClearUpper ? "MOVSD " : "MOVLPD",
 778                               Matcher::regName[reg_lo], offset);
 779       else         st->print("MOVSD  [ESP + #%d],%s",
 780                               offset, Matcher::regName[reg_lo]);
 781     } else {
 782       if (is_load) st->print("MOVSS  %s,[ESP + #%d]",
 783                               Matcher::regName[reg_lo], offset);
 784       else         st->print("MOVSS  [ESP + #%d],%s",
 785                               offset, Matcher::regName[reg_lo]);
 786     }
 787 #endif
 788   }
 789   int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
 790   // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes.
 791   return size+5+offset_size;
 792 }
 793 
 794 
 795 static int impl_movx_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 796                             int src_hi, int dst_hi, int size, outputStream* st ) {
 797   if (cbuf) {
 798     MacroAssembler _masm(cbuf);
 799     if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
 800       __ movdbl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 801                 as_XMMRegister(Matcher::_regEncode[src_lo]));
 802     } else {
 803       __ movflt(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 804                 as_XMMRegister(Matcher::_regEncode[src_lo]));
 805     }
 806 #ifndef PRODUCT
 807   } else if (!do_size) {
 808     if (size != 0) st->print("\n\t");
 809     if (UseXmmRegToRegMoveAll) {//Use movaps,movapd to move between xmm registers
 810       if (src_lo+1 == src_hi && dst_lo+1 == dst_hi) { // double move?
 811         st->print("MOVAPD %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 812       } else {
 813         st->print("MOVAPS %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 814       }
 815     } else {
 816       if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double move?
 817         st->print("MOVSD  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 818       } else {
 819         st->print("MOVSS  %s,%s",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
 820       }
 821     }
 822 #endif
 823   }
 824   // VEX_2bytes prefix is used if UseAVX > 0, and it takes the same 2 bytes.
 825   // Only MOVAPS SSE prefix uses 1 byte.
 826   int sz = 4;
 827   if (!(src_lo+1 == src_hi && dst_lo+1 == dst_hi) &&
 828       UseXmmRegToRegMoveAll && (UseAVX == 0)) sz = 3;
 829   return size + sz;
 830 }
 831 
 832 static int impl_movgpr2x_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 833                             int src_hi, int dst_hi, int size, outputStream* st ) {
 834   // 32-bit
 835   if (cbuf) {
 836     MacroAssembler _masm(cbuf);
 837     __ movdl(as_XMMRegister(Matcher::_regEncode[dst_lo]),
 838              as_Register(Matcher::_regEncode[src_lo]));
 839 #ifndef PRODUCT
 840   } else if (!do_size) {
 841     st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
 842 #endif
 843   }
 844   return 4;
 845 }
 846 
 847 
 848 static int impl_movx2gpr_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
 849                                  int src_hi, int dst_hi, int size, outputStream* st ) {
 850   // 32-bit
 851   if (cbuf) {
 852     MacroAssembler _masm(cbuf);
 853     __ movdl(as_Register(Matcher::_regEncode[dst_lo]),
 854              as_XMMRegister(Matcher::_regEncode[src_lo]));
 855 #ifndef PRODUCT
 856   } else if (!do_size) {
 857     st->print("movdl   %s, %s\t# spill", Matcher::regName[dst_lo], Matcher::regName[src_lo]);
 858 #endif
 859   }
 860   return 4;
 861 }
 862 
 863 static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int size, outputStream* st ) {
 864   if( cbuf ) {
 865     emit_opcode(*cbuf, 0x8B );
 866     emit_rm    (*cbuf, 0x3, Matcher::_regEncode[dst], Matcher::_regEncode[src] );
 867 #ifndef PRODUCT
 868   } else if( !do_size ) {
 869     if( size != 0 ) st->print("\n\t");
 870     st->print("MOV    %s,%s",Matcher::regName[dst],Matcher::regName[src]);
 871 #endif
 872   }
 873   return size+2;
 874 }
 875 
 876 static int impl_fp_store_helper( CodeBuffer *cbuf, bool do_size, int src_lo, int src_hi, int dst_lo, int dst_hi,
 877                                  int offset, int size, outputStream* st ) {
 878   if( src_lo != FPR1L_num ) {      // Move value to top of FP stack, if not already there
 879     if( cbuf ) {
 880       emit_opcode( *cbuf, 0xD9 );  // FLD (i.e., push it)
 881       emit_d8( *cbuf, 0xC0-1+Matcher::_regEncode[src_lo] );
 882 #ifndef PRODUCT
 883     } else if( !do_size ) {
 884       if( size != 0 ) st->print("\n\t");
 885       st->print("FLD    %s",Matcher::regName[src_lo]);
 886 #endif
 887     }
 888     size += 2;
 889   }
 890 
 891   int st_op = (src_lo != FPR1L_num) ? EBX_num /*store & pop*/ : EDX_num /*store no pop*/;
 892   const char *op_str;
 893   int op;
 894   if( src_lo+1 == src_hi && dst_lo+1 == dst_hi ) { // double store?
 895     op_str = (src_lo != FPR1L_num) ? "FSTP_D" : "FST_D ";
 896     op = 0xDD;
 897   } else {                   // 32-bit store
 898     op_str = (src_lo != FPR1L_num) ? "FSTP_S" : "FST_S ";
 899     op = 0xD9;
 900     assert( !OptoReg::is_valid(src_hi) && !OptoReg::is_valid(dst_hi), "no non-adjacent float-stores" );
 901   }
 902 
 903   return impl_helper(cbuf,do_size,false,offset,st_op,op,op_str,size, st);
 904 }
 905 
 906 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream* st ) const {
 907   // Get registers to move
 908   OptoReg::Name src_second = ra_->get_reg_second(in(1));
 909   OptoReg::Name src_first = ra_->get_reg_first(in(1));
 910   OptoReg::Name dst_second = ra_->get_reg_second(this );
 911   OptoReg::Name dst_first = ra_->get_reg_first(this );
 912 
 913   enum RC src_second_rc = rc_class(src_second);
 914   enum RC src_first_rc = rc_class(src_first);
 915   enum RC dst_second_rc = rc_class(dst_second);
 916   enum RC dst_first_rc = rc_class(dst_first);
 917 
 918   assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );
 919 
 920   // Generate spill code!
 921   int size = 0;
 922 
 923   if( src_first == dst_first && src_second == dst_second )
 924     return size;            // Self copy, no move
 925 
 926   // --------------------------------------
 927   // Check for mem-mem move.  push/pop to move.
 928   if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
 929     if( src_second == dst_first ) { // overlapping stack copy ranges
 930       assert( src_second_rc == rc_stack && dst_second_rc == rc_stack, "we only expect a stk-stk copy here" );
 931       size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
 932       size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
 933       src_second_rc = dst_second_rc = rc_bad;  // flag as already moved the second bits
 934     }
 935     // move low bits
 936     size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),ESI_num,0xFF,"PUSH  ",size, st);
 937     size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),EAX_num,0x8F,"POP   ",size, st);
 938     if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) { // mov second bits
 939       size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),ESI_num,0xFF,"PUSH  ",size, st);
 940       size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),EAX_num,0x8F,"POP   ",size, st);
 941     }
 942     return size;
 943   }
 944 
 945   // --------------------------------------
 946   // Check for integer reg-reg copy
 947   if( src_first_rc == rc_int && dst_first_rc == rc_int )
 948     size = impl_mov_helper(cbuf,do_size,src_first,dst_first,size, st);
 949 
 950   // Check for integer store
 951   if( src_first_rc == rc_int && dst_first_rc == rc_stack )
 952     size = impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first,0x89,"MOV ",size, st);
 953 
 954   // Check for integer load
 955   if( dst_first_rc == rc_int && src_first_rc == rc_stack )
 956     size = impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first,0x8B,"MOV ",size, st);
 957 
 958   // Check for integer reg-xmm reg copy
 959   if( src_first_rc == rc_int && dst_first_rc == rc_xmm ) {
 960     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
 961             "no 64 bit integer-float reg moves" );
 962     return impl_movgpr2x_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
 963   }
 964   // --------------------------------------
 965   // Check for float reg-reg copy
 966   if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
 967     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
 968             (src_first+1 == src_second && dst_first+1 == dst_second), "no non-adjacent float-moves" );
 969     if( cbuf ) {
 970 
 971       // Note the mucking with the register encode to compensate for the 0/1
 972       // indexing issue mentioned in a comment in the reg_def sections
 973       // for FPR registers many lines above here.
 974 
 975       if( src_first != FPR1L_num ) {
 976         emit_opcode  (*cbuf, 0xD9 );           // FLD    ST(i)
 977         emit_d8      (*cbuf, 0xC0+Matcher::_regEncode[src_first]-1 );
 978         emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
 979         emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
 980      } else {
 981         emit_opcode  (*cbuf, 0xDD );           // FST    ST(i)
 982         emit_d8      (*cbuf, 0xD0+Matcher::_regEncode[dst_first]-1 );
 983      }
 984 #ifndef PRODUCT
 985     } else if( !do_size ) {
 986       if( size != 0 ) st->print("\n\t");
 987       if( src_first != FPR1L_num ) st->print("FLD    %s\n\tFSTP   %s",Matcher::regName[src_first],Matcher::regName[dst_first]);
 988       else                      st->print(             "FST    %s",                            Matcher::regName[dst_first]);
 989 #endif
 990     }
 991     return size + ((src_first != FPR1L_num) ? 2+2 : 2);
 992   }
 993 
 994   // Check for float store
 995   if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
 996     return impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,ra_->reg2offset(dst_first),size, st);
 997   }
 998 
 999   // Check for float load
1000   if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
1001     int offset = ra_->reg2offset(src_first);
1002     const char *op_str;
1003     int op;
1004     if( src_first+1 == src_second && dst_first+1 == dst_second ) { // double load?
1005       op_str = "FLD_D";
1006       op = 0xDD;
1007     } else {                   // 32-bit load
1008       op_str = "FLD_S";
1009       op = 0xD9;
1010       assert( src_second_rc == rc_bad && dst_second_rc == rc_bad, "no non-adjacent float-loads" );
1011     }
1012     if( cbuf ) {
1013       emit_opcode  (*cbuf, op );
1014       encode_RegMem(*cbuf, 0x0, ESP_enc, 0x4, 0, offset, false);
1015       emit_opcode  (*cbuf, 0xDD );           // FSTP   ST(i)
1016       emit_d8      (*cbuf, 0xD8+Matcher::_regEncode[dst_first] );
1017 #ifndef PRODUCT
1018     } else if( !do_size ) {
1019       if( size != 0 ) st->print("\n\t");
1020       st->print("%s  ST,[ESP + #%d]\n\tFSTP   %s",op_str, offset,Matcher::regName[dst_first]);
1021 #endif
1022     }
1023     int offset_size = (offset == 0) ? 0 : ((offset <= 127) ? 1 : 4);
1024     return size + 3+offset_size+2;
1025   }
1026 
1027   // Check for xmm reg-reg copy
1028   if( src_first_rc == rc_xmm && dst_first_rc == rc_xmm ) {
1029     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad) ||
1030             (src_first+1 == src_second && dst_first+1 == dst_second),
1031             "no non-adjacent float-moves" );
1032     return impl_movx_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
1033   }
1034 
1035   // Check for xmm reg-integer reg copy
1036   if( src_first_rc == rc_xmm && dst_first_rc == rc_int ) {
1037     assert( (src_second_rc == rc_bad && dst_second_rc == rc_bad),
1038             "no 64 bit float-integer reg moves" );
1039     return impl_movx2gpr_helper(cbuf,do_size,src_first,dst_first,src_second, dst_second, size, st);
1040   }
1041 
1042   // Check for xmm store
1043   if( src_first_rc == rc_xmm && dst_first_rc == rc_stack ) {
1044     return impl_x_helper(cbuf,do_size,false,ra_->reg2offset(dst_first),src_first, src_second, size, st);
1045   }
1046 
1047   // Check for float xmm load
1048   if( dst_first_rc == rc_xmm && src_first_rc == rc_stack ) {
1049     return impl_x_helper(cbuf,do_size,true ,ra_->reg2offset(src_first),dst_first, dst_second, size, st);
1050   }
1051 
1052   // Copy from float reg to xmm reg
1053   if( dst_first_rc == rc_xmm && src_first_rc == rc_float ) {
1054     // copy to the top of stack from floating point reg
1055     // and use LEA to preserve flags
1056     if( cbuf ) {
1057       emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP-8]
1058       emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
1059       emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
1060       emit_d8(*cbuf,0xF8);
1061 #ifndef PRODUCT
1062     } else if( !do_size ) {
1063       if( size != 0 ) st->print("\n\t");
1064       st->print("LEA    ESP,[ESP-8]");
1065 #endif
1066     }
1067     size += 4;
1068 
1069     size = impl_fp_store_helper(cbuf,do_size,src_first,src_second,dst_first,dst_second,0,size, st);
1070 
1071     // Copy from the temp memory to the xmm reg.
1072     size = impl_x_helper(cbuf,do_size,true ,0,dst_first, dst_second, size, st);
1073 
1074     if( cbuf ) {
1075       emit_opcode(*cbuf,0x8D);  // LEA  ESP,[ESP+8]
1076       emit_rm(*cbuf, 0x1, ESP_enc, 0x04);
1077       emit_rm(*cbuf, 0x0, 0x04, ESP_enc);
1078       emit_d8(*cbuf,0x08);
1079 #ifndef PRODUCT
1080     } else if( !do_size ) {
1081       if( size != 0 ) st->print("\n\t");
1082       st->print("LEA    ESP,[ESP+8]");
1083 #endif
1084     }
1085     size += 4;
1086     return size;
1087   }
1088 
1089   assert( size > 0, "missed a case" );
1090 
1091   // --------------------------------------------------------------------
1092   // Check for second bits still needing moving.
1093   if( src_second == dst_second )
1094     return size;               // Self copy; no move
1095   assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );
1096 
1097   // Check for second word int-int move
1098   if( src_second_rc == rc_int && dst_second_rc == rc_int )
1099     return impl_mov_helper(cbuf,do_size,src_second,dst_second,size, st);
1100 
1101   // Check for second word integer store
1102   if( src_second_rc == rc_int && dst_second_rc == rc_stack )
1103     return impl_helper(cbuf,do_size,false,ra_->reg2offset(dst_second),src_second,0x89,"MOV ",size, st);
1104 
1105   // Check for second word integer load
1106   if( dst_second_rc == rc_int && src_second_rc == rc_stack )
1107     return impl_helper(cbuf,do_size,true ,ra_->reg2offset(src_second),dst_second,0x8B,"MOV ",size, st);
1108 
1109 
1110   Unimplemented();
1111 }
1112 
1113 #ifndef PRODUCT
1114 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const {
1115   implementation( NULL, ra_, false, st );
1116 }
1117 #endif
1118 
1119 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1120   implementation( &cbuf, ra_, false, NULL );
1121 }
1122 
1123 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1124   return implementation( NULL, ra_, true, NULL );
1125 }
1126 
1127 
1128 //=============================================================================
1129 #ifndef PRODUCT
1130 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
1131   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1132   int reg = ra_->get_reg_first(this);
1133   st->print("LEA    %s,[ESP + #%d]",Matcher::regName[reg],offset);
1134 }
1135 #endif
1136 
1137 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1138   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1139   int reg = ra_->get_encode(this);
1140   if( offset >= 128 ) {
1141     emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
1142     emit_rm(cbuf, 0x2, reg, 0x04);
1143     emit_rm(cbuf, 0x0, 0x04, ESP_enc);
1144     emit_d32(cbuf, offset);
1145   }
1146   else {
1147     emit_opcode(cbuf, 0x8D);      // LEA  reg,[SP+offset]
1148     emit_rm(cbuf, 0x1, reg, 0x04);
1149     emit_rm(cbuf, 0x0, 0x04, ESP_enc);
1150     emit_d8(cbuf, offset);
1151   }
1152 }
1153 
1154 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1155   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1156   if( offset >= 128 ) {
1157     return 7;
1158   }
1159   else {
1160     return 4;
1161   }
1162 }
1163 
1164 //=============================================================================
1165 
1166 // emit call stub, compiled java to interpreter
1167 void emit_java_to_interp(CodeBuffer &cbuf ) {
1168   // Stub is fixed up when the corresponding call is converted from calling
1169   // compiled code to calling interpreted code.
1170   // mov rbx,0
1171   // jmp -1
1172 
1173   address mark = cbuf.insts_mark();  // get mark within main instrs section
1174 
1175   // Note that the code buffer's insts_mark is always relative to insts.
1176   // That's why we must use the macroassembler to generate a stub.
1177   MacroAssembler _masm(&cbuf);
1178 
1179   address base =
1180   __ start_a_stub(Compile::MAX_stubs_size);
1181   if (base == NULL)  return;  // CodeBuffer::expand failed
1182   // static stub relocation stores the instruction address of the call
1183   __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM32);
1184   // static stub relocation also tags the methodOop in the code-stream.
1185   __ movoop(rbx, (jobject)NULL);  // method is zapped till fixup time
1186   // This is recognized as unresolved by relocs/nativeInst/ic code
1187   __ jump(RuntimeAddress(__ pc()));
1188 
1189   __ end_a_stub();
1190   // Update current stubs pointer and restore insts_end.
1191 }
1192 // size of call stub, compiled java to interpretor
1193 uint size_java_to_interp() {
1194   return 10;  // movl; jmp
1195 }
1196 // relocation entries for call stub, compiled java to interpretor
1197 uint reloc_java_to_interp() {
1198   return 4;  // 3 in emit_java_to_interp + 1 in Java_Static_Call
1199 }
1200 
1201 //=============================================================================
1202 #ifndef PRODUCT
1203 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream* st ) const {
1204   st->print_cr(  "CMP    EAX,[ECX+4]\t# Inline cache check");
1205   st->print_cr("\tJNE    SharedRuntime::handle_ic_miss_stub");
1206   st->print_cr("\tNOP");
1207   st->print_cr("\tNOP");
1208   if( !OptoBreakpoint )
1209     st->print_cr("\tNOP");
1210 }
1211 #endif
1212 
1213 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1214   MacroAssembler masm(&cbuf);
1215 #ifdef ASSERT
1216   uint insts_size = cbuf.insts_size();
1217 #endif
1218   masm.cmpptr(rax, Address(rcx, oopDesc::klass_offset_in_bytes()));
1219   masm.jump_cc(Assembler::notEqual,
1220                RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1221   /* WARNING these NOPs are critical so that verified entry point is properly
1222      aligned for patching by NativeJump::patch_verified_entry() */
1223   int nops_cnt = 2;
1224   if( !OptoBreakpoint ) // Leave space for int3
1225      nops_cnt += 1;
1226   masm.nop(nops_cnt);
1227 
1228   assert(cbuf.insts_size() - insts_size == size(ra_), "checking code size of inline cache node");
1229 }
1230 
1231 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1232   return OptoBreakpoint ? 11 : 12;
1233 }
1234 
1235 
1236 //=============================================================================
1237 uint size_exception_handler() {
1238   // NativeCall instruction size is the same as NativeJump.
1239   // exception handler starts out as jump and can be patched to
1240   // a call be deoptimization.  (4932387)
1241   // Note that this value is also credited (in output.cpp) to
1242   // the size of the code section.
1243   return NativeJump::instruction_size;
1244 }
1245 
1246 // Emit exception handler code.  Stuff framesize into a register
1247 // and call a VM stub routine.
1248 int emit_exception_handler(CodeBuffer& cbuf) {
1249 
1250   // Note that the code buffer's insts_mark is always relative to insts.
1251   // That's why we must use the macroassembler to generate a handler.
1252   MacroAssembler _masm(&cbuf);
1253   address base =
1254   __ start_a_stub(size_exception_handler());
1255   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1256   int offset = __ offset();
1257   __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
1258   assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1259   __ end_a_stub();
1260   return offset;
1261 }
1262 
1263 uint size_deopt_handler() {
1264   // NativeCall instruction size is the same as NativeJump.
1265   // exception handler starts out as jump and can be patched to
1266   // a call be deoptimization.  (4932387)
1267   // Note that this value is also credited (in output.cpp) to
1268   // the size of the code section.
1269   return 5 + NativeJump::instruction_size; // pushl(); jmp;
1270 }
1271 
1272 // Emit deopt handler code.
1273 int emit_deopt_handler(CodeBuffer& cbuf) {
1274 
1275   // Note that the code buffer's insts_mark is always relative to insts.
1276   // That's why we must use the macroassembler to generate a handler.
1277   MacroAssembler _masm(&cbuf);
1278   address base =
1279   __ start_a_stub(size_exception_handler());
1280   if (base == NULL)  return 0;  // CodeBuffer::expand failed
1281   int offset = __ offset();
1282   InternalAddress here(__ pc());
1283   __ pushptr(here.addr());
1284 
1285   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1286   assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1287   __ end_a_stub();
1288   return offset;
1289 }
1290 
1291 
1292 const bool Matcher::match_rule_supported(int opcode) {
1293   if (!has_match_rule(opcode))
1294     return false;
1295 
1296   switch (opcode) {
1297     case Op_PopCountI:
1298     case Op_PopCountL:
1299       if (!UsePopCountInstruction)
1300         return false;
1301     break;
1302   }
1303   
1304   return true;  // Per default match rules are supported.
1305 }
1306 
1307 int Matcher::regnum_to_fpu_offset(int regnum) {
1308   return regnum - 32; // The FP registers are in the second chunk
1309 }
1310 
1311 // This is UltraSparc specific, true just means we have fast l2f conversion
1312 const bool Matcher::convL2FSupported(void) {
1313   return true;
1314 }
1315 
1316 // Vector width in bytes
1317 const uint Matcher::vector_width_in_bytes(void) {
1318   return UseSSE >= 2 ? 8 : 0;
1319 }
1320 
1321 // Vector ideal reg
1322 const uint Matcher::vector_ideal_reg(void) {
1323   return Op_RegD;
1324 }
1325 
1326 // Is this branch offset short enough that a short branch can be used?
1327 //
1328 // NOTE: If the platform does not provide any short branch variants, then
1329 //       this method should return false for offset 0.
1330 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
1331   // The passed offset is relative to address of the branch.
1332   // On 86 a branch displacement is calculated relative to address
1333   // of a next instruction.
1334   offset -= br_size;
1335 
1336   // the short version of jmpConUCF2 contains multiple branches,
1337   // making the reach slightly less
1338   if (rule == jmpConUCF2_rule)
1339     return (-126 <= offset && offset <= 125);
1340   return (-128 <= offset && offset <= 127);
1341 }
1342 
1343 const bool Matcher::isSimpleConstant64(jlong value) {
1344   // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1345   return false;
1346 }
1347 
1348 // The ecx parameter to rep stos for the ClearArray node is in dwords.
1349 const bool Matcher::init_array_count_is_in_bytes = false;
1350 
1351 // Threshold size for cleararray.
1352 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1353 
1354 // Needs 2 CMOV's for longs.
1355 const int Matcher::long_cmove_cost() { return 1; }
1356 
1357 // No CMOVF/CMOVD with SSE/SSE2
1358 const int Matcher::float_cmove_cost() { return (UseSSE>=1) ? ConditionalMoveLimit : 0; }
1359 
1360 // Should the Matcher clone shifts on addressing modes, expecting them to
1361 // be subsumed into complex addressing expressions or compute them into
1362 // registers?  True for Intel but false for most RISCs
1363 const bool Matcher::clone_shift_expressions = true;
1364 
1365 // Do we need to mask the count passed to shift instructions or does
1366 // the cpu only look at the lower 5/6 bits anyway?
1367 const bool Matcher::need_masked_shift_count = false;
1368 
1369 bool Matcher::narrow_oop_use_complex_address() {
1370   ShouldNotCallThis();
1371   return true;
1372 }
1373 
1374 
1375 // Is it better to copy float constants, or load them directly from memory?
1376 // Intel can load a float constant from a direct address, requiring no
1377 // extra registers.  Most RISCs will have to materialize an address into a
1378 // register first, so they would do better to copy the constant from stack.
1379 const bool Matcher::rematerialize_float_constants = true;
1380 
1381 // If CPU can load and store mis-aligned doubles directly then no fixup is
1382 // needed.  Else we split the double into 2 integer pieces and move it
1383 // piece-by-piece.  Only happens when passing doubles into C code as the
1384 // Java calling convention forces doubles to be aligned.
1385 const bool Matcher::misaligned_doubles_ok = true;
1386 
1387 
1388 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
1389   // Get the memory operand from the node
1390   uint numopnds = node->num_opnds();        // Virtual call for number of operands
1391   uint skipped  = node->oper_input_base();  // Sum of leaves skipped so far
1392   assert( idx >= skipped, "idx too low in pd_implicit_null_fixup" );
1393   uint opcnt     = 1;                 // First operand
1394   uint num_edges = node->_opnds[1]->num_edges(); // leaves for first operand
1395   while( idx >= skipped+num_edges ) {
1396     skipped += num_edges;
1397     opcnt++;                          // Bump operand count
1398     assert( opcnt < numopnds, "Accessing non-existent operand" );
1399     num_edges = node->_opnds[opcnt]->num_edges(); // leaves for next operand
1400   }
1401 
1402   MachOper *memory = node->_opnds[opcnt];
1403   MachOper *new_memory = NULL;
1404   switch (memory->opcode()) {
1405   case DIRECT:
1406   case INDOFFSET32X:
1407     // No transformation necessary.
1408     return;
1409   case INDIRECT:
1410     new_memory = new (C) indirect_win95_safeOper( );
1411     break;
1412   case INDOFFSET8:
1413     new_memory = new (C) indOffset8_win95_safeOper(memory->disp(NULL, NULL, 0));
1414     break;
1415   case INDOFFSET32:
1416     new_memory = new (C) indOffset32_win95_safeOper(memory->disp(NULL, NULL, 0));
1417     break;
1418   case INDINDEXOFFSET:
1419     new_memory = new (C) indIndexOffset_win95_safeOper(memory->disp(NULL, NULL, 0));
1420     break;
1421   case INDINDEXSCALE:
1422     new_memory = new (C) indIndexScale_win95_safeOper(memory->scale());
1423     break;
1424   case INDINDEXSCALEOFFSET:
1425     new_memory = new (C) indIndexScaleOffset_win95_safeOper(memory->scale(), memory->disp(NULL, NULL, 0));
1426     break;
1427   case LOAD_LONG_INDIRECT:
1428   case LOAD_LONG_INDOFFSET32:
1429     // Does not use EBP as address register, use { EDX, EBX, EDI, ESI}
1430     return;
1431   default:
1432     assert(false, "unexpected memory operand in pd_implicit_null_fixup()");
1433     return;
1434   }
1435   node->_opnds[opcnt] = new_memory;
1436 }
1437 
1438 // Advertise here if the CPU requires explicit rounding operations
1439 // to implement the UseStrictFP mode.
1440 const bool Matcher::strict_fp_requires_explicit_rounding = true;
1441 
1442 // Are floats conerted to double when stored to stack during deoptimization?
1443 // On x32 it is stored with convertion only when FPU is used for floats.
1444 bool Matcher::float_in_double() { return (UseSSE == 0); }
1445 
1446 // Do ints take an entire long register or just half?
1447 const bool Matcher::int_in_long = false;
1448 
1449 // Return whether or not this register is ever used as an argument.  This
1450 // function is used on startup to build the trampoline stubs in generateOptoStub.
1451 // Registers not mentioned will be killed by the VM call in the trampoline, and
1452 // arguments in those registers not be available to the callee.
1453 bool Matcher::can_be_java_arg( int reg ) {
1454   if(  reg == ECX_num   || reg == EDX_num   ) return true;
1455   if( (reg == XMM0a_num || reg == XMM1a_num) && UseSSE>=1 ) return true;
1456   if( (reg == XMM0b_num || reg == XMM1b_num) && UseSSE>=2 ) return true;
1457   return false;
1458 }
1459 
1460 bool Matcher::is_spillable_arg( int reg ) {
1461   return can_be_java_arg(reg);
1462 }
1463 
1464 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
1465   // Use hardware integer DIV instruction when
1466   // it is faster than a code which use multiply.
1467   // Only when constant divisor fits into 32 bit
1468   // (min_jint is excluded to get only correct
1469   // positive 32 bit values from negative).
1470   return VM_Version::has_fast_idiv() &&
1471          (divisor == (int)divisor && divisor != min_jint);
1472 }
1473 
1474 // Register for DIVI projection of divmodI
1475 RegMask Matcher::divI_proj_mask() {
1476   return EAX_REG_mask();
1477 }
1478 
1479 // Register for MODI projection of divmodI
1480 RegMask Matcher::modI_proj_mask() {
1481   return EDX_REG_mask();
1482 }
1483 
1484 // Register for DIVL projection of divmodL
1485 RegMask Matcher::divL_proj_mask() {
1486   ShouldNotReachHere();
1487   return RegMask();
1488 }
1489 
1490 // Register for MODL projection of divmodL
1491 RegMask Matcher::modL_proj_mask() {
1492   ShouldNotReachHere();
1493   return RegMask();
1494 }
1495 
1496 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
1497   return EBP_REG_mask();
1498 }
1499 
1500 // Returns true if the high 32 bits of the value is known to be zero.
1501 bool is_operand_hi32_zero(Node* n) {
1502   int opc = n->Opcode();
1503   if (opc == Op_LoadUI2L) {
1504     return true;
1505   }
1506   if (opc == Op_AndL) {
1507     Node* o2 = n->in(2);
1508     if (o2->is_Con() && (o2->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
1509       return true;
1510     }
1511   }
1512   if (opc == Op_ConL && (n->get_long() & 0xFFFFFFFF00000000LL) == 0LL) {
1513     return true;
1514   }
1515   return false;
1516 }
1517 
1518 %}
1519 
1520 //----------ENCODING BLOCK-----------------------------------------------------
1521 // This block specifies the encoding classes used by the compiler to output
1522 // byte streams.  Encoding classes generate functions which are called by
1523 // Machine Instruction Nodes in order to generate the bit encoding of the
1524 // instruction.  Operands specify their base encoding interface with the
1525 // interface keyword.  There are currently supported four interfaces,
1526 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER.  REG_INTER causes an
1527 // operand to generate a function which returns its register number when
1528 // queried.   CONST_INTER causes an operand to generate a function which
1529 // returns the value of the constant when queried.  MEMORY_INTER causes an
1530 // operand to generate four functions which return the Base Register, the
1531 // Index Register, the Scale Value, and the Offset Value of the operand when
1532 // queried.  COND_INTER causes an operand to generate six functions which
1533 // return the encoding code (ie - encoding bits for the instruction)
1534 // associated with each basic boolean condition for a conditional instruction.
1535 // Instructions specify two basic values for encoding.  They use the
1536 // ins_encode keyword to specify their encoding class (which must be one of
1537 // the class names specified in the encoding block), and they use the
1538 // opcode keyword to specify, in order, their primary, secondary, and
1539 // tertiary opcode.  Only the opcode sections which a particular instruction
1540 // needs for encoding need to be specified.
1541 encode %{
1542   // Build emit functions for each basic byte or larger field in the intel
1543   // encoding scheme (opcode, rm, sib, immediate), and call them from C++
1544   // code in the enc_class source block.  Emit functions will live in the
1545   // main source block for now.  In future, we can generalize this by
1546   // adding a syntax that specifies the sizes of fields in an order,
1547   // so that the adlc can build the emit functions automagically
1548 
1549   // Emit primary opcode
1550   enc_class OpcP %{
1551     emit_opcode(cbuf, $primary);
1552   %}
1553 
1554   // Emit secondary opcode
1555   enc_class OpcS %{
1556     emit_opcode(cbuf, $secondary);
1557   %}
1558 
1559   // Emit opcode directly
1560   enc_class Opcode(immI d8) %{
1561     emit_opcode(cbuf, $d8$$constant);
1562   %}
1563 
1564   enc_class SizePrefix %{
1565     emit_opcode(cbuf,0x66);
1566   %}
1567 
1568   enc_class RegReg (eRegI dst, eRegI src) %{    // RegReg(Many)
1569     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1570   %}
1571 
1572   enc_class OpcRegReg (immI opcode, eRegI dst, eRegI src) %{    // OpcRegReg(Many)
1573     emit_opcode(cbuf,$opcode$$constant);
1574     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1575   %}
1576 
1577   enc_class mov_r32_imm0( eRegI dst ) %{
1578     emit_opcode( cbuf, 0xB8 + $dst$$reg ); // 0xB8+ rd   -- MOV r32  ,imm32
1579     emit_d32   ( cbuf, 0x0  );             //                         imm32==0x0
1580   %}
1581 
1582   enc_class cdq_enc %{
1583     // Full implementation of Java idiv and irem; checks for
1584     // special case as described in JVM spec., p.243 & p.271.
1585     //
1586     //         normal case                           special case
1587     //
1588     // input : rax,: dividend                         min_int
1589     //         reg: divisor                          -1
1590     //
1591     // output: rax,: quotient  (= rax, idiv reg)       min_int
1592     //         rdx: remainder (= rax, irem reg)       0
1593     //
1594     //  Code sequnce:
1595     //
1596     //  81 F8 00 00 00 80    cmp         rax,80000000h
1597     //  0F 85 0B 00 00 00    jne         normal_case
1598     //  33 D2                xor         rdx,edx
1599     //  83 F9 FF             cmp         rcx,0FFh
1600     //  0F 84 03 00 00 00    je          done
1601     //                  normal_case:
1602     //  99                   cdq
1603     //  F7 F9                idiv        rax,ecx
1604     //                  done:
1605     //
1606     emit_opcode(cbuf,0x81); emit_d8(cbuf,0xF8);
1607     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);
1608     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x80);                     // cmp rax,80000000h
1609     emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x85);
1610     emit_opcode(cbuf,0x0B); emit_d8(cbuf,0x00);
1611     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // jne normal_case
1612     emit_opcode(cbuf,0x33); emit_d8(cbuf,0xD2);                     // xor rdx,edx
1613     emit_opcode(cbuf,0x83); emit_d8(cbuf,0xF9); emit_d8(cbuf,0xFF); // cmp rcx,0FFh
1614     emit_opcode(cbuf,0x0F); emit_d8(cbuf,0x84);
1615     emit_opcode(cbuf,0x03); emit_d8(cbuf,0x00);
1616     emit_opcode(cbuf,0x00); emit_d8(cbuf,0x00);                     // je done
1617     // normal_case:
1618     emit_opcode(cbuf,0x99);                                         // cdq
1619     // idiv (note: must be emitted by the user of this rule)
1620     // normal:
1621   %}
1622 
1623   // Dense encoding for older common ops
1624   enc_class Opc_plus(immI opcode, eRegI reg) %{
1625     emit_opcode(cbuf, $opcode$$constant + $reg$$reg);
1626   %}
1627 
1628 
1629   // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
1630   enc_class OpcSE (immI imm) %{ // Emit primary opcode and set sign-extend bit
1631     // Check for 8-bit immediate, and set sign extend bit in opcode
1632     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1633       emit_opcode(cbuf, $primary | 0x02);
1634     }
1635     else {                          // If 32-bit immediate
1636       emit_opcode(cbuf, $primary);
1637     }
1638   %}
1639 
1640   enc_class OpcSErm (eRegI dst, immI imm) %{    // OpcSEr/m
1641     // Emit primary opcode and set sign-extend bit
1642     // Check for 8-bit immediate, and set sign extend bit in opcode
1643     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1644       emit_opcode(cbuf, $primary | 0x02);    }
1645     else {                          // If 32-bit immediate
1646       emit_opcode(cbuf, $primary);
1647     }
1648     // Emit r/m byte with secondary opcode, after primary opcode.
1649     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1650   %}
1651 
1652   enc_class Con8or32 (immI imm) %{    // Con8or32(storeImmI), 8 or 32 bits
1653     // Check for 8-bit immediate, and set sign extend bit in opcode
1654     if (($imm$$constant >= -128) && ($imm$$constant <= 127)) {
1655       $$$emit8$imm$$constant;
1656     }
1657     else {                          // If 32-bit immediate
1658       // Output immediate
1659       $$$emit32$imm$$constant;
1660     }
1661   %}
1662 
1663   enc_class Long_OpcSErm_Lo(eRegL dst, immL imm) %{
1664     // Emit primary opcode and set sign-extend bit
1665     // Check for 8-bit immediate, and set sign extend bit in opcode
1666     int con = (int)$imm$$constant; // Throw away top bits
1667     emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
1668     // Emit r/m byte with secondary opcode, after primary opcode.
1669     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1670     if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
1671     else                               emit_d32(cbuf,con);
1672   %}
1673 
1674   enc_class Long_OpcSErm_Hi(eRegL dst, immL imm) %{
1675     // Emit primary opcode and set sign-extend bit
1676     // Check for 8-bit immediate, and set sign extend bit in opcode
1677     int con = (int)($imm$$constant >> 32); // Throw away bottom bits
1678     emit_opcode(cbuf, ((con >= -128) && (con <= 127)) ? ($primary | 0x02) : $primary);
1679     // Emit r/m byte with tertiary opcode, after primary opcode.
1680     emit_rm(cbuf, 0x3, $tertiary, HIGH_FROM_LOW($dst$$reg));
1681     if ((con >= -128) && (con <= 127)) emit_d8 (cbuf,con);
1682     else                               emit_d32(cbuf,con);
1683   %}
1684 
1685   enc_class OpcSReg (eRegI dst) %{    // BSWAP
1686     emit_cc(cbuf, $secondary, $dst$$reg );
1687   %}
1688 
1689   enc_class bswap_long_bytes(eRegL dst) %{ // BSWAP
1690     int destlo = $dst$$reg;
1691     int desthi = HIGH_FROM_LOW(destlo);
1692     // bswap lo
1693     emit_opcode(cbuf, 0x0F);
1694     emit_cc(cbuf, 0xC8, destlo);
1695     // bswap hi
1696     emit_opcode(cbuf, 0x0F);
1697     emit_cc(cbuf, 0xC8, desthi);
1698     // xchg lo and hi
1699     emit_opcode(cbuf, 0x87);
1700     emit_rm(cbuf, 0x3, destlo, desthi);
1701   %}
1702 
1703   enc_class RegOpc (eRegI div) %{    // IDIV, IMOD, JMP indirect, ...
1704     emit_rm(cbuf, 0x3, $secondary, $div$$reg );
1705   %}
1706 
1707   enc_class enc_cmov(cmpOp cop ) %{ // CMOV
1708     $$$emit8$primary;
1709     emit_cc(cbuf, $secondary, $cop$$cmpcode);
1710   %}
1711 
1712   enc_class enc_cmov_dpr(cmpOp cop, regDPR src ) %{ // CMOV
1713     int op = 0xDA00 + $cop$$cmpcode + ($src$$reg-1);
1714     emit_d8(cbuf, op >> 8 );
1715     emit_d8(cbuf, op & 255);
1716   %}
1717 
1718   // emulate a CMOV with a conditional branch around a MOV
1719   enc_class enc_cmov_branch( cmpOp cop, immI brOffs ) %{ // CMOV
1720     // Invert sense of branch from sense of CMOV
1721     emit_cc( cbuf, 0x70, ($cop$$cmpcode^1) );
1722     emit_d8( cbuf, $brOffs$$constant );
1723   %}
1724 
1725   enc_class enc_PartialSubtypeCheck( ) %{
1726     Register Redi = as_Register(EDI_enc); // result register
1727     Register Reax = as_Register(EAX_enc); // super class
1728     Register Recx = as_Register(ECX_enc); // killed
1729     Register Resi = as_Register(ESI_enc); // sub class
1730     Label miss;
1731 
1732     MacroAssembler _masm(&cbuf);
1733     __ check_klass_subtype_slow_path(Resi, Reax, Recx, Redi,
1734                                      NULL, &miss,
1735                                      /*set_cond_codes:*/ true);
1736     if ($primary) {
1737       __ xorptr(Redi, Redi);
1738     }
1739     __ bind(miss);
1740   %}
1741 
1742   enc_class FFree_Float_Stack_All %{    // Free_Float_Stack_All
1743     MacroAssembler masm(&cbuf);
1744     int start = masm.offset();
1745     if (UseSSE >= 2) {
1746       if (VerifyFPU) {
1747         masm.verify_FPU(0, "must be empty in SSE2+ mode");
1748       }
1749     } else {
1750       // External c_calling_convention expects the FPU stack to be 'clean'.
1751       // Compiled code leaves it dirty.  Do cleanup now.
1752       masm.empty_FPU_stack();
1753     }
1754     if (sizeof_FFree_Float_Stack_All == -1) {
1755       sizeof_FFree_Float_Stack_All = masm.offset() - start;
1756     } else {
1757       assert(masm.offset() - start == sizeof_FFree_Float_Stack_All, "wrong size");
1758     }
1759   %}
1760 
1761   enc_class Verify_FPU_For_Leaf %{
1762     if( VerifyFPU ) {
1763       MacroAssembler masm(&cbuf);
1764       masm.verify_FPU( -3, "Returning from Runtime Leaf call");
1765     }
1766   %}
1767 
1768   enc_class Java_To_Runtime (method meth) %{    // CALL Java_To_Runtime, Java_To_Runtime_Leaf
1769     // This is the instruction starting address for relocation info.
1770     cbuf.set_insts_mark();
1771     $$$emit8$primary;
1772     // CALL directly to the runtime
1773     emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1774                 runtime_call_Relocation::spec(), RELOC_IMM32 );
1775 
1776     if (UseSSE >= 2) {
1777       MacroAssembler _masm(&cbuf);
1778       BasicType rt = tf()->return_type();
1779 
1780       if ((rt == T_FLOAT || rt == T_DOUBLE) && !return_value_is_used()) {
1781         // A C runtime call where the return value is unused.  In SSE2+
1782         // mode the result needs to be removed from the FPU stack.  It's
1783         // likely that this function call could be removed by the
1784         // optimizer if the C function is a pure function.
1785         __ ffree(0);
1786       } else if (rt == T_FLOAT) {
1787         __ lea(rsp, Address(rsp, -4));
1788         __ fstp_s(Address(rsp, 0));
1789         __ movflt(xmm0, Address(rsp, 0));
1790         __ lea(rsp, Address(rsp,  4));
1791       } else if (rt == T_DOUBLE) {
1792         __ lea(rsp, Address(rsp, -8));
1793         __ fstp_d(Address(rsp, 0));
1794         __ movdbl(xmm0, Address(rsp, 0));
1795         __ lea(rsp, Address(rsp,  8));
1796       }
1797     }
1798   %}
1799 
1800 
1801   enc_class pre_call_FPU %{
1802     // If method sets FPU control word restore it here
1803     debug_only(int off0 = cbuf.insts_size());
1804     if( Compile::current()->in_24_bit_fp_mode() ) {
1805       MacroAssembler masm(&cbuf);
1806       masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1807     }
1808     debug_only(int off1 = cbuf.insts_size());
1809     assert(off1 - off0 == pre_call_FPU_size(), "correct size prediction");
1810   %}
1811 
1812   enc_class post_call_FPU %{
1813     // If method sets FPU control word do it here also
1814     if( Compile::current()->in_24_bit_fp_mode() ) {
1815       MacroAssembler masm(&cbuf);
1816       masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
1817     }
1818   %}
1819 
1820   enc_class Java_Static_Call (method meth) %{    // JAVA STATIC CALL
1821     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
1822     // who we intended to call.
1823     cbuf.set_insts_mark();
1824     $$$emit8$primary;
1825     if ( !_method ) {
1826       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1827                      runtime_call_Relocation::spec(), RELOC_IMM32 );
1828     } else if(_optimized_virtual) {
1829       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1830                      opt_virtual_call_Relocation::spec(), RELOC_IMM32 );
1831     } else {
1832       emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1833                      static_call_Relocation::spec(), RELOC_IMM32 );
1834     }
1835     if( _method ) {  // Emit stub for static call
1836       emit_java_to_interp(cbuf);
1837     }
1838   %}
1839 
1840   enc_class Java_Dynamic_Call (method meth) %{    // JAVA DYNAMIC CALL
1841     // !!!!!
1842     // Generate  "Mov EAX,0x00", placeholder instruction to load oop-info
1843     // emit_call_dynamic_prologue( cbuf );
1844     cbuf.set_insts_mark();
1845     emit_opcode(cbuf, 0xB8 + EAX_enc);        // mov    EAX,-1
1846     emit_d32_reloc(cbuf, (int)Universe::non_oop_word(), oop_Relocation::spec_for_immediate(), RELOC_IMM32);
1847     address  virtual_call_oop_addr = cbuf.insts_mark();
1848     // CALL to fixup routine.  Fixup routine uses ScopeDesc info to determine
1849     // who we intended to call.
1850     cbuf.set_insts_mark();
1851     $$$emit8$primary;
1852     emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4),
1853                 virtual_call_Relocation::spec(virtual_call_oop_addr), RELOC_IMM32 );
1854   %}
1855 
1856   enc_class Java_Compiled_Call (method meth) %{    // JAVA COMPILED CALL
1857     int disp = in_bytes(methodOopDesc::from_compiled_offset());
1858     assert( -128 <= disp && disp <= 127, "compiled_code_offset isn't small");
1859 
1860     // CALL *[EAX+in_bytes(methodOopDesc::from_compiled_code_entry_point_offset())]
1861     cbuf.set_insts_mark();
1862     $$$emit8$primary;
1863     emit_rm(cbuf, 0x01, $secondary, EAX_enc );  // R/M byte
1864     emit_d8(cbuf, disp);             // Displacement
1865 
1866   %}
1867 
1868 //   Following encoding is no longer used, but may be restored if calling
1869 //   convention changes significantly.
1870 //   Became: Xor_Reg(EBP), Java_To_Runtime( labl )
1871 //
1872 //   enc_class Java_Interpreter_Call (label labl) %{    // JAVA INTERPRETER CALL
1873 //     // int ic_reg     = Matcher::inline_cache_reg();
1874 //     // int ic_encode  = Matcher::_regEncode[ic_reg];
1875 //     // int imo_reg    = Matcher::interpreter_method_oop_reg();
1876 //     // int imo_encode = Matcher::_regEncode[imo_reg];
1877 //
1878 //     // // Interpreter expects method_oop in EBX, currently a callee-saved register,
1879 //     // // so we load it immediately before the call
1880 //     // emit_opcode(cbuf, 0x8B);                     // MOV    imo_reg,ic_reg  # method_oop
1881 //     // emit_rm(cbuf, 0x03, imo_encode, ic_encode ); // R/M byte
1882 //
1883 //     // xor rbp,ebp
1884 //     emit_opcode(cbuf, 0x33);
1885 //     emit_rm(cbuf, 0x3, EBP_enc, EBP_enc);
1886 //
1887 //     // CALL to interpreter.
1888 //     cbuf.set_insts_mark();
1889 //     $$$emit8$primary;
1890 //     emit_d32_reloc(cbuf, ($labl$$label - (int)(cbuf.insts_end()) - 4),
1891 //                 runtime_call_Relocation::spec(), RELOC_IMM32 );
1892 //   %}
1893 
1894   enc_class RegOpcImm (eRegI dst, immI8 shift) %{    // SHL, SAR, SHR
1895     $$$emit8$primary;
1896     emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
1897     $$$emit8$shift$$constant;
1898   %}
1899 
1900   enc_class LdImmI (eRegI dst, immI src) %{    // Load Immediate
1901     // Load immediate does not have a zero or sign extended version
1902     // for 8-bit immediates
1903     emit_opcode(cbuf, 0xB8 + $dst$$reg);
1904     $$$emit32$src$$constant;
1905   %}
1906 
1907   enc_class LdImmP (eRegI dst, immI src) %{    // Load Immediate
1908     // Load immediate does not have a zero or sign extended version
1909     // for 8-bit immediates
1910     emit_opcode(cbuf, $primary + $dst$$reg);
1911     $$$emit32$src$$constant;
1912   %}
1913 
1914   enc_class LdImmL_Lo( eRegL dst, immL src) %{    // Load Immediate
1915     // Load immediate does not have a zero or sign extended version
1916     // for 8-bit immediates
1917     int dst_enc = $dst$$reg;
1918     int src_con = $src$$constant & 0x0FFFFFFFFL;
1919     if (src_con == 0) {
1920       // xor dst, dst
1921       emit_opcode(cbuf, 0x33);
1922       emit_rm(cbuf, 0x3, dst_enc, dst_enc);
1923     } else {
1924       emit_opcode(cbuf, $primary + dst_enc);
1925       emit_d32(cbuf, src_con);
1926     }
1927   %}
1928 
1929   enc_class LdImmL_Hi( eRegL dst, immL src) %{    // Load Immediate
1930     // Load immediate does not have a zero or sign extended version
1931     // for 8-bit immediates
1932     int dst_enc = $dst$$reg + 2;
1933     int src_con = ((julong)($src$$constant)) >> 32;
1934     if (src_con == 0) {
1935       // xor dst, dst
1936       emit_opcode(cbuf, 0x33);
1937       emit_rm(cbuf, 0x3, dst_enc, dst_enc);
1938     } else {
1939       emit_opcode(cbuf, $primary + dst_enc);
1940       emit_d32(cbuf, src_con);
1941     }
1942   %}
1943 
1944 
1945   // Encode a reg-reg copy.  If it is useless, then empty encoding.
1946   enc_class enc_Copy( eRegI dst, eRegI src ) %{
1947     encode_Copy( cbuf, $dst$$reg, $src$$reg );
1948   %}
1949 
1950   enc_class enc_CopyL_Lo( eRegI dst, eRegL src ) %{
1951     encode_Copy( cbuf, $dst$$reg, $src$$reg );
1952   %}
1953 
1954   enc_class RegReg (eRegI dst, eRegI src) %{    // RegReg(Many)
1955     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1956   %}
1957 
1958   enc_class RegReg_Lo(eRegL dst, eRegL src) %{    // RegReg(Many)
1959     $$$emit8$primary;
1960     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1961   %}
1962 
1963   enc_class RegReg_Hi(eRegL dst, eRegL src) %{    // RegReg(Many)
1964     $$$emit8$secondary;
1965     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
1966   %}
1967 
1968   enc_class RegReg_Lo2(eRegL dst, eRegL src) %{    // RegReg(Many)
1969     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
1970   %}
1971 
1972   enc_class RegReg_Hi2(eRegL dst, eRegL src) %{    // RegReg(Many)
1973     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg));
1974   %}
1975 
1976   enc_class RegReg_HiLo( eRegL src, eRegI dst ) %{
1977     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($src$$reg));
1978   %}
1979 
1980   enc_class Con32 (immI src) %{    // Con32(storeImmI)
1981     // Output immediate
1982     $$$emit32$src$$constant;
1983   %}
1984 
1985   enc_class Con32FPR_as_bits(immFPR src) %{        // storeF_imm
1986     // Output Float immediate bits
1987     jfloat jf = $src$$constant;
1988     int    jf_as_bits = jint_cast( jf );
1989     emit_d32(cbuf, jf_as_bits);
1990   %}
1991 
1992   enc_class Con32F_as_bits(immF src) %{      // storeX_imm
1993     // Output Float immediate bits
1994     jfloat jf = $src$$constant;
1995     int    jf_as_bits = jint_cast( jf );
1996     emit_d32(cbuf, jf_as_bits);
1997   %}
1998 
1999   enc_class Con16 (immI src) %{    // Con16(storeImmI)
2000     // Output immediate
2001     $$$emit16$src$$constant;
2002   %}
2003 
2004   enc_class Con_d32(immI src) %{
2005     emit_d32(cbuf,$src$$constant);
2006   %}
2007 
2008   enc_class conmemref (eRegP t1) %{    // Con32(storeImmI)
2009     // Output immediate memory reference
2010     emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2011     emit_d32(cbuf, 0x00);
2012   %}
2013 
2014   enc_class lock_prefix( ) %{
2015     if( os::is_MP() )
2016       emit_opcode(cbuf,0xF0);         // [Lock]
2017   %}
2018 
2019   // Cmp-xchg long value.
2020   // Note: we need to swap rbx, and rcx before and after the
2021   //       cmpxchg8 instruction because the instruction uses
2022   //       rcx as the high order word of the new value to store but
2023   //       our register encoding uses rbx,.
2024   enc_class enc_cmpxchg8(eSIRegP mem_ptr) %{
2025 
2026     // XCHG  rbx,ecx
2027     emit_opcode(cbuf,0x87);
2028     emit_opcode(cbuf,0xD9);
2029     // [Lock]
2030     if( os::is_MP() )
2031       emit_opcode(cbuf,0xF0);
2032     // CMPXCHG8 [Eptr]
2033     emit_opcode(cbuf,0x0F);
2034     emit_opcode(cbuf,0xC7);
2035     emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
2036     // XCHG  rbx,ecx
2037     emit_opcode(cbuf,0x87);
2038     emit_opcode(cbuf,0xD9);
2039   %}
2040 
2041   enc_class enc_cmpxchg(eSIRegP mem_ptr) %{
2042     // [Lock]
2043     if( os::is_MP() )
2044       emit_opcode(cbuf,0xF0);
2045 
2046     // CMPXCHG [Eptr]
2047     emit_opcode(cbuf,0x0F);
2048     emit_opcode(cbuf,0xB1);
2049     emit_rm( cbuf, 0x0, 1, $mem_ptr$$reg );
2050   %}
2051 
2052   enc_class enc_flags_ne_to_boolean( iRegI res ) %{
2053     int res_encoding = $res$$reg;
2054 
2055     // MOV  res,0
2056     emit_opcode( cbuf, 0xB8 + res_encoding);
2057     emit_d32( cbuf, 0 );
2058     // JNE,s  fail
2059     emit_opcode(cbuf,0x75);
2060     emit_d8(cbuf, 5 );
2061     // MOV  res,1
2062     emit_opcode( cbuf, 0xB8 + res_encoding);
2063     emit_d32( cbuf, 1 );
2064     // fail:
2065   %}
2066 
2067   enc_class set_instruction_start( ) %{
2068     cbuf.set_insts_mark();            // Mark start of opcode for reloc info in mem operand
2069   %}
2070 
2071   enc_class RegMem (eRegI ereg, memory mem) %{    // emit_reg_mem
2072     int reg_encoding = $ereg$$reg;
2073     int base  = $mem$$base;
2074     int index = $mem$$index;
2075     int scale = $mem$$scale;
2076     int displace = $mem$$disp;
2077     bool disp_is_oop = $mem->disp_is_oop();
2078     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_is_oop);
2079   %}
2080 
2081   enc_class RegMem_Hi(eRegL ereg, memory mem) %{    // emit_reg_mem
2082     int reg_encoding = HIGH_FROM_LOW($ereg$$reg);  // Hi register of pair, computed from lo
2083     int base  = $mem$$base;
2084     int index = $mem$$index;
2085     int scale = $mem$$scale;
2086     int displace = $mem$$disp + 4;      // Offset is 4 further in memory
2087     assert( !$mem->disp_is_oop(), "Cannot add 4 to oop" );
2088     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, false/*disp_is_oop*/);
2089   %}
2090 
2091   enc_class move_long_small_shift( eRegL dst, immI_1_31 cnt ) %{
2092     int r1, r2;
2093     if( $tertiary == 0xA4 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
2094     else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }
2095     emit_opcode(cbuf,0x0F);
2096     emit_opcode(cbuf,$tertiary);
2097     emit_rm(cbuf, 0x3, r1, r2);
2098     emit_d8(cbuf,$cnt$$constant);
2099     emit_d8(cbuf,$primary);
2100     emit_rm(cbuf, 0x3, $secondary, r1);
2101     emit_d8(cbuf,$cnt$$constant);
2102   %}
2103 
2104   enc_class move_long_big_shift_sign( eRegL dst, immI_32_63 cnt ) %{
2105     emit_opcode( cbuf, 0x8B ); // Move
2106     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
2107     if( $cnt$$constant > 32 ) { // Shift, if not by zero
2108       emit_d8(cbuf,$primary);
2109       emit_rm(cbuf, 0x3, $secondary, $dst$$reg);
2110       emit_d8(cbuf,$cnt$$constant-32);
2111     }
2112     emit_d8(cbuf,$primary);
2113     emit_rm(cbuf, 0x3, $secondary, HIGH_FROM_LOW($dst$$reg));
2114     emit_d8(cbuf,31);
2115   %}
2116 
2117   enc_class move_long_big_shift_clr( eRegL dst, immI_32_63 cnt ) %{
2118     int r1, r2;
2119     if( $secondary == 0x5 ) { r1 = $dst$$reg;  r2 = HIGH_FROM_LOW($dst$$reg); }
2120     else                    { r2 = $dst$$reg;  r1 = HIGH_FROM_LOW($dst$$reg); }
2121 
2122     emit_opcode( cbuf, 0x8B ); // Move r1,r2
2123     emit_rm(cbuf, 0x3, r1, r2);
2124     if( $cnt$$constant > 32 ) { // Shift, if not by zero
2125       emit_opcode(cbuf,$primary);
2126       emit_rm(cbuf, 0x3, $secondary, r1);
2127       emit_d8(cbuf,$cnt$$constant-32);
2128     }
2129     emit_opcode(cbuf,0x33);  // XOR r2,r2
2130     emit_rm(cbuf, 0x3, r2, r2);
2131   %}
2132 
2133   // Clone of RegMem but accepts an extra parameter to access each
2134   // half of a double in memory; it never needs relocation info.
2135   enc_class Mov_MemD_half_to_Reg (immI opcode, memory mem, immI disp_for_half, eRegI rm_reg) %{
2136     emit_opcode(cbuf,$opcode$$constant);
2137     int reg_encoding = $rm_reg$$reg;
2138     int base     = $mem$$base;
2139     int index    = $mem$$index;
2140     int scale    = $mem$$scale;
2141     int displace = $mem$$disp + $disp_for_half$$constant;
2142     bool disp_is_oop = false;
2143     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_is_oop);
2144   %}
2145 
2146   // !!!!! Special Custom Code used by MemMove, and stack access instructions !!!!!
2147   //
2148   // Clone of RegMem except the RM-byte's reg/opcode field is an ADLC-time constant
2149   // and it never needs relocation information.
2150   // Frequently used to move data between FPU's Stack Top and memory.
2151   enc_class RMopc_Mem_no_oop (immI rm_opcode, memory mem) %{
2152     int rm_byte_opcode = $rm_opcode$$constant;
2153     int base     = $mem$$base;
2154     int index    = $mem$$index;
2155     int scale    = $mem$$scale;
2156     int displace = $mem$$disp;
2157     assert( !$mem->disp_is_oop(), "No oops here because no relo info allowed" );
2158     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, false);
2159   %}
2160 
2161   enc_class RMopc_Mem (immI rm_opcode, memory mem) %{
2162     int rm_byte_opcode = $rm_opcode$$constant;
2163     int base     = $mem$$base;
2164     int index    = $mem$$index;
2165     int scale    = $mem$$scale;
2166     int displace = $mem$$disp;
2167     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when working with static globals
2168     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_is_oop);
2169   %}
2170 
2171   enc_class RegLea (eRegI dst, eRegI src0, immI src1 ) %{    // emit_reg_lea
2172     int reg_encoding = $dst$$reg;
2173     int base         = $src0$$reg;      // 0xFFFFFFFF indicates no base
2174     int index        = 0x04;            // 0x04 indicates no index
2175     int scale        = 0x00;            // 0x00 indicates no scale
2176     int displace     = $src1$$constant; // 0x00 indicates no displacement
2177     bool disp_is_oop = false;
2178     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_is_oop);
2179   %}
2180 
2181   enc_class min_enc (eRegI dst, eRegI src) %{    // MIN
2182     // Compare dst,src
2183     emit_opcode(cbuf,0x3B);
2184     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2185     // jmp dst < src around move
2186     emit_opcode(cbuf,0x7C);
2187     emit_d8(cbuf,2);
2188     // move dst,src
2189     emit_opcode(cbuf,0x8B);
2190     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2191   %}
2192 
2193   enc_class max_enc (eRegI dst, eRegI src) %{    // MAX
2194     // Compare dst,src
2195     emit_opcode(cbuf,0x3B);
2196     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2197     // jmp dst > src around move
2198     emit_opcode(cbuf,0x7F);
2199     emit_d8(cbuf,2);
2200     // move dst,src
2201     emit_opcode(cbuf,0x8B);
2202     emit_rm(cbuf, 0x3, $dst$$reg, $src$$reg);
2203   %}
2204 
2205   enc_class enc_FPR_store(memory mem, regDPR src) %{
2206     // If src is FPR1, we can just FST to store it.
2207     // Else we need to FLD it to FPR1, then FSTP to store/pop it.
2208     int reg_encoding = 0x2; // Just store
2209     int base  = $mem$$base;
2210     int index = $mem$$index;
2211     int scale = $mem$$scale;
2212     int displace = $mem$$disp;
2213     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when working with static globals
2214     if( $src$$reg != FPR1L_enc ) {
2215       reg_encoding = 0x3;  // Store & pop
2216       emit_opcode( cbuf, 0xD9 ); // FLD (i.e., push it)
2217       emit_d8( cbuf, 0xC0-1+$src$$reg );
2218     }
2219     cbuf.set_insts_mark();       // Mark start of opcode for reloc info in mem operand
2220     emit_opcode(cbuf,$primary);
2221     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_is_oop);
2222   %}
2223 
2224   enc_class neg_reg(eRegI dst) %{
2225     // NEG $dst
2226     emit_opcode(cbuf,0xF7);
2227     emit_rm(cbuf, 0x3, 0x03, $dst$$reg );
2228   %}
2229 
2230   enc_class setLT_reg(eCXRegI dst) %{
2231     // SETLT $dst
2232     emit_opcode(cbuf,0x0F);
2233     emit_opcode(cbuf,0x9C);
2234     emit_rm( cbuf, 0x3, 0x4, $dst$$reg );
2235   %}
2236 
2237   enc_class enc_cmpLTP(ncxRegI p, ncxRegI q, ncxRegI y, eCXRegI tmp) %{    // cadd_cmpLT
2238     int tmpReg = $tmp$$reg;
2239 
2240     // SUB $p,$q
2241     emit_opcode(cbuf,0x2B);
2242     emit_rm(cbuf, 0x3, $p$$reg, $q$$reg);
2243     // SBB $tmp,$tmp
2244     emit_opcode(cbuf,0x1B);
2245     emit_rm(cbuf, 0x3, tmpReg, tmpReg);
2246     // AND $tmp,$y
2247     emit_opcode(cbuf,0x23);
2248     emit_rm(cbuf, 0x3, tmpReg, $y$$reg);
2249     // ADD $p,$tmp
2250     emit_opcode(cbuf,0x03);
2251     emit_rm(cbuf, 0x3, $p$$reg, tmpReg);
2252   %}
2253 
2254   enc_class enc_cmpLTP_mem(eRegI p, eRegI q, memory mem, eCXRegI tmp) %{    // cadd_cmpLT
2255     int tmpReg = $tmp$$reg;
2256 
2257     // SUB $p,$q
2258     emit_opcode(cbuf,0x2B);
2259     emit_rm(cbuf, 0x3, $p$$reg, $q$$reg);
2260     // SBB $tmp,$tmp
2261     emit_opcode(cbuf,0x1B);
2262     emit_rm(cbuf, 0x3, tmpReg, tmpReg);
2263     // AND $tmp,$y
2264     cbuf.set_insts_mark();       // Mark start of opcode for reloc info in mem operand
2265     emit_opcode(cbuf,0x23);
2266     int reg_encoding = tmpReg;
2267     int base  = $mem$$base;
2268     int index = $mem$$index;
2269     int scale = $mem$$scale;
2270     int displace = $mem$$disp;
2271     bool disp_is_oop = $mem->disp_is_oop();
2272     encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, disp_is_oop);
2273     // ADD $p,$tmp
2274     emit_opcode(cbuf,0x03);
2275     emit_rm(cbuf, 0x3, $p$$reg, tmpReg);
2276   %}
2277 
2278   enc_class shift_left_long( eRegL dst, eCXRegI shift ) %{
2279     // TEST shift,32
2280     emit_opcode(cbuf,0xF7);
2281     emit_rm(cbuf, 0x3, 0, ECX_enc);
2282     emit_d32(cbuf,0x20);
2283     // JEQ,s small
2284     emit_opcode(cbuf, 0x74);
2285     emit_d8(cbuf, 0x04);
2286     // MOV    $dst.hi,$dst.lo
2287     emit_opcode( cbuf, 0x8B );
2288     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
2289     // CLR    $dst.lo
2290     emit_opcode(cbuf, 0x33);
2291     emit_rm(cbuf, 0x3, $dst$$reg, $dst$$reg);
2292 // small:
2293     // SHLD   $dst.hi,$dst.lo,$shift
2294     emit_opcode(cbuf,0x0F);
2295     emit_opcode(cbuf,0xA5);
2296     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg));
2297     // SHL    $dst.lo,$shift"
2298     emit_opcode(cbuf,0xD3);
2299     emit_rm(cbuf, 0x3, 0x4, $dst$$reg );
2300   %}
2301 
2302   enc_class shift_right_long( eRegL dst, eCXRegI shift ) %{
2303     // TEST shift,32
2304     emit_opcode(cbuf,0xF7);
2305     emit_rm(cbuf, 0x3, 0, ECX_enc);
2306     emit_d32(cbuf,0x20);
2307     // JEQ,s small
2308     emit_opcode(cbuf, 0x74);
2309     emit_d8(cbuf, 0x04);
2310     // MOV    $dst.lo,$dst.hi
2311     emit_opcode( cbuf, 0x8B );
2312     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
2313     // CLR    $dst.hi
2314     emit_opcode(cbuf, 0x33);
2315     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($dst$$reg));
2316 // small:
2317     // SHRD   $dst.lo,$dst.hi,$shift
2318     emit_opcode(cbuf,0x0F);
2319     emit_opcode(cbuf,0xAD);
2320     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
2321     // SHR    $dst.hi,$shift"
2322     emit_opcode(cbuf,0xD3);
2323     emit_rm(cbuf, 0x3, 0x5, HIGH_FROM_LOW($dst$$reg) );
2324   %}
2325 
2326   enc_class shift_right_arith_long( eRegL dst, eCXRegI shift ) %{
2327     // TEST shift,32
2328     emit_opcode(cbuf,0xF7);
2329     emit_rm(cbuf, 0x3, 0, ECX_enc);
2330     emit_d32(cbuf,0x20);
2331     // JEQ,s small
2332     emit_opcode(cbuf, 0x74);
2333     emit_d8(cbuf, 0x05);
2334     // MOV    $dst.lo,$dst.hi
2335     emit_opcode( cbuf, 0x8B );
2336     emit_rm(cbuf, 0x3, $dst$$reg, HIGH_FROM_LOW($dst$$reg) );
2337     // SAR    $dst.hi,31
2338     emit_opcode(cbuf, 0xC1);
2339     emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW($dst$$reg) );
2340     emit_d8(cbuf, 0x1F );
2341 // small:
2342     // SHRD   $dst.lo,$dst.hi,$shift
2343     emit_opcode(cbuf,0x0F);
2344     emit_opcode(cbuf,0xAD);
2345     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg);
2346     // SAR    $dst.hi,$shift"
2347     emit_opcode(cbuf,0xD3);
2348     emit_rm(cbuf, 0x3, 0x7, HIGH_FROM_LOW($dst$$reg) );
2349   %}
2350 
2351 
2352   // ----------------- Encodings for floating point unit -----------------
2353   // May leave result in FPU-TOS or FPU reg depending on opcodes
2354   enc_class OpcReg_FPR(regFPR src) %{    // FMUL, FDIV
2355     $$$emit8$primary;
2356     emit_rm(cbuf, 0x3, $secondary, $src$$reg );
2357   %}
2358 
2359   // Pop argument in FPR0 with FSTP ST(0)
2360   enc_class PopFPU() %{
2361     emit_opcode( cbuf, 0xDD );
2362     emit_d8( cbuf, 0xD8 );
2363   %}
2364 
2365   // !!!!! equivalent to Pop_Reg_F
2366   enc_class Pop_Reg_DPR( regDPR dst ) %{
2367     emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
2368     emit_d8( cbuf, 0xD8+$dst$$reg );
2369   %}
2370 
2371   enc_class Push_Reg_DPR( regDPR dst ) %{
2372     emit_opcode( cbuf, 0xD9 );
2373     emit_d8( cbuf, 0xC0-1+$dst$$reg );   // FLD ST(i-1)
2374   %}
2375 
2376   enc_class strictfp_bias1( regDPR dst ) %{
2377     emit_opcode( cbuf, 0xDB );           // FLD m80real
2378     emit_opcode( cbuf, 0x2D );
2379     emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias1() );
2380     emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
2381     emit_opcode( cbuf, 0xC8+$dst$$reg );
2382   %}
2383 
2384   enc_class strictfp_bias2( regDPR dst ) %{
2385     emit_opcode( cbuf, 0xDB );           // FLD m80real
2386     emit_opcode( cbuf, 0x2D );
2387     emit_d32( cbuf, (int)StubRoutines::addr_fpu_subnormal_bias2() );
2388     emit_opcode( cbuf, 0xDE );           // FMULP ST(dst), ST0
2389     emit_opcode( cbuf, 0xC8+$dst$$reg );
2390   %}
2391 
2392   // Special case for moving an integer register to a stack slot.
2393   enc_class OpcPRegSS( stackSlotI dst, eRegI src ) %{ // RegSS
2394     store_to_stackslot( cbuf, $primary, $src$$reg, $dst$$disp );
2395   %}
2396 
2397   // Special case for moving a register to a stack slot.
2398   enc_class RegSS( stackSlotI dst, eRegI src ) %{ // RegSS
2399     // Opcode already emitted
2400     emit_rm( cbuf, 0x02, $src$$reg, ESP_enc );   // R/M byte
2401     emit_rm( cbuf, 0x00, ESP_enc, ESP_enc);          // SIB byte
2402     emit_d32(cbuf, $dst$$disp);   // Displacement
2403   %}
2404 
2405   // Push the integer in stackSlot 'src' onto FP-stack
2406   enc_class Push_Mem_I( memory src ) %{    // FILD   [ESP+src]
2407     store_to_stackslot( cbuf, $primary, $secondary, $src$$disp );
2408   %}
2409 
2410   // Push FPU's TOS float to a stack-slot, and pop FPU-stack
2411   enc_class Pop_Mem_FPR( stackSlotF dst ) %{ // FSTP_S [ESP+dst]
2412     store_to_stackslot( cbuf, 0xD9, 0x03, $dst$$disp );
2413   %}
2414 
2415   // Same as Pop_Mem_F except for opcode
2416   // Push FPU's TOS double to a stack-slot, and pop FPU-stack
2417   enc_class Pop_Mem_DPR( stackSlotD dst ) %{ // FSTP_D [ESP+dst]
2418     store_to_stackslot( cbuf, 0xDD, 0x03, $dst$$disp );
2419   %}
2420 
2421   enc_class Pop_Reg_FPR( regFPR dst ) %{
2422     emit_opcode( cbuf, 0xDD );           // FSTP   ST(i)
2423     emit_d8( cbuf, 0xD8+$dst$$reg );
2424   %}
2425 
2426   enc_class Push_Reg_FPR( regFPR dst ) %{
2427     emit_opcode( cbuf, 0xD9 );           // FLD    ST(i-1)
2428     emit_d8( cbuf, 0xC0-1+$dst$$reg );
2429   %}
2430 
2431   // Push FPU's float to a stack-slot, and pop FPU-stack
2432   enc_class Pop_Mem_Reg_FPR( stackSlotF dst, regFPR src ) %{
2433     int pop = 0x02;
2434     if ($src$$reg != FPR1L_enc) {
2435       emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
2436       emit_d8( cbuf, 0xC0-1+$src$$reg );
2437       pop = 0x03;
2438     }
2439     store_to_stackslot( cbuf, 0xD9, pop, $dst$$disp ); // FST<P>_S  [ESP+dst]
2440   %}
2441 
2442   // Push FPU's double to a stack-slot, and pop FPU-stack
2443   enc_class Pop_Mem_Reg_DPR( stackSlotD dst, regDPR src ) %{
2444     int pop = 0x02;
2445     if ($src$$reg != FPR1L_enc) {
2446       emit_opcode( cbuf, 0xD9 );         // FLD    ST(i-1)
2447       emit_d8( cbuf, 0xC0-1+$src$$reg );
2448       pop = 0x03;
2449     }
2450     store_to_stackslot( cbuf, 0xDD, pop, $dst$$disp ); // FST<P>_D  [ESP+dst]
2451   %}
2452 
2453   // Push FPU's double to a FPU-stack-slot, and pop FPU-stack
2454   enc_class Pop_Reg_Reg_DPR( regDPR dst, regFPR src ) %{
2455     int pop = 0xD0 - 1; // -1 since we skip FLD
2456     if ($src$$reg != FPR1L_enc) {
2457       emit_opcode( cbuf, 0xD9 );         // FLD    ST(src-1)
2458       emit_d8( cbuf, 0xC0-1+$src$$reg );
2459       pop = 0xD8;
2460     }
2461     emit_opcode( cbuf, 0xDD );
2462     emit_d8( cbuf, pop+$dst$$reg );      // FST<P> ST(i)
2463   %}
2464 
2465 
2466   enc_class Push_Reg_Mod_DPR( regDPR dst, regDPR src) %{
2467     // load dst in FPR0
2468     emit_opcode( cbuf, 0xD9 );
2469     emit_d8( cbuf, 0xC0-1+$dst$$reg );
2470     if ($src$$reg != FPR1L_enc) {
2471       // fincstp
2472       emit_opcode (cbuf, 0xD9);
2473       emit_opcode (cbuf, 0xF7);
2474       // swap src with FPR1:
2475       // FXCH FPR1 with src
2476       emit_opcode(cbuf, 0xD9);
2477       emit_d8(cbuf, 0xC8-1+$src$$reg );
2478       // fdecstp
2479       emit_opcode (cbuf, 0xD9);
2480       emit_opcode (cbuf, 0xF6);
2481     }
2482   %}
2483 
2484   enc_class Push_ModD_encoding(regD src0, regD src1) %{
2485     MacroAssembler _masm(&cbuf);
2486     __ subptr(rsp, 8);
2487     __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
2488     __ fld_d(Address(rsp, 0));
2489     __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
2490     __ fld_d(Address(rsp, 0));
2491   %}
2492 
2493   enc_class Push_ModF_encoding(regF src0, regF src1) %{
2494     MacroAssembler _masm(&cbuf);
2495     __ subptr(rsp, 4);
2496     __ movflt(Address(rsp, 0), $src1$$XMMRegister);
2497     __ fld_s(Address(rsp, 0));
2498     __ movflt(Address(rsp, 0), $src0$$XMMRegister);
2499     __ fld_s(Address(rsp, 0));
2500   %}
2501 
2502   enc_class Push_ResultD(regD dst) %{
2503     MacroAssembler _masm(&cbuf);
2504     __ fstp_d(Address(rsp, 0));
2505     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
2506     __ addptr(rsp, 8);
2507   %}
2508 
2509   enc_class Push_ResultF(regF dst, immI d8) %{
2510     MacroAssembler _masm(&cbuf);
2511     __ fstp_s(Address(rsp, 0));
2512     __ movflt($dst$$XMMRegister, Address(rsp, 0));
2513     __ addptr(rsp, $d8$$constant);
2514   %}
2515 
2516   enc_class Push_SrcD(regD src) %{
2517     MacroAssembler _masm(&cbuf);
2518     __ subptr(rsp, 8);
2519     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
2520     __ fld_d(Address(rsp, 0));
2521   %}
2522 
2523   enc_class push_stack_temp_qword() %{
2524     MacroAssembler _masm(&cbuf);
2525     __ subptr(rsp, 8);
2526   %}
2527 
2528   enc_class pop_stack_temp_qword() %{
2529     MacroAssembler _masm(&cbuf);
2530     __ addptr(rsp, 8);
2531   %}
2532 
2533   enc_class push_xmm_to_fpr1(regD src) %{
2534     MacroAssembler _masm(&cbuf);
2535     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
2536     __ fld_d(Address(rsp, 0));
2537   %}
2538 
2539   enc_class Push_Result_Mod_DPR( regDPR src) %{
2540     if ($src$$reg != FPR1L_enc) {
2541       // fincstp
2542       emit_opcode (cbuf, 0xD9);
2543       emit_opcode (cbuf, 0xF7);
2544       // FXCH FPR1 with src
2545       emit_opcode(cbuf, 0xD9);
2546       emit_d8(cbuf, 0xC8-1+$src$$reg );
2547       // fdecstp
2548       emit_opcode (cbuf, 0xD9);
2549       emit_opcode (cbuf, 0xF6);
2550     }
2551     // // following asm replaced with Pop_Reg_F or Pop_Mem_F
2552     // // FSTP   FPR$dst$$reg
2553     // emit_opcode( cbuf, 0xDD );
2554     // emit_d8( cbuf, 0xD8+$dst$$reg );
2555   %}
2556 
2557   enc_class fnstsw_sahf_skip_parity() %{
2558     // fnstsw ax
2559     emit_opcode( cbuf, 0xDF );
2560     emit_opcode( cbuf, 0xE0 );
2561     // sahf
2562     emit_opcode( cbuf, 0x9E );
2563     // jnp  ::skip
2564     emit_opcode( cbuf, 0x7B );
2565     emit_opcode( cbuf, 0x05 );
2566   %}
2567 
2568   enc_class emitModDPR() %{
2569     // fprem must be iterative
2570     // :: loop
2571     // fprem
2572     emit_opcode( cbuf, 0xD9 );
2573     emit_opcode( cbuf, 0xF8 );
2574     // wait
2575     emit_opcode( cbuf, 0x9b );
2576     // fnstsw ax
2577     emit_opcode( cbuf, 0xDF );
2578     emit_opcode( cbuf, 0xE0 );
2579     // sahf
2580     emit_opcode( cbuf, 0x9E );
2581     // jp  ::loop
2582     emit_opcode( cbuf, 0x0F );
2583     emit_opcode( cbuf, 0x8A );
2584     emit_opcode( cbuf, 0xF4 );
2585     emit_opcode( cbuf, 0xFF );
2586     emit_opcode( cbuf, 0xFF );
2587     emit_opcode( cbuf, 0xFF );
2588   %}
2589 
2590   enc_class fpu_flags() %{
2591     // fnstsw_ax
2592     emit_opcode( cbuf, 0xDF);
2593     emit_opcode( cbuf, 0xE0);
2594     // test ax,0x0400
2595     emit_opcode( cbuf, 0x66 );   // operand-size prefix for 16-bit immediate
2596     emit_opcode( cbuf, 0xA9 );
2597     emit_d16   ( cbuf, 0x0400 );
2598     // // // This sequence works, but stalls for 12-16 cycles on PPro
2599     // // test rax,0x0400
2600     // emit_opcode( cbuf, 0xA9 );
2601     // emit_d32   ( cbuf, 0x00000400 );
2602     //
2603     // jz exit (no unordered comparison)
2604     emit_opcode( cbuf, 0x74 );
2605     emit_d8    ( cbuf, 0x02 );
2606     // mov ah,1 - treat as LT case (set carry flag)
2607     emit_opcode( cbuf, 0xB4 );
2608     emit_d8    ( cbuf, 0x01 );
2609     // sahf
2610     emit_opcode( cbuf, 0x9E);
2611   %}
2612 
2613   enc_class cmpF_P6_fixup() %{
2614     // Fixup the integer flags in case comparison involved a NaN
2615     //
2616     // JNP exit (no unordered comparison, P-flag is set by NaN)
2617     emit_opcode( cbuf, 0x7B );
2618     emit_d8    ( cbuf, 0x03 );
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     // NOP     // target for branch to avoid branch to branch
2625     emit_opcode( cbuf, 0x90);
2626   %}
2627 
2628 //     fnstsw_ax();
2629 //     sahf();
2630 //     movl(dst, nan_result);
2631 //     jcc(Assembler::parity, exit);
2632 //     movl(dst, less_result);
2633 //     jcc(Assembler::below, exit);
2634 //     movl(dst, equal_result);
2635 //     jcc(Assembler::equal, exit);
2636 //     movl(dst, greater_result);
2637 
2638 // less_result     =  1;
2639 // greater_result  = -1;
2640 // equal_result    = 0;
2641 // nan_result      = -1;
2642 
2643   enc_class CmpF_Result(eRegI dst) %{
2644     // fnstsw_ax();
2645     emit_opcode( cbuf, 0xDF);
2646     emit_opcode( cbuf, 0xE0);
2647     // sahf
2648     emit_opcode( cbuf, 0x9E);
2649     // movl(dst, nan_result);
2650     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2651     emit_d32( cbuf, -1 );
2652     // jcc(Assembler::parity, exit);
2653     emit_opcode( cbuf, 0x7A );
2654     emit_d8    ( cbuf, 0x13 );
2655     // movl(dst, less_result);
2656     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2657     emit_d32( cbuf, -1 );
2658     // jcc(Assembler::below, exit);
2659     emit_opcode( cbuf, 0x72 );
2660     emit_d8    ( cbuf, 0x0C );
2661     // movl(dst, equal_result);
2662     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2663     emit_d32( cbuf, 0 );
2664     // jcc(Assembler::equal, exit);
2665     emit_opcode( cbuf, 0x74 );
2666     emit_d8    ( cbuf, 0x05 );
2667     // movl(dst, greater_result);
2668     emit_opcode( cbuf, 0xB8 + $dst$$reg);
2669     emit_d32( cbuf, 1 );
2670   %}
2671 
2672 
2673   // Compare the longs and set flags
2674   // BROKEN!  Do Not use as-is
2675   enc_class cmpl_test( eRegL src1, eRegL src2 ) %{
2676     // CMP    $src1.hi,$src2.hi
2677     emit_opcode( cbuf, 0x3B );
2678     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
2679     // JNE,s  done
2680     emit_opcode(cbuf,0x75);
2681     emit_d8(cbuf, 2 );
2682     // CMP    $src1.lo,$src2.lo
2683     emit_opcode( cbuf, 0x3B );
2684     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2685 // done:
2686   %}
2687 
2688   enc_class convert_int_long( regL dst, eRegI src ) %{
2689     // mov $dst.lo,$src
2690     int dst_encoding = $dst$$reg;
2691     int src_encoding = $src$$reg;
2692     encode_Copy( cbuf, dst_encoding  , src_encoding );
2693     // mov $dst.hi,$src
2694     encode_Copy( cbuf, HIGH_FROM_LOW(dst_encoding), src_encoding );
2695     // sar $dst.hi,31
2696     emit_opcode( cbuf, 0xC1 );
2697     emit_rm(cbuf, 0x3, 7, HIGH_FROM_LOW(dst_encoding) );
2698     emit_d8(cbuf, 0x1F );
2699   %}
2700 
2701   enc_class convert_long_double( eRegL src ) %{
2702     // push $src.hi
2703     emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
2704     // push $src.lo
2705     emit_opcode(cbuf, 0x50+$src$$reg  );
2706     // fild 64-bits at [SP]
2707     emit_opcode(cbuf,0xdf);
2708     emit_d8(cbuf, 0x6C);
2709     emit_d8(cbuf, 0x24);
2710     emit_d8(cbuf, 0x00);
2711     // pop stack
2712     emit_opcode(cbuf, 0x83); // add  SP, #8
2713     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2714     emit_d8(cbuf, 0x8);
2715   %}
2716 
2717   enc_class multiply_con_and_shift_high( eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr ) %{
2718     // IMUL   EDX:EAX,$src1
2719     emit_opcode( cbuf, 0xF7 );
2720     emit_rm( cbuf, 0x3, 0x5, $src1$$reg );
2721     // SAR    EDX,$cnt-32
2722     int shift_count = ((int)$cnt$$constant) - 32;
2723     if (shift_count > 0) {
2724       emit_opcode(cbuf, 0xC1);
2725       emit_rm(cbuf, 0x3, 7, $dst$$reg );
2726       emit_d8(cbuf, shift_count);
2727     }
2728   %}
2729 
2730   // this version doesn't have add sp, 8
2731   enc_class convert_long_double2( eRegL src ) %{
2732     // push $src.hi
2733     emit_opcode(cbuf, 0x50+HIGH_FROM_LOW($src$$reg));
2734     // push $src.lo
2735     emit_opcode(cbuf, 0x50+$src$$reg  );
2736     // fild 64-bits at [SP]
2737     emit_opcode(cbuf,0xdf);
2738     emit_d8(cbuf, 0x6C);
2739     emit_d8(cbuf, 0x24);
2740     emit_d8(cbuf, 0x00);
2741   %}
2742 
2743   enc_class long_int_multiply( eADXRegL dst, nadxRegI src) %{
2744     // Basic idea: long = (long)int * (long)int
2745     // IMUL EDX:EAX, src
2746     emit_opcode( cbuf, 0xF7 );
2747     emit_rm( cbuf, 0x3, 0x5, $src$$reg);
2748   %}
2749 
2750   enc_class long_uint_multiply( eADXRegL dst, nadxRegI src) %{
2751     // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
2752     // MUL EDX:EAX, src
2753     emit_opcode( cbuf, 0xF7 );
2754     emit_rm( cbuf, 0x3, 0x4, $src$$reg);
2755   %}
2756 
2757   enc_class long_multiply( eADXRegL dst, eRegL src, eRegI tmp ) %{
2758     // Basic idea: lo(result) = lo(x_lo * y_lo)
2759     //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
2760     // MOV    $tmp,$src.lo
2761     encode_Copy( cbuf, $tmp$$reg, $src$$reg );
2762     // IMUL   $tmp,EDX
2763     emit_opcode( cbuf, 0x0F );
2764     emit_opcode( cbuf, 0xAF );
2765     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2766     // MOV    EDX,$src.hi
2767     encode_Copy( cbuf, HIGH_FROM_LOW($dst$$reg), HIGH_FROM_LOW($src$$reg) );
2768     // IMUL   EDX,EAX
2769     emit_opcode( cbuf, 0x0F );
2770     emit_opcode( cbuf, 0xAF );
2771     emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $dst$$reg );
2772     // ADD    $tmp,EDX
2773     emit_opcode( cbuf, 0x03 );
2774     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2775     // MUL   EDX:EAX,$src.lo
2776     emit_opcode( cbuf, 0xF7 );
2777     emit_rm( cbuf, 0x3, 0x4, $src$$reg );
2778     // ADD    EDX,ESI
2779     emit_opcode( cbuf, 0x03 );
2780     emit_rm( cbuf, 0x3, HIGH_FROM_LOW($dst$$reg), $tmp$$reg );
2781   %}
2782 
2783   enc_class long_multiply_con( eADXRegL dst, immL_127 src, eRegI tmp ) %{
2784     // Basic idea: lo(result) = lo(src * y_lo)
2785     //             hi(result) = hi(src * y_lo) + lo(src * y_hi)
2786     // IMUL   $tmp,EDX,$src
2787     emit_opcode( cbuf, 0x6B );
2788     emit_rm( cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($dst$$reg) );
2789     emit_d8( cbuf, (int)$src$$constant );
2790     // MOV    EDX,$src
2791     emit_opcode(cbuf, 0xB8 + EDX_enc);
2792     emit_d32( cbuf, (int)$src$$constant );
2793     // MUL   EDX:EAX,EDX
2794     emit_opcode( cbuf, 0xF7 );
2795     emit_rm( cbuf, 0x3, 0x4, EDX_enc );
2796     // ADD    EDX,ESI
2797     emit_opcode( cbuf, 0x03 );
2798     emit_rm( cbuf, 0x3, EDX_enc, $tmp$$reg );
2799   %}
2800 
2801   enc_class long_div( eRegL src1, eRegL src2 ) %{
2802     // PUSH src1.hi
2803     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
2804     // PUSH src1.lo
2805     emit_opcode(cbuf,               0x50+$src1$$reg  );
2806     // PUSH src2.hi
2807     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
2808     // PUSH src2.lo
2809     emit_opcode(cbuf,               0x50+$src2$$reg  );
2810     // CALL directly to the runtime
2811     cbuf.set_insts_mark();
2812     emit_opcode(cbuf,0xE8);       // Call into runtime
2813     emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::ldiv) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
2814     // Restore stack
2815     emit_opcode(cbuf, 0x83); // add  SP, #framesize
2816     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2817     emit_d8(cbuf, 4*4);
2818   %}
2819 
2820   enc_class long_mod( eRegL src1, eRegL src2 ) %{
2821     // PUSH src1.hi
2822     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src1$$reg) );
2823     // PUSH src1.lo
2824     emit_opcode(cbuf,               0x50+$src1$$reg  );
2825     // PUSH src2.hi
2826     emit_opcode(cbuf, HIGH_FROM_LOW(0x50+$src2$$reg) );
2827     // PUSH src2.lo
2828     emit_opcode(cbuf,               0x50+$src2$$reg  );
2829     // CALL directly to the runtime
2830     cbuf.set_insts_mark();
2831     emit_opcode(cbuf,0xE8);       // Call into runtime
2832     emit_d32_reloc(cbuf, (CAST_FROM_FN_PTR(address, SharedRuntime::lrem ) - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
2833     // Restore stack
2834     emit_opcode(cbuf, 0x83); // add  SP, #framesize
2835     emit_rm(cbuf, 0x3, 0x00, ESP_enc);
2836     emit_d8(cbuf, 4*4);
2837   %}
2838 
2839   enc_class long_cmp_flags0( eRegL src, eRegI tmp ) %{
2840     // MOV   $tmp,$src.lo
2841     emit_opcode(cbuf, 0x8B);
2842     emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg);
2843     // OR    $tmp,$src.hi
2844     emit_opcode(cbuf, 0x0B);
2845     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg));
2846   %}
2847 
2848   enc_class long_cmp_flags1( eRegL src1, eRegL src2 ) %{
2849     // CMP    $src1.lo,$src2.lo
2850     emit_opcode( cbuf, 0x3B );
2851     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2852     // JNE,s  skip
2853     emit_cc(cbuf, 0x70, 0x5);
2854     emit_d8(cbuf,2);
2855     // CMP    $src1.hi,$src2.hi
2856     emit_opcode( cbuf, 0x3B );
2857     emit_rm(cbuf, 0x3, HIGH_FROM_LOW($src1$$reg), HIGH_FROM_LOW($src2$$reg) );
2858   %}
2859 
2860   enc_class long_cmp_flags2( eRegL src1, eRegL src2, eRegI tmp ) %{
2861     // CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits
2862     emit_opcode( cbuf, 0x3B );
2863     emit_rm(cbuf, 0x3, $src1$$reg, $src2$$reg );
2864     // MOV    $tmp,$src1.hi
2865     emit_opcode( cbuf, 0x8B );
2866     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src1$$reg) );
2867     // SBB   $tmp,$src2.hi\t! Compute flags for long compare
2868     emit_opcode( cbuf, 0x1B );
2869     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src2$$reg) );
2870   %}
2871 
2872   enc_class long_cmp_flags3( eRegL src, eRegI tmp ) %{
2873     // XOR    $tmp,$tmp
2874     emit_opcode(cbuf,0x33);  // XOR
2875     emit_rm(cbuf,0x3, $tmp$$reg, $tmp$$reg);
2876     // CMP    $tmp,$src.lo
2877     emit_opcode( cbuf, 0x3B );
2878     emit_rm(cbuf, 0x3, $tmp$$reg, $src$$reg );
2879     // SBB    $tmp,$src.hi
2880     emit_opcode( cbuf, 0x1B );
2881     emit_rm(cbuf, 0x3, $tmp$$reg, HIGH_FROM_LOW($src$$reg) );
2882   %}
2883 
2884  // Sniff, sniff... smells like Gnu Superoptimizer
2885   enc_class neg_long( eRegL dst ) %{
2886     emit_opcode(cbuf,0xF7);    // NEG hi
2887     emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
2888     emit_opcode(cbuf,0xF7);    // NEG lo
2889     emit_rm    (cbuf,0x3, 0x3,               $dst$$reg );
2890     emit_opcode(cbuf,0x83);    // SBB hi,0
2891     emit_rm    (cbuf,0x3, 0x3, HIGH_FROM_LOW($dst$$reg));
2892     emit_d8    (cbuf,0 );
2893   %}
2894 
2895 
2896   // Because the transitions from emitted code to the runtime
2897   // monitorenter/exit helper stubs are so slow it's critical that
2898   // we inline both the stack-locking fast-path and the inflated fast path.
2899   //
2900   // See also: cmpFastLock and cmpFastUnlock.
2901   //
2902   // What follows is a specialized inline transliteration of the code
2903   // in slow_enter() and slow_exit().  If we're concerned about I$ bloat
2904   // another option would be to emit TrySlowEnter and TrySlowExit methods
2905   // at startup-time.  These methods would accept arguments as
2906   // (rax,=Obj, rbx=Self, rcx=box, rdx=Scratch) and return success-failure
2907   // indications in the icc.ZFlag.  Fast_Lock and Fast_Unlock would simply
2908   // marshal the arguments and emit calls to TrySlowEnter and TrySlowExit.
2909   // In practice, however, the # of lock sites is bounded and is usually small.
2910   // Besides the call overhead, TrySlowEnter and TrySlowExit might suffer
2911   // if the processor uses simple bimodal branch predictors keyed by EIP
2912   // Since the helper routines would be called from multiple synchronization
2913   // sites.
2914   //
2915   // An even better approach would be write "MonitorEnter()" and "MonitorExit()"
2916   // in java - using j.u.c and unsafe - and just bind the lock and unlock sites
2917   // to those specialized methods.  That'd give us a mostly platform-independent
2918   // implementation that the JITs could optimize and inline at their pleasure.
2919   // Done correctly, the only time we'd need to cross to native could would be
2920   // to park() or unpark() threads.  We'd also need a few more unsafe operators
2921   // to (a) prevent compiler-JIT reordering of non-volatile accesses, and
2922   // (b) explicit barriers or fence operations.
2923   //
2924   // TODO:
2925   //
2926   // *  Arrange for C2 to pass "Self" into Fast_Lock and Fast_Unlock in one of the registers (scr).
2927   //    This avoids manifesting the Self pointer in the Fast_Lock and Fast_Unlock terminals.
2928   //    Given TLAB allocation, Self is usually manifested in a register, so passing it into
2929   //    the lock operators would typically be faster than reifying Self.
2930   //
2931   // *  Ideally I'd define the primitives as:
2932   //       fast_lock   (nax Obj, nax box, EAX tmp, nax scr) where box, tmp and scr are KILLED.
2933   //       fast_unlock (nax Obj, EAX box, nax tmp) where box and tmp are KILLED
2934   //    Unfortunately ADLC bugs prevent us from expressing the ideal form.
2935   //    Instead, we're stuck with a rather awkward and brittle register assignments below.
2936   //    Furthermore the register assignments are overconstrained, possibly resulting in
2937   //    sub-optimal code near the synchronization site.
2938   //
2939   // *  Eliminate the sp-proximity tests and just use "== Self" tests instead.
2940   //    Alternately, use a better sp-proximity test.
2941   //
2942   // *  Currently ObjectMonitor._Owner can hold either an sp value or a (THREAD *) value.
2943   //    Either one is sufficient to uniquely identify a thread.
2944   //    TODO: eliminate use of sp in _owner and use get_thread(tr) instead.
2945   //
2946   // *  Intrinsify notify() and notifyAll() for the common cases where the
2947   //    object is locked by the calling thread but the waitlist is empty.
2948   //    avoid the expensive JNI call to JVM_Notify() and JVM_NotifyAll().
2949   //
2950   // *  use jccb and jmpb instead of jcc and jmp to improve code density.
2951   //    But beware of excessive branch density on AMD Opterons.
2952   //
2953   // *  Both Fast_Lock and Fast_Unlock set the ICC.ZF to indicate success
2954   //    or failure of the fast-path.  If the fast-path fails then we pass
2955   //    control to the slow-path, typically in C.  In Fast_Lock and
2956   //    Fast_Unlock we often branch to DONE_LABEL, just to find that C2
2957   //    will emit a conditional branch immediately after the node.
2958   //    So we have branches to branches and lots of ICC.ZF games.
2959   //    Instead, it might be better to have C2 pass a "FailureLabel"
2960   //    into Fast_Lock and Fast_Unlock.  In the case of success, control
2961   //    will drop through the node.  ICC.ZF is undefined at exit.
2962   //    In the case of failure, the node will branch directly to the
2963   //    FailureLabel
2964 
2965 
2966   // obj: object to lock
2967   // box: on-stack box address (displaced header location) - KILLED
2968   // rax,: tmp -- KILLED
2969   // scr: tmp -- KILLED
2970   enc_class Fast_Lock( eRegP obj, eRegP box, eAXRegI tmp, eRegP scr ) %{
2971 
2972     Register objReg = as_Register($obj$$reg);
2973     Register boxReg = as_Register($box$$reg);
2974     Register tmpReg = as_Register($tmp$$reg);
2975     Register scrReg = as_Register($scr$$reg);
2976 
2977     // Ensure the register assignents are disjoint
2978     guarantee (objReg != boxReg, "") ;
2979     guarantee (objReg != tmpReg, "") ;
2980     guarantee (objReg != scrReg, "") ;
2981     guarantee (boxReg != tmpReg, "") ;
2982     guarantee (boxReg != scrReg, "") ;
2983     guarantee (tmpReg == as_Register(EAX_enc), "") ;
2984 
2985     MacroAssembler masm(&cbuf);
2986 
2987     if (_counters != NULL) {
2988       masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
2989     }
2990     if (EmitSync & 1) {
2991         // set box->dhw = unused_mark (3)
2992         // Force all sync thru slow-path: slow_enter() and slow_exit() 
2993         masm.movptr (Address(boxReg, 0), int32_t(markOopDesc::unused_mark())) ;             
2994         masm.cmpptr (rsp, (int32_t)0) ;                        
2995     } else 
2996     if (EmitSync & 2) { 
2997         Label DONE_LABEL ;           
2998         if (UseBiasedLocking) {
2999            // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3000            masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3001         }
3002 
3003         masm.movptr(tmpReg, Address(objReg, 0)) ;          // fetch markword 
3004         masm.orptr (tmpReg, 0x1);
3005         masm.movptr(Address(boxReg, 0), tmpReg);           // Anticipate successful CAS 
3006         if (os::is_MP()) { masm.lock();  }
3007         masm.cmpxchgptr(boxReg, Address(objReg, 0));          // Updates tmpReg
3008         masm.jcc(Assembler::equal, DONE_LABEL);
3009         // Recursive locking
3010         masm.subptr(tmpReg, rsp);
3011         masm.andptr(tmpReg, (int32_t) 0xFFFFF003 );
3012         masm.movptr(Address(boxReg, 0), tmpReg);
3013         masm.bind(DONE_LABEL) ; 
3014     } else {  
3015       // Possible cases that we'll encounter in fast_lock 
3016       // ------------------------------------------------
3017       // * Inflated
3018       //    -- unlocked
3019       //    -- Locked
3020       //       = by self
3021       //       = by other
3022       // * biased
3023       //    -- by Self
3024       //    -- by other
3025       // * neutral
3026       // * stack-locked
3027       //    -- by self
3028       //       = sp-proximity test hits
3029       //       = sp-proximity test generates false-negative
3030       //    -- by other
3031       //
3032 
3033       Label IsInflated, DONE_LABEL, PopDone ;
3034 
3035       // TODO: optimize away redundant LDs of obj->mark and improve the markword triage
3036       // order to reduce the number of conditional branches in the most common cases.
3037       // Beware -- there's a subtle invariant that fetch of the markword
3038       // at [FETCH], below, will never observe a biased encoding (*101b).
3039       // If this invariant is not held we risk exclusion (safety) failure.
3040       if (UseBiasedLocking && !UseOptoBiasInlining) {
3041         masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3042       }
3043 
3044       masm.movptr(tmpReg, Address(objReg, 0)) ;         // [FETCH]
3045       masm.testptr(tmpReg, 0x02) ;                      // Inflated v (Stack-locked or neutral)
3046       masm.jccb  (Assembler::notZero, IsInflated) ;
3047 
3048       // Attempt stack-locking ...
3049       masm.orptr (tmpReg, 0x1);
3050       masm.movptr(Address(boxReg, 0), tmpReg);          // Anticipate successful CAS
3051       if (os::is_MP()) { masm.lock();  }
3052       masm.cmpxchgptr(boxReg, Address(objReg, 0));           // Updates tmpReg
3053       if (_counters != NULL) {
3054         masm.cond_inc32(Assembler::equal,
3055                         ExternalAddress((address)_counters->fast_path_entry_count_addr()));
3056       }
3057       masm.jccb (Assembler::equal, DONE_LABEL);
3058 
3059       // Recursive locking
3060       masm.subptr(tmpReg, rsp);
3061       masm.andptr(tmpReg, 0xFFFFF003 );
3062       masm.movptr(Address(boxReg, 0), tmpReg);
3063       if (_counters != NULL) {
3064         masm.cond_inc32(Assembler::equal,
3065                         ExternalAddress((address)_counters->fast_path_entry_count_addr()));
3066       }
3067       masm.jmp  (DONE_LABEL) ;
3068 
3069       masm.bind (IsInflated) ;
3070 
3071       // The object is inflated.
3072       //
3073       // TODO-FIXME: eliminate the ugly use of manifest constants:
3074       //   Use markOopDesc::monitor_value instead of "2".
3075       //   use markOop::unused_mark() instead of "3".
3076       // The tmpReg value is an objectMonitor reference ORed with
3077       // markOopDesc::monitor_value (2).   We can either convert tmpReg to an
3078       // objectmonitor pointer by masking off the "2" bit or we can just
3079       // use tmpReg as an objectmonitor pointer but bias the objectmonitor
3080       // field offsets with "-2" to compensate for and annul the low-order tag bit.
3081       //
3082       // I use the latter as it avoids AGI stalls.
3083       // As such, we write "mov r, [tmpReg+OFFSETOF(Owner)-2]"
3084       // instead of "mov r, [tmpReg+OFFSETOF(Owner)]".
3085       //
3086       #define OFFSET_SKEWED(f) ((ObjectMonitor::f ## _offset_in_bytes())-2)
3087 
3088       // boxReg refers to the on-stack BasicLock in the current frame.
3089       // We'd like to write:
3090       //   set box->_displaced_header = markOop::unused_mark().  Any non-0 value suffices.
3091       // This is convenient but results a ST-before-CAS penalty.  The following CAS suffers
3092       // additional latency as we have another ST in the store buffer that must drain.
3093 
3094       if (EmitSync & 8192) { 
3095          masm.movptr(Address(boxReg, 0), 3) ;            // results in ST-before-CAS penalty
3096          masm.get_thread (scrReg) ; 
3097          masm.movptr(boxReg, tmpReg);                    // consider: LEA box, [tmp-2] 
3098          masm.movptr(tmpReg, NULL_WORD);                 // consider: xor vs mov
3099          if (os::is_MP()) { masm.lock(); } 
3100          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3101       } else 
3102       if ((EmitSync & 128) == 0) {                      // avoid ST-before-CAS
3103          masm.movptr(scrReg, boxReg) ; 
3104          masm.movptr(boxReg, tmpReg);                   // consider: LEA box, [tmp-2] 
3105 
3106          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
3107          if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3108             // prefetchw [eax + Offset(_owner)-2]
3109             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
3110          }
3111 
3112          if ((EmitSync & 64) == 0) {
3113            // Optimistic form: consider XORL tmpReg,tmpReg
3114            masm.movptr(tmpReg, NULL_WORD) ; 
3115          } else { 
3116            // Can suffer RTS->RTO upgrades on shared or cold $ lines
3117            // Test-And-CAS instead of CAS
3118            masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
3119            masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
3120            masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
3121          }
3122 
3123          // Appears unlocked - try to swing _owner from null to non-null.
3124          // Ideally, I'd manifest "Self" with get_thread and then attempt
3125          // to CAS the register containing Self into m->Owner.
3126          // But we don't have enough registers, so instead we can either try to CAS
3127          // rsp or the address of the box (in scr) into &m->owner.  If the CAS succeeds
3128          // we later store "Self" into m->Owner.  Transiently storing a stack address
3129          // (rsp or the address of the box) into  m->owner is harmless.
3130          // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
3131          if (os::is_MP()) { masm.lock();  }
3132          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; 
3133          masm.movptr(Address(scrReg, 0), 3) ;          // box->_displaced_header = 3
3134          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3135          masm.get_thread (scrReg) ;                    // beware: clobbers ICCs
3136          masm.movptr(Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2), scrReg) ; 
3137          masm.xorptr(boxReg, boxReg) ;                 // set icc.ZFlag = 1 to indicate success
3138                        
3139          // If the CAS fails we can either retry or pass control to the slow-path.  
3140          // We use the latter tactic.  
3141          // Pass the CAS result in the icc.ZFlag into DONE_LABEL
3142          // If the CAS was successful ...
3143          //   Self has acquired the lock
3144          //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
3145          // Intentional fall-through into DONE_LABEL ...
3146       } else {
3147          masm.movptr(Address(boxReg, 0), 3) ;       // results in ST-before-CAS penalty
3148          masm.movptr(boxReg, tmpReg) ; 
3149 
3150          // Using a prefetchw helps avoid later RTS->RTO upgrades and cache probes
3151          if ((EmitSync & 2048) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3152             // prefetchw [eax + Offset(_owner)-2]
3153             masm.prefetchw(Address(rax, ObjectMonitor::owner_offset_in_bytes()-2));
3154          }
3155 
3156          if ((EmitSync & 64) == 0) {
3157            // Optimistic form
3158            masm.xorptr  (tmpReg, tmpReg) ; 
3159          } else { 
3160            // Can suffer RTS->RTO upgrades on shared or cold $ lines
3161            masm.movptr(tmpReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;   // rax, = m->_owner
3162            masm.testptr(tmpReg, tmpReg) ;                   // Locked ? 
3163            masm.jccb  (Assembler::notZero, DONE_LABEL) ;                   
3164          }
3165 
3166          // Appears unlocked - try to swing _owner from null to non-null.
3167          // Use either "Self" (in scr) or rsp as thread identity in _owner.
3168          // Invariant: tmpReg == 0.  tmpReg is EAX which is the implicit cmpxchg comparand.
3169          masm.get_thread (scrReg) ;
3170          if (os::is_MP()) { masm.lock(); }
3171          masm.cmpxchgptr(scrReg, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3172 
3173          // If the CAS fails we can either retry or pass control to the slow-path.
3174          // We use the latter tactic.
3175          // Pass the CAS result in the icc.ZFlag into DONE_LABEL
3176          // If the CAS was successful ...
3177          //   Self has acquired the lock
3178          //   Invariant: m->_recursions should already be 0, so we don't need to explicitly set it.
3179          // Intentional fall-through into DONE_LABEL ...
3180       }
3181 
3182       // DONE_LABEL is a hot target - we'd really like to place it at the
3183       // start of cache line by padding with NOPs.
3184       // See the AMD and Intel software optimization manuals for the
3185       // most efficient "long" NOP encodings.
3186       // Unfortunately none of our alignment mechanisms suffice.
3187       masm.bind(DONE_LABEL);
3188 
3189       // Avoid branch-to-branch on AMD processors
3190       // This appears to be superstition.
3191       if (EmitSync & 32) masm.nop() ;
3192 
3193 
3194       // At DONE_LABEL the icc ZFlag is set as follows ...
3195       // Fast_Unlock uses the same protocol.
3196       // ZFlag == 1 -> Success
3197       // ZFlag == 0 -> Failure - force control through the slow-path
3198     }
3199   %}
3200 
3201   // obj: object to unlock
3202   // box: box address (displaced header location), killed.  Must be EAX.
3203   // rbx,: killed tmp; cannot be obj nor box.
3204   //
3205   // Some commentary on balanced locking:
3206   //
3207   // Fast_Lock and Fast_Unlock are emitted only for provably balanced lock sites.
3208   // Methods that don't have provably balanced locking are forced to run in the
3209   // interpreter - such methods won't be compiled to use fast_lock and fast_unlock.
3210   // The interpreter provides two properties:
3211   // I1:  At return-time the interpreter automatically and quietly unlocks any
3212   //      objects acquired the current activation (frame).  Recall that the
3213   //      interpreter maintains an on-stack list of locks currently held by
3214   //      a frame.
3215   // I2:  If a method attempts to unlock an object that is not held by the
3216   //      the frame the interpreter throws IMSX.
3217   //
3218   // Lets say A(), which has provably balanced locking, acquires O and then calls B().
3219   // B() doesn't have provably balanced locking so it runs in the interpreter.
3220   // Control returns to A() and A() unlocks O.  By I1 and I2, above, we know that O
3221   // is still locked by A().
3222   //
3223   // The only other source of unbalanced locking would be JNI.  The "Java Native Interface:
3224   // Programmer's Guide and Specification" claims that an object locked by jni_monitorenter
3225   // should not be unlocked by "normal" java-level locking and vice-versa.  The specification
3226   // doesn't specify what will occur if a program engages in such mixed-mode locking, however.
3227 
3228   enc_class Fast_Unlock( nabxRegP obj, eAXRegP box, eRegP tmp) %{
3229 
3230     Register objReg = as_Register($obj$$reg);
3231     Register boxReg = as_Register($box$$reg);
3232     Register tmpReg = as_Register($tmp$$reg);
3233 
3234     guarantee (objReg != boxReg, "") ;
3235     guarantee (objReg != tmpReg, "") ;
3236     guarantee (boxReg != tmpReg, "") ;
3237     guarantee (boxReg == as_Register(EAX_enc), "") ;
3238     MacroAssembler masm(&cbuf);
3239 
3240     if (EmitSync & 4) {
3241       // Disable - inhibit all inlining.  Force control through the slow-path
3242       masm.cmpptr (rsp, 0) ; 
3243     } else 
3244     if (EmitSync & 8) {
3245       Label DONE_LABEL ;
3246       if (UseBiasedLocking) {
3247          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3248       }
3249       // classic stack-locking code ...
3250       masm.movptr(tmpReg, Address(boxReg, 0)) ;
3251       masm.testptr(tmpReg, tmpReg) ;
3252       masm.jcc   (Assembler::zero, DONE_LABEL) ;
3253       if (os::is_MP()) { masm.lock(); }
3254       masm.cmpxchgptr(tmpReg, Address(objReg, 0));          // Uses EAX which is box
3255       masm.bind(DONE_LABEL);
3256     } else {
3257       Label DONE_LABEL, Stacked, CheckSucc, Inflated ;
3258 
3259       // Critically, the biased locking test must have precedence over
3260       // and appear before the (box->dhw == 0) recursive stack-lock test.
3261       if (UseBiasedLocking && !UseOptoBiasInlining) {
3262          masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3263       }
3264       
3265       masm.cmpptr(Address(boxReg, 0), 0) ;            // Examine the displaced header
3266       masm.movptr(tmpReg, Address(objReg, 0)) ;       // Examine the object's markword
3267       masm.jccb  (Assembler::zero, DONE_LABEL) ;      // 0 indicates recursive stack-lock
3268 
3269       masm.testptr(tmpReg, 0x02) ;                     // Inflated? 
3270       masm.jccb  (Assembler::zero, Stacked) ;
3271 
3272       masm.bind  (Inflated) ;
3273       // It's inflated.
3274       // Despite our balanced locking property we still check that m->_owner == Self
3275       // as java routines or native JNI code called by this thread might
3276       // have released the lock.
3277       // Refer to the comments in synchronizer.cpp for how we might encode extra
3278       // state in _succ so we can avoid fetching EntryList|cxq.
3279       //
3280       // I'd like to add more cases in fast_lock() and fast_unlock() --
3281       // such as recursive enter and exit -- but we have to be wary of
3282       // I$ bloat, T$ effects and BP$ effects.
3283       //
3284       // If there's no contention try a 1-0 exit.  That is, exit without
3285       // a costly MEMBAR or CAS.  See synchronizer.cpp for details on how
3286       // we detect and recover from the race that the 1-0 exit admits.
3287       //
3288       // Conceptually Fast_Unlock() must execute a STST|LDST "release" barrier
3289       // before it STs null into _owner, releasing the lock.  Updates
3290       // to data protected by the critical section must be visible before
3291       // we drop the lock (and thus before any other thread could acquire
3292       // the lock and observe the fields protected by the lock).
3293       // IA32's memory-model is SPO, so STs are ordered with respect to
3294       // each other and there's no need for an explicit barrier (fence).
3295       // See also http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
3296 
3297       masm.get_thread (boxReg) ;
3298       if ((EmitSync & 4096) && VM_Version::supports_3dnow_prefetch() && os::is_MP()) {
3299         // prefetchw [ebx + Offset(_owner)-2]
3300         masm.prefetchw(Address(rbx, ObjectMonitor::owner_offset_in_bytes()-2));
3301       }
3302 
3303       // Note that we could employ various encoding schemes to reduce
3304       // the number of loads below (currently 4) to just 2 or 3.
3305       // Refer to the comments in synchronizer.cpp.
3306       // In practice the chain of fetches doesn't seem to impact performance, however.
3307       if ((EmitSync & 65536) == 0 && (EmitSync & 256)) {
3308          // Attempt to reduce branch density - AMD's branch predictor.
3309          masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
3310          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3311          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
3312          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
3313          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3314          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3315          masm.jmpb  (DONE_LABEL) ; 
3316       } else { 
3317          masm.xorptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;  
3318          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3319          masm.jccb  (Assembler::notZero, DONE_LABEL) ; 
3320          masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; 
3321          masm.orptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; 
3322          masm.jccb  (Assembler::notZero, CheckSucc) ; 
3323          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3324          masm.jmpb  (DONE_LABEL) ; 
3325       }
3326 
3327       // The Following code fragment (EmitSync & 65536) improves the performance of
3328       // contended applications and contended synchronization microbenchmarks.
3329       // Unfortunately the emission of the code - even though not executed - causes regressions
3330       // in scimark and jetstream, evidently because of $ effects.  Replacing the code
3331       // with an equal number of never-executed NOPs results in the same regression.
3332       // We leave it off by default.
3333 
3334       if ((EmitSync & 65536) != 0) {
3335          Label LSuccess, LGoSlowPath ;
3336 
3337          masm.bind  (CheckSucc) ;
3338 
3339          // Optional pre-test ... it's safe to elide this
3340          if ((EmitSync & 16) == 0) { 
3341             masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
3342             masm.jccb  (Assembler::zero, LGoSlowPath) ; 
3343          }
3344 
3345          // We have a classic Dekker-style idiom:
3346          //    ST m->_owner = 0 ; MEMBAR; LD m->_succ
3347          // There are a number of ways to implement the barrier:
3348          // (1) lock:andl &m->_owner, 0
3349          //     is fast, but mask doesn't currently support the "ANDL M,IMM32" form.
3350          //     LOCK: ANDL [ebx+Offset(_Owner)-2], 0
3351          //     Encodes as 81 31 OFF32 IMM32 or 83 63 OFF8 IMM8
3352          // (2) If supported, an explicit MFENCE is appealing.
3353          //     In older IA32 processors MFENCE is slower than lock:add or xchg
3354          //     particularly if the write-buffer is full as might be the case if
3355          //     if stores closely precede the fence or fence-equivalent instruction.
3356          //     In more modern implementations MFENCE appears faster, however.
3357          // (3) In lieu of an explicit fence, use lock:addl to the top-of-stack
3358          //     The $lines underlying the top-of-stack should be in M-state.
3359          //     The locked add instruction is serializing, of course.
3360          // (4) Use xchg, which is serializing
3361          //     mov boxReg, 0; xchgl boxReg, [tmpReg + Offset(_owner)-2] also works
3362          // (5) ST m->_owner = 0 and then execute lock:orl &m->_succ, 0.
3363          //     The integer condition codes will tell us if succ was 0.
3364          //     Since _succ and _owner should reside in the same $line and
3365          //     we just stored into _owner, it's likely that the $line
3366          //     remains in M-state for the lock:orl.
3367          //
3368          // We currently use (3), although it's likely that switching to (2)
3369          // is correct for the future.
3370             
3371          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), NULL_WORD) ; 
3372          if (os::is_MP()) { 
3373             if (VM_Version::supports_sse2() && 1 == FenceInstruction) { 
3374               masm.mfence();
3375             } else { 
3376               masm.lock () ; masm.addptr(Address(rsp, 0), 0) ; 
3377             }
3378          }
3379          // Ratify _succ remains non-null
3380          masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), 0) ; 
3381          masm.jccb  (Assembler::notZero, LSuccess) ; 
3382 
3383          masm.xorptr(boxReg, boxReg) ;                  // box is really EAX
3384          if (os::is_MP()) { masm.lock(); }
3385          masm.cmpxchgptr(rsp, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3386          masm.jccb  (Assembler::notEqual, LSuccess) ;
3387          // Since we're low on registers we installed rsp as a placeholding in _owner.
3388          // Now install Self over rsp.  This is safe as we're transitioning from
3389          // non-null to non=null
3390          masm.get_thread (boxReg) ;
3391          masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), boxReg) ;
3392          // Intentional fall-through into LGoSlowPath ...
3393 
3394          masm.bind  (LGoSlowPath) ; 
3395          masm.orptr(boxReg, 1) ;                      // set ICC.ZF=0 to indicate failure
3396          masm.jmpb  (DONE_LABEL) ; 
3397 
3398          masm.bind  (LSuccess) ; 
3399          masm.xorptr(boxReg, boxReg) ;                 // set ICC.ZF=1 to indicate success
3400          masm.jmpb  (DONE_LABEL) ; 
3401       }
3402 
3403       masm.bind (Stacked) ;
3404       // It's not inflated and it's not recursively stack-locked and it's not biased.
3405       // It must be stack-locked.
3406       // Try to reset the header to displaced header.
3407       // The "box" value on the stack is stable, so we can reload
3408       // and be assured we observe the same value as above.
3409       masm.movptr(tmpReg, Address(boxReg, 0)) ;
3410       if (os::is_MP()) {   masm.lock();    }
3411       masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses EAX which is box
3412       // Intention fall-thru into DONE_LABEL
3413 
3414 
3415       // DONE_LABEL is a hot target - we'd really like to place it at the
3416       // start of cache line by padding with NOPs.
3417       // See the AMD and Intel software optimization manuals for the
3418       // most efficient "long" NOP encodings.
3419       // Unfortunately none of our alignment mechanisms suffice.
3420       if ((EmitSync & 65536) == 0) {
3421          masm.bind (CheckSucc) ;
3422       }
3423       masm.bind(DONE_LABEL);
3424 
3425       // Avoid branch to branch on AMD processors
3426       if (EmitSync & 32768) { masm.nop() ; }
3427     }
3428   %}
3429 
3430 
3431   enc_class enc_pop_rdx() %{
3432     emit_opcode(cbuf,0x5A);
3433   %}
3434 
3435   enc_class enc_rethrow() %{
3436     cbuf.set_insts_mark();
3437     emit_opcode(cbuf, 0xE9);        // jmp    entry
3438     emit_d32_reloc(cbuf, (int)OptoRuntime::rethrow_stub() - ((int)cbuf.insts_end())-4,
3439                    runtime_call_Relocation::spec(), RELOC_IMM32 );
3440   %}
3441 
3442 
3443   // Convert a double to an int.  Java semantics require we do complex
3444   // manglelations in the corner cases.  So we set the rounding mode to
3445   // 'zero', store the darned double down as an int, and reset the
3446   // rounding mode to 'nearest'.  The hardware throws an exception which
3447   // patches up the correct value directly to the stack.
3448   enc_class DPR2I_encoding( regDPR src ) %{
3449     // Flip to round-to-zero mode.  We attempted to allow invalid-op
3450     // exceptions here, so that a NAN or other corner-case value will
3451     // thrown an exception (but normal values get converted at full speed).
3452     // However, I2C adapters and other float-stack manglers leave pending
3453     // invalid-op exceptions hanging.  We would have to clear them before
3454     // enabling them and that is more expensive than just testing for the
3455     // invalid value Intel stores down in the corner cases.
3456     emit_opcode(cbuf,0xD9);            // FLDCW  trunc
3457     emit_opcode(cbuf,0x2D);
3458     emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
3459     // Allocate a word
3460     emit_opcode(cbuf,0x83);            // SUB ESP,4
3461     emit_opcode(cbuf,0xEC);
3462     emit_d8(cbuf,0x04);
3463     // Encoding assumes a double has been pushed into FPR0.
3464     // Store down the double as an int, popping the FPU stack
3465     emit_opcode(cbuf,0xDB);            // FISTP [ESP]
3466     emit_opcode(cbuf,0x1C);
3467     emit_d8(cbuf,0x24);
3468     // Restore the rounding mode; mask the exception
3469     emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
3470     emit_opcode(cbuf,0x2D);
3471     emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
3472         ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
3473         : (int)StubRoutines::addr_fpu_cntrl_wrd_std());
3474 
3475     // Load the converted int; adjust CPU stack
3476     emit_opcode(cbuf,0x58);       // POP EAX
3477     emit_opcode(cbuf,0x3D);       // CMP EAX,imm
3478     emit_d32   (cbuf,0x80000000); //         0x80000000
3479     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3480     emit_d8    (cbuf,0x07);       // Size of slow_call
3481     // Push src onto stack slow-path
3482     emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
3483     emit_d8    (cbuf,0xC0-1+$src$$reg );
3484     // CALL directly to the runtime
3485     cbuf.set_insts_mark();
3486     emit_opcode(cbuf,0xE8);       // Call into runtime
3487     emit_d32_reloc(cbuf, (StubRoutines::d2i_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
3488     // Carry on here...
3489   %}
3490 
3491   enc_class DPR2L_encoding( regDPR src ) %{
3492     emit_opcode(cbuf,0xD9);            // FLDCW  trunc
3493     emit_opcode(cbuf,0x2D);
3494     emit_d32(cbuf,(int)StubRoutines::addr_fpu_cntrl_wrd_trunc());
3495     // Allocate a word
3496     emit_opcode(cbuf,0x83);            // SUB ESP,8
3497     emit_opcode(cbuf,0xEC);
3498     emit_d8(cbuf,0x08);
3499     // Encoding assumes a double has been pushed into FPR0.
3500     // Store down the double as a long, popping the FPU stack
3501     emit_opcode(cbuf,0xDF);            // FISTP [ESP]
3502     emit_opcode(cbuf,0x3C);
3503     emit_d8(cbuf,0x24);
3504     // Restore the rounding mode; mask the exception
3505     emit_opcode(cbuf,0xD9);            // FLDCW   std/24-bit mode
3506     emit_opcode(cbuf,0x2D);
3507     emit_d32( cbuf, Compile::current()->in_24_bit_fp_mode()
3508         ? (int)StubRoutines::addr_fpu_cntrl_wrd_24()
3509         : (int)StubRoutines::addr_fpu_cntrl_wrd_std());
3510 
3511     // Load the converted int; adjust CPU stack
3512     emit_opcode(cbuf,0x58);       // POP EAX
3513     emit_opcode(cbuf,0x5A);       // POP EDX
3514     emit_opcode(cbuf,0x81);       // CMP EDX,imm
3515     emit_d8    (cbuf,0xFA);       // rdx
3516     emit_d32   (cbuf,0x80000000); //         0x80000000
3517     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3518     emit_d8    (cbuf,0x07+4);     // Size of slow_call
3519     emit_opcode(cbuf,0x85);       // TEST EAX,EAX
3520     emit_opcode(cbuf,0xC0);       // 2/rax,/rax,
3521     emit_opcode(cbuf,0x75);       // JNE around_slow_call
3522     emit_d8    (cbuf,0x07);       // Size of slow_call
3523     // Push src onto stack slow-path
3524     emit_opcode(cbuf,0xD9 );      // FLD     ST(i)
3525     emit_d8    (cbuf,0xC0-1+$src$$reg );
3526     // CALL directly to the runtime
3527     cbuf.set_insts_mark();
3528     emit_opcode(cbuf,0xE8);       // Call into runtime
3529     emit_d32_reloc(cbuf, (StubRoutines::d2l_wrapper() - cbuf.insts_end()) - 4, runtime_call_Relocation::spec(), RELOC_IMM32 );
3530     // Carry on here...
3531   %}
3532 
3533   enc_class FMul_ST_reg( eRegFPR src1 ) %{
3534     // Operand was loaded from memory into fp ST (stack top)
3535     // FMUL   ST,$src  /* D8 C8+i */
3536     emit_opcode(cbuf, 0xD8);
3537     emit_opcode(cbuf, 0xC8 + $src1$$reg);
3538   %}
3539 
3540   enc_class FAdd_ST_reg( eRegFPR src2 ) %{
3541     // FADDP  ST,src2  /* D8 C0+i */
3542     emit_opcode(cbuf, 0xD8);
3543     emit_opcode(cbuf, 0xC0 + $src2$$reg);
3544     //could use FADDP  src2,fpST  /* DE C0+i */
3545   %}
3546 
3547   enc_class FAddP_reg_ST( eRegFPR src2 ) %{
3548     // FADDP  src2,ST  /* DE C0+i */
3549     emit_opcode(cbuf, 0xDE);
3550     emit_opcode(cbuf, 0xC0 + $src2$$reg);
3551   %}
3552 
3553   enc_class subFPR_divFPR_encode( eRegFPR src1, eRegFPR src2) %{
3554     // Operand has been loaded into fp ST (stack top)
3555       // FSUB   ST,$src1
3556       emit_opcode(cbuf, 0xD8);
3557       emit_opcode(cbuf, 0xE0 + $src1$$reg);
3558 
3559       // FDIV
3560       emit_opcode(cbuf, 0xD8);
3561       emit_opcode(cbuf, 0xF0 + $src2$$reg);
3562   %}
3563 
3564   enc_class MulFAddF (eRegFPR src1, eRegFPR src2) %{
3565     // Operand was loaded from memory into fp ST (stack top)
3566     // FADD   ST,$src  /* D8 C0+i */
3567     emit_opcode(cbuf, 0xD8);
3568     emit_opcode(cbuf, 0xC0 + $src1$$reg);
3569 
3570     // FMUL  ST,src2  /* D8 C*+i */
3571     emit_opcode(cbuf, 0xD8);
3572     emit_opcode(cbuf, 0xC8 + $src2$$reg);
3573   %}
3574 
3575 
3576   enc_class MulFAddFreverse (eRegFPR src1, eRegFPR src2) %{
3577     // Operand was loaded from memory into fp ST (stack top)
3578     // FADD   ST,$src  /* D8 C0+i */
3579     emit_opcode(cbuf, 0xD8);
3580     emit_opcode(cbuf, 0xC0 + $src1$$reg);
3581 
3582     // FMULP  src2,ST  /* DE C8+i */
3583     emit_opcode(cbuf, 0xDE);
3584     emit_opcode(cbuf, 0xC8 + $src2$$reg);
3585   %}
3586 
3587   // Atomically load the volatile long
3588   enc_class enc_loadL_volatile( memory mem, stackSlotL dst ) %{
3589     emit_opcode(cbuf,0xDF);
3590     int rm_byte_opcode = 0x05;
3591     int base     = $mem$$base;
3592     int index    = $mem$$index;
3593     int scale    = $mem$$scale;
3594     int displace = $mem$$disp;
3595     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when working with static globals
3596     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_is_oop);
3597     store_to_stackslot( cbuf, 0x0DF, 0x07, $dst$$disp );
3598   %}
3599 
3600   // Volatile Store Long.  Must be atomic, so move it into
3601   // the FP TOS and then do a 64-bit FIST.  Has to probe the
3602   // target address before the store (for null-ptr checks)
3603   // so the memory operand is used twice in the encoding.
3604   enc_class enc_storeL_volatile( memory mem, stackSlotL src ) %{
3605     store_to_stackslot( cbuf, 0x0DF, 0x05, $src$$disp );
3606     cbuf.set_insts_mark();            // Mark start of FIST in case $mem has an oop
3607     emit_opcode(cbuf,0xDF);
3608     int rm_byte_opcode = 0x07;
3609     int base     = $mem$$base;
3610     int index    = $mem$$index;
3611     int scale    = $mem$$scale;
3612     int displace = $mem$$disp;
3613     bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when working with static globals
3614     encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, disp_is_oop);
3615   %}
3616 
3617   // Safepoint Poll.  This polls the safepoint page, and causes an
3618   // exception if it is not readable. Unfortunately, it kills the condition code
3619   // in the process
3620   // We current use TESTL [spp],EDI
3621   // A better choice might be TESTB [spp + pagesize() - CacheLineSize()],0
3622 
3623   enc_class Safepoint_Poll() %{
3624     cbuf.relocate(cbuf.insts_mark(), relocInfo::poll_type, 0);
3625     emit_opcode(cbuf,0x85);
3626     emit_rm (cbuf, 0x0, 0x7, 0x5);
3627     emit_d32(cbuf, (intptr_t)os::get_polling_page());
3628   %}
3629 %}
3630 
3631 
3632 //----------FRAME--------------------------------------------------------------
3633 // Definition of frame structure and management information.
3634 //
3635 //  S T A C K   L A Y O U T    Allocators stack-slot number
3636 //                             |   (to get allocators register number
3637 //  G  Owned by    |        |  v    add OptoReg::stack0())
3638 //  r   CALLER     |        |
3639 //  o     |        +--------+      pad to even-align allocators stack-slot
3640 //  w     V        |  pad0  |        numbers; owned by CALLER
3641 //  t   -----------+--------+----> Matcher::_in_arg_limit, unaligned
3642 //  h     ^        |   in   |  5
3643 //        |        |  args  |  4   Holes in incoming args owned by SELF
3644 //  |     |        |        |  3
3645 //  |     |        +--------+
3646 //  V     |        | old out|      Empty on Intel, window on Sparc
3647 //        |    old |preserve|      Must be even aligned.
3648 //        |     SP-+--------+----> Matcher::_old_SP, even aligned
3649 //        |        |   in   |  3   area for Intel ret address
3650 //     Owned by    |preserve|      Empty on Sparc.
3651 //       SELF      +--------+
3652 //        |        |  pad2  |  2   pad to align old SP
3653 //        |        +--------+  1
3654 //        |        | locks  |  0
3655 //        |        +--------+----> OptoReg::stack0(), even aligned
3656 //        |        |  pad1  | 11   pad to align new SP
3657 //        |        +--------+
3658 //        |        |        | 10
3659 //        |        | spills |  9   spills
3660 //        V        |        |  8   (pad0 slot for callee)
3661 //      -----------+--------+----> Matcher::_out_arg_limit, unaligned
3662 //        ^        |  out   |  7
3663 //        |        |  args  |  6   Holes in outgoing args owned by CALLEE
3664 //     Owned by    +--------+
3665 //      CALLEE     | new out|  6   Empty on Intel, window on Sparc
3666 //        |    new |preserve|      Must be even-aligned.
3667 //        |     SP-+--------+----> Matcher::_new_SP, even aligned
3668 //        |        |        |
3669 //
3670 // Note 1: Only region 8-11 is determined by the allocator.  Region 0-5 is
3671 //         known from SELF's arguments and the Java calling convention.
3672 //         Region 6-7 is determined per call site.
3673 // Note 2: If the calling convention leaves holes in the incoming argument
3674 //         area, those holes are owned by SELF.  Holes in the outgoing area
3675 //         are owned by the CALLEE.  Holes should not be nessecary in the
3676 //         incoming area, as the Java calling convention is completely under
3677 //         the control of the AD file.  Doubles can be sorted and packed to
3678 //         avoid holes.  Holes in the outgoing arguments may be nessecary for
3679 //         varargs C calling conventions.
3680 // Note 3: Region 0-3 is even aligned, with pad2 as needed.  Region 3-5 is
3681 //         even aligned with pad0 as needed.
3682 //         Region 6 is even aligned.  Region 6-7 is NOT even aligned;
3683 //         region 6-11 is even aligned; it may be padded out more so that
3684 //         the region from SP to FP meets the minimum stack alignment.
3685 
3686 frame %{
3687   // What direction does stack grow in (assumed to be same for C & Java)
3688   stack_direction(TOWARDS_LOW);
3689 
3690   // These three registers define part of the calling convention
3691   // between compiled code and the interpreter.
3692   inline_cache_reg(EAX);                // Inline Cache Register
3693   interpreter_method_oop_reg(EBX);      // Method Oop Register when calling interpreter
3694 
3695   // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
3696   cisc_spilling_operand_name(indOffset32);
3697 
3698   // Number of stack slots consumed by locking an object
3699   sync_stack_slots(1);
3700 
3701   // Compiled code's Frame Pointer
3702   frame_pointer(ESP);
3703   // Interpreter stores its frame pointer in a register which is
3704   // stored to the stack by I2CAdaptors.
3705   // I2CAdaptors convert from interpreted java to compiled java.
3706   interpreter_frame_pointer(EBP);
3707 
3708   // Stack alignment requirement
3709   // Alignment size in bytes (128-bit -> 16 bytes)
3710   stack_alignment(StackAlignmentInBytes);
3711 
3712   // Number of stack slots between incoming argument block and the start of
3713   // a new frame.  The PROLOG must add this many slots to the stack.  The
3714   // EPILOG must remove this many slots.  Intel needs one slot for
3715   // return address and one for rbp, (must save rbp)
3716   in_preserve_stack_slots(2+VerifyStackAtCalls);
3717 
3718   // Number of outgoing stack slots killed above the out_preserve_stack_slots
3719   // for calls to C.  Supports the var-args backing area for register parms.
3720   varargs_C_out_slots_killed(0);
3721 
3722   // The after-PROLOG location of the return address.  Location of
3723   // return address specifies a type (REG or STACK) and a number
3724   // representing the register number (i.e. - use a register name) or
3725   // stack slot.
3726   // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
3727   // Otherwise, it is above the locks and verification slot and alignment word
3728   return_addr(STACK - 1 +
3729               round_to((Compile::current()->in_preserve_stack_slots() +
3730                         Compile::current()->fixed_slots()),
3731                        stack_alignment_in_slots()));
3732 
3733   // Body of function which returns an integer array locating
3734   // arguments either in registers or in stack slots.  Passed an array
3735   // of ideal registers called "sig" and a "length" count.  Stack-slot
3736   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3737   // arguments for a CALLEE.  Incoming stack arguments are
3738   // automatically biased by the preserve_stack_slots field above.
3739   calling_convention %{
3740     // No difference between ingoing/outgoing just pass false
3741     SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
3742   %}
3743 
3744 
3745   // Body of function which returns an integer array locating
3746   // arguments either in registers or in stack slots.  Passed an array
3747   // of ideal registers called "sig" and a "length" count.  Stack-slot
3748   // offsets are based on outgoing arguments, i.e. a CALLER setting up
3749   // arguments for a CALLEE.  Incoming stack arguments are
3750   // automatically biased by the preserve_stack_slots field above.
3751   c_calling_convention %{
3752     // This is obviously always outgoing
3753     (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
3754   %}
3755 
3756   // Location of C & interpreter return values
3757   c_return_value %{
3758     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3759     static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
3760     static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };
3761 
3762     // in SSE2+ mode we want to keep the FPU stack clean so pretend
3763     // that C functions return float and double results in XMM0.
3764     if( ideal_reg == Op_RegD && UseSSE>=2 )
3765       return OptoRegPair(XMM0b_num,XMM0a_num);
3766     if( ideal_reg == Op_RegF && UseSSE>=2 )
3767       return OptoRegPair(OptoReg::Bad,XMM0a_num);
3768 
3769     return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
3770   %}
3771 
3772   // Location of return values
3773   return_value %{
3774     assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3775     static int lo[Op_RegL+1] = { 0, 0, OptoReg::Bad, EAX_num,      EAX_num,      FPR1L_num,    FPR1L_num, EAX_num };
3776     static int hi[Op_RegL+1] = { 0, 0, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, FPR1H_num, EDX_num };
3777     if( ideal_reg == Op_RegD && UseSSE>=2 )
3778       return OptoRegPair(XMM0b_num,XMM0a_num);
3779     if( ideal_reg == Op_RegF && UseSSE>=1 )
3780       return OptoRegPair(OptoReg::Bad,XMM0a_num);
3781     return OptoRegPair(hi[ideal_reg],lo[ideal_reg]);
3782   %}
3783 
3784 %}
3785 
3786 //----------ATTRIBUTES---------------------------------------------------------
3787 //----------Operand Attributes-------------------------------------------------
3788 op_attrib op_cost(0);        // Required cost attribute
3789 
3790 //----------Instruction Attributes---------------------------------------------
3791 ins_attrib ins_cost(100);       // Required cost attribute
3792 ins_attrib ins_size(8);         // Required size attribute (in bits)
3793 ins_attrib ins_short_branch(0); // Required flag: is this instruction a
3794                                 // non-matching short branch variant of some
3795                                                             // long branch?
3796 ins_attrib ins_alignment(1);    // Required alignment attribute (must be a power of 2)
3797                                 // specifies the alignment that some part of the instruction (not
3798                                 // necessarily the start) requires.  If > 1, a compute_padding()
3799                                 // function must be provided for the instruction
3800 
3801 //----------OPERANDS-----------------------------------------------------------
3802 // Operand definitions must precede instruction definitions for correct parsing
3803 // in the ADLC because operands constitute user defined types which are used in
3804 // instruction definitions.
3805 
3806 //----------Simple Operands----------------------------------------------------
3807 // Immediate Operands
3808 // Integer Immediate
3809 operand immI() %{
3810   match(ConI);
3811 
3812   op_cost(10);
3813   format %{ %}
3814   interface(CONST_INTER);
3815 %}
3816 
3817 // Constant for test vs zero
3818 operand immI0() %{
3819   predicate(n->get_int() == 0);
3820   match(ConI);
3821 
3822   op_cost(0);
3823   format %{ %}
3824   interface(CONST_INTER);
3825 %}
3826 
3827 // Constant for increment
3828 operand immI1() %{
3829   predicate(n->get_int() == 1);
3830   match(ConI);
3831 
3832   op_cost(0);
3833   format %{ %}
3834   interface(CONST_INTER);
3835 %}
3836 
3837 // Constant for decrement
3838 operand immI_M1() %{
3839   predicate(n->get_int() == -1);
3840   match(ConI);
3841 
3842   op_cost(0);
3843   format %{ %}
3844   interface(CONST_INTER);
3845 %}
3846 
3847 // Valid scale values for addressing modes
3848 operand immI2() %{
3849   predicate(0 <= n->get_int() && (n->get_int() <= 3));
3850   match(ConI);
3851 
3852   format %{ %}
3853   interface(CONST_INTER);
3854 %}
3855 
3856 operand immI8() %{
3857   predicate((-128 <= n->get_int()) && (n->get_int() <= 127));
3858   match(ConI);
3859 
3860   op_cost(5);
3861   format %{ %}
3862   interface(CONST_INTER);
3863 %}
3864 
3865 operand immI16() %{
3866   predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
3867   match(ConI);
3868 
3869   op_cost(10);
3870   format %{ %}
3871   interface(CONST_INTER);
3872 %}
3873 
3874 // Constant for long shifts
3875 operand immI_32() %{
3876   predicate( n->get_int() == 32 );
3877   match(ConI);
3878 
3879   op_cost(0);
3880   format %{ %}
3881   interface(CONST_INTER);
3882 %}
3883 
3884 operand immI_1_31() %{
3885   predicate( n->get_int() >= 1 && n->get_int() <= 31 );
3886   match(ConI);
3887 
3888   op_cost(0);
3889   format %{ %}
3890   interface(CONST_INTER);
3891 %}
3892 
3893 operand immI_32_63() %{
3894   predicate( n->get_int() >= 32 && n->get_int() <= 63 );
3895   match(ConI);
3896   op_cost(0);
3897 
3898   format %{ %}
3899   interface(CONST_INTER);
3900 %}
3901 
3902 operand immI_1() %{
3903   predicate( n->get_int() == 1 );
3904   match(ConI);
3905 
3906   op_cost(0);
3907   format %{ %}
3908   interface(CONST_INTER);
3909 %}
3910 
3911 operand immI_2() %{
3912   predicate( n->get_int() == 2 );
3913   match(ConI);
3914 
3915   op_cost(0);
3916   format %{ %}
3917   interface(CONST_INTER);
3918 %}
3919 
3920 operand immI_3() %{
3921   predicate( n->get_int() == 3 );
3922   match(ConI);
3923 
3924   op_cost(0);
3925   format %{ %}
3926   interface(CONST_INTER);
3927 %}
3928 
3929 // Pointer Immediate
3930 operand immP() %{
3931   match(ConP);
3932 
3933   op_cost(10);
3934   format %{ %}
3935   interface(CONST_INTER);
3936 %}
3937 
3938 // NULL Pointer Immediate
3939 operand immP0() %{
3940   predicate( n->get_ptr() == 0 );
3941   match(ConP);
3942   op_cost(0);
3943 
3944   format %{ %}
3945   interface(CONST_INTER);
3946 %}
3947 
3948 // Long Immediate
3949 operand immL() %{
3950   match(ConL);
3951 
3952   op_cost(20);
3953   format %{ %}
3954   interface(CONST_INTER);
3955 %}
3956 
3957 // Long Immediate zero
3958 operand immL0() %{
3959   predicate( n->get_long() == 0L );
3960   match(ConL);
3961   op_cost(0);
3962 
3963   format %{ %}
3964   interface(CONST_INTER);
3965 %}
3966 
3967 // Long Immediate zero
3968 operand immL_M1() %{
3969   predicate( n->get_long() == -1L );
3970   match(ConL);
3971   op_cost(0);
3972 
3973   format %{ %}
3974   interface(CONST_INTER);
3975 %}
3976 
3977 // Long immediate from 0 to 127.
3978 // Used for a shorter form of long mul by 10.
3979 operand immL_127() %{
3980   predicate((0 <= n->get_long()) && (n->get_long() <= 127));
3981   match(ConL);
3982   op_cost(0);
3983 
3984   format %{ %}
3985   interface(CONST_INTER);
3986 %}
3987 
3988 // Long Immediate: low 32-bit mask
3989 operand immL_32bits() %{
3990   predicate(n->get_long() == 0xFFFFFFFFL);
3991   match(ConL);
3992   op_cost(0);
3993 
3994   format %{ %}
3995   interface(CONST_INTER);
3996 %}
3997 
3998 // Long Immediate: low 32-bit mask
3999 operand immL32() %{
4000   predicate(n->get_long() == (int)(n->get_long()));
4001   match(ConL);
4002   op_cost(20);
4003 
4004   format %{ %}
4005   interface(CONST_INTER);
4006 %}
4007 
4008 //Double Immediate zero
4009 operand immDPR0() %{
4010   // Do additional (and counter-intuitive) test against NaN to work around VC++
4011   // bug that generates code such that NaNs compare equal to 0.0
4012   predicate( UseSSE<=1 && n->getd() == 0.0 && !g_isnan(n->getd()) );
4013   match(ConD);
4014 
4015   op_cost(5);
4016   format %{ %}
4017   interface(CONST_INTER);
4018 %}
4019 
4020 // Double Immediate one
4021 operand immDPR1() %{
4022   predicate( UseSSE<=1 && n->getd() == 1.0 );
4023   match(ConD);
4024 
4025   op_cost(5);
4026   format %{ %}
4027   interface(CONST_INTER);
4028 %}
4029 
4030 // Double Immediate
4031 operand immDPR() %{
4032   predicate(UseSSE<=1);
4033   match(ConD);
4034 
4035   op_cost(5);
4036   format %{ %}
4037   interface(CONST_INTER);
4038 %}
4039 
4040 operand immD() %{
4041   predicate(UseSSE>=2);
4042   match(ConD);
4043 
4044   op_cost(5);
4045   format %{ %}
4046   interface(CONST_INTER);
4047 %}
4048 
4049 // Double Immediate zero
4050 operand immD0() %{
4051   // Do additional (and counter-intuitive) test against NaN to work around VC++
4052   // bug that generates code such that NaNs compare equal to 0.0 AND do not
4053   // compare equal to -0.0.
4054   predicate( UseSSE>=2 && jlong_cast(n->getd()) == 0 );
4055   match(ConD);
4056 
4057   format %{ %}
4058   interface(CONST_INTER);
4059 %}
4060 
4061 // Float Immediate zero
4062 operand immFPR0() %{
4063   predicate(UseSSE == 0 && n->getf() == 0.0F);
4064   match(ConF);
4065 
4066   op_cost(5);
4067   format %{ %}
4068   interface(CONST_INTER);
4069 %}
4070 
4071 // Float Immediate one
4072 operand immFPR1() %{
4073   predicate(UseSSE == 0 && n->getf() == 1.0F);
4074   match(ConF);
4075 
4076   op_cost(5);
4077   format %{ %}
4078   interface(CONST_INTER);
4079 %}
4080 
4081 // Float Immediate
4082 operand immFPR() %{
4083   predicate( UseSSE == 0 );
4084   match(ConF);
4085 
4086   op_cost(5);
4087   format %{ %}
4088   interface(CONST_INTER);
4089 %}
4090 
4091 // Float Immediate
4092 operand immF() %{
4093   predicate(UseSSE >= 1);
4094   match(ConF);
4095 
4096   op_cost(5);
4097   format %{ %}
4098   interface(CONST_INTER);
4099 %}
4100 
4101 // Float Immediate zero.  Zero and not -0.0
4102 operand immF0() %{
4103   predicate( UseSSE >= 1 && jint_cast(n->getf()) == 0 );
4104   match(ConF);
4105 
4106   op_cost(5);
4107   format %{ %}
4108   interface(CONST_INTER);
4109 %}
4110 
4111 // Immediates for special shifts (sign extend)
4112 
4113 // Constants for increment
4114 operand immI_16() %{
4115   predicate( n->get_int() == 16 );
4116   match(ConI);
4117 
4118   format %{ %}
4119   interface(CONST_INTER);
4120 %}
4121 
4122 operand immI_24() %{
4123   predicate( n->get_int() == 24 );
4124   match(ConI);
4125 
4126   format %{ %}
4127   interface(CONST_INTER);
4128 %}
4129 
4130 // Constant for byte-wide masking
4131 operand immI_255() %{
4132   predicate( n->get_int() == 255 );
4133   match(ConI);
4134 
4135   format %{ %}
4136   interface(CONST_INTER);
4137 %}
4138 
4139 // Constant for short-wide masking
4140 operand immI_65535() %{
4141   predicate(n->get_int() == 65535);
4142   match(ConI);
4143 
4144   format %{ %}
4145   interface(CONST_INTER);
4146 %}
4147 
4148 // Register Operands
4149 // Integer Register
4150 operand eRegI() %{
4151   constraint(ALLOC_IN_RC(e_reg));
4152   match(RegI);
4153   match(xRegI);
4154   match(eAXRegI);
4155   match(eBXRegI);
4156   match(eCXRegI);
4157   match(eDXRegI);
4158   match(eDIRegI);
4159   match(eSIRegI);
4160 
4161   format %{ %}
4162   interface(REG_INTER);
4163 %}
4164 
4165 // Subset of Integer Register
4166 operand xRegI(eRegI reg) %{
4167   constraint(ALLOC_IN_RC(x_reg));
4168   match(reg);
4169   match(eAXRegI);
4170   match(eBXRegI);
4171   match(eCXRegI);
4172   match(eDXRegI);
4173 
4174   format %{ %}
4175   interface(REG_INTER);
4176 %}
4177 
4178 // Special Registers
4179 operand eAXRegI(xRegI reg) %{
4180   constraint(ALLOC_IN_RC(eax_reg));
4181   match(reg);
4182   match(eRegI);
4183 
4184   format %{ "EAX" %}
4185   interface(REG_INTER);
4186 %}
4187 
4188 // Special Registers
4189 operand eBXRegI(xRegI reg) %{
4190   constraint(ALLOC_IN_RC(ebx_reg));
4191   match(reg);
4192   match(eRegI);
4193 
4194   format %{ "EBX" %}
4195   interface(REG_INTER);
4196 %}
4197 
4198 operand eCXRegI(xRegI reg) %{
4199   constraint(ALLOC_IN_RC(ecx_reg));
4200   match(reg);
4201   match(eRegI);
4202 
4203   format %{ "ECX" %}
4204   interface(REG_INTER);
4205 %}
4206 
4207 operand eDXRegI(xRegI reg) %{
4208   constraint(ALLOC_IN_RC(edx_reg));
4209   match(reg);
4210   match(eRegI);
4211 
4212   format %{ "EDX" %}
4213   interface(REG_INTER);
4214 %}
4215 
4216 operand eDIRegI(xRegI reg) %{
4217   constraint(ALLOC_IN_RC(edi_reg));
4218   match(reg);
4219   match(eRegI);
4220 
4221   format %{ "EDI" %}
4222   interface(REG_INTER);
4223 %}
4224 
4225 operand naxRegI() %{
4226   constraint(ALLOC_IN_RC(nax_reg));
4227   match(RegI);
4228   match(eCXRegI);
4229   match(eDXRegI);
4230   match(eSIRegI);
4231   match(eDIRegI);
4232 
4233   format %{ %}
4234   interface(REG_INTER);
4235 %}
4236 
4237 operand nadxRegI() %{
4238   constraint(ALLOC_IN_RC(nadx_reg));
4239   match(RegI);
4240   match(eBXRegI);
4241   match(eCXRegI);
4242   match(eSIRegI);
4243   match(eDIRegI);
4244 
4245   format %{ %}
4246   interface(REG_INTER);
4247 %}
4248 
4249 operand ncxRegI() %{
4250   constraint(ALLOC_IN_RC(ncx_reg));
4251   match(RegI);
4252   match(eAXRegI);
4253   match(eDXRegI);
4254   match(eSIRegI);
4255   match(eDIRegI);
4256 
4257   format %{ %}
4258   interface(REG_INTER);
4259 %}
4260 
4261 // // This operand was used by cmpFastUnlock, but conflicted with 'object' reg
4262 // //
4263 operand eSIRegI(xRegI reg) %{
4264    constraint(ALLOC_IN_RC(esi_reg));
4265    match(reg);
4266    match(eRegI);
4267 
4268    format %{ "ESI" %}
4269    interface(REG_INTER);
4270 %}
4271 
4272 // Pointer Register
4273 operand anyRegP() %{
4274   constraint(ALLOC_IN_RC(any_reg));
4275   match(RegP);
4276   match(eAXRegP);
4277   match(eBXRegP);
4278   match(eCXRegP);
4279   match(eDIRegP);
4280   match(eRegP);
4281 
4282   format %{ %}
4283   interface(REG_INTER);
4284 %}
4285 
4286 operand eRegP() %{
4287   constraint(ALLOC_IN_RC(e_reg));
4288   match(RegP);
4289   match(eAXRegP);
4290   match(eBXRegP);
4291   match(eCXRegP);
4292   match(eDIRegP);
4293 
4294   format %{ %}
4295   interface(REG_INTER);
4296 %}
4297 
4298 // On windows95, EBP is not safe to use for implicit null tests.
4299 operand eRegP_no_EBP() %{
4300   constraint(ALLOC_IN_RC(e_reg_no_rbp));
4301   match(RegP);
4302   match(eAXRegP);
4303   match(eBXRegP);
4304   match(eCXRegP);
4305   match(eDIRegP);
4306 
4307   op_cost(100);
4308   format %{ %}
4309   interface(REG_INTER);
4310 %}
4311 
4312 operand naxRegP() %{
4313   constraint(ALLOC_IN_RC(nax_reg));
4314   match(RegP);
4315   match(eBXRegP);
4316   match(eDXRegP);
4317   match(eCXRegP);
4318   match(eSIRegP);
4319   match(eDIRegP);
4320 
4321   format %{ %}
4322   interface(REG_INTER);
4323 %}
4324 
4325 operand nabxRegP() %{
4326   constraint(ALLOC_IN_RC(nabx_reg));
4327   match(RegP);
4328   match(eCXRegP);
4329   match(eDXRegP);
4330   match(eSIRegP);
4331   match(eDIRegP);
4332 
4333   format %{ %}
4334   interface(REG_INTER);
4335 %}
4336 
4337 operand pRegP() %{
4338   constraint(ALLOC_IN_RC(p_reg));
4339   match(RegP);
4340   match(eBXRegP);
4341   match(eDXRegP);
4342   match(eSIRegP);
4343   match(eDIRegP);
4344 
4345   format %{ %}
4346   interface(REG_INTER);
4347 %}
4348 
4349 // Special Registers
4350 // Return a pointer value
4351 operand eAXRegP(eRegP reg) %{
4352   constraint(ALLOC_IN_RC(eax_reg));
4353   match(reg);
4354   format %{ "EAX" %}
4355   interface(REG_INTER);
4356 %}
4357 
4358 // Used in AtomicAdd
4359 operand eBXRegP(eRegP reg) %{
4360   constraint(ALLOC_IN_RC(ebx_reg));
4361   match(reg);
4362   format %{ "EBX" %}
4363   interface(REG_INTER);
4364 %}
4365 
4366 // Tail-call (interprocedural jump) to interpreter
4367 operand eCXRegP(eRegP reg) %{
4368   constraint(ALLOC_IN_RC(ecx_reg));
4369   match(reg);
4370   format %{ "ECX" %}
4371   interface(REG_INTER);
4372 %}
4373 
4374 operand eSIRegP(eRegP reg) %{
4375   constraint(ALLOC_IN_RC(esi_reg));
4376   match(reg);
4377   format %{ "ESI" %}
4378   interface(REG_INTER);
4379 %}
4380 
4381 // Used in rep stosw
4382 operand eDIRegP(eRegP reg) %{
4383   constraint(ALLOC_IN_RC(edi_reg));
4384   match(reg);
4385   format %{ "EDI" %}
4386   interface(REG_INTER);
4387 %}
4388 
4389 operand eBPRegP() %{
4390   constraint(ALLOC_IN_RC(ebp_reg));
4391   match(RegP);
4392   format %{ "EBP" %}
4393   interface(REG_INTER);
4394 %}
4395 
4396 operand eRegL() %{
4397   constraint(ALLOC_IN_RC(long_reg));
4398   match(RegL);
4399   match(eADXRegL);
4400 
4401   format %{ %}
4402   interface(REG_INTER);
4403 %}
4404 
4405 operand eADXRegL( eRegL reg ) %{
4406   constraint(ALLOC_IN_RC(eadx_reg));
4407   match(reg);
4408 
4409   format %{ "EDX:EAX" %}
4410   interface(REG_INTER);
4411 %}
4412 
4413 operand eBCXRegL( eRegL reg ) %{
4414   constraint(ALLOC_IN_RC(ebcx_reg));
4415   match(reg);
4416 
4417   format %{ "EBX:ECX" %}
4418   interface(REG_INTER);
4419 %}
4420 
4421 // Special case for integer high multiply
4422 operand eADXRegL_low_only() %{
4423   constraint(ALLOC_IN_RC(eadx_reg));
4424   match(RegL);
4425 
4426   format %{ "EAX" %}
4427   interface(REG_INTER);
4428 %}
4429 
4430 // Flags register, used as output of compare instructions
4431 operand eFlagsReg() %{
4432   constraint(ALLOC_IN_RC(int_flags));
4433   match(RegFlags);
4434 
4435   format %{ "EFLAGS" %}
4436   interface(REG_INTER);
4437 %}
4438 
4439 // Flags register, used as output of FLOATING POINT compare instructions
4440 operand eFlagsRegU() %{
4441   constraint(ALLOC_IN_RC(int_flags));
4442   match(RegFlags);
4443 
4444   format %{ "EFLAGS_U" %}
4445   interface(REG_INTER);
4446 %}
4447 
4448 operand eFlagsRegUCF() %{
4449   constraint(ALLOC_IN_RC(int_flags));
4450   match(RegFlags);
4451   predicate(false);
4452 
4453   format %{ "EFLAGS_U_CF" %}
4454   interface(REG_INTER);
4455 %}
4456 
4457 // Condition Code Register used by long compare
4458 operand flagsReg_long_LTGE() %{
4459   constraint(ALLOC_IN_RC(int_flags));
4460   match(RegFlags);
4461   format %{ "FLAGS_LTGE" %}
4462   interface(REG_INTER);
4463 %}
4464 operand flagsReg_long_EQNE() %{
4465   constraint(ALLOC_IN_RC(int_flags));
4466   match(RegFlags);
4467   format %{ "FLAGS_EQNE" %}
4468   interface(REG_INTER);
4469 %}
4470 operand flagsReg_long_LEGT() %{
4471   constraint(ALLOC_IN_RC(int_flags));
4472   match(RegFlags);
4473   format %{ "FLAGS_LEGT" %}
4474   interface(REG_INTER);
4475 %}
4476 
4477 // Float register operands
4478 operand regDPR() %{
4479   predicate( UseSSE < 2 );
4480   constraint(ALLOC_IN_RC(dbl_reg));
4481   match(RegD);
4482   match(regDPR1);
4483   match(regDPR2);
4484   format %{ %}
4485   interface(REG_INTER);
4486 %}
4487 
4488 operand regDPR1(regDPR reg) %{
4489   predicate( UseSSE < 2 );
4490   constraint(ALLOC_IN_RC(dbl_reg0));
4491   match(reg);
4492   format %{ "FPR1" %}
4493   interface(REG_INTER);
4494 %}
4495 
4496 operand regDPR2(regDPR reg) %{
4497   predicate( UseSSE < 2 );
4498   constraint(ALLOC_IN_RC(dbl_reg1));
4499   match(reg);
4500   format %{ "FPR2" %}
4501   interface(REG_INTER);
4502 %}
4503 
4504 operand regnotDPR1(regDPR reg) %{
4505   predicate( UseSSE < 2 );
4506   constraint(ALLOC_IN_RC(dbl_notreg0));
4507   match(reg);
4508   format %{ %}
4509   interface(REG_INTER);
4510 %}
4511 
4512 // XMM Double register operands
4513 operand regD() %{
4514   predicate( UseSSE>=2 );
4515   constraint(ALLOC_IN_RC(xdb_reg));
4516   match(RegD);
4517   match(regD6);
4518   match(regD7);
4519   format %{ %}
4520   interface(REG_INTER);
4521 %}
4522 
4523 // XMM6 double register operands
4524 operand regD6(regD reg) %{
4525   predicate( UseSSE>=2 );
4526   constraint(ALLOC_IN_RC(xdb_reg6));
4527   match(reg);
4528   format %{ "XMM6" %}
4529   interface(REG_INTER);
4530 %}
4531 
4532 // XMM7 double register operands
4533 operand regD7(regD reg) %{
4534   predicate( UseSSE>=2 );
4535   constraint(ALLOC_IN_RC(xdb_reg7));
4536   match(reg);
4537   format %{ "XMM7" %}
4538   interface(REG_INTER);
4539 %}
4540 
4541 // Float register operands
4542 operand regFPR() %{
4543   predicate( UseSSE < 2 );
4544   constraint(ALLOC_IN_RC(flt_reg));
4545   match(RegF);
4546   match(regFPR1);
4547   format %{ %}
4548   interface(REG_INTER);
4549 %}
4550 
4551 // Float register operands
4552 operand regFPR1(regFPR reg) %{
4553   predicate( UseSSE < 2 );
4554   constraint(ALLOC_IN_RC(flt_reg0));
4555   match(reg);
4556   format %{ "FPR1" %}
4557   interface(REG_INTER);
4558 %}
4559 
4560 // XMM register operands
4561 operand regF() %{
4562   predicate( UseSSE>=1 );
4563   constraint(ALLOC_IN_RC(xmm_reg));
4564   match(RegF);
4565   format %{ %}
4566   interface(REG_INTER);
4567 %}
4568 
4569 
4570 //----------Memory Operands----------------------------------------------------
4571 // Direct Memory Operand
4572 operand direct(immP addr) %{
4573   match(addr);
4574 
4575   format %{ "[$addr]" %}
4576   interface(MEMORY_INTER) %{
4577     base(0xFFFFFFFF);
4578     index(0x4);
4579     scale(0x0);
4580     disp($addr);
4581   %}
4582 %}
4583 
4584 // Indirect Memory Operand
4585 operand indirect(eRegP reg) %{
4586   constraint(ALLOC_IN_RC(e_reg));
4587   match(reg);
4588 
4589   format %{ "[$reg]" %}
4590   interface(MEMORY_INTER) %{
4591     base($reg);
4592     index(0x4);
4593     scale(0x0);
4594     disp(0x0);
4595   %}
4596 %}
4597 
4598 // Indirect Memory Plus Short Offset Operand
4599 operand indOffset8(eRegP reg, immI8 off) %{
4600   match(AddP reg off);
4601 
4602   format %{ "[$reg + $off]" %}
4603   interface(MEMORY_INTER) %{
4604     base($reg);
4605     index(0x4);
4606     scale(0x0);
4607     disp($off);
4608   %}
4609 %}
4610 
4611 // Indirect Memory Plus Long Offset Operand
4612 operand indOffset32(eRegP reg, immI off) %{
4613   match(AddP reg off);
4614 
4615   format %{ "[$reg + $off]" %}
4616   interface(MEMORY_INTER) %{
4617     base($reg);
4618     index(0x4);
4619     scale(0x0);
4620     disp($off);
4621   %}
4622 %}
4623 
4624 // Indirect Memory Plus Long Offset Operand
4625 operand indOffset32X(eRegI reg, immP off) %{
4626   match(AddP off reg);
4627 
4628   format %{ "[$reg + $off]" %}
4629   interface(MEMORY_INTER) %{
4630     base($reg);
4631     index(0x4);
4632     scale(0x0);
4633     disp($off);
4634   %}
4635 %}
4636 
4637 // Indirect Memory Plus Index Register Plus Offset Operand
4638 operand indIndexOffset(eRegP reg, eRegI ireg, immI off) %{
4639   match(AddP (AddP reg ireg) off);
4640 
4641   op_cost(10);
4642   format %{"[$reg + $off + $ireg]" %}
4643   interface(MEMORY_INTER) %{
4644     base($reg);
4645     index($ireg);
4646     scale(0x0);
4647     disp($off);
4648   %}
4649 %}
4650 
4651 // Indirect Memory Plus Index Register Plus Offset Operand
4652 operand indIndex(eRegP reg, eRegI ireg) %{
4653   match(AddP reg ireg);
4654 
4655   op_cost(10);
4656   format %{"[$reg + $ireg]" %}
4657   interface(MEMORY_INTER) %{
4658     base($reg);
4659     index($ireg);
4660     scale(0x0);
4661     disp(0x0);
4662   %}
4663 %}
4664 
4665 // // -------------------------------------------------------------------------
4666 // // 486 architecture doesn't support "scale * index + offset" with out a base
4667 // // -------------------------------------------------------------------------
4668 // // Scaled Memory Operands
4669 // // Indirect Memory Times Scale Plus Offset Operand
4670 // operand indScaleOffset(immP off, eRegI ireg, immI2 scale) %{
4671 //   match(AddP off (LShiftI ireg scale));
4672 //
4673 //   op_cost(10);
4674 //   format %{"[$off + $ireg << $scale]" %}
4675 //   interface(MEMORY_INTER) %{
4676 //     base(0x4);
4677 //     index($ireg);
4678 //     scale($scale);
4679 //     disp($off);
4680 //   %}
4681 // %}
4682 
4683 // Indirect Memory Times Scale Plus Index Register
4684 operand indIndexScale(eRegP reg, eRegI ireg, immI2 scale) %{
4685   match(AddP reg (LShiftI ireg scale));
4686 
4687   op_cost(10);
4688   format %{"[$reg + $ireg << $scale]" %}
4689   interface(MEMORY_INTER) %{
4690     base($reg);
4691     index($ireg);
4692     scale($scale);
4693     disp(0x0);
4694   %}
4695 %}
4696 
4697 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4698 operand indIndexScaleOffset(eRegP reg, immI off, eRegI ireg, immI2 scale) %{
4699   match(AddP (AddP reg (LShiftI ireg scale)) off);
4700 
4701   op_cost(10);
4702   format %{"[$reg + $off + $ireg << $scale]" %}
4703   interface(MEMORY_INTER) %{
4704     base($reg);
4705     index($ireg);
4706     scale($scale);
4707     disp($off);
4708   %}
4709 %}
4710 
4711 //----------Load Long Memory Operands------------------------------------------
4712 // The load-long idiom will use it's address expression again after loading
4713 // the first word of the long.  If the load-long destination overlaps with
4714 // registers used in the addressing expression, the 2nd half will be loaded
4715 // from a clobbered address.  Fix this by requiring that load-long use
4716 // address registers that do not overlap with the load-long target.
4717 
4718 // load-long support
4719 operand load_long_RegP() %{
4720   constraint(ALLOC_IN_RC(esi_reg));
4721   match(RegP);
4722   match(eSIRegP);
4723   op_cost(100);
4724   format %{  %}
4725   interface(REG_INTER);
4726 %}
4727 
4728 // Indirect Memory Operand Long
4729 operand load_long_indirect(load_long_RegP reg) %{
4730   constraint(ALLOC_IN_RC(esi_reg));
4731   match(reg);
4732 
4733   format %{ "[$reg]" %}
4734   interface(MEMORY_INTER) %{
4735     base($reg);
4736     index(0x4);
4737     scale(0x0);
4738     disp(0x0);
4739   %}
4740 %}
4741 
4742 // Indirect Memory Plus Long Offset Operand
4743 operand load_long_indOffset32(load_long_RegP reg, immI off) %{
4744   match(AddP reg off);
4745 
4746   format %{ "[$reg + $off]" %}
4747   interface(MEMORY_INTER) %{
4748     base($reg);
4749     index(0x4);
4750     scale(0x0);
4751     disp($off);
4752   %}
4753 %}
4754 
4755 opclass load_long_memory(load_long_indirect, load_long_indOffset32);
4756 
4757 
4758 //----------Special Memory Operands--------------------------------------------
4759 // Stack Slot Operand - This operand is used for loading and storing temporary
4760 //                      values on the stack where a match requires a value to
4761 //                      flow through memory.
4762 operand stackSlotP(sRegP reg) %{
4763   constraint(ALLOC_IN_RC(stack_slots));
4764   // No match rule because this operand is only generated in matching
4765   format %{ "[$reg]" %}
4766   interface(MEMORY_INTER) %{
4767     base(0x4);   // ESP
4768     index(0x4);  // No Index
4769     scale(0x0);  // No Scale
4770     disp($reg);  // Stack Offset
4771   %}
4772 %}
4773 
4774 operand stackSlotI(sRegI reg) %{
4775   constraint(ALLOC_IN_RC(stack_slots));
4776   // No match rule because this operand is only generated in matching
4777   format %{ "[$reg]" %}
4778   interface(MEMORY_INTER) %{
4779     base(0x4);   // ESP
4780     index(0x4);  // No Index
4781     scale(0x0);  // No Scale
4782     disp($reg);  // Stack Offset
4783   %}
4784 %}
4785 
4786 operand stackSlotF(sRegF reg) %{
4787   constraint(ALLOC_IN_RC(stack_slots));
4788   // No match rule because this operand is only generated in matching
4789   format %{ "[$reg]" %}
4790   interface(MEMORY_INTER) %{
4791     base(0x4);   // ESP
4792     index(0x4);  // No Index
4793     scale(0x0);  // No Scale
4794     disp($reg);  // Stack Offset
4795   %}
4796 %}
4797 
4798 operand stackSlotD(sRegD reg) %{
4799   constraint(ALLOC_IN_RC(stack_slots));
4800   // No match rule because this operand is only generated in matching
4801   format %{ "[$reg]" %}
4802   interface(MEMORY_INTER) %{
4803     base(0x4);   // ESP
4804     index(0x4);  // No Index
4805     scale(0x0);  // No Scale
4806     disp($reg);  // Stack Offset
4807   %}
4808 %}
4809 
4810 operand stackSlotL(sRegL reg) %{
4811   constraint(ALLOC_IN_RC(stack_slots));
4812   // No match rule because this operand is only generated in matching
4813   format %{ "[$reg]" %}
4814   interface(MEMORY_INTER) %{
4815     base(0x4);   // ESP
4816     index(0x4);  // No Index
4817     scale(0x0);  // No Scale
4818     disp($reg);  // Stack Offset
4819   %}
4820 %}
4821 
4822 //----------Memory Operands - Win95 Implicit Null Variants----------------
4823 // Indirect Memory Operand
4824 operand indirect_win95_safe(eRegP_no_EBP reg)
4825 %{
4826   constraint(ALLOC_IN_RC(e_reg));
4827   match(reg);
4828 
4829   op_cost(100);
4830   format %{ "[$reg]" %}
4831   interface(MEMORY_INTER) %{
4832     base($reg);
4833     index(0x4);
4834     scale(0x0);
4835     disp(0x0);
4836   %}
4837 %}
4838 
4839 // Indirect Memory Plus Short Offset Operand
4840 operand indOffset8_win95_safe(eRegP_no_EBP reg, immI8 off)
4841 %{
4842   match(AddP reg off);
4843 
4844   op_cost(100);
4845   format %{ "[$reg + $off]" %}
4846   interface(MEMORY_INTER) %{
4847     base($reg);
4848     index(0x4);
4849     scale(0x0);
4850     disp($off);
4851   %}
4852 %}
4853 
4854 // Indirect Memory Plus Long Offset Operand
4855 operand indOffset32_win95_safe(eRegP_no_EBP reg, immI off)
4856 %{
4857   match(AddP reg off);
4858 
4859   op_cost(100);
4860   format %{ "[$reg + $off]" %}
4861   interface(MEMORY_INTER) %{
4862     base($reg);
4863     index(0x4);
4864     scale(0x0);
4865     disp($off);
4866   %}
4867 %}
4868 
4869 // Indirect Memory Plus Index Register Plus Offset Operand
4870 operand indIndexOffset_win95_safe(eRegP_no_EBP reg, eRegI ireg, immI off)
4871 %{
4872   match(AddP (AddP reg ireg) off);
4873 
4874   op_cost(100);
4875   format %{"[$reg + $off + $ireg]" %}
4876   interface(MEMORY_INTER) %{
4877     base($reg);
4878     index($ireg);
4879     scale(0x0);
4880     disp($off);
4881   %}
4882 %}
4883 
4884 // Indirect Memory Times Scale Plus Index Register
4885 operand indIndexScale_win95_safe(eRegP_no_EBP reg, eRegI ireg, immI2 scale)
4886 %{
4887   match(AddP reg (LShiftI ireg scale));
4888 
4889   op_cost(100);
4890   format %{"[$reg + $ireg << $scale]" %}
4891   interface(MEMORY_INTER) %{
4892     base($reg);
4893     index($ireg);
4894     scale($scale);
4895     disp(0x0);
4896   %}
4897 %}
4898 
4899 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4900 operand indIndexScaleOffset_win95_safe(eRegP_no_EBP reg, immI off, eRegI ireg, immI2 scale)
4901 %{
4902   match(AddP (AddP reg (LShiftI ireg scale)) off);
4903 
4904   op_cost(100);
4905   format %{"[$reg + $off + $ireg << $scale]" %}
4906   interface(MEMORY_INTER) %{
4907     base($reg);
4908     index($ireg);
4909     scale($scale);
4910     disp($off);
4911   %}
4912 %}
4913 
4914 //----------Conditional Branch Operands----------------------------------------
4915 // Comparison Op  - This is the operation of the comparison, and is limited to
4916 //                  the following set of codes:
4917 //                  L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
4918 //
4919 // Other attributes of the comparison, such as unsignedness, are specified
4920 // by the comparison instruction that sets a condition code flags register.
4921 // That result is represented by a flags operand whose subtype is appropriate
4922 // to the unsignedness (etc.) of the comparison.
4923 //
4924 // Later, the instruction which matches both the Comparison Op (a Bool) and
4925 // the flags (produced by the Cmp) specifies the coding of the comparison op
4926 // by matching a specific subtype of Bool operand below, such as cmpOpU.
4927 
4928 // Comparision Code
4929 operand cmpOp() %{
4930   match(Bool);
4931 
4932   format %{ "" %}
4933   interface(COND_INTER) %{
4934     equal(0x4, "e");
4935     not_equal(0x5, "ne");
4936     less(0xC, "l");
4937     greater_equal(0xD, "ge");
4938     less_equal(0xE, "le");
4939     greater(0xF, "g");
4940   %}
4941 %}
4942 
4943 // Comparison Code, unsigned compare.  Used by FP also, with
4944 // C2 (unordered) turned into GT or LT already.  The other bits
4945 // C0 and C3 are turned into Carry & Zero flags.
4946 operand cmpOpU() %{
4947   match(Bool);
4948 
4949   format %{ "" %}
4950   interface(COND_INTER) %{
4951     equal(0x4, "e");
4952     not_equal(0x5, "ne");
4953     less(0x2, "b");
4954     greater_equal(0x3, "nb");
4955     less_equal(0x6, "be");
4956     greater(0x7, "nbe");
4957   %}
4958 %}
4959 
4960 // Floating comparisons that don't require any fixup for the unordered case
4961 operand cmpOpUCF() %{
4962   match(Bool);
4963   predicate(n->as_Bool()->_test._test == BoolTest::lt ||
4964             n->as_Bool()->_test._test == BoolTest::ge ||
4965             n->as_Bool()->_test._test == BoolTest::le ||
4966             n->as_Bool()->_test._test == BoolTest::gt);
4967   format %{ "" %}
4968   interface(COND_INTER) %{
4969     equal(0x4, "e");
4970     not_equal(0x5, "ne");
4971     less(0x2, "b");
4972     greater_equal(0x3, "nb");
4973     less_equal(0x6, "be");
4974     greater(0x7, "nbe");
4975   %}
4976 %}
4977 
4978 
4979 // Floating comparisons that can be fixed up with extra conditional jumps
4980 operand cmpOpUCF2() %{
4981   match(Bool);
4982   predicate(n->as_Bool()->_test._test == BoolTest::ne ||
4983             n->as_Bool()->_test._test == BoolTest::eq);
4984   format %{ "" %}
4985   interface(COND_INTER) %{
4986     equal(0x4, "e");
4987     not_equal(0x5, "ne");
4988     less(0x2, "b");
4989     greater_equal(0x3, "nb");
4990     less_equal(0x6, "be");
4991     greater(0x7, "nbe");
4992   %}
4993 %}
4994 
4995 // Comparison Code for FP conditional move
4996 operand cmpOp_fcmov() %{
4997   match(Bool);
4998 
4999   format %{ "" %}
5000   interface(COND_INTER) %{
5001     equal        (0x0C8);
5002     not_equal    (0x1C8);
5003     less         (0x0C0);
5004     greater_equal(0x1C0);
5005     less_equal   (0x0D0);
5006     greater      (0x1D0);
5007   %}
5008 %}
5009 
5010 // Comparision Code used in long compares
5011 operand cmpOp_commute() %{
5012   match(Bool);
5013 
5014   format %{ "" %}
5015   interface(COND_INTER) %{
5016     equal(0x4, "e");
5017     not_equal(0x5, "ne");
5018     less(0xF, "g");
5019     greater_equal(0xE, "le");
5020     less_equal(0xD, "ge");
5021     greater(0xC, "l");
5022   %}
5023 %}
5024 
5025 //----------OPERAND CLASSES----------------------------------------------------
5026 // Operand Classes are groups of operands that are used as to simplify
5027 // instruction definitions by not requiring the AD writer to specify separate
5028 // instructions for every form of operand when the instruction accepts
5029 // multiple operand types with the same basic encoding and format.  The classic
5030 // case of this is memory operands.
5031 
5032 opclass memory(direct, indirect, indOffset8, indOffset32, indOffset32X, indIndexOffset,
5033                indIndex, indIndexScale, indIndexScaleOffset);
5034 
5035 // Long memory operations are encoded in 2 instructions and a +4 offset.
5036 // This means some kind of offset is always required and you cannot use
5037 // an oop as the offset (done when working on static globals).
5038 opclass long_memory(direct, indirect, indOffset8, indOffset32, indIndexOffset,
5039                     indIndex, indIndexScale, indIndexScaleOffset);
5040 
5041 
5042 //----------PIPELINE-----------------------------------------------------------
5043 // Rules which define the behavior of the target architectures pipeline.
5044 pipeline %{
5045 
5046 //----------ATTRIBUTES---------------------------------------------------------
5047 attributes %{
5048   variable_size_instructions;        // Fixed size instructions
5049   max_instructions_per_bundle = 3;   // Up to 3 instructions per bundle
5050   instruction_unit_size = 1;         // An instruction is 1 bytes long
5051   instruction_fetch_unit_size = 16;  // The processor fetches one line
5052   instruction_fetch_units = 1;       // of 16 bytes
5053 
5054   // List of nop instructions
5055   nops( MachNop );
5056 %}
5057 
5058 //----------RESOURCES----------------------------------------------------------
5059 // Resources are the functional units available to the machine
5060 
5061 // Generic P2/P3 pipeline
5062 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5063 // 3 instructions decoded per cycle.
5064 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5065 // 2 ALU op, only ALU0 handles mul/div instructions.
5066 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5067            MS0, MS1, MEM = MS0 | MS1,
5068            BR, FPU,
5069            ALU0, ALU1, ALU = ALU0 | ALU1 );
5070 
5071 //----------PIPELINE DESCRIPTION-----------------------------------------------
5072 // Pipeline Description specifies the stages in the machine's pipeline
5073 
5074 // Generic P2/P3 pipeline
5075 pipe_desc(S0, S1, S2, S3, S4, S5);
5076 
5077 //----------PIPELINE CLASSES---------------------------------------------------
5078 // Pipeline Classes describe the stages in which input and output are
5079 // referenced by the hardware pipeline.
5080 
5081 // Naming convention: ialu or fpu
5082 // Then: _reg
5083 // Then: _reg if there is a 2nd register
5084 // Then: _long if it's a pair of instructions implementing a long
5085 // Then: _fat if it requires the big decoder
5086 //   Or: _mem if it requires the big decoder and a memory unit.
5087 
5088 // Integer ALU reg operation
5089 pipe_class ialu_reg(eRegI dst) %{
5090     single_instruction;
5091     dst    : S4(write);
5092     dst    : S3(read);
5093     DECODE : S0;        // any decoder
5094     ALU    : S3;        // any alu
5095 %}
5096 
5097 // Long ALU reg operation
5098 pipe_class ialu_reg_long(eRegL dst) %{
5099     instruction_count(2);
5100     dst    : S4(write);
5101     dst    : S3(read);
5102     DECODE : S0(2);     // any 2 decoders
5103     ALU    : S3(2);     // both alus
5104 %}
5105 
5106 // Integer ALU reg operation using big decoder
5107 pipe_class ialu_reg_fat(eRegI dst) %{
5108     single_instruction;
5109     dst    : S4(write);
5110     dst    : S3(read);
5111     D0     : S0;        // big decoder only
5112     ALU    : S3;        // any alu
5113 %}
5114 
5115 // Long ALU reg operation using big decoder
5116 pipe_class ialu_reg_long_fat(eRegL dst) %{
5117     instruction_count(2);
5118     dst    : S4(write);
5119     dst    : S3(read);
5120     D0     : S0(2);     // big decoder only; twice
5121     ALU    : S3(2);     // any 2 alus
5122 %}
5123 
5124 // Integer ALU reg-reg operation
5125 pipe_class ialu_reg_reg(eRegI dst, eRegI src) %{
5126     single_instruction;
5127     dst    : S4(write);
5128     src    : S3(read);
5129     DECODE : S0;        // any decoder
5130     ALU    : S3;        // any alu
5131 %}
5132 
5133 // Long ALU reg-reg operation
5134 pipe_class ialu_reg_reg_long(eRegL dst, eRegL src) %{
5135     instruction_count(2);
5136     dst    : S4(write);
5137     src    : S3(read);
5138     DECODE : S0(2);     // any 2 decoders
5139     ALU    : S3(2);     // both alus
5140 %}
5141 
5142 // Integer ALU reg-reg operation
5143 pipe_class ialu_reg_reg_fat(eRegI dst, memory src) %{
5144     single_instruction;
5145     dst    : S4(write);
5146     src    : S3(read);
5147     D0     : S0;        // big decoder only
5148     ALU    : S3;        // any alu
5149 %}
5150 
5151 // Long ALU reg-reg operation
5152 pipe_class ialu_reg_reg_long_fat(eRegL dst, eRegL src) %{
5153     instruction_count(2);
5154     dst    : S4(write);
5155     src    : S3(read);
5156     D0     : S0(2);     // big decoder only; twice
5157     ALU    : S3(2);     // both alus
5158 %}
5159 
5160 // Integer ALU reg-mem operation
5161 pipe_class ialu_reg_mem(eRegI dst, memory mem) %{
5162     single_instruction;
5163     dst    : S5(write);
5164     mem    : S3(read);
5165     D0     : S0;        // big decoder only
5166     ALU    : S4;        // any alu
5167     MEM    : S3;        // any mem
5168 %}
5169 
5170 // Long ALU reg-mem operation
5171 pipe_class ialu_reg_long_mem(eRegL dst, load_long_memory mem) %{
5172     instruction_count(2);
5173     dst    : S5(write);
5174     mem    : S3(read);
5175     D0     : S0(2);     // big decoder only; twice
5176     ALU    : S4(2);     // any 2 alus
5177     MEM    : S3(2);     // both mems
5178 %}
5179 
5180 // Integer mem operation (prefetch)
5181 pipe_class ialu_mem(memory mem)
5182 %{
5183     single_instruction;
5184     mem    : S3(read);
5185     D0     : S0;        // big decoder only
5186     MEM    : S3;        // any mem
5187 %}
5188 
5189 // Integer Store to Memory
5190 pipe_class ialu_mem_reg(memory mem, eRegI src) %{
5191     single_instruction;
5192     mem    : S3(read);
5193     src    : S5(read);
5194     D0     : S0;        // big decoder only
5195     ALU    : S4;        // any alu
5196     MEM    : S3;
5197 %}
5198 
5199 // Long Store to Memory
5200 pipe_class ialu_mem_long_reg(memory mem, eRegL src) %{
5201     instruction_count(2);
5202     mem    : S3(read);
5203     src    : S5(read);
5204     D0     : S0(2);     // big decoder only; twice
5205     ALU    : S4(2);     // any 2 alus
5206     MEM    : S3(2);     // Both mems
5207 %}
5208 
5209 // Integer Store to Memory
5210 pipe_class ialu_mem_imm(memory mem) %{
5211     single_instruction;
5212     mem    : S3(read);
5213     D0     : S0;        // big decoder only
5214     ALU    : S4;        // any alu
5215     MEM    : S3;
5216 %}
5217 
5218 // Integer ALU0 reg-reg operation
5219 pipe_class ialu_reg_reg_alu0(eRegI dst, eRegI src) %{
5220     single_instruction;
5221     dst    : S4(write);
5222     src    : S3(read);
5223     D0     : S0;        // Big decoder only
5224     ALU0   : S3;        // only alu0
5225 %}
5226 
5227 // Integer ALU0 reg-mem operation
5228 pipe_class ialu_reg_mem_alu0(eRegI dst, memory mem) %{
5229     single_instruction;
5230     dst    : S5(write);
5231     mem    : S3(read);
5232     D0     : S0;        // big decoder only
5233     ALU0   : S4;        // ALU0 only
5234     MEM    : S3;        // any mem
5235 %}
5236 
5237 // Integer ALU reg-reg operation
5238 pipe_class ialu_cr_reg_reg(eFlagsReg cr, eRegI src1, eRegI src2) %{
5239     single_instruction;
5240     cr     : S4(write);
5241     src1   : S3(read);
5242     src2   : S3(read);
5243     DECODE : S0;        // any decoder
5244     ALU    : S3;        // any alu
5245 %}
5246 
5247 // Integer ALU reg-imm operation
5248 pipe_class ialu_cr_reg_imm(eFlagsReg cr, eRegI src1) %{
5249     single_instruction;
5250     cr     : S4(write);
5251     src1   : S3(read);
5252     DECODE : S0;        // any decoder
5253     ALU    : S3;        // any alu
5254 %}
5255 
5256 // Integer ALU reg-mem operation
5257 pipe_class ialu_cr_reg_mem(eFlagsReg cr, eRegI src1, memory src2) %{
5258     single_instruction;
5259     cr     : S4(write);
5260     src1   : S3(read);
5261     src2   : S3(read);
5262     D0     : S0;        // big decoder only
5263     ALU    : S4;        // any alu
5264     MEM    : S3;
5265 %}
5266 
5267 // Conditional move reg-reg
5268 pipe_class pipe_cmplt( eRegI p, eRegI q, eRegI y ) %{
5269     instruction_count(4);
5270     y      : S4(read);
5271     q      : S3(read);
5272     p      : S3(read);
5273     DECODE : S0(4);     // any decoder
5274 %}
5275 
5276 // Conditional move reg-reg
5277 pipe_class pipe_cmov_reg( eRegI dst, eRegI src, eFlagsReg cr ) %{
5278     single_instruction;
5279     dst    : S4(write);
5280     src    : S3(read);
5281     cr     : S3(read);
5282     DECODE : S0;        // any decoder
5283 %}
5284 
5285 // Conditional move reg-mem
5286 pipe_class pipe_cmov_mem( eFlagsReg cr, eRegI dst, memory src) %{
5287     single_instruction;
5288     dst    : S4(write);
5289     src    : S3(read);
5290     cr     : S3(read);
5291     DECODE : S0;        // any decoder
5292     MEM    : S3;
5293 %}
5294 
5295 // Conditional move reg-reg long
5296 pipe_class pipe_cmov_reg_long( eFlagsReg cr, eRegL dst, eRegL src) %{
5297     single_instruction;
5298     dst    : S4(write);
5299     src    : S3(read);
5300     cr     : S3(read);
5301     DECODE : S0(2);     // any 2 decoders
5302 %}
5303 
5304 // Conditional move double reg-reg
5305 pipe_class pipe_cmovDPR_reg( eFlagsReg cr, regDPR1 dst, regDPR src) %{
5306     single_instruction;
5307     dst    : S4(write);
5308     src    : S3(read);
5309     cr     : S3(read);
5310     DECODE : S0;        // any decoder
5311 %}
5312 
5313 // Float reg-reg operation
5314 pipe_class fpu_reg(regDPR dst) %{
5315     instruction_count(2);
5316     dst    : S3(read);
5317     DECODE : S0(2);     // any 2 decoders
5318     FPU    : S3;
5319 %}
5320 
5321 // Float reg-reg operation
5322 pipe_class fpu_reg_reg(regDPR dst, regDPR src) %{
5323     instruction_count(2);
5324     dst    : S4(write);
5325     src    : S3(read);
5326     DECODE : S0(2);     // any 2 decoders
5327     FPU    : S3;
5328 %}
5329 
5330 // Float reg-reg operation
5331 pipe_class fpu_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2) %{
5332     instruction_count(3);
5333     dst    : S4(write);
5334     src1   : S3(read);
5335     src2   : S3(read);
5336     DECODE : S0(3);     // any 3 decoders
5337     FPU    : S3(2);
5338 %}
5339 
5340 // Float reg-reg operation
5341 pipe_class fpu_reg_reg_reg_reg(regDPR dst, regDPR src1, regDPR src2, regDPR src3) %{
5342     instruction_count(4);
5343     dst    : S4(write);
5344     src1   : S3(read);
5345     src2   : S3(read);
5346     src3   : S3(read);
5347     DECODE : S0(4);     // any 3 decoders
5348     FPU    : S3(2);
5349 %}
5350 
5351 // Float reg-reg operation
5352 pipe_class fpu_reg_mem_reg_reg(regDPR dst, memory src1, regDPR src2, regDPR src3) %{
5353     instruction_count(4);
5354     dst    : S4(write);
5355     src1   : S3(read);
5356     src2   : S3(read);
5357     src3   : S3(read);
5358     DECODE : S1(3);     // any 3 decoders
5359     D0     : S0;        // Big decoder only
5360     FPU    : S3(2);
5361     MEM    : S3;
5362 %}
5363 
5364 // Float reg-mem operation
5365 pipe_class fpu_reg_mem(regDPR dst, memory mem) %{
5366     instruction_count(2);
5367     dst    : S5(write);
5368     mem    : S3(read);
5369     D0     : S0;        // big decoder only
5370     DECODE : S1;        // any decoder for FPU POP
5371     FPU    : S4;
5372     MEM    : S3;        // any mem
5373 %}
5374 
5375 // Float reg-mem operation
5376 pipe_class fpu_reg_reg_mem(regDPR dst, regDPR src1, memory mem) %{
5377     instruction_count(3);
5378     dst    : S5(write);
5379     src1   : S3(read);
5380     mem    : S3(read);
5381     D0     : S0;        // big decoder only
5382     DECODE : S1(2);     // any decoder for FPU POP
5383     FPU    : S4;
5384     MEM    : S3;        // any mem
5385 %}
5386 
5387 // Float mem-reg operation
5388 pipe_class fpu_mem_reg(memory mem, regDPR src) %{
5389     instruction_count(2);
5390     src    : S5(read);
5391     mem    : S3(read);
5392     DECODE : S0;        // any decoder for FPU PUSH
5393     D0     : S1;        // big decoder only
5394     FPU    : S4;
5395     MEM    : S3;        // any mem
5396 %}
5397 
5398 pipe_class fpu_mem_reg_reg(memory mem, regDPR src1, regDPR src2) %{
5399     instruction_count(3);
5400     src1   : S3(read);
5401     src2   : S3(read);
5402     mem    : S3(read);
5403     DECODE : S0(2);     // any decoder for FPU PUSH
5404     D0     : S1;        // big decoder only
5405     FPU    : S4;
5406     MEM    : S3;        // any mem
5407 %}
5408 
5409 pipe_class fpu_mem_reg_mem(memory mem, regDPR src1, memory src2) %{
5410     instruction_count(3);
5411     src1   : S3(read);
5412     src2   : S3(read);
5413     mem    : S4(read);
5414     DECODE : S0;        // any decoder for FPU PUSH
5415     D0     : S0(2);     // big decoder only
5416     FPU    : S4;
5417     MEM    : S3(2);     // any mem
5418 %}
5419 
5420 pipe_class fpu_mem_mem(memory dst, memory src1) %{
5421     instruction_count(2);
5422     src1   : S3(read);
5423     dst    : S4(read);
5424     D0     : S0(2);     // big decoder only
5425     MEM    : S3(2);     // any mem
5426 %}
5427 
5428 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2) %{
5429     instruction_count(3);
5430     src1   : S3(read);
5431     src2   : S3(read);
5432     dst    : S4(read);
5433     D0     : S0(3);     // big decoder only
5434     FPU    : S4;
5435     MEM    : S3(3);     // any mem
5436 %}
5437 
5438 pipe_class fpu_mem_reg_con(memory mem, regDPR src1) %{
5439     instruction_count(3);
5440     src1   : S4(read);
5441     mem    : S4(read);
5442     DECODE : S0;        // any decoder for FPU PUSH
5443     D0     : S0(2);     // big decoder only
5444     FPU    : S4;
5445     MEM    : S3(2);     // any mem
5446 %}
5447 
5448 // Float load constant
5449 pipe_class fpu_reg_con(regDPR dst) %{
5450     instruction_count(2);
5451     dst    : S5(write);
5452     D0     : S0;        // big decoder only for the load
5453     DECODE : S1;        // any decoder for FPU POP
5454     FPU    : S4;
5455     MEM    : S3;        // any mem
5456 %}
5457 
5458 // Float load constant
5459 pipe_class fpu_reg_reg_con(regDPR dst, regDPR src) %{
5460     instruction_count(3);
5461     dst    : S5(write);
5462     src    : S3(read);
5463     D0     : S0;        // big decoder only for the load
5464     DECODE : S1(2);     // any decoder for FPU POP
5465     FPU    : S4;
5466     MEM    : S3;        // any mem
5467 %}
5468 
5469 // UnConditional branch
5470 pipe_class pipe_jmp( label labl ) %{
5471     single_instruction;
5472     BR   : S3;
5473 %}
5474 
5475 // Conditional branch
5476 pipe_class pipe_jcc( cmpOp cmp, eFlagsReg cr, label labl ) %{
5477     single_instruction;
5478     cr    : S1(read);
5479     BR    : S3;
5480 %}
5481 
5482 // Allocation idiom
5483 pipe_class pipe_cmpxchg( eRegP dst, eRegP heap_ptr ) %{
5484     instruction_count(1); force_serialization;
5485     fixed_latency(6);
5486     heap_ptr : S3(read);
5487     DECODE   : S0(3);
5488     D0       : S2;
5489     MEM      : S3;
5490     ALU      : S3(2);
5491     dst      : S5(write);
5492     BR       : S5;
5493 %}
5494 
5495 // Generic big/slow expanded idiom
5496 pipe_class pipe_slow(  ) %{
5497     instruction_count(10); multiple_bundles; force_serialization;
5498     fixed_latency(100);
5499     D0  : S0(2);
5500     MEM : S3(2);
5501 %}
5502 
5503 // The real do-nothing guy
5504 pipe_class empty( ) %{
5505     instruction_count(0);
5506 %}
5507 
5508 // Define the class for the Nop node
5509 define %{
5510    MachNop = empty;
5511 %}
5512 
5513 %}
5514 
5515 //----------INSTRUCTIONS-------------------------------------------------------
5516 //
5517 // match      -- States which machine-independent subtree may be replaced
5518 //               by this instruction.
5519 // ins_cost   -- The estimated cost of this instruction is used by instruction
5520 //               selection to identify a minimum cost tree of machine
5521 //               instructions that matches a tree of machine-independent
5522 //               instructions.
5523 // format     -- A string providing the disassembly for this instruction.
5524 //               The value of an instruction's operand may be inserted
5525 //               by referring to it with a '$' prefix.
5526 // opcode     -- Three instruction opcodes may be provided.  These are referred
5527 //               to within an encode class as $primary, $secondary, and $tertiary
5528 //               respectively.  The primary opcode is commonly used to
5529 //               indicate the type of machine instruction, while secondary
5530 //               and tertiary are often used for prefix options or addressing
5531 //               modes.
5532 // ins_encode -- A list of encode classes with parameters. The encode class
5533 //               name must have been defined in an 'enc_class' specification
5534 //               in the encode section of the architecture description.
5535 
5536 //----------BSWAP-Instruction--------------------------------------------------
5537 instruct bytes_reverse_int(eRegI dst) %{
5538   match(Set dst (ReverseBytesI dst));
5539 
5540   format %{ "BSWAP  $dst" %}
5541   opcode(0x0F, 0xC8);
5542   ins_encode( OpcP, OpcSReg(dst) );
5543   ins_pipe( ialu_reg );
5544 %}
5545 
5546 instruct bytes_reverse_long(eRegL dst) %{
5547   match(Set dst (ReverseBytesL dst));
5548 
5549   format %{ "BSWAP  $dst.lo\n\t"
5550             "BSWAP  $dst.hi\n\t"
5551             "XCHG   $dst.lo $dst.hi" %}
5552 
5553   ins_cost(125);
5554   ins_encode( bswap_long_bytes(dst) );
5555   ins_pipe( ialu_reg_reg);
5556 %}
5557 
5558 instruct bytes_reverse_unsigned_short(eRegI dst) %{
5559   match(Set dst (ReverseBytesUS dst));
5560 
5561   format %{ "BSWAP  $dst\n\t" 
5562             "SHR    $dst,16\n\t" %}
5563   ins_encode %{
5564     __ bswapl($dst$$Register);
5565     __ shrl($dst$$Register, 16); 
5566   %}
5567   ins_pipe( ialu_reg );
5568 %}
5569 
5570 instruct bytes_reverse_short(eRegI dst) %{
5571   match(Set dst (ReverseBytesS dst));
5572 
5573   format %{ "BSWAP  $dst\n\t" 
5574             "SAR    $dst,16\n\t" %}
5575   ins_encode %{
5576     __ bswapl($dst$$Register);
5577     __ sarl($dst$$Register, 16); 
5578   %}
5579   ins_pipe( ialu_reg );
5580 %}
5581 
5582 
5583 //---------- Zeros Count Instructions ------------------------------------------
5584 
5585 instruct countLeadingZerosI(eRegI dst, eRegI src, eFlagsReg cr) %{
5586   predicate(UseCountLeadingZerosInstruction);
5587   match(Set dst (CountLeadingZerosI src));
5588   effect(KILL cr);
5589 
5590   format %{ "LZCNT  $dst, $src\t# count leading zeros (int)" %}
5591   ins_encode %{
5592     __ lzcntl($dst$$Register, $src$$Register);
5593   %}
5594   ins_pipe(ialu_reg);
5595 %}
5596 
5597 instruct countLeadingZerosI_bsr(eRegI dst, eRegI src, eFlagsReg cr) %{
5598   predicate(!UseCountLeadingZerosInstruction);
5599   match(Set dst (CountLeadingZerosI src));
5600   effect(KILL cr);
5601 
5602   format %{ "BSR    $dst, $src\t# count leading zeros (int)\n\t"
5603             "JNZ    skip\n\t"
5604             "MOV    $dst, -1\n"
5605       "skip:\n\t"
5606             "NEG    $dst\n\t"
5607             "ADD    $dst, 31" %}
5608   ins_encode %{
5609     Register Rdst = $dst$$Register;
5610     Register Rsrc = $src$$Register;
5611     Label skip;
5612     __ bsrl(Rdst, Rsrc);
5613     __ jccb(Assembler::notZero, skip);
5614     __ movl(Rdst, -1);
5615     __ bind(skip);
5616     __ negl(Rdst);
5617     __ addl(Rdst, BitsPerInt - 1);
5618   %}
5619   ins_pipe(ialu_reg);
5620 %}
5621 
5622 instruct countLeadingZerosL(eRegI dst, eRegL src, eFlagsReg cr) %{
5623   predicate(UseCountLeadingZerosInstruction);
5624   match(Set dst (CountLeadingZerosL src));
5625   effect(TEMP dst, KILL cr);
5626 
5627   format %{ "LZCNT  $dst, $src.hi\t# count leading zeros (long)\n\t"
5628             "JNC    done\n\t"
5629             "LZCNT  $dst, $src.lo\n\t"
5630             "ADD    $dst, 32\n"
5631       "done:" %}
5632   ins_encode %{
5633     Register Rdst = $dst$$Register;
5634     Register Rsrc = $src$$Register;
5635     Label done;
5636     __ lzcntl(Rdst, HIGH_FROM_LOW(Rsrc));
5637     __ jccb(Assembler::carryClear, done);
5638     __ lzcntl(Rdst, Rsrc);
5639     __ addl(Rdst, BitsPerInt);
5640     __ bind(done);
5641   %}
5642   ins_pipe(ialu_reg);
5643 %}
5644 
5645 instruct countLeadingZerosL_bsr(eRegI dst, eRegL src, eFlagsReg cr) %{
5646   predicate(!UseCountLeadingZerosInstruction);
5647   match(Set dst (CountLeadingZerosL src));
5648   effect(TEMP dst, KILL cr);
5649 
5650   format %{ "BSR    $dst, $src.hi\t# count leading zeros (long)\n\t"
5651             "JZ     msw_is_zero\n\t"
5652             "ADD    $dst, 32\n\t"
5653             "JMP    not_zero\n"
5654       "msw_is_zero:\n\t"
5655             "BSR    $dst, $src.lo\n\t"
5656             "JNZ    not_zero\n\t"
5657             "MOV    $dst, -1\n"
5658       "not_zero:\n\t"
5659             "NEG    $dst\n\t"
5660             "ADD    $dst, 63\n" %}
5661  ins_encode %{
5662     Register Rdst = $dst$$Register;
5663     Register Rsrc = $src$$Register;
5664     Label msw_is_zero;
5665     Label not_zero;
5666     __ bsrl(Rdst, HIGH_FROM_LOW(Rsrc));
5667     __ jccb(Assembler::zero, msw_is_zero);
5668     __ addl(Rdst, BitsPerInt);
5669     __ jmpb(not_zero);
5670     __ bind(msw_is_zero);
5671     __ bsrl(Rdst, Rsrc);
5672     __ jccb(Assembler::notZero, not_zero);
5673     __ movl(Rdst, -1);
5674     __ bind(not_zero);
5675     __ negl(Rdst);
5676     __ addl(Rdst, BitsPerLong - 1);
5677   %}
5678   ins_pipe(ialu_reg);
5679 %}
5680 
5681 instruct countTrailingZerosI(eRegI dst, eRegI src, eFlagsReg cr) %{
5682   match(Set dst (CountTrailingZerosI src));
5683   effect(KILL cr);
5684 
5685   format %{ "BSF    $dst, $src\t# count trailing zeros (int)\n\t"
5686             "JNZ    done\n\t"
5687             "MOV    $dst, 32\n"
5688       "done:" %}
5689   ins_encode %{
5690     Register Rdst = $dst$$Register;
5691     Label done;
5692     __ bsfl(Rdst, $src$$Register);
5693     __ jccb(Assembler::notZero, done);
5694     __ movl(Rdst, BitsPerInt);
5695     __ bind(done);
5696   %}
5697   ins_pipe(ialu_reg);
5698 %}
5699 
5700 instruct countTrailingZerosL(eRegI dst, eRegL src, eFlagsReg cr) %{
5701   match(Set dst (CountTrailingZerosL src));
5702   effect(TEMP dst, KILL cr);
5703 
5704   format %{ "BSF    $dst, $src.lo\t# count trailing zeros (long)\n\t"
5705             "JNZ    done\n\t"
5706             "BSF    $dst, $src.hi\n\t"
5707             "JNZ    msw_not_zero\n\t"
5708             "MOV    $dst, 32\n"
5709       "msw_not_zero:\n\t"
5710             "ADD    $dst, 32\n"
5711       "done:" %}
5712   ins_encode %{
5713     Register Rdst = $dst$$Register;
5714     Register Rsrc = $src$$Register;
5715     Label msw_not_zero;
5716     Label done;
5717     __ bsfl(Rdst, Rsrc);
5718     __ jccb(Assembler::notZero, done);
5719     __ bsfl(Rdst, HIGH_FROM_LOW(Rsrc));
5720     __ jccb(Assembler::notZero, msw_not_zero);
5721     __ movl(Rdst, BitsPerInt);
5722     __ bind(msw_not_zero);
5723     __ addl(Rdst, BitsPerInt);
5724     __ bind(done);
5725   %}
5726   ins_pipe(ialu_reg);
5727 %}
5728 
5729 
5730 //---------- Population Count Instructions -------------------------------------
5731 
5732 instruct popCountI(eRegI dst, eRegI src) %{
5733   predicate(UsePopCountInstruction);
5734   match(Set dst (PopCountI src));
5735 
5736   format %{ "POPCNT $dst, $src" %}
5737   ins_encode %{
5738     __ popcntl($dst$$Register, $src$$Register);
5739   %}
5740   ins_pipe(ialu_reg);
5741 %}
5742 
5743 instruct popCountI_mem(eRegI dst, memory mem) %{
5744   predicate(UsePopCountInstruction);
5745   match(Set dst (PopCountI (LoadI mem)));
5746 
5747   format %{ "POPCNT $dst, $mem" %}
5748   ins_encode %{
5749     __ popcntl($dst$$Register, $mem$$Address);
5750   %}
5751   ins_pipe(ialu_reg);
5752 %}
5753 
5754 // Note: Long.bitCount(long) returns an int.
5755 instruct popCountL(eRegI dst, eRegL src, eRegI tmp, eFlagsReg cr) %{
5756   predicate(UsePopCountInstruction);
5757   match(Set dst (PopCountL src));
5758   effect(KILL cr, TEMP tmp, TEMP dst);
5759 
5760   format %{ "POPCNT $dst, $src.lo\n\t"
5761             "POPCNT $tmp, $src.hi\n\t"
5762             "ADD    $dst, $tmp" %}
5763   ins_encode %{
5764     __ popcntl($dst$$Register, $src$$Register);
5765     __ popcntl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
5766     __ addl($dst$$Register, $tmp$$Register);
5767   %}
5768   ins_pipe(ialu_reg);
5769 %}
5770 
5771 // Note: Long.bitCount(long) returns an int.
5772 instruct popCountL_mem(eRegI dst, memory mem, eRegI tmp, eFlagsReg cr) %{
5773   predicate(UsePopCountInstruction);
5774   match(Set dst (PopCountL (LoadL mem)));
5775   effect(KILL cr, TEMP tmp, TEMP dst);
5776 
5777   format %{ "POPCNT $dst, $mem\n\t"
5778             "POPCNT $tmp, $mem+4\n\t"
5779             "ADD    $dst, $tmp" %}
5780   ins_encode %{
5781     //__ popcntl($dst$$Register, $mem$$Address$$first);
5782     //__ popcntl($tmp$$Register, $mem$$Address$$second);
5783     __ popcntl($dst$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, false));
5784     __ popcntl($tmp$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, false));
5785     __ addl($dst$$Register, $tmp$$Register);
5786   %}
5787   ins_pipe(ialu_reg);
5788 %}
5789 
5790 
5791 //----------Load/Store/Move Instructions---------------------------------------
5792 //----------Load Instructions--------------------------------------------------
5793 // Load Byte (8bit signed)
5794 instruct loadB(xRegI dst, memory mem) %{
5795   match(Set dst (LoadB mem));
5796 
5797   ins_cost(125);
5798   format %{ "MOVSX8 $dst,$mem\t# byte" %}
5799 
5800   ins_encode %{
5801     __ movsbl($dst$$Register, $mem$$Address);
5802   %}
5803 
5804   ins_pipe(ialu_reg_mem);
5805 %}
5806 
5807 // Load Byte (8bit signed) into Long Register
5808 instruct loadB2L(eRegL dst, memory mem, eFlagsReg cr) %{
5809   match(Set dst (ConvI2L (LoadB mem)));
5810   effect(KILL cr);
5811 
5812   ins_cost(375);
5813   format %{ "MOVSX8 $dst.lo,$mem\t# byte -> long\n\t"
5814             "MOV    $dst.hi,$dst.lo\n\t"
5815             "SAR    $dst.hi,7" %}
5816 
5817   ins_encode %{
5818     __ movsbl($dst$$Register, $mem$$Address);
5819     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
5820     __ sarl(HIGH_FROM_LOW($dst$$Register), 7); // 24+1 MSB are already signed extended.
5821   %}
5822 
5823   ins_pipe(ialu_reg_mem);
5824 %}
5825 
5826 // Load Unsigned Byte (8bit UNsigned)
5827 instruct loadUB(xRegI dst, memory mem) %{
5828   match(Set dst (LoadUB mem));
5829 
5830   ins_cost(125);
5831   format %{ "MOVZX8 $dst,$mem\t# ubyte -> int" %}
5832 
5833   ins_encode %{
5834     __ movzbl($dst$$Register, $mem$$Address);
5835   %}
5836 
5837   ins_pipe(ialu_reg_mem);
5838 %}
5839 
5840 // Load Unsigned Byte (8 bit UNsigned) into Long Register
5841 instruct loadUB2L(eRegL dst, memory mem, eFlagsReg cr) %{
5842   match(Set dst (ConvI2L (LoadUB mem)));
5843   effect(KILL cr);
5844 
5845   ins_cost(250);
5846   format %{ "MOVZX8 $dst.lo,$mem\t# ubyte -> long\n\t"
5847             "XOR    $dst.hi,$dst.hi" %}
5848 
5849   ins_encode %{
5850     Register Rdst = $dst$$Register;
5851     __ movzbl(Rdst, $mem$$Address);
5852     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5853   %}
5854 
5855   ins_pipe(ialu_reg_mem);
5856 %}
5857 
5858 // Load Unsigned Byte (8 bit UNsigned) with mask into Long Register
5859 instruct loadUB2L_immI8(eRegL dst, memory mem, immI8 mask, eFlagsReg cr) %{
5860   match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5861   effect(KILL cr);
5862 
5863   format %{ "MOVZX8 $dst.lo,$mem\t# ubyte & 8-bit mask -> long\n\t"
5864             "XOR    $dst.hi,$dst.hi\n\t"
5865             "AND    $dst.lo,$mask" %}
5866   ins_encode %{
5867     Register Rdst = $dst$$Register;
5868     __ movzbl(Rdst, $mem$$Address);
5869     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5870     __ andl(Rdst, $mask$$constant);
5871   %}
5872   ins_pipe(ialu_reg_mem);
5873 %}
5874 
5875 // Load Short (16bit signed)
5876 instruct loadS(eRegI dst, memory mem) %{
5877   match(Set dst (LoadS mem));
5878 
5879   ins_cost(125);
5880   format %{ "MOVSX  $dst,$mem\t# short" %}
5881 
5882   ins_encode %{
5883     __ movswl($dst$$Register, $mem$$Address);
5884   %}
5885 
5886   ins_pipe(ialu_reg_mem);
5887 %}
5888 
5889 // Load Short (16 bit signed) to Byte (8 bit signed)
5890 instruct loadS2B(eRegI dst, memory mem, immI_24 twentyfour) %{
5891   match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5892 
5893   ins_cost(125);
5894   format %{ "MOVSX  $dst, $mem\t# short -> byte" %}
5895   ins_encode %{
5896     __ movsbl($dst$$Register, $mem$$Address);
5897   %}
5898   ins_pipe(ialu_reg_mem);
5899 %}
5900 
5901 // Load Short (16bit signed) into Long Register
5902 instruct loadS2L(eRegL dst, memory mem, eFlagsReg cr) %{
5903   match(Set dst (ConvI2L (LoadS mem)));
5904   effect(KILL cr);
5905 
5906   ins_cost(375);
5907   format %{ "MOVSX  $dst.lo,$mem\t# short -> long\n\t"
5908             "MOV    $dst.hi,$dst.lo\n\t"
5909             "SAR    $dst.hi,15" %}
5910 
5911   ins_encode %{
5912     __ movswl($dst$$Register, $mem$$Address);
5913     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
5914     __ sarl(HIGH_FROM_LOW($dst$$Register), 15); // 16+1 MSB are already signed extended.
5915   %}
5916 
5917   ins_pipe(ialu_reg_mem);
5918 %}
5919 
5920 // Load Unsigned Short/Char (16bit unsigned)
5921 instruct loadUS(eRegI dst, memory mem) %{
5922   match(Set dst (LoadUS mem));
5923 
5924   ins_cost(125);
5925   format %{ "MOVZX  $dst,$mem\t# ushort/char -> int" %}
5926 
5927   ins_encode %{
5928     __ movzwl($dst$$Register, $mem$$Address);
5929   %}
5930 
5931   ins_pipe(ialu_reg_mem);
5932 %}
5933 
5934 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5935 instruct loadUS2B(eRegI dst, memory mem, immI_24 twentyfour) %{
5936   match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5937 
5938   ins_cost(125);
5939   format %{ "MOVSX  $dst, $mem\t# ushort -> byte" %}
5940   ins_encode %{
5941     __ movsbl($dst$$Register, $mem$$Address);
5942   %}
5943   ins_pipe(ialu_reg_mem);
5944 %}
5945 
5946 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
5947 instruct loadUS2L(eRegL dst, memory mem, eFlagsReg cr) %{
5948   match(Set dst (ConvI2L (LoadUS mem)));
5949   effect(KILL cr);
5950 
5951   ins_cost(250);
5952   format %{ "MOVZX  $dst.lo,$mem\t# ushort/char -> long\n\t"
5953             "XOR    $dst.hi,$dst.hi" %}
5954 
5955   ins_encode %{
5956     __ movzwl($dst$$Register, $mem$$Address);
5957     __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
5958   %}
5959 
5960   ins_pipe(ialu_reg_mem);
5961 %}
5962 
5963 // Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
5964 instruct loadUS2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
5965   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5966   effect(KILL cr);
5967 
5968   format %{ "MOVZX8 $dst.lo,$mem\t# ushort/char & 0xFF -> long\n\t"
5969             "XOR    $dst.hi,$dst.hi" %}
5970   ins_encode %{
5971     Register Rdst = $dst$$Register;
5972     __ movzbl(Rdst, $mem$$Address);
5973     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5974   %}
5975   ins_pipe(ialu_reg_mem);
5976 %}
5977 
5978 // Load Unsigned Short/Char (16 bit UNsigned) with a 16-bit mask into Long Register
5979 instruct loadUS2L_immI16(eRegL dst, memory mem, immI16 mask, eFlagsReg cr) %{
5980   match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5981   effect(KILL cr);
5982 
5983   format %{ "MOVZX  $dst.lo, $mem\t# ushort/char & 16-bit mask -> long\n\t"
5984             "XOR    $dst.hi,$dst.hi\n\t"
5985             "AND    $dst.lo,$mask" %}
5986   ins_encode %{
5987     Register Rdst = $dst$$Register;
5988     __ movzwl(Rdst, $mem$$Address);
5989     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
5990     __ andl(Rdst, $mask$$constant);
5991   %}
5992   ins_pipe(ialu_reg_mem);
5993 %}
5994 
5995 // Load Integer
5996 instruct loadI(eRegI dst, memory mem) %{
5997   match(Set dst (LoadI mem));
5998 
5999   ins_cost(125);
6000   format %{ "MOV    $dst,$mem\t# int" %}
6001 
6002   ins_encode %{
6003     __ movl($dst$$Register, $mem$$Address);
6004   %}
6005 
6006   ins_pipe(ialu_reg_mem);
6007 %}
6008 
6009 // Load Integer (32 bit signed) to Byte (8 bit signed)
6010 instruct loadI2B(eRegI dst, memory mem, immI_24 twentyfour) %{
6011   match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
6012 
6013   ins_cost(125);
6014   format %{ "MOVSX  $dst, $mem\t# int -> byte" %}
6015   ins_encode %{
6016     __ movsbl($dst$$Register, $mem$$Address);
6017   %}
6018   ins_pipe(ialu_reg_mem);
6019 %}
6020 
6021 // Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
6022 instruct loadI2UB(eRegI dst, memory mem, immI_255 mask) %{
6023   match(Set dst (AndI (LoadI mem) mask));
6024 
6025   ins_cost(125);
6026   format %{ "MOVZX  $dst, $mem\t# int -> ubyte" %}
6027   ins_encode %{
6028     __ movzbl($dst$$Register, $mem$$Address);
6029   %}
6030   ins_pipe(ialu_reg_mem);
6031 %}
6032 
6033 // Load Integer (32 bit signed) to Short (16 bit signed)
6034 instruct loadI2S(eRegI dst, memory mem, immI_16 sixteen) %{
6035   match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
6036 
6037   ins_cost(125);
6038   format %{ "MOVSX  $dst, $mem\t# int -> short" %}
6039   ins_encode %{
6040     __ movswl($dst$$Register, $mem$$Address);
6041   %}
6042   ins_pipe(ialu_reg_mem);
6043 %}
6044 
6045 // Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
6046 instruct loadI2US(eRegI dst, memory mem, immI_65535 mask) %{
6047   match(Set dst (AndI (LoadI mem) mask));
6048 
6049   ins_cost(125);
6050   format %{ "MOVZX  $dst, $mem\t# int -> ushort/char" %}
6051   ins_encode %{
6052     __ movzwl($dst$$Register, $mem$$Address);
6053   %}
6054   ins_pipe(ialu_reg_mem);
6055 %}
6056 
6057 // Load Integer into Long Register
6058 instruct loadI2L(eRegL dst, memory mem, eFlagsReg cr) %{
6059   match(Set dst (ConvI2L (LoadI mem)));
6060   effect(KILL cr);
6061 
6062   ins_cost(375);
6063   format %{ "MOV    $dst.lo,$mem\t# int -> long\n\t"
6064             "MOV    $dst.hi,$dst.lo\n\t"
6065             "SAR    $dst.hi,31" %}
6066 
6067   ins_encode %{
6068     __ movl($dst$$Register, $mem$$Address);
6069     __ movl(HIGH_FROM_LOW($dst$$Register), $dst$$Register); // This is always a different register.
6070     __ sarl(HIGH_FROM_LOW($dst$$Register), 31);
6071   %}
6072 
6073   ins_pipe(ialu_reg_mem);
6074 %}
6075 
6076 // Load Integer with mask 0xFF into Long Register
6077 instruct loadI2L_immI_255(eRegL dst, memory mem, immI_255 mask, eFlagsReg cr) %{
6078   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6079   effect(KILL cr);
6080 
6081   format %{ "MOVZX8 $dst.lo,$mem\t# int & 0xFF -> long\n\t"
6082             "XOR    $dst.hi,$dst.hi" %}
6083   ins_encode %{
6084     Register Rdst = $dst$$Register;
6085     __ movzbl(Rdst, $mem$$Address);
6086     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6087   %}
6088   ins_pipe(ialu_reg_mem);
6089 %}
6090 
6091 // Load Integer with mask 0xFFFF into Long Register
6092 instruct loadI2L_immI_65535(eRegL dst, memory mem, immI_65535 mask, eFlagsReg cr) %{
6093   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6094   effect(KILL cr);
6095 
6096   format %{ "MOVZX  $dst.lo,$mem\t# int & 0xFFFF -> long\n\t"
6097             "XOR    $dst.hi,$dst.hi" %}
6098   ins_encode %{
6099     Register Rdst = $dst$$Register;
6100     __ movzwl(Rdst, $mem$$Address);
6101     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6102   %}
6103   ins_pipe(ialu_reg_mem);
6104 %}
6105 
6106 // Load Integer with 32-bit mask into Long Register
6107 instruct loadI2L_immI(eRegL dst, memory mem, immI mask, eFlagsReg cr) %{
6108   match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6109   effect(KILL cr);
6110 
6111   format %{ "MOV    $dst.lo,$mem\t# int & 32-bit mask -> long\n\t"
6112             "XOR    $dst.hi,$dst.hi\n\t"
6113             "AND    $dst.lo,$mask" %}
6114   ins_encode %{
6115     Register Rdst = $dst$$Register;
6116     __ movl(Rdst, $mem$$Address);
6117     __ xorl(HIGH_FROM_LOW(Rdst), HIGH_FROM_LOW(Rdst));
6118     __ andl(Rdst, $mask$$constant);
6119   %}
6120   ins_pipe(ialu_reg_mem);
6121 %}
6122 
6123 // Load Unsigned Integer into Long Register
6124 instruct loadUI2L(eRegL dst, memory mem, eFlagsReg cr) %{
6125   match(Set dst (LoadUI2L mem));
6126   effect(KILL cr);
6127 
6128   ins_cost(250);
6129   format %{ "MOV    $dst.lo,$mem\t# uint -> long\n\t"
6130             "XOR    $dst.hi,$dst.hi" %}
6131 
6132   ins_encode %{
6133     __ movl($dst$$Register, $mem$$Address);
6134     __ xorl(HIGH_FROM_LOW($dst$$Register), HIGH_FROM_LOW($dst$$Register));
6135   %}
6136 
6137   ins_pipe(ialu_reg_mem);
6138 %}
6139 
6140 // Load Long.  Cannot clobber address while loading, so restrict address
6141 // register to ESI
6142 instruct loadL(eRegL dst, load_long_memory mem) %{
6143   predicate(!((LoadLNode*)n)->require_atomic_access());
6144   match(Set dst (LoadL mem));
6145 
6146   ins_cost(250);
6147   format %{ "MOV    $dst.lo,$mem\t# long\n\t"
6148             "MOV    $dst.hi,$mem+4" %}
6149 
6150   ins_encode %{
6151     Address Amemlo = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, false);
6152     Address Amemhi = Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, false);
6153     __ movl($dst$$Register, Amemlo);
6154     __ movl(HIGH_FROM_LOW($dst$$Register), Amemhi);
6155   %}
6156 
6157   ins_pipe(ialu_reg_long_mem);
6158 %}
6159 
6160 // Volatile Load Long.  Must be atomic, so do 64-bit FILD
6161 // then store it down to the stack and reload on the int
6162 // side.
6163 instruct loadL_volatile(stackSlotL dst, memory mem) %{
6164   predicate(UseSSE<=1 && ((LoadLNode*)n)->require_atomic_access());
6165   match(Set dst (LoadL mem));
6166 
6167   ins_cost(200);
6168   format %{ "FILD   $mem\t# Atomic volatile long load\n\t"
6169             "FISTp  $dst" %}
6170   ins_encode(enc_loadL_volatile(mem,dst));
6171   ins_pipe( fpu_reg_mem );
6172 %}
6173 
6174 instruct loadLX_volatile(stackSlotL dst, memory mem, regD tmp) %{
6175   predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
6176   match(Set dst (LoadL mem));
6177   effect(TEMP tmp);
6178   ins_cost(180);
6179   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
6180             "MOVSD  $dst,$tmp" %}
6181   ins_encode %{
6182     __ movdbl($tmp$$XMMRegister, $mem$$Address);
6183     __ movdbl(Address(rsp, $dst$$disp), $tmp$$XMMRegister);
6184   %}
6185   ins_pipe( pipe_slow );
6186 %}
6187 
6188 instruct loadLX_reg_volatile(eRegL dst, memory mem, regD tmp) %{
6189   predicate(UseSSE>=2 && ((LoadLNode*)n)->require_atomic_access());
6190   match(Set dst (LoadL mem));
6191   effect(TEMP tmp);
6192   ins_cost(160);
6193   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
6194             "MOVD   $dst.lo,$tmp\n\t"
6195             "PSRLQ  $tmp,32\n\t"
6196             "MOVD   $dst.hi,$tmp" %}
6197   ins_encode %{
6198     __ movdbl($tmp$$XMMRegister, $mem$$Address);
6199     __ movdl($dst$$Register, $tmp$$XMMRegister);
6200     __ psrlq($tmp$$XMMRegister, 32);
6201     __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
6202   %}
6203   ins_pipe( pipe_slow );
6204 %}
6205 
6206 // Load Range
6207 instruct loadRange(eRegI dst, memory mem) %{
6208   match(Set dst (LoadRange mem));
6209 
6210   ins_cost(125);
6211   format %{ "MOV    $dst,$mem" %}
6212   opcode(0x8B);
6213   ins_encode( OpcP, RegMem(dst,mem));
6214   ins_pipe( ialu_reg_mem );
6215 %}
6216 
6217 
6218 // Load Pointer
6219 instruct loadP(eRegP dst, memory mem) %{
6220   match(Set dst (LoadP mem));
6221 
6222   ins_cost(125);
6223   format %{ "MOV    $dst,$mem" %}
6224   opcode(0x8B);
6225   ins_encode( OpcP, RegMem(dst,mem));
6226   ins_pipe( ialu_reg_mem );
6227 %}
6228 
6229 // Load Klass Pointer
6230 instruct loadKlass(eRegP dst, memory mem) %{
6231   match(Set dst (LoadKlass mem));
6232 
6233   ins_cost(125);
6234   format %{ "MOV    $dst,$mem" %}
6235   opcode(0x8B);
6236   ins_encode( OpcP, RegMem(dst,mem));
6237   ins_pipe( ialu_reg_mem );
6238 %}
6239 
6240 // Load Double
6241 instruct loadDPR(regDPR dst, memory mem) %{
6242   predicate(UseSSE<=1);
6243   match(Set dst (LoadD mem));
6244 
6245   ins_cost(150);
6246   format %{ "FLD_D  ST,$mem\n\t"
6247             "FSTP   $dst" %}
6248   opcode(0xDD);               /* DD /0 */
6249   ins_encode( OpcP, RMopc_Mem(0x00,mem),
6250               Pop_Reg_DPR(dst) );
6251   ins_pipe( fpu_reg_mem );
6252 %}
6253 
6254 // Load Double to XMM
6255 instruct loadD(regD dst, memory mem) %{
6256   predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
6257   match(Set dst (LoadD mem));
6258   ins_cost(145);
6259   format %{ "MOVSD  $dst,$mem" %}
6260   ins_encode %{
6261     __ movdbl ($dst$$XMMRegister, $mem$$Address);
6262   %}
6263   ins_pipe( pipe_slow );
6264 %}
6265 
6266 instruct loadD_partial(regD dst, memory mem) %{
6267   predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
6268   match(Set dst (LoadD mem));
6269   ins_cost(145);
6270   format %{ "MOVLPD $dst,$mem" %}
6271   ins_encode %{
6272     __ movdbl ($dst$$XMMRegister, $mem$$Address);
6273   %}
6274   ins_pipe( pipe_slow );
6275 %}
6276 
6277 // Load to XMM register (single-precision floating point)
6278 // MOVSS instruction
6279 instruct loadF(regF dst, memory mem) %{
6280   predicate(UseSSE>=1);
6281   match(Set dst (LoadF mem));
6282   ins_cost(145);
6283   format %{ "MOVSS  $dst,$mem" %}
6284   ins_encode %{
6285     __ movflt ($dst$$XMMRegister, $mem$$Address);
6286   %}
6287   ins_pipe( pipe_slow );
6288 %}
6289 
6290 // Load Float
6291 instruct loadFPR(regFPR dst, memory mem) %{
6292   predicate(UseSSE==0);
6293   match(Set dst (LoadF mem));
6294 
6295   ins_cost(150);
6296   format %{ "FLD_S  ST,$mem\n\t"
6297             "FSTP   $dst" %}
6298   opcode(0xD9);               /* D9 /0 */
6299   ins_encode( OpcP, RMopc_Mem(0x00,mem),
6300               Pop_Reg_FPR(dst) );
6301   ins_pipe( fpu_reg_mem );
6302 %}
6303 
6304 // Load Aligned Packed Byte to XMM register
6305 instruct loadA8B(regD dst, memory mem) %{
6306   predicate(UseSSE>=1);
6307   match(Set dst (Load8B mem));
6308   ins_cost(125);
6309   format %{ "MOVQ  $dst,$mem\t! packed8B" %}
6310   ins_encode %{
6311     __ movq($dst$$XMMRegister, $mem$$Address);
6312   %}
6313   ins_pipe( pipe_slow );
6314 %}
6315 
6316 // Load Aligned Packed Short to XMM register
6317 instruct loadA4S(regD dst, memory mem) %{
6318   predicate(UseSSE>=1);
6319   match(Set dst (Load4S mem));
6320   ins_cost(125);
6321   format %{ "MOVQ  $dst,$mem\t! packed4S" %}
6322   ins_encode %{
6323     __ movq($dst$$XMMRegister, $mem$$Address);
6324   %}
6325   ins_pipe( pipe_slow );
6326 %}
6327 
6328 // Load Aligned Packed Char to XMM register
6329 instruct loadA4C(regD dst, memory mem) %{
6330   predicate(UseSSE>=1);
6331   match(Set dst (Load4C mem));
6332   ins_cost(125);
6333   format %{ "MOVQ  $dst,$mem\t! packed4C" %}
6334   ins_encode %{
6335     __ movq($dst$$XMMRegister, $mem$$Address);
6336   %}
6337   ins_pipe( pipe_slow );
6338 %}
6339 
6340 // Load Aligned Packed Integer to XMM register
6341 instruct load2IU(regD dst, memory mem) %{
6342   predicate(UseSSE>=1);
6343   match(Set dst (Load2I mem));
6344   ins_cost(125);
6345   format %{ "MOVQ  $dst,$mem\t! packed2I" %}
6346   ins_encode %{
6347     __ movq($dst$$XMMRegister, $mem$$Address);
6348   %}
6349   ins_pipe( pipe_slow );
6350 %}
6351 
6352 // Load Aligned Packed Single to XMM
6353 instruct loadA2F(regD dst, memory mem) %{
6354   predicate(UseSSE>=1);
6355   match(Set dst (Load2F mem));
6356   ins_cost(145);
6357   format %{ "MOVQ  $dst,$mem\t! packed2F" %}
6358   ins_encode %{
6359     __ movq($dst$$XMMRegister, $mem$$Address);
6360   %}
6361   ins_pipe( pipe_slow );
6362 %}
6363 
6364 // Load Effective Address
6365 instruct leaP8(eRegP dst, indOffset8 mem) %{
6366   match(Set dst mem);
6367 
6368   ins_cost(110);
6369   format %{ "LEA    $dst,$mem" %}
6370   opcode(0x8D);
6371   ins_encode( OpcP, RegMem(dst,mem));
6372   ins_pipe( ialu_reg_reg_fat );
6373 %}
6374 
6375 instruct leaP32(eRegP dst, indOffset32 mem) %{
6376   match(Set dst mem);
6377 
6378   ins_cost(110);
6379   format %{ "LEA    $dst,$mem" %}
6380   opcode(0x8D);
6381   ins_encode( OpcP, RegMem(dst,mem));
6382   ins_pipe( ialu_reg_reg_fat );
6383 %}
6384 
6385 instruct leaPIdxOff(eRegP dst, indIndexOffset mem) %{
6386   match(Set dst mem);
6387 
6388   ins_cost(110);
6389   format %{ "LEA    $dst,$mem" %}
6390   opcode(0x8D);
6391   ins_encode( OpcP, RegMem(dst,mem));
6392   ins_pipe( ialu_reg_reg_fat );
6393 %}
6394 
6395 instruct leaPIdxScale(eRegP dst, indIndexScale mem) %{
6396   match(Set dst mem);
6397 
6398   ins_cost(110);
6399   format %{ "LEA    $dst,$mem" %}
6400   opcode(0x8D);
6401   ins_encode( OpcP, RegMem(dst,mem));
6402   ins_pipe( ialu_reg_reg_fat );
6403 %}
6404 
6405 instruct leaPIdxScaleOff(eRegP dst, indIndexScaleOffset mem) %{
6406   match(Set dst mem);
6407 
6408   ins_cost(110);
6409   format %{ "LEA    $dst,$mem" %}
6410   opcode(0x8D);
6411   ins_encode( OpcP, RegMem(dst,mem));
6412   ins_pipe( ialu_reg_reg_fat );
6413 %}
6414 
6415 // Load Constant
6416 instruct loadConI(eRegI dst, immI src) %{
6417   match(Set dst src);
6418 
6419   format %{ "MOV    $dst,$src" %}
6420   ins_encode( LdImmI(dst, src) );
6421   ins_pipe( ialu_reg_fat );
6422 %}
6423 
6424 // Load Constant zero
6425 instruct loadConI0(eRegI dst, immI0 src, eFlagsReg cr) %{
6426   match(Set dst src);
6427   effect(KILL cr);
6428 
6429   ins_cost(50);
6430   format %{ "XOR    $dst,$dst" %}
6431   opcode(0x33);  /* + rd */
6432   ins_encode( OpcP, RegReg( dst, dst ) );
6433   ins_pipe( ialu_reg );
6434 %}
6435 
6436 instruct loadConP(eRegP dst, immP src) %{
6437   match(Set dst src);
6438 
6439   format %{ "MOV    $dst,$src" %}
6440   opcode(0xB8);  /* + rd */
6441   ins_encode( LdImmP(dst, src) );
6442   ins_pipe( ialu_reg_fat );
6443 %}
6444 
6445 instruct loadConL(eRegL dst, immL src, eFlagsReg cr) %{
6446   match(Set dst src);
6447   effect(KILL cr);
6448   ins_cost(200);
6449   format %{ "MOV    $dst.lo,$src.lo\n\t"
6450             "MOV    $dst.hi,$src.hi" %}
6451   opcode(0xB8);
6452   ins_encode( LdImmL_Lo(dst, src), LdImmL_Hi(dst, src) );
6453   ins_pipe( ialu_reg_long_fat );
6454 %}
6455 
6456 instruct loadConL0(eRegL dst, immL0 src, eFlagsReg cr) %{
6457   match(Set dst src);
6458   effect(KILL cr);
6459   ins_cost(150);
6460   format %{ "XOR    $dst.lo,$dst.lo\n\t"
6461             "XOR    $dst.hi,$dst.hi" %}
6462   opcode(0x33,0x33);
6463   ins_encode( RegReg_Lo(dst,dst), RegReg_Hi(dst, dst) );
6464   ins_pipe( ialu_reg_long );
6465 %}
6466 
6467 // The instruction usage is guarded by predicate in operand immFPR().
6468 instruct loadConFPR(regFPR dst, immFPR con) %{
6469   match(Set dst con);
6470   ins_cost(125);
6471   format %{ "FLD_S  ST,[$constantaddress]\t# load from constant table: float=$con\n\t"
6472             "FSTP   $dst" %}
6473   ins_encode %{
6474     __ fld_s($constantaddress($con));
6475     __ fstp_d($dst$$reg);
6476   %}
6477   ins_pipe(fpu_reg_con);
6478 %}
6479 
6480 // The instruction usage is guarded by predicate in operand immFPR0().
6481 instruct loadConFPR0(regFPR dst, immFPR0 con) %{
6482   match(Set dst con);
6483   ins_cost(125);
6484   format %{ "FLDZ   ST\n\t"
6485             "FSTP   $dst" %}
6486   ins_encode %{
6487     __ fldz();
6488     __ fstp_d($dst$$reg);
6489   %}
6490   ins_pipe(fpu_reg_con);
6491 %}
6492 
6493 // The instruction usage is guarded by predicate in operand immFPR1().
6494 instruct loadConFPR1(regFPR dst, immFPR1 con) %{
6495   match(Set dst con);
6496   ins_cost(125);
6497   format %{ "FLD1   ST\n\t"
6498             "FSTP   $dst" %}
6499   ins_encode %{
6500     __ fld1();
6501     __ fstp_d($dst$$reg);
6502   %}
6503   ins_pipe(fpu_reg_con);
6504 %}
6505 
6506 // The instruction usage is guarded by predicate in operand immF().
6507 instruct loadConF(regF dst, immF con) %{
6508   match(Set dst con);
6509   ins_cost(125);
6510   format %{ "MOVSS  $dst,[$constantaddress]\t# load from constant table: float=$con" %}
6511   ins_encode %{
6512     __ movflt($dst$$XMMRegister, $constantaddress($con));
6513   %}
6514   ins_pipe(pipe_slow);
6515 %}
6516 
6517 // The instruction usage is guarded by predicate in operand immF0().
6518 instruct loadConF0(regF dst, immF0 src) %{
6519   match(Set dst src);
6520   ins_cost(100);
6521   format %{ "XORPS  $dst,$dst\t# float 0.0" %}
6522   ins_encode %{
6523     __ xorps($dst$$XMMRegister, $dst$$XMMRegister);
6524   %}
6525   ins_pipe(pipe_slow);
6526 %}
6527 
6528 // The instruction usage is guarded by predicate in operand immDPR().
6529 instruct loadConDPR(regDPR dst, immDPR con) %{
6530   match(Set dst con);
6531   ins_cost(125);
6532 
6533   format %{ "FLD_D  ST,[$constantaddress]\t# load from constant table: double=$con\n\t"
6534             "FSTP   $dst" %}
6535   ins_encode %{
6536     __ fld_d($constantaddress($con));
6537     __ fstp_d($dst$$reg);
6538   %}
6539   ins_pipe(fpu_reg_con);
6540 %}
6541 
6542 // The instruction usage is guarded by predicate in operand immDPR0().
6543 instruct loadConDPR0(regDPR dst, immDPR0 con) %{
6544   match(Set dst con);
6545   ins_cost(125);
6546 
6547   format %{ "FLDZ   ST\n\t"
6548             "FSTP   $dst" %}
6549   ins_encode %{
6550     __ fldz();
6551     __ fstp_d($dst$$reg);
6552   %}
6553   ins_pipe(fpu_reg_con);
6554 %}
6555 
6556 // The instruction usage is guarded by predicate in operand immDPR1().
6557 instruct loadConDPR1(regDPR dst, immDPR1 con) %{
6558   match(Set dst con);
6559   ins_cost(125);
6560 
6561   format %{ "FLD1   ST\n\t"
6562             "FSTP   $dst" %}
6563   ins_encode %{
6564     __ fld1();
6565     __ fstp_d($dst$$reg);
6566   %}
6567   ins_pipe(fpu_reg_con);
6568 %}
6569 
6570 // The instruction usage is guarded by predicate in operand immD().
6571 instruct loadConD(regD dst, immD con) %{
6572   match(Set dst con);
6573   ins_cost(125);
6574   format %{ "MOVSD  $dst,[$constantaddress]\t# load from constant table: double=$con" %}
6575   ins_encode %{
6576     __ movdbl($dst$$XMMRegister, $constantaddress($con));
6577   %}
6578   ins_pipe(pipe_slow);
6579 %}
6580 
6581 // The instruction usage is guarded by predicate in operand immD0().
6582 instruct loadConD0(regD dst, immD0 src) %{
6583   match(Set dst src);
6584   ins_cost(100);
6585   format %{ "XORPD  $dst,$dst\t# double 0.0" %}
6586   ins_encode %{
6587     __ xorpd ($dst$$XMMRegister, $dst$$XMMRegister);
6588   %}
6589   ins_pipe( pipe_slow );
6590 %}
6591 
6592 // Load Stack Slot
6593 instruct loadSSI(eRegI dst, stackSlotI src) %{
6594   match(Set dst src);
6595   ins_cost(125);
6596 
6597   format %{ "MOV    $dst,$src" %}
6598   opcode(0x8B);
6599   ins_encode( OpcP, RegMem(dst,src));
6600   ins_pipe( ialu_reg_mem );
6601 %}
6602 
6603 instruct loadSSL(eRegL dst, stackSlotL src) %{
6604   match(Set dst src);
6605 
6606   ins_cost(200);
6607   format %{ "MOV    $dst,$src.lo\n\t"
6608             "MOV    $dst+4,$src.hi" %}
6609   opcode(0x8B, 0x8B);
6610   ins_encode( OpcP, RegMem( dst, src ), OpcS, RegMem_Hi( dst, src ) );
6611   ins_pipe( ialu_mem_long_reg );
6612 %}
6613 
6614 // Load Stack Slot
6615 instruct loadSSP(eRegP dst, stackSlotP src) %{
6616   match(Set dst src);
6617   ins_cost(125);
6618 
6619   format %{ "MOV    $dst,$src" %}
6620   opcode(0x8B);
6621   ins_encode( OpcP, RegMem(dst,src));
6622   ins_pipe( ialu_reg_mem );
6623 %}
6624 
6625 // Load Stack Slot
6626 instruct loadSSF(regFPR dst, stackSlotF src) %{
6627   match(Set dst src);
6628   ins_cost(125);
6629 
6630   format %{ "FLD_S  $src\n\t"
6631             "FSTP   $dst" %}
6632   opcode(0xD9);               /* D9 /0, FLD m32real */
6633   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
6634               Pop_Reg_FPR(dst) );
6635   ins_pipe( fpu_reg_mem );
6636 %}
6637 
6638 // Load Stack Slot
6639 instruct loadSSD(regDPR dst, stackSlotD src) %{
6640   match(Set dst src);
6641   ins_cost(125);
6642 
6643   format %{ "FLD_D  $src\n\t"
6644             "FSTP   $dst" %}
6645   opcode(0xDD);               /* DD /0, FLD m64real */
6646   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
6647               Pop_Reg_DPR(dst) );
6648   ins_pipe( fpu_reg_mem );
6649 %}
6650 
6651 // Prefetch instructions.
6652 // Must be safe to execute with invalid address (cannot fault).
6653 
6654 instruct prefetchr0( memory mem ) %{
6655   predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
6656   match(PrefetchRead mem);
6657   ins_cost(0);
6658   size(0);
6659   format %{ "PREFETCHR (non-SSE is empty encoding)" %}
6660   ins_encode();
6661   ins_pipe(empty);
6662 %}
6663 
6664 instruct prefetchr( memory mem ) %{
6665   predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch() || ReadPrefetchInstr==3);
6666   match(PrefetchRead mem);
6667   ins_cost(100);
6668 
6669   format %{ "PREFETCHR $mem\t! Prefetch into level 1 cache for read" %}
6670   ins_encode %{
6671     __ prefetchr($mem$$Address);
6672   %}
6673   ins_pipe(ialu_mem);
6674 %}
6675 
6676 instruct prefetchrNTA( memory mem ) %{
6677   predicate(UseSSE>=1 && ReadPrefetchInstr==0);
6678   match(PrefetchRead mem);
6679   ins_cost(100);
6680 
6681   format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for read" %}
6682   ins_encode %{
6683     __ prefetchnta($mem$$Address);
6684   %}
6685   ins_pipe(ialu_mem);
6686 %}
6687 
6688 instruct prefetchrT0( memory mem ) %{
6689   predicate(UseSSE>=1 && ReadPrefetchInstr==1);
6690   match(PrefetchRead mem);
6691   ins_cost(100);
6692 
6693   format %{ "PREFETCHT0 $mem\t! Prefetch into L1 and L2 caches for read" %}
6694   ins_encode %{
6695     __ prefetcht0($mem$$Address);
6696   %}
6697   ins_pipe(ialu_mem);
6698 %}
6699 
6700 instruct prefetchrT2( memory mem ) %{
6701   predicate(UseSSE>=1 && ReadPrefetchInstr==2);
6702   match(PrefetchRead mem);
6703   ins_cost(100);
6704 
6705   format %{ "PREFETCHT2 $mem\t! Prefetch into L2 cache for read" %}
6706   ins_encode %{
6707     __ prefetcht2($mem$$Address);
6708   %}
6709   ins_pipe(ialu_mem);
6710 %}
6711 
6712 instruct prefetchw0( memory mem ) %{
6713   predicate(UseSSE==0 && !VM_Version::supports_3dnow_prefetch());
6714   match(PrefetchWrite mem);
6715   ins_cost(0);
6716   size(0);
6717   format %{ "Prefetch (non-SSE is empty encoding)" %}
6718   ins_encode();
6719   ins_pipe(empty);
6720 %}
6721 
6722 instruct prefetchw( memory mem ) %{
6723   predicate(UseSSE==0 && VM_Version::supports_3dnow_prefetch());
6724   match( PrefetchWrite mem );
6725   ins_cost(100);
6726 
6727   format %{ "PREFETCHW $mem\t! Prefetch into L1 cache and mark modified" %}
6728   ins_encode %{
6729     __ prefetchw($mem$$Address);
6730   %}
6731   ins_pipe(ialu_mem);
6732 %}
6733 
6734 instruct prefetchwNTA( memory mem ) %{
6735   predicate(UseSSE>=1);
6736   match(PrefetchWrite mem);
6737   ins_cost(100);
6738 
6739   format %{ "PREFETCHNTA $mem\t! Prefetch into non-temporal cache for write" %}
6740   ins_encode %{
6741     __ prefetchnta($mem$$Address);
6742   %}
6743   ins_pipe(ialu_mem);
6744 %}
6745 
6746 // Prefetch instructions for allocation.
6747 
6748 instruct prefetchAlloc0( memory mem ) %{
6749   predicate(UseSSE==0 && AllocatePrefetchInstr!=3);
6750   match(PrefetchAllocation mem);
6751   ins_cost(0);
6752   size(0);
6753   format %{ "Prefetch allocation (non-SSE is empty encoding)" %}
6754   ins_encode();
6755   ins_pipe(empty);
6756 %}
6757 
6758 instruct prefetchAlloc( memory mem ) %{
6759   predicate(AllocatePrefetchInstr==3);
6760   match( PrefetchAllocation mem );
6761   ins_cost(100);
6762 
6763   format %{ "PREFETCHW $mem\t! Prefetch allocation into L1 cache and mark modified" %}
6764   ins_encode %{
6765     __ prefetchw($mem$$Address);
6766   %}
6767   ins_pipe(ialu_mem);
6768 %}
6769 
6770 instruct prefetchAllocNTA( memory mem ) %{
6771   predicate(UseSSE>=1 && AllocatePrefetchInstr==0);
6772   match(PrefetchAllocation mem);
6773   ins_cost(100);
6774 
6775   format %{ "PREFETCHNTA $mem\t! Prefetch allocation into non-temporal cache for write" %}
6776   ins_encode %{
6777     __ prefetchnta($mem$$Address);
6778   %}
6779   ins_pipe(ialu_mem);
6780 %}
6781 
6782 instruct prefetchAllocT0( memory mem ) %{
6783   predicate(UseSSE>=1 && AllocatePrefetchInstr==1);
6784   match(PrefetchAllocation mem);
6785   ins_cost(100);
6786 
6787   format %{ "PREFETCHT0 $mem\t! Prefetch allocation into L1 and L2 caches for write" %}
6788   ins_encode %{
6789     __ prefetcht0($mem$$Address);
6790   %}
6791   ins_pipe(ialu_mem);
6792 %}
6793 
6794 instruct prefetchAllocT2( memory mem ) %{
6795   predicate(UseSSE>=1 && AllocatePrefetchInstr==2);
6796   match(PrefetchAllocation mem);
6797   ins_cost(100);
6798 
6799   format %{ "PREFETCHT2 $mem\t! Prefetch allocation into L2 cache for write" %}
6800   ins_encode %{
6801     __ prefetcht2($mem$$Address);
6802   %}
6803   ins_pipe(ialu_mem);
6804 %}
6805 
6806 //----------Store Instructions-------------------------------------------------
6807 
6808 // Store Byte
6809 instruct storeB(memory mem, xRegI src) %{
6810   match(Set mem (StoreB mem src));
6811 
6812   ins_cost(125);
6813   format %{ "MOV8   $mem,$src" %}
6814   opcode(0x88);
6815   ins_encode( OpcP, RegMem( src, mem ) );
6816   ins_pipe( ialu_mem_reg );
6817 %}
6818 
6819 // Store Char/Short
6820 instruct storeC(memory mem, eRegI src) %{
6821   match(Set mem (StoreC mem src));
6822 
6823   ins_cost(125);
6824   format %{ "MOV16  $mem,$src" %}
6825   opcode(0x89, 0x66);
6826   ins_encode( OpcS, OpcP, RegMem( src, mem ) );
6827   ins_pipe( ialu_mem_reg );
6828 %}
6829 
6830 // Store Integer
6831 instruct storeI(memory mem, eRegI src) %{
6832   match(Set mem (StoreI mem src));
6833 
6834   ins_cost(125);
6835   format %{ "MOV    $mem,$src" %}
6836   opcode(0x89);
6837   ins_encode( OpcP, RegMem( src, mem ) );
6838   ins_pipe( ialu_mem_reg );
6839 %}
6840 
6841 // Store Long
6842 instruct storeL(long_memory mem, eRegL src) %{
6843   predicate(!((StoreLNode*)n)->require_atomic_access());
6844   match(Set mem (StoreL mem src));
6845 
6846   ins_cost(200);
6847   format %{ "MOV    $mem,$src.lo\n\t"
6848             "MOV    $mem+4,$src.hi" %}
6849   opcode(0x89, 0x89);
6850   ins_encode( OpcP, RegMem( src, mem ), OpcS, RegMem_Hi( src, mem ) );
6851   ins_pipe( ialu_mem_long_reg );
6852 %}
6853 
6854 // Store Long to Integer
6855 instruct storeL2I(memory mem, eRegL src) %{
6856   match(Set mem (StoreI mem (ConvL2I src)));
6857 
6858   format %{ "MOV    $mem,$src.lo\t# long -> int" %}
6859   ins_encode %{
6860     __ movl($mem$$Address, $src$$Register);
6861   %}
6862   ins_pipe(ialu_mem_reg);
6863 %}
6864 
6865 // Volatile Store Long.  Must be atomic, so move it into
6866 // the FP TOS and then do a 64-bit FIST.  Has to probe the
6867 // target address before the store (for null-ptr checks)
6868 // so the memory operand is used twice in the encoding.
6869 instruct storeL_volatile(memory mem, stackSlotL src, eFlagsReg cr ) %{
6870   predicate(UseSSE<=1 && ((StoreLNode*)n)->require_atomic_access());
6871   match(Set mem (StoreL mem src));
6872   effect( KILL cr );
6873   ins_cost(400);
6874   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6875             "FILD   $src\n\t"
6876             "FISTp  $mem\t # 64-bit atomic volatile long store" %}
6877   opcode(0x3B);
6878   ins_encode( OpcP, RegMem( EAX, mem ), enc_storeL_volatile(mem,src));
6879   ins_pipe( fpu_reg_mem );
6880 %}
6881 
6882 instruct storeLX_volatile(memory mem, stackSlotL src, regD tmp, eFlagsReg cr) %{
6883   predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
6884   match(Set mem (StoreL mem src));
6885   effect( TEMP tmp, KILL cr );
6886   ins_cost(380);
6887   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6888             "MOVSD  $tmp,$src\n\t"
6889             "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
6890   ins_encode %{
6891     __ cmpl(rax, $mem$$Address);
6892     __ movdbl($tmp$$XMMRegister, Address(rsp, $src$$disp));
6893     __ movdbl($mem$$Address, $tmp$$XMMRegister);
6894   %}
6895   ins_pipe( pipe_slow );
6896 %}
6897 
6898 instruct storeLX_reg_volatile(memory mem, eRegL src, regD tmp2, regD tmp, eFlagsReg cr) %{
6899   predicate(UseSSE>=2 && ((StoreLNode*)n)->require_atomic_access());
6900   match(Set mem (StoreL mem src));
6901   effect( TEMP tmp2 , TEMP tmp, KILL cr );
6902   ins_cost(360);
6903   format %{ "CMP    $mem,EAX\t# Probe address for implicit null check\n\t"
6904             "MOVD   $tmp,$src.lo\n\t"
6905             "MOVD   $tmp2,$src.hi\n\t"
6906             "PUNPCKLDQ $tmp,$tmp2\n\t"
6907             "MOVSD  $mem,$tmp\t # 64-bit atomic volatile long store" %}
6908   ins_encode %{
6909     __ cmpl(rax, $mem$$Address);
6910     __ movdl($tmp$$XMMRegister, $src$$Register);
6911     __ movdl($tmp2$$XMMRegister, HIGH_FROM_LOW($src$$Register));
6912     __ punpckldq($tmp$$XMMRegister, $tmp2$$XMMRegister);
6913     __ movdbl($mem$$Address, $tmp$$XMMRegister);
6914   %}
6915   ins_pipe( pipe_slow );
6916 %}
6917 
6918 // Store Pointer; for storing unknown oops and raw pointers
6919 instruct storeP(memory mem, anyRegP src) %{
6920   match(Set mem (StoreP mem src));
6921 
6922   ins_cost(125);
6923   format %{ "MOV    $mem,$src" %}
6924   opcode(0x89);
6925   ins_encode( OpcP, RegMem( src, mem ) );
6926   ins_pipe( ialu_mem_reg );
6927 %}
6928 
6929 // Store Integer Immediate
6930 instruct storeImmI(memory mem, immI src) %{
6931   match(Set mem (StoreI mem src));
6932 
6933   ins_cost(150);
6934   format %{ "MOV    $mem,$src" %}
6935   opcode(0xC7);               /* C7 /0 */
6936   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
6937   ins_pipe( ialu_mem_imm );
6938 %}
6939 
6940 // Store Short/Char Immediate
6941 instruct storeImmI16(memory mem, immI16 src) %{
6942   predicate(UseStoreImmI16);
6943   match(Set mem (StoreC mem src));
6944 
6945   ins_cost(150);
6946   format %{ "MOV16  $mem,$src" %}
6947   opcode(0xC7);     /* C7 /0 Same as 32 store immediate with prefix */
6948   ins_encode( SizePrefix, OpcP, RMopc_Mem(0x00,mem),  Con16( src ));
6949   ins_pipe( ialu_mem_imm );
6950 %}
6951 
6952 // Store Pointer Immediate; null pointers or constant oops that do not
6953 // need card-mark barriers.
6954 instruct storeImmP(memory mem, immP src) %{
6955   match(Set mem (StoreP mem src));
6956 
6957   ins_cost(150);
6958   format %{ "MOV    $mem,$src" %}
6959   opcode(0xC7);               /* C7 /0 */
6960   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32( src ));
6961   ins_pipe( ialu_mem_imm );
6962 %}
6963 
6964 // Store Byte Immediate
6965 instruct storeImmB(memory mem, immI8 src) %{
6966   match(Set mem (StoreB mem src));
6967 
6968   ins_cost(150);
6969   format %{ "MOV8   $mem,$src" %}
6970   opcode(0xC6);               /* C6 /0 */
6971   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
6972   ins_pipe( ialu_mem_imm );
6973 %}
6974 
6975 // Store Aligned Packed Byte XMM register to memory
6976 instruct storeA8B(memory mem, regD src) %{
6977   predicate(UseSSE>=1);
6978   match(Set mem (Store8B mem src));
6979   ins_cost(145);
6980   format %{ "MOVQ  $mem,$src\t! packed8B" %}
6981   ins_encode %{
6982     __ movq($mem$$Address, $src$$XMMRegister);
6983   %}
6984   ins_pipe( pipe_slow );
6985 %}
6986 
6987 // Store Aligned Packed Char/Short XMM register to memory
6988 instruct storeA4C(memory mem, regD src) %{
6989   predicate(UseSSE>=1);
6990   match(Set mem (Store4C mem src));
6991   ins_cost(145);
6992   format %{ "MOVQ  $mem,$src\t! packed4C" %}
6993   ins_encode %{
6994     __ movq($mem$$Address, $src$$XMMRegister);
6995   %}
6996   ins_pipe( pipe_slow );
6997 %}
6998 
6999 // Store Aligned Packed Integer XMM register to memory
7000 instruct storeA2I(memory mem, regD src) %{
7001   predicate(UseSSE>=1);
7002   match(Set mem (Store2I mem src));
7003   ins_cost(145);
7004   format %{ "MOVQ  $mem,$src\t! packed2I" %}
7005   ins_encode %{
7006     __ movq($mem$$Address, $src$$XMMRegister);
7007   %}
7008   ins_pipe( pipe_slow );
7009 %}
7010 
7011 // Store CMS card-mark Immediate
7012 instruct storeImmCM(memory mem, immI8 src) %{
7013   match(Set mem (StoreCM mem src));
7014 
7015   ins_cost(150);
7016   format %{ "MOV8   $mem,$src\t! CMS card-mark imm0" %}
7017   opcode(0xC6);               /* C6 /0 */
7018   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con8or32( src ));
7019   ins_pipe( ialu_mem_imm );
7020 %}
7021 
7022 // Store Double
7023 instruct storeDPR( memory mem, regDPR1 src) %{
7024   predicate(UseSSE<=1);
7025   match(Set mem (StoreD mem src));
7026 
7027   ins_cost(100);
7028   format %{ "FST_D  $mem,$src" %}
7029   opcode(0xDD);       /* DD /2 */
7030   ins_encode( enc_FPR_store(mem,src) );
7031   ins_pipe( fpu_mem_reg );
7032 %}
7033 
7034 // Store double does rounding on x86
7035 instruct storeDPR_rounded( memory mem, regDPR1 src) %{
7036   predicate(UseSSE<=1);
7037   match(Set mem (StoreD mem (RoundDouble src)));
7038 
7039   ins_cost(100);
7040   format %{ "FST_D  $mem,$src\t# round" %}
7041   opcode(0xDD);       /* DD /2 */
7042   ins_encode( enc_FPR_store(mem,src) );
7043   ins_pipe( fpu_mem_reg );
7044 %}
7045 
7046 // Store XMM register to memory (double-precision floating points)
7047 // MOVSD instruction
7048 instruct storeD(memory mem, regD src) %{
7049   predicate(UseSSE>=2);
7050   match(Set mem (StoreD mem src));
7051   ins_cost(95);
7052   format %{ "MOVSD  $mem,$src" %}
7053   ins_encode %{
7054     __ movdbl($mem$$Address, $src$$XMMRegister);
7055   %}
7056   ins_pipe( pipe_slow );
7057 %}
7058 
7059 // Store XMM register to memory (single-precision floating point)
7060 // MOVSS instruction
7061 instruct storeF(memory mem, regF src) %{
7062   predicate(UseSSE>=1);
7063   match(Set mem (StoreF mem src));
7064   ins_cost(95);
7065   format %{ "MOVSS  $mem,$src" %}
7066   ins_encode %{
7067     __ movflt($mem$$Address, $src$$XMMRegister);
7068   %}
7069   ins_pipe( pipe_slow );
7070 %}
7071 
7072 // Store Aligned Packed Single Float XMM register to memory
7073 instruct storeA2F(memory mem, regD src) %{
7074   predicate(UseSSE>=1);
7075   match(Set mem (Store2F mem src));
7076   ins_cost(145);
7077   format %{ "MOVQ  $mem,$src\t! packed2F" %}
7078   ins_encode %{
7079     __ movq($mem$$Address, $src$$XMMRegister);
7080   %}
7081   ins_pipe( pipe_slow );
7082 %}
7083 
7084 // Store Float
7085 instruct storeFPR( memory mem, regFPR1 src) %{
7086   predicate(UseSSE==0);
7087   match(Set mem (StoreF mem src));
7088 
7089   ins_cost(100);
7090   format %{ "FST_S  $mem,$src" %}
7091   opcode(0xD9);       /* D9 /2 */
7092   ins_encode( enc_FPR_store(mem,src) );
7093   ins_pipe( fpu_mem_reg );
7094 %}
7095 
7096 // Store Float does rounding on x86
7097 instruct storeFPR_rounded( memory mem, regFPR1 src) %{
7098   predicate(UseSSE==0);
7099   match(Set mem (StoreF mem (RoundFloat src)));
7100 
7101   ins_cost(100);
7102   format %{ "FST_S  $mem,$src\t# round" %}
7103   opcode(0xD9);       /* D9 /2 */
7104   ins_encode( enc_FPR_store(mem,src) );
7105   ins_pipe( fpu_mem_reg );
7106 %}
7107 
7108 // Store Float does rounding on x86
7109 instruct storeFPR_Drounded( memory mem, regDPR1 src) %{
7110   predicate(UseSSE<=1);
7111   match(Set mem (StoreF mem (ConvD2F src)));
7112 
7113   ins_cost(100);
7114   format %{ "FST_S  $mem,$src\t# D-round" %}
7115   opcode(0xD9);       /* D9 /2 */
7116   ins_encode( enc_FPR_store(mem,src) );
7117   ins_pipe( fpu_mem_reg );
7118 %}
7119 
7120 // Store immediate Float value (it is faster than store from FPU register)
7121 // The instruction usage is guarded by predicate in operand immFPR().
7122 instruct storeFPR_imm( memory mem, immFPR src) %{
7123   match(Set mem (StoreF mem src));
7124 
7125   ins_cost(50);
7126   format %{ "MOV    $mem,$src\t# store float" %}
7127   opcode(0xC7);               /* C7 /0 */
7128   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32FPR_as_bits( src ));
7129   ins_pipe( ialu_mem_imm );
7130 %}
7131 
7132 // Store immediate Float value (it is faster than store from XMM register)
7133 // The instruction usage is guarded by predicate in operand immF().
7134 instruct storeF_imm( memory mem, immF src) %{
7135   match(Set mem (StoreF mem src));
7136 
7137   ins_cost(50);
7138   format %{ "MOV    $mem,$src\t# store float" %}
7139   opcode(0xC7);               /* C7 /0 */
7140   ins_encode( OpcP, RMopc_Mem(0x00,mem),  Con32F_as_bits( src ));
7141   ins_pipe( ialu_mem_imm );
7142 %}
7143 
7144 // Store Integer to stack slot
7145 instruct storeSSI(stackSlotI dst, eRegI src) %{
7146   match(Set dst src);
7147 
7148   ins_cost(100);
7149   format %{ "MOV    $dst,$src" %}
7150   opcode(0x89);
7151   ins_encode( OpcPRegSS( dst, src ) );
7152   ins_pipe( ialu_mem_reg );
7153 %}
7154 
7155 // Store Integer to stack slot
7156 instruct storeSSP(stackSlotP dst, eRegP src) %{
7157   match(Set dst src);
7158 
7159   ins_cost(100);
7160   format %{ "MOV    $dst,$src" %}
7161   opcode(0x89);
7162   ins_encode( OpcPRegSS( dst, src ) );
7163   ins_pipe( ialu_mem_reg );
7164 %}
7165 
7166 // Store Long to stack slot
7167 instruct storeSSL(stackSlotL dst, eRegL src) %{
7168   match(Set dst src);
7169 
7170   ins_cost(200);
7171   format %{ "MOV    $dst,$src.lo\n\t"
7172             "MOV    $dst+4,$src.hi" %}
7173   opcode(0x89, 0x89);
7174   ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
7175   ins_pipe( ialu_mem_long_reg );
7176 %}
7177 
7178 //----------MemBar Instructions-----------------------------------------------
7179 // Memory barrier flavors
7180 
7181 instruct membar_acquire() %{
7182   match(MemBarAcquire);
7183   ins_cost(400);
7184 
7185   size(0);
7186   format %{ "MEMBAR-acquire ! (empty encoding)" %}
7187   ins_encode();
7188   ins_pipe(empty);
7189 %}
7190 
7191 instruct membar_acquire_lock() %{
7192   match(MemBarAcquireLock);
7193   ins_cost(0);
7194 
7195   size(0);
7196   format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7197   ins_encode( );
7198   ins_pipe(empty);
7199 %}
7200 
7201 instruct membar_release() %{
7202   match(MemBarRelease);
7203   ins_cost(400);
7204 
7205   size(0);
7206   format %{ "MEMBAR-release ! (empty encoding)" %}
7207   ins_encode( );
7208   ins_pipe(empty);
7209 %}
7210 
7211 instruct membar_release_lock() %{
7212   match(MemBarReleaseLock);
7213   ins_cost(0);
7214 
7215   size(0);
7216   format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7217   ins_encode( );
7218   ins_pipe(empty);
7219 %}
7220 
7221 instruct membar_volatile(eFlagsReg cr) %{
7222   match(MemBarVolatile);
7223   effect(KILL cr);
7224   ins_cost(400);
7225 
7226   format %{ 
7227     $$template
7228     if (os::is_MP()) {
7229       $$emit$$"LOCK ADDL [ESP + #0], 0\t! membar_volatile"
7230     } else {
7231       $$emit$$"MEMBAR-volatile ! (empty encoding)"
7232     }
7233   %}
7234   ins_encode %{
7235     __ membar(Assembler::StoreLoad);
7236   %}
7237   ins_pipe(pipe_slow);
7238 %}
7239 
7240 instruct unnecessary_membar_volatile() %{
7241   match(MemBarVolatile);
7242   predicate(Matcher::post_store_load_barrier(n));
7243   ins_cost(0);
7244 
7245   size(0);
7246   format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7247   ins_encode( );
7248   ins_pipe(empty);
7249 %}
7250 
7251 instruct membar_storestore() %{
7252   match(MemBarStoreStore);
7253   ins_cost(0);
7254 
7255   size(0);
7256   format %{ "MEMBAR-storestore (empty encoding)" %}
7257   ins_encode( );
7258   ins_pipe(empty);
7259 %}
7260 
7261 //----------Move Instructions--------------------------------------------------
7262 instruct castX2P(eAXRegP dst, eAXRegI src) %{
7263   match(Set dst (CastX2P src));
7264   format %{ "# X2P  $dst, $src" %}
7265   ins_encode( /*empty encoding*/ );
7266   ins_cost(0);
7267   ins_pipe(empty);
7268 %}
7269 
7270 instruct castP2X(eRegI dst, eRegP src ) %{
7271   match(Set dst (CastP2X src));
7272   ins_cost(50);
7273   format %{ "MOV    $dst, $src\t# CastP2X" %}
7274   ins_encode( enc_Copy( dst, src) );
7275   ins_pipe( ialu_reg_reg );
7276 %}
7277 
7278 //----------Conditional Move---------------------------------------------------
7279 // Conditional move
7280 instruct jmovI_reg(cmpOp cop, eFlagsReg cr, eRegI dst, eRegI src) %{
7281   predicate(!VM_Version::supports_cmov() );
7282   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7283   ins_cost(200);
7284   format %{ "J$cop,us skip\t# signed cmove\n\t"
7285             "MOV    $dst,$src\n"
7286       "skip:" %}
7287   ins_encode %{
7288     Label Lskip;
7289     // Invert sense of branch from sense of CMOV
7290     __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
7291     __ movl($dst$$Register, $src$$Register);
7292     __ bind(Lskip);
7293   %}
7294   ins_pipe( pipe_cmov_reg );
7295 %}
7296 
7297 instruct jmovI_regU(cmpOpU cop, eFlagsRegU cr, eRegI dst, eRegI src) %{
7298   predicate(!VM_Version::supports_cmov() );
7299   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7300   ins_cost(200);
7301   format %{ "J$cop,us skip\t# unsigned cmove\n\t"
7302             "MOV    $dst,$src\n"
7303       "skip:" %}
7304   ins_encode %{
7305     Label Lskip;
7306     // Invert sense of branch from sense of CMOV
7307     __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
7308     __ movl($dst$$Register, $src$$Register);
7309     __ bind(Lskip);
7310   %}
7311   ins_pipe( pipe_cmov_reg );
7312 %}
7313 
7314 instruct cmovI_reg(eRegI dst, eRegI src, eFlagsReg cr, cmpOp cop ) %{
7315   predicate(VM_Version::supports_cmov() );
7316   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7317   ins_cost(200);
7318   format %{ "CMOV$cop $dst,$src" %}
7319   opcode(0x0F,0x40);
7320   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7321   ins_pipe( pipe_cmov_reg );
7322 %}
7323 
7324 instruct cmovI_regU( cmpOpU cop, eFlagsRegU cr, eRegI dst, eRegI src ) %{
7325   predicate(VM_Version::supports_cmov() );
7326   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7327   ins_cost(200);
7328   format %{ "CMOV$cop $dst,$src" %}
7329   opcode(0x0F,0x40);
7330   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7331   ins_pipe( pipe_cmov_reg );
7332 %}
7333 
7334 instruct cmovI_regUCF( cmpOpUCF cop, eFlagsRegUCF cr, eRegI dst, eRegI src ) %{
7335   predicate(VM_Version::supports_cmov() );
7336   match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7337   ins_cost(200);
7338   expand %{
7339     cmovI_regU(cop, cr, dst, src);
7340   %}
7341 %}
7342 
7343 // Conditional move
7344 instruct cmovI_mem(cmpOp cop, eFlagsReg cr, eRegI dst, memory src) %{
7345   predicate(VM_Version::supports_cmov() );
7346   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7347   ins_cost(250);
7348   format %{ "CMOV$cop $dst,$src" %}
7349   opcode(0x0F,0x40);
7350   ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7351   ins_pipe( pipe_cmov_mem );
7352 %}
7353 
7354 // Conditional move
7355 instruct cmovI_memU(cmpOpU cop, eFlagsRegU cr, eRegI dst, memory src) %{
7356   predicate(VM_Version::supports_cmov() );
7357   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7358   ins_cost(250);
7359   format %{ "CMOV$cop $dst,$src" %}
7360   opcode(0x0F,0x40);
7361   ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7362   ins_pipe( pipe_cmov_mem );
7363 %}
7364 
7365 instruct cmovI_memUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegI dst, memory src) %{
7366   predicate(VM_Version::supports_cmov() );
7367   match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7368   ins_cost(250);
7369   expand %{
7370     cmovI_memU(cop, cr, dst, src);
7371   %}
7372 %}
7373 
7374 // Conditional move
7375 instruct cmovP_reg(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
7376   predicate(VM_Version::supports_cmov() );
7377   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7378   ins_cost(200);
7379   format %{ "CMOV$cop $dst,$src\t# ptr" %}
7380   opcode(0x0F,0x40);
7381   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7382   ins_pipe( pipe_cmov_reg );
7383 %}
7384 
7385 // Conditional move (non-P6 version)
7386 // Note:  a CMoveP is generated for  stubs and native wrappers
7387 //        regardless of whether we are on a P6, so we
7388 //        emulate a cmov here
7389 instruct cmovP_reg_nonP6(eRegP dst, eRegP src, eFlagsReg cr, cmpOp cop ) %{
7390   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7391   ins_cost(300);
7392   format %{ "Jn$cop   skip\n\t"
7393           "MOV    $dst,$src\t# pointer\n"
7394       "skip:" %}
7395   opcode(0x8b);
7396   ins_encode( enc_cmov_branch(cop, 0x2), OpcP, RegReg(dst, src));
7397   ins_pipe( pipe_cmov_reg );
7398 %}
7399 
7400 // Conditional move
7401 instruct cmovP_regU(cmpOpU cop, eFlagsRegU cr, eRegP dst, eRegP src ) %{
7402   predicate(VM_Version::supports_cmov() );
7403   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7404   ins_cost(200);
7405   format %{ "CMOV$cop $dst,$src\t# ptr" %}
7406   opcode(0x0F,0x40);
7407   ins_encode( enc_cmov(cop), RegReg( dst, src ) );
7408   ins_pipe( pipe_cmov_reg );
7409 %}
7410 
7411 instruct cmovP_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegP dst, eRegP src ) %{
7412   predicate(VM_Version::supports_cmov() );
7413   match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7414   ins_cost(200);
7415   expand %{
7416     cmovP_regU(cop, cr, dst, src);
7417   %}
7418 %}
7419 
7420 // DISABLED: Requires the ADLC to emit a bottom_type call that
7421 // correctly meets the two pointer arguments; one is an incoming
7422 // register but the other is a memory operand.  ALSO appears to
7423 // be buggy with implicit null checks.
7424 //
7425 //// Conditional move
7426 //instruct cmovP_mem(cmpOp cop, eFlagsReg cr, eRegP dst, memory src) %{
7427 //  predicate(VM_Version::supports_cmov() );
7428 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7429 //  ins_cost(250);
7430 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7431 //  opcode(0x0F,0x40);
7432 //  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7433 //  ins_pipe( pipe_cmov_mem );
7434 //%}
7435 //
7436 //// Conditional move
7437 //instruct cmovP_memU(cmpOpU cop, eFlagsRegU cr, eRegP dst, memory src) %{
7438 //  predicate(VM_Version::supports_cmov() );
7439 //  match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7440 //  ins_cost(250);
7441 //  format %{ "CMOV$cop $dst,$src\t# ptr" %}
7442 //  opcode(0x0F,0x40);
7443 //  ins_encode( enc_cmov(cop), RegMem( dst, src ) );
7444 //  ins_pipe( pipe_cmov_mem );
7445 //%}
7446 
7447 // Conditional move
7448 instruct fcmovDPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regDPR1 dst, regDPR src) %{
7449   predicate(UseSSE<=1);
7450   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7451   ins_cost(200);
7452   format %{ "FCMOV$cop $dst,$src\t# double" %}
7453   opcode(0xDA);
7454   ins_encode( enc_cmov_dpr(cop,src) );
7455   ins_pipe( pipe_cmovDPR_reg );
7456 %}
7457 
7458 // Conditional move
7459 instruct fcmovFPR_regU(cmpOp_fcmov cop, eFlagsRegU cr, regFPR1 dst, regFPR src) %{
7460   predicate(UseSSE==0);
7461   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7462   ins_cost(200);
7463   format %{ "FCMOV$cop $dst,$src\t# float" %}
7464   opcode(0xDA);
7465   ins_encode( enc_cmov_dpr(cop,src) );
7466   ins_pipe( pipe_cmovDPR_reg );
7467 %}
7468 
7469 // Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
7470 instruct fcmovDPR_regS(cmpOp cop, eFlagsReg cr, regDPR dst, regDPR src) %{
7471   predicate(UseSSE<=1);
7472   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7473   ins_cost(200);
7474   format %{ "Jn$cop   skip\n\t"
7475             "MOV    $dst,$src\t# double\n"
7476       "skip:" %}
7477   opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
7478   ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_DPR(src), OpcP, RegOpc(dst) );
7479   ins_pipe( pipe_cmovDPR_reg );
7480 %}
7481 
7482 // Float CMOV on Intel doesn't handle *signed* compares, only unsigned.
7483 instruct fcmovFPR_regS(cmpOp cop, eFlagsReg cr, regFPR dst, regFPR src) %{
7484   predicate(UseSSE==0);
7485   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7486   ins_cost(200);
7487   format %{ "Jn$cop    skip\n\t"
7488             "MOV    $dst,$src\t# float\n"
7489       "skip:" %}
7490   opcode (0xdd, 0x3);     /* DD D8+i or DD /3 */
7491   ins_encode( enc_cmov_branch( cop, 0x4 ), Push_Reg_FPR(src), OpcP, RegOpc(dst) );
7492   ins_pipe( pipe_cmovDPR_reg );
7493 %}
7494 
7495 // No CMOVE with SSE/SSE2
7496 instruct fcmovF_regS(cmpOp cop, eFlagsReg cr, regF dst, regF src) %{
7497   predicate (UseSSE>=1);
7498   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7499   ins_cost(200);
7500   format %{ "Jn$cop   skip\n\t"
7501             "MOVSS  $dst,$src\t# float\n"
7502       "skip:" %}
7503   ins_encode %{
7504     Label skip;
7505     // Invert sense of branch from sense of CMOV
7506     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7507     __ movflt($dst$$XMMRegister, $src$$XMMRegister);
7508     __ bind(skip);
7509   %}
7510   ins_pipe( pipe_slow );
7511 %}
7512 
7513 // No CMOVE with SSE/SSE2
7514 instruct fcmovD_regS(cmpOp cop, eFlagsReg cr, regD dst, regD src) %{
7515   predicate (UseSSE>=2);
7516   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7517   ins_cost(200);
7518   format %{ "Jn$cop   skip\n\t"
7519             "MOVSD  $dst,$src\t# float\n"
7520       "skip:" %}
7521   ins_encode %{
7522     Label skip;
7523     // Invert sense of branch from sense of CMOV
7524     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7525     __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
7526     __ bind(skip);
7527   %}
7528   ins_pipe( pipe_slow );
7529 %}
7530 
7531 // unsigned version
7532 instruct fcmovF_regU(cmpOpU cop, eFlagsRegU cr, regF dst, regF src) %{
7533   predicate (UseSSE>=1);
7534   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7535   ins_cost(200);
7536   format %{ "Jn$cop   skip\n\t"
7537             "MOVSS  $dst,$src\t# float\n"
7538       "skip:" %}
7539   ins_encode %{
7540     Label skip;
7541     // Invert sense of branch from sense of CMOV
7542     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7543     __ movflt($dst$$XMMRegister, $src$$XMMRegister);
7544     __ bind(skip);
7545   %}
7546   ins_pipe( pipe_slow );
7547 %}
7548 
7549 instruct fcmovF_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regF dst, regF src) %{
7550   predicate (UseSSE>=1);
7551   match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7552   ins_cost(200);
7553   expand %{
7554     fcmovF_regU(cop, cr, dst, src);
7555   %}
7556 %}
7557 
7558 // unsigned version
7559 instruct fcmovD_regU(cmpOpU cop, eFlagsRegU cr, regD dst, regD src) %{
7560   predicate (UseSSE>=2);
7561   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7562   ins_cost(200);
7563   format %{ "Jn$cop   skip\n\t"
7564             "MOVSD  $dst,$src\t# float\n"
7565       "skip:" %}
7566   ins_encode %{
7567     Label skip;
7568     // Invert sense of branch from sense of CMOV
7569     __ jccb((Assembler::Condition)($cop$$cmpcode^1), skip);
7570     __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
7571     __ bind(skip);
7572   %}
7573   ins_pipe( pipe_slow );
7574 %}
7575 
7576 instruct fcmovD_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, regD dst, regD src) %{
7577   predicate (UseSSE>=2);
7578   match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7579   ins_cost(200);
7580   expand %{
7581     fcmovD_regU(cop, cr, dst, src);
7582   %}
7583 %}
7584 
7585 instruct cmovL_reg(cmpOp cop, eFlagsReg cr, eRegL dst, eRegL src) %{
7586   predicate(VM_Version::supports_cmov() );
7587   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7588   ins_cost(200);
7589   format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
7590             "CMOV$cop $dst.hi,$src.hi" %}
7591   opcode(0x0F,0x40);
7592   ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
7593   ins_pipe( pipe_cmov_reg_long );
7594 %}
7595 
7596 instruct cmovL_regU(cmpOpU cop, eFlagsRegU cr, eRegL dst, eRegL src) %{
7597   predicate(VM_Version::supports_cmov() );
7598   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7599   ins_cost(200);
7600   format %{ "CMOV$cop $dst.lo,$src.lo\n\t"
7601             "CMOV$cop $dst.hi,$src.hi" %}
7602   opcode(0x0F,0x40);
7603   ins_encode( enc_cmov(cop), RegReg_Lo2( dst, src ), enc_cmov(cop), RegReg_Hi2( dst, src ) );
7604   ins_pipe( pipe_cmov_reg_long );
7605 %}
7606 
7607 instruct cmovL_regUCF(cmpOpUCF cop, eFlagsRegUCF cr, eRegL dst, eRegL src) %{
7608   predicate(VM_Version::supports_cmov() );
7609   match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7610   ins_cost(200);
7611   expand %{
7612     cmovL_regU(cop, cr, dst, src);
7613   %}
7614 %}
7615 
7616 //----------Arithmetic Instructions--------------------------------------------
7617 //----------Addition Instructions----------------------------------------------
7618 // Integer Addition Instructions
7619 instruct addI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
7620   match(Set dst (AddI dst src));
7621   effect(KILL cr);
7622 
7623   size(2);
7624   format %{ "ADD    $dst,$src" %}
7625   opcode(0x03);
7626   ins_encode( OpcP, RegReg( dst, src) );
7627   ins_pipe( ialu_reg_reg );
7628 %}
7629 
7630 instruct addI_eReg_imm(eRegI dst, immI src, eFlagsReg cr) %{
7631   match(Set dst (AddI dst src));
7632   effect(KILL cr);
7633 
7634   format %{ "ADD    $dst,$src" %}
7635   opcode(0x81, 0x00); /* /0 id */
7636   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7637   ins_pipe( ialu_reg );
7638 %}
7639 
7640 instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
7641   predicate(UseIncDec);
7642   match(Set dst (AddI dst src));
7643   effect(KILL cr);
7644 
7645   size(1);
7646   format %{ "INC    $dst" %}
7647   opcode(0x40); /*  */
7648   ins_encode( Opc_plus( primary, dst ) );
7649   ins_pipe( ialu_reg );
7650 %}
7651 
7652 instruct leaI_eReg_immI(eRegI dst, eRegI src0, immI src1) %{
7653   match(Set dst (AddI src0 src1));
7654   ins_cost(110);
7655 
7656   format %{ "LEA    $dst,[$src0 + $src1]" %}
7657   opcode(0x8D); /* 0x8D /r */
7658   ins_encode( OpcP, RegLea( dst, src0, src1 ) );
7659   ins_pipe( ialu_reg_reg );
7660 %}
7661 
7662 instruct leaP_eReg_immI(eRegP dst, eRegP src0, immI src1) %{
7663   match(Set dst (AddP src0 src1));
7664   ins_cost(110);
7665 
7666   format %{ "LEA    $dst,[$src0 + $src1]\t# ptr" %}
7667   opcode(0x8D); /* 0x8D /r */
7668   ins_encode( OpcP, RegLea( dst, src0, src1 ) );
7669   ins_pipe( ialu_reg_reg );
7670 %}
7671 
7672 instruct decI_eReg(eRegI dst, immI_M1 src, eFlagsReg cr) %{
7673   predicate(UseIncDec);
7674   match(Set dst (AddI dst src));
7675   effect(KILL cr);
7676 
7677   size(1);
7678   format %{ "DEC    $dst" %}
7679   opcode(0x48); /*  */
7680   ins_encode( Opc_plus( primary, dst ) );
7681   ins_pipe( ialu_reg );
7682 %}
7683 
7684 instruct addP_eReg(eRegP dst, eRegI src, eFlagsReg cr) %{
7685   match(Set dst (AddP dst src));
7686   effect(KILL cr);
7687 
7688   size(2);
7689   format %{ "ADD    $dst,$src" %}
7690   opcode(0x03);
7691   ins_encode( OpcP, RegReg( dst, src) );
7692   ins_pipe( ialu_reg_reg );
7693 %}
7694 
7695 instruct addP_eReg_imm(eRegP dst, immI src, eFlagsReg cr) %{
7696   match(Set dst (AddP dst src));
7697   effect(KILL cr);
7698 
7699   format %{ "ADD    $dst,$src" %}
7700   opcode(0x81,0x00); /* Opcode 81 /0 id */
7701   // ins_encode( RegImm( dst, src) );
7702   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7703   ins_pipe( ialu_reg );
7704 %}
7705 
7706 instruct addI_eReg_mem(eRegI dst, memory src, eFlagsReg cr) %{
7707   match(Set dst (AddI dst (LoadI src)));
7708   effect(KILL cr);
7709 
7710   ins_cost(125);
7711   format %{ "ADD    $dst,$src" %}
7712   opcode(0x03);
7713   ins_encode( OpcP, RegMem( dst, src) );
7714   ins_pipe( ialu_reg_mem );
7715 %}
7716 
7717 instruct addI_mem_eReg(memory dst, eRegI src, eFlagsReg cr) %{
7718   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7719   effect(KILL cr);
7720 
7721   ins_cost(150);
7722   format %{ "ADD    $dst,$src" %}
7723   opcode(0x01);  /* Opcode 01 /r */
7724   ins_encode( OpcP, RegMem( src, dst ) );
7725   ins_pipe( ialu_mem_reg );
7726 %}
7727 
7728 // Add Memory with Immediate
7729 instruct addI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
7730   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7731   effect(KILL cr);
7732 
7733   ins_cost(125);
7734   format %{ "ADD    $dst,$src" %}
7735   opcode(0x81);               /* Opcode 81 /0 id */
7736   ins_encode( OpcSE( src ), RMopc_Mem(0x00,dst), Con8or32( src ) );
7737   ins_pipe( ialu_mem_imm );
7738 %}
7739 
7740 instruct incI_mem(memory dst, immI1 src, eFlagsReg cr) %{
7741   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7742   effect(KILL cr);
7743 
7744   ins_cost(125);
7745   format %{ "INC    $dst" %}
7746   opcode(0xFF);               /* Opcode FF /0 */
7747   ins_encode( OpcP, RMopc_Mem(0x00,dst));
7748   ins_pipe( ialu_mem_imm );
7749 %}
7750 
7751 instruct decI_mem(memory dst, immI_M1 src, eFlagsReg cr) %{
7752   match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7753   effect(KILL cr);
7754 
7755   ins_cost(125);
7756   format %{ "DEC    $dst" %}
7757   opcode(0xFF);               /* Opcode FF /1 */
7758   ins_encode( OpcP, RMopc_Mem(0x01,dst));
7759   ins_pipe( ialu_mem_imm );
7760 %}
7761 
7762 
7763 instruct checkCastPP( eRegP dst ) %{
7764   match(Set dst (CheckCastPP dst));
7765 
7766   size(0);
7767   format %{ "#checkcastPP of $dst" %}
7768   ins_encode( /*empty encoding*/ );
7769   ins_pipe( empty );
7770 %}
7771 
7772 instruct castPP( eRegP dst ) %{
7773   match(Set dst (CastPP dst));
7774   format %{ "#castPP of $dst" %}
7775   ins_encode( /*empty encoding*/ );
7776   ins_pipe( empty );
7777 %}
7778 
7779 instruct castII( eRegI dst ) %{
7780   match(Set dst (CastII dst));
7781   format %{ "#castII of $dst" %}
7782   ins_encode( /*empty encoding*/ );
7783   ins_cost(0);
7784   ins_pipe( empty );
7785 %}
7786 
7787 
7788 // Load-locked - same as a regular pointer load when used with compare-swap
7789 instruct loadPLocked(eRegP dst, memory mem) %{
7790   match(Set dst (LoadPLocked mem));
7791 
7792   ins_cost(125);
7793   format %{ "MOV    $dst,$mem\t# Load ptr. locked" %}
7794   opcode(0x8B);
7795   ins_encode( OpcP, RegMem(dst,mem));
7796   ins_pipe( ialu_reg_mem );
7797 %}
7798 
7799 // LoadLong-locked - same as a volatile long load when used with compare-swap
7800 instruct loadLLocked(stackSlotL dst, memory mem) %{
7801   predicate(UseSSE<=1);
7802   match(Set dst (LoadLLocked mem));
7803 
7804   ins_cost(200);
7805   format %{ "FILD   $mem\t# Atomic volatile long load\n\t"
7806             "FISTp  $dst" %}
7807   ins_encode(enc_loadL_volatile(mem,dst));
7808   ins_pipe( fpu_reg_mem );
7809 %}
7810 
7811 instruct loadLX_Locked(stackSlotL dst, memory mem, regD tmp) %{
7812   predicate(UseSSE>=2);
7813   match(Set dst (LoadLLocked mem));
7814   effect(TEMP tmp);
7815   ins_cost(180);
7816   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
7817             "MOVSD  $dst,$tmp" %}
7818   ins_encode %{
7819     __ movdbl($tmp$$XMMRegister, $mem$$Address);
7820     __ movdbl(Address(rsp, $dst$$disp), $tmp$$XMMRegister);
7821   %}
7822   ins_pipe( pipe_slow );
7823 %}
7824 
7825 instruct loadLX_reg_Locked(eRegL dst, memory mem, regD tmp) %{
7826   predicate(UseSSE>=2);
7827   match(Set dst (LoadLLocked mem));
7828   effect(TEMP tmp);
7829   ins_cost(160);
7830   format %{ "MOVSD  $tmp,$mem\t# Atomic volatile long load\n\t"
7831             "MOVD   $dst.lo,$tmp\n\t"
7832             "PSRLQ  $tmp,32\n\t"
7833             "MOVD   $dst.hi,$tmp" %}
7834   ins_encode %{
7835     __ movdbl($tmp$$XMMRegister, $mem$$Address);
7836     __ movdl($dst$$Register, $tmp$$XMMRegister);
7837     __ psrlq($tmp$$XMMRegister, 32);
7838     __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
7839   %}
7840   ins_pipe( pipe_slow );
7841 %}
7842 
7843 // Conditional-store of the updated heap-top.
7844 // Used during allocation of the shared heap.
7845 // Sets flags (EQ) on success.  Implemented with a CMPXCHG on Intel.
7846 instruct storePConditional( memory heap_top_ptr, eAXRegP oldval, eRegP newval, eFlagsReg cr ) %{
7847   match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
7848   // EAX is killed if there is contention, but then it's also unused.
7849   // In the common case of no contention, EAX holds the new oop address.
7850   format %{ "CMPXCHG $heap_top_ptr,$newval\t# If EAX==$heap_top_ptr Then store $newval into $heap_top_ptr" %}
7851   ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval,heap_top_ptr) );
7852   ins_pipe( pipe_cmpxchg );
7853 %}
7854 
7855 // Conditional-store of an int value.
7856 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG on Intel.
7857 instruct storeIConditional( memory mem, eAXRegI oldval, eRegI newval, eFlagsReg cr ) %{
7858   match(Set cr (StoreIConditional mem (Binary oldval newval)));
7859   effect(KILL oldval);
7860   format %{ "CMPXCHG $mem,$newval\t# If EAX==$mem Then store $newval into $mem" %}
7861   ins_encode( lock_prefix, Opcode(0x0F), Opcode(0xB1), RegMem(newval, mem) );
7862   ins_pipe( pipe_cmpxchg );
7863 %}
7864 
7865 // Conditional-store of a long value.
7866 // ZF flag is set on success, reset otherwise.  Implemented with a CMPXCHG8 on Intel.
7867 instruct storeLConditional( memory mem, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
7868   match(Set cr (StoreLConditional mem (Binary oldval newval)));
7869   effect(KILL oldval);
7870   format %{ "XCHG   EBX,ECX\t# correct order for CMPXCHG8 instruction\n\t"
7871             "CMPXCHG8 $mem,ECX:EBX\t# If EDX:EAX==$mem Then store ECX:EBX into $mem\n\t"
7872             "XCHG   EBX,ECX"
7873   %}
7874   ins_encode %{
7875     // Note: we need to swap rbx, and rcx before and after the
7876     //       cmpxchg8 instruction because the instruction uses
7877     //       rcx as the high order word of the new value to store but
7878     //       our register encoding uses rbx.
7879     __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
7880     if( os::is_MP() )
7881       __ lock();
7882     __ cmpxchg8($mem$$Address);
7883     __ xchgl(as_Register(EBX_enc), as_Register(ECX_enc));
7884   %}
7885   ins_pipe( pipe_cmpxchg );
7886 %}
7887 
7888 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7889 
7890 instruct compareAndSwapL( eRegI res, eSIRegP mem_ptr, eADXRegL oldval, eBCXRegL newval, eFlagsReg cr ) %{
7891   match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7892   effect(KILL cr, KILL oldval);
7893   format %{ "CMPXCHG8 [$mem_ptr],$newval\t# If EDX:EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7894             "MOV    $res,0\n\t"
7895             "JNE,s  fail\n\t"
7896             "MOV    $res,1\n"
7897           "fail:" %}
7898   ins_encode( enc_cmpxchg8(mem_ptr),
7899               enc_flags_ne_to_boolean(res) );
7900   ins_pipe( pipe_cmpxchg );
7901 %}
7902 
7903 instruct compareAndSwapP( eRegI res,  pRegP mem_ptr, eAXRegP oldval, eCXRegP newval, eFlagsReg cr) %{
7904   match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7905   effect(KILL cr, KILL oldval);
7906   format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7907             "MOV    $res,0\n\t"
7908             "JNE,s  fail\n\t"
7909             "MOV    $res,1\n"
7910           "fail:" %}
7911   ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
7912   ins_pipe( pipe_cmpxchg );
7913 %}
7914 
7915 instruct compareAndSwapI( eRegI res, pRegP mem_ptr, eAXRegI oldval, eCXRegI newval, eFlagsReg cr) %{
7916   match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7917   effect(KILL cr, KILL oldval);
7918   format %{ "CMPXCHG [$mem_ptr],$newval\t# If EAX==[$mem_ptr] Then store $newval into [$mem_ptr]\n\t"
7919             "MOV    $res,0\n\t"
7920             "JNE,s  fail\n\t"
7921             "MOV    $res,1\n"
7922           "fail:" %}
7923   ins_encode( enc_cmpxchg(mem_ptr), enc_flags_ne_to_boolean(res) );
7924   ins_pipe( pipe_cmpxchg );
7925 %}
7926 
7927 //----------Subtraction Instructions-------------------------------------------
7928 // Integer Subtraction Instructions
7929 instruct subI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
7930   match(Set dst (SubI dst src));
7931   effect(KILL cr);
7932 
7933   size(2);
7934   format %{ "SUB    $dst,$src" %}
7935   opcode(0x2B);
7936   ins_encode( OpcP, RegReg( dst, src) );
7937   ins_pipe( ialu_reg_reg );
7938 %}
7939 
7940 instruct subI_eReg_imm(eRegI dst, immI src, eFlagsReg cr) %{
7941   match(Set dst (SubI dst src));
7942   effect(KILL cr);
7943 
7944   format %{ "SUB    $dst,$src" %}
7945   opcode(0x81,0x05);  /* Opcode 81 /5 */
7946   // ins_encode( RegImm( dst, src) );
7947   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
7948   ins_pipe( ialu_reg );
7949 %}
7950 
7951 instruct subI_eReg_mem(eRegI dst, memory src, eFlagsReg cr) %{
7952   match(Set dst (SubI dst (LoadI src)));
7953   effect(KILL cr);
7954 
7955   ins_cost(125);
7956   format %{ "SUB    $dst,$src" %}
7957   opcode(0x2B);
7958   ins_encode( OpcP, RegMem( dst, src) );
7959   ins_pipe( ialu_reg_mem );
7960 %}
7961 
7962 instruct subI_mem_eReg(memory dst, eRegI src, eFlagsReg cr) %{
7963   match(Set dst (StoreI dst (SubI (LoadI dst) src)));
7964   effect(KILL cr);
7965 
7966   ins_cost(150);
7967   format %{ "SUB    $dst,$src" %}
7968   opcode(0x29);  /* Opcode 29 /r */
7969   ins_encode( OpcP, RegMem( src, dst ) );
7970   ins_pipe( ialu_mem_reg );
7971 %}
7972 
7973 // Subtract from a pointer
7974 instruct subP_eReg(eRegP dst, eRegI src, immI0 zero, eFlagsReg cr) %{
7975   match(Set dst (AddP dst (SubI zero src)));
7976   effect(KILL cr);
7977 
7978   size(2);
7979   format %{ "SUB    $dst,$src" %}
7980   opcode(0x2B);
7981   ins_encode( OpcP, RegReg( dst, src) );
7982   ins_pipe( ialu_reg_reg );
7983 %}
7984 
7985 instruct negI_eReg(eRegI dst, immI0 zero, eFlagsReg cr) %{
7986   match(Set dst (SubI zero dst));
7987   effect(KILL cr);
7988 
7989   size(2);
7990   format %{ "NEG    $dst" %}
7991   opcode(0xF7,0x03);  // Opcode F7 /3
7992   ins_encode( OpcP, RegOpc( dst ) );
7993   ins_pipe( ialu_reg );
7994 %}
7995 
7996 
7997 //----------Multiplication/Division Instructions-------------------------------
7998 // Integer Multiplication Instructions
7999 // Multiply Register
8000 instruct mulI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
8001   match(Set dst (MulI dst src));
8002   effect(KILL cr);
8003 
8004   size(3);
8005   ins_cost(300);
8006   format %{ "IMUL   $dst,$src" %}
8007   opcode(0xAF, 0x0F);
8008   ins_encode( OpcS, OpcP, RegReg( dst, src) );
8009   ins_pipe( ialu_reg_reg_alu0 );
8010 %}
8011 
8012 // Multiply 32-bit Immediate
8013 instruct mulI_eReg_imm(eRegI dst, eRegI src, immI imm, eFlagsReg cr) %{
8014   match(Set dst (MulI src imm));
8015   effect(KILL cr);
8016 
8017   ins_cost(300);
8018   format %{ "IMUL   $dst,$src,$imm" %}
8019   opcode(0x69);  /* 69 /r id */
8020   ins_encode( OpcSE(imm), RegReg( dst, src ), Con8or32( imm ) );
8021   ins_pipe( ialu_reg_reg_alu0 );
8022 %}
8023 
8024 instruct loadConL_low_only(eADXRegL_low_only dst, immL32 src, eFlagsReg cr) %{
8025   match(Set dst src);
8026   effect(KILL cr);
8027 
8028   // Note that this is artificially increased to make it more expensive than loadConL
8029   ins_cost(250);
8030   format %{ "MOV    EAX,$src\t// low word only" %}
8031   opcode(0xB8);
8032   ins_encode( LdImmL_Lo(dst, src) );
8033   ins_pipe( ialu_reg_fat );
8034 %}
8035 
8036 // Multiply by 32-bit Immediate, taking the shifted high order results
8037 //  (special case for shift by 32)
8038 instruct mulI_imm_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32 cnt, eFlagsReg cr) %{
8039   match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
8040   predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
8041              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
8042              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
8043   effect(USE src1, KILL cr);
8044 
8045   // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
8046   ins_cost(0*100 + 1*400 - 150);
8047   format %{ "IMUL   EDX:EAX,$src1" %}
8048   ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
8049   ins_pipe( pipe_slow );
8050 %}
8051 
8052 // Multiply by 32-bit Immediate, taking the shifted high order results
8053 instruct mulI_imm_RShift_high(eDXRegI dst, nadxRegI src1, eADXRegL_low_only src2, immI_32_63 cnt, eFlagsReg cr) %{
8054   match(Set dst (ConvL2I (RShiftL (MulL (ConvI2L src1) src2) cnt)));
8055   predicate( _kids[0]->_kids[0]->_kids[1]->_leaf->Opcode() == Op_ConL &&
8056              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() >= min_jint &&
8057              _kids[0]->_kids[0]->_kids[1]->_leaf->as_Type()->type()->is_long()->get_con() <= max_jint );
8058   effect(USE src1, KILL cr);
8059 
8060   // Note that this is adjusted by 150 to compensate for the overcosting of loadConL_low_only
8061   ins_cost(1*100 + 1*400 - 150);
8062   format %{ "IMUL   EDX:EAX,$src1\n\t"
8063             "SAR    EDX,$cnt-32" %}
8064   ins_encode( multiply_con_and_shift_high( dst, src1, src2, cnt, cr ) );
8065   ins_pipe( pipe_slow );
8066 %}
8067 
8068 // Multiply Memory 32-bit Immediate
8069 instruct mulI_mem_imm(eRegI dst, memory src, immI imm, eFlagsReg cr) %{
8070   match(Set dst (MulI (LoadI src) imm));
8071   effect(KILL cr);
8072 
8073   ins_cost(300);
8074   format %{ "IMUL   $dst,$src,$imm" %}
8075   opcode(0x69);  /* 69 /r id */
8076   ins_encode( OpcSE(imm), RegMem( dst, src ), Con8or32( imm ) );
8077   ins_pipe( ialu_reg_mem_alu0 );
8078 %}
8079 
8080 // Multiply Memory
8081 instruct mulI(eRegI dst, memory src, eFlagsReg cr) %{
8082   match(Set dst (MulI dst (LoadI src)));
8083   effect(KILL cr);
8084 
8085   ins_cost(350);
8086   format %{ "IMUL   $dst,$src" %}
8087   opcode(0xAF, 0x0F);
8088   ins_encode( OpcS, OpcP, RegMem( dst, src) );
8089   ins_pipe( ialu_reg_mem_alu0 );
8090 %}
8091 
8092 // Multiply Register Int to Long
8093 instruct mulI2L(eADXRegL dst, eAXRegI src, nadxRegI src1, eFlagsReg flags) %{
8094   // Basic Idea: long = (long)int * (long)int
8095   match(Set dst (MulL (ConvI2L src) (ConvI2L src1)));
8096   effect(DEF dst, USE src, USE src1, KILL flags);
8097 
8098   ins_cost(300);
8099   format %{ "IMUL   $dst,$src1" %}
8100 
8101   ins_encode( long_int_multiply( dst, src1 ) );
8102   ins_pipe( ialu_reg_reg_alu0 );
8103 %}
8104 
8105 instruct mulIS_eReg(eADXRegL dst, immL_32bits mask, eFlagsReg flags, eAXRegI src, nadxRegI src1) %{
8106   // Basic Idea:  long = (int & 0xffffffffL) * (int & 0xffffffffL)
8107   match(Set dst (MulL (AndL (ConvI2L src) mask) (AndL (ConvI2L src1) mask)));
8108   effect(KILL flags);
8109 
8110   ins_cost(300);
8111   format %{ "MUL    $dst,$src1" %}
8112 
8113   ins_encode( long_uint_multiply(dst, src1) );
8114   ins_pipe( ialu_reg_reg_alu0 );
8115 %}
8116 
8117 // Multiply Register Long
8118 instruct mulL_eReg(eADXRegL dst, eRegL src, eRegI tmp, eFlagsReg cr) %{
8119   match(Set dst (MulL dst src));
8120   effect(KILL cr, TEMP tmp);
8121   ins_cost(4*100+3*400);
8122 // Basic idea: lo(result) = lo(x_lo * y_lo)
8123 //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
8124   format %{ "MOV    $tmp,$src.lo\n\t"
8125             "IMUL   $tmp,EDX\n\t"
8126             "MOV    EDX,$src.hi\n\t"
8127             "IMUL   EDX,EAX\n\t"
8128             "ADD    $tmp,EDX\n\t"
8129             "MUL    EDX:EAX,$src.lo\n\t"
8130             "ADD    EDX,$tmp" %}
8131   ins_encode( long_multiply( dst, src, tmp ) );
8132   ins_pipe( pipe_slow );
8133 %}
8134 
8135 // Multiply Register Long where the left operand's high 32 bits are zero
8136 instruct mulL_eReg_lhi0(eADXRegL dst, eRegL src, eRegI tmp, eFlagsReg cr) %{
8137   predicate(is_operand_hi32_zero(n->in(1)));
8138   match(Set dst (MulL dst src));
8139   effect(KILL cr, TEMP tmp);
8140   ins_cost(2*100+2*400);
8141 // Basic idea: lo(result) = lo(x_lo * y_lo)
8142 //             hi(result) = hi(x_lo * y_lo) + lo(x_lo * y_hi) where lo(x_hi * y_lo) = 0 because x_hi = 0
8143   format %{ "MOV    $tmp,$src.hi\n\t"
8144             "IMUL   $tmp,EAX\n\t"
8145             "MUL    EDX:EAX,$src.lo\n\t"
8146             "ADD    EDX,$tmp" %}
8147   ins_encode %{
8148     __ movl($tmp$$Register, HIGH_FROM_LOW($src$$Register));
8149     __ imull($tmp$$Register, rax);
8150     __ mull($src$$Register);
8151     __ addl(rdx, $tmp$$Register);
8152   %}
8153   ins_pipe( pipe_slow );
8154 %}
8155 
8156 // Multiply Register Long where the right operand's high 32 bits are zero
8157 instruct mulL_eReg_rhi0(eADXRegL dst, eRegL src, eRegI tmp, eFlagsReg cr) %{
8158   predicate(is_operand_hi32_zero(n->in(2)));
8159   match(Set dst (MulL dst src));
8160   effect(KILL cr, TEMP tmp);
8161   ins_cost(2*100+2*400);
8162 // Basic idea: lo(result) = lo(x_lo * y_lo)
8163 //             hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) where lo(x_lo * y_hi) = 0 because y_hi = 0
8164   format %{ "MOV    $tmp,$src.lo\n\t"
8165             "IMUL   $tmp,EDX\n\t"
8166             "MUL    EDX:EAX,$src.lo\n\t"
8167             "ADD    EDX,$tmp" %}
8168   ins_encode %{
8169     __ movl($tmp$$Register, $src$$Register);
8170     __ imull($tmp$$Register, rdx);
8171     __ mull($src$$Register);
8172     __ addl(rdx, $tmp$$Register);
8173   %}
8174   ins_pipe( pipe_slow );
8175 %}
8176 
8177 // Multiply Register Long where the left and the right operands' high 32 bits are zero
8178 instruct mulL_eReg_hi0(eADXRegL dst, eRegL src, eFlagsReg cr) %{
8179   predicate(is_operand_hi32_zero(n->in(1)) && is_operand_hi32_zero(n->in(2)));
8180   match(Set dst (MulL dst src));
8181   effect(KILL cr);
8182   ins_cost(1*400);
8183 // Basic idea: lo(result) = lo(x_lo * y_lo)
8184 //             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
8185   format %{ "MUL    EDX:EAX,$src.lo\n\t" %}
8186   ins_encode %{
8187     __ mull($src$$Register);
8188   %}
8189   ins_pipe( pipe_slow );
8190 %}
8191 
8192 // Multiply Register Long by small constant
8193 instruct mulL_eReg_con(eADXRegL dst, immL_127 src, eRegI tmp, eFlagsReg cr) %{
8194   match(Set dst (MulL dst src));
8195   effect(KILL cr, TEMP tmp);
8196   ins_cost(2*100+2*400);
8197   size(12);
8198 // Basic idea: lo(result) = lo(src * EAX)
8199 //             hi(result) = hi(src * EAX) + lo(src * EDX)
8200   format %{ "IMUL   $tmp,EDX,$src\n\t"
8201             "MOV    EDX,$src\n\t"
8202             "MUL    EDX\t# EDX*EAX -> EDX:EAX\n\t"
8203             "ADD    EDX,$tmp" %}
8204   ins_encode( long_multiply_con( dst, src, tmp ) );
8205   ins_pipe( pipe_slow );
8206 %}
8207 
8208 // Integer DIV with Register
8209 instruct divI_eReg(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
8210   match(Set rax (DivI rax div));
8211   effect(KILL rdx, KILL cr);
8212   size(26);
8213   ins_cost(30*100+10*100);
8214   format %{ "CMP    EAX,0x80000000\n\t"
8215             "JNE,s  normal\n\t"
8216             "XOR    EDX,EDX\n\t"
8217             "CMP    ECX,-1\n\t"
8218             "JE,s   done\n"
8219     "normal: CDQ\n\t"
8220             "IDIV   $div\n\t"
8221     "done:"        %}
8222   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8223   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8224   ins_pipe( ialu_reg_reg_alu0 );
8225 %}
8226 
8227 // Divide Register Long
8228 instruct divL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
8229   match(Set dst (DivL src1 src2));
8230   effect( KILL cr, KILL cx, KILL bx );
8231   ins_cost(10000);
8232   format %{ "PUSH   $src1.hi\n\t"
8233             "PUSH   $src1.lo\n\t"
8234             "PUSH   $src2.hi\n\t"
8235             "PUSH   $src2.lo\n\t"
8236             "CALL   SharedRuntime::ldiv\n\t"
8237             "ADD    ESP,16" %}
8238   ins_encode( long_div(src1,src2) );
8239   ins_pipe( pipe_slow );
8240 %}
8241 
8242 // Integer DIVMOD with Register, both quotient and mod results
8243 instruct divModI_eReg_divmod(eAXRegI rax, eDXRegI rdx, eCXRegI div, eFlagsReg cr) %{
8244   match(DivModI rax div);
8245   effect(KILL cr);
8246   size(26);
8247   ins_cost(30*100+10*100);
8248   format %{ "CMP    EAX,0x80000000\n\t"
8249             "JNE,s  normal\n\t"
8250             "XOR    EDX,EDX\n\t"
8251             "CMP    ECX,-1\n\t"
8252             "JE,s   done\n"
8253     "normal: CDQ\n\t"
8254             "IDIV   $div\n\t"
8255     "done:"        %}
8256   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8257   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8258   ins_pipe( pipe_slow );
8259 %}
8260 
8261 // Integer MOD with Register
8262 instruct modI_eReg(eDXRegI rdx, eAXRegI rax, eCXRegI div, eFlagsReg cr) %{
8263   match(Set rdx (ModI rax div));
8264   effect(KILL rax, KILL cr);
8265 
8266   size(26);
8267   ins_cost(300);
8268   format %{ "CDQ\n\t"
8269             "IDIV   $div" %}
8270   opcode(0xF7, 0x7);  /* Opcode F7 /7 */
8271   ins_encode( cdq_enc, OpcP, RegOpc(div) );
8272   ins_pipe( ialu_reg_reg_alu0 );
8273 %}
8274 
8275 // Remainder Register Long
8276 instruct modL_eReg( eADXRegL dst, eRegL src1, eRegL src2, eFlagsReg cr, eCXRegI cx, eBXRegI bx ) %{
8277   match(Set dst (ModL src1 src2));
8278   effect( KILL cr, KILL cx, KILL bx );
8279   ins_cost(10000);
8280   format %{ "PUSH   $src1.hi\n\t"
8281             "PUSH   $src1.lo\n\t"
8282             "PUSH   $src2.hi\n\t"
8283             "PUSH   $src2.lo\n\t"
8284             "CALL   SharedRuntime::lrem\n\t"
8285             "ADD    ESP,16" %}
8286   ins_encode( long_mod(src1,src2) );
8287   ins_pipe( pipe_slow );
8288 %}
8289 
8290 // Divide Register Long (no special case since divisor != -1)
8291 instruct divL_eReg_imm32( eADXRegL dst, immL32 imm, eRegI tmp, eRegI tmp2, eFlagsReg cr ) %{
8292   match(Set dst (DivL dst imm));
8293   effect( TEMP tmp, TEMP tmp2, KILL cr );
8294   ins_cost(1000);
8295   format %{ "MOV    $tmp,abs($imm) # ldiv EDX:EAX,$imm\n\t"
8296             "XOR    $tmp2,$tmp2\n\t"
8297             "CMP    $tmp,EDX\n\t"
8298             "JA,s   fast\n\t"
8299             "MOV    $tmp2,EAX\n\t"
8300             "MOV    EAX,EDX\n\t"
8301             "MOV    EDX,0\n\t"
8302             "JLE,s  pos\n\t"
8303             "LNEG   EAX : $tmp2\n\t"
8304             "DIV    $tmp # unsigned division\n\t"
8305             "XCHG   EAX,$tmp2\n\t"
8306             "DIV    $tmp\n\t"
8307             "LNEG   $tmp2 : EAX\n\t"
8308             "JMP,s  done\n"
8309     "pos:\n\t"
8310             "DIV    $tmp\n\t"
8311             "XCHG   EAX,$tmp2\n"
8312     "fast:\n\t"
8313             "DIV    $tmp\n"
8314     "done:\n\t"
8315             "MOV    EDX,$tmp2\n\t"
8316             "NEG    EDX:EAX # if $imm < 0" %}
8317   ins_encode %{
8318     int con = (int)$imm$$constant;
8319     assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
8320     int pcon = (con > 0) ? con : -con;
8321     Label Lfast, Lpos, Ldone;
8322 
8323     __ movl($tmp$$Register, pcon);
8324     __ xorl($tmp2$$Register,$tmp2$$Register);
8325     __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
8326     __ jccb(Assembler::above, Lfast); // result fits into 32 bit
8327 
8328     __ movl($tmp2$$Register, $dst$$Register); // save
8329     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8330     __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
8331     __ jccb(Assembler::lessEqual, Lpos); // result is positive
8332 
8333     // Negative dividend.
8334     // convert value to positive to use unsigned division
8335     __ lneg($dst$$Register, $tmp2$$Register);
8336     __ divl($tmp$$Register);
8337     __ xchgl($dst$$Register, $tmp2$$Register);
8338     __ divl($tmp$$Register);
8339     // revert result back to negative
8340     __ lneg($tmp2$$Register, $dst$$Register);
8341     __ jmpb(Ldone);
8342 
8343     __ bind(Lpos);
8344     __ divl($tmp$$Register); // Use unsigned division
8345     __ xchgl($dst$$Register, $tmp2$$Register);
8346     // Fallthrow for final divide, tmp2 has 32 bit hi result
8347 
8348     __ bind(Lfast);
8349     // fast path: src is positive
8350     __ divl($tmp$$Register); // Use unsigned division
8351 
8352     __ bind(Ldone);
8353     __ movl(HIGH_FROM_LOW($dst$$Register),$tmp2$$Register);
8354     if (con < 0) {
8355       __ lneg(HIGH_FROM_LOW($dst$$Register), $dst$$Register);
8356     }
8357   %}
8358   ins_pipe( pipe_slow );
8359 %}
8360 
8361 // Remainder Register Long (remainder fit into 32 bits)
8362 instruct modL_eReg_imm32( eADXRegL dst, immL32 imm, eRegI tmp, eRegI tmp2, eFlagsReg cr ) %{
8363   match(Set dst (ModL dst imm));
8364   effect( TEMP tmp, TEMP tmp2, KILL cr );
8365   ins_cost(1000);
8366   format %{ "MOV    $tmp,abs($imm) # lrem EDX:EAX,$imm\n\t"
8367             "CMP    $tmp,EDX\n\t"
8368             "JA,s   fast\n\t"
8369             "MOV    $tmp2,EAX\n\t"
8370             "MOV    EAX,EDX\n\t"
8371             "MOV    EDX,0\n\t"
8372             "JLE,s  pos\n\t"
8373             "LNEG   EAX : $tmp2\n\t"
8374             "DIV    $tmp # unsigned division\n\t"
8375             "MOV    EAX,$tmp2\n\t"
8376             "DIV    $tmp\n\t"
8377             "NEG    EDX\n\t"
8378             "JMP,s  done\n"
8379     "pos:\n\t"
8380             "DIV    $tmp\n\t"
8381             "MOV    EAX,$tmp2\n"
8382     "fast:\n\t"
8383             "DIV    $tmp\n"
8384     "done:\n\t"
8385             "MOV    EAX,EDX\n\t"
8386             "SAR    EDX,31\n\t" %}
8387   ins_encode %{
8388     int con = (int)$imm$$constant;
8389     assert(con != 0 && con != -1 && con != min_jint, "wrong divisor");
8390     int pcon = (con > 0) ? con : -con;
8391     Label  Lfast, Lpos, Ldone;
8392 
8393     __ movl($tmp$$Register, pcon);
8394     __ cmpl($tmp$$Register, HIGH_FROM_LOW($dst$$Register));
8395     __ jccb(Assembler::above, Lfast); // src is positive and result fits into 32 bit
8396 
8397     __ movl($tmp2$$Register, $dst$$Register); // save
8398     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8399     __ movl(HIGH_FROM_LOW($dst$$Register),0); // preserve flags
8400     __ jccb(Assembler::lessEqual, Lpos); // result is positive
8401 
8402     // Negative dividend.
8403     // convert value to positive to use unsigned division
8404     __ lneg($dst$$Register, $tmp2$$Register);
8405     __ divl($tmp$$Register);
8406     __ movl($dst$$Register, $tmp2$$Register);
8407     __ divl($tmp$$Register);
8408     // revert remainder back to negative
8409     __ negl(HIGH_FROM_LOW($dst$$Register));
8410     __ jmpb(Ldone);
8411 
8412     __ bind(Lpos);
8413     __ divl($tmp$$Register);
8414     __ movl($dst$$Register, $tmp2$$Register);
8415 
8416     __ bind(Lfast);
8417     // fast path: src is positive
8418     __ divl($tmp$$Register);
8419 
8420     __ bind(Ldone);
8421     __ movl($dst$$Register, HIGH_FROM_LOW($dst$$Register));
8422     __ sarl(HIGH_FROM_LOW($dst$$Register), 31); // result sign
8423 
8424   %}
8425   ins_pipe( pipe_slow );
8426 %}
8427 
8428 // Integer Shift Instructions
8429 // Shift Left by one
8430 instruct shlI_eReg_1(eRegI dst, immI1 shift, eFlagsReg cr) %{
8431   match(Set dst (LShiftI dst shift));
8432   effect(KILL cr);
8433 
8434   size(2);
8435   format %{ "SHL    $dst,$shift" %}
8436   opcode(0xD1, 0x4);  /* D1 /4 */
8437   ins_encode( OpcP, RegOpc( dst ) );
8438   ins_pipe( ialu_reg );
8439 %}
8440 
8441 // Shift Left by 8-bit immediate
8442 instruct salI_eReg_imm(eRegI dst, immI8 shift, eFlagsReg cr) %{
8443   match(Set dst (LShiftI dst shift));
8444   effect(KILL cr);
8445 
8446   size(3);
8447   format %{ "SHL    $dst,$shift" %}
8448   opcode(0xC1, 0x4);  /* C1 /4 ib */
8449   ins_encode( RegOpcImm( dst, shift) );
8450   ins_pipe( ialu_reg );
8451 %}
8452 
8453 // Shift Left by variable
8454 instruct salI_eReg_CL(eRegI dst, eCXRegI shift, eFlagsReg cr) %{
8455   match(Set dst (LShiftI dst shift));
8456   effect(KILL cr);
8457 
8458   size(2);
8459   format %{ "SHL    $dst,$shift" %}
8460   opcode(0xD3, 0x4);  /* D3 /4 */
8461   ins_encode( OpcP, RegOpc( dst ) );
8462   ins_pipe( ialu_reg_reg );
8463 %}
8464 
8465 // Arithmetic shift right by one
8466 instruct sarI_eReg_1(eRegI dst, immI1 shift, eFlagsReg cr) %{
8467   match(Set dst (RShiftI dst shift));
8468   effect(KILL cr);
8469 
8470   size(2);
8471   format %{ "SAR    $dst,$shift" %}
8472   opcode(0xD1, 0x7);  /* D1 /7 */
8473   ins_encode( OpcP, RegOpc( dst ) );
8474   ins_pipe( ialu_reg );
8475 %}
8476 
8477 // Arithmetic shift right by one
8478 instruct sarI_mem_1(memory dst, immI1 shift, eFlagsReg cr) %{
8479   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8480   effect(KILL cr);
8481   format %{ "SAR    $dst,$shift" %}
8482   opcode(0xD1, 0x7);  /* D1 /7 */
8483   ins_encode( OpcP, RMopc_Mem(secondary,dst) );
8484   ins_pipe( ialu_mem_imm );
8485 %}
8486 
8487 // Arithmetic Shift Right by 8-bit immediate
8488 instruct sarI_eReg_imm(eRegI dst, immI8 shift, eFlagsReg cr) %{
8489   match(Set dst (RShiftI dst shift));
8490   effect(KILL cr);
8491 
8492   size(3);
8493   format %{ "SAR    $dst,$shift" %}
8494   opcode(0xC1, 0x7);  /* C1 /7 ib */
8495   ins_encode( RegOpcImm( dst, shift ) );
8496   ins_pipe( ialu_mem_imm );
8497 %}
8498 
8499 // Arithmetic Shift Right by 8-bit immediate
8500 instruct sarI_mem_imm(memory dst, immI8 shift, eFlagsReg cr) %{
8501   match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8502   effect(KILL cr);
8503 
8504   format %{ "SAR    $dst,$shift" %}
8505   opcode(0xC1, 0x7);  /* C1 /7 ib */
8506   ins_encode( OpcP, RMopc_Mem(secondary, dst ), Con8or32( shift ) );
8507   ins_pipe( ialu_mem_imm );
8508 %}
8509 
8510 // Arithmetic Shift Right by variable
8511 instruct sarI_eReg_CL(eRegI dst, eCXRegI shift, eFlagsReg cr) %{
8512   match(Set dst (RShiftI dst shift));
8513   effect(KILL cr);
8514 
8515   size(2);
8516   format %{ "SAR    $dst,$shift" %}
8517   opcode(0xD3, 0x7);  /* D3 /7 */
8518   ins_encode( OpcP, RegOpc( dst ) );
8519   ins_pipe( ialu_reg_reg );
8520 %}
8521 
8522 // Logical shift right by one
8523 instruct shrI_eReg_1(eRegI dst, immI1 shift, eFlagsReg cr) %{
8524   match(Set dst (URShiftI dst shift));
8525   effect(KILL cr);
8526 
8527   size(2);
8528   format %{ "SHR    $dst,$shift" %}
8529   opcode(0xD1, 0x5);  /* D1 /5 */
8530   ins_encode( OpcP, RegOpc( dst ) );
8531   ins_pipe( ialu_reg );
8532 %}
8533 
8534 // Logical Shift Right by 8-bit immediate
8535 instruct shrI_eReg_imm(eRegI dst, immI8 shift, eFlagsReg cr) %{
8536   match(Set dst (URShiftI dst shift));
8537   effect(KILL cr);
8538 
8539   size(3);
8540   format %{ "SHR    $dst,$shift" %}
8541   opcode(0xC1, 0x5);  /* C1 /5 ib */
8542   ins_encode( RegOpcImm( dst, shift) );
8543   ins_pipe( ialu_reg );
8544 %}
8545 
8546 
8547 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
8548 // This idiom is used by the compiler for the i2b bytecode.
8549 instruct i2b(eRegI dst, xRegI src, immI_24 twentyfour) %{
8550   match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
8551 
8552   size(3);
8553   format %{ "MOVSX  $dst,$src :8" %}
8554   ins_encode %{
8555     __ movsbl($dst$$Register, $src$$Register);
8556   %}
8557   ins_pipe(ialu_reg_reg);
8558 %}
8559 
8560 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
8561 // This idiom is used by the compiler the i2s bytecode.
8562 instruct i2s(eRegI dst, xRegI src, immI_16 sixteen) %{
8563   match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
8564 
8565   size(3);
8566   format %{ "MOVSX  $dst,$src :16" %}
8567   ins_encode %{
8568     __ movswl($dst$$Register, $src$$Register);
8569   %}
8570   ins_pipe(ialu_reg_reg);
8571 %}
8572 
8573 
8574 // Logical Shift Right by variable
8575 instruct shrI_eReg_CL(eRegI dst, eCXRegI shift, eFlagsReg cr) %{
8576   match(Set dst (URShiftI dst shift));
8577   effect(KILL cr);
8578 
8579   size(2);
8580   format %{ "SHR    $dst,$shift" %}
8581   opcode(0xD3, 0x5);  /* D3 /5 */
8582   ins_encode( OpcP, RegOpc( dst ) );
8583   ins_pipe( ialu_reg_reg );
8584 %}
8585 
8586 
8587 //----------Logical Instructions-----------------------------------------------
8588 //----------Integer Logical Instructions---------------------------------------
8589 // And Instructions
8590 // And Register with Register
8591 instruct andI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
8592   match(Set dst (AndI dst src));
8593   effect(KILL cr);
8594 
8595   size(2);
8596   format %{ "AND    $dst,$src" %}
8597   opcode(0x23);
8598   ins_encode( OpcP, RegReg( dst, src) );
8599   ins_pipe( ialu_reg_reg );
8600 %}
8601 
8602 // And Register with Immediate
8603 instruct andI_eReg_imm(eRegI dst, immI src, eFlagsReg cr) %{
8604   match(Set dst (AndI dst src));
8605   effect(KILL cr);
8606 
8607   format %{ "AND    $dst,$src" %}
8608   opcode(0x81,0x04);  /* Opcode 81 /4 */
8609   // ins_encode( RegImm( dst, src) );
8610   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8611   ins_pipe( ialu_reg );
8612 %}
8613 
8614 // And Register with Memory
8615 instruct andI_eReg_mem(eRegI dst, memory src, eFlagsReg cr) %{
8616   match(Set dst (AndI dst (LoadI src)));
8617   effect(KILL cr);
8618 
8619   ins_cost(125);
8620   format %{ "AND    $dst,$src" %}
8621   opcode(0x23);
8622   ins_encode( OpcP, RegMem( dst, src) );
8623   ins_pipe( ialu_reg_mem );
8624 %}
8625 
8626 // And Memory with Register
8627 instruct andI_mem_eReg(memory dst, eRegI src, eFlagsReg cr) %{
8628   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
8629   effect(KILL cr);
8630 
8631   ins_cost(150);
8632   format %{ "AND    $dst,$src" %}
8633   opcode(0x21);  /* Opcode 21 /r */
8634   ins_encode( OpcP, RegMem( src, dst ) );
8635   ins_pipe( ialu_mem_reg );
8636 %}
8637 
8638 // And Memory with Immediate
8639 instruct andI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8640   match(Set dst (StoreI dst (AndI (LoadI dst) src)));
8641   effect(KILL cr);
8642 
8643   ins_cost(125);
8644   format %{ "AND    $dst,$src" %}
8645   opcode(0x81, 0x4);  /* Opcode 81 /4 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 // Or Instructions
8652 // Or Register with Register
8653 instruct orI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
8654   match(Set dst (OrI dst src));
8655   effect(KILL cr);
8656 
8657   size(2);
8658   format %{ "OR     $dst,$src" %}
8659   opcode(0x0B);
8660   ins_encode( OpcP, RegReg( dst, src) );
8661   ins_pipe( ialu_reg_reg );
8662 %}
8663 
8664 instruct orI_eReg_castP2X(eRegI dst, eRegP src, eFlagsReg cr) %{
8665   match(Set dst (OrI dst (CastP2X src)));
8666   effect(KILL cr);
8667 
8668   size(2);
8669   format %{ "OR     $dst,$src" %}
8670   opcode(0x0B);
8671   ins_encode( OpcP, RegReg( dst, src) );
8672   ins_pipe( ialu_reg_reg );
8673 %}
8674 
8675 
8676 // Or Register with Immediate
8677 instruct orI_eReg_imm(eRegI dst, immI src, eFlagsReg cr) %{
8678   match(Set dst (OrI dst src));
8679   effect(KILL cr);
8680 
8681   format %{ "OR     $dst,$src" %}
8682   opcode(0x81,0x01);  /* Opcode 81 /1 id */
8683   // ins_encode( RegImm( dst, src) );
8684   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8685   ins_pipe( ialu_reg );
8686 %}
8687 
8688 // Or Register with Memory
8689 instruct orI_eReg_mem(eRegI dst, memory src, eFlagsReg cr) %{
8690   match(Set dst (OrI dst (LoadI src)));
8691   effect(KILL cr);
8692 
8693   ins_cost(125);
8694   format %{ "OR     $dst,$src" %}
8695   opcode(0x0B);
8696   ins_encode( OpcP, RegMem( dst, src) );
8697   ins_pipe( ialu_reg_mem );
8698 %}
8699 
8700 // Or Memory with Register
8701 instruct orI_mem_eReg(memory dst, eRegI src, eFlagsReg cr) %{
8702   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
8703   effect(KILL cr);
8704 
8705   ins_cost(150);
8706   format %{ "OR     $dst,$src" %}
8707   opcode(0x09);  /* Opcode 09 /r */
8708   ins_encode( OpcP, RegMem( src, dst ) );
8709   ins_pipe( ialu_mem_reg );
8710 %}
8711 
8712 // Or Memory with Immediate
8713 instruct orI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8714   match(Set dst (StoreI dst (OrI (LoadI dst) src)));
8715   effect(KILL cr);
8716 
8717   ins_cost(125);
8718   format %{ "OR     $dst,$src" %}
8719   opcode(0x81,0x1);  /* Opcode 81 /1 id */
8720   // ins_encode( MemImm( dst, src) );
8721   ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
8722   ins_pipe( ialu_mem_imm );
8723 %}
8724 
8725 // ROL/ROR
8726 // ROL expand
8727 instruct rolI_eReg_imm1(eRegI dst, immI1 shift, eFlagsReg cr) %{
8728   effect(USE_DEF dst, USE shift, KILL cr);
8729 
8730   format %{ "ROL    $dst, $shift" %}
8731   opcode(0xD1, 0x0); /* Opcode D1 /0 */
8732   ins_encode( OpcP, RegOpc( dst ));
8733   ins_pipe( ialu_reg );
8734 %}
8735 
8736 instruct rolI_eReg_imm8(eRegI dst, immI8 shift, eFlagsReg cr) %{
8737   effect(USE_DEF dst, USE shift, KILL cr);
8738 
8739   format %{ "ROL    $dst, $shift" %}
8740   opcode(0xC1, 0x0); /*Opcode /C1  /0  */
8741   ins_encode( RegOpcImm(dst, shift) );
8742   ins_pipe(ialu_reg);
8743 %}
8744 
8745 instruct rolI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr) %{
8746   effect(USE_DEF dst, USE shift, KILL cr);
8747 
8748   format %{ "ROL    $dst, $shift" %}
8749   opcode(0xD3, 0x0);    /* Opcode D3 /0 */
8750   ins_encode(OpcP, RegOpc(dst));
8751   ins_pipe( ialu_reg_reg );
8752 %}
8753 // end of ROL expand
8754 
8755 // ROL 32bit by one once
8756 instruct rolI_eReg_i1(eRegI dst, immI1 lshift, immI_M1 rshift, eFlagsReg cr) %{
8757   match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8758 
8759   expand %{
8760     rolI_eReg_imm1(dst, lshift, cr);
8761   %}
8762 %}
8763 
8764 // ROL 32bit var by imm8 once
8765 instruct rolI_eReg_i8(eRegI dst, immI8 lshift, immI8 rshift, eFlagsReg cr) %{
8766   predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8767   match(Set dst ( OrI (LShiftI dst lshift) (URShiftI dst rshift)));
8768 
8769   expand %{
8770     rolI_eReg_imm8(dst, lshift, cr);
8771   %}
8772 %}
8773 
8774 // ROL 32bit var by var once
8775 instruct rolI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
8776   match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
8777 
8778   expand %{
8779     rolI_eReg_CL(dst, shift, cr);
8780   %}
8781 %}
8782 
8783 // ROL 32bit var by var once
8784 instruct rolI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
8785   match(Set dst ( OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
8786 
8787   expand %{
8788     rolI_eReg_CL(dst, shift, cr);
8789   %}
8790 %}
8791 
8792 // ROR expand
8793 instruct rorI_eReg_imm1(eRegI dst, immI1 shift, eFlagsReg cr) %{
8794   effect(USE_DEF dst, USE shift, KILL cr);
8795 
8796   format %{ "ROR    $dst, $shift" %}
8797   opcode(0xD1,0x1);  /* Opcode D1 /1 */
8798   ins_encode( OpcP, RegOpc( dst ) );
8799   ins_pipe( ialu_reg );
8800 %}
8801 
8802 instruct rorI_eReg_imm8(eRegI dst, immI8 shift, eFlagsReg cr) %{
8803   effect (USE_DEF dst, USE shift, KILL cr);
8804 
8805   format %{ "ROR    $dst, $shift" %}
8806   opcode(0xC1, 0x1); /* Opcode /C1 /1 ib */
8807   ins_encode( RegOpcImm(dst, shift) );
8808   ins_pipe( ialu_reg );
8809 %}
8810 
8811 instruct rorI_eReg_CL(ncxRegI dst, eCXRegI shift, eFlagsReg cr)%{
8812   effect(USE_DEF dst, USE shift, KILL cr);
8813 
8814   format %{ "ROR    $dst, $shift" %}
8815   opcode(0xD3, 0x1);    /* Opcode D3 /1 */
8816   ins_encode(OpcP, RegOpc(dst));
8817   ins_pipe( ialu_reg_reg );
8818 %}
8819 // end of ROR expand
8820 
8821 // ROR right once
8822 instruct rorI_eReg_i1(eRegI dst, immI1 rshift, immI_M1 lshift, eFlagsReg cr) %{
8823   match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8824 
8825   expand %{
8826     rorI_eReg_imm1(dst, rshift, cr);
8827   %}
8828 %}
8829 
8830 // ROR 32bit by immI8 once
8831 instruct rorI_eReg_i8(eRegI dst, immI8 rshift, immI8 lshift, eFlagsReg cr) %{
8832   predicate(  0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
8833   match(Set dst ( OrI (URShiftI dst rshift) (LShiftI dst lshift)));
8834 
8835   expand %{
8836     rorI_eReg_imm8(dst, rshift, cr);
8837   %}
8838 %}
8839 
8840 // ROR 32bit var by var once
8841 instruct rorI_eReg_Var_C0(ncxRegI dst, eCXRegI shift, immI0 zero, eFlagsReg cr) %{
8842   match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
8843 
8844   expand %{
8845     rorI_eReg_CL(dst, shift, cr);
8846   %}
8847 %}
8848 
8849 // ROR 32bit var by var once
8850 instruct rorI_eReg_Var_C32(ncxRegI dst, eCXRegI shift, immI_32 c32, eFlagsReg cr) %{
8851   match(Set dst ( OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
8852 
8853   expand %{
8854     rorI_eReg_CL(dst, shift, cr);
8855   %}
8856 %}
8857 
8858 // Xor Instructions
8859 // Xor Register with Register
8860 instruct xorI_eReg(eRegI dst, eRegI src, eFlagsReg cr) %{
8861   match(Set dst (XorI dst src));
8862   effect(KILL cr);
8863 
8864   size(2);
8865   format %{ "XOR    $dst,$src" %}
8866   opcode(0x33);
8867   ins_encode( OpcP, RegReg( dst, src) );
8868   ins_pipe( ialu_reg_reg );
8869 %}
8870 
8871 // Xor Register with Immediate -1
8872 instruct xorI_eReg_im1(eRegI dst, immI_M1 imm) %{
8873   match(Set dst (XorI dst imm));  
8874 
8875   size(2);
8876   format %{ "NOT    $dst" %}  
8877   ins_encode %{
8878      __ notl($dst$$Register);
8879   %}
8880   ins_pipe( ialu_reg );
8881 %}
8882 
8883 // Xor Register with Immediate
8884 instruct xorI_eReg_imm(eRegI dst, immI src, eFlagsReg cr) %{
8885   match(Set dst (XorI dst src));
8886   effect(KILL cr);
8887 
8888   format %{ "XOR    $dst,$src" %}
8889   opcode(0x81,0x06);  /* Opcode 81 /6 id */
8890   // ins_encode( RegImm( dst, src) );
8891   ins_encode( OpcSErm( dst, src ), Con8or32( src ) );
8892   ins_pipe( ialu_reg );
8893 %}
8894 
8895 // Xor Register with Memory
8896 instruct xorI_eReg_mem(eRegI dst, memory src, eFlagsReg cr) %{
8897   match(Set dst (XorI dst (LoadI src)));
8898   effect(KILL cr);
8899 
8900   ins_cost(125);
8901   format %{ "XOR    $dst,$src" %}
8902   opcode(0x33);
8903   ins_encode( OpcP, RegMem(dst, src) );
8904   ins_pipe( ialu_reg_mem );
8905 %}
8906 
8907 // Xor Memory with Register
8908 instruct xorI_mem_eReg(memory dst, eRegI src, eFlagsReg cr) %{
8909   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
8910   effect(KILL cr);
8911 
8912   ins_cost(150);
8913   format %{ "XOR    $dst,$src" %}
8914   opcode(0x31);  /* Opcode 31 /r */
8915   ins_encode( OpcP, RegMem( src, dst ) );
8916   ins_pipe( ialu_mem_reg );
8917 %}
8918 
8919 // Xor Memory with Immediate
8920 instruct xorI_mem_imm(memory dst, immI src, eFlagsReg cr) %{
8921   match(Set dst (StoreI dst (XorI (LoadI dst) src)));
8922   effect(KILL cr);
8923 
8924   ins_cost(125);
8925   format %{ "XOR    $dst,$src" %}
8926   opcode(0x81,0x6);  /* Opcode 81 /6 id */
8927   ins_encode( OpcSE( src ), RMopc_Mem(secondary, dst ), Con8or32( src ) );
8928   ins_pipe( ialu_mem_imm );
8929 %}
8930 
8931 //----------Convert Int to Boolean---------------------------------------------
8932 
8933 instruct movI_nocopy(eRegI dst, eRegI src) %{
8934   effect( DEF dst, USE src );
8935   format %{ "MOV    $dst,$src" %}
8936   ins_encode( enc_Copy( dst, src) );
8937   ins_pipe( ialu_reg_reg );
8938 %}
8939 
8940 instruct ci2b( eRegI dst, eRegI src, eFlagsReg cr ) %{
8941   effect( USE_DEF dst, USE src, KILL cr );
8942 
8943   size(4);
8944   format %{ "NEG    $dst\n\t"
8945             "ADC    $dst,$src" %}
8946   ins_encode( neg_reg(dst),
8947               OpcRegReg(0x13,dst,src) );
8948   ins_pipe( ialu_reg_reg_long );
8949 %}
8950 
8951 instruct convI2B( eRegI dst, eRegI src, eFlagsReg cr ) %{
8952   match(Set dst (Conv2B src));
8953 
8954   expand %{
8955     movI_nocopy(dst,src);
8956     ci2b(dst,src,cr);
8957   %}
8958 %}
8959 
8960 instruct movP_nocopy(eRegI dst, eRegP src) %{
8961   effect( DEF dst, USE src );
8962   format %{ "MOV    $dst,$src" %}
8963   ins_encode( enc_Copy( dst, src) );
8964   ins_pipe( ialu_reg_reg );
8965 %}
8966 
8967 instruct cp2b( eRegI dst, eRegP src, eFlagsReg cr ) %{
8968   effect( USE_DEF dst, USE src, KILL cr );
8969   format %{ "NEG    $dst\n\t"
8970             "ADC    $dst,$src" %}
8971   ins_encode( neg_reg(dst),
8972               OpcRegReg(0x13,dst,src) );
8973   ins_pipe( ialu_reg_reg_long );
8974 %}
8975 
8976 instruct convP2B( eRegI dst, eRegP src, eFlagsReg cr ) %{
8977   match(Set dst (Conv2B src));
8978 
8979   expand %{
8980     movP_nocopy(dst,src);
8981     cp2b(dst,src,cr);
8982   %}
8983 %}
8984 
8985 instruct cmpLTMask( eCXRegI dst, ncxRegI p, ncxRegI q, eFlagsReg cr ) %{
8986   match(Set dst (CmpLTMask p q));
8987   effect( KILL cr );
8988   ins_cost(400);
8989 
8990   // SETlt can only use low byte of EAX,EBX, ECX, or EDX as destination
8991   format %{ "XOR    $dst,$dst\n\t"
8992             "CMP    $p,$q\n\t"
8993             "SETlt  $dst\n\t"
8994             "NEG    $dst" %}
8995   ins_encode( OpcRegReg(0x33,dst,dst),
8996               OpcRegReg(0x3B,p,q),
8997               setLT_reg(dst), neg_reg(dst) );
8998   ins_pipe( pipe_slow );
8999 %}
9000 
9001 instruct cmpLTMask0( eRegI dst, immI0 zero, eFlagsReg cr ) %{
9002   match(Set dst (CmpLTMask dst zero));
9003   effect( DEF dst, KILL cr );
9004   ins_cost(100);
9005 
9006   format %{ "SAR    $dst,31" %}
9007   opcode(0xC1, 0x7);  /* C1 /7 ib */
9008   ins_encode( RegOpcImm( dst, 0x1F ) );
9009   ins_pipe( ialu_reg );
9010 %}
9011 
9012 
9013 instruct cadd_cmpLTMask( ncxRegI p, ncxRegI q, ncxRegI y, eCXRegI tmp, eFlagsReg cr ) %{
9014   match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
9015   effect( KILL tmp, KILL cr );
9016   ins_cost(400);
9017   // annoyingly, $tmp has no edges so you cant ask for it in
9018   // any format or encoding
9019   format %{ "SUB    $p,$q\n\t"
9020             "SBB    ECX,ECX\n\t"
9021             "AND    ECX,$y\n\t"
9022             "ADD    $p,ECX" %}
9023   ins_encode( enc_cmpLTP(p,q,y,tmp) );
9024   ins_pipe( pipe_cmplt );
9025 %}
9026 
9027 /* If I enable this, I encourage spilling in the inner loop of compress.
9028 instruct cadd_cmpLTMask_mem( ncxRegI p, ncxRegI q, memory y, eCXRegI tmp, eFlagsReg cr ) %{
9029   match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
9030   effect( USE_KILL tmp, KILL cr );
9031   ins_cost(400);
9032 
9033   format %{ "SUB    $p,$q\n\t"
9034             "SBB    ECX,ECX\n\t"
9035             "AND    ECX,$y\n\t"
9036             "ADD    $p,ECX" %}
9037   ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
9038 %}
9039 */
9040 
9041 //----------Long Instructions------------------------------------------------
9042 // Add Long Register with Register
9043 instruct addL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9044   match(Set dst (AddL dst src));
9045   effect(KILL cr);
9046   ins_cost(200);
9047   format %{ "ADD    $dst.lo,$src.lo\n\t"
9048             "ADC    $dst.hi,$src.hi" %}
9049   opcode(0x03, 0x13);
9050   ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
9051   ins_pipe( ialu_reg_reg_long );
9052 %}
9053 
9054 // Add Long Register with Immediate
9055 instruct addL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9056   match(Set dst (AddL dst src));
9057   effect(KILL cr);
9058   format %{ "ADD    $dst.lo,$src.lo\n\t"
9059             "ADC    $dst.hi,$src.hi" %}
9060   opcode(0x81,0x00,0x02);  /* Opcode 81 /0, 81 /2 */
9061   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9062   ins_pipe( ialu_reg_long );
9063 %}
9064 
9065 // Add Long Register with Memory
9066 instruct addL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9067   match(Set dst (AddL dst (LoadL mem)));
9068   effect(KILL cr);
9069   ins_cost(125);
9070   format %{ "ADD    $dst.lo,$mem\n\t"
9071             "ADC    $dst.hi,$mem+4" %}
9072   opcode(0x03, 0x13);
9073   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9074   ins_pipe( ialu_reg_long_mem );
9075 %}
9076 
9077 // Subtract Long Register with Register.
9078 instruct subL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9079   match(Set dst (SubL dst src));
9080   effect(KILL cr);
9081   ins_cost(200);
9082   format %{ "SUB    $dst.lo,$src.lo\n\t"
9083             "SBB    $dst.hi,$src.hi" %}
9084   opcode(0x2B, 0x1B);
9085   ins_encode( RegReg_Lo(dst, src), RegReg_Hi(dst,src) );
9086   ins_pipe( ialu_reg_reg_long );
9087 %}
9088 
9089 // Subtract Long Register with Immediate
9090 instruct subL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9091   match(Set dst (SubL dst src));
9092   effect(KILL cr);
9093   format %{ "SUB    $dst.lo,$src.lo\n\t"
9094             "SBB    $dst.hi,$src.hi" %}
9095   opcode(0x81,0x05,0x03);  /* Opcode 81 /5, 81 /3 */
9096   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9097   ins_pipe( ialu_reg_long );
9098 %}
9099 
9100 // Subtract Long Register with Memory
9101 instruct subL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9102   match(Set dst (SubL dst (LoadL mem)));
9103   effect(KILL cr);
9104   ins_cost(125);
9105   format %{ "SUB    $dst.lo,$mem\n\t"
9106             "SBB    $dst.hi,$mem+4" %}
9107   opcode(0x2B, 0x1B);
9108   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9109   ins_pipe( ialu_reg_long_mem );
9110 %}
9111 
9112 instruct negL_eReg(eRegL dst, immL0 zero, eFlagsReg cr) %{
9113   match(Set dst (SubL zero dst));
9114   effect(KILL cr);
9115   ins_cost(300);
9116   format %{ "NEG    $dst.hi\n\tNEG    $dst.lo\n\tSBB    $dst.hi,0" %}
9117   ins_encode( neg_long(dst) );
9118   ins_pipe( ialu_reg_reg_long );
9119 %}
9120 
9121 // And Long Register with Register
9122 instruct andL_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9123   match(Set dst (AndL dst src));
9124   effect(KILL cr);
9125   format %{ "AND    $dst.lo,$src.lo\n\t"
9126             "AND    $dst.hi,$src.hi" %}
9127   opcode(0x23,0x23);
9128   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9129   ins_pipe( ialu_reg_reg_long );
9130 %}
9131 
9132 // And Long Register with Immediate
9133 instruct andL_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9134   match(Set dst (AndL dst src));
9135   effect(KILL cr);
9136   format %{ "AND    $dst.lo,$src.lo\n\t"
9137             "AND    $dst.hi,$src.hi" %}
9138   opcode(0x81,0x04,0x04);  /* Opcode 81 /4, 81 /4 */
9139   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9140   ins_pipe( ialu_reg_long );
9141 %}
9142 
9143 // And Long Register with Memory
9144 instruct andL_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9145   match(Set dst (AndL dst (LoadL mem)));
9146   effect(KILL cr);
9147   ins_cost(125);
9148   format %{ "AND    $dst.lo,$mem\n\t"
9149             "AND    $dst.hi,$mem+4" %}
9150   opcode(0x23, 0x23);
9151   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9152   ins_pipe( ialu_reg_long_mem );
9153 %}
9154 
9155 // Or Long Register with Register
9156 instruct orl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9157   match(Set dst (OrL dst src));
9158   effect(KILL cr);
9159   format %{ "OR     $dst.lo,$src.lo\n\t"
9160             "OR     $dst.hi,$src.hi" %}
9161   opcode(0x0B,0x0B);
9162   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9163   ins_pipe( ialu_reg_reg_long );
9164 %}
9165 
9166 // Or Long Register with Immediate
9167 instruct orl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9168   match(Set dst (OrL dst src));
9169   effect(KILL cr);
9170   format %{ "OR     $dst.lo,$src.lo\n\t"
9171             "OR     $dst.hi,$src.hi" %}
9172   opcode(0x81,0x01,0x01);  /* Opcode 81 /1, 81 /1 */
9173   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9174   ins_pipe( ialu_reg_long );
9175 %}
9176 
9177 // Or Long Register with Memory
9178 instruct orl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9179   match(Set dst (OrL dst (LoadL mem)));
9180   effect(KILL cr);
9181   ins_cost(125);
9182   format %{ "OR     $dst.lo,$mem\n\t"
9183             "OR     $dst.hi,$mem+4" %}
9184   opcode(0x0B,0x0B);
9185   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9186   ins_pipe( ialu_reg_long_mem );
9187 %}
9188 
9189 // Xor Long Register with Register
9190 instruct xorl_eReg(eRegL dst, eRegL src, eFlagsReg cr) %{
9191   match(Set dst (XorL dst src));
9192   effect(KILL cr);
9193   format %{ "XOR    $dst.lo,$src.lo\n\t"
9194             "XOR    $dst.hi,$src.hi" %}
9195   opcode(0x33,0x33);
9196   ins_encode( RegReg_Lo( dst, src), RegReg_Hi( dst, src) );
9197   ins_pipe( ialu_reg_reg_long );
9198 %}
9199 
9200 // Xor Long Register with Immediate -1
9201 instruct xorl_eReg_im1(eRegL dst, immL_M1 imm) %{
9202   match(Set dst (XorL dst imm));  
9203   format %{ "NOT    $dst.lo\n\t"
9204             "NOT    $dst.hi" %}
9205   ins_encode %{
9206      __ notl($dst$$Register);
9207      __ notl(HIGH_FROM_LOW($dst$$Register));
9208   %}
9209   ins_pipe( ialu_reg_long );
9210 %}
9211 
9212 // Xor Long Register with Immediate
9213 instruct xorl_eReg_imm(eRegL dst, immL src, eFlagsReg cr) %{
9214   match(Set dst (XorL dst src));
9215   effect(KILL cr);
9216   format %{ "XOR    $dst.lo,$src.lo\n\t"
9217             "XOR    $dst.hi,$src.hi" %}
9218   opcode(0x81,0x06,0x06);  /* Opcode 81 /6, 81 /6 */
9219   ins_encode( Long_OpcSErm_Lo( dst, src ), Long_OpcSErm_Hi( dst, src ) );
9220   ins_pipe( ialu_reg_long );
9221 %}
9222 
9223 // Xor Long Register with Memory
9224 instruct xorl_eReg_mem(eRegL dst, load_long_memory mem, eFlagsReg cr) %{
9225   match(Set dst (XorL dst (LoadL mem)));
9226   effect(KILL cr);
9227   ins_cost(125);
9228   format %{ "XOR    $dst.lo,$mem\n\t"
9229             "XOR    $dst.hi,$mem+4" %}
9230   opcode(0x33,0x33);
9231   ins_encode( OpcP, RegMem( dst, mem), OpcS, RegMem_Hi(dst,mem) );
9232   ins_pipe( ialu_reg_long_mem );
9233 %}
9234 
9235 // Shift Left Long by 1
9236 instruct shlL_eReg_1(eRegL dst, immI_1 cnt, eFlagsReg cr) %{
9237   predicate(UseNewLongLShift);
9238   match(Set dst (LShiftL dst cnt));
9239   effect(KILL cr);
9240   ins_cost(100);
9241   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9242             "ADC    $dst.hi,$dst.hi" %}
9243   ins_encode %{
9244     __ addl($dst$$Register,$dst$$Register);
9245     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9246   %}
9247   ins_pipe( ialu_reg_long );
9248 %}
9249 
9250 // Shift Left Long by 2
9251 instruct shlL_eReg_2(eRegL dst, immI_2 cnt, eFlagsReg cr) %{
9252   predicate(UseNewLongLShift);
9253   match(Set dst (LShiftL dst cnt));
9254   effect(KILL cr);
9255   ins_cost(100);
9256   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9257             "ADC    $dst.hi,$dst.hi\n\t" 
9258             "ADD    $dst.lo,$dst.lo\n\t"
9259             "ADC    $dst.hi,$dst.hi" %}
9260   ins_encode %{
9261     __ addl($dst$$Register,$dst$$Register);
9262     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9263     __ addl($dst$$Register,$dst$$Register);
9264     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9265   %}
9266   ins_pipe( ialu_reg_long );
9267 %}
9268 
9269 // Shift Left Long by 3
9270 instruct shlL_eReg_3(eRegL dst, immI_3 cnt, eFlagsReg cr) %{
9271   predicate(UseNewLongLShift);
9272   match(Set dst (LShiftL dst cnt));
9273   effect(KILL cr);
9274   ins_cost(100);
9275   format %{ "ADD    $dst.lo,$dst.lo\n\t"
9276             "ADC    $dst.hi,$dst.hi\n\t" 
9277             "ADD    $dst.lo,$dst.lo\n\t"
9278             "ADC    $dst.hi,$dst.hi\n\t" 
9279             "ADD    $dst.lo,$dst.lo\n\t"
9280             "ADC    $dst.hi,$dst.hi" %}
9281   ins_encode %{
9282     __ addl($dst$$Register,$dst$$Register);
9283     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9284     __ addl($dst$$Register,$dst$$Register);
9285     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9286     __ addl($dst$$Register,$dst$$Register);
9287     __ adcl(HIGH_FROM_LOW($dst$$Register),HIGH_FROM_LOW($dst$$Register));
9288   %}
9289   ins_pipe( ialu_reg_long );
9290 %}
9291 
9292 // Shift Left Long by 1-31
9293 instruct shlL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9294   match(Set dst (LShiftL dst cnt));
9295   effect(KILL cr);
9296   ins_cost(200);
9297   format %{ "SHLD   $dst.hi,$dst.lo,$cnt\n\t"
9298             "SHL    $dst.lo,$cnt" %}
9299   opcode(0xC1, 0x4, 0xA4);  /* 0F/A4, then C1 /4 ib */
9300   ins_encode( move_long_small_shift(dst,cnt) );
9301   ins_pipe( ialu_reg_long );
9302 %}
9303 
9304 // Shift Left Long by 32-63
9305 instruct shlL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9306   match(Set dst (LShiftL dst cnt));
9307   effect(KILL cr);
9308   ins_cost(300);
9309   format %{ "MOV    $dst.hi,$dst.lo\n"
9310           "\tSHL    $dst.hi,$cnt-32\n"
9311           "\tXOR    $dst.lo,$dst.lo" %}
9312   opcode(0xC1, 0x4);  /* C1 /4 ib */
9313   ins_encode( move_long_big_shift_clr(dst,cnt) );
9314   ins_pipe( ialu_reg_long );
9315 %}
9316 
9317 // Shift Left Long by variable
9318 instruct salL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9319   match(Set dst (LShiftL dst shift));
9320   effect(KILL cr);
9321   ins_cost(500+200);
9322   size(17);
9323   format %{ "TEST   $shift,32\n\t"
9324             "JEQ,s  small\n\t"
9325             "MOV    $dst.hi,$dst.lo\n\t"
9326             "XOR    $dst.lo,$dst.lo\n"
9327     "small:\tSHLD   $dst.hi,$dst.lo,$shift\n\t"
9328             "SHL    $dst.lo,$shift" %}
9329   ins_encode( shift_left_long( dst, shift ) );
9330   ins_pipe( pipe_slow );
9331 %}
9332 
9333 // Shift Right Long by 1-31
9334 instruct shrL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9335   match(Set dst (URShiftL dst cnt));
9336   effect(KILL cr);
9337   ins_cost(200);
9338   format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
9339             "SHR    $dst.hi,$cnt" %}
9340   opcode(0xC1, 0x5, 0xAC);  /* 0F/AC, then C1 /5 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 shrL_eReg_32_63(eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9347   match(Set dst (URShiftL dst cnt));
9348   effect(KILL cr);
9349   ins_cost(300);
9350   format %{ "MOV    $dst.lo,$dst.hi\n"
9351           "\tSHR    $dst.lo,$cnt-32\n"
9352           "\tXOR    $dst.hi,$dst.hi" %}
9353   opcode(0xC1, 0x5);  /* C1 /5 ib */
9354   ins_encode( move_long_big_shift_clr(dst,cnt) );
9355   ins_pipe( ialu_reg_long );
9356 %}
9357 
9358 // Shift Right Long by variable
9359 instruct shrL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9360   match(Set dst (URShiftL dst shift));
9361   effect(KILL cr);
9362   ins_cost(600);
9363   size(17);
9364   format %{ "TEST   $shift,32\n\t"
9365             "JEQ,s  small\n\t"
9366             "MOV    $dst.lo,$dst.hi\n\t"
9367             "XOR    $dst.hi,$dst.hi\n"
9368     "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
9369             "SHR    $dst.hi,$shift" %}
9370   ins_encode( shift_right_long( dst, shift ) );
9371   ins_pipe( pipe_slow );
9372 %}
9373 
9374 // Shift Right Long by 1-31
9375 instruct sarL_eReg_1_31(eRegL dst, immI_1_31 cnt, eFlagsReg cr) %{
9376   match(Set dst (RShiftL dst cnt));
9377   effect(KILL cr);
9378   ins_cost(200);
9379   format %{ "SHRD   $dst.lo,$dst.hi,$cnt\n\t"
9380             "SAR    $dst.hi,$cnt" %}
9381   opcode(0xC1, 0x7, 0xAC);  /* 0F/AC, then C1 /7 ib */
9382   ins_encode( move_long_small_shift(dst,cnt) );
9383   ins_pipe( ialu_reg_long );
9384 %}
9385 
9386 // Shift Right Long by 32-63
9387 instruct sarL_eReg_32_63( eRegL dst, immI_32_63 cnt, eFlagsReg cr) %{
9388   match(Set dst (RShiftL dst cnt));
9389   effect(KILL cr);
9390   ins_cost(300);
9391   format %{ "MOV    $dst.lo,$dst.hi\n"
9392           "\tSAR    $dst.lo,$cnt-32\n"
9393           "\tSAR    $dst.hi,31" %}
9394   opcode(0xC1, 0x7);  /* C1 /7 ib */
9395   ins_encode( move_long_big_shift_sign(dst,cnt) );
9396   ins_pipe( ialu_reg_long );
9397 %}
9398 
9399 // Shift Right arithmetic Long by variable
9400 instruct sarL_eReg_CL(eRegL dst, eCXRegI shift, eFlagsReg cr) %{
9401   match(Set dst (RShiftL dst shift));
9402   effect(KILL cr);
9403   ins_cost(600);
9404   size(18);
9405   format %{ "TEST   $shift,32\n\t"
9406             "JEQ,s  small\n\t"
9407             "MOV    $dst.lo,$dst.hi\n\t"
9408             "SAR    $dst.hi,31\n"
9409     "small:\tSHRD   $dst.lo,$dst.hi,$shift\n\t"
9410             "SAR    $dst.hi,$shift" %}
9411   ins_encode( shift_right_arith_long( dst, shift ) );
9412   ins_pipe( pipe_slow );
9413 %}
9414 
9415 
9416 //----------Double Instructions------------------------------------------------
9417 // Double Math
9418 
9419 // Compare & branch
9420 
9421 // P6 version of float compare, sets condition codes in EFLAGS
9422 instruct cmpDPR_cc_P6(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
9423   predicate(VM_Version::supports_cmov() && UseSSE <=1);
9424   match(Set cr (CmpD src1 src2));
9425   effect(KILL rax);
9426   ins_cost(150);
9427   format %{ "FLD    $src1\n\t"
9428             "FUCOMIP ST,$src2  // P6 instruction\n\t"
9429             "JNP    exit\n\t"
9430             "MOV    ah,1       // saw a NaN, set CF\n\t"
9431             "SAHF\n"
9432      "exit:\tNOP               // avoid branch to branch" %}
9433   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
9434   ins_encode( Push_Reg_DPR(src1),
9435               OpcP, RegOpc(src2),
9436               cmpF_P6_fixup );
9437   ins_pipe( pipe_slow );
9438 %}
9439 
9440 instruct cmpDPR_cc_P6CF(eFlagsRegUCF cr, regDPR src1, regDPR src2) %{
9441   predicate(VM_Version::supports_cmov() && UseSSE <=1);
9442   match(Set cr (CmpD src1 src2));
9443   ins_cost(150);
9444   format %{ "FLD    $src1\n\t"
9445             "FUCOMIP ST,$src2  // P6 instruction" %}
9446   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
9447   ins_encode( Push_Reg_DPR(src1),
9448               OpcP, RegOpc(src2));
9449   ins_pipe( pipe_slow );
9450 %}
9451 
9452 // Compare & branch
9453 instruct cmpDPR_cc(eFlagsRegU cr, regDPR src1, regDPR src2, eAXRegI rax) %{
9454   predicate(UseSSE<=1);
9455   match(Set cr (CmpD src1 src2));
9456   effect(KILL rax);
9457   ins_cost(200);
9458   format %{ "FLD    $src1\n\t"
9459             "FCOMp  $src2\n\t"
9460             "FNSTSW AX\n\t"
9461             "TEST   AX,0x400\n\t"
9462             "JZ,s   flags\n\t"
9463             "MOV    AH,1\t# unordered treat as LT\n"
9464     "flags:\tSAHF" %}
9465   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
9466   ins_encode( Push_Reg_DPR(src1),
9467               OpcP, RegOpc(src2),
9468               fpu_flags);
9469   ins_pipe( pipe_slow );
9470 %}
9471 
9472 // Compare vs zero into -1,0,1
9473 instruct cmpDPR_0(eRegI dst, regDPR src1, immDPR0 zero, eAXRegI rax, eFlagsReg cr) %{
9474   predicate(UseSSE<=1);
9475   match(Set dst (CmpD3 src1 zero));
9476   effect(KILL cr, KILL rax);
9477   ins_cost(280);
9478   format %{ "FTSTD  $dst,$src1" %}
9479   opcode(0xE4, 0xD9);
9480   ins_encode( Push_Reg_DPR(src1),
9481               OpcS, OpcP, PopFPU,
9482               CmpF_Result(dst));
9483   ins_pipe( pipe_slow );
9484 %}
9485 
9486 // Compare into -1,0,1
9487 instruct cmpDPR_reg(eRegI dst, regDPR src1, regDPR src2, eAXRegI rax, eFlagsReg cr) %{
9488   predicate(UseSSE<=1);
9489   match(Set dst (CmpD3 src1 src2));
9490   effect(KILL cr, KILL rax);
9491   ins_cost(300);
9492   format %{ "FCMPD  $dst,$src1,$src2" %}
9493   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
9494   ins_encode( Push_Reg_DPR(src1),
9495               OpcP, RegOpc(src2),
9496               CmpF_Result(dst));
9497   ins_pipe( pipe_slow );
9498 %}
9499 
9500 // float compare and set condition codes in EFLAGS by XMM regs
9501 instruct cmpD_cc(eFlagsRegU cr, regD src1, regD src2) %{
9502   predicate(UseSSE>=2);
9503   match(Set cr (CmpD src1 src2));
9504   ins_cost(145);
9505   format %{ "UCOMISD $src1,$src2\n\t"
9506             "JNP,s   exit\n\t"
9507             "PUSHF\t# saw NaN, set CF\n\t"
9508             "AND     [rsp], #0xffffff2b\n\t"
9509             "POPF\n"
9510     "exit:" %}
9511   ins_encode %{
9512     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9513     emit_cmpfp_fixup(_masm);
9514   %}
9515   ins_pipe( pipe_slow );
9516 %}
9517 
9518 instruct cmpD_ccCF(eFlagsRegUCF cr, regD src1, regD src2) %{
9519   predicate(UseSSE>=2);
9520   match(Set cr (CmpD src1 src2));
9521   ins_cost(100);
9522   format %{ "UCOMISD $src1,$src2" %}
9523   ins_encode %{
9524     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9525   %}
9526   ins_pipe( pipe_slow );
9527 %}
9528 
9529 // float compare and set condition codes in EFLAGS by XMM regs
9530 instruct cmpD_ccmem(eFlagsRegU cr, regD src1, memory src2) %{
9531   predicate(UseSSE>=2);
9532   match(Set cr (CmpD src1 (LoadD src2)));
9533   ins_cost(145);
9534   format %{ "UCOMISD $src1,$src2\n\t"
9535             "JNP,s   exit\n\t"
9536             "PUSHF\t# saw NaN, set CF\n\t"
9537             "AND     [rsp], #0xffffff2b\n\t"
9538             "POPF\n"
9539     "exit:" %}
9540   ins_encode %{
9541     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9542     emit_cmpfp_fixup(_masm);
9543   %}
9544   ins_pipe( pipe_slow );
9545 %}
9546 
9547 instruct cmpD_ccmemCF(eFlagsRegUCF cr, regD src1, memory src2) %{
9548   predicate(UseSSE>=2);
9549   match(Set cr (CmpD src1 (LoadD src2)));
9550   ins_cost(100);
9551   format %{ "UCOMISD $src1,$src2" %}
9552   ins_encode %{
9553     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9554   %}
9555   ins_pipe( pipe_slow );
9556 %}
9557 
9558 // Compare into -1,0,1 in XMM
9559 instruct cmpD_reg(xRegI dst, regD src1, regD src2, eFlagsReg cr) %{
9560   predicate(UseSSE>=2);
9561   match(Set dst (CmpD3 src1 src2));
9562   effect(KILL cr);
9563   ins_cost(255);
9564   format %{ "UCOMISD $src1, $src2\n\t"
9565             "MOV     $dst, #-1\n\t"
9566             "JP,s    done\n\t"
9567             "JB,s    done\n\t"
9568             "SETNE   $dst\n\t"
9569             "MOVZB   $dst, $dst\n"
9570     "done:" %}
9571   ins_encode %{
9572     __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
9573     emit_cmpfp3(_masm, $dst$$Register);
9574   %}
9575   ins_pipe( pipe_slow );
9576 %}
9577 
9578 // Compare into -1,0,1 in XMM and memory
9579 instruct cmpD_regmem(xRegI dst, regD src1, memory src2, eFlagsReg cr) %{
9580   predicate(UseSSE>=2);
9581   match(Set dst (CmpD3 src1 (LoadD src2)));
9582   effect(KILL cr);
9583   ins_cost(275);
9584   format %{ "UCOMISD $src1, $src2\n\t"
9585             "MOV     $dst, #-1\n\t"
9586             "JP,s    done\n\t"
9587             "JB,s    done\n\t"
9588             "SETNE   $dst\n\t"
9589             "MOVZB   $dst, $dst\n"
9590     "done:" %}
9591   ins_encode %{
9592     __ ucomisd($src1$$XMMRegister, $src2$$Address);
9593     emit_cmpfp3(_masm, $dst$$Register);
9594   %}
9595   ins_pipe( pipe_slow );
9596 %}
9597 
9598 
9599 instruct subDPR_reg(regDPR dst, regDPR src) %{
9600   predicate (UseSSE <=1);
9601   match(Set dst (SubD dst src));
9602 
9603   format %{ "FLD    $src\n\t"
9604             "DSUBp  $dst,ST" %}
9605   opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
9606   ins_cost(150);
9607   ins_encode( Push_Reg_DPR(src),
9608               OpcP, RegOpc(dst) );
9609   ins_pipe( fpu_reg_reg );
9610 %}
9611 
9612 instruct subDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9613   predicate (UseSSE <=1);
9614   match(Set dst (RoundDouble (SubD src1 src2)));
9615   ins_cost(250);
9616 
9617   format %{ "FLD    $src2\n\t"
9618             "DSUB   ST,$src1\n\t"
9619             "FSTP_D $dst\t# D-round" %}
9620   opcode(0xD8, 0x5);
9621   ins_encode( Push_Reg_DPR(src2),
9622               OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
9623   ins_pipe( fpu_mem_reg_reg );
9624 %}
9625 
9626 
9627 instruct subDPR_reg_mem(regDPR dst, memory src) %{
9628   predicate (UseSSE <=1);
9629   match(Set dst (SubD dst (LoadD src)));
9630   ins_cost(150);
9631 
9632   format %{ "FLD    $src\n\t"
9633             "DSUBp  $dst,ST" %}
9634   opcode(0xDE, 0x5, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
9635   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9636               OpcP, RegOpc(dst) );
9637   ins_pipe( fpu_reg_mem );
9638 %}
9639 
9640 instruct absDPR_reg(regDPR1 dst, regDPR1 src) %{
9641   predicate (UseSSE<=1);
9642   match(Set dst (AbsD src));
9643   ins_cost(100);
9644   format %{ "FABS" %}
9645   opcode(0xE1, 0xD9);
9646   ins_encode( OpcS, OpcP );
9647   ins_pipe( fpu_reg_reg );
9648 %}
9649 
9650 instruct negDPR_reg(regDPR1 dst, regDPR1 src) %{
9651   predicate(UseSSE<=1);
9652   match(Set dst (NegD src));
9653   ins_cost(100);
9654   format %{ "FCHS" %}
9655   opcode(0xE0, 0xD9);
9656   ins_encode( OpcS, OpcP );
9657   ins_pipe( fpu_reg_reg );
9658 %}
9659 
9660 instruct addDPR_reg(regDPR dst, regDPR src) %{
9661   predicate(UseSSE<=1);
9662   match(Set dst (AddD dst src));
9663   format %{ "FLD    $src\n\t"
9664             "DADD   $dst,ST" %}
9665   size(4);
9666   ins_cost(150);
9667   opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
9668   ins_encode( Push_Reg_DPR(src),
9669               OpcP, RegOpc(dst) );
9670   ins_pipe( fpu_reg_reg );
9671 %}
9672 
9673 
9674 instruct addDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9675   predicate(UseSSE<=1);
9676   match(Set dst (RoundDouble (AddD src1 src2)));
9677   ins_cost(250);
9678 
9679   format %{ "FLD    $src2\n\t"
9680             "DADD   ST,$src1\n\t"
9681             "FSTP_D $dst\t# D-round" %}
9682   opcode(0xD8, 0x0); /* D8 C0+i or D8 /0*/
9683   ins_encode( Push_Reg_DPR(src2),
9684               OpcP, RegOpc(src1), Pop_Mem_DPR(dst) );
9685   ins_pipe( fpu_mem_reg_reg );
9686 %}
9687 
9688 
9689 instruct addDPR_reg_mem(regDPR dst, memory src) %{
9690   predicate(UseSSE<=1);
9691   match(Set dst (AddD dst (LoadD src)));
9692   ins_cost(150);
9693 
9694   format %{ "FLD    $src\n\t"
9695             "DADDp  $dst,ST" %}
9696   opcode(0xDE, 0x0, 0xDD); /* DE C0+i */  /* LoadD  DD /0 */
9697   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9698               OpcP, RegOpc(dst) );
9699   ins_pipe( fpu_reg_mem );
9700 %}
9701 
9702 // add-to-memory
9703 instruct addDPR_mem_reg(memory dst, regDPR src) %{
9704   predicate(UseSSE<=1);
9705   match(Set dst (StoreD dst (RoundDouble (AddD (LoadD dst) src))));
9706   ins_cost(150);
9707 
9708   format %{ "FLD_D  $dst\n\t"
9709             "DADD   ST,$src\n\t"
9710             "FST_D  $dst" %}
9711   opcode(0xDD, 0x0);
9712   ins_encode( Opcode(0xDD), RMopc_Mem(0x00,dst),
9713               Opcode(0xD8), RegOpc(src),
9714               set_instruction_start,
9715               Opcode(0xDD), RMopc_Mem(0x03,dst) );
9716   ins_pipe( fpu_reg_mem );
9717 %}
9718 
9719 instruct addDPR_reg_imm1(regDPR dst, immDPR1 con) %{
9720   predicate(UseSSE<=1);
9721   match(Set dst (AddD dst con));
9722   ins_cost(125);
9723   format %{ "FLD1\n\t"
9724             "DADDp  $dst,ST" %}
9725   ins_encode %{
9726     __ fld1();
9727     __ faddp($dst$$reg);
9728   %}
9729   ins_pipe(fpu_reg);
9730 %}
9731 
9732 instruct addDPR_reg_imm(regDPR dst, immDPR con) %{
9733   predicate(UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
9734   match(Set dst (AddD dst con));
9735   ins_cost(200);
9736   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9737             "DADDp  $dst,ST" %}
9738   ins_encode %{
9739     __ fld_d($constantaddress($con));
9740     __ faddp($dst$$reg);
9741   %}
9742   ins_pipe(fpu_reg_mem);
9743 %}
9744 
9745 instruct addDPR_reg_imm_round(stackSlotD dst, regDPR src, immDPR con) %{
9746   predicate(UseSSE<=1 && _kids[0]->_kids[1]->_leaf->getd() != 0.0 && _kids[0]->_kids[1]->_leaf->getd() != 1.0 );
9747   match(Set dst (RoundDouble (AddD src con)));
9748   ins_cost(200);
9749   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9750             "DADD   ST,$src\n\t"
9751             "FSTP_D $dst\t# D-round" %}
9752   ins_encode %{
9753     __ fld_d($constantaddress($con));
9754     __ fadd($src$$reg);
9755     __ fstp_d(Address(rsp, $dst$$disp));
9756   %}
9757   ins_pipe(fpu_mem_reg_con);
9758 %}
9759 
9760 instruct mulDPR_reg(regDPR dst, regDPR src) %{
9761   predicate(UseSSE<=1);
9762   match(Set dst (MulD dst src));
9763   format %{ "FLD    $src\n\t"
9764             "DMULp  $dst,ST" %}
9765   opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
9766   ins_cost(150);
9767   ins_encode( Push_Reg_DPR(src),
9768               OpcP, RegOpc(dst) );
9769   ins_pipe( fpu_reg_reg );
9770 %}
9771 
9772 // Strict FP instruction biases argument before multiply then
9773 // biases result to avoid double rounding of subnormals.
9774 //
9775 // scale arg1 by multiplying arg1 by 2^(-15360)
9776 // load arg2
9777 // multiply scaled arg1 by arg2
9778 // rescale product by 2^(15360)
9779 //
9780 instruct strictfp_mulDPR_reg(regDPR1 dst, regnotDPR1 src) %{
9781   predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
9782   match(Set dst (MulD dst src));
9783   ins_cost(1);   // Select this instruction for all strict FP double multiplies
9784 
9785   format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
9786             "DMULp  $dst,ST\n\t"
9787             "FLD    $src\n\t"
9788             "DMULp  $dst,ST\n\t"
9789             "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
9790             "DMULp  $dst,ST\n\t" %}
9791   opcode(0xDE, 0x1); /* DE C8+i or DE /1*/
9792   ins_encode( strictfp_bias1(dst),
9793               Push_Reg_DPR(src),
9794               OpcP, RegOpc(dst),
9795               strictfp_bias2(dst) );
9796   ins_pipe( fpu_reg_reg );
9797 %}
9798 
9799 instruct mulDPR_reg_imm(regDPR dst, immDPR con) %{
9800   predicate( UseSSE<=1 && _kids[1]->_leaf->getd() != 0.0 && _kids[1]->_leaf->getd() != 1.0 );
9801   match(Set dst (MulD dst con));
9802   ins_cost(200);
9803   format %{ "FLD_D  [$constantaddress]\t# load from constant table: double=$con\n\t"
9804             "DMULp  $dst,ST" %}
9805   ins_encode %{
9806     __ fld_d($constantaddress($con));
9807     __ fmulp($dst$$reg);
9808   %}
9809   ins_pipe(fpu_reg_mem);
9810 %}
9811 
9812 
9813 instruct mulDPR_reg_mem(regDPR dst, memory src) %{
9814   predicate( UseSSE<=1 );
9815   match(Set dst (MulD dst (LoadD src)));
9816   ins_cost(200);
9817   format %{ "FLD_D  $src\n\t"
9818             "DMULp  $dst,ST" %}
9819   opcode(0xDE, 0x1, 0xDD); /* DE C8+i or DE /1*/  /* LoadD  DD /0 */
9820   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
9821               OpcP, RegOpc(dst) );
9822   ins_pipe( fpu_reg_mem );
9823 %}
9824 
9825 //
9826 // Cisc-alternate to reg-reg multiply
9827 instruct mulDPR_reg_mem_cisc(regDPR dst, regDPR src, memory mem) %{
9828   predicate( UseSSE<=1 );
9829   match(Set dst (MulD src (LoadD mem)));
9830   ins_cost(250);
9831   format %{ "FLD_D  $mem\n\t"
9832             "DMUL   ST,$src\n\t"
9833             "FSTP_D $dst" %}
9834   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadD D9 /0 */
9835   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem),
9836               OpcReg_FPR(src),
9837               Pop_Reg_DPR(dst) );
9838   ins_pipe( fpu_reg_reg_mem );
9839 %}
9840 
9841 
9842 // MACRO3 -- addDPR a mulDPR
9843 // This instruction is a '2-address' instruction in that the result goes
9844 // back to src2.  This eliminates a move from the macro; possibly the
9845 // register allocator will have to add it back (and maybe not).
9846 instruct addDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
9847   predicate( UseSSE<=1 );
9848   match(Set src2 (AddD (MulD src0 src1) src2));
9849   format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
9850             "DMUL   ST,$src1\n\t"
9851             "DADDp  $src2,ST" %}
9852   ins_cost(250);
9853   opcode(0xDD); /* LoadD DD /0 */
9854   ins_encode( Push_Reg_FPR(src0),
9855               FMul_ST_reg(src1),
9856               FAddP_reg_ST(src2) );
9857   ins_pipe( fpu_reg_reg_reg );
9858 %}
9859 
9860 
9861 // MACRO3 -- subDPR a mulDPR
9862 instruct subDPR_mulDPR_reg(regDPR src2, regDPR src1, regDPR src0) %{
9863   predicate( UseSSE<=1 );
9864   match(Set src2 (SubD (MulD src0 src1) src2));
9865   format %{ "FLD    $src0\t# ===MACRO3d===\n\t"
9866             "DMUL   ST,$src1\n\t"
9867             "DSUBRp $src2,ST" %}
9868   ins_cost(250);
9869   ins_encode( Push_Reg_FPR(src0),
9870               FMul_ST_reg(src1),
9871               Opcode(0xDE), Opc_plus(0xE0,src2));
9872   ins_pipe( fpu_reg_reg_reg );
9873 %}
9874 
9875 
9876 instruct divDPR_reg(regDPR dst, regDPR src) %{
9877   predicate( UseSSE<=1 );
9878   match(Set dst (DivD dst src));
9879 
9880   format %{ "FLD    $src\n\t"
9881             "FDIVp  $dst,ST" %}
9882   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
9883   ins_cost(150);
9884   ins_encode( Push_Reg_DPR(src),
9885               OpcP, RegOpc(dst) );
9886   ins_pipe( fpu_reg_reg );
9887 %}
9888 
9889 // Strict FP instruction biases argument before division then
9890 // biases result, to avoid double rounding of subnormals.
9891 //
9892 // scale dividend by multiplying dividend by 2^(-15360)
9893 // load divisor
9894 // divide scaled dividend by divisor
9895 // rescale quotient by 2^(15360)
9896 //
9897 instruct strictfp_divDPR_reg(regDPR1 dst, regnotDPR1 src) %{
9898   predicate (UseSSE<=1);
9899   match(Set dst (DivD dst src));
9900   predicate( UseSSE<=1 && Compile::current()->has_method() && Compile::current()->method()->is_strict() );
9901   ins_cost(01);
9902 
9903   format %{ "FLD    StubRoutines::_fpu_subnormal_bias1\n\t"
9904             "DMULp  $dst,ST\n\t"
9905             "FLD    $src\n\t"
9906             "FDIVp  $dst,ST\n\t"
9907             "FLD    StubRoutines::_fpu_subnormal_bias2\n\t"
9908             "DMULp  $dst,ST\n\t" %}
9909   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
9910   ins_encode( strictfp_bias1(dst),
9911               Push_Reg_DPR(src),
9912               OpcP, RegOpc(dst),
9913               strictfp_bias2(dst) );
9914   ins_pipe( fpu_reg_reg );
9915 %}
9916 
9917 instruct divDPR_reg_round(stackSlotD dst, regDPR src1, regDPR src2) %{
9918   predicate( UseSSE<=1 && !(Compile::current()->has_method() && Compile::current()->method()->is_strict()) );
9919   match(Set dst (RoundDouble (DivD src1 src2)));
9920 
9921   format %{ "FLD    $src1\n\t"
9922             "FDIV   ST,$src2\n\t"
9923             "FSTP_D $dst\t# D-round" %}
9924   opcode(0xD8, 0x6); /* D8 F0+i or D8 /6 */
9925   ins_encode( Push_Reg_DPR(src1),
9926               OpcP, RegOpc(src2), Pop_Mem_DPR(dst) );
9927   ins_pipe( fpu_mem_reg_reg );
9928 %}
9929 
9930 
9931 instruct modDPR_reg(regDPR dst, regDPR src, eAXRegI rax, eFlagsReg cr) %{
9932   predicate(UseSSE<=1);
9933   match(Set dst (ModD dst src));
9934   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
9935 
9936   format %{ "DMOD   $dst,$src" %}
9937   ins_cost(250);
9938   ins_encode(Push_Reg_Mod_DPR(dst, src),
9939               emitModDPR(),
9940               Push_Result_Mod_DPR(src),
9941               Pop_Reg_DPR(dst));
9942   ins_pipe( pipe_slow );
9943 %}
9944 
9945 instruct modD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eFlagsReg cr) %{
9946   predicate(UseSSE>=2);
9947   match(Set dst (ModD src0 src1));
9948   effect(KILL rax, KILL cr);
9949 
9950   format %{ "SUB    ESP,8\t # DMOD\n"
9951           "\tMOVSD  [ESP+0],$src1\n"
9952           "\tFLD_D  [ESP+0]\n"
9953           "\tMOVSD  [ESP+0],$src0\n"
9954           "\tFLD_D  [ESP+0]\n"
9955      "loop:\tFPREM\n"
9956           "\tFWAIT\n"
9957           "\tFNSTSW AX\n"
9958           "\tSAHF\n"
9959           "\tJP     loop\n"
9960           "\tFSTP_D [ESP+0]\n"
9961           "\tMOVSD  $dst,[ESP+0]\n"
9962           "\tADD    ESP,8\n"
9963           "\tFSTP   ST0\t # Restore FPU Stack"
9964     %}
9965   ins_cost(250);
9966   ins_encode( Push_ModD_encoding(src0, src1), emitModDPR(), Push_ResultD(dst), PopFPU);
9967   ins_pipe( pipe_slow );
9968 %}
9969 
9970 instruct sinDPR_reg(regDPR1 dst, regDPR1 src) %{
9971   predicate (UseSSE<=1);
9972   match(Set dst (SinD src));
9973   ins_cost(1800);
9974   format %{ "DSIN   $dst" %}
9975   opcode(0xD9, 0xFE);
9976   ins_encode( OpcP, OpcS );
9977   ins_pipe( pipe_slow );
9978 %}
9979 
9980 instruct sinD_reg(regD dst, eFlagsReg cr) %{
9981   predicate (UseSSE>=2);
9982   match(Set dst (SinD dst));
9983   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
9984   ins_cost(1800);
9985   format %{ "DSIN   $dst" %}
9986   opcode(0xD9, 0xFE);
9987   ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
9988   ins_pipe( pipe_slow );
9989 %}
9990 
9991 instruct cosDPR_reg(regDPR1 dst, regDPR1 src) %{
9992   predicate (UseSSE<=1);
9993   match(Set dst (CosD src));
9994   ins_cost(1800);
9995   format %{ "DCOS   $dst" %}
9996   opcode(0xD9, 0xFF);
9997   ins_encode( OpcP, OpcS );
9998   ins_pipe( pipe_slow );
9999 %}
10000 
10001 instruct cosD_reg(regD dst, eFlagsReg cr) %{
10002   predicate (UseSSE>=2);
10003   match(Set dst (CosD dst));
10004   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
10005   ins_cost(1800);
10006   format %{ "DCOS   $dst" %}
10007   opcode(0xD9, 0xFF);
10008   ins_encode( Push_SrcD(dst), OpcP, OpcS, Push_ResultD(dst) );
10009   ins_pipe( pipe_slow );
10010 %}
10011 
10012 instruct tanDPR_reg(regDPR1 dst, regDPR1 src) %{
10013   predicate (UseSSE<=1);
10014   match(Set dst(TanD src));
10015   format %{ "DTAN   $dst" %}
10016   ins_encode( Opcode(0xD9), Opcode(0xF2),    // fptan
10017               Opcode(0xDD), Opcode(0xD8));   // fstp st
10018   ins_pipe( pipe_slow );
10019 %}
10020 
10021 instruct tanD_reg(regD dst, eFlagsReg cr) %{
10022   predicate (UseSSE>=2);
10023   match(Set dst(TanD dst));
10024   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
10025   format %{ "DTAN   $dst" %}
10026   ins_encode( Push_SrcD(dst),
10027               Opcode(0xD9), Opcode(0xF2),    // fptan
10028               Opcode(0xDD), Opcode(0xD8),   // fstp st
10029               Push_ResultD(dst) );
10030   ins_pipe( pipe_slow );
10031 %}
10032 
10033 instruct atanDPR_reg(regDPR dst, regDPR src) %{
10034   predicate (UseSSE<=1);
10035   match(Set dst(AtanD dst src));
10036   format %{ "DATA   $dst,$src" %}
10037   opcode(0xD9, 0xF3);
10038   ins_encode( Push_Reg_DPR(src),
10039               OpcP, OpcS, RegOpc(dst) );
10040   ins_pipe( pipe_slow );
10041 %}
10042 
10043 instruct atanD_reg(regD dst, regD src, eFlagsReg cr) %{
10044   predicate (UseSSE>=2);
10045   match(Set dst(AtanD dst src));
10046   effect(KILL cr); // Push_{Src|Result}D() uses "{SUB|ADD} ESP,8"
10047   format %{ "DATA   $dst,$src" %}
10048   opcode(0xD9, 0xF3);
10049   ins_encode( Push_SrcD(src),
10050               OpcP, OpcS, Push_ResultD(dst) );
10051   ins_pipe( pipe_slow );
10052 %}
10053 
10054 instruct sqrtDPR_reg(regDPR dst, regDPR src) %{
10055   predicate (UseSSE<=1);
10056   match(Set dst (SqrtD src));
10057   format %{ "DSQRT  $dst,$src" %}
10058   opcode(0xFA, 0xD9);
10059   ins_encode( Push_Reg_DPR(src),
10060               OpcS, OpcP, Pop_Reg_DPR(dst) );
10061   ins_pipe( pipe_slow );
10062 %}
10063 
10064 instruct powDPR_reg(regDPR X, regDPR1 Y, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10065   predicate (UseSSE<=1);
10066   match(Set Y (PowD X Y));  // Raise X to the Yth power
10067   effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
10068   format %{ "fast_pow $X $Y -> $Y  // KILL $rax, $rcx, $rdx" %}
10069   ins_encode %{
10070     __ subptr(rsp, 8);
10071     __ fld_s($X$$reg - 1);
10072     __ fast_pow();
10073     __ addptr(rsp, 8);
10074   %}
10075   ins_pipe( pipe_slow );
10076 %}
10077 
10078 instruct powD_reg(regD dst, regD src0, regD src1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10079   predicate (UseSSE>=2);
10080   match(Set dst (PowD src0 src1));  // Raise src0 to the src1'th power
10081   effect(KILL rax, KILL rdx, KILL rcx, KILL cr);
10082   format %{ "fast_pow $src0 $src1 -> $dst  // KILL $rax, $rcx, $rdx" %}
10083   ins_encode %{
10084     __ subptr(rsp, 8);
10085     __ movdbl(Address(rsp, 0), $src1$$XMMRegister);
10086     __ fld_d(Address(rsp, 0));
10087     __ movdbl(Address(rsp, 0), $src0$$XMMRegister);
10088     __ fld_d(Address(rsp, 0));
10089     __ fast_pow();
10090     __ fstp_d(Address(rsp, 0));
10091     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
10092     __ addptr(rsp, 8);
10093   %}
10094   ins_pipe( pipe_slow );
10095 %}
10096 
10097 
10098 instruct expDPR_reg(regDPR1 dpr1, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10099   predicate (UseSSE<=1);
10100   match(Set dpr1 (ExpD dpr1));
10101   effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
10102   format %{ "fast_exp $dpr1 -> $dpr1  // KILL $rax, $rcx, $rdx" %}
10103   ins_encode %{
10104     __ fast_exp();
10105   %}
10106   ins_pipe( pipe_slow );
10107 %}
10108 
10109 instruct expD_reg(regD dst, regD src, eAXRegI rax, eDXRegI rdx, eCXRegI rcx, eFlagsReg cr) %{
10110   predicate (UseSSE>=2);
10111   match(Set dst (ExpD src));
10112   effect(KILL rax, KILL rcx, KILL rdx, KILL cr);
10113   format %{ "fast_exp $dst -> $src  // KILL $rax, $rcx, $rdx" %}
10114   ins_encode %{
10115     __ subptr(rsp, 8);
10116     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
10117     __ fld_d(Address(rsp, 0));
10118     __ fast_exp();
10119     __ fstp_d(Address(rsp, 0));
10120     __ movdbl($dst$$XMMRegister, Address(rsp, 0));
10121     __ addptr(rsp, 8);
10122   %}
10123   ins_pipe( pipe_slow );
10124 %}
10125 
10126 instruct log10DPR_reg(regDPR1 dst, regDPR1 src) %{
10127   predicate (UseSSE<=1);
10128   // The source Double operand on FPU stack
10129   match(Set dst (Log10D src));
10130   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
10131   // fxch         ; swap ST(0) with ST(1)
10132   // fyl2x        ; compute log_10(2) * log_2(x)
10133   format %{ "FLDLG2 \t\t\t#Log10\n\t"
10134             "FXCH   \n\t"
10135             "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
10136          %}
10137   ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
10138               Opcode(0xD9), Opcode(0xC9),   // fxch
10139               Opcode(0xD9), Opcode(0xF1));  // fyl2x
10140 
10141   ins_pipe( pipe_slow );
10142 %}
10143 
10144 instruct log10D_reg(regD dst, regD src, eFlagsReg cr) %{
10145   predicate (UseSSE>=2);
10146   effect(KILL cr);
10147   match(Set dst (Log10D src));
10148   // fldlg2       ; push log_10(2) on the FPU stack; full 80-bit number
10149   // fyl2x        ; compute log_10(2) * log_2(x)
10150   format %{ "FLDLG2 \t\t\t#Log10\n\t"
10151             "FYL2X  \t\t\t# Q=Log10*Log_2(x)"
10152          %}
10153   ins_encode( Opcode(0xD9), Opcode(0xEC),   // fldlg2
10154               Push_SrcD(src),
10155               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10156               Push_ResultD(dst));
10157 
10158   ins_pipe( pipe_slow );
10159 %}
10160 
10161 instruct logDPR_reg(regDPR1 dst, regDPR1 src) %{
10162   predicate (UseSSE<=1);
10163   // The source Double operand on FPU stack
10164   match(Set dst (LogD src));
10165   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
10166   // fxch         ; swap ST(0) with ST(1)
10167   // fyl2x        ; compute log_e(2) * log_2(x)
10168   format %{ "FLDLN2 \t\t\t#Log_e\n\t"
10169             "FXCH   \n\t"
10170             "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
10171          %}
10172   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
10173               Opcode(0xD9), Opcode(0xC9),   // fxch
10174               Opcode(0xD9), Opcode(0xF1));  // fyl2x
10175 
10176   ins_pipe( pipe_slow );
10177 %}
10178 
10179 instruct logD_reg(regD dst, regD src, eFlagsReg cr) %{
10180   predicate (UseSSE>=2);
10181   effect(KILL cr);
10182   // The source and result Double operands in XMM registers
10183   match(Set dst (LogD src));
10184   // fldln2       ; push log_e(2) on the FPU stack; full 80-bit number
10185   // fyl2x        ; compute log_e(2) * log_2(x)
10186   format %{ "FLDLN2 \t\t\t#Log_e\n\t"
10187             "FYL2X  \t\t\t# Q=Log_e*Log_2(x)"
10188          %}
10189   ins_encode( Opcode(0xD9), Opcode(0xED),   // fldln2
10190               Push_SrcD(src),
10191               Opcode(0xD9), Opcode(0xF1),   // fyl2x
10192               Push_ResultD(dst));
10193   ins_pipe( pipe_slow );
10194 %}
10195 
10196 //-------------Float Instructions-------------------------------
10197 // Float Math
10198 
10199 // Code for float compare:
10200 //     fcompp();
10201 //     fwait(); fnstsw_ax();
10202 //     sahf();
10203 //     movl(dst, unordered_result);
10204 //     jcc(Assembler::parity, exit);
10205 //     movl(dst, less_result);
10206 //     jcc(Assembler::below, exit);
10207 //     movl(dst, equal_result);
10208 //     jcc(Assembler::equal, exit);
10209 //     movl(dst, greater_result);
10210 //   exit:
10211 
10212 // P6 version of float compare, sets condition codes in EFLAGS
10213 instruct cmpFPR_cc_P6(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
10214   predicate(VM_Version::supports_cmov() && UseSSE == 0);
10215   match(Set cr (CmpF src1 src2));
10216   effect(KILL rax);
10217   ins_cost(150);
10218   format %{ "FLD    $src1\n\t"
10219             "FUCOMIP ST,$src2  // P6 instruction\n\t"
10220             "JNP    exit\n\t"
10221             "MOV    ah,1       // saw a NaN, set CF (treat as LT)\n\t"
10222             "SAHF\n"
10223      "exit:\tNOP               // avoid branch to branch" %}
10224   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
10225   ins_encode( Push_Reg_DPR(src1),
10226               OpcP, RegOpc(src2),
10227               cmpF_P6_fixup );
10228   ins_pipe( pipe_slow );
10229 %}
10230 
10231 instruct cmpFPR_cc_P6CF(eFlagsRegUCF cr, regFPR src1, regFPR src2) %{
10232   predicate(VM_Version::supports_cmov() && UseSSE == 0);
10233   match(Set cr (CmpF src1 src2));
10234   ins_cost(100);
10235   format %{ "FLD    $src1\n\t"
10236             "FUCOMIP ST,$src2  // P6 instruction" %}
10237   opcode(0xDF, 0x05); /* DF E8+i or DF /5 */
10238   ins_encode( Push_Reg_DPR(src1),
10239               OpcP, RegOpc(src2));
10240   ins_pipe( pipe_slow );
10241 %}
10242 
10243 
10244 // Compare & branch
10245 instruct cmpFPR_cc(eFlagsRegU cr, regFPR src1, regFPR src2, eAXRegI rax) %{
10246   predicate(UseSSE == 0);
10247   match(Set cr (CmpF src1 src2));
10248   effect(KILL rax);
10249   ins_cost(200);
10250   format %{ "FLD    $src1\n\t"
10251             "FCOMp  $src2\n\t"
10252             "FNSTSW AX\n\t"
10253             "TEST   AX,0x400\n\t"
10254             "JZ,s   flags\n\t"
10255             "MOV    AH,1\t# unordered treat as LT\n"
10256     "flags:\tSAHF" %}
10257   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
10258   ins_encode( Push_Reg_DPR(src1),
10259               OpcP, RegOpc(src2),
10260               fpu_flags);
10261   ins_pipe( pipe_slow );
10262 %}
10263 
10264 // Compare vs zero into -1,0,1
10265 instruct cmpFPR_0(eRegI dst, regFPR src1, immFPR0 zero, eAXRegI rax, eFlagsReg cr) %{
10266   predicate(UseSSE == 0);
10267   match(Set dst (CmpF3 src1 zero));
10268   effect(KILL cr, KILL rax);
10269   ins_cost(280);
10270   format %{ "FTSTF  $dst,$src1" %}
10271   opcode(0xE4, 0xD9);
10272   ins_encode( Push_Reg_DPR(src1),
10273               OpcS, OpcP, PopFPU,
10274               CmpF_Result(dst));
10275   ins_pipe( pipe_slow );
10276 %}
10277 
10278 // Compare into -1,0,1
10279 instruct cmpFPR_reg(eRegI dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
10280   predicate(UseSSE == 0);
10281   match(Set dst (CmpF3 src1 src2));
10282   effect(KILL cr, KILL rax);
10283   ins_cost(300);
10284   format %{ "FCMPF  $dst,$src1,$src2" %}
10285   opcode(0xD8, 0x3); /* D8 D8+i or D8 /3 */
10286   ins_encode( Push_Reg_DPR(src1),
10287               OpcP, RegOpc(src2),
10288               CmpF_Result(dst));
10289   ins_pipe( pipe_slow );
10290 %}
10291 
10292 // float compare and set condition codes in EFLAGS by XMM regs
10293 instruct cmpF_cc(eFlagsRegU cr, regF src1, regF src2) %{
10294   predicate(UseSSE>=1);
10295   match(Set cr (CmpF src1 src2));
10296   ins_cost(145);
10297   format %{ "UCOMISS $src1,$src2\n\t"
10298             "JNP,s   exit\n\t"
10299             "PUSHF\t# saw NaN, set CF\n\t"
10300             "AND     [rsp], #0xffffff2b\n\t"
10301             "POPF\n"
10302     "exit:" %}
10303   ins_encode %{
10304     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10305     emit_cmpfp_fixup(_masm);
10306   %}
10307   ins_pipe( pipe_slow );
10308 %}
10309 
10310 instruct cmpF_ccCF(eFlagsRegUCF cr, regF src1, regF src2) %{
10311   predicate(UseSSE>=1);
10312   match(Set cr (CmpF src1 src2));
10313   ins_cost(100);
10314   format %{ "UCOMISS $src1,$src2" %}
10315   ins_encode %{
10316     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10317   %}
10318   ins_pipe( pipe_slow );
10319 %}
10320 
10321 // float compare and set condition codes in EFLAGS by XMM regs
10322 instruct cmpF_ccmem(eFlagsRegU cr, regF src1, memory src2) %{
10323   predicate(UseSSE>=1);
10324   match(Set cr (CmpF src1 (LoadF src2)));
10325   ins_cost(165);
10326   format %{ "UCOMISS $src1,$src2\n\t"
10327             "JNP,s   exit\n\t"
10328             "PUSHF\t# saw NaN, set CF\n\t"
10329             "AND     [rsp], #0xffffff2b\n\t"
10330             "POPF\n"
10331     "exit:" %}
10332   ins_encode %{
10333     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10334     emit_cmpfp_fixup(_masm);
10335   %}
10336   ins_pipe( pipe_slow );
10337 %}
10338 
10339 instruct cmpF_ccmemCF(eFlagsRegUCF cr, regF src1, memory src2) %{
10340   predicate(UseSSE>=1);
10341   match(Set cr (CmpF src1 (LoadF src2)));
10342   ins_cost(100);
10343   format %{ "UCOMISS $src1,$src2" %}
10344   ins_encode %{
10345     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10346   %}
10347   ins_pipe( pipe_slow );
10348 %}
10349 
10350 // Compare into -1,0,1 in XMM
10351 instruct cmpF_reg(xRegI dst, regF src1, regF src2, eFlagsReg cr) %{
10352   predicate(UseSSE>=1);
10353   match(Set dst (CmpF3 src1 src2));
10354   effect(KILL cr);
10355   ins_cost(255);
10356   format %{ "UCOMISS $src1, $src2\n\t"
10357             "MOV     $dst, #-1\n\t"
10358             "JP,s    done\n\t"
10359             "JB,s    done\n\t"
10360             "SETNE   $dst\n\t"
10361             "MOVZB   $dst, $dst\n"
10362     "done:" %}
10363   ins_encode %{
10364     __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10365     emit_cmpfp3(_masm, $dst$$Register);
10366   %}
10367   ins_pipe( pipe_slow );
10368 %}
10369 
10370 // Compare into -1,0,1 in XMM and memory
10371 instruct cmpF_regmem(xRegI dst, regF src1, memory src2, eFlagsReg cr) %{
10372   predicate(UseSSE>=1);
10373   match(Set dst (CmpF3 src1 (LoadF src2)));
10374   effect(KILL cr);
10375   ins_cost(275);
10376   format %{ "UCOMISS $src1, $src2\n\t"
10377             "MOV     $dst, #-1\n\t"
10378             "JP,s    done\n\t"
10379             "JB,s    done\n\t"
10380             "SETNE   $dst\n\t"
10381             "MOVZB   $dst, $dst\n"
10382     "done:" %}
10383   ins_encode %{
10384     __ ucomiss($src1$$XMMRegister, $src2$$Address);
10385     emit_cmpfp3(_masm, $dst$$Register);
10386   %}
10387   ins_pipe( pipe_slow );
10388 %}
10389 
10390 // Spill to obtain 24-bit precision
10391 instruct subFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10392   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10393   match(Set dst (SubF src1 src2));
10394 
10395   format %{ "FSUB   $dst,$src1 - $src2" %}
10396   opcode(0xD8, 0x4); /* D8 E0+i or D8 /4 mod==0x3 ;; result in TOS */
10397   ins_encode( Push_Reg_FPR(src1),
10398               OpcReg_FPR(src2),
10399               Pop_Mem_FPR(dst) );
10400   ins_pipe( fpu_mem_reg_reg );
10401 %}
10402 //
10403 // This instruction does not round to 24-bits
10404 instruct subFPR_reg(regFPR dst, regFPR src) %{
10405   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10406   match(Set dst (SubF dst src));
10407 
10408   format %{ "FSUB   $dst,$src" %}
10409   opcode(0xDE, 0x5); /* DE E8+i  or DE /5 */
10410   ins_encode( Push_Reg_FPR(src),
10411               OpcP, RegOpc(dst) );
10412   ins_pipe( fpu_reg_reg );
10413 %}
10414 
10415 // Spill to obtain 24-bit precision
10416 instruct addFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10417   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10418   match(Set dst (AddF src1 src2));
10419 
10420   format %{ "FADD   $dst,$src1,$src2" %}
10421   opcode(0xD8, 0x0); /* D8 C0+i */
10422   ins_encode( Push_Reg_FPR(src2),
10423               OpcReg_FPR(src1),
10424               Pop_Mem_FPR(dst) );
10425   ins_pipe( fpu_mem_reg_reg );
10426 %}
10427 //
10428 // This instruction does not round to 24-bits
10429 instruct addFPR_reg(regFPR dst, regFPR src) %{
10430   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10431   match(Set dst (AddF dst src));
10432 
10433   format %{ "FLD    $src\n\t"
10434             "FADDp  $dst,ST" %}
10435   opcode(0xDE, 0x0); /* DE C0+i or DE /0*/
10436   ins_encode( Push_Reg_FPR(src),
10437               OpcP, RegOpc(dst) );
10438   ins_pipe( fpu_reg_reg );
10439 %}
10440 
10441 instruct absFPR_reg(regFPR1 dst, regFPR1 src) %{
10442   predicate(UseSSE==0);
10443   match(Set dst (AbsF src));
10444   ins_cost(100);
10445   format %{ "FABS" %}
10446   opcode(0xE1, 0xD9);
10447   ins_encode( OpcS, OpcP );
10448   ins_pipe( fpu_reg_reg );
10449 %}
10450 
10451 instruct negFPR_reg(regFPR1 dst, regFPR1 src) %{
10452   predicate(UseSSE==0);
10453   match(Set dst (NegF src));
10454   ins_cost(100);
10455   format %{ "FCHS" %}
10456   opcode(0xE0, 0xD9);
10457   ins_encode( OpcS, OpcP );
10458   ins_pipe( fpu_reg_reg );
10459 %}
10460 
10461 // Cisc-alternate to addFPR_reg
10462 // Spill to obtain 24-bit precision
10463 instruct addFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
10464   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10465   match(Set dst (AddF src1 (LoadF src2)));
10466 
10467   format %{ "FLD    $src2\n\t"
10468             "FADD   ST,$src1\n\t"
10469             "FSTP_S $dst" %}
10470   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10471   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10472               OpcReg_FPR(src1),
10473               Pop_Mem_FPR(dst) );
10474   ins_pipe( fpu_mem_reg_mem );
10475 %}
10476 //
10477 // Cisc-alternate to addFPR_reg
10478 // This instruction does not round to 24-bits
10479 instruct addFPR_reg_mem(regFPR dst, memory src) %{
10480   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10481   match(Set dst (AddF dst (LoadF src)));
10482 
10483   format %{ "FADD   $dst,$src" %}
10484   opcode(0xDE, 0x0, 0xD9); /* DE C0+i or DE /0*/  /* LoadF  D9 /0 */
10485   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src),
10486               OpcP, RegOpc(dst) );
10487   ins_pipe( fpu_reg_mem );
10488 %}
10489 
10490 // // Following two instructions for _222_mpegaudio
10491 // Spill to obtain 24-bit precision
10492 instruct addFPR24_mem_reg(stackSlotF dst, regFPR src2, memory src1 ) %{
10493   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10494   match(Set dst (AddF src1 src2));
10495 
10496   format %{ "FADD   $dst,$src1,$src2" %}
10497   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10498   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src1),
10499               OpcReg_FPR(src2),
10500               Pop_Mem_FPR(dst) );
10501   ins_pipe( fpu_mem_reg_mem );
10502 %}
10503 
10504 // Cisc-spill variant
10505 // Spill to obtain 24-bit precision
10506 instruct addFPR24_mem_cisc(stackSlotF dst, memory src1, memory src2) %{
10507   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10508   match(Set dst (AddF src1 (LoadF src2)));
10509 
10510   format %{ "FADD   $dst,$src1,$src2 cisc" %}
10511   opcode(0xD8, 0x0, 0xD9); /* D8 C0+i */  /* LoadF  D9 /0 */
10512   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10513               set_instruction_start,
10514               OpcP, RMopc_Mem(secondary,src1),
10515               Pop_Mem_FPR(dst) );
10516   ins_pipe( fpu_mem_mem_mem );
10517 %}
10518 
10519 // Spill to obtain 24-bit precision
10520 instruct addFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
10521   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10522   match(Set dst (AddF src1 src2));
10523 
10524   format %{ "FADD   $dst,$src1,$src2" %}
10525   opcode(0xD8, 0x0, 0xD9); /* D8 /0 */  /* LoadF  D9 /0 */
10526   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10527               set_instruction_start,
10528               OpcP, RMopc_Mem(secondary,src1),
10529               Pop_Mem_FPR(dst) );
10530   ins_pipe( fpu_mem_mem_mem );
10531 %}
10532 
10533 
10534 // Spill to obtain 24-bit precision
10535 instruct addFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
10536   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10537   match(Set dst (AddF src con));
10538   format %{ "FLD    $src\n\t"
10539             "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10540             "FSTP_S $dst"  %}
10541   ins_encode %{
10542     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10543     __ fadd_s($constantaddress($con));
10544     __ fstp_s(Address(rsp, $dst$$disp));
10545   %}
10546   ins_pipe(fpu_mem_reg_con);
10547 %}
10548 //
10549 // This instruction does not round to 24-bits
10550 instruct addFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
10551   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10552   match(Set dst (AddF src con));
10553   format %{ "FLD    $src\n\t"
10554             "FADD_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10555             "FSTP   $dst"  %}
10556   ins_encode %{
10557     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10558     __ fadd_s($constantaddress($con));
10559     __ fstp_d($dst$$reg);
10560   %}
10561   ins_pipe(fpu_reg_reg_con);
10562 %}
10563 
10564 // Spill to obtain 24-bit precision
10565 instruct mulFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10566   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10567   match(Set dst (MulF src1 src2));
10568 
10569   format %{ "FLD    $src1\n\t"
10570             "FMUL   $src2\n\t"
10571             "FSTP_S $dst"  %}
10572   opcode(0xD8, 0x1); /* D8 C8+i or D8 /1 ;; result in TOS */
10573   ins_encode( Push_Reg_FPR(src1),
10574               OpcReg_FPR(src2),
10575               Pop_Mem_FPR(dst) );
10576   ins_pipe( fpu_mem_reg_reg );
10577 %}
10578 //
10579 // This instruction does not round to 24-bits
10580 instruct mulFPR_reg(regFPR dst, regFPR src1, regFPR src2) %{
10581   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10582   match(Set dst (MulF src1 src2));
10583 
10584   format %{ "FLD    $src1\n\t"
10585             "FMUL   $src2\n\t"
10586             "FSTP_S $dst"  %}
10587   opcode(0xD8, 0x1); /* D8 C8+i */
10588   ins_encode( Push_Reg_FPR(src2),
10589               OpcReg_FPR(src1),
10590               Pop_Reg_FPR(dst) );
10591   ins_pipe( fpu_reg_reg_reg );
10592 %}
10593 
10594 
10595 // Spill to obtain 24-bit precision
10596 // Cisc-alternate to reg-reg multiply
10597 instruct mulFPR24_reg_mem(stackSlotF dst, regFPR src1, memory src2) %{
10598   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10599   match(Set dst (MulF src1 (LoadF src2)));
10600 
10601   format %{ "FLD_S  $src2\n\t"
10602             "FMUL   $src1\n\t"
10603             "FSTP_S $dst"  %}
10604   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or DE /1*/  /* LoadF D9 /0 */
10605   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10606               OpcReg_FPR(src1),
10607               Pop_Mem_FPR(dst) );
10608   ins_pipe( fpu_mem_reg_mem );
10609 %}
10610 //
10611 // This instruction does not round to 24-bits
10612 // Cisc-alternate to reg-reg multiply
10613 instruct mulFPR_reg_mem(regFPR dst, regFPR src1, memory src2) %{
10614   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10615   match(Set dst (MulF src1 (LoadF src2)));
10616 
10617   format %{ "FMUL   $dst,$src1,$src2" %}
10618   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i */  /* LoadF D9 /0 */
10619   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10620               OpcReg_FPR(src1),
10621               Pop_Reg_FPR(dst) );
10622   ins_pipe( fpu_reg_reg_mem );
10623 %}
10624 
10625 // Spill to obtain 24-bit precision
10626 instruct mulFPR24_mem_mem(stackSlotF dst, memory src1, memory src2) %{
10627   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10628   match(Set dst (MulF src1 src2));
10629 
10630   format %{ "FMUL   $dst,$src1,$src2" %}
10631   opcode(0xD8, 0x1, 0xD9); /* D8 /1 */  /* LoadF D9 /0 */
10632   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,src2),
10633               set_instruction_start,
10634               OpcP, RMopc_Mem(secondary,src1),
10635               Pop_Mem_FPR(dst) );
10636   ins_pipe( fpu_mem_mem_mem );
10637 %}
10638 
10639 // Spill to obtain 24-bit precision
10640 instruct mulFPR24_reg_imm(stackSlotF dst, regFPR src, immFPR con) %{
10641   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10642   match(Set dst (MulF src con));
10643 
10644   format %{ "FLD    $src\n\t"
10645             "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10646             "FSTP_S $dst"  %}
10647   ins_encode %{
10648     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10649     __ fmul_s($constantaddress($con));
10650     __ fstp_s(Address(rsp, $dst$$disp));
10651   %}
10652   ins_pipe(fpu_mem_reg_con);
10653 %}
10654 //
10655 // This instruction does not round to 24-bits
10656 instruct mulFPR_reg_imm(regFPR dst, regFPR src, immFPR con) %{
10657   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10658   match(Set dst (MulF src con));
10659 
10660   format %{ "FLD    $src\n\t"
10661             "FMUL_S [$constantaddress]\t# load from constant table: float=$con\n\t"
10662             "FSTP   $dst"  %}
10663   ins_encode %{
10664     __ fld_s($src$$reg - 1);  // FLD ST(i-1)
10665     __ fmul_s($constantaddress($con));
10666     __ fstp_d($dst$$reg);
10667   %}
10668   ins_pipe(fpu_reg_reg_con);
10669 %}
10670 
10671 
10672 //
10673 // MACRO1 -- subsume unshared load into mulFPR
10674 // This instruction does not round to 24-bits
10675 instruct mulFPR_reg_load1(regFPR dst, regFPR src, memory mem1 ) %{
10676   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10677   match(Set dst (MulF (LoadF mem1) src));
10678 
10679   format %{ "FLD    $mem1    ===MACRO1===\n\t"
10680             "FMUL   ST,$src\n\t"
10681             "FSTP   $dst" %}
10682   opcode(0xD8, 0x1, 0xD9); /* D8 C8+i or D8 /1 */  /* LoadF D9 /0 */
10683   ins_encode( Opcode(tertiary), RMopc_Mem(0x00,mem1),
10684               OpcReg_FPR(src),
10685               Pop_Reg_FPR(dst) );
10686   ins_pipe( fpu_reg_reg_mem );
10687 %}
10688 //
10689 // MACRO2 -- addFPR a mulFPR which subsumed an unshared load
10690 // This instruction does not round to 24-bits
10691 instruct addFPR_mulFPR_reg_load1(regFPR dst, memory mem1, regFPR src1, regFPR src2) %{
10692   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10693   match(Set dst (AddF (MulF (LoadF mem1) src1) src2));
10694   ins_cost(95);
10695 
10696   format %{ "FLD    $mem1     ===MACRO2===\n\t"
10697             "FMUL   ST,$src1  subsume mulFPR left load\n\t"
10698             "FADD   ST,$src2\n\t"
10699             "FSTP   $dst" %}
10700   opcode(0xD9); /* LoadF D9 /0 */
10701   ins_encode( OpcP, RMopc_Mem(0x00,mem1),
10702               FMul_ST_reg(src1),
10703               FAdd_ST_reg(src2),
10704               Pop_Reg_FPR(dst) );
10705   ins_pipe( fpu_reg_mem_reg_reg );
10706 %}
10707 
10708 // MACRO3 -- addFPR a mulFPR
10709 // This instruction does not round to 24-bits.  It is a '2-address'
10710 // instruction in that the result goes back to src2.  This eliminates
10711 // a move from the macro; possibly the register allocator will have
10712 // to add it back (and maybe not).
10713 instruct addFPR_mulFPR_reg(regFPR src2, regFPR src1, regFPR src0) %{
10714   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10715   match(Set src2 (AddF (MulF src0 src1) src2));
10716 
10717   format %{ "FLD    $src0     ===MACRO3===\n\t"
10718             "FMUL   ST,$src1\n\t"
10719             "FADDP  $src2,ST" %}
10720   opcode(0xD9); /* LoadF D9 /0 */
10721   ins_encode( Push_Reg_FPR(src0),
10722               FMul_ST_reg(src1),
10723               FAddP_reg_ST(src2) );
10724   ins_pipe( fpu_reg_reg_reg );
10725 %}
10726 
10727 // MACRO4 -- divFPR subFPR
10728 // This instruction does not round to 24-bits
10729 instruct subFPR_divFPR_reg(regFPR dst, regFPR src1, regFPR src2, regFPR src3) %{
10730   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10731   match(Set dst (DivF (SubF src2 src1) src3));
10732 
10733   format %{ "FLD    $src2   ===MACRO4===\n\t"
10734             "FSUB   ST,$src1\n\t"
10735             "FDIV   ST,$src3\n\t"
10736             "FSTP  $dst" %}
10737   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
10738   ins_encode( Push_Reg_FPR(src2),
10739               subFPR_divFPR_encode(src1,src3),
10740               Pop_Reg_FPR(dst) );
10741   ins_pipe( fpu_reg_reg_reg_reg );
10742 %}
10743 
10744 // Spill to obtain 24-bit precision
10745 instruct divFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2) %{
10746   predicate(UseSSE==0 && Compile::current()->select_24_bit_instr());
10747   match(Set dst (DivF src1 src2));
10748 
10749   format %{ "FDIV   $dst,$src1,$src2" %}
10750   opcode(0xD8, 0x6); /* D8 F0+i or DE /6*/
10751   ins_encode( Push_Reg_FPR(src1),
10752               OpcReg_FPR(src2),
10753               Pop_Mem_FPR(dst) );
10754   ins_pipe( fpu_mem_reg_reg );
10755 %}
10756 //
10757 // This instruction does not round to 24-bits
10758 instruct divFPR_reg(regFPR dst, regFPR src) %{
10759   predicate(UseSSE==0 && !Compile::current()->select_24_bit_instr());
10760   match(Set dst (DivF dst src));
10761 
10762   format %{ "FDIV   $dst,$src" %}
10763   opcode(0xDE, 0x7); /* DE F8+i or DE /7*/
10764   ins_encode( Push_Reg_FPR(src),
10765               OpcP, RegOpc(dst) );
10766   ins_pipe( fpu_reg_reg );
10767 %}
10768 
10769 
10770 // Spill to obtain 24-bit precision
10771 instruct modFPR24_reg(stackSlotF dst, regFPR src1, regFPR src2, eAXRegI rax, eFlagsReg cr) %{
10772   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
10773   match(Set dst (ModF src1 src2));
10774   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
10775 
10776   format %{ "FMOD   $dst,$src1,$src2" %}
10777   ins_encode( Push_Reg_Mod_DPR(src1, src2),
10778               emitModDPR(),
10779               Push_Result_Mod_DPR(src2),
10780               Pop_Mem_FPR(dst));
10781   ins_pipe( pipe_slow );
10782 %}
10783 //
10784 // This instruction does not round to 24-bits
10785 instruct modFPR_reg(regFPR dst, regFPR src, eAXRegI rax, eFlagsReg cr) %{
10786   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
10787   match(Set dst (ModF dst src));
10788   effect(KILL rax, KILL cr); // emitModDPR() uses EAX and EFLAGS
10789 
10790   format %{ "FMOD   $dst,$src" %}
10791   ins_encode(Push_Reg_Mod_DPR(dst, src),
10792               emitModDPR(),
10793               Push_Result_Mod_DPR(src),
10794               Pop_Reg_FPR(dst));
10795   ins_pipe( pipe_slow );
10796 %}
10797 
10798 instruct modF_reg(regF dst, regF src0, regF src1, eAXRegI rax, eFlagsReg cr) %{
10799   predicate(UseSSE>=1);
10800   match(Set dst (ModF src0 src1));
10801   effect(KILL rax, KILL cr);
10802   format %{ "SUB    ESP,4\t # FMOD\n"
10803           "\tMOVSS  [ESP+0],$src1\n"
10804           "\tFLD_S  [ESP+0]\n"
10805           "\tMOVSS  [ESP+0],$src0\n"
10806           "\tFLD_S  [ESP+0]\n"
10807      "loop:\tFPREM\n"
10808           "\tFWAIT\n"
10809           "\tFNSTSW AX\n"
10810           "\tSAHF\n"
10811           "\tJP     loop\n"
10812           "\tFSTP_S [ESP+0]\n"
10813           "\tMOVSS  $dst,[ESP+0]\n"
10814           "\tADD    ESP,4\n"
10815           "\tFSTP   ST0\t # Restore FPU Stack"
10816     %}
10817   ins_cost(250);
10818   ins_encode( Push_ModF_encoding(src0, src1), emitModDPR(), Push_ResultF(dst,0x4), PopFPU);
10819   ins_pipe( pipe_slow );
10820 %}
10821 
10822 
10823 //----------Arithmetic Conversion Instructions---------------------------------
10824 // The conversions operations are all Alpha sorted.  Please keep it that way!
10825 
10826 instruct roundFloat_mem_reg(stackSlotF dst, regFPR src) %{
10827   predicate(UseSSE==0);
10828   match(Set dst (RoundFloat src));
10829   ins_cost(125);
10830   format %{ "FST_S  $dst,$src\t# F-round" %}
10831   ins_encode( Pop_Mem_Reg_FPR(dst, src) );
10832   ins_pipe( fpu_mem_reg );
10833 %}
10834 
10835 instruct roundDouble_mem_reg(stackSlotD dst, regDPR src) %{
10836   predicate(UseSSE<=1);
10837   match(Set dst (RoundDouble src));
10838   ins_cost(125);
10839   format %{ "FST_D  $dst,$src\t# D-round" %}
10840   ins_encode( Pop_Mem_Reg_DPR(dst, src) );
10841   ins_pipe( fpu_mem_reg );
10842 %}
10843 
10844 // Force rounding to 24-bit precision and 6-bit exponent
10845 instruct convDPR2FPR_reg(stackSlotF dst, regDPR src) %{
10846   predicate(UseSSE==0);
10847   match(Set dst (ConvD2F src));
10848   format %{ "FST_S  $dst,$src\t# F-round" %}
10849   expand %{
10850     roundFloat_mem_reg(dst,src);
10851   %}
10852 %}
10853 
10854 // Force rounding to 24-bit precision and 6-bit exponent
10855 instruct convDPR2F_reg(regF dst, regDPR src, eFlagsReg cr) %{
10856   predicate(UseSSE==1);
10857   match(Set dst (ConvD2F src));
10858   effect( KILL cr );
10859   format %{ "SUB    ESP,4\n\t"
10860             "FST_S  [ESP],$src\t# F-round\n\t"
10861             "MOVSS  $dst,[ESP]\n\t"
10862             "ADD ESP,4" %}
10863   ins_encode %{
10864     __ subptr(rsp, 4);
10865     if ($src$$reg != FPR1L_enc) {
10866       __ fld_s($src$$reg-1);
10867       __ fstp_s(Address(rsp, 0));
10868     } else {
10869       __ fst_s(Address(rsp, 0));
10870     }
10871     __ movflt($dst$$XMMRegister, Address(rsp, 0));
10872     __ addptr(rsp, 4);
10873   %}
10874   ins_pipe( pipe_slow );
10875 %}
10876 
10877 // Force rounding double precision to single precision
10878 instruct convD2F_reg(regF dst, regD src) %{
10879   predicate(UseSSE>=2);
10880   match(Set dst (ConvD2F src));
10881   format %{ "CVTSD2SS $dst,$src\t# F-round" %}
10882   ins_encode %{
10883     __ cvtsd2ss ($dst$$XMMRegister, $src$$XMMRegister);
10884   %}
10885   ins_pipe( pipe_slow );
10886 %}
10887 
10888 instruct convFPR2DPR_reg_reg(regDPR dst, regFPR src) %{
10889   predicate(UseSSE==0);
10890   match(Set dst (ConvF2D src));
10891   format %{ "FST_S  $dst,$src\t# D-round" %}
10892   ins_encode( Pop_Reg_Reg_DPR(dst, src));
10893   ins_pipe( fpu_reg_reg );
10894 %}
10895 
10896 instruct convFPR2D_reg(stackSlotD dst, regFPR src) %{
10897   predicate(UseSSE==1);
10898   match(Set dst (ConvF2D src));
10899   format %{ "FST_D  $dst,$src\t# D-round" %}
10900   expand %{
10901     roundDouble_mem_reg(dst,src);
10902   %}
10903 %}
10904 
10905 instruct convF2DPR_reg(regDPR dst, regF src, eFlagsReg cr) %{
10906   predicate(UseSSE==1);
10907   match(Set dst (ConvF2D src));
10908   effect( KILL cr );
10909   format %{ "SUB    ESP,4\n\t"
10910             "MOVSS  [ESP] $src\n\t"
10911             "FLD_S  [ESP]\n\t"
10912             "ADD    ESP,4\n\t"
10913             "FSTP   $dst\t# D-round" %}
10914   ins_encode %{
10915     __ subptr(rsp, 4);
10916     __ movflt(Address(rsp, 0), $src$$XMMRegister);
10917     __ fld_s(Address(rsp, 0));
10918     __ addptr(rsp, 4);
10919     __ fstp_d($dst$$reg);
10920   %}
10921   ins_pipe( pipe_slow );
10922 %}
10923 
10924 instruct convF2D_reg(regD dst, regF src) %{
10925   predicate(UseSSE>=2);
10926   match(Set dst (ConvF2D src));
10927   format %{ "CVTSS2SD $dst,$src\t# D-round" %}
10928   ins_encode %{
10929     __ cvtss2sd ($dst$$XMMRegister, $src$$XMMRegister);
10930   %}
10931   ins_pipe( pipe_slow );
10932 %}
10933 
10934 // Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
10935 instruct convDPR2I_reg_reg( eAXRegI dst, eDXRegI tmp, regDPR src, eFlagsReg cr ) %{
10936   predicate(UseSSE<=1);
10937   match(Set dst (ConvD2I src));
10938   effect( KILL tmp, KILL cr );
10939   format %{ "FLD    $src\t# Convert double to int \n\t"
10940             "FLDCW  trunc mode\n\t"
10941             "SUB    ESP,4\n\t"
10942             "FISTp  [ESP + #0]\n\t"
10943             "FLDCW  std/24-bit mode\n\t"
10944             "POP    EAX\n\t"
10945             "CMP    EAX,0x80000000\n\t"
10946             "JNE,s  fast\n\t"
10947             "FLD_D  $src\n\t"
10948             "CALL   d2i_wrapper\n"
10949       "fast:" %}
10950   ins_encode( Push_Reg_DPR(src), DPR2I_encoding(src) );
10951   ins_pipe( pipe_slow );
10952 %}
10953 
10954 // Convert a double to an int.  If the double is a NAN, stuff a zero in instead.
10955 instruct convD2I_reg_reg( eAXRegI dst, eDXRegI tmp, regD src, eFlagsReg cr ) %{
10956   predicate(UseSSE>=2);
10957   match(Set dst (ConvD2I src));
10958   effect( KILL tmp, KILL cr );
10959   format %{ "CVTTSD2SI $dst, $src\n\t"
10960             "CMP    $dst,0x80000000\n\t"
10961             "JNE,s  fast\n\t"
10962             "SUB    ESP, 8\n\t"
10963             "MOVSD  [ESP], $src\n\t"
10964             "FLD_D  [ESP]\n\t"
10965             "ADD    ESP, 8\n\t"
10966             "CALL   d2i_wrapper\n"
10967       "fast:" %}
10968   ins_encode %{
10969     Label fast;
10970     __ cvttsd2sil($dst$$Register, $src$$XMMRegister);
10971     __ cmpl($dst$$Register, 0x80000000);
10972     __ jccb(Assembler::notEqual, fast);
10973     __ subptr(rsp, 8);
10974     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
10975     __ fld_d(Address(rsp, 0));
10976     __ addptr(rsp, 8);
10977     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
10978     __ bind(fast);
10979   %}
10980   ins_pipe( pipe_slow );
10981 %}
10982 
10983 instruct convDPR2L_reg_reg( eADXRegL dst, regDPR src, eFlagsReg cr ) %{
10984   predicate(UseSSE<=1);
10985   match(Set dst (ConvD2L src));
10986   effect( KILL cr );
10987   format %{ "FLD    $src\t# Convert double to long\n\t"
10988             "FLDCW  trunc mode\n\t"
10989             "SUB    ESP,8\n\t"
10990             "FISTp  [ESP + #0]\n\t"
10991             "FLDCW  std/24-bit mode\n\t"
10992             "POP    EAX\n\t"
10993             "POP    EDX\n\t"
10994             "CMP    EDX,0x80000000\n\t"
10995             "JNE,s  fast\n\t"
10996             "TEST   EAX,EAX\n\t"
10997             "JNE,s  fast\n\t"
10998             "FLD    $src\n\t"
10999             "CALL   d2l_wrapper\n"
11000       "fast:" %}
11001   ins_encode( Push_Reg_DPR(src),  DPR2L_encoding(src) );
11002   ins_pipe( pipe_slow );
11003 %}
11004 
11005 // XMM lacks a float/double->long conversion, so use the old FPU stack.
11006 instruct convD2L_reg_reg( eADXRegL dst, regD src, eFlagsReg cr ) %{
11007   predicate (UseSSE>=2);
11008   match(Set dst (ConvD2L src));
11009   effect( KILL cr );
11010   format %{ "SUB    ESP,8\t# Convert double to long\n\t"
11011             "MOVSD  [ESP],$src\n\t"
11012             "FLD_D  [ESP]\n\t"
11013             "FLDCW  trunc mode\n\t"
11014             "FISTp  [ESP + #0]\n\t"
11015             "FLDCW  std/24-bit mode\n\t"
11016             "POP    EAX\n\t"
11017             "POP    EDX\n\t"
11018             "CMP    EDX,0x80000000\n\t"
11019             "JNE,s  fast\n\t"
11020             "TEST   EAX,EAX\n\t"
11021             "JNE,s  fast\n\t"
11022             "SUB    ESP,8\n\t"
11023             "MOVSD  [ESP],$src\n\t"
11024             "FLD_D  [ESP]\n\t"
11025             "ADD    ESP,8\n\t"
11026             "CALL   d2l_wrapper\n"
11027       "fast:" %}
11028   ins_encode %{
11029     Label fast;
11030     __ subptr(rsp, 8);
11031     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
11032     __ fld_d(Address(rsp, 0));
11033     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
11034     __ fistp_d(Address(rsp, 0));
11035     // Restore the rounding mode, mask the exception
11036     if (Compile::current()->in_24_bit_fp_mode()) {
11037       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
11038     } else {
11039       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
11040     }
11041     // Load the converted long, adjust CPU stack
11042     __ pop(rax);
11043     __ pop(rdx);
11044     __ cmpl(rdx, 0x80000000);
11045     __ jccb(Assembler::notEqual, fast);
11046     __ testl(rax, rax);
11047     __ jccb(Assembler::notEqual, fast);
11048     __ subptr(rsp, 8);
11049     __ movdbl(Address(rsp, 0), $src$$XMMRegister);
11050     __ fld_d(Address(rsp, 0));
11051     __ addptr(rsp, 8);
11052     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
11053     __ bind(fast);
11054   %}
11055   ins_pipe( pipe_slow );
11056 %}
11057 
11058 // Convert a double to an int.  Java semantics require we do complex
11059 // manglations in the corner cases.  So we set the rounding mode to
11060 // 'zero', store the darned double down as an int, and reset the
11061 // rounding mode to 'nearest'.  The hardware stores a flag value down
11062 // if we would overflow or converted a NAN; we check for this and
11063 // and go the slow path if needed.
11064 instruct convFPR2I_reg_reg(eAXRegI dst, eDXRegI tmp, regFPR src, eFlagsReg cr ) %{
11065   predicate(UseSSE==0);
11066   match(Set dst (ConvF2I src));
11067   effect( KILL tmp, KILL cr );
11068   format %{ "FLD    $src\t# Convert float to int \n\t"
11069             "FLDCW  trunc mode\n\t"
11070             "SUB    ESP,4\n\t"
11071             "FISTp  [ESP + #0]\n\t"
11072             "FLDCW  std/24-bit mode\n\t"
11073             "POP    EAX\n\t"
11074             "CMP    EAX,0x80000000\n\t"
11075             "JNE,s  fast\n\t"
11076             "FLD    $src\n\t"
11077             "CALL   d2i_wrapper\n"
11078       "fast:" %}
11079   // DPR2I_encoding works for FPR2I
11080   ins_encode( Push_Reg_FPR(src), DPR2I_encoding(src) );
11081   ins_pipe( pipe_slow );
11082 %}
11083 
11084 // Convert a float in xmm to an int reg.
11085 instruct convF2I_reg(eAXRegI dst, eDXRegI tmp, regF src, eFlagsReg cr ) %{
11086   predicate(UseSSE>=1);
11087   match(Set dst (ConvF2I src));
11088   effect( KILL tmp, KILL cr );
11089   format %{ "CVTTSS2SI $dst, $src\n\t"
11090             "CMP    $dst,0x80000000\n\t"
11091             "JNE,s  fast\n\t"
11092             "SUB    ESP, 4\n\t"
11093             "MOVSS  [ESP], $src\n\t"
11094             "FLD    [ESP]\n\t"
11095             "ADD    ESP, 4\n\t"
11096             "CALL   d2i_wrapper\n"
11097       "fast:" %}
11098   ins_encode %{
11099     Label fast;
11100     __ cvttss2sil($dst$$Register, $src$$XMMRegister);
11101     __ cmpl($dst$$Register, 0x80000000);
11102     __ jccb(Assembler::notEqual, fast);
11103     __ subptr(rsp, 4);
11104     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11105     __ fld_s(Address(rsp, 0));
11106     __ addptr(rsp, 4);
11107     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2i_wrapper())));
11108     __ bind(fast);
11109   %}
11110   ins_pipe( pipe_slow );
11111 %}
11112 
11113 instruct convFPR2L_reg_reg( eADXRegL dst, regFPR src, eFlagsReg cr ) %{
11114   predicate(UseSSE==0);
11115   match(Set dst (ConvF2L src));
11116   effect( KILL cr );
11117   format %{ "FLD    $src\t# Convert float to long\n\t"
11118             "FLDCW  trunc mode\n\t"
11119             "SUB    ESP,8\n\t"
11120             "FISTp  [ESP + #0]\n\t"
11121             "FLDCW  std/24-bit mode\n\t"
11122             "POP    EAX\n\t"
11123             "POP    EDX\n\t"
11124             "CMP    EDX,0x80000000\n\t"
11125             "JNE,s  fast\n\t"
11126             "TEST   EAX,EAX\n\t"
11127             "JNE,s  fast\n\t"
11128             "FLD    $src\n\t"
11129             "CALL   d2l_wrapper\n"
11130       "fast:" %}
11131   // DPR2L_encoding works for FPR2L
11132   ins_encode( Push_Reg_FPR(src), DPR2L_encoding(src) );
11133   ins_pipe( pipe_slow );
11134 %}
11135 
11136 // XMM lacks a float/double->long conversion, so use the old FPU stack.
11137 instruct convF2L_reg_reg( eADXRegL dst, regF src, eFlagsReg cr ) %{
11138   predicate (UseSSE>=1);
11139   match(Set dst (ConvF2L src));
11140   effect( KILL cr );
11141   format %{ "SUB    ESP,8\t# Convert float to long\n\t"
11142             "MOVSS  [ESP],$src\n\t"
11143             "FLD_S  [ESP]\n\t"
11144             "FLDCW  trunc mode\n\t"
11145             "FISTp  [ESP + #0]\n\t"
11146             "FLDCW  std/24-bit mode\n\t"
11147             "POP    EAX\n\t"
11148             "POP    EDX\n\t"
11149             "CMP    EDX,0x80000000\n\t"
11150             "JNE,s  fast\n\t"
11151             "TEST   EAX,EAX\n\t"
11152             "JNE,s  fast\n\t"
11153             "SUB    ESP,4\t# Convert float to long\n\t"
11154             "MOVSS  [ESP],$src\n\t"
11155             "FLD_S  [ESP]\n\t"
11156             "ADD    ESP,4\n\t"
11157             "CALL   d2l_wrapper\n"
11158       "fast:" %}
11159   ins_encode %{
11160     Label fast;
11161     __ subptr(rsp, 8);
11162     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11163     __ fld_s(Address(rsp, 0));
11164     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
11165     __ fistp_d(Address(rsp, 0));
11166     // Restore the rounding mode, mask the exception
11167     if (Compile::current()->in_24_bit_fp_mode()) {
11168       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
11169     } else {
11170       __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
11171     }
11172     // Load the converted long, adjust CPU stack
11173     __ pop(rax);
11174     __ pop(rdx);
11175     __ cmpl(rdx, 0x80000000);
11176     __ jccb(Assembler::notEqual, fast);
11177     __ testl(rax, rax);
11178     __ jccb(Assembler::notEqual, fast);
11179     __ subptr(rsp, 4);
11180     __ movflt(Address(rsp, 0), $src$$XMMRegister);
11181     __ fld_s(Address(rsp, 0));
11182     __ addptr(rsp, 4);
11183     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::d2l_wrapper())));
11184     __ bind(fast);
11185   %}
11186   ins_pipe( pipe_slow );
11187 %}
11188 
11189 instruct convI2DPR_reg(regDPR dst, stackSlotI src) %{
11190   predicate( UseSSE<=1 );
11191   match(Set dst (ConvI2D src));
11192   format %{ "FILD   $src\n\t"
11193             "FSTP   $dst" %}
11194   opcode(0xDB, 0x0);  /* DB /0 */
11195   ins_encode(Push_Mem_I(src), Pop_Reg_DPR(dst));
11196   ins_pipe( fpu_reg_mem );
11197 %}
11198 
11199 instruct convI2D_reg(regD dst, eRegI src) %{
11200   predicate( UseSSE>=2 && !UseXmmI2D );
11201   match(Set dst (ConvI2D src));
11202   format %{ "CVTSI2SD $dst,$src" %}
11203   ins_encode %{
11204     __ cvtsi2sdl ($dst$$XMMRegister, $src$$Register);
11205   %}
11206   ins_pipe( pipe_slow );
11207 %}
11208 
11209 instruct convI2D_mem(regD dst, memory mem) %{
11210   predicate( UseSSE>=2 );
11211   match(Set dst (ConvI2D (LoadI mem)));
11212   format %{ "CVTSI2SD $dst,$mem" %}
11213   ins_encode %{
11214     __ cvtsi2sdl ($dst$$XMMRegister, $mem$$Address);
11215   %}
11216   ins_pipe( pipe_slow );
11217 %}
11218 
11219 instruct convXI2D_reg(regD dst, eRegI src)
11220 %{
11221   predicate( UseSSE>=2 && UseXmmI2D );
11222   match(Set dst (ConvI2D src));
11223 
11224   format %{ "MOVD  $dst,$src\n\t"
11225             "CVTDQ2PD $dst,$dst\t# i2d" %}
11226   ins_encode %{
11227     __ movdl($dst$$XMMRegister, $src$$Register);
11228     __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11229   %}
11230   ins_pipe(pipe_slow); // XXX
11231 %}
11232 
11233 instruct convI2DPR_mem(regDPR dst, memory mem) %{
11234   predicate( UseSSE<=1 && !Compile::current()->select_24_bit_instr());
11235   match(Set dst (ConvI2D (LoadI mem)));
11236   format %{ "FILD   $mem\n\t"
11237             "FSTP   $dst" %}
11238   opcode(0xDB);      /* DB /0 */
11239   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11240               Pop_Reg_DPR(dst));
11241   ins_pipe( fpu_reg_mem );
11242 %}
11243 
11244 // Convert a byte to a float; no rounding step needed.
11245 instruct conv24I2FPR_reg(regFPR dst, stackSlotI src) %{
11246   predicate( UseSSE==0 && n->in(1)->Opcode() == Op_AndI && n->in(1)->in(2)->is_Con() && n->in(1)->in(2)->get_int() == 255 );
11247   match(Set dst (ConvI2F src));
11248   format %{ "FILD   $src\n\t"
11249             "FSTP   $dst" %}
11250 
11251   opcode(0xDB, 0x0);  /* DB /0 */
11252   ins_encode(Push_Mem_I(src), Pop_Reg_FPR(dst));
11253   ins_pipe( fpu_reg_mem );
11254 %}
11255 
11256 // In 24-bit mode, force exponent rounding by storing back out
11257 instruct convI2FPR_SSF(stackSlotF dst, stackSlotI src) %{
11258   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
11259   match(Set dst (ConvI2F src));
11260   ins_cost(200);
11261   format %{ "FILD   $src\n\t"
11262             "FSTP_S $dst" %}
11263   opcode(0xDB, 0x0);  /* DB /0 */
11264   ins_encode( Push_Mem_I(src),
11265               Pop_Mem_FPR(dst));
11266   ins_pipe( fpu_mem_mem );
11267 %}
11268 
11269 // In 24-bit mode, force exponent rounding by storing back out
11270 instruct convI2FPR_SSF_mem(stackSlotF dst, memory mem) %{
11271   predicate( UseSSE==0 && Compile::current()->select_24_bit_instr());
11272   match(Set dst (ConvI2F (LoadI mem)));
11273   ins_cost(200);
11274   format %{ "FILD   $mem\n\t"
11275             "FSTP_S $dst" %}
11276   opcode(0xDB);  /* DB /0 */
11277   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11278               Pop_Mem_FPR(dst));
11279   ins_pipe( fpu_mem_mem );
11280 %}
11281 
11282 // This instruction does not round to 24-bits
11283 instruct convI2FPR_reg(regFPR dst, stackSlotI src) %{
11284   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
11285   match(Set dst (ConvI2F src));
11286   format %{ "FILD   $src\n\t"
11287             "FSTP   $dst" %}
11288   opcode(0xDB, 0x0);  /* DB /0 */
11289   ins_encode( Push_Mem_I(src),
11290               Pop_Reg_FPR(dst));
11291   ins_pipe( fpu_reg_mem );
11292 %}
11293 
11294 // This instruction does not round to 24-bits
11295 instruct convI2FPR_mem(regFPR dst, memory mem) %{
11296   predicate( UseSSE==0 && !Compile::current()->select_24_bit_instr());
11297   match(Set dst (ConvI2F (LoadI mem)));
11298   format %{ "FILD   $mem\n\t"
11299             "FSTP   $dst" %}
11300   opcode(0xDB);      /* DB /0 */
11301   ins_encode( OpcP, RMopc_Mem(0x00,mem),
11302               Pop_Reg_FPR(dst));
11303   ins_pipe( fpu_reg_mem );
11304 %}
11305 
11306 // Convert an int to a float in xmm; no rounding step needed.
11307 instruct convI2F_reg(regF dst, eRegI src) %{
11308   predicate( UseSSE==1 || UseSSE>=2 && !UseXmmI2F );
11309   match(Set dst (ConvI2F src));
11310   format %{ "CVTSI2SS $dst, $src" %}
11311   ins_encode %{
11312     __ cvtsi2ssl ($dst$$XMMRegister, $src$$Register);
11313   %}
11314   ins_pipe( pipe_slow );
11315 %}
11316 
11317  instruct convXI2F_reg(regF dst, eRegI src)
11318 %{
11319   predicate( UseSSE>=2 && UseXmmI2F );
11320   match(Set dst (ConvI2F src));
11321 
11322   format %{ "MOVD  $dst,$src\n\t"
11323             "CVTDQ2PS $dst,$dst\t# i2f" %}
11324   ins_encode %{
11325     __ movdl($dst$$XMMRegister, $src$$Register);
11326     __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11327   %}
11328   ins_pipe(pipe_slow); // XXX
11329 %}
11330 
11331 instruct convI2L_reg( eRegL dst, eRegI src, eFlagsReg cr) %{
11332   match(Set dst (ConvI2L src));
11333   effect(KILL cr);
11334   ins_cost(375);
11335   format %{ "MOV    $dst.lo,$src\n\t"
11336             "MOV    $dst.hi,$src\n\t"
11337             "SAR    $dst.hi,31" %}
11338   ins_encode(convert_int_long(dst,src));
11339   ins_pipe( ialu_reg_reg_long );
11340 %}
11341 
11342 // Zero-extend convert int to long
11343 instruct convI2L_reg_zex(eRegL dst, eRegI src, immL_32bits mask, eFlagsReg flags ) %{
11344   match(Set dst (AndL (ConvI2L src) mask) );
11345   effect( KILL flags );
11346   ins_cost(250);
11347   format %{ "MOV    $dst.lo,$src\n\t"
11348             "XOR    $dst.hi,$dst.hi" %}
11349   opcode(0x33); // XOR
11350   ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
11351   ins_pipe( ialu_reg_reg_long );
11352 %}
11353 
11354 // Zero-extend long
11355 instruct zerox_long(eRegL dst, eRegL src, immL_32bits mask, eFlagsReg flags ) %{
11356   match(Set dst (AndL src mask) );
11357   effect( KILL flags );
11358   ins_cost(250);
11359   format %{ "MOV    $dst.lo,$src.lo\n\t"
11360             "XOR    $dst.hi,$dst.hi\n\t" %}
11361   opcode(0x33); // XOR
11362   ins_encode(enc_Copy(dst,src), OpcP, RegReg_Hi2(dst,dst) );
11363   ins_pipe( ialu_reg_reg_long );
11364 %}
11365 
11366 instruct convL2DPR_reg( stackSlotD dst, eRegL src, eFlagsReg cr) %{
11367   predicate (UseSSE<=1);
11368   match(Set dst (ConvL2D src));
11369   effect( KILL cr );
11370   format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
11371             "PUSH   $src.lo\n\t"
11372             "FILD   ST,[ESP + #0]\n\t"
11373             "ADD    ESP,8\n\t"
11374             "FSTP_D $dst\t# D-round" %}
11375   opcode(0xDF, 0x5);  /* DF /5 */
11376   ins_encode(convert_long_double(src), Pop_Mem_DPR(dst));
11377   ins_pipe( pipe_slow );
11378 %}
11379 
11380 instruct convL2D_reg( regD dst, eRegL src, eFlagsReg cr) %{
11381   predicate (UseSSE>=2);
11382   match(Set dst (ConvL2D src));
11383   effect( KILL cr );
11384   format %{ "PUSH   $src.hi\t# Convert long to double\n\t"
11385             "PUSH   $src.lo\n\t"
11386             "FILD_D [ESP]\n\t"
11387             "FSTP_D [ESP]\n\t"
11388             "MOVSD  $dst,[ESP]\n\t"
11389             "ADD    ESP,8" %}
11390   opcode(0xDF, 0x5);  /* DF /5 */
11391   ins_encode(convert_long_double2(src), Push_ResultD(dst));
11392   ins_pipe( pipe_slow );
11393 %}
11394 
11395 instruct convL2F_reg( regF dst, eRegL src, eFlagsReg cr) %{
11396   predicate (UseSSE>=1);
11397   match(Set dst (ConvL2F src));
11398   effect( KILL cr );
11399   format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
11400             "PUSH   $src.lo\n\t"
11401             "FILD_D [ESP]\n\t"
11402             "FSTP_S [ESP]\n\t"
11403             "MOVSS  $dst,[ESP]\n\t"
11404             "ADD    ESP,8" %}
11405   opcode(0xDF, 0x5);  /* DF /5 */
11406   ins_encode(convert_long_double2(src), Push_ResultF(dst,0x8));
11407   ins_pipe( pipe_slow );
11408 %}
11409 
11410 instruct convL2FPR_reg( stackSlotF dst, eRegL src, eFlagsReg cr) %{
11411   match(Set dst (ConvL2F src));
11412   effect( KILL cr );
11413   format %{ "PUSH   $src.hi\t# Convert long to single float\n\t"
11414             "PUSH   $src.lo\n\t"
11415             "FILD   ST,[ESP + #0]\n\t"
11416             "ADD    ESP,8\n\t"
11417             "FSTP_S $dst\t# F-round" %}
11418   opcode(0xDF, 0x5);  /* DF /5 */
11419   ins_encode(convert_long_double(src), Pop_Mem_FPR(dst));
11420   ins_pipe( pipe_slow );
11421 %}
11422 
11423 instruct convL2I_reg( eRegI dst, eRegL src ) %{
11424   match(Set dst (ConvL2I src));
11425   effect( DEF dst, USE src );
11426   format %{ "MOV    $dst,$src.lo" %}
11427   ins_encode(enc_CopyL_Lo(dst,src));
11428   ins_pipe( ialu_reg_reg );
11429 %}
11430 
11431 
11432 instruct MoveF2I_stack_reg(eRegI dst, stackSlotF src) %{
11433   match(Set dst (MoveF2I src));
11434   effect( DEF dst, USE src );
11435   ins_cost(100);
11436   format %{ "MOV    $dst,$src\t# MoveF2I_stack_reg" %}
11437   ins_encode %{
11438     __ movl($dst$$Register, Address(rsp, $src$$disp));
11439   %}
11440   ins_pipe( ialu_reg_mem );
11441 %}
11442 
11443 instruct MoveFPR2I_reg_stack(stackSlotI dst, regFPR src) %{
11444   predicate(UseSSE==0);
11445   match(Set dst (MoveF2I src));
11446   effect( DEF dst, USE src );
11447 
11448   ins_cost(125);
11449   format %{ "FST_S  $dst,$src\t# MoveF2I_reg_stack" %}
11450   ins_encode( Pop_Mem_Reg_FPR(dst, src) );
11451   ins_pipe( fpu_mem_reg );
11452 %}
11453 
11454 instruct MoveF2I_reg_stack_sse(stackSlotI dst, regF src) %{
11455   predicate(UseSSE>=1);
11456   match(Set dst (MoveF2I src));
11457   effect( DEF dst, USE src );
11458 
11459   ins_cost(95);
11460   format %{ "MOVSS  $dst,$src\t# MoveF2I_reg_stack_sse" %}
11461   ins_encode %{
11462     __ movflt(Address(rsp, $dst$$disp), $src$$XMMRegister);
11463   %}
11464   ins_pipe( pipe_slow );
11465 %}
11466 
11467 instruct MoveF2I_reg_reg_sse(eRegI dst, regF src) %{
11468   predicate(UseSSE>=2);
11469   match(Set dst (MoveF2I src));
11470   effect( DEF dst, USE src );
11471   ins_cost(85);
11472   format %{ "MOVD   $dst,$src\t# MoveF2I_reg_reg_sse" %}
11473   ins_encode %{
11474     __ movdl($dst$$Register, $src$$XMMRegister);
11475   %}
11476   ins_pipe( pipe_slow );
11477 %}
11478 
11479 instruct MoveI2F_reg_stack(stackSlotF dst, eRegI src) %{
11480   match(Set dst (MoveI2F src));
11481   effect( DEF dst, USE src );
11482 
11483   ins_cost(100);
11484   format %{ "MOV    $dst,$src\t# MoveI2F_reg_stack" %}
11485   ins_encode %{
11486     __ movl(Address(rsp, $dst$$disp), $src$$Register);
11487   %}
11488   ins_pipe( ialu_mem_reg );
11489 %}
11490 
11491 
11492 instruct MoveI2FPR_stack_reg(regFPR dst, stackSlotI src) %{
11493   predicate(UseSSE==0);
11494   match(Set dst (MoveI2F src));
11495   effect(DEF dst, USE src);
11496 
11497   ins_cost(125);
11498   format %{ "FLD_S  $src\n\t"
11499             "FSTP   $dst\t# MoveI2F_stack_reg" %}
11500   opcode(0xD9);               /* D9 /0, FLD m32real */
11501   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
11502               Pop_Reg_FPR(dst) );
11503   ins_pipe( fpu_reg_mem );
11504 %}
11505 
11506 instruct MoveI2F_stack_reg_sse(regF dst, stackSlotI src) %{
11507   predicate(UseSSE>=1);
11508   match(Set dst (MoveI2F src));
11509   effect( DEF dst, USE src );
11510 
11511   ins_cost(95);
11512   format %{ "MOVSS  $dst,$src\t# MoveI2F_stack_reg_sse" %}
11513   ins_encode %{
11514     __ movflt($dst$$XMMRegister, Address(rsp, $src$$disp));
11515   %}
11516   ins_pipe( pipe_slow );
11517 %}
11518 
11519 instruct MoveI2F_reg_reg_sse(regF dst, eRegI src) %{
11520   predicate(UseSSE>=2);
11521   match(Set dst (MoveI2F src));
11522   effect( DEF dst, USE src );
11523 
11524   ins_cost(85);
11525   format %{ "MOVD   $dst,$src\t# MoveI2F_reg_reg_sse" %}
11526   ins_encode %{
11527     __ movdl($dst$$XMMRegister, $src$$Register);
11528   %}
11529   ins_pipe( pipe_slow );
11530 %}
11531 
11532 instruct MoveD2L_stack_reg(eRegL dst, stackSlotD src) %{
11533   match(Set dst (MoveD2L src));
11534   effect(DEF dst, USE src);
11535 
11536   ins_cost(250);
11537   format %{ "MOV    $dst.lo,$src\n\t"
11538             "MOV    $dst.hi,$src+4\t# MoveD2L_stack_reg" %}
11539   opcode(0x8B, 0x8B);
11540   ins_encode( OpcP, RegMem(dst,src), OpcS, RegMem_Hi(dst,src));
11541   ins_pipe( ialu_mem_long_reg );
11542 %}
11543 
11544 instruct MoveDPR2L_reg_stack(stackSlotL dst, regDPR src) %{
11545   predicate(UseSSE<=1);
11546   match(Set dst (MoveD2L src));
11547   effect(DEF dst, USE src);
11548 
11549   ins_cost(125);
11550   format %{ "FST_D  $dst,$src\t# MoveD2L_reg_stack" %}
11551   ins_encode( Pop_Mem_Reg_DPR(dst, src) );
11552   ins_pipe( fpu_mem_reg );
11553 %}
11554 
11555 instruct MoveD2L_reg_stack_sse(stackSlotL dst, regD src) %{
11556   predicate(UseSSE>=2);
11557   match(Set dst (MoveD2L src));
11558   effect(DEF dst, USE src);
11559   ins_cost(95);
11560   format %{ "MOVSD  $dst,$src\t# MoveD2L_reg_stack_sse" %}
11561   ins_encode %{
11562     __ movdbl(Address(rsp, $dst$$disp), $src$$XMMRegister);
11563   %}
11564   ins_pipe( pipe_slow );
11565 %}
11566 
11567 instruct MoveD2L_reg_reg_sse(eRegL dst, regD src, regD tmp) %{
11568   predicate(UseSSE>=2);
11569   match(Set dst (MoveD2L src));
11570   effect(DEF dst, USE src, TEMP tmp);
11571   ins_cost(85);
11572   format %{ "MOVD   $dst.lo,$src\n\t"
11573             "PSHUFLW $tmp,$src,0x4E\n\t"
11574             "MOVD   $dst.hi,$tmp\t# MoveD2L_reg_reg_sse" %}
11575   ins_encode %{
11576     __ movdl($dst$$Register, $src$$XMMRegister);
11577     __ pshuflw($tmp$$XMMRegister, $src$$XMMRegister, 0x4e);
11578     __ movdl(HIGH_FROM_LOW($dst$$Register), $tmp$$XMMRegister);
11579   %}
11580   ins_pipe( pipe_slow );
11581 %}
11582 
11583 instruct MoveL2D_reg_stack(stackSlotD dst, eRegL src) %{
11584   match(Set dst (MoveL2D src));
11585   effect(DEF dst, USE src);
11586 
11587   ins_cost(200);
11588   format %{ "MOV    $dst,$src.lo\n\t"
11589             "MOV    $dst+4,$src.hi\t# MoveL2D_reg_stack" %}
11590   opcode(0x89, 0x89);
11591   ins_encode( OpcP, RegMem( src, dst ), OpcS, RegMem_Hi( src, dst ) );
11592   ins_pipe( ialu_mem_long_reg );
11593 %}
11594 
11595 
11596 instruct MoveL2DPR_stack_reg(regDPR dst, stackSlotL src) %{
11597   predicate(UseSSE<=1);
11598   match(Set dst (MoveL2D src));
11599   effect(DEF dst, USE src);
11600   ins_cost(125);
11601 
11602   format %{ "FLD_D  $src\n\t"
11603             "FSTP   $dst\t# MoveL2D_stack_reg" %}
11604   opcode(0xDD);               /* DD /0, FLD m64real */
11605   ins_encode( OpcP, RMopc_Mem_no_oop(0x00,src),
11606               Pop_Reg_DPR(dst) );
11607   ins_pipe( fpu_reg_mem );
11608 %}
11609 
11610 
11611 instruct MoveL2D_stack_reg_sse(regD dst, stackSlotL src) %{
11612   predicate(UseSSE>=2 && UseXmmLoadAndClearUpper);
11613   match(Set dst (MoveL2D src));
11614   effect(DEF dst, USE src);
11615 
11616   ins_cost(95);
11617   format %{ "MOVSD  $dst,$src\t# MoveL2D_stack_reg_sse" %}
11618   ins_encode %{
11619     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
11620   %}
11621   ins_pipe( pipe_slow );
11622 %}
11623 
11624 instruct MoveL2D_stack_reg_sse_partial(regD dst, stackSlotL src) %{
11625   predicate(UseSSE>=2 && !UseXmmLoadAndClearUpper);
11626   match(Set dst (MoveL2D src));
11627   effect(DEF dst, USE src);
11628 
11629   ins_cost(95);
11630   format %{ "MOVLPD $dst,$src\t# MoveL2D_stack_reg_sse" %}
11631   ins_encode %{
11632     __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
11633   %}
11634   ins_pipe( pipe_slow );
11635 %}
11636 
11637 instruct MoveL2D_reg_reg_sse(regD dst, eRegL src, regD tmp) %{
11638   predicate(UseSSE>=2);
11639   match(Set dst (MoveL2D src));
11640   effect(TEMP dst, USE src, TEMP tmp);
11641   ins_cost(85);
11642   format %{ "MOVD   $dst,$src.lo\n\t"
11643             "MOVD   $tmp,$src.hi\n\t"
11644             "PUNPCKLDQ $dst,$tmp\t# MoveL2D_reg_reg_sse" %}
11645   ins_encode %{
11646     __ movdl($dst$$XMMRegister, $src$$Register);
11647     __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
11648     __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
11649   %}
11650   ins_pipe( pipe_slow );
11651 %}
11652 
11653 // Replicate scalar to packed byte (1 byte) values in xmm
11654 instruct Repl8B_reg(regD dst, regD src) %{
11655   predicate(UseSSE>=2);
11656   match(Set dst (Replicate8B src));
11657   format %{ "MOVDQA  $dst,$src\n\t"
11658             "PUNPCKLBW $dst,$dst\n\t"
11659             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11660   ins_encode %{
11661     if ($dst$$reg != $src$$reg) {
11662       __ movdqa($dst$$XMMRegister, $src$$XMMRegister);
11663     }
11664     __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
11665     __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
11666   %}
11667   ins_pipe( pipe_slow );
11668 %}
11669 
11670 // Replicate scalar to packed byte (1 byte) values in xmm
11671 instruct Repl8B_eRegI(regD dst, eRegI src) %{
11672   predicate(UseSSE>=2);
11673   match(Set dst (Replicate8B src));
11674   format %{ "MOVD    $dst,$src\n\t"
11675             "PUNPCKLBW $dst,$dst\n\t"
11676             "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11677   ins_encode %{
11678     __ movdl($dst$$XMMRegister, $src$$Register);
11679     __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
11680     __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
11681   %}
11682   ins_pipe( pipe_slow );
11683 %}
11684 
11685 // Replicate scalar zero to packed byte (1 byte) values in xmm
11686 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11687   predicate(UseSSE>=2);
11688   match(Set dst (Replicate8B zero));
11689   format %{ "PXOR  $dst,$dst\t! replicate8B" %}
11690   ins_encode %{
11691     __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
11692   %}
11693   ins_pipe( fpu_reg_reg );
11694 %}
11695 
11696 // Replicate scalar to packed shore (2 byte) values in xmm
11697 instruct Repl4S_reg(regD dst, regD src) %{
11698   predicate(UseSSE>=2);
11699   match(Set dst (Replicate4S src));
11700   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11701   ins_encode %{
11702     __ pshuflw($dst$$XMMRegister, $src$$XMMRegister, 0x00);
11703   %}
11704   ins_pipe( fpu_reg_reg );
11705 %}
11706 
11707 // Replicate scalar to packed shore (2 byte) values in xmm
11708 instruct Repl4S_eRegI(regD dst, eRegI src) %{
11709   predicate(UseSSE>=2);
11710   match(Set dst (Replicate4S src));
11711   format %{ "MOVD    $dst,$src\n\t"
11712             "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11713   ins_encode %{
11714     __ movdl($dst$$XMMRegister, $src$$Register);
11715     __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
11716   %}
11717   ins_pipe( fpu_reg_reg );
11718 %}
11719 
11720 // Replicate scalar zero to packed short (2 byte) values in xmm
11721 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11722   predicate(UseSSE>=2);
11723   match(Set dst (Replicate4S zero));
11724   format %{ "PXOR  $dst,$dst\t! replicate4S" %}
11725   ins_encode %{
11726     __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
11727   %}
11728   ins_pipe( fpu_reg_reg );
11729 %}
11730 
11731 // Replicate scalar to packed char (2 byte) values in xmm
11732 instruct Repl4C_reg(regD dst, regD src) %{
11733   predicate(UseSSE>=2);
11734   match(Set dst (Replicate4C src));
11735   format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11736   ins_encode %{
11737     __ pshuflw($dst$$XMMRegister, $src$$XMMRegister, 0x00);
11738   %}
11739   ins_pipe( fpu_reg_reg );
11740 %}
11741 
11742 // Replicate scalar to packed char (2 byte) values in xmm
11743 instruct Repl4C_eRegI(regD dst, eRegI src) %{
11744   predicate(UseSSE>=2);
11745   match(Set dst (Replicate4C src));
11746   format %{ "MOVD    $dst,$src\n\t"
11747             "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11748   ins_encode %{
11749     __ movdl($dst$$XMMRegister, $src$$Register);
11750     __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
11751   %}
11752   ins_pipe( fpu_reg_reg );
11753 %}
11754 
11755 // Replicate scalar zero to packed char (2 byte) values in xmm
11756 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11757   predicate(UseSSE>=2);
11758   match(Set dst (Replicate4C zero));
11759   format %{ "PXOR  $dst,$dst\t! replicate4C" %}
11760   ins_encode %{
11761     __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
11762   %}
11763   ins_pipe( fpu_reg_reg );
11764 %}
11765 
11766 // Replicate scalar to packed integer (4 byte) values in xmm
11767 instruct Repl2I_reg(regD dst, regD src) %{
11768   predicate(UseSSE>=2);
11769   match(Set dst (Replicate2I src));
11770   format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11771   ins_encode %{
11772     __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
11773   %}
11774   ins_pipe( fpu_reg_reg );
11775 %}
11776 
11777 // Replicate scalar to packed integer (4 byte) values in xmm
11778 instruct Repl2I_eRegI(regD dst, eRegI src) %{
11779   predicate(UseSSE>=2);
11780   match(Set dst (Replicate2I src));
11781   format %{ "MOVD   $dst,$src\n\t"
11782             "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11783   ins_encode %{
11784     __ movdl($dst$$XMMRegister, $src$$Register);
11785     __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
11786   %}
11787   ins_pipe( fpu_reg_reg );
11788 %}
11789 
11790 // Replicate scalar zero to packed integer (2 byte) values in xmm
11791 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11792   predicate(UseSSE>=2);
11793   match(Set dst (Replicate2I zero));
11794   format %{ "PXOR  $dst,$dst\t! replicate2I" %}
11795   ins_encode %{
11796     __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
11797   %}
11798   ins_pipe( fpu_reg_reg );
11799 %}
11800 
11801 // Replicate scalar to packed single precision floating point values in xmm
11802 instruct Repl2F_reg(regD dst, regD src) %{
11803   predicate(UseSSE>=2);
11804   match(Set dst (Replicate2F src));
11805   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11806   ins_encode %{
11807     __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0xe0);
11808   %}
11809   ins_pipe( fpu_reg_reg );
11810 %}
11811 
11812 // Replicate scalar to packed single precision floating point values in xmm
11813 instruct Repl2F_regF(regD dst, regF src) %{
11814   predicate(UseSSE>=2);
11815   match(Set dst (Replicate2F src));
11816   format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11817   ins_encode %{
11818     __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0xe0);
11819   %}
11820   ins_pipe( fpu_reg_reg );
11821 %}
11822 
11823 // Replicate scalar to packed single precision floating point values in xmm
11824 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11825   predicate(UseSSE>=2);
11826   match(Set dst (Replicate2F zero));
11827   format %{ "PXOR  $dst,$dst\t! replicate2F" %}
11828   ins_encode %{
11829     __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
11830   %}
11831   ins_pipe( fpu_reg_reg );
11832 %}
11833 
11834 // =======================================================================
11835 // fast clearing of an array
11836 instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
11837   match(Set dummy (ClearArray cnt base));
11838   effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11839   format %{ "SHL    ECX,1\t# Convert doublewords to words\n\t"
11840             "XOR    EAX,EAX\n\t"
11841             "REP STOS\t# store EAX into [EDI++] while ECX--" %}
11842   opcode(0,0x4);
11843   ins_encode( Opcode(0xD1), RegOpc(ECX),
11844               OpcRegReg(0x33,EAX,EAX),
11845               Opcode(0xF3), Opcode(0xAB) );
11846   ins_pipe( pipe_slow );
11847 %}
11848 
11849 instruct string_compare(eDIRegP str1, eCXRegI cnt1, eSIRegP str2, eDXRegI cnt2,
11850                         eAXRegI result, regD tmp1, eFlagsReg cr) %{
11851   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11852   effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
11853 
11854   format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result   // KILL $tmp1" %}
11855   ins_encode %{
11856     __ string_compare($str1$$Register, $str2$$Register,
11857                       $cnt1$$Register, $cnt2$$Register, $result$$Register,
11858                       $tmp1$$XMMRegister);
11859   %}
11860   ins_pipe( pipe_slow );
11861 %}
11862 
11863 // fast string equals
11864 instruct string_equals(eDIRegP str1, eSIRegP str2, eCXRegI cnt, eAXRegI result,
11865                        regD tmp1, regD tmp2, eBXRegI tmp3, eFlagsReg cr) %{
11866   match(Set result (StrEquals (Binary str1 str2) cnt));
11867   effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);
11868 
11869   format %{ "String Equals $str1,$str2,$cnt -> $result    // KILL $tmp1, $tmp2, $tmp3" %}
11870   ins_encode %{
11871     __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
11872                           $cnt$$Register, $result$$Register, $tmp3$$Register,
11873                           $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11874   %}
11875   ins_pipe( pipe_slow );
11876 %}
11877 
11878 // fast search of substring with known size.
11879 instruct string_indexof_con(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, immI int_cnt2,
11880                             eBXRegI result, regD vec, eAXRegI cnt2, eCXRegI tmp, eFlagsReg cr) %{
11881   predicate(UseSSE42Intrinsics);
11882   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
11883   effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, KILL cnt2, KILL tmp, KILL cr);
11884 
11885   format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result   // KILL $vec, $cnt1, $cnt2, $tmp" %}
11886   ins_encode %{
11887     int icnt2 = (int)$int_cnt2$$constant;
11888     if (icnt2 >= 8) {
11889       // IndexOf for constant substrings with size >= 8 elements
11890       // which don't need to be loaded through stack.
11891       __ string_indexofC8($str1$$Register, $str2$$Register,
11892                           $cnt1$$Register, $cnt2$$Register,
11893                           icnt2, $result$$Register,
11894                           $vec$$XMMRegister, $tmp$$Register);
11895     } else {
11896       // Small strings are loaded through stack if they cross page boundary.
11897       __ string_indexof($str1$$Register, $str2$$Register,
11898                         $cnt1$$Register, $cnt2$$Register,
11899                         icnt2, $result$$Register,
11900                         $vec$$XMMRegister, $tmp$$Register);
11901     }
11902   %}
11903   ins_pipe( pipe_slow );
11904 %}
11905 
11906 instruct string_indexof(eDIRegP str1, eDXRegI cnt1, eSIRegP str2, eAXRegI cnt2,
11907                         eBXRegI result, regD vec, eCXRegI tmp, eFlagsReg cr) %{
11908   predicate(UseSSE42Intrinsics);
11909   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
11910   effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp, KILL cr);
11911 
11912   format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result   // KILL all" %}
11913   ins_encode %{
11914     __ string_indexof($str1$$Register, $str2$$Register,
11915                       $cnt1$$Register, $cnt2$$Register,
11916                       (-1), $result$$Register,
11917                       $vec$$XMMRegister, $tmp$$Register);
11918   %}
11919   ins_pipe( pipe_slow );
11920 %}
11921 
11922 // fast array equals
11923 instruct array_equals(eDIRegP ary1, eSIRegP ary2, eAXRegI result,
11924                       regD tmp1, regD tmp2, eCXRegI tmp3, eBXRegI tmp4, eFlagsReg cr)
11925 %{
11926   match(Set result (AryEq ary1 ary2));
11927   effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
11928   //ins_cost(300);
11929 
11930   format %{ "Array Equals $ary1,$ary2 -> $result   // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
11931   ins_encode %{
11932     __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
11933                           $tmp3$$Register, $result$$Register, $tmp4$$Register,
11934                           $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11935   %}
11936   ins_pipe( pipe_slow );
11937 %}
11938 
11939 //----------Control Flow Instructions------------------------------------------
11940 // Signed compare Instructions
11941 instruct compI_eReg(eFlagsReg cr, eRegI op1, eRegI op2) %{
11942   match(Set cr (CmpI op1 op2));
11943   effect( DEF cr, USE op1, USE op2 );
11944   format %{ "CMP    $op1,$op2" %}
11945   opcode(0x3B);  /* Opcode 3B /r */
11946   ins_encode( OpcP, RegReg( op1, op2) );
11947   ins_pipe( ialu_cr_reg_reg );
11948 %}
11949 
11950 instruct compI_eReg_imm(eFlagsReg cr, eRegI op1, immI op2) %{
11951   match(Set cr (CmpI op1 op2));
11952   effect( DEF cr, USE op1 );
11953   format %{ "CMP    $op1,$op2" %}
11954   opcode(0x81,0x07);  /* Opcode 81 /7 */
11955   // ins_encode( RegImm( op1, op2) );  /* Was CmpImm */
11956   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
11957   ins_pipe( ialu_cr_reg_imm );
11958 %}
11959 
11960 // Cisc-spilled version of cmpI_eReg
11961 instruct compI_eReg_mem(eFlagsReg cr, eRegI op1, memory op2) %{
11962   match(Set cr (CmpI op1 (LoadI op2)));
11963 
11964   format %{ "CMP    $op1,$op2" %}
11965   ins_cost(500);
11966   opcode(0x3B);  /* Opcode 3B /r */
11967   ins_encode( OpcP, RegMem( op1, op2) );
11968   ins_pipe( ialu_cr_reg_mem );
11969 %}
11970 
11971 instruct testI_reg( eFlagsReg cr, eRegI src, immI0 zero ) %{
11972   match(Set cr (CmpI src zero));
11973   effect( DEF cr, USE src );
11974 
11975   format %{ "TEST   $src,$src" %}
11976   opcode(0x85);
11977   ins_encode( OpcP, RegReg( src, src ) );
11978   ins_pipe( ialu_cr_reg_imm );
11979 %}
11980 
11981 instruct testI_reg_imm( eFlagsReg cr, eRegI src, immI con, immI0 zero ) %{
11982   match(Set cr (CmpI (AndI src con) zero));
11983 
11984   format %{ "TEST   $src,$con" %}
11985   opcode(0xF7,0x00);
11986   ins_encode( OpcP, RegOpc(src), Con32(con) );
11987   ins_pipe( ialu_cr_reg_imm );
11988 %}
11989 
11990 instruct testI_reg_mem( eFlagsReg cr, eRegI src, memory mem, immI0 zero ) %{
11991   match(Set cr (CmpI (AndI src mem) zero));
11992 
11993   format %{ "TEST   $src,$mem" %}
11994   opcode(0x85);
11995   ins_encode( OpcP, RegMem( src, mem ) );
11996   ins_pipe( ialu_cr_reg_mem );
11997 %}
11998 
11999 // Unsigned compare Instructions; really, same as signed except they
12000 // produce an eFlagsRegU instead of eFlagsReg.
12001 instruct compU_eReg(eFlagsRegU cr, eRegI op1, eRegI op2) %{
12002   match(Set cr (CmpU op1 op2));
12003 
12004   format %{ "CMPu   $op1,$op2" %}
12005   opcode(0x3B);  /* Opcode 3B /r */
12006   ins_encode( OpcP, RegReg( op1, op2) );
12007   ins_pipe( ialu_cr_reg_reg );
12008 %}
12009 
12010 instruct compU_eReg_imm(eFlagsRegU cr, eRegI op1, immI op2) %{
12011   match(Set cr (CmpU op1 op2));
12012 
12013   format %{ "CMPu   $op1,$op2" %}
12014   opcode(0x81,0x07);  /* Opcode 81 /7 */
12015   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
12016   ins_pipe( ialu_cr_reg_imm );
12017 %}
12018 
12019 // // Cisc-spilled version of cmpU_eReg
12020 instruct compU_eReg_mem(eFlagsRegU cr, eRegI op1, memory op2) %{
12021   match(Set cr (CmpU op1 (LoadI op2)));
12022 
12023   format %{ "CMPu   $op1,$op2" %}
12024   ins_cost(500);
12025   opcode(0x3B);  /* Opcode 3B /r */
12026   ins_encode( OpcP, RegMem( op1, op2) );
12027   ins_pipe( ialu_cr_reg_mem );
12028 %}
12029 
12030 // // Cisc-spilled version of cmpU_eReg
12031 //instruct compU_mem_eReg(eFlagsRegU cr, memory op1, eRegI op2) %{
12032 //  match(Set cr (CmpU (LoadI op1) op2));
12033 //
12034 //  format %{ "CMPu   $op1,$op2" %}
12035 //  ins_cost(500);
12036 //  opcode(0x39);  /* Opcode 39 /r */
12037 //  ins_encode( OpcP, RegMem( op1, op2) );
12038 //%}
12039 
12040 instruct testU_reg( eFlagsRegU cr, eRegI src, immI0 zero ) %{
12041   match(Set cr (CmpU src zero));
12042 
12043   format %{ "TESTu  $src,$src" %}
12044   opcode(0x85);
12045   ins_encode( OpcP, RegReg( src, src ) );
12046   ins_pipe( ialu_cr_reg_imm );
12047 %}
12048 
12049 // Unsigned pointer compare Instructions
12050 instruct compP_eReg(eFlagsRegU cr, eRegP op1, eRegP op2) %{
12051   match(Set cr (CmpP op1 op2));
12052 
12053   format %{ "CMPu   $op1,$op2" %}
12054   opcode(0x3B);  /* Opcode 3B /r */
12055   ins_encode( OpcP, RegReg( op1, op2) );
12056   ins_pipe( ialu_cr_reg_reg );
12057 %}
12058 
12059 instruct compP_eReg_imm(eFlagsRegU cr, eRegP op1, immP op2) %{
12060   match(Set cr (CmpP op1 op2));
12061 
12062   format %{ "CMPu   $op1,$op2" %}
12063   opcode(0x81,0x07);  /* Opcode 81 /7 */
12064   ins_encode( OpcSErm( op1, op2 ), Con8or32( op2 ) );
12065   ins_pipe( ialu_cr_reg_imm );
12066 %}
12067 
12068 // // Cisc-spilled version of cmpP_eReg
12069 instruct compP_eReg_mem(eFlagsRegU cr, eRegP op1, memory op2) %{
12070   match(Set cr (CmpP op1 (LoadP op2)));
12071 
12072   format %{ "CMPu   $op1,$op2" %}
12073   ins_cost(500);
12074   opcode(0x3B);  /* Opcode 3B /r */
12075   ins_encode( OpcP, RegMem( op1, op2) );
12076   ins_pipe( ialu_cr_reg_mem );
12077 %}
12078 
12079 // // Cisc-spilled version of cmpP_eReg
12080 //instruct compP_mem_eReg(eFlagsRegU cr, memory op1, eRegP op2) %{
12081 //  match(Set cr (CmpP (LoadP op1) op2));
12082 //
12083 //  format %{ "CMPu   $op1,$op2" %}
12084 //  ins_cost(500);
12085 //  opcode(0x39);  /* Opcode 39 /r */
12086 //  ins_encode( OpcP, RegMem( op1, op2) );
12087 //%}
12088 
12089 // Compare raw pointer (used in out-of-heap check).
12090 // Only works because non-oop pointers must be raw pointers
12091 // and raw pointers have no anti-dependencies.
12092 instruct compP_mem_eReg( eFlagsRegU cr, eRegP op1, memory op2 ) %{
12093   predicate( !n->in(2)->in(2)->bottom_type()->isa_oop_ptr() );
12094   match(Set cr (CmpP op1 (LoadP op2)));
12095 
12096   format %{ "CMPu   $op1,$op2" %}
12097   opcode(0x3B);  /* Opcode 3B /r */
12098   ins_encode( OpcP, RegMem( op1, op2) );
12099   ins_pipe( ialu_cr_reg_mem );
12100 %}
12101 
12102 //
12103 // This will generate a signed flags result. This should be ok
12104 // since any compare to a zero should be eq/neq.
12105 instruct testP_reg( eFlagsReg cr, eRegP src, immP0 zero ) %{
12106   match(Set cr (CmpP src zero));
12107 
12108   format %{ "TEST   $src,$src" %}
12109   opcode(0x85);
12110   ins_encode( OpcP, RegReg( src, src ) );
12111   ins_pipe( ialu_cr_reg_imm );
12112 %}
12113 
12114 // Cisc-spilled version of testP_reg
12115 // This will generate a signed flags result. This should be ok
12116 // since any compare to a zero should be eq/neq.
12117 instruct testP_Reg_mem( eFlagsReg cr, memory op, immI0 zero ) %{
12118   match(Set cr (CmpP (LoadP op) zero));
12119 
12120   format %{ "TEST   $op,0xFFFFFFFF" %}
12121   ins_cost(500);
12122   opcode(0xF7);               /* Opcode F7 /0 */
12123   ins_encode( OpcP, RMopc_Mem(0x00,op), Con_d32(0xFFFFFFFF) );
12124   ins_pipe( ialu_cr_reg_imm );
12125 %}
12126 
12127 // Yanked all unsigned pointer compare operations.
12128 // Pointer compares are done with CmpP which is already unsigned.
12129 
12130 //----------Max and Min--------------------------------------------------------
12131 // Min Instructions
12132 ////
12133 //   *** Min and Max using the conditional move are slower than the
12134 //   *** branch version on a Pentium III.
12135 // // Conditional move for min
12136 //instruct cmovI_reg_lt( eRegI op2, eRegI op1, eFlagsReg cr ) %{
12137 //  effect( USE_DEF op2, USE op1, USE cr );
12138 //  format %{ "CMOVlt $op2,$op1\t! min" %}
12139 //  opcode(0x4C,0x0F);
12140 //  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
12141 //  ins_pipe( pipe_cmov_reg );
12142 //%}
12143 //
12144 //// Min Register with Register (P6 version)
12145 //instruct minI_eReg_p6( eRegI op1, eRegI op2 ) %{
12146 //  predicate(VM_Version::supports_cmov() );
12147 //  match(Set op2 (MinI op1 op2));
12148 //  ins_cost(200);
12149 //  expand %{
12150 //    eFlagsReg cr;
12151 //    compI_eReg(cr,op1,op2);
12152 //    cmovI_reg_lt(op2,op1,cr);
12153 //  %}
12154 //%}
12155 
12156 // Min Register with Register (generic version)
12157 instruct minI_eReg(eRegI dst, eRegI src, eFlagsReg flags) %{
12158   match(Set dst (MinI dst src));
12159   effect(KILL flags);
12160   ins_cost(300);
12161 
12162   format %{ "MIN    $dst,$src" %}
12163   opcode(0xCC);
12164   ins_encode( min_enc(dst,src) );
12165   ins_pipe( pipe_slow );
12166 %}
12167 
12168 // Max Register with Register
12169 //   *** Min and Max using the conditional move are slower than the
12170 //   *** branch version on a Pentium III.
12171 // // Conditional move for max
12172 //instruct cmovI_reg_gt( eRegI op2, eRegI op1, eFlagsReg cr ) %{
12173 //  effect( USE_DEF op2, USE op1, USE cr );
12174 //  format %{ "CMOVgt $op2,$op1\t! max" %}
12175 //  opcode(0x4F,0x0F);
12176 //  ins_encode( OpcS, OpcP, RegReg( op2, op1 ) );
12177 //  ins_pipe( pipe_cmov_reg );
12178 //%}
12179 //
12180 // // Max Register with Register (P6 version)
12181 //instruct maxI_eReg_p6( eRegI op1, eRegI op2 ) %{
12182 //  predicate(VM_Version::supports_cmov() );
12183 //  match(Set op2 (MaxI op1 op2));
12184 //  ins_cost(200);
12185 //  expand %{
12186 //    eFlagsReg cr;
12187 //    compI_eReg(cr,op1,op2);
12188 //    cmovI_reg_gt(op2,op1,cr);
12189 //  %}
12190 //%}
12191 
12192 // Max Register with Register (generic version)
12193 instruct maxI_eReg(eRegI dst, eRegI src, eFlagsReg flags) %{
12194   match(Set dst (MaxI dst src));
12195   effect(KILL flags);
12196   ins_cost(300);
12197 
12198   format %{ "MAX    $dst,$src" %}
12199   opcode(0xCC);
12200   ins_encode( max_enc(dst,src) );
12201   ins_pipe( pipe_slow );
12202 %}
12203 
12204 // ============================================================================
12205 // Counted Loop limit node which represents exact final iterator value.
12206 // Note: the resulting value should fit into integer range since
12207 // counted loops have limit check on overflow.
12208 instruct loopLimit_eReg(eAXRegI limit, nadxRegI init, immI stride, eDXRegI limit_hi, nadxRegI tmp, eFlagsReg flags) %{
12209   match(Set limit (LoopLimit (Binary init limit) stride));
12210   effect(TEMP limit_hi, TEMP tmp, KILL flags);
12211   ins_cost(300);
12212 
12213   format %{ "loopLimit $init,$limit,$stride  # $limit = $init + $stride *( $limit - $init + $stride -1)/ $stride, kills $limit_hi" %}
12214   ins_encode %{
12215     int strd = (int)$stride$$constant;
12216     assert(strd != 1 && strd != -1, "sanity");
12217     int m1 = (strd > 0) ? 1 : -1;
12218     // Convert limit to long (EAX:EDX)
12219     __ cdql();
12220     // Convert init to long (init:tmp)
12221     __ movl($tmp$$Register, $init$$Register);
12222     __ sarl($tmp$$Register, 31);
12223     // $limit - $init
12224     __ subl($limit$$Register, $init$$Register);
12225     __ sbbl($limit_hi$$Register, $tmp$$Register);
12226     // + ($stride - 1)
12227     if (strd > 0) {
12228       __ addl($limit$$Register, (strd - 1));
12229       __ adcl($limit_hi$$Register, 0);
12230       __ movl($tmp$$Register, strd);
12231     } else {
12232       __ addl($limit$$Register, (strd + 1));
12233       __ adcl($limit_hi$$Register, -1);
12234       __ lneg($limit_hi$$Register, $limit$$Register);
12235       __ movl($tmp$$Register, -strd);
12236     }
12237     // signed devision: (EAX:EDX) / pos_stride
12238     __ idivl($tmp$$Register);
12239     if (strd < 0) {
12240       // restore sign
12241       __ negl($tmp$$Register);
12242     }
12243     // (EAX) * stride
12244     __ mull($tmp$$Register);
12245     // + init (ignore upper bits)
12246     __ addl($limit$$Register, $init$$Register);
12247   %}
12248   ins_pipe( pipe_slow );
12249 %}
12250 
12251 // ============================================================================
12252 // Branch Instructions
12253 // Jump Table
12254 instruct jumpXtnd(eRegI switch_val) %{
12255   match(Jump switch_val);
12256   ins_cost(350);
12257   format %{  "JMP    [$constantaddress](,$switch_val,1)\n\t" %}
12258   ins_encode %{
12259     // Jump to Address(table_base + switch_reg)
12260     Address index(noreg, $switch_val$$Register, Address::times_1);
12261     __ jump(ArrayAddress($constantaddress, index));
12262   %}
12263   ins_pipe(pipe_jmp);
12264 %}
12265 
12266 // Jump Direct - Label defines a relative address from JMP+1
12267 instruct jmpDir(label labl) %{
12268   match(Goto);
12269   effect(USE labl);
12270 
12271   ins_cost(300);
12272   format %{ "JMP    $labl" %}
12273   size(5);
12274   ins_encode %{
12275     Label* L = $labl$$label;
12276     __ jmp(*L, false); // Always long jump
12277   %}
12278   ins_pipe( pipe_jmp );
12279 %}
12280 
12281 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12282 instruct jmpCon(cmpOp cop, eFlagsReg cr, label labl) %{
12283   match(If cop cr);
12284   effect(USE labl);
12285 
12286   ins_cost(300);
12287   format %{ "J$cop    $labl" %}
12288   size(6);
12289   ins_encode %{
12290     Label* L = $labl$$label;
12291     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12292   %}
12293   ins_pipe( pipe_jcc );
12294 %}
12295 
12296 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12297 instruct jmpLoopEnd(cmpOp cop, eFlagsReg cr, label labl) %{
12298   match(CountedLoopEnd cop cr);
12299   effect(USE labl);
12300 
12301   ins_cost(300);
12302   format %{ "J$cop    $labl\t# Loop end" %}
12303   size(6);
12304   ins_encode %{
12305     Label* L = $labl$$label;
12306     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12307   %}
12308   ins_pipe( pipe_jcc );
12309 %}
12310 
12311 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12312 instruct jmpLoopEndU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12313   match(CountedLoopEnd cop cmp);
12314   effect(USE labl);
12315 
12316   ins_cost(300);
12317   format %{ "J$cop,u  $labl\t# Loop end" %}
12318   size(6);
12319   ins_encode %{
12320     Label* L = $labl$$label;
12321     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12322   %}
12323   ins_pipe( pipe_jcc );
12324 %}
12325 
12326 instruct jmpLoopEndUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12327   match(CountedLoopEnd cop cmp);
12328   effect(USE labl);
12329 
12330   ins_cost(200);
12331   format %{ "J$cop,u  $labl\t# Loop end" %}
12332   size(6);
12333   ins_encode %{
12334     Label* L = $labl$$label;
12335     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12336   %}
12337   ins_pipe( pipe_jcc );
12338 %}
12339 
12340 // Jump Direct Conditional - using unsigned comparison
12341 instruct jmpConU(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12342   match(If cop cmp);
12343   effect(USE labl);
12344 
12345   ins_cost(300);
12346   format %{ "J$cop,u  $labl" %}
12347   size(6);
12348   ins_encode %{
12349     Label* L = $labl$$label;
12350     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12351   %}
12352   ins_pipe(pipe_jcc);
12353 %}
12354 
12355 instruct jmpConUCF(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12356   match(If cop cmp);
12357   effect(USE labl);
12358 
12359   ins_cost(200);
12360   format %{ "J$cop,u  $labl" %}
12361   size(6);
12362   ins_encode %{
12363     Label* L = $labl$$label;
12364     __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
12365   %}
12366   ins_pipe(pipe_jcc);
12367 %}
12368 
12369 instruct jmpConUCF2(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
12370   match(If cop cmp);
12371   effect(USE labl);
12372 
12373   ins_cost(200);
12374   format %{ $$template
12375     if ($cop$$cmpcode == Assembler::notEqual) {
12376       $$emit$$"JP,u   $labl\n\t"
12377       $$emit$$"J$cop,u   $labl"
12378     } else {
12379       $$emit$$"JP,u   done\n\t"
12380       $$emit$$"J$cop,u   $labl\n\t"
12381       $$emit$$"done:"
12382     }
12383   %}
12384   ins_encode %{
12385     Label* l = $labl$$label;
12386     if ($cop$$cmpcode == Assembler::notEqual) {
12387       __ jcc(Assembler::parity, *l, false);
12388       __ jcc(Assembler::notEqual, *l, false);
12389     } else if ($cop$$cmpcode == Assembler::equal) {
12390       Label done;
12391       __ jccb(Assembler::parity, done);
12392       __ jcc(Assembler::equal, *l, false);
12393       __ bind(done);
12394     } else {
12395        ShouldNotReachHere();
12396     }
12397   %}
12398   ins_pipe(pipe_jcc);
12399 %}
12400 
12401 // ============================================================================
12402 // The 2nd slow-half of a subtype check.  Scan the subklass's 2ndary superklass
12403 // array for an instance of the superklass.  Set a hidden internal cache on a
12404 // hit (cache is checked with exposed code in gen_subtype_check()).  Return
12405 // NZ for a miss or zero for a hit.  The encoding ALSO sets flags.
12406 instruct partialSubtypeCheck( eDIRegP result, eSIRegP sub, eAXRegP super, eCXRegI rcx, eFlagsReg cr ) %{
12407   match(Set result (PartialSubtypeCheck sub super));
12408   effect( KILL rcx, KILL cr );
12409 
12410   ins_cost(1100);  // slightly larger than the next version
12411   format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
12412             "MOV    ECX,[EDI+arrayKlass::length]\t# length to scan\n\t"
12413             "ADD    EDI,arrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
12414             "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
12415             "JNE,s  miss\t\t# Missed: EDI not-zero\n\t"
12416             "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache\n\t"
12417             "XOR    $result,$result\t\t Hit: EDI zero\n\t"
12418      "miss:\t" %}
12419 
12420   opcode(0x1); // Force a XOR of EDI
12421   ins_encode( enc_PartialSubtypeCheck() );
12422   ins_pipe( pipe_slow );
12423 %}
12424 
12425 instruct partialSubtypeCheck_vs_Zero( eFlagsReg cr, eSIRegP sub, eAXRegP super, eCXRegI rcx, eDIRegP result, immP0 zero ) %{
12426   match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12427   effect( KILL rcx, KILL result );
12428 
12429   ins_cost(1000);
12430   format %{ "MOV    EDI,[$sub+Klass::secondary_supers]\n\t"
12431             "MOV    ECX,[EDI+arrayKlass::length]\t# length to scan\n\t"
12432             "ADD    EDI,arrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t"
12433             "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
12434             "JNE,s  miss\t\t# Missed: flags NZ\n\t"
12435             "MOV    [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache, flags Z\n\t"
12436      "miss:\t" %}
12437 
12438   opcode(0x0);  // No need to XOR EDI
12439   ins_encode( enc_PartialSubtypeCheck() );
12440   ins_pipe( pipe_slow );
12441 %}
12442 
12443 // ============================================================================
12444 // Branch Instructions -- short offset versions
12445 //
12446 // These instructions are used to replace jumps of a long offset (the default
12447 // match) with jumps of a shorter offset.  These instructions are all tagged
12448 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12449 // match rules in general matching.  Instead, the ADLC generates a conversion
12450 // method in the MachNode which can be used to do in-place replacement of the
12451 // long variant with the shorter variant.  The compiler will determine if a
12452 // branch can be taken by the is_short_branch_offset() predicate in the machine
12453 // specific code section of the file.
12454 
12455 // Jump Direct - Label defines a relative address from JMP+1
12456 instruct jmpDir_short(label labl) %{
12457   match(Goto);
12458   effect(USE labl);
12459 
12460   ins_cost(300);
12461   format %{ "JMP,s  $labl" %}
12462   size(2);
12463   ins_encode %{
12464     Label* L = $labl$$label;
12465     __ jmpb(*L);
12466   %}
12467   ins_pipe( pipe_jmp );
12468   ins_short_branch(1);
12469 %}
12470 
12471 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12472 instruct jmpCon_short(cmpOp cop, eFlagsReg cr, label labl) %{
12473   match(If cop cr);
12474   effect(USE labl);
12475 
12476   ins_cost(300);
12477   format %{ "J$cop,s  $labl" %}
12478   size(2);
12479   ins_encode %{
12480     Label* L = $labl$$label;
12481     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12482   %}
12483   ins_pipe( pipe_jcc );
12484   ins_short_branch(1);
12485 %}
12486 
12487 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12488 instruct jmpLoopEnd_short(cmpOp cop, eFlagsReg cr, label labl) %{
12489   match(CountedLoopEnd cop cr);
12490   effect(USE labl);
12491 
12492   ins_cost(300);
12493   format %{ "J$cop,s  $labl\t# Loop end" %}
12494   size(2);
12495   ins_encode %{
12496     Label* L = $labl$$label;
12497     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12498   %}
12499   ins_pipe( pipe_jcc );
12500   ins_short_branch(1);
12501 %}
12502 
12503 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12504 instruct jmpLoopEndU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12505   match(CountedLoopEnd cop cmp);
12506   effect(USE labl);
12507 
12508   ins_cost(300);
12509   format %{ "J$cop,us $labl\t# Loop end" %}
12510   size(2);
12511   ins_encode %{
12512     Label* L = $labl$$label;
12513     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12514   %}
12515   ins_pipe( pipe_jcc );
12516   ins_short_branch(1);
12517 %}
12518 
12519 instruct jmpLoopEndUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12520   match(CountedLoopEnd cop cmp);
12521   effect(USE labl);
12522 
12523   ins_cost(300);
12524   format %{ "J$cop,us $labl\t# Loop end" %}
12525   size(2);
12526   ins_encode %{
12527     Label* L = $labl$$label;
12528     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12529   %}
12530   ins_pipe( pipe_jcc );
12531   ins_short_branch(1);
12532 %}
12533 
12534 // Jump Direct Conditional - using unsigned comparison
12535 instruct jmpConU_short(cmpOpU cop, eFlagsRegU cmp, label labl) %{
12536   match(If cop cmp);
12537   effect(USE labl);
12538 
12539   ins_cost(300);
12540   format %{ "J$cop,us $labl" %}
12541   size(2);
12542   ins_encode %{
12543     Label* L = $labl$$label;
12544     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12545   %}
12546   ins_pipe( pipe_jcc );
12547   ins_short_branch(1);
12548 %}
12549 
12550 instruct jmpConUCF_short(cmpOpUCF cop, eFlagsRegUCF cmp, label labl) %{
12551   match(If cop cmp);
12552   effect(USE labl);
12553 
12554   ins_cost(300);
12555   format %{ "J$cop,us $labl" %}
12556   size(2);
12557   ins_encode %{
12558     Label* L = $labl$$label;
12559     __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
12560   %}
12561   ins_pipe( pipe_jcc );
12562   ins_short_branch(1);
12563 %}
12564 
12565 instruct jmpConUCF2_short(cmpOpUCF2 cop, eFlagsRegUCF cmp, label labl) %{
12566   match(If cop cmp);
12567   effect(USE labl);
12568 
12569   ins_cost(300);
12570   format %{ $$template
12571     if ($cop$$cmpcode == Assembler::notEqual) {
12572       $$emit$$"JP,u,s   $labl\n\t"
12573       $$emit$$"J$cop,u,s   $labl"
12574     } else {
12575       $$emit$$"JP,u,s   done\n\t"
12576       $$emit$$"J$cop,u,s  $labl\n\t"
12577       $$emit$$"done:"
12578     }
12579   %}
12580   size(4);
12581   ins_encode %{
12582     Label* l = $labl$$label;
12583     if ($cop$$cmpcode == Assembler::notEqual) {
12584       __ jccb(Assembler::parity, *l);
12585       __ jccb(Assembler::notEqual, *l);
12586     } else if ($cop$$cmpcode == Assembler::equal) {
12587       Label done;
12588       __ jccb(Assembler::parity, done);
12589       __ jccb(Assembler::equal, *l);
12590       __ bind(done);
12591     } else {
12592        ShouldNotReachHere();
12593     }
12594   %}
12595   ins_pipe(pipe_jcc);
12596   ins_short_branch(1);
12597 %}
12598 
12599 // ============================================================================
12600 // Long Compare
12601 //
12602 // Currently we hold longs in 2 registers.  Comparing such values efficiently
12603 // is tricky.  The flavor of compare used depends on whether we are testing
12604 // for LT, LE, or EQ.  For a simple LT test we can check just the sign bit.
12605 // The GE test is the negated LT test.  The LE test can be had by commuting
12606 // the operands (yielding a GE test) and then negating; negate again for the
12607 // GT test.  The EQ test is done by ORcc'ing the high and low halves, and the
12608 // NE test is negated from that.
12609 
12610 // Due to a shortcoming in the ADLC, it mixes up expressions like:
12611 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)).  Note the
12612 // difference between 'Y' and '0L'.  The tree-matches for the CmpI sections
12613 // are collapsed internally in the ADLC's dfa-gen code.  The match for
12614 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
12615 // foo match ends up with the wrong leaf.  One fix is to not match both
12616 // reg-reg and reg-zero forms of long-compare.  This is unfortunate because
12617 // both forms beat the trinary form of long-compare and both are very useful
12618 // on Intel which has so few registers.
12619 
12620 // Manifest a CmpL result in an integer register.  Very painful.
12621 // This is the test to avoid.
12622 instruct cmpL3_reg_reg(eSIRegI dst, eRegL src1, eRegL src2, eFlagsReg flags ) %{
12623   match(Set dst (CmpL3 src1 src2));
12624   effect( KILL flags );
12625   ins_cost(1000);
12626   format %{ "XOR    $dst,$dst\n\t"
12627             "CMP    $src1.hi,$src2.hi\n\t"
12628             "JLT,s  m_one\n\t"
12629             "JGT,s  p_one\n\t"
12630             "CMP    $src1.lo,$src2.lo\n\t"
12631             "JB,s   m_one\n\t"
12632             "JEQ,s  done\n"
12633     "p_one:\tINC    $dst\n\t"
12634             "JMP,s  done\n"
12635     "m_one:\tDEC    $dst\n"
12636      "done:" %}
12637   ins_encode %{
12638     Label p_one, m_one, done;
12639     __ xorptr($dst$$Register, $dst$$Register);
12640     __ cmpl(HIGH_FROM_LOW($src1$$Register), HIGH_FROM_LOW($src2$$Register));
12641     __ jccb(Assembler::less,    m_one);
12642     __ jccb(Assembler::greater, p_one);
12643     __ cmpl($src1$$Register, $src2$$Register);
12644     __ jccb(Assembler::below,   m_one);
12645     __ jccb(Assembler::equal,   done);
12646     __ bind(p_one);
12647     __ incrementl($dst$$Register);
12648     __ jmpb(done);
12649     __ bind(m_one);
12650     __ decrementl($dst$$Register);
12651     __ bind(done);
12652   %}
12653   ins_pipe( pipe_slow );
12654 %}
12655 
12656 //======
12657 // Manifest a CmpL result in the normal flags.  Only good for LT or GE
12658 // compares.  Can be used for LE or GT compares by reversing arguments.
12659 // NOT GOOD FOR EQ/NE tests.
12660 instruct cmpL_zero_flags_LTGE( flagsReg_long_LTGE flags, eRegL src, immL0 zero ) %{
12661   match( Set flags (CmpL src zero ));
12662   ins_cost(100);
12663   format %{ "TEST   $src.hi,$src.hi" %}
12664   opcode(0x85);
12665   ins_encode( OpcP, RegReg_Hi2( src, src ) );
12666   ins_pipe( ialu_cr_reg_reg );
12667 %}
12668 
12669 // Manifest a CmpL result in the normal flags.  Only good for LT or GE
12670 // compares.  Can be used for LE or GT compares by reversing arguments.
12671 // NOT GOOD FOR EQ/NE tests.
12672 instruct cmpL_reg_flags_LTGE( flagsReg_long_LTGE flags, eRegL src1, eRegL src2, eRegI tmp ) %{
12673   match( Set flags (CmpL src1 src2 ));
12674   effect( TEMP tmp );
12675   ins_cost(300);
12676   format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
12677             "MOV    $tmp,$src1.hi\n\t"
12678             "SBB    $tmp,$src2.hi\t! Compute flags for long compare" %}
12679   ins_encode( long_cmp_flags2( src1, src2, tmp ) );
12680   ins_pipe( ialu_cr_reg_reg );
12681 %}
12682 
12683 // Long compares reg < zero/req OR reg >= zero/req.
12684 // Just a wrapper for a normal branch, plus the predicate test.
12685 instruct cmpL_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, label labl) %{
12686   match(If cmp flags);
12687   effect(USE labl);
12688   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::lt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ge );
12689   expand %{
12690     jmpCon(cmp,flags,labl);    // JLT or JGE...
12691   %}
12692 %}
12693 
12694 // Compare 2 longs and CMOVE longs.
12695 instruct cmovLL_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, eRegL src) %{
12696   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12697   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 ));
12698   ins_cost(400);
12699   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12700             "CMOV$cmp $dst.hi,$src.hi" %}
12701   opcode(0x0F,0x40);
12702   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12703   ins_pipe( pipe_cmov_reg_long );
12704 %}
12705 
12706 instruct cmovLL_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegL dst, load_long_memory src) %{
12707   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12708   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 ));
12709   ins_cost(500);
12710   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12711             "CMOV$cmp $dst.hi,$src.hi" %}
12712   opcode(0x0F,0x40);
12713   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12714   ins_pipe( pipe_cmov_reg_long );
12715 %}
12716 
12717 // Compare 2 longs and CMOVE ints.
12718 instruct cmovII_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegI dst, eRegI src) %{
12719   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 ));
12720   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12721   ins_cost(200);
12722   format %{ "CMOV$cmp $dst,$src" %}
12723   opcode(0x0F,0x40);
12724   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12725   ins_pipe( pipe_cmov_reg );
12726 %}
12727 
12728 instruct cmovII_mem_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegI dst, memory src) %{
12729   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 ));
12730   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12731   ins_cost(250);
12732   format %{ "CMOV$cmp $dst,$src" %}
12733   opcode(0x0F,0x40);
12734   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12735   ins_pipe( pipe_cmov_mem );
12736 %}
12737 
12738 // Compare 2 longs and CMOVE ints.
12739 instruct cmovPP_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, eRegP dst, eRegP src) %{
12740   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 ));
12741   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
12742   ins_cost(200);
12743   format %{ "CMOV$cmp $dst,$src" %}
12744   opcode(0x0F,0x40);
12745   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12746   ins_pipe( pipe_cmov_reg );
12747 %}
12748 
12749 // Compare 2 longs and CMOVE doubles
12750 instruct cmovDDPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regDPR dst, regDPR src) %{
12751   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 );
12752   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12753   ins_cost(200);
12754   expand %{
12755     fcmovDPR_regS(cmp,flags,dst,src);
12756   %}
12757 %}
12758 
12759 // Compare 2 longs and CMOVE doubles
12760 instruct cmovDD_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regD dst, regD src) %{
12761   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 );
12762   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12763   ins_cost(200);
12764   expand %{
12765     fcmovD_regS(cmp,flags,dst,src);
12766   %}
12767 %}
12768 
12769 instruct cmovFFPR_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regFPR dst, regFPR src) %{
12770   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 );
12771   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12772   ins_cost(200);
12773   expand %{
12774     fcmovFPR_regS(cmp,flags,dst,src);
12775   %}
12776 %}
12777 
12778 instruct cmovFF_reg_LTGE(cmpOp cmp, flagsReg_long_LTGE flags, regF dst, regF src) %{
12779   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 );
12780   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12781   ins_cost(200);
12782   expand %{
12783     fcmovF_regS(cmp,flags,dst,src);
12784   %}
12785 %}
12786 
12787 //======
12788 // Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
12789 instruct cmpL_zero_flags_EQNE( flagsReg_long_EQNE flags, eRegL src, immL0 zero, eRegI tmp ) %{
12790   match( Set flags (CmpL src zero ));
12791   effect(TEMP tmp);
12792   ins_cost(200);
12793   format %{ "MOV    $tmp,$src.lo\n\t"
12794             "OR     $tmp,$src.hi\t! Long is EQ/NE 0?" %}
12795   ins_encode( long_cmp_flags0( src, tmp ) );
12796   ins_pipe( ialu_reg_reg_long );
12797 %}
12798 
12799 // Manifest a CmpL result in the normal flags.  Only good for EQ/NE compares.
12800 instruct cmpL_reg_flags_EQNE( flagsReg_long_EQNE flags, eRegL src1, eRegL src2 ) %{
12801   match( Set flags (CmpL src1 src2 ));
12802   ins_cost(200+300);
12803   format %{ "CMP    $src1.lo,$src2.lo\t! Long compare; set flags for low bits\n\t"
12804             "JNE,s  skip\n\t"
12805             "CMP    $src1.hi,$src2.hi\n\t"
12806      "skip:\t" %}
12807   ins_encode( long_cmp_flags1( src1, src2 ) );
12808   ins_pipe( ialu_cr_reg_reg );
12809 %}
12810 
12811 // Long compare reg == zero/reg OR reg != zero/reg
12812 // Just a wrapper for a normal branch, plus the predicate test.
12813 instruct cmpL_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, label labl) %{
12814   match(If cmp flags);
12815   effect(USE labl);
12816   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::eq || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne );
12817   expand %{
12818     jmpCon(cmp,flags,labl);    // JEQ or JNE...
12819   %}
12820 %}
12821 
12822 // Compare 2 longs and CMOVE longs.
12823 instruct cmovLL_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, eRegL src) %{
12824   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12825   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 ));
12826   ins_cost(400);
12827   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12828             "CMOV$cmp $dst.hi,$src.hi" %}
12829   opcode(0x0F,0x40);
12830   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12831   ins_pipe( pipe_cmov_reg_long );
12832 %}
12833 
12834 instruct cmovLL_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegL dst, load_long_memory src) %{
12835   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12836   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 ));
12837   ins_cost(500);
12838   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12839             "CMOV$cmp $dst.hi,$src.hi" %}
12840   opcode(0x0F,0x40);
12841   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12842   ins_pipe( pipe_cmov_reg_long );
12843 %}
12844 
12845 // Compare 2 longs and CMOVE ints.
12846 instruct cmovII_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegI dst, eRegI src) %{
12847   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 ));
12848   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12849   ins_cost(200);
12850   format %{ "CMOV$cmp $dst,$src" %}
12851   opcode(0x0F,0x40);
12852   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12853   ins_pipe( pipe_cmov_reg );
12854 %}
12855 
12856 instruct cmovII_mem_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegI dst, memory src) %{
12857   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 ));
12858   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12859   ins_cost(250);
12860   format %{ "CMOV$cmp $dst,$src" %}
12861   opcode(0x0F,0x40);
12862   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12863   ins_pipe( pipe_cmov_mem );
12864 %}
12865 
12866 // Compare 2 longs and CMOVE ints.
12867 instruct cmovPP_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, eRegP dst, eRegP src) %{
12868   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 ));
12869   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
12870   ins_cost(200);
12871   format %{ "CMOV$cmp $dst,$src" %}
12872   opcode(0x0F,0x40);
12873   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12874   ins_pipe( pipe_cmov_reg );
12875 %}
12876 
12877 // Compare 2 longs and CMOVE doubles
12878 instruct cmovDDPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regDPR dst, regDPR src) %{
12879   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 );
12880   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12881   ins_cost(200);
12882   expand %{
12883     fcmovDPR_regS(cmp,flags,dst,src);
12884   %}
12885 %}
12886 
12887 // Compare 2 longs and CMOVE doubles
12888 instruct cmovDD_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regD dst, regD src) %{
12889   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 );
12890   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
12891   ins_cost(200);
12892   expand %{
12893     fcmovD_regS(cmp,flags,dst,src);
12894   %}
12895 %}
12896 
12897 instruct cmovFFPR_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regFPR dst, regFPR src) %{
12898   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 );
12899   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12900   ins_cost(200);
12901   expand %{
12902     fcmovFPR_regS(cmp,flags,dst,src);
12903   %}
12904 %}
12905 
12906 instruct cmovFF_reg_EQNE(cmpOp cmp, flagsReg_long_EQNE flags, regF dst, regF src) %{
12907   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 );
12908   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
12909   ins_cost(200);
12910   expand %{
12911     fcmovF_regS(cmp,flags,dst,src);
12912   %}
12913 %}
12914 
12915 //======
12916 // Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
12917 // Same as cmpL_reg_flags_LEGT except must negate src
12918 instruct cmpL_zero_flags_LEGT( flagsReg_long_LEGT flags, eRegL src, immL0 zero, eRegI tmp ) %{
12919   match( Set flags (CmpL src zero ));
12920   effect( TEMP tmp );
12921   ins_cost(300);
12922   format %{ "XOR    $tmp,$tmp\t# Long compare for -$src < 0, use commuted test\n\t"
12923             "CMP    $tmp,$src.lo\n\t"
12924             "SBB    $tmp,$src.hi\n\t" %}
12925   ins_encode( long_cmp_flags3(src, tmp) );
12926   ins_pipe( ialu_reg_reg_long );
12927 %}
12928 
12929 // Manifest a CmpL result in the normal flags.  Only good for LE or GT compares.
12930 // Same as cmpL_reg_flags_LTGE except operands swapped.  Swapping operands
12931 // requires a commuted test to get the same result.
12932 instruct cmpL_reg_flags_LEGT( flagsReg_long_LEGT flags, eRegL src1, eRegL src2, eRegI tmp ) %{
12933   match( Set flags (CmpL src1 src2 ));
12934   effect( TEMP tmp );
12935   ins_cost(300);
12936   format %{ "CMP    $src2.lo,$src1.lo\t! Long compare, swapped operands, use with commuted test\n\t"
12937             "MOV    $tmp,$src2.hi\n\t"
12938             "SBB    $tmp,$src1.hi\t! Compute flags for long compare" %}
12939   ins_encode( long_cmp_flags2( src2, src1, tmp ) );
12940   ins_pipe( ialu_cr_reg_reg );
12941 %}
12942 
12943 // Long compares reg < zero/req OR reg >= zero/req.
12944 // Just a wrapper for a normal branch, plus the predicate test
12945 instruct cmpL_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, label labl) %{
12946   match(If cmp flags);
12947   effect(USE labl);
12948   predicate( _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::gt || _kids[0]->_leaf->as_Bool()->_test._test == BoolTest::le );
12949   ins_cost(300);
12950   expand %{
12951     jmpCon(cmp,flags,labl);    // JGT or JLE...
12952   %}
12953 %}
12954 
12955 // Compare 2 longs and CMOVE longs.
12956 instruct cmovLL_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, eRegL src) %{
12957   match(Set dst (CMoveL (Binary cmp flags) (Binary dst src)));
12958   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 ));
12959   ins_cost(400);
12960   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12961             "CMOV$cmp $dst.hi,$src.hi" %}
12962   opcode(0x0F,0x40);
12963   ins_encode( enc_cmov(cmp), RegReg_Lo2( dst, src ), enc_cmov(cmp), RegReg_Hi2( dst, src ) );
12964   ins_pipe( pipe_cmov_reg_long );
12965 %}
12966 
12967 instruct cmovLL_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegL dst, load_long_memory src) %{
12968   match(Set dst (CMoveL (Binary cmp flags) (Binary dst (LoadL src))));
12969   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 ));
12970   ins_cost(500);
12971   format %{ "CMOV$cmp $dst.lo,$src.lo\n\t"
12972             "CMOV$cmp $dst.hi,$src.hi+4" %}
12973   opcode(0x0F,0x40);
12974   ins_encode( enc_cmov(cmp), RegMem(dst, src), enc_cmov(cmp), RegMem_Hi(dst, src) );
12975   ins_pipe( pipe_cmov_reg_long );
12976 %}
12977 
12978 // Compare 2 longs and CMOVE ints.
12979 instruct cmovII_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegI dst, eRegI src) %{
12980   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 ));
12981   match(Set dst (CMoveI (Binary cmp flags) (Binary dst src)));
12982   ins_cost(200);
12983   format %{ "CMOV$cmp $dst,$src" %}
12984   opcode(0x0F,0x40);
12985   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
12986   ins_pipe( pipe_cmov_reg );
12987 %}
12988 
12989 instruct cmovII_mem_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegI dst, memory src) %{
12990   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 ));
12991   match(Set dst (CMoveI (Binary cmp flags) (Binary dst (LoadI src))));
12992   ins_cost(250);
12993   format %{ "CMOV$cmp $dst,$src" %}
12994   opcode(0x0F,0x40);
12995   ins_encode( enc_cmov(cmp), RegMem( dst, src ) );
12996   ins_pipe( pipe_cmov_mem );
12997 %}
12998 
12999 // Compare 2 longs and CMOVE ptrs.
13000 instruct cmovPP_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, eRegP dst, eRegP src) %{
13001   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 ));
13002   match(Set dst (CMoveP (Binary cmp flags) (Binary dst src)));
13003   ins_cost(200);
13004   format %{ "CMOV$cmp $dst,$src" %}
13005   opcode(0x0F,0x40);
13006   ins_encode( enc_cmov(cmp), RegReg( dst, src ) );
13007   ins_pipe( pipe_cmov_reg );
13008 %}
13009 
13010 // Compare 2 longs and CMOVE doubles
13011 instruct cmovDDPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regDPR dst, regDPR src) %{
13012   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 );
13013   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
13014   ins_cost(200);
13015   expand %{
13016     fcmovDPR_regS(cmp,flags,dst,src);
13017   %}
13018 %}
13019 
13020 // Compare 2 longs and CMOVE doubles
13021 instruct cmovDD_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regD dst, regD src) %{
13022   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 );
13023   match(Set dst (CMoveD (Binary cmp flags) (Binary dst src)));
13024   ins_cost(200);
13025   expand %{
13026     fcmovD_regS(cmp,flags,dst,src);
13027   %}
13028 %}
13029 
13030 instruct cmovFFPR_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regFPR dst, regFPR src) %{
13031   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 );
13032   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
13033   ins_cost(200);
13034   expand %{
13035     fcmovFPR_regS(cmp,flags,dst,src);
13036   %}
13037 %}
13038 
13039 
13040 instruct cmovFF_reg_LEGT(cmpOp_commute cmp, flagsReg_long_LEGT flags, regF dst, regF src) %{
13041   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 );
13042   match(Set dst (CMoveF (Binary cmp flags) (Binary dst src)));
13043   ins_cost(200);
13044   expand %{
13045     fcmovF_regS(cmp,flags,dst,src);
13046   %}
13047 %}
13048 
13049 
13050 // ============================================================================
13051 // Procedure Call/Return Instructions
13052 // Call Java Static Instruction
13053 // Note: If this code changes, the corresponding ret_addr_offset() and
13054 //       compute_padding() functions will have to be adjusted.
13055 instruct CallStaticJavaDirect(method meth) %{
13056   match(CallStaticJava);
13057   predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
13058   effect(USE meth);
13059 
13060   ins_cost(300);
13061   format %{ "CALL,static " %}
13062   opcode(0xE8); /* E8 cd */
13063   ins_encode( pre_call_FPU,
13064               Java_Static_Call( meth ),
13065               call_epilog,
13066               post_call_FPU );
13067   ins_pipe( pipe_slow );
13068   ins_alignment(4);
13069 %}
13070 
13071 // Call Java Static Instruction (method handle version)
13072 // Note: If this code changes, the corresponding ret_addr_offset() and
13073 //       compute_padding() functions will have to be adjusted.
13074 instruct CallStaticJavaHandle(method meth, eBPRegP ebp_mh_SP_save) %{
13075   match(CallStaticJava);
13076   predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
13077   effect(USE meth);
13078   // EBP is saved by all callees (for interpreter stack correction).
13079   // We use it here for a similar purpose, in {preserve,restore}_SP.
13080 
13081   ins_cost(300);
13082   format %{ "CALL,static/MethodHandle " %}
13083   opcode(0xE8); /* E8 cd */
13084   ins_encode( pre_call_FPU,
13085               preserve_SP,
13086               Java_Static_Call( meth ),
13087               restore_SP,
13088               call_epilog,
13089               post_call_FPU );
13090   ins_pipe( pipe_slow );
13091   ins_alignment(4);
13092 %}
13093 
13094 // Call Java Dynamic Instruction
13095 // Note: If this code changes, the corresponding ret_addr_offset() and
13096 //       compute_padding() functions will have to be adjusted.
13097 instruct CallDynamicJavaDirect(method meth) %{
13098   match(CallDynamicJava);
13099   effect(USE meth);
13100 
13101   ins_cost(300);
13102   format %{ "MOV    EAX,(oop)-1\n\t"
13103             "CALL,dynamic" %}
13104   opcode(0xE8); /* E8 cd */
13105   ins_encode( pre_call_FPU,
13106               Java_Dynamic_Call( meth ),
13107               call_epilog,
13108               post_call_FPU );
13109   ins_pipe( pipe_slow );
13110   ins_alignment(4);
13111 %}
13112 
13113 // Call Runtime Instruction
13114 instruct CallRuntimeDirect(method meth) %{
13115   match(CallRuntime );
13116   effect(USE meth);
13117 
13118   ins_cost(300);
13119   format %{ "CALL,runtime " %}
13120   opcode(0xE8); /* E8 cd */
13121   // Use FFREEs to clear entries in float stack
13122   ins_encode( pre_call_FPU,
13123               FFree_Float_Stack_All,
13124               Java_To_Runtime( meth ),
13125               post_call_FPU );
13126   ins_pipe( pipe_slow );
13127 %}
13128 
13129 // Call runtime without safepoint
13130 instruct CallLeafDirect(method meth) %{
13131   match(CallLeaf);
13132   effect(USE meth);
13133 
13134   ins_cost(300);
13135   format %{ "CALL_LEAF,runtime " %}
13136   opcode(0xE8); /* E8 cd */
13137   ins_encode( pre_call_FPU,
13138               FFree_Float_Stack_All,
13139               Java_To_Runtime( meth ),
13140               Verify_FPU_For_Leaf, post_call_FPU );
13141   ins_pipe( pipe_slow );
13142 %}
13143 
13144 instruct CallLeafNoFPDirect(method meth) %{
13145   match(CallLeafNoFP);
13146   effect(USE meth);
13147 
13148   ins_cost(300);
13149   format %{ "CALL_LEAF_NOFP,runtime " %}
13150   opcode(0xE8); /* E8 cd */
13151   ins_encode(Java_To_Runtime(meth));
13152   ins_pipe( pipe_slow );
13153 %}
13154 
13155 
13156 // Return Instruction
13157 // Remove the return address & jump to it.
13158 instruct Ret() %{
13159   match(Return);
13160   format %{ "RET" %}
13161   opcode(0xC3);
13162   ins_encode(OpcP);
13163   ins_pipe( pipe_jmp );
13164 %}
13165 
13166 // Tail Call; Jump from runtime stub to Java code.
13167 // Also known as an 'interprocedural jump'.
13168 // Target of jump will eventually return to caller.
13169 // TailJump below removes the return address.
13170 instruct TailCalljmpInd(eRegP_no_EBP jump_target, eBXRegP method_oop) %{
13171   match(TailCall jump_target method_oop );
13172   ins_cost(300);
13173   format %{ "JMP    $jump_target \t# EBX holds method oop" %}
13174   opcode(0xFF, 0x4);  /* Opcode FF /4 */
13175   ins_encode( OpcP, RegOpc(jump_target) );
13176   ins_pipe( pipe_jmp );
13177 %}
13178 
13179 
13180 // Tail Jump; remove the return address; jump to target.
13181 // TailCall above leaves the return address around.
13182 instruct tailjmpInd(eRegP_no_EBP jump_target, eAXRegP ex_oop) %{
13183   match( TailJump jump_target ex_oop );
13184   ins_cost(300);
13185   format %{ "POP    EDX\t# pop return address into dummy\n\t"
13186             "JMP    $jump_target " %}
13187   opcode(0xFF, 0x4);  /* Opcode FF /4 */
13188   ins_encode( enc_pop_rdx,
13189               OpcP, RegOpc(jump_target) );
13190   ins_pipe( pipe_jmp );
13191 %}
13192 
13193 // Create exception oop: created by stack-crawling runtime code.
13194 // Created exception is now available to this handler, and is setup
13195 // just prior to jumping to this handler.  No code emitted.
13196 instruct CreateException( eAXRegP ex_oop )
13197 %{
13198   match(Set ex_oop (CreateEx));
13199 
13200   size(0);
13201   // use the following format syntax
13202   format %{ "# exception oop is in EAX; no code emitted" %}
13203   ins_encode();
13204   ins_pipe( empty );
13205 %}
13206 
13207 
13208 // Rethrow exception:
13209 // The exception oop will come in the first argument position.
13210 // Then JUMP (not call) to the rethrow stub code.
13211 instruct RethrowException()
13212 %{
13213   match(Rethrow);
13214 
13215   // use the following format syntax
13216   format %{ "JMP    rethrow_stub" %}
13217   ins_encode(enc_rethrow);
13218   ins_pipe( pipe_jmp );
13219 %}
13220 
13221 // inlined locking and unlocking
13222 
13223 
13224 instruct cmpFastLock( eFlagsReg cr, eRegP object, eBXRegP box, eAXRegI tmp, eRegP scr) %{
13225   match( Set cr (FastLock object box) );
13226   effect( TEMP tmp, TEMP scr, USE_KILL box );
13227   ins_cost(300);
13228   format %{ "FASTLOCK $object,$box\t! kills $box,$tmp,$scr" %}
13229   ins_encode( Fast_Lock(object,box,tmp,scr) );
13230   ins_pipe( pipe_slow );
13231 %}
13232 
13233 instruct cmpFastUnlock( eFlagsReg cr, eRegP object, eAXRegP box, eRegP tmp ) %{
13234   match( Set cr (FastUnlock object box) );
13235   effect( TEMP tmp, USE_KILL box );
13236   ins_cost(300);
13237   format %{ "FASTUNLOCK $object,$box\t! kills $box,$tmp" %}
13238   ins_encode( Fast_Unlock(object,box,tmp) );
13239   ins_pipe( pipe_slow );
13240 %}
13241 
13242 
13243 
13244 // ============================================================================
13245 // Safepoint Instruction
13246 instruct safePoint_poll(eFlagsReg cr) %{
13247   match(SafePoint);
13248   effect(KILL cr);
13249 
13250   // TODO-FIXME: we currently poll at offset 0 of the safepoint polling page.
13251   // On SPARC that might be acceptable as we can generate the address with
13252   // just a sethi, saving an or.  By polling at offset 0 we can end up
13253   // putting additional pressure on the index-0 in the D$.  Because of
13254   // alignment (just like the situation at hand) the lower indices tend
13255   // to see more traffic.  It'd be better to change the polling address
13256   // to offset 0 of the last $line in the polling page.
13257 
13258   format %{ "TSTL   #polladdr,EAX\t! Safepoint: poll for GC" %}
13259   ins_cost(125);
13260   size(6) ;
13261   ins_encode( Safepoint_Poll() );
13262   ins_pipe( ialu_reg_mem );
13263 %}
13264 
13265 
13266 // ============================================================================
13267 // This name is KNOWN by the ADLC and cannot be changed.
13268 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type
13269 // for this guy.
13270 instruct tlsLoadP(eRegP dst, eFlagsReg cr) %{
13271   match(Set dst (ThreadLocal));
13272   effect(DEF dst, KILL cr);
13273 
13274   format %{ "MOV    $dst, Thread::current()" %}
13275   ins_encode %{
13276     Register dstReg = as_Register($dst$$reg);
13277     __ get_thread(dstReg);
13278   %}
13279   ins_pipe( ialu_reg_fat );
13280 %}
13281 
13282 
13283 
13284 //----------PEEPHOLE RULES-----------------------------------------------------
13285 // These must follow all instruction definitions as they use the names
13286 // defined in the instructions definitions.
13287 //
13288 // peepmatch ( root_instr_name [preceding_instruction]* );
13289 //
13290 // peepconstraint %{
13291 // (instruction_number.operand_name relational_op instruction_number.operand_name
13292 //  [, ...] );
13293 // // instruction numbers are zero-based using left to right order in peepmatch
13294 //
13295 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
13296 // // provide an instruction_number.operand_name for each operand that appears
13297 // // in the replacement instruction's match rule
13298 //
13299 // ---------VM FLAGS---------------------------------------------------------
13300 //
13301 // All peephole optimizations can be turned off using -XX:-OptoPeephole
13302 //
13303 // Each peephole rule is given an identifying number starting with zero and
13304 // increasing by one in the order seen by the parser.  An individual peephole
13305 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
13306 // on the command-line.
13307 //
13308 // ---------CURRENT LIMITATIONS----------------------------------------------
13309 //
13310 // Only match adjacent instructions in same basic block
13311 // Only equality constraints
13312 // Only constraints between operands, not (0.dest_reg == EAX_enc)
13313 // Only one replacement instruction
13314 //
13315 // ---------EXAMPLE----------------------------------------------------------
13316 //
13317 // // pertinent parts of existing instructions in architecture description
13318 // instruct movI(eRegI dst, eRegI src) %{
13319 //   match(Set dst (CopyI src));
13320 // %}
13321 //
13322 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
13323 //   match(Set dst (AddI dst src));
13324 //   effect(KILL cr);
13325 // %}
13326 //
13327 // // Change (inc mov) to lea
13328 // peephole %{
13329 //   // increment preceeded by register-register move
13330 //   peepmatch ( incI_eReg movI );
13331 //   // require that the destination register of the increment
13332 //   // match the destination register of the move
13333 //   peepconstraint ( 0.dst == 1.dst );
13334 //   // construct a replacement instruction that sets
13335 //   // the destination to ( move's source register + one )
13336 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13337 // %}
13338 //
13339 // Implementation no longer uses movX instructions since
13340 // machine-independent system no longer uses CopyX nodes.
13341 //
13342 // peephole %{
13343 //   peepmatch ( incI_eReg movI );
13344 //   peepconstraint ( 0.dst == 1.dst );
13345 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13346 // %}
13347 //
13348 // peephole %{
13349 //   peepmatch ( decI_eReg movI );
13350 //   peepconstraint ( 0.dst == 1.dst );
13351 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13352 // %}
13353 //
13354 // peephole %{
13355 //   peepmatch ( addI_eReg_imm movI );
13356 //   peepconstraint ( 0.dst == 1.dst );
13357 //   peepreplace ( leaI_eReg_immI( 0.dst 1.src 0.src ) );
13358 // %}
13359 //
13360 // peephole %{
13361 //   peepmatch ( addP_eReg_imm movP );
13362 //   peepconstraint ( 0.dst == 1.dst );
13363 //   peepreplace ( leaP_eReg_immI( 0.dst 1.src 0.src ) );
13364 // %}
13365 
13366 // // Change load of spilled value to only a spill
13367 // instruct storeI(memory mem, eRegI src) %{
13368 //   match(Set mem (StoreI mem src));
13369 // %}
13370 //
13371 // instruct loadI(eRegI dst, memory mem) %{
13372 //   match(Set dst (LoadI mem));
13373 // %}
13374 //
13375 peephole %{
13376   peepmatch ( loadI storeI );
13377   peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
13378   peepreplace ( storeI( 1.mem 1.mem 1.src ) );
13379 %}
13380 
13381 //----------SMARTSPILL RULES---------------------------------------------------
13382 // These must follow all instruction definitions as they use the names
13383 // defined in the instructions definitions.