Mercurial > hg > truffle
annotate src/cpu/x86/vm/x86_64.ad @ 643:c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
Summary: bitCount() should use POPC on SPARC processors where POPC is implemented directly in hardware.
Reviewed-by: kvn, never
| author | twisti |
|---|---|
| date | Fri, 13 Mar 2009 11:35:17 -0700 |
| parents | 660978a2a31a |
| children | c517646eef23 |
| rev | line source |
|---|---|
| 0 | 1 // |
| 624 | 2 // Copyright 2003-2009 Sun Microsystems, Inc. All Rights Reserved. |
| 0 | 3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 // | |
| 5 // This code is free software; you can redistribute it and/or modify it | |
| 6 // under the terms of the GNU General Public License version 2 only, as | |
| 7 // published by the Free Software Foundation. | |
| 8 // | |
| 9 // This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 // version 2 for more details (a copy is included in the LICENSE file that | |
| 13 // accompanied this code). | |
| 14 // | |
| 15 // You should have received a copy of the GNU General Public License version | |
| 16 // 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 // | |
| 19 // Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 // CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 // have any questions. | |
| 22 // | |
| 23 // | |
| 24 | |
| 25 // AMD64 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 // R8-R15 must be encoded with REX. (RSP, RBP, RSI, RDI need REX when | |
| 64 // used as byte registers) | |
| 65 | |
| 66 // Previously set RBX, RSI, and RDI as save-on-entry for java code | |
| 67 // Turn off SOE in java-code due to frequent use of uncommon-traps. | |
| 68 // Now that allocator is better, turn on RSI and RDI as SOE registers. | |
| 69 | |
| 70 reg_def RAX (SOC, SOC, Op_RegI, 0, rax->as_VMReg()); | |
| 71 reg_def RAX_H(SOC, SOC, Op_RegI, 0, rax->as_VMReg()->next()); | |
| 72 | |
| 73 reg_def RCX (SOC, SOC, Op_RegI, 1, rcx->as_VMReg()); | |
| 74 reg_def RCX_H(SOC, SOC, Op_RegI, 1, rcx->as_VMReg()->next()); | |
| 75 | |
| 76 reg_def RDX (SOC, SOC, Op_RegI, 2, rdx->as_VMReg()); | |
| 77 reg_def RDX_H(SOC, SOC, Op_RegI, 2, rdx->as_VMReg()->next()); | |
| 78 | |
| 79 reg_def RBX (SOC, SOE, Op_RegI, 3, rbx->as_VMReg()); | |
| 80 reg_def RBX_H(SOC, SOE, Op_RegI, 3, rbx->as_VMReg()->next()); | |
| 81 | |
| 82 reg_def RSP (NS, NS, Op_RegI, 4, rsp->as_VMReg()); | |
| 83 reg_def RSP_H(NS, NS, Op_RegI, 4, rsp->as_VMReg()->next()); | |
| 84 | |
| 85 // now that adapter frames are gone RBP is always saved and restored by the prolog/epilog code | |
| 86 reg_def RBP (NS, SOE, Op_RegI, 5, rbp->as_VMReg()); | |
| 87 reg_def RBP_H(NS, SOE, Op_RegI, 5, rbp->as_VMReg()->next()); | |
| 88 | |
| 89 #ifdef _WIN64 | |
| 90 | |
| 91 reg_def RSI (SOC, SOE, Op_RegI, 6, rsi->as_VMReg()); | |
| 92 reg_def RSI_H(SOC, SOE, Op_RegI, 6, rsi->as_VMReg()->next()); | |
| 93 | |
| 94 reg_def RDI (SOC, SOE, Op_RegI, 7, rdi->as_VMReg()); | |
| 95 reg_def RDI_H(SOC, SOE, Op_RegI, 7, rdi->as_VMReg()->next()); | |
| 96 | |
| 97 #else | |
| 98 | |
| 99 reg_def RSI (SOC, SOC, Op_RegI, 6, rsi->as_VMReg()); | |
| 100 reg_def RSI_H(SOC, SOC, Op_RegI, 6, rsi->as_VMReg()->next()); | |
| 101 | |
| 102 reg_def RDI (SOC, SOC, Op_RegI, 7, rdi->as_VMReg()); | |
| 103 reg_def RDI_H(SOC, SOC, Op_RegI, 7, rdi->as_VMReg()->next()); | |
| 104 | |
| 105 #endif | |
| 106 | |
| 107 reg_def R8 (SOC, SOC, Op_RegI, 8, r8->as_VMReg()); | |
| 108 reg_def R8_H (SOC, SOC, Op_RegI, 8, r8->as_VMReg()->next()); | |
| 109 | |
| 110 reg_def R9 (SOC, SOC, Op_RegI, 9, r9->as_VMReg()); | |
| 111 reg_def R9_H (SOC, SOC, Op_RegI, 9, r9->as_VMReg()->next()); | |
| 112 | |
| 113 reg_def R10 (SOC, SOC, Op_RegI, 10, r10->as_VMReg()); | |
| 114 reg_def R10_H(SOC, SOC, Op_RegI, 10, r10->as_VMReg()->next()); | |
| 115 | |
| 116 reg_def R11 (SOC, SOC, Op_RegI, 11, r11->as_VMReg()); | |
| 117 reg_def R11_H(SOC, SOC, Op_RegI, 11, r11->as_VMReg()->next()); | |
| 118 | |
| 119 reg_def R12 (SOC, SOE, Op_RegI, 12, r12->as_VMReg()); | |
| 120 reg_def R12_H(SOC, SOE, Op_RegI, 12, r12->as_VMReg()->next()); | |
| 121 | |
| 122 reg_def R13 (SOC, SOE, Op_RegI, 13, r13->as_VMReg()); | |
| 123 reg_def R13_H(SOC, SOE, Op_RegI, 13, r13->as_VMReg()->next()); | |
| 124 | |
| 125 reg_def R14 (SOC, SOE, Op_RegI, 14, r14->as_VMReg()); | |
| 126 reg_def R14_H(SOC, SOE, Op_RegI, 14, r14->as_VMReg()->next()); | |
| 127 | |
| 128 reg_def R15 (SOC, SOE, Op_RegI, 15, r15->as_VMReg()); | |
| 129 reg_def R15_H(SOC, SOE, Op_RegI, 15, r15->as_VMReg()->next()); | |
| 130 | |
| 131 | |
| 132 // Floating Point Registers | |
| 133 | |
| 134 // XMM registers. 128-bit registers or 4 words each, labeled (a)-d. | |
| 135 // Word a in each register holds a Float, words ab hold a Double. We | |
| 136 // currently do not use the SIMD capabilities, so registers cd are | |
| 137 // unused at the moment. | |
| 138 // XMM8-XMM15 must be encoded with REX. | |
| 139 // Linux ABI: No register preserved across function calls | |
| 140 // XMM0-XMM7 might hold parameters | |
| 141 // Windows ABI: XMM6-XMM15 preserved across function calls | |
| 142 // XMM0-XMM3 might hold parameters | |
| 143 | |
| 144 reg_def XMM0 (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()); | |
| 145 reg_def XMM0_H (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next()); | |
| 146 | |
| 147 reg_def XMM1 (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()); | |
| 148 reg_def XMM1_H (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next()); | |
| 149 | |
| 150 reg_def XMM2 (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()); | |
| 151 reg_def XMM2_H (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next()); | |
| 152 | |
| 153 reg_def XMM3 (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()); | |
| 154 reg_def XMM3_H (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next()); | |
| 155 | |
| 156 reg_def XMM4 (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()); | |
| 157 reg_def XMM4_H (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next()); | |
| 158 | |
| 159 reg_def XMM5 (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()); | |
| 160 reg_def XMM5_H (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next()); | |
| 161 | |
| 162 #ifdef _WIN64 | |
| 163 | |
| 164 reg_def XMM6 (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg()); | |
| 165 reg_def XMM6_H (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg()->next()); | |
| 166 | |
| 167 reg_def XMM7 (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg()); | |
| 168 reg_def XMM7_H (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg()->next()); | |
| 169 | |
| 170 reg_def XMM8 (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg()); | |
| 171 reg_def XMM8_H (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg()->next()); | |
| 172 | |
| 173 reg_def XMM9 (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg()); | |
| 174 reg_def XMM9_H (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg()->next()); | |
| 175 | |
| 176 reg_def XMM10 (SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()); | |
| 177 reg_def XMM10_H(SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()->next()); | |
| 178 | |
| 179 reg_def XMM11 (SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()); | |
| 180 reg_def XMM11_H(SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()->next()); | |
| 181 | |
| 182 reg_def XMM12 (SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()); | |
| 183 reg_def XMM12_H(SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()->next()); | |
| 184 | |
| 185 reg_def XMM13 (SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()); | |
| 186 reg_def XMM13_H(SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()->next()); | |
| 187 | |
| 188 reg_def XMM14 (SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()); | |
| 189 reg_def XMM14_H(SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()->next()); | |
| 190 | |
| 191 reg_def XMM15 (SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()); | |
| 192 reg_def XMM15_H(SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()->next()); | |
| 193 | |
| 194 #else | |
| 195 | |
| 196 reg_def XMM6 (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()); | |
| 197 reg_def XMM6_H (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next()); | |
| 198 | |
| 199 reg_def XMM7 (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()); | |
| 200 reg_def XMM7_H (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next()); | |
| 201 | |
| 202 reg_def XMM8 (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()); | |
| 203 reg_def XMM8_H (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next()); | |
| 204 | |
| 205 reg_def XMM9 (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()); | |
| 206 reg_def XMM9_H (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next()); | |
| 207 | |
| 208 reg_def XMM10 (SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()); | |
| 209 reg_def XMM10_H(SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next()); | |
| 210 | |
| 211 reg_def XMM11 (SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()); | |
| 212 reg_def XMM11_H(SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next()); | |
| 213 | |
| 214 reg_def XMM12 (SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()); | |
| 215 reg_def XMM12_H(SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next()); | |
| 216 | |
| 217 reg_def XMM13 (SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()); | |
| 218 reg_def XMM13_H(SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next()); | |
| 219 | |
| 220 reg_def XMM14 (SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()); | |
| 221 reg_def XMM14_H(SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next()); | |
| 222 | |
| 223 reg_def XMM15 (SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()); | |
| 224 reg_def XMM15_H(SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next()); | |
| 225 | |
| 226 #endif // _WIN64 | |
| 227 | |
| 228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad()); | |
| 229 | |
| 230 // Specify priority of register selection within phases of register | |
| 231 // allocation. Highest priority is first. A useful heuristic is to | |
| 232 // give registers a low priority when they are required by machine | |
| 233 // instructions, like EAX and EDX on I486, and choose no-save registers | |
| 234 // before save-on-call, & save-on-call before save-on-entry. Registers | |
| 235 // which participate in fixed calling sequences should come last. | |
| 236 // Registers which are used as pairs must fall on an even boundary. | |
| 237 | |
| 238 alloc_class chunk0(R10, R10_H, | |
| 239 R11, R11_H, | |
| 240 R8, R8_H, | |
| 241 R9, R9_H, | |
| 242 R12, R12_H, | |
| 243 RCX, RCX_H, | |
| 244 RBX, RBX_H, | |
| 245 RDI, RDI_H, | |
| 246 RDX, RDX_H, | |
| 247 RSI, RSI_H, | |
| 248 RAX, RAX_H, | |
| 249 RBP, RBP_H, | |
| 250 R13, R13_H, | |
| 251 R14, R14_H, | |
| 252 R15, R15_H, | |
| 253 RSP, RSP_H); | |
| 254 | |
| 255 // XXX probably use 8-15 first on Linux | |
| 256 alloc_class chunk1(XMM0, XMM0_H, | |
| 257 XMM1, XMM1_H, | |
| 258 XMM2, XMM2_H, | |
| 259 XMM3, XMM3_H, | |
| 260 XMM4, XMM4_H, | |
| 261 XMM5, XMM5_H, | |
| 262 XMM6, XMM6_H, | |
| 263 XMM7, XMM7_H, | |
| 264 XMM8, XMM8_H, | |
| 265 XMM9, XMM9_H, | |
| 266 XMM10, XMM10_H, | |
| 267 XMM11, XMM11_H, | |
| 268 XMM12, XMM12_H, | |
| 269 XMM13, XMM13_H, | |
| 270 XMM14, XMM14_H, | |
| 271 XMM15, XMM15_H); | |
| 272 | |
| 273 alloc_class chunk2(RFLAGS); | |
| 274 | |
| 275 | |
| 276 //----------Architecture Description Register Classes-------------------------- | |
| 277 // Several register classes are automatically defined based upon information in | |
| 278 // this architecture description. | |
| 279 // 1) reg_class inline_cache_reg ( /* as def'd in frame section */ ) | |
| 280 // 2) reg_class compiler_method_oop_reg ( /* as def'd in frame section */ ) | |
| 281 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ ) | |
| 282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ ) | |
| 283 // | |
| 284 | |
| 285 // Class for all pointer registers (including RSP) | |
| 286 reg_class any_reg(RAX, RAX_H, | |
| 287 RDX, RDX_H, | |
| 288 RBP, RBP_H, | |
| 289 RDI, RDI_H, | |
| 290 RSI, RSI_H, | |
| 291 RCX, RCX_H, | |
| 292 RBX, RBX_H, | |
| 293 RSP, RSP_H, | |
| 294 R8, R8_H, | |
| 295 R9, R9_H, | |
| 296 R10, R10_H, | |
| 297 R11, R11_H, | |
| 298 R12, R12_H, | |
| 299 R13, R13_H, | |
| 300 R14, R14_H, | |
| 301 R15, R15_H); | |
| 302 | |
| 303 // Class for all pointer registers except RSP | |
| 304 reg_class ptr_reg(RAX, RAX_H, | |
| 305 RDX, RDX_H, | |
| 306 RBP, RBP_H, | |
| 307 RDI, RDI_H, | |
| 308 RSI, RSI_H, | |
| 309 RCX, RCX_H, | |
| 310 RBX, RBX_H, | |
| 311 R8, R8_H, | |
| 312 R9, R9_H, | |
| 313 R10, R10_H, | |
| 314 R11, R11_H, | |
| 315 R13, R13_H, | |
| 316 R14, R14_H); | |
| 317 | |
| 318 // Class for all pointer registers except RAX and RSP | |
| 319 reg_class ptr_no_rax_reg(RDX, RDX_H, | |
| 320 RBP, RBP_H, | |
| 321 RDI, RDI_H, | |
| 322 RSI, RSI_H, | |
| 323 RCX, RCX_H, | |
| 324 RBX, RBX_H, | |
| 325 R8, R8_H, | |
| 326 R9, R9_H, | |
| 327 R10, R10_H, | |
| 328 R11, R11_H, | |
| 329 R13, R13_H, | |
| 330 R14, R14_H); | |
| 331 | |
| 332 reg_class ptr_no_rbp_reg(RDX, RDX_H, | |
| 333 RAX, RAX_H, | |
| 334 RDI, RDI_H, | |
| 335 RSI, RSI_H, | |
| 336 RCX, RCX_H, | |
| 337 RBX, RBX_H, | |
| 338 R8, R8_H, | |
| 339 R9, R9_H, | |
| 340 R10, R10_H, | |
| 341 R11, R11_H, | |
| 342 R13, R13_H, | |
| 343 R14, R14_H); | |
| 344 | |
| 345 // Class for all pointer registers except RAX, RBX and RSP | |
| 346 reg_class ptr_no_rax_rbx_reg(RDX, RDX_H, | |
| 347 RBP, RBP_H, | |
| 348 RDI, RDI_H, | |
| 349 RSI, RSI_H, | |
| 350 RCX, RCX_H, | |
| 351 R8, R8_H, | |
| 352 R9, R9_H, | |
| 353 R10, R10_H, | |
| 354 R11, R11_H, | |
| 355 R13, R13_H, | |
| 356 R14, R14_H); | |
| 357 | |
| 358 // Singleton class for RAX pointer register | |
| 359 reg_class ptr_rax_reg(RAX, RAX_H); | |
| 360 | |
| 361 // Singleton class for RBX pointer register | |
| 362 reg_class ptr_rbx_reg(RBX, RBX_H); | |
| 363 | |
| 364 // Singleton class for RSI pointer register | |
| 365 reg_class ptr_rsi_reg(RSI, RSI_H); | |
| 366 | |
| 367 // Singleton class for RDI pointer register | |
| 368 reg_class ptr_rdi_reg(RDI, RDI_H); | |
| 369 | |
| 370 // Singleton class for RBP pointer register | |
| 371 reg_class ptr_rbp_reg(RBP, RBP_H); | |
| 372 | |
| 373 // Singleton class for stack pointer | |
| 374 reg_class ptr_rsp_reg(RSP, RSP_H); | |
| 375 | |
| 376 // Singleton class for TLS pointer | |
| 377 reg_class ptr_r15_reg(R15, R15_H); | |
| 378 | |
| 379 // Class for all long registers (except RSP) | |
| 380 reg_class long_reg(RAX, RAX_H, | |
| 381 RDX, RDX_H, | |
| 382 RBP, RBP_H, | |
| 383 RDI, RDI_H, | |
| 384 RSI, RSI_H, | |
| 385 RCX, RCX_H, | |
| 386 RBX, RBX_H, | |
| 387 R8, R8_H, | |
| 388 R9, R9_H, | |
| 389 R10, R10_H, | |
| 390 R11, R11_H, | |
| 391 R13, R13_H, | |
| 392 R14, R14_H); | |
| 393 | |
| 394 // Class for all long registers except RAX, RDX (and RSP) | |
| 395 reg_class long_no_rax_rdx_reg(RBP, RBP_H, | |
| 396 RDI, RDI_H, | |
| 397 RSI, RSI_H, | |
| 398 RCX, RCX_H, | |
| 399 RBX, RBX_H, | |
| 400 R8, R8_H, | |
| 401 R9, R9_H, | |
| 402 R10, R10_H, | |
| 403 R11, R11_H, | |
| 404 R13, R13_H, | |
| 405 R14, R14_H); | |
| 406 | |
| 407 // Class for all long registers except RCX (and RSP) | |
| 408 reg_class long_no_rcx_reg(RBP, RBP_H, | |
| 409 RDI, RDI_H, | |
| 410 RSI, RSI_H, | |
| 411 RAX, RAX_H, | |
| 412 RDX, RDX_H, | |
| 413 RBX, RBX_H, | |
| 414 R8, R8_H, | |
| 415 R9, R9_H, | |
| 416 R10, R10_H, | |
| 417 R11, R11_H, | |
| 418 R13, R13_H, | |
| 419 R14, R14_H); | |
| 420 | |
| 421 // Class for all long registers except RAX (and RSP) | |
| 422 reg_class long_no_rax_reg(RBP, RBP_H, | |
| 423 RDX, RDX_H, | |
| 424 RDI, RDI_H, | |
| 425 RSI, RSI_H, | |
| 426 RCX, RCX_H, | |
| 427 RBX, RBX_H, | |
| 428 R8, R8_H, | |
| 429 R9, R9_H, | |
| 430 R10, R10_H, | |
| 431 R11, R11_H, | |
| 432 R13, R13_H, | |
| 433 R14, R14_H); | |
| 434 | |
| 435 // Singleton class for RAX long register | |
| 436 reg_class long_rax_reg(RAX, RAX_H); | |
| 437 | |
| 438 // Singleton class for RCX long register | |
| 439 reg_class long_rcx_reg(RCX, RCX_H); | |
| 440 | |
| 441 // Singleton class for RDX long register | |
| 442 reg_class long_rdx_reg(RDX, RDX_H); | |
| 443 | |
| 444 // Class for all int registers (except RSP) | |
| 445 reg_class int_reg(RAX, | |
| 446 RDX, | |
| 447 RBP, | |
| 448 RDI, | |
| 449 RSI, | |
| 450 RCX, | |
| 451 RBX, | |
| 452 R8, | |
| 453 R9, | |
| 454 R10, | |
| 455 R11, | |
| 456 R13, | |
| 457 R14); | |
| 458 | |
| 459 // Class for all int registers except RCX (and RSP) | |
| 460 reg_class int_no_rcx_reg(RAX, | |
| 461 RDX, | |
| 462 RBP, | |
| 463 RDI, | |
| 464 RSI, | |
| 465 RBX, | |
| 466 R8, | |
| 467 R9, | |
| 468 R10, | |
| 469 R11, | |
| 470 R13, | |
| 471 R14); | |
| 472 | |
| 473 // Class for all int registers except RAX, RDX (and RSP) | |
| 474 reg_class int_no_rax_rdx_reg(RBP, | |
| 304 | 475 RDI, |
| 0 | 476 RSI, |
| 477 RCX, | |
| 478 RBX, | |
| 479 R8, | |
| 480 R9, | |
| 481 R10, | |
| 482 R11, | |
| 483 R13, | |
| 484 R14); | |
| 485 | |
| 486 // Singleton class for RAX int register | |
| 487 reg_class int_rax_reg(RAX); | |
| 488 | |
| 489 // Singleton class for RBX int register | |
| 490 reg_class int_rbx_reg(RBX); | |
| 491 | |
| 492 // Singleton class for RCX int register | |
| 493 reg_class int_rcx_reg(RCX); | |
| 494 | |
| 495 // Singleton class for RCX int register | |
| 496 reg_class int_rdx_reg(RDX); | |
| 497 | |
| 498 // Singleton class for RCX int register | |
| 499 reg_class int_rdi_reg(RDI); | |
| 500 | |
| 501 // Singleton class for instruction pointer | |
| 502 // reg_class ip_reg(RIP); | |
| 503 | |
| 504 // Singleton class for condition codes | |
| 505 reg_class int_flags(RFLAGS); | |
| 506 | |
| 507 // Class for all float registers | |
| 508 reg_class float_reg(XMM0, | |
| 509 XMM1, | |
| 510 XMM2, | |
| 511 XMM3, | |
| 512 XMM4, | |
| 513 XMM5, | |
| 514 XMM6, | |
| 515 XMM7, | |
| 516 XMM8, | |
| 517 XMM9, | |
| 518 XMM10, | |
| 519 XMM11, | |
| 520 XMM12, | |
| 521 XMM13, | |
| 522 XMM14, | |
| 523 XMM15); | |
| 524 | |
| 525 // Class for all double registers | |
| 526 reg_class double_reg(XMM0, XMM0_H, | |
| 527 XMM1, XMM1_H, | |
| 528 XMM2, XMM2_H, | |
| 529 XMM3, XMM3_H, | |
| 530 XMM4, XMM4_H, | |
| 531 XMM5, XMM5_H, | |
| 532 XMM6, XMM6_H, | |
| 533 XMM7, XMM7_H, | |
| 534 XMM8, XMM8_H, | |
| 535 XMM9, XMM9_H, | |
| 536 XMM10, XMM10_H, | |
| 537 XMM11, XMM11_H, | |
| 538 XMM12, XMM12_H, | |
| 539 XMM13, XMM13_H, | |
| 540 XMM14, XMM14_H, | |
| 541 XMM15, XMM15_H); | |
| 542 %} | |
| 543 | |
| 544 | |
| 545 //----------SOURCE BLOCK------------------------------------------------------- | |
| 546 // This is a block of C++ code which provides values, functions, and | |
| 547 // definitions necessary in the rest of the architecture description | |
| 548 source %{ | |
| 304 | 549 #define RELOC_IMM64 Assembler::imm_operand |
| 0 | 550 #define RELOC_DISP32 Assembler::disp32_operand |
| 551 | |
| 552 #define __ _masm. | |
| 553 | |
| 554 // !!!!! Special hack to get all types of calls to specify the byte offset | |
| 555 // from the start of the call to the point where the return address | |
| 556 // will point. | |
| 557 int MachCallStaticJavaNode::ret_addr_offset() | |
| 558 { | |
| 559 return 5; // 5 bytes from start of call to where return address points | |
| 560 } | |
| 561 | |
| 562 int MachCallDynamicJavaNode::ret_addr_offset() | |
| 563 { | |
| 564 return 15; // 15 bytes from start of call to where return address points | |
| 565 } | |
| 566 | |
| 567 // In os_cpu .ad file | |
| 568 // int MachCallRuntimeNode::ret_addr_offset() | |
| 569 | |
| 570 // Indicate if the safepoint node needs the polling page as an input. | |
| 571 // Since amd64 does not have absolute addressing but RIP-relative | |
| 572 // addressing and the polling page is within 2G, it doesn't. | |
| 573 bool SafePointNode::needs_polling_address_input() | |
| 574 { | |
| 575 return false; | |
| 576 } | |
| 577 | |
| 578 // | |
| 579 // Compute padding required for nodes which need alignment | |
| 580 // | |
| 581 | |
| 582 // The address of the call instruction needs to be 4-byte aligned to | |
| 583 // ensure that it does not span a cache line so that it can be patched. | |
| 584 int CallStaticJavaDirectNode::compute_padding(int current_offset) const | |
| 585 { | |
| 586 current_offset += 1; // skip call opcode byte | |
| 587 return round_to(current_offset, alignment_required()) - current_offset; | |
| 588 } | |
| 589 | |
| 590 // The address of the call instruction needs to be 4-byte aligned to | |
| 591 // ensure that it does not span a cache line so that it can be patched. | |
| 592 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const | |
| 593 { | |
| 594 current_offset += 11; // skip movq instruction + call opcode byte | |
| 595 return round_to(current_offset, alignment_required()) - current_offset; | |
| 596 } | |
| 597 | |
| 598 #ifndef PRODUCT | |
| 599 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const | |
| 600 { | |
| 601 st->print("INT3"); | |
| 602 } | |
| 603 #endif | |
| 604 | |
| 605 // EMIT_RM() | |
| 606 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) | |
| 607 { | |
| 608 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3); | |
| 609 *(cbuf.code_end()) = c; | |
| 610 cbuf.set_code_end(cbuf.code_end() + 1); | |
| 611 } | |
| 612 | |
| 613 // EMIT_CC() | |
| 614 void emit_cc(CodeBuffer &cbuf, int f1, int f2) | |
| 615 { | |
| 616 unsigned char c = (unsigned char) (f1 | f2); | |
| 617 *(cbuf.code_end()) = c; | |
| 618 cbuf.set_code_end(cbuf.code_end() + 1); | |
| 619 } | |
| 620 | |
| 621 // EMIT_OPCODE() | |
| 622 void emit_opcode(CodeBuffer &cbuf, int code) | |
| 623 { | |
| 624 *(cbuf.code_end()) = (unsigned char) code; | |
| 625 cbuf.set_code_end(cbuf.code_end() + 1); | |
| 626 } | |
| 627 | |
| 628 // EMIT_OPCODE() w/ relocation information | |
| 629 void emit_opcode(CodeBuffer &cbuf, | |
| 630 int code, relocInfo::relocType reloc, int offset, int format) | |
| 631 { | |
| 632 cbuf.relocate(cbuf.inst_mark() + offset, reloc, format); | |
| 633 emit_opcode(cbuf, code); | |
| 634 } | |
| 635 | |
| 636 // EMIT_D8() | |
| 637 void emit_d8(CodeBuffer &cbuf, int d8) | |
| 638 { | |
| 639 *(cbuf.code_end()) = (unsigned char) d8; | |
| 640 cbuf.set_code_end(cbuf.code_end() + 1); | |
| 641 } | |
| 642 | |
| 643 // EMIT_D16() | |
| 644 void emit_d16(CodeBuffer &cbuf, int d16) | |
| 645 { | |
| 646 *((short *)(cbuf.code_end())) = d16; | |
| 647 cbuf.set_code_end(cbuf.code_end() + 2); | |
| 648 } | |
| 649 | |
| 650 // EMIT_D32() | |
| 651 void emit_d32(CodeBuffer &cbuf, int d32) | |
| 652 { | |
| 653 *((int *)(cbuf.code_end())) = d32; | |
| 654 cbuf.set_code_end(cbuf.code_end() + 4); | |
| 655 } | |
| 656 | |
| 657 // EMIT_D64() | |
| 658 void emit_d64(CodeBuffer &cbuf, int64_t d64) | |
| 659 { | |
| 660 *((int64_t*) (cbuf.code_end())) = d64; | |
| 661 cbuf.set_code_end(cbuf.code_end() + 8); | |
| 662 } | |
| 663 | |
| 664 // emit 32 bit value and construct relocation entry from relocInfo::relocType | |
| 665 void emit_d32_reloc(CodeBuffer& cbuf, | |
| 666 int d32, | |
| 667 relocInfo::relocType reloc, | |
| 668 int format) | |
| 669 { | |
| 670 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc"); | |
| 671 cbuf.relocate(cbuf.inst_mark(), reloc, format); | |
| 672 | |
| 673 *((int*) (cbuf.code_end())) = d32; | |
| 674 cbuf.set_code_end(cbuf.code_end() + 4); | |
| 675 } | |
| 676 | |
| 677 // emit 32 bit value and construct relocation entry from RelocationHolder | |
| 678 void emit_d32_reloc(CodeBuffer& cbuf, | |
| 679 int d32, | |
| 680 RelocationHolder const& rspec, | |
| 681 int format) | |
| 682 { | |
| 683 #ifdef ASSERT | |
| 684 if (rspec.reloc()->type() == relocInfo::oop_type && | |
| 685 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) { | |
| 686 assert(oop((intptr_t)d32)->is_oop() && oop((intptr_t)d32)->is_perm(), "cannot embed non-perm oops in code"); | |
| 687 } | |
| 688 #endif | |
| 689 cbuf.relocate(cbuf.inst_mark(), rspec, format); | |
| 690 | |
| 691 *((int* )(cbuf.code_end())) = d32; | |
| 692 cbuf.set_code_end(cbuf.code_end() + 4); | |
| 693 } | |
| 694 | |
| 695 void emit_d32_reloc(CodeBuffer& cbuf, address addr) { | |
| 696 address next_ip = cbuf.code_end() + 4; | |
| 697 emit_d32_reloc(cbuf, (int) (addr - next_ip), | |
| 698 external_word_Relocation::spec(addr), | |
| 699 RELOC_DISP32); | |
| 700 } | |
| 701 | |
| 702 | |
| 703 // emit 64 bit value and construct relocation entry from relocInfo::relocType | |
| 704 void emit_d64_reloc(CodeBuffer& cbuf, | |
| 705 int64_t d64, | |
| 706 relocInfo::relocType reloc, | |
| 707 int format) | |
| 708 { | |
| 709 cbuf.relocate(cbuf.inst_mark(), reloc, format); | |
| 710 | |
| 711 *((int64_t*) (cbuf.code_end())) = d64; | |
| 712 cbuf.set_code_end(cbuf.code_end() + 8); | |
| 713 } | |
| 714 | |
| 715 // emit 64 bit value and construct relocation entry from RelocationHolder | |
| 716 void emit_d64_reloc(CodeBuffer& cbuf, | |
| 717 int64_t d64, | |
| 718 RelocationHolder const& rspec, | |
| 719 int format) | |
| 720 { | |
| 721 #ifdef ASSERT | |
| 722 if (rspec.reloc()->type() == relocInfo::oop_type && | |
| 723 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) { | |
| 724 assert(oop(d64)->is_oop() && oop(d64)->is_perm(), | |
| 725 "cannot embed non-perm oops in code"); | |
| 726 } | |
| 727 #endif | |
| 728 cbuf.relocate(cbuf.inst_mark(), rspec, format); | |
| 729 | |
| 730 *((int64_t*) (cbuf.code_end())) = d64; | |
| 731 cbuf.set_code_end(cbuf.code_end() + 8); | |
| 732 } | |
| 733 | |
| 734 // Access stack slot for load or store | |
| 735 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp) | |
| 736 { | |
| 737 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src]) | |
| 738 if (-0x80 <= disp && disp < 0x80) { | |
| 739 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte | |
| 740 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte | |
| 741 emit_d8(cbuf, disp); // Displacement // R/M byte | |
| 742 } else { | |
| 743 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte | |
| 744 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte | |
| 745 emit_d32(cbuf, disp); // Displacement // R/M byte | |
| 746 } | |
| 747 } | |
| 748 | |
| 749 // rRegI ereg, memory mem) %{ // emit_reg_mem | |
| 750 void encode_RegMem(CodeBuffer &cbuf, | |
| 751 int reg, | |
| 752 int base, int index, int scale, int disp, bool disp_is_oop) | |
| 753 { | |
| 754 assert(!disp_is_oop, "cannot have disp"); | |
| 755 int regenc = reg & 7; | |
| 756 int baseenc = base & 7; | |
| 757 int indexenc = index & 7; | |
| 758 | |
| 759 // There is no index & no scale, use form without SIB byte | |
| 760 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) { | |
| 761 // If no displacement, mode is 0x0; unless base is [RBP] or [R13] | |
| 762 if (disp == 0 && base != RBP_enc && base != R13_enc) { | |
| 763 emit_rm(cbuf, 0x0, regenc, baseenc); // * | |
| 764 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) { | |
| 765 // If 8-bit displacement, mode 0x1 | |
| 766 emit_rm(cbuf, 0x1, regenc, baseenc); // * | |
| 767 emit_d8(cbuf, disp); | |
| 768 } else { | |
| 769 // If 32-bit displacement | |
| 770 if (base == -1) { // Special flag for absolute address | |
| 771 emit_rm(cbuf, 0x0, regenc, 0x5); // * | |
| 772 if (disp_is_oop) { | |
| 773 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32); | |
| 774 } else { | |
| 775 emit_d32(cbuf, disp); | |
| 776 } | |
| 777 } else { | |
| 778 // Normal base + offset | |
| 779 emit_rm(cbuf, 0x2, regenc, baseenc); // * | |
| 780 if (disp_is_oop) { | |
| 781 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32); | |
| 782 } else { | |
| 783 emit_d32(cbuf, disp); | |
| 784 } | |
| 785 } | |
| 786 } | |
| 787 } else { | |
| 788 // Else, encode with the SIB byte | |
| 789 // If no displacement, mode is 0x0; unless base is [RBP] or [R13] | |
| 790 if (disp == 0 && base != RBP_enc && base != R13_enc) { | |
| 791 // If no displacement | |
| 792 emit_rm(cbuf, 0x0, regenc, 0x4); // * | |
| 793 emit_rm(cbuf, scale, indexenc, baseenc); | |
| 794 } else { | |
| 795 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) { | |
| 796 // If 8-bit displacement, mode 0x1 | |
| 797 emit_rm(cbuf, 0x1, regenc, 0x4); // * | |
| 798 emit_rm(cbuf, scale, indexenc, baseenc); | |
| 799 emit_d8(cbuf, disp); | |
| 800 } else { | |
| 801 // If 32-bit displacement | |
| 802 if (base == 0x04 ) { | |
| 803 emit_rm(cbuf, 0x2, regenc, 0x4); | |
| 804 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid??? | |
| 805 } else { | |
| 806 emit_rm(cbuf, 0x2, regenc, 0x4); | |
| 807 emit_rm(cbuf, scale, indexenc, baseenc); // * | |
| 808 } | |
| 809 if (disp_is_oop) { | |
| 810 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32); | |
| 811 } else { | |
| 812 emit_d32(cbuf, disp); | |
| 813 } | |
| 814 } | |
| 815 } | |
| 816 } | |
| 817 } | |
| 818 | |
| 819 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc) | |
| 820 { | |
| 821 if (dstenc != srcenc) { | |
| 822 if (dstenc < 8) { | |
| 823 if (srcenc >= 8) { | |
| 824 emit_opcode(cbuf, Assembler::REX_B); | |
| 825 srcenc -= 8; | |
| 826 } | |
| 827 } else { | |
| 828 if (srcenc < 8) { | |
| 829 emit_opcode(cbuf, Assembler::REX_R); | |
| 830 } else { | |
| 831 emit_opcode(cbuf, Assembler::REX_RB); | |
| 832 srcenc -= 8; | |
| 833 } | |
| 834 dstenc -= 8; | |
| 835 } | |
| 836 | |
| 837 emit_opcode(cbuf, 0x8B); | |
| 838 emit_rm(cbuf, 0x3, dstenc, srcenc); | |
| 839 } | |
| 840 } | |
| 841 | |
| 842 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) { | |
| 843 if( dst_encoding == src_encoding ) { | |
| 844 // reg-reg copy, use an empty encoding | |
| 845 } else { | |
| 846 MacroAssembler _masm(&cbuf); | |
| 847 | |
| 848 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding)); | |
| 849 } | |
| 850 } | |
| 851 | |
| 852 | |
| 853 //============================================================================= | |
| 854 #ifndef PRODUCT | |
| 855 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const | |
| 856 { | |
| 857 Compile* C = ra_->C; | |
| 858 | |
| 859 int framesize = C->frame_slots() << LogBytesPerInt; | |
| 860 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); | |
| 861 // Remove wordSize for return adr already pushed | |
| 862 // and another for the RBP we are going to save | |
| 863 framesize -= 2*wordSize; | |
| 864 bool need_nop = true; | |
| 865 | |
| 866 // Calls to C2R adapters often do not accept exceptional returns. | |
| 867 // We require that their callers must bang for them. But be | |
| 868 // careful, because some VM calls (such as call site linkage) can | |
| 869 // use several kilobytes of stack. But the stack safety zone should | |
| 870 // account for that. See bugs 4446381, 4468289, 4497237. | |
| 871 if (C->need_stack_bang(framesize)) { | |
| 872 st->print_cr("# stack bang"); st->print("\t"); | |
| 873 need_nop = false; | |
| 874 } | |
| 875 st->print_cr("pushq rbp"); st->print("\t"); | |
| 876 | |
| 877 if (VerifyStackAtCalls) { | |
| 878 // Majik cookie to verify stack depth | |
| 879 st->print_cr("pushq 0xffffffffbadb100d" | |
| 880 "\t# Majik cookie for stack depth check"); | |
| 881 st->print("\t"); | |
| 882 framesize -= wordSize; // Remove 2 for cookie | |
| 883 need_nop = false; | |
| 884 } | |
| 885 | |
| 886 if (framesize) { | |
| 887 st->print("subq rsp, #%d\t# Create frame", framesize); | |
| 888 if (framesize < 0x80 && need_nop) { | |
| 889 st->print("\n\tnop\t# nop for patch_verified_entry"); | |
| 890 } | |
| 891 } | |
| 892 } | |
| 893 #endif | |
| 894 | |
| 895 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const | |
| 896 { | |
| 897 Compile* C = ra_->C; | |
| 898 | |
| 899 // WARNING: Initial instruction MUST be 5 bytes or longer so that | |
| 900 // NativeJump::patch_verified_entry will be able to patch out the entry | |
| 901 // code safely. The fldcw is ok at 6 bytes, the push to verify stack | |
| 902 // depth is ok at 5 bytes, the frame allocation can be either 3 or | |
| 903 // 6 bytes. So if we don't do the fldcw or the push then we must | |
| 904 // use the 6 byte frame allocation even if we have no frame. :-( | |
| 905 // If method sets FPU control word do it now | |
| 906 | |
| 907 int framesize = C->frame_slots() << LogBytesPerInt; | |
| 908 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); | |
| 909 // Remove wordSize for return adr already pushed | |
| 910 // and another for the RBP we are going to save | |
| 911 framesize -= 2*wordSize; | |
| 912 bool need_nop = true; | |
| 913 | |
| 914 // Calls to C2R adapters often do not accept exceptional returns. | |
| 915 // We require that their callers must bang for them. But be | |
| 916 // careful, because some VM calls (such as call site linkage) can | |
| 917 // use several kilobytes of stack. But the stack safety zone should | |
| 918 // account for that. See bugs 4446381, 4468289, 4497237. | |
| 919 if (C->need_stack_bang(framesize)) { | |
| 920 MacroAssembler masm(&cbuf); | |
| 921 masm.generate_stack_overflow_check(framesize); | |
| 922 need_nop = false; | |
| 923 } | |
| 924 | |
| 925 // We always push rbp so that on return to interpreter rbp will be | |
| 926 // restored correctly and we can correct the stack. | |
| 927 emit_opcode(cbuf, 0x50 | RBP_enc); | |
| 928 | |
| 929 if (VerifyStackAtCalls) { | |
| 930 // Majik cookie to verify stack depth | |
| 931 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d | |
| 932 emit_d32(cbuf, 0xbadb100d); | |
| 933 framesize -= wordSize; // Remove 2 for cookie | |
| 934 need_nop = false; | |
| 935 } | |
| 936 | |
| 937 if (framesize) { | |
| 938 emit_opcode(cbuf, Assembler::REX_W); | |
| 939 if (framesize < 0x80) { | |
| 940 emit_opcode(cbuf, 0x83); // sub SP,#framesize | |
| 941 emit_rm(cbuf, 0x3, 0x05, RSP_enc); | |
| 942 emit_d8(cbuf, framesize); | |
| 943 if (need_nop) { | |
| 944 emit_opcode(cbuf, 0x90); // nop | |
| 945 } | |
| 946 } else { | |
| 947 emit_opcode(cbuf, 0x81); // sub SP,#framesize | |
| 948 emit_rm(cbuf, 0x3, 0x05, RSP_enc); | |
| 949 emit_d32(cbuf, framesize); | |
| 950 } | |
| 951 } | |
| 952 | |
| 953 C->set_frame_complete(cbuf.code_end() - cbuf.code_begin()); | |
| 954 | |
| 955 #ifdef ASSERT | |
| 956 if (VerifyStackAtCalls) { | |
| 957 Label L; | |
| 958 MacroAssembler masm(&cbuf); | |
| 304 | 959 masm.push(rax); |
| 960 masm.mov(rax, rsp); | |
| 961 masm.andptr(rax, StackAlignmentInBytes-1); | |
| 962 masm.cmpptr(rax, StackAlignmentInBytes-wordSize); | |
| 963 masm.pop(rax); | |
| 0 | 964 masm.jcc(Assembler::equal, L); |
| 965 masm.stop("Stack is not properly aligned!"); | |
| 966 masm.bind(L); | |
| 967 } | |
| 968 #endif | |
| 969 } | |
| 970 | |
| 971 uint MachPrologNode::size(PhaseRegAlloc* ra_) const | |
| 972 { | |
| 973 return MachNode::size(ra_); // too many variables; just compute it | |
| 974 // the hard way | |
| 975 } | |
| 976 | |
| 977 int MachPrologNode::reloc() const | |
| 978 { | |
| 979 return 0; // a large enough number | |
| 980 } | |
| 981 | |
| 982 //============================================================================= | |
| 983 #ifndef PRODUCT | |
| 984 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const | |
| 985 { | |
| 986 Compile* C = ra_->C; | |
| 987 int framesize = C->frame_slots() << LogBytesPerInt; | |
| 988 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); | |
| 989 // Remove word for return adr already pushed | |
| 990 // and RBP | |
| 991 framesize -= 2*wordSize; | |
| 992 | |
| 993 if (framesize) { | |
| 994 st->print_cr("addq\trsp, %d\t# Destroy frame", framesize); | |
| 995 st->print("\t"); | |
| 996 } | |
| 997 | |
| 998 st->print_cr("popq\trbp"); | |
| 999 if (do_polling() && C->is_method_compilation()) { | |
| 1000 st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t" | |
| 1001 "# Safepoint: poll for GC"); | |
| 1002 st->print("\t"); | |
| 1003 } | |
| 1004 } | |
| 1005 #endif | |
| 1006 | |
| 1007 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const | |
| 1008 { | |
| 1009 Compile* C = ra_->C; | |
| 1010 int framesize = C->frame_slots() << LogBytesPerInt; | |
| 1011 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); | |
| 1012 // Remove word for return adr already pushed | |
| 1013 // and RBP | |
| 1014 framesize -= 2*wordSize; | |
| 1015 | |
| 1016 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here | |
| 1017 | |
| 1018 if (framesize) { | |
| 1019 emit_opcode(cbuf, Assembler::REX_W); | |
| 1020 if (framesize < 0x80) { | |
| 1021 emit_opcode(cbuf, 0x83); // addq rsp, #framesize | |
| 1022 emit_rm(cbuf, 0x3, 0x00, RSP_enc); | |
| 1023 emit_d8(cbuf, framesize); | |
| 1024 } else { | |
| 1025 emit_opcode(cbuf, 0x81); // addq rsp, #framesize | |
| 1026 emit_rm(cbuf, 0x3, 0x00, RSP_enc); | |
| 1027 emit_d32(cbuf, framesize); | |
| 1028 } | |
| 1029 } | |
| 1030 | |
| 1031 // popq rbp | |
| 1032 emit_opcode(cbuf, 0x58 | RBP_enc); | |
| 1033 | |
| 1034 if (do_polling() && C->is_method_compilation()) { | |
| 1035 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes | |
| 1036 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 1037 cbuf.set_inst_mark(); | |
| 1038 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX | |
| 1039 emit_opcode(cbuf, 0x85); // testl | |
| 1040 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5 | |
| 1041 // cbuf.inst_mark() is beginning of instruction | |
| 1042 emit_d32_reloc(cbuf, os::get_polling_page()); | |
| 1043 // relocInfo::poll_return_type, | |
| 1044 } | |
| 1045 } | |
| 1046 | |
| 1047 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const | |
| 1048 { | |
| 1049 Compile* C = ra_->C; | |
| 1050 int framesize = C->frame_slots() << LogBytesPerInt; | |
| 1051 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); | |
| 1052 // Remove word for return adr already pushed | |
| 1053 // and RBP | |
| 1054 framesize -= 2*wordSize; | |
| 1055 | |
| 1056 uint size = 0; | |
| 1057 | |
| 1058 if (do_polling() && C->is_method_compilation()) { | |
| 1059 size += 6; | |
| 1060 } | |
| 1061 | |
| 1062 // count popq rbp | |
| 1063 size++; | |
| 1064 | |
| 1065 if (framesize) { | |
| 1066 if (framesize < 0x80) { | |
| 1067 size += 4; | |
| 1068 } else if (framesize) { | |
| 1069 size += 7; | |
| 1070 } | |
| 1071 } | |
| 1072 | |
| 1073 return size; | |
| 1074 } | |
| 1075 | |
| 1076 int MachEpilogNode::reloc() const | |
| 1077 { | |
| 1078 return 2; // a large enough number | |
| 1079 } | |
| 1080 | |
| 1081 const Pipeline* MachEpilogNode::pipeline() const | |
| 1082 { | |
| 1083 return MachNode::pipeline_class(); | |
| 1084 } | |
| 1085 | |
| 1086 int MachEpilogNode::safepoint_offset() const | |
| 1087 { | |
| 1088 return 0; | |
| 1089 } | |
| 1090 | |
| 1091 //============================================================================= | |
| 1092 | |
| 1093 enum RC { | |
| 1094 rc_bad, | |
| 1095 rc_int, | |
| 1096 rc_float, | |
| 1097 rc_stack | |
| 1098 }; | |
| 1099 | |
| 1100 static enum RC rc_class(OptoReg::Name reg) | |
| 1101 { | |
| 1102 if( !OptoReg::is_valid(reg) ) return rc_bad; | |
| 1103 | |
| 1104 if (OptoReg::is_stack(reg)) return rc_stack; | |
| 1105 | |
| 1106 VMReg r = OptoReg::as_VMReg(reg); | |
| 1107 | |
| 1108 if (r->is_Register()) return rc_int; | |
| 1109 | |
| 1110 assert(r->is_XMMRegister(), "must be"); | |
| 1111 return rc_float; | |
| 1112 } | |
| 1113 | |
| 1114 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf, | |
| 1115 PhaseRegAlloc* ra_, | |
| 1116 bool do_size, | |
| 1117 outputStream* st) const | |
| 1118 { | |
| 1119 | |
| 1120 // Get registers to move | |
| 1121 OptoReg::Name src_second = ra_->get_reg_second(in(1)); | |
| 1122 OptoReg::Name src_first = ra_->get_reg_first(in(1)); | |
| 1123 OptoReg::Name dst_second = ra_->get_reg_second(this); | |
| 1124 OptoReg::Name dst_first = ra_->get_reg_first(this); | |
| 1125 | |
| 1126 enum RC src_second_rc = rc_class(src_second); | |
| 1127 enum RC src_first_rc = rc_class(src_first); | |
| 1128 enum RC dst_second_rc = rc_class(dst_second); | |
| 1129 enum RC dst_first_rc = rc_class(dst_first); | |
| 1130 | |
| 1131 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), | |
| 1132 "must move at least 1 register" ); | |
| 1133 | |
| 1134 if (src_first == dst_first && src_second == dst_second) { | |
| 1135 // Self copy, no move | |
| 1136 return 0; | |
| 1137 } else if (src_first_rc == rc_stack) { | |
| 1138 // mem -> | |
| 1139 if (dst_first_rc == rc_stack) { | |
| 1140 // mem -> mem | |
| 1141 assert(src_second != dst_first, "overlap"); | |
| 1142 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1143 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1144 // 64-bit | |
| 1145 int src_offset = ra_->reg2offset(src_first); | |
| 1146 int dst_offset = ra_->reg2offset(dst_first); | |
| 1147 if (cbuf) { | |
| 1148 emit_opcode(*cbuf, 0xFF); | |
| 1149 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false); | |
| 1150 | |
| 1151 emit_opcode(*cbuf, 0x8F); | |
| 1152 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false); | |
| 1153 | |
| 1154 #ifndef PRODUCT | |
| 1155 } else if (!do_size) { | |
| 1156 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t" | |
| 1157 "popq [rsp + #%d]", | |
| 1158 src_offset, | |
| 1159 dst_offset); | |
| 1160 #endif | |
| 1161 } | |
| 1162 return | |
| 1163 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + | |
| 1164 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)); | |
| 1165 } else { | |
| 1166 // 32-bit | |
| 1167 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1168 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1169 // No pushl/popl, so: | |
| 1170 int src_offset = ra_->reg2offset(src_first); | |
| 1171 int dst_offset = ra_->reg2offset(dst_first); | |
| 1172 if (cbuf) { | |
| 1173 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1174 emit_opcode(*cbuf, 0x89); | |
| 1175 emit_opcode(*cbuf, 0x44); | |
| 1176 emit_opcode(*cbuf, 0x24); | |
| 1177 emit_opcode(*cbuf, 0xF8); | |
| 1178 | |
| 1179 emit_opcode(*cbuf, 0x8B); | |
| 1180 encode_RegMem(*cbuf, | |
| 1181 RAX_enc, | |
| 1182 RSP_enc, 0x4, 0, src_offset, | |
| 1183 false); | |
| 1184 | |
| 1185 emit_opcode(*cbuf, 0x89); | |
| 1186 encode_RegMem(*cbuf, | |
| 1187 RAX_enc, | |
| 1188 RSP_enc, 0x4, 0, dst_offset, | |
| 1189 false); | |
| 1190 | |
| 1191 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1192 emit_opcode(*cbuf, 0x8B); | |
| 1193 emit_opcode(*cbuf, 0x44); | |
| 1194 emit_opcode(*cbuf, 0x24); | |
| 1195 emit_opcode(*cbuf, 0xF8); | |
| 1196 | |
| 1197 #ifndef PRODUCT | |
| 1198 } else if (!do_size) { | |
| 1199 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t" | |
| 1200 "movl rax, [rsp + #%d]\n\t" | |
| 1201 "movl [rsp + #%d], rax\n\t" | |
| 1202 "movq rax, [rsp - #8]", | |
| 1203 src_offset, | |
| 1204 dst_offset); | |
| 1205 #endif | |
| 1206 } | |
| 1207 return | |
| 1208 5 + // movq | |
| 1209 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl | |
| 1210 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl | |
| 1211 5; // movq | |
| 1212 } | |
| 1213 } else if (dst_first_rc == rc_int) { | |
| 1214 // mem -> gpr | |
| 1215 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1216 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1217 // 64-bit | |
| 1218 int offset = ra_->reg2offset(src_first); | |
| 1219 if (cbuf) { | |
| 1220 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1221 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1222 } else { | |
| 1223 emit_opcode(*cbuf, Assembler::REX_WR); | |
| 1224 } | |
| 1225 emit_opcode(*cbuf, 0x8B); | |
| 1226 encode_RegMem(*cbuf, | |
| 1227 Matcher::_regEncode[dst_first], | |
| 1228 RSP_enc, 0x4, 0, offset, | |
| 1229 false); | |
| 1230 #ifndef PRODUCT | |
| 1231 } else if (!do_size) { | |
| 1232 st->print("movq %s, [rsp + #%d]\t# spill", | |
| 1233 Matcher::regName[dst_first], | |
| 1234 offset); | |
| 1235 #endif | |
| 1236 } | |
| 1237 return | |
| 1238 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX | |
| 1239 } else { | |
| 1240 // 32-bit | |
| 1241 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1242 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1243 int offset = ra_->reg2offset(src_first); | |
| 1244 if (cbuf) { | |
| 1245 if (Matcher::_regEncode[dst_first] >= 8) { | |
| 1246 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1247 } | |
| 1248 emit_opcode(*cbuf, 0x8B); | |
| 1249 encode_RegMem(*cbuf, | |
| 1250 Matcher::_regEncode[dst_first], | |
| 1251 RSP_enc, 0x4, 0, offset, | |
| 1252 false); | |
| 1253 #ifndef PRODUCT | |
| 1254 } else if (!do_size) { | |
| 1255 st->print("movl %s, [rsp + #%d]\t# spill", | |
| 1256 Matcher::regName[dst_first], | |
| 1257 offset); | |
| 1258 #endif | |
| 1259 } | |
| 1260 return | |
| 1261 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1262 ((Matcher::_regEncode[dst_first] < 8) | |
| 1263 ? 3 | |
| 1264 : 4); // REX | |
| 1265 } | |
| 1266 } else if (dst_first_rc == rc_float) { | |
| 1267 // mem-> xmm | |
| 1268 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1269 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1270 // 64-bit | |
| 1271 int offset = ra_->reg2offset(src_first); | |
| 1272 if (cbuf) { | |
| 1273 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66); | |
| 1274 if (Matcher::_regEncode[dst_first] >= 8) { | |
| 1275 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1276 } | |
| 1277 emit_opcode(*cbuf, 0x0F); | |
| 1278 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12); | |
| 1279 encode_RegMem(*cbuf, | |
| 1280 Matcher::_regEncode[dst_first], | |
| 1281 RSP_enc, 0x4, 0, offset, | |
| 1282 false); | |
| 1283 #ifndef PRODUCT | |
| 1284 } else if (!do_size) { | |
| 1285 st->print("%s %s, [rsp + #%d]\t# spill", | |
| 1286 UseXmmLoadAndClearUpper ? "movsd " : "movlpd", | |
| 1287 Matcher::regName[dst_first], | |
| 1288 offset); | |
| 1289 #endif | |
| 1290 } | |
| 1291 return | |
| 1292 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1293 ((Matcher::_regEncode[dst_first] < 8) | |
| 1294 ? 5 | |
| 1295 : 6); // REX | |
| 1296 } else { | |
| 1297 // 32-bit | |
| 1298 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1299 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1300 int offset = ra_->reg2offset(src_first); | |
| 1301 if (cbuf) { | |
| 1302 emit_opcode(*cbuf, 0xF3); | |
| 1303 if (Matcher::_regEncode[dst_first] >= 8) { | |
| 1304 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1305 } | |
| 1306 emit_opcode(*cbuf, 0x0F); | |
| 1307 emit_opcode(*cbuf, 0x10); | |
| 1308 encode_RegMem(*cbuf, | |
| 1309 Matcher::_regEncode[dst_first], | |
| 1310 RSP_enc, 0x4, 0, offset, | |
| 1311 false); | |
| 1312 #ifndef PRODUCT | |
| 1313 } else if (!do_size) { | |
| 1314 st->print("movss %s, [rsp + #%d]\t# spill", | |
| 1315 Matcher::regName[dst_first], | |
| 1316 offset); | |
| 1317 #endif | |
| 1318 } | |
| 1319 return | |
| 1320 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1321 ((Matcher::_regEncode[dst_first] < 8) | |
| 1322 ? 5 | |
| 1323 : 6); // REX | |
| 1324 } | |
| 1325 } | |
| 1326 } else if (src_first_rc == rc_int) { | |
| 1327 // gpr -> | |
| 1328 if (dst_first_rc == rc_stack) { | |
| 1329 // gpr -> mem | |
| 1330 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1331 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1332 // 64-bit | |
| 1333 int offset = ra_->reg2offset(dst_first); | |
| 1334 if (cbuf) { | |
| 1335 if (Matcher::_regEncode[src_first] < 8) { | |
| 1336 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1337 } else { | |
| 1338 emit_opcode(*cbuf, Assembler::REX_WR); | |
| 1339 } | |
| 1340 emit_opcode(*cbuf, 0x89); | |
| 1341 encode_RegMem(*cbuf, | |
| 1342 Matcher::_regEncode[src_first], | |
| 1343 RSP_enc, 0x4, 0, offset, | |
| 1344 false); | |
| 1345 #ifndef PRODUCT | |
| 1346 } else if (!do_size) { | |
| 1347 st->print("movq [rsp + #%d], %s\t# spill", | |
| 1348 offset, | |
| 1349 Matcher::regName[src_first]); | |
| 1350 #endif | |
| 1351 } | |
| 1352 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX | |
| 1353 } else { | |
| 1354 // 32-bit | |
| 1355 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1356 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1357 int offset = ra_->reg2offset(dst_first); | |
| 1358 if (cbuf) { | |
| 1359 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1360 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1361 } | |
| 1362 emit_opcode(*cbuf, 0x89); | |
| 1363 encode_RegMem(*cbuf, | |
| 1364 Matcher::_regEncode[src_first], | |
| 1365 RSP_enc, 0x4, 0, offset, | |
| 1366 false); | |
| 1367 #ifndef PRODUCT | |
| 1368 } else if (!do_size) { | |
| 1369 st->print("movl [rsp + #%d], %s\t# spill", | |
| 1370 offset, | |
| 1371 Matcher::regName[src_first]); | |
| 1372 #endif | |
| 1373 } | |
| 1374 return | |
| 1375 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1376 ((Matcher::_regEncode[src_first] < 8) | |
| 1377 ? 3 | |
| 1378 : 4); // REX | |
| 1379 } | |
| 1380 } else if (dst_first_rc == rc_int) { | |
| 1381 // gpr -> gpr | |
| 1382 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1383 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1384 // 64-bit | |
| 1385 if (cbuf) { | |
| 1386 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1387 if (Matcher::_regEncode[src_first] < 8) { | |
| 1388 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1389 } else { | |
| 1390 emit_opcode(*cbuf, Assembler::REX_WB); | |
| 1391 } | |
| 1392 } else { | |
| 1393 if (Matcher::_regEncode[src_first] < 8) { | |
| 1394 emit_opcode(*cbuf, Assembler::REX_WR); | |
| 1395 } else { | |
| 1396 emit_opcode(*cbuf, Assembler::REX_WRB); | |
| 1397 } | |
| 1398 } | |
| 1399 emit_opcode(*cbuf, 0x8B); | |
| 1400 emit_rm(*cbuf, 0x3, | |
| 1401 Matcher::_regEncode[dst_first] & 7, | |
| 1402 Matcher::_regEncode[src_first] & 7); | |
| 1403 #ifndef PRODUCT | |
| 1404 } else if (!do_size) { | |
| 1405 st->print("movq %s, %s\t# spill", | |
| 1406 Matcher::regName[dst_first], | |
| 1407 Matcher::regName[src_first]); | |
| 1408 #endif | |
| 1409 } | |
| 1410 return 3; // REX | |
| 1411 } else { | |
| 1412 // 32-bit | |
| 1413 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1414 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1415 if (cbuf) { | |
| 1416 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1417 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1418 emit_opcode(*cbuf, Assembler::REX_B); | |
| 1419 } | |
| 1420 } else { | |
| 1421 if (Matcher::_regEncode[src_first] < 8) { | |
| 1422 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1423 } else { | |
| 1424 emit_opcode(*cbuf, Assembler::REX_RB); | |
| 1425 } | |
| 1426 } | |
| 1427 emit_opcode(*cbuf, 0x8B); | |
| 1428 emit_rm(*cbuf, 0x3, | |
| 1429 Matcher::_regEncode[dst_first] & 7, | |
| 1430 Matcher::_regEncode[src_first] & 7); | |
| 1431 #ifndef PRODUCT | |
| 1432 } else if (!do_size) { | |
| 1433 st->print("movl %s, %s\t# spill", | |
| 1434 Matcher::regName[dst_first], | |
| 1435 Matcher::regName[src_first]); | |
| 1436 #endif | |
| 1437 } | |
| 1438 return | |
| 1439 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8) | |
| 1440 ? 2 | |
| 1441 : 3; // REX | |
| 1442 } | |
| 1443 } else if (dst_first_rc == rc_float) { | |
| 1444 // gpr -> xmm | |
| 1445 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1446 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1447 // 64-bit | |
| 1448 if (cbuf) { | |
| 1449 emit_opcode(*cbuf, 0x66); | |
| 1450 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1451 if (Matcher::_regEncode[src_first] < 8) { | |
| 1452 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1453 } else { | |
| 1454 emit_opcode(*cbuf, Assembler::REX_WB); | |
| 1455 } | |
| 1456 } else { | |
| 1457 if (Matcher::_regEncode[src_first] < 8) { | |
| 1458 emit_opcode(*cbuf, Assembler::REX_WR); | |
| 1459 } else { | |
| 1460 emit_opcode(*cbuf, Assembler::REX_WRB); | |
| 1461 } | |
| 1462 } | |
| 1463 emit_opcode(*cbuf, 0x0F); | |
| 1464 emit_opcode(*cbuf, 0x6E); | |
| 1465 emit_rm(*cbuf, 0x3, | |
| 1466 Matcher::_regEncode[dst_first] & 7, | |
| 1467 Matcher::_regEncode[src_first] & 7); | |
| 1468 #ifndef PRODUCT | |
| 1469 } else if (!do_size) { | |
| 1470 st->print("movdq %s, %s\t# spill", | |
| 1471 Matcher::regName[dst_first], | |
| 1472 Matcher::regName[src_first]); | |
| 1473 #endif | |
| 1474 } | |
| 1475 return 5; // REX | |
| 1476 } else { | |
| 1477 // 32-bit | |
| 1478 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1479 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1480 if (cbuf) { | |
| 1481 emit_opcode(*cbuf, 0x66); | |
| 1482 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1483 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1484 emit_opcode(*cbuf, Assembler::REX_B); | |
| 1485 } | |
| 1486 } else { | |
| 1487 if (Matcher::_regEncode[src_first] < 8) { | |
| 1488 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1489 } else { | |
| 1490 emit_opcode(*cbuf, Assembler::REX_RB); | |
| 1491 } | |
| 1492 } | |
| 1493 emit_opcode(*cbuf, 0x0F); | |
| 1494 emit_opcode(*cbuf, 0x6E); | |
| 1495 emit_rm(*cbuf, 0x3, | |
| 1496 Matcher::_regEncode[dst_first] & 7, | |
| 1497 Matcher::_regEncode[src_first] & 7); | |
| 1498 #ifndef PRODUCT | |
| 1499 } else if (!do_size) { | |
| 1500 st->print("movdl %s, %s\t# spill", | |
| 1501 Matcher::regName[dst_first], | |
| 1502 Matcher::regName[src_first]); | |
| 1503 #endif | |
| 1504 } | |
| 1505 return | |
| 1506 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8) | |
| 1507 ? 4 | |
| 1508 : 5; // REX | |
| 1509 } | |
| 1510 } | |
| 1511 } else if (src_first_rc == rc_float) { | |
| 1512 // xmm -> | |
| 1513 if (dst_first_rc == rc_stack) { | |
| 1514 // xmm -> mem | |
| 1515 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1516 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1517 // 64-bit | |
| 1518 int offset = ra_->reg2offset(dst_first); | |
| 1519 if (cbuf) { | |
| 1520 emit_opcode(*cbuf, 0xF2); | |
| 1521 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1522 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1523 } | |
| 1524 emit_opcode(*cbuf, 0x0F); | |
| 1525 emit_opcode(*cbuf, 0x11); | |
| 1526 encode_RegMem(*cbuf, | |
| 1527 Matcher::_regEncode[src_first], | |
| 1528 RSP_enc, 0x4, 0, offset, | |
| 1529 false); | |
| 1530 #ifndef PRODUCT | |
| 1531 } else if (!do_size) { | |
| 1532 st->print("movsd [rsp + #%d], %s\t# spill", | |
| 1533 offset, | |
| 1534 Matcher::regName[src_first]); | |
| 1535 #endif | |
| 1536 } | |
| 1537 return | |
| 1538 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1539 ((Matcher::_regEncode[src_first] < 8) | |
| 1540 ? 5 | |
| 1541 : 6); // REX | |
| 1542 } else { | |
| 1543 // 32-bit | |
| 1544 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1545 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1546 int offset = ra_->reg2offset(dst_first); | |
| 1547 if (cbuf) { | |
| 1548 emit_opcode(*cbuf, 0xF3); | |
| 1549 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1550 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1551 } | |
| 1552 emit_opcode(*cbuf, 0x0F); | |
| 1553 emit_opcode(*cbuf, 0x11); | |
| 1554 encode_RegMem(*cbuf, | |
| 1555 Matcher::_regEncode[src_first], | |
| 1556 RSP_enc, 0x4, 0, offset, | |
| 1557 false); | |
| 1558 #ifndef PRODUCT | |
| 1559 } else if (!do_size) { | |
| 1560 st->print("movss [rsp + #%d], %s\t# spill", | |
| 1561 offset, | |
| 1562 Matcher::regName[src_first]); | |
| 1563 #endif | |
| 1564 } | |
| 1565 return | |
| 1566 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + | |
| 1567 ((Matcher::_regEncode[src_first] < 8) | |
| 1568 ? 5 | |
| 1569 : 6); // REX | |
| 1570 } | |
| 1571 } else if (dst_first_rc == rc_int) { | |
| 1572 // xmm -> gpr | |
| 1573 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1574 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1575 // 64-bit | |
| 1576 if (cbuf) { | |
| 1577 emit_opcode(*cbuf, 0x66); | |
| 1578 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1579 if (Matcher::_regEncode[src_first] < 8) { | |
| 1580 emit_opcode(*cbuf, Assembler::REX_W); | |
| 1581 } else { | |
| 1582 emit_opcode(*cbuf, Assembler::REX_WR); // attention! | |
| 1583 } | |
| 1584 } else { | |
| 1585 if (Matcher::_regEncode[src_first] < 8) { | |
| 1586 emit_opcode(*cbuf, Assembler::REX_WB); // attention! | |
| 1587 } else { | |
| 1588 emit_opcode(*cbuf, Assembler::REX_WRB); | |
| 1589 } | |
| 1590 } | |
| 1591 emit_opcode(*cbuf, 0x0F); | |
| 1592 emit_opcode(*cbuf, 0x7E); | |
| 1593 emit_rm(*cbuf, 0x3, | |
| 1594 Matcher::_regEncode[dst_first] & 7, | |
| 1595 Matcher::_regEncode[src_first] & 7); | |
| 1596 #ifndef PRODUCT | |
| 1597 } else if (!do_size) { | |
| 1598 st->print("movdq %s, %s\t# spill", | |
| 1599 Matcher::regName[dst_first], | |
| 1600 Matcher::regName[src_first]); | |
| 1601 #endif | |
| 1602 } | |
| 1603 return 5; // REX | |
| 1604 } else { | |
| 1605 // 32-bit | |
| 1606 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1607 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1608 if (cbuf) { | |
| 1609 emit_opcode(*cbuf, 0x66); | |
| 1610 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1611 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1612 emit_opcode(*cbuf, Assembler::REX_R); // attention! | |
| 1613 } | |
| 1614 } else { | |
| 1615 if (Matcher::_regEncode[src_first] < 8) { | |
| 1616 emit_opcode(*cbuf, Assembler::REX_B); // attention! | |
| 1617 } else { | |
| 1618 emit_opcode(*cbuf, Assembler::REX_RB); | |
| 1619 } | |
| 1620 } | |
| 1621 emit_opcode(*cbuf, 0x0F); | |
| 1622 emit_opcode(*cbuf, 0x7E); | |
| 1623 emit_rm(*cbuf, 0x3, | |
| 1624 Matcher::_regEncode[dst_first] & 7, | |
| 1625 Matcher::_regEncode[src_first] & 7); | |
| 1626 #ifndef PRODUCT | |
| 1627 } else if (!do_size) { | |
| 1628 st->print("movdl %s, %s\t# spill", | |
| 1629 Matcher::regName[dst_first], | |
| 1630 Matcher::regName[src_first]); | |
| 1631 #endif | |
| 1632 } | |
| 1633 return | |
| 1634 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8) | |
| 1635 ? 4 | |
| 1636 : 5; // REX | |
| 1637 } | |
| 1638 } else if (dst_first_rc == rc_float) { | |
| 1639 // xmm -> xmm | |
| 1640 if ((src_first & 1) == 0 && src_first + 1 == src_second && | |
| 1641 (dst_first & 1) == 0 && dst_first + 1 == dst_second) { | |
| 1642 // 64-bit | |
| 1643 if (cbuf) { | |
| 1644 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2); | |
| 1645 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1646 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1647 emit_opcode(*cbuf, Assembler::REX_B); | |
| 1648 } | |
| 1649 } else { | |
| 1650 if (Matcher::_regEncode[src_first] < 8) { | |
| 1651 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1652 } else { | |
| 1653 emit_opcode(*cbuf, Assembler::REX_RB); | |
| 1654 } | |
| 1655 } | |
| 1656 emit_opcode(*cbuf, 0x0F); | |
| 1657 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10); | |
| 1658 emit_rm(*cbuf, 0x3, | |
| 1659 Matcher::_regEncode[dst_first] & 7, | |
| 1660 Matcher::_regEncode[src_first] & 7); | |
| 1661 #ifndef PRODUCT | |
| 1662 } else if (!do_size) { | |
| 1663 st->print("%s %s, %s\t# spill", | |
| 1664 UseXmmRegToRegMoveAll ? "movapd" : "movsd ", | |
| 1665 Matcher::regName[dst_first], | |
| 1666 Matcher::regName[src_first]); | |
| 1667 #endif | |
| 1668 } | |
| 1669 return | |
| 1670 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8) | |
| 1671 ? 4 | |
| 1672 : 5; // REX | |
| 1673 } else { | |
| 1674 // 32-bit | |
| 1675 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform"); | |
| 1676 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform"); | |
| 1677 if (cbuf) { | |
| 1678 if (!UseXmmRegToRegMoveAll) | |
| 1679 emit_opcode(*cbuf, 0xF3); | |
| 1680 if (Matcher::_regEncode[dst_first] < 8) { | |
| 1681 if (Matcher::_regEncode[src_first] >= 8) { | |
| 1682 emit_opcode(*cbuf, Assembler::REX_B); | |
| 1683 } | |
| 1684 } else { | |
| 1685 if (Matcher::_regEncode[src_first] < 8) { | |
| 1686 emit_opcode(*cbuf, Assembler::REX_R); | |
| 1687 } else { | |
| 1688 emit_opcode(*cbuf, Assembler::REX_RB); | |
| 1689 } | |
| 1690 } | |
| 1691 emit_opcode(*cbuf, 0x0F); | |
| 1692 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10); | |
| 1693 emit_rm(*cbuf, 0x3, | |
| 1694 Matcher::_regEncode[dst_first] & 7, | |
| 1695 Matcher::_regEncode[src_first] & 7); | |
| 1696 #ifndef PRODUCT | |
| 1697 } else if (!do_size) { | |
| 1698 st->print("%s %s, %s\t# spill", | |
| 1699 UseXmmRegToRegMoveAll ? "movaps" : "movss ", | |
| 1700 Matcher::regName[dst_first], | |
| 1701 Matcher::regName[src_first]); | |
| 1702 #endif | |
| 1703 } | |
| 1704 return | |
| 1705 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8) | |
| 1706 ? (UseXmmRegToRegMoveAll ? 3 : 4) | |
| 1707 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX | |
| 1708 } | |
| 1709 } | |
| 1710 } | |
| 1711 | |
| 1712 assert(0," foo "); | |
| 1713 Unimplemented(); | |
| 1714 | |
| 1715 return 0; | |
| 1716 } | |
| 1717 | |
| 1718 #ifndef PRODUCT | |
| 1719 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const | |
| 1720 { | |
| 1721 implementation(NULL, ra_, false, st); | |
| 1722 } | |
| 1723 #endif | |
| 1724 | |
| 1725 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const | |
| 1726 { | |
| 1727 implementation(&cbuf, ra_, false, NULL); | |
| 1728 } | |
| 1729 | |
| 1730 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const | |
| 1731 { | |
| 1732 return implementation(NULL, ra_, true, NULL); | |
| 1733 } | |
| 1734 | |
| 1735 //============================================================================= | |
| 1736 #ifndef PRODUCT | |
| 1737 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const | |
| 1738 { | |
| 1739 st->print("nop \t# %d bytes pad for loops and calls", _count); | |
| 1740 } | |
| 1741 #endif | |
| 1742 | |
| 1743 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const | |
| 1744 { | |
| 1745 MacroAssembler _masm(&cbuf); | |
| 1746 __ nop(_count); | |
| 1747 } | |
| 1748 | |
| 1749 uint MachNopNode::size(PhaseRegAlloc*) const | |
| 1750 { | |
| 1751 return _count; | |
| 1752 } | |
| 1753 | |
| 1754 | |
| 1755 //============================================================================= | |
| 1756 #ifndef PRODUCT | |
| 1757 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const | |
| 1758 { | |
| 1759 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()); | |
| 1760 int reg = ra_->get_reg_first(this); | |
| 1761 st->print("leaq %s, [rsp + #%d]\t# box lock", | |
| 1762 Matcher::regName[reg], offset); | |
| 1763 } | |
| 1764 #endif | |
| 1765 | |
| 1766 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const | |
| 1767 { | |
| 1768 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()); | |
| 1769 int reg = ra_->get_encode(this); | |
| 1770 if (offset >= 0x80) { | |
| 1771 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR); | |
| 1772 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset] | |
| 1773 emit_rm(cbuf, 0x2, reg & 7, 0x04); | |
| 1774 emit_rm(cbuf, 0x0, 0x04, RSP_enc); | |
| 1775 emit_d32(cbuf, offset); | |
| 1776 } else { | |
| 1777 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR); | |
| 1778 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset] | |
| 1779 emit_rm(cbuf, 0x1, reg & 7, 0x04); | |
| 1780 emit_rm(cbuf, 0x0, 0x04, RSP_enc); | |
| 1781 emit_d8(cbuf, offset); | |
| 1782 } | |
| 1783 } | |
| 1784 | |
| 1785 uint BoxLockNode::size(PhaseRegAlloc *ra_) const | |
| 1786 { | |
| 1787 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()); | |
| 1788 return (offset < 0x80) ? 5 : 8; // REX | |
| 1789 } | |
| 1790 | |
| 1791 //============================================================================= | |
| 1792 | |
| 1793 // emit call stub, compiled java to interpreter | |
| 1794 void emit_java_to_interp(CodeBuffer& cbuf) | |
| 1795 { | |
| 1796 // Stub is fixed up when the corresponding call is converted from | |
| 1797 // calling compiled code to calling interpreted code. | |
| 1798 // movq rbx, 0 | |
| 1799 // jmp -5 # to self | |
| 1800 | |
| 1801 address mark = cbuf.inst_mark(); // get mark within main instrs section | |
| 1802 | |
| 1803 // Note that the code buffer's inst_mark is always relative to insts. | |
| 1804 // That's why we must use the macroassembler to generate a stub. | |
| 1805 MacroAssembler _masm(&cbuf); | |
| 1806 | |
| 1807 address base = | |
| 1808 __ start_a_stub(Compile::MAX_stubs_size); | |
| 1809 if (base == NULL) return; // CodeBuffer::expand failed | |
| 1810 // static stub relocation stores the instruction address of the call | |
| 1811 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64); | |
| 1812 // static stub relocation also tags the methodOop in the code-stream. | |
| 1813 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time | |
| 304 | 1814 // This is recognized as unresolved by relocs/nativeinst/ic code |
| 0 | 1815 __ jump(RuntimeAddress(__ pc())); |
| 1816 | |
| 1817 // Update current stubs pointer and restore code_end. | |
| 1818 __ end_a_stub(); | |
| 1819 } | |
| 1820 | |
| 1821 // size of call stub, compiled java to interpretor | |
| 1822 uint size_java_to_interp() | |
| 1823 { | |
| 1824 return 15; // movq (1+1+8); jmp (1+4) | |
| 1825 } | |
| 1826 | |
| 1827 // relocation entries for call stub, compiled java to interpretor | |
| 1828 uint reloc_java_to_interp() | |
| 1829 { | |
| 1830 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call | |
| 1831 } | |
| 1832 | |
| 1833 //============================================================================= | |
| 1834 #ifndef PRODUCT | |
| 1835 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const | |
| 1836 { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1837 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1838 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes()); |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1839 if (Universe::narrow_oop_shift() != 0) { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1840 st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]"); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1841 } |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1842 st->print_cr("cmpq rax, rscratch1\t # Inline cache check"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1843 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1844 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t" |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1845 "# Inline cache check", oopDesc::klass_offset_in_bytes()); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1846 } |
| 0 | 1847 st->print_cr("\tjne SharedRuntime::_ic_miss_stub"); |
| 1848 st->print_cr("\tnop"); | |
| 1849 if (!OptoBreakpoint) { | |
| 1850 st->print_cr("\tnop"); | |
| 1851 } | |
| 1852 } | |
| 1853 #endif | |
| 1854 | |
| 1855 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const | |
| 1856 { | |
| 1857 MacroAssembler masm(&cbuf); | |
| 1858 #ifdef ASSERT | |
| 1859 uint code_size = cbuf.code_size(); | |
| 1860 #endif | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1861 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1862 masm.load_klass(rscratch1, j_rarg0); |
| 304 | 1863 masm.cmpptr(rax, rscratch1); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1864 } else { |
| 304 | 1865 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes())); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1866 } |
| 0 | 1867 |
| 1868 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub())); | |
| 1869 | |
| 1870 /* WARNING these NOPs are critical so that verified entry point is properly | |
| 1871 aligned for patching by NativeJump::patch_verified_entry() */ | |
| 1872 int nops_cnt = 1; | |
| 1873 if (!OptoBreakpoint) { | |
| 1874 // Leave space for int3 | |
| 1875 nops_cnt += 1; | |
| 1876 } | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1877 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1878 // ??? divisible by 4 is aligned? |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1879 nops_cnt += 1; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1880 } |
| 0 | 1881 masm.nop(nops_cnt); |
| 1882 | |
| 1883 assert(cbuf.code_size() - code_size == size(ra_), | |
| 1884 "checking code size of inline cache node"); | |
| 1885 } | |
| 1886 | |
| 1887 uint MachUEPNode::size(PhaseRegAlloc* ra_) const | |
| 1888 { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1889 if (UseCompressedOops) { |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1890 if (Universe::narrow_oop_shift() == 0) { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1891 return OptoBreakpoint ? 15 : 16; |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1892 } else { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1893 return OptoBreakpoint ? 19 : 20; |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1894 } |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1895 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1896 return OptoBreakpoint ? 11 : 12; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1897 } |
| 0 | 1898 } |
| 1899 | |
| 1900 | |
| 1901 //============================================================================= | |
| 1902 uint size_exception_handler() | |
| 1903 { | |
| 1904 // NativeCall instruction size is the same as NativeJump. | |
| 1905 // Note that this value is also credited (in output.cpp) to | |
| 1906 // the size of the code section. | |
| 1907 return NativeJump::instruction_size; | |
| 1908 } | |
| 1909 | |
| 1910 // Emit exception handler code. | |
| 1911 int emit_exception_handler(CodeBuffer& cbuf) | |
| 1912 { | |
| 1913 | |
| 1914 // Note that the code buffer's inst_mark is always relative to insts. | |
| 1915 // That's why we must use the macroassembler to generate a handler. | |
| 1916 MacroAssembler _masm(&cbuf); | |
| 1917 address base = | |
| 1918 __ start_a_stub(size_exception_handler()); | |
| 1919 if (base == NULL) return 0; // CodeBuffer::expand failed | |
| 1920 int offset = __ offset(); | |
| 1921 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin())); | |
| 1922 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow"); | |
| 1923 __ end_a_stub(); | |
| 1924 return offset; | |
| 1925 } | |
| 1926 | |
| 1927 uint size_deopt_handler() | |
| 1928 { | |
| 1929 // three 5 byte instructions | |
| 1930 return 15; | |
| 1931 } | |
| 1932 | |
| 1933 // Emit deopt handler code. | |
| 1934 int emit_deopt_handler(CodeBuffer& cbuf) | |
| 1935 { | |
| 1936 | |
| 1937 // Note that the code buffer's inst_mark is always relative to insts. | |
| 1938 // That's why we must use the macroassembler to generate a handler. | |
| 1939 MacroAssembler _masm(&cbuf); | |
| 1940 address base = | |
| 1941 __ start_a_stub(size_deopt_handler()); | |
| 1942 if (base == NULL) return 0; // CodeBuffer::expand failed | |
| 1943 int offset = __ offset(); | |
| 1944 address the_pc = (address) __ pc(); | |
| 1945 Label next; | |
| 1946 // push a "the_pc" on the stack without destroying any registers | |
| 1947 // as they all may be live. | |
| 1948 | |
| 1949 // push address of "next" | |
| 1950 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32 | |
| 1951 __ bind(next); | |
| 1952 // adjust it so it matches "the_pc" | |
| 304 | 1953 __ subptr(Address(rsp, 0), __ offset() - offset); |
| 0 | 1954 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack())); |
| 1955 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow"); | |
| 1956 __ end_a_stub(); | |
| 1957 return offset; | |
| 1958 } | |
| 1959 | |
| 1960 static void emit_double_constant(CodeBuffer& cbuf, double x) { | |
| 1961 int mark = cbuf.insts()->mark_off(); | |
| 1962 MacroAssembler _masm(&cbuf); | |
| 1963 address double_address = __ double_constant(x); | |
| 1964 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift | |
| 1965 emit_d32_reloc(cbuf, | |
| 1966 (int) (double_address - cbuf.code_end() - 4), | |
| 1967 internal_word_Relocation::spec(double_address), | |
| 1968 RELOC_DISP32); | |
| 1969 } | |
| 1970 | |
| 1971 static void emit_float_constant(CodeBuffer& cbuf, float x) { | |
| 1972 int mark = cbuf.insts()->mark_off(); | |
| 1973 MacroAssembler _masm(&cbuf); | |
| 1974 address float_address = __ float_constant(x); | |
| 1975 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift | |
| 1976 emit_d32_reloc(cbuf, | |
| 1977 (int) (float_address - cbuf.code_end() - 4), | |
| 1978 internal_word_Relocation::spec(float_address), | |
| 1979 RELOC_DISP32); | |
| 1980 } | |
| 1981 | |
| 1982 | |
| 1983 int Matcher::regnum_to_fpu_offset(int regnum) | |
| 1984 { | |
| 1985 return regnum - 32; // The FP registers are in the second chunk | |
| 1986 } | |
| 1987 | |
| 1988 // This is UltraSparc specific, true just means we have fast l2f conversion | |
| 1989 const bool Matcher::convL2FSupported(void) { | |
| 1990 return true; | |
| 1991 } | |
| 1992 | |
| 1993 // Vector width in bytes | |
| 1994 const uint Matcher::vector_width_in_bytes(void) { | |
| 1995 return 8; | |
| 1996 } | |
| 1997 | |
| 1998 // Vector ideal reg | |
| 1999 const uint Matcher::vector_ideal_reg(void) { | |
| 2000 return Op_RegD; | |
| 2001 } | |
| 2002 | |
| 2003 // Is this branch offset short enough that a short branch can be used? | |
| 2004 // | |
| 2005 // NOTE: If the platform does not provide any short branch variants, then | |
| 2006 // this method should return false for offset 0. | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2007 bool Matcher::is_short_branch_offset(int rule, int offset) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2008 // the short version of jmpConUCF2 contains multiple branches, |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2009 // making the reach slightly less |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2010 if (rule == jmpConUCF2_rule) |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2011 return (-126 <= offset && offset <= 125); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
2012 return (-128 <= offset && offset <= 127); |
| 0 | 2013 } |
| 2014 | |
| 2015 const bool Matcher::isSimpleConstant64(jlong value) { | |
| 2016 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?. | |
| 2017 //return value == (int) value; // Cf. storeImmL and immL32. | |
| 2018 | |
| 2019 // Probably always true, even if a temp register is required. | |
| 2020 return true; | |
| 2021 } | |
| 2022 | |
| 2023 // The ecx parameter to rep stosq for the ClearArray node is in words. | |
| 2024 const bool Matcher::init_array_count_is_in_bytes = false; | |
| 2025 | |
| 2026 // Threshold size for cleararray. | |
| 2027 const int Matcher::init_array_short_size = 8 * BytesPerLong; | |
| 2028 | |
| 2029 // Should the Matcher clone shifts on addressing modes, expecting them | |
| 2030 // to be subsumed into complex addressing expressions or compute them | |
| 2031 // into registers? True for Intel but false for most RISCs | |
| 2032 const bool Matcher::clone_shift_expressions = true; | |
| 2033 | |
| 2034 // Is it better to copy float constants, or load them directly from | |
| 2035 // memory? Intel can load a float constant from a direct address, | |
| 2036 // requiring no extra registers. Most RISCs will have to materialize | |
| 2037 // an address into a register first, so they would do better to copy | |
| 2038 // the constant from stack. | |
| 2039 const bool Matcher::rematerialize_float_constants = true; // XXX | |
| 2040 | |
| 2041 // If CPU can load and store mis-aligned doubles directly then no | |
| 2042 // fixup is needed. Else we split the double into 2 integer pieces | |
| 2043 // and move it piece-by-piece. Only happens when passing doubles into | |
| 2044 // C code as the Java calling convention forces doubles to be aligned. | |
| 2045 const bool Matcher::misaligned_doubles_ok = true; | |
| 2046 | |
| 2047 // No-op on amd64 | |
| 2048 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {} | |
| 2049 | |
| 2050 // Advertise here if the CPU requires explicit rounding operations to | |
| 2051 // implement the UseStrictFP mode. | |
| 2052 const bool Matcher::strict_fp_requires_explicit_rounding = true; | |
| 2053 | |
| 2054 // Do floats take an entire double register or just half? | |
| 2055 const bool Matcher::float_in_double = true; | |
| 2056 // Do ints take an entire long register or just half? | |
| 2057 const bool Matcher::int_in_long = true; | |
| 2058 | |
| 2059 // Return whether or not this register is ever used as an argument. | |
| 2060 // This function is used on startup to build the trampoline stubs in | |
| 2061 // generateOptoStub. Registers not mentioned will be killed by the VM | |
| 2062 // call in the trampoline, and arguments in those registers not be | |
| 2063 // available to the callee. | |
| 2064 bool Matcher::can_be_java_arg(int reg) | |
| 2065 { | |
| 2066 return | |
| 2067 reg == RDI_num || reg == RDI_H_num || | |
| 2068 reg == RSI_num || reg == RSI_H_num || | |
| 2069 reg == RDX_num || reg == RDX_H_num || | |
| 2070 reg == RCX_num || reg == RCX_H_num || | |
| 2071 reg == R8_num || reg == R8_H_num || | |
| 2072 reg == R9_num || reg == R9_H_num || | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2073 reg == R12_num || reg == R12_H_num || |
| 0 | 2074 reg == XMM0_num || reg == XMM0_H_num || |
| 2075 reg == XMM1_num || reg == XMM1_H_num || | |
| 2076 reg == XMM2_num || reg == XMM2_H_num || | |
| 2077 reg == XMM3_num || reg == XMM3_H_num || | |
| 2078 reg == XMM4_num || reg == XMM4_H_num || | |
| 2079 reg == XMM5_num || reg == XMM5_H_num || | |
| 2080 reg == XMM6_num || reg == XMM6_H_num || | |
| 2081 reg == XMM7_num || reg == XMM7_H_num; | |
| 2082 } | |
| 2083 | |
| 2084 bool Matcher::is_spillable_arg(int reg) | |
| 2085 { | |
| 2086 return can_be_java_arg(reg); | |
| 2087 } | |
| 2088 | |
| 2089 // Register for DIVI projection of divmodI | |
| 2090 RegMask Matcher::divI_proj_mask() { | |
| 2091 return INT_RAX_REG_mask; | |
| 2092 } | |
| 2093 | |
| 2094 // Register for MODI projection of divmodI | |
| 2095 RegMask Matcher::modI_proj_mask() { | |
| 2096 return INT_RDX_REG_mask; | |
| 2097 } | |
| 2098 | |
| 2099 // Register for DIVL projection of divmodL | |
| 2100 RegMask Matcher::divL_proj_mask() { | |
| 2101 return LONG_RAX_REG_mask; | |
| 2102 } | |
| 2103 | |
| 2104 // Register for MODL projection of divmodL | |
| 2105 RegMask Matcher::modL_proj_mask() { | |
| 2106 return LONG_RDX_REG_mask; | |
| 2107 } | |
| 2108 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2109 static Address build_address(int b, int i, int s, int d) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2110 Register index = as_Register(i); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2111 Address::ScaleFactor scale = (Address::ScaleFactor)s; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2112 if (index == rsp) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2113 index = noreg; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2114 scale = Address::no_scale; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2115 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2116 Address addr(as_Register(b), index, scale, d); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2117 return addr; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2118 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2119 |
| 0 | 2120 %} |
| 2121 | |
| 2122 //----------ENCODING BLOCK----------------------------------------------------- | |
| 2123 // This block specifies the encoding classes used by the compiler to | |
| 2124 // output byte streams. Encoding classes are parameterized macros | |
| 2125 // used by Machine Instruction Nodes in order to generate the bit | |
| 2126 // encoding of the instruction. Operands specify their base encoding | |
| 2127 // interface with the interface keyword. There are currently | |
| 2128 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, & | |
| 2129 // COND_INTER. REG_INTER causes an operand to generate a function | |
| 2130 // which returns its register number when queried. CONST_INTER causes | |
| 2131 // an operand to generate a function which returns the value of the | |
| 2132 // constant when queried. MEMORY_INTER causes an operand to generate | |
| 2133 // four functions which return the Base Register, the Index Register, | |
| 2134 // the Scale Value, and the Offset Value of the operand when queried. | |
| 2135 // COND_INTER causes an operand to generate six functions which return | |
| 2136 // the encoding code (ie - encoding bits for the instruction) | |
| 2137 // associated with each basic boolean condition for a conditional | |
| 2138 // instruction. | |
| 2139 // | |
| 2140 // Instructions specify two basic values for encoding. Again, a | |
| 2141 // function is available to check if the constant displacement is an | |
| 2142 // oop. They use the ins_encode keyword to specify their encoding | |
| 2143 // classes (which must be a sequence of enc_class names, and their | |
| 2144 // parameters, specified in the encoding block), and they use the | |
| 2145 // opcode keyword to specify, in order, their primary, secondary, and | |
| 2146 // tertiary opcode. Only the opcode sections which a particular | |
| 2147 // instruction needs for encoding need to be specified. | |
| 2148 encode %{ | |
| 2149 // Build emit functions for each basic byte or larger field in the | |
| 2150 // intel encoding scheme (opcode, rm, sib, immediate), and call them | |
| 2151 // from C++ code in the enc_class source block. Emit functions will | |
| 2152 // live in the main source block for now. In future, we can | |
| 2153 // generalize this by adding a syntax that specifies the sizes of | |
| 2154 // fields in an order, so that the adlc can build the emit functions | |
| 2155 // automagically | |
| 2156 | |
| 2157 // Emit primary opcode | |
| 2158 enc_class OpcP | |
| 2159 %{ | |
| 2160 emit_opcode(cbuf, $primary); | |
| 2161 %} | |
| 2162 | |
| 2163 // Emit secondary opcode | |
| 2164 enc_class OpcS | |
| 2165 %{ | |
| 2166 emit_opcode(cbuf, $secondary); | |
| 2167 %} | |
| 2168 | |
| 2169 // Emit tertiary opcode | |
| 2170 enc_class OpcT | |
| 2171 %{ | |
| 2172 emit_opcode(cbuf, $tertiary); | |
| 2173 %} | |
| 2174 | |
| 2175 // Emit opcode directly | |
| 2176 enc_class Opcode(immI d8) | |
| 2177 %{ | |
| 2178 emit_opcode(cbuf, $d8$$constant); | |
| 2179 %} | |
| 2180 | |
| 2181 // Emit size prefix | |
| 2182 enc_class SizePrefix | |
| 2183 %{ | |
| 2184 emit_opcode(cbuf, 0x66); | |
| 2185 %} | |
| 2186 | |
| 2187 enc_class reg(rRegI reg) | |
| 2188 %{ | |
| 2189 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7); | |
| 2190 %} | |
| 2191 | |
| 2192 enc_class reg_reg(rRegI dst, rRegI src) | |
| 2193 %{ | |
| 2194 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7); | |
| 2195 %} | |
| 2196 | |
| 2197 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src) | |
| 2198 %{ | |
| 2199 emit_opcode(cbuf, $opcode$$constant); | |
| 2200 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7); | |
| 2201 %} | |
| 2202 | |
| 2203 enc_class cmpfp_fixup() | |
| 2204 %{ | |
| 2205 // jnp,s exit | |
| 2206 emit_opcode(cbuf, 0x7B); | |
| 2207 emit_d8(cbuf, 0x0A); | |
| 2208 | |
| 2209 // pushfq | |
| 2210 emit_opcode(cbuf, 0x9C); | |
| 2211 | |
| 2212 // andq $0xffffff2b, (%rsp) | |
| 2213 emit_opcode(cbuf, Assembler::REX_W); | |
| 2214 emit_opcode(cbuf, 0x81); | |
| 2215 emit_opcode(cbuf, 0x24); | |
| 2216 emit_opcode(cbuf, 0x24); | |
| 2217 emit_d32(cbuf, 0xffffff2b); | |
| 2218 | |
| 2219 // popfq | |
| 2220 emit_opcode(cbuf, 0x9D); | |
| 2221 | |
| 2222 // nop (target for branch to avoid branch to branch) | |
| 2223 emit_opcode(cbuf, 0x90); | |
| 2224 %} | |
| 2225 | |
| 2226 enc_class cmpfp3(rRegI dst) | |
| 2227 %{ | |
| 2228 int dstenc = $dst$$reg; | |
| 2229 | |
| 2230 // movl $dst, -1 | |
| 2231 if (dstenc >= 8) { | |
| 2232 emit_opcode(cbuf, Assembler::REX_B); | |
| 2233 } | |
| 2234 emit_opcode(cbuf, 0xB8 | (dstenc & 7)); | |
| 2235 emit_d32(cbuf, -1); | |
| 2236 | |
| 2237 // jp,s done | |
| 2238 emit_opcode(cbuf, 0x7A); | |
| 2239 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A); | |
| 2240 | |
| 2241 // jb,s done | |
| 2242 emit_opcode(cbuf, 0x72); | |
| 2243 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08); | |
| 2244 | |
| 2245 // setne $dst | |
| 2246 if (dstenc >= 4) { | |
| 2247 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B); | |
| 2248 } | |
| 2249 emit_opcode(cbuf, 0x0F); | |
| 2250 emit_opcode(cbuf, 0x95); | |
| 2251 emit_opcode(cbuf, 0xC0 | (dstenc & 7)); | |
| 2252 | |
| 2253 // movzbl $dst, $dst | |
| 2254 if (dstenc >= 4) { | |
| 2255 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB); | |
| 2256 } | |
| 2257 emit_opcode(cbuf, 0x0F); | |
| 2258 emit_opcode(cbuf, 0xB6); | |
| 2259 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7); | |
| 2260 %} | |
| 2261 | |
| 2262 enc_class cdql_enc(no_rax_rdx_RegI div) | |
| 2263 %{ | |
| 2264 // Full implementation of Java idiv and irem; checks for | |
| 2265 // special case as described in JVM spec., p.243 & p.271. | |
| 2266 // | |
| 2267 // normal case special case | |
| 2268 // | |
| 2269 // input : rax: dividend min_int | |
| 2270 // reg: divisor -1 | |
| 2271 // | |
| 2272 // output: rax: quotient (= rax idiv reg) min_int | |
| 2273 // rdx: remainder (= rax irem reg) 0 | |
| 2274 // | |
| 2275 // Code sequnce: | |
| 2276 // | |
| 2277 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax | |
| 2278 // 5: 75 07/08 jne e <normal> | |
| 2279 // 7: 33 d2 xor %edx,%edx | |
| 2280 // [div >= 8 -> offset + 1] | |
| 2281 // [REX_B] | |
| 2282 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div | |
| 2283 // c: 74 03/04 je 11 <done> | |
| 2284 // 000000000000000e <normal>: | |
| 2285 // e: 99 cltd | |
| 2286 // [div >= 8 -> offset + 1] | |
| 2287 // [REX_B] | |
| 2288 // f: f7 f9 idiv $div | |
| 2289 // 0000000000000011 <done>: | |
| 2290 | |
| 2291 // cmp $0x80000000,%eax | |
| 2292 emit_opcode(cbuf, 0x3d); | |
| 2293 emit_d8(cbuf, 0x00); | |
| 2294 emit_d8(cbuf, 0x00); | |
| 2295 emit_d8(cbuf, 0x00); | |
| 2296 emit_d8(cbuf, 0x80); | |
| 2297 | |
| 2298 // jne e <normal> | |
| 2299 emit_opcode(cbuf, 0x75); | |
| 2300 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08); | |
| 2301 | |
| 2302 // xor %edx,%edx | |
| 2303 emit_opcode(cbuf, 0x33); | |
| 2304 emit_d8(cbuf, 0xD2); | |
| 2305 | |
| 2306 // cmp $0xffffffffffffffff,%ecx | |
| 2307 if ($div$$reg >= 8) { | |
| 2308 emit_opcode(cbuf, Assembler::REX_B); | |
| 2309 } | |
| 2310 emit_opcode(cbuf, 0x83); | |
| 2311 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7); | |
| 2312 emit_d8(cbuf, 0xFF); | |
| 2313 | |
| 2314 // je 11 <done> | |
| 2315 emit_opcode(cbuf, 0x74); | |
| 2316 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04); | |
| 2317 | |
| 2318 // <normal> | |
| 2319 // cltd | |
| 2320 emit_opcode(cbuf, 0x99); | |
| 2321 | |
| 2322 // idivl (note: must be emitted by the user of this rule) | |
| 2323 // <done> | |
| 2324 %} | |
| 2325 | |
| 2326 enc_class cdqq_enc(no_rax_rdx_RegL div) | |
| 2327 %{ | |
| 2328 // Full implementation of Java ldiv and lrem; checks for | |
| 2329 // special case as described in JVM spec., p.243 & p.271. | |
| 2330 // | |
| 2331 // normal case special case | |
| 2332 // | |
| 2333 // input : rax: dividend min_long | |
| 2334 // reg: divisor -1 | |
| 2335 // | |
| 2336 // output: rax: quotient (= rax idiv reg) min_long | |
| 2337 // rdx: remainder (= rax irem reg) 0 | |
| 2338 // | |
| 2339 // Code sequnce: | |
| 2340 // | |
| 2341 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx | |
| 2342 // 7: 00 00 80 | |
| 2343 // a: 48 39 d0 cmp %rdx,%rax | |
| 2344 // d: 75 08 jne 17 <normal> | |
| 2345 // f: 33 d2 xor %edx,%edx | |
| 2346 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div | |
| 2347 // 15: 74 05 je 1c <done> | |
| 2348 // 0000000000000017 <normal>: | |
| 2349 // 17: 48 99 cqto | |
| 2350 // 19: 48 f7 f9 idiv $div | |
| 2351 // 000000000000001c <done>: | |
| 2352 | |
| 2353 // mov $0x8000000000000000,%rdx | |
| 2354 emit_opcode(cbuf, Assembler::REX_W); | |
| 2355 emit_opcode(cbuf, 0xBA); | |
| 2356 emit_d8(cbuf, 0x00); | |
| 2357 emit_d8(cbuf, 0x00); | |
| 2358 emit_d8(cbuf, 0x00); | |
| 2359 emit_d8(cbuf, 0x00); | |
| 2360 emit_d8(cbuf, 0x00); | |
| 2361 emit_d8(cbuf, 0x00); | |
| 2362 emit_d8(cbuf, 0x00); | |
| 2363 emit_d8(cbuf, 0x80); | |
| 2364 | |
| 2365 // cmp %rdx,%rax | |
| 2366 emit_opcode(cbuf, Assembler::REX_W); | |
| 2367 emit_opcode(cbuf, 0x39); | |
| 2368 emit_d8(cbuf, 0xD0); | |
| 2369 | |
| 2370 // jne 17 <normal> | |
| 2371 emit_opcode(cbuf, 0x75); | |
| 2372 emit_d8(cbuf, 0x08); | |
| 2373 | |
| 2374 // xor %edx,%edx | |
| 2375 emit_opcode(cbuf, 0x33); | |
| 2376 emit_d8(cbuf, 0xD2); | |
| 2377 | |
| 2378 // cmp $0xffffffffffffffff,$div | |
| 2379 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB); | |
| 2380 emit_opcode(cbuf, 0x83); | |
| 2381 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7); | |
| 2382 emit_d8(cbuf, 0xFF); | |
| 2383 | |
| 2384 // je 1e <done> | |
| 2385 emit_opcode(cbuf, 0x74); | |
| 2386 emit_d8(cbuf, 0x05); | |
| 2387 | |
| 2388 // <normal> | |
| 2389 // cqto | |
| 2390 emit_opcode(cbuf, Assembler::REX_W); | |
| 2391 emit_opcode(cbuf, 0x99); | |
| 2392 | |
| 2393 // idivq (note: must be emitted by the user of this rule) | |
| 2394 // <done> | |
| 2395 %} | |
| 2396 | |
| 2397 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension | |
| 2398 enc_class OpcSE(immI imm) | |
| 2399 %{ | |
| 2400 // Emit primary opcode and set sign-extend bit | |
| 2401 // Check for 8-bit immediate, and set sign extend bit in opcode | |
| 2402 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) { | |
| 2403 emit_opcode(cbuf, $primary | 0x02); | |
| 2404 } else { | |
| 2405 // 32-bit immediate | |
| 2406 emit_opcode(cbuf, $primary); | |
| 2407 } | |
| 2408 %} | |
| 2409 | |
| 2410 enc_class OpcSErm(rRegI dst, immI imm) | |
| 2411 %{ | |
| 2412 // OpcSEr/m | |
| 2413 int dstenc = $dst$$reg; | |
| 2414 if (dstenc >= 8) { | |
| 2415 emit_opcode(cbuf, Assembler::REX_B); | |
| 2416 dstenc -= 8; | |
| 2417 } | |
| 2418 // Emit primary opcode and set sign-extend bit | |
| 2419 // Check for 8-bit immediate, and set sign extend bit in opcode | |
| 2420 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) { | |
| 2421 emit_opcode(cbuf, $primary | 0x02); | |
| 2422 } else { | |
| 2423 // 32-bit immediate | |
| 2424 emit_opcode(cbuf, $primary); | |
| 2425 } | |
| 2426 // Emit r/m byte with secondary opcode, after primary opcode. | |
| 2427 emit_rm(cbuf, 0x3, $secondary, dstenc); | |
| 2428 %} | |
| 2429 | |
| 2430 enc_class OpcSErm_wide(rRegL dst, immI imm) | |
| 2431 %{ | |
| 2432 // OpcSEr/m | |
| 2433 int dstenc = $dst$$reg; | |
| 2434 if (dstenc < 8) { | |
| 2435 emit_opcode(cbuf, Assembler::REX_W); | |
| 2436 } else { | |
| 2437 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2438 dstenc -= 8; | |
| 2439 } | |
| 2440 // Emit primary opcode and set sign-extend bit | |
| 2441 // Check for 8-bit immediate, and set sign extend bit in opcode | |
| 2442 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) { | |
| 2443 emit_opcode(cbuf, $primary | 0x02); | |
| 2444 } else { | |
| 2445 // 32-bit immediate | |
| 2446 emit_opcode(cbuf, $primary); | |
| 2447 } | |
| 2448 // Emit r/m byte with secondary opcode, after primary opcode. | |
| 2449 emit_rm(cbuf, 0x3, $secondary, dstenc); | |
| 2450 %} | |
| 2451 | |
| 2452 enc_class Con8or32(immI imm) | |
| 2453 %{ | |
| 2454 // Check for 8-bit immediate, and set sign extend bit in opcode | |
| 2455 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) { | |
| 2456 $$$emit8$imm$$constant; | |
| 2457 } else { | |
| 2458 // 32-bit immediate | |
| 2459 $$$emit32$imm$$constant; | |
| 2460 } | |
| 2461 %} | |
| 2462 | |
| 2463 enc_class Lbl(label labl) | |
| 2464 %{ | |
| 2465 // JMP, CALL | |
| 2466 Label* l = $labl$$label; | |
| 2467 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0); | |
| 2468 %} | |
| 2469 | |
| 2470 enc_class LblShort(label labl) | |
| 2471 %{ | |
| 2472 // JMP, CALL | |
| 2473 Label* l = $labl$$label; | |
| 2474 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0; | |
| 2475 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp"); | |
| 2476 emit_d8(cbuf, disp); | |
| 2477 %} | |
| 2478 | |
| 2479 enc_class opc2_reg(rRegI dst) | |
| 2480 %{ | |
| 2481 // BSWAP | |
| 2482 emit_cc(cbuf, $secondary, $dst$$reg); | |
| 2483 %} | |
| 2484 | |
| 2485 enc_class opc3_reg(rRegI dst) | |
| 2486 %{ | |
| 2487 // BSWAP | |
| 2488 emit_cc(cbuf, $tertiary, $dst$$reg); | |
| 2489 %} | |
| 2490 | |
| 2491 enc_class reg_opc(rRegI div) | |
| 2492 %{ | |
| 2493 // INC, DEC, IDIV, IMOD, JMP indirect, ... | |
| 2494 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7); | |
| 2495 %} | |
| 2496 | |
| 2497 enc_class Jcc(cmpOp cop, label labl) | |
| 2498 %{ | |
| 2499 // JCC | |
| 2500 Label* l = $labl$$label; | |
| 2501 $$$emit8$primary; | |
| 2502 emit_cc(cbuf, $secondary, $cop$$cmpcode); | |
| 2503 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0); | |
| 2504 %} | |
| 2505 | |
| 2506 enc_class JccShort (cmpOp cop, label labl) | |
| 2507 %{ | |
| 2508 // JCC | |
| 2509 Label *l = $labl$$label; | |
| 2510 emit_cc(cbuf, $primary, $cop$$cmpcode); | |
| 2511 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0; | |
| 2512 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp"); | |
| 2513 emit_d8(cbuf, disp); | |
| 2514 %} | |
| 2515 | |
| 2516 enc_class enc_cmov(cmpOp cop) | |
| 2517 %{ | |
| 2518 // CMOV | |
| 2519 $$$emit8$primary; | |
| 2520 emit_cc(cbuf, $secondary, $cop$$cmpcode); | |
| 2521 %} | |
| 2522 | |
| 2523 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src) | |
| 2524 %{ | |
| 2525 // Invert sense of branch from sense of cmov | |
| 2526 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1); | |
| 2527 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8) | |
| 2528 ? (UseXmmRegToRegMoveAll ? 3 : 4) | |
| 2529 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX | |
| 2530 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src) | |
| 2531 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3); | |
| 2532 if ($dst$$reg < 8) { | |
| 2533 if ($src$$reg >= 8) { | |
| 2534 emit_opcode(cbuf, Assembler::REX_B); | |
| 2535 } | |
| 2536 } else { | |
| 2537 if ($src$$reg < 8) { | |
| 2538 emit_opcode(cbuf, Assembler::REX_R); | |
| 2539 } else { | |
| 2540 emit_opcode(cbuf, Assembler::REX_RB); | |
| 2541 } | |
| 2542 } | |
| 2543 emit_opcode(cbuf, 0x0F); | |
| 2544 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10); | |
| 2545 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7); | |
| 2546 %} | |
| 2547 | |
| 2548 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src) | |
| 2549 %{ | |
| 2550 // Invert sense of branch from sense of cmov | |
| 2551 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1); | |
| 2552 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX | |
| 2553 | |
| 2554 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src) | |
| 2555 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2); | |
| 2556 if ($dst$$reg < 8) { | |
| 2557 if ($src$$reg >= 8) { | |
| 2558 emit_opcode(cbuf, Assembler::REX_B); | |
| 2559 } | |
| 2560 } else { | |
| 2561 if ($src$$reg < 8) { | |
| 2562 emit_opcode(cbuf, Assembler::REX_R); | |
| 2563 } else { | |
| 2564 emit_opcode(cbuf, Assembler::REX_RB); | |
| 2565 } | |
| 2566 } | |
| 2567 emit_opcode(cbuf, 0x0F); | |
| 2568 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10); | |
| 2569 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7); | |
| 2570 %} | |
| 2571 | |
| 2572 enc_class enc_PartialSubtypeCheck() | |
| 2573 %{ | |
| 2574 Register Rrdi = as_Register(RDI_enc); // result register | |
| 2575 Register Rrax = as_Register(RAX_enc); // super class | |
| 2576 Register Rrcx = as_Register(RCX_enc); // killed | |
| 2577 Register Rrsi = as_Register(RSI_enc); // sub class | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2578 Label hit, miss, cmiss; |
| 0 | 2579 |
| 2580 MacroAssembler _masm(&cbuf); | |
| 2581 // Compare super with sub directly, since super is not in its own SSA. | |
| 2582 // The compiler used to emit this test, but we fold it in here, | |
| 2583 // to allow platform-specific tweaking on sparc. | |
| 304 | 2584 __ cmpptr(Rrax, Rrsi); |
| 0 | 2585 __ jcc(Assembler::equal, hit); |
| 2586 #ifndef PRODUCT | |
| 2587 __ lea(Rrcx, ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr)); | |
| 2588 __ incrementl(Address(Rrcx, 0)); | |
| 2589 #endif //PRODUCT | |
| 304 | 2590 __ movptr(Rrdi, Address(Rrsi, |
| 2591 sizeof(oopDesc) + | |
| 0 | 2592 Klass::secondary_supers_offset_in_bytes())); |
| 2593 __ movl(Rrcx, Address(Rrdi, arrayOopDesc::length_offset_in_bytes())); | |
| 304 | 2594 __ addptr(Rrdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2595 if (UseCompressedOops) { |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2596 __ push(Rrax); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2597 __ encode_heap_oop(Rrax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2598 __ repne_scanl(); |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2599 __ pop(Rrax); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2600 __ jccb(Assembler::notEqual, miss); |
| 304 | 2601 __ movptr(Address(Rrsi, |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2602 sizeof(oopDesc) + |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2603 Klass::secondary_super_cache_offset_in_bytes()), |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2604 Rrax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2605 __ jmp(hit); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2606 } else { |
| 304 | 2607 __ repne_scan(); |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2608 __ jccb(Assembler::notEqual, miss); |
| 304 | 2609 __ movptr(Address(Rrsi, |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2610 sizeof(oopDesc) + |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2611 Klass::secondary_super_cache_offset_in_bytes()), |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2612 Rrax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
2613 } |
| 0 | 2614 __ bind(hit); |
| 2615 if ($primary) { | |
| 304 | 2616 __ xorptr(Rrdi, Rrdi); |
| 0 | 2617 } |
| 2618 __ bind(miss); | |
| 2619 %} | |
| 2620 | |
| 2621 enc_class Java_To_Interpreter(method meth) | |
| 2622 %{ | |
| 2623 // CALL Java_To_Interpreter | |
| 2624 // This is the instruction starting address for relocation info. | |
| 2625 cbuf.set_inst_mark(); | |
| 2626 $$$emit8$primary; | |
| 2627 // CALL directly to the runtime | |
| 2628 emit_d32_reloc(cbuf, | |
| 2629 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4), | |
| 2630 runtime_call_Relocation::spec(), | |
| 2631 RELOC_DISP32); | |
| 2632 %} | |
| 2633 | |
| 2634 enc_class Java_Static_Call(method meth) | |
| 2635 %{ | |
| 2636 // JAVA STATIC CALL | |
| 2637 // CALL to fixup routine. Fixup routine uses ScopeDesc info to | |
| 2638 // determine who we intended to call. | |
| 2639 cbuf.set_inst_mark(); | |
| 2640 $$$emit8$primary; | |
| 2641 | |
| 2642 if (!_method) { | |
| 2643 emit_d32_reloc(cbuf, | |
| 2644 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4), | |
| 2645 runtime_call_Relocation::spec(), | |
| 2646 RELOC_DISP32); | |
| 2647 } else if (_optimized_virtual) { | |
| 2648 emit_d32_reloc(cbuf, | |
| 2649 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4), | |
| 2650 opt_virtual_call_Relocation::spec(), | |
| 2651 RELOC_DISP32); | |
| 2652 } else { | |
| 2653 emit_d32_reloc(cbuf, | |
| 2654 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4), | |
| 2655 static_call_Relocation::spec(), | |
| 2656 RELOC_DISP32); | |
| 2657 } | |
| 2658 if (_method) { | |
| 2659 // Emit stub for static call | |
| 2660 emit_java_to_interp(cbuf); | |
| 2661 } | |
| 2662 %} | |
| 2663 | |
| 2664 enc_class Java_Dynamic_Call(method meth) | |
| 2665 %{ | |
| 2666 // JAVA DYNAMIC CALL | |
| 2667 // !!!!! | |
| 2668 // Generate "movq rax, -1", placeholder instruction to load oop-info | |
| 2669 // emit_call_dynamic_prologue( cbuf ); | |
| 2670 cbuf.set_inst_mark(); | |
| 2671 | |
| 2672 // movq rax, -1 | |
| 2673 emit_opcode(cbuf, Assembler::REX_W); | |
| 2674 emit_opcode(cbuf, 0xB8 | RAX_enc); | |
| 2675 emit_d64_reloc(cbuf, | |
| 2676 (int64_t) Universe::non_oop_word(), | |
| 2677 oop_Relocation::spec_for_immediate(), RELOC_IMM64); | |
| 2678 address virtual_call_oop_addr = cbuf.inst_mark(); | |
| 2679 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine | |
| 2680 // who we intended to call. | |
| 2681 cbuf.set_inst_mark(); | |
| 2682 $$$emit8$primary; | |
| 2683 emit_d32_reloc(cbuf, | |
| 2684 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4), | |
| 2685 virtual_call_Relocation::spec(virtual_call_oop_addr), | |
| 2686 RELOC_DISP32); | |
| 2687 %} | |
| 2688 | |
| 2689 enc_class Java_Compiled_Call(method meth) | |
| 2690 %{ | |
| 2691 // JAVA COMPILED CALL | |
| 2692 int disp = in_bytes(methodOopDesc:: from_compiled_offset()); | |
| 2693 | |
| 2694 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!! | |
| 2695 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small"); | |
| 2696 | |
| 2697 // callq *disp(%rax) | |
| 2698 cbuf.set_inst_mark(); | |
| 2699 $$$emit8$primary; | |
| 2700 if (disp < 0x80) { | |
| 2701 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte | |
| 2702 emit_d8(cbuf, disp); // Displacement | |
| 2703 } else { | |
| 2704 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte | |
| 2705 emit_d32(cbuf, disp); // Displacement | |
| 2706 } | |
| 2707 %} | |
| 2708 | |
| 2709 enc_class reg_opc_imm(rRegI dst, immI8 shift) | |
| 2710 %{ | |
| 2711 // SAL, SAR, SHR | |
| 2712 int dstenc = $dst$$reg; | |
| 2713 if (dstenc >= 8) { | |
| 2714 emit_opcode(cbuf, Assembler::REX_B); | |
| 2715 dstenc -= 8; | |
| 2716 } | |
| 2717 $$$emit8$primary; | |
| 2718 emit_rm(cbuf, 0x3, $secondary, dstenc); | |
| 2719 $$$emit8$shift$$constant; | |
| 2720 %} | |
| 2721 | |
| 2722 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift) | |
| 2723 %{ | |
| 2724 // SAL, SAR, SHR | |
| 2725 int dstenc = $dst$$reg; | |
| 2726 if (dstenc < 8) { | |
| 2727 emit_opcode(cbuf, Assembler::REX_W); | |
| 2728 } else { | |
| 2729 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2730 dstenc -= 8; | |
| 2731 } | |
| 2732 $$$emit8$primary; | |
| 2733 emit_rm(cbuf, 0x3, $secondary, dstenc); | |
| 2734 $$$emit8$shift$$constant; | |
| 2735 %} | |
| 2736 | |
| 2737 enc_class load_immI(rRegI dst, immI src) | |
| 2738 %{ | |
| 2739 int dstenc = $dst$$reg; | |
| 2740 if (dstenc >= 8) { | |
| 2741 emit_opcode(cbuf, Assembler::REX_B); | |
| 2742 dstenc -= 8; | |
| 2743 } | |
| 2744 emit_opcode(cbuf, 0xB8 | dstenc); | |
| 2745 $$$emit32$src$$constant; | |
| 2746 %} | |
| 2747 | |
| 2748 enc_class load_immL(rRegL dst, immL src) | |
| 2749 %{ | |
| 2750 int dstenc = $dst$$reg; | |
| 2751 if (dstenc < 8) { | |
| 2752 emit_opcode(cbuf, Assembler::REX_W); | |
| 2753 } else { | |
| 2754 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2755 dstenc -= 8; | |
| 2756 } | |
| 2757 emit_opcode(cbuf, 0xB8 | dstenc); | |
| 2758 emit_d64(cbuf, $src$$constant); | |
| 2759 %} | |
| 2760 | |
| 2761 enc_class load_immUL32(rRegL dst, immUL32 src) | |
| 2762 %{ | |
| 2763 // same as load_immI, but this time we care about zeroes in the high word | |
| 2764 int dstenc = $dst$$reg; | |
| 2765 if (dstenc >= 8) { | |
| 2766 emit_opcode(cbuf, Assembler::REX_B); | |
| 2767 dstenc -= 8; | |
| 2768 } | |
| 2769 emit_opcode(cbuf, 0xB8 | dstenc); | |
| 2770 $$$emit32$src$$constant; | |
| 2771 %} | |
| 2772 | |
| 2773 enc_class load_immL32(rRegL dst, immL32 src) | |
| 2774 %{ | |
| 2775 int dstenc = $dst$$reg; | |
| 2776 if (dstenc < 8) { | |
| 2777 emit_opcode(cbuf, Assembler::REX_W); | |
| 2778 } else { | |
| 2779 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2780 dstenc -= 8; | |
| 2781 } | |
| 2782 emit_opcode(cbuf, 0xC7); | |
| 2783 emit_rm(cbuf, 0x03, 0x00, dstenc); | |
| 2784 $$$emit32$src$$constant; | |
| 2785 %} | |
| 2786 | |
| 2787 enc_class load_immP31(rRegP dst, immP32 src) | |
| 2788 %{ | |
| 2789 // same as load_immI, but this time we care about zeroes in the high word | |
| 2790 int dstenc = $dst$$reg; | |
| 2791 if (dstenc >= 8) { | |
| 2792 emit_opcode(cbuf, Assembler::REX_B); | |
| 2793 dstenc -= 8; | |
| 2794 } | |
| 2795 emit_opcode(cbuf, 0xB8 | dstenc); | |
| 2796 $$$emit32$src$$constant; | |
| 2797 %} | |
| 2798 | |
| 2799 enc_class load_immP(rRegP dst, immP src) | |
| 2800 %{ | |
| 2801 int dstenc = $dst$$reg; | |
| 2802 if (dstenc < 8) { | |
| 2803 emit_opcode(cbuf, Assembler::REX_W); | |
| 2804 } else { | |
| 2805 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2806 dstenc -= 8; | |
| 2807 } | |
| 2808 emit_opcode(cbuf, 0xB8 | dstenc); | |
| 2809 // This next line should be generated from ADLC | |
| 2810 if ($src->constant_is_oop()) { | |
| 2811 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64); | |
| 2812 } else { | |
| 2813 emit_d64(cbuf, $src$$constant); | |
| 2814 } | |
| 2815 %} | |
| 2816 | |
| 2817 enc_class load_immF(regF dst, immF con) | |
| 2818 %{ | |
| 2819 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 2820 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101 | |
| 2821 emit_float_constant(cbuf, $con$$constant); | |
| 2822 %} | |
| 2823 | |
| 2824 enc_class load_immD(regD dst, immD con) | |
| 2825 %{ | |
| 2826 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 2827 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101 | |
| 2828 emit_double_constant(cbuf, $con$$constant); | |
| 2829 %} | |
| 2830 | |
| 2831 enc_class load_conF (regF dst, immF con) %{ // Load float constant | |
| 2832 emit_opcode(cbuf, 0xF3); | |
| 2833 if ($dst$$reg >= 8) { | |
| 2834 emit_opcode(cbuf, Assembler::REX_R); | |
| 2835 } | |
| 2836 emit_opcode(cbuf, 0x0F); | |
| 2837 emit_opcode(cbuf, 0x10); | |
| 2838 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101 | |
| 2839 emit_float_constant(cbuf, $con$$constant); | |
| 2840 %} | |
| 2841 | |
| 2842 enc_class load_conD (regD dst, immD con) %{ // Load double constant | |
| 2843 // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con) | |
| 2844 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66); | |
| 2845 if ($dst$$reg >= 8) { | |
| 2846 emit_opcode(cbuf, Assembler::REX_R); | |
| 2847 } | |
| 2848 emit_opcode(cbuf, 0x0F); | |
| 2849 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12); | |
| 2850 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101 | |
| 2851 emit_double_constant(cbuf, $con$$constant); | |
| 2852 %} | |
| 2853 | |
| 2854 // Encode a reg-reg copy. If it is useless, then empty encoding. | |
| 2855 enc_class enc_copy(rRegI dst, rRegI src) | |
| 2856 %{ | |
| 2857 encode_copy(cbuf, $dst$$reg, $src$$reg); | |
| 2858 %} | |
| 2859 | |
| 2860 // Encode xmm reg-reg copy. If it is useless, then empty encoding. | |
| 2861 enc_class enc_CopyXD( RegD dst, RegD src ) %{ | |
| 2862 encode_CopyXD( cbuf, $dst$$reg, $src$$reg ); | |
| 2863 %} | |
| 2864 | |
| 2865 enc_class enc_copy_always(rRegI dst, rRegI src) | |
| 2866 %{ | |
| 2867 int srcenc = $src$$reg; | |
| 2868 int dstenc = $dst$$reg; | |
| 2869 | |
| 2870 if (dstenc < 8) { | |
| 2871 if (srcenc >= 8) { | |
| 2872 emit_opcode(cbuf, Assembler::REX_B); | |
| 2873 srcenc -= 8; | |
| 2874 } | |
| 2875 } else { | |
| 2876 if (srcenc < 8) { | |
| 2877 emit_opcode(cbuf, Assembler::REX_R); | |
| 2878 } else { | |
| 2879 emit_opcode(cbuf, Assembler::REX_RB); | |
| 2880 srcenc -= 8; | |
| 2881 } | |
| 2882 dstenc -= 8; | |
| 2883 } | |
| 2884 | |
| 2885 emit_opcode(cbuf, 0x8B); | |
| 2886 emit_rm(cbuf, 0x3, dstenc, srcenc); | |
| 2887 %} | |
| 2888 | |
| 2889 enc_class enc_copy_wide(rRegL dst, rRegL src) | |
| 2890 %{ | |
| 2891 int srcenc = $src$$reg; | |
| 2892 int dstenc = $dst$$reg; | |
| 2893 | |
| 2894 if (dstenc != srcenc) { | |
| 2895 if (dstenc < 8) { | |
| 2896 if (srcenc < 8) { | |
| 2897 emit_opcode(cbuf, Assembler::REX_W); | |
| 2898 } else { | |
| 2899 emit_opcode(cbuf, Assembler::REX_WB); | |
| 2900 srcenc -= 8; | |
| 2901 } | |
| 2902 } else { | |
| 2903 if (srcenc < 8) { | |
| 2904 emit_opcode(cbuf, Assembler::REX_WR); | |
| 2905 } else { | |
| 2906 emit_opcode(cbuf, Assembler::REX_WRB); | |
| 2907 srcenc -= 8; | |
| 2908 } | |
| 2909 dstenc -= 8; | |
| 2910 } | |
| 2911 emit_opcode(cbuf, 0x8B); | |
| 2912 emit_rm(cbuf, 0x3, dstenc, srcenc); | |
| 2913 } | |
| 2914 %} | |
| 2915 | |
| 2916 enc_class Con32(immI src) | |
| 2917 %{ | |
| 2918 // Output immediate | |
| 2919 $$$emit32$src$$constant; | |
| 2920 %} | |
| 2921 | |
| 2922 enc_class Con64(immL src) | |
| 2923 %{ | |
| 2924 // Output immediate | |
| 2925 emit_d64($src$$constant); | |
| 2926 %} | |
| 2927 | |
| 2928 enc_class Con32F_as_bits(immF src) | |
| 2929 %{ | |
| 2930 // Output Float immediate bits | |
| 2931 jfloat jf = $src$$constant; | |
| 2932 jint jf_as_bits = jint_cast(jf); | |
| 2933 emit_d32(cbuf, jf_as_bits); | |
| 2934 %} | |
| 2935 | |
| 2936 enc_class Con16(immI src) | |
| 2937 %{ | |
| 2938 // Output immediate | |
| 2939 $$$emit16$src$$constant; | |
| 2940 %} | |
| 2941 | |
| 2942 // How is this different from Con32??? XXX | |
| 2943 enc_class Con_d32(immI src) | |
| 2944 %{ | |
| 2945 emit_d32(cbuf,$src$$constant); | |
| 2946 %} | |
| 2947 | |
| 2948 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI) | |
| 2949 // Output immediate memory reference | |
| 2950 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 ); | |
| 2951 emit_d32(cbuf, 0x00); | |
| 2952 %} | |
| 2953 | |
| 2954 enc_class jump_enc(rRegL switch_val, rRegI dest) %{ | |
| 2955 MacroAssembler masm(&cbuf); | |
| 2956 | |
| 2957 Register switch_reg = as_Register($switch_val$$reg); | |
| 2958 Register dest_reg = as_Register($dest$$reg); | |
| 2959 address table_base = masm.address_table_constant(_index2label); | |
| 2960 | |
| 2961 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10 | |
| 2962 // to do that and the compiler is using that register as one it can allocate. | |
| 2963 // So we build it all by hand. | |
| 2964 // Address index(noreg, switch_reg, Address::times_1); | |
| 2965 // ArrayAddress dispatch(table, index); | |
| 2966 | |
| 2967 Address dispatch(dest_reg, switch_reg, Address::times_1); | |
| 2968 | |
| 2969 masm.lea(dest_reg, InternalAddress(table_base)); | |
| 2970 masm.jmp(dispatch); | |
| 2971 %} | |
| 2972 | |
| 2973 enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{ | |
| 2974 MacroAssembler masm(&cbuf); | |
| 2975 | |
| 2976 Register switch_reg = as_Register($switch_val$$reg); | |
| 2977 Register dest_reg = as_Register($dest$$reg); | |
| 2978 address table_base = masm.address_table_constant(_index2label); | |
| 2979 | |
| 2980 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10 | |
| 2981 // to do that and the compiler is using that register as one it can allocate. | |
| 2982 // So we build it all by hand. | |
| 2983 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant); | |
| 2984 // ArrayAddress dispatch(table, index); | |
| 2985 | |
| 2986 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant); | |
| 2987 | |
| 2988 masm.lea(dest_reg, InternalAddress(table_base)); | |
| 2989 masm.jmp(dispatch); | |
| 2990 %} | |
| 2991 | |
| 2992 enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{ | |
| 2993 MacroAssembler masm(&cbuf); | |
| 2994 | |
| 2995 Register switch_reg = as_Register($switch_val$$reg); | |
| 2996 Register dest_reg = as_Register($dest$$reg); | |
| 2997 address table_base = masm.address_table_constant(_index2label); | |
| 2998 | |
| 2999 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10 | |
| 3000 // to do that and the compiler is using that register as one it can allocate. | |
| 3001 // So we build it all by hand. | |
| 3002 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant); | |
| 3003 // ArrayAddress dispatch(table, index); | |
| 3004 | |
| 3005 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant); | |
| 3006 masm.lea(dest_reg, InternalAddress(table_base)); | |
| 3007 masm.jmp(dispatch); | |
| 3008 | |
| 3009 %} | |
| 3010 | |
| 3011 enc_class lock_prefix() | |
| 3012 %{ | |
| 3013 if (os::is_MP()) { | |
| 3014 emit_opcode(cbuf, 0xF0); // lock | |
| 3015 } | |
| 3016 %} | |
| 3017 | |
| 3018 enc_class REX_mem(memory mem) | |
| 3019 %{ | |
| 3020 if ($mem$$base >= 8) { | |
| 3021 if ($mem$$index < 8) { | |
| 3022 emit_opcode(cbuf, Assembler::REX_B); | |
| 3023 } else { | |
| 3024 emit_opcode(cbuf, Assembler::REX_XB); | |
| 3025 } | |
| 3026 } else { | |
| 3027 if ($mem$$index >= 8) { | |
| 3028 emit_opcode(cbuf, Assembler::REX_X); | |
| 3029 } | |
| 3030 } | |
| 3031 %} | |
| 3032 | |
| 3033 enc_class REX_mem_wide(memory mem) | |
| 3034 %{ | |
| 3035 if ($mem$$base >= 8) { | |
| 3036 if ($mem$$index < 8) { | |
| 3037 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3038 } else { | |
| 3039 emit_opcode(cbuf, Assembler::REX_WXB); | |
| 3040 } | |
| 3041 } else { | |
| 3042 if ($mem$$index < 8) { | |
| 3043 emit_opcode(cbuf, Assembler::REX_W); | |
| 3044 } else { | |
| 3045 emit_opcode(cbuf, Assembler::REX_WX); | |
| 3046 } | |
| 3047 } | |
| 3048 %} | |
| 3049 | |
| 3050 // for byte regs | |
| 3051 enc_class REX_breg(rRegI reg) | |
| 3052 %{ | |
| 3053 if ($reg$$reg >= 4) { | |
| 3054 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B); | |
| 3055 } | |
| 3056 %} | |
| 3057 | |
| 3058 // for byte regs | |
| 3059 enc_class REX_reg_breg(rRegI dst, rRegI src) | |
| 3060 %{ | |
| 3061 if ($dst$$reg < 8) { | |
| 3062 if ($src$$reg >= 4) { | |
| 3063 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B); | |
| 3064 } | |
| 3065 } else { | |
| 3066 if ($src$$reg < 8) { | |
| 3067 emit_opcode(cbuf, Assembler::REX_R); | |
| 3068 } else { | |
| 3069 emit_opcode(cbuf, Assembler::REX_RB); | |
| 3070 } | |
| 3071 } | |
| 3072 %} | |
| 3073 | |
| 3074 // for byte regs | |
| 3075 enc_class REX_breg_mem(rRegI reg, memory mem) | |
| 3076 %{ | |
| 3077 if ($reg$$reg < 8) { | |
| 3078 if ($mem$$base < 8) { | |
| 3079 if ($mem$$index >= 8) { | |
| 3080 emit_opcode(cbuf, Assembler::REX_X); | |
| 3081 } else if ($reg$$reg >= 4) { | |
| 3082 emit_opcode(cbuf, Assembler::REX); | |
| 3083 } | |
| 3084 } else { | |
| 3085 if ($mem$$index < 8) { | |
| 3086 emit_opcode(cbuf, Assembler::REX_B); | |
| 3087 } else { | |
| 3088 emit_opcode(cbuf, Assembler::REX_XB); | |
| 3089 } | |
| 3090 } | |
| 3091 } else { | |
| 3092 if ($mem$$base < 8) { | |
| 3093 if ($mem$$index < 8) { | |
| 3094 emit_opcode(cbuf, Assembler::REX_R); | |
| 3095 } else { | |
| 3096 emit_opcode(cbuf, Assembler::REX_RX); | |
| 3097 } | |
| 3098 } else { | |
| 3099 if ($mem$$index < 8) { | |
| 3100 emit_opcode(cbuf, Assembler::REX_RB); | |
| 3101 } else { | |
| 3102 emit_opcode(cbuf, Assembler::REX_RXB); | |
| 3103 } | |
| 3104 } | |
| 3105 } | |
| 3106 %} | |
| 3107 | |
| 3108 enc_class REX_reg(rRegI reg) | |
| 3109 %{ | |
| 3110 if ($reg$$reg >= 8) { | |
| 3111 emit_opcode(cbuf, Assembler::REX_B); | |
| 3112 } | |
| 3113 %} | |
| 3114 | |
| 3115 enc_class REX_reg_wide(rRegI reg) | |
| 3116 %{ | |
| 3117 if ($reg$$reg < 8) { | |
| 3118 emit_opcode(cbuf, Assembler::REX_W); | |
| 3119 } else { | |
| 3120 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3121 } | |
| 3122 %} | |
| 3123 | |
| 3124 enc_class REX_reg_reg(rRegI dst, rRegI src) | |
| 3125 %{ | |
| 3126 if ($dst$$reg < 8) { | |
| 3127 if ($src$$reg >= 8) { | |
| 3128 emit_opcode(cbuf, Assembler::REX_B); | |
| 3129 } | |
| 3130 } else { | |
| 3131 if ($src$$reg < 8) { | |
| 3132 emit_opcode(cbuf, Assembler::REX_R); | |
| 3133 } else { | |
| 3134 emit_opcode(cbuf, Assembler::REX_RB); | |
| 3135 } | |
| 3136 } | |
| 3137 %} | |
| 3138 | |
| 3139 enc_class REX_reg_reg_wide(rRegI dst, rRegI src) | |
| 3140 %{ | |
| 3141 if ($dst$$reg < 8) { | |
| 3142 if ($src$$reg < 8) { | |
| 3143 emit_opcode(cbuf, Assembler::REX_W); | |
| 3144 } else { | |
| 3145 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3146 } | |
| 3147 } else { | |
| 3148 if ($src$$reg < 8) { | |
| 3149 emit_opcode(cbuf, Assembler::REX_WR); | |
| 3150 } else { | |
| 3151 emit_opcode(cbuf, Assembler::REX_WRB); | |
| 3152 } | |
| 3153 } | |
| 3154 %} | |
| 3155 | |
| 3156 enc_class REX_reg_mem(rRegI reg, memory mem) | |
| 3157 %{ | |
| 3158 if ($reg$$reg < 8) { | |
| 3159 if ($mem$$base < 8) { | |
| 3160 if ($mem$$index >= 8) { | |
| 3161 emit_opcode(cbuf, Assembler::REX_X); | |
| 3162 } | |
| 3163 } else { | |
| 3164 if ($mem$$index < 8) { | |
| 3165 emit_opcode(cbuf, Assembler::REX_B); | |
| 3166 } else { | |
| 3167 emit_opcode(cbuf, Assembler::REX_XB); | |
| 3168 } | |
| 3169 } | |
| 3170 } else { | |
| 3171 if ($mem$$base < 8) { | |
| 3172 if ($mem$$index < 8) { | |
| 3173 emit_opcode(cbuf, Assembler::REX_R); | |
| 3174 } else { | |
| 3175 emit_opcode(cbuf, Assembler::REX_RX); | |
| 3176 } | |
| 3177 } else { | |
| 3178 if ($mem$$index < 8) { | |
| 3179 emit_opcode(cbuf, Assembler::REX_RB); | |
| 3180 } else { | |
| 3181 emit_opcode(cbuf, Assembler::REX_RXB); | |
| 3182 } | |
| 3183 } | |
| 3184 } | |
| 3185 %} | |
| 3186 | |
| 3187 enc_class REX_reg_mem_wide(rRegL reg, memory mem) | |
| 3188 %{ | |
| 3189 if ($reg$$reg < 8) { | |
| 3190 if ($mem$$base < 8) { | |
| 3191 if ($mem$$index < 8) { | |
| 3192 emit_opcode(cbuf, Assembler::REX_W); | |
| 3193 } else { | |
| 3194 emit_opcode(cbuf, Assembler::REX_WX); | |
| 3195 } | |
| 3196 } else { | |
| 3197 if ($mem$$index < 8) { | |
| 3198 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3199 } else { | |
| 3200 emit_opcode(cbuf, Assembler::REX_WXB); | |
| 3201 } | |
| 3202 } | |
| 3203 } else { | |
| 3204 if ($mem$$base < 8) { | |
| 3205 if ($mem$$index < 8) { | |
| 3206 emit_opcode(cbuf, Assembler::REX_WR); | |
| 3207 } else { | |
| 3208 emit_opcode(cbuf, Assembler::REX_WRX); | |
| 3209 } | |
| 3210 } else { | |
| 3211 if ($mem$$index < 8) { | |
| 3212 emit_opcode(cbuf, Assembler::REX_WRB); | |
| 3213 } else { | |
| 3214 emit_opcode(cbuf, Assembler::REX_WRXB); | |
| 3215 } | |
| 3216 } | |
| 3217 } | |
| 3218 %} | |
| 3219 | |
| 3220 enc_class reg_mem(rRegI ereg, memory mem) | |
| 3221 %{ | |
| 3222 // High registers handle in encode_RegMem | |
| 3223 int reg = $ereg$$reg; | |
| 3224 int base = $mem$$base; | |
| 3225 int index = $mem$$index; | |
| 3226 int scale = $mem$$scale; | |
| 3227 int disp = $mem$$disp; | |
| 3228 bool disp_is_oop = $mem->disp_is_oop(); | |
| 3229 | |
| 3230 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop); | |
| 3231 %} | |
| 3232 | |
| 3233 enc_class RM_opc_mem(immI rm_opcode, memory mem) | |
| 3234 %{ | |
| 3235 int rm_byte_opcode = $rm_opcode$$constant; | |
| 3236 | |
| 3237 // High registers handle in encode_RegMem | |
| 3238 int base = $mem$$base; | |
| 3239 int index = $mem$$index; | |
| 3240 int scale = $mem$$scale; | |
| 3241 int displace = $mem$$disp; | |
| 3242 | |
| 3243 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when | |
| 3244 // working with static | |
| 3245 // globals | |
| 3246 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace, | |
| 3247 disp_is_oop); | |
| 3248 %} | |
| 3249 | |
| 3250 enc_class reg_lea(rRegI dst, rRegI src0, immI src1) | |
| 3251 %{ | |
| 3252 int reg_encoding = $dst$$reg; | |
| 3253 int base = $src0$$reg; // 0xFFFFFFFF indicates no base | |
| 3254 int index = 0x04; // 0x04 indicates no index | |
| 3255 int scale = 0x00; // 0x00 indicates no scale | |
| 3256 int displace = $src1$$constant; // 0x00 indicates no displacement | |
| 3257 bool disp_is_oop = false; | |
| 3258 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace, | |
| 3259 disp_is_oop); | |
| 3260 %} | |
| 3261 | |
| 3262 enc_class neg_reg(rRegI dst) | |
| 3263 %{ | |
| 3264 int dstenc = $dst$$reg; | |
| 3265 if (dstenc >= 8) { | |
| 3266 emit_opcode(cbuf, Assembler::REX_B); | |
| 3267 dstenc -= 8; | |
| 3268 } | |
| 3269 // NEG $dst | |
| 3270 emit_opcode(cbuf, 0xF7); | |
| 3271 emit_rm(cbuf, 0x3, 0x03, dstenc); | |
| 3272 %} | |
| 3273 | |
| 3274 enc_class neg_reg_wide(rRegI dst) | |
| 3275 %{ | |
| 3276 int dstenc = $dst$$reg; | |
| 3277 if (dstenc < 8) { | |
| 3278 emit_opcode(cbuf, Assembler::REX_W); | |
| 3279 } else { | |
| 3280 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3281 dstenc -= 8; | |
| 3282 } | |
| 3283 // NEG $dst | |
| 3284 emit_opcode(cbuf, 0xF7); | |
| 3285 emit_rm(cbuf, 0x3, 0x03, dstenc); | |
| 3286 %} | |
| 3287 | |
| 3288 enc_class setLT_reg(rRegI dst) | |
| 3289 %{ | |
| 3290 int dstenc = $dst$$reg; | |
| 3291 if (dstenc >= 8) { | |
| 3292 emit_opcode(cbuf, Assembler::REX_B); | |
| 3293 dstenc -= 8; | |
| 3294 } else if (dstenc >= 4) { | |
| 3295 emit_opcode(cbuf, Assembler::REX); | |
| 3296 } | |
| 3297 // SETLT $dst | |
| 3298 emit_opcode(cbuf, 0x0F); | |
| 3299 emit_opcode(cbuf, 0x9C); | |
| 3300 emit_rm(cbuf, 0x3, 0x0, dstenc); | |
| 3301 %} | |
| 3302 | |
| 3303 enc_class setNZ_reg(rRegI dst) | |
| 3304 %{ | |
| 3305 int dstenc = $dst$$reg; | |
| 3306 if (dstenc >= 8) { | |
| 3307 emit_opcode(cbuf, Assembler::REX_B); | |
| 3308 dstenc -= 8; | |
| 3309 } else if (dstenc >= 4) { | |
| 3310 emit_opcode(cbuf, Assembler::REX); | |
| 3311 } | |
| 3312 // SETNZ $dst | |
| 3313 emit_opcode(cbuf, 0x0F); | |
| 3314 emit_opcode(cbuf, 0x95); | |
| 3315 emit_rm(cbuf, 0x3, 0x0, dstenc); | |
| 3316 %} | |
| 3317 | |
| 3318 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y, | |
| 3319 rcx_RegI tmp) | |
| 3320 %{ | |
| 3321 // cadd_cmpLT | |
| 3322 | |
| 3323 int tmpReg = $tmp$$reg; | |
| 3324 | |
| 3325 int penc = $p$$reg; | |
| 3326 int qenc = $q$$reg; | |
| 3327 int yenc = $y$$reg; | |
| 3328 | |
| 3329 // subl $p,$q | |
| 3330 if (penc < 8) { | |
| 3331 if (qenc >= 8) { | |
| 3332 emit_opcode(cbuf, Assembler::REX_B); | |
| 3333 } | |
| 3334 } else { | |
| 3335 if (qenc < 8) { | |
| 3336 emit_opcode(cbuf, Assembler::REX_R); | |
| 3337 } else { | |
| 3338 emit_opcode(cbuf, Assembler::REX_RB); | |
| 3339 } | |
| 3340 } | |
| 3341 emit_opcode(cbuf, 0x2B); | |
| 3342 emit_rm(cbuf, 0x3, penc & 7, qenc & 7); | |
| 3343 | |
| 3344 // sbbl $tmp, $tmp | |
| 3345 emit_opcode(cbuf, 0x1B); | |
| 3346 emit_rm(cbuf, 0x3, tmpReg, tmpReg); | |
| 3347 | |
| 3348 // andl $tmp, $y | |
| 3349 if (yenc >= 8) { | |
| 3350 emit_opcode(cbuf, Assembler::REX_B); | |
| 3351 } | |
| 3352 emit_opcode(cbuf, 0x23); | |
| 3353 emit_rm(cbuf, 0x3, tmpReg, yenc & 7); | |
| 3354 | |
| 3355 // addl $p,$tmp | |
| 3356 if (penc >= 8) { | |
| 3357 emit_opcode(cbuf, Assembler::REX_R); | |
| 3358 } | |
| 3359 emit_opcode(cbuf, 0x03); | |
| 3360 emit_rm(cbuf, 0x3, penc & 7, tmpReg); | |
| 3361 %} | |
| 3362 | |
| 3363 // Compare the lonogs and set -1, 0, or 1 into dst | |
| 3364 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst) | |
| 3365 %{ | |
| 3366 int src1enc = $src1$$reg; | |
| 3367 int src2enc = $src2$$reg; | |
| 3368 int dstenc = $dst$$reg; | |
| 3369 | |
| 3370 // cmpq $src1, $src2 | |
| 3371 if (src1enc < 8) { | |
| 3372 if (src2enc < 8) { | |
| 3373 emit_opcode(cbuf, Assembler::REX_W); | |
| 3374 } else { | |
| 3375 emit_opcode(cbuf, Assembler::REX_WB); | |
| 3376 } | |
| 3377 } else { | |
| 3378 if (src2enc < 8) { | |
| 3379 emit_opcode(cbuf, Assembler::REX_WR); | |
| 3380 } else { | |
| 3381 emit_opcode(cbuf, Assembler::REX_WRB); | |
| 3382 } | |
| 3383 } | |
| 3384 emit_opcode(cbuf, 0x3B); | |
| 3385 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7); | |
| 3386 | |
| 3387 // movl $dst, -1 | |
| 3388 if (dstenc >= 8) { | |
| 3389 emit_opcode(cbuf, Assembler::REX_B); | |
| 3390 } | |
| 3391 emit_opcode(cbuf, 0xB8 | (dstenc & 7)); | |
| 3392 emit_d32(cbuf, -1); | |
| 3393 | |
| 3394 // jl,s done | |
| 3395 emit_opcode(cbuf, 0x7C); | |
| 3396 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08); | |
| 3397 | |
| 3398 // setne $dst | |
| 3399 if (dstenc >= 4) { | |
| 3400 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B); | |
| 3401 } | |
| 3402 emit_opcode(cbuf, 0x0F); | |
| 3403 emit_opcode(cbuf, 0x95); | |
| 3404 emit_opcode(cbuf, 0xC0 | (dstenc & 7)); | |
| 3405 | |
| 3406 // movzbl $dst, $dst | |
| 3407 if (dstenc >= 4) { | |
| 3408 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB); | |
| 3409 } | |
| 3410 emit_opcode(cbuf, 0x0F); | |
| 3411 emit_opcode(cbuf, 0xB6); | |
| 3412 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7); | |
| 3413 %} | |
| 3414 | |
| 3415 enc_class Push_ResultXD(regD dst) %{ | |
| 3416 int dstenc = $dst$$reg; | |
| 3417 | |
| 3418 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP] | |
| 3419 | |
| 3420 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp] | |
| 3421 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66); | |
| 3422 if (dstenc >= 8) { | |
| 3423 emit_opcode(cbuf, Assembler::REX_R); | |
| 3424 } | |
| 3425 emit_opcode (cbuf, 0x0F ); | |
| 3426 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 ); | |
| 3427 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false); | |
| 3428 | |
| 3429 // add rsp,8 | |
| 3430 emit_opcode(cbuf, Assembler::REX_W); | |
| 3431 emit_opcode(cbuf,0x83); | |
| 3432 emit_rm(cbuf,0x3, 0x0, RSP_enc); | |
| 3433 emit_d8(cbuf,0x08); | |
| 3434 %} | |
| 3435 | |
| 3436 enc_class Push_SrcXD(regD src) %{ | |
| 3437 int srcenc = $src$$reg; | |
| 3438 | |
| 3439 // subq rsp,#8 | |
| 3440 emit_opcode(cbuf, Assembler::REX_W); | |
| 3441 emit_opcode(cbuf, 0x83); | |
| 3442 emit_rm(cbuf, 0x3, 0x5, RSP_enc); | |
| 3443 emit_d8(cbuf, 0x8); | |
| 3444 | |
| 3445 // movsd [rsp],src | |
| 3446 emit_opcode(cbuf, 0xF2); | |
| 3447 if (srcenc >= 8) { | |
| 3448 emit_opcode(cbuf, Assembler::REX_R); | |
| 3449 } | |
| 3450 emit_opcode(cbuf, 0x0F); | |
| 3451 emit_opcode(cbuf, 0x11); | |
| 3452 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); | |
| 3453 | |
| 3454 // fldd [rsp] | |
| 3455 emit_opcode(cbuf, 0x66); | |
| 3456 emit_opcode(cbuf, 0xDD); | |
| 3457 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false); | |
| 3458 %} | |
| 3459 | |
| 3460 | |
| 3461 enc_class movq_ld(regD dst, memory mem) %{ | |
| 3462 MacroAssembler _masm(&cbuf); | |
| 624 | 3463 __ movq($dst$$XMMRegister, $mem$$Address); |
| 0 | 3464 %} |
| 3465 | |
| 3466 enc_class movq_st(memory mem, regD src) %{ | |
| 3467 MacroAssembler _masm(&cbuf); | |
| 624 | 3468 __ movq($mem$$Address, $src$$XMMRegister); |
| 0 | 3469 %} |
| 3470 | |
| 3471 enc_class pshufd_8x8(regF dst, regF src) %{ | |
| 3472 MacroAssembler _masm(&cbuf); | |
| 3473 | |
| 3474 encode_CopyXD(cbuf, $dst$$reg, $src$$reg); | |
| 3475 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg)); | |
| 3476 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00); | |
| 3477 %} | |
| 3478 | |
| 3479 enc_class pshufd_4x16(regF dst, regF src) %{ | |
| 3480 MacroAssembler _masm(&cbuf); | |
| 3481 | |
| 3482 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00); | |
| 3483 %} | |
| 3484 | |
| 3485 enc_class pshufd(regD dst, regD src, int mode) %{ | |
| 3486 MacroAssembler _masm(&cbuf); | |
| 3487 | |
| 3488 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode); | |
| 3489 %} | |
| 3490 | |
| 3491 enc_class pxor(regD dst, regD src) %{ | |
| 3492 MacroAssembler _masm(&cbuf); | |
| 3493 | |
| 3494 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg)); | |
| 3495 %} | |
| 3496 | |
| 3497 enc_class mov_i2x(regD dst, rRegI src) %{ | |
| 3498 MacroAssembler _masm(&cbuf); | |
| 3499 | |
| 3500 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg)); | |
| 3501 %} | |
| 3502 | |
| 3503 // obj: object to lock | |
| 3504 // box: box address (header location) -- killed | |
| 3505 // tmp: rax -- killed | |
| 3506 // scr: rbx -- killed | |
| 3507 // | |
| 3508 // What follows is a direct transliteration of fast_lock() and fast_unlock() | |
| 3509 // from i486.ad. See that file for comments. | |
| 3510 // TODO: where possible switch from movq (r, 0) to movl(r,0) and | |
| 3511 // use the shorter encoding. (Movl clears the high-order 32-bits). | |
| 3512 | |
| 3513 | |
| 3514 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr) | |
| 3515 %{ | |
| 3516 Register objReg = as_Register((int)$obj$$reg); | |
| 3517 Register boxReg = as_Register((int)$box$$reg); | |
| 3518 Register tmpReg = as_Register($tmp$$reg); | |
| 3519 Register scrReg = as_Register($scr$$reg); | |
| 3520 MacroAssembler masm(&cbuf); | |
| 3521 | |
| 3522 // Verify uniqueness of register assignments -- necessary but not sufficient | |
| 3523 assert (objReg != boxReg && objReg != tmpReg && | |
| 3524 objReg != scrReg && tmpReg != scrReg, "invariant") ; | |
| 3525 | |
| 3526 if (_counters != NULL) { | |
| 3527 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr())); | |
| 3528 } | |
| 3529 if (EmitSync & 1) { | |
| 304 | 3530 // Without cast to int32_t a movptr will destroy r10 which is typically obj |
| 3531 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; | |
| 3532 masm.cmpptr(rsp, (int32_t)NULL_WORD) ; | |
| 0 | 3533 } else |
| 3534 if (EmitSync & 2) { | |
| 3535 Label DONE_LABEL; | |
| 3536 if (UseBiasedLocking) { | |
| 3537 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument. | |
| 3538 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters); | |
| 3539 } | |
| 304 | 3540 // QQQ was movl... |
| 3541 masm.movptr(tmpReg, 0x1); | |
| 3542 masm.orptr(tmpReg, Address(objReg, 0)); | |
| 3543 masm.movptr(Address(boxReg, 0), tmpReg); | |
| 0 | 3544 if (os::is_MP()) { |
| 3545 masm.lock(); | |
| 3546 } | |
| 304 | 3547 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg |
| 0 | 3548 masm.jcc(Assembler::equal, DONE_LABEL); |
| 3549 | |
| 3550 // Recursive locking | |
| 304 | 3551 masm.subptr(tmpReg, rsp); |
| 3552 masm.andptr(tmpReg, 7 - os::vm_page_size()); | |
| 3553 masm.movptr(Address(boxReg, 0), tmpReg); | |
| 0 | 3554 |
| 3555 masm.bind(DONE_LABEL); | |
| 3556 masm.nop(); // avoid branch to branch | |
| 3557 } else { | |
| 3558 Label DONE_LABEL, IsInflated, Egress; | |
| 3559 | |
| 304 | 3560 masm.movptr(tmpReg, Address(objReg, 0)) ; |
| 3561 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased | |
| 3562 masm.jcc (Assembler::notZero, IsInflated) ; | |
| 3563 | |
| 0 | 3564 // it's stack-locked, biased or neutral |
| 3565 // TODO: optimize markword triage order to reduce the number of | |
| 3566 // conditional branches in the most common cases. | |
| 3567 // Beware -- there's a subtle invariant that fetch of the markword | |
| 3568 // at [FETCH], below, will never observe a biased encoding (*101b). | |
| 3569 // If this invariant is not held we'll suffer exclusion (safety) failure. | |
| 3570 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3571 if (UseBiasedLocking && !UseOptoBiasInlining) { |
| 0 | 3572 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters); |
| 304 | 3573 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH] |
| 0 | 3574 } |
| 3575 | |
| 304 | 3576 // was q will it destroy high? |
| 3577 masm.orl (tmpReg, 1) ; | |
| 3578 masm.movptr(Address(boxReg, 0), tmpReg) ; | |
| 3579 if (os::is_MP()) { masm.lock(); } | |
| 3580 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg | |
| 0 | 3581 if (_counters != NULL) { |
| 3582 masm.cond_inc32(Assembler::equal, | |
| 3583 ExternalAddress((address) _counters->fast_path_entry_count_addr())); | |
| 3584 } | |
| 3585 masm.jcc (Assembler::equal, DONE_LABEL); | |
| 3586 | |
| 3587 // Recursive locking | |
| 304 | 3588 masm.subptr(tmpReg, rsp); |
| 3589 masm.andptr(tmpReg, 7 - os::vm_page_size()); | |
| 3590 masm.movptr(Address(boxReg, 0), tmpReg); | |
| 0 | 3591 if (_counters != NULL) { |
| 3592 masm.cond_inc32(Assembler::equal, | |
| 3593 ExternalAddress((address) _counters->fast_path_entry_count_addr())); | |
| 3594 } | |
| 3595 masm.jmp (DONE_LABEL) ; | |
| 3596 | |
| 3597 masm.bind (IsInflated) ; | |
| 3598 // It's inflated | |
| 3599 | |
| 3600 // TODO: someday avoid the ST-before-CAS penalty by | |
| 3601 // relocating (deferring) the following ST. | |
| 3602 // We should also think about trying a CAS without having | |
| 3603 // fetched _owner. If the CAS is successful we may | |
| 3604 // avoid an RTO->RTS upgrade on the $line. | |
| 304 | 3605 // Without cast to int32_t a movptr will destroy r10 which is typically obj |
| 3606 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ; | |
| 3607 | |
| 3608 masm.mov (boxReg, tmpReg) ; | |
| 3609 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; | |
| 3610 masm.testptr(tmpReg, tmpReg) ; | |
| 3611 masm.jcc (Assembler::notZero, DONE_LABEL) ; | |
| 0 | 3612 |
| 3613 // It's inflated and appears unlocked | |
| 304 | 3614 if (os::is_MP()) { masm.lock(); } |
| 3615 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; | |
| 0 | 3616 // Intentional fall-through into DONE_LABEL ... |
| 3617 | |
| 3618 masm.bind (DONE_LABEL) ; | |
| 3619 masm.nop () ; // avoid jmp to jmp | |
| 3620 } | |
| 3621 %} | |
| 3622 | |
| 3623 // obj: object to unlock | |
| 3624 // box: box address (displaced header location), killed | |
| 3625 // RBX: killed tmp; cannot be obj nor box | |
| 3626 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp) | |
| 3627 %{ | |
| 3628 | |
| 3629 Register objReg = as_Register($obj$$reg); | |
| 3630 Register boxReg = as_Register($box$$reg); | |
| 3631 Register tmpReg = as_Register($tmp$$reg); | |
| 3632 MacroAssembler masm(&cbuf); | |
| 3633 | |
| 304 | 3634 if (EmitSync & 4) { |
| 3635 masm.cmpptr(rsp, 0) ; | |
| 0 | 3636 } else |
| 3637 if (EmitSync & 8) { | |
| 3638 Label DONE_LABEL; | |
| 3639 if (UseBiasedLocking) { | |
| 3640 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL); | |
| 3641 } | |
| 3642 | |
| 3643 // Check whether the displaced header is 0 | |
| 3644 //(=> recursive unlock) | |
| 304 | 3645 masm.movptr(tmpReg, Address(boxReg, 0)); |
| 3646 masm.testptr(tmpReg, tmpReg); | |
| 0 | 3647 masm.jcc(Assembler::zero, DONE_LABEL); |
| 3648 | |
| 3649 // If not recursive lock, reset the header to displaced header | |
| 3650 if (os::is_MP()) { | |
| 3651 masm.lock(); | |
| 3652 } | |
| 304 | 3653 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box |
| 0 | 3654 masm.bind(DONE_LABEL); |
| 3655 masm.nop(); // avoid branch to branch | |
| 3656 } else { | |
| 3657 Label DONE_LABEL, Stacked, CheckSucc ; | |
| 3658 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3659 if (UseBiasedLocking && !UseOptoBiasInlining) { |
| 0 | 3660 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL); |
| 3661 } | |
| 304 | 3662 |
| 3663 masm.movptr(tmpReg, Address(objReg, 0)) ; | |
| 3664 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ; | |
| 3665 masm.jcc (Assembler::zero, DONE_LABEL) ; | |
| 3666 masm.testl (tmpReg, 0x02) ; | |
| 3667 masm.jcc (Assembler::zero, Stacked) ; | |
| 3668 | |
| 0 | 3669 // It's inflated |
| 304 | 3670 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ; |
| 3671 masm.xorptr(boxReg, r15_thread) ; | |
| 3672 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ; | |
| 3673 masm.jcc (Assembler::notZero, DONE_LABEL) ; | |
| 3674 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ; | |
| 3675 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ; | |
| 3676 masm.jcc (Assembler::notZero, CheckSucc) ; | |
| 3677 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ; | |
| 3678 masm.jmp (DONE_LABEL) ; | |
| 3679 | |
| 3680 if ((EmitSync & 65536) == 0) { | |
| 0 | 3681 Label LSuccess, LGoSlowPath ; |
| 3682 masm.bind (CheckSucc) ; | |
| 304 | 3683 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ; |
| 0 | 3684 masm.jcc (Assembler::zero, LGoSlowPath) ; |
| 3685 | |
| 3686 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the | |
| 3687 // the explicit ST;MEMBAR combination, but masm doesn't currently support | |
| 3688 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc | |
| 3689 // are all faster when the write buffer is populated. | |
| 304 | 3690 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ; |
| 0 | 3691 if (os::is_MP()) { |
| 304 | 3692 masm.lock () ; masm.addl (Address(rsp, 0), 0) ; |
| 0 | 3693 } |
| 304 | 3694 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ; |
| 0 | 3695 masm.jcc (Assembler::notZero, LSuccess) ; |
| 3696 | |
| 304 | 3697 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX |
| 0 | 3698 if (os::is_MP()) { masm.lock(); } |
| 304 | 3699 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)); |
| 0 | 3700 masm.jcc (Assembler::notEqual, LSuccess) ; |
| 3701 // Intentional fall-through into slow-path | |
| 3702 | |
| 3703 masm.bind (LGoSlowPath) ; | |
| 3704 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure | |
| 3705 masm.jmp (DONE_LABEL) ; | |
| 3706 | |
| 3707 masm.bind (LSuccess) ; | |
| 3708 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success | |
| 3709 masm.jmp (DONE_LABEL) ; | |
| 3710 } | |
| 3711 | |
| 304 | 3712 masm.bind (Stacked) ; |
| 3713 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch | |
| 3714 if (os::is_MP()) { masm.lock(); } | |
| 3715 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box | |
| 0 | 3716 |
| 3717 if (EmitSync & 65536) { | |
| 3718 masm.bind (CheckSucc) ; | |
| 3719 } | |
| 3720 masm.bind(DONE_LABEL); | |
| 3721 if (EmitSync & 32768) { | |
| 3722 masm.nop(); // avoid branch to branch | |
| 3723 } | |
| 3724 } | |
| 3725 %} | |
| 3726 | |
| 3727 enc_class enc_String_Compare() | |
| 3728 %{ | |
| 3729 Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL, | |
| 3730 POP_LABEL, DONE_LABEL, CONT_LABEL, | |
| 3731 WHILE_HEAD_LABEL; | |
| 3732 MacroAssembler masm(&cbuf); | |
| 3733 | |
| 3734 // Get the first character position in both strings | |
| 3735 // [8] char array, [12] offset, [16] count | |
| 3736 int value_offset = java_lang_String::value_offset_in_bytes(); | |
| 3737 int offset_offset = java_lang_String::offset_offset_in_bytes(); | |
| 3738 int count_offset = java_lang_String::count_offset_in_bytes(); | |
| 3739 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR); | |
| 3740 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
3741 masm.load_heap_oop(rax, Address(rsi, value_offset)); |
| 0 | 3742 masm.movl(rcx, Address(rsi, offset_offset)); |
| 304 | 3743 masm.lea(rax, Address(rax, rcx, Address::times_2, base_offset)); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
3744 masm.load_heap_oop(rbx, Address(rdi, value_offset)); |
| 0 | 3745 masm.movl(rcx, Address(rdi, offset_offset)); |
| 304 | 3746 masm.lea(rbx, Address(rbx, rcx, Address::times_2, base_offset)); |
| 0 | 3747 |
| 3748 // Compute the minimum of the string lengths(rsi) and the | |
| 3749 // difference of the string lengths (stack) | |
| 3750 | |
| 3751 masm.movl(rdi, Address(rdi, count_offset)); | |
| 3752 masm.movl(rsi, Address(rsi, count_offset)); | |
| 3753 masm.movl(rcx, rdi); | |
| 3754 masm.subl(rdi, rsi); | |
| 304 | 3755 masm.push(rdi); |
| 3756 masm.cmov(Assembler::lessEqual, rsi, rcx); | |
| 0 | 3757 |
| 3758 // Is the minimum length zero? | |
| 3759 masm.bind(RCX_GOOD_LABEL); | |
| 3760 masm.testl(rsi, rsi); | |
| 3761 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL); | |
| 3762 | |
| 3763 // Load first characters | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3764 masm.load_unsigned_short(rcx, Address(rbx, 0)); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3765 masm.load_unsigned_short(rdi, Address(rax, 0)); |
| 0 | 3766 |
| 3767 // Compare first characters | |
| 3768 masm.subl(rcx, rdi); | |
| 3769 masm.jcc(Assembler::notZero, POP_LABEL); | |
| 3770 masm.decrementl(rsi); | |
| 3771 masm.jcc(Assembler::zero, LENGTH_DIFF_LABEL); | |
| 3772 | |
| 3773 { | |
| 3774 // Check after comparing first character to see if strings are equivalent | |
| 3775 Label LSkip2; | |
| 3776 // Check if the strings start at same location | |
| 304 | 3777 masm.cmpptr(rbx, rax); |
| 0 | 3778 masm.jcc(Assembler::notEqual, LSkip2); |
| 3779 | |
| 3780 // Check if the length difference is zero (from stack) | |
| 3781 masm.cmpl(Address(rsp, 0), 0x0); | |
| 3782 masm.jcc(Assembler::equal, LENGTH_DIFF_LABEL); | |
| 3783 | |
| 3784 // Strings might not be equivalent | |
| 3785 masm.bind(LSkip2); | |
| 3786 } | |
| 3787 | |
| 3788 // Shift RAX and RBX to the end of the arrays, negate min | |
| 304 | 3789 masm.lea(rax, Address(rax, rsi, Address::times_2, 2)); |
| 3790 masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2)); | |
| 3791 masm.negptr(rsi); | |
| 0 | 3792 |
| 3793 // Compare the rest of the characters | |
| 3794 masm.bind(WHILE_HEAD_LABEL); | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3795 masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0)); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3796 masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0)); |
| 0 | 3797 masm.subl(rcx, rdi); |
| 3798 masm.jcc(Assembler::notZero, POP_LABEL); | |
| 304 | 3799 masm.increment(rsi); |
| 0 | 3800 masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL); |
| 3801 | |
| 3802 // Strings are equal up to min length. Return the length difference. | |
| 3803 masm.bind(LENGTH_DIFF_LABEL); | |
| 304 | 3804 masm.pop(rcx); |
| 0 | 3805 masm.jmp(DONE_LABEL); |
| 3806 | |
| 3807 // Discard the stored length difference | |
| 3808 masm.bind(POP_LABEL); | |
| 304 | 3809 masm.addptr(rsp, 8); |
| 3810 | |
| 0 | 3811 // That's it |
| 3812 masm.bind(DONE_LABEL); | |
| 3813 %} | |
| 3814 | |
|
169
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3815 enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, rbx_RegI tmp2, rcx_RegI result) %{ |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3816 Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP; |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3817 MacroAssembler masm(&cbuf); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3818 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3819 Register ary1Reg = as_Register($ary1$$reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3820 Register ary2Reg = as_Register($ary2$$reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3821 Register tmp1Reg = as_Register($tmp1$$reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3822 Register tmp2Reg = as_Register($tmp2$$reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3823 Register resultReg = as_Register($result$$reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3824 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3825 int length_offset = arrayOopDesc::length_offset_in_bytes(); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3826 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3827 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3828 // Check the input args |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3829 masm.cmpq(ary1Reg, ary2Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3830 masm.jcc(Assembler::equal, TRUE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3831 masm.testq(ary1Reg, ary1Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3832 masm.jcc(Assembler::zero, FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3833 masm.testq(ary2Reg, ary2Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3834 masm.jcc(Assembler::zero, FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3835 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3836 // Check the lengths |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3837 masm.movl(tmp2Reg, Address(ary1Reg, length_offset)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3838 masm.movl(resultReg, Address(ary2Reg, length_offset)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3839 masm.cmpl(tmp2Reg, resultReg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3840 masm.jcc(Assembler::notEqual, FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3841 masm.testl(resultReg, resultReg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3842 masm.jcc(Assembler::zero, TRUE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3843 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3844 // Get the number of 4 byte vectors to compare |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3845 masm.shrl(resultReg, 1); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3846 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3847 // Check for odd-length arrays |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3848 masm.andl(tmp2Reg, 1); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3849 masm.testl(tmp2Reg, tmp2Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3850 masm.jcc(Assembler::zero, COMPARE_LOOP_HDR); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3851 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3852 // Compare 2-byte "tail" at end of arrays |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3853 masm.load_unsigned_short(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset)); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
3854 masm.load_unsigned_short(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset)); |
|
169
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3855 masm.cmpl(tmp1Reg, tmp2Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3856 masm.jcc(Assembler::notEqual, FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3857 masm.testl(resultReg, resultReg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3858 masm.jcc(Assembler::zero, TRUE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3859 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3860 // Setup compare loop |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3861 masm.bind(COMPARE_LOOP_HDR); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3862 // Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3863 masm.leaq(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3864 masm.leaq(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3865 masm.negq(resultReg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3866 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3867 // 4-byte-wide compare loop |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3868 masm.bind(COMPARE_LOOP); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3869 masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3870 masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3871 masm.cmpl(ary1Reg, ary2Reg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3872 masm.jcc(Assembler::notEqual, FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3873 masm.incrementq(resultReg); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3874 masm.jcc(Assembler::notZero, COMPARE_LOOP); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3875 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3876 masm.bind(TRUE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3877 masm.movl(resultReg, 1); // return true |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3878 masm.jmp(DONE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3879 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3880 masm.bind(FALSE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3881 masm.xorl(resultReg, resultReg); // return false |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3882 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3883 // That's it |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3884 masm.bind(DONE_LABEL); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3885 %} |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
3886 |
| 0 | 3887 enc_class enc_rethrow() |
| 3888 %{ | |
| 3889 cbuf.set_inst_mark(); | |
| 3890 emit_opcode(cbuf, 0xE9); // jmp entry | |
| 3891 emit_d32_reloc(cbuf, | |
| 3892 (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4), | |
| 3893 runtime_call_Relocation::spec(), | |
| 3894 RELOC_DISP32); | |
| 3895 %} | |
| 3896 | |
| 3897 enc_class absF_encoding(regF dst) | |
| 3898 %{ | |
| 3899 int dstenc = $dst$$reg; | |
| 304 | 3900 address signmask_address = (address) StubRoutines::x86::float_sign_mask(); |
| 0 | 3901 |
| 3902 cbuf.set_inst_mark(); | |
| 3903 if (dstenc >= 8) { | |
| 3904 emit_opcode(cbuf, Assembler::REX_R); | |
| 3905 dstenc -= 8; | |
| 3906 } | |
| 3907 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 3908 emit_opcode(cbuf, 0x0F); | |
| 3909 emit_opcode(cbuf, 0x54); | |
| 3910 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101 | |
| 3911 emit_d32_reloc(cbuf, signmask_address); | |
| 3912 %} | |
| 3913 | |
| 3914 enc_class absD_encoding(regD dst) | |
| 3915 %{ | |
| 3916 int dstenc = $dst$$reg; | |
| 304 | 3917 address signmask_address = (address) StubRoutines::x86::double_sign_mask(); |
| 0 | 3918 |
| 3919 cbuf.set_inst_mark(); | |
| 3920 emit_opcode(cbuf, 0x66); | |
| 3921 if (dstenc >= 8) { | |
| 3922 emit_opcode(cbuf, Assembler::REX_R); | |
| 3923 dstenc -= 8; | |
| 3924 } | |
| 3925 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 3926 emit_opcode(cbuf, 0x0F); | |
| 3927 emit_opcode(cbuf, 0x54); | |
| 3928 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101 | |
| 3929 emit_d32_reloc(cbuf, signmask_address); | |
| 3930 %} | |
| 3931 | |
| 3932 enc_class negF_encoding(regF dst) | |
| 3933 %{ | |
| 3934 int dstenc = $dst$$reg; | |
| 304 | 3935 address signflip_address = (address) StubRoutines::x86::float_sign_flip(); |
| 0 | 3936 |
| 3937 cbuf.set_inst_mark(); | |
| 3938 if (dstenc >= 8) { | |
| 3939 emit_opcode(cbuf, Assembler::REX_R); | |
| 3940 dstenc -= 8; | |
| 3941 } | |
| 3942 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 3943 emit_opcode(cbuf, 0x0F); | |
| 3944 emit_opcode(cbuf, 0x57); | |
| 3945 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101 | |
| 3946 emit_d32_reloc(cbuf, signflip_address); | |
| 3947 %} | |
| 3948 | |
| 3949 enc_class negD_encoding(regD dst) | |
| 3950 %{ | |
| 3951 int dstenc = $dst$$reg; | |
| 304 | 3952 address signflip_address = (address) StubRoutines::x86::double_sign_flip(); |
| 0 | 3953 |
| 3954 cbuf.set_inst_mark(); | |
| 3955 emit_opcode(cbuf, 0x66); | |
| 3956 if (dstenc >= 8) { | |
| 3957 emit_opcode(cbuf, Assembler::REX_R); | |
| 3958 dstenc -= 8; | |
| 3959 } | |
| 3960 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 3961 emit_opcode(cbuf, 0x0F); | |
| 3962 emit_opcode(cbuf, 0x57); | |
| 3963 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101 | |
| 3964 emit_d32_reloc(cbuf, signflip_address); | |
| 3965 %} | |
| 3966 | |
| 3967 enc_class f2i_fixup(rRegI dst, regF src) | |
| 3968 %{ | |
| 3969 int dstenc = $dst$$reg; | |
| 3970 int srcenc = $src$$reg; | |
| 3971 | |
| 3972 // cmpl $dst, #0x80000000 | |
| 3973 if (dstenc >= 8) { | |
| 3974 emit_opcode(cbuf, Assembler::REX_B); | |
| 3975 } | |
| 3976 emit_opcode(cbuf, 0x81); | |
| 3977 emit_rm(cbuf, 0x3, 0x7, dstenc & 7); | |
| 3978 emit_d32(cbuf, 0x80000000); | |
| 3979 | |
| 3980 // jne,s done | |
| 3981 emit_opcode(cbuf, 0x75); | |
| 3982 if (srcenc < 8 && dstenc < 8) { | |
| 3983 emit_d8(cbuf, 0xF); | |
| 3984 } else if (srcenc >= 8 && dstenc >= 8) { | |
| 3985 emit_d8(cbuf, 0x11); | |
| 3986 } else { | |
| 3987 emit_d8(cbuf, 0x10); | |
| 3988 } | |
| 3989 | |
| 3990 // subq rsp, #8 | |
| 3991 emit_opcode(cbuf, Assembler::REX_W); | |
| 3992 emit_opcode(cbuf, 0x83); | |
| 3993 emit_rm(cbuf, 0x3, 0x5, RSP_enc); | |
| 3994 emit_d8(cbuf, 8); | |
| 3995 | |
| 3996 // movss [rsp], $src | |
| 3997 emit_opcode(cbuf, 0xF3); | |
| 3998 if (srcenc >= 8) { | |
| 3999 emit_opcode(cbuf, Assembler::REX_R); | |
| 4000 } | |
| 4001 emit_opcode(cbuf, 0x0F); | |
| 4002 emit_opcode(cbuf, 0x11); | |
| 4003 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes | |
| 4004 | |
| 4005 // call f2i_fixup | |
| 4006 cbuf.set_inst_mark(); | |
| 4007 emit_opcode(cbuf, 0xE8); | |
| 4008 emit_d32_reloc(cbuf, | |
| 4009 (int) | |
| 304 | 4010 (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4), |
| 0 | 4011 runtime_call_Relocation::spec(), |
| 4012 RELOC_DISP32); | |
| 4013 | |
| 4014 // popq $dst | |
| 4015 if (dstenc >= 8) { | |
| 4016 emit_opcode(cbuf, Assembler::REX_B); | |
| 4017 } | |
| 4018 emit_opcode(cbuf, 0x58 | (dstenc & 7)); | |
| 4019 | |
| 4020 // done: | |
| 4021 %} | |
| 4022 | |
| 4023 enc_class f2l_fixup(rRegL dst, regF src) | |
| 4024 %{ | |
| 4025 int dstenc = $dst$$reg; | |
| 4026 int srcenc = $src$$reg; | |
| 304 | 4027 address const_address = (address) StubRoutines::x86::double_sign_flip(); |
| 0 | 4028 |
| 4029 // cmpq $dst, [0x8000000000000000] | |
| 4030 cbuf.set_inst_mark(); | |
| 4031 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR); | |
| 4032 emit_opcode(cbuf, 0x39); | |
| 4033 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 4034 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101 | |
| 4035 emit_d32_reloc(cbuf, const_address); | |
| 4036 | |
| 4037 | |
| 4038 // jne,s done | |
| 4039 emit_opcode(cbuf, 0x75); | |
| 4040 if (srcenc < 8 && dstenc < 8) { | |
| 4041 emit_d8(cbuf, 0xF); | |
| 4042 } else if (srcenc >= 8 && dstenc >= 8) { | |
| 4043 emit_d8(cbuf, 0x11); | |
| 4044 } else { | |
| 4045 emit_d8(cbuf, 0x10); | |
| 4046 } | |
| 4047 | |
| 4048 // subq rsp, #8 | |
| 4049 emit_opcode(cbuf, Assembler::REX_W); | |
| 4050 emit_opcode(cbuf, 0x83); | |
| 4051 emit_rm(cbuf, 0x3, 0x5, RSP_enc); | |
| 4052 emit_d8(cbuf, 8); | |
| 4053 | |
| 4054 // movss [rsp], $src | |
| 4055 emit_opcode(cbuf, 0xF3); | |
| 4056 if (srcenc >= 8) { | |
| 4057 emit_opcode(cbuf, Assembler::REX_R); | |
| 4058 } | |
| 4059 emit_opcode(cbuf, 0x0F); | |
| 4060 emit_opcode(cbuf, 0x11); | |
| 4061 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes | |
| 4062 | |
| 4063 // call f2l_fixup | |
| 4064 cbuf.set_inst_mark(); | |
| 4065 emit_opcode(cbuf, 0xE8); | |
| 4066 emit_d32_reloc(cbuf, | |
| 4067 (int) | |
| 304 | 4068 (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4), |
| 0 | 4069 runtime_call_Relocation::spec(), |
| 4070 RELOC_DISP32); | |
| 4071 | |
| 4072 // popq $dst | |
| 4073 if (dstenc >= 8) { | |
| 4074 emit_opcode(cbuf, Assembler::REX_B); | |
| 4075 } | |
| 4076 emit_opcode(cbuf, 0x58 | (dstenc & 7)); | |
| 4077 | |
| 4078 // done: | |
| 4079 %} | |
| 4080 | |
| 4081 enc_class d2i_fixup(rRegI dst, regD src) | |
| 4082 %{ | |
| 4083 int dstenc = $dst$$reg; | |
| 4084 int srcenc = $src$$reg; | |
| 4085 | |
| 4086 // cmpl $dst, #0x80000000 | |
| 4087 if (dstenc >= 8) { | |
| 4088 emit_opcode(cbuf, Assembler::REX_B); | |
| 4089 } | |
| 4090 emit_opcode(cbuf, 0x81); | |
| 4091 emit_rm(cbuf, 0x3, 0x7, dstenc & 7); | |
| 4092 emit_d32(cbuf, 0x80000000); | |
| 4093 | |
| 4094 // jne,s done | |
| 4095 emit_opcode(cbuf, 0x75); | |
| 4096 if (srcenc < 8 && dstenc < 8) { | |
| 4097 emit_d8(cbuf, 0xF); | |
| 4098 } else if (srcenc >= 8 && dstenc >= 8) { | |
| 4099 emit_d8(cbuf, 0x11); | |
| 4100 } else { | |
| 4101 emit_d8(cbuf, 0x10); | |
| 4102 } | |
| 4103 | |
| 4104 // subq rsp, #8 | |
| 4105 emit_opcode(cbuf, Assembler::REX_W); | |
| 4106 emit_opcode(cbuf, 0x83); | |
| 4107 emit_rm(cbuf, 0x3, 0x5, RSP_enc); | |
| 4108 emit_d8(cbuf, 8); | |
| 4109 | |
| 4110 // movsd [rsp], $src | |
| 4111 emit_opcode(cbuf, 0xF2); | |
| 4112 if (srcenc >= 8) { | |
| 4113 emit_opcode(cbuf, Assembler::REX_R); | |
| 4114 } | |
| 4115 emit_opcode(cbuf, 0x0F); | |
| 4116 emit_opcode(cbuf, 0x11); | |
| 4117 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes | |
| 4118 | |
| 4119 // call d2i_fixup | |
| 4120 cbuf.set_inst_mark(); | |
| 4121 emit_opcode(cbuf, 0xE8); | |
| 4122 emit_d32_reloc(cbuf, | |
| 4123 (int) | |
| 304 | 4124 (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4), |
| 0 | 4125 runtime_call_Relocation::spec(), |
| 4126 RELOC_DISP32); | |
| 4127 | |
| 4128 // popq $dst | |
| 4129 if (dstenc >= 8) { | |
| 4130 emit_opcode(cbuf, Assembler::REX_B); | |
| 4131 } | |
| 4132 emit_opcode(cbuf, 0x58 | (dstenc & 7)); | |
| 4133 | |
| 4134 // done: | |
| 4135 %} | |
| 4136 | |
| 4137 enc_class d2l_fixup(rRegL dst, regD src) | |
| 4138 %{ | |
| 4139 int dstenc = $dst$$reg; | |
| 4140 int srcenc = $src$$reg; | |
| 304 | 4141 address const_address = (address) StubRoutines::x86::double_sign_flip(); |
| 0 | 4142 |
| 4143 // cmpq $dst, [0x8000000000000000] | |
| 4144 cbuf.set_inst_mark(); | |
| 4145 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR); | |
| 4146 emit_opcode(cbuf, 0x39); | |
| 4147 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 4148 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101 | |
| 4149 emit_d32_reloc(cbuf, const_address); | |
| 4150 | |
| 4151 | |
| 4152 // jne,s done | |
| 4153 emit_opcode(cbuf, 0x75); | |
| 4154 if (srcenc < 8 && dstenc < 8) { | |
| 4155 emit_d8(cbuf, 0xF); | |
| 4156 } else if (srcenc >= 8 && dstenc >= 8) { | |
| 4157 emit_d8(cbuf, 0x11); | |
| 4158 } else { | |
| 4159 emit_d8(cbuf, 0x10); | |
| 4160 } | |
| 4161 | |
| 4162 // subq rsp, #8 | |
| 4163 emit_opcode(cbuf, Assembler::REX_W); | |
| 4164 emit_opcode(cbuf, 0x83); | |
| 4165 emit_rm(cbuf, 0x3, 0x5, RSP_enc); | |
| 4166 emit_d8(cbuf, 8); | |
| 4167 | |
| 4168 // movsd [rsp], $src | |
| 4169 emit_opcode(cbuf, 0xF2); | |
| 4170 if (srcenc >= 8) { | |
| 4171 emit_opcode(cbuf, Assembler::REX_R); | |
| 4172 } | |
| 4173 emit_opcode(cbuf, 0x0F); | |
| 4174 emit_opcode(cbuf, 0x11); | |
| 4175 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes | |
| 4176 | |
| 4177 // call d2l_fixup | |
| 4178 cbuf.set_inst_mark(); | |
| 4179 emit_opcode(cbuf, 0xE8); | |
| 4180 emit_d32_reloc(cbuf, | |
| 4181 (int) | |
| 304 | 4182 (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4), |
| 0 | 4183 runtime_call_Relocation::spec(), |
| 4184 RELOC_DISP32); | |
| 4185 | |
| 4186 // popq $dst | |
| 4187 if (dstenc >= 8) { | |
| 4188 emit_opcode(cbuf, Assembler::REX_B); | |
| 4189 } | |
| 4190 emit_opcode(cbuf, 0x58 | (dstenc & 7)); | |
| 4191 | |
| 4192 // done: | |
| 4193 %} | |
| 4194 | |
| 4195 enc_class enc_membar_acquire | |
| 4196 %{ | |
| 4197 // [jk] not needed currently, if you enable this and it really | |
| 4198 // emits code don't forget to the remove the "size(0)" line in | |
| 4199 // membar_acquire() | |
| 4200 // MacroAssembler masm(&cbuf); | |
| 4201 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore | | |
| 4202 // Assembler::LoadLoad)); | |
| 4203 %} | |
| 4204 | |
| 4205 enc_class enc_membar_release | |
| 4206 %{ | |
| 4207 // [jk] not needed currently, if you enable this and it really | |
| 4208 // emits code don't forget to the remove the "size(0)" line in | |
| 4209 // membar_release() | |
| 4210 // MacroAssembler masm(&cbuf); | |
| 4211 // masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore | | |
| 4212 // Assembler::StoreStore)); | |
| 4213 %} | |
| 4214 | |
| 4215 enc_class enc_membar_volatile | |
| 4216 %{ | |
| 4217 MacroAssembler masm(&cbuf); | |
| 4218 masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad | | |
| 4219 Assembler::StoreStore)); | |
| 4220 %} | |
| 4221 | |
| 4222 // Safepoint Poll. This polls the safepoint page, and causes an | |
| 4223 // exception if it is not readable. Unfortunately, it kills | |
| 4224 // RFLAGS in the process. | |
| 4225 enc_class enc_safepoint_poll | |
| 4226 %{ | |
| 4227 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes | |
| 4228 // XXX reg_mem doesn't support RIP-relative addressing yet | |
| 4229 cbuf.set_inst_mark(); | |
| 4230 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX | |
| 4231 emit_opcode(cbuf, 0x85); // testl | |
| 4232 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5 | |
| 4233 // cbuf.inst_mark() is beginning of instruction | |
| 4234 emit_d32_reloc(cbuf, os::get_polling_page()); | |
| 4235 // relocInfo::poll_type, | |
| 4236 %} | |
| 4237 %} | |
| 4238 | |
| 4239 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4240 |
| 0 | 4241 //----------FRAME-------------------------------------------------------------- |
| 4242 // Definition of frame structure and management information. | |
| 4243 // | |
| 4244 // S T A C K L A Y O U T Allocators stack-slot number | |
| 4245 // | (to get allocators register number | |
| 4246 // G Owned by | | v add OptoReg::stack0()) | |
| 4247 // r CALLER | | | |
| 4248 // o | +--------+ pad to even-align allocators stack-slot | |
| 4249 // w V | pad0 | numbers; owned by CALLER | |
| 4250 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned | |
| 4251 // h ^ | in | 5 | |
| 4252 // | | args | 4 Holes in incoming args owned by SELF | |
| 4253 // | | | | 3 | |
| 4254 // | | +--------+ | |
| 4255 // V | | old out| Empty on Intel, window on Sparc | |
| 4256 // | old |preserve| Must be even aligned. | |
| 4257 // | SP-+--------+----> Matcher::_old_SP, even aligned | |
| 4258 // | | in | 3 area for Intel ret address | |
| 4259 // Owned by |preserve| Empty on Sparc. | |
| 4260 // SELF +--------+ | |
| 4261 // | | pad2 | 2 pad to align old SP | |
| 4262 // | +--------+ 1 | |
| 4263 // | | locks | 0 | |
| 4264 // | +--------+----> OptoReg::stack0(), even aligned | |
| 4265 // | | pad1 | 11 pad to align new SP | |
| 4266 // | +--------+ | |
| 4267 // | | | 10 | |
| 4268 // | | spills | 9 spills | |
| 4269 // V | | 8 (pad0 slot for callee) | |
| 4270 // -----------+--------+----> Matcher::_out_arg_limit, unaligned | |
| 4271 // ^ | out | 7 | |
| 4272 // | | args | 6 Holes in outgoing args owned by CALLEE | |
| 4273 // Owned by +--------+ | |
| 4274 // CALLEE | new out| 6 Empty on Intel, window on Sparc | |
| 4275 // | new |preserve| Must be even-aligned. | |
| 4276 // | SP-+--------+----> Matcher::_new_SP, even aligned | |
| 4277 // | | | | |
| 4278 // | |
| 4279 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is | |
| 4280 // known from SELF's arguments and the Java calling convention. | |
| 4281 // Region 6-7 is determined per call site. | |
| 4282 // Note 2: If the calling convention leaves holes in the incoming argument | |
| 4283 // area, those holes are owned by SELF. Holes in the outgoing area | |
| 4284 // are owned by the CALLEE. Holes should not be nessecary in the | |
| 4285 // incoming area, as the Java calling convention is completely under | |
| 4286 // the control of the AD file. Doubles can be sorted and packed to | |
| 4287 // avoid holes. Holes in the outgoing arguments may be nessecary for | |
| 4288 // varargs C calling conventions. | |
| 4289 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is | |
| 4290 // even aligned with pad0 as needed. | |
| 4291 // Region 6 is even aligned. Region 6-7 is NOT even aligned; | |
| 4292 // region 6-11 is even aligned; it may be padded out more so that | |
| 4293 // the region from SP to FP meets the minimum stack alignment. | |
| 4294 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack | |
| 4295 // alignment. Region 11, pad1, may be dynamically extended so that | |
| 4296 // SP meets the minimum alignment. | |
| 4297 | |
| 4298 frame | |
| 4299 %{ | |
| 4300 // What direction does stack grow in (assumed to be same for C & Java) | |
| 4301 stack_direction(TOWARDS_LOW); | |
| 4302 | |
| 4303 // These three registers define part of the calling convention | |
| 4304 // between compiled code and the interpreter. | |
| 4305 inline_cache_reg(RAX); // Inline Cache Register | |
| 4306 interpreter_method_oop_reg(RBX); // Method Oop Register when | |
| 4307 // calling interpreter | |
| 4308 | |
| 4309 // Optional: name the operand used by cisc-spilling to access | |
| 4310 // [stack_pointer + offset] | |
| 4311 cisc_spilling_operand_name(indOffset32); | |
| 4312 | |
| 4313 // Number of stack slots consumed by locking an object | |
| 4314 sync_stack_slots(2); | |
| 4315 | |
| 4316 // Compiled code's Frame Pointer | |
| 4317 frame_pointer(RSP); | |
| 4318 | |
| 4319 // Interpreter stores its frame pointer in a register which is | |
| 4320 // stored to the stack by I2CAdaptors. | |
| 4321 // I2CAdaptors convert from interpreted java to compiled java. | |
| 4322 interpreter_frame_pointer(RBP); | |
| 4323 | |
| 4324 // Stack alignment requirement | |
| 4325 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes) | |
| 4326 | |
| 4327 // Number of stack slots between incoming argument block and the start of | |
| 4328 // a new frame. The PROLOG must add this many slots to the stack. The | |
| 4329 // EPILOG must remove this many slots. amd64 needs two slots for | |
| 4330 // return address. | |
| 4331 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls); | |
| 4332 | |
| 4333 // Number of outgoing stack slots killed above the out_preserve_stack_slots | |
| 4334 // for calls to C. Supports the var-args backing area for register parms. | |
| 4335 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt); | |
| 4336 | |
| 4337 // The after-PROLOG location of the return address. Location of | |
| 4338 // return address specifies a type (REG or STACK) and a number | |
| 4339 // representing the register number (i.e. - use a register name) or | |
| 4340 // stack slot. | |
| 4341 // Ret Addr is on stack in slot 0 if no locks or verification or alignment. | |
| 4342 // Otherwise, it is above the locks and verification slot and alignment word | |
| 4343 return_addr(STACK - 2 + | |
| 4344 round_to(2 + 2 * VerifyStackAtCalls + | |
| 4345 Compile::current()->fixed_slots(), | |
| 4346 WordsPerLong * 2)); | |
| 4347 | |
| 4348 // Body of function which returns an integer array locating | |
| 4349 // arguments either in registers or in stack slots. Passed an array | |
| 4350 // of ideal registers called "sig" and a "length" count. Stack-slot | |
| 4351 // offsets are based on outgoing arguments, i.e. a CALLER setting up | |
| 4352 // arguments for a CALLEE. Incoming stack arguments are | |
| 4353 // automatically biased by the preserve_stack_slots field above. | |
| 4354 | |
| 4355 calling_convention | |
| 4356 %{ | |
| 4357 // No difference between ingoing/outgoing just pass false | |
| 4358 SharedRuntime::java_calling_convention(sig_bt, regs, length, false); | |
| 4359 %} | |
| 4360 | |
| 4361 c_calling_convention | |
| 4362 %{ | |
| 4363 // This is obviously always outgoing | |
| 4364 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length); | |
| 4365 %} | |
| 4366 | |
| 4367 // Location of compiled Java return values. Same as C for now. | |
| 4368 return_value | |
| 4369 %{ | |
| 4370 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, | |
| 4371 "only return normal values"); | |
| 4372 | |
| 4373 static const int lo[Op_RegL + 1] = { | |
| 4374 0, | |
| 4375 0, | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4376 RAX_num, // Op_RegN |
| 0 | 4377 RAX_num, // Op_RegI |
| 4378 RAX_num, // Op_RegP | |
| 4379 XMM0_num, // Op_RegF | |
| 4380 XMM0_num, // Op_RegD | |
| 4381 RAX_num // Op_RegL | |
| 4382 }; | |
| 4383 static const int hi[Op_RegL + 1] = { | |
| 4384 0, | |
| 4385 0, | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4386 OptoReg::Bad, // Op_RegN |
| 0 | 4387 OptoReg::Bad, // Op_RegI |
| 4388 RAX_H_num, // Op_RegP | |
| 4389 OptoReg::Bad, // Op_RegF | |
| 4390 XMM0_H_num, // Op_RegD | |
| 4391 RAX_H_num // Op_RegL | |
| 4392 }; | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4393 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type"); |
| 0 | 4394 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]); |
| 4395 %} | |
| 4396 %} | |
| 4397 | |
| 4398 //----------ATTRIBUTES--------------------------------------------------------- | |
| 4399 //----------Operand Attributes------------------------------------------------- | |
| 4400 op_attrib op_cost(0); // Required cost attribute | |
| 4401 | |
| 4402 //----------Instruction Attributes--------------------------------------------- | |
| 4403 ins_attrib ins_cost(100); // Required cost attribute | |
| 4404 ins_attrib ins_size(8); // Required size attribute (in bits) | |
| 4405 ins_attrib ins_pc_relative(0); // Required PC Relative flag | |
| 4406 ins_attrib ins_short_branch(0); // Required flag: is this instruction | |
| 4407 // a non-matching short branch variant | |
| 4408 // of some long branch? | |
| 4409 ins_attrib ins_alignment(1); // Required alignment attribute (must | |
| 4410 // be a power of 2) specifies the | |
| 4411 // alignment that some part of the | |
| 4412 // instruction (not necessarily the | |
| 4413 // start) requires. If > 1, a | |
| 4414 // compute_padding() function must be | |
| 4415 // provided for the instruction | |
| 4416 | |
| 4417 //----------OPERANDS----------------------------------------------------------- | |
| 4418 // Operand definitions must precede instruction definitions for correct parsing | |
| 4419 // in the ADLC because operands constitute user defined types which are used in | |
| 4420 // instruction definitions. | |
| 4421 | |
| 4422 //----------Simple Operands---------------------------------------------------- | |
| 4423 // Immediate Operands | |
| 4424 // Integer Immediate | |
| 4425 operand immI() | |
| 4426 %{ | |
| 4427 match(ConI); | |
| 4428 | |
| 4429 op_cost(10); | |
| 4430 format %{ %} | |
| 4431 interface(CONST_INTER); | |
| 4432 %} | |
| 4433 | |
| 4434 // Constant for test vs zero | |
| 4435 operand immI0() | |
| 4436 %{ | |
| 4437 predicate(n->get_int() == 0); | |
| 4438 match(ConI); | |
| 4439 | |
| 4440 op_cost(0); | |
| 4441 format %{ %} | |
| 4442 interface(CONST_INTER); | |
| 4443 %} | |
| 4444 | |
| 4445 // Constant for increment | |
| 4446 operand immI1() | |
| 4447 %{ | |
| 4448 predicate(n->get_int() == 1); | |
| 4449 match(ConI); | |
| 4450 | |
| 4451 op_cost(0); | |
| 4452 format %{ %} | |
| 4453 interface(CONST_INTER); | |
| 4454 %} | |
| 4455 | |
| 4456 // Constant for decrement | |
| 4457 operand immI_M1() | |
| 4458 %{ | |
| 4459 predicate(n->get_int() == -1); | |
| 4460 match(ConI); | |
| 4461 | |
| 4462 op_cost(0); | |
| 4463 format %{ %} | |
| 4464 interface(CONST_INTER); | |
| 4465 %} | |
| 4466 | |
| 4467 // Valid scale values for addressing modes | |
| 4468 operand immI2() | |
| 4469 %{ | |
| 4470 predicate(0 <= n->get_int() && (n->get_int() <= 3)); | |
| 4471 match(ConI); | |
| 4472 | |
| 4473 format %{ %} | |
| 4474 interface(CONST_INTER); | |
| 4475 %} | |
| 4476 | |
| 4477 operand immI8() | |
| 4478 %{ | |
| 4479 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80)); | |
| 4480 match(ConI); | |
| 4481 | |
| 4482 op_cost(5); | |
| 4483 format %{ %} | |
| 4484 interface(CONST_INTER); | |
| 4485 %} | |
| 4486 | |
| 4487 operand immI16() | |
| 4488 %{ | |
| 4489 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767)); | |
| 4490 match(ConI); | |
| 4491 | |
| 4492 op_cost(10); | |
| 4493 format %{ %} | |
| 4494 interface(CONST_INTER); | |
| 4495 %} | |
| 4496 | |
| 4497 // Constant for long shifts | |
| 4498 operand immI_32() | |
| 4499 %{ | |
| 4500 predicate( n->get_int() == 32 ); | |
| 4501 match(ConI); | |
| 4502 | |
| 4503 op_cost(0); | |
| 4504 format %{ %} | |
| 4505 interface(CONST_INTER); | |
| 4506 %} | |
| 4507 | |
| 4508 // Constant for long shifts | |
| 4509 operand immI_64() | |
| 4510 %{ | |
| 4511 predicate( n->get_int() == 64 ); | |
| 4512 match(ConI); | |
| 4513 | |
| 4514 op_cost(0); | |
| 4515 format %{ %} | |
| 4516 interface(CONST_INTER); | |
| 4517 %} | |
| 4518 | |
| 4519 // Pointer Immediate | |
| 4520 operand immP() | |
| 4521 %{ | |
| 4522 match(ConP); | |
| 4523 | |
| 4524 op_cost(10); | |
| 4525 format %{ %} | |
| 4526 interface(CONST_INTER); | |
| 4527 %} | |
| 4528 | |
| 4529 // NULL Pointer Immediate | |
| 4530 operand immP0() | |
| 4531 %{ | |
| 4532 predicate(n->get_ptr() == 0); | |
| 4533 match(ConP); | |
| 4534 | |
| 4535 op_cost(5); | |
| 4536 format %{ %} | |
| 4537 interface(CONST_INTER); | |
| 4538 %} | |
| 4539 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4540 // Pointer Immediate |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4541 operand immN() %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4542 match(ConN); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4543 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4544 op_cost(10); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4545 format %{ %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4546 interface(CONST_INTER); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4547 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4548 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4549 // NULL Pointer Immediate |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4550 operand immN0() %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4551 predicate(n->get_narrowcon() == 0); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4552 match(ConN); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4553 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4554 op_cost(5); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4555 format %{ %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4556 interface(CONST_INTER); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4557 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4558 |
| 0 | 4559 operand immP31() |
| 4560 %{ | |
| 4561 predicate(!n->as_Type()->type()->isa_oopptr() | |
| 4562 && (n->get_ptr() >> 31) == 0); | |
| 4563 match(ConP); | |
| 4564 | |
| 4565 op_cost(5); | |
| 4566 format %{ %} | |
| 4567 interface(CONST_INTER); | |
| 4568 %} | |
| 4569 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4570 |
| 0 | 4571 // Long Immediate |
| 4572 operand immL() | |
| 4573 %{ | |
| 4574 match(ConL); | |
| 4575 | |
| 4576 op_cost(20); | |
| 4577 format %{ %} | |
| 4578 interface(CONST_INTER); | |
| 4579 %} | |
| 4580 | |
| 4581 // Long Immediate 8-bit | |
| 4582 operand immL8() | |
| 4583 %{ | |
| 4584 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L); | |
| 4585 match(ConL); | |
| 4586 | |
| 4587 op_cost(5); | |
| 4588 format %{ %} | |
| 4589 interface(CONST_INTER); | |
| 4590 %} | |
| 4591 | |
| 4592 // Long Immediate 32-bit unsigned | |
| 4593 operand immUL32() | |
| 4594 %{ | |
| 4595 predicate(n->get_long() == (unsigned int) (n->get_long())); | |
| 4596 match(ConL); | |
| 4597 | |
| 4598 op_cost(10); | |
| 4599 format %{ %} | |
| 4600 interface(CONST_INTER); | |
| 4601 %} | |
| 4602 | |
| 4603 // Long Immediate 32-bit signed | |
| 4604 operand immL32() | |
| 4605 %{ | |
| 4606 predicate(n->get_long() == (int) (n->get_long())); | |
| 4607 match(ConL); | |
| 4608 | |
| 4609 op_cost(15); | |
| 4610 format %{ %} | |
| 4611 interface(CONST_INTER); | |
| 4612 %} | |
| 4613 | |
| 4614 // Long Immediate zero | |
| 4615 operand immL0() | |
| 4616 %{ | |
| 4617 predicate(n->get_long() == 0L); | |
| 4618 match(ConL); | |
| 4619 | |
| 4620 op_cost(10); | |
| 4621 format %{ %} | |
| 4622 interface(CONST_INTER); | |
| 4623 %} | |
| 4624 | |
| 4625 // Constant for increment | |
| 4626 operand immL1() | |
| 4627 %{ | |
| 4628 predicate(n->get_long() == 1); | |
| 4629 match(ConL); | |
| 4630 | |
| 4631 format %{ %} | |
| 4632 interface(CONST_INTER); | |
| 4633 %} | |
| 4634 | |
| 4635 // Constant for decrement | |
| 4636 operand immL_M1() | |
| 4637 %{ | |
| 4638 predicate(n->get_long() == -1); | |
| 4639 match(ConL); | |
| 4640 | |
| 4641 format %{ %} | |
| 4642 interface(CONST_INTER); | |
| 4643 %} | |
| 4644 | |
| 4645 // Long Immediate: the value 10 | |
| 4646 operand immL10() | |
| 4647 %{ | |
| 4648 predicate(n->get_long() == 10); | |
| 4649 match(ConL); | |
| 4650 | |
| 4651 format %{ %} | |
| 4652 interface(CONST_INTER); | |
| 4653 %} | |
| 4654 | |
| 4655 // Long immediate from 0 to 127. | |
| 4656 // Used for a shorter form of long mul by 10. | |
| 4657 operand immL_127() | |
| 4658 %{ | |
| 4659 predicate(0 <= n->get_long() && n->get_long() < 0x80); | |
| 4660 match(ConL); | |
| 4661 | |
| 4662 op_cost(10); | |
| 4663 format %{ %} | |
| 4664 interface(CONST_INTER); | |
| 4665 %} | |
| 4666 | |
| 4667 // Long Immediate: low 32-bit mask | |
| 4668 operand immL_32bits() | |
| 4669 %{ | |
| 4670 predicate(n->get_long() == 0xFFFFFFFFL); | |
| 4671 match(ConL); | |
| 4672 op_cost(20); | |
| 4673 | |
| 4674 format %{ %} | |
| 4675 interface(CONST_INTER); | |
| 4676 %} | |
| 4677 | |
| 4678 // Float Immediate zero | |
| 4679 operand immF0() | |
| 4680 %{ | |
| 4681 predicate(jint_cast(n->getf()) == 0); | |
| 4682 match(ConF); | |
| 4683 | |
| 4684 op_cost(5); | |
| 4685 format %{ %} | |
| 4686 interface(CONST_INTER); | |
| 4687 %} | |
| 4688 | |
| 4689 // Float Immediate | |
| 4690 operand immF() | |
| 4691 %{ | |
| 4692 match(ConF); | |
| 4693 | |
| 4694 op_cost(15); | |
| 4695 format %{ %} | |
| 4696 interface(CONST_INTER); | |
| 4697 %} | |
| 4698 | |
| 4699 // Double Immediate zero | |
| 4700 operand immD0() | |
| 4701 %{ | |
| 4702 predicate(jlong_cast(n->getd()) == 0); | |
| 4703 match(ConD); | |
| 4704 | |
| 4705 op_cost(5); | |
| 4706 format %{ %} | |
| 4707 interface(CONST_INTER); | |
| 4708 %} | |
| 4709 | |
| 4710 // Double Immediate | |
| 4711 operand immD() | |
| 4712 %{ | |
| 4713 match(ConD); | |
| 4714 | |
| 4715 op_cost(15); | |
| 4716 format %{ %} | |
| 4717 interface(CONST_INTER); | |
| 4718 %} | |
| 4719 | |
| 4720 // Immediates for special shifts (sign extend) | |
| 4721 | |
| 4722 // Constants for increment | |
| 4723 operand immI_16() | |
| 4724 %{ | |
| 4725 predicate(n->get_int() == 16); | |
| 4726 match(ConI); | |
| 4727 | |
| 4728 format %{ %} | |
| 4729 interface(CONST_INTER); | |
| 4730 %} | |
| 4731 | |
| 4732 operand immI_24() | |
| 4733 %{ | |
| 4734 predicate(n->get_int() == 24); | |
| 4735 match(ConI); | |
| 4736 | |
| 4737 format %{ %} | |
| 4738 interface(CONST_INTER); | |
| 4739 %} | |
| 4740 | |
| 4741 // Constant for byte-wide masking | |
| 4742 operand immI_255() | |
| 4743 %{ | |
| 4744 predicate(n->get_int() == 255); | |
| 4745 match(ConI); | |
| 4746 | |
| 4747 format %{ %} | |
| 4748 interface(CONST_INTER); | |
| 4749 %} | |
| 4750 | |
| 4751 // Constant for short-wide masking | |
| 4752 operand immI_65535() | |
| 4753 %{ | |
| 4754 predicate(n->get_int() == 65535); | |
| 4755 match(ConI); | |
| 4756 | |
| 4757 format %{ %} | |
| 4758 interface(CONST_INTER); | |
| 4759 %} | |
| 4760 | |
| 4761 // Constant for byte-wide masking | |
| 4762 operand immL_255() | |
| 4763 %{ | |
| 4764 predicate(n->get_long() == 255); | |
| 4765 match(ConL); | |
| 4766 | |
| 4767 format %{ %} | |
| 4768 interface(CONST_INTER); | |
| 4769 %} | |
| 4770 | |
| 4771 // Constant for short-wide masking | |
| 4772 operand immL_65535() | |
| 4773 %{ | |
| 4774 predicate(n->get_long() == 65535); | |
| 4775 match(ConL); | |
| 4776 | |
| 4777 format %{ %} | |
| 4778 interface(CONST_INTER); | |
| 4779 %} | |
| 4780 | |
| 4781 // Register Operands | |
| 4782 // Integer Register | |
| 4783 operand rRegI() | |
| 4784 %{ | |
| 4785 constraint(ALLOC_IN_RC(int_reg)); | |
| 4786 match(RegI); | |
| 4787 | |
| 4788 match(rax_RegI); | |
| 4789 match(rbx_RegI); | |
| 4790 match(rcx_RegI); | |
| 4791 match(rdx_RegI); | |
| 4792 match(rdi_RegI); | |
| 4793 | |
| 4794 format %{ %} | |
| 4795 interface(REG_INTER); | |
| 4796 %} | |
| 4797 | |
| 4798 // Special Registers | |
| 4799 operand rax_RegI() | |
| 4800 %{ | |
| 4801 constraint(ALLOC_IN_RC(int_rax_reg)); | |
| 4802 match(RegI); | |
| 4803 match(rRegI); | |
| 4804 | |
| 4805 format %{ "RAX" %} | |
| 4806 interface(REG_INTER); | |
| 4807 %} | |
| 4808 | |
| 4809 // Special Registers | |
| 4810 operand rbx_RegI() | |
| 4811 %{ | |
| 4812 constraint(ALLOC_IN_RC(int_rbx_reg)); | |
| 4813 match(RegI); | |
| 4814 match(rRegI); | |
| 4815 | |
| 4816 format %{ "RBX" %} | |
| 4817 interface(REG_INTER); | |
| 4818 %} | |
| 4819 | |
| 4820 operand rcx_RegI() | |
| 4821 %{ | |
| 4822 constraint(ALLOC_IN_RC(int_rcx_reg)); | |
| 4823 match(RegI); | |
| 4824 match(rRegI); | |
| 4825 | |
| 4826 format %{ "RCX" %} | |
| 4827 interface(REG_INTER); | |
| 4828 %} | |
| 4829 | |
| 4830 operand rdx_RegI() | |
| 4831 %{ | |
| 4832 constraint(ALLOC_IN_RC(int_rdx_reg)); | |
| 4833 match(RegI); | |
| 4834 match(rRegI); | |
| 4835 | |
| 4836 format %{ "RDX" %} | |
| 4837 interface(REG_INTER); | |
| 4838 %} | |
| 4839 | |
| 4840 operand rdi_RegI() | |
| 4841 %{ | |
| 4842 constraint(ALLOC_IN_RC(int_rdi_reg)); | |
| 4843 match(RegI); | |
| 4844 match(rRegI); | |
| 4845 | |
| 4846 format %{ "RDI" %} | |
| 4847 interface(REG_INTER); | |
| 4848 %} | |
| 4849 | |
| 4850 operand no_rcx_RegI() | |
| 4851 %{ | |
| 4852 constraint(ALLOC_IN_RC(int_no_rcx_reg)); | |
| 4853 match(RegI); | |
| 4854 match(rax_RegI); | |
| 4855 match(rbx_RegI); | |
| 4856 match(rdx_RegI); | |
| 4857 match(rdi_RegI); | |
| 4858 | |
| 4859 format %{ %} | |
| 4860 interface(REG_INTER); | |
| 4861 %} | |
| 4862 | |
| 4863 operand no_rax_rdx_RegI() | |
| 4864 %{ | |
| 4865 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg)); | |
| 4866 match(RegI); | |
| 4867 match(rbx_RegI); | |
| 4868 match(rcx_RegI); | |
| 4869 match(rdi_RegI); | |
| 4870 | |
| 4871 format %{ %} | |
| 4872 interface(REG_INTER); | |
| 4873 %} | |
| 4874 | |
| 4875 // Pointer Register | |
| 4876 operand any_RegP() | |
| 4877 %{ | |
| 4878 constraint(ALLOC_IN_RC(any_reg)); | |
| 4879 match(RegP); | |
| 4880 match(rax_RegP); | |
| 4881 match(rbx_RegP); | |
| 4882 match(rdi_RegP); | |
| 4883 match(rsi_RegP); | |
| 4884 match(rbp_RegP); | |
| 4885 match(r15_RegP); | |
| 4886 match(rRegP); | |
| 4887 | |
| 4888 format %{ %} | |
| 4889 interface(REG_INTER); | |
| 4890 %} | |
| 4891 | |
| 4892 operand rRegP() | |
| 4893 %{ | |
| 4894 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 4895 match(RegP); | |
| 4896 match(rax_RegP); | |
| 4897 match(rbx_RegP); | |
| 4898 match(rdi_RegP); | |
| 4899 match(rsi_RegP); | |
| 4900 match(rbp_RegP); | |
| 4901 match(r15_RegP); // See Q&A below about r15_RegP. | |
| 4902 | |
| 4903 format %{ %} | |
| 4904 interface(REG_INTER); | |
| 4905 %} | |
| 4906 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4907 operand rRegN() %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4908 constraint(ALLOC_IN_RC(int_reg)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4909 match(RegN); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4910 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4911 format %{ %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4912 interface(REG_INTER); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4913 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4914 |
| 0 | 4915 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP? |
| 4916 // Answer: Operand match rules govern the DFA as it processes instruction inputs. | |
| 4917 // It's fine for an instruction input which expects rRegP to match a r15_RegP. | |
| 4918 // The output of an instruction is controlled by the allocator, which respects | |
| 4919 // register class masks, not match rules. Unless an instruction mentions | |
| 4920 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered | |
| 4921 // by the allocator as an input. | |
| 4922 | |
| 4923 operand no_rax_RegP() | |
| 4924 %{ | |
| 4925 constraint(ALLOC_IN_RC(ptr_no_rax_reg)); | |
| 4926 match(RegP); | |
| 4927 match(rbx_RegP); | |
| 4928 match(rsi_RegP); | |
| 4929 match(rdi_RegP); | |
| 4930 | |
| 4931 format %{ %} | |
| 4932 interface(REG_INTER); | |
| 4933 %} | |
| 4934 | |
| 4935 operand no_rbp_RegP() | |
| 4936 %{ | |
| 4937 constraint(ALLOC_IN_RC(ptr_no_rbp_reg)); | |
| 4938 match(RegP); | |
| 4939 match(rbx_RegP); | |
| 4940 match(rsi_RegP); | |
| 4941 match(rdi_RegP); | |
| 4942 | |
| 4943 format %{ %} | |
| 4944 interface(REG_INTER); | |
| 4945 %} | |
| 4946 | |
| 4947 operand no_rax_rbx_RegP() | |
| 4948 %{ | |
| 4949 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg)); | |
| 4950 match(RegP); | |
| 4951 match(rsi_RegP); | |
| 4952 match(rdi_RegP); | |
| 4953 | |
| 4954 format %{ %} | |
| 4955 interface(REG_INTER); | |
| 4956 %} | |
| 4957 | |
| 4958 // Special Registers | |
| 4959 // Return a pointer value | |
| 4960 operand rax_RegP() | |
| 4961 %{ | |
| 4962 constraint(ALLOC_IN_RC(ptr_rax_reg)); | |
| 4963 match(RegP); | |
| 4964 match(rRegP); | |
| 4965 | |
| 4966 format %{ %} | |
| 4967 interface(REG_INTER); | |
| 4968 %} | |
| 4969 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4970 // Special Registers |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4971 // Return a compressed pointer value |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4972 operand rax_RegN() |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4973 %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4974 constraint(ALLOC_IN_RC(int_rax_reg)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4975 match(RegN); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4976 match(rRegN); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4977 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4978 format %{ %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4979 interface(REG_INTER); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4980 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
4981 |
| 0 | 4982 // Used in AtomicAdd |
| 4983 operand rbx_RegP() | |
| 4984 %{ | |
| 4985 constraint(ALLOC_IN_RC(ptr_rbx_reg)); | |
| 4986 match(RegP); | |
| 4987 match(rRegP); | |
| 4988 | |
| 4989 format %{ %} | |
| 4990 interface(REG_INTER); | |
| 4991 %} | |
| 4992 | |
| 4993 operand rsi_RegP() | |
| 4994 %{ | |
| 4995 constraint(ALLOC_IN_RC(ptr_rsi_reg)); | |
| 4996 match(RegP); | |
| 4997 match(rRegP); | |
| 4998 | |
| 4999 format %{ %} | |
| 5000 interface(REG_INTER); | |
| 5001 %} | |
| 5002 | |
| 5003 // Used in rep stosq | |
| 5004 operand rdi_RegP() | |
| 5005 %{ | |
| 5006 constraint(ALLOC_IN_RC(ptr_rdi_reg)); | |
| 5007 match(RegP); | |
| 5008 match(rRegP); | |
| 5009 | |
| 5010 format %{ %} | |
| 5011 interface(REG_INTER); | |
| 5012 %} | |
| 5013 | |
| 5014 operand rbp_RegP() | |
| 5015 %{ | |
| 5016 constraint(ALLOC_IN_RC(ptr_rbp_reg)); | |
| 5017 match(RegP); | |
| 5018 match(rRegP); | |
| 5019 | |
| 5020 format %{ %} | |
| 5021 interface(REG_INTER); | |
| 5022 %} | |
| 5023 | |
| 5024 operand r15_RegP() | |
| 5025 %{ | |
| 5026 constraint(ALLOC_IN_RC(ptr_r15_reg)); | |
| 5027 match(RegP); | |
| 5028 match(rRegP); | |
| 5029 | |
| 5030 format %{ %} | |
| 5031 interface(REG_INTER); | |
| 5032 %} | |
| 5033 | |
| 5034 operand rRegL() | |
| 5035 %{ | |
| 5036 constraint(ALLOC_IN_RC(long_reg)); | |
| 5037 match(RegL); | |
| 5038 match(rax_RegL); | |
| 5039 match(rdx_RegL); | |
| 5040 | |
| 5041 format %{ %} | |
| 5042 interface(REG_INTER); | |
| 5043 %} | |
| 5044 | |
| 5045 // Special Registers | |
| 5046 operand no_rax_rdx_RegL() | |
| 5047 %{ | |
| 5048 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg)); | |
| 5049 match(RegL); | |
| 5050 match(rRegL); | |
| 5051 | |
| 5052 format %{ %} | |
| 5053 interface(REG_INTER); | |
| 5054 %} | |
| 5055 | |
| 5056 operand no_rax_RegL() | |
| 5057 %{ | |
| 5058 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg)); | |
| 5059 match(RegL); | |
| 5060 match(rRegL); | |
| 5061 match(rdx_RegL); | |
| 5062 | |
| 5063 format %{ %} | |
| 5064 interface(REG_INTER); | |
| 5065 %} | |
| 5066 | |
| 5067 operand no_rcx_RegL() | |
| 5068 %{ | |
| 5069 constraint(ALLOC_IN_RC(long_no_rcx_reg)); | |
| 5070 match(RegL); | |
| 5071 match(rRegL); | |
| 5072 | |
| 5073 format %{ %} | |
| 5074 interface(REG_INTER); | |
| 5075 %} | |
| 5076 | |
| 5077 operand rax_RegL() | |
| 5078 %{ | |
| 5079 constraint(ALLOC_IN_RC(long_rax_reg)); | |
| 5080 match(RegL); | |
| 5081 match(rRegL); | |
| 5082 | |
| 5083 format %{ "RAX" %} | |
| 5084 interface(REG_INTER); | |
| 5085 %} | |
| 5086 | |
| 5087 operand rcx_RegL() | |
| 5088 %{ | |
| 5089 constraint(ALLOC_IN_RC(long_rcx_reg)); | |
| 5090 match(RegL); | |
| 5091 match(rRegL); | |
| 5092 | |
| 5093 format %{ %} | |
| 5094 interface(REG_INTER); | |
| 5095 %} | |
| 5096 | |
| 5097 operand rdx_RegL() | |
| 5098 %{ | |
| 5099 constraint(ALLOC_IN_RC(long_rdx_reg)); | |
| 5100 match(RegL); | |
| 5101 match(rRegL); | |
| 5102 | |
| 5103 format %{ %} | |
| 5104 interface(REG_INTER); | |
| 5105 %} | |
| 5106 | |
| 5107 // Flags register, used as output of compare instructions | |
| 5108 operand rFlagsReg() | |
| 5109 %{ | |
| 5110 constraint(ALLOC_IN_RC(int_flags)); | |
| 5111 match(RegFlags); | |
| 5112 | |
| 5113 format %{ "RFLAGS" %} | |
| 5114 interface(REG_INTER); | |
| 5115 %} | |
| 5116 | |
| 5117 // Flags register, used as output of FLOATING POINT compare instructions | |
| 5118 operand rFlagsRegU() | |
| 5119 %{ | |
| 5120 constraint(ALLOC_IN_RC(int_flags)); | |
| 5121 match(RegFlags); | |
| 5122 | |
| 5123 format %{ "RFLAGS_U" %} | |
| 5124 interface(REG_INTER); | |
| 5125 %} | |
| 5126 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5127 operand rFlagsRegUCF() %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5128 constraint(ALLOC_IN_RC(int_flags)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5129 match(RegFlags); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5130 predicate(false); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5131 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5132 format %{ "RFLAGS_U_CF" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5133 interface(REG_INTER); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5134 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5135 |
| 0 | 5136 // Float register operands |
| 5137 operand regF() | |
| 5138 %{ | |
| 5139 constraint(ALLOC_IN_RC(float_reg)); | |
| 5140 match(RegF); | |
| 5141 | |
| 5142 format %{ %} | |
| 5143 interface(REG_INTER); | |
| 5144 %} | |
| 5145 | |
| 5146 // Double register operands | |
| 5147 operand regD() | |
| 5148 %{ | |
| 5149 constraint(ALLOC_IN_RC(double_reg)); | |
| 5150 match(RegD); | |
| 5151 | |
| 5152 format %{ %} | |
| 5153 interface(REG_INTER); | |
| 5154 %} | |
| 5155 | |
| 5156 | |
| 5157 //----------Memory Operands---------------------------------------------------- | |
| 5158 // Direct Memory Operand | |
| 5159 // operand direct(immP addr) | |
| 5160 // %{ | |
| 5161 // match(addr); | |
| 5162 | |
| 5163 // format %{ "[$addr]" %} | |
| 5164 // interface(MEMORY_INTER) %{ | |
| 5165 // base(0xFFFFFFFF); | |
| 5166 // index(0x4); | |
| 5167 // scale(0x0); | |
| 5168 // disp($addr); | |
| 5169 // %} | |
| 5170 // %} | |
| 5171 | |
| 5172 // Indirect Memory Operand | |
| 5173 operand indirect(any_RegP reg) | |
| 5174 %{ | |
| 5175 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5176 match(reg); | |
| 5177 | |
| 5178 format %{ "[$reg]" %} | |
| 5179 interface(MEMORY_INTER) %{ | |
| 5180 base($reg); | |
| 5181 index(0x4); | |
| 5182 scale(0x0); | |
| 5183 disp(0x0); | |
| 5184 %} | |
| 5185 %} | |
| 5186 | |
| 5187 // Indirect Memory Plus Short Offset Operand | |
| 5188 operand indOffset8(any_RegP reg, immL8 off) | |
| 5189 %{ | |
| 5190 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5191 match(AddP reg off); | |
| 5192 | |
| 5193 format %{ "[$reg + $off (8-bit)]" %} | |
| 5194 interface(MEMORY_INTER) %{ | |
| 5195 base($reg); | |
| 5196 index(0x4); | |
| 5197 scale(0x0); | |
| 5198 disp($off); | |
| 5199 %} | |
| 5200 %} | |
| 5201 | |
| 5202 // Indirect Memory Plus Long Offset Operand | |
| 5203 operand indOffset32(any_RegP reg, immL32 off) | |
| 5204 %{ | |
| 5205 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5206 match(AddP reg off); | |
| 5207 | |
| 5208 format %{ "[$reg + $off (32-bit)]" %} | |
| 5209 interface(MEMORY_INTER) %{ | |
| 5210 base($reg); | |
| 5211 index(0x4); | |
| 5212 scale(0x0); | |
| 5213 disp($off); | |
| 5214 %} | |
| 5215 %} | |
| 5216 | |
| 5217 // Indirect Memory Plus Index Register Plus Offset Operand | |
| 5218 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off) | |
| 5219 %{ | |
| 5220 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5221 match(AddP (AddP reg lreg) off); | |
| 5222 | |
| 5223 op_cost(10); | |
| 5224 format %{"[$reg + $off + $lreg]" %} | |
| 5225 interface(MEMORY_INTER) %{ | |
| 5226 base($reg); | |
| 5227 index($lreg); | |
| 5228 scale(0x0); | |
| 5229 disp($off); | |
| 5230 %} | |
| 5231 %} | |
| 5232 | |
| 5233 // Indirect Memory Plus Index Register Plus Offset Operand | |
| 5234 operand indIndex(any_RegP reg, rRegL lreg) | |
| 5235 %{ | |
| 5236 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5237 match(AddP reg lreg); | |
| 5238 | |
| 5239 op_cost(10); | |
| 5240 format %{"[$reg + $lreg]" %} | |
| 5241 interface(MEMORY_INTER) %{ | |
| 5242 base($reg); | |
| 5243 index($lreg); | |
| 5244 scale(0x0); | |
| 5245 disp(0x0); | |
| 5246 %} | |
| 5247 %} | |
| 5248 | |
| 5249 // Indirect Memory Times Scale Plus Index Register | |
| 5250 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale) | |
| 5251 %{ | |
| 5252 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5253 match(AddP reg (LShiftL lreg scale)); | |
| 5254 | |
| 5255 op_cost(10); | |
| 5256 format %{"[$reg + $lreg << $scale]" %} | |
| 5257 interface(MEMORY_INTER) %{ | |
| 5258 base($reg); | |
| 5259 index($lreg); | |
| 5260 scale($scale); | |
| 5261 disp(0x0); | |
| 5262 %} | |
| 5263 %} | |
| 5264 | |
| 5265 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand | |
| 5266 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale) | |
| 5267 %{ | |
| 5268 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5269 match(AddP (AddP reg (LShiftL lreg scale)) off); | |
| 5270 | |
| 5271 op_cost(10); | |
| 5272 format %{"[$reg + $off + $lreg << $scale]" %} | |
| 5273 interface(MEMORY_INTER) %{ | |
| 5274 base($reg); | |
| 5275 index($lreg); | |
| 5276 scale($scale); | |
| 5277 disp($off); | |
| 5278 %} | |
| 5279 %} | |
| 5280 | |
| 5281 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand | |
| 5282 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale) | |
| 5283 %{ | |
| 5284 constraint(ALLOC_IN_RC(ptr_reg)); | |
| 5285 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0); | |
| 5286 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off); | |
| 5287 | |
| 5288 op_cost(10); | |
| 5289 format %{"[$reg + $off + $idx << $scale]" %} | |
| 5290 interface(MEMORY_INTER) %{ | |
| 5291 base($reg); | |
| 5292 index($idx); | |
| 5293 scale($scale); | |
| 5294 disp($off); | |
| 5295 %} | |
| 5296 %} | |
| 5297 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5298 // Indirect Narrow Oop Plus Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5299 // Note: x86 architecture doesn't support "scale * index + offset" without a base |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5300 // we can't free r12 even with Universe::narrow_oop_base() == NULL. |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5301 operand indCompressedOopOffset(rRegN reg, immL32 off) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5302 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5303 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5304 match(AddP (DecodeN reg) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5305 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5306 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5307 format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5308 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5309 base(0xc); // R12 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5310 index($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5311 scale(0x3); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5312 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5313 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5314 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5315 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5316 // Indirect Memory Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5317 operand indirectNarrow(rRegN reg) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5318 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5319 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5320 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5321 match(DecodeN reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5322 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5323 format %{ "[$reg]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5324 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5325 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5326 index(0x4); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5327 scale(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5328 disp(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5329 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5330 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5331 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5332 // Indirect Memory Plus Short Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5333 operand indOffset8Narrow(rRegN reg, immL8 off) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5334 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5335 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5336 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5337 match(AddP (DecodeN reg) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5338 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5339 format %{ "[$reg + $off (8-bit)]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5340 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5341 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5342 index(0x4); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5343 scale(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5344 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5345 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5346 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5347 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5348 // Indirect Memory Plus Long Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5349 operand indOffset32Narrow(rRegN reg, immL32 off) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5350 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5351 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5352 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5353 match(AddP (DecodeN reg) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5354 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5355 format %{ "[$reg + $off (32-bit)]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5356 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5357 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5358 index(0x4); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5359 scale(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5360 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5361 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5362 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5363 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5364 // Indirect Memory Plus Index Register Plus Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5365 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5366 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5367 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5368 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5369 match(AddP (AddP (DecodeN reg) lreg) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5370 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5371 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5372 format %{"[$reg + $off + $lreg]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5373 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5374 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5375 index($lreg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5376 scale(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5377 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5378 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5379 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5380 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5381 // Indirect Memory Plus Index Register Plus Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5382 operand indIndexNarrow(rRegN reg, rRegL lreg) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5383 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5384 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5385 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5386 match(AddP (DecodeN reg) lreg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5387 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5388 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5389 format %{"[$reg + $lreg]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5390 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5391 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5392 index($lreg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5393 scale(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5394 disp(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5395 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5396 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5397 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5398 // Indirect Memory Times Scale Plus Index Register |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5399 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5400 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5401 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5402 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5403 match(AddP (DecodeN reg) (LShiftL lreg scale)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5404 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5405 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5406 format %{"[$reg + $lreg << $scale]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5407 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5408 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5409 index($lreg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5410 scale($scale); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5411 disp(0x0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5412 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5413 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5414 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5415 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5416 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5417 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5418 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5419 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5420 match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5421 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5422 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5423 format %{"[$reg + $off + $lreg << $scale]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5424 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5425 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5426 index($lreg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5427 scale($scale); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5428 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5429 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5430 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5431 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5432 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5433 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5434 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5435 constraint(ALLOC_IN_RC(ptr_reg)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5436 predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5437 match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5438 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5439 op_cost(10); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5440 format %{"[$reg + $off + $idx << $scale]" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5441 interface(MEMORY_INTER) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5442 base($reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5443 index($idx); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5444 scale($scale); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5445 disp($off); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5446 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5447 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5448 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5449 |
| 0 | 5450 //----------Special Memory Operands-------------------------------------------- |
| 5451 // Stack Slot Operand - This operand is used for loading and storing temporary | |
| 5452 // values on the stack where a match requires a value to | |
| 5453 // flow through memory. | |
| 5454 operand stackSlotP(sRegP reg) | |
| 5455 %{ | |
| 5456 constraint(ALLOC_IN_RC(stack_slots)); | |
| 5457 // No match rule because this operand is only generated in matching | |
| 5458 | |
| 5459 format %{ "[$reg]" %} | |
| 5460 interface(MEMORY_INTER) %{ | |
| 5461 base(0x4); // RSP | |
| 5462 index(0x4); // No Index | |
| 5463 scale(0x0); // No Scale | |
| 5464 disp($reg); // Stack Offset | |
| 5465 %} | |
| 5466 %} | |
| 5467 | |
| 5468 operand stackSlotI(sRegI reg) | |
| 5469 %{ | |
| 5470 constraint(ALLOC_IN_RC(stack_slots)); | |
| 5471 // No match rule because this operand is only generated in matching | |
| 5472 | |
| 5473 format %{ "[$reg]" %} | |
| 5474 interface(MEMORY_INTER) %{ | |
| 5475 base(0x4); // RSP | |
| 5476 index(0x4); // No Index | |
| 5477 scale(0x0); // No Scale | |
| 5478 disp($reg); // Stack Offset | |
| 5479 %} | |
| 5480 %} | |
| 5481 | |
| 5482 operand stackSlotF(sRegF reg) | |
| 5483 %{ | |
| 5484 constraint(ALLOC_IN_RC(stack_slots)); | |
| 5485 // No match rule because this operand is only generated in matching | |
| 5486 | |
| 5487 format %{ "[$reg]" %} | |
| 5488 interface(MEMORY_INTER) %{ | |
| 5489 base(0x4); // RSP | |
| 5490 index(0x4); // No Index | |
| 5491 scale(0x0); // No Scale | |
| 5492 disp($reg); // Stack Offset | |
| 5493 %} | |
| 5494 %} | |
| 5495 | |
| 5496 operand stackSlotD(sRegD reg) | |
| 5497 %{ | |
| 5498 constraint(ALLOC_IN_RC(stack_slots)); | |
| 5499 // No match rule because this operand is only generated in matching | |
| 5500 | |
| 5501 format %{ "[$reg]" %} | |
| 5502 interface(MEMORY_INTER) %{ | |
| 5503 base(0x4); // RSP | |
| 5504 index(0x4); // No Index | |
| 5505 scale(0x0); // No Scale | |
| 5506 disp($reg); // Stack Offset | |
| 5507 %} | |
| 5508 %} | |
| 5509 operand stackSlotL(sRegL reg) | |
| 5510 %{ | |
| 5511 constraint(ALLOC_IN_RC(stack_slots)); | |
| 5512 // No match rule because this operand is only generated in matching | |
| 5513 | |
| 5514 format %{ "[$reg]" %} | |
| 5515 interface(MEMORY_INTER) %{ | |
| 5516 base(0x4); // RSP | |
| 5517 index(0x4); // No Index | |
| 5518 scale(0x0); // No Scale | |
| 5519 disp($reg); // Stack Offset | |
| 5520 %} | |
| 5521 %} | |
| 5522 | |
| 5523 //----------Conditional Branch Operands---------------------------------------- | |
| 5524 // Comparison Op - This is the operation of the comparison, and is limited to | |
| 5525 // the following set of codes: | |
| 5526 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=) | |
| 5527 // | |
| 5528 // Other attributes of the comparison, such as unsignedness, are specified | |
| 5529 // by the comparison instruction that sets a condition code flags register. | |
| 5530 // That result is represented by a flags operand whose subtype is appropriate | |
| 5531 // to the unsignedness (etc.) of the comparison. | |
| 5532 // | |
| 5533 // Later, the instruction which matches both the Comparison Op (a Bool) and | |
| 5534 // the flags (produced by the Cmp) specifies the coding of the comparison op | |
| 5535 // by matching a specific subtype of Bool operand below, such as cmpOpU. | |
| 5536 | |
| 5537 // Comparision Code | |
| 5538 operand cmpOp() | |
| 5539 %{ | |
| 5540 match(Bool); | |
| 5541 | |
| 5542 format %{ "" %} | |
| 5543 interface(COND_INTER) %{ | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5544 equal(0x4, "e"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5545 not_equal(0x5, "ne"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5546 less(0xC, "l"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5547 greater_equal(0xD, "ge"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5548 less_equal(0xE, "le"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5549 greater(0xF, "g"); |
| 0 | 5550 %} |
| 5551 %} | |
| 5552 | |
| 5553 // Comparison Code, unsigned compare. Used by FP also, with | |
| 5554 // C2 (unordered) turned into GT or LT already. The other bits | |
| 5555 // C0 and C3 are turned into Carry & Zero flags. | |
| 5556 operand cmpOpU() | |
| 5557 %{ | |
| 5558 match(Bool); | |
| 5559 | |
| 5560 format %{ "" %} | |
| 5561 interface(COND_INTER) %{ | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5562 equal(0x4, "e"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5563 not_equal(0x5, "ne"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5564 less(0x2, "b"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5565 greater_equal(0x3, "nb"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5566 less_equal(0x6, "be"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5567 greater(0x7, "nbe"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5568 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5569 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5570 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5571 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5572 // Floating comparisons that don't require any fixup for the unordered case |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5573 operand cmpOpUCF() %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5574 match(Bool); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5575 predicate(n->as_Bool()->_test._test == BoolTest::lt || |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5576 n->as_Bool()->_test._test == BoolTest::ge || |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5577 n->as_Bool()->_test._test == BoolTest::le || |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5578 n->as_Bool()->_test._test == BoolTest::gt); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5579 format %{ "" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5580 interface(COND_INTER) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5581 equal(0x4, "e"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5582 not_equal(0x5, "ne"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5583 less(0x2, "b"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5584 greater_equal(0x3, "nb"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5585 less_equal(0x6, "be"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5586 greater(0x7, "nbe"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5587 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5588 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5589 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5590 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5591 // Floating comparisons that can be fixed up with extra conditional jumps |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5592 operand cmpOpUCF2() %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5593 match(Bool); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5594 predicate(n->as_Bool()->_test._test == BoolTest::ne || |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5595 n->as_Bool()->_test._test == BoolTest::eq); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5596 format %{ "" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5597 interface(COND_INTER) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5598 equal(0x4, "e"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5599 not_equal(0x5, "ne"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5600 less(0x2, "b"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5601 greater_equal(0x3, "nb"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5602 less_equal(0x6, "be"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
5603 greater(0x7, "nbe"); |
| 0 | 5604 %} |
| 5605 %} | |
| 5606 | |
| 5607 | |
| 5608 //----------OPERAND CLASSES---------------------------------------------------- | |
| 5609 // Operand Classes are groups of operands that are used as to simplify | |
| 605 | 5610 // instruction definitions by not requiring the AD writer to specify separate |
| 0 | 5611 // instructions for every form of operand when the instruction accepts |
| 5612 // multiple operand types with the same basic encoding and format. The classic | |
| 5613 // case of this is memory operands. | |
| 5614 | |
| 5615 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex, | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
5616 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset, |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5617 indCompressedOopOffset, |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5618 indirectNarrow, indOffset8Narrow, indOffset32Narrow, |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5619 indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow, |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
5620 indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow); |
| 0 | 5621 |
| 5622 //----------PIPELINE----------------------------------------------------------- | |
| 5623 // Rules which define the behavior of the target architectures pipeline. | |
| 5624 pipeline %{ | |
| 5625 | |
| 5626 //----------ATTRIBUTES--------------------------------------------------------- | |
| 5627 attributes %{ | |
| 5628 variable_size_instructions; // Fixed size instructions | |
| 5629 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle | |
| 5630 instruction_unit_size = 1; // An instruction is 1 bytes long | |
| 5631 instruction_fetch_unit_size = 16; // The processor fetches one line | |
| 5632 instruction_fetch_units = 1; // of 16 bytes | |
| 5633 | |
| 5634 // List of nop instructions | |
| 5635 nops( MachNop ); | |
| 5636 %} | |
| 5637 | |
| 5638 //----------RESOURCES---------------------------------------------------------- | |
| 5639 // Resources are the functional units available to the machine | |
| 5640 | |
| 5641 // Generic P2/P3 pipeline | |
| 5642 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of | |
| 5643 // 3 instructions decoded per cycle. | |
| 5644 // 2 load/store ops per cycle, 1 branch, 1 FPU, | |
| 5645 // 3 ALU op, only ALU0 handles mul instructions. | |
| 5646 resources( D0, D1, D2, DECODE = D0 | D1 | D2, | |
| 5647 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2, | |
| 5648 BR, FPU, | |
| 5649 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2); | |
| 5650 | |
| 5651 //----------PIPELINE DESCRIPTION----------------------------------------------- | |
| 5652 // Pipeline Description specifies the stages in the machine's pipeline | |
| 5653 | |
| 5654 // Generic P2/P3 pipeline | |
| 5655 pipe_desc(S0, S1, S2, S3, S4, S5); | |
| 5656 | |
| 5657 //----------PIPELINE CLASSES--------------------------------------------------- | |
| 5658 // Pipeline Classes describe the stages in which input and output are | |
| 5659 // referenced by the hardware pipeline. | |
| 5660 | |
| 5661 // Naming convention: ialu or fpu | |
| 5662 // Then: _reg | |
| 5663 // Then: _reg if there is a 2nd register | |
| 5664 // Then: _long if it's a pair of instructions implementing a long | |
| 5665 // Then: _fat if it requires the big decoder | |
| 5666 // Or: _mem if it requires the big decoder and a memory unit. | |
| 5667 | |
| 5668 // Integer ALU reg operation | |
| 5669 pipe_class ialu_reg(rRegI dst) | |
| 5670 %{ | |
| 5671 single_instruction; | |
| 5672 dst : S4(write); | |
| 5673 dst : S3(read); | |
| 5674 DECODE : S0; // any decoder | |
| 5675 ALU : S3; // any alu | |
| 5676 %} | |
| 5677 | |
| 5678 // Long ALU reg operation | |
| 5679 pipe_class ialu_reg_long(rRegL dst) | |
| 5680 %{ | |
| 5681 instruction_count(2); | |
| 5682 dst : S4(write); | |
| 5683 dst : S3(read); | |
| 5684 DECODE : S0(2); // any 2 decoders | |
| 5685 ALU : S3(2); // both alus | |
| 5686 %} | |
| 5687 | |
| 5688 // Integer ALU reg operation using big decoder | |
| 5689 pipe_class ialu_reg_fat(rRegI dst) | |
| 5690 %{ | |
| 5691 single_instruction; | |
| 5692 dst : S4(write); | |
| 5693 dst : S3(read); | |
| 5694 D0 : S0; // big decoder only | |
| 5695 ALU : S3; // any alu | |
| 5696 %} | |
| 5697 | |
| 5698 // Long ALU reg operation using big decoder | |
| 5699 pipe_class ialu_reg_long_fat(rRegL dst) | |
| 5700 %{ | |
| 5701 instruction_count(2); | |
| 5702 dst : S4(write); | |
| 5703 dst : S3(read); | |
| 5704 D0 : S0(2); // big decoder only; twice | |
| 5705 ALU : S3(2); // any 2 alus | |
| 5706 %} | |
| 5707 | |
| 5708 // Integer ALU reg-reg operation | |
| 5709 pipe_class ialu_reg_reg(rRegI dst, rRegI src) | |
| 5710 %{ | |
| 5711 single_instruction; | |
| 5712 dst : S4(write); | |
| 5713 src : S3(read); | |
| 5714 DECODE : S0; // any decoder | |
| 5715 ALU : S3; // any alu | |
| 5716 %} | |
| 5717 | |
| 5718 // Long ALU reg-reg operation | |
| 5719 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src) | |
| 5720 %{ | |
| 5721 instruction_count(2); | |
| 5722 dst : S4(write); | |
| 5723 src : S3(read); | |
| 5724 DECODE : S0(2); // any 2 decoders | |
| 5725 ALU : S3(2); // both alus | |
| 5726 %} | |
| 5727 | |
| 5728 // Integer ALU reg-reg operation | |
| 5729 pipe_class ialu_reg_reg_fat(rRegI dst, memory src) | |
| 5730 %{ | |
| 5731 single_instruction; | |
| 5732 dst : S4(write); | |
| 5733 src : S3(read); | |
| 5734 D0 : S0; // big decoder only | |
| 5735 ALU : S3; // any alu | |
| 5736 %} | |
| 5737 | |
| 5738 // Long ALU reg-reg operation | |
| 5739 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src) | |
| 5740 %{ | |
| 5741 instruction_count(2); | |
| 5742 dst : S4(write); | |
| 5743 src : S3(read); | |
| 5744 D0 : S0(2); // big decoder only; twice | |
| 5745 ALU : S3(2); // both alus | |
| 5746 %} | |
| 5747 | |
| 5748 // Integer ALU reg-mem operation | |
| 5749 pipe_class ialu_reg_mem(rRegI dst, memory mem) | |
| 5750 %{ | |
| 5751 single_instruction; | |
| 5752 dst : S5(write); | |
| 5753 mem : S3(read); | |
| 5754 D0 : S0; // big decoder only | |
| 5755 ALU : S4; // any alu | |
| 5756 MEM : S3; // any mem | |
| 5757 %} | |
| 5758 | |
| 5759 // Integer mem operation (prefetch) | |
| 5760 pipe_class ialu_mem(memory mem) | |
| 5761 %{ | |
| 5762 single_instruction; | |
| 5763 mem : S3(read); | |
| 5764 D0 : S0; // big decoder only | |
| 5765 MEM : S3; // any mem | |
| 5766 %} | |
| 5767 | |
| 5768 // Integer Store to Memory | |
| 5769 pipe_class ialu_mem_reg(memory mem, rRegI src) | |
| 5770 %{ | |
| 5771 single_instruction; | |
| 5772 mem : S3(read); | |
| 5773 src : S5(read); | |
| 5774 D0 : S0; // big decoder only | |
| 5775 ALU : S4; // any alu | |
| 5776 MEM : S3; | |
| 5777 %} | |
| 5778 | |
| 5779 // // Long Store to Memory | |
| 5780 // pipe_class ialu_mem_long_reg(memory mem, rRegL src) | |
| 5781 // %{ | |
| 5782 // instruction_count(2); | |
| 5783 // mem : S3(read); | |
| 5784 // src : S5(read); | |
| 5785 // D0 : S0(2); // big decoder only; twice | |
| 5786 // ALU : S4(2); // any 2 alus | |
| 5787 // MEM : S3(2); // Both mems | |
| 5788 // %} | |
| 5789 | |
| 5790 // Integer Store to Memory | |
| 5791 pipe_class ialu_mem_imm(memory mem) | |
| 5792 %{ | |
| 5793 single_instruction; | |
| 5794 mem : S3(read); | |
| 5795 D0 : S0; // big decoder only | |
| 5796 ALU : S4; // any alu | |
| 5797 MEM : S3; | |
| 5798 %} | |
| 5799 | |
| 5800 // Integer ALU0 reg-reg operation | |
| 5801 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src) | |
| 5802 %{ | |
| 5803 single_instruction; | |
| 5804 dst : S4(write); | |
| 5805 src : S3(read); | |
| 5806 D0 : S0; // Big decoder only | |
| 5807 ALU0 : S3; // only alu0 | |
| 5808 %} | |
| 5809 | |
| 5810 // Integer ALU0 reg-mem operation | |
| 5811 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem) | |
| 5812 %{ | |
| 5813 single_instruction; | |
| 5814 dst : S5(write); | |
| 5815 mem : S3(read); | |
| 5816 D0 : S0; // big decoder only | |
| 5817 ALU0 : S4; // ALU0 only | |
| 5818 MEM : S3; // any mem | |
| 5819 %} | |
| 5820 | |
| 5821 // Integer ALU reg-reg operation | |
| 5822 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2) | |
| 5823 %{ | |
| 5824 single_instruction; | |
| 5825 cr : S4(write); | |
| 5826 src1 : S3(read); | |
| 5827 src2 : S3(read); | |
| 5828 DECODE : S0; // any decoder | |
| 5829 ALU : S3; // any alu | |
| 5830 %} | |
| 5831 | |
| 5832 // Integer ALU reg-imm operation | |
| 5833 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1) | |
| 5834 %{ | |
| 5835 single_instruction; | |
| 5836 cr : S4(write); | |
| 5837 src1 : S3(read); | |
| 5838 DECODE : S0; // any decoder | |
| 5839 ALU : S3; // any alu | |
| 5840 %} | |
| 5841 | |
| 5842 // Integer ALU reg-mem operation | |
| 5843 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2) | |
| 5844 %{ | |
| 5845 single_instruction; | |
| 5846 cr : S4(write); | |
| 5847 src1 : S3(read); | |
| 5848 src2 : S3(read); | |
| 5849 D0 : S0; // big decoder only | |
| 5850 ALU : S4; // any alu | |
| 5851 MEM : S3; | |
| 5852 %} | |
| 5853 | |
| 5854 // Conditional move reg-reg | |
| 5855 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y) | |
| 5856 %{ | |
| 5857 instruction_count(4); | |
| 5858 y : S4(read); | |
| 5859 q : S3(read); | |
| 5860 p : S3(read); | |
| 5861 DECODE : S0(4); // any decoder | |
| 5862 %} | |
| 5863 | |
| 5864 // Conditional move reg-reg | |
| 5865 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr) | |
| 5866 %{ | |
| 5867 single_instruction; | |
| 5868 dst : S4(write); | |
| 5869 src : S3(read); | |
| 5870 cr : S3(read); | |
| 5871 DECODE : S0; // any decoder | |
| 5872 %} | |
| 5873 | |
| 5874 // Conditional move reg-mem | |
| 5875 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src) | |
| 5876 %{ | |
| 5877 single_instruction; | |
| 5878 dst : S4(write); | |
| 5879 src : S3(read); | |
| 5880 cr : S3(read); | |
| 5881 DECODE : S0; // any decoder | |
| 5882 MEM : S3; | |
| 5883 %} | |
| 5884 | |
| 5885 // Conditional move reg-reg long | |
| 5886 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src) | |
| 5887 %{ | |
| 5888 single_instruction; | |
| 5889 dst : S4(write); | |
| 5890 src : S3(read); | |
| 5891 cr : S3(read); | |
| 5892 DECODE : S0(2); // any 2 decoders | |
| 5893 %} | |
| 5894 | |
| 5895 // XXX | |
| 5896 // // Conditional move double reg-reg | |
| 5897 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src) | |
| 5898 // %{ | |
| 5899 // single_instruction; | |
| 5900 // dst : S4(write); | |
| 5901 // src : S3(read); | |
| 5902 // cr : S3(read); | |
| 5903 // DECODE : S0; // any decoder | |
| 5904 // %} | |
| 5905 | |
| 5906 // Float reg-reg operation | |
| 5907 pipe_class fpu_reg(regD dst) | |
| 5908 %{ | |
| 5909 instruction_count(2); | |
| 5910 dst : S3(read); | |
| 5911 DECODE : S0(2); // any 2 decoders | |
| 5912 FPU : S3; | |
| 5913 %} | |
| 5914 | |
| 5915 // Float reg-reg operation | |
| 5916 pipe_class fpu_reg_reg(regD dst, regD src) | |
| 5917 %{ | |
| 5918 instruction_count(2); | |
| 5919 dst : S4(write); | |
| 5920 src : S3(read); | |
| 5921 DECODE : S0(2); // any 2 decoders | |
| 5922 FPU : S3; | |
| 5923 %} | |
| 5924 | |
| 5925 // Float reg-reg operation | |
| 5926 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2) | |
| 5927 %{ | |
| 5928 instruction_count(3); | |
| 5929 dst : S4(write); | |
| 5930 src1 : S3(read); | |
| 5931 src2 : S3(read); | |
| 5932 DECODE : S0(3); // any 3 decoders | |
| 5933 FPU : S3(2); | |
| 5934 %} | |
| 5935 | |
| 5936 // Float reg-reg operation | |
| 5937 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3) | |
| 5938 %{ | |
| 5939 instruction_count(4); | |
| 5940 dst : S4(write); | |
| 5941 src1 : S3(read); | |
| 5942 src2 : S3(read); | |
| 5943 src3 : S3(read); | |
| 5944 DECODE : S0(4); // any 3 decoders | |
| 5945 FPU : S3(2); | |
| 5946 %} | |
| 5947 | |
| 5948 // Float reg-reg operation | |
| 5949 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3) | |
| 5950 %{ | |
| 5951 instruction_count(4); | |
| 5952 dst : S4(write); | |
| 5953 src1 : S3(read); | |
| 5954 src2 : S3(read); | |
| 5955 src3 : S3(read); | |
| 5956 DECODE : S1(3); // any 3 decoders | |
| 5957 D0 : S0; // Big decoder only | |
| 5958 FPU : S3(2); | |
| 5959 MEM : S3; | |
| 5960 %} | |
| 5961 | |
| 5962 // Float reg-mem operation | |
| 5963 pipe_class fpu_reg_mem(regD dst, memory mem) | |
| 5964 %{ | |
| 5965 instruction_count(2); | |
| 5966 dst : S5(write); | |
| 5967 mem : S3(read); | |
| 5968 D0 : S0; // big decoder only | |
| 5969 DECODE : S1; // any decoder for FPU POP | |
| 5970 FPU : S4; | |
| 5971 MEM : S3; // any mem | |
| 5972 %} | |
| 5973 | |
| 5974 // Float reg-mem operation | |
| 5975 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem) | |
| 5976 %{ | |
| 5977 instruction_count(3); | |
| 5978 dst : S5(write); | |
| 5979 src1 : S3(read); | |
| 5980 mem : S3(read); | |
| 5981 D0 : S0; // big decoder only | |
| 5982 DECODE : S1(2); // any decoder for FPU POP | |
| 5983 FPU : S4; | |
| 5984 MEM : S3; // any mem | |
| 5985 %} | |
| 5986 | |
| 5987 // Float mem-reg operation | |
| 5988 pipe_class fpu_mem_reg(memory mem, regD src) | |
| 5989 %{ | |
| 5990 instruction_count(2); | |
| 5991 src : S5(read); | |
| 5992 mem : S3(read); | |
| 5993 DECODE : S0; // any decoder for FPU PUSH | |
| 5994 D0 : S1; // big decoder only | |
| 5995 FPU : S4; | |
| 5996 MEM : S3; // any mem | |
| 5997 %} | |
| 5998 | |
| 5999 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2) | |
| 6000 %{ | |
| 6001 instruction_count(3); | |
| 6002 src1 : S3(read); | |
| 6003 src2 : S3(read); | |
| 6004 mem : S3(read); | |
| 6005 DECODE : S0(2); // any decoder for FPU PUSH | |
| 6006 D0 : S1; // big decoder only | |
| 6007 FPU : S4; | |
| 6008 MEM : S3; // any mem | |
| 6009 %} | |
| 6010 | |
| 6011 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2) | |
| 6012 %{ | |
| 6013 instruction_count(3); | |
| 6014 src1 : S3(read); | |
| 6015 src2 : S3(read); | |
| 6016 mem : S4(read); | |
| 6017 DECODE : S0; // any decoder for FPU PUSH | |
| 6018 D0 : S0(2); // big decoder only | |
| 6019 FPU : S4; | |
| 6020 MEM : S3(2); // any mem | |
| 6021 %} | |
| 6022 | |
| 6023 pipe_class fpu_mem_mem(memory dst, memory src1) | |
| 6024 %{ | |
| 6025 instruction_count(2); | |
| 6026 src1 : S3(read); | |
| 6027 dst : S4(read); | |
| 6028 D0 : S0(2); // big decoder only | |
| 6029 MEM : S3(2); // any mem | |
| 6030 %} | |
| 6031 | |
| 6032 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2) | |
| 6033 %{ | |
| 6034 instruction_count(3); | |
| 6035 src1 : S3(read); | |
| 6036 src2 : S3(read); | |
| 6037 dst : S4(read); | |
| 6038 D0 : S0(3); // big decoder only | |
| 6039 FPU : S4; | |
| 6040 MEM : S3(3); // any mem | |
| 6041 %} | |
| 6042 | |
| 6043 pipe_class fpu_mem_reg_con(memory mem, regD src1) | |
| 6044 %{ | |
| 6045 instruction_count(3); | |
| 6046 src1 : S4(read); | |
| 6047 mem : S4(read); | |
| 6048 DECODE : S0; // any decoder for FPU PUSH | |
| 6049 D0 : S0(2); // big decoder only | |
| 6050 FPU : S4; | |
| 6051 MEM : S3(2); // any mem | |
| 6052 %} | |
| 6053 | |
| 6054 // Float load constant | |
| 6055 pipe_class fpu_reg_con(regD dst) | |
| 6056 %{ | |
| 6057 instruction_count(2); | |
| 6058 dst : S5(write); | |
| 6059 D0 : S0; // big decoder only for the load | |
| 6060 DECODE : S1; // any decoder for FPU POP | |
| 6061 FPU : S4; | |
| 6062 MEM : S3; // any mem | |
| 6063 %} | |
| 6064 | |
| 6065 // Float load constant | |
| 6066 pipe_class fpu_reg_reg_con(regD dst, regD src) | |
| 6067 %{ | |
| 6068 instruction_count(3); | |
| 6069 dst : S5(write); | |
| 6070 src : S3(read); | |
| 6071 D0 : S0; // big decoder only for the load | |
| 6072 DECODE : S1(2); // any decoder for FPU POP | |
| 6073 FPU : S4; | |
| 6074 MEM : S3; // any mem | |
| 6075 %} | |
| 6076 | |
| 6077 // UnConditional branch | |
| 6078 pipe_class pipe_jmp(label labl) | |
| 6079 %{ | |
| 6080 single_instruction; | |
| 6081 BR : S3; | |
| 6082 %} | |
| 6083 | |
| 6084 // Conditional branch | |
| 6085 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl) | |
| 6086 %{ | |
| 6087 single_instruction; | |
| 6088 cr : S1(read); | |
| 6089 BR : S3; | |
| 6090 %} | |
| 6091 | |
| 6092 // Allocation idiom | |
| 6093 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr) | |
| 6094 %{ | |
| 6095 instruction_count(1); force_serialization; | |
| 6096 fixed_latency(6); | |
| 6097 heap_ptr : S3(read); | |
| 6098 DECODE : S0(3); | |
| 6099 D0 : S2; | |
| 6100 MEM : S3; | |
| 6101 ALU : S3(2); | |
| 6102 dst : S5(write); | |
| 6103 BR : S5; | |
| 6104 %} | |
| 6105 | |
| 6106 // Generic big/slow expanded idiom | |
| 6107 pipe_class pipe_slow() | |
| 6108 %{ | |
| 6109 instruction_count(10); multiple_bundles; force_serialization; | |
| 6110 fixed_latency(100); | |
| 6111 D0 : S0(2); | |
| 6112 MEM : S3(2); | |
| 6113 %} | |
| 6114 | |
| 6115 // The real do-nothing guy | |
| 6116 pipe_class empty() | |
| 6117 %{ | |
| 6118 instruction_count(0); | |
| 6119 %} | |
| 6120 | |
| 6121 // Define the class for the Nop node | |
| 6122 define | |
| 6123 %{ | |
| 6124 MachNop = empty; | |
| 6125 %} | |
| 6126 | |
| 6127 %} | |
| 6128 | |
| 6129 //----------INSTRUCTIONS------------------------------------------------------- | |
| 6130 // | |
| 6131 // match -- States which machine-independent subtree may be replaced | |
| 6132 // by this instruction. | |
| 6133 // ins_cost -- The estimated cost of this instruction is used by instruction | |
| 6134 // selection to identify a minimum cost tree of machine | |
| 6135 // instructions that matches a tree of machine-independent | |
| 6136 // instructions. | |
| 6137 // format -- A string providing the disassembly for this instruction. | |
| 6138 // The value of an instruction's operand may be inserted | |
| 6139 // by referring to it with a '$' prefix. | |
| 6140 // opcode -- Three instruction opcodes may be provided. These are referred | |
| 6141 // to within an encode class as $primary, $secondary, and $tertiary | |
| 6142 // rrspectively. The primary opcode is commonly used to | |
| 6143 // indicate the type of machine instruction, while secondary | |
| 6144 // and tertiary are often used for prefix options or addressing | |
| 6145 // modes. | |
| 6146 // ins_encode -- A list of encode classes with parameters. The encode class | |
| 6147 // name must have been defined in an 'enc_class' specification | |
| 6148 // in the encode section of the architecture description. | |
| 6149 | |
| 6150 | |
| 6151 //----------Load/Store/Move Instructions--------------------------------------- | |
| 6152 //----------Load Instructions-------------------------------------------------- | |
| 6153 | |
| 6154 // Load Byte (8 bit signed) | |
| 6155 instruct loadB(rRegI dst, memory mem) | |
| 6156 %{ | |
| 6157 match(Set dst (LoadB mem)); | |
| 6158 | |
| 6159 ins_cost(125); | |
| 6160 format %{ "movsbl $dst, $mem\t# byte" %} | |
| 624 | 6161 |
| 6162 ins_encode %{ | |
| 6163 __ movsbl($dst$$Register, $mem$$Address); | |
| 6164 %} | |
| 6165 | |
| 0 | 6166 ins_pipe(ialu_reg_mem); |
| 6167 %} | |
| 6168 | |
| 624 | 6169 // Load Byte (8 bit signed) into Long Register |
| 6170 instruct loadB2L(rRegL dst, memory mem) | |
| 6171 %{ | |
| 6172 match(Set dst (ConvI2L (LoadB mem))); | |
| 6173 | |
| 6174 ins_cost(125); | |
| 6175 format %{ "movsbq $dst, $mem\t# byte -> long" %} | |
| 6176 | |
| 6177 ins_encode %{ | |
| 6178 __ movsbq($dst$$Register, $mem$$Address); | |
| 6179 %} | |
| 6180 | |
| 6181 ins_pipe(ialu_reg_mem); | |
| 6182 %} | |
| 6183 | |
| 6184 // Load Unsigned Byte (8 bit UNsigned) | |
| 6185 instruct loadUB(rRegI dst, memory mem) | |
| 6186 %{ | |
| 6187 match(Set dst (LoadUB mem)); | |
| 0 | 6188 |
| 6189 ins_cost(125); | |
| 6190 format %{ "movzbl $dst, $mem\t# ubyte" %} | |
| 624 | 6191 |
| 6192 ins_encode %{ | |
| 6193 __ movzbl($dst$$Register, $mem$$Address); | |
| 6194 %} | |
| 6195 | |
| 0 | 6196 ins_pipe(ialu_reg_mem); |
| 6197 %} | |
| 6198 | |
| 624 | 6199 // Load Unsigned Byte (8 bit UNsigned) into Long Register |
| 6200 instruct loadUB2L(rRegL dst, memory mem) | |
| 6201 %{ | |
| 6202 match(Set dst (ConvI2L (LoadUB mem))); | |
| 6203 | |
| 6204 ins_cost(125); | |
| 6205 format %{ "movzbq $dst, $mem\t# ubyte -> long" %} | |
| 6206 | |
| 6207 ins_encode %{ | |
| 6208 __ movzbq($dst$$Register, $mem$$Address); | |
| 6209 %} | |
| 6210 | |
| 6211 ins_pipe(ialu_reg_mem); | |
| 6212 %} | |
| 0 | 6213 |
| 6214 // Load Short (16 bit signed) | |
| 6215 instruct loadS(rRegI dst, memory mem) | |
| 6216 %{ | |
| 6217 match(Set dst (LoadS mem)); | |
| 6218 | |
| 624 | 6219 ins_cost(125); |
| 0 | 6220 format %{ "movswl $dst, $mem\t# short" %} |
| 624 | 6221 |
| 6222 ins_encode %{ | |
| 6223 __ movswl($dst$$Register, $mem$$Address); | |
| 6224 %} | |
| 6225 | |
| 0 | 6226 ins_pipe(ialu_reg_mem); |
| 6227 %} | |
| 6228 | |
| 624 | 6229 // Load Short (16 bit signed) into Long Register |
| 6230 instruct loadS2L(rRegL dst, memory mem) | |
| 6231 %{ | |
| 6232 match(Set dst (ConvI2L (LoadS mem))); | |
| 6233 | |
| 6234 ins_cost(125); | |
| 6235 format %{ "movswq $dst, $mem\t# short -> long" %} | |
| 6236 | |
| 6237 ins_encode %{ | |
| 6238 __ movswq($dst$$Register, $mem$$Address); | |
| 6239 %} | |
| 6240 | |
| 6241 ins_pipe(ialu_reg_mem); | |
| 6242 %} | |
| 0 | 6243 |
|
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
420
diff
changeset
|
6244 // Load Unsigned Short/Char (16 bit UNsigned) |
|
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
420
diff
changeset
|
6245 instruct loadUS(rRegI dst, memory mem) |
|
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
420
diff
changeset
|
6246 %{ |
|
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
420
diff
changeset
|
6247 match(Set dst (LoadUS mem)); |
| 0 | 6248 |
| 6249 ins_cost(125); | |
|
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
420
diff
changeset
|
6250 format %{ "movzwl $dst, $mem\t# ushort/char" %} |
| 624 | 6251 |
| 6252 ins_encode %{ | |
| 6253 __ movzwl($dst$$Register, $mem$$Address); | |
| 6254 %} | |
| 6255 | |
| 0 | 6256 ins_pipe(ialu_reg_mem); |
| 6257 %} | |
| 6258 | |
| 624 | 6259 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register |
| 6260 instruct loadUS2L(rRegL dst, memory mem) | |
| 6261 %{ | |
| 6262 match(Set dst (ConvI2L (LoadUS mem))); | |
| 6263 | |
| 6264 ins_cost(125); | |
| 6265 format %{ "movzwq $dst, $mem\t# ushort/char -> long" %} | |
| 6266 | |
| 6267 ins_encode %{ | |
| 6268 __ movzwq($dst$$Register, $mem$$Address); | |
| 6269 %} | |
| 6270 | |
| 6271 ins_pipe(ialu_reg_mem); | |
| 6272 %} | |
| 0 | 6273 |
| 6274 // Load Integer | |
| 6275 instruct loadI(rRegI dst, memory mem) | |
| 6276 %{ | |
| 6277 match(Set dst (LoadI mem)); | |
| 6278 | |
| 624 | 6279 ins_cost(125); |
| 0 | 6280 format %{ "movl $dst, $mem\t# int" %} |
| 624 | 6281 |
| 6282 ins_encode %{ | |
| 6283 __ movl($dst$$Register, $mem$$Address); | |
| 6284 %} | |
| 6285 | |
| 6286 ins_pipe(ialu_reg_mem); | |
| 6287 %} | |
| 6288 | |
| 6289 // Load Integer into Long Register | |
| 6290 instruct loadI2L(rRegL dst, memory mem) | |
| 6291 %{ | |
| 6292 match(Set dst (ConvI2L (LoadI mem))); | |
| 6293 | |
| 6294 ins_cost(125); | |
| 6295 format %{ "movslq $dst, $mem\t# int -> long" %} | |
| 6296 | |
| 6297 ins_encode %{ | |
| 6298 __ movslq($dst$$Register, $mem$$Address); | |
| 6299 %} | |
| 6300 | |
| 6301 ins_pipe(ialu_reg_mem); | |
| 6302 %} | |
| 6303 | |
| 6304 // Load Unsigned Integer into Long Register | |
| 6305 instruct loadUI2L(rRegL dst, memory mem) | |
| 6306 %{ | |
| 6307 match(Set dst (LoadUI2L mem)); | |
| 6308 | |
| 6309 ins_cost(125); | |
| 6310 format %{ "movl $dst, $mem\t# uint -> long" %} | |
| 6311 | |
| 6312 ins_encode %{ | |
| 6313 __ movl($dst$$Register, $mem$$Address); | |
| 6314 %} | |
| 6315 | |
| 0 | 6316 ins_pipe(ialu_reg_mem); |
| 6317 %} | |
| 6318 | |
| 6319 // Load Long | |
| 6320 instruct loadL(rRegL dst, memory mem) | |
| 6321 %{ | |
| 6322 match(Set dst (LoadL mem)); | |
| 6323 | |
| 624 | 6324 ins_cost(125); |
| 0 | 6325 format %{ "movq $dst, $mem\t# long" %} |
| 624 | 6326 |
| 6327 ins_encode %{ | |
| 6328 __ movq($dst$$Register, $mem$$Address); | |
| 6329 %} | |
| 6330 | |
| 0 | 6331 ins_pipe(ialu_reg_mem); // XXX |
| 6332 %} | |
| 6333 | |
| 6334 // Load Range | |
| 6335 instruct loadRange(rRegI dst, memory mem) | |
| 6336 %{ | |
| 6337 match(Set dst (LoadRange mem)); | |
| 6338 | |
| 6339 ins_cost(125); // XXX | |
| 6340 format %{ "movl $dst, $mem\t# range" %} | |
| 6341 opcode(0x8B); | |
| 6342 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6343 ins_pipe(ialu_reg_mem); | |
| 6344 %} | |
| 6345 | |
| 6346 // Load Pointer | |
| 6347 instruct loadP(rRegP dst, memory mem) | |
| 6348 %{ | |
| 6349 match(Set dst (LoadP mem)); | |
| 6350 | |
| 6351 ins_cost(125); // XXX | |
| 6352 format %{ "movq $dst, $mem\t# ptr" %} | |
| 6353 opcode(0x8B); | |
| 6354 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6355 ins_pipe(ialu_reg_mem); // XXX | |
| 6356 %} | |
| 6357 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6358 // Load Compressed Pointer |
| 163 | 6359 instruct loadN(rRegN dst, memory mem) |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6360 %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6361 match(Set dst (LoadN mem)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6362 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6363 ins_cost(125); // XXX |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6364 format %{ "movl $dst, $mem\t# compressed ptr" %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6365 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6366 __ movl($dst$$Register, $mem$$Address); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6367 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6368 ins_pipe(ialu_reg_mem); // XXX |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6369 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6370 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6371 |
| 0 | 6372 // Load Klass Pointer |
| 6373 instruct loadKlass(rRegP dst, memory mem) | |
| 6374 %{ | |
| 6375 match(Set dst (LoadKlass mem)); | |
| 6376 | |
| 6377 ins_cost(125); // XXX | |
| 6378 format %{ "movq $dst, $mem\t# class" %} | |
| 6379 opcode(0x8B); | |
| 6380 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6381 ins_pipe(ialu_reg_mem); // XXX | |
| 6382 %} | |
| 6383 | |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6384 // Load narrow Klass Pointer |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6385 instruct loadNKlass(rRegN dst, memory mem) |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6386 %{ |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6387 match(Set dst (LoadNKlass mem)); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6388 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6389 ins_cost(125); // XXX |
|
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
169
diff
changeset
|
6390 format %{ "movl $dst, $mem\t# compressed klass ptr" %} |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6391 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6392 __ movl($dst$$Register, $mem$$Address); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6393 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6394 ins_pipe(ialu_reg_mem); // XXX |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6395 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6396 |
| 0 | 6397 // Load Float |
| 6398 instruct loadF(regF dst, memory mem) | |
| 6399 %{ | |
| 6400 match(Set dst (LoadF mem)); | |
| 6401 | |
| 6402 ins_cost(145); // XXX | |
| 6403 format %{ "movss $dst, $mem\t# float" %} | |
| 6404 opcode(0xF3, 0x0F, 0x10); | |
| 6405 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem)); | |
| 6406 ins_pipe(pipe_slow); // XXX | |
| 6407 %} | |
| 6408 | |
| 6409 // Load Double | |
| 6410 instruct loadD_partial(regD dst, memory mem) | |
| 6411 %{ | |
| 6412 predicate(!UseXmmLoadAndClearUpper); | |
| 6413 match(Set dst (LoadD mem)); | |
| 6414 | |
| 6415 ins_cost(145); // XXX | |
| 6416 format %{ "movlpd $dst, $mem\t# double" %} | |
| 6417 opcode(0x66, 0x0F, 0x12); | |
| 6418 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem)); | |
| 6419 ins_pipe(pipe_slow); // XXX | |
| 6420 %} | |
| 6421 | |
| 6422 instruct loadD(regD dst, memory mem) | |
| 6423 %{ | |
| 6424 predicate(UseXmmLoadAndClearUpper); | |
| 6425 match(Set dst (LoadD mem)); | |
| 6426 | |
| 6427 ins_cost(145); // XXX | |
| 6428 format %{ "movsd $dst, $mem\t# double" %} | |
| 6429 opcode(0xF2, 0x0F, 0x10); | |
| 6430 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem)); | |
| 6431 ins_pipe(pipe_slow); // XXX | |
| 6432 %} | |
| 6433 | |
| 6434 // Load Aligned Packed Byte to XMM register | |
| 6435 instruct loadA8B(regD dst, memory mem) %{ | |
| 6436 match(Set dst (Load8B mem)); | |
| 6437 ins_cost(125); | |
| 6438 format %{ "MOVQ $dst,$mem\t! packed8B" %} | |
| 6439 ins_encode( movq_ld(dst, mem)); | |
| 6440 ins_pipe( pipe_slow ); | |
| 6441 %} | |
| 6442 | |
| 6443 // Load Aligned Packed Short to XMM register | |
| 6444 instruct loadA4S(regD dst, memory mem) %{ | |
| 6445 match(Set dst (Load4S mem)); | |
| 6446 ins_cost(125); | |
| 6447 format %{ "MOVQ $dst,$mem\t! packed4S" %} | |
| 6448 ins_encode( movq_ld(dst, mem)); | |
| 6449 ins_pipe( pipe_slow ); | |
| 6450 %} | |
| 6451 | |
| 6452 // Load Aligned Packed Char to XMM register | |
| 6453 instruct loadA4C(regD dst, memory mem) %{ | |
| 6454 match(Set dst (Load4C mem)); | |
| 6455 ins_cost(125); | |
| 6456 format %{ "MOVQ $dst,$mem\t! packed4C" %} | |
| 6457 ins_encode( movq_ld(dst, mem)); | |
| 6458 ins_pipe( pipe_slow ); | |
| 6459 %} | |
| 6460 | |
| 6461 // Load Aligned Packed Integer to XMM register | |
| 6462 instruct load2IU(regD dst, memory mem) %{ | |
| 6463 match(Set dst (Load2I mem)); | |
| 6464 ins_cost(125); | |
| 6465 format %{ "MOVQ $dst,$mem\t! packed2I" %} | |
| 6466 ins_encode( movq_ld(dst, mem)); | |
| 6467 ins_pipe( pipe_slow ); | |
| 6468 %} | |
| 6469 | |
| 6470 // Load Aligned Packed Single to XMM | |
| 6471 instruct loadA2F(regD dst, memory mem) %{ | |
| 6472 match(Set dst (Load2F mem)); | |
| 6473 ins_cost(145); | |
| 6474 format %{ "MOVQ $dst,$mem\t! packed2F" %} | |
| 6475 ins_encode( movq_ld(dst, mem)); | |
| 6476 ins_pipe( pipe_slow ); | |
| 6477 %} | |
| 6478 | |
| 6479 // Load Effective Address | |
| 6480 instruct leaP8(rRegP dst, indOffset8 mem) | |
| 6481 %{ | |
| 6482 match(Set dst mem); | |
| 6483 | |
| 6484 ins_cost(110); // XXX | |
| 6485 format %{ "leaq $dst, $mem\t# ptr 8" %} | |
| 6486 opcode(0x8D); | |
| 6487 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6488 ins_pipe(ialu_reg_reg_fat); | |
| 6489 %} | |
| 6490 | |
| 6491 instruct leaP32(rRegP dst, indOffset32 mem) | |
| 6492 %{ | |
| 6493 match(Set dst mem); | |
| 6494 | |
| 6495 ins_cost(110); | |
| 6496 format %{ "leaq $dst, $mem\t# ptr 32" %} | |
| 6497 opcode(0x8D); | |
| 6498 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6499 ins_pipe(ialu_reg_reg_fat); | |
| 6500 %} | |
| 6501 | |
| 6502 // instruct leaPIdx(rRegP dst, indIndex mem) | |
| 6503 // %{ | |
| 6504 // match(Set dst mem); | |
| 6505 | |
| 6506 // ins_cost(110); | |
| 6507 // format %{ "leaq $dst, $mem\t# ptr idx" %} | |
| 6508 // opcode(0x8D); | |
| 6509 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6510 // ins_pipe(ialu_reg_reg_fat); | |
| 6511 // %} | |
| 6512 | |
| 6513 instruct leaPIdxOff(rRegP dst, indIndexOffset mem) | |
| 6514 %{ | |
| 6515 match(Set dst mem); | |
| 6516 | |
| 6517 ins_cost(110); | |
| 6518 format %{ "leaq $dst, $mem\t# ptr idxoff" %} | |
| 6519 opcode(0x8D); | |
| 6520 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6521 ins_pipe(ialu_reg_reg_fat); | |
| 6522 %} | |
| 6523 | |
| 6524 instruct leaPIdxScale(rRegP dst, indIndexScale mem) | |
| 6525 %{ | |
| 6526 match(Set dst mem); | |
| 6527 | |
| 6528 ins_cost(110); | |
| 6529 format %{ "leaq $dst, $mem\t# ptr idxscale" %} | |
| 6530 opcode(0x8D); | |
| 6531 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6532 ins_pipe(ialu_reg_reg_fat); | |
| 6533 %} | |
| 6534 | |
| 6535 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem) | |
| 6536 %{ | |
| 6537 match(Set dst mem); | |
| 6538 | |
| 6539 ins_cost(110); | |
| 6540 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %} | |
| 6541 opcode(0x8D); | |
| 6542 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 6543 ins_pipe(ialu_reg_reg_fat); | |
| 6544 %} | |
| 6545 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6546 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6547 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6548 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6549 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6550 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6551 format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6552 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6553 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6554 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6555 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6556 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6557 // Load Effective Address which uses Narrow (32-bits) oop |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6558 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6559 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6560 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6561 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6562 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6563 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6564 format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6565 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6566 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6567 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6568 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6569 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6570 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6571 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6572 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6573 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6574 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6575 ins_cost(110); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6576 format %{ "leaq $dst, $mem\t# ptr off8narrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6577 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6578 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6579 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6580 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6581 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6582 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6583 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6584 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6585 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6586 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6587 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6588 format %{ "leaq $dst, $mem\t# ptr off32narrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6589 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6590 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6591 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6592 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6593 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6594 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6595 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6596 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6597 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6598 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6599 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6600 format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6601 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6602 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6603 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6604 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6605 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6606 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6607 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6608 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6609 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6610 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6611 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6612 format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6613 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6614 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6615 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6616 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6617 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6618 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6619 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6620 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6621 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6622 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6623 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6624 format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6625 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6626 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6627 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6628 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6629 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6630 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6631 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6632 predicate(Universe::narrow_oop_shift() == 0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6633 match(Set dst mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6634 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6635 ins_cost(110); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6636 format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6637 opcode(0x8D); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6638 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6639 ins_pipe(ialu_reg_reg_fat); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6640 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6641 |
| 0 | 6642 instruct loadConI(rRegI dst, immI src) |
| 6643 %{ | |
| 6644 match(Set dst src); | |
| 6645 | |
| 6646 format %{ "movl $dst, $src\t# int" %} | |
| 6647 ins_encode(load_immI(dst, src)); | |
| 6648 ins_pipe(ialu_reg_fat); // XXX | |
| 6649 %} | |
| 6650 | |
| 6651 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr) | |
| 6652 %{ | |
| 6653 match(Set dst src); | |
| 6654 effect(KILL cr); | |
| 6655 | |
| 6656 ins_cost(50); | |
| 6657 format %{ "xorl $dst, $dst\t# int" %} | |
| 6658 opcode(0x33); /* + rd */ | |
| 6659 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst)); | |
| 6660 ins_pipe(ialu_reg); | |
| 6661 %} | |
| 6662 | |
| 6663 instruct loadConL(rRegL dst, immL src) | |
| 6664 %{ | |
| 6665 match(Set dst src); | |
| 6666 | |
| 6667 ins_cost(150); | |
| 6668 format %{ "movq $dst, $src\t# long" %} | |
| 6669 ins_encode(load_immL(dst, src)); | |
| 6670 ins_pipe(ialu_reg); | |
| 6671 %} | |
| 6672 | |
| 6673 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr) | |
| 6674 %{ | |
| 6675 match(Set dst src); | |
| 6676 effect(KILL cr); | |
| 6677 | |
| 6678 ins_cost(50); | |
| 6679 format %{ "xorl $dst, $dst\t# long" %} | |
| 6680 opcode(0x33); /* + rd */ | |
| 6681 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst)); | |
| 6682 ins_pipe(ialu_reg); // XXX | |
| 6683 %} | |
| 6684 | |
| 6685 instruct loadConUL32(rRegL dst, immUL32 src) | |
| 6686 %{ | |
| 6687 match(Set dst src); | |
| 6688 | |
| 6689 ins_cost(60); | |
| 6690 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %} | |
| 6691 ins_encode(load_immUL32(dst, src)); | |
| 6692 ins_pipe(ialu_reg); | |
| 6693 %} | |
| 6694 | |
| 6695 instruct loadConL32(rRegL dst, immL32 src) | |
| 6696 %{ | |
| 6697 match(Set dst src); | |
| 6698 | |
| 6699 ins_cost(70); | |
| 6700 format %{ "movq $dst, $src\t# long (32-bit)" %} | |
| 6701 ins_encode(load_immL32(dst, src)); | |
| 6702 ins_pipe(ialu_reg); | |
| 6703 %} | |
| 6704 | |
| 6705 instruct loadConP(rRegP dst, immP src) | |
| 6706 %{ | |
| 6707 match(Set dst src); | |
| 6708 | |
| 6709 format %{ "movq $dst, $src\t# ptr" %} | |
| 6710 ins_encode(load_immP(dst, src)); | |
| 6711 ins_pipe(ialu_reg_fat); // XXX | |
| 6712 %} | |
| 6713 | |
| 6714 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr) | |
| 6715 %{ | |
| 6716 match(Set dst src); | |
| 6717 effect(KILL cr); | |
| 6718 | |
| 6719 ins_cost(50); | |
| 6720 format %{ "xorl $dst, $dst\t# ptr" %} | |
| 6721 opcode(0x33); /* + rd */ | |
| 6722 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst)); | |
| 6723 ins_pipe(ialu_reg); | |
| 6724 %} | |
| 6725 | |
| 6726 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr) | |
| 6727 %{ | |
| 6728 match(Set dst src); | |
| 6729 effect(KILL cr); | |
| 6730 | |
| 6731 ins_cost(60); | |
| 6732 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %} | |
| 6733 ins_encode(load_immP31(dst, src)); | |
| 6734 ins_pipe(ialu_reg); | |
| 6735 %} | |
| 6736 | |
| 6737 instruct loadConF(regF dst, immF src) | |
| 6738 %{ | |
| 6739 match(Set dst src); | |
| 6740 ins_cost(125); | |
| 6741 | |
| 6742 format %{ "movss $dst, [$src]" %} | |
| 6743 ins_encode(load_conF(dst, src)); | |
| 6744 ins_pipe(pipe_slow); | |
| 6745 %} | |
| 6746 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6747 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6748 match(Set dst src); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6749 effect(KILL cr); |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6750 format %{ "xorq $dst, $src\t# compressed NULL ptr" %} |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6751 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6752 __ xorq($dst$$Register, $dst$$Register); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6753 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6754 ins_pipe(ialu_reg); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6755 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6756 |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6757 instruct loadConN(rRegN dst, immN src) %{ |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6758 match(Set dst src); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6759 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6760 ins_cost(125); |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6761 format %{ "movl $dst, $src\t# compressed ptr" %} |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6762 ins_encode %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6763 address con = (address)$src$$constant; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6764 if (con == NULL) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6765 ShouldNotReachHere(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6766 } else { |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
6767 __ set_narrow_oop($dst$$Register, (jobject)$src$$constant); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6768 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6769 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6770 ins_pipe(ialu_reg_fat); // XXX |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6771 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
6772 |
| 0 | 6773 instruct loadConF0(regF dst, immF0 src) |
| 6774 %{ | |
| 6775 match(Set dst src); | |
| 6776 ins_cost(100); | |
| 6777 | |
| 6778 format %{ "xorps $dst, $dst\t# float 0.0" %} | |
| 6779 opcode(0x0F, 0x57); | |
| 6780 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); | |
| 6781 ins_pipe(pipe_slow); | |
| 6782 %} | |
| 6783 | |
| 6784 // Use the same format since predicate() can not be used here. | |
| 6785 instruct loadConD(regD dst, immD src) | |
| 6786 %{ | |
| 6787 match(Set dst src); | |
| 6788 ins_cost(125); | |
| 6789 | |
| 6790 format %{ "movsd $dst, [$src]" %} | |
| 6791 ins_encode(load_conD(dst, src)); | |
| 6792 ins_pipe(pipe_slow); | |
| 6793 %} | |
| 6794 | |
| 6795 instruct loadConD0(regD dst, immD0 src) | |
| 6796 %{ | |
| 6797 match(Set dst src); | |
| 6798 ins_cost(100); | |
| 6799 | |
| 6800 format %{ "xorpd $dst, $dst\t# double 0.0" %} | |
| 6801 opcode(0x66, 0x0F, 0x57); | |
| 6802 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst)); | |
| 6803 ins_pipe(pipe_slow); | |
| 6804 %} | |
| 6805 | |
| 6806 instruct loadSSI(rRegI dst, stackSlotI src) | |
| 6807 %{ | |
| 6808 match(Set dst src); | |
| 6809 | |
| 6810 ins_cost(125); | |
| 6811 format %{ "movl $dst, $src\t# int stk" %} | |
| 6812 opcode(0x8B); | |
| 6813 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 6814 ins_pipe(ialu_reg_mem); | |
| 6815 %} | |
| 6816 | |
| 6817 instruct loadSSL(rRegL dst, stackSlotL src) | |
| 6818 %{ | |
| 6819 match(Set dst src); | |
| 6820 | |
| 6821 ins_cost(125); | |
| 6822 format %{ "movq $dst, $src\t# long stk" %} | |
| 6823 opcode(0x8B); | |
| 6824 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 6825 ins_pipe(ialu_reg_mem); | |
| 6826 %} | |
| 6827 | |
| 6828 instruct loadSSP(rRegP dst, stackSlotP src) | |
| 6829 %{ | |
| 6830 match(Set dst src); | |
| 6831 | |
| 6832 ins_cost(125); | |
| 6833 format %{ "movq $dst, $src\t# ptr stk" %} | |
| 6834 opcode(0x8B); | |
| 6835 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 6836 ins_pipe(ialu_reg_mem); | |
| 6837 %} | |
| 6838 | |
| 6839 instruct loadSSF(regF dst, stackSlotF src) | |
| 6840 %{ | |
| 6841 match(Set dst src); | |
| 6842 | |
| 6843 ins_cost(125); | |
| 6844 format %{ "movss $dst, $src\t# float stk" %} | |
| 6845 opcode(0xF3, 0x0F, 0x10); | |
| 6846 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 6847 ins_pipe(pipe_slow); // XXX | |
| 6848 %} | |
| 6849 | |
| 6850 // Use the same format since predicate() can not be used here. | |
| 6851 instruct loadSSD(regD dst, stackSlotD src) | |
| 6852 %{ | |
| 6853 match(Set dst src); | |
| 6854 | |
| 6855 ins_cost(125); | |
| 6856 format %{ "movsd $dst, $src\t# double stk" %} | |
| 6857 ins_encode %{ | |
| 6858 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp)); | |
| 6859 %} | |
| 6860 ins_pipe(pipe_slow); // XXX | |
| 6861 %} | |
| 6862 | |
| 6863 // Prefetch instructions. | |
| 6864 // Must be safe to execute with invalid address (cannot fault). | |
| 6865 | |
| 6866 instruct prefetchr( memory mem ) %{ | |
| 6867 predicate(ReadPrefetchInstr==3); | |
| 6868 match(PrefetchRead mem); | |
| 6869 ins_cost(125); | |
| 6870 | |
| 6871 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %} | |
| 6872 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */ | |
| 6873 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem)); | |
| 6874 ins_pipe(ialu_mem); | |
| 6875 %} | |
| 6876 | |
| 6877 instruct prefetchrNTA( memory mem ) %{ | |
| 6878 predicate(ReadPrefetchInstr==0); | |
| 6879 match(PrefetchRead mem); | |
| 6880 ins_cost(125); | |
| 6881 | |
| 6882 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %} | |
| 6883 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */ | |
| 6884 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem)); | |
| 6885 ins_pipe(ialu_mem); | |
| 6886 %} | |
| 6887 | |
| 6888 instruct prefetchrT0( memory mem ) %{ | |
| 6889 predicate(ReadPrefetchInstr==1); | |
| 6890 match(PrefetchRead mem); | |
| 6891 ins_cost(125); | |
| 6892 | |
| 6893 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %} | |
| 6894 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */ | |
| 6895 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem)); | |
| 6896 ins_pipe(ialu_mem); | |
| 6897 %} | |
| 6898 | |
| 6899 instruct prefetchrT2( memory mem ) %{ | |
| 6900 predicate(ReadPrefetchInstr==2); | |
| 6901 match(PrefetchRead mem); | |
| 6902 ins_cost(125); | |
| 6903 | |
| 6904 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %} | |
| 6905 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */ | |
| 6906 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem)); | |
| 6907 ins_pipe(ialu_mem); | |
| 6908 %} | |
| 6909 | |
| 6910 instruct prefetchw( memory mem ) %{ | |
| 6911 predicate(AllocatePrefetchInstr==3); | |
| 6912 match(PrefetchWrite mem); | |
| 6913 ins_cost(125); | |
| 6914 | |
| 6915 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %} | |
| 6916 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */ | |
| 6917 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem)); | |
| 6918 ins_pipe(ialu_mem); | |
| 6919 %} | |
| 6920 | |
| 6921 instruct prefetchwNTA( memory mem ) %{ | |
| 6922 predicate(AllocatePrefetchInstr==0); | |
| 6923 match(PrefetchWrite mem); | |
| 6924 ins_cost(125); | |
| 6925 | |
| 6926 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %} | |
| 6927 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */ | |
| 6928 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem)); | |
| 6929 ins_pipe(ialu_mem); | |
| 6930 %} | |
| 6931 | |
| 6932 instruct prefetchwT0( memory mem ) %{ | |
| 6933 predicate(AllocatePrefetchInstr==1); | |
| 6934 match(PrefetchWrite mem); | |
| 6935 ins_cost(125); | |
| 6936 | |
| 6937 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %} | |
| 6938 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */ | |
| 6939 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem)); | |
| 6940 ins_pipe(ialu_mem); | |
| 6941 %} | |
| 6942 | |
| 6943 instruct prefetchwT2( memory mem ) %{ | |
| 6944 predicate(AllocatePrefetchInstr==2); | |
| 6945 match(PrefetchWrite mem); | |
| 6946 ins_cost(125); | |
| 6947 | |
| 6948 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %} | |
| 6949 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */ | |
| 6950 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem)); | |
| 6951 ins_pipe(ialu_mem); | |
| 6952 %} | |
| 6953 | |
| 6954 //----------Store Instructions------------------------------------------------- | |
| 6955 | |
| 6956 // Store Byte | |
| 6957 instruct storeB(memory mem, rRegI src) | |
| 6958 %{ | |
| 6959 match(Set mem (StoreB mem src)); | |
| 6960 | |
| 6961 ins_cost(125); // XXX | |
| 6962 format %{ "movb $mem, $src\t# byte" %} | |
| 6963 opcode(0x88); | |
| 6964 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem)); | |
| 6965 ins_pipe(ialu_mem_reg); | |
| 6966 %} | |
| 6967 | |
| 6968 // Store Char/Short | |
| 6969 instruct storeC(memory mem, rRegI src) | |
| 6970 %{ | |
| 6971 match(Set mem (StoreC mem src)); | |
| 6972 | |
| 6973 ins_cost(125); // XXX | |
| 6974 format %{ "movw $mem, $src\t# char/short" %} | |
| 6975 opcode(0x89); | |
| 6976 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem)); | |
| 6977 ins_pipe(ialu_mem_reg); | |
| 6978 %} | |
| 6979 | |
| 6980 // Store Integer | |
| 6981 instruct storeI(memory mem, rRegI src) | |
| 6982 %{ | |
| 6983 match(Set mem (StoreI mem src)); | |
| 6984 | |
| 6985 ins_cost(125); // XXX | |
| 6986 format %{ "movl $mem, $src\t# int" %} | |
| 6987 opcode(0x89); | |
| 6988 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem)); | |
| 6989 ins_pipe(ialu_mem_reg); | |
| 6990 %} | |
| 6991 | |
| 6992 // Store Long | |
| 6993 instruct storeL(memory mem, rRegL src) | |
| 6994 %{ | |
| 6995 match(Set mem (StoreL mem src)); | |
| 6996 | |
| 6997 ins_cost(125); // XXX | |
| 6998 format %{ "movq $mem, $src\t# long" %} | |
| 6999 opcode(0x89); | |
| 7000 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem)); | |
| 7001 ins_pipe(ialu_mem_reg); // XXX | |
| 7002 %} | |
| 7003 | |
| 7004 // Store Pointer | |
| 7005 instruct storeP(memory mem, any_RegP src) | |
| 7006 %{ | |
| 7007 match(Set mem (StoreP mem src)); | |
| 7008 | |
| 7009 ins_cost(125); // XXX | |
| 7010 format %{ "movq $mem, $src\t# ptr" %} | |
| 7011 opcode(0x89); | |
| 7012 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem)); | |
| 7013 ins_pipe(ialu_mem_reg); | |
| 7014 %} | |
| 7015 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7016 instruct storeImmP0(memory mem, immP0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7017 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7018 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7019 match(Set mem (StoreP mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7020 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7021 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7022 format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7023 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7024 __ movq($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7025 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7026 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7027 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7028 |
| 0 | 7029 // Store NULL Pointer, mark word, or other simple pointer constant. |
| 7030 instruct storeImmP(memory mem, immP31 src) | |
| 7031 %{ | |
| 7032 match(Set mem (StoreP mem src)); | |
| 7033 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7034 ins_cost(150); // XXX |
| 0 | 7035 format %{ "movq $mem, $src\t# ptr" %} |
| 7036 opcode(0xC7); /* C7 /0 */ | |
| 7037 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src)); | |
| 7038 ins_pipe(ialu_mem_imm); | |
| 7039 %} | |
| 7040 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7041 // Store Compressed Pointer |
| 163 | 7042 instruct storeN(memory mem, rRegN src) |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7043 %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7044 match(Set mem (StoreN mem src)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7045 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7046 ins_cost(125); // XXX |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7047 format %{ "movl $mem, $src\t# compressed ptr" %} |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7048 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7049 __ movl($mem$$Address, $src$$Register); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7050 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7051 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7052 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7053 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7054 instruct storeImmN0(memory mem, immN0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7055 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7056 predicate(Universe::narrow_oop_base() == NULL); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7057 match(Set mem (StoreN mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7058 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7059 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7060 format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7061 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7062 __ movl($mem$$Address, r12); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7063 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7064 ins_pipe(ialu_mem_reg); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7065 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7066 |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7067 instruct storeImmN(memory mem, immN src) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7068 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7069 match(Set mem (StoreN mem src)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7070 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7071 ins_cost(150); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7072 format %{ "movl $mem, $src\t# compressed ptr" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7073 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7074 address con = (address)$src$$constant; |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7075 if (con == NULL) { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7076 __ movl($mem$$Address, (int32_t)0); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7077 } else { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7078 __ set_narrow_oop($mem$$Address, (jobject)$src$$constant); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7079 } |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7080 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7081 ins_pipe(ialu_mem_imm); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7082 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7083 |
| 0 | 7084 // Store Integer Immediate |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7085 instruct storeImmI0(memory mem, immI0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7086 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7087 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7088 match(Set mem (StoreI mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7089 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7090 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7091 format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7092 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7093 __ movl($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7094 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7095 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7096 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7097 |
| 0 | 7098 instruct storeImmI(memory mem, immI src) |
| 7099 %{ | |
| 7100 match(Set mem (StoreI mem src)); | |
| 7101 | |
| 7102 ins_cost(150); | |
| 7103 format %{ "movl $mem, $src\t# int" %} | |
| 7104 opcode(0xC7); /* C7 /0 */ | |
| 7105 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src)); | |
| 7106 ins_pipe(ialu_mem_imm); | |
| 7107 %} | |
| 7108 | |
| 7109 // Store Long Immediate | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7110 instruct storeImmL0(memory mem, immL0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7111 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7112 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7113 match(Set mem (StoreL mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7114 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7115 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7116 format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7117 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7118 __ movq($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7119 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7120 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7121 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7122 |
| 0 | 7123 instruct storeImmL(memory mem, immL32 src) |
| 7124 %{ | |
| 7125 match(Set mem (StoreL mem src)); | |
| 7126 | |
| 7127 ins_cost(150); | |
| 7128 format %{ "movq $mem, $src\t# long" %} | |
| 7129 opcode(0xC7); /* C7 /0 */ | |
| 7130 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src)); | |
| 7131 ins_pipe(ialu_mem_imm); | |
| 7132 %} | |
| 7133 | |
| 7134 // Store Short/Char Immediate | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7135 instruct storeImmC0(memory mem, immI0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7136 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7137 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7138 match(Set mem (StoreC mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7139 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7140 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7141 format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7142 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7143 __ movw($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7144 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7145 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7146 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7147 |
| 0 | 7148 instruct storeImmI16(memory mem, immI16 src) |
| 7149 %{ | |
| 7150 predicate(UseStoreImmI16); | |
| 7151 match(Set mem (StoreC mem src)); | |
| 7152 | |
| 7153 ins_cost(150); | |
| 7154 format %{ "movw $mem, $src\t# short/char" %} | |
| 7155 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */ | |
| 7156 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src)); | |
| 7157 ins_pipe(ialu_mem_imm); | |
| 7158 %} | |
| 7159 | |
| 7160 // Store Byte Immediate | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7161 instruct storeImmB0(memory mem, immI0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7162 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7163 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7164 match(Set mem (StoreB mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7165 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7166 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7167 format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7168 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7169 __ movb($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7170 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7171 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7172 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7173 |
| 0 | 7174 instruct storeImmB(memory mem, immI8 src) |
| 7175 %{ | |
| 7176 match(Set mem (StoreB mem src)); | |
| 7177 | |
| 7178 ins_cost(150); // XXX | |
| 7179 format %{ "movb $mem, $src\t# byte" %} | |
| 7180 opcode(0xC6); /* C6 /0 */ | |
| 7181 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src)); | |
| 7182 ins_pipe(ialu_mem_imm); | |
| 7183 %} | |
| 7184 | |
| 7185 // Store Aligned Packed Byte XMM register to memory | |
| 7186 instruct storeA8B(memory mem, regD src) %{ | |
| 7187 match(Set mem (Store8B mem src)); | |
| 7188 ins_cost(145); | |
| 7189 format %{ "MOVQ $mem,$src\t! packed8B" %} | |
| 7190 ins_encode( movq_st(mem, src)); | |
| 7191 ins_pipe( pipe_slow ); | |
| 7192 %} | |
| 7193 | |
| 7194 // Store Aligned Packed Char/Short XMM register to memory | |
| 7195 instruct storeA4C(memory mem, regD src) %{ | |
| 7196 match(Set mem (Store4C mem src)); | |
| 7197 ins_cost(145); | |
| 7198 format %{ "MOVQ $mem,$src\t! packed4C" %} | |
| 7199 ins_encode( movq_st(mem, src)); | |
| 7200 ins_pipe( pipe_slow ); | |
| 7201 %} | |
| 7202 | |
| 7203 // Store Aligned Packed Integer XMM register to memory | |
| 7204 instruct storeA2I(memory mem, regD src) %{ | |
| 7205 match(Set mem (Store2I mem src)); | |
| 7206 ins_cost(145); | |
| 7207 format %{ "MOVQ $mem,$src\t! packed2I" %} | |
| 7208 ins_encode( movq_st(mem, src)); | |
| 7209 ins_pipe( pipe_slow ); | |
| 7210 %} | |
| 7211 | |
| 7212 // Store CMS card-mark Immediate | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7213 instruct storeImmCM0_reg(memory mem, immI0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7214 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7215 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7216 match(Set mem (StoreCM mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7217 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7218 ins_cost(125); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7219 format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7220 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7221 __ movb($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7222 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7223 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7224 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7225 |
| 0 | 7226 instruct storeImmCM0(memory mem, immI0 src) |
| 7227 %{ | |
| 7228 match(Set mem (StoreCM mem src)); | |
| 7229 | |
| 7230 ins_cost(150); // XXX | |
| 7231 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %} | |
| 7232 opcode(0xC6); /* C6 /0 */ | |
| 7233 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src)); | |
| 7234 ins_pipe(ialu_mem_imm); | |
| 7235 %} | |
| 7236 | |
| 7237 // Store Aligned Packed Single Float XMM register to memory | |
| 7238 instruct storeA2F(memory mem, regD src) %{ | |
| 7239 match(Set mem (Store2F mem src)); | |
| 7240 ins_cost(145); | |
| 7241 format %{ "MOVQ $mem,$src\t! packed2F" %} | |
| 7242 ins_encode( movq_st(mem, src)); | |
| 7243 ins_pipe( pipe_slow ); | |
| 7244 %} | |
| 7245 | |
| 7246 // Store Float | |
| 7247 instruct storeF(memory mem, regF src) | |
| 7248 %{ | |
| 7249 match(Set mem (StoreF mem src)); | |
| 7250 | |
| 7251 ins_cost(95); // XXX | |
| 7252 format %{ "movss $mem, $src\t# float" %} | |
| 7253 opcode(0xF3, 0x0F, 0x11); | |
| 7254 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem)); | |
| 7255 ins_pipe(pipe_slow); // XXX | |
| 7256 %} | |
| 7257 | |
| 7258 // Store immediate Float value (it is faster than store from XMM register) | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7259 instruct storeF0(memory mem, immF0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7260 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7261 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7262 match(Set mem (StoreF mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7263 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7264 ins_cost(25); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7265 format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7266 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7267 __ movl($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7268 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7269 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7270 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7271 |
| 0 | 7272 instruct storeF_imm(memory mem, immF src) |
| 7273 %{ | |
| 7274 match(Set mem (StoreF mem src)); | |
| 7275 | |
| 7276 ins_cost(50); | |
| 7277 format %{ "movl $mem, $src\t# float" %} | |
| 7278 opcode(0xC7); /* C7 /0 */ | |
| 7279 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src)); | |
| 7280 ins_pipe(ialu_mem_imm); | |
| 7281 %} | |
| 7282 | |
| 7283 // Store Double | |
| 7284 instruct storeD(memory mem, regD src) | |
| 7285 %{ | |
| 7286 match(Set mem (StoreD mem src)); | |
| 7287 | |
| 7288 ins_cost(95); // XXX | |
| 7289 format %{ "movsd $mem, $src\t# double" %} | |
| 7290 opcode(0xF2, 0x0F, 0x11); | |
| 7291 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem)); | |
| 7292 ins_pipe(pipe_slow); // XXX | |
| 7293 %} | |
| 7294 | |
| 7295 // Store immediate double 0.0 (it is faster than store from XMM register) | |
| 7296 instruct storeD0_imm(memory mem, immD0 src) | |
| 7297 %{ | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7298 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL)); |
| 0 | 7299 match(Set mem (StoreD mem src)); |
| 7300 | |
| 7301 ins_cost(50); | |
| 7302 format %{ "movq $mem, $src\t# double 0." %} | |
| 7303 opcode(0xC7); /* C7 /0 */ | |
| 7304 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src)); | |
| 7305 ins_pipe(ialu_mem_imm); | |
| 7306 %} | |
| 7307 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7308 instruct storeD0(memory mem, immD0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7309 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7310 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7311 match(Set mem (StoreD mem zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7312 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7313 ins_cost(25); // XXX |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7314 format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7315 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7316 __ movq($mem$$Address, r12); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7317 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7318 ins_pipe(ialu_mem_reg); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7319 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7320 |
| 0 | 7321 instruct storeSSI(stackSlotI dst, rRegI src) |
| 7322 %{ | |
| 7323 match(Set dst src); | |
| 7324 | |
| 7325 ins_cost(100); | |
| 7326 format %{ "movl $dst, $src\t# int stk" %} | |
| 7327 opcode(0x89); | |
| 7328 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 7329 ins_pipe( ialu_mem_reg ); | |
| 7330 %} | |
| 7331 | |
| 7332 instruct storeSSL(stackSlotL dst, rRegL src) | |
| 7333 %{ | |
| 7334 match(Set dst src); | |
| 7335 | |
| 7336 ins_cost(100); | |
| 7337 format %{ "movq $dst, $src\t# long stk" %} | |
| 7338 opcode(0x89); | |
| 7339 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 7340 ins_pipe(ialu_mem_reg); | |
| 7341 %} | |
| 7342 | |
| 7343 instruct storeSSP(stackSlotP dst, rRegP src) | |
| 7344 %{ | |
| 7345 match(Set dst src); | |
| 7346 | |
| 7347 ins_cost(100); | |
| 7348 format %{ "movq $dst, $src\t# ptr stk" %} | |
| 7349 opcode(0x89); | |
| 7350 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 7351 ins_pipe(ialu_mem_reg); | |
| 7352 %} | |
| 7353 | |
| 7354 instruct storeSSF(stackSlotF dst, regF src) | |
| 7355 %{ | |
| 7356 match(Set dst src); | |
| 7357 | |
| 7358 ins_cost(95); // XXX | |
| 7359 format %{ "movss $dst, $src\t# float stk" %} | |
| 7360 opcode(0xF3, 0x0F, 0x11); | |
| 7361 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst)); | |
| 7362 ins_pipe(pipe_slow); // XXX | |
| 7363 %} | |
| 7364 | |
| 7365 instruct storeSSD(stackSlotD dst, regD src) | |
| 7366 %{ | |
| 7367 match(Set dst src); | |
| 7368 | |
| 7369 ins_cost(95); // XXX | |
| 7370 format %{ "movsd $dst, $src\t# double stk" %} | |
| 7371 opcode(0xF2, 0x0F, 0x11); | |
| 7372 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst)); | |
| 7373 ins_pipe(pipe_slow); // XXX | |
| 7374 %} | |
| 7375 | |
| 7376 //----------BSWAP Instructions------------------------------------------------- | |
| 7377 instruct bytes_reverse_int(rRegI dst) %{ | |
| 7378 match(Set dst (ReverseBytesI dst)); | |
| 7379 | |
| 7380 format %{ "bswapl $dst" %} | |
| 7381 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */ | |
| 7382 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) ); | |
| 7383 ins_pipe( ialu_reg ); | |
| 7384 %} | |
| 7385 | |
| 7386 instruct bytes_reverse_long(rRegL dst) %{ | |
| 7387 match(Set dst (ReverseBytesL dst)); | |
| 7388 | |
| 7389 format %{ "bswapq $dst" %} | |
| 7390 | |
| 7391 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */ | |
| 7392 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) ); | |
| 7393 ins_pipe( ialu_reg); | |
| 7394 %} | |
| 7395 | |
| 7396 instruct loadI_reversed(rRegI dst, memory src) %{ | |
| 7397 match(Set dst (ReverseBytesI (LoadI src))); | |
| 7398 | |
| 7399 format %{ "bswap_movl $dst, $src" %} | |
| 7400 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */ | |
| 7401 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst)); | |
| 7402 ins_pipe( ialu_reg_mem ); | |
| 7403 %} | |
| 7404 | |
| 7405 instruct loadL_reversed(rRegL dst, memory src) %{ | |
| 7406 match(Set dst (ReverseBytesL (LoadL src))); | |
| 7407 | |
| 7408 format %{ "bswap_movq $dst, $src" %} | |
| 7409 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */ | |
| 7410 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst)); | |
| 7411 ins_pipe( ialu_reg_mem ); | |
| 7412 %} | |
| 7413 | |
| 7414 instruct storeI_reversed(memory dst, rRegI src) %{ | |
| 7415 match(Set dst (StoreI dst (ReverseBytesI src))); | |
| 7416 | |
| 7417 format %{ "movl_bswap $dst, $src" %} | |
| 7418 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */ | |
| 7419 ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) ); | |
| 7420 ins_pipe( ialu_mem_reg ); | |
| 7421 %} | |
| 7422 | |
| 7423 instruct storeL_reversed(memory dst, rRegL src) %{ | |
| 7424 match(Set dst (StoreL dst (ReverseBytesL src))); | |
| 7425 | |
| 7426 format %{ "movq_bswap $dst, $src" %} | |
| 7427 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */ | |
| 7428 ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) ); | |
| 7429 ins_pipe( ialu_mem_reg ); | |
| 7430 %} | |
| 7431 | |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7432 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7433 //---------- Population Count Instructions ------------------------------------- |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7434 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7435 instruct popCountI(rRegI dst, rRegI src) %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7436 predicate(UsePopCountInstruction); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7437 match(Set dst (PopCountI src)); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7438 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7439 format %{ "popcnt $dst, $src" %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7440 ins_encode %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7441 __ popcntl($dst$$Register, $src$$Register); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7442 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7443 ins_pipe(ialu_reg); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7444 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7445 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7446 instruct popCountI_mem(rRegI dst, memory mem) %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7447 predicate(UsePopCountInstruction); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7448 match(Set dst (PopCountI (LoadI mem))); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7449 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7450 format %{ "popcnt $dst, $mem" %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7451 ins_encode %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7452 __ popcntl($dst$$Register, $mem$$Address); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7453 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7454 ins_pipe(ialu_reg); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7455 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7456 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7457 // Note: Long.bitCount(long) returns an int. |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7458 instruct popCountL(rRegI dst, rRegL src) %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7459 predicate(UsePopCountInstruction); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7460 match(Set dst (PopCountL src)); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7461 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7462 format %{ "popcnt $dst, $src" %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7463 ins_encode %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7464 __ popcntq($dst$$Register, $src$$Register); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7465 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7466 ins_pipe(ialu_reg); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7467 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7468 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7469 // Note: Long.bitCount(long) returns an int. |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7470 instruct popCountL_mem(rRegI dst, memory mem) %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7471 predicate(UsePopCountInstruction); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7472 match(Set dst (PopCountL (LoadL mem))); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7473 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7474 format %{ "popcnt $dst, $mem" %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7475 ins_encode %{ |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7476 __ popcntq($dst$$Register, $mem$$Address); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7477 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7478 ins_pipe(ialu_reg); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7479 %} |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7480 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
7481 |
| 0 | 7482 //----------MemBar Instructions----------------------------------------------- |
| 7483 // Memory barrier flavors | |
| 7484 | |
| 7485 instruct membar_acquire() | |
| 7486 %{ | |
| 7487 match(MemBarAcquire); | |
| 7488 ins_cost(0); | |
| 7489 | |
| 7490 size(0); | |
| 7491 format %{ "MEMBAR-acquire" %} | |
| 7492 ins_encode(); | |
| 7493 ins_pipe(empty); | |
| 7494 %} | |
| 7495 | |
| 7496 instruct membar_acquire_lock() | |
| 7497 %{ | |
| 7498 match(MemBarAcquire); | |
| 7499 predicate(Matcher::prior_fast_lock(n)); | |
| 7500 ins_cost(0); | |
| 7501 | |
| 7502 size(0); | |
| 7503 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %} | |
| 7504 ins_encode(); | |
| 7505 ins_pipe(empty); | |
| 7506 %} | |
| 7507 | |
| 7508 instruct membar_release() | |
| 7509 %{ | |
| 7510 match(MemBarRelease); | |
| 7511 ins_cost(0); | |
| 7512 | |
| 7513 size(0); | |
| 7514 format %{ "MEMBAR-release" %} | |
| 7515 ins_encode(); | |
| 7516 ins_pipe(empty); | |
| 7517 %} | |
| 7518 | |
| 7519 instruct membar_release_lock() | |
| 7520 %{ | |
| 7521 match(MemBarRelease); | |
| 7522 predicate(Matcher::post_fast_unlock(n)); | |
| 7523 ins_cost(0); | |
| 7524 | |
| 7525 size(0); | |
| 7526 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %} | |
| 7527 ins_encode(); | |
| 7528 ins_pipe(empty); | |
| 7529 %} | |
| 7530 | |
| 7531 instruct membar_volatile() | |
| 7532 %{ | |
| 7533 match(MemBarVolatile); | |
| 7534 ins_cost(400); | |
| 7535 | |
| 7536 format %{ "MEMBAR-volatile" %} | |
| 7537 ins_encode(enc_membar_volatile); | |
| 7538 ins_pipe(pipe_slow); | |
| 7539 %} | |
| 7540 | |
| 7541 instruct unnecessary_membar_volatile() | |
| 7542 %{ | |
| 7543 match(MemBarVolatile); | |
| 7544 predicate(Matcher::post_store_load_barrier(n)); | |
| 7545 ins_cost(0); | |
| 7546 | |
| 7547 size(0); | |
| 7548 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %} | |
| 7549 ins_encode(); | |
| 7550 ins_pipe(empty); | |
| 7551 %} | |
| 7552 | |
| 7553 //----------Move Instructions-------------------------------------------------- | |
| 7554 | |
| 7555 instruct castX2P(rRegP dst, rRegL src) | |
| 7556 %{ | |
| 7557 match(Set dst (CastX2P src)); | |
| 7558 | |
| 7559 format %{ "movq $dst, $src\t# long->ptr" %} | |
| 7560 ins_encode(enc_copy_wide(dst, src)); | |
| 7561 ins_pipe(ialu_reg_reg); // XXX | |
| 7562 %} | |
| 7563 | |
| 7564 instruct castP2X(rRegL dst, rRegP src) | |
| 7565 %{ | |
| 7566 match(Set dst (CastP2X src)); | |
| 7567 | |
| 7568 format %{ "movq $dst, $src\t# ptr -> long" %} | |
| 7569 ins_encode(enc_copy_wide(dst, src)); | |
| 7570 ins_pipe(ialu_reg_reg); // XXX | |
| 7571 %} | |
| 7572 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7573 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7574 // Convert oop pointer into compressed form |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7575 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{ |
|
221
1e026f8da827
6710487: More than half of JDI Regression tests hang with COOPs in -Xcomp mode
kvn
parents:
182
diff
changeset
|
7576 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7577 match(Set dst (EncodeP src)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7578 effect(KILL cr); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7579 format %{ "encode_heap_oop $dst,$src" %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7580 ins_encode %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7581 Register s = $src$$Register; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7582 Register d = $dst$$Register; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7583 if (s != d) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7584 __ movq(d, s); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7585 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7586 __ encode_heap_oop(d); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7587 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7588 ins_pipe(ialu_reg_long); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7589 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7590 |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7591 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{ |
|
221
1e026f8da827
6710487: More than half of JDI Regression tests hang with COOPs in -Xcomp mode
kvn
parents:
182
diff
changeset
|
7592 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull); |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7593 match(Set dst (EncodeP src)); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7594 effect(KILL cr); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7595 format %{ "encode_heap_oop_not_null $dst,$src" %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7596 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7597 __ encode_heap_oop_not_null($dst$$Register, $src$$Register); |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7598 %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7599 ins_pipe(ialu_reg_long); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7600 %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7601 |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7602 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{ |
|
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
169
diff
changeset
|
7603 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull && |
|
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
169
diff
changeset
|
7604 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7605 match(Set dst (DecodeN src)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7606 effect(KILL cr); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7607 format %{ "decode_heap_oop $dst,$src" %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7608 ins_encode %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7609 Register s = $src$$Register; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7610 Register d = $dst$$Register; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7611 if (s != d) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7612 __ movq(d, s); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7613 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7614 __ decode_heap_oop(d); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7615 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7616 ins_pipe(ialu_reg_long); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7617 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7618 |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7619 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{ |
|
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
169
diff
changeset
|
7620 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull || |
|
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
169
diff
changeset
|
7621 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant); |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7622 match(Set dst (DecodeN src)); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7623 format %{ "decode_heap_oop_not_null $dst,$src" %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7624 ins_encode %{ |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7625 Register s = $src$$Register; |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7626 Register d = $dst$$Register; |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7627 if (s != d) { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7628 __ decode_heap_oop_not_null(d, s); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7629 } else { |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7630 __ decode_heap_oop_not_null(d); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
7631 } |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7632 %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7633 ins_pipe(ialu_reg_long); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7634 %} |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
7635 |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
7636 |
| 0 | 7637 //----------Conditional Move--------------------------------------------------- |
| 7638 // Jump | |
| 7639 // dummy instruction for generating temp registers | |
| 7640 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{ | |
| 7641 match(Jump (LShiftL switch_val shift)); | |
| 7642 ins_cost(350); | |
| 7643 predicate(false); | |
| 7644 effect(TEMP dest); | |
| 7645 | |
| 7646 format %{ "leaq $dest, table_base\n\t" | |
| 7647 "jmp [$dest + $switch_val << $shift]\n\t" %} | |
| 7648 ins_encode(jump_enc_offset(switch_val, shift, dest)); | |
| 7649 ins_pipe(pipe_jmp); | |
| 7650 ins_pc_relative(1); | |
| 7651 %} | |
| 7652 | |
| 7653 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{ | |
| 7654 match(Jump (AddL (LShiftL switch_val shift) offset)); | |
| 7655 ins_cost(350); | |
| 7656 effect(TEMP dest); | |
| 7657 | |
| 7658 format %{ "leaq $dest, table_base\n\t" | |
| 7659 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %} | |
| 7660 ins_encode(jump_enc_addr(switch_val, shift, offset, dest)); | |
| 7661 ins_pipe(pipe_jmp); | |
| 7662 ins_pc_relative(1); | |
| 7663 %} | |
| 7664 | |
| 7665 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{ | |
| 7666 match(Jump switch_val); | |
| 7667 ins_cost(350); | |
| 7668 effect(TEMP dest); | |
| 7669 | |
| 7670 format %{ "leaq $dest, table_base\n\t" | |
| 7671 "jmp [$dest + $switch_val]\n\t" %} | |
| 7672 ins_encode(jump_enc(switch_val, dest)); | |
| 7673 ins_pipe(pipe_jmp); | |
| 7674 ins_pc_relative(1); | |
| 7675 %} | |
| 7676 | |
| 7677 // Conditional move | |
| 7678 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop) | |
| 7679 %{ | |
| 7680 match(Set dst (CMoveI (Binary cop cr) (Binary dst src))); | |
| 7681 | |
| 7682 ins_cost(200); // XXX | |
| 7683 format %{ "cmovl$cop $dst, $src\t# signed, int" %} | |
| 7684 opcode(0x0F, 0x40); | |
| 7685 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7686 ins_pipe(pipe_cmov_reg); | |
| 7687 %} | |
| 7688 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7689 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{ |
| 0 | 7690 match(Set dst (CMoveI (Binary cop cr) (Binary dst src))); |
| 7691 | |
| 7692 ins_cost(200); // XXX | |
| 7693 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %} | |
| 7694 opcode(0x0F, 0x40); | |
| 7695 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7696 ins_pipe(pipe_cmov_reg); | |
| 7697 %} | |
| 7698 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7699 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7700 match(Set dst (CMoveI (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7701 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7702 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7703 cmovI_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7704 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7705 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7706 |
| 0 | 7707 // Conditional move |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7708 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{ |
| 0 | 7709 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src)))); |
| 7710 | |
| 7711 ins_cost(250); // XXX | |
| 7712 format %{ "cmovl$cop $dst, $src\t# signed, int" %} | |
| 7713 opcode(0x0F, 0x40); | |
| 7714 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src)); | |
| 7715 ins_pipe(pipe_cmov_mem); | |
| 7716 %} | |
| 7717 | |
| 7718 // Conditional move | |
| 7719 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src) | |
| 7720 %{ | |
| 7721 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src)))); | |
| 7722 | |
| 7723 ins_cost(250); // XXX | |
| 7724 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %} | |
| 7725 opcode(0x0F, 0x40); | |
| 7726 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src)); | |
| 7727 ins_pipe(pipe_cmov_mem); | |
| 7728 %} | |
| 7729 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7730 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7731 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src)))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7732 ins_cost(250); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7733 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7734 cmovI_memU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7735 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7736 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7737 |
| 0 | 7738 // Conditional move |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7739 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop) |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7740 %{ |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7741 match(Set dst (CMoveN (Binary cop cr) (Binary dst src))); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7742 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7743 ins_cost(200); // XXX |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7744 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7745 opcode(0x0F, 0x40); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7746 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src)); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7747 ins_pipe(pipe_cmov_reg); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7748 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7749 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7750 // Conditional move |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7751 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src) |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7752 %{ |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7753 match(Set dst (CMoveN (Binary cop cr) (Binary dst src))); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7754 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7755 ins_cost(200); // XXX |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7756 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7757 opcode(0x0F, 0x40); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7758 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src)); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7759 ins_pipe(pipe_cmov_reg); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7760 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7761 |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7762 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7763 match(Set dst (CMoveN (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7764 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7765 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7766 cmovN_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7767 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7768 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7769 |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
7770 // Conditional move |
| 0 | 7771 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop) |
| 7772 %{ | |
| 7773 match(Set dst (CMoveP (Binary cop cr) (Binary dst src))); | |
| 7774 | |
| 7775 ins_cost(200); // XXX | |
| 7776 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %} | |
| 7777 opcode(0x0F, 0x40); | |
| 7778 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7779 ins_pipe(pipe_cmov_reg); // XXX | |
| 7780 %} | |
| 7781 | |
| 7782 // Conditional move | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7783 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src) |
| 0 | 7784 %{ |
| 7785 match(Set dst (CMoveP (Binary cop cr) (Binary dst src))); | |
| 7786 | |
| 7787 ins_cost(200); // XXX | |
| 7788 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %} | |
| 7789 opcode(0x0F, 0x40); | |
| 7790 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7791 ins_pipe(pipe_cmov_reg); // XXX | |
| 7792 %} | |
| 7793 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7794 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7795 match(Set dst (CMoveP (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7796 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7797 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7798 cmovP_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7799 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7800 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7801 |
| 0 | 7802 // DISABLED: Requires the ADLC to emit a bottom_type call that |
| 7803 // correctly meets the two pointer arguments; one is an incoming | |
| 7804 // register but the other is a memory operand. ALSO appears to | |
| 7805 // be buggy with implicit null checks. | |
| 7806 // | |
| 7807 //// Conditional move | |
| 7808 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src) | |
| 7809 //%{ | |
| 7810 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src)))); | |
| 7811 // ins_cost(250); | |
| 7812 // format %{ "CMOV$cop $dst,$src\t# ptr" %} | |
| 7813 // opcode(0x0F,0x40); | |
| 7814 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) ); | |
| 7815 // ins_pipe( pipe_cmov_mem ); | |
| 7816 //%} | |
| 7817 // | |
| 7818 //// Conditional move | |
| 7819 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src) | |
| 7820 //%{ | |
| 7821 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src)))); | |
| 7822 // ins_cost(250); | |
| 7823 // format %{ "CMOV$cop $dst,$src\t# ptr" %} | |
| 7824 // opcode(0x0F,0x40); | |
| 7825 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) ); | |
| 7826 // ins_pipe( pipe_cmov_mem ); | |
| 7827 //%} | |
| 7828 | |
| 7829 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src) | |
| 7830 %{ | |
| 7831 match(Set dst (CMoveL (Binary cop cr) (Binary dst src))); | |
| 7832 | |
| 7833 ins_cost(200); // XXX | |
| 7834 format %{ "cmovq$cop $dst, $src\t# signed, long" %} | |
| 7835 opcode(0x0F, 0x40); | |
| 7836 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7837 ins_pipe(pipe_cmov_reg); // XXX | |
| 7838 %} | |
| 7839 | |
| 7840 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src) | |
| 7841 %{ | |
| 7842 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src)))); | |
| 7843 | |
| 7844 ins_cost(200); // XXX | |
| 7845 format %{ "cmovq$cop $dst, $src\t# signed, long" %} | |
| 7846 opcode(0x0F, 0x40); | |
| 7847 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src)); | |
| 7848 ins_pipe(pipe_cmov_mem); // XXX | |
| 7849 %} | |
| 7850 | |
| 7851 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src) | |
| 7852 %{ | |
| 7853 match(Set dst (CMoveL (Binary cop cr) (Binary dst src))); | |
| 7854 | |
| 7855 ins_cost(200); // XXX | |
| 7856 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %} | |
| 7857 opcode(0x0F, 0x40); | |
| 7858 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src)); | |
| 7859 ins_pipe(pipe_cmov_reg); // XXX | |
| 7860 %} | |
| 7861 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7862 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7863 match(Set dst (CMoveL (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7864 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7865 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7866 cmovL_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7867 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7868 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7869 |
| 0 | 7870 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src) |
| 7871 %{ | |
| 7872 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src)))); | |
| 7873 | |
| 7874 ins_cost(200); // XXX | |
| 7875 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %} | |
| 7876 opcode(0x0F, 0x40); | |
| 7877 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src)); | |
| 7878 ins_pipe(pipe_cmov_mem); // XXX | |
| 7879 %} | |
| 7880 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7881 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7882 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src)))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7883 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7884 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7885 cmovL_memU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7886 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7887 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7888 |
| 0 | 7889 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src) |
| 7890 %{ | |
| 7891 match(Set dst (CMoveF (Binary cop cr) (Binary dst src))); | |
| 7892 | |
| 7893 ins_cost(200); // XXX | |
| 7894 format %{ "jn$cop skip\t# signed cmove float\n\t" | |
| 7895 "movss $dst, $src\n" | |
| 7896 "skip:" %} | |
| 7897 ins_encode(enc_cmovf_branch(cop, dst, src)); | |
| 7898 ins_pipe(pipe_slow); | |
| 7899 %} | |
| 7900 | |
| 7901 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src) | |
| 7902 // %{ | |
| 7903 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src)))); | |
| 7904 | |
| 7905 // ins_cost(200); // XXX | |
| 7906 // format %{ "jn$cop skip\t# signed cmove float\n\t" | |
| 7907 // "movss $dst, $src\n" | |
| 7908 // "skip:" %} | |
| 7909 // ins_encode(enc_cmovf_mem_branch(cop, dst, src)); | |
| 7910 // ins_pipe(pipe_slow); | |
| 7911 // %} | |
| 7912 | |
| 7913 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src) | |
| 7914 %{ | |
| 7915 match(Set dst (CMoveF (Binary cop cr) (Binary dst src))); | |
| 7916 | |
| 7917 ins_cost(200); // XXX | |
| 7918 format %{ "jn$cop skip\t# unsigned cmove float\n\t" | |
| 7919 "movss $dst, $src\n" | |
| 7920 "skip:" %} | |
| 7921 ins_encode(enc_cmovf_branch(cop, dst, src)); | |
| 7922 ins_pipe(pipe_slow); | |
| 7923 %} | |
| 7924 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7925 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7926 match(Set dst (CMoveF (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7927 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7928 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7929 cmovF_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7930 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7931 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7932 |
| 0 | 7933 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src) |
| 7934 %{ | |
| 7935 match(Set dst (CMoveD (Binary cop cr) (Binary dst src))); | |
| 7936 | |
| 7937 ins_cost(200); // XXX | |
| 7938 format %{ "jn$cop skip\t# signed cmove double\n\t" | |
| 7939 "movsd $dst, $src\n" | |
| 7940 "skip:" %} | |
| 7941 ins_encode(enc_cmovd_branch(cop, dst, src)); | |
| 7942 ins_pipe(pipe_slow); | |
| 7943 %} | |
| 7944 | |
| 7945 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src) | |
| 7946 %{ | |
| 7947 match(Set dst (CMoveD (Binary cop cr) (Binary dst src))); | |
| 7948 | |
| 7949 ins_cost(200); // XXX | |
| 7950 format %{ "jn$cop skip\t# unsigned cmove double\n\t" | |
| 7951 "movsd $dst, $src\n" | |
| 7952 "skip:" %} | |
| 7953 ins_encode(enc_cmovd_branch(cop, dst, src)); | |
| 7954 ins_pipe(pipe_slow); | |
| 7955 %} | |
| 7956 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7957 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7958 match(Set dst (CMoveD (Binary cop cr) (Binary dst src))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7959 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7960 expand %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7961 cmovD_regU(cop, cr, dst, src); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7962 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7963 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
7964 |
| 0 | 7965 //----------Arithmetic Instructions-------------------------------------------- |
| 7966 //----------Addition Instructions---------------------------------------------- | |
| 7967 | |
| 7968 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 7969 %{ | |
| 7970 match(Set dst (AddI dst src)); | |
| 7971 effect(KILL cr); | |
| 7972 | |
| 7973 format %{ "addl $dst, $src\t# int" %} | |
| 7974 opcode(0x03); | |
| 7975 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src)); | |
| 7976 ins_pipe(ialu_reg_reg); | |
| 7977 %} | |
| 7978 | |
| 7979 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr) | |
| 7980 %{ | |
| 7981 match(Set dst (AddI dst src)); | |
| 7982 effect(KILL cr); | |
| 7983 | |
| 7984 format %{ "addl $dst, $src\t# int" %} | |
| 7985 opcode(0x81, 0x00); /* /0 id */ | |
| 7986 ins_encode(OpcSErm(dst, src), Con8or32(src)); | |
| 7987 ins_pipe( ialu_reg ); | |
| 7988 %} | |
| 7989 | |
| 7990 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 7991 %{ | |
| 7992 match(Set dst (AddI dst (LoadI src))); | |
| 7993 effect(KILL cr); | |
| 7994 | |
| 7995 ins_cost(125); // XXX | |
| 7996 format %{ "addl $dst, $src\t# int" %} | |
| 7997 opcode(0x03); | |
| 7998 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 7999 ins_pipe(ialu_reg_mem); | |
| 8000 %} | |
| 8001 | |
| 8002 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr) | |
| 8003 %{ | |
| 8004 match(Set dst (StoreI dst (AddI (LoadI dst) src))); | |
| 8005 effect(KILL cr); | |
| 8006 | |
| 8007 ins_cost(150); // XXX | |
| 8008 format %{ "addl $dst, $src\t# int" %} | |
| 8009 opcode(0x01); /* Opcode 01 /r */ | |
| 8010 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 8011 ins_pipe(ialu_mem_reg); | |
| 8012 %} | |
| 8013 | |
| 8014 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr) | |
| 8015 %{ | |
| 8016 match(Set dst (StoreI dst (AddI (LoadI dst) src))); | |
| 8017 effect(KILL cr); | |
| 8018 | |
| 8019 ins_cost(125); // XXX | |
| 8020 format %{ "addl $dst, $src\t# int" %} | |
| 8021 opcode(0x81); /* Opcode 81 /0 id */ | |
| 8022 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src)); | |
| 8023 ins_pipe(ialu_mem_imm); | |
| 8024 %} | |
| 8025 | |
| 8026 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr) | |
| 8027 %{ | |
| 8028 predicate(UseIncDec); | |
| 8029 match(Set dst (AddI dst src)); | |
| 8030 effect(KILL cr); | |
| 8031 | |
| 8032 format %{ "incl $dst\t# int" %} | |
| 8033 opcode(0xFF, 0x00); // FF /0 | |
| 8034 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 8035 ins_pipe(ialu_reg); | |
| 8036 %} | |
| 8037 | |
| 8038 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr) | |
| 8039 %{ | |
| 8040 predicate(UseIncDec); | |
| 8041 match(Set dst (StoreI dst (AddI (LoadI dst) src))); | |
| 8042 effect(KILL cr); | |
| 8043 | |
| 8044 ins_cost(125); // XXX | |
| 8045 format %{ "incl $dst\t# int" %} | |
| 8046 opcode(0xFF); /* Opcode FF /0 */ | |
| 8047 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst)); | |
| 8048 ins_pipe(ialu_mem_imm); | |
| 8049 %} | |
| 8050 | |
| 8051 // XXX why does that use AddI | |
| 8052 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr) | |
| 8053 %{ | |
| 8054 predicate(UseIncDec); | |
| 8055 match(Set dst (AddI dst src)); | |
| 8056 effect(KILL cr); | |
| 8057 | |
| 8058 format %{ "decl $dst\t# int" %} | |
| 8059 opcode(0xFF, 0x01); // FF /1 | |
| 8060 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 8061 ins_pipe(ialu_reg); | |
| 8062 %} | |
| 8063 | |
| 8064 // XXX why does that use AddI | |
| 8065 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr) | |
| 8066 %{ | |
| 8067 predicate(UseIncDec); | |
| 8068 match(Set dst (StoreI dst (AddI (LoadI dst) src))); | |
| 8069 effect(KILL cr); | |
| 8070 | |
| 8071 ins_cost(125); // XXX | |
| 8072 format %{ "decl $dst\t# int" %} | |
| 8073 opcode(0xFF); /* Opcode FF /1 */ | |
| 8074 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst)); | |
| 8075 ins_pipe(ialu_mem_imm); | |
| 8076 %} | |
| 8077 | |
| 8078 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1) | |
| 8079 %{ | |
| 8080 match(Set dst (AddI src0 src1)); | |
| 8081 | |
| 8082 ins_cost(110); | |
| 8083 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %} | |
| 8084 opcode(0x8D); /* 0x8D /r */ | |
| 8085 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX | |
| 8086 ins_pipe(ialu_reg_reg); | |
| 8087 %} | |
| 8088 | |
| 8089 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 8090 %{ | |
| 8091 match(Set dst (AddL dst src)); | |
| 8092 effect(KILL cr); | |
| 8093 | |
| 8094 format %{ "addq $dst, $src\t# long" %} | |
| 8095 opcode(0x03); | |
| 8096 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 8097 ins_pipe(ialu_reg_reg); | |
| 8098 %} | |
| 8099 | |
| 8100 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr) | |
| 8101 %{ | |
| 8102 match(Set dst (AddL dst src)); | |
| 8103 effect(KILL cr); | |
| 8104 | |
| 8105 format %{ "addq $dst, $src\t# long" %} | |
| 8106 opcode(0x81, 0x00); /* /0 id */ | |
| 8107 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 8108 ins_pipe( ialu_reg ); | |
| 8109 %} | |
| 8110 | |
| 8111 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 8112 %{ | |
| 8113 match(Set dst (AddL dst (LoadL src))); | |
| 8114 effect(KILL cr); | |
| 8115 | |
| 8116 ins_cost(125); // XXX | |
| 8117 format %{ "addq $dst, $src\t# long" %} | |
| 8118 opcode(0x03); | |
| 8119 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 8120 ins_pipe(ialu_reg_mem); | |
| 8121 %} | |
| 8122 | |
| 8123 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr) | |
| 8124 %{ | |
| 8125 match(Set dst (StoreL dst (AddL (LoadL dst) src))); | |
| 8126 effect(KILL cr); | |
| 8127 | |
| 8128 ins_cost(150); // XXX | |
| 8129 format %{ "addq $dst, $src\t# long" %} | |
| 8130 opcode(0x01); /* Opcode 01 /r */ | |
| 8131 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 8132 ins_pipe(ialu_mem_reg); | |
| 8133 %} | |
| 8134 | |
| 8135 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr) | |
| 8136 %{ | |
| 8137 match(Set dst (StoreL dst (AddL (LoadL dst) src))); | |
| 8138 effect(KILL cr); | |
| 8139 | |
| 8140 ins_cost(125); // XXX | |
| 8141 format %{ "addq $dst, $src\t# long" %} | |
| 8142 opcode(0x81); /* Opcode 81 /0 id */ | |
| 8143 ins_encode(REX_mem_wide(dst), | |
| 8144 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src)); | |
| 8145 ins_pipe(ialu_mem_imm); | |
| 8146 %} | |
| 8147 | |
| 8148 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr) | |
| 8149 %{ | |
| 8150 predicate(UseIncDec); | |
| 8151 match(Set dst (AddL dst src)); | |
| 8152 effect(KILL cr); | |
| 8153 | |
| 8154 format %{ "incq $dst\t# long" %} | |
| 8155 opcode(0xFF, 0x00); // FF /0 | |
| 8156 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 8157 ins_pipe(ialu_reg); | |
| 8158 %} | |
| 8159 | |
| 8160 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr) | |
| 8161 %{ | |
| 8162 predicate(UseIncDec); | |
| 8163 match(Set dst (StoreL dst (AddL (LoadL dst) src))); | |
| 8164 effect(KILL cr); | |
| 8165 | |
| 8166 ins_cost(125); // XXX | |
| 8167 format %{ "incq $dst\t# long" %} | |
| 8168 opcode(0xFF); /* Opcode FF /0 */ | |
| 8169 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst)); | |
| 8170 ins_pipe(ialu_mem_imm); | |
| 8171 %} | |
| 8172 | |
| 8173 // XXX why does that use AddL | |
| 8174 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr) | |
| 8175 %{ | |
| 8176 predicate(UseIncDec); | |
| 8177 match(Set dst (AddL dst src)); | |
| 8178 effect(KILL cr); | |
| 8179 | |
| 8180 format %{ "decq $dst\t# long" %} | |
| 8181 opcode(0xFF, 0x01); // FF /1 | |
| 8182 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 8183 ins_pipe(ialu_reg); | |
| 8184 %} | |
| 8185 | |
| 8186 // XXX why does that use AddL | |
| 8187 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr) | |
| 8188 %{ | |
| 8189 predicate(UseIncDec); | |
| 8190 match(Set dst (StoreL dst (AddL (LoadL dst) src))); | |
| 8191 effect(KILL cr); | |
| 8192 | |
| 8193 ins_cost(125); // XXX | |
| 8194 format %{ "decq $dst\t# long" %} | |
| 8195 opcode(0xFF); /* Opcode FF /1 */ | |
| 8196 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst)); | |
| 8197 ins_pipe(ialu_mem_imm); | |
| 8198 %} | |
| 8199 | |
| 8200 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1) | |
| 8201 %{ | |
| 8202 match(Set dst (AddL src0 src1)); | |
| 8203 | |
| 8204 ins_cost(110); | |
| 8205 format %{ "leaq $dst, [$src0 + $src1]\t# long" %} | |
| 8206 opcode(0x8D); /* 0x8D /r */ | |
| 8207 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX | |
| 8208 ins_pipe(ialu_reg_reg); | |
| 8209 %} | |
| 8210 | |
| 8211 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr) | |
| 8212 %{ | |
| 8213 match(Set dst (AddP dst src)); | |
| 8214 effect(KILL cr); | |
| 8215 | |
| 8216 format %{ "addq $dst, $src\t# ptr" %} | |
| 8217 opcode(0x03); | |
| 8218 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 8219 ins_pipe(ialu_reg_reg); | |
| 8220 %} | |
| 8221 | |
| 8222 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr) | |
| 8223 %{ | |
| 8224 match(Set dst (AddP dst src)); | |
| 8225 effect(KILL cr); | |
| 8226 | |
| 8227 format %{ "addq $dst, $src\t# ptr" %} | |
| 8228 opcode(0x81, 0x00); /* /0 id */ | |
| 8229 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 8230 ins_pipe( ialu_reg ); | |
| 8231 %} | |
| 8232 | |
| 8233 // XXX addP mem ops ???? | |
| 8234 | |
| 8235 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1) | |
| 8236 %{ | |
| 8237 match(Set dst (AddP src0 src1)); | |
| 8238 | |
| 8239 ins_cost(110); | |
| 8240 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %} | |
| 8241 opcode(0x8D); /* 0x8D /r */ | |
| 8242 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX | |
| 8243 ins_pipe(ialu_reg_reg); | |
| 8244 %} | |
| 8245 | |
| 8246 instruct checkCastPP(rRegP dst) | |
| 8247 %{ | |
| 8248 match(Set dst (CheckCastPP dst)); | |
| 8249 | |
| 8250 size(0); | |
| 8251 format %{ "# checkcastPP of $dst" %} | |
| 8252 ins_encode(/* empty encoding */); | |
| 8253 ins_pipe(empty); | |
| 8254 %} | |
| 8255 | |
| 8256 instruct castPP(rRegP dst) | |
| 8257 %{ | |
| 8258 match(Set dst (CastPP dst)); | |
| 8259 | |
| 8260 size(0); | |
| 8261 format %{ "# castPP of $dst" %} | |
| 8262 ins_encode(/* empty encoding */); | |
| 8263 ins_pipe(empty); | |
| 8264 %} | |
| 8265 | |
| 8266 instruct castII(rRegI dst) | |
| 8267 %{ | |
| 8268 match(Set dst (CastII dst)); | |
| 8269 | |
| 8270 size(0); | |
| 8271 format %{ "# castII of $dst" %} | |
| 8272 ins_encode(/* empty encoding */); | |
| 8273 ins_cost(0); | |
| 8274 ins_pipe(empty); | |
| 8275 %} | |
| 8276 | |
| 8277 // LoadP-locked same as a regular LoadP when used with compare-swap | |
| 8278 instruct loadPLocked(rRegP dst, memory mem) | |
| 8279 %{ | |
| 8280 match(Set dst (LoadPLocked mem)); | |
| 8281 | |
| 8282 ins_cost(125); // XXX | |
| 8283 format %{ "movq $dst, $mem\t# ptr locked" %} | |
| 8284 opcode(0x8B); | |
| 8285 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 8286 ins_pipe(ialu_reg_mem); // XXX | |
| 8287 %} | |
| 8288 | |
| 8289 // LoadL-locked - same as a regular LoadL when used with compare-swap | |
| 8290 instruct loadLLocked(rRegL dst, memory mem) | |
| 8291 %{ | |
| 8292 match(Set dst (LoadLLocked mem)); | |
| 8293 | |
| 8294 ins_cost(125); // XXX | |
| 8295 format %{ "movq $dst, $mem\t# long locked" %} | |
| 8296 opcode(0x8B); | |
| 8297 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); | |
| 8298 ins_pipe(ialu_reg_mem); // XXX | |
| 8299 %} | |
| 8300 | |
| 8301 // Conditional-store of the updated heap-top. | |
| 8302 // Used during allocation of the shared heap. | |
| 8303 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel. | |
| 8304 | |
| 8305 instruct storePConditional(memory heap_top_ptr, | |
| 8306 rax_RegP oldval, rRegP newval, | |
| 8307 rFlagsReg cr) | |
| 8308 %{ | |
| 8309 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval))); | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8310 |
| 0 | 8311 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) " |
| 8312 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %} | |
| 8313 opcode(0x0F, 0xB1); | |
| 8314 ins_encode(lock_prefix, | |
| 8315 REX_reg_mem_wide(newval, heap_top_ptr), | |
| 8316 OpcP, OpcS, | |
| 8317 reg_mem(newval, heap_top_ptr)); | |
| 8318 ins_pipe(pipe_cmpxchg); | |
| 8319 %} | |
| 8320 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8321 // Conditional-store of an int value. |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8322 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG. |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8323 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr) |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8324 %{ |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8325 match(Set cr (StoreIConditional mem (Binary oldval newval))); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8326 effect(KILL oldval); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8327 |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8328 format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %} |
| 0 | 8329 opcode(0x0F, 0xB1); |
| 8330 ins_encode(lock_prefix, | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8331 REX_reg_mem(newval, mem), |
| 0 | 8332 OpcP, OpcS, |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8333 reg_mem(newval, mem)); |
| 0 | 8334 ins_pipe(pipe_cmpxchg); |
| 8335 %} | |
| 8336 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8337 // Conditional-store of a long value. |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8338 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG. |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8339 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr) |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8340 %{ |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8341 match(Set cr (StoreLConditional mem (Binary oldval newval))); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8342 effect(KILL oldval); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8343 |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8344 format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %} |
| 0 | 8345 opcode(0x0F, 0xB1); |
| 8346 ins_encode(lock_prefix, | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8347 REX_reg_mem_wide(newval, mem), |
| 0 | 8348 OpcP, OpcS, |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8349 reg_mem(newval, mem)); |
| 0 | 8350 ins_pipe(pipe_cmpxchg); |
| 8351 %} | |
| 8352 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8353 |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
8354 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them |
| 0 | 8355 instruct compareAndSwapP(rRegI res, |
| 8356 memory mem_ptr, | |
| 8357 rax_RegP oldval, rRegP newval, | |
| 8358 rFlagsReg cr) | |
| 8359 %{ | |
| 8360 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval))); | |
| 8361 effect(KILL cr, KILL oldval); | |
| 8362 | |
| 8363 format %{ "cmpxchgq $mem_ptr,$newval\t# " | |
| 8364 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t" | |
| 8365 "sete $res\n\t" | |
| 8366 "movzbl $res, $res" %} | |
| 8367 opcode(0x0F, 0xB1); | |
| 8368 ins_encode(lock_prefix, | |
| 8369 REX_reg_mem_wide(newval, mem_ptr), | |
| 8370 OpcP, OpcS, | |
| 8371 reg_mem(newval, mem_ptr), | |
| 8372 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete | |
| 8373 REX_reg_breg(res, res), // movzbl | |
| 8374 Opcode(0xF), Opcode(0xB6), reg_reg(res, res)); | |
| 8375 ins_pipe( pipe_cmpxchg ); | |
| 8376 %} | |
| 8377 | |
| 8378 instruct compareAndSwapL(rRegI res, | |
| 8379 memory mem_ptr, | |
| 8380 rax_RegL oldval, rRegL newval, | |
| 8381 rFlagsReg cr) | |
| 8382 %{ | |
| 8383 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval))); | |
| 8384 effect(KILL cr, KILL oldval); | |
| 8385 | |
| 8386 format %{ "cmpxchgq $mem_ptr,$newval\t# " | |
| 8387 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t" | |
| 8388 "sete $res\n\t" | |
| 8389 "movzbl $res, $res" %} | |
| 8390 opcode(0x0F, 0xB1); | |
| 8391 ins_encode(lock_prefix, | |
| 8392 REX_reg_mem_wide(newval, mem_ptr), | |
| 8393 OpcP, OpcS, | |
| 8394 reg_mem(newval, mem_ptr), | |
| 8395 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete | |
| 8396 REX_reg_breg(res, res), // movzbl | |
| 8397 Opcode(0xF), Opcode(0xB6), reg_reg(res, res)); | |
| 8398 ins_pipe( pipe_cmpxchg ); | |
| 8399 %} | |
| 8400 | |
| 8401 instruct compareAndSwapI(rRegI res, | |
| 8402 memory mem_ptr, | |
| 8403 rax_RegI oldval, rRegI newval, | |
| 8404 rFlagsReg cr) | |
| 8405 %{ | |
| 8406 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval))); | |
| 8407 effect(KILL cr, KILL oldval); | |
| 8408 | |
| 8409 format %{ "cmpxchgl $mem_ptr,$newval\t# " | |
| 8410 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t" | |
| 8411 "sete $res\n\t" | |
| 8412 "movzbl $res, $res" %} | |
| 8413 opcode(0x0F, 0xB1); | |
| 8414 ins_encode(lock_prefix, | |
| 8415 REX_reg_mem(newval, mem_ptr), | |
| 8416 OpcP, OpcS, | |
| 8417 reg_mem(newval, mem_ptr), | |
| 8418 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete | |
| 8419 REX_reg_breg(res, res), // movzbl | |
| 8420 Opcode(0xF), Opcode(0xB6), reg_reg(res, res)); | |
| 8421 ins_pipe( pipe_cmpxchg ); | |
| 8422 %} | |
| 8423 | |
| 8424 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8425 instruct compareAndSwapN(rRegI res, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8426 memory mem_ptr, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8427 rax_RegN oldval, rRegN newval, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8428 rFlagsReg cr) %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8429 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval))); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8430 effect(KILL cr, KILL oldval); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8431 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8432 format %{ "cmpxchgl $mem_ptr,$newval\t# " |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8433 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t" |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8434 "sete $res\n\t" |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8435 "movzbl $res, $res" %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8436 opcode(0x0F, 0xB1); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8437 ins_encode(lock_prefix, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8438 REX_reg_mem(newval, mem_ptr), |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8439 OpcP, OpcS, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8440 reg_mem(newval, mem_ptr), |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8441 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8442 REX_reg_breg(res, res), // movzbl |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8443 Opcode(0xF), Opcode(0xB6), reg_reg(res, res)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8444 ins_pipe( pipe_cmpxchg ); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8445 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
8446 |
| 0 | 8447 //----------Subtraction Instructions------------------------------------------- |
| 8448 | |
| 8449 // Integer Subtraction Instructions | |
| 8450 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 8451 %{ | |
| 8452 match(Set dst (SubI dst src)); | |
| 8453 effect(KILL cr); | |
| 8454 | |
| 8455 format %{ "subl $dst, $src\t# int" %} | |
| 8456 opcode(0x2B); | |
| 8457 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src)); | |
| 8458 ins_pipe(ialu_reg_reg); | |
| 8459 %} | |
| 8460 | |
| 8461 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr) | |
| 8462 %{ | |
| 8463 match(Set dst (SubI dst src)); | |
| 8464 effect(KILL cr); | |
| 8465 | |
| 8466 format %{ "subl $dst, $src\t# int" %} | |
| 8467 opcode(0x81, 0x05); /* Opcode 81 /5 */ | |
| 8468 ins_encode(OpcSErm(dst, src), Con8or32(src)); | |
| 8469 ins_pipe(ialu_reg); | |
| 8470 %} | |
| 8471 | |
| 8472 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 8473 %{ | |
| 8474 match(Set dst (SubI dst (LoadI src))); | |
| 8475 effect(KILL cr); | |
| 8476 | |
| 8477 ins_cost(125); | |
| 8478 format %{ "subl $dst, $src\t# int" %} | |
| 8479 opcode(0x2B); | |
| 8480 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 8481 ins_pipe(ialu_reg_mem); | |
| 8482 %} | |
| 8483 | |
| 8484 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr) | |
| 8485 %{ | |
| 8486 match(Set dst (StoreI dst (SubI (LoadI dst) src))); | |
| 8487 effect(KILL cr); | |
| 8488 | |
| 8489 ins_cost(150); | |
| 8490 format %{ "subl $dst, $src\t# int" %} | |
| 8491 opcode(0x29); /* Opcode 29 /r */ | |
| 8492 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 8493 ins_pipe(ialu_mem_reg); | |
| 8494 %} | |
| 8495 | |
| 8496 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr) | |
| 8497 %{ | |
| 8498 match(Set dst (StoreI dst (SubI (LoadI dst) src))); | |
| 8499 effect(KILL cr); | |
| 8500 | |
| 8501 ins_cost(125); // XXX | |
| 8502 format %{ "subl $dst, $src\t# int" %} | |
| 8503 opcode(0x81); /* Opcode 81 /5 id */ | |
| 8504 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src)); | |
| 8505 ins_pipe(ialu_mem_imm); | |
| 8506 %} | |
| 8507 | |
| 8508 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 8509 %{ | |
| 8510 match(Set dst (SubL dst src)); | |
| 8511 effect(KILL cr); | |
| 8512 | |
| 8513 format %{ "subq $dst, $src\t# long" %} | |
| 8514 opcode(0x2B); | |
| 8515 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 8516 ins_pipe(ialu_reg_reg); | |
| 8517 %} | |
| 8518 | |
| 8519 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr) | |
| 8520 %{ | |
| 8521 match(Set dst (SubL dst src)); | |
| 8522 effect(KILL cr); | |
| 8523 | |
| 8524 format %{ "subq $dst, $src\t# long" %} | |
| 8525 opcode(0x81, 0x05); /* Opcode 81 /5 */ | |
| 8526 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 8527 ins_pipe(ialu_reg); | |
| 8528 %} | |
| 8529 | |
| 8530 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 8531 %{ | |
| 8532 match(Set dst (SubL dst (LoadL src))); | |
| 8533 effect(KILL cr); | |
| 8534 | |
| 8535 ins_cost(125); | |
| 8536 format %{ "subq $dst, $src\t# long" %} | |
| 8537 opcode(0x2B); | |
| 8538 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 8539 ins_pipe(ialu_reg_mem); | |
| 8540 %} | |
| 8541 | |
| 8542 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr) | |
| 8543 %{ | |
| 8544 match(Set dst (StoreL dst (SubL (LoadL dst) src))); | |
| 8545 effect(KILL cr); | |
| 8546 | |
| 8547 ins_cost(150); | |
| 8548 format %{ "subq $dst, $src\t# long" %} | |
| 8549 opcode(0x29); /* Opcode 29 /r */ | |
| 8550 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 8551 ins_pipe(ialu_mem_reg); | |
| 8552 %} | |
| 8553 | |
| 8554 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr) | |
| 8555 %{ | |
| 8556 match(Set dst (StoreL dst (SubL (LoadL dst) src))); | |
| 8557 effect(KILL cr); | |
| 8558 | |
| 8559 ins_cost(125); // XXX | |
| 8560 format %{ "subq $dst, $src\t# long" %} | |
| 8561 opcode(0x81); /* Opcode 81 /5 id */ | |
| 8562 ins_encode(REX_mem_wide(dst), | |
| 8563 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src)); | |
| 8564 ins_pipe(ialu_mem_imm); | |
| 8565 %} | |
| 8566 | |
| 8567 // Subtract from a pointer | |
| 8568 // XXX hmpf??? | |
| 8569 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr) | |
| 8570 %{ | |
| 8571 match(Set dst (AddP dst (SubI zero src))); | |
| 8572 effect(KILL cr); | |
| 8573 | |
| 8574 format %{ "subq $dst, $src\t# ptr - int" %} | |
| 8575 opcode(0x2B); | |
| 8576 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 8577 ins_pipe(ialu_reg_reg); | |
| 8578 %} | |
| 8579 | |
| 8580 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr) | |
| 8581 %{ | |
| 8582 match(Set dst (SubI zero dst)); | |
| 8583 effect(KILL cr); | |
| 8584 | |
| 8585 format %{ "negl $dst\t# int" %} | |
| 8586 opcode(0xF7, 0x03); // Opcode F7 /3 | |
| 8587 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 8588 ins_pipe(ialu_reg); | |
| 8589 %} | |
| 8590 | |
| 8591 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr) | |
| 8592 %{ | |
| 8593 match(Set dst (StoreI dst (SubI zero (LoadI dst)))); | |
| 8594 effect(KILL cr); | |
| 8595 | |
| 8596 format %{ "negl $dst\t# int" %} | |
| 8597 opcode(0xF7, 0x03); // Opcode F7 /3 | |
| 8598 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 8599 ins_pipe(ialu_reg); | |
| 8600 %} | |
| 8601 | |
| 8602 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr) | |
| 8603 %{ | |
| 8604 match(Set dst (SubL zero dst)); | |
| 8605 effect(KILL cr); | |
| 8606 | |
| 8607 format %{ "negq $dst\t# long" %} | |
| 8608 opcode(0xF7, 0x03); // Opcode F7 /3 | |
| 8609 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 8610 ins_pipe(ialu_reg); | |
| 8611 %} | |
| 8612 | |
| 8613 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr) | |
| 8614 %{ | |
| 8615 match(Set dst (StoreL dst (SubL zero (LoadL dst)))); | |
| 8616 effect(KILL cr); | |
| 8617 | |
| 8618 format %{ "negq $dst\t# long" %} | |
| 8619 opcode(0xF7, 0x03); // Opcode F7 /3 | |
| 8620 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 8621 ins_pipe(ialu_reg); | |
| 8622 %} | |
| 8623 | |
| 8624 | |
| 8625 //----------Multiplication/Division Instructions------------------------------- | |
| 8626 // Integer Multiplication Instructions | |
| 8627 // Multiply Register | |
| 8628 | |
| 8629 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 8630 %{ | |
| 8631 match(Set dst (MulI dst src)); | |
| 8632 effect(KILL cr); | |
| 8633 | |
| 8634 ins_cost(300); | |
| 8635 format %{ "imull $dst, $src\t# int" %} | |
| 8636 opcode(0x0F, 0xAF); | |
| 8637 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 8638 ins_pipe(ialu_reg_reg_alu0); | |
| 8639 %} | |
| 8640 | |
| 8641 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr) | |
| 8642 %{ | |
| 8643 match(Set dst (MulI src imm)); | |
| 8644 effect(KILL cr); | |
| 8645 | |
| 8646 ins_cost(300); | |
| 8647 format %{ "imull $dst, $src, $imm\t# int" %} | |
| 8648 opcode(0x69); /* 69 /r id */ | |
| 8649 ins_encode(REX_reg_reg(dst, src), | |
| 8650 OpcSE(imm), reg_reg(dst, src), Con8or32(imm)); | |
| 8651 ins_pipe(ialu_reg_reg_alu0); | |
| 8652 %} | |
| 8653 | |
| 8654 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 8655 %{ | |
| 8656 match(Set dst (MulI dst (LoadI src))); | |
| 8657 effect(KILL cr); | |
| 8658 | |
| 8659 ins_cost(350); | |
| 8660 format %{ "imull $dst, $src\t# int" %} | |
| 8661 opcode(0x0F, 0xAF); | |
| 8662 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src)); | |
| 8663 ins_pipe(ialu_reg_mem_alu0); | |
| 8664 %} | |
| 8665 | |
| 8666 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr) | |
| 8667 %{ | |
| 8668 match(Set dst (MulI (LoadI src) imm)); | |
| 8669 effect(KILL cr); | |
| 8670 | |
| 8671 ins_cost(300); | |
| 8672 format %{ "imull $dst, $src, $imm\t# int" %} | |
| 8673 opcode(0x69); /* 69 /r id */ | |
| 8674 ins_encode(REX_reg_mem(dst, src), | |
| 8675 OpcSE(imm), reg_mem(dst, src), Con8or32(imm)); | |
| 8676 ins_pipe(ialu_reg_mem_alu0); | |
| 8677 %} | |
| 8678 | |
| 8679 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 8680 %{ | |
| 8681 match(Set dst (MulL dst src)); | |
| 8682 effect(KILL cr); | |
| 8683 | |
| 8684 ins_cost(300); | |
| 8685 format %{ "imulq $dst, $src\t# long" %} | |
| 8686 opcode(0x0F, 0xAF); | |
| 8687 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 8688 ins_pipe(ialu_reg_reg_alu0); | |
| 8689 %} | |
| 8690 | |
| 8691 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr) | |
| 8692 %{ | |
| 8693 match(Set dst (MulL src imm)); | |
| 8694 effect(KILL cr); | |
| 8695 | |
| 8696 ins_cost(300); | |
| 8697 format %{ "imulq $dst, $src, $imm\t# long" %} | |
| 8698 opcode(0x69); /* 69 /r id */ | |
| 8699 ins_encode(REX_reg_reg_wide(dst, src), | |
| 8700 OpcSE(imm), reg_reg(dst, src), Con8or32(imm)); | |
| 8701 ins_pipe(ialu_reg_reg_alu0); | |
| 8702 %} | |
| 8703 | |
| 8704 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 8705 %{ | |
| 8706 match(Set dst (MulL dst (LoadL src))); | |
| 8707 effect(KILL cr); | |
| 8708 | |
| 8709 ins_cost(350); | |
| 8710 format %{ "imulq $dst, $src\t# long" %} | |
| 8711 opcode(0x0F, 0xAF); | |
| 8712 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src)); | |
| 8713 ins_pipe(ialu_reg_mem_alu0); | |
| 8714 %} | |
| 8715 | |
| 8716 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr) | |
| 8717 %{ | |
| 8718 match(Set dst (MulL (LoadL src) imm)); | |
| 8719 effect(KILL cr); | |
| 8720 | |
| 8721 ins_cost(300); | |
| 8722 format %{ "imulq $dst, $src, $imm\t# long" %} | |
| 8723 opcode(0x69); /* 69 /r id */ | |
| 8724 ins_encode(REX_reg_mem_wide(dst, src), | |
| 8725 OpcSE(imm), reg_mem(dst, src), Con8or32(imm)); | |
| 8726 ins_pipe(ialu_reg_mem_alu0); | |
| 8727 %} | |
| 8728 | |
| 145 | 8729 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr) |
| 8730 %{ | |
| 8731 match(Set dst (MulHiL src rax)); | |
| 8732 effect(USE_KILL rax, KILL cr); | |
| 8733 | |
| 8734 ins_cost(300); | |
| 8735 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %} | |
| 8736 opcode(0xF7, 0x5); /* Opcode F7 /5 */ | |
| 8737 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src)); | |
| 8738 ins_pipe(ialu_reg_reg_alu0); | |
| 8739 %} | |
| 8740 | |
| 0 | 8741 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div, |
| 8742 rFlagsReg cr) | |
| 8743 %{ | |
| 8744 match(Set rax (DivI rax div)); | |
| 8745 effect(KILL rdx, KILL cr); | |
| 8746 | |
| 8747 ins_cost(30*100+10*100); // XXX | |
| 8748 format %{ "cmpl rax, 0x80000000\t# idiv\n\t" | |
| 8749 "jne,s normal\n\t" | |
| 8750 "xorl rdx, rdx\n\t" | |
| 8751 "cmpl $div, -1\n\t" | |
| 8752 "je,s done\n" | |
| 8753 "normal: cdql\n\t" | |
| 8754 "idivl $div\n" | |
| 8755 "done:" %} | |
| 8756 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8757 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div)); | |
| 8758 ins_pipe(ialu_reg_reg_alu0); | |
| 8759 %} | |
| 8760 | |
| 8761 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div, | |
| 8762 rFlagsReg cr) | |
| 8763 %{ | |
| 8764 match(Set rax (DivL rax div)); | |
| 8765 effect(KILL rdx, KILL cr); | |
| 8766 | |
| 8767 ins_cost(30*100+10*100); // XXX | |
| 8768 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t" | |
| 8769 "cmpq rax, rdx\n\t" | |
| 8770 "jne,s normal\n\t" | |
| 8771 "xorl rdx, rdx\n\t" | |
| 8772 "cmpq $div, -1\n\t" | |
| 8773 "je,s done\n" | |
| 8774 "normal: cdqq\n\t" | |
| 8775 "idivq $div\n" | |
| 8776 "done:" %} | |
| 8777 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8778 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div)); | |
| 8779 ins_pipe(ialu_reg_reg_alu0); | |
| 8780 %} | |
| 8781 | |
| 8782 // Integer DIVMOD with Register, both quotient and mod results | |
| 8783 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div, | |
| 8784 rFlagsReg cr) | |
| 8785 %{ | |
| 8786 match(DivModI rax div); | |
| 8787 effect(KILL cr); | |
| 8788 | |
| 8789 ins_cost(30*100+10*100); // XXX | |
| 8790 format %{ "cmpl rax, 0x80000000\t# idiv\n\t" | |
| 8791 "jne,s normal\n\t" | |
| 8792 "xorl rdx, rdx\n\t" | |
| 8793 "cmpl $div, -1\n\t" | |
| 8794 "je,s done\n" | |
| 8795 "normal: cdql\n\t" | |
| 8796 "idivl $div\n" | |
| 8797 "done:" %} | |
| 8798 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8799 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div)); | |
| 8800 ins_pipe(pipe_slow); | |
| 8801 %} | |
| 8802 | |
| 8803 // Long DIVMOD with Register, both quotient and mod results | |
| 8804 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div, | |
| 8805 rFlagsReg cr) | |
| 8806 %{ | |
| 8807 match(DivModL rax div); | |
| 8808 effect(KILL cr); | |
| 8809 | |
| 8810 ins_cost(30*100+10*100); // XXX | |
| 8811 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t" | |
| 8812 "cmpq rax, rdx\n\t" | |
| 8813 "jne,s normal\n\t" | |
| 8814 "xorl rdx, rdx\n\t" | |
| 8815 "cmpq $div, -1\n\t" | |
| 8816 "je,s done\n" | |
| 8817 "normal: cdqq\n\t" | |
| 8818 "idivq $div\n" | |
| 8819 "done:" %} | |
| 8820 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8821 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div)); | |
| 8822 ins_pipe(pipe_slow); | |
| 8823 %} | |
| 8824 | |
| 8825 //----------- DivL-By-Constant-Expansions-------------------------------------- | |
| 8826 // DivI cases are handled by the compiler | |
| 8827 | |
| 605 | 8828 // Magic constant, reciprocal of 10 |
| 0 | 8829 instruct loadConL_0x6666666666666667(rRegL dst) |
| 8830 %{ | |
| 8831 effect(DEF dst); | |
| 8832 | |
| 8833 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %} | |
| 8834 ins_encode(load_immL(dst, 0x6666666666666667)); | |
| 8835 ins_pipe(ialu_reg); | |
| 8836 %} | |
| 8837 | |
| 8838 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr) | |
| 8839 %{ | |
| 8840 effect(DEF dst, USE src, USE_KILL rax, KILL cr); | |
| 8841 | |
| 8842 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %} | |
| 8843 opcode(0xF7, 0x5); /* Opcode F7 /5 */ | |
| 8844 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src)); | |
| 8845 ins_pipe(ialu_reg_reg_alu0); | |
| 8846 %} | |
| 8847 | |
| 8848 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr) | |
| 8849 %{ | |
| 8850 effect(USE_DEF dst, KILL cr); | |
| 8851 | |
| 8852 format %{ "sarq $dst, #63\t# Used in div-by-10" %} | |
| 8853 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 8854 ins_encode(reg_opc_imm_wide(dst, 0x3F)); | |
| 8855 ins_pipe(ialu_reg); | |
| 8856 %} | |
| 8857 | |
| 8858 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr) | |
| 8859 %{ | |
| 8860 effect(USE_DEF dst, KILL cr); | |
| 8861 | |
| 8862 format %{ "sarq $dst, #2\t# Used in div-by-10" %} | |
| 8863 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 8864 ins_encode(reg_opc_imm_wide(dst, 0x2)); | |
| 8865 ins_pipe(ialu_reg); | |
| 8866 %} | |
| 8867 | |
| 8868 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div) | |
| 8869 %{ | |
| 8870 match(Set dst (DivL src div)); | |
| 8871 | |
| 8872 ins_cost((5+8)*100); | |
| 8873 expand %{ | |
| 8874 rax_RegL rax; // Killed temp | |
| 8875 rFlagsReg cr; // Killed | |
| 8876 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667 | |
| 8877 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src | |
| 8878 sarL_rReg_63(src, cr); // sarq src, 63 | |
| 8879 sarL_rReg_2(dst, cr); // sarq rdx, 2 | |
| 8880 subL_rReg(dst, src, cr); // subl rdx, src | |
| 8881 %} | |
| 8882 %} | |
| 8883 | |
| 8884 //----------------------------------------------------------------------------- | |
| 8885 | |
| 8886 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div, | |
| 8887 rFlagsReg cr) | |
| 8888 %{ | |
| 8889 match(Set rdx (ModI rax div)); | |
| 8890 effect(KILL rax, KILL cr); | |
| 8891 | |
| 8892 ins_cost(300); // XXX | |
| 8893 format %{ "cmpl rax, 0x80000000\t# irem\n\t" | |
| 8894 "jne,s normal\n\t" | |
| 8895 "xorl rdx, rdx\n\t" | |
| 8896 "cmpl $div, -1\n\t" | |
| 8897 "je,s done\n" | |
| 8898 "normal: cdql\n\t" | |
| 8899 "idivl $div\n" | |
| 8900 "done:" %} | |
| 8901 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8902 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div)); | |
| 8903 ins_pipe(ialu_reg_reg_alu0); | |
| 8904 %} | |
| 8905 | |
| 8906 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div, | |
| 8907 rFlagsReg cr) | |
| 8908 %{ | |
| 8909 match(Set rdx (ModL rax div)); | |
| 8910 effect(KILL rax, KILL cr); | |
| 8911 | |
| 8912 ins_cost(300); // XXX | |
| 8913 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t" | |
| 8914 "cmpq rax, rdx\n\t" | |
| 8915 "jne,s normal\n\t" | |
| 8916 "xorl rdx, rdx\n\t" | |
| 8917 "cmpq $div, -1\n\t" | |
| 8918 "je,s done\n" | |
| 8919 "normal: cdqq\n\t" | |
| 8920 "idivq $div\n" | |
| 8921 "done:" %} | |
| 8922 opcode(0xF7, 0x7); /* Opcode F7 /7 */ | |
| 8923 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div)); | |
| 8924 ins_pipe(ialu_reg_reg_alu0); | |
| 8925 %} | |
| 8926 | |
| 8927 // Integer Shift Instructions | |
| 8928 // Shift Left by one | |
| 8929 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr) | |
| 8930 %{ | |
| 8931 match(Set dst (LShiftI dst shift)); | |
| 8932 effect(KILL cr); | |
| 8933 | |
| 8934 format %{ "sall $dst, $shift" %} | |
| 8935 opcode(0xD1, 0x4); /* D1 /4 */ | |
| 8936 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 8937 ins_pipe(ialu_reg); | |
| 8938 %} | |
| 8939 | |
| 8940 // Shift Left by one | |
| 8941 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 8942 %{ | |
| 8943 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift))); | |
| 8944 effect(KILL cr); | |
| 8945 | |
| 8946 format %{ "sall $dst, $shift\t" %} | |
| 8947 opcode(0xD1, 0x4); /* D1 /4 */ | |
| 8948 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 8949 ins_pipe(ialu_mem_imm); | |
| 8950 %} | |
| 8951 | |
| 8952 // Shift Left by 8-bit immediate | |
| 8953 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr) | |
| 8954 %{ | |
| 8955 match(Set dst (LShiftI dst shift)); | |
| 8956 effect(KILL cr); | |
| 8957 | |
| 8958 format %{ "sall $dst, $shift" %} | |
| 8959 opcode(0xC1, 0x4); /* C1 /4 ib */ | |
| 8960 ins_encode(reg_opc_imm(dst, shift)); | |
| 8961 ins_pipe(ialu_reg); | |
| 8962 %} | |
| 8963 | |
| 8964 // Shift Left by 8-bit immediate | |
| 8965 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 8966 %{ | |
| 8967 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift))); | |
| 8968 effect(KILL cr); | |
| 8969 | |
| 8970 format %{ "sall $dst, $shift" %} | |
| 8971 opcode(0xC1, 0x4); /* C1 /4 ib */ | |
| 8972 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 8973 ins_pipe(ialu_mem_imm); | |
| 8974 %} | |
| 8975 | |
| 8976 // Shift Left by variable | |
| 8977 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr) | |
| 8978 %{ | |
| 8979 match(Set dst (LShiftI dst shift)); | |
| 8980 effect(KILL cr); | |
| 8981 | |
| 8982 format %{ "sall $dst, $shift" %} | |
| 8983 opcode(0xD3, 0x4); /* D3 /4 */ | |
| 8984 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 8985 ins_pipe(ialu_reg_reg); | |
| 8986 %} | |
| 8987 | |
| 8988 // Shift Left by variable | |
| 8989 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 8990 %{ | |
| 8991 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift))); | |
| 8992 effect(KILL cr); | |
| 8993 | |
| 8994 format %{ "sall $dst, $shift" %} | |
| 8995 opcode(0xD3, 0x4); /* D3 /4 */ | |
| 8996 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 8997 ins_pipe(ialu_mem_reg); | |
| 8998 %} | |
| 8999 | |
| 9000 // Arithmetic shift right by one | |
| 9001 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr) | |
| 9002 %{ | |
| 9003 match(Set dst (RShiftI dst shift)); | |
| 9004 effect(KILL cr); | |
| 9005 | |
| 9006 format %{ "sarl $dst, $shift" %} | |
| 9007 opcode(0xD1, 0x7); /* D1 /7 */ | |
| 9008 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9009 ins_pipe(ialu_reg); | |
| 9010 %} | |
| 9011 | |
| 9012 // Arithmetic shift right by one | |
| 9013 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 9014 %{ | |
| 9015 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift))); | |
| 9016 effect(KILL cr); | |
| 9017 | |
| 9018 format %{ "sarl $dst, $shift" %} | |
| 9019 opcode(0xD1, 0x7); /* D1 /7 */ | |
| 9020 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9021 ins_pipe(ialu_mem_imm); | |
| 9022 %} | |
| 9023 | |
| 9024 // Arithmetic Shift Right by 8-bit immediate | |
| 9025 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr) | |
| 9026 %{ | |
| 9027 match(Set dst (RShiftI dst shift)); | |
| 9028 effect(KILL cr); | |
| 9029 | |
| 9030 format %{ "sarl $dst, $shift" %} | |
| 9031 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 9032 ins_encode(reg_opc_imm(dst, shift)); | |
| 9033 ins_pipe(ialu_mem_imm); | |
| 9034 %} | |
| 9035 | |
| 9036 // Arithmetic Shift Right by 8-bit immediate | |
| 9037 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 9038 %{ | |
| 9039 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift))); | |
| 9040 effect(KILL cr); | |
| 9041 | |
| 9042 format %{ "sarl $dst, $shift" %} | |
| 9043 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 9044 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 9045 ins_pipe(ialu_mem_imm); | |
| 9046 %} | |
| 9047 | |
| 9048 // Arithmetic Shift Right by variable | |
| 9049 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr) | |
| 9050 %{ | |
| 9051 match(Set dst (RShiftI dst shift)); | |
| 9052 effect(KILL cr); | |
| 9053 | |
| 9054 format %{ "sarl $dst, $shift" %} | |
| 9055 opcode(0xD3, 0x7); /* D3 /7 */ | |
| 9056 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9057 ins_pipe(ialu_reg_reg); | |
| 9058 %} | |
| 9059 | |
| 9060 // Arithmetic Shift Right by variable | |
| 9061 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 9062 %{ | |
| 9063 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift))); | |
| 9064 effect(KILL cr); | |
| 9065 | |
| 9066 format %{ "sarl $dst, $shift" %} | |
| 9067 opcode(0xD3, 0x7); /* D3 /7 */ | |
| 9068 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9069 ins_pipe(ialu_mem_reg); | |
| 9070 %} | |
| 9071 | |
| 9072 // Logical shift right by one | |
| 9073 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr) | |
| 9074 %{ | |
| 9075 match(Set dst (URShiftI dst shift)); | |
| 9076 effect(KILL cr); | |
| 9077 | |
| 9078 format %{ "shrl $dst, $shift" %} | |
| 9079 opcode(0xD1, 0x5); /* D1 /5 */ | |
| 9080 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9081 ins_pipe(ialu_reg); | |
| 9082 %} | |
| 9083 | |
| 9084 // Logical shift right by one | |
| 9085 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 9086 %{ | |
| 9087 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift))); | |
| 9088 effect(KILL cr); | |
| 9089 | |
| 9090 format %{ "shrl $dst, $shift" %} | |
| 9091 opcode(0xD1, 0x5); /* D1 /5 */ | |
| 9092 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9093 ins_pipe(ialu_mem_imm); | |
| 9094 %} | |
| 9095 | |
| 9096 // Logical Shift Right by 8-bit immediate | |
| 9097 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr) | |
| 9098 %{ | |
| 9099 match(Set dst (URShiftI dst shift)); | |
| 9100 effect(KILL cr); | |
| 9101 | |
| 9102 format %{ "shrl $dst, $shift" %} | |
| 9103 opcode(0xC1, 0x5); /* C1 /5 ib */ | |
| 9104 ins_encode(reg_opc_imm(dst, shift)); | |
| 9105 ins_pipe(ialu_reg); | |
| 9106 %} | |
| 9107 | |
| 9108 // Logical Shift Right by 8-bit immediate | |
| 9109 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 9110 %{ | |
| 9111 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift))); | |
| 9112 effect(KILL cr); | |
| 9113 | |
| 9114 format %{ "shrl $dst, $shift" %} | |
| 9115 opcode(0xC1, 0x5); /* C1 /5 ib */ | |
| 9116 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 9117 ins_pipe(ialu_mem_imm); | |
| 9118 %} | |
| 9119 | |
| 9120 // Logical Shift Right by variable | |
| 9121 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr) | |
| 9122 %{ | |
| 9123 match(Set dst (URShiftI dst shift)); | |
| 9124 effect(KILL cr); | |
| 9125 | |
| 9126 format %{ "shrl $dst, $shift" %} | |
| 9127 opcode(0xD3, 0x5); /* D3 /5 */ | |
| 9128 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9129 ins_pipe(ialu_reg_reg); | |
| 9130 %} | |
| 9131 | |
| 9132 // Logical Shift Right by variable | |
| 9133 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 9134 %{ | |
| 9135 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift))); | |
| 9136 effect(KILL cr); | |
| 9137 | |
| 9138 format %{ "shrl $dst, $shift" %} | |
| 9139 opcode(0xD3, 0x5); /* D3 /5 */ | |
| 9140 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9141 ins_pipe(ialu_mem_reg); | |
| 9142 %} | |
| 9143 | |
| 9144 // Long Shift Instructions | |
| 9145 // Shift Left by one | |
| 9146 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr) | |
| 9147 %{ | |
| 9148 match(Set dst (LShiftL dst shift)); | |
| 9149 effect(KILL cr); | |
| 9150 | |
| 9151 format %{ "salq $dst, $shift" %} | |
| 9152 opcode(0xD1, 0x4); /* D1 /4 */ | |
| 9153 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9154 ins_pipe(ialu_reg); | |
| 9155 %} | |
| 9156 | |
| 9157 // Shift Left by one | |
| 9158 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 9159 %{ | |
| 9160 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift))); | |
| 9161 effect(KILL cr); | |
| 9162 | |
| 9163 format %{ "salq $dst, $shift" %} | |
| 9164 opcode(0xD1, 0x4); /* D1 /4 */ | |
| 9165 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9166 ins_pipe(ialu_mem_imm); | |
| 9167 %} | |
| 9168 | |
| 9169 // Shift Left by 8-bit immediate | |
| 9170 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr) | |
| 9171 %{ | |
| 9172 match(Set dst (LShiftL dst shift)); | |
| 9173 effect(KILL cr); | |
| 9174 | |
| 9175 format %{ "salq $dst, $shift" %} | |
| 9176 opcode(0xC1, 0x4); /* C1 /4 ib */ | |
| 9177 ins_encode(reg_opc_imm_wide(dst, shift)); | |
| 9178 ins_pipe(ialu_reg); | |
| 9179 %} | |
| 9180 | |
| 9181 // Shift Left by 8-bit immediate | |
| 9182 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 9183 %{ | |
| 9184 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift))); | |
| 9185 effect(KILL cr); | |
| 9186 | |
| 9187 format %{ "salq $dst, $shift" %} | |
| 9188 opcode(0xC1, 0x4); /* C1 /4 ib */ | |
| 9189 ins_encode(REX_mem_wide(dst), OpcP, | |
| 9190 RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 9191 ins_pipe(ialu_mem_imm); | |
| 9192 %} | |
| 9193 | |
| 9194 // Shift Left by variable | |
| 9195 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr) | |
| 9196 %{ | |
| 9197 match(Set dst (LShiftL dst shift)); | |
| 9198 effect(KILL cr); | |
| 9199 | |
| 9200 format %{ "salq $dst, $shift" %} | |
| 9201 opcode(0xD3, 0x4); /* D3 /4 */ | |
| 9202 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9203 ins_pipe(ialu_reg_reg); | |
| 9204 %} | |
| 9205 | |
| 9206 // Shift Left by variable | |
| 9207 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 9208 %{ | |
| 9209 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift))); | |
| 9210 effect(KILL cr); | |
| 9211 | |
| 9212 format %{ "salq $dst, $shift" %} | |
| 9213 opcode(0xD3, 0x4); /* D3 /4 */ | |
| 9214 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9215 ins_pipe(ialu_mem_reg); | |
| 9216 %} | |
| 9217 | |
| 9218 // Arithmetic shift right by one | |
| 9219 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr) | |
| 9220 %{ | |
| 9221 match(Set dst (RShiftL dst shift)); | |
| 9222 effect(KILL cr); | |
| 9223 | |
| 9224 format %{ "sarq $dst, $shift" %} | |
| 9225 opcode(0xD1, 0x7); /* D1 /7 */ | |
| 9226 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9227 ins_pipe(ialu_reg); | |
| 9228 %} | |
| 9229 | |
| 9230 // Arithmetic shift right by one | |
| 9231 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 9232 %{ | |
| 9233 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift))); | |
| 9234 effect(KILL cr); | |
| 9235 | |
| 9236 format %{ "sarq $dst, $shift" %} | |
| 9237 opcode(0xD1, 0x7); /* D1 /7 */ | |
| 9238 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9239 ins_pipe(ialu_mem_imm); | |
| 9240 %} | |
| 9241 | |
| 9242 // Arithmetic Shift Right by 8-bit immediate | |
| 9243 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr) | |
| 9244 %{ | |
| 9245 match(Set dst (RShiftL dst shift)); | |
| 9246 effect(KILL cr); | |
| 9247 | |
| 9248 format %{ "sarq $dst, $shift" %} | |
| 9249 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 9250 ins_encode(reg_opc_imm_wide(dst, shift)); | |
| 9251 ins_pipe(ialu_mem_imm); | |
| 9252 %} | |
| 9253 | |
| 9254 // Arithmetic Shift Right by 8-bit immediate | |
| 9255 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 9256 %{ | |
| 9257 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift))); | |
| 9258 effect(KILL cr); | |
| 9259 | |
| 9260 format %{ "sarq $dst, $shift" %} | |
| 9261 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 9262 ins_encode(REX_mem_wide(dst), OpcP, | |
| 9263 RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 9264 ins_pipe(ialu_mem_imm); | |
| 9265 %} | |
| 9266 | |
| 9267 // Arithmetic Shift Right by variable | |
| 9268 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr) | |
| 9269 %{ | |
| 9270 match(Set dst (RShiftL dst shift)); | |
| 9271 effect(KILL cr); | |
| 9272 | |
| 9273 format %{ "sarq $dst, $shift" %} | |
| 9274 opcode(0xD3, 0x7); /* D3 /7 */ | |
| 9275 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9276 ins_pipe(ialu_reg_reg); | |
| 9277 %} | |
| 9278 | |
| 9279 // Arithmetic Shift Right by variable | |
| 9280 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 9281 %{ | |
| 9282 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift))); | |
| 9283 effect(KILL cr); | |
| 9284 | |
| 9285 format %{ "sarq $dst, $shift" %} | |
| 9286 opcode(0xD3, 0x7); /* D3 /7 */ | |
| 9287 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9288 ins_pipe(ialu_mem_reg); | |
| 9289 %} | |
| 9290 | |
| 9291 // Logical shift right by one | |
| 9292 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr) | |
| 9293 %{ | |
| 9294 match(Set dst (URShiftL dst shift)); | |
| 9295 effect(KILL cr); | |
| 9296 | |
| 9297 format %{ "shrq $dst, $shift" %} | |
| 9298 opcode(0xD1, 0x5); /* D1 /5 */ | |
| 9299 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst )); | |
| 9300 ins_pipe(ialu_reg); | |
| 9301 %} | |
| 9302 | |
| 9303 // Logical shift right by one | |
| 9304 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr) | |
| 9305 %{ | |
| 9306 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift))); | |
| 9307 effect(KILL cr); | |
| 9308 | |
| 9309 format %{ "shrq $dst, $shift" %} | |
| 9310 opcode(0xD1, 0x5); /* D1 /5 */ | |
| 9311 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9312 ins_pipe(ialu_mem_imm); | |
| 9313 %} | |
| 9314 | |
| 9315 // Logical Shift Right by 8-bit immediate | |
| 9316 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr) | |
| 9317 %{ | |
| 9318 match(Set dst (URShiftL dst shift)); | |
| 9319 effect(KILL cr); | |
| 9320 | |
| 9321 format %{ "shrq $dst, $shift" %} | |
| 9322 opcode(0xC1, 0x5); /* C1 /5 ib */ | |
| 9323 ins_encode(reg_opc_imm_wide(dst, shift)); | |
| 9324 ins_pipe(ialu_reg); | |
| 9325 %} | |
| 9326 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
9327 |
| 0 | 9328 // Logical Shift Right by 8-bit immediate |
| 9329 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr) | |
| 9330 %{ | |
| 9331 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift))); | |
| 9332 effect(KILL cr); | |
| 9333 | |
| 9334 format %{ "shrq $dst, $shift" %} | |
| 9335 opcode(0xC1, 0x5); /* C1 /5 ib */ | |
| 9336 ins_encode(REX_mem_wide(dst), OpcP, | |
| 9337 RM_opc_mem(secondary, dst), Con8or32(shift)); | |
| 9338 ins_pipe(ialu_mem_imm); | |
| 9339 %} | |
| 9340 | |
| 9341 // Logical Shift Right by variable | |
| 9342 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr) | |
| 9343 %{ | |
| 9344 match(Set dst (URShiftL dst shift)); | |
| 9345 effect(KILL cr); | |
| 9346 | |
| 9347 format %{ "shrq $dst, $shift" %} | |
| 9348 opcode(0xD3, 0x5); /* D3 /5 */ | |
| 9349 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9350 ins_pipe(ialu_reg_reg); | |
| 9351 %} | |
| 9352 | |
| 9353 // Logical Shift Right by variable | |
| 9354 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr) | |
| 9355 %{ | |
| 9356 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift))); | |
| 9357 effect(KILL cr); | |
| 9358 | |
| 9359 format %{ "shrq $dst, $shift" %} | |
| 9360 opcode(0xD3, 0x5); /* D3 /5 */ | |
| 9361 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst)); | |
| 9362 ins_pipe(ialu_mem_reg); | |
| 9363 %} | |
| 9364 | |
| 9365 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24. | |
| 9366 // This idiom is used by the compiler for the i2b bytecode. | |
| 9367 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour) | |
| 9368 %{ | |
| 9369 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour)); | |
| 9370 | |
| 9371 format %{ "movsbl $dst, $src\t# i2b" %} | |
| 9372 opcode(0x0F, 0xBE); | |
| 9373 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 9374 ins_pipe(ialu_reg_reg); | |
| 9375 %} | |
| 9376 | |
| 9377 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16. | |
| 9378 // This idiom is used by the compiler the i2s bytecode. | |
| 9379 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen) | |
| 9380 %{ | |
| 9381 match(Set dst (RShiftI (LShiftI src sixteen) sixteen)); | |
| 9382 | |
| 9383 format %{ "movswl $dst, $src\t# i2s" %} | |
| 9384 opcode(0x0F, 0xBF); | |
| 9385 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 9386 ins_pipe(ialu_reg_reg); | |
| 9387 %} | |
| 9388 | |
| 9389 // ROL/ROR instructions | |
| 9390 | |
| 9391 // ROL expand | |
| 9392 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{ | |
| 9393 effect(KILL cr, USE_DEF dst); | |
| 9394 | |
| 9395 format %{ "roll $dst" %} | |
| 9396 opcode(0xD1, 0x0); /* Opcode D1 /0 */ | |
| 9397 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9398 ins_pipe(ialu_reg); | |
| 9399 %} | |
| 9400 | |
| 9401 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{ | |
| 9402 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9403 | |
| 9404 format %{ "roll $dst, $shift" %} | |
| 9405 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */ | |
| 9406 ins_encode( reg_opc_imm(dst, shift) ); | |
| 9407 ins_pipe(ialu_reg); | |
| 9408 %} | |
| 9409 | |
| 9410 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr) | |
| 9411 %{ | |
| 9412 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9413 | |
| 9414 format %{ "roll $dst, $shift" %} | |
| 9415 opcode(0xD3, 0x0); /* Opcode D3 /0 */ | |
| 9416 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9417 ins_pipe(ialu_reg_reg); | |
| 9418 %} | |
| 9419 // end of ROL expand | |
| 9420 | |
| 9421 // Rotate Left by one | |
| 9422 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr) | |
| 9423 %{ | |
| 9424 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift))); | |
| 9425 | |
| 9426 expand %{ | |
| 9427 rolI_rReg_imm1(dst, cr); | |
| 9428 %} | |
| 9429 %} | |
| 9430 | |
| 9431 // Rotate Left by 8-bit immediate | |
| 9432 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr) | |
| 9433 %{ | |
| 9434 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f)); | |
| 9435 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift))); | |
| 9436 | |
| 9437 expand %{ | |
| 9438 rolI_rReg_imm8(dst, lshift, cr); | |
| 9439 %} | |
| 9440 %} | |
| 9441 | |
| 9442 // Rotate Left by variable | |
| 9443 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr) | |
| 9444 %{ | |
| 9445 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift)))); | |
| 9446 | |
| 9447 expand %{ | |
| 9448 rolI_rReg_CL(dst, shift, cr); | |
| 9449 %} | |
| 9450 %} | |
| 9451 | |
| 9452 // Rotate Left by variable | |
| 9453 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr) | |
| 9454 %{ | |
| 9455 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift)))); | |
| 9456 | |
| 9457 expand %{ | |
| 9458 rolI_rReg_CL(dst, shift, cr); | |
| 9459 %} | |
| 9460 %} | |
| 9461 | |
| 9462 // ROR expand | |
| 9463 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr) | |
| 9464 %{ | |
| 9465 effect(USE_DEF dst, KILL cr); | |
| 9466 | |
| 9467 format %{ "rorl $dst" %} | |
| 9468 opcode(0xD1, 0x1); /* D1 /1 */ | |
| 9469 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9470 ins_pipe(ialu_reg); | |
| 9471 %} | |
| 9472 | |
| 9473 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) | |
| 9474 %{ | |
| 9475 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9476 | |
| 9477 format %{ "rorl $dst, $shift" %} | |
| 9478 opcode(0xC1, 0x1); /* C1 /1 ib */ | |
| 9479 ins_encode(reg_opc_imm(dst, shift)); | |
| 9480 ins_pipe(ialu_reg); | |
| 9481 %} | |
| 9482 | |
| 9483 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr) | |
| 9484 %{ | |
| 9485 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9486 | |
| 9487 format %{ "rorl $dst, $shift" %} | |
| 9488 opcode(0xD3, 0x1); /* D3 /1 */ | |
| 9489 ins_encode(REX_reg(dst), OpcP, reg_opc(dst)); | |
| 9490 ins_pipe(ialu_reg_reg); | |
| 9491 %} | |
| 9492 // end of ROR expand | |
| 9493 | |
| 9494 // Rotate Right by one | |
| 9495 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr) | |
| 9496 %{ | |
| 9497 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift))); | |
| 9498 | |
| 9499 expand %{ | |
| 9500 rorI_rReg_imm1(dst, cr); | |
| 9501 %} | |
| 9502 %} | |
| 9503 | |
| 9504 // Rotate Right by 8-bit immediate | |
| 9505 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr) | |
| 9506 %{ | |
| 9507 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f)); | |
| 9508 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift))); | |
| 9509 | |
| 9510 expand %{ | |
| 9511 rorI_rReg_imm8(dst, rshift, cr); | |
| 9512 %} | |
| 9513 %} | |
| 9514 | |
| 9515 // Rotate Right by variable | |
| 9516 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr) | |
| 9517 %{ | |
| 9518 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift)))); | |
| 9519 | |
| 9520 expand %{ | |
| 9521 rorI_rReg_CL(dst, shift, cr); | |
| 9522 %} | |
| 9523 %} | |
| 9524 | |
| 9525 // Rotate Right by variable | |
| 9526 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr) | |
| 9527 %{ | |
| 9528 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift)))); | |
| 9529 | |
| 9530 expand %{ | |
| 9531 rorI_rReg_CL(dst, shift, cr); | |
| 9532 %} | |
| 9533 %} | |
| 9534 | |
| 9535 // for long rotate | |
| 9536 // ROL expand | |
| 9537 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{ | |
| 9538 effect(USE_DEF dst, KILL cr); | |
| 9539 | |
| 9540 format %{ "rolq $dst" %} | |
| 9541 opcode(0xD1, 0x0); /* Opcode D1 /0 */ | |
| 9542 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9543 ins_pipe(ialu_reg); | |
| 9544 %} | |
| 9545 | |
| 9546 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{ | |
| 9547 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9548 | |
| 9549 format %{ "rolq $dst, $shift" %} | |
| 9550 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */ | |
| 9551 ins_encode( reg_opc_imm_wide(dst, shift) ); | |
| 9552 ins_pipe(ialu_reg); | |
| 9553 %} | |
| 9554 | |
| 9555 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr) | |
| 9556 %{ | |
| 9557 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9558 | |
| 9559 format %{ "rolq $dst, $shift" %} | |
| 9560 opcode(0xD3, 0x0); /* Opcode D3 /0 */ | |
| 9561 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9562 ins_pipe(ialu_reg_reg); | |
| 9563 %} | |
| 9564 // end of ROL expand | |
| 9565 | |
| 9566 // Rotate Left by one | |
| 9567 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr) | |
| 9568 %{ | |
| 9569 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift))); | |
| 9570 | |
| 9571 expand %{ | |
| 9572 rolL_rReg_imm1(dst, cr); | |
| 9573 %} | |
| 9574 %} | |
| 9575 | |
| 9576 // Rotate Left by 8-bit immediate | |
| 9577 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr) | |
| 9578 %{ | |
| 9579 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f)); | |
| 9580 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift))); | |
| 9581 | |
| 9582 expand %{ | |
| 9583 rolL_rReg_imm8(dst, lshift, cr); | |
| 9584 %} | |
| 9585 %} | |
| 9586 | |
| 9587 // Rotate Left by variable | |
| 9588 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr) | |
| 9589 %{ | |
| 9590 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift)))); | |
| 9591 | |
| 9592 expand %{ | |
| 9593 rolL_rReg_CL(dst, shift, cr); | |
| 9594 %} | |
| 9595 %} | |
| 9596 | |
| 9597 // Rotate Left by variable | |
| 9598 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr) | |
| 9599 %{ | |
| 9600 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift)))); | |
| 9601 | |
| 9602 expand %{ | |
| 9603 rolL_rReg_CL(dst, shift, cr); | |
| 9604 %} | |
| 9605 %} | |
| 9606 | |
| 9607 // ROR expand | |
| 9608 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr) | |
| 9609 %{ | |
| 9610 effect(USE_DEF dst, KILL cr); | |
| 9611 | |
| 9612 format %{ "rorq $dst" %} | |
| 9613 opcode(0xD1, 0x1); /* D1 /1 */ | |
| 9614 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9615 ins_pipe(ialu_reg); | |
| 9616 %} | |
| 9617 | |
| 9618 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) | |
| 9619 %{ | |
| 9620 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9621 | |
| 9622 format %{ "rorq $dst, $shift" %} | |
| 9623 opcode(0xC1, 0x1); /* C1 /1 ib */ | |
| 9624 ins_encode(reg_opc_imm_wide(dst, shift)); | |
| 9625 ins_pipe(ialu_reg); | |
| 9626 %} | |
| 9627 | |
| 9628 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr) | |
| 9629 %{ | |
| 9630 effect(USE_DEF dst, USE shift, KILL cr); | |
| 9631 | |
| 9632 format %{ "rorq $dst, $shift" %} | |
| 9633 opcode(0xD3, 0x1); /* D3 /1 */ | |
| 9634 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst)); | |
| 9635 ins_pipe(ialu_reg_reg); | |
| 9636 %} | |
| 9637 // end of ROR expand | |
| 9638 | |
| 9639 // Rotate Right by one | |
| 9640 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr) | |
| 9641 %{ | |
| 9642 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift))); | |
| 9643 | |
| 9644 expand %{ | |
| 9645 rorL_rReg_imm1(dst, cr); | |
| 9646 %} | |
| 9647 %} | |
| 9648 | |
| 9649 // Rotate Right by 8-bit immediate | |
| 9650 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr) | |
| 9651 %{ | |
| 9652 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f)); | |
| 9653 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift))); | |
| 9654 | |
| 9655 expand %{ | |
| 9656 rorL_rReg_imm8(dst, rshift, cr); | |
| 9657 %} | |
| 9658 %} | |
| 9659 | |
| 9660 // Rotate Right by variable | |
| 9661 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr) | |
| 9662 %{ | |
| 9663 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift)))); | |
| 9664 | |
| 9665 expand %{ | |
| 9666 rorL_rReg_CL(dst, shift, cr); | |
| 9667 %} | |
| 9668 %} | |
| 9669 | |
| 9670 // Rotate Right by variable | |
| 9671 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr) | |
| 9672 %{ | |
| 9673 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift)))); | |
| 9674 | |
| 9675 expand %{ | |
| 9676 rorL_rReg_CL(dst, shift, cr); | |
| 9677 %} | |
| 9678 %} | |
| 9679 | |
| 9680 // Logical Instructions | |
| 9681 | |
| 9682 // Integer Logical Instructions | |
| 9683 | |
| 9684 // And Instructions | |
| 9685 // And Register with Register | |
| 9686 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 9687 %{ | |
| 9688 match(Set dst (AndI dst src)); | |
| 9689 effect(KILL cr); | |
| 9690 | |
| 9691 format %{ "andl $dst, $src\t# int" %} | |
| 9692 opcode(0x23); | |
| 9693 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src)); | |
| 9694 ins_pipe(ialu_reg_reg); | |
| 9695 %} | |
| 9696 | |
| 9697 // And Register with Immediate 255 | |
| 9698 instruct andI_rReg_imm255(rRegI dst, immI_255 src) | |
| 9699 %{ | |
| 9700 match(Set dst (AndI dst src)); | |
| 9701 | |
| 9702 format %{ "movzbl $dst, $dst\t# int & 0xFF" %} | |
| 9703 opcode(0x0F, 0xB6); | |
| 9704 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); | |
| 9705 ins_pipe(ialu_reg); | |
| 9706 %} | |
| 9707 | |
| 9708 // And Register with Immediate 255 and promote to long | |
| 9709 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask) | |
| 9710 %{ | |
| 9711 match(Set dst (ConvI2L (AndI src mask))); | |
| 9712 | |
| 9713 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %} | |
| 9714 opcode(0x0F, 0xB6); | |
| 9715 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 9716 ins_pipe(ialu_reg); | |
| 9717 %} | |
| 9718 | |
| 9719 // And Register with Immediate 65535 | |
| 9720 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src) | |
| 9721 %{ | |
| 9722 match(Set dst (AndI dst src)); | |
| 9723 | |
| 9724 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %} | |
| 9725 opcode(0x0F, 0xB7); | |
| 9726 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); | |
| 9727 ins_pipe(ialu_reg); | |
| 9728 %} | |
| 9729 | |
| 9730 // And Register with Immediate 65535 and promote to long | |
| 9731 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask) | |
| 9732 %{ | |
| 9733 match(Set dst (ConvI2L (AndI src mask))); | |
| 9734 | |
| 9735 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %} | |
| 9736 opcode(0x0F, 0xB7); | |
| 9737 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 9738 ins_pipe(ialu_reg); | |
| 9739 %} | |
| 9740 | |
| 9741 // And Register with Immediate | |
| 9742 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr) | |
| 9743 %{ | |
| 9744 match(Set dst (AndI dst src)); | |
| 9745 effect(KILL cr); | |
| 9746 | |
| 9747 format %{ "andl $dst, $src\t# int" %} | |
| 9748 opcode(0x81, 0x04); /* Opcode 81 /4 */ | |
| 9749 ins_encode(OpcSErm(dst, src), Con8or32(src)); | |
| 9750 ins_pipe(ialu_reg); | |
| 9751 %} | |
| 9752 | |
| 9753 // And Register with Memory | |
| 9754 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 9755 %{ | |
| 9756 match(Set dst (AndI dst (LoadI src))); | |
| 9757 effect(KILL cr); | |
| 9758 | |
| 9759 ins_cost(125); | |
| 9760 format %{ "andl $dst, $src\t# int" %} | |
| 9761 opcode(0x23); | |
| 9762 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 9763 ins_pipe(ialu_reg_mem); | |
| 9764 %} | |
| 9765 | |
| 9766 // And Memory with Register | |
| 9767 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr) | |
| 9768 %{ | |
| 9769 match(Set dst (StoreI dst (AndI (LoadI dst) src))); | |
| 9770 effect(KILL cr); | |
| 9771 | |
| 9772 ins_cost(150); | |
| 9773 format %{ "andl $dst, $src\t# int" %} | |
| 9774 opcode(0x21); /* Opcode 21 /r */ | |
| 9775 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 9776 ins_pipe(ialu_mem_reg); | |
| 9777 %} | |
| 9778 | |
| 9779 // And Memory with Immediate | |
| 9780 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr) | |
| 9781 %{ | |
| 9782 match(Set dst (StoreI dst (AndI (LoadI dst) src))); | |
| 9783 effect(KILL cr); | |
| 9784 | |
| 9785 ins_cost(125); | |
| 9786 format %{ "andl $dst, $src\t# int" %} | |
| 9787 opcode(0x81, 0x4); /* Opcode 81 /4 id */ | |
| 9788 ins_encode(REX_mem(dst), OpcSE(src), | |
| 9789 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 9790 ins_pipe(ialu_mem_imm); | |
| 9791 %} | |
| 9792 | |
| 9793 // Or Instructions | |
| 9794 // Or Register with Register | |
| 9795 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 9796 %{ | |
| 9797 match(Set dst (OrI dst src)); | |
| 9798 effect(KILL cr); | |
| 9799 | |
| 9800 format %{ "orl $dst, $src\t# int" %} | |
| 9801 opcode(0x0B); | |
| 9802 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src)); | |
| 9803 ins_pipe(ialu_reg_reg); | |
| 9804 %} | |
| 9805 | |
| 9806 // Or Register with Immediate | |
| 9807 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr) | |
| 9808 %{ | |
| 9809 match(Set dst (OrI dst src)); | |
| 9810 effect(KILL cr); | |
| 9811 | |
| 9812 format %{ "orl $dst, $src\t# int" %} | |
| 9813 opcode(0x81, 0x01); /* Opcode 81 /1 id */ | |
| 9814 ins_encode(OpcSErm(dst, src), Con8or32(src)); | |
| 9815 ins_pipe(ialu_reg); | |
| 9816 %} | |
| 9817 | |
| 9818 // Or Register with Memory | |
| 9819 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 9820 %{ | |
| 9821 match(Set dst (OrI dst (LoadI src))); | |
| 9822 effect(KILL cr); | |
| 9823 | |
| 9824 ins_cost(125); | |
| 9825 format %{ "orl $dst, $src\t# int" %} | |
| 9826 opcode(0x0B); | |
| 9827 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 9828 ins_pipe(ialu_reg_mem); | |
| 9829 %} | |
| 9830 | |
| 9831 // Or Memory with Register | |
| 9832 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr) | |
| 9833 %{ | |
| 9834 match(Set dst (StoreI dst (OrI (LoadI dst) src))); | |
| 9835 effect(KILL cr); | |
| 9836 | |
| 9837 ins_cost(150); | |
| 9838 format %{ "orl $dst, $src\t# int" %} | |
| 9839 opcode(0x09); /* Opcode 09 /r */ | |
| 9840 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 9841 ins_pipe(ialu_mem_reg); | |
| 9842 %} | |
| 9843 | |
| 9844 // Or Memory with Immediate | |
| 9845 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr) | |
| 9846 %{ | |
| 9847 match(Set dst (StoreI dst (OrI (LoadI dst) src))); | |
| 9848 effect(KILL cr); | |
| 9849 | |
| 9850 ins_cost(125); | |
| 9851 format %{ "orl $dst, $src\t# int" %} | |
| 9852 opcode(0x81, 0x1); /* Opcode 81 /1 id */ | |
| 9853 ins_encode(REX_mem(dst), OpcSE(src), | |
| 9854 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 9855 ins_pipe(ialu_mem_imm); | |
| 9856 %} | |
| 9857 | |
| 9858 // Xor Instructions | |
| 9859 // Xor Register with Register | |
| 9860 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr) | |
| 9861 %{ | |
| 9862 match(Set dst (XorI dst src)); | |
| 9863 effect(KILL cr); | |
| 9864 | |
| 9865 format %{ "xorl $dst, $src\t# int" %} | |
| 9866 opcode(0x33); | |
| 9867 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src)); | |
| 9868 ins_pipe(ialu_reg_reg); | |
| 9869 %} | |
| 9870 | |
| 403 | 9871 // Xor Register with Immediate -1 |
| 9872 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{ | |
| 9873 match(Set dst (XorI dst imm)); | |
| 9874 | |
| 9875 format %{ "not $dst" %} | |
| 9876 ins_encode %{ | |
| 9877 __ notl($dst$$Register); | |
| 9878 %} | |
| 9879 ins_pipe(ialu_reg); | |
| 9880 %} | |
| 9881 | |
| 0 | 9882 // Xor Register with Immediate |
| 9883 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr) | |
| 9884 %{ | |
| 9885 match(Set dst (XorI dst src)); | |
| 9886 effect(KILL cr); | |
| 9887 | |
| 9888 format %{ "xorl $dst, $src\t# int" %} | |
| 9889 opcode(0x81, 0x06); /* Opcode 81 /6 id */ | |
| 9890 ins_encode(OpcSErm(dst, src), Con8or32(src)); | |
| 9891 ins_pipe(ialu_reg); | |
| 9892 %} | |
| 9893 | |
| 9894 // Xor Register with Memory | |
| 9895 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr) | |
| 9896 %{ | |
| 9897 match(Set dst (XorI dst (LoadI src))); | |
| 9898 effect(KILL cr); | |
| 9899 | |
| 9900 ins_cost(125); | |
| 9901 format %{ "xorl $dst, $src\t# int" %} | |
| 9902 opcode(0x33); | |
| 9903 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 9904 ins_pipe(ialu_reg_mem); | |
| 9905 %} | |
| 9906 | |
| 9907 // Xor Memory with Register | |
| 9908 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr) | |
| 9909 %{ | |
| 9910 match(Set dst (StoreI dst (XorI (LoadI dst) src))); | |
| 9911 effect(KILL cr); | |
| 9912 | |
| 9913 ins_cost(150); | |
| 9914 format %{ "xorl $dst, $src\t# int" %} | |
| 9915 opcode(0x31); /* Opcode 31 /r */ | |
| 9916 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 9917 ins_pipe(ialu_mem_reg); | |
| 9918 %} | |
| 9919 | |
| 9920 // Xor Memory with Immediate | |
| 9921 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr) | |
| 9922 %{ | |
| 9923 match(Set dst (StoreI dst (XorI (LoadI dst) src))); | |
| 9924 effect(KILL cr); | |
| 9925 | |
| 9926 ins_cost(125); | |
| 9927 format %{ "xorl $dst, $src\t# int" %} | |
| 9928 opcode(0x81, 0x6); /* Opcode 81 /6 id */ | |
| 9929 ins_encode(REX_mem(dst), OpcSE(src), | |
| 9930 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 9931 ins_pipe(ialu_mem_imm); | |
| 9932 %} | |
| 9933 | |
| 9934 | |
| 9935 // Long Logical Instructions | |
| 9936 | |
| 9937 // And Instructions | |
| 9938 // And Register with Register | |
| 9939 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 9940 %{ | |
| 9941 match(Set dst (AndL dst src)); | |
| 9942 effect(KILL cr); | |
| 9943 | |
| 9944 format %{ "andq $dst, $src\t# long" %} | |
| 9945 opcode(0x23); | |
| 9946 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 9947 ins_pipe(ialu_reg_reg); | |
| 9948 %} | |
| 9949 | |
| 9950 // And Register with Immediate 255 | |
| 9951 instruct andL_rReg_imm255(rRegL dst, immL_255 src) | |
| 9952 %{ | |
| 9953 match(Set dst (AndL dst src)); | |
| 9954 | |
|
569
2cacccded90f
6805950: Typos in andL_rReg_imm instructions in x86_64.ad
twisti
parents:
558
diff
changeset
|
9955 format %{ "movzbq $dst, $dst\t# long & 0xFF" %} |
| 0 | 9956 opcode(0x0F, 0xB6); |
| 9957 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); | |
| 9958 ins_pipe(ialu_reg); | |
| 9959 %} | |
| 9960 | |
| 9961 // And Register with Immediate 65535 | |
|
569
2cacccded90f
6805950: Typos in andL_rReg_imm instructions in x86_64.ad
twisti
parents:
558
diff
changeset
|
9962 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src) |
| 0 | 9963 %{ |
| 9964 match(Set dst (AndL dst src)); | |
| 9965 | |
| 9966 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %} | |
| 9967 opcode(0x0F, 0xB7); | |
| 9968 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); | |
| 9969 ins_pipe(ialu_reg); | |
| 9970 %} | |
| 9971 | |
| 9972 // And Register with Immediate | |
| 9973 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr) | |
| 9974 %{ | |
| 9975 match(Set dst (AndL dst src)); | |
| 9976 effect(KILL cr); | |
| 9977 | |
| 9978 format %{ "andq $dst, $src\t# long" %} | |
| 9979 opcode(0x81, 0x04); /* Opcode 81 /4 */ | |
| 9980 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 9981 ins_pipe(ialu_reg); | |
| 9982 %} | |
| 9983 | |
| 9984 // And Register with Memory | |
| 9985 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 9986 %{ | |
| 9987 match(Set dst (AndL dst (LoadL src))); | |
| 9988 effect(KILL cr); | |
| 9989 | |
| 9990 ins_cost(125); | |
| 9991 format %{ "andq $dst, $src\t# long" %} | |
| 9992 opcode(0x23); | |
| 9993 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 9994 ins_pipe(ialu_reg_mem); | |
| 9995 %} | |
| 9996 | |
| 9997 // And Memory with Register | |
| 9998 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr) | |
| 9999 %{ | |
| 10000 match(Set dst (StoreL dst (AndL (LoadL dst) src))); | |
| 10001 effect(KILL cr); | |
| 10002 | |
| 10003 ins_cost(150); | |
| 10004 format %{ "andq $dst, $src\t# long" %} | |
| 10005 opcode(0x21); /* Opcode 21 /r */ | |
| 10006 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 10007 ins_pipe(ialu_mem_reg); | |
| 10008 %} | |
| 10009 | |
| 10010 // And Memory with Immediate | |
| 10011 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr) | |
| 10012 %{ | |
| 10013 match(Set dst (StoreL dst (AndL (LoadL dst) src))); | |
| 10014 effect(KILL cr); | |
| 10015 | |
| 10016 ins_cost(125); | |
| 10017 format %{ "andq $dst, $src\t# long" %} | |
| 10018 opcode(0x81, 0x4); /* Opcode 81 /4 id */ | |
| 10019 ins_encode(REX_mem_wide(dst), OpcSE(src), | |
| 10020 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 10021 ins_pipe(ialu_mem_imm); | |
| 10022 %} | |
| 10023 | |
| 10024 // Or Instructions | |
| 10025 // Or Register with Register | |
| 10026 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 10027 %{ | |
| 10028 match(Set dst (OrL dst src)); | |
| 10029 effect(KILL cr); | |
| 10030 | |
| 10031 format %{ "orq $dst, $src\t# long" %} | |
| 10032 opcode(0x0B); | |
| 10033 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 10034 ins_pipe(ialu_reg_reg); | |
| 10035 %} | |
| 10036 | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10037 // Use any_RegP to match R15 (TLS register) without spilling. |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10038 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{ |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10039 match(Set dst (OrL dst (CastP2X src))); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10040 effect(KILL cr); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10041 |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10042 format %{ "orq $dst, $src\t# long" %} |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10043 opcode(0x0B); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10044 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10045 ins_pipe(ialu_reg_reg); |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10046 %} |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10047 |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
10048 |
| 0 | 10049 // Or Register with Immediate |
| 10050 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr) | |
| 10051 %{ | |
| 10052 match(Set dst (OrL dst src)); | |
| 10053 effect(KILL cr); | |
| 10054 | |
| 10055 format %{ "orq $dst, $src\t# long" %} | |
| 10056 opcode(0x81, 0x01); /* Opcode 81 /1 id */ | |
| 10057 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 10058 ins_pipe(ialu_reg); | |
| 10059 %} | |
| 10060 | |
| 10061 // Or Register with Memory | |
| 10062 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 10063 %{ | |
| 10064 match(Set dst (OrL dst (LoadL src))); | |
| 10065 effect(KILL cr); | |
| 10066 | |
| 10067 ins_cost(125); | |
| 10068 format %{ "orq $dst, $src\t# long" %} | |
| 10069 opcode(0x0B); | |
| 10070 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 10071 ins_pipe(ialu_reg_mem); | |
| 10072 %} | |
| 10073 | |
| 10074 // Or Memory with Register | |
| 10075 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr) | |
| 10076 %{ | |
| 10077 match(Set dst (StoreL dst (OrL (LoadL dst) src))); | |
| 10078 effect(KILL cr); | |
| 10079 | |
| 10080 ins_cost(150); | |
| 10081 format %{ "orq $dst, $src\t# long" %} | |
| 10082 opcode(0x09); /* Opcode 09 /r */ | |
| 10083 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 10084 ins_pipe(ialu_mem_reg); | |
| 10085 %} | |
| 10086 | |
| 10087 // Or Memory with Immediate | |
| 10088 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr) | |
| 10089 %{ | |
| 10090 match(Set dst (StoreL dst (OrL (LoadL dst) src))); | |
| 10091 effect(KILL cr); | |
| 10092 | |
| 10093 ins_cost(125); | |
| 10094 format %{ "orq $dst, $src\t# long" %} | |
| 10095 opcode(0x81, 0x1); /* Opcode 81 /1 id */ | |
| 10096 ins_encode(REX_mem_wide(dst), OpcSE(src), | |
| 10097 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 10098 ins_pipe(ialu_mem_imm); | |
| 10099 %} | |
| 10100 | |
| 10101 // Xor Instructions | |
| 10102 // Xor Register with Register | |
| 10103 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr) | |
| 10104 %{ | |
| 10105 match(Set dst (XorL dst src)); | |
| 10106 effect(KILL cr); | |
| 10107 | |
| 10108 format %{ "xorq $dst, $src\t# long" %} | |
| 10109 opcode(0x33); | |
| 10110 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src)); | |
| 10111 ins_pipe(ialu_reg_reg); | |
| 10112 %} | |
| 10113 | |
| 403 | 10114 // Xor Register with Immediate -1 |
| 10115 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{ | |
| 10116 match(Set dst (XorL dst imm)); | |
| 10117 | |
| 10118 format %{ "notq $dst" %} | |
| 10119 ins_encode %{ | |
| 10120 __ notq($dst$$Register); | |
| 10121 %} | |
| 10122 ins_pipe(ialu_reg); | |
| 10123 %} | |
| 10124 | |
| 0 | 10125 // Xor Register with Immediate |
| 10126 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr) | |
| 10127 %{ | |
| 10128 match(Set dst (XorL dst src)); | |
| 10129 effect(KILL cr); | |
| 10130 | |
| 10131 format %{ "xorq $dst, $src\t# long" %} | |
| 10132 opcode(0x81, 0x06); /* Opcode 81 /6 id */ | |
| 10133 ins_encode(OpcSErm_wide(dst, src), Con8or32(src)); | |
| 10134 ins_pipe(ialu_reg); | |
| 10135 %} | |
| 10136 | |
| 10137 // Xor Register with Memory | |
| 10138 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr) | |
| 10139 %{ | |
| 10140 match(Set dst (XorL dst (LoadL src))); | |
| 10141 effect(KILL cr); | |
| 10142 | |
| 10143 ins_cost(125); | |
| 10144 format %{ "xorq $dst, $src\t# long" %} | |
| 10145 opcode(0x33); | |
| 10146 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 10147 ins_pipe(ialu_reg_mem); | |
| 10148 %} | |
| 10149 | |
| 10150 // Xor Memory with Register | |
| 10151 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr) | |
| 10152 %{ | |
| 10153 match(Set dst (StoreL dst (XorL (LoadL dst) src))); | |
| 10154 effect(KILL cr); | |
| 10155 | |
| 10156 ins_cost(150); | |
| 10157 format %{ "xorq $dst, $src\t# long" %} | |
| 10158 opcode(0x31); /* Opcode 31 /r */ | |
| 10159 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 10160 ins_pipe(ialu_mem_reg); | |
| 10161 %} | |
| 10162 | |
| 10163 // Xor Memory with Immediate | |
| 10164 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr) | |
| 10165 %{ | |
| 10166 match(Set dst (StoreL dst (XorL (LoadL dst) src))); | |
| 10167 effect(KILL cr); | |
| 10168 | |
| 10169 ins_cost(125); | |
| 10170 format %{ "xorq $dst, $src\t# long" %} | |
| 10171 opcode(0x81, 0x6); /* Opcode 81 /6 id */ | |
| 10172 ins_encode(REX_mem_wide(dst), OpcSE(src), | |
| 10173 RM_opc_mem(secondary, dst), Con8or32(src)); | |
| 10174 ins_pipe(ialu_mem_imm); | |
| 10175 %} | |
| 10176 | |
| 10177 // Convert Int to Boolean | |
| 10178 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr) | |
| 10179 %{ | |
| 10180 match(Set dst (Conv2B src)); | |
| 10181 effect(KILL cr); | |
| 10182 | |
| 10183 format %{ "testl $src, $src\t# ci2b\n\t" | |
| 10184 "setnz $dst\n\t" | |
| 10185 "movzbl $dst, $dst" %} | |
| 10186 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl | |
| 10187 setNZ_reg(dst), | |
| 10188 REX_reg_breg(dst, dst), // movzbl | |
| 10189 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst)); | |
| 10190 ins_pipe(pipe_slow); // XXX | |
| 10191 %} | |
| 10192 | |
| 10193 // Convert Pointer to Boolean | |
| 10194 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr) | |
| 10195 %{ | |
| 10196 match(Set dst (Conv2B src)); | |
| 10197 effect(KILL cr); | |
| 10198 | |
| 10199 format %{ "testq $src, $src\t# cp2b\n\t" | |
| 10200 "setnz $dst\n\t" | |
| 10201 "movzbl $dst, $dst" %} | |
| 10202 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq | |
| 10203 setNZ_reg(dst), | |
| 10204 REX_reg_breg(dst, dst), // movzbl | |
| 10205 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst)); | |
| 10206 ins_pipe(pipe_slow); // XXX | |
| 10207 %} | |
| 10208 | |
| 10209 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr) | |
| 10210 %{ | |
| 10211 match(Set dst (CmpLTMask p q)); | |
| 10212 effect(KILL cr); | |
| 10213 | |
| 10214 ins_cost(400); // XXX | |
| 10215 format %{ "cmpl $p, $q\t# cmpLTMask\n\t" | |
| 10216 "setlt $dst\n\t" | |
| 10217 "movzbl $dst, $dst\n\t" | |
| 10218 "negl $dst" %} | |
| 10219 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl | |
| 10220 setLT_reg(dst), | |
| 10221 REX_reg_breg(dst, dst), // movzbl | |
| 10222 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst), | |
| 10223 neg_reg(dst)); | |
| 10224 ins_pipe(pipe_slow); | |
| 10225 %} | |
| 10226 | |
| 10227 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr) | |
| 10228 %{ | |
| 10229 match(Set dst (CmpLTMask dst zero)); | |
| 10230 effect(KILL cr); | |
| 10231 | |
| 10232 ins_cost(100); // XXX | |
| 10233 format %{ "sarl $dst, #31\t# cmpLTMask0" %} | |
| 10234 opcode(0xC1, 0x7); /* C1 /7 ib */ | |
| 10235 ins_encode(reg_opc_imm(dst, 0x1F)); | |
| 10236 ins_pipe(ialu_reg); | |
| 10237 %} | |
| 10238 | |
| 10239 | |
| 10240 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y, | |
| 10241 rRegI tmp, | |
| 10242 rFlagsReg cr) | |
| 10243 %{ | |
| 10244 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q))); | |
| 10245 effect(TEMP tmp, KILL cr); | |
| 10246 | |
| 10247 ins_cost(400); // XXX | |
| 10248 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t" | |
| 10249 "sbbl $tmp, $tmp\n\t" | |
| 10250 "andl $tmp, $y\n\t" | |
| 10251 "addl $p, $tmp" %} | |
| 10252 ins_encode(enc_cmpLTP(p, q, y, tmp)); | |
| 10253 ins_pipe(pipe_cmplt); | |
| 10254 %} | |
| 10255 | |
| 10256 /* If I enable this, I encourage spilling in the inner loop of compress. | |
| 10257 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr ) | |
| 10258 %{ | |
| 10259 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q))); | |
| 10260 effect( TEMP tmp, KILL cr ); | |
| 10261 ins_cost(400); | |
| 10262 | |
| 10263 format %{ "SUB $p,$q\n\t" | |
| 10264 "SBB RCX,RCX\n\t" | |
| 10265 "AND RCX,$y\n\t" | |
| 10266 "ADD $p,RCX" %} | |
| 10267 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) ); | |
| 10268 %} | |
| 10269 */ | |
| 10270 | |
| 10271 //---------- FP Instructions------------------------------------------------ | |
| 10272 | |
| 10273 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2) | |
| 10274 %{ | |
| 10275 match(Set cr (CmpF src1 src2)); | |
| 10276 | |
| 10277 ins_cost(145); | |
| 10278 format %{ "ucomiss $src1, $src2\n\t" | |
| 10279 "jnp,s exit\n\t" | |
| 10280 "pushfq\t# saw NaN, set CF\n\t" | |
| 10281 "andq [rsp], #0xffffff2b\n\t" | |
| 10282 "popfq\n" | |
| 10283 "exit: nop\t# avoid branch to branch" %} | |
| 10284 opcode(0x0F, 0x2E); | |
| 10285 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2), | |
| 10286 cmpfp_fixup); | |
| 10287 ins_pipe(pipe_slow); | |
| 10288 %} | |
| 10289 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10290 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10291 match(Set cr (CmpF src1 src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10292 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10293 ins_cost(145); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10294 format %{ "ucomiss $src1, $src2" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10295 ins_encode %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10296 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10297 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10298 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10299 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10300 |
| 0 | 10301 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2) |
| 10302 %{ | |
| 10303 match(Set cr (CmpF src1 (LoadF src2))); | |
| 10304 | |
| 10305 ins_cost(145); | |
| 10306 format %{ "ucomiss $src1, $src2\n\t" | |
| 10307 "jnp,s exit\n\t" | |
| 10308 "pushfq\t# saw NaN, set CF\n\t" | |
| 10309 "andq [rsp], #0xffffff2b\n\t" | |
| 10310 "popfq\n" | |
| 10311 "exit: nop\t# avoid branch to branch" %} | |
| 10312 opcode(0x0F, 0x2E); | |
| 10313 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2), | |
| 10314 cmpfp_fixup); | |
| 10315 ins_pipe(pipe_slow); | |
| 10316 %} | |
| 10317 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10318 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10319 match(Set cr (CmpF src1 (LoadF src2))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10320 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10321 ins_cost(100); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10322 format %{ "ucomiss $src1, $src2" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10323 opcode(0x0F, 0x2E); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10324 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10325 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10326 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10327 |
| 0 | 10328 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2) |
| 10329 %{ | |
| 10330 match(Set cr (CmpF src1 src2)); | |
| 10331 | |
| 10332 ins_cost(145); | |
| 10333 format %{ "ucomiss $src1, $src2\n\t" | |
| 10334 "jnp,s exit\n\t" | |
| 10335 "pushfq\t# saw NaN, set CF\n\t" | |
| 10336 "andq [rsp], #0xffffff2b\n\t" | |
| 10337 "popfq\n" | |
| 10338 "exit: nop\t# avoid branch to branch" %} | |
| 10339 opcode(0x0F, 0x2E); | |
| 10340 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2), | |
| 10341 cmpfp_fixup); | |
| 10342 ins_pipe(pipe_slow); | |
| 10343 %} | |
| 10344 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10345 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10346 match(Set cr (CmpF src1 src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10347 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10348 ins_cost(100); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10349 format %{ "ucomiss $src1, $src2" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10350 opcode(0x0F, 0x2E); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10351 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10352 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10353 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10354 |
| 0 | 10355 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2) |
| 10356 %{ | |
| 10357 match(Set cr (CmpD src1 src2)); | |
| 10358 | |
| 10359 ins_cost(145); | |
| 10360 format %{ "ucomisd $src1, $src2\n\t" | |
| 10361 "jnp,s exit\n\t" | |
| 10362 "pushfq\t# saw NaN, set CF\n\t" | |
| 10363 "andq [rsp], #0xffffff2b\n\t" | |
| 10364 "popfq\n" | |
| 10365 "exit: nop\t# avoid branch to branch" %} | |
| 10366 opcode(0x66, 0x0F, 0x2E); | |
| 10367 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2), | |
| 10368 cmpfp_fixup); | |
| 10369 ins_pipe(pipe_slow); | |
| 10370 %} | |
| 10371 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10372 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10373 match(Set cr (CmpD src1 src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10374 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10375 ins_cost(100); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10376 format %{ "ucomisd $src1, $src2 test" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10377 ins_encode %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10378 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10379 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10380 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10381 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10382 |
| 0 | 10383 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2) |
| 10384 %{ | |
| 10385 match(Set cr (CmpD src1 (LoadD src2))); | |
| 10386 | |
| 10387 ins_cost(145); | |
| 10388 format %{ "ucomisd $src1, $src2\n\t" | |
| 10389 "jnp,s exit\n\t" | |
| 10390 "pushfq\t# saw NaN, set CF\n\t" | |
| 10391 "andq [rsp], #0xffffff2b\n\t" | |
| 10392 "popfq\n" | |
| 10393 "exit: nop\t# avoid branch to branch" %} | |
| 10394 opcode(0x66, 0x0F, 0x2E); | |
| 10395 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2), | |
| 10396 cmpfp_fixup); | |
| 10397 ins_pipe(pipe_slow); | |
| 10398 %} | |
| 10399 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10400 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10401 match(Set cr (CmpD src1 (LoadD src2))); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10402 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10403 ins_cost(100); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10404 format %{ "ucomisd $src1, $src2" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10405 opcode(0x66, 0x0F, 0x2E); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10406 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10407 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10408 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10409 |
| 0 | 10410 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2) |
| 10411 %{ | |
| 10412 match(Set cr (CmpD src1 src2)); | |
| 10413 | |
| 10414 ins_cost(145); | |
| 10415 format %{ "ucomisd $src1, [$src2]\n\t" | |
| 10416 "jnp,s exit\n\t" | |
| 10417 "pushfq\t# saw NaN, set CF\n\t" | |
| 10418 "andq [rsp], #0xffffff2b\n\t" | |
| 10419 "popfq\n" | |
| 10420 "exit: nop\t# avoid branch to branch" %} | |
| 10421 opcode(0x66, 0x0F, 0x2E); | |
| 10422 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2), | |
| 10423 cmpfp_fixup); | |
| 10424 ins_pipe(pipe_slow); | |
| 10425 %} | |
| 10426 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10427 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10428 match(Set cr (CmpD src1 src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10429 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10430 ins_cost(100); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10431 format %{ "ucomisd $src1, [$src2]" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10432 opcode(0x66, 0x0F, 0x2E); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10433 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10434 ins_pipe(pipe_slow); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10435 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
10436 |
| 0 | 10437 // Compare into -1,0,1 |
| 10438 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr) | |
| 10439 %{ | |
| 10440 match(Set dst (CmpF3 src1 src2)); | |
| 10441 effect(KILL cr); | |
| 10442 | |
| 10443 ins_cost(275); | |
| 10444 format %{ "ucomiss $src1, $src2\n\t" | |
| 10445 "movl $dst, #-1\n\t" | |
| 10446 "jp,s done\n\t" | |
| 10447 "jb,s done\n\t" | |
| 10448 "setne $dst\n\t" | |
| 10449 "movzbl $dst, $dst\n" | |
| 10450 "done:" %} | |
| 10451 | |
| 10452 opcode(0x0F, 0x2E); | |
| 10453 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2), | |
| 10454 cmpfp3(dst)); | |
| 10455 ins_pipe(pipe_slow); | |
| 10456 %} | |
| 10457 | |
| 10458 // Compare into -1,0,1 | |
| 10459 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr) | |
| 10460 %{ | |
| 10461 match(Set dst (CmpF3 src1 (LoadF src2))); | |
| 10462 effect(KILL cr); | |
| 10463 | |
| 10464 ins_cost(275); | |
| 10465 format %{ "ucomiss $src1, $src2\n\t" | |
| 10466 "movl $dst, #-1\n\t" | |
| 10467 "jp,s done\n\t" | |
| 10468 "jb,s done\n\t" | |
| 10469 "setne $dst\n\t" | |
| 10470 "movzbl $dst, $dst\n" | |
| 10471 "done:" %} | |
| 10472 | |
| 10473 opcode(0x0F, 0x2E); | |
| 10474 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2), | |
| 10475 cmpfp3(dst)); | |
| 10476 ins_pipe(pipe_slow); | |
| 10477 %} | |
| 10478 | |
| 10479 // Compare into -1,0,1 | |
| 10480 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr) | |
| 10481 %{ | |
| 10482 match(Set dst (CmpF3 src1 src2)); | |
| 10483 effect(KILL cr); | |
| 10484 | |
| 10485 ins_cost(275); | |
| 10486 format %{ "ucomiss $src1, [$src2]\n\t" | |
| 10487 "movl $dst, #-1\n\t" | |
| 10488 "jp,s done\n\t" | |
| 10489 "jb,s done\n\t" | |
| 10490 "setne $dst\n\t" | |
| 10491 "movzbl $dst, $dst\n" | |
| 10492 "done:" %} | |
| 10493 | |
| 10494 opcode(0x0F, 0x2E); | |
| 10495 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2), | |
| 10496 cmpfp3(dst)); | |
| 10497 ins_pipe(pipe_slow); | |
| 10498 %} | |
| 10499 | |
| 10500 // Compare into -1,0,1 | |
| 10501 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr) | |
| 10502 %{ | |
| 10503 match(Set dst (CmpD3 src1 src2)); | |
| 10504 effect(KILL cr); | |
| 10505 | |
| 10506 ins_cost(275); | |
| 10507 format %{ "ucomisd $src1, $src2\n\t" | |
| 10508 "movl $dst, #-1\n\t" | |
| 10509 "jp,s done\n\t" | |
| 10510 "jb,s done\n\t" | |
| 10511 "setne $dst\n\t" | |
| 10512 "movzbl $dst, $dst\n" | |
| 10513 "done:" %} | |
| 10514 | |
| 10515 opcode(0x66, 0x0F, 0x2E); | |
| 10516 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2), | |
| 10517 cmpfp3(dst)); | |
| 10518 ins_pipe(pipe_slow); | |
| 10519 %} | |
| 10520 | |
| 10521 // Compare into -1,0,1 | |
| 10522 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr) | |
| 10523 %{ | |
| 10524 match(Set dst (CmpD3 src1 (LoadD src2))); | |
| 10525 effect(KILL cr); | |
| 10526 | |
| 10527 ins_cost(275); | |
| 10528 format %{ "ucomisd $src1, $src2\n\t" | |
| 10529 "movl $dst, #-1\n\t" | |
| 10530 "jp,s done\n\t" | |
| 10531 "jb,s done\n\t" | |
| 10532 "setne $dst\n\t" | |
| 10533 "movzbl $dst, $dst\n" | |
| 10534 "done:" %} | |
| 10535 | |
| 10536 opcode(0x66, 0x0F, 0x2E); | |
| 10537 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2), | |
| 10538 cmpfp3(dst)); | |
| 10539 ins_pipe(pipe_slow); | |
| 10540 %} | |
| 10541 | |
| 10542 // Compare into -1,0,1 | |
| 10543 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr) | |
| 10544 %{ | |
| 10545 match(Set dst (CmpD3 src1 src2)); | |
| 10546 effect(KILL cr); | |
| 10547 | |
| 10548 ins_cost(275); | |
| 10549 format %{ "ucomisd $src1, [$src2]\n\t" | |
| 10550 "movl $dst, #-1\n\t" | |
| 10551 "jp,s done\n\t" | |
| 10552 "jb,s done\n\t" | |
| 10553 "setne $dst\n\t" | |
| 10554 "movzbl $dst, $dst\n" | |
| 10555 "done:" %} | |
| 10556 | |
| 10557 opcode(0x66, 0x0F, 0x2E); | |
| 10558 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2), | |
| 10559 cmpfp3(dst)); | |
| 10560 ins_pipe(pipe_slow); | |
| 10561 %} | |
| 10562 | |
| 10563 instruct addF_reg(regF dst, regF src) | |
| 10564 %{ | |
| 10565 match(Set dst (AddF dst src)); | |
| 10566 | |
| 10567 format %{ "addss $dst, $src" %} | |
| 10568 ins_cost(150); // XXX | |
| 10569 opcode(0xF3, 0x0F, 0x58); | |
| 10570 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10571 ins_pipe(pipe_slow); | |
| 10572 %} | |
| 10573 | |
| 10574 instruct addF_mem(regF dst, memory src) | |
| 10575 %{ | |
| 10576 match(Set dst (AddF dst (LoadF src))); | |
| 10577 | |
| 10578 format %{ "addss $dst, $src" %} | |
| 10579 ins_cost(150); // XXX | |
| 10580 opcode(0xF3, 0x0F, 0x58); | |
| 10581 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10582 ins_pipe(pipe_slow); | |
| 10583 %} | |
| 10584 | |
| 10585 instruct addF_imm(regF dst, immF src) | |
| 10586 %{ | |
| 10587 match(Set dst (AddF dst src)); | |
| 10588 | |
| 10589 format %{ "addss $dst, [$src]" %} | |
| 10590 ins_cost(150); // XXX | |
| 10591 opcode(0xF3, 0x0F, 0x58); | |
| 10592 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src)); | |
| 10593 ins_pipe(pipe_slow); | |
| 10594 %} | |
| 10595 | |
| 10596 instruct addD_reg(regD dst, regD src) | |
| 10597 %{ | |
| 10598 match(Set dst (AddD dst src)); | |
| 10599 | |
| 10600 format %{ "addsd $dst, $src" %} | |
| 10601 ins_cost(150); // XXX | |
| 10602 opcode(0xF2, 0x0F, 0x58); | |
| 10603 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10604 ins_pipe(pipe_slow); | |
| 10605 %} | |
| 10606 | |
| 10607 instruct addD_mem(regD dst, memory src) | |
| 10608 %{ | |
| 10609 match(Set dst (AddD dst (LoadD src))); | |
| 10610 | |
| 10611 format %{ "addsd $dst, $src" %} | |
| 10612 ins_cost(150); // XXX | |
| 10613 opcode(0xF2, 0x0F, 0x58); | |
| 10614 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10615 ins_pipe(pipe_slow); | |
| 10616 %} | |
| 10617 | |
| 10618 instruct addD_imm(regD dst, immD src) | |
| 10619 %{ | |
| 10620 match(Set dst (AddD dst src)); | |
| 10621 | |
| 10622 format %{ "addsd $dst, [$src]" %} | |
| 10623 ins_cost(150); // XXX | |
| 10624 opcode(0xF2, 0x0F, 0x58); | |
| 10625 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src)); | |
| 10626 ins_pipe(pipe_slow); | |
| 10627 %} | |
| 10628 | |
| 10629 instruct subF_reg(regF dst, regF src) | |
| 10630 %{ | |
| 10631 match(Set dst (SubF dst src)); | |
| 10632 | |
| 10633 format %{ "subss $dst, $src" %} | |
| 10634 ins_cost(150); // XXX | |
| 10635 opcode(0xF3, 0x0F, 0x5C); | |
| 10636 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10637 ins_pipe(pipe_slow); | |
| 10638 %} | |
| 10639 | |
| 10640 instruct subF_mem(regF dst, memory src) | |
| 10641 %{ | |
| 10642 match(Set dst (SubF dst (LoadF src))); | |
| 10643 | |
| 10644 format %{ "subss $dst, $src" %} | |
| 10645 ins_cost(150); // XXX | |
| 10646 opcode(0xF3, 0x0F, 0x5C); | |
| 10647 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10648 ins_pipe(pipe_slow); | |
| 10649 %} | |
| 10650 | |
| 10651 instruct subF_imm(regF dst, immF src) | |
| 10652 %{ | |
| 10653 match(Set dst (SubF dst src)); | |
| 10654 | |
| 10655 format %{ "subss $dst, [$src]" %} | |
| 10656 ins_cost(150); // XXX | |
| 10657 opcode(0xF3, 0x0F, 0x5C); | |
| 10658 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src)); | |
| 10659 ins_pipe(pipe_slow); | |
| 10660 %} | |
| 10661 | |
| 10662 instruct subD_reg(regD dst, regD src) | |
| 10663 %{ | |
| 10664 match(Set dst (SubD dst src)); | |
| 10665 | |
| 10666 format %{ "subsd $dst, $src" %} | |
| 10667 ins_cost(150); // XXX | |
| 10668 opcode(0xF2, 0x0F, 0x5C); | |
| 10669 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10670 ins_pipe(pipe_slow); | |
| 10671 %} | |
| 10672 | |
| 10673 instruct subD_mem(regD dst, memory src) | |
| 10674 %{ | |
| 10675 match(Set dst (SubD dst (LoadD src))); | |
| 10676 | |
| 10677 format %{ "subsd $dst, $src" %} | |
| 10678 ins_cost(150); // XXX | |
| 10679 opcode(0xF2, 0x0F, 0x5C); | |
| 10680 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10681 ins_pipe(pipe_slow); | |
| 10682 %} | |
| 10683 | |
| 10684 instruct subD_imm(regD dst, immD src) | |
| 10685 %{ | |
| 10686 match(Set dst (SubD dst src)); | |
| 10687 | |
| 10688 format %{ "subsd $dst, [$src]" %} | |
| 10689 ins_cost(150); // XXX | |
| 10690 opcode(0xF2, 0x0F, 0x5C); | |
| 10691 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src)); | |
| 10692 ins_pipe(pipe_slow); | |
| 10693 %} | |
| 10694 | |
| 10695 instruct mulF_reg(regF dst, regF src) | |
| 10696 %{ | |
| 10697 match(Set dst (MulF dst src)); | |
| 10698 | |
| 10699 format %{ "mulss $dst, $src" %} | |
| 10700 ins_cost(150); // XXX | |
| 10701 opcode(0xF3, 0x0F, 0x59); | |
| 10702 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10703 ins_pipe(pipe_slow); | |
| 10704 %} | |
| 10705 | |
| 10706 instruct mulF_mem(regF dst, memory src) | |
| 10707 %{ | |
| 10708 match(Set dst (MulF dst (LoadF src))); | |
| 10709 | |
| 10710 format %{ "mulss $dst, $src" %} | |
| 10711 ins_cost(150); // XXX | |
| 10712 opcode(0xF3, 0x0F, 0x59); | |
| 10713 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10714 ins_pipe(pipe_slow); | |
| 10715 %} | |
| 10716 | |
| 10717 instruct mulF_imm(regF dst, immF src) | |
| 10718 %{ | |
| 10719 match(Set dst (MulF dst src)); | |
| 10720 | |
| 10721 format %{ "mulss $dst, [$src]" %} | |
| 10722 ins_cost(150); // XXX | |
| 10723 opcode(0xF3, 0x0F, 0x59); | |
| 10724 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src)); | |
| 10725 ins_pipe(pipe_slow); | |
| 10726 %} | |
| 10727 | |
| 10728 instruct mulD_reg(regD dst, regD src) | |
| 10729 %{ | |
| 10730 match(Set dst (MulD dst src)); | |
| 10731 | |
| 10732 format %{ "mulsd $dst, $src" %} | |
| 10733 ins_cost(150); // XXX | |
| 10734 opcode(0xF2, 0x0F, 0x59); | |
| 10735 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10736 ins_pipe(pipe_slow); | |
| 10737 %} | |
| 10738 | |
| 10739 instruct mulD_mem(regD dst, memory src) | |
| 10740 %{ | |
| 10741 match(Set dst (MulD dst (LoadD src))); | |
| 10742 | |
| 10743 format %{ "mulsd $dst, $src" %} | |
| 10744 ins_cost(150); // XXX | |
| 10745 opcode(0xF2, 0x0F, 0x59); | |
| 10746 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10747 ins_pipe(pipe_slow); | |
| 10748 %} | |
| 10749 | |
| 10750 instruct mulD_imm(regD dst, immD src) | |
| 10751 %{ | |
| 10752 match(Set dst (MulD dst src)); | |
| 10753 | |
| 10754 format %{ "mulsd $dst, [$src]" %} | |
| 10755 ins_cost(150); // XXX | |
| 10756 opcode(0xF2, 0x0F, 0x59); | |
| 10757 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src)); | |
| 10758 ins_pipe(pipe_slow); | |
| 10759 %} | |
| 10760 | |
| 10761 instruct divF_reg(regF dst, regF src) | |
| 10762 %{ | |
| 10763 match(Set dst (DivF dst src)); | |
| 10764 | |
| 10765 format %{ "divss $dst, $src" %} | |
| 10766 ins_cost(150); // XXX | |
| 10767 opcode(0xF3, 0x0F, 0x5E); | |
| 10768 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10769 ins_pipe(pipe_slow); | |
| 10770 %} | |
| 10771 | |
| 10772 instruct divF_mem(regF dst, memory src) | |
| 10773 %{ | |
| 10774 match(Set dst (DivF dst (LoadF src))); | |
| 10775 | |
| 10776 format %{ "divss $dst, $src" %} | |
| 10777 ins_cost(150); // XXX | |
| 10778 opcode(0xF3, 0x0F, 0x5E); | |
| 10779 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10780 ins_pipe(pipe_slow); | |
| 10781 %} | |
| 10782 | |
| 10783 instruct divF_imm(regF dst, immF src) | |
| 10784 %{ | |
| 10785 match(Set dst (DivF dst src)); | |
| 10786 | |
| 10787 format %{ "divss $dst, [$src]" %} | |
| 10788 ins_cost(150); // XXX | |
| 10789 opcode(0xF3, 0x0F, 0x5E); | |
| 10790 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src)); | |
| 10791 ins_pipe(pipe_slow); | |
| 10792 %} | |
| 10793 | |
| 10794 instruct divD_reg(regD dst, regD src) | |
| 10795 %{ | |
| 10796 match(Set dst (DivD dst src)); | |
| 10797 | |
| 10798 format %{ "divsd $dst, $src" %} | |
| 10799 ins_cost(150); // XXX | |
| 10800 opcode(0xF2, 0x0F, 0x5E); | |
| 10801 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10802 ins_pipe(pipe_slow); | |
| 10803 %} | |
| 10804 | |
| 10805 instruct divD_mem(regD dst, memory src) | |
| 10806 %{ | |
| 10807 match(Set dst (DivD dst (LoadD src))); | |
| 10808 | |
| 10809 format %{ "divsd $dst, $src" %} | |
| 10810 ins_cost(150); // XXX | |
| 10811 opcode(0xF2, 0x0F, 0x5E); | |
| 10812 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10813 ins_pipe(pipe_slow); | |
| 10814 %} | |
| 10815 | |
| 10816 instruct divD_imm(regD dst, immD src) | |
| 10817 %{ | |
| 10818 match(Set dst (DivD dst src)); | |
| 10819 | |
| 10820 format %{ "divsd $dst, [$src]" %} | |
| 10821 ins_cost(150); // XXX | |
| 10822 opcode(0xF2, 0x0F, 0x5E); | |
| 10823 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src)); | |
| 10824 ins_pipe(pipe_slow); | |
| 10825 %} | |
| 10826 | |
| 10827 instruct sqrtF_reg(regF dst, regF src) | |
| 10828 %{ | |
| 10829 match(Set dst (ConvD2F (SqrtD (ConvF2D src)))); | |
| 10830 | |
| 10831 format %{ "sqrtss $dst, $src" %} | |
| 10832 ins_cost(150); // XXX | |
| 10833 opcode(0xF3, 0x0F, 0x51); | |
| 10834 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10835 ins_pipe(pipe_slow); | |
| 10836 %} | |
| 10837 | |
| 10838 instruct sqrtF_mem(regF dst, memory src) | |
| 10839 %{ | |
| 10840 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src))))); | |
| 10841 | |
| 10842 format %{ "sqrtss $dst, $src" %} | |
| 10843 ins_cost(150); // XXX | |
| 10844 opcode(0xF3, 0x0F, 0x51); | |
| 10845 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10846 ins_pipe(pipe_slow); | |
| 10847 %} | |
| 10848 | |
| 10849 instruct sqrtF_imm(regF dst, immF src) | |
| 10850 %{ | |
| 10851 match(Set dst (ConvD2F (SqrtD (ConvF2D src)))); | |
| 10852 | |
| 10853 format %{ "sqrtss $dst, [$src]" %} | |
| 10854 ins_cost(150); // XXX | |
| 10855 opcode(0xF3, 0x0F, 0x51); | |
| 10856 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src)); | |
| 10857 ins_pipe(pipe_slow); | |
| 10858 %} | |
| 10859 | |
| 10860 instruct sqrtD_reg(regD dst, regD src) | |
| 10861 %{ | |
| 10862 match(Set dst (SqrtD src)); | |
| 10863 | |
| 10864 format %{ "sqrtsd $dst, $src" %} | |
| 10865 ins_cost(150); // XXX | |
| 10866 opcode(0xF2, 0x0F, 0x51); | |
| 10867 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 10868 ins_pipe(pipe_slow); | |
| 10869 %} | |
| 10870 | |
| 10871 instruct sqrtD_mem(regD dst, memory src) | |
| 10872 %{ | |
| 10873 match(Set dst (SqrtD (LoadD src))); | |
| 10874 | |
| 10875 format %{ "sqrtsd $dst, $src" %} | |
| 10876 ins_cost(150); // XXX | |
| 10877 opcode(0xF2, 0x0F, 0x51); | |
| 10878 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 10879 ins_pipe(pipe_slow); | |
| 10880 %} | |
| 10881 | |
| 10882 instruct sqrtD_imm(regD dst, immD src) | |
| 10883 %{ | |
| 10884 match(Set dst (SqrtD src)); | |
| 10885 | |
| 10886 format %{ "sqrtsd $dst, [$src]" %} | |
| 10887 ins_cost(150); // XXX | |
| 10888 opcode(0xF2, 0x0F, 0x51); | |
| 10889 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src)); | |
| 10890 ins_pipe(pipe_slow); | |
| 10891 %} | |
| 10892 | |
| 10893 instruct absF_reg(regF dst) | |
| 10894 %{ | |
| 10895 match(Set dst (AbsF dst)); | |
| 10896 | |
| 10897 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %} | |
| 10898 ins_encode(absF_encoding(dst)); | |
| 10899 ins_pipe(pipe_slow); | |
| 10900 %} | |
| 10901 | |
| 10902 instruct absD_reg(regD dst) | |
| 10903 %{ | |
| 10904 match(Set dst (AbsD dst)); | |
| 10905 | |
| 10906 format %{ "andpd $dst, [0x7fffffffffffffff]\t" | |
| 10907 "# abs double by sign masking" %} | |
| 10908 ins_encode(absD_encoding(dst)); | |
| 10909 ins_pipe(pipe_slow); | |
| 10910 %} | |
| 10911 | |
| 10912 instruct negF_reg(regF dst) | |
| 10913 %{ | |
| 10914 match(Set dst (NegF dst)); | |
| 10915 | |
| 10916 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %} | |
| 10917 ins_encode(negF_encoding(dst)); | |
| 10918 ins_pipe(pipe_slow); | |
| 10919 %} | |
| 10920 | |
| 10921 instruct negD_reg(regD dst) | |
| 10922 %{ | |
| 10923 match(Set dst (NegD dst)); | |
| 10924 | |
| 10925 format %{ "xorpd $dst, [0x8000000000000000]\t" | |
| 10926 "# neg double by sign flipping" %} | |
| 10927 ins_encode(negD_encoding(dst)); | |
| 10928 ins_pipe(pipe_slow); | |
| 10929 %} | |
| 10930 | |
| 10931 // -----------Trig and Trancendental Instructions------------------------------ | |
| 10932 instruct cosD_reg(regD dst) %{ | |
| 10933 match(Set dst (CosD dst)); | |
| 10934 | |
| 10935 format %{ "dcos $dst\n\t" %} | |
| 10936 opcode(0xD9, 0xFF); | |
| 10937 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) ); | |
| 10938 ins_pipe( pipe_slow ); | |
| 10939 %} | |
| 10940 | |
| 10941 instruct sinD_reg(regD dst) %{ | |
| 10942 match(Set dst (SinD dst)); | |
| 10943 | |
| 10944 format %{ "dsin $dst\n\t" %} | |
| 10945 opcode(0xD9, 0xFE); | |
| 10946 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) ); | |
| 10947 ins_pipe( pipe_slow ); | |
| 10948 %} | |
| 10949 | |
| 10950 instruct tanD_reg(regD dst) %{ | |
| 10951 match(Set dst (TanD dst)); | |
| 10952 | |
| 10953 format %{ "dtan $dst\n\t" %} | |
| 10954 ins_encode( Push_SrcXD(dst), | |
| 10955 Opcode(0xD9), Opcode(0xF2), //fptan | |
| 10956 Opcode(0xDD), Opcode(0xD8), //fstp st | |
| 10957 Push_ResultXD(dst) ); | |
| 10958 ins_pipe( pipe_slow ); | |
| 10959 %} | |
| 10960 | |
| 10961 instruct log10D_reg(regD dst) %{ | |
| 10962 // The source and result Double operands in XMM registers | |
| 10963 match(Set dst (Log10D dst)); | |
| 10964 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number | |
| 10965 // fyl2x ; compute log_10(2) * log_2(x) | |
| 10966 format %{ "fldlg2\t\t\t#Log10\n\t" | |
| 10967 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t" | |
| 10968 %} | |
| 10969 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2 | |
| 10970 Push_SrcXD(dst), | |
| 10971 Opcode(0xD9), Opcode(0xF1), // fyl2x | |
| 10972 Push_ResultXD(dst)); | |
| 10973 | |
| 10974 ins_pipe( pipe_slow ); | |
| 10975 %} | |
| 10976 | |
| 10977 instruct logD_reg(regD dst) %{ | |
| 10978 // The source and result Double operands in XMM registers | |
| 10979 match(Set dst (LogD dst)); | |
| 10980 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number | |
| 10981 // fyl2x ; compute log_e(2) * log_2(x) | |
| 10982 format %{ "fldln2\t\t\t#Log_e\n\t" | |
| 10983 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t" | |
| 10984 %} | |
| 10985 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2 | |
| 10986 Push_SrcXD(dst), | |
| 10987 Opcode(0xD9), Opcode(0xF1), // fyl2x | |
| 10988 Push_ResultXD(dst)); | |
| 10989 ins_pipe( pipe_slow ); | |
| 10990 %} | |
| 10991 | |
| 10992 | |
| 10993 | |
| 10994 //----------Arithmetic Conversion Instructions--------------------------------- | |
| 10995 | |
| 10996 instruct roundFloat_nop(regF dst) | |
| 10997 %{ | |
| 10998 match(Set dst (RoundFloat dst)); | |
| 10999 | |
| 11000 ins_cost(0); | |
| 11001 ins_encode(); | |
| 11002 ins_pipe(empty); | |
| 11003 %} | |
| 11004 | |
| 11005 instruct roundDouble_nop(regD dst) | |
| 11006 %{ | |
| 11007 match(Set dst (RoundDouble dst)); | |
| 11008 | |
| 11009 ins_cost(0); | |
| 11010 ins_encode(); | |
| 11011 ins_pipe(empty); | |
| 11012 %} | |
| 11013 | |
| 11014 instruct convF2D_reg_reg(regD dst, regF src) | |
| 11015 %{ | |
| 11016 match(Set dst (ConvF2D src)); | |
| 11017 | |
| 11018 format %{ "cvtss2sd $dst, $src" %} | |
| 11019 opcode(0xF3, 0x0F, 0x5A); | |
| 11020 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11021 ins_pipe(pipe_slow); // XXX | |
| 11022 %} | |
| 11023 | |
| 11024 instruct convF2D_reg_mem(regD dst, memory src) | |
| 11025 %{ | |
| 11026 match(Set dst (ConvF2D (LoadF src))); | |
| 11027 | |
| 11028 format %{ "cvtss2sd $dst, $src" %} | |
| 11029 opcode(0xF3, 0x0F, 0x5A); | |
| 11030 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11031 ins_pipe(pipe_slow); // XXX | |
| 11032 %} | |
| 11033 | |
| 11034 instruct convD2F_reg_reg(regF dst, regD src) | |
| 11035 %{ | |
| 11036 match(Set dst (ConvD2F src)); | |
| 11037 | |
| 11038 format %{ "cvtsd2ss $dst, $src" %} | |
| 11039 opcode(0xF2, 0x0F, 0x5A); | |
| 11040 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11041 ins_pipe(pipe_slow); // XXX | |
| 11042 %} | |
| 11043 | |
| 11044 instruct convD2F_reg_mem(regF dst, memory src) | |
| 11045 %{ | |
| 11046 match(Set dst (ConvD2F (LoadD src))); | |
| 11047 | |
| 11048 format %{ "cvtsd2ss $dst, $src" %} | |
| 11049 opcode(0xF2, 0x0F, 0x5A); | |
| 11050 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11051 ins_pipe(pipe_slow); // XXX | |
| 11052 %} | |
| 11053 | |
| 11054 // XXX do mem variants | |
| 11055 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr) | |
| 11056 %{ | |
| 11057 match(Set dst (ConvF2I src)); | |
| 11058 effect(KILL cr); | |
| 11059 | |
| 11060 format %{ "cvttss2sil $dst, $src\t# f2i\n\t" | |
| 11061 "cmpl $dst, #0x80000000\n\t" | |
| 11062 "jne,s done\n\t" | |
| 11063 "subq rsp, #8\n\t" | |
| 11064 "movss [rsp], $src\n\t" | |
| 11065 "call f2i_fixup\n\t" | |
| 11066 "popq $dst\n" | |
| 11067 "done: "%} | |
| 11068 opcode(0xF3, 0x0F, 0x2C); | |
| 11069 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src), | |
| 11070 f2i_fixup(dst, src)); | |
| 11071 ins_pipe(pipe_slow); | |
| 11072 %} | |
| 11073 | |
| 11074 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr) | |
| 11075 %{ | |
| 11076 match(Set dst (ConvF2L src)); | |
| 11077 effect(KILL cr); | |
| 11078 | |
| 11079 format %{ "cvttss2siq $dst, $src\t# f2l\n\t" | |
| 11080 "cmpq $dst, [0x8000000000000000]\n\t" | |
| 11081 "jne,s done\n\t" | |
| 11082 "subq rsp, #8\n\t" | |
| 11083 "movss [rsp], $src\n\t" | |
| 11084 "call f2l_fixup\n\t" | |
| 11085 "popq $dst\n" | |
| 11086 "done: "%} | |
| 11087 opcode(0xF3, 0x0F, 0x2C); | |
| 11088 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src), | |
| 11089 f2l_fixup(dst, src)); | |
| 11090 ins_pipe(pipe_slow); | |
| 11091 %} | |
| 11092 | |
| 11093 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr) | |
| 11094 %{ | |
| 11095 match(Set dst (ConvD2I src)); | |
| 11096 effect(KILL cr); | |
| 11097 | |
| 11098 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t" | |
| 11099 "cmpl $dst, #0x80000000\n\t" | |
| 11100 "jne,s done\n\t" | |
| 11101 "subq rsp, #8\n\t" | |
| 11102 "movsd [rsp], $src\n\t" | |
| 11103 "call d2i_fixup\n\t" | |
| 11104 "popq $dst\n" | |
| 11105 "done: "%} | |
| 11106 opcode(0xF2, 0x0F, 0x2C); | |
| 11107 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src), | |
| 11108 d2i_fixup(dst, src)); | |
| 11109 ins_pipe(pipe_slow); | |
| 11110 %} | |
| 11111 | |
| 11112 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr) | |
| 11113 %{ | |
| 11114 match(Set dst (ConvD2L src)); | |
| 11115 effect(KILL cr); | |
| 11116 | |
| 11117 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t" | |
| 11118 "cmpq $dst, [0x8000000000000000]\n\t" | |
| 11119 "jne,s done\n\t" | |
| 11120 "subq rsp, #8\n\t" | |
| 11121 "movsd [rsp], $src\n\t" | |
| 11122 "call d2l_fixup\n\t" | |
| 11123 "popq $dst\n" | |
| 11124 "done: "%} | |
| 11125 opcode(0xF2, 0x0F, 0x2C); | |
| 11126 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src), | |
| 11127 d2l_fixup(dst, src)); | |
| 11128 ins_pipe(pipe_slow); | |
| 11129 %} | |
| 11130 | |
| 11131 instruct convI2F_reg_reg(regF dst, rRegI src) | |
| 11132 %{ | |
| 71 | 11133 predicate(!UseXmmI2F); |
| 0 | 11134 match(Set dst (ConvI2F src)); |
| 11135 | |
| 11136 format %{ "cvtsi2ssl $dst, $src\t# i2f" %} | |
| 11137 opcode(0xF3, 0x0F, 0x2A); | |
| 11138 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11139 ins_pipe(pipe_slow); // XXX | |
| 11140 %} | |
| 11141 | |
| 11142 instruct convI2F_reg_mem(regF dst, memory src) | |
| 11143 %{ | |
| 11144 match(Set dst (ConvI2F (LoadI src))); | |
| 11145 | |
| 11146 format %{ "cvtsi2ssl $dst, $src\t# i2f" %} | |
| 11147 opcode(0xF3, 0x0F, 0x2A); | |
| 11148 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11149 ins_pipe(pipe_slow); // XXX | |
| 11150 %} | |
| 11151 | |
| 11152 instruct convI2D_reg_reg(regD dst, rRegI src) | |
| 11153 %{ | |
| 71 | 11154 predicate(!UseXmmI2D); |
| 0 | 11155 match(Set dst (ConvI2D src)); |
| 11156 | |
| 11157 format %{ "cvtsi2sdl $dst, $src\t# i2d" %} | |
| 11158 opcode(0xF2, 0x0F, 0x2A); | |
| 11159 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11160 ins_pipe(pipe_slow); // XXX | |
| 11161 %} | |
| 11162 | |
| 11163 instruct convI2D_reg_mem(regD dst, memory src) | |
| 11164 %{ | |
| 11165 match(Set dst (ConvI2D (LoadI src))); | |
| 11166 | |
| 11167 format %{ "cvtsi2sdl $dst, $src\t# i2d" %} | |
| 11168 opcode(0xF2, 0x0F, 0x2A); | |
| 11169 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11170 ins_pipe(pipe_slow); // XXX | |
| 11171 %} | |
| 11172 | |
| 71 | 11173 instruct convXI2F_reg(regF dst, rRegI src) |
| 11174 %{ | |
| 11175 predicate(UseXmmI2F); | |
| 11176 match(Set dst (ConvI2F src)); | |
| 11177 | |
| 11178 format %{ "movdl $dst, $src\n\t" | |
| 11179 "cvtdq2psl $dst, $dst\t# i2f" %} | |
| 11180 ins_encode %{ | |
| 11181 __ movdl($dst$$XMMRegister, $src$$Register); | |
| 11182 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister); | |
| 11183 %} | |
| 11184 ins_pipe(pipe_slow); // XXX | |
| 11185 %} | |
| 11186 | |
| 11187 instruct convXI2D_reg(regD dst, rRegI src) | |
| 11188 %{ | |
| 11189 predicate(UseXmmI2D); | |
| 11190 match(Set dst (ConvI2D src)); | |
| 11191 | |
| 11192 format %{ "movdl $dst, $src\n\t" | |
| 11193 "cvtdq2pdl $dst, $dst\t# i2d" %} | |
| 11194 ins_encode %{ | |
| 11195 __ movdl($dst$$XMMRegister, $src$$Register); | |
| 11196 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister); | |
| 11197 %} | |
| 11198 ins_pipe(pipe_slow); // XXX | |
| 11199 %} | |
| 11200 | |
| 0 | 11201 instruct convL2F_reg_reg(regF dst, rRegL src) |
| 11202 %{ | |
| 11203 match(Set dst (ConvL2F src)); | |
| 11204 | |
| 11205 format %{ "cvtsi2ssq $dst, $src\t# l2f" %} | |
| 11206 opcode(0xF3, 0x0F, 0x2A); | |
| 11207 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11208 ins_pipe(pipe_slow); // XXX | |
| 11209 %} | |
| 11210 | |
| 11211 instruct convL2F_reg_mem(regF dst, memory src) | |
| 11212 %{ | |
| 11213 match(Set dst (ConvL2F (LoadL src))); | |
| 11214 | |
| 11215 format %{ "cvtsi2ssq $dst, $src\t# l2f" %} | |
| 11216 opcode(0xF3, 0x0F, 0x2A); | |
| 11217 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11218 ins_pipe(pipe_slow); // XXX | |
| 11219 %} | |
| 11220 | |
| 11221 instruct convL2D_reg_reg(regD dst, rRegL src) | |
| 11222 %{ | |
| 11223 match(Set dst (ConvL2D src)); | |
| 11224 | |
| 11225 format %{ "cvtsi2sdq $dst, $src\t# l2d" %} | |
| 11226 opcode(0xF2, 0x0F, 0x2A); | |
| 11227 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src)); | |
| 11228 ins_pipe(pipe_slow); // XXX | |
| 11229 %} | |
| 11230 | |
| 11231 instruct convL2D_reg_mem(regD dst, memory src) | |
| 11232 %{ | |
| 11233 match(Set dst (ConvL2D (LoadL src))); | |
| 11234 | |
| 11235 format %{ "cvtsi2sdq $dst, $src\t# l2d" %} | |
| 11236 opcode(0xF2, 0x0F, 0x2A); | |
| 11237 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11238 ins_pipe(pipe_slow); // XXX | |
| 11239 %} | |
| 11240 | |
| 11241 instruct convI2L_reg_reg(rRegL dst, rRegI src) | |
| 11242 %{ | |
| 11243 match(Set dst (ConvI2L src)); | |
| 11244 | |
| 11245 ins_cost(125); | |
| 11246 format %{ "movslq $dst, $src\t# i2l" %} | |
| 11247 opcode(0x63); // needs REX.W | |
| 11248 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src)); | |
| 11249 ins_pipe(ialu_reg_reg); | |
| 11250 %} | |
| 11251 | |
| 11252 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src) | |
| 11253 // %{ | |
| 11254 // match(Set dst (ConvI2L src)); | |
| 11255 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 && | |
| 11256 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0); | |
| 11257 // predicate(((const TypeNode*) n)->type()->is_long()->_hi == | |
| 11258 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi && | |
| 11259 // ((const TypeNode*) n)->type()->is_long()->_lo == | |
| 11260 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo); | |
| 11261 | |
| 11262 // format %{ "movl $dst, $src\t# unsigned i2l" %} | |
| 11263 // ins_encode(enc_copy(dst, src)); | |
| 11264 // // opcode(0x63); // needs REX.W | |
| 11265 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src)); | |
| 11266 // ins_pipe(ialu_reg_reg); | |
| 11267 // %} | |
| 11268 | |
| 11269 // Zero-extend convert int to long | |
| 11270 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask) | |
| 11271 %{ | |
| 11272 match(Set dst (AndL (ConvI2L src) mask)); | |
| 11273 | |
| 11274 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %} | |
| 11275 ins_encode(enc_copy(dst, src)); | |
| 11276 ins_pipe(ialu_reg_reg); | |
| 11277 %} | |
| 11278 | |
| 11279 // Zero-extend convert int to long | |
| 11280 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask) | |
| 11281 %{ | |
| 11282 match(Set dst (AndL (ConvI2L (LoadI src)) mask)); | |
| 11283 | |
| 11284 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %} | |
| 11285 opcode(0x8B); | |
| 11286 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 11287 ins_pipe(ialu_reg_mem); | |
| 11288 %} | |
| 11289 | |
| 11290 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask) | |
| 11291 %{ | |
| 11292 match(Set dst (AndL src mask)); | |
| 11293 | |
| 11294 format %{ "movl $dst, $src\t# zero-extend long" %} | |
| 11295 ins_encode(enc_copy_always(dst, src)); | |
| 11296 ins_pipe(ialu_reg_reg); | |
| 11297 %} | |
| 11298 | |
| 11299 instruct convL2I_reg_reg(rRegI dst, rRegL src) | |
| 11300 %{ | |
| 11301 match(Set dst (ConvL2I src)); | |
| 11302 | |
| 11303 format %{ "movl $dst, $src\t# l2i" %} | |
| 11304 ins_encode(enc_copy_always(dst, src)); | |
| 11305 ins_pipe(ialu_reg_reg); | |
| 11306 %} | |
| 11307 | |
| 11308 | |
| 11309 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{ | |
| 11310 match(Set dst (MoveF2I src)); | |
| 11311 effect(DEF dst, USE src); | |
| 11312 | |
| 11313 ins_cost(125); | |
| 11314 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %} | |
| 11315 opcode(0x8B); | |
| 11316 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src)); | |
| 11317 ins_pipe(ialu_reg_mem); | |
| 11318 %} | |
| 11319 | |
| 11320 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{ | |
| 11321 match(Set dst (MoveI2F src)); | |
| 11322 effect(DEF dst, USE src); | |
| 11323 | |
| 11324 ins_cost(125); | |
| 11325 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %} | |
| 11326 opcode(0xF3, 0x0F, 0x10); | |
| 11327 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11328 ins_pipe(pipe_slow); | |
| 11329 %} | |
| 11330 | |
| 11331 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{ | |
| 11332 match(Set dst (MoveD2L src)); | |
| 11333 effect(DEF dst, USE src); | |
| 11334 | |
| 11335 ins_cost(125); | |
| 11336 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %} | |
| 11337 opcode(0x8B); | |
| 11338 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src)); | |
| 11339 ins_pipe(ialu_reg_mem); | |
| 11340 %} | |
| 11341 | |
| 11342 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{ | |
| 11343 predicate(!UseXmmLoadAndClearUpper); | |
| 11344 match(Set dst (MoveL2D src)); | |
| 11345 effect(DEF dst, USE src); | |
| 11346 | |
| 11347 ins_cost(125); | |
| 11348 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %} | |
| 11349 opcode(0x66, 0x0F, 0x12); | |
| 11350 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11351 ins_pipe(pipe_slow); | |
| 11352 %} | |
| 11353 | |
| 11354 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{ | |
| 11355 predicate(UseXmmLoadAndClearUpper); | |
| 11356 match(Set dst (MoveL2D src)); | |
| 11357 effect(DEF dst, USE src); | |
| 11358 | |
| 11359 ins_cost(125); | |
| 11360 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %} | |
| 11361 opcode(0xF2, 0x0F, 0x10); | |
| 11362 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src)); | |
| 11363 ins_pipe(pipe_slow); | |
| 11364 %} | |
| 11365 | |
| 11366 | |
| 11367 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{ | |
| 11368 match(Set dst (MoveF2I src)); | |
| 11369 effect(DEF dst, USE src); | |
| 11370 | |
| 11371 ins_cost(95); // XXX | |
| 11372 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %} | |
| 11373 opcode(0xF3, 0x0F, 0x11); | |
| 11374 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst)); | |
| 11375 ins_pipe(pipe_slow); | |
| 11376 %} | |
| 11377 | |
| 11378 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{ | |
| 11379 match(Set dst (MoveI2F src)); | |
| 11380 effect(DEF dst, USE src); | |
| 11381 | |
| 11382 ins_cost(100); | |
| 11383 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %} | |
| 11384 opcode(0x89); | |
| 11385 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst)); | |
| 11386 ins_pipe( ialu_mem_reg ); | |
| 11387 %} | |
| 11388 | |
| 11389 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{ | |
| 11390 match(Set dst (MoveD2L src)); | |
| 11391 effect(DEF dst, USE src); | |
| 11392 | |
| 11393 ins_cost(95); // XXX | |
| 11394 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %} | |
| 11395 opcode(0xF2, 0x0F, 0x11); | |
| 11396 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst)); | |
| 11397 ins_pipe(pipe_slow); | |
| 11398 %} | |
| 11399 | |
| 11400 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{ | |
| 11401 match(Set dst (MoveL2D src)); | |
| 11402 effect(DEF dst, USE src); | |
| 11403 | |
| 11404 ins_cost(100); | |
| 11405 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %} | |
| 11406 opcode(0x89); | |
| 11407 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst)); | |
| 11408 ins_pipe(ialu_mem_reg); | |
| 11409 %} | |
| 11410 | |
| 11411 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{ | |
| 11412 match(Set dst (MoveF2I src)); | |
| 11413 effect(DEF dst, USE src); | |
| 11414 ins_cost(85); | |
| 11415 format %{ "movd $dst,$src\t# MoveF2I" %} | |
| 11416 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %} | |
| 11417 ins_pipe( pipe_slow ); | |
| 11418 %} | |
| 11419 | |
| 11420 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{ | |
| 11421 match(Set dst (MoveD2L src)); | |
| 11422 effect(DEF dst, USE src); | |
| 11423 ins_cost(85); | |
| 11424 format %{ "movd $dst,$src\t# MoveD2L" %} | |
| 11425 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %} | |
| 11426 ins_pipe( pipe_slow ); | |
| 11427 %} | |
| 11428 | |
| 11429 // The next instructions have long latency and use Int unit. Set high cost. | |
| 11430 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{ | |
| 11431 match(Set dst (MoveI2F src)); | |
| 11432 effect(DEF dst, USE src); | |
| 11433 ins_cost(300); | |
| 11434 format %{ "movd $dst,$src\t# MoveI2F" %} | |
| 11435 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %} | |
| 11436 ins_pipe( pipe_slow ); | |
| 11437 %} | |
| 11438 | |
| 11439 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{ | |
| 11440 match(Set dst (MoveL2D src)); | |
| 11441 effect(DEF dst, USE src); | |
| 11442 ins_cost(300); | |
| 11443 format %{ "movd $dst,$src\t# MoveL2D" %} | |
| 11444 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %} | |
| 11445 ins_pipe( pipe_slow ); | |
| 11446 %} | |
| 11447 | |
| 11448 // Replicate scalar to packed byte (1 byte) values in xmm | |
| 11449 instruct Repl8B_reg(regD dst, regD src) %{ | |
| 11450 match(Set dst (Replicate8B src)); | |
| 11451 format %{ "MOVDQA $dst,$src\n\t" | |
| 11452 "PUNPCKLBW $dst,$dst\n\t" | |
| 11453 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %} | |
| 11454 ins_encode( pshufd_8x8(dst, src)); | |
| 11455 ins_pipe( pipe_slow ); | |
| 11456 %} | |
| 11457 | |
| 11458 // Replicate scalar to packed byte (1 byte) values in xmm | |
| 11459 instruct Repl8B_rRegI(regD dst, rRegI src) %{ | |
| 11460 match(Set dst (Replicate8B src)); | |
| 11461 format %{ "MOVD $dst,$src\n\t" | |
| 11462 "PUNPCKLBW $dst,$dst\n\t" | |
| 11463 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %} | |
| 11464 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst)); | |
| 11465 ins_pipe( pipe_slow ); | |
| 11466 %} | |
| 11467 | |
| 11468 // Replicate scalar zero to packed byte (1 byte) values in xmm | |
| 11469 instruct Repl8B_immI0(regD dst, immI0 zero) %{ | |
| 11470 match(Set dst (Replicate8B zero)); | |
| 11471 format %{ "PXOR $dst,$dst\t! replicate8B" %} | |
| 11472 ins_encode( pxor(dst, dst)); | |
| 11473 ins_pipe( fpu_reg_reg ); | |
| 11474 %} | |
| 11475 | |
| 11476 // Replicate scalar to packed shore (2 byte) values in xmm | |
| 11477 instruct Repl4S_reg(regD dst, regD src) %{ | |
| 11478 match(Set dst (Replicate4S src)); | |
| 11479 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %} | |
| 11480 ins_encode( pshufd_4x16(dst, src)); | |
| 11481 ins_pipe( fpu_reg_reg ); | |
| 11482 %} | |
| 11483 | |
| 11484 // Replicate scalar to packed shore (2 byte) values in xmm | |
| 11485 instruct Repl4S_rRegI(regD dst, rRegI src) %{ | |
| 11486 match(Set dst (Replicate4S src)); | |
| 11487 format %{ "MOVD $dst,$src\n\t" | |
| 11488 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %} | |
| 11489 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst)); | |
| 11490 ins_pipe( fpu_reg_reg ); | |
| 11491 %} | |
| 11492 | |
| 11493 // Replicate scalar zero to packed short (2 byte) values in xmm | |
| 11494 instruct Repl4S_immI0(regD dst, immI0 zero) %{ | |
| 11495 match(Set dst (Replicate4S zero)); | |
| 11496 format %{ "PXOR $dst,$dst\t! replicate4S" %} | |
| 11497 ins_encode( pxor(dst, dst)); | |
| 11498 ins_pipe( fpu_reg_reg ); | |
| 11499 %} | |
| 11500 | |
| 11501 // Replicate scalar to packed char (2 byte) values in xmm | |
| 11502 instruct Repl4C_reg(regD dst, regD src) %{ | |
| 11503 match(Set dst (Replicate4C src)); | |
| 11504 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %} | |
| 11505 ins_encode( pshufd_4x16(dst, src)); | |
| 11506 ins_pipe( fpu_reg_reg ); | |
| 11507 %} | |
| 11508 | |
| 11509 // Replicate scalar to packed char (2 byte) values in xmm | |
| 11510 instruct Repl4C_rRegI(regD dst, rRegI src) %{ | |
| 11511 match(Set dst (Replicate4C src)); | |
| 11512 format %{ "MOVD $dst,$src\n\t" | |
| 11513 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %} | |
| 11514 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst)); | |
| 11515 ins_pipe( fpu_reg_reg ); | |
| 11516 %} | |
| 11517 | |
| 11518 // Replicate scalar zero to packed char (2 byte) values in xmm | |
| 11519 instruct Repl4C_immI0(regD dst, immI0 zero) %{ | |
| 11520 match(Set dst (Replicate4C zero)); | |
| 11521 format %{ "PXOR $dst,$dst\t! replicate4C" %} | |
| 11522 ins_encode( pxor(dst, dst)); | |
| 11523 ins_pipe( fpu_reg_reg ); | |
| 11524 %} | |
| 11525 | |
| 11526 // Replicate scalar to packed integer (4 byte) values in xmm | |
| 11527 instruct Repl2I_reg(regD dst, regD src) %{ | |
| 11528 match(Set dst (Replicate2I src)); | |
| 11529 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %} | |
| 11530 ins_encode( pshufd(dst, src, 0x00)); | |
| 11531 ins_pipe( fpu_reg_reg ); | |
| 11532 %} | |
| 11533 | |
| 11534 // Replicate scalar to packed integer (4 byte) values in xmm | |
| 11535 instruct Repl2I_rRegI(regD dst, rRegI src) %{ | |
| 11536 match(Set dst (Replicate2I src)); | |
| 11537 format %{ "MOVD $dst,$src\n\t" | |
| 11538 "PSHUFD $dst,$dst,0x00\t! replicate2I" %} | |
| 11539 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00)); | |
| 11540 ins_pipe( fpu_reg_reg ); | |
| 11541 %} | |
| 11542 | |
| 11543 // Replicate scalar zero to packed integer (2 byte) values in xmm | |
| 11544 instruct Repl2I_immI0(regD dst, immI0 zero) %{ | |
| 11545 match(Set dst (Replicate2I zero)); | |
| 11546 format %{ "PXOR $dst,$dst\t! replicate2I" %} | |
| 11547 ins_encode( pxor(dst, dst)); | |
| 11548 ins_pipe( fpu_reg_reg ); | |
| 11549 %} | |
| 11550 | |
| 11551 // Replicate scalar to packed single precision floating point values in xmm | |
| 11552 instruct Repl2F_reg(regD dst, regD src) %{ | |
| 11553 match(Set dst (Replicate2F src)); | |
| 11554 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %} | |
| 11555 ins_encode( pshufd(dst, src, 0xe0)); | |
| 11556 ins_pipe( fpu_reg_reg ); | |
| 11557 %} | |
| 11558 | |
| 11559 // Replicate scalar to packed single precision floating point values in xmm | |
| 11560 instruct Repl2F_regF(regD dst, regF src) %{ | |
| 11561 match(Set dst (Replicate2F src)); | |
| 11562 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %} | |
| 11563 ins_encode( pshufd(dst, src, 0xe0)); | |
| 11564 ins_pipe( fpu_reg_reg ); | |
| 11565 %} | |
| 11566 | |
| 11567 // Replicate scalar to packed single precision floating point values in xmm | |
| 11568 instruct Repl2F_immF0(regD dst, immF0 zero) %{ | |
| 11569 match(Set dst (Replicate2F zero)); | |
| 11570 format %{ "PXOR $dst,$dst\t! replicate2F" %} | |
| 11571 ins_encode( pxor(dst, dst)); | |
| 11572 ins_pipe( fpu_reg_reg ); | |
| 11573 %} | |
| 11574 | |
| 11575 | |
| 11576 // ======================================================================= | |
| 11577 // fast clearing of an array | |
| 11578 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy, | |
| 11579 rFlagsReg cr) | |
| 11580 %{ | |
| 11581 match(Set dummy (ClearArray cnt base)); | |
| 11582 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr); | |
| 11583 | |
| 11584 format %{ "xorl rax, rax\t# ClearArray:\n\t" | |
| 11585 "rep stosq\t# Store rax to *rdi++ while rcx--" %} | |
| 11586 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax | |
| 11587 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos | |
| 11588 ins_pipe(pipe_slow); | |
| 11589 %} | |
| 11590 | |
| 11591 instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1, | |
| 11592 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr) | |
| 11593 %{ | |
| 11594 match(Set result (StrComp str1 str2)); | |
| 11595 effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr); | |
| 11596 //ins_cost(300); | |
| 11597 | |
| 11598 format %{ "String Compare $str1, $str2 -> $result // XXX KILL RAX, RBX" %} | |
| 11599 ins_encode( enc_String_Compare() ); | |
| 11600 ins_pipe( pipe_slow ); | |
| 11601 %} | |
| 11602 | |
|
169
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11603 // fast array equals |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11604 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11605 rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr) %{ |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11606 match(Set result (AryEq ary1 ary2)); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11607 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11608 //ins_cost(300); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11609 |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11610 format %{ "Array Equals $ary1,$ary2 -> $result // KILL RAX, RBX" %} |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11611 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) ); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11612 ins_pipe( pipe_slow ); |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11613 %} |
|
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
168
diff
changeset
|
11614 |
| 0 | 11615 //----------Control Flow Instructions------------------------------------------ |
| 11616 // Signed compare Instructions | |
| 11617 | |
| 11618 // XXX more variants!! | |
| 11619 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2) | |
| 11620 %{ | |
| 11621 match(Set cr (CmpI op1 op2)); | |
| 11622 effect(DEF cr, USE op1, USE op2); | |
| 11623 | |
| 11624 format %{ "cmpl $op1, $op2" %} | |
| 11625 opcode(0x3B); /* Opcode 3B /r */ | |
| 11626 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2)); | |
| 11627 ins_pipe(ialu_cr_reg_reg); | |
| 11628 %} | |
| 11629 | |
| 11630 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2) | |
| 11631 %{ | |
| 11632 match(Set cr (CmpI op1 op2)); | |
| 11633 | |
| 11634 format %{ "cmpl $op1, $op2" %} | |
| 11635 opcode(0x81, 0x07); /* Opcode 81 /7 */ | |
| 11636 ins_encode(OpcSErm(op1, op2), Con8or32(op2)); | |
| 11637 ins_pipe(ialu_cr_reg_imm); | |
| 11638 %} | |
| 11639 | |
| 11640 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2) | |
| 11641 %{ | |
| 11642 match(Set cr (CmpI op1 (LoadI op2))); | |
| 11643 | |
| 11644 ins_cost(500); // XXX | |
| 11645 format %{ "cmpl $op1, $op2" %} | |
| 11646 opcode(0x3B); /* Opcode 3B /r */ | |
| 11647 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2)); | |
| 11648 ins_pipe(ialu_cr_reg_mem); | |
| 11649 %} | |
| 11650 | |
| 11651 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero) | |
| 11652 %{ | |
| 11653 match(Set cr (CmpI src zero)); | |
| 11654 | |
| 11655 format %{ "testl $src, $src" %} | |
| 11656 opcode(0x85); | |
| 11657 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src)); | |
| 11658 ins_pipe(ialu_cr_reg_imm); | |
| 11659 %} | |
| 11660 | |
| 11661 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero) | |
| 11662 %{ | |
| 11663 match(Set cr (CmpI (AndI src con) zero)); | |
| 11664 | |
| 11665 format %{ "testl $src, $con" %} | |
| 11666 opcode(0xF7, 0x00); | |
| 11667 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con)); | |
| 11668 ins_pipe(ialu_cr_reg_imm); | |
| 11669 %} | |
| 11670 | |
| 11671 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero) | |
| 11672 %{ | |
| 11673 match(Set cr (CmpI (AndI src (LoadI mem)) zero)); | |
| 11674 | |
| 11675 format %{ "testl $src, $mem" %} | |
| 11676 opcode(0x85); | |
| 11677 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem)); | |
| 11678 ins_pipe(ialu_cr_reg_mem); | |
| 11679 %} | |
| 11680 | |
| 11681 // Unsigned compare Instructions; really, same as signed except they | |
| 11682 // produce an rFlagsRegU instead of rFlagsReg. | |
| 11683 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2) | |
| 11684 %{ | |
| 11685 match(Set cr (CmpU op1 op2)); | |
| 11686 | |
| 11687 format %{ "cmpl $op1, $op2\t# unsigned" %} | |
| 11688 opcode(0x3B); /* Opcode 3B /r */ | |
| 11689 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2)); | |
| 11690 ins_pipe(ialu_cr_reg_reg); | |
| 11691 %} | |
| 11692 | |
| 11693 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2) | |
| 11694 %{ | |
| 11695 match(Set cr (CmpU op1 op2)); | |
| 11696 | |
| 11697 format %{ "cmpl $op1, $op2\t# unsigned" %} | |
| 11698 opcode(0x81,0x07); /* Opcode 81 /7 */ | |
| 11699 ins_encode(OpcSErm(op1, op2), Con8or32(op2)); | |
| 11700 ins_pipe(ialu_cr_reg_imm); | |
| 11701 %} | |
| 11702 | |
| 11703 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2) | |
| 11704 %{ | |
| 11705 match(Set cr (CmpU op1 (LoadI op2))); | |
| 11706 | |
| 11707 ins_cost(500); // XXX | |
| 11708 format %{ "cmpl $op1, $op2\t# unsigned" %} | |
| 11709 opcode(0x3B); /* Opcode 3B /r */ | |
| 11710 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2)); | |
| 11711 ins_pipe(ialu_cr_reg_mem); | |
| 11712 %} | |
| 11713 | |
| 11714 // // // Cisc-spilled version of cmpU_rReg | |
| 11715 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2) | |
| 11716 // //%{ | |
| 11717 // // match(Set cr (CmpU (LoadI op1) op2)); | |
| 11718 // // | |
| 11719 // // format %{ "CMPu $op1,$op2" %} | |
| 11720 // // ins_cost(500); | |
| 11721 // // opcode(0x39); /* Opcode 39 /r */ | |
| 11722 // // ins_encode( OpcP, reg_mem( op1, op2) ); | |
| 11723 // //%} | |
| 11724 | |
| 11725 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero) | |
| 11726 %{ | |
| 11727 match(Set cr (CmpU src zero)); | |
| 11728 | |
| 11729 format %{ "testl $src, $src\t# unsigned" %} | |
| 11730 opcode(0x85); | |
| 11731 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src)); | |
| 11732 ins_pipe(ialu_cr_reg_imm); | |
| 11733 %} | |
| 11734 | |
| 11735 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2) | |
| 11736 %{ | |
| 11737 match(Set cr (CmpP op1 op2)); | |
| 11738 | |
| 11739 format %{ "cmpq $op1, $op2\t# ptr" %} | |
| 11740 opcode(0x3B); /* Opcode 3B /r */ | |
| 11741 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2)); | |
| 11742 ins_pipe(ialu_cr_reg_reg); | |
| 11743 %} | |
| 11744 | |
| 11745 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2) | |
| 11746 %{ | |
| 11747 match(Set cr (CmpP op1 (LoadP op2))); | |
| 11748 | |
| 11749 ins_cost(500); // XXX | |
| 11750 format %{ "cmpq $op1, $op2\t# ptr" %} | |
| 11751 opcode(0x3B); /* Opcode 3B /r */ | |
| 11752 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2)); | |
| 11753 ins_pipe(ialu_cr_reg_mem); | |
| 11754 %} | |
| 11755 | |
| 11756 // // // Cisc-spilled version of cmpP_rReg | |
| 11757 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2) | |
| 11758 // //%{ | |
| 11759 // // match(Set cr (CmpP (LoadP op1) op2)); | |
| 11760 // // | |
| 11761 // // format %{ "CMPu $op1,$op2" %} | |
| 11762 // // ins_cost(500); | |
| 11763 // // opcode(0x39); /* Opcode 39 /r */ | |
| 11764 // // ins_encode( OpcP, reg_mem( op1, op2) ); | |
| 11765 // //%} | |
| 11766 | |
| 11767 // XXX this is generalized by compP_rReg_mem??? | |
| 11768 // Compare raw pointer (used in out-of-heap check). | |
| 11769 // Only works because non-oop pointers must be raw pointers | |
| 11770 // and raw pointers have no anti-dependencies. | |
| 11771 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2) | |
| 11772 %{ | |
| 11773 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr()); | |
| 11774 match(Set cr (CmpP op1 (LoadP op2))); | |
| 11775 | |
| 11776 format %{ "cmpq $op1, $op2\t# raw ptr" %} | |
| 11777 opcode(0x3B); /* Opcode 3B /r */ | |
| 11778 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2)); | |
| 11779 ins_pipe(ialu_cr_reg_mem); | |
| 11780 %} | |
| 11781 | |
| 11782 // This will generate a signed flags result. This should be OK since | |
| 11783 // any compare to a zero should be eq/neq. | |
| 11784 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero) | |
| 11785 %{ | |
| 11786 match(Set cr (CmpP src zero)); | |
| 11787 | |
| 11788 format %{ "testq $src, $src\t# ptr" %} | |
| 11789 opcode(0x85); | |
| 11790 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src)); | |
| 11791 ins_pipe(ialu_cr_reg_imm); | |
| 11792 %} | |
| 11793 | |
| 11794 // This will generate a signed flags result. This should be OK since | |
| 11795 // any compare to a zero should be eq/neq. | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11796 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11797 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11798 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL)); |
| 0 | 11799 match(Set cr (CmpP (LoadP op) zero)); |
| 11800 | |
| 11801 ins_cost(500); // XXX | |
| 11802 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %} | |
| 11803 opcode(0xF7); /* Opcode F7 /0 */ | |
| 11804 ins_encode(REX_mem_wide(op), | |
| 11805 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF)); | |
| 11806 ins_pipe(ialu_cr_reg_imm); | |
| 11807 %} | |
| 11808 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11809 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11810 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11811 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11812 match(Set cr (CmpP (LoadP mem) zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11813 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11814 format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11815 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11816 __ cmpq(r12, $mem$$Address); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11817 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11818 ins_pipe(ialu_cr_reg_mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11819 %} |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11820 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11821 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2) |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11822 %{ |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11823 match(Set cr (CmpN op1 op2)); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11824 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11825 format %{ "cmpl $op1, $op2\t# compressed ptr" %} |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11826 ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %} |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11827 ins_pipe(ialu_cr_reg_reg); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11828 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11829 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11830 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem) |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11831 %{ |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11832 match(Set cr (CmpN src (LoadN mem))); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11833 |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11834 format %{ "cmpl $src, $mem\t# compressed ptr" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11835 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11836 __ cmpl($src$$Register, $mem$$Address); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11837 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11838 ins_pipe(ialu_cr_reg_mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11839 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11840 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11841 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11842 match(Set cr (CmpN op1 op2)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11843 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11844 format %{ "cmpl $op1, $op2\t# compressed ptr" %} |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11845 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11846 __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11847 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11848 ins_pipe(ialu_cr_reg_imm); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11849 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11850 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11851 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11852 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11853 match(Set cr (CmpN src (LoadN mem))); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11854 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11855 format %{ "cmpl $mem, $src\t# compressed ptr" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11856 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11857 __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant); |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11858 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11859 ins_pipe(ialu_cr_reg_mem); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11860 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11861 |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11862 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{ |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11863 match(Set cr (CmpN src zero)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11864 |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11865 format %{ "testl $src, $src\t# compressed ptr" %} |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11866 ins_encode %{ __ testl($src$$Register, $src$$Register); %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11867 ins_pipe(ialu_cr_reg_imm); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11868 %} |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
11869 |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11870 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11871 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11872 predicate(Universe::narrow_oop_base() != NULL); |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11873 match(Set cr (CmpN (LoadN mem) zero)); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11874 |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11875 ins_cost(500); // XXX |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11876 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11877 ins_encode %{ |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11878 __ cmpl($mem$$Address, (int)0xFFFFFFFF); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11879 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11880 ins_pipe(ialu_cr_reg_mem); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11881 %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11882 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11883 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero) |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11884 %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11885 predicate(Universe::narrow_oop_base() == NULL); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11886 match(Set cr (CmpN (LoadN mem) zero)); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11887 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11888 format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %} |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11889 ins_encode %{ |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
11890 __ cmpl(r12, $mem$$Address); |
|
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11891 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11892 ins_pipe(ialu_cr_reg_mem); |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11893 %} |
|
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
11894 |
| 0 | 11895 // Yanked all unsigned pointer compare operations. |
| 11896 // Pointer compares are done with CmpP which is already unsigned. | |
| 11897 | |
| 11898 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2) | |
| 11899 %{ | |
| 11900 match(Set cr (CmpL op1 op2)); | |
| 11901 | |
| 11902 format %{ "cmpq $op1, $op2" %} | |
| 11903 opcode(0x3B); /* Opcode 3B /r */ | |
| 11904 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2)); | |
| 11905 ins_pipe(ialu_cr_reg_reg); | |
| 11906 %} | |
| 11907 | |
| 11908 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2) | |
| 11909 %{ | |
| 11910 match(Set cr (CmpL op1 op2)); | |
| 11911 | |
| 11912 format %{ "cmpq $op1, $op2" %} | |
| 11913 opcode(0x81, 0x07); /* Opcode 81 /7 */ | |
| 11914 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2)); | |
| 11915 ins_pipe(ialu_cr_reg_imm); | |
| 11916 %} | |
| 11917 | |
| 11918 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2) | |
| 11919 %{ | |
| 11920 match(Set cr (CmpL op1 (LoadL op2))); | |
| 11921 | |
| 11922 format %{ "cmpq $op1, $op2" %} | |
| 11923 opcode(0x3B); /* Opcode 3B /r */ | |
| 11924 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2)); | |
| 11925 ins_pipe(ialu_cr_reg_mem); | |
| 11926 %} | |
| 11927 | |
| 11928 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero) | |
| 11929 %{ | |
| 11930 match(Set cr (CmpL src zero)); | |
| 11931 | |
| 11932 format %{ "testq $src, $src" %} | |
| 11933 opcode(0x85); | |
| 11934 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src)); | |
| 11935 ins_pipe(ialu_cr_reg_imm); | |
| 11936 %} | |
| 11937 | |
| 11938 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero) | |
| 11939 %{ | |
| 11940 match(Set cr (CmpL (AndL src con) zero)); | |
| 11941 | |
| 11942 format %{ "testq $src, $con\t# long" %} | |
| 11943 opcode(0xF7, 0x00); | |
| 11944 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con)); | |
| 11945 ins_pipe(ialu_cr_reg_imm); | |
| 11946 %} | |
| 11947 | |
| 11948 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero) | |
| 11949 %{ | |
| 11950 match(Set cr (CmpL (AndL src (LoadL mem)) zero)); | |
| 11951 | |
| 11952 format %{ "testq $src, $mem" %} | |
| 11953 opcode(0x85); | |
| 11954 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem)); | |
| 11955 ins_pipe(ialu_cr_reg_mem); | |
| 11956 %} | |
| 11957 | |
| 11958 // Manifest a CmpL result in an integer register. Very painful. | |
| 11959 // This is the test to avoid. | |
| 11960 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags) | |
| 11961 %{ | |
| 11962 match(Set dst (CmpL3 src1 src2)); | |
| 11963 effect(KILL flags); | |
| 11964 | |
| 11965 ins_cost(275); // XXX | |
| 11966 format %{ "cmpq $src1, $src2\t# CmpL3\n\t" | |
| 11967 "movl $dst, -1\n\t" | |
| 11968 "jl,s done\n\t" | |
| 11969 "setne $dst\n\t" | |
| 11970 "movzbl $dst, $dst\n\t" | |
| 11971 "done:" %} | |
| 11972 ins_encode(cmpl3_flag(src1, src2, dst)); | |
| 11973 ins_pipe(pipe_slow); | |
| 11974 %} | |
| 11975 | |
| 11976 //----------Max and Min-------------------------------------------------------- | |
| 11977 // Min Instructions | |
| 11978 | |
| 11979 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr) | |
| 11980 %{ | |
| 11981 effect(USE_DEF dst, USE src, USE cr); | |
| 11982 | |
| 11983 format %{ "cmovlgt $dst, $src\t# min" %} | |
| 11984 opcode(0x0F, 0x4F); | |
| 11985 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 11986 ins_pipe(pipe_cmov_reg); | |
| 11987 %} | |
| 11988 | |
| 11989 | |
| 11990 instruct minI_rReg(rRegI dst, rRegI src) | |
| 11991 %{ | |
| 11992 match(Set dst (MinI dst src)); | |
| 11993 | |
| 11994 ins_cost(200); | |
| 11995 expand %{ | |
| 11996 rFlagsReg cr; | |
| 11997 compI_rReg(cr, dst, src); | |
| 11998 cmovI_reg_g(dst, src, cr); | |
| 11999 %} | |
| 12000 %} | |
| 12001 | |
| 12002 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr) | |
| 12003 %{ | |
| 12004 effect(USE_DEF dst, USE src, USE cr); | |
| 12005 | |
| 12006 format %{ "cmovllt $dst, $src\t# max" %} | |
| 12007 opcode(0x0F, 0x4C); | |
| 12008 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src)); | |
| 12009 ins_pipe(pipe_cmov_reg); | |
| 12010 %} | |
| 12011 | |
| 12012 | |
| 12013 instruct maxI_rReg(rRegI dst, rRegI src) | |
| 12014 %{ | |
| 12015 match(Set dst (MaxI dst src)); | |
| 12016 | |
| 12017 ins_cost(200); | |
| 12018 expand %{ | |
| 12019 rFlagsReg cr; | |
| 12020 compI_rReg(cr, dst, src); | |
| 12021 cmovI_reg_l(dst, src, cr); | |
| 12022 %} | |
| 12023 %} | |
| 12024 | |
| 12025 // ============================================================================ | |
| 12026 // Branch Instructions | |
| 12027 | |
| 12028 // Jump Direct - Label defines a relative address from JMP+1 | |
| 12029 instruct jmpDir(label labl) | |
| 12030 %{ | |
| 12031 match(Goto); | |
| 12032 effect(USE labl); | |
| 12033 | |
| 12034 ins_cost(300); | |
| 12035 format %{ "jmp $labl" %} | |
| 12036 size(5); | |
| 12037 opcode(0xE9); | |
| 12038 ins_encode(OpcP, Lbl(labl)); | |
| 12039 ins_pipe(pipe_jmp); | |
| 12040 ins_pc_relative(1); | |
| 12041 %} | |
| 12042 | |
| 12043 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
| 12044 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl) | |
| 12045 %{ | |
| 12046 match(If cop cr); | |
| 12047 effect(USE labl); | |
| 12048 | |
| 12049 ins_cost(300); | |
| 12050 format %{ "j$cop $labl" %} | |
| 12051 size(6); | |
| 12052 opcode(0x0F, 0x80); | |
| 12053 ins_encode(Jcc(cop, labl)); | |
| 12054 ins_pipe(pipe_jcc); | |
| 12055 ins_pc_relative(1); | |
| 12056 %} | |
| 12057 | |
| 12058 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
| 12059 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl) | |
| 12060 %{ | |
| 12061 match(CountedLoopEnd cop cr); | |
| 12062 effect(USE labl); | |
| 12063 | |
| 12064 ins_cost(300); | |
| 12065 format %{ "j$cop $labl\t# loop end" %} | |
| 12066 size(6); | |
| 12067 opcode(0x0F, 0x80); | |
| 12068 ins_encode(Jcc(cop, labl)); | |
| 12069 ins_pipe(pipe_jcc); | |
| 12070 ins_pc_relative(1); | |
| 12071 %} | |
| 12072 | |
| 12073 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12074 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{ |
| 0 | 12075 match(CountedLoopEnd cop cmp); |
| 12076 effect(USE labl); | |
| 12077 | |
| 12078 ins_cost(300); | |
| 12079 format %{ "j$cop,u $labl\t# loop end" %} | |
| 12080 size(6); | |
| 12081 opcode(0x0F, 0x80); | |
| 12082 ins_encode(Jcc(cop, labl)); | |
| 12083 ins_pipe(pipe_jcc); | |
| 12084 ins_pc_relative(1); | |
| 12085 %} | |
| 12086 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12087 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12088 match(CountedLoopEnd cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12089 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12090 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12091 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12092 format %{ "j$cop,u $labl\t# loop end" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12093 size(6); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12094 opcode(0x0F, 0x80); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12095 ins_encode(Jcc(cop, labl)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12096 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12097 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12098 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12099 |
| 0 | 12100 // Jump Direct Conditional - using unsigned comparison |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12101 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{ |
| 0 | 12102 match(If cop cmp); |
| 12103 effect(USE labl); | |
| 12104 | |
| 12105 ins_cost(300); | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12106 format %{ "j$cop,u $labl" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12107 size(6); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12108 opcode(0x0F, 0x80); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12109 ins_encode(Jcc(cop, labl)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12110 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12111 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12112 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12113 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12114 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12115 match(If cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12116 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12117 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12118 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12119 format %{ "j$cop,u $labl" %} |
| 0 | 12120 size(6); |
| 12121 opcode(0x0F, 0x80); | |
| 12122 ins_encode(Jcc(cop, labl)); | |
| 12123 ins_pipe(pipe_jcc); | |
| 12124 ins_pc_relative(1); | |
| 12125 %} | |
| 12126 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12127 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12128 match(If cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12129 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12130 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12131 ins_cost(200); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12132 format %{ $$template |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12133 if ($cop$$cmpcode == Assembler::notEqual) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12134 $$emit$$"jp,u $labl\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12135 $$emit$$"j$cop,u $labl" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12136 } else { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12137 $$emit$$"jp,u done\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12138 $$emit$$"j$cop,u $labl\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12139 $$emit$$"done:" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12140 } |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12141 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12142 size(12); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12143 opcode(0x0F, 0x80); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12144 ins_encode %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12145 Label* l = $labl$$label; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12146 $$$emit8$primary; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12147 emit_cc(cbuf, $secondary, Assembler::parity); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12148 int parity_disp = -1; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12149 if ($cop$$cmpcode == Assembler::notEqual) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12150 // the two jumps 6 bytes apart so the jump distances are too |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12151 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12152 } else if ($cop$$cmpcode == Assembler::equal) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12153 parity_disp = 6; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12154 } else { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12155 ShouldNotReachHere(); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12156 } |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12157 emit_d32(cbuf, parity_disp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12158 $$$emit8$primary; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12159 emit_cc(cbuf, $secondary, $cop$$cmpcode); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12160 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12161 emit_d32(cbuf, disp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12162 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12163 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12164 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12165 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12166 |
| 0 | 12167 // ============================================================================ |
| 12168 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary | |
| 12169 // superklass array for an instance of the superklass. Set a hidden | |
| 12170 // internal cache on a hit (cache is checked with exposed code in | |
| 12171 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The | |
| 12172 // encoding ALSO sets flags. | |
| 12173 | |
| 12174 instruct partialSubtypeCheck(rdi_RegP result, | |
| 12175 rsi_RegP sub, rax_RegP super, rcx_RegI rcx, | |
| 12176 rFlagsReg cr) | |
| 12177 %{ | |
| 12178 match(Set result (PartialSubtypeCheck sub super)); | |
| 12179 effect(KILL rcx, KILL cr); | |
| 12180 | |
| 12181 ins_cost(1100); // slightly larger than the next version | |
| 12182 format %{ "cmpq rax, rsi\n\t" | |
| 12183 "jeq,s hit\n\t" | |
| 12184 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" | |
| 12185 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t" | |
| 12186 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t" | |
| 12187 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t" | |
| 12188 "jne,s miss\t\t# Missed: rdi not-zero\n\t" | |
| 12189 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t" | |
| 12190 "hit:\n\t" | |
| 12191 "xorq $result, $result\t\t Hit: rdi zero\n\t" | |
| 12192 "miss:\t" %} | |
| 12193 | |
| 12194 opcode(0x1); // Force a XOR of RDI | |
| 12195 ins_encode(enc_PartialSubtypeCheck()); | |
| 12196 ins_pipe(pipe_slow); | |
| 12197 %} | |
| 12198 | |
| 12199 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr, | |
| 12200 rsi_RegP sub, rax_RegP super, rcx_RegI rcx, | |
| 12201 immP0 zero, | |
| 12202 rdi_RegP result) | |
| 12203 %{ | |
| 12204 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero)); | |
| 12205 effect(KILL rcx, KILL result); | |
| 12206 | |
| 12207 ins_cost(1000); | |
| 12208 format %{ "cmpq rax, rsi\n\t" | |
| 12209 "jeq,s miss\t# Actually a hit; we are done.\n\t" | |
| 12210 "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" | |
| 12211 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t" | |
| 12212 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t" | |
| 12213 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t" | |
| 12214 "jne,s miss\t\t# Missed: flags nz\n\t" | |
| 12215 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t" | |
| 12216 "miss:\t" %} | |
| 12217 | |
| 12218 opcode(0x0); // No need to XOR RDI | |
| 12219 ins_encode(enc_PartialSubtypeCheck()); | |
| 12220 ins_pipe(pipe_slow); | |
| 12221 %} | |
| 12222 | |
| 12223 // ============================================================================ | |
| 12224 // Branch Instructions -- short offset versions | |
| 12225 // | |
| 12226 // These instructions are used to replace jumps of a long offset (the default | |
| 12227 // match) with jumps of a shorter offset. These instructions are all tagged | |
| 12228 // with the ins_short_branch attribute, which causes the ADLC to suppress the | |
| 12229 // match rules in general matching. Instead, the ADLC generates a conversion | |
| 12230 // method in the MachNode which can be used to do in-place replacement of the | |
| 12231 // long variant with the shorter variant. The compiler will determine if a | |
| 12232 // branch can be taken by the is_short_branch_offset() predicate in the machine | |
| 12233 // specific code section of the file. | |
| 12234 | |
| 12235 // Jump Direct - Label defines a relative address from JMP+1 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12236 instruct jmpDir_short(label labl) %{ |
| 0 | 12237 match(Goto); |
| 12238 effect(USE labl); | |
| 12239 | |
| 12240 ins_cost(300); | |
| 12241 format %{ "jmp,s $labl" %} | |
| 12242 size(2); | |
| 12243 opcode(0xEB); | |
| 12244 ins_encode(OpcP, LblShort(labl)); | |
| 12245 ins_pipe(pipe_jmp); | |
| 12246 ins_pc_relative(1); | |
| 12247 ins_short_branch(1); | |
| 12248 %} | |
| 12249 | |
| 12250 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12251 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{ |
| 0 | 12252 match(If cop cr); |
| 12253 effect(USE labl); | |
| 12254 | |
| 12255 ins_cost(300); | |
| 12256 format %{ "j$cop,s $labl" %} | |
| 12257 size(2); | |
| 12258 opcode(0x70); | |
| 12259 ins_encode(JccShort(cop, labl)); | |
| 12260 ins_pipe(pipe_jcc); | |
| 12261 ins_pc_relative(1); | |
| 12262 ins_short_branch(1); | |
| 12263 %} | |
| 12264 | |
| 12265 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12266 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{ |
| 0 | 12267 match(CountedLoopEnd cop cr); |
| 12268 effect(USE labl); | |
| 12269 | |
| 12270 ins_cost(300); | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12271 format %{ "j$cop,s $labl\t# loop end" %} |
| 0 | 12272 size(2); |
| 12273 opcode(0x70); | |
| 12274 ins_encode(JccShort(cop, labl)); | |
| 12275 ins_pipe(pipe_jcc); | |
| 12276 ins_pc_relative(1); | |
| 12277 ins_short_branch(1); | |
| 12278 %} | |
| 12279 | |
| 12280 // Jump Direct Conditional - Label defines a relative address from Jcc+1 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12281 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12282 match(CountedLoopEnd cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12283 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12284 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12285 ins_cost(300); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12286 format %{ "j$cop,us $labl\t# loop end" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12287 size(2); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12288 opcode(0x70); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12289 ins_encode(JccShort(cop, labl)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12290 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12291 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12292 ins_short_branch(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12293 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12294 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12295 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{ |
| 0 | 12296 match(CountedLoopEnd cop cmp); |
| 12297 effect(USE labl); | |
| 12298 | |
| 12299 ins_cost(300); | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12300 format %{ "j$cop,us $labl\t# loop end" %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12301 size(2); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12302 opcode(0x70); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12303 ins_encode(JccShort(cop, labl)); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12304 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12305 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12306 ins_short_branch(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12307 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12308 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12309 // Jump Direct Conditional - using unsigned comparison |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12310 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12311 match(If cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12312 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12313 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12314 ins_cost(300); |
| 0 | 12315 format %{ "j$cop,us $labl" %} |
| 12316 size(2); | |
| 12317 opcode(0x70); | |
| 12318 ins_encode(JccShort(cop, labl)); | |
| 12319 ins_pipe(pipe_jcc); | |
| 12320 ins_pc_relative(1); | |
| 12321 ins_short_branch(1); | |
| 12322 %} | |
| 12323 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12324 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{ |
| 0 | 12325 match(If cop cmp); |
| 12326 effect(USE labl); | |
| 12327 | |
| 12328 ins_cost(300); | |
| 12329 format %{ "j$cop,us $labl" %} | |
| 12330 size(2); | |
| 12331 opcode(0x70); | |
| 12332 ins_encode(JccShort(cop, labl)); | |
| 12333 ins_pipe(pipe_jcc); | |
| 12334 ins_pc_relative(1); | |
| 12335 ins_short_branch(1); | |
| 12336 %} | |
| 12337 | |
|
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12338 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12339 match(If cop cmp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12340 effect(USE labl); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12341 |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12342 ins_cost(300); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12343 format %{ $$template |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12344 if ($cop$$cmpcode == Assembler::notEqual) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12345 $$emit$$"jp,u,s $labl\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12346 $$emit$$"j$cop,u,s $labl" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12347 } else { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12348 $$emit$$"jp,u,s done\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12349 $$emit$$"j$cop,u,s $labl\n\t" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12350 $$emit$$"done:" |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12351 } |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12352 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12353 size(4); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12354 opcode(0x70); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12355 ins_encode %{ |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12356 Label* l = $labl$$label; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12357 emit_cc(cbuf, $primary, Assembler::parity); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12358 int parity_disp = -1; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12359 if ($cop$$cmpcode == Assembler::notEqual) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12360 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12361 } else if ($cop$$cmpcode == Assembler::equal) { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12362 parity_disp = 2; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12363 } else { |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12364 ShouldNotReachHere(); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12365 } |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12366 emit_d8(cbuf, parity_disp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12367 emit_cc(cbuf, $primary, $cop$$cmpcode); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12368 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0; |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12369 emit_d8(cbuf, disp); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12370 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12371 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp"); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12372 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12373 ins_pipe(pipe_jcc); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12374 ins_pc_relative(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12375 ins_short_branch(1); |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12376 %} |
|
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
403
diff
changeset
|
12377 |
| 0 | 12378 // ============================================================================ |
| 12379 // inlined locking and unlocking | |
| 12380 | |
| 12381 instruct cmpFastLock(rFlagsReg cr, | |
| 12382 rRegP object, rRegP box, rax_RegI tmp, rRegP scr) | |
| 12383 %{ | |
| 12384 match(Set cr (FastLock object box)); | |
| 12385 effect(TEMP tmp, TEMP scr); | |
| 12386 | |
| 12387 ins_cost(300); | |
| 12388 format %{ "fastlock $object,$box,$tmp,$scr" %} | |
| 12389 ins_encode(Fast_Lock(object, box, tmp, scr)); | |
| 12390 ins_pipe(pipe_slow); | |
| 12391 ins_pc_relative(1); | |
| 12392 %} | |
| 12393 | |
| 12394 instruct cmpFastUnlock(rFlagsReg cr, | |
| 12395 rRegP object, rax_RegP box, rRegP tmp) | |
| 12396 %{ | |
| 12397 match(Set cr (FastUnlock object box)); | |
| 12398 effect(TEMP tmp); | |
| 12399 | |
| 12400 ins_cost(300); | |
| 12401 format %{ "fastunlock $object, $box, $tmp" %} | |
| 12402 ins_encode(Fast_Unlock(object, box, tmp)); | |
| 12403 ins_pipe(pipe_slow); | |
| 12404 ins_pc_relative(1); | |
| 12405 %} | |
| 12406 | |
| 12407 | |
| 12408 // ============================================================================ | |
| 12409 // Safepoint Instructions | |
| 12410 instruct safePoint_poll(rFlagsReg cr) | |
| 12411 %{ | |
| 12412 match(SafePoint); | |
| 12413 effect(KILL cr); | |
| 12414 | |
| 12415 format %{ "testl rax, [rip + #offset_to_poll_page]\t" | |
| 12416 "# Safepoint: poll for GC" %} | |
| 12417 size(6); // Opcode + ModRM + Disp32 == 6 bytes | |
| 12418 ins_cost(125); | |
| 12419 ins_encode(enc_safepoint_poll); | |
| 12420 ins_pipe(ialu_reg_mem); | |
| 12421 %} | |
| 12422 | |
| 12423 // ============================================================================ | |
| 12424 // Procedure Call/Return Instructions | |
| 12425 // Call Java Static Instruction | |
| 12426 // Note: If this code changes, the corresponding ret_addr_offset() and | |
| 12427 // compute_padding() functions will have to be adjusted. | |
| 12428 instruct CallStaticJavaDirect(method meth) | |
| 12429 %{ | |
| 12430 match(CallStaticJava); | |
| 12431 effect(USE meth); | |
| 12432 | |
| 12433 ins_cost(300); | |
| 12434 format %{ "call,static " %} | |
| 12435 opcode(0xE8); /* E8 cd */ | |
| 12436 ins_encode(Java_Static_Call(meth), call_epilog); | |
| 12437 ins_pipe(pipe_slow); | |
| 12438 ins_pc_relative(1); | |
| 12439 ins_alignment(4); | |
| 12440 %} | |
| 12441 | |
| 12442 // Call Java Dynamic Instruction | |
| 12443 // Note: If this code changes, the corresponding ret_addr_offset() and | |
| 12444 // compute_padding() functions will have to be adjusted. | |
| 12445 instruct CallDynamicJavaDirect(method meth) | |
| 12446 %{ | |
| 12447 match(CallDynamicJava); | |
| 12448 effect(USE meth); | |
| 12449 | |
| 12450 ins_cost(300); | |
| 12451 format %{ "movq rax, #Universe::non_oop_word()\n\t" | |
| 12452 "call,dynamic " %} | |
| 12453 opcode(0xE8); /* E8 cd */ | |
| 12454 ins_encode(Java_Dynamic_Call(meth), call_epilog); | |
| 12455 ins_pipe(pipe_slow); | |
| 12456 ins_pc_relative(1); | |
| 12457 ins_alignment(4); | |
| 12458 %} | |
| 12459 | |
| 12460 // Call Runtime Instruction | |
| 12461 instruct CallRuntimeDirect(method meth) | |
| 12462 %{ | |
| 12463 match(CallRuntime); | |
| 12464 effect(USE meth); | |
| 12465 | |
| 12466 ins_cost(300); | |
| 12467 format %{ "call,runtime " %} | |
| 12468 opcode(0xE8); /* E8 cd */ | |
| 12469 ins_encode(Java_To_Runtime(meth)); | |
| 12470 ins_pipe(pipe_slow); | |
| 12471 ins_pc_relative(1); | |
| 12472 %} | |
| 12473 | |
| 12474 // Call runtime without safepoint | |
| 12475 instruct CallLeafDirect(method meth) | |
| 12476 %{ | |
| 12477 match(CallLeaf); | |
| 12478 effect(USE meth); | |
| 12479 | |
| 12480 ins_cost(300); | |
| 12481 format %{ "call_leaf,runtime " %} | |
| 12482 opcode(0xE8); /* E8 cd */ | |
| 12483 ins_encode(Java_To_Runtime(meth)); | |
| 12484 ins_pipe(pipe_slow); | |
| 12485 ins_pc_relative(1); | |
| 12486 %} | |
| 12487 | |
| 12488 // Call runtime without safepoint | |
| 12489 instruct CallLeafNoFPDirect(method meth) | |
| 12490 %{ | |
| 12491 match(CallLeafNoFP); | |
| 12492 effect(USE meth); | |
| 12493 | |
| 12494 ins_cost(300); | |
| 12495 format %{ "call_leaf_nofp,runtime " %} | |
| 12496 opcode(0xE8); /* E8 cd */ | |
| 12497 ins_encode(Java_To_Runtime(meth)); | |
| 12498 ins_pipe(pipe_slow); | |
| 12499 ins_pc_relative(1); | |
| 12500 %} | |
| 12501 | |
| 12502 // Return Instruction | |
| 12503 // Remove the return address & jump to it. | |
| 12504 // Notice: We always emit a nop after a ret to make sure there is room | |
| 12505 // for safepoint patching | |
| 12506 instruct Ret() | |
| 12507 %{ | |
| 12508 match(Return); | |
| 12509 | |
| 12510 format %{ "ret" %} | |
| 12511 opcode(0xC3); | |
| 12512 ins_encode(OpcP); | |
| 12513 ins_pipe(pipe_jmp); | |
| 12514 %} | |
| 12515 | |
| 12516 // Tail Call; Jump from runtime stub to Java code. | |
| 12517 // Also known as an 'interprocedural jump'. | |
| 12518 // Target of jump will eventually return to caller. | |
| 12519 // TailJump below removes the return address. | |
| 12520 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop) | |
| 12521 %{ | |
| 12522 match(TailCall jump_target method_oop); | |
| 12523 | |
| 12524 ins_cost(300); | |
| 12525 format %{ "jmp $jump_target\t# rbx holds method oop" %} | |
| 12526 opcode(0xFF, 0x4); /* Opcode FF /4 */ | |
| 12527 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target)); | |
| 12528 ins_pipe(pipe_jmp); | |
| 12529 %} | |
| 12530 | |
| 12531 // Tail Jump; remove the return address; jump to target. | |
| 12532 // TailCall above leaves the return address around. | |
| 12533 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop) | |
| 12534 %{ | |
| 12535 match(TailJump jump_target ex_oop); | |
| 12536 | |
| 12537 ins_cost(300); | |
| 12538 format %{ "popq rdx\t# pop return address\n\t" | |
| 12539 "jmp $jump_target" %} | |
| 12540 opcode(0xFF, 0x4); /* Opcode FF /4 */ | |
| 12541 ins_encode(Opcode(0x5a), // popq rdx | |
| 12542 REX_reg(jump_target), OpcP, reg_opc(jump_target)); | |
| 12543 ins_pipe(pipe_jmp); | |
| 12544 %} | |
| 12545 | |
| 12546 // Create exception oop: created by stack-crawling runtime code. | |
| 12547 // Created exception is now available to this handler, and is setup | |
| 12548 // just prior to jumping to this handler. No code emitted. | |
| 12549 instruct CreateException(rax_RegP ex_oop) | |
| 12550 %{ | |
| 12551 match(Set ex_oop (CreateEx)); | |
| 12552 | |
| 12553 size(0); | |
| 12554 // use the following format syntax | |
| 12555 format %{ "# exception oop is in rax; no code emitted" %} | |
| 12556 ins_encode(); | |
| 12557 ins_pipe(empty); | |
| 12558 %} | |
| 12559 | |
| 12560 // Rethrow exception: | |
| 12561 // The exception oop will come in the first argument position. | |
| 12562 // Then JUMP (not call) to the rethrow stub code. | |
| 12563 instruct RethrowException() | |
| 12564 %{ | |
| 12565 match(Rethrow); | |
| 12566 | |
| 12567 // use the following format syntax | |
| 12568 format %{ "jmp rethrow_stub" %} | |
| 12569 ins_encode(enc_rethrow); | |
| 12570 ins_pipe(pipe_jmp); | |
| 12571 %} | |
| 12572 | |
| 12573 | |
| 12574 //----------PEEPHOLE RULES----------------------------------------------------- | |
| 12575 // These must follow all instruction definitions as they use the names | |
| 12576 // defined in the instructions definitions. | |
| 12577 // | |
| 605 | 12578 // peepmatch ( root_instr_name [preceding_instruction]* ); |
| 0 | 12579 // |
| 12580 // peepconstraint %{ | |
| 12581 // (instruction_number.operand_name relational_op instruction_number.operand_name | |
| 12582 // [, ...] ); | |
| 12583 // // instruction numbers are zero-based using left to right order in peepmatch | |
| 12584 // | |
| 12585 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) ); | |
| 12586 // // provide an instruction_number.operand_name for each operand that appears | |
| 12587 // // in the replacement instruction's match rule | |
| 12588 // | |
| 12589 // ---------VM FLAGS--------------------------------------------------------- | |
| 12590 // | |
| 12591 // All peephole optimizations can be turned off using -XX:-OptoPeephole | |
| 12592 // | |
| 12593 // Each peephole rule is given an identifying number starting with zero and | |
| 12594 // increasing by one in the order seen by the parser. An individual peephole | |
| 12595 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=# | |
| 12596 // on the command-line. | |
| 12597 // | |
| 12598 // ---------CURRENT LIMITATIONS---------------------------------------------- | |
| 12599 // | |
| 12600 // Only match adjacent instructions in same basic block | |
| 12601 // Only equality constraints | |
| 12602 // Only constraints between operands, not (0.dest_reg == RAX_enc) | |
| 12603 // Only one replacement instruction | |
| 12604 // | |
| 12605 // ---------EXAMPLE---------------------------------------------------------- | |
| 12606 // | |
| 12607 // // pertinent parts of existing instructions in architecture description | |
| 12608 // instruct movI(rRegI dst, rRegI src) | |
| 12609 // %{ | |
| 12610 // match(Set dst (CopyI src)); | |
| 12611 // %} | |
| 12612 // | |
| 12613 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr) | |
| 12614 // %{ | |
| 12615 // match(Set dst (AddI dst src)); | |
| 12616 // effect(KILL cr); | |
| 12617 // %} | |
| 12618 // | |
| 12619 // // Change (inc mov) to lea | |
| 12620 // peephole %{ | |
| 12621 // // increment preceeded by register-register move | |
| 12622 // peepmatch ( incI_rReg movI ); | |
| 12623 // // require that the destination register of the increment | |
| 12624 // // match the destination register of the move | |
| 12625 // peepconstraint ( 0.dst == 1.dst ); | |
| 12626 // // construct a replacement instruction that sets | |
| 12627 // // the destination to ( move's source register + one ) | |
| 12628 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) ); | |
| 12629 // %} | |
| 12630 // | |
| 12631 | |
| 12632 // Implementation no longer uses movX instructions since | |
| 12633 // machine-independent system no longer uses CopyX nodes. | |
| 12634 // | |
| 12635 // peephole | |
| 12636 // %{ | |
| 12637 // peepmatch (incI_rReg movI); | |
| 12638 // peepconstraint (0.dst == 1.dst); | |
| 12639 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src)); | |
| 12640 // %} | |
| 12641 | |
| 12642 // peephole | |
| 12643 // %{ | |
| 12644 // peepmatch (decI_rReg movI); | |
| 12645 // peepconstraint (0.dst == 1.dst); | |
| 12646 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src)); | |
| 12647 // %} | |
| 12648 | |
| 12649 // peephole | |
| 12650 // %{ | |
| 12651 // peepmatch (addI_rReg_imm movI); | |
| 12652 // peepconstraint (0.dst == 1.dst); | |
| 12653 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src)); | |
| 12654 // %} | |
| 12655 | |
| 12656 // peephole | |
| 12657 // %{ | |
| 12658 // peepmatch (incL_rReg movL); | |
| 12659 // peepconstraint (0.dst == 1.dst); | |
| 12660 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src)); | |
| 12661 // %} | |
| 12662 | |
| 12663 // peephole | |
| 12664 // %{ | |
| 12665 // peepmatch (decL_rReg movL); | |
| 12666 // peepconstraint (0.dst == 1.dst); | |
| 12667 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src)); | |
| 12668 // %} | |
| 12669 | |
| 12670 // peephole | |
| 12671 // %{ | |
| 12672 // peepmatch (addL_rReg_imm movL); | |
| 12673 // peepconstraint (0.dst == 1.dst); | |
| 12674 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src)); | |
| 12675 // %} | |
| 12676 | |
| 12677 // peephole | |
| 12678 // %{ | |
| 12679 // peepmatch (addP_rReg_imm movP); | |
| 12680 // peepconstraint (0.dst == 1.dst); | |
| 12681 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src)); | |
| 12682 // %} | |
| 12683 | |
| 12684 // // Change load of spilled value to only a spill | |
| 12685 // instruct storeI(memory mem, rRegI src) | |
| 12686 // %{ | |
| 12687 // match(Set mem (StoreI mem src)); | |
| 12688 // %} | |
| 12689 // | |
| 12690 // instruct loadI(rRegI dst, memory mem) | |
| 12691 // %{ | |
| 12692 // match(Set dst (LoadI mem)); | |
| 12693 // %} | |
| 12694 // | |
| 12695 | |
| 12696 peephole | |
| 12697 %{ | |
| 12698 peepmatch (loadI storeI); | |
| 12699 peepconstraint (1.src == 0.dst, 1.mem == 0.mem); | |
| 12700 peepreplace (storeI(1.mem 1.mem 1.src)); | |
| 12701 %} | |
| 12702 | |
| 12703 peephole | |
| 12704 %{ | |
| 12705 peepmatch (loadL storeL); | |
| 12706 peepconstraint (1.src == 0.dst, 1.mem == 0.mem); | |
| 12707 peepreplace (storeL(1.mem 1.mem 1.src)); | |
| 12708 %} | |
| 12709 | |
| 12710 //----------SMARTSPILL RULES--------------------------------------------------- | |
| 12711 // These must follow all instruction definitions as they use the names | |
| 12712 // defined in the instructions definitions. |