Mercurial > hg > truffle
annotate src/cpu/x86/vm/assembler_x86.hpp @ 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 1997-2009 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 class BiasedLockingCounters; | |
| 26 | |
| 27 // Contains all the definitions needed for x86 assembly code generation. | |
| 28 | |
| 29 // Calling convention | |
| 30 class Argument VALUE_OBJ_CLASS_SPEC { | |
| 31 public: | |
| 32 enum { | |
| 33 #ifdef _LP64 | |
| 34 #ifdef _WIN64 | |
| 35 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) | |
| 36 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) | |
| 37 #else | |
| 38 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) | |
| 39 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) | |
| 40 #endif // _WIN64 | |
| 41 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... | |
| 42 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... | |
| 43 #else | |
| 44 n_register_parameters = 0 // 0 registers used to pass arguments | |
| 45 #endif // _LP64 | |
| 46 }; | |
| 47 }; | |
| 48 | |
| 49 | |
| 50 #ifdef _LP64 | |
| 51 // Symbolically name the register arguments used by the c calling convention. | |
| 52 // Windows is different from linux/solaris. So much for standards... | |
| 53 | |
| 54 #ifdef _WIN64 | |
| 55 | |
| 56 REGISTER_DECLARATION(Register, c_rarg0, rcx); | |
| 57 REGISTER_DECLARATION(Register, c_rarg1, rdx); | |
| 58 REGISTER_DECLARATION(Register, c_rarg2, r8); | |
| 59 REGISTER_DECLARATION(Register, c_rarg3, r9); | |
| 60 | |
| 304 | 61 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); |
| 62 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
| 63 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
| 64 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
| 0 | 65 |
| 66 #else | |
| 67 | |
| 68 REGISTER_DECLARATION(Register, c_rarg0, rdi); | |
| 69 REGISTER_DECLARATION(Register, c_rarg1, rsi); | |
| 70 REGISTER_DECLARATION(Register, c_rarg2, rdx); | |
| 71 REGISTER_DECLARATION(Register, c_rarg3, rcx); | |
| 72 REGISTER_DECLARATION(Register, c_rarg4, r8); | |
| 73 REGISTER_DECLARATION(Register, c_rarg5, r9); | |
| 74 | |
| 304 | 75 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); |
| 76 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
| 77 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
| 78 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
| 79 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); | |
| 80 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); | |
| 81 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); | |
| 82 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); | |
| 0 | 83 |
| 84 #endif // _WIN64 | |
| 85 | |
| 86 // Symbolically name the register arguments used by the Java calling convention. | |
| 87 // We have control over the convention for java so we can do what we please. | |
| 88 // What pleases us is to offset the java calling convention so that when | |
| 89 // we call a suitable jni method the arguments are lined up and we don't | |
| 90 // have to do little shuffling. A suitable jni method is non-static and a | |
| 91 // small number of arguments (two fewer args on windows) | |
| 92 // | |
| 93 // |-------------------------------------------------------| | |
| 94 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | | |
| 95 // |-------------------------------------------------------| | |
| 96 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) | |
| 97 // | rdi rsi rdx rcx r8 r9 | solaris/linux | |
| 98 // |-------------------------------------------------------| | |
| 99 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | | |
| 100 // |-------------------------------------------------------| | |
| 101 | |
| 102 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); | |
| 103 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); | |
| 104 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); | |
| 105 // Windows runs out of register args here | |
| 106 #ifdef _WIN64 | |
| 107 REGISTER_DECLARATION(Register, j_rarg3, rdi); | |
| 108 REGISTER_DECLARATION(Register, j_rarg4, rsi); | |
| 109 #else | |
| 110 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); | |
| 111 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); | |
| 112 #endif /* _WIN64 */ | |
| 113 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); | |
| 114 | |
| 304 | 115 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); |
| 116 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); | |
| 117 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); | |
| 118 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); | |
| 119 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); | |
| 120 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); | |
| 121 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); | |
| 122 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); | |
| 0 | 123 |
| 124 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile | |
| 125 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile | |
| 126 | |
| 304 | 127 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved |
| 0 | 128 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved |
| 129 | |
| 304 | 130 #else |
| 131 // rscratch1 will apear in 32bit code that is dead but of course must compile | |
| 132 // Using noreg ensures if the dead code is incorrectly live and executed it | |
| 133 // will cause an assertion failure | |
| 134 #define rscratch1 noreg | |
| 135 | |
| 0 | 136 #endif // _LP64 |
| 137 | |
| 138 // Address is an abstraction used to represent a memory location | |
| 139 // using any of the amd64 addressing modes with one object. | |
| 140 // | |
| 141 // Note: A register location is represented via a Register, not | |
| 142 // via an address for efficiency & simplicity reasons. | |
| 143 | |
| 144 class ArrayAddress; | |
| 145 | |
| 146 class Address VALUE_OBJ_CLASS_SPEC { | |
| 147 public: | |
| 148 enum ScaleFactor { | |
| 149 no_scale = -1, | |
| 150 times_1 = 0, | |
| 151 times_2 = 1, | |
| 152 times_4 = 2, | |
| 304 | 153 times_8 = 3, |
| 154 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) | |
| 0 | 155 }; |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
156 static ScaleFactor times(int size) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
157 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size"); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
158 if (size == 8) return times_8; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
159 if (size == 4) return times_4; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
160 if (size == 2) return times_2; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
161 return times_1; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
162 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
163 static int scale_size(ScaleFactor scale) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
164 assert(scale != no_scale, ""); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
165 assert(((1 << (int)times_1) == 1 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
166 (1 << (int)times_2) == 2 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
167 (1 << (int)times_4) == 4 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
168 (1 << (int)times_8) == 8), ""); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
169 return (1 << (int)scale); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
170 } |
| 0 | 171 |
| 172 private: | |
| 173 Register _base; | |
| 174 Register _index; | |
| 175 ScaleFactor _scale; | |
| 176 int _disp; | |
| 177 RelocationHolder _rspec; | |
| 178 | |
| 304 | 179 // Easily misused constructors make them private |
| 180 // %%% can we make these go away? | |
| 181 NOT_LP64(Address(address loc, RelocationHolder spec);) | |
| 182 Address(int disp, address loc, relocInfo::relocType rtype); | |
| 183 Address(int disp, address loc, RelocationHolder spec); | |
| 0 | 184 |
| 185 public: | |
| 304 | 186 |
| 187 int disp() { return _disp; } | |
| 0 | 188 // creation |
| 189 Address() | |
| 190 : _base(noreg), | |
| 191 _index(noreg), | |
| 192 _scale(no_scale), | |
| 193 _disp(0) { | |
| 194 } | |
| 195 | |
| 196 // No default displacement otherwise Register can be implicitly | |
| 197 // converted to 0(Register) which is quite a different animal. | |
| 198 | |
| 199 Address(Register base, int disp) | |
| 200 : _base(base), | |
| 201 _index(noreg), | |
| 202 _scale(no_scale), | |
| 203 _disp(disp) { | |
| 204 } | |
| 205 | |
| 206 Address(Register base, Register index, ScaleFactor scale, int disp = 0) | |
| 207 : _base (base), | |
| 208 _index(index), | |
| 209 _scale(scale), | |
| 210 _disp (disp) { | |
| 211 assert(!index->is_valid() == (scale == Address::no_scale), | |
| 212 "inconsistent address"); | |
| 213 } | |
| 214 | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
215 Address(Register base, RegisterConstant index, ScaleFactor scale = times_1, int disp = 0) |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
216 : _base (base), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
217 _index(index.register_or_noreg()), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
218 _scale(scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
219 _disp (disp + (index.constant_or_zero() * scale_size(scale))) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
220 if (!index.is_register()) scale = Address::no_scale; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
221 assert(!_index->is_valid() == (scale == Address::no_scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
222 "inconsistent address"); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
223 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
224 |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
225 Address plus_disp(int disp) const { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
226 Address a = (*this); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
227 a._disp += disp; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
228 return a; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
229 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
230 |
| 0 | 231 // The following two overloads are used in connection with the |
| 232 // ByteSize type (see sizes.hpp). They simplify the use of | |
| 233 // ByteSize'd arguments in assembly code. Note that their equivalent | |
| 234 // for the optimized build are the member functions with int disp | |
| 235 // argument since ByteSize is mapped to an int type in that case. | |
| 236 // | |
| 237 // Note: DO NOT introduce similar overloaded functions for WordSize | |
| 238 // arguments as in the optimized mode, both ByteSize and WordSize | |
| 239 // are mapped to the same type and thus the compiler cannot make a | |
| 240 // distinction anymore (=> compiler errors). | |
| 241 | |
| 242 #ifdef ASSERT | |
| 243 Address(Register base, ByteSize disp) | |
| 244 : _base(base), | |
| 245 _index(noreg), | |
| 246 _scale(no_scale), | |
| 247 _disp(in_bytes(disp)) { | |
| 248 } | |
| 249 | |
| 250 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) | |
| 251 : _base(base), | |
| 252 _index(index), | |
| 253 _scale(scale), | |
| 254 _disp(in_bytes(disp)) { | |
| 255 assert(!index->is_valid() == (scale == Address::no_scale), | |
| 256 "inconsistent address"); | |
| 257 } | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
258 |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
259 Address(Register base, RegisterConstant index, ScaleFactor scale, ByteSize disp) |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
260 : _base (base), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
261 _index(index.register_or_noreg()), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
262 _scale(scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
263 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
264 if (!index.is_register()) scale = Address::no_scale; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
265 assert(!_index->is_valid() == (scale == Address::no_scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
266 "inconsistent address"); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
267 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
268 |
| 0 | 269 #endif // ASSERT |
| 270 | |
| 271 // accessors | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
272 bool uses(Register reg) const { return _base == reg || _index == reg; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
273 Register base() const { return _base; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
274 Register index() const { return _index; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
275 ScaleFactor scale() const { return _scale; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
276 int disp() const { return _disp; } |
| 0 | 277 |
| 278 // Convert the raw encoding form into the form expected by the constructor for | |
| 279 // Address. An index of 4 (rsp) corresponds to having no index, so convert | |
| 280 // that to noreg for the Address constructor. | |
| 624 | 281 static Address make_raw(int base, int index, int scale, int disp, bool disp_is_oop); |
| 0 | 282 |
| 283 static Address make_array(ArrayAddress); | |
| 284 | |
| 285 private: | |
| 286 bool base_needs_rex() const { | |
| 287 return _base != noreg && _base->encoding() >= 8; | |
| 288 } | |
| 289 | |
| 290 bool index_needs_rex() const { | |
| 291 return _index != noreg &&_index->encoding() >= 8; | |
| 292 } | |
| 293 | |
| 294 relocInfo::relocType reloc() const { return _rspec.type(); } | |
| 295 | |
| 296 friend class Assembler; | |
| 297 friend class MacroAssembler; | |
| 298 friend class LIR_Assembler; // base/index/scale/disp | |
| 299 }; | |
| 300 | |
| 301 // | |
| 302 // AddressLiteral has been split out from Address because operands of this type | |
| 303 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out | |
| 304 // the few instructions that need to deal with address literals are unique and the | |
| 305 // MacroAssembler does not have to implement every instruction in the Assembler | |
| 306 // in order to search for address literals that may need special handling depending | |
| 307 // on the instruction and the platform. As small step on the way to merging i486/amd64 | |
| 308 // directories. | |
| 309 // | |
| 310 class AddressLiteral VALUE_OBJ_CLASS_SPEC { | |
| 311 friend class ArrayAddress; | |
| 312 RelocationHolder _rspec; | |
| 313 // Typically we use AddressLiterals we want to use their rval | |
| 314 // However in some situations we want the lval (effect address) of the item. | |
| 315 // We provide a special factory for making those lvals. | |
| 316 bool _is_lval; | |
| 317 | |
| 318 // If the target is far we'll need to load the ea of this to | |
| 319 // a register to reach it. Otherwise if near we can do rip | |
| 320 // relative addressing. | |
| 321 | |
| 322 address _target; | |
| 323 | |
| 324 protected: | |
| 325 // creation | |
| 326 AddressLiteral() | |
| 327 : _is_lval(false), | |
| 328 _target(NULL) | |
| 329 {} | |
| 330 | |
| 331 public: | |
| 332 | |
| 333 | |
| 334 AddressLiteral(address target, relocInfo::relocType rtype); | |
| 335 | |
| 336 AddressLiteral(address target, RelocationHolder const& rspec) | |
| 337 : _rspec(rspec), | |
| 338 _is_lval(false), | |
| 339 _target(target) | |
| 340 {} | |
| 341 | |
| 342 AddressLiteral addr() { | |
| 343 AddressLiteral ret = *this; | |
| 344 ret._is_lval = true; | |
| 345 return ret; | |
| 346 } | |
| 347 | |
| 348 | |
| 349 private: | |
| 350 | |
| 351 address target() { return _target; } | |
| 352 bool is_lval() { return _is_lval; } | |
| 353 | |
| 354 relocInfo::relocType reloc() const { return _rspec.type(); } | |
| 355 const RelocationHolder& rspec() const { return _rspec; } | |
| 356 | |
| 357 friend class Assembler; | |
| 358 friend class MacroAssembler; | |
| 359 friend class Address; | |
| 360 friend class LIR_Assembler; | |
| 361 }; | |
| 362 | |
| 363 // Convience classes | |
| 364 class RuntimeAddress: public AddressLiteral { | |
| 365 | |
| 366 public: | |
| 367 | |
| 368 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} | |
| 369 | |
| 370 }; | |
| 371 | |
| 372 class OopAddress: public AddressLiteral { | |
| 373 | |
| 374 public: | |
| 375 | |
| 376 OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){} | |
| 377 | |
| 378 }; | |
| 379 | |
| 380 class ExternalAddress: public AddressLiteral { | |
| 381 | |
| 382 public: | |
| 383 | |
| 384 ExternalAddress(address target) : AddressLiteral(target, relocInfo::external_word_type){} | |
| 385 | |
| 386 }; | |
| 387 | |
| 388 class InternalAddress: public AddressLiteral { | |
| 389 | |
| 390 public: | |
| 391 | |
| 392 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} | |
| 393 | |
| 394 }; | |
| 395 | |
| 396 // x86 can do array addressing as a single operation since disp can be an absolute | |
| 397 // address amd64 can't. We create a class that expresses the concept but does extra | |
| 398 // magic on amd64 to get the final result | |
| 399 | |
| 400 class ArrayAddress VALUE_OBJ_CLASS_SPEC { | |
| 401 private: | |
| 402 | |
| 403 AddressLiteral _base; | |
| 404 Address _index; | |
| 405 | |
| 406 public: | |
| 407 | |
| 408 ArrayAddress() {}; | |
| 409 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; | |
| 410 AddressLiteral base() { return _base; } | |
| 411 Address index() { return _index; } | |
| 412 | |
| 413 }; | |
| 414 | |
| 304 | 415 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); |
| 0 | 416 |
| 417 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction | |
| 418 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write | |
| 419 // is what you get. The Assembler is generating code into a CodeBuffer. | |
| 420 | |
| 421 class Assembler : public AbstractAssembler { | |
| 422 friend class AbstractAssembler; // for the non-virtual hack | |
| 423 friend class LIR_Assembler; // as_Address() | |
| 304 | 424 friend class StubGenerator; |
| 0 | 425 |
| 426 public: | |
| 427 enum Condition { // The x86 condition codes used for conditional jumps/moves. | |
| 428 zero = 0x4, | |
| 429 notZero = 0x5, | |
| 430 equal = 0x4, | |
| 431 notEqual = 0x5, | |
| 432 less = 0xc, | |
| 433 lessEqual = 0xe, | |
| 434 greater = 0xf, | |
| 435 greaterEqual = 0xd, | |
| 436 below = 0x2, | |
| 437 belowEqual = 0x6, | |
| 438 above = 0x7, | |
| 439 aboveEqual = 0x3, | |
| 440 overflow = 0x0, | |
| 441 noOverflow = 0x1, | |
| 442 carrySet = 0x2, | |
| 443 carryClear = 0x3, | |
| 444 negative = 0x8, | |
| 445 positive = 0x9, | |
| 446 parity = 0xa, | |
| 447 noParity = 0xb | |
| 448 }; | |
| 449 | |
| 450 enum Prefix { | |
| 451 // segment overrides | |
| 452 CS_segment = 0x2e, | |
| 453 SS_segment = 0x36, | |
| 454 DS_segment = 0x3e, | |
| 455 ES_segment = 0x26, | |
| 456 FS_segment = 0x64, | |
| 457 GS_segment = 0x65, | |
| 458 | |
| 459 REX = 0x40, | |
| 460 | |
| 461 REX_B = 0x41, | |
| 462 REX_X = 0x42, | |
| 463 REX_XB = 0x43, | |
| 464 REX_R = 0x44, | |
| 465 REX_RB = 0x45, | |
| 466 REX_RX = 0x46, | |
| 467 REX_RXB = 0x47, | |
| 468 | |
| 469 REX_W = 0x48, | |
| 470 | |
| 471 REX_WB = 0x49, | |
| 472 REX_WX = 0x4A, | |
| 473 REX_WXB = 0x4B, | |
| 474 REX_WR = 0x4C, | |
| 475 REX_WRB = 0x4D, | |
| 476 REX_WRX = 0x4E, | |
| 477 REX_WRXB = 0x4F | |
| 478 }; | |
| 479 | |
| 480 enum WhichOperand { | |
| 481 // input to locate_operand, and format code for relocations | |
| 304 | 482 imm_operand = 0, // embedded 32-bit|64-bit immediate operand |
| 0 | 483 disp32_operand = 1, // embedded 32-bit displacement or address |
| 484 call32_operand = 2, // embedded 32-bit self-relative displacement | |
| 304 | 485 #ifndef _LP64 |
| 0 | 486 _WhichOperand_limit = 3 |
| 304 | 487 #else |
| 488 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop | |
| 489 _WhichOperand_limit = 4 | |
| 490 #endif | |
| 0 | 491 }; |
| 492 | |
| 304 | 493 |
| 494 | |
| 495 // NOTE: The general philopsophy of the declarations here is that 64bit versions | |
| 496 // of instructions are freely declared without the need for wrapping them an ifdef. | |
| 497 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) | |
| 498 // In the .cpp file the implementations are wrapped so that they are dropped out | |
| 499 // of the resulting jvm. This is done mostly to keep the footprint of KERNEL | |
| 500 // to the size it was prior to merging up the 32bit and 64bit assemblers. | |
| 501 // | |
| 502 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction | |
| 503 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. | |
| 504 | |
| 505 private: | |
| 506 | |
| 507 | |
| 508 // 64bit prefixes | |
| 509 int prefix_and_encode(int reg_enc, bool byteinst = false); | |
| 510 int prefixq_and_encode(int reg_enc); | |
| 511 | |
| 512 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); | |
| 513 int prefixq_and_encode(int dst_enc, int src_enc); | |
| 514 | |
| 515 void prefix(Register reg); | |
| 516 void prefix(Address adr); | |
| 517 void prefixq(Address adr); | |
| 518 | |
| 519 void prefix(Address adr, Register reg, bool byteinst = false); | |
| 520 void prefixq(Address adr, Register reg); | |
| 521 | |
| 522 void prefix(Address adr, XMMRegister reg); | |
| 523 | |
| 524 void prefetch_prefix(Address src); | |
| 525 | |
| 526 // Helper functions for groups of instructions | |
| 527 void emit_arith_b(int op1, int op2, Register dst, int imm8); | |
| 528 | |
| 529 void emit_arith(int op1, int op2, Register dst, int32_t imm32); | |
| 530 // only 32bit?? | |
| 531 void emit_arith(int op1, int op2, Register dst, jobject obj); | |
| 532 void emit_arith(int op1, int op2, Register dst, Register src); | |
| 533 | |
| 534 void emit_operand(Register reg, | |
| 535 Register base, Register index, Address::ScaleFactor scale, | |
| 536 int disp, | |
| 537 RelocationHolder const& rspec, | |
| 538 int rip_relative_correction = 0); | |
| 539 | |
| 540 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); | |
| 541 | |
| 542 // operands that only take the original 32bit registers | |
| 543 void emit_operand32(Register reg, Address adr); | |
| 544 | |
| 545 void emit_operand(XMMRegister reg, | |
| 546 Register base, Register index, Address::ScaleFactor scale, | |
| 547 int disp, | |
| 548 RelocationHolder const& rspec); | |
| 549 | |
| 550 void emit_operand(XMMRegister reg, Address adr); | |
| 551 | |
| 552 void emit_operand(MMXRegister reg, Address adr); | |
| 553 | |
| 554 // workaround gcc (3.2.1-7) bug | |
| 555 void emit_operand(Address adr, MMXRegister reg); | |
| 556 | |
| 557 | |
| 558 // Immediate-to-memory forms | |
| 559 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); | |
| 560 | |
| 561 void emit_farith(int b1, int b2, int i); | |
| 562 | |
| 563 | |
| 564 protected: | |
| 565 #ifdef ASSERT | |
| 566 void check_relocation(RelocationHolder const& rspec, int format); | |
| 567 #endif | |
| 568 | |
| 569 inline void emit_long64(jlong x); | |
| 570 | |
| 571 void emit_data(jint data, relocInfo::relocType rtype, int format); | |
| 572 void emit_data(jint data, RelocationHolder const& rspec, int format); | |
| 573 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); | |
| 574 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); | |
| 575 | |
| 576 | |
| 577 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); | |
| 578 | |
| 579 // These are all easily abused and hence protected | |
| 580 | |
| 581 // 32BIT ONLY SECTION | |
| 582 #ifndef _LP64 | |
| 583 // Make these disappear in 64bit mode since they would never be correct | |
| 584 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 585 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 586 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
587 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY |
| 304 | 588 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY |
| 589 | |
| 590 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 591 #else | |
| 592 // 64BIT ONLY SECTION | |
| 593 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
594 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
595 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
596 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
597 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
598 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
599 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); |
| 304 | 600 #endif // _LP64 |
| 601 | |
| 602 // These are unique in that we are ensured by the caller that the 32bit | |
| 603 // relative in these instructions will always be able to reach the potentially | |
| 604 // 64bit address described by entry. Since they can take a 64bit address they | |
| 605 // don't have the 32 suffix like the other instructions in this class. | |
| 606 | |
| 607 void call_literal(address entry, RelocationHolder const& rspec); | |
| 608 void jmp_literal(address entry, RelocationHolder const& rspec); | |
| 609 | |
| 610 // Avoid using directly section | |
| 611 // Instructions in this section are actually usable by anyone without danger | |
| 612 // of failure but have performance issues that are addressed my enhanced | |
| 613 // instructions which will do the proper thing base on the particular cpu. | |
| 614 // We protect them because we don't trust you... | |
| 615 | |
| 616 // Don't use next inc() and dec() methods directly. INC & DEC instructions | |
| 617 // could cause a partial flag stall since they don't set CF flag. | |
| 618 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods | |
| 619 // which call inc() & dec() or add() & sub() in accordance with | |
| 620 // the product flag UseIncDec value. | |
| 621 | |
| 622 void decl(Register dst); | |
| 623 void decl(Address dst); | |
| 624 void decq(Register dst); | |
| 625 void decq(Address dst); | |
| 626 | |
| 627 void incl(Register dst); | |
| 628 void incl(Address dst); | |
| 629 void incq(Register dst); | |
| 630 void incq(Address dst); | |
| 631 | |
| 632 // New cpus require use of movsd and movss to avoid partial register stall | |
| 633 // when loading from memory. But for old Opteron use movlpd instead of movsd. | |
| 634 // The selection is done in MacroAssembler::movdbl() and movflt(). | |
| 635 | |
| 636 // Move Scalar Single-Precision Floating-Point Values | |
| 637 void movss(XMMRegister dst, Address src); | |
| 638 void movss(XMMRegister dst, XMMRegister src); | |
| 639 void movss(Address dst, XMMRegister src); | |
| 640 | |
| 641 // Move Scalar Double-Precision Floating-Point Values | |
| 642 void movsd(XMMRegister dst, Address src); | |
| 643 void movsd(XMMRegister dst, XMMRegister src); | |
| 644 void movsd(Address dst, XMMRegister src); | |
| 645 void movlpd(XMMRegister dst, Address src); | |
| 646 | |
| 647 // New cpus require use of movaps and movapd to avoid partial register stall | |
| 648 // when moving between registers. | |
| 649 void movaps(XMMRegister dst, XMMRegister src); | |
| 650 void movapd(XMMRegister dst, XMMRegister src); | |
| 651 | |
| 652 // End avoid using directly | |
| 653 | |
| 654 | |
| 655 // Instruction prefixes | |
| 656 void prefix(Prefix p); | |
| 657 | |
| 0 | 658 public: |
| 659 | |
| 660 // Creation | |
| 661 Assembler(CodeBuffer* code) : AbstractAssembler(code) {} | |
| 662 | |
| 663 // Decoding | |
| 664 static address locate_operand(address inst, WhichOperand which); | |
| 665 static address locate_next_instruction(address inst); | |
| 666 | |
| 304 | 667 // Utilities |
| 668 | |
| 669 #ifdef _LP64 | |
| 670 static bool is_simm(int64_t x, int nbits) { return -( CONST64(1) << (nbits-1) ) <= x && x < ( CONST64(1) << (nbits-1) ); } | |
| 671 static bool is_simm32(int64_t x) { return x == (int64_t)(int32_t)x; } | |
| 672 #else | |
| 673 static bool is_simm(int32_t x, int nbits) { return -( 1 << (nbits-1) ) <= x && x < ( 1 << (nbits-1) ); } | |
| 674 static bool is_simm32(int32_t x) { return true; } | |
| 675 #endif // LP64 | |
| 676 | |
| 677 // Generic instructions | |
| 678 // Does 32bit or 64bit as needed for the platform. In some sense these | |
| 679 // belong in macro assembler but there is no need for both varieties to exist | |
| 680 | |
| 681 void lea(Register dst, Address src); | |
| 682 | |
| 683 void mov(Register dst, Register src); | |
| 684 | |
| 685 void pusha(); | |
| 686 void popa(); | |
| 687 | |
| 688 void pushf(); | |
| 689 void popf(); | |
| 690 | |
| 691 void push(int32_t imm32); | |
| 692 | |
| 693 void push(Register src); | |
| 694 | |
| 695 void pop(Register dst); | |
| 696 | |
| 697 // These are dummies to prevent surprise implicit conversions to Register | |
| 698 void push(void* v); | |
| 699 void pop(void* v); | |
| 700 | |
| 701 | |
| 702 // These do register sized moves/scans | |
| 703 void rep_mov(); | |
| 704 void rep_set(); | |
| 705 void repne_scan(); | |
| 706 #ifdef _LP64 | |
| 707 void repne_scanl(); | |
| 708 #endif | |
| 709 | |
| 710 // Vanilla instructions in lexical order | |
| 711 | |
| 712 void adcl(Register dst, int32_t imm32); | |
| 0 | 713 void adcl(Register dst, Address src); |
| 714 void adcl(Register dst, Register src); | |
| 715 | |
| 304 | 716 void adcq(Register dst, int32_t imm32); |
| 717 void adcq(Register dst, Address src); | |
| 718 void adcq(Register dst, Register src); | |
| 719 | |
| 720 | |
| 721 void addl(Address dst, int32_t imm32); | |
| 0 | 722 void addl(Address dst, Register src); |
| 304 | 723 void addl(Register dst, int32_t imm32); |
| 0 | 724 void addl(Register dst, Address src); |
| 725 void addl(Register dst, Register src); | |
| 726 | |
| 304 | 727 void addq(Address dst, int32_t imm32); |
| 728 void addq(Address dst, Register src); | |
| 729 void addq(Register dst, int32_t imm32); | |
| 730 void addq(Register dst, Address src); | |
| 731 void addq(Register dst, Register src); | |
| 732 | |
| 733 | |
| 0 | 734 void addr_nop_4(); |
| 735 void addr_nop_5(); | |
| 736 void addr_nop_7(); | |
| 737 void addr_nop_8(); | |
| 738 | |
| 304 | 739 // Add Scalar Double-Precision Floating-Point Values |
| 740 void addsd(XMMRegister dst, Address src); | |
| 741 void addsd(XMMRegister dst, XMMRegister src); | |
| 742 | |
| 743 // Add Scalar Single-Precision Floating-Point Values | |
| 744 void addss(XMMRegister dst, Address src); | |
| 745 void addss(XMMRegister dst, XMMRegister src); | |
| 746 | |
| 747 void andl(Register dst, int32_t imm32); | |
| 748 void andl(Register dst, Address src); | |
| 749 void andl(Register dst, Register src); | |
| 750 | |
| 751 void andq(Register dst, int32_t imm32); | |
| 752 void andq(Register dst, Address src); | |
| 753 void andq(Register dst, Register src); | |
| 754 | |
| 755 | |
| 756 // Bitwise Logical AND of Packed Double-Precision Floating-Point Values | |
| 757 void andpd(XMMRegister dst, Address src); | |
| 758 void andpd(XMMRegister dst, XMMRegister src); | |
| 759 | |
| 760 void bswapl(Register reg); | |
| 761 | |
| 762 void bswapq(Register reg); | |
| 763 | |
| 0 | 764 void call(Label& L, relocInfo::relocType rtype); |
| 765 void call(Register reg); // push pc; pc <- reg | |
| 766 void call(Address adr); // push pc; pc <- adr | |
| 767 | |
| 304 | 768 void cdql(); |
| 769 | |
| 770 void cdqq(); | |
| 771 | |
| 772 void cld() { emit_byte(0xfc); } | |
| 773 | |
| 774 void clflush(Address adr); | |
| 775 | |
| 776 void cmovl(Condition cc, Register dst, Register src); | |
| 777 void cmovl(Condition cc, Register dst, Address src); | |
| 778 | |
| 779 void cmovq(Condition cc, Register dst, Register src); | |
| 780 void cmovq(Condition cc, Register dst, Address src); | |
| 781 | |
| 782 | |
| 783 void cmpb(Address dst, int imm8); | |
| 784 | |
| 785 void cmpl(Address dst, int32_t imm32); | |
| 786 | |
| 787 void cmpl(Register dst, int32_t imm32); | |
| 788 void cmpl(Register dst, Register src); | |
| 789 void cmpl(Register dst, Address src); | |
| 790 | |
| 791 void cmpq(Address dst, int32_t imm32); | |
| 792 void cmpq(Address dst, Register src); | |
| 793 | |
| 794 void cmpq(Register dst, int32_t imm32); | |
| 795 void cmpq(Register dst, Register src); | |
| 796 void cmpq(Register dst, Address src); | |
| 797 | |
| 798 // these are dummies used to catch attempting to convert NULL to Register | |
| 799 void cmpl(Register dst, void* junk); // dummy | |
| 800 void cmpq(Register dst, void* junk); // dummy | |
| 801 | |
| 802 void cmpw(Address dst, int imm16); | |
| 803 | |
| 804 void cmpxchg8 (Address adr); | |
| 805 | |
| 806 void cmpxchgl(Register reg, Address adr); | |
| 807 | |
| 808 void cmpxchgq(Register reg, Address adr); | |
| 809 | |
| 810 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS | |
| 811 void comisd(XMMRegister dst, Address src); | |
| 812 | |
| 813 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS | |
| 814 void comiss(XMMRegister dst, Address src); | |
| 815 | |
| 816 // Identify processor type and features | |
| 817 void cpuid() { | |
| 818 emit_byte(0x0F); | |
| 819 emit_byte(0xA2); | |
| 820 } | |
| 821 | |
| 822 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value | |
| 823 void cvtsd2ss(XMMRegister dst, XMMRegister src); | |
| 824 | |
| 825 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value | |
| 826 void cvtsi2sdl(XMMRegister dst, Register src); | |
| 827 void cvtsi2sdq(XMMRegister dst, Register src); | |
| 828 | |
| 829 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value | |
| 830 void cvtsi2ssl(XMMRegister dst, Register src); | |
| 831 void cvtsi2ssq(XMMRegister dst, Register src); | |
| 832 | |
| 833 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value | |
| 834 void cvtdq2pd(XMMRegister dst, XMMRegister src); | |
| 835 | |
| 836 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value | |
| 837 void cvtdq2ps(XMMRegister dst, XMMRegister src); | |
| 838 | |
| 839 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value | |
| 840 void cvtss2sd(XMMRegister dst, XMMRegister src); | |
| 841 | |
| 842 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer | |
| 843 void cvttsd2sil(Register dst, Address src); | |
| 844 void cvttsd2sil(Register dst, XMMRegister src); | |
| 845 void cvttsd2siq(Register dst, XMMRegister src); | |
| 846 | |
| 847 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer | |
| 848 void cvttss2sil(Register dst, XMMRegister src); | |
| 849 void cvttss2siq(Register dst, XMMRegister src); | |
| 850 | |
| 851 // Divide Scalar Double-Precision Floating-Point Values | |
| 852 void divsd(XMMRegister dst, Address src); | |
| 853 void divsd(XMMRegister dst, XMMRegister src); | |
| 854 | |
| 855 // Divide Scalar Single-Precision Floating-Point Values | |
| 856 void divss(XMMRegister dst, Address src); | |
| 857 void divss(XMMRegister dst, XMMRegister src); | |
| 858 | |
| 859 void emms(); | |
| 860 | |
| 861 void fabs(); | |
| 862 | |
| 863 void fadd(int i); | |
| 864 | |
| 865 void fadd_d(Address src); | |
| 866 void fadd_s(Address src); | |
| 867 | |
| 868 // "Alternate" versions of x87 instructions place result down in FPU | |
| 869 // stack instead of on TOS | |
| 870 | |
| 871 void fadda(int i); // "alternate" fadd | |
| 872 void faddp(int i = 1); | |
| 873 | |
| 874 void fchs(); | |
| 875 | |
| 876 void fcom(int i); | |
| 877 | |
| 878 void fcomp(int i = 1); | |
| 879 void fcomp_d(Address src); | |
| 880 void fcomp_s(Address src); | |
| 881 | |
| 882 void fcompp(); | |
| 883 | |
| 884 void fcos(); | |
| 885 | |
| 886 void fdecstp(); | |
| 887 | |
| 888 void fdiv(int i); | |
| 889 void fdiv_d(Address src); | |
| 890 void fdivr_s(Address src); | |
| 891 void fdiva(int i); // "alternate" fdiv | |
| 892 void fdivp(int i = 1); | |
| 893 | |
| 894 void fdivr(int i); | |
| 895 void fdivr_d(Address src); | |
| 896 void fdiv_s(Address src); | |
| 897 | |
| 898 void fdivra(int i); // "alternate" reversed fdiv | |
| 899 | |
| 900 void fdivrp(int i = 1); | |
| 901 | |
| 902 void ffree(int i = 0); | |
| 903 | |
| 904 void fild_d(Address adr); | |
| 905 void fild_s(Address adr); | |
| 906 | |
| 907 void fincstp(); | |
| 908 | |
| 909 void finit(); | |
| 910 | |
| 911 void fist_s (Address adr); | |
| 912 void fistp_d(Address adr); | |
| 913 void fistp_s(Address adr); | |
| 914 | |
| 915 void fld1(); | |
| 916 | |
| 917 void fld_d(Address adr); | |
| 918 void fld_s(Address adr); | |
| 919 void fld_s(int index); | |
| 920 void fld_x(Address adr); // extended-precision (80-bit) format | |
| 921 | |
| 922 void fldcw(Address src); | |
| 923 | |
| 924 void fldenv(Address src); | |
| 925 | |
| 926 void fldlg2(); | |
| 927 | |
| 928 void fldln2(); | |
| 929 | |
| 930 void fldz(); | |
| 931 | |
| 932 void flog(); | |
| 933 void flog10(); | |
| 934 | |
| 935 void fmul(int i); | |
| 936 | |
| 937 void fmul_d(Address src); | |
| 938 void fmul_s(Address src); | |
| 939 | |
| 940 void fmula(int i); // "alternate" fmul | |
| 941 | |
| 942 void fmulp(int i = 1); | |
| 943 | |
| 944 void fnsave(Address dst); | |
| 945 | |
| 946 void fnstcw(Address src); | |
| 947 | |
| 948 void fnstsw_ax(); | |
| 949 | |
| 950 void fprem(); | |
| 951 void fprem1(); | |
| 952 | |
| 953 void frstor(Address src); | |
| 954 | |
| 955 void fsin(); | |
| 956 | |
| 957 void fsqrt(); | |
| 958 | |
| 959 void fst_d(Address adr); | |
| 960 void fst_s(Address adr); | |
| 961 | |
| 962 void fstp_d(Address adr); | |
| 963 void fstp_d(int index); | |
| 964 void fstp_s(Address adr); | |
| 965 void fstp_x(Address adr); // extended-precision (80-bit) format | |
| 966 | |
| 967 void fsub(int i); | |
| 968 void fsub_d(Address src); | |
| 969 void fsub_s(Address src); | |
| 970 | |
| 971 void fsuba(int i); // "alternate" fsub | |
| 972 | |
| 973 void fsubp(int i = 1); | |
| 974 | |
| 975 void fsubr(int i); | |
| 976 void fsubr_d(Address src); | |
| 977 void fsubr_s(Address src); | |
| 978 | |
| 979 void fsubra(int i); // "alternate" reversed fsub | |
| 980 | |
| 981 void fsubrp(int i = 1); | |
| 982 | |
| 983 void ftan(); | |
| 984 | |
| 985 void ftst(); | |
| 986 | |
| 987 void fucomi(int i = 1); | |
| 988 void fucomip(int i = 1); | |
| 989 | |
| 990 void fwait(); | |
| 991 | |
| 992 void fxch(int i = 1); | |
| 993 | |
| 994 void fxrstor(Address src); | |
| 995 | |
| 996 void fxsave(Address dst); | |
| 997 | |
| 998 void fyl2x(); | |
| 999 | |
| 1000 void hlt(); | |
| 1001 | |
| 1002 void idivl(Register src); | |
| 1003 | |
| 1004 void idivq(Register src); | |
| 1005 | |
| 1006 void imull(Register dst, Register src); | |
| 1007 void imull(Register dst, Register src, int value); | |
| 1008 | |
| 1009 void imulq(Register dst, Register src); | |
| 1010 void imulq(Register dst, Register src, int value); | |
| 1011 | |
| 0 | 1012 |
| 1013 // jcc is the generic conditional branch generator to run- | |
| 1014 // time routines, jcc is used for branches to labels. jcc | |
| 1015 // takes a branch opcode (cc) and a label (L) and generates | |
| 1016 // either a backward branch or a forward branch and links it | |
| 1017 // to the label fixup chain. Usage: | |
| 1018 // | |
| 1019 // Label L; // unbound label | |
| 1020 // jcc(cc, L); // forward branch to unbound label | |
| 1021 // bind(L); // bind label to the current pc | |
| 1022 // jcc(cc, L); // backward branch to bound label | |
| 1023 // bind(L); // illegal: a label may be bound only once | |
| 1024 // | |
| 1025 // Note: The same Label can be used for forward and backward branches | |
| 1026 // but it may be bound only once. | |
| 1027 | |
| 1028 void jcc(Condition cc, Label& L, | |
| 1029 relocInfo::relocType rtype = relocInfo::none); | |
| 1030 | |
| 1031 // Conditional jump to a 8-bit offset to L. | |
| 1032 // WARNING: be very careful using this for forward jumps. If the label is | |
| 1033 // not bound within an 8-bit offset of this instruction, a run-time error | |
| 1034 // will occur. | |
| 1035 void jccb(Condition cc, Label& L); | |
| 1036 | |
| 304 | 1037 void jmp(Address entry); // pc <- entry |
| 1038 | |
| 1039 // Label operations & relative jumps (PPUM Appendix D) | |
| 1040 void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none); // unconditional jump to L | |
| 1041 | |
| 1042 void jmp(Register entry); // pc <- entry | |
| 1043 | |
| 1044 // Unconditional 8-bit offset jump to L. | |
| 1045 // WARNING: be very careful using this for forward jumps. If the label is | |
| 1046 // not bound within an 8-bit offset of this instruction, a run-time error | |
| 1047 // will occur. | |
| 1048 void jmpb(Label& L); | |
| 1049 | |
| 1050 void ldmxcsr( Address src ); | |
| 1051 | |
| 1052 void leal(Register dst, Address src); | |
| 1053 | |
| 1054 void leaq(Register dst, Address src); | |
| 1055 | |
| 1056 void lfence() { | |
| 1057 emit_byte(0x0F); | |
| 1058 emit_byte(0xAE); | |
| 1059 emit_byte(0xE8); | |
| 1060 } | |
| 1061 | |
| 1062 void lock(); | |
| 1063 | |
| 1064 enum Membar_mask_bits { | |
| 1065 StoreStore = 1 << 3, | |
| 1066 LoadStore = 1 << 2, | |
| 1067 StoreLoad = 1 << 1, | |
| 1068 LoadLoad = 1 << 0 | |
| 1069 }; | |
| 1070 | |
| 1071 // Serializes memory. | |
| 1072 void membar(Membar_mask_bits order_constraint) { | |
| 1073 // We only have to handle StoreLoad and LoadLoad | |
| 1074 if (order_constraint & StoreLoad) { | |
| 1075 // MFENCE subsumes LFENCE | |
| 1076 mfence(); | |
| 1077 } /* [jk] not needed currently: else if (order_constraint & LoadLoad) { | |
| 1078 lfence(); | |
| 1079 } */ | |
| 1080 } | |
| 1081 | |
| 1082 void mfence(); | |
| 1083 | |
| 1084 // Moves | |
| 1085 | |
| 1086 void mov64(Register dst, int64_t imm64); | |
| 1087 | |
| 1088 void movb(Address dst, Register src); | |
| 1089 void movb(Address dst, int imm8); | |
| 1090 void movb(Register dst, Address src); | |
| 1091 | |
| 1092 void movdl(XMMRegister dst, Register src); | |
| 1093 void movdl(Register dst, XMMRegister src); | |
| 1094 | |
| 1095 // Move Double Quadword | |
| 1096 void movdq(XMMRegister dst, Register src); | |
| 1097 void movdq(Register dst, XMMRegister src); | |
| 1098 | |
| 1099 // Move Aligned Double Quadword | |
| 1100 void movdqa(Address dst, XMMRegister src); | |
| 1101 void movdqa(XMMRegister dst, Address src); | |
| 1102 void movdqa(XMMRegister dst, XMMRegister src); | |
| 1103 | |
| 405 | 1104 // Move Unaligned Double Quadword |
| 1105 void movdqu(Address dst, XMMRegister src); | |
| 1106 void movdqu(XMMRegister dst, Address src); | |
| 1107 void movdqu(XMMRegister dst, XMMRegister src); | |
| 1108 | |
| 304 | 1109 void movl(Register dst, int32_t imm32); |
| 1110 void movl(Address dst, int32_t imm32); | |
| 1111 void movl(Register dst, Register src); | |
| 1112 void movl(Register dst, Address src); | |
| 1113 void movl(Address dst, Register src); | |
| 1114 | |
| 1115 // These dummies prevent using movl from converting a zero (like NULL) into Register | |
| 1116 // by giving the compiler two choices it can't resolve | |
| 1117 | |
| 1118 void movl(Address dst, void* junk); | |
| 1119 void movl(Register dst, void* junk); | |
| 1120 | |
| 1121 #ifdef _LP64 | |
| 1122 void movq(Register dst, Register src); | |
| 1123 void movq(Register dst, Address src); | |
| 1124 void movq(Address dst, Register src); | |
| 1125 #endif | |
| 1126 | |
| 1127 void movq(Address dst, MMXRegister src ); | |
| 1128 void movq(MMXRegister dst, Address src ); | |
| 1129 | |
| 1130 #ifdef _LP64 | |
| 1131 // These dummies prevent using movq from converting a zero (like NULL) into Register | |
| 1132 // by giving the compiler two choices it can't resolve | |
| 1133 | |
| 1134 void movq(Address dst, void* dummy); | |
| 1135 void movq(Register dst, void* dummy); | |
| 1136 #endif | |
| 1137 | |
| 1138 // Move Quadword | |
| 1139 void movq(Address dst, XMMRegister src); | |
| 1140 void movq(XMMRegister dst, Address src); | |
| 1141 | |
| 1142 void movsbl(Register dst, Address src); | |
| 1143 void movsbl(Register dst, Register src); | |
| 1144 | |
| 1145 #ifdef _LP64 | |
| 624 | 1146 void movsbq(Register dst, Address src); |
| 1147 void movsbq(Register dst, Register src); | |
| 1148 | |
| 304 | 1149 // Move signed 32bit immediate to 64bit extending sign |
| 1150 void movslq(Address dst, int32_t imm64); | |
| 1151 void movslq(Register dst, int32_t imm64); | |
| 1152 | |
| 1153 void movslq(Register dst, Address src); | |
| 1154 void movslq(Register dst, Register src); | |
| 1155 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous | |
| 1156 #endif | |
| 1157 | |
| 1158 void movswl(Register dst, Address src); | |
| 1159 void movswl(Register dst, Register src); | |
| 1160 | |
| 624 | 1161 #ifdef _LP64 |
| 1162 void movswq(Register dst, Address src); | |
| 1163 void movswq(Register dst, Register src); | |
| 1164 #endif | |
| 1165 | |
| 304 | 1166 void movw(Address dst, int imm16); |
| 1167 void movw(Register dst, Address src); | |
| 1168 void movw(Address dst, Register src); | |
| 1169 | |
| 1170 void movzbl(Register dst, Address src); | |
| 1171 void movzbl(Register dst, Register src); | |
| 1172 | |
| 624 | 1173 #ifdef _LP64 |
| 1174 void movzbq(Register dst, Address src); | |
| 1175 void movzbq(Register dst, Register src); | |
| 1176 #endif | |
| 1177 | |
| 304 | 1178 void movzwl(Register dst, Address src); |
| 1179 void movzwl(Register dst, Register src); | |
| 1180 | |
| 624 | 1181 #ifdef _LP64 |
| 1182 void movzwq(Register dst, Address src); | |
| 1183 void movzwq(Register dst, Register src); | |
| 1184 #endif | |
| 1185 | |
| 304 | 1186 void mull(Address src); |
| 1187 void mull(Register src); | |
| 1188 | |
| 1189 // Multiply Scalar Double-Precision Floating-Point Values | |
| 1190 void mulsd(XMMRegister dst, Address src); | |
| 1191 void mulsd(XMMRegister dst, XMMRegister src); | |
| 1192 | |
| 1193 // Multiply Scalar Single-Precision Floating-Point Values | |
| 1194 void mulss(XMMRegister dst, Address src); | |
| 1195 void mulss(XMMRegister dst, XMMRegister src); | |
| 1196 | |
| 1197 void negl(Register dst); | |
| 1198 | |
| 1199 #ifdef _LP64 | |
| 1200 void negq(Register dst); | |
| 1201 #endif | |
| 1202 | |
| 1203 void nop(int i = 1); | |
| 1204 | |
| 1205 void notl(Register dst); | |
| 1206 | |
| 1207 #ifdef _LP64 | |
| 1208 void notq(Register dst); | |
| 1209 #endif | |
| 1210 | |
| 1211 void orl(Address dst, int32_t imm32); | |
| 1212 void orl(Register dst, int32_t imm32); | |
| 1213 void orl(Register dst, Address src); | |
| 1214 void orl(Register dst, Register src); | |
| 1215 | |
| 1216 void orq(Address dst, int32_t imm32); | |
| 1217 void orq(Register dst, int32_t imm32); | |
| 1218 void orq(Register dst, Address src); | |
| 1219 void orq(Register dst, Register src); | |
| 1220 | |
| 1221 void popl(Address dst); | |
| 1222 | |
| 1223 #ifdef _LP64 | |
| 1224 void popq(Address dst); | |
| 1225 #endif | |
| 1226 | |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1227 void popcntl(Register dst, Address src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1228 void popcntl(Register dst, Register src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1229 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1230 #ifdef _LP64 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1231 void popcntq(Register dst, Address src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1232 void popcntq(Register dst, Register src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1233 #endif |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1234 |
| 304 | 1235 // Prefetches (SSE, SSE2, 3DNOW only) |
| 1236 | |
| 1237 void prefetchnta(Address src); | |
| 1238 void prefetchr(Address src); | |
| 1239 void prefetcht0(Address src); | |
| 1240 void prefetcht1(Address src); | |
| 1241 void prefetcht2(Address src); | |
| 1242 void prefetchw(Address src); | |
| 1243 | |
| 1244 // Shuffle Packed Doublewords | |
| 1245 void pshufd(XMMRegister dst, XMMRegister src, int mode); | |
| 1246 void pshufd(XMMRegister dst, Address src, int mode); | |
| 1247 | |
| 1248 // Shuffle Packed Low Words | |
| 1249 void pshuflw(XMMRegister dst, XMMRegister src, int mode); | |
| 1250 void pshuflw(XMMRegister dst, Address src, int mode); | |
| 1251 | |
| 1252 // Shift Right Logical Quadword Immediate | |
| 1253 void psrlq(XMMRegister dst, int shift); | |
| 1254 | |
| 1255 // Interleave Low Bytes | |
| 1256 void punpcklbw(XMMRegister dst, XMMRegister src); | |
| 1257 | |
| 1258 void pushl(Address src); | |
| 1259 | |
| 1260 void pushq(Address src); | |
| 1261 | |
| 1262 // Xor Packed Byte Integer Values | |
| 1263 void pxor(XMMRegister dst, Address src); | |
| 1264 void pxor(XMMRegister dst, XMMRegister src); | |
| 1265 | |
| 1266 void rcll(Register dst, int imm8); | |
| 1267 | |
| 1268 void rclq(Register dst, int imm8); | |
| 1269 | |
| 1270 void ret(int imm16); | |
| 0 | 1271 |
| 1272 void sahf(); | |
| 1273 | |
| 304 | 1274 void sarl(Register dst, int imm8); |
| 1275 void sarl(Register dst); | |
| 1276 | |
| 1277 void sarq(Register dst, int imm8); | |
| 1278 void sarq(Register dst); | |
| 1279 | |
| 1280 void sbbl(Address dst, int32_t imm32); | |
| 1281 void sbbl(Register dst, int32_t imm32); | |
| 1282 void sbbl(Register dst, Address src); | |
| 1283 void sbbl(Register dst, Register src); | |
| 1284 | |
| 1285 void sbbq(Address dst, int32_t imm32); | |
| 1286 void sbbq(Register dst, int32_t imm32); | |
| 1287 void sbbq(Register dst, Address src); | |
| 1288 void sbbq(Register dst, Register src); | |
| 1289 | |
| 1290 void setb(Condition cc, Register dst); | |
| 1291 | |
| 1292 void shldl(Register dst, Register src); | |
| 1293 | |
| 1294 void shll(Register dst, int imm8); | |
| 1295 void shll(Register dst); | |
| 1296 | |
| 1297 void shlq(Register dst, int imm8); | |
| 1298 void shlq(Register dst); | |
| 1299 | |
| 1300 void shrdl(Register dst, Register src); | |
| 1301 | |
| 1302 void shrl(Register dst, int imm8); | |
| 1303 void shrl(Register dst); | |
| 1304 | |
| 1305 void shrq(Register dst, int imm8); | |
| 1306 void shrq(Register dst); | |
| 1307 | |
| 1308 void smovl(); // QQQ generic? | |
| 1309 | |
| 1310 // Compute Square Root of Scalar Double-Precision Floating-Point Value | |
| 1311 void sqrtsd(XMMRegister dst, Address src); | |
| 1312 void sqrtsd(XMMRegister dst, XMMRegister src); | |
| 1313 | |
| 1314 void std() { emit_byte(0xfd); } | |
| 1315 | |
| 1316 void stmxcsr( Address dst ); | |
| 1317 | |
| 1318 void subl(Address dst, int32_t imm32); | |
| 1319 void subl(Address dst, Register src); | |
| 1320 void subl(Register dst, int32_t imm32); | |
| 1321 void subl(Register dst, Address src); | |
| 1322 void subl(Register dst, Register src); | |
| 1323 | |
| 1324 void subq(Address dst, int32_t imm32); | |
| 1325 void subq(Address dst, Register src); | |
| 1326 void subq(Register dst, int32_t imm32); | |
| 1327 void subq(Register dst, Address src); | |
| 1328 void subq(Register dst, Register src); | |
| 1329 | |
| 1330 | |
| 1331 // Subtract Scalar Double-Precision Floating-Point Values | |
| 1332 void subsd(XMMRegister dst, Address src); | |
| 0 | 1333 void subsd(XMMRegister dst, XMMRegister src); |
| 1334 | |
| 304 | 1335 // Subtract Scalar Single-Precision Floating-Point Values |
| 1336 void subss(XMMRegister dst, Address src); | |
| 1337 void subss(XMMRegister dst, XMMRegister src); | |
| 1338 | |
| 1339 void testb(Register dst, int imm8); | |
| 1340 | |
| 1341 void testl(Register dst, int32_t imm32); | |
| 1342 void testl(Register dst, Register src); | |
| 1343 void testl(Register dst, Address src); | |
| 1344 | |
| 1345 void testq(Register dst, int32_t imm32); | |
| 1346 void testq(Register dst, Register src); | |
| 1347 | |
| 1348 | |
| 1349 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS | |
| 1350 void ucomisd(XMMRegister dst, Address src); | |
| 0 | 1351 void ucomisd(XMMRegister dst, XMMRegister src); |
| 1352 | |
| 304 | 1353 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS |
| 1354 void ucomiss(XMMRegister dst, Address src); | |
| 1355 void ucomiss(XMMRegister dst, XMMRegister src); | |
| 1356 | |
| 1357 void xaddl(Address dst, Register src); | |
| 1358 | |
| 1359 void xaddq(Address dst, Register src); | |
| 1360 | |
| 1361 void xchgl(Register reg, Address adr); | |
| 1362 void xchgl(Register dst, Register src); | |
| 1363 | |
| 1364 void xchgq(Register reg, Address adr); | |
| 1365 void xchgq(Register dst, Register src); | |
| 1366 | |
| 1367 void xorl(Register dst, int32_t imm32); | |
| 1368 void xorl(Register dst, Address src); | |
| 1369 void xorl(Register dst, Register src); | |
| 1370 | |
| 1371 void xorq(Register dst, Address src); | |
| 1372 void xorq(Register dst, Register src); | |
| 1373 | |
| 1374 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values | |
| 1375 void xorpd(XMMRegister dst, Address src); | |
| 1376 void xorpd(XMMRegister dst, XMMRegister src); | |
| 1377 | |
| 1378 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values | |
| 1379 void xorps(XMMRegister dst, Address src); | |
| 0 | 1380 void xorps(XMMRegister dst, XMMRegister src); |
| 304 | 1381 |
| 1382 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 | |
| 0 | 1383 }; |
| 1384 | |
| 1385 | |
| 1386 // MacroAssembler extends Assembler by frequently used macros. | |
| 1387 // | |
| 1388 // Instructions for which a 'better' code sequence exists depending | |
| 1389 // on arguments should also go in here. | |
| 1390 | |
| 1391 class MacroAssembler: public Assembler { | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1392 friend class LIR_Assembler; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1393 friend class Runtime1; // as_Address() |
| 0 | 1394 protected: |
| 1395 | |
| 1396 Address as_Address(AddressLiteral adr); | |
| 1397 Address as_Address(ArrayAddress adr); | |
| 1398 | |
| 1399 // Support for VM calls | |
| 1400 // | |
| 1401 // This is the base routine called by the different versions of call_VM_leaf. The interpreter | |
| 1402 // may customize this version by overriding it for its purposes (e.g., to save/restore | |
| 1403 // additional registers when doing a VM call). | |
| 1404 #ifdef CC_INTERP | |
| 1405 // c++ interpreter never wants to use interp_masm version of call_VM | |
| 1406 #define VIRTUAL | |
| 1407 #else | |
| 1408 #define VIRTUAL virtual | |
| 1409 #endif | |
| 1410 | |
| 1411 VIRTUAL void call_VM_leaf_base( | |
| 1412 address entry_point, // the entry point | |
| 1413 int number_of_arguments // the number of arguments to pop after the call | |
| 1414 ); | |
| 1415 | |
| 1416 // This is the base routine called by the different versions of call_VM. The interpreter | |
| 1417 // may customize this version by overriding it for its purposes (e.g., to save/restore | |
| 1418 // additional registers when doing a VM call). | |
| 1419 // | |
| 1420 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base | |
| 1421 // returns the register which contains the thread upon return. If a thread register has been | |
| 1422 // specified, the return value will correspond to that register. If no last_java_sp is specified | |
| 1423 // (noreg) than rsp will be used instead. | |
| 1424 VIRTUAL void call_VM_base( // returns the register containing the thread upon return | |
| 1425 Register oop_result, // where an oop-result ends up if any; use noreg otherwise | |
| 1426 Register java_thread, // the thread if computed before ; use noreg otherwise | |
| 1427 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise | |
| 1428 address entry_point, // the entry point | |
| 1429 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call | |
| 1430 bool check_exceptions // whether to check for pending exceptions after return | |
| 1431 ); | |
| 1432 | |
| 1433 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. | |
| 1434 // The implementation is only non-empty for the InterpreterMacroAssembler, | |
| 1435 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. | |
| 1436 virtual void check_and_handle_popframe(Register java_thread); | |
| 1437 virtual void check_and_handle_earlyret(Register java_thread); | |
| 1438 | |
| 1439 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); | |
| 1440 | |
| 1441 // helpers for FPU flag access | |
| 1442 // tmp is a temporary register, if none is available use noreg | |
| 1443 void save_rax (Register tmp); | |
| 1444 void restore_rax(Register tmp); | |
| 1445 | |
| 1446 public: | |
| 1447 MacroAssembler(CodeBuffer* code) : Assembler(code) {} | |
| 1448 | |
| 1449 // Support for NULL-checks | |
| 1450 // | |
| 1451 // Generates code that causes a NULL OS exception if the content of reg is NULL. | |
| 1452 // If the accessed location is M[reg + offset] and the offset is known, provide the | |
| 1453 // offset. No explicit code generation is needed if the offset is within a certain | |
| 1454 // range (0 <= offset <= page_size). | |
| 1455 | |
| 1456 void null_check(Register reg, int offset = -1); | |
|
168
7793bd37a336
6705887: Compressed Oops: generate x64 addressing and implicit null checks with narrow oops
kvn
parents:
71
diff
changeset
|
1457 static bool needs_explicit_null_check(intptr_t offset); |
| 0 | 1458 |
| 1459 // Required platform-specific helpers for Label::patch_instructions. | |
| 1460 // They _shadow_ the declarations in AbstractAssembler, which are undefined. | |
| 1461 void pd_patch_instruction(address branch, address target); | |
| 1462 #ifndef PRODUCT | |
| 1463 static void pd_print_patched_instruction(address branch); | |
| 1464 #endif | |
| 1465 | |
| 1466 // The following 4 methods return the offset of the appropriate move instruction | |
| 1467 | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1468 // Support for fast byte/short loading with zero extension (depending on particular CPU) |
| 0 | 1469 int load_unsigned_byte(Register dst, Address src); |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1470 int load_unsigned_short(Register dst, Address src); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1471 |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1472 // Support for fast byte/short loading with sign extension (depending on particular CPU) |
| 0 | 1473 int load_signed_byte(Register dst, Address src); |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1474 int load_signed_short(Register dst, Address src); |
| 0 | 1475 |
| 1476 // Support for sign-extension (hi:lo = extend_sign(lo)) | |
| 1477 void extend_sign(Register hi, Register lo); | |
| 1478 | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1479 // Loading values by size and signed-ness |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1480 void load_sized_value(Register dst, Address src, int size_in_bytes, bool is_signed); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1481 |
| 0 | 1482 // Support for inc/dec with optimal instruction selection depending on value |
| 304 | 1483 |
| 1484 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } | |
| 1485 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } | |
| 1486 | |
| 1487 void decrementl(Address dst, int value = 1); | |
| 1488 void decrementl(Register reg, int value = 1); | |
| 1489 | |
| 1490 void decrementq(Register reg, int value = 1); | |
| 1491 void decrementq(Address dst, int value = 1); | |
| 1492 | |
| 1493 void incrementl(Address dst, int value = 1); | |
| 1494 void incrementl(Register reg, int value = 1); | |
| 1495 | |
| 1496 void incrementq(Register reg, int value = 1); | |
| 1497 void incrementq(Address dst, int value = 1); | |
| 1498 | |
| 0 | 1499 |
| 1500 // Support optimal SSE move instructions. | |
| 1501 void movflt(XMMRegister dst, XMMRegister src) { | |
| 1502 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } | |
| 1503 else { movss (dst, src); return; } | |
| 1504 } | |
| 1505 void movflt(XMMRegister dst, Address src) { movss(dst, src); } | |
| 1506 void movflt(XMMRegister dst, AddressLiteral src); | |
| 1507 void movflt(Address dst, XMMRegister src) { movss(dst, src); } | |
| 1508 | |
| 1509 void movdbl(XMMRegister dst, XMMRegister src) { | |
| 1510 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } | |
| 1511 else { movsd (dst, src); return; } | |
| 1512 } | |
| 1513 | |
| 1514 void movdbl(XMMRegister dst, AddressLiteral src); | |
| 1515 | |
| 1516 void movdbl(XMMRegister dst, Address src) { | |
| 1517 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } | |
| 1518 else { movlpd(dst, src); return; } | |
| 1519 } | |
| 1520 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } | |
| 1521 | |
| 304 | 1522 void incrementl(AddressLiteral dst); |
| 1523 void incrementl(ArrayAddress dst); | |
| 0 | 1524 |
| 1525 // Alignment | |
| 1526 void align(int modulus); | |
| 1527 | |
| 1528 // Misc | |
| 1529 void fat_nop(); // 5 byte nop | |
| 1530 | |
| 1531 // Stack frame creation/removal | |
| 1532 void enter(); | |
| 1533 void leave(); | |
| 1534 | |
| 1535 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) | |
| 1536 // The pointer will be loaded into the thread register. | |
| 1537 void get_thread(Register thread); | |
| 1538 | |
| 362 | 1539 |
| 0 | 1540 // Support for VM calls |
| 1541 // | |
| 1542 // It is imperative that all calls into the VM are handled via the call_VM macros. | |
| 1543 // They make sure that the stack linkage is setup correctly. call_VM's correspond | |
| 1544 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. | |
| 1545 | |
| 304 | 1546 |
| 1547 void call_VM(Register oop_result, | |
| 1548 address entry_point, | |
| 1549 bool check_exceptions = true); | |
| 1550 void call_VM(Register oop_result, | |
| 1551 address entry_point, | |
| 1552 Register arg_1, | |
| 1553 bool check_exceptions = true); | |
| 1554 void call_VM(Register oop_result, | |
| 1555 address entry_point, | |
| 1556 Register arg_1, Register arg_2, | |
| 1557 bool check_exceptions = true); | |
| 1558 void call_VM(Register oop_result, | |
| 1559 address entry_point, | |
| 1560 Register arg_1, Register arg_2, Register arg_3, | |
| 1561 bool check_exceptions = true); | |
| 1562 | |
| 1563 // Overloadings with last_Java_sp | |
| 1564 void call_VM(Register oop_result, | |
| 1565 Register last_java_sp, | |
| 1566 address entry_point, | |
| 1567 int number_of_arguments = 0, | |
| 1568 bool check_exceptions = true); | |
| 1569 void call_VM(Register oop_result, | |
| 1570 Register last_java_sp, | |
| 1571 address entry_point, | |
| 1572 Register arg_1, bool | |
| 1573 check_exceptions = true); | |
| 1574 void call_VM(Register oop_result, | |
| 1575 Register last_java_sp, | |
| 1576 address entry_point, | |
| 1577 Register arg_1, Register arg_2, | |
| 1578 bool check_exceptions = true); | |
| 1579 void call_VM(Register oop_result, | |
| 1580 Register last_java_sp, | |
| 1581 address entry_point, | |
| 1582 Register arg_1, Register arg_2, Register arg_3, | |
| 1583 bool check_exceptions = true); | |
| 1584 | |
| 1585 void call_VM_leaf(address entry_point, | |
| 1586 int number_of_arguments = 0); | |
| 1587 void call_VM_leaf(address entry_point, | |
| 1588 Register arg_1); | |
| 1589 void call_VM_leaf(address entry_point, | |
| 1590 Register arg_1, Register arg_2); | |
| 1591 void call_VM_leaf(address entry_point, | |
| 1592 Register arg_1, Register arg_2, Register arg_3); | |
| 0 | 1593 |
| 1594 // last Java Frame (fills frame anchor) | |
| 304 | 1595 void set_last_Java_frame(Register thread, |
| 1596 Register last_java_sp, | |
| 1597 Register last_java_fp, | |
| 1598 address last_java_pc); | |
| 1599 | |
| 1600 // thread in the default location (r15_thread on 64bit) | |
| 1601 void set_last_Java_frame(Register last_java_sp, | |
| 1602 Register last_java_fp, | |
| 1603 address last_java_pc); | |
| 1604 | |
| 0 | 1605 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); |
| 1606 | |
| 304 | 1607 // thread in the default location (r15_thread on 64bit) |
| 1608 void reset_last_Java_frame(bool clear_fp, bool clear_pc); | |
| 1609 | |
| 0 | 1610 // Stores |
| 1611 void store_check(Register obj); // store check for obj - register is destroyed afterwards | |
| 1612 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) | |
| 1613 | |
| 362 | 1614 void g1_write_barrier_pre(Register obj, |
| 1615 #ifndef _LP64 | |
| 1616 Register thread, | |
| 1617 #endif | |
| 1618 Register tmp, | |
| 1619 Register tmp2, | |
| 1620 bool tosca_live); | |
| 1621 void g1_write_barrier_post(Register store_addr, | |
| 1622 Register new_val, | |
| 1623 #ifndef _LP64 | |
| 1624 Register thread, | |
| 1625 #endif | |
| 1626 Register tmp, | |
| 1627 Register tmp2); | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1628 |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1629 |
| 0 | 1630 // split store_check(Register obj) to enhance instruction interleaving |
| 1631 void store_check_part_1(Register obj); | |
| 1632 void store_check_part_2(Register obj); | |
| 1633 | |
| 1634 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 | |
| 1635 void c2bool(Register x); | |
| 1636 | |
| 1637 // C++ bool manipulation | |
| 1638 | |
| 1639 void movbool(Register dst, Address src); | |
| 1640 void movbool(Address dst, bool boolconst); | |
| 1641 void movbool(Address dst, Register src); | |
| 1642 void testbool(Register dst); | |
| 1643 | |
| 304 | 1644 // oop manipulations |
| 1645 void load_klass(Register dst, Register src); | |
| 1646 void store_klass(Register dst, Register src); | |
| 1647 | |
| 1648 void load_prototype_header(Register dst, Register src); | |
| 1649 | |
| 1650 #ifdef _LP64 | |
| 1651 void store_klass_gap(Register dst, Register src); | |
| 1652 | |
| 1653 void load_heap_oop(Register dst, Address src); | |
| 1654 void store_heap_oop(Address dst, Register src); | |
| 1655 void encode_heap_oop(Register r); | |
| 1656 void decode_heap_oop(Register r); | |
| 1657 void encode_heap_oop_not_null(Register r); | |
| 1658 void decode_heap_oop_not_null(Register r); | |
| 1659 void encode_heap_oop_not_null(Register dst, Register src); | |
| 1660 void decode_heap_oop_not_null(Register dst, Register src); | |
| 1661 | |
| 1662 void set_narrow_oop(Register dst, jobject obj); | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1663 void set_narrow_oop(Address dst, jobject obj); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1664 void cmp_narrow_oop(Register dst, jobject obj); |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1665 void cmp_narrow_oop(Address dst, jobject obj); |
| 304 | 1666 |
| 1667 // if heap base register is used - reinit it with the correct value | |
| 1668 void reinit_heapbase(); | |
| 1669 #endif // _LP64 | |
| 1670 | |
| 1671 // Int division/remainder for Java | |
| 0 | 1672 // (as idivl, but checks for special case as described in JVM spec.) |
| 1673 // returns idivl instruction offset for implicit exception handling | |
| 1674 int corrected_idivl(Register reg); | |
| 1675 | |
| 304 | 1676 // Long division/remainder for Java |
| 1677 // (as idivq, but checks for special case as described in JVM spec.) | |
| 1678 // returns idivq instruction offset for implicit exception handling | |
| 1679 int corrected_idivq(Register reg); | |
| 1680 | |
| 0 | 1681 void int3(); |
| 1682 | |
| 304 | 1683 // Long operation macros for a 32bit cpu |
| 0 | 1684 // Long negation for Java |
| 1685 void lneg(Register hi, Register lo); | |
| 1686 | |
| 1687 // Long multiplication for Java | |
| 304 | 1688 // (destroys contents of eax, ebx, ecx and edx) |
| 0 | 1689 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y |
| 1690 | |
| 1691 // Long shifts for Java | |
| 1692 // (semantics as described in JVM spec.) | |
| 1693 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) | |
| 1694 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) | |
| 1695 | |
| 1696 // Long compare for Java | |
| 1697 // (semantics as described in JVM spec.) | |
| 1698 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) | |
| 1699 | |
| 304 | 1700 |
| 1701 // misc | |
| 1702 | |
| 1703 // Sign extension | |
| 1704 void sign_extend_short(Register reg); | |
| 1705 void sign_extend_byte(Register reg); | |
| 1706 | |
| 1707 // Division by power of 2, rounding towards 0 | |
| 1708 void division_with_shift(Register reg, int shift_value); | |
| 1709 | |
| 0 | 1710 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: |
| 1711 // | |
| 1712 // CF (corresponds to C0) if x < y | |
| 1713 // PF (corresponds to C2) if unordered | |
| 1714 // ZF (corresponds to C3) if x = y | |
| 1715 // | |
| 1716 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). | |
| 1717 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) | |
| 1718 void fcmp(Register tmp); | |
| 1719 // Variant of the above which allows y to be further down the stack | |
| 1720 // and which only pops x and y if specified. If pop_right is | |
| 1721 // specified then pop_left must also be specified. | |
| 1722 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); | |
| 1723 | |
| 1724 // Floating-point comparison for Java | |
| 1725 // Compares the top-most stack entries on the FPU stack and stores the result in dst. | |
| 1726 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). | |
| 1727 // (semantics as described in JVM spec.) | |
| 1728 void fcmp2int(Register dst, bool unordered_is_less); | |
| 1729 // Variant of the above which allows y to be further down the stack | |
| 1730 // and which only pops x and y if specified. If pop_right is | |
| 1731 // specified then pop_left must also be specified. | |
| 1732 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); | |
| 1733 | |
| 1734 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) | |
| 1735 // tmp is a temporary register, if none is available use noreg | |
| 1736 void fremr(Register tmp); | |
| 1737 | |
| 1738 | |
| 1739 // same as fcmp2int, but using SSE2 | |
| 1740 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); | |
| 1741 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); | |
| 1742 | |
| 1743 // Inlined sin/cos generator for Java; must not use CPU instruction | |
| 1744 // directly on Intel as it does not have high enough precision | |
| 1745 // outside of the range [-pi/4, pi/4]. Extra argument indicate the | |
| 1746 // number of FPU stack slots in use; all but the topmost will | |
| 1747 // require saving if a slow case is necessary. Assumes argument is | |
| 1748 // on FP TOS; result is on FP TOS. No cpu registers are changed by | |
| 1749 // this code. | |
| 1750 void trigfunc(char trig, int num_fpu_regs_in_use = 1); | |
| 1751 | |
| 1752 // branch to L if FPU flag C2 is set/not set | |
| 1753 // tmp is a temporary register, if none is available use noreg | |
| 1754 void jC2 (Register tmp, Label& L); | |
| 1755 void jnC2(Register tmp, Label& L); | |
| 1756 | |
| 1757 // Pop ST (ffree & fincstp combined) | |
| 1758 void fpop(); | |
| 1759 | |
| 1760 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack | |
| 1761 void push_fTOS(); | |
| 1762 | |
| 1763 // pops double TOS element from CPU stack and pushes on FPU stack | |
| 1764 void pop_fTOS(); | |
| 1765 | |
| 1766 void empty_FPU_stack(); | |
| 1767 | |
| 1768 void push_IU_state(); | |
| 1769 void pop_IU_state(); | |
| 1770 | |
| 1771 void push_FPU_state(); | |
| 1772 void pop_FPU_state(); | |
| 1773 | |
| 1774 void push_CPU_state(); | |
| 1775 void pop_CPU_state(); | |
| 1776 | |
| 1777 // Round up to a power of two | |
| 1778 void round_to(Register reg, int modulus); | |
| 1779 | |
| 1780 // Callee saved registers handling | |
| 1781 void push_callee_saved_registers(); | |
| 1782 void pop_callee_saved_registers(); | |
| 1783 | |
| 1784 // allocation | |
| 1785 void eden_allocate( | |
| 1786 Register obj, // result: pointer to object after successful allocation | |
| 1787 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 1788 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 1789 Register t1, // temp register | |
| 1790 Label& slow_case // continuation point if fast allocation fails | |
| 1791 ); | |
| 1792 void tlab_allocate( | |
| 1793 Register obj, // result: pointer to object after successful allocation | |
| 1794 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 1795 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 1796 Register t1, // temp register | |
| 1797 Register t2, // temp register | |
| 1798 Label& slow_case // continuation point if fast allocation fails | |
| 1799 ); | |
| 1800 void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); | |
| 1801 | |
|
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1802 // interface method calling |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1803 void lookup_interface_method(Register recv_klass, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1804 Register intf_klass, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1805 RegisterConstant itable_index, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1806 Register method_result, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1807 Register scan_temp, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1808 Label& no_such_interface); |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
1809 |
| 0 | 1810 //---- |
| 1811 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 | |
| 1812 | |
| 1813 // Debugging | |
| 304 | 1814 |
| 1815 // only if +VerifyOops | |
| 1816 void verify_oop(Register reg, const char* s = "broken oop"); | |
| 0 | 1817 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); |
| 1818 | |
| 304 | 1819 // only if +VerifyFPU |
| 1820 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); | |
| 1821 | |
| 1822 // prints msg, dumps registers and stops execution | |
| 1823 void stop(const char* msg); | |
| 1824 | |
| 1825 // prints msg and continues | |
| 1826 void warn(const char* msg); | |
| 1827 | |
| 1828 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); | |
| 1829 static void debug64(char* msg, int64_t pc, int64_t regs[]); | |
| 1830 | |
| 0 | 1831 void os_breakpoint(); |
| 304 | 1832 |
| 0 | 1833 void untested() { stop("untested"); } |
| 304 | 1834 |
| 0 | 1835 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, sizeof(b), "unimplemented: %s", what); stop(b); } |
| 304 | 1836 |
| 0 | 1837 void should_not_reach_here() { stop("should not reach here"); } |
| 304 | 1838 |
| 0 | 1839 void print_CPU_state(); |
| 1840 | |
| 1841 // Stack overflow checking | |
| 1842 void bang_stack_with_offset(int offset) { | |
| 1843 // stack grows down, caller passes positive offset | |
| 1844 assert(offset > 0, "must bang with negative offset"); | |
| 1845 movl(Address(rsp, (-offset)), rax); | |
| 1846 } | |
| 1847 | |
| 1848 // Writes to stack successive pages until offset reached to check for | |
| 1849 // stack overflow + shadow pages. Also, clobbers tmp | |
| 1850 void bang_stack_size(Register size, Register tmp); | |
| 1851 | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1852 virtual RegisterConstant delayed_value(intptr_t* delayed_value_addr, |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1853 Register tmp, |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1854 int offset); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1855 |
| 0 | 1856 // Support for serializing memory accesses between threads |
| 1857 void serialize_memory(Register thread, Register tmp); | |
| 1858 | |
| 1859 void verify_tlab(); | |
| 1860 | |
| 1861 // Biased locking support | |
| 1862 // lock_reg and obj_reg must be loaded up with the appropriate values. | |
| 1863 // swap_reg must be rax, and is killed. | |
| 1864 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will | |
| 1865 // be killed; if not supplied, push/pop will be used internally to | |
| 1866 // allocate a temporary (inefficient, avoid if possible). | |
| 1867 // Optional slow case is for implementations (interpreter and C1) which branch to | |
| 1868 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. | |
| 1869 // Returns offset of first potentially-faulting instruction for null | |
| 1870 // check info (currently consumed only by C1). If | |
| 1871 // swap_reg_contains_mark is true then returns -1 as it is assumed | |
| 1872 // the calling code has already passed any potential faults. | |
|
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
1873 int biased_locking_enter(Register lock_reg, Register obj_reg, |
|
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
1874 Register swap_reg, Register tmp_reg, |
| 0 | 1875 bool swap_reg_contains_mark, |
| 1876 Label& done, Label* slow_case = NULL, | |
| 1877 BiasedLockingCounters* counters = NULL); | |
| 1878 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); | |
| 1879 | |
| 1880 | |
| 1881 Condition negate_condition(Condition cond); | |
| 1882 | |
| 1883 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit | |
| 1884 // operands. In general the names are modified to avoid hiding the instruction in Assembler | |
| 1885 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers | |
| 1886 // here in MacroAssembler. The major exception to this rule is call | |
| 1887 | |
| 1888 // Arithmetics | |
| 1889 | |
| 304 | 1890 |
| 1891 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } | |
| 1892 void addptr(Address dst, Register src); | |
| 1893 | |
| 1894 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } | |
| 1895 void addptr(Register dst, int32_t src); | |
| 1896 void addptr(Register dst, Register src); | |
| 1897 | |
| 1898 void andptr(Register dst, int32_t src); | |
| 1899 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } | |
| 1900 | |
| 1901 void cmp8(AddressLiteral src1, int imm); | |
| 1902 | |
| 1903 // renamed to drag out the casting of address to int32_t/intptr_t | |
| 0 | 1904 void cmp32(Register src1, int32_t imm); |
| 1905 | |
| 1906 void cmp32(AddressLiteral src1, int32_t imm); | |
| 1907 // compare reg - mem, or reg - &mem | |
| 1908 void cmp32(Register src1, AddressLiteral src2); | |
| 1909 | |
| 1910 void cmp32(Register src1, Address src2); | |
| 1911 | |
| 304 | 1912 #ifndef _LP64 |
| 1913 void cmpoop(Address dst, jobject obj); | |
| 1914 void cmpoop(Register dst, jobject obj); | |
| 1915 #endif // _LP64 | |
| 1916 | |
| 0 | 1917 // NOTE src2 must be the lval. This is NOT an mem-mem compare |
| 1918 void cmpptr(Address src1, AddressLiteral src2); | |
| 1919 | |
| 1920 void cmpptr(Register src1, AddressLiteral src2); | |
| 1921 | |
| 304 | 1922 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } |
| 1923 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 1924 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 1925 | |
| 1926 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 1927 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 1928 | |
| 1929 // cmp64 to avoild hiding cmpq | |
| 1930 void cmp64(Register src1, AddressLiteral src); | |
| 1931 | |
| 1932 void cmpxchgptr(Register reg, Address adr); | |
| 1933 | |
| 1934 void locked_cmpxchgptr(Register reg, AddressLiteral adr); | |
| 1935 | |
| 1936 | |
| 1937 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } | |
| 1938 | |
| 1939 | |
| 1940 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } | |
| 1941 | |
| 1942 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } | |
| 1943 | |
| 1944 void shlptr(Register dst, int32_t shift); | |
| 1945 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } | |
| 1946 | |
| 1947 void shrptr(Register dst, int32_t shift); | |
| 1948 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } | |
| 1949 | |
| 1950 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } | |
| 1951 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } | |
| 1952 | |
| 1953 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } | |
| 1954 | |
| 1955 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } | |
| 1956 void subptr(Register dst, int32_t src); | |
| 1957 void subptr(Register dst, Register src); | |
| 1958 | |
| 1959 | |
| 1960 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } | |
| 1961 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } | |
| 1962 | |
| 1963 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } | |
| 1964 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } | |
| 1965 | |
| 1966 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } | |
| 1967 | |
| 1968 | |
| 0 | 1969 |
| 1970 // Helper functions for statistics gathering. | |
| 1971 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. | |
| 1972 void cond_inc32(Condition cond, AddressLiteral counter_addr); | |
| 1973 // Unconditional atomic increment. | |
| 1974 void atomic_incl(AddressLiteral counter_addr); | |
| 1975 | |
| 1976 void lea(Register dst, AddressLiteral adr); | |
| 1977 void lea(Address dst, AddressLiteral adr); | |
| 304 | 1978 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } |
| 1979 | |
| 1980 void leal32(Register dst, Address src) { leal(dst, src); } | |
| 1981 | |
| 1982 void test32(Register src1, AddressLiteral src2); | |
| 1983 | |
| 1984 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 1985 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 1986 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 1987 | |
| 1988 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } | |
| 1989 void testptr(Register src1, Register src2); | |
| 1990 | |
| 1991 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } | |
| 1992 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } | |
| 0 | 1993 |
| 1994 // Calls | |
| 1995 | |
| 1996 void call(Label& L, relocInfo::relocType rtype); | |
| 1997 void call(Register entry); | |
| 1998 | |
| 1999 // NOTE: this call tranfers to the effective address of entry NOT | |
| 2000 // the address contained by entry. This is because this is more natural | |
| 2001 // for jumps/calls. | |
| 2002 void call(AddressLiteral entry); | |
| 2003 | |
| 2004 // Jumps | |
| 2005 | |
| 2006 // NOTE: these jumps tranfer to the effective address of dst NOT | |
| 2007 // the address contained by dst. This is because this is more natural | |
| 2008 // for jumps/calls. | |
| 2009 void jump(AddressLiteral dst); | |
| 2010 void jump_cc(Condition cc, AddressLiteral dst); | |
| 2011 | |
| 2012 // 32bit can do a case table jump in one instruction but we no longer allow the base | |
| 2013 // to be installed in the Address class. This jump will tranfers to the address | |
| 2014 // contained in the location described by entry (not the address of entry) | |
| 2015 void jump(ArrayAddress entry); | |
| 2016 | |
| 2017 // Floating | |
| 2018 | |
| 2019 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } | |
| 2020 void andpd(XMMRegister dst, AddressLiteral src); | |
| 2021 | |
| 2022 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } | |
| 2023 void comiss(XMMRegister dst, AddressLiteral src); | |
| 2024 | |
| 2025 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } | |
| 2026 void comisd(XMMRegister dst, AddressLiteral src); | |
| 2027 | |
| 2028 void fldcw(Address src) { Assembler::fldcw(src); } | |
| 2029 void fldcw(AddressLiteral src); | |
| 2030 | |
| 2031 void fld_s(int index) { Assembler::fld_s(index); } | |
| 2032 void fld_s(Address src) { Assembler::fld_s(src); } | |
| 2033 void fld_s(AddressLiteral src); | |
| 2034 | |
| 2035 void fld_d(Address src) { Assembler::fld_d(src); } | |
| 2036 void fld_d(AddressLiteral src); | |
| 2037 | |
| 2038 void fld_x(Address src) { Assembler::fld_x(src); } | |
| 2039 void fld_x(AddressLiteral src); | |
| 2040 | |
| 2041 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } | |
| 2042 void ldmxcsr(AddressLiteral src); | |
| 2043 | |
| 304 | 2044 private: |
| 2045 // these are private because users should be doing movflt/movdbl | |
| 2046 | |
| 0 | 2047 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } |
| 2048 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } | |
| 2049 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } | |
| 2050 void movss(XMMRegister dst, AddressLiteral src); | |
| 2051 | |
| 304 | 2052 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } |
| 2053 void movlpd(XMMRegister dst, AddressLiteral src); | |
| 2054 | |
| 2055 public: | |
| 2056 | |
| 0 | 2057 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } |
| 2058 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } | |
| 2059 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } | |
| 2060 void movsd(XMMRegister dst, AddressLiteral src); | |
| 2061 | |
| 2062 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } | |
| 2063 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } | |
| 2064 void ucomiss(XMMRegister dst, AddressLiteral src); | |
| 2065 | |
| 2066 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } | |
| 2067 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } | |
| 2068 void ucomisd(XMMRegister dst, AddressLiteral src); | |
| 2069 | |
| 2070 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values | |
| 2071 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); } | |
| 2072 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } | |
| 2073 void xorpd(XMMRegister dst, AddressLiteral src); | |
| 2074 | |
| 2075 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values | |
| 2076 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); } | |
| 2077 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } | |
| 2078 void xorps(XMMRegister dst, AddressLiteral src); | |
| 2079 | |
| 2080 // Data | |
| 2081 | |
| 304 | 2082 void cmov(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } |
| 2083 | |
| 2084 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } | |
| 2085 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } | |
| 2086 | |
| 0 | 2087 void movoop(Register dst, jobject obj); |
| 2088 void movoop(Address dst, jobject obj); | |
| 2089 | |
| 2090 void movptr(ArrayAddress dst, Register src); | |
| 2091 // can this do an lea? | |
| 2092 void movptr(Register dst, ArrayAddress src); | |
| 2093 | |
| 304 | 2094 void movptr(Register dst, Address src); |
| 2095 | |
| 0 | 2096 void movptr(Register dst, AddressLiteral src); |
| 2097 | |
| 304 | 2098 void movptr(Register dst, intptr_t src); |
| 2099 void movptr(Register dst, Register src); | |
| 2100 void movptr(Address dst, intptr_t src); | |
| 2101 | |
| 2102 void movptr(Address dst, Register src); | |
| 2103 | |
| 2104 #ifdef _LP64 | |
| 2105 // Generally the next two are only used for moving NULL | |
| 2106 // Although there are situations in initializing the mark word where | |
| 2107 // they could be used. They are dangerous. | |
| 2108 | |
| 2109 // They only exist on LP64 so that int32_t and intptr_t are not the same | |
| 2110 // and we have ambiguous declarations. | |
| 2111 | |
| 2112 void movptr(Address dst, int32_t imm32); | |
| 2113 void movptr(Register dst, int32_t imm32); | |
| 2114 #endif // _LP64 | |
| 2115 | |
| 0 | 2116 // to avoid hiding movl |
| 2117 void mov32(AddressLiteral dst, Register src); | |
| 2118 void mov32(Register dst, AddressLiteral src); | |
| 304 | 2119 |
| 0 | 2120 // to avoid hiding movb |
| 2121 void movbyte(ArrayAddress dst, int src); | |
| 2122 | |
| 2123 // Can push value or effective address | |
| 2124 void pushptr(AddressLiteral src); | |
| 2125 | |
| 304 | 2126 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } |
| 2127 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } | |
| 2128 | |
| 2129 void pushoop(jobject obj); | |
| 2130 | |
| 2131 // sign extend as need a l to ptr sized element | |
| 2132 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } | |
| 2133 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } | |
| 2134 | |
| 2135 | |
| 0 | 2136 #undef VIRTUAL |
| 2137 | |
| 2138 }; | |
| 2139 | |
| 2140 /** | |
| 2141 * class SkipIfEqual: | |
| 2142 * | |
| 2143 * Instantiating this class will result in assembly code being output that will | |
| 2144 * jump around any code emitted between the creation of the instance and it's | |
| 2145 * automatic destruction at the end of a scope block, depending on the value of | |
| 2146 * the flag passed to the constructor, which will be checked at run-time. | |
| 2147 */ | |
| 2148 class SkipIfEqual { | |
| 2149 private: | |
| 2150 MacroAssembler* _masm; | |
| 2151 Label _label; | |
| 2152 | |
| 2153 public: | |
| 2154 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); | |
| 2155 ~SkipIfEqual(); | |
| 2156 }; | |
| 2157 | |
| 2158 #ifdef ASSERT | |
| 2159 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } | |
| 2160 #endif |