Mercurial > hg > truffle
annotate src/cpu/x86/vm/assembler_x86.hpp @ 6614:006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
Summary: Added asm encoding and mach nodes for vector arithmetic instructions on x86.
Reviewed-by: roland
| author | kvn |
|---|---|
| date | Mon, 20 Aug 2012 09:07:21 -0700 |
| parents | 1d7922586cf6 |
| children | da91efe96a93 |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. 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 * | |
|
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1503
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1503
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1503
diff
changeset
|
21 * questions. |
| 0 | 22 * |
| 23 */ | |
| 24 | |
| 1972 | 25 #ifndef CPU_X86_VM_ASSEMBLER_X86_HPP |
| 26 #define CPU_X86_VM_ASSEMBLER_X86_HPP | |
| 27 | |
| 0 | 28 class BiasedLockingCounters; |
| 29 | |
| 30 // Contains all the definitions needed for x86 assembly code generation. | |
| 31 | |
| 32 // Calling convention | |
| 33 class Argument VALUE_OBJ_CLASS_SPEC { | |
| 34 public: | |
| 35 enum { | |
| 36 #ifdef _LP64 | |
| 37 #ifdef _WIN64 | |
| 38 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) | |
| 39 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) | |
| 40 #else | |
| 41 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) | |
| 42 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) | |
| 43 #endif // _WIN64 | |
| 44 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... | |
| 45 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... | |
| 46 #else | |
| 47 n_register_parameters = 0 // 0 registers used to pass arguments | |
| 48 #endif // _LP64 | |
| 49 }; | |
| 50 }; | |
| 51 | |
| 52 | |
| 53 #ifdef _LP64 | |
| 54 // Symbolically name the register arguments used by the c calling convention. | |
| 55 // Windows is different from linux/solaris. So much for standards... | |
| 56 | |
| 57 #ifdef _WIN64 | |
| 58 | |
| 59 REGISTER_DECLARATION(Register, c_rarg0, rcx); | |
| 60 REGISTER_DECLARATION(Register, c_rarg1, rdx); | |
| 61 REGISTER_DECLARATION(Register, c_rarg2, r8); | |
| 62 REGISTER_DECLARATION(Register, c_rarg3, r9); | |
| 63 | |
| 304 | 64 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); |
| 65 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
| 66 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
| 67 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
| 0 | 68 |
| 69 #else | |
| 70 | |
| 71 REGISTER_DECLARATION(Register, c_rarg0, rdi); | |
| 72 REGISTER_DECLARATION(Register, c_rarg1, rsi); | |
| 73 REGISTER_DECLARATION(Register, c_rarg2, rdx); | |
| 74 REGISTER_DECLARATION(Register, c_rarg3, rcx); | |
| 75 REGISTER_DECLARATION(Register, c_rarg4, r8); | |
| 76 REGISTER_DECLARATION(Register, c_rarg5, r9); | |
| 77 | |
| 304 | 78 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); |
| 79 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
| 80 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
| 81 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
| 82 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); | |
| 83 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); | |
| 84 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); | |
| 85 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); | |
| 0 | 86 |
| 87 #endif // _WIN64 | |
| 88 | |
| 89 // Symbolically name the register arguments used by the Java calling convention. | |
| 90 // We have control over the convention for java so we can do what we please. | |
| 91 // What pleases us is to offset the java calling convention so that when | |
| 92 // we call a suitable jni method the arguments are lined up and we don't | |
| 93 // have to do little shuffling. A suitable jni method is non-static and a | |
| 94 // small number of arguments (two fewer args on windows) | |
| 95 // | |
| 96 // |-------------------------------------------------------| | |
| 97 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | | |
| 98 // |-------------------------------------------------------| | |
| 99 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) | |
| 100 // | rdi rsi rdx rcx r8 r9 | solaris/linux | |
| 101 // |-------------------------------------------------------| | |
| 102 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | | |
| 103 // |-------------------------------------------------------| | |
| 104 | |
| 105 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); | |
| 106 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); | |
| 107 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); | |
| 108 // Windows runs out of register args here | |
| 109 #ifdef _WIN64 | |
| 110 REGISTER_DECLARATION(Register, j_rarg3, rdi); | |
| 111 REGISTER_DECLARATION(Register, j_rarg4, rsi); | |
| 112 #else | |
| 113 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); | |
| 114 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); | |
| 115 #endif /* _WIN64 */ | |
| 116 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); | |
| 117 | |
| 304 | 118 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); |
| 119 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); | |
| 120 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); | |
| 121 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); | |
| 122 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); | |
| 123 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); | |
| 124 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); | |
| 125 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); | |
| 0 | 126 |
| 127 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile | |
| 128 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile | |
| 129 | |
| 304 | 130 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved |
| 0 | 131 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved |
| 132 | |
| 304 | 133 #else |
| 134 // rscratch1 will apear in 32bit code that is dead but of course must compile | |
| 135 // Using noreg ensures if the dead code is incorrectly live and executed it | |
| 136 // will cause an assertion failure | |
| 137 #define rscratch1 noreg | |
| 2002 | 138 #define rscratch2 noreg |
| 304 | 139 |
| 0 | 140 #endif // _LP64 |
| 141 | |
| 1564 | 142 // JSR 292 fixed register usages: |
| 143 REGISTER_DECLARATION(Register, rbp_mh_SP_save, rbp); | |
| 144 | |
| 0 | 145 // Address is an abstraction used to represent a memory location |
| 146 // using any of the amd64 addressing modes with one object. | |
| 147 // | |
| 148 // Note: A register location is represented via a Register, not | |
| 149 // via an address for efficiency & simplicity reasons. | |
| 150 | |
| 151 class ArrayAddress; | |
| 152 | |
| 153 class Address VALUE_OBJ_CLASS_SPEC { | |
| 154 public: | |
| 155 enum ScaleFactor { | |
| 156 no_scale = -1, | |
| 157 times_1 = 0, | |
| 158 times_2 = 1, | |
| 159 times_4 = 2, | |
| 304 | 160 times_8 = 3, |
| 161 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) | |
| 0 | 162 }; |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
163 static ScaleFactor times(int size) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
164 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
|
165 if (size == 8) return times_8; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
166 if (size == 4) return times_4; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
167 if (size == 2) return times_2; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
168 return times_1; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
169 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
170 static int scale_size(ScaleFactor scale) { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
171 assert(scale != no_scale, ""); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
172 assert(((1 << (int)times_1) == 1 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
173 (1 << (int)times_2) == 2 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
174 (1 << (int)times_4) == 4 && |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
175 (1 << (int)times_8) == 8), ""); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
176 return (1 << (int)scale); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
177 } |
| 0 | 178 |
| 179 private: | |
| 180 Register _base; | |
| 181 Register _index; | |
| 182 ScaleFactor _scale; | |
| 183 int _disp; | |
| 184 RelocationHolder _rspec; | |
| 185 | |
| 304 | 186 // Easily misused constructors make them private |
| 187 // %%% can we make these go away? | |
| 188 NOT_LP64(Address(address loc, RelocationHolder spec);) | |
| 189 Address(int disp, address loc, relocInfo::relocType rtype); | |
| 190 Address(int disp, address loc, RelocationHolder spec); | |
| 0 | 191 |
| 192 public: | |
| 304 | 193 |
| 194 int disp() { return _disp; } | |
| 0 | 195 // creation |
| 196 Address() | |
| 197 : _base(noreg), | |
| 198 _index(noreg), | |
| 199 _scale(no_scale), | |
| 200 _disp(0) { | |
| 201 } | |
| 202 | |
| 203 // No default displacement otherwise Register can be implicitly | |
| 204 // converted to 0(Register) which is quite a different animal. | |
| 205 | |
| 206 Address(Register base, int disp) | |
| 207 : _base(base), | |
| 208 _index(noreg), | |
| 209 _scale(no_scale), | |
| 210 _disp(disp) { | |
| 211 } | |
| 212 | |
| 213 Address(Register base, Register index, ScaleFactor scale, int disp = 0) | |
| 214 : _base (base), | |
| 215 _index(index), | |
| 216 _scale(scale), | |
| 217 _disp (disp) { | |
| 218 assert(!index->is_valid() == (scale == Address::no_scale), | |
| 219 "inconsistent address"); | |
| 220 } | |
| 221 | |
|
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
222 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
223 : _base (base), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
224 _index(index.register_or_noreg()), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
225 _scale(scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
226 _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
|
227 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
|
228 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
|
229 "inconsistent address"); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
230 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
231 |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
232 Address plus_disp(int disp) const { |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
233 Address a = (*this); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
234 a._disp += disp; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
235 return a; |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
236 } |
|
3363
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
237 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
238 Address a = (*this); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
239 a._disp += disp.constant_or_zero() * scale_size(scale); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
240 if (disp.is_register()) { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
241 assert(!a.index()->is_valid(), "competing indexes"); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
242 a._index = disp.as_register(); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
243 a._scale = scale; |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
244 } |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
245 return a; |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
246 } |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
247 bool is_same_address(Address a) const { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
248 // disregard _rspec |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
249 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale; |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
250 } |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
251 |
| 0 | 252 // The following two overloads are used in connection with the |
| 253 // ByteSize type (see sizes.hpp). They simplify the use of | |
| 254 // ByteSize'd arguments in assembly code. Note that their equivalent | |
| 255 // for the optimized build are the member functions with int disp | |
| 256 // argument since ByteSize is mapped to an int type in that case. | |
| 257 // | |
| 258 // Note: DO NOT introduce similar overloaded functions for WordSize | |
| 259 // arguments as in the optimized mode, both ByteSize and WordSize | |
| 260 // are mapped to the same type and thus the compiler cannot make a | |
| 261 // distinction anymore (=> compiler errors). | |
| 262 | |
| 263 #ifdef ASSERT | |
| 264 Address(Register base, ByteSize disp) | |
| 265 : _base(base), | |
| 266 _index(noreg), | |
| 267 _scale(no_scale), | |
| 268 _disp(in_bytes(disp)) { | |
| 269 } | |
| 270 | |
| 271 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) | |
| 272 : _base(base), | |
| 273 _index(index), | |
| 274 _scale(scale), | |
| 275 _disp(in_bytes(disp)) { | |
| 276 assert(!index->is_valid() == (scale == Address::no_scale), | |
| 277 "inconsistent address"); | |
| 278 } | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
279 |
|
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
280 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
281 : _base (base), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
282 _index(index.register_or_noreg()), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
283 _scale(scale), |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
284 _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
|
285 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
|
286 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
|
287 "inconsistent address"); |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
288 } |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
289 |
| 0 | 290 #endif // ASSERT |
| 291 | |
| 292 // accessors | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
293 bool uses(Register reg) const { return _base == reg || _index == reg; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
294 Register base() const { return _base; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
295 Register index() const { return _index; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
296 ScaleFactor scale() const { return _scale; } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
297 int disp() const { return _disp; } |
| 0 | 298 |
| 299 // Convert the raw encoding form into the form expected by the constructor for | |
| 300 // Address. An index of 4 (rsp) corresponds to having no index, so convert | |
| 301 // that to noreg for the Address constructor. | |
| 624 | 302 static Address make_raw(int base, int index, int scale, int disp, bool disp_is_oop); |
| 0 | 303 |
| 304 static Address make_array(ArrayAddress); | |
| 305 | |
| 306 private: | |
| 307 bool base_needs_rex() const { | |
| 308 return _base != noreg && _base->encoding() >= 8; | |
| 309 } | |
| 310 | |
| 311 bool index_needs_rex() const { | |
| 312 return _index != noreg &&_index->encoding() >= 8; | |
| 313 } | |
| 314 | |
| 315 relocInfo::relocType reloc() const { return _rspec.type(); } | |
| 316 | |
| 317 friend class Assembler; | |
| 318 friend class MacroAssembler; | |
| 319 friend class LIR_Assembler; // base/index/scale/disp | |
| 320 }; | |
| 321 | |
| 322 // | |
| 323 // AddressLiteral has been split out from Address because operands of this type | |
| 324 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out | |
| 325 // the few instructions that need to deal with address literals are unique and the | |
| 326 // MacroAssembler does not have to implement every instruction in the Assembler | |
| 327 // in order to search for address literals that may need special handling depending | |
| 328 // on the instruction and the platform. As small step on the way to merging i486/amd64 | |
| 329 // directories. | |
| 330 // | |
| 331 class AddressLiteral VALUE_OBJ_CLASS_SPEC { | |
| 332 friend class ArrayAddress; | |
| 333 RelocationHolder _rspec; | |
| 334 // Typically we use AddressLiterals we want to use their rval | |
| 335 // However in some situations we want the lval (effect address) of the item. | |
| 336 // We provide a special factory for making those lvals. | |
| 337 bool _is_lval; | |
| 338 | |
| 339 // If the target is far we'll need to load the ea of this to | |
| 340 // a register to reach it. Otherwise if near we can do rip | |
| 341 // relative addressing. | |
| 342 | |
| 343 address _target; | |
| 344 | |
| 345 protected: | |
| 346 // creation | |
| 347 AddressLiteral() | |
| 348 : _is_lval(false), | |
| 349 _target(NULL) | |
| 350 {} | |
| 351 | |
| 352 public: | |
| 353 | |
| 354 | |
| 355 AddressLiteral(address target, relocInfo::relocType rtype); | |
| 356 | |
| 357 AddressLiteral(address target, RelocationHolder const& rspec) | |
| 358 : _rspec(rspec), | |
| 359 _is_lval(false), | |
| 360 _target(target) | |
| 361 {} | |
| 362 | |
| 363 AddressLiteral addr() { | |
| 364 AddressLiteral ret = *this; | |
| 365 ret._is_lval = true; | |
| 366 return ret; | |
| 367 } | |
| 368 | |
| 369 | |
| 370 private: | |
| 371 | |
| 372 address target() { return _target; } | |
| 373 bool is_lval() { return _is_lval; } | |
| 374 | |
| 375 relocInfo::relocType reloc() const { return _rspec.type(); } | |
| 376 const RelocationHolder& rspec() const { return _rspec; } | |
| 377 | |
| 378 friend class Assembler; | |
| 379 friend class MacroAssembler; | |
| 380 friend class Address; | |
| 381 friend class LIR_Assembler; | |
| 382 }; | |
| 383 | |
| 384 // Convience classes | |
| 385 class RuntimeAddress: public AddressLiteral { | |
| 386 | |
| 387 public: | |
| 388 | |
| 389 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} | |
| 390 | |
| 391 }; | |
| 392 | |
| 393 class OopAddress: public AddressLiteral { | |
| 394 | |
| 395 public: | |
| 396 | |
| 397 OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){} | |
| 398 | |
| 399 }; | |
| 400 | |
| 401 class ExternalAddress: public AddressLiteral { | |
|
2455
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
402 private: |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
403 static relocInfo::relocType reloc_for_target(address target) { |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
404 // Sometimes ExternalAddress is used for values which aren't |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
405 // exactly addresses, like the card table base. |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
406 // external_word_type can't be used for values in the first page |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
407 // so just skip the reloc in that case. |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
408 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
409 } |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
410 |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
411 public: |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
412 |
|
479b4b4b6950
6777083: assert(target != __null,"must not be null")
never
parents:
2415
diff
changeset
|
413 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} |
| 0 | 414 |
| 415 }; | |
| 416 | |
| 417 class InternalAddress: public AddressLiteral { | |
| 418 | |
| 419 public: | |
| 420 | |
| 421 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} | |
| 422 | |
| 423 }; | |
| 424 | |
| 425 // x86 can do array addressing as a single operation since disp can be an absolute | |
| 426 // address amd64 can't. We create a class that expresses the concept but does extra | |
| 427 // magic on amd64 to get the final result | |
| 428 | |
| 429 class ArrayAddress VALUE_OBJ_CLASS_SPEC { | |
| 430 private: | |
| 431 | |
| 432 AddressLiteral _base; | |
| 433 Address _index; | |
| 434 | |
| 435 public: | |
| 436 | |
| 437 ArrayAddress() {}; | |
| 438 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; | |
| 439 AddressLiteral base() { return _base; } | |
| 440 Address index() { return _index; } | |
| 441 | |
| 442 }; | |
| 443 | |
| 304 | 444 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); |
| 0 | 445 |
| 446 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction | |
| 447 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write | |
| 448 // is what you get. The Assembler is generating code into a CodeBuffer. | |
| 449 | |
| 450 class Assembler : public AbstractAssembler { | |
| 451 friend class AbstractAssembler; // for the non-virtual hack | |
| 452 friend class LIR_Assembler; // as_Address() | |
| 304 | 453 friend class StubGenerator; |
| 0 | 454 |
| 455 public: | |
| 456 enum Condition { // The x86 condition codes used for conditional jumps/moves. | |
| 457 zero = 0x4, | |
| 458 notZero = 0x5, | |
| 459 equal = 0x4, | |
| 460 notEqual = 0x5, | |
| 461 less = 0xc, | |
| 462 lessEqual = 0xe, | |
| 463 greater = 0xf, | |
| 464 greaterEqual = 0xd, | |
| 465 below = 0x2, | |
| 466 belowEqual = 0x6, | |
| 467 above = 0x7, | |
| 468 aboveEqual = 0x3, | |
| 469 overflow = 0x0, | |
| 470 noOverflow = 0x1, | |
| 471 carrySet = 0x2, | |
| 472 carryClear = 0x3, | |
| 473 negative = 0x8, | |
| 474 positive = 0x9, | |
| 475 parity = 0xa, | |
| 476 noParity = 0xb | |
| 477 }; | |
| 478 | |
| 479 enum Prefix { | |
| 480 // segment overrides | |
| 481 CS_segment = 0x2e, | |
| 482 SS_segment = 0x36, | |
| 483 DS_segment = 0x3e, | |
| 484 ES_segment = 0x26, | |
| 485 FS_segment = 0x64, | |
| 486 GS_segment = 0x65, | |
| 487 | |
| 488 REX = 0x40, | |
| 489 | |
| 490 REX_B = 0x41, | |
| 491 REX_X = 0x42, | |
| 492 REX_XB = 0x43, | |
| 493 REX_R = 0x44, | |
| 494 REX_RB = 0x45, | |
| 495 REX_RX = 0x46, | |
| 496 REX_RXB = 0x47, | |
| 497 | |
| 498 REX_W = 0x48, | |
| 499 | |
| 500 REX_WB = 0x49, | |
| 501 REX_WX = 0x4A, | |
| 502 REX_WXB = 0x4B, | |
| 503 REX_WR = 0x4C, | |
| 504 REX_WRB = 0x4D, | |
| 505 REX_WRX = 0x4E, | |
| 4759 | 506 REX_WRXB = 0x4F, |
| 507 | |
| 508 VEX_3bytes = 0xC4, | |
| 509 VEX_2bytes = 0xC5 | |
| 510 }; | |
| 511 | |
| 512 enum VexPrefix { | |
| 513 VEX_B = 0x20, | |
| 514 VEX_X = 0x40, | |
| 515 VEX_R = 0x80, | |
| 516 VEX_W = 0x80 | |
| 517 }; | |
| 518 | |
| 519 enum VexSimdPrefix { | |
| 520 VEX_SIMD_NONE = 0x0, | |
| 521 VEX_SIMD_66 = 0x1, | |
| 522 VEX_SIMD_F3 = 0x2, | |
| 523 VEX_SIMD_F2 = 0x3 | |
| 524 }; | |
| 525 | |
| 526 enum VexOpcode { | |
| 527 VEX_OPCODE_NONE = 0x0, | |
| 528 VEX_OPCODE_0F = 0x1, | |
| 529 VEX_OPCODE_0F_38 = 0x2, | |
| 530 VEX_OPCODE_0F_3A = 0x3 | |
| 0 | 531 }; |
| 532 | |
| 533 enum WhichOperand { | |
| 534 // input to locate_operand, and format code for relocations | |
| 304 | 535 imm_operand = 0, // embedded 32-bit|64-bit immediate operand |
| 0 | 536 disp32_operand = 1, // embedded 32-bit displacement or address |
| 537 call32_operand = 2, // embedded 32-bit self-relative displacement | |
| 304 | 538 #ifndef _LP64 |
| 0 | 539 _WhichOperand_limit = 3 |
| 304 | 540 #else |
| 541 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop | |
| 542 _WhichOperand_limit = 4 | |
| 543 #endif | |
| 0 | 544 }; |
| 545 | |
| 304 | 546 |
| 547 | |
| 548 // NOTE: The general philopsophy of the declarations here is that 64bit versions | |
| 549 // of instructions are freely declared without the need for wrapping them an ifdef. | |
| 550 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) | |
| 551 // In the .cpp file the implementations are wrapped so that they are dropped out | |
| 552 // of the resulting jvm. This is done mostly to keep the footprint of KERNEL | |
| 553 // to the size it was prior to merging up the 32bit and 64bit assemblers. | |
| 554 // | |
| 555 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction | |
| 556 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. | |
| 557 | |
| 558 private: | |
| 559 | |
| 560 | |
| 561 // 64bit prefixes | |
| 562 int prefix_and_encode(int reg_enc, bool byteinst = false); | |
| 563 int prefixq_and_encode(int reg_enc); | |
| 564 | |
| 565 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); | |
| 566 int prefixq_and_encode(int dst_enc, int src_enc); | |
| 567 | |
| 568 void prefix(Register reg); | |
| 569 void prefix(Address adr); | |
| 570 void prefixq(Address adr); | |
| 571 | |
| 572 void prefix(Address adr, Register reg, bool byteinst = false); | |
| 4759 | 573 void prefix(Address adr, XMMRegister reg); |
| 304 | 574 void prefixq(Address adr, Register reg); |
| 4759 | 575 void prefixq(Address adr, XMMRegister reg); |
| 304 | 576 |
| 577 void prefetch_prefix(Address src); | |
| 578 | |
| 4759 | 579 void rex_prefix(Address adr, XMMRegister xreg, |
| 580 VexSimdPrefix pre, VexOpcode opc, bool rex_w); | |
| 581 int rex_prefix_and_encode(int dst_enc, int src_enc, | |
| 582 VexSimdPrefix pre, VexOpcode opc, bool rex_w); | |
| 583 | |
| 584 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, | |
| 585 int nds_enc, VexSimdPrefix pre, VexOpcode opc, | |
| 586 bool vector256); | |
| 587 | |
| 588 void vex_prefix(Address adr, int nds_enc, int xreg_enc, | |
| 589 VexSimdPrefix pre, VexOpcode opc, | |
| 590 bool vex_w, bool vector256); | |
| 591 | |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
592 void vex_prefix(XMMRegister dst, XMMRegister nds, Address src, |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
593 VexSimdPrefix pre, bool vector256 = false) { |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
594 int dst_enc = dst->encoding(); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
595 int nds_enc = nds->is_valid() ? nds->encoding() : 0; |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
596 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, false, vector256); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
597 } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
598 |
| 4759 | 599 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, |
| 600 VexSimdPrefix pre, VexOpcode opc, | |
| 601 bool vex_w, bool vector256); | |
| 602 | |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
603 int vex_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
604 VexSimdPrefix pre, bool vector256 = false, |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
605 VexOpcode opc = VEX_OPCODE_0F) { |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
606 int src_enc = src->encoding(); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
607 int dst_enc = dst->encoding(); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
608 int nds_enc = nds->is_valid() ? nds->encoding() : 0; |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
609 return vex_prefix_and_encode(dst_enc, nds_enc, src_enc, pre, opc, false, vector256); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
610 } |
| 4759 | 611 |
| 612 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, | |
| 613 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, | |
| 614 bool rex_w = false, bool vector256 = false); | |
| 615 | |
| 616 void simd_prefix(XMMRegister dst, Address src, | |
| 617 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { | |
| 618 simd_prefix(dst, xnoreg, src, pre, opc); | |
| 619 } | |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
620 |
| 4759 | 621 void simd_prefix(Address dst, XMMRegister src, VexSimdPrefix pre) { |
| 622 simd_prefix(src, dst, pre); | |
| 623 } | |
| 624 void simd_prefix_q(XMMRegister dst, XMMRegister nds, Address src, | |
| 625 VexSimdPrefix pre) { | |
| 626 bool rex_w = true; | |
| 627 simd_prefix(dst, nds, src, pre, VEX_OPCODE_0F, rex_w); | |
| 628 } | |
| 629 | |
| 630 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, | |
| 631 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, | |
| 632 bool rex_w = false, bool vector256 = false); | |
| 633 | |
| 634 // Move/convert 32-bit integer value. | |
| 635 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, Register src, | |
| 636 VexSimdPrefix pre) { | |
| 637 // It is OK to cast from Register to XMMRegister to pass argument here | |
| 638 // since only encoding is used in simd_prefix_and_encode() and number of | |
| 639 // Gen and Xmm registers are the same. | |
| 640 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre); | |
| 641 } | |
| 642 int simd_prefix_and_encode(XMMRegister dst, Register src, VexSimdPrefix pre) { | |
| 643 return simd_prefix_and_encode(dst, xnoreg, src, pre); | |
| 644 } | |
| 645 int simd_prefix_and_encode(Register dst, XMMRegister src, | |
| 646 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { | |
| 647 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc); | |
| 648 } | |
| 649 | |
| 650 // Move/convert 64-bit integer value. | |
| 651 int simd_prefix_and_encode_q(XMMRegister dst, XMMRegister nds, Register src, | |
| 652 VexSimdPrefix pre) { | |
| 653 bool rex_w = true; | |
| 654 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, VEX_OPCODE_0F, rex_w); | |
| 655 } | |
| 656 int simd_prefix_and_encode_q(XMMRegister dst, Register src, VexSimdPrefix pre) { | |
| 657 return simd_prefix_and_encode_q(dst, xnoreg, src, pre); | |
| 658 } | |
| 659 int simd_prefix_and_encode_q(Register dst, XMMRegister src, | |
| 660 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { | |
| 661 bool rex_w = true; | |
| 662 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc, rex_w); | |
| 663 } | |
| 664 | |
| 304 | 665 // Helper functions for groups of instructions |
| 666 void emit_arith_b(int op1, int op2, Register dst, int imm8); | |
| 667 | |
| 668 void emit_arith(int op1, int op2, Register dst, int32_t imm32); | |
|
4947
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
669 // Force generation of a 4 byte immediate value even if it fits into 8bit |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
670 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32); |
| 304 | 671 // only 32bit?? |
| 672 void emit_arith(int op1, int op2, Register dst, jobject obj); | |
| 673 void emit_arith(int op1, int op2, Register dst, Register src); | |
| 674 | |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
675 void emit_simd_arith(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
676 void emit_simd_arith(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
677 void emit_simd_arith_nonds(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
678 void emit_simd_arith_nonds(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
679 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
680 Address src, VexSimdPrefix pre, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
681 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
682 XMMRegister src, VexSimdPrefix pre, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
683 |
| 304 | 684 void emit_operand(Register reg, |
| 685 Register base, Register index, Address::ScaleFactor scale, | |
| 686 int disp, | |
| 687 RelocationHolder const& rspec, | |
| 688 int rip_relative_correction = 0); | |
| 689 | |
| 690 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); | |
| 691 | |
| 692 // operands that only take the original 32bit registers | |
| 693 void emit_operand32(Register reg, Address adr); | |
| 694 | |
| 695 void emit_operand(XMMRegister reg, | |
| 696 Register base, Register index, Address::ScaleFactor scale, | |
| 697 int disp, | |
| 698 RelocationHolder const& rspec); | |
| 699 | |
| 700 void emit_operand(XMMRegister reg, Address adr); | |
| 701 | |
| 702 void emit_operand(MMXRegister reg, Address adr); | |
| 703 | |
| 704 // workaround gcc (3.2.1-7) bug | |
| 705 void emit_operand(Address adr, MMXRegister reg); | |
| 706 | |
| 707 | |
| 708 // Immediate-to-memory forms | |
| 709 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); | |
| 710 | |
| 711 void emit_farith(int b1, int b2, int i); | |
| 712 | |
| 713 | |
| 714 protected: | |
| 715 #ifdef ASSERT | |
| 716 void check_relocation(RelocationHolder const& rspec, int format); | |
| 717 #endif | |
| 718 | |
| 719 inline void emit_long64(jlong x); | |
| 720 | |
| 721 void emit_data(jint data, relocInfo::relocType rtype, int format); | |
| 722 void emit_data(jint data, RelocationHolder const& rspec, int format); | |
| 723 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); | |
| 724 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); | |
| 725 | |
| 726 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); | |
| 727 | |
| 728 // These are all easily abused and hence protected | |
| 729 | |
| 730 // 32BIT ONLY SECTION | |
| 731 #ifndef _LP64 | |
| 732 // Make these disappear in 64bit mode since they would never be correct | |
| 733 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 734 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 735 | |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
736 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY |
| 304 | 737 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY |
| 738 | |
| 739 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY | |
| 740 #else | |
| 741 // 64BIT ONLY SECTION | |
| 742 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
|
743 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
744 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
|
745 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
|
746 |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
747 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
|
748 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); |
| 304 | 749 #endif // _LP64 |
| 750 | |
| 751 // These are unique in that we are ensured by the caller that the 32bit | |
| 752 // relative in these instructions will always be able to reach the potentially | |
| 753 // 64bit address described by entry. Since they can take a 64bit address they | |
| 754 // don't have the 32 suffix like the other instructions in this class. | |
| 755 | |
| 756 void call_literal(address entry, RelocationHolder const& rspec); | |
| 757 void jmp_literal(address entry, RelocationHolder const& rspec); | |
| 758 | |
| 759 // Avoid using directly section | |
| 760 // Instructions in this section are actually usable by anyone without danger | |
| 761 // of failure but have performance issues that are addressed my enhanced | |
| 762 // instructions which will do the proper thing base on the particular cpu. | |
| 763 // We protect them because we don't trust you... | |
| 764 | |
| 765 // Don't use next inc() and dec() methods directly. INC & DEC instructions | |
| 766 // could cause a partial flag stall since they don't set CF flag. | |
| 767 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods | |
| 768 // which call inc() & dec() or add() & sub() in accordance with | |
| 769 // the product flag UseIncDec value. | |
| 770 | |
| 771 void decl(Register dst); | |
| 772 void decl(Address dst); | |
| 773 void decq(Register dst); | |
| 774 void decq(Address dst); | |
| 775 | |
| 776 void incl(Register dst); | |
| 777 void incl(Address dst); | |
| 778 void incq(Register dst); | |
| 779 void incq(Address dst); | |
| 780 | |
| 781 // New cpus require use of movsd and movss to avoid partial register stall | |
| 782 // when loading from memory. But for old Opteron use movlpd instead of movsd. | |
| 783 // The selection is done in MacroAssembler::movdbl() and movflt(). | |
| 784 | |
| 785 // Move Scalar Single-Precision Floating-Point Values | |
| 786 void movss(XMMRegister dst, Address src); | |
| 787 void movss(XMMRegister dst, XMMRegister src); | |
| 788 void movss(Address dst, XMMRegister src); | |
| 789 | |
| 790 // Move Scalar Double-Precision Floating-Point Values | |
| 791 void movsd(XMMRegister dst, Address src); | |
| 792 void movsd(XMMRegister dst, XMMRegister src); | |
| 793 void movsd(Address dst, XMMRegister src); | |
| 794 void movlpd(XMMRegister dst, Address src); | |
| 795 | |
| 796 // New cpus require use of movaps and movapd to avoid partial register stall | |
| 797 // when moving between registers. | |
| 798 void movaps(XMMRegister dst, XMMRegister src); | |
| 799 void movapd(XMMRegister dst, XMMRegister src); | |
| 800 | |
| 801 // End avoid using directly | |
| 802 | |
| 803 | |
| 804 // Instruction prefixes | |
| 805 void prefix(Prefix p); | |
| 806 | |
| 0 | 807 public: |
| 808 | |
| 809 // Creation | |
| 810 Assembler(CodeBuffer* code) : AbstractAssembler(code) {} | |
| 811 | |
| 812 // Decoding | |
| 813 static address locate_operand(address inst, WhichOperand which); | |
| 814 static address locate_next_instruction(address inst); | |
| 815 | |
| 304 | 816 // Utilities |
|
2404
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
817 static bool is_polling_page_far() NOT_LP64({ return false;}); |
|
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
818 |
| 304 | 819 // Generic instructions |
| 820 // Does 32bit or 64bit as needed for the platform. In some sense these | |
| 821 // belong in macro assembler but there is no need for both varieties to exist | |
| 822 | |
| 823 void lea(Register dst, Address src); | |
| 824 | |
| 825 void mov(Register dst, Register src); | |
| 826 | |
| 827 void pusha(); | |
| 828 void popa(); | |
| 829 | |
| 830 void pushf(); | |
| 831 void popf(); | |
| 832 | |
| 833 void push(int32_t imm32); | |
| 834 | |
| 835 void push(Register src); | |
| 836 | |
| 837 void pop(Register dst); | |
| 838 | |
| 839 // These are dummies to prevent surprise implicit conversions to Register | |
| 840 void push(void* v); | |
| 841 void pop(void* v); | |
| 842 | |
| 843 // These do register sized moves/scans | |
| 844 void rep_mov(); | |
| 845 void rep_set(); | |
| 846 void repne_scan(); | |
| 847 #ifdef _LP64 | |
| 848 void repne_scanl(); | |
| 849 #endif | |
| 850 | |
| 851 // Vanilla instructions in lexical order | |
| 852 | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
853 void adcl(Address dst, int32_t imm32); |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
854 void adcl(Address dst, Register src); |
| 304 | 855 void adcl(Register dst, int32_t imm32); |
| 0 | 856 void adcl(Register dst, Address src); |
| 857 void adcl(Register dst, Register src); | |
| 858 | |
| 304 | 859 void adcq(Register dst, int32_t imm32); |
| 860 void adcq(Register dst, Address src); | |
| 861 void adcq(Register dst, Register src); | |
| 862 | |
| 863 void addl(Address dst, int32_t imm32); | |
| 0 | 864 void addl(Address dst, Register src); |
| 304 | 865 void addl(Register dst, int32_t imm32); |
| 0 | 866 void addl(Register dst, Address src); |
| 867 void addl(Register dst, Register src); | |
| 868 | |
| 304 | 869 void addq(Address dst, int32_t imm32); |
| 870 void addq(Address dst, Register src); | |
| 871 void addq(Register dst, int32_t imm32); | |
| 872 void addq(Register dst, Address src); | |
| 873 void addq(Register dst, Register src); | |
| 874 | |
| 0 | 875 void addr_nop_4(); |
| 876 void addr_nop_5(); | |
| 877 void addr_nop_7(); | |
| 878 void addr_nop_8(); | |
| 879 | |
| 304 | 880 // Add Scalar Double-Precision Floating-Point Values |
| 881 void addsd(XMMRegister dst, Address src); | |
| 882 void addsd(XMMRegister dst, XMMRegister src); | |
| 883 | |
| 884 // Add Scalar Single-Precision Floating-Point Values | |
| 885 void addss(XMMRegister dst, Address src); | |
| 886 void addss(XMMRegister dst, XMMRegister src); | |
| 887 | |
| 4759 | 888 void andl(Address dst, int32_t imm32); |
| 304 | 889 void andl(Register dst, int32_t imm32); |
| 890 void andl(Register dst, Address src); | |
| 891 void andl(Register dst, Register src); | |
| 892 | |
| 3783 | 893 void andq(Address dst, int32_t imm32); |
| 304 | 894 void andq(Register dst, int32_t imm32); |
| 895 void andq(Register dst, Address src); | |
| 896 void andq(Register dst, Register src); | |
| 897 | |
|
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
898 void bsfl(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
899 void bsrl(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
900 |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
901 #ifdef _LP64 |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
902 void bsfq(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
903 void bsrq(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
904 #endif |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
905 |
| 304 | 906 void bswapl(Register reg); |
| 907 | |
| 908 void bswapq(Register reg); | |
| 909 | |
| 0 | 910 void call(Label& L, relocInfo::relocType rtype); |
| 911 void call(Register reg); // push pc; pc <- reg | |
| 912 void call(Address adr); // push pc; pc <- adr | |
| 913 | |
| 304 | 914 void cdql(); |
| 915 | |
| 916 void cdqq(); | |
| 917 | |
| 918 void cld() { emit_byte(0xfc); } | |
| 919 | |
| 920 void clflush(Address adr); | |
| 921 | |
| 922 void cmovl(Condition cc, Register dst, Register src); | |
| 923 void cmovl(Condition cc, Register dst, Address src); | |
| 924 | |
| 925 void cmovq(Condition cc, Register dst, Register src); | |
| 926 void cmovq(Condition cc, Register dst, Address src); | |
| 927 | |
| 928 | |
| 929 void cmpb(Address dst, int imm8); | |
| 930 | |
| 931 void cmpl(Address dst, int32_t imm32); | |
| 932 | |
| 933 void cmpl(Register dst, int32_t imm32); | |
| 934 void cmpl(Register dst, Register src); | |
| 935 void cmpl(Register dst, Address src); | |
| 936 | |
| 937 void cmpq(Address dst, int32_t imm32); | |
| 938 void cmpq(Address dst, Register src); | |
| 939 | |
| 940 void cmpq(Register dst, int32_t imm32); | |
| 941 void cmpq(Register dst, Register src); | |
| 942 void cmpq(Register dst, Address src); | |
| 943 | |
| 944 // these are dummies used to catch attempting to convert NULL to Register | |
| 945 void cmpl(Register dst, void* junk); // dummy | |
| 946 void cmpq(Register dst, void* junk); // dummy | |
| 947 | |
| 948 void cmpw(Address dst, int imm16); | |
| 949 | |
| 950 void cmpxchg8 (Address adr); | |
| 951 | |
| 952 void cmpxchgl(Register reg, Address adr); | |
| 953 | |
| 954 void cmpxchgq(Register reg, Address adr); | |
| 955 | |
| 956 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS | |
| 957 void comisd(XMMRegister dst, Address src); | |
| 4759 | 958 void comisd(XMMRegister dst, XMMRegister src); |
| 304 | 959 |
| 960 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS | |
| 961 void comiss(XMMRegister dst, Address src); | |
| 4759 | 962 void comiss(XMMRegister dst, XMMRegister src); |
| 304 | 963 |
| 964 // Identify processor type and features | |
| 965 void cpuid() { | |
| 966 emit_byte(0x0F); | |
| 967 emit_byte(0xA2); | |
| 968 } | |
| 969 | |
| 970 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value | |
| 971 void cvtsd2ss(XMMRegister dst, XMMRegister src); | |
| 4759 | 972 void cvtsd2ss(XMMRegister dst, Address src); |
| 304 | 973 |
| 974 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value | |
| 975 void cvtsi2sdl(XMMRegister dst, Register src); | |
| 4759 | 976 void cvtsi2sdl(XMMRegister dst, Address src); |
| 304 | 977 void cvtsi2sdq(XMMRegister dst, Register src); |
| 4759 | 978 void cvtsi2sdq(XMMRegister dst, Address src); |
| 304 | 979 |
| 980 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value | |
| 981 void cvtsi2ssl(XMMRegister dst, Register src); | |
| 4759 | 982 void cvtsi2ssl(XMMRegister dst, Address src); |
| 304 | 983 void cvtsi2ssq(XMMRegister dst, Register src); |
| 4759 | 984 void cvtsi2ssq(XMMRegister dst, Address src); |
| 304 | 985 |
| 986 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value | |
| 987 void cvtdq2pd(XMMRegister dst, XMMRegister src); | |
| 988 | |
| 989 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value | |
| 990 void cvtdq2ps(XMMRegister dst, XMMRegister src); | |
| 991 | |
| 992 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value | |
| 993 void cvtss2sd(XMMRegister dst, XMMRegister src); | |
| 4759 | 994 void cvtss2sd(XMMRegister dst, Address src); |
| 304 | 995 |
| 996 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer | |
| 997 void cvttsd2sil(Register dst, Address src); | |
| 998 void cvttsd2sil(Register dst, XMMRegister src); | |
| 999 void cvttsd2siq(Register dst, XMMRegister src); | |
| 1000 | |
| 1001 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer | |
| 1002 void cvttss2sil(Register dst, XMMRegister src); | |
| 1003 void cvttss2siq(Register dst, XMMRegister src); | |
| 1004 | |
| 1005 // Divide Scalar Double-Precision Floating-Point Values | |
| 1006 void divsd(XMMRegister dst, Address src); | |
| 1007 void divsd(XMMRegister dst, XMMRegister src); | |
| 1008 | |
| 1009 // Divide Scalar Single-Precision Floating-Point Values | |
| 1010 void divss(XMMRegister dst, Address src); | |
| 1011 void divss(XMMRegister dst, XMMRegister src); | |
| 1012 | |
| 1013 void emms(); | |
| 1014 | |
| 1015 void fabs(); | |
| 1016 | |
| 1017 void fadd(int i); | |
| 1018 | |
| 1019 void fadd_d(Address src); | |
| 1020 void fadd_s(Address src); | |
| 1021 | |
| 1022 // "Alternate" versions of x87 instructions place result down in FPU | |
| 1023 // stack instead of on TOS | |
| 1024 | |
| 1025 void fadda(int i); // "alternate" fadd | |
| 1026 void faddp(int i = 1); | |
| 1027 | |
| 1028 void fchs(); | |
| 1029 | |
| 1030 void fcom(int i); | |
| 1031 | |
| 1032 void fcomp(int i = 1); | |
| 1033 void fcomp_d(Address src); | |
| 1034 void fcomp_s(Address src); | |
| 1035 | |
| 1036 void fcompp(); | |
| 1037 | |
| 1038 void fcos(); | |
| 1039 | |
| 1040 void fdecstp(); | |
| 1041 | |
| 1042 void fdiv(int i); | |
| 1043 void fdiv_d(Address src); | |
| 1044 void fdivr_s(Address src); | |
| 1045 void fdiva(int i); // "alternate" fdiv | |
| 1046 void fdivp(int i = 1); | |
| 1047 | |
| 1048 void fdivr(int i); | |
| 1049 void fdivr_d(Address src); | |
| 1050 void fdiv_s(Address src); | |
| 1051 | |
| 1052 void fdivra(int i); // "alternate" reversed fdiv | |
| 1053 | |
| 1054 void fdivrp(int i = 1); | |
| 1055 | |
| 1056 void ffree(int i = 0); | |
| 1057 | |
| 1058 void fild_d(Address adr); | |
| 1059 void fild_s(Address adr); | |
| 1060 | |
| 1061 void fincstp(); | |
| 1062 | |
| 1063 void finit(); | |
| 1064 | |
| 1065 void fist_s (Address adr); | |
| 1066 void fistp_d(Address adr); | |
| 1067 void fistp_s(Address adr); | |
| 1068 | |
| 1069 void fld1(); | |
| 1070 | |
| 1071 void fld_d(Address adr); | |
| 1072 void fld_s(Address adr); | |
| 1073 void fld_s(int index); | |
| 1074 void fld_x(Address adr); // extended-precision (80-bit) format | |
| 1075 | |
| 1076 void fldcw(Address src); | |
| 1077 | |
| 1078 void fldenv(Address src); | |
| 1079 | |
| 1080 void fldlg2(); | |
| 1081 | |
| 1082 void fldln2(); | |
| 1083 | |
| 1084 void fldz(); | |
| 1085 | |
| 1086 void flog(); | |
| 1087 void flog10(); | |
| 1088 | |
| 1089 void fmul(int i); | |
| 1090 | |
| 1091 void fmul_d(Address src); | |
| 1092 void fmul_s(Address src); | |
| 1093 | |
| 1094 void fmula(int i); // "alternate" fmul | |
| 1095 | |
| 1096 void fmulp(int i = 1); | |
| 1097 | |
| 1098 void fnsave(Address dst); | |
| 1099 | |
| 1100 void fnstcw(Address src); | |
| 1101 | |
| 1102 void fnstsw_ax(); | |
| 1103 | |
| 1104 void fprem(); | |
| 1105 void fprem1(); | |
| 1106 | |
| 1107 void frstor(Address src); | |
| 1108 | |
| 1109 void fsin(); | |
| 1110 | |
| 1111 void fsqrt(); | |
| 1112 | |
| 1113 void fst_d(Address adr); | |
| 1114 void fst_s(Address adr); | |
| 1115 | |
| 1116 void fstp_d(Address adr); | |
| 1117 void fstp_d(int index); | |
| 1118 void fstp_s(Address adr); | |
| 1119 void fstp_x(Address adr); // extended-precision (80-bit) format | |
| 1120 | |
| 1121 void fsub(int i); | |
| 1122 void fsub_d(Address src); | |
| 1123 void fsub_s(Address src); | |
| 1124 | |
| 1125 void fsuba(int i); // "alternate" fsub | |
| 1126 | |
| 1127 void fsubp(int i = 1); | |
| 1128 | |
| 1129 void fsubr(int i); | |
| 1130 void fsubr_d(Address src); | |
| 1131 void fsubr_s(Address src); | |
| 1132 | |
| 1133 void fsubra(int i); // "alternate" reversed fsub | |
| 1134 | |
| 1135 void fsubrp(int i = 1); | |
| 1136 | |
| 1137 void ftan(); | |
| 1138 | |
| 1139 void ftst(); | |
| 1140 | |
| 1141 void fucomi(int i = 1); | |
| 1142 void fucomip(int i = 1); | |
| 1143 | |
| 1144 void fwait(); | |
| 1145 | |
| 1146 void fxch(int i = 1); | |
| 1147 | |
| 1148 void fxrstor(Address src); | |
| 1149 | |
| 1150 void fxsave(Address dst); | |
| 1151 | |
| 1152 void fyl2x(); | |
|
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
1153 void frndint(); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
1154 void f2xm1(); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
1155 void fldl2e(); |
| 304 | 1156 |
| 1157 void hlt(); | |
| 1158 | |
| 1159 void idivl(Register src); | |
| 1920 | 1160 void divl(Register src); // Unsigned division |
| 304 | 1161 |
| 1162 void idivq(Register src); | |
| 1163 | |
| 1164 void imull(Register dst, Register src); | |
| 1165 void imull(Register dst, Register src, int value); | |
| 1166 | |
| 1167 void imulq(Register dst, Register src); | |
| 1168 void imulq(Register dst, Register src, int value); | |
| 1169 | |
| 0 | 1170 |
| 1171 // jcc is the generic conditional branch generator to run- | |
| 1172 // time routines, jcc is used for branches to labels. jcc | |
| 1173 // takes a branch opcode (cc) and a label (L) and generates | |
| 1174 // either a backward branch or a forward branch and links it | |
| 1175 // to the label fixup chain. Usage: | |
| 1176 // | |
| 1177 // Label L; // unbound label | |
| 1178 // jcc(cc, L); // forward branch to unbound label | |
| 1179 // bind(L); // bind label to the current pc | |
| 1180 // jcc(cc, L); // backward branch to bound label | |
| 1181 // bind(L); // illegal: a label may be bound only once | |
| 1182 // | |
| 1183 // Note: The same Label can be used for forward and backward branches | |
| 1184 // but it may be bound only once. | |
| 1185 | |
| 3851 | 1186 void jcc(Condition cc, Label& L, bool maybe_short = true); |
| 0 | 1187 |
| 1188 // Conditional jump to a 8-bit offset to L. | |
| 1189 // WARNING: be very careful using this for forward jumps. If the label is | |
| 1190 // not bound within an 8-bit offset of this instruction, a run-time error | |
| 1191 // will occur. | |
| 1192 void jccb(Condition cc, Label& L); | |
| 1193 | |
| 304 | 1194 void jmp(Address entry); // pc <- entry |
| 1195 | |
| 1196 // Label operations & relative jumps (PPUM Appendix D) | |
| 3851 | 1197 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L |
| 304 | 1198 |
| 1199 void jmp(Register entry); // pc <- entry | |
| 1200 | |
| 1201 // Unconditional 8-bit offset jump to L. | |
| 1202 // WARNING: be very careful using this for forward jumps. If the label is | |
| 1203 // not bound within an 8-bit offset of this instruction, a run-time error | |
| 1204 // will occur. | |
| 1205 void jmpb(Label& L); | |
| 1206 | |
| 1207 void ldmxcsr( Address src ); | |
| 1208 | |
| 1209 void leal(Register dst, Address src); | |
| 1210 | |
| 1211 void leaq(Register dst, Address src); | |
| 1212 | |
| 1213 void lfence() { | |
| 1214 emit_byte(0x0F); | |
| 1215 emit_byte(0xAE); | |
| 1216 emit_byte(0xE8); | |
| 1217 } | |
| 1218 | |
| 1219 void lock(); | |
| 1220 | |
|
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1221 void lzcntl(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1222 |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1223 #ifdef _LP64 |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1224 void lzcntq(Register dst, Register src); |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1225 #endif |
|
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1226 |
| 304 | 1227 enum Membar_mask_bits { |
| 1228 StoreStore = 1 << 3, | |
| 1229 LoadStore = 1 << 2, | |
| 1230 StoreLoad = 1 << 1, | |
| 1231 LoadLoad = 1 << 0 | |
| 1232 }; | |
| 1233 | |
|
671
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1234 // Serializes memory and blows flags |
| 304 | 1235 void membar(Membar_mask_bits order_constraint) { |
|
671
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1236 if (os::is_MP()) { |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1237 // We only have to handle StoreLoad |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1238 if (order_constraint & StoreLoad) { |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1239 // All usable chips support "locked" instructions which suffice |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1240 // as barriers, and are much faster than the alternative of |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1241 // using cpuid instruction. We use here a locked add [esp],0. |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1242 // This is conveniently otherwise a no-op except for blowing |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1243 // flags. |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1244 // Any change to this code may need to revisit other places in |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1245 // the code where this idiom is used, in particular the |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1246 // orderAccess code. |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1247 lock(); |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1248 addl(Address(rsp, 0), 0);// Assert the lock# signal here |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1249 } |
|
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1250 } |
| 304 | 1251 } |
| 1252 | |
| 1253 void mfence(); | |
| 1254 | |
| 1255 // Moves | |
| 1256 | |
| 1257 void mov64(Register dst, int64_t imm64); | |
| 1258 | |
| 1259 void movb(Address dst, Register src); | |
| 1260 void movb(Address dst, int imm8); | |
| 1261 void movb(Register dst, Address src); | |
| 1262 | |
| 1263 void movdl(XMMRegister dst, Register src); | |
| 1264 void movdl(Register dst, XMMRegister src); | |
|
2320
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
1265 void movdl(XMMRegister dst, Address src); |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1266 void movdl(Address dst, XMMRegister src); |
| 304 | 1267 |
| 1268 // Move Double Quadword | |
| 1269 void movdq(XMMRegister dst, Register src); | |
| 1270 void movdq(Register dst, XMMRegister src); | |
| 1271 | |
| 1272 // Move Aligned Double Quadword | |
| 1273 void movdqa(XMMRegister dst, XMMRegister src); | |
| 1274 | |
| 405 | 1275 // Move Unaligned Double Quadword |
| 1276 void movdqu(Address dst, XMMRegister src); | |
| 1277 void movdqu(XMMRegister dst, Address src); | |
| 1278 void movdqu(XMMRegister dst, XMMRegister src); | |
| 1279 | |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1280 // Move Unaligned 256bit Vector |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1281 void vmovdqu(Address dst, XMMRegister src); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1282 void vmovdqu(XMMRegister dst, Address src); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1283 void vmovdqu(XMMRegister dst, XMMRegister src); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1284 |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1285 // Move lower 64bit to high 64bit in 128bit register |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1286 void movlhps(XMMRegister dst, XMMRegister src); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1287 |
| 304 | 1288 void movl(Register dst, int32_t imm32); |
| 1289 void movl(Address dst, int32_t imm32); | |
| 1290 void movl(Register dst, Register src); | |
| 1291 void movl(Register dst, Address src); | |
| 1292 void movl(Address dst, Register src); | |
| 1293 | |
| 1294 // These dummies prevent using movl from converting a zero (like NULL) into Register | |
| 1295 // by giving the compiler two choices it can't resolve | |
| 1296 | |
| 1297 void movl(Address dst, void* junk); | |
| 1298 void movl(Register dst, void* junk); | |
| 1299 | |
| 1300 #ifdef _LP64 | |
| 1301 void movq(Register dst, Register src); | |
| 1302 void movq(Register dst, Address src); | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
1303 void movq(Address dst, Register src); |
| 304 | 1304 #endif |
| 1305 | |
| 1306 void movq(Address dst, MMXRegister src ); | |
| 1307 void movq(MMXRegister dst, Address src ); | |
| 1308 | |
| 1309 #ifdef _LP64 | |
| 1310 // These dummies prevent using movq from converting a zero (like NULL) into Register | |
| 1311 // by giving the compiler two choices it can't resolve | |
| 1312 | |
| 1313 void movq(Address dst, void* dummy); | |
| 1314 void movq(Register dst, void* dummy); | |
| 1315 #endif | |
| 1316 | |
| 1317 // Move Quadword | |
| 1318 void movq(Address dst, XMMRegister src); | |
| 1319 void movq(XMMRegister dst, Address src); | |
| 1320 | |
| 1321 void movsbl(Register dst, Address src); | |
| 1322 void movsbl(Register dst, Register src); | |
| 1323 | |
| 1324 #ifdef _LP64 | |
| 624 | 1325 void movsbq(Register dst, Address src); |
| 1326 void movsbq(Register dst, Register src); | |
| 1327 | |
| 304 | 1328 // Move signed 32bit immediate to 64bit extending sign |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
1329 void movslq(Address dst, int32_t imm64); |
| 304 | 1330 void movslq(Register dst, int32_t imm64); |
| 1331 | |
| 1332 void movslq(Register dst, Address src); | |
| 1333 void movslq(Register dst, Register src); | |
| 1334 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous | |
| 1335 #endif | |
| 1336 | |
| 1337 void movswl(Register dst, Address src); | |
| 1338 void movswl(Register dst, Register src); | |
| 1339 | |
| 624 | 1340 #ifdef _LP64 |
| 1341 void movswq(Register dst, Address src); | |
| 1342 void movswq(Register dst, Register src); | |
| 1343 #endif | |
| 1344 | |
| 304 | 1345 void movw(Address dst, int imm16); |
| 1346 void movw(Register dst, Address src); | |
| 1347 void movw(Address dst, Register src); | |
| 1348 | |
| 1349 void movzbl(Register dst, Address src); | |
| 1350 void movzbl(Register dst, Register src); | |
| 1351 | |
| 624 | 1352 #ifdef _LP64 |
| 1353 void movzbq(Register dst, Address src); | |
| 1354 void movzbq(Register dst, Register src); | |
| 1355 #endif | |
| 1356 | |
| 304 | 1357 void movzwl(Register dst, Address src); |
| 1358 void movzwl(Register dst, Register src); | |
| 1359 | |
| 624 | 1360 #ifdef _LP64 |
| 1361 void movzwq(Register dst, Address src); | |
| 1362 void movzwq(Register dst, Register src); | |
| 1363 #endif | |
| 1364 | |
| 304 | 1365 void mull(Address src); |
| 1366 void mull(Register src); | |
| 1367 | |
| 1368 // Multiply Scalar Double-Precision Floating-Point Values | |
| 1369 void mulsd(XMMRegister dst, Address src); | |
| 1370 void mulsd(XMMRegister dst, XMMRegister src); | |
| 1371 | |
| 1372 // Multiply Scalar Single-Precision Floating-Point Values | |
| 1373 void mulss(XMMRegister dst, Address src); | |
| 1374 void mulss(XMMRegister dst, XMMRegister src); | |
| 1375 | |
| 1376 void negl(Register dst); | |
| 1377 | |
| 1378 #ifdef _LP64 | |
| 1379 void negq(Register dst); | |
| 1380 #endif | |
| 1381 | |
| 1382 void nop(int i = 1); | |
| 1383 | |
| 1384 void notl(Register dst); | |
| 1385 | |
| 1386 #ifdef _LP64 | |
| 1387 void notq(Register dst); | |
| 1388 #endif | |
| 1389 | |
| 1390 void orl(Address dst, int32_t imm32); | |
| 1391 void orl(Register dst, int32_t imm32); | |
| 1392 void orl(Register dst, Address src); | |
| 1393 void orl(Register dst, Register src); | |
| 1394 | |
| 1395 void orq(Address dst, int32_t imm32); | |
| 1396 void orq(Register dst, int32_t imm32); | |
| 1397 void orq(Register dst, Address src); | |
| 1398 void orq(Register dst, Register src); | |
| 1399 | |
| 4759 | 1400 // Pack with unsigned saturation |
| 1401 void packuswb(XMMRegister dst, XMMRegister src); | |
| 1402 void packuswb(XMMRegister dst, Address src); | |
| 1403 | |
| 681 | 1404 // SSE4.2 string instructions |
| 1405 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); | |
| 1406 void pcmpestri(XMMRegister xmm1, Address src, int imm8); | |
| 1407 | |
| 4759 | 1408 // SSE4.1 packed move |
| 1409 void pmovzxbw(XMMRegister dst, XMMRegister src); | |
| 1410 void pmovzxbw(XMMRegister dst, Address src); | |
| 1411 | |
| 1060 | 1412 #ifndef _LP64 // no 32bit push/pop on amd64 |
| 304 | 1413 void popl(Address dst); |
| 1060 | 1414 #endif |
| 304 | 1415 |
| 1416 #ifdef _LP64 | |
| 1417 void popq(Address dst); | |
| 1418 #endif | |
| 1419 | |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1420 void popcntl(Register dst, Address src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1421 void popcntl(Register dst, Register src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1422 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1423 #ifdef _LP64 |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1424 void popcntq(Register dst, Address src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1425 void popcntq(Register dst, Register src); |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1426 #endif |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
1427 |
| 304 | 1428 // Prefetches (SSE, SSE2, 3DNOW only) |
| 1429 | |
| 1430 void prefetchnta(Address src); | |
| 1431 void prefetchr(Address src); | |
| 1432 void prefetcht0(Address src); | |
| 1433 void prefetcht1(Address src); | |
| 1434 void prefetcht2(Address src); | |
| 1435 void prefetchw(Address src); | |
| 1436 | |
| 1437 // Shuffle Packed Doublewords | |
| 1438 void pshufd(XMMRegister dst, XMMRegister src, int mode); | |
| 1439 void pshufd(XMMRegister dst, Address src, int mode); | |
| 1440 | |
| 1441 // Shuffle Packed Low Words | |
| 1442 void pshuflw(XMMRegister dst, XMMRegister src, int mode); | |
| 1443 void pshuflw(XMMRegister dst, Address src, int mode); | |
| 1444 | |
|
2320
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
1445 // Shift Right by bytes Logical DoubleQuadword Immediate |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
1446 void psrldq(XMMRegister dst, int shift); |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
1447 |
| 681 | 1448 // Logical Compare Double Quadword |
| 1449 void ptest(XMMRegister dst, XMMRegister src); | |
| 1450 void ptest(XMMRegister dst, Address src); | |
| 1451 | |
| 304 | 1452 // Interleave Low Bytes |
| 1453 void punpcklbw(XMMRegister dst, XMMRegister src); | |
| 4759 | 1454 void punpcklbw(XMMRegister dst, Address src); |
| 1455 | |
| 1456 // Interleave Low Doublewords | |
| 1457 void punpckldq(XMMRegister dst, XMMRegister src); | |
| 1458 void punpckldq(XMMRegister dst, Address src); | |
| 304 | 1459 |
| 6225 | 1460 // Interleave Low Quadwords |
| 1461 void punpcklqdq(XMMRegister dst, XMMRegister src); | |
| 1462 | |
| 1060 | 1463 #ifndef _LP64 // no 32bit push/pop on amd64 |
| 304 | 1464 void pushl(Address src); |
| 1060 | 1465 #endif |
| 304 | 1466 |
| 1467 void pushq(Address src); | |
| 1468 | |
| 1469 void rcll(Register dst, int imm8); | |
| 1470 | |
| 1471 void rclq(Register dst, int imm8); | |
| 1472 | |
| 1473 void ret(int imm16); | |
| 0 | 1474 |
| 1475 void sahf(); | |
| 1476 | |
| 304 | 1477 void sarl(Register dst, int imm8); |
| 1478 void sarl(Register dst); | |
| 1479 | |
| 1480 void sarq(Register dst, int imm8); | |
| 1481 void sarq(Register dst); | |
| 1482 | |
| 1483 void sbbl(Address dst, int32_t imm32); | |
| 1484 void sbbl(Register dst, int32_t imm32); | |
| 1485 void sbbl(Register dst, Address src); | |
| 1486 void sbbl(Register dst, Register src); | |
| 1487 | |
| 1488 void sbbq(Address dst, int32_t imm32); | |
| 1489 void sbbq(Register dst, int32_t imm32); | |
| 1490 void sbbq(Register dst, Address src); | |
| 1491 void sbbq(Register dst, Register src); | |
| 1492 | |
| 1493 void setb(Condition cc, Register dst); | |
| 1494 | |
| 1495 void shldl(Register dst, Register src); | |
| 1496 | |
| 1497 void shll(Register dst, int imm8); | |
| 1498 void shll(Register dst); | |
| 1499 | |
| 1500 void shlq(Register dst, int imm8); | |
| 1501 void shlq(Register dst); | |
| 1502 | |
| 1503 void shrdl(Register dst, Register src); | |
| 1504 | |
| 1505 void shrl(Register dst, int imm8); | |
| 1506 void shrl(Register dst); | |
| 1507 | |
| 1508 void shrq(Register dst, int imm8); | |
| 1509 void shrq(Register dst); | |
| 1510 | |
| 1511 void smovl(); // QQQ generic? | |
| 1512 | |
| 1513 // Compute Square Root of Scalar Double-Precision Floating-Point Value | |
| 1514 void sqrtsd(XMMRegister dst, Address src); | |
| 1515 void sqrtsd(XMMRegister dst, XMMRegister src); | |
| 1516 | |
| 2008 | 1517 // Compute Square Root of Scalar Single-Precision Floating-Point Value |
| 1518 void sqrtss(XMMRegister dst, Address src); | |
| 1519 void sqrtss(XMMRegister dst, XMMRegister src); | |
| 1520 | |
| 304 | 1521 void std() { emit_byte(0xfd); } |
| 1522 | |
| 1523 void stmxcsr( Address dst ); | |
| 1524 | |
| 1525 void subl(Address dst, int32_t imm32); | |
| 1526 void subl(Address dst, Register src); | |
| 1527 void subl(Register dst, int32_t imm32); | |
| 1528 void subl(Register dst, Address src); | |
| 1529 void subl(Register dst, Register src); | |
| 1530 | |
| 1531 void subq(Address dst, int32_t imm32); | |
| 1532 void subq(Address dst, Register src); | |
| 1533 void subq(Register dst, int32_t imm32); | |
| 1534 void subq(Register dst, Address src); | |
| 1535 void subq(Register dst, Register src); | |
| 1536 | |
|
4947
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
1537 // Force generation of a 4 byte immediate value even if it fits into 8bit |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
1538 void subl_imm32(Register dst, int32_t imm32); |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
1539 void subq_imm32(Register dst, int32_t imm32); |
| 304 | 1540 |
| 1541 // Subtract Scalar Double-Precision Floating-Point Values | |
| 1542 void subsd(XMMRegister dst, Address src); | |
| 0 | 1543 void subsd(XMMRegister dst, XMMRegister src); |
| 1544 | |
| 304 | 1545 // Subtract Scalar Single-Precision Floating-Point Values |
| 1546 void subss(XMMRegister dst, Address src); | |
| 1547 void subss(XMMRegister dst, XMMRegister src); | |
| 1548 | |
| 1549 void testb(Register dst, int imm8); | |
| 1550 | |
| 1551 void testl(Register dst, int32_t imm32); | |
| 1552 void testl(Register dst, Register src); | |
| 1553 void testl(Register dst, Address src); | |
| 1554 | |
| 1555 void testq(Register dst, int32_t imm32); | |
| 1556 void testq(Register dst, Register src); | |
| 1557 | |
| 1558 | |
| 1559 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS | |
| 1560 void ucomisd(XMMRegister dst, Address src); | |
| 0 | 1561 void ucomisd(XMMRegister dst, XMMRegister src); |
| 1562 | |
| 304 | 1563 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS |
| 1564 void ucomiss(XMMRegister dst, Address src); | |
| 1565 void ucomiss(XMMRegister dst, XMMRegister src); | |
| 1566 | |
| 1567 void xaddl(Address dst, Register src); | |
| 1568 | |
| 1569 void xaddq(Address dst, Register src); | |
| 1570 | |
| 1571 void xchgl(Register reg, Address adr); | |
| 1572 void xchgl(Register dst, Register src); | |
| 1573 | |
| 1574 void xchgq(Register reg, Address adr); | |
| 1575 void xchgq(Register dst, Register src); | |
| 1576 | |
| 4759 | 1577 // Get Value of Extended Control Register |
| 1578 void xgetbv() { | |
| 1579 emit_byte(0x0F); | |
| 1580 emit_byte(0x01); | |
| 1581 emit_byte(0xD0); | |
| 1582 } | |
| 1583 | |
| 304 | 1584 void xorl(Register dst, int32_t imm32); |
| 1585 void xorl(Register dst, Address src); | |
| 1586 void xorl(Register dst, Register src); | |
| 1587 | |
| 1588 void xorq(Register dst, Address src); | |
| 1589 void xorq(Register dst, Register src); | |
| 1590 | |
| 1591 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 | |
| 4759 | 1592 |
| 6225 | 1593 // AVX 3-operands scalar instructions (encoded with VEX prefix) |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1594 |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1595 void vaddsd(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1596 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1597 void vaddss(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1598 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1599 void vdivsd(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1600 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1601 void vdivss(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1602 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1603 void vmulsd(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1604 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1605 void vmulss(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1606 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1607 void vsubsd(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1608 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1609 void vsubss(XMMRegister dst, XMMRegister nds, Address src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1610 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1611 |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1612 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1613 //====================VECTOR ARITHMETIC===================================== |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1614 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1615 // Add Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1616 void addpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1617 void addps(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1618 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1619 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1620 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1621 void vaddps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1622 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1623 // Subtract Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1624 void subpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1625 void subps(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1626 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1627 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1628 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1629 void vsubps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1630 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1631 // Multiply Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1632 void mulpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1633 void mulps(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1634 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1635 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1636 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1637 void vmulps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1638 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1639 // Divide Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1640 void divpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1641 void divps(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1642 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1643 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1644 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1645 void vdivps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1646 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1647 // Bitwise Logical AND of Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1648 void andpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1649 void andps(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1650 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1651 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1652 void vandpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1653 void vandps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1654 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1655 // Bitwise Logical XOR of Packed Floating-Point Values |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1656 void xorpd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1657 void xorps(XMMRegister dst, XMMRegister src); |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1658 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1659 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1660 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1661 void vxorps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1662 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1663 // Add packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1664 void paddb(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1665 void paddw(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1666 void paddd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1667 void paddq(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1668 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1669 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1670 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1671 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1672 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1673 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1674 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1675 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1676 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1677 // Sub packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1678 void psubb(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1679 void psubw(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1680 void psubd(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1681 void psubq(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1682 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1683 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1684 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1685 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1686 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1687 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1688 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1689 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1690 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1691 // Multiply packed integers (only shorts and ints) |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1692 void pmullw(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1693 void pmulld(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1694 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1695 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1696 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1697 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1698 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1699 // Shift left packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1700 void psllw(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1701 void pslld(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1702 void psllq(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1703 void psllw(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1704 void pslld(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1705 void psllq(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1706 void vpsllw(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1707 void vpslld(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1708 void vpsllq(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1709 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1710 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1711 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1712 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1713 // Logical shift right packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1714 void psrlw(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1715 void psrld(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1716 void psrlq(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1717 void psrlw(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1718 void psrld(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1719 void psrlq(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1720 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1721 void vpsrld(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1722 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1723 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1724 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1725 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1726 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1727 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs) |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1728 void psraw(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1729 void psrad(XMMRegister dst, int shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1730 void psraw(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1731 void psrad(XMMRegister dst, XMMRegister shift); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1732 void vpsraw(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1733 void vpsrad(XMMRegister dst, XMMRegister src, int shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1734 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1735 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1736 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1737 // And packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1738 void pand(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1739 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1740 void vpand(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1741 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1742 // Or packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1743 void por(XMMRegister dst, XMMRegister src); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1744 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1745 void vpor(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1746 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1747 // Xor packed integers |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1748 void pxor(XMMRegister dst, XMMRegister src); |
| 6225 | 1749 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1750 void vpxor(XMMRegister dst, XMMRegister nds, Address src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1751 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
1752 // Copy low 128bit into high 128bit of YMM registers. |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1753 void vinsertf128h(XMMRegister dst, XMMRegister nds, XMMRegister src); |
| 6225 | 1754 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src); |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1755 |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1756 // AVX instruction which is used to clear upper 128 bits of YMM registers and |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1757 // to avoid transaction penalty between AVX and SSE states. There is no |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1758 // penalty if legacy SSE instructions are encoded using VEX prefix because |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1759 // they always clear upper 128 bits. It should be used before calling |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1760 // runtime code and native libraries. |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
1761 void vzeroupper(); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
1762 |
| 4759 | 1763 protected: |
| 1764 // Next instructions require address alignment 16 bytes SSE mode. | |
| 1765 // They should be called only from corresponding MacroAssembler instructions. | |
| 1766 void andpd(XMMRegister dst, Address src); | |
| 1767 void andps(XMMRegister dst, Address src); | |
| 1768 void xorpd(XMMRegister dst, Address src); | |
| 1769 void xorps(XMMRegister dst, Address src); | |
| 1770 | |
| 0 | 1771 }; |
| 1772 | |
| 1773 | |
| 1774 // MacroAssembler extends Assembler by frequently used macros. | |
| 1775 // | |
| 1776 // Instructions for which a 'better' code sequence exists depending | |
| 1777 // on arguments should also go in here. | |
| 1778 | |
| 1779 class MacroAssembler: public Assembler { | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1780 friend class LIR_Assembler; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
1781 friend class Runtime1; // as_Address() |
|
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
1782 |
| 0 | 1783 protected: |
| 1784 | |
| 1785 Address as_Address(AddressLiteral adr); | |
| 1786 Address as_Address(ArrayAddress adr); | |
| 1787 | |
| 1788 // Support for VM calls | |
| 1789 // | |
| 1790 // This is the base routine called by the different versions of call_VM_leaf. The interpreter | |
| 1791 // may customize this version by overriding it for its purposes (e.g., to save/restore | |
| 1792 // additional registers when doing a VM call). | |
| 1793 #ifdef CC_INTERP | |
| 1794 // c++ interpreter never wants to use interp_masm version of call_VM | |
| 1795 #define VIRTUAL | |
| 1796 #else | |
| 1797 #define VIRTUAL virtual | |
| 1798 #endif | |
| 1799 | |
| 1800 VIRTUAL void call_VM_leaf_base( | |
| 1801 address entry_point, // the entry point | |
| 1802 int number_of_arguments // the number of arguments to pop after the call | |
| 1803 ); | |
| 1804 | |
| 1805 // This is the base routine called by the different versions of call_VM. The interpreter | |
| 1806 // may customize this version by overriding it for its purposes (e.g., to save/restore | |
| 1807 // additional registers when doing a VM call). | |
| 1808 // | |
| 1809 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base | |
| 1810 // returns the register which contains the thread upon return. If a thread register has been | |
| 1811 // specified, the return value will correspond to that register. If no last_java_sp is specified | |
| 1812 // (noreg) than rsp will be used instead. | |
| 1813 VIRTUAL void call_VM_base( // returns the register containing the thread upon return | |
| 1814 Register oop_result, // where an oop-result ends up if any; use noreg otherwise | |
| 1815 Register java_thread, // the thread if computed before ; use noreg otherwise | |
| 1816 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise | |
| 1817 address entry_point, // the entry point | |
| 1818 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call | |
| 1819 bool check_exceptions // whether to check for pending exceptions after return | |
| 1820 ); | |
| 1821 | |
| 1822 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. | |
| 1823 // The implementation is only non-empty for the InterpreterMacroAssembler, | |
| 1824 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. | |
| 1825 virtual void check_and_handle_popframe(Register java_thread); | |
| 1826 virtual void check_and_handle_earlyret(Register java_thread); | |
| 1827 | |
| 1828 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); | |
| 1829 | |
| 1830 // helpers for FPU flag access | |
| 1831 // tmp is a temporary register, if none is available use noreg | |
| 1832 void save_rax (Register tmp); | |
| 1833 void restore_rax(Register tmp); | |
| 1834 | |
| 1835 public: | |
| 1836 MacroAssembler(CodeBuffer* code) : Assembler(code) {} | |
| 1837 | |
| 1838 // Support for NULL-checks | |
| 1839 // | |
| 1840 // Generates code that causes a NULL OS exception if the content of reg is NULL. | |
| 1841 // If the accessed location is M[reg + offset] and the offset is known, provide the | |
| 1842 // offset. No explicit code generation is needed if the offset is within a certain | |
| 1843 // range (0 <= offset <= page_size). | |
| 1844 | |
| 1845 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
|
1846 static bool needs_explicit_null_check(intptr_t offset); |
| 0 | 1847 |
| 1848 // Required platform-specific helpers for Label::patch_instructions. | |
| 1849 // They _shadow_ the declarations in AbstractAssembler, which are undefined. | |
| 1850 void pd_patch_instruction(address branch, address target); | |
| 1851 #ifndef PRODUCT | |
| 1852 static void pd_print_patched_instruction(address branch); | |
| 1853 #endif | |
| 1854 | |
| 1855 // The following 4 methods return the offset of the appropriate move instruction | |
| 1856 | |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1857 // Support for fast byte/short loading with zero extension (depending on particular CPU) |
| 0 | 1858 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
|
1859 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
|
1860 |
|
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1861 // Support for fast byte/short loading with sign extension (depending on particular CPU) |
| 0 | 1862 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
|
1863 int load_signed_short(Register dst, Address src); |
| 0 | 1864 |
| 1865 // Support for sign-extension (hi:lo = extend_sign(lo)) | |
| 1866 void extend_sign(Register hi, Register lo); | |
| 1867 | |
|
2258
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2100
diff
changeset
|
1868 // Load and store values by size and signed-ness |
|
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2100
diff
changeset
|
1869 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); |
|
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2100
diff
changeset
|
1870 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
1871 |
| 0 | 1872 // Support for inc/dec with optimal instruction selection depending on value |
| 304 | 1873 |
| 1874 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } | |
| 1875 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } | |
| 1876 | |
| 1877 void decrementl(Address dst, int value = 1); | |
| 1878 void decrementl(Register reg, int value = 1); | |
| 1879 | |
| 1880 void decrementq(Register reg, int value = 1); | |
| 1881 void decrementq(Address dst, int value = 1); | |
| 1882 | |
| 1883 void incrementl(Address dst, int value = 1); | |
| 1884 void incrementl(Register reg, int value = 1); | |
| 1885 | |
| 1886 void incrementq(Register reg, int value = 1); | |
| 1887 void incrementq(Address dst, int value = 1); | |
| 1888 | |
| 0 | 1889 |
| 1890 // Support optimal SSE move instructions. | |
| 1891 void movflt(XMMRegister dst, XMMRegister src) { | |
| 1892 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } | |
| 1893 else { movss (dst, src); return; } | |
| 1894 } | |
| 1895 void movflt(XMMRegister dst, Address src) { movss(dst, src); } | |
| 1896 void movflt(XMMRegister dst, AddressLiteral src); | |
| 1897 void movflt(Address dst, XMMRegister src) { movss(dst, src); } | |
| 1898 | |
| 1899 void movdbl(XMMRegister dst, XMMRegister src) { | |
| 1900 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } | |
| 1901 else { movsd (dst, src); return; } | |
| 1902 } | |
| 1903 | |
| 1904 void movdbl(XMMRegister dst, AddressLiteral src); | |
| 1905 | |
| 1906 void movdbl(XMMRegister dst, Address src) { | |
| 1907 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } | |
| 1908 else { movlpd(dst, src); return; } | |
| 1909 } | |
| 1910 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } | |
| 1911 | |
| 304 | 1912 void incrementl(AddressLiteral dst); |
| 1913 void incrementl(ArrayAddress dst); | |
| 0 | 1914 |
| 1915 // Alignment | |
| 1916 void align(int modulus); | |
| 1917 | |
|
4947
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
1918 // A 5 byte nop that is safe for patching (see patch_verified_entry) |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
1919 void fat_nop(); |
| 0 | 1920 |
| 1921 // Stack frame creation/removal | |
| 1922 void enter(); | |
| 1923 void leave(); | |
| 1924 | |
| 1925 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) | |
| 1926 // The pointer will be loaded into the thread register. | |
| 1927 void get_thread(Register thread); | |
| 1928 | |
| 362 | 1929 |
| 0 | 1930 // Support for VM calls |
| 1931 // | |
| 1932 // It is imperative that all calls into the VM are handled via the call_VM macros. | |
| 1933 // They make sure that the stack linkage is setup correctly. call_VM's correspond | |
| 1934 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. | |
| 1935 | |
| 304 | 1936 |
| 1937 void call_VM(Register oop_result, | |
| 1938 address entry_point, | |
| 1939 bool check_exceptions = true); | |
| 1940 void call_VM(Register oop_result, | |
| 1941 address entry_point, | |
| 1942 Register arg_1, | |
| 1943 bool check_exceptions = true); | |
| 1944 void call_VM(Register oop_result, | |
| 1945 address entry_point, | |
| 1946 Register arg_1, Register arg_2, | |
| 1947 bool check_exceptions = true); | |
| 1948 void call_VM(Register oop_result, | |
| 1949 address entry_point, | |
| 1950 Register arg_1, Register arg_2, Register arg_3, | |
| 1951 bool check_exceptions = true); | |
| 1952 | |
| 1953 // Overloadings with last_Java_sp | |
| 1954 void call_VM(Register oop_result, | |
| 1955 Register last_java_sp, | |
| 1956 address entry_point, | |
| 1957 int number_of_arguments = 0, | |
| 1958 bool check_exceptions = true); | |
| 1959 void call_VM(Register oop_result, | |
| 1960 Register last_java_sp, | |
| 1961 address entry_point, | |
| 1962 Register arg_1, bool | |
| 1963 check_exceptions = true); | |
| 1964 void call_VM(Register oop_result, | |
| 1965 Register last_java_sp, | |
| 1966 address entry_point, | |
| 1967 Register arg_1, Register arg_2, | |
| 1968 bool check_exceptions = true); | |
| 1969 void call_VM(Register oop_result, | |
| 1970 Register last_java_sp, | |
| 1971 address entry_point, | |
| 1972 Register arg_1, Register arg_2, Register arg_3, | |
| 1973 bool check_exceptions = true); | |
| 1974 | |
|
3755
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1975 // These always tightly bind to MacroAssembler::call_VM_base |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1976 // bypassing the virtual implementation |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1977 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1978 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1979 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1980 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true); |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1981 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true); |
|
5cf771a79037
7047697: MethodHandle.invokeExact call for wrong method causes VM failure if run with -Xcomp
jrose
parents:
3363
diff
changeset
|
1982 |
| 304 | 1983 void call_VM_leaf(address entry_point, |
| 1984 int number_of_arguments = 0); | |
| 1985 void call_VM_leaf(address entry_point, | |
| 1986 Register arg_1); | |
| 1987 void call_VM_leaf(address entry_point, | |
| 1988 Register arg_1, Register arg_2); | |
| 1989 void call_VM_leaf(address entry_point, | |
| 1990 Register arg_1, Register arg_2, Register arg_3); | |
| 0 | 1991 |
|
3336
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1992 // These always tightly bind to MacroAssembler::call_VM_leaf_base |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1993 // bypassing the virtual implementation |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1994 void super_call_VM_leaf(address entry_point); |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1995 void super_call_VM_leaf(address entry_point, Register arg_1); |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1996 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1997 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1998 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); |
|
2e038ad0c1d0
7009361: JSR 292 Invalid value on stack on solaris-sparc with -Xcomp
never
parents:
3255
diff
changeset
|
1999 |
| 0 | 2000 // last Java Frame (fills frame anchor) |
| 304 | 2001 void set_last_Java_frame(Register thread, |
| 2002 Register last_java_sp, | |
| 2003 Register last_java_fp, | |
| 2004 address last_java_pc); | |
| 2005 | |
| 2006 // thread in the default location (r15_thread on 64bit) | |
| 2007 void set_last_Java_frame(Register last_java_sp, | |
| 2008 Register last_java_fp, | |
| 2009 address last_java_pc); | |
| 2010 | |
| 0 | 2011 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); |
| 2012 | |
| 304 | 2013 // thread in the default location (r15_thread on 64bit) |
| 2014 void reset_last_Java_frame(bool clear_fp, bool clear_pc); | |
| 2015 | |
| 0 | 2016 // Stores |
| 2017 void store_check(Register obj); // store check for obj - register is destroyed afterwards | |
| 2018 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) | |
| 2019 | |
|
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2020 #ifndef SERIALGC |
|
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2021 |
| 362 | 2022 void g1_write_barrier_pre(Register obj, |
|
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2023 Register pre_val, |
| 362 | 2024 Register thread, |
| 2025 Register tmp, | |
|
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2026 bool tosca_live, |
|
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2027 bool expand_call); |
|
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2028 |
| 362 | 2029 void g1_write_barrier_post(Register store_addr, |
| 2030 Register new_val, | |
| 2031 Register thread, | |
| 2032 Register tmp, | |
| 2033 Register tmp2); | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
2034 |
|
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2320
diff
changeset
|
2035 #endif // SERIALGC |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
71
diff
changeset
|
2036 |
| 0 | 2037 // split store_check(Register obj) to enhance instruction interleaving |
| 2038 void store_check_part_1(Register obj); | |
| 2039 void store_check_part_2(Register obj); | |
| 2040 | |
| 2041 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 | |
| 2042 void c2bool(Register x); | |
| 2043 | |
| 2044 // C++ bool manipulation | |
| 2045 | |
| 2046 void movbool(Register dst, Address src); | |
| 2047 void movbool(Address dst, bool boolconst); | |
| 2048 void movbool(Address dst, Register src); | |
| 2049 void testbool(Register dst); | |
| 2050 | |
| 304 | 2051 // oop manipulations |
| 2052 void load_klass(Register dst, Register src); | |
| 2053 void store_klass(Register dst, Register src); | |
| 2054 | |
| 1846 | 2055 void load_heap_oop(Register dst, Address src); |
|
2464
d86923d96dca
7034967: C1: assert(false) failed: error (assembler_sparc.cpp:2043)
iveresov
parents:
2455
diff
changeset
|
2056 void load_heap_oop_not_null(Register dst, Address src); |
| 1846 | 2057 void store_heap_oop(Address dst, Register src); |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2058 void cmp_heap_oop(Register src1, Address src2, Register tmp = noreg); |
| 1846 | 2059 |
| 2060 // Used for storing NULL. All other oop constants should be | |
| 2061 // stored using routines that take a jobject. | |
| 2062 void store_heap_oop_null(Address dst); | |
| 2063 | |
| 304 | 2064 void load_prototype_header(Register dst, Register src); |
| 2065 | |
| 2066 #ifdef _LP64 | |
| 2067 void store_klass_gap(Register dst, Register src); | |
| 2068 | |
|
1047
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2069 // This dummy is to prevent a call to store_heap_oop from |
|
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2070 // converting a zero (like NULL) into a Register by giving |
|
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2071 // the compiler two choices it can't resolve |
|
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2072 |
|
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2073 void store_heap_oop(Address dst, void* dummy); |
|
beb8f45ee9f0
6889740: G1: OpenDS fails with "unhandled exception in compiled code"
johnc
parents:
986
diff
changeset
|
2074 |
| 304 | 2075 void encode_heap_oop(Register r); |
| 2076 void decode_heap_oop(Register r); | |
| 2077 void encode_heap_oop_not_null(Register r); | |
| 2078 void decode_heap_oop_not_null(Register r); | |
| 2079 void encode_heap_oop_not_null(Register dst, Register src); | |
| 2080 void decode_heap_oop_not_null(Register dst, Register src); | |
| 2081 | |
| 2082 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
|
2083 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
|
2084 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
|
2085 void cmp_narrow_oop(Address dst, jobject obj); |
| 304 | 2086 |
| 2087 // if heap base register is used - reinit it with the correct value | |
| 2088 void reinit_heapbase(); | |
|
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1579
diff
changeset
|
2089 |
|
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1579
diff
changeset
|
2090 DEBUG_ONLY(void verify_heapbase(const char* msg);) |
|
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1579
diff
changeset
|
2091 |
| 304 | 2092 #endif // _LP64 |
| 2093 | |
| 2094 // Int division/remainder for Java | |
| 0 | 2095 // (as idivl, but checks for special case as described in JVM spec.) |
| 2096 // returns idivl instruction offset for implicit exception handling | |
| 2097 int corrected_idivl(Register reg); | |
| 2098 | |
| 304 | 2099 // Long division/remainder for Java |
| 2100 // (as idivq, but checks for special case as described in JVM spec.) | |
| 2101 // returns idivq instruction offset for implicit exception handling | |
| 2102 int corrected_idivq(Register reg); | |
| 2103 | |
| 0 | 2104 void int3(); |
| 2105 | |
| 304 | 2106 // Long operation macros for a 32bit cpu |
| 0 | 2107 // Long negation for Java |
| 2108 void lneg(Register hi, Register lo); | |
| 2109 | |
| 2110 // Long multiplication for Java | |
| 304 | 2111 // (destroys contents of eax, ebx, ecx and edx) |
| 0 | 2112 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y |
| 2113 | |
| 2114 // Long shifts for Java | |
| 2115 // (semantics as described in JVM spec.) | |
| 2116 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) | |
| 2117 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) | |
| 2118 | |
| 2119 // Long compare for Java | |
| 2120 // (semantics as described in JVM spec.) | |
| 2121 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) | |
| 2122 | |
| 304 | 2123 |
| 2124 // misc | |
| 2125 | |
| 2126 // Sign extension | |
| 2127 void sign_extend_short(Register reg); | |
| 2128 void sign_extend_byte(Register reg); | |
| 2129 | |
| 2130 // Division by power of 2, rounding towards 0 | |
| 2131 void division_with_shift(Register reg, int shift_value); | |
| 2132 | |
| 0 | 2133 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: |
| 2134 // | |
| 2135 // CF (corresponds to C0) if x < y | |
| 2136 // PF (corresponds to C2) if unordered | |
| 2137 // ZF (corresponds to C3) if x = y | |
| 2138 // | |
| 2139 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). | |
| 2140 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) | |
| 2141 void fcmp(Register tmp); | |
| 2142 // Variant of the above which allows y to be further down the stack | |
| 2143 // and which only pops x and y if specified. If pop_right is | |
| 2144 // specified then pop_left must also be specified. | |
| 2145 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); | |
| 2146 | |
| 2147 // Floating-point comparison for Java | |
| 2148 // Compares the top-most stack entries on the FPU stack and stores the result in dst. | |
| 2149 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). | |
| 2150 // (semantics as described in JVM spec.) | |
| 2151 void fcmp2int(Register dst, bool unordered_is_less); | |
| 2152 // Variant of the above which allows y to be further down the stack | |
| 2153 // and which only pops x and y if specified. If pop_right is | |
| 2154 // specified then pop_left must also be specified. | |
| 2155 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); | |
| 2156 | |
| 2157 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) | |
| 2158 // tmp is a temporary register, if none is available use noreg | |
| 2159 void fremr(Register tmp); | |
| 2160 | |
| 2161 | |
| 2162 // same as fcmp2int, but using SSE2 | |
| 2163 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); | |
| 2164 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); | |
| 2165 | |
| 2166 // Inlined sin/cos generator for Java; must not use CPU instruction | |
| 2167 // directly on Intel as it does not have high enough precision | |
| 2168 // outside of the range [-pi/4, pi/4]. Extra argument indicate the | |
| 2169 // number of FPU stack slots in use; all but the topmost will | |
| 2170 // require saving if a slow case is necessary. Assumes argument is | |
| 2171 // on FP TOS; result is on FP TOS. No cpu registers are changed by | |
| 2172 // this code. | |
| 2173 void trigfunc(char trig, int num_fpu_regs_in_use = 1); | |
| 2174 | |
| 2175 // branch to L if FPU flag C2 is set/not set | |
| 2176 // tmp is a temporary register, if none is available use noreg | |
| 2177 void jC2 (Register tmp, Label& L); | |
| 2178 void jnC2(Register tmp, Label& L); | |
| 2179 | |
| 2180 // Pop ST (ffree & fincstp combined) | |
| 2181 void fpop(); | |
| 2182 | |
| 2183 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack | |
| 2184 void push_fTOS(); | |
| 2185 | |
| 2186 // pops double TOS element from CPU stack and pushes on FPU stack | |
| 2187 void pop_fTOS(); | |
| 2188 | |
| 2189 void empty_FPU_stack(); | |
| 2190 | |
| 2191 void push_IU_state(); | |
| 2192 void pop_IU_state(); | |
| 2193 | |
| 2194 void push_FPU_state(); | |
| 2195 void pop_FPU_state(); | |
| 2196 | |
| 2197 void push_CPU_state(); | |
| 2198 void pop_CPU_state(); | |
| 2199 | |
| 2200 // Round up to a power of two | |
| 2201 void round_to(Register reg, int modulus); | |
| 2202 | |
| 2203 // Callee saved registers handling | |
| 2204 void push_callee_saved_registers(); | |
| 2205 void pop_callee_saved_registers(); | |
| 2206 | |
| 2207 // allocation | |
| 2208 void eden_allocate( | |
| 2209 Register obj, // result: pointer to object after successful allocation | |
| 2210 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 2211 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 2212 Register t1, // temp register | |
| 2213 Label& slow_case // continuation point if fast allocation fails | |
| 2214 ); | |
| 2215 void tlab_allocate( | |
| 2216 Register obj, // result: pointer to object after successful allocation | |
| 2217 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 2218 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 2219 Register t1, // temp register | |
| 2220 Register t2, // temp register | |
| 2221 Label& slow_case // continuation point if fast allocation fails | |
| 2222 ); | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2223 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2224 void incr_allocated_bytes(Register thread, |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2225 Register var_size_in_bytes, int con_size_in_bytes, |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2226 Register t1 = noreg); |
| 0 | 2227 |
|
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2228 // interface method calling |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2229 void lookup_interface_method(Register recv_klass, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2230 Register intf_klass, |
|
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2231 RegisterOrConstant itable_index, |
|
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2232 Register method_result, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2233 Register scan_temp, |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2234 Label& no_such_interface); |
|
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2235 |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2236 // virtual method calling |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2237 void lookup_virtual_method(Register recv_klass, |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2238 RegisterOrConstant vtable_index, |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2239 Register method_result); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2240 |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2241 // Test sub_klass against super_klass, with fast and slow paths. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2242 |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2243 // The fast path produces a tri-state answer: yes / no / maybe-slow. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2244 // One of the three labels can be NULL, meaning take the fall-through. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2245 // If super_check_offset is -1, the value is loaded up from super_klass. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2246 // No registers are killed, except temp_reg. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2247 void check_klass_subtype_fast_path(Register sub_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2248 Register super_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2249 Register temp_reg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2250 Label* L_success, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2251 Label* L_failure, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2252 Label* L_slow_path, |
|
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2253 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2254 |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2255 // The rest of the type check; must be wired to a corresponding fast path. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2256 // It does not repeat the fast path logic, so don't use it standalone. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2257 // The temp_reg and temp2_reg can be noreg, if no temps are available. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2258 // Updates the sub's secondary super cache as necessary. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2259 // If set_cond_codes, condition codes will be Z on success, NZ on failure. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2260 void check_klass_subtype_slow_path(Register sub_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2261 Register super_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2262 Register temp_reg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2263 Register temp2_reg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2264 Label* L_success, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2265 Label* L_failure, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2266 bool set_cond_codes = false); |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2267 |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2268 // Simplified, combined version, good for typical uses. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2269 // Falls through on failure. |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2270 void check_klass_subtype(Register sub_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2271 Register super_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2272 Register temp_reg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2273 Label& L_success); |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
2274 |
| 710 | 2275 // method handles (JSR 292) |
| 2276 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); | |
| 2277 | |
| 0 | 2278 //---- |
| 2279 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 | |
| 2280 | |
| 2281 // Debugging | |
| 304 | 2282 |
| 2283 // only if +VerifyOops | |
| 2284 void verify_oop(Register reg, const char* s = "broken oop"); | |
| 0 | 2285 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); |
| 2286 | |
| 304 | 2287 // only if +VerifyFPU |
| 2288 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); | |
| 2289 | |
| 2290 // prints msg, dumps registers and stops execution | |
| 2291 void stop(const char* msg); | |
| 2292 | |
| 2293 // prints msg and continues | |
| 2294 void warn(const char* msg); | |
| 2295 | |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2296 // dumps registers and other state |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2297 void print_state(); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2298 |
| 304 | 2299 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); |
| 2300 static void debug64(char* msg, int64_t pc, int64_t regs[]); | |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2301 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6225
diff
changeset
|
2302 static void print_state64(int64_t pc, int64_t regs[]); |
| 304 | 2303 |
| 0 | 2304 void os_breakpoint(); |
| 304 | 2305 |
| 0 | 2306 void untested() { stop("untested"); } |
| 304 | 2307 |
| 1846 | 2308 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } |
| 304 | 2309 |
| 0 | 2310 void should_not_reach_here() { stop("should not reach here"); } |
| 304 | 2311 |
| 0 | 2312 void print_CPU_state(); |
| 2313 | |
| 2314 // Stack overflow checking | |
| 2315 void bang_stack_with_offset(int offset) { | |
| 2316 // stack grows down, caller passes positive offset | |
| 2317 assert(offset > 0, "must bang with negative offset"); | |
| 2318 movl(Address(rsp, (-offset)), rax); | |
| 2319 } | |
| 2320 | |
| 2321 // Writes to stack successive pages until offset reached to check for | |
| 2322 // stack overflow + shadow pages. Also, clobbers tmp | |
| 2323 void bang_stack_size(Register size, Register tmp); | |
| 2324 | |
|
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2325 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, |
|
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2326 Register tmp, |
|
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2327 int offset); |
|
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2328 |
| 0 | 2329 // Support for serializing memory accesses between threads |
| 2330 void serialize_memory(Register thread, Register tmp); | |
| 2331 | |
| 2332 void verify_tlab(); | |
| 2333 | |
| 2334 // Biased locking support | |
| 2335 // lock_reg and obj_reg must be loaded up with the appropriate values. | |
| 2336 // swap_reg must be rax, and is killed. | |
| 2337 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will | |
| 2338 // be killed; if not supplied, push/pop will be used internally to | |
| 2339 // allocate a temporary (inefficient, avoid if possible). | |
| 2340 // Optional slow case is for implementations (interpreter and C1) which branch to | |
| 2341 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. | |
| 2342 // Returns offset of first potentially-faulting instruction for null | |
| 2343 // check info (currently consumed only by C1). If | |
| 2344 // swap_reg_contains_mark is true then returns -1 as it is assumed | |
| 2345 // 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
|
2346 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
|
2347 Register swap_reg, Register tmp_reg, |
| 0 | 2348 bool swap_reg_contains_mark, |
| 2349 Label& done, Label* slow_case = NULL, | |
| 2350 BiasedLockingCounters* counters = NULL); | |
| 2351 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); | |
| 2352 | |
| 2353 | |
| 2354 Condition negate_condition(Condition cond); | |
| 2355 | |
| 2356 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit | |
| 2357 // operands. In general the names are modified to avoid hiding the instruction in Assembler | |
| 2358 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers | |
| 2359 // here in MacroAssembler. The major exception to this rule is call | |
| 2360 | |
| 2361 // Arithmetics | |
| 2362 | |
| 304 | 2363 |
| 2364 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } | |
| 2365 void addptr(Address dst, Register src); | |
| 2366 | |
| 2367 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } | |
| 2368 void addptr(Register dst, int32_t src); | |
| 2369 void addptr(Register dst, Register src); | |
|
3363
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2370 void addptr(Register dst, RegisterOrConstant src) { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2371 if (src.is_constant()) addptr(dst, (int) src.as_constant()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2372 else addptr(dst, src.as_register()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2373 } |
| 304 | 2374 |
| 2375 void andptr(Register dst, int32_t src); | |
| 2376 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } | |
| 2377 | |
| 2378 void cmp8(AddressLiteral src1, int imm); | |
| 2379 | |
| 2380 // renamed to drag out the casting of address to int32_t/intptr_t | |
| 0 | 2381 void cmp32(Register src1, int32_t imm); |
| 2382 | |
| 2383 void cmp32(AddressLiteral src1, int32_t imm); | |
| 2384 // compare reg - mem, or reg - &mem | |
| 2385 void cmp32(Register src1, AddressLiteral src2); | |
| 2386 | |
| 2387 void cmp32(Register src1, Address src2); | |
| 2388 | |
| 304 | 2389 #ifndef _LP64 |
| 2390 void cmpoop(Address dst, jobject obj); | |
| 2391 void cmpoop(Register dst, jobject obj); | |
| 2392 #endif // _LP64 | |
| 2393 | |
| 0 | 2394 // NOTE src2 must be the lval. This is NOT an mem-mem compare |
| 2395 void cmpptr(Address src1, AddressLiteral src2); | |
| 2396 | |
| 2397 void cmpptr(Register src1, AddressLiteral src2); | |
| 2398 | |
| 304 | 2399 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } |
| 2400 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 2401 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 2402 | |
| 2403 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 2404 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } | |
| 2405 | |
| 2406 // cmp64 to avoild hiding cmpq | |
| 2407 void cmp64(Register src1, AddressLiteral src); | |
| 2408 | |
| 2409 void cmpxchgptr(Register reg, Address adr); | |
| 2410 | |
| 2411 void locked_cmpxchgptr(Register reg, AddressLiteral adr); | |
| 2412 | |
| 2413 | |
| 2414 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } | |
| 2415 | |
| 2416 | |
| 2417 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } | |
| 2418 | |
| 2419 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } | |
| 2420 | |
| 2421 void shlptr(Register dst, int32_t shift); | |
| 2422 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } | |
| 2423 | |
| 2424 void shrptr(Register dst, int32_t shift); | |
| 2425 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } | |
| 2426 | |
| 2427 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } | |
| 2428 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } | |
| 2429 | |
| 2430 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } | |
| 2431 | |
| 2432 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } | |
| 2433 void subptr(Register dst, int32_t src); | |
|
4947
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
2434 // Force generation of a 4 byte immediate value even if it fits into 8bit |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
2435 void subptr_imm32(Register dst, int32_t src); |
| 304 | 2436 void subptr(Register dst, Register src); |
|
3363
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2437 void subptr(Register dst, RegisterOrConstant src) { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2438 if (src.is_constant()) subptr(dst, (int) src.as_constant()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2439 else subptr(dst, src.as_register()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2440 } |
| 304 | 2441 |
| 2442 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } | |
| 2443 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } | |
| 2444 | |
| 2445 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } | |
| 2446 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } | |
| 2447 | |
| 2448 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } | |
| 2449 | |
| 2450 | |
| 0 | 2451 |
| 2452 // Helper functions for statistics gathering. | |
| 2453 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. | |
| 2454 void cond_inc32(Condition cond, AddressLiteral counter_addr); | |
| 2455 // Unconditional atomic increment. | |
| 2456 void atomic_incl(AddressLiteral counter_addr); | |
| 2457 | |
| 2458 void lea(Register dst, AddressLiteral adr); | |
| 2459 void lea(Address dst, AddressLiteral adr); | |
| 304 | 2460 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } |
| 2461 | |
| 2462 void leal32(Register dst, Address src) { leal(dst, src); } | |
| 2463 | |
|
2404
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
2464 // Import other testl() methods from the parent class or else |
|
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
2465 // they will be hidden by the following overriding declaration. |
|
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
2466 using Assembler::testl; |
|
b40d4fa697bf
6964776: c2 should ensure the polling page is reachable on 64 bit
iveresov
parents:
2320
diff
changeset
|
2467 void testl(Register dst, AddressLiteral src); |
| 304 | 2468 |
| 2469 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 2470 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 2471 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } | |
| 2472 | |
| 2473 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } | |
| 2474 void testptr(Register src1, Register src2); | |
| 2475 | |
| 2476 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } | |
| 2477 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } | |
| 0 | 2478 |
| 2479 // Calls | |
| 2480 | |
| 2481 void call(Label& L, relocInfo::relocType rtype); | |
| 2482 void call(Register entry); | |
| 2483 | |
| 2484 // NOTE: this call tranfers to the effective address of entry NOT | |
| 2485 // the address contained by entry. This is because this is more natural | |
| 2486 // for jumps/calls. | |
| 2487 void call(AddressLiteral entry); | |
| 2488 | |
| 2489 // Jumps | |
| 2490 | |
| 2491 // NOTE: these jumps tranfer to the effective address of dst NOT | |
| 2492 // the address contained by dst. This is because this is more natural | |
| 2493 // for jumps/calls. | |
| 2494 void jump(AddressLiteral dst); | |
| 2495 void jump_cc(Condition cc, AddressLiteral dst); | |
| 2496 | |
| 2497 // 32bit can do a case table jump in one instruction but we no longer allow the base | |
| 2498 // to be installed in the Address class. This jump will tranfers to the address | |
| 2499 // contained in the location described by entry (not the address of entry) | |
| 2500 void jump(ArrayAddress entry); | |
| 2501 | |
| 2502 // Floating | |
| 2503 | |
| 2504 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } | |
| 2505 void andpd(XMMRegister dst, AddressLiteral src); | |
| 2506 | |
| 4759 | 2507 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); } |
| 2508 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); } | |
| 2509 void andps(XMMRegister dst, AddressLiteral src); | |
| 2510 | |
| 2511 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); } | |
| 0 | 2512 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } |
| 2513 void comiss(XMMRegister dst, AddressLiteral src); | |
| 2514 | |
| 4759 | 2515 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); } |
| 0 | 2516 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } |
| 2517 void comisd(XMMRegister dst, AddressLiteral src); | |
| 2518 | |
| 2008 | 2519 void fadd_s(Address src) { Assembler::fadd_s(src); } |
| 2520 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); } | |
| 2521 | |
| 0 | 2522 void fldcw(Address src) { Assembler::fldcw(src); } |
| 2523 void fldcw(AddressLiteral src); | |
| 2524 | |
| 2525 void fld_s(int index) { Assembler::fld_s(index); } | |
| 2526 void fld_s(Address src) { Assembler::fld_s(src); } | |
| 2527 void fld_s(AddressLiteral src); | |
| 2528 | |
| 2529 void fld_d(Address src) { Assembler::fld_d(src); } | |
| 2530 void fld_d(AddressLiteral src); | |
| 2531 | |
| 2532 void fld_x(Address src) { Assembler::fld_x(src); } | |
| 2533 void fld_x(AddressLiteral src); | |
| 2534 | |
| 2008 | 2535 void fmul_s(Address src) { Assembler::fmul_s(src); } |
| 2536 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); } | |
| 2537 | |
| 0 | 2538 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } |
| 2539 void ldmxcsr(AddressLiteral src); | |
| 2540 | |
|
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2541 // compute pow(x,y) and exp(x) with x86 instructions. Don't cover |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2542 // all corner cases and may result in NaN and require fallback to a |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2543 // runtime call. |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2544 void fast_pow(); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2545 void fast_exp(); |
|
6141
e7715c222897
7174532: jdk/test/java/lang/Math/WorstCaseTests.java failing on x86
roland
parents:
6084
diff
changeset
|
2546 void increase_precision(); |
|
e7715c222897
7174532: jdk/test/java/lang/Math/WorstCaseTests.java failing on x86
roland
parents:
6084
diff
changeset
|
2547 void restore_precision(); |
|
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2548 |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2549 // computes exp(x). Fallback to runtime call included. |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2550 void exp_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(true, num_fpu_regs_in_use); } |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2551 // computes pow(x,y). Fallback to runtime call included. |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2552 void pow_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(false, num_fpu_regs_in_use); } |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2553 |
| 304 | 2554 private: |
|
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2555 |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2556 // call runtime as a fallback for trig functions and pow/exp. |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2557 void fp_runtime_fallback(address runtime_entry, int nb_args, int num_fpu_regs_in_use); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2558 |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2559 // computes 2^(Ylog2X); Ylog2X in ST(0) |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2560 void pow_exp_core_encoding(); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2561 |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2562 // computes pow(x,y) or exp(x). Fallback to runtime call included. |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2563 void pow_or_exp(bool is_exp, int num_fpu_regs_in_use); |
|
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4947
diff
changeset
|
2564 |
| 304 | 2565 // these are private because users should be doing movflt/movdbl |
| 2566 | |
| 0 | 2567 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } |
| 2568 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } | |
| 2569 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } | |
| 2570 void movss(XMMRegister dst, AddressLiteral src); | |
| 2571 | |
| 4759 | 2572 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } |
| 304 | 2573 void movlpd(XMMRegister dst, AddressLiteral src); |
| 2574 | |
| 2575 public: | |
| 2576 | |
| 2008 | 2577 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); } |
| 2578 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); } | |
| 4759 | 2579 void addsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2580 |
| 2581 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); } | |
| 2582 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); } | |
| 4759 | 2583 void addss(XMMRegister dst, AddressLiteral src); |
| 2008 | 2584 |
| 2585 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); } | |
| 2586 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); } | |
| 4759 | 2587 void divsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2588 |
| 2589 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); } | |
| 2590 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); } | |
| 4759 | 2591 void divss(XMMRegister dst, AddressLiteral src); |
| 2008 | 2592 |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2593 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2594 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } |
|
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2595 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } |
| 4759 | 2596 void movsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2597 |
| 2598 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); } | |
| 2599 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); } | |
| 4759 | 2600 void mulsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2601 |
| 2602 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); } | |
| 2603 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); } | |
| 4759 | 2604 void mulss(XMMRegister dst, AddressLiteral src); |
| 2008 | 2605 |
| 2606 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); } | |
| 2607 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); } | |
| 4759 | 2608 void sqrtsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2609 |
| 2610 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); } | |
| 2611 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); } | |
| 4759 | 2612 void sqrtss(XMMRegister dst, AddressLiteral src); |
| 2008 | 2613 |
| 2614 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); } | |
| 2615 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); } | |
| 4759 | 2616 void subsd(XMMRegister dst, AddressLiteral src); |
| 2008 | 2617 |
| 2618 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); } | |
| 2619 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); } | |
| 4759 | 2620 void subss(XMMRegister dst, AddressLiteral src); |
| 0 | 2621 |
| 2622 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } | |
| 4759 | 2623 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } |
| 0 | 2624 void ucomiss(XMMRegister dst, AddressLiteral src); |
| 2625 | |
| 2626 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } | |
| 4759 | 2627 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } |
| 0 | 2628 void ucomisd(XMMRegister dst, AddressLiteral src); |
| 2629 | |
| 2630 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values | |
| 2631 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); } | |
| 2632 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } | |
| 2633 void xorpd(XMMRegister dst, AddressLiteral src); | |
| 2634 | |
| 2635 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values | |
| 2636 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); } | |
| 2637 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } | |
| 2638 void xorps(XMMRegister dst, AddressLiteral src); | |
| 2639 | |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2640 // AVX 3-operands instructions |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2641 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2642 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2643 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2644 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2645 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2646 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2647 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2648 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2649 |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2650 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2651 void vandpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2652 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2653 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2654 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2655 void vandps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2656 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2657 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2658 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2659 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2660 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2661 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2662 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2663 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2664 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2665 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2666 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2667 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2668 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2669 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2670 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2671 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2672 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2673 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2674 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2675 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2676 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2677 |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2678 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2679 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); } |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2680 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src); |
|
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2681 |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
2682 // AVX Vector instructions |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
2683 |
|
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
2684 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); } |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2685 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2686 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2687 |
|
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
6141
diff
changeset
|
2688 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); } |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2689 void vxorps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); } |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2690 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2691 |
| 6225 | 2692 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { |
| 2693 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2 | |
| 2694 Assembler::vpxor(dst, nds, src, vector256); | |
| 2695 else | |
| 2696 Assembler::vxorpd(dst, nds, src, vector256); | |
| 2697 } | |
|
6614
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2698 void vpxor(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2699 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2 |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2700 Assembler::vpxor(dst, nds, src, vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2701 else |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2702 Assembler::vxorpd(dst, nds, src, vector256); |
|
006050192a5a
6340864: Implement vectorization optimizations in hotspot-server
kvn
parents:
6266
diff
changeset
|
2703 } |
| 6225 | 2704 |
| 2705 // Move packed integer values from low 128 bit to hign 128 bit in 256 bit vector. | |
| 2706 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src) { | |
| 2707 if (UseAVX > 1) // vinserti128h is available only in AVX2 | |
| 2708 Assembler::vinserti128h(dst, nds, src); | |
| 2709 else | |
| 2710 Assembler::vinsertf128h(dst, nds, src); | |
| 2711 } | |
|
4761
65149e74c706
7121648: Use 3-operands SIMD instructions on x86 with AVX
kvn
parents:
4759
diff
changeset
|
2712 |
| 0 | 2713 // Data |
| 2714 | |
|
2415
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2715 void cmov32( Condition cc, Register dst, Address src); |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2716 void cmov32( Condition cc, Register dst, Register src); |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2717 |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2718 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); } |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2719 |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2720 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } |
|
09f96c3ff1ad
7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934
twisti
parents:
2404
diff
changeset
|
2721 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } |
| 304 | 2722 |
| 0 | 2723 void movoop(Register dst, jobject obj); |
| 2724 void movoop(Address dst, jobject obj); | |
| 2725 | |
| 2726 void movptr(ArrayAddress dst, Register src); | |
| 2727 // can this do an lea? | |
| 2728 void movptr(Register dst, ArrayAddress src); | |
| 2729 | |
| 304 | 2730 void movptr(Register dst, Address src); |
| 2731 | |
| 0 | 2732 void movptr(Register dst, AddressLiteral src); |
| 2733 | |
| 304 | 2734 void movptr(Register dst, intptr_t src); |
| 2735 void movptr(Register dst, Register src); | |
| 2736 void movptr(Address dst, intptr_t src); | |
| 2737 | |
| 2738 void movptr(Address dst, Register src); | |
| 2739 | |
|
3363
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2740 void movptr(Register dst, RegisterOrConstant src) { |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2741 if (src.is_constant()) movptr(dst, src.as_constant()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2742 else movptr(dst, src.as_register()); |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2743 } |
|
167b70ff3abc
6939861: JVM should handle more conversion operations
never
parents:
3336
diff
changeset
|
2744 |
| 304 | 2745 #ifdef _LP64 |
| 2746 // Generally the next two are only used for moving NULL | |
| 2747 // Although there are situations in initializing the mark word where | |
| 2748 // they could be used. They are dangerous. | |
| 2749 | |
| 2750 // They only exist on LP64 so that int32_t and intptr_t are not the same | |
| 2751 // and we have ambiguous declarations. | |
| 2752 | |
| 2753 void movptr(Address dst, int32_t imm32); | |
| 2754 void movptr(Register dst, int32_t imm32); | |
| 2755 #endif // _LP64 | |
| 2756 | |
| 0 | 2757 // to avoid hiding movl |
| 2758 void mov32(AddressLiteral dst, Register src); | |
| 2759 void mov32(Register dst, AddressLiteral src); | |
| 304 | 2760 |
| 0 | 2761 // to avoid hiding movb |
| 2762 void movbyte(ArrayAddress dst, int src); | |
| 2763 | |
| 6225 | 2764 // Import other mov() methods from the parent class or else |
| 2765 // they will be hidden by the following overriding declaration. | |
| 2766 using Assembler::movdl; | |
| 2767 using Assembler::movq; | |
| 2768 void movdl(XMMRegister dst, AddressLiteral src); | |
| 2769 void movq(XMMRegister dst, AddressLiteral src); | |
| 2770 | |
| 0 | 2771 // Can push value or effective address |
| 2772 void pushptr(AddressLiteral src); | |
| 2773 | |
| 304 | 2774 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } |
| 2775 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } | |
| 2776 | |
| 2777 void pushoop(jobject obj); | |
| 2778 | |
| 2779 // sign extend as need a l to ptr sized element | |
| 2780 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } | |
| 2781 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } | |
| 2782 | |
|
4947
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
2783 // C2 compiled method's prolog code. |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
2784 void verified_entry(int framesize, bool stack_bang, bool fp_mode_24b); |
|
fd8114661503
7125136: SIGILL on linux amd64 in gc/ArrayJuggle/Juggle29
kvn
parents:
4761
diff
changeset
|
2785 |
|
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2786 // IndexOf strings. |
|
2320
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2787 // Small strings are loaded through stack if they cross page boundary. |
|
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2788 void string_indexof(Register str1, Register str2, |
|
2320
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2789 Register cnt1, Register cnt2, |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2790 int int_cnt2, Register result, |
|
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2791 XMMRegister vec, Register tmp); |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2792 |
|
2320
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2793 // IndexOf for constant substrings with size >= 8 elements |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2794 // which don't need to be loaded through stack. |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2795 void string_indexofC8(Register str1, Register str2, |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2796 Register cnt1, Register cnt2, |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2797 int int_cnt2, Register result, |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2798 XMMRegister vec, Register tmp); |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2799 |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2800 // Smallest code: we don't need to load through stack, |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2801 // check string tail. |
|
41d4973cf100
6942326: x86 code in string_indexof() could read beyond reserved heap space
kvn
parents:
2262
diff
changeset
|
2802 |
|
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2803 // Compare strings. |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2804 void string_compare(Register str1, Register str2, |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2805 Register cnt1, Register cnt2, Register result, |
| 2262 | 2806 XMMRegister vec1); |
|
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2807 |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2808 // Compare char[] arrays. |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2809 void char_arrays_equals(bool is_array_equ, Register ary1, Register ary2, |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2810 Register limit, Register result, Register chr, |
|
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
775
diff
changeset
|
2811 XMMRegister vec1, XMMRegister vec2); |
| 304 | 2812 |
| 1763 | 2813 // Fill primitive arrays |
| 2814 void generate_fill(BasicType t, bool aligned, | |
| 2815 Register to, Register value, Register count, | |
| 2816 Register rtmp, XMMRegister xtmp); | |
| 2817 | |
| 0 | 2818 #undef VIRTUAL |
| 2819 | |
| 2820 }; | |
| 2821 | |
| 2822 /** | |
| 2823 * class SkipIfEqual: | |
| 2824 * | |
| 2825 * Instantiating this class will result in assembly code being output that will | |
| 2826 * jump around any code emitted between the creation of the instance and it's | |
| 2827 * automatic destruction at the end of a scope block, depending on the value of | |
| 2828 * the flag passed to the constructor, which will be checked at run-time. | |
| 2829 */ | |
| 2830 class SkipIfEqual { | |
| 2831 private: | |
| 2832 MacroAssembler* _masm; | |
| 2833 Label _label; | |
| 2834 | |
| 2835 public: | |
| 2836 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); | |
| 2837 ~SkipIfEqual(); | |
| 2838 }; | |
| 2839 | |
| 2840 #ifdef ASSERT | |
| 2841 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } | |
| 2842 #endif | |
| 1972 | 2843 |
| 2844 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP |