1 /* 2 * Copyright (c) 2000, 2019, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef CPU_X86_GLOBALS_X86_HPP 26 #define CPU_X86_GLOBALS_X86_HPP 27 28 #include "utilities/globalDefinitions.hpp" 29 #include "utilities/macros.hpp" 30 31 // Sets the default values for platform dependent flags used by the runtime system. 32 // (see globals.hpp) 33 34 define_pd_global(bool, ImplicitNullChecks, true); // Generate code for implicit null checks 35 define_pd_global(bool, TrapBasedNullChecks, false); // Not needed on x86. 36 define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap NULLs passed to check cast 37 38 define_pd_global(uintx, CodeCacheSegmentSize, 64 TIERED_ONLY(+64)); // Tiered compilation has large code-entry alignment. 39 // See 4827828 for this change. There is no globals_core_i486.hpp. I can't 40 // assign a different value for C2 without touching a number of files. Use 41 // #ifdef to minimize the change as it's late in Mantis. -- FIXME. 42 // c1 doesn't have this problem because the fix to 4858033 assures us 43 // the the vep is aligned at CodeEntryAlignment whereas c2 only aligns 44 // the uep and the vep doesn't get real alignment but just slops on by 45 // only assured that the entry instruction meets the 5 byte size requirement. 46 #if COMPILER2_OR_JVMCI 47 define_pd_global(intx, CodeEntryAlignment, 32); 48 #else 49 define_pd_global(intx, CodeEntryAlignment, 16); 50 #endif // COMPILER2_OR_JVMCI 51 define_pd_global(intx, OptoLoopAlignment, 16); 52 define_pd_global(intx, InlineFrequencyCount, 100); 53 define_pd_global(intx, InlineSmallCode, 1000); 54 55 #define DEFAULT_STACK_YELLOW_PAGES (NOT_WINDOWS(2) WINDOWS_ONLY(3)) 56 #define DEFAULT_STACK_RED_PAGES (1) 57 #define DEFAULT_STACK_RESERVED_PAGES (NOT_WINDOWS(1) WINDOWS_ONLY(0)) 58 59 #define MIN_STACK_YELLOW_PAGES DEFAULT_STACK_YELLOW_PAGES 60 #define MIN_STACK_RED_PAGES DEFAULT_STACK_RED_PAGES 61 #define MIN_STACK_RESERVED_PAGES (0) 62 63 #ifdef _LP64 64 // Java_java_net_SocketOutputStream_socketWrite0() uses a 64k buffer on the 65 // stack if compiled for unix and LP64. To pass stack overflow tests we need 66 // 20 shadow pages. 67 #define DEFAULT_STACK_SHADOW_PAGES (NOT_WIN64(20) WIN64_ONLY(7) DEBUG_ONLY(+2)) 68 // For those clients that do not use write socket, we allow 69 // the min range value to be below that of the default 70 #define MIN_STACK_SHADOW_PAGES (NOT_WIN64(10) WIN64_ONLY(7) DEBUG_ONLY(+2)) 71 #else 72 #define DEFAULT_STACK_SHADOW_PAGES (4 DEBUG_ONLY(+5)) 73 #define MIN_STACK_SHADOW_PAGES DEFAULT_STACK_SHADOW_PAGES 74 #endif // _LP64 75 76 define_pd_global(intx, StackYellowPages, DEFAULT_STACK_YELLOW_PAGES); 77 define_pd_global(intx, StackRedPages, DEFAULT_STACK_RED_PAGES); 78 define_pd_global(intx, StackShadowPages, DEFAULT_STACK_SHADOW_PAGES); 79 define_pd_global(intx, StackReservedPages, DEFAULT_STACK_RESERVED_PAGES); 80 81 define_pd_global(bool, RewriteBytecodes, true); 82 define_pd_global(bool, RewriteFrequentPairs, true); 83 84 // GC Ergo Flags 85 define_pd_global(size_t, CMSYoungGenPerWorker, 64*M); // default max size of CMS young gen, per GC worker thread 86 87 define_pd_global(uintx, TypeProfileLevel, 111); 88 89 define_pd_global(bool, CompactStrings, true); 90 91 define_pd_global(bool, PreserveFramePointer, false); 92 93 define_pd_global(intx, InitArrayShortSize, 8*BytesPerLong); 94 95 #if defined(_LP64) || defined(_WINDOWS) 96 define_pd_global(bool, ThreadLocalHandshakes, true); 97 // ObjectMonitor ref_count is implemented in LP64 C2 fast_lock() 98 // and fast_unlock() so we don't need the handshake by default. 99 #ifdef _LP64 100 define_pd_global(bool, HandshakeAfterDeflateIdleMonitors, false); 101 #else 102 define_pd_global(bool, HandshakeAfterDeflateIdleMonitors, true); 103 #endif 104 #else 105 // get_thread() is slow on linux 32 bit, therefore off by default 106 define_pd_global(bool, ThreadLocalHandshakes, false); 107 // ObjectMonitor ref_count not implemented in C2 fast_lock() or 108 // fast_unlock() so use a handshake for safety. 109 // Will use a safepoint instead of a handshake on this platform. 110 define_pd_global(bool, HandshakeAfterDeflateIdleMonitors, true); 111 #endif 112 113 #define ARCH_FLAGS(develop, \ 114 product, \ 115 diagnostic, \ 116 experimental, \ 117 notproduct, \ 118 range, \ 119 constraint, \ 120 writeable) \ 121 \ 122 develop(bool, IEEEPrecision, true, \ 123 "Enables IEEE precision (for INTEL only)") \ 124 \ 125 product(bool, UseStoreImmI16, true, \ 126 "Use store immediate 16-bits value instruction on x86") \ 127 \ 128 product(intx, UseAVX, 3, \ 129 "Highest supported AVX instructions set on x86/x64") \ 130 range(0, 99) \ 131 \ 132 product(bool, UseCLMUL, false, \ 133 "Control whether CLMUL instructions can be used on x86/x64") \ 134 \ 135 diagnostic(bool, UseIncDec, true, \ 136 "Use INC, DEC instructions on x86") \ 137 \ 138 product(bool, UseNewLongLShift, false, \ 139 "Use optimized bitwise shift left") \ 140 \ 141 product(bool, UseAddressNop, false, \ 142 "Use '0F 1F [addr]' NOP instructions on x86 cpus") \ 143 \ 144 product(bool, UseXmmLoadAndClearUpper, true, \ 145 "Load low part of XMM register and clear upper part") \ 146 \ 147 product(bool, UseXmmRegToRegMoveAll, false, \ 148 "Copy all XMM register bits when moving value between registers") \ 149 \ 150 product(bool, UseXmmI2D, false, \ 151 "Use SSE2 CVTDQ2PD instruction to convert Integer to Double") \ 152 \ 153 product(bool, UseXmmI2F, false, \ 154 "Use SSE2 CVTDQ2PS instruction to convert Integer to Float") \ 155 \ 156 product(bool, UseUnalignedLoadStores, false, \ 157 "Use SSE2 MOVDQU instruction for Arraycopy") \ 158 \ 159 product(bool, UseXMMForObjInit, false, \ 160 "Use XMM/YMM MOVDQU instruction for Object Initialization") \ 161 \ 162 product(bool, UseFastStosb, false, \ 163 "Use fast-string operation for zeroing: rep stosb") \ 164 \ 165 /* Use Restricted Transactional Memory for lock eliding */ \ 166 product(bool, UseRTMLocking, false, \ 167 "Enable RTM lock eliding for inflated locks in compiled code") \ 168 \ 169 experimental(bool, UseRTMForStackLocks, false, \ 170 "Enable RTM lock eliding for stack locks in compiled code") \ 171 \ 172 product(bool, UseRTMDeopt, false, \ 173 "Perform deopt and recompilation based on RTM abort ratio") \ 174 \ 175 product(int, RTMRetryCount, 5, \ 176 "Number of RTM retries on lock abort or busy") \ 177 range(0, max_jint) \ 178 \ 179 experimental(int, RTMSpinLoopCount, 100, \ 180 "Spin count for lock to become free before RTM retry") \ 181 range(0, max_jint) \ 182 \ 183 experimental(int, RTMAbortThreshold, 1000, \ 184 "Calculate abort ratio after this number of aborts") \ 185 range(0, max_jint) \ 186 \ 187 experimental(int, RTMLockingThreshold, 10000, \ 188 "Lock count at which to do RTM lock eliding without " \ 189 "abort ratio calculation") \ 190 range(0, max_jint) \ 191 \ 192 experimental(int, RTMAbortRatio, 50, \ 193 "Lock abort ratio at which to stop use RTM lock eliding") \ 194 range(0, 100) /* natural range */ \ 195 \ 196 experimental(int, RTMTotalCountIncrRate, 64, \ 197 "Increment total RTM attempted lock count once every n times") \ 198 range(1, max_jint) \ 199 constraint(RTMTotalCountIncrRateConstraintFunc,AfterErgo) \ 200 \ 201 experimental(intx, RTMLockingCalculationDelay, 0, \ 202 "Number of milliseconds to wait before start calculating aborts " \ 203 "for RTM locking") \ 204 \ 205 experimental(bool, UseRTMXendForLockBusy, true, \ 206 "Use RTM Xend instead of Xabort when lock busy") \ 207 \ 208 /* assembler */ \ 209 product(bool, UseCountLeadingZerosInstruction, false, \ 210 "Use count leading zeros instruction") \ 211 \ 212 product(bool, UseCountTrailingZerosInstruction, false, \ 213 "Use count trailing zeros instruction") \ 214 \ 215 product(bool, UseSSE42Intrinsics, false, \ 216 "SSE4.2 versions of intrinsics") \ 217 \ 218 product(bool, UseBMI1Instructions, false, \ 219 "Use BMI1 instructions") \ 220 \ 221 product(bool, UseBMI2Instructions, false, \ 222 "Use BMI2 instructions") \ 223 \ 224 diagnostic(bool, UseLibmIntrinsic, true, \ 225 "Use Libm Intrinsics") \ 226 \ 227 /* Minimum array size in bytes to use AVX512 intrinsics */ \ 228 /* for copy, inflate and fill which don't bail out early based on any */ \ 229 /* condition. When this value is set to zero compare operations like */ \ 230 /* compare, vectorizedMismatch, compress can also use AVX512 intrinsics.*/\ 231 diagnostic(int, AVX3Threshold, 4096, \ 232 "Minimum array size in bytes to use AVX512 intrinsics" \ 233 "for copy, inflate and fill. When this value is set as zero" \ 234 "compare operations can also use AVX512 intrinsics.") \ 235 range(0, max_jint) 236 #endif // CPU_X86_GLOBALS_X86_HPP