| File: | jdk/src/hotspot/share/opto/lcm.cpp |
| Warning: | line 810, column 12 Array access (from variable 'save_policy') results in a null pointer dereference |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* | ||||||
| 2 | * Copyright (c) 1998, 2021, 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 | #include "precompiled.hpp" | ||||||
| 26 | #include "asm/macroAssembler.inline.hpp" | ||||||
| 27 | #include "gc/shared/gc_globals.hpp" | ||||||
| 28 | #include "memory/allocation.inline.hpp" | ||||||
| 29 | #include "oops/compressedOops.hpp" | ||||||
| 30 | #include "opto/ad.hpp" | ||||||
| 31 | #include "opto/block.hpp" | ||||||
| 32 | #include "opto/c2compiler.hpp" | ||||||
| 33 | #include "opto/callnode.hpp" | ||||||
| 34 | #include "opto/cfgnode.hpp" | ||||||
| 35 | #include "opto/machnode.hpp" | ||||||
| 36 | #include "opto/runtime.hpp" | ||||||
| 37 | #include "opto/chaitin.hpp" | ||||||
| 38 | #include "runtime/sharedRuntime.hpp" | ||||||
| 39 | |||||||
| 40 | // Optimization - Graph Style | ||||||
| 41 | |||||||
| 42 | // Check whether val is not-null-decoded compressed oop, | ||||||
| 43 | // i.e. will grab into the base of the heap if it represents NULL. | ||||||
| 44 | static bool accesses_heap_base_zone(Node *val) { | ||||||
| 45 | if (CompressedOops::base() != NULL__null) { // Implies UseCompressedOops. | ||||||
| 46 | if (val && val->is_Mach()) { | ||||||
| 47 | if (val->as_Mach()->ideal_Opcode() == Op_DecodeN) { | ||||||
| 48 | // This assumes all Decodes with TypePtr::NotNull are matched to nodes that | ||||||
| 49 | // decode NULL to point to the heap base (Decode_NN). | ||||||
| 50 | if (val->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull) { | ||||||
| 51 | return true; | ||||||
| 52 | } | ||||||
| 53 | } | ||||||
| 54 | // Must recognize load operation with Decode matched in memory operand. | ||||||
| 55 | // We should not reach here exept for PPC/AIX, as os::zero_page_read_protected() | ||||||
| 56 | // returns true everywhere else. On PPC, no such memory operands | ||||||
| 57 | // exist, therefore we did not yet implement a check for such operands. | ||||||
| 58 | NOT_AIX(Unimplemented())do { (*g_assert_poison) = 'X';; report_unimplemented("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 58); ::breakpoint(); } while (0); | ||||||
| 59 | } | ||||||
| 60 | } | ||||||
| 61 | return false; | ||||||
| 62 | } | ||||||
| 63 | |||||||
| 64 | static bool needs_explicit_null_check_for_read(Node *val) { | ||||||
| 65 | // On some OSes (AIX) the page at address 0 is only write protected. | ||||||
| 66 | // If so, only Store operations will trap. | ||||||
| 67 | if (os::zero_page_read_protected()) { | ||||||
| 68 | return false; // Implicit null check will work. | ||||||
| 69 | } | ||||||
| 70 | // Also a read accessing the base of a heap-based compressed heap will trap. | ||||||
| 71 | if (accesses_heap_base_zone(val) && // Hits the base zone page. | ||||||
| 72 | CompressedOops::use_implicit_null_checks()) { // Base zone page is protected. | ||||||
| 73 | return false; | ||||||
| 74 | } | ||||||
| 75 | |||||||
| 76 | return true; | ||||||
| 77 | } | ||||||
| 78 | |||||||
| 79 | //------------------------------implicit_null_check---------------------------- | ||||||
| 80 | // Detect implicit-null-check opportunities. Basically, find NULL checks | ||||||
| 81 | // with suitable memory ops nearby. Use the memory op to do the NULL check. | ||||||
| 82 | // I can generate a memory op if there is not one nearby. | ||||||
| 83 | // The proj is the control projection for the not-null case. | ||||||
| 84 | // The val is the pointer being checked for nullness or | ||||||
| 85 | // decodeHeapOop_not_null node if it did not fold into address. | ||||||
| 86 | void PhaseCFG::implicit_null_check(Block* block, Node *proj, Node *val, int allowed_reasons) { | ||||||
| 87 | // Assume if null check need for 0 offset then always needed | ||||||
| 88 | // Intel solaris doesn't support any null checks yet and no | ||||||
| 89 | // mechanism exists (yet) to set the switches at an os_cpu level | ||||||
| 90 | if( !ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(0)) return; | ||||||
| 91 | |||||||
| 92 | // Make sure the ptr-is-null path appears to be uncommon! | ||||||
| 93 | float f = block->end()->as_MachIf()->_prob; | ||||||
| 94 | if( proj->Opcode() == Op_IfTrue ) f = 1.0f - f; | ||||||
| 95 | if( f > PROB_UNLIKELY_MAG(4)(1e-4f) ) return; | ||||||
| 96 | |||||||
| 97 | uint bidx = 0; // Capture index of value into memop | ||||||
| 98 | bool was_store; // Memory op is a store op | ||||||
| 99 | |||||||
| 100 | // Get the successor block for if the test ptr is non-null | ||||||
| 101 | Block* not_null_block; // this one goes with the proj | ||||||
| 102 | Block* null_block; | ||||||
| 103 | if (block->get_node(block->number_of_nodes()-1) == proj) { | ||||||
| 104 | null_block = block->_succs[0]; | ||||||
| 105 | not_null_block = block->_succs[1]; | ||||||
| 106 | } else { | ||||||
| 107 | assert(block->get_node(block->number_of_nodes()-2) == proj, "proj is one or the other")do { if (!(block->get_node(block->number_of_nodes()-2) == proj)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 107, "assert(" "block->get_node(block->number_of_nodes()-2) == proj" ") failed", "proj is one or the other"); ::breakpoint(); } } while (0); | ||||||
| 108 | not_null_block = block->_succs[0]; | ||||||
| 109 | null_block = block->_succs[1]; | ||||||
| 110 | } | ||||||
| 111 | while (null_block->is_Empty() == Block::empty_with_goto) { | ||||||
| 112 | null_block = null_block->_succs[0]; | ||||||
| 113 | } | ||||||
| 114 | |||||||
| 115 | // Search the exception block for an uncommon trap. | ||||||
| 116 | // (See Parse::do_if and Parse::do_ifnull for the reason | ||||||
| 117 | // we need an uncommon trap. Briefly, we need a way to | ||||||
| 118 | // detect failure of this optimization, as in 6366351.) | ||||||
| 119 | { | ||||||
| 120 | bool found_trap = false; | ||||||
| 121 | for (uint i1 = 0; i1 < null_block->number_of_nodes(); i1++) { | ||||||
| 122 | Node* nn = null_block->get_node(i1); | ||||||
| 123 | if (nn->is_MachCall() && | ||||||
| 124 | nn->as_MachCall()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point()) { | ||||||
| 125 | const Type* trtype = nn->in(TypeFunc::Parms)->bottom_type(); | ||||||
| 126 | if (trtype->isa_int() && trtype->is_int()->is_con()) { | ||||||
| 127 | jint tr_con = trtype->is_int()->get_con(); | ||||||
| 128 | Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(tr_con); | ||||||
| 129 | Deoptimization::DeoptAction action = Deoptimization::trap_request_action(tr_con); | ||||||
| 130 | assert((int)reason < (int)BitsPerInt, "recode bit map")do { if (!((int)reason < (int)BitsPerInt)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 130, "assert(" "(int)reason < (int)BitsPerInt" ") failed" , "recode bit map"); ::breakpoint(); } } while (0); | ||||||
| 131 | if (is_set_nth_bit(allowed_reasons, (int) reason) | ||||||
| 132 | && action != Deoptimization::Action_none) { | ||||||
| 133 | // This uncommon trap is sure to recompile, eventually. | ||||||
| 134 | // When that happens, C->too_many_traps will prevent | ||||||
| 135 | // this transformation from happening again. | ||||||
| 136 | found_trap = true; | ||||||
| 137 | } | ||||||
| 138 | } | ||||||
| 139 | break; | ||||||
| 140 | } | ||||||
| 141 | } | ||||||
| 142 | if (!found_trap) { | ||||||
| 143 | // We did not find an uncommon trap. | ||||||
| 144 | return; | ||||||
| 145 | } | ||||||
| 146 | } | ||||||
| 147 | |||||||
| 148 | // Check for decodeHeapOop_not_null node which did not fold into address | ||||||
| 149 | bool is_decoden = ((intptr_t)val) & 1; | ||||||
| 150 | val = (Node*)(((intptr_t)val) & ~1); | ||||||
| 151 | |||||||
| 152 | assert(!is_decoden || (val->in(0) == NULL) && val->is_Mach() &&do { if (!(!is_decoden || (val->in(0) == __null) && val->is_Mach() && (val->as_Mach()->ideal_Opcode () == Op_DecodeN))) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 153, "assert(" "!is_decoden || (val->in(0) == __null) && val->is_Mach() && (val->as_Mach()->ideal_Opcode() == Op_DecodeN)" ") failed", "sanity"); ::breakpoint(); } } while (0) | ||||||
| 153 | (val->as_Mach()->ideal_Opcode() == Op_DecodeN), "sanity")do { if (!(!is_decoden || (val->in(0) == __null) && val->is_Mach() && (val->as_Mach()->ideal_Opcode () == Op_DecodeN))) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 153, "assert(" "!is_decoden || (val->in(0) == __null) && val->is_Mach() && (val->as_Mach()->ideal_Opcode() == Op_DecodeN)" ") failed", "sanity"); ::breakpoint(); } } while (0); | ||||||
| 154 | |||||||
| 155 | // Search the successor block for a load or store who's base value is also | ||||||
| 156 | // the tested value. There may be several. | ||||||
| 157 | MachNode *best = NULL__null; // Best found so far | ||||||
| 158 | for (DUIterator i = val->outs(); val->has_out(i); i++) { | ||||||
| 159 | Node *m = val->out(i); | ||||||
| 160 | if( !m->is_Mach() ) continue; | ||||||
| 161 | MachNode *mach = m->as_Mach(); | ||||||
| 162 | was_store = false; | ||||||
| 163 | int iop = mach->ideal_Opcode(); | ||||||
| 164 | switch( iop ) { | ||||||
| 165 | case Op_LoadB: | ||||||
| 166 | case Op_LoadUB: | ||||||
| 167 | case Op_LoadUS: | ||||||
| 168 | case Op_LoadD: | ||||||
| 169 | case Op_LoadF: | ||||||
| 170 | case Op_LoadI: | ||||||
| 171 | case Op_LoadL: | ||||||
| 172 | case Op_LoadP: | ||||||
| 173 | case Op_LoadN: | ||||||
| 174 | case Op_LoadS: | ||||||
| 175 | case Op_LoadKlass: | ||||||
| 176 | case Op_LoadNKlass: | ||||||
| 177 | case Op_LoadRange: | ||||||
| 178 | case Op_LoadD_unaligned: | ||||||
| 179 | case Op_LoadL_unaligned: | ||||||
| 180 | assert(mach->in(2) == val, "should be address")do { if (!(mach->in(2) == val)) { (*g_assert_poison) = 'X' ;; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 180, "assert(" "mach->in(2) == val" ") failed", "should be address" ); ::breakpoint(); } } while (0); | ||||||
| 181 | break; | ||||||
| 182 | case Op_StoreB: | ||||||
| 183 | case Op_StoreC: | ||||||
| 184 | case Op_StoreCM: | ||||||
| 185 | case Op_StoreD: | ||||||
| 186 | case Op_StoreF: | ||||||
| 187 | case Op_StoreI: | ||||||
| 188 | case Op_StoreL: | ||||||
| 189 | case Op_StoreP: | ||||||
| 190 | case Op_StoreN: | ||||||
| 191 | case Op_StoreNKlass: | ||||||
| 192 | was_store = true; // Memory op is a store op | ||||||
| 193 | // Stores will have their address in slot 2 (memory in slot 1). | ||||||
| 194 | // If the value being nul-checked is in another slot, it means we | ||||||
| 195 | // are storing the checked value, which does NOT check the value! | ||||||
| 196 | if( mach->in(2) != val ) continue; | ||||||
| 197 | break; // Found a memory op? | ||||||
| 198 | case Op_StrComp: | ||||||
| 199 | case Op_StrEquals: | ||||||
| 200 | case Op_StrIndexOf: | ||||||
| 201 | case Op_StrIndexOfChar: | ||||||
| 202 | case Op_AryEq: | ||||||
| 203 | case Op_StrInflatedCopy: | ||||||
| 204 | case Op_StrCompressedCopy: | ||||||
| 205 | case Op_EncodeISOArray: | ||||||
| 206 | case Op_HasNegatives: | ||||||
| 207 | // Not a legit memory op for implicit null check regardless of | ||||||
| 208 | // embedded loads | ||||||
| 209 | continue; | ||||||
| 210 | default: // Also check for embedded loads | ||||||
| 211 | if( !mach->needs_anti_dependence_check() ) | ||||||
| 212 | continue; // Not an memory op; skip it | ||||||
| 213 | if( must_clone[iop] ) { | ||||||
| 214 | // Do not move nodes which produce flags because | ||||||
| 215 | // RA will try to clone it to place near branch and | ||||||
| 216 | // it will cause recompilation, see clone_node(). | ||||||
| 217 | continue; | ||||||
| 218 | } | ||||||
| 219 | { | ||||||
| 220 | // Check that value is used in memory address in | ||||||
| 221 | // instructions with embedded load (CmpP val1,(val2+off)). | ||||||
| 222 | Node* base; | ||||||
| 223 | Node* index; | ||||||
| 224 | const MachOper* oper = mach->memory_inputs(base, index); | ||||||
| 225 | if (oper == NULL__null || oper == (MachOper*)-1) { | ||||||
| 226 | continue; // Not an memory op; skip it | ||||||
| 227 | } | ||||||
| 228 | if (val == base || | ||||||
| 229 | (val == index && val->bottom_type()->isa_narrowoop())) { | ||||||
| 230 | break; // Found it | ||||||
| 231 | } else { | ||||||
| 232 | continue; // Skip it | ||||||
| 233 | } | ||||||
| 234 | } | ||||||
| 235 | break; | ||||||
| 236 | } | ||||||
| 237 | |||||||
| 238 | // On some OSes (AIX) the page at address 0 is only write protected. | ||||||
| 239 | // If so, only Store operations will trap. | ||||||
| 240 | // But a read accessing the base of a heap-based compressed heap will trap. | ||||||
| 241 | if (!was_store && needs_explicit_null_check_for_read(val)) { | ||||||
| 242 | continue; | ||||||
| 243 | } | ||||||
| 244 | |||||||
| 245 | // Check that node's control edge is not-null block's head or dominates it, | ||||||
| 246 | // otherwise we can't hoist it because there are other control dependencies. | ||||||
| 247 | Node* ctrl = mach->in(0); | ||||||
| 248 | if (ctrl != NULL__null && !(ctrl == not_null_block->head() || | ||||||
| 249 | get_block_for_node(ctrl)->dominates(not_null_block))) { | ||||||
| 250 | continue; | ||||||
| 251 | } | ||||||
| 252 | |||||||
| 253 | // check if the offset is not too high for implicit exception | ||||||
| 254 | { | ||||||
| 255 | intptr_t offset = 0; | ||||||
| 256 | const TypePtr *adr_type = NULL__null; // Do not need this return value here | ||||||
| 257 | const Node* base = mach->get_base_and_disp(offset, adr_type); | ||||||
| 258 | if (base == NULL__null || base == NodeSentinel(Node*)-1) { | ||||||
| 259 | // Narrow oop address doesn't have base, only index. | ||||||
| 260 | // Give up if offset is beyond page size or if heap base is not protected. | ||||||
| 261 | if (val->bottom_type()->isa_narrowoop() && | ||||||
| 262 | (MacroAssembler::needs_explicit_null_check(offset) || | ||||||
| 263 | !CompressedOops::use_implicit_null_checks())) | ||||||
| 264 | continue; | ||||||
| 265 | // cannot reason about it; is probably not implicit null exception | ||||||
| 266 | } else { | ||||||
| 267 | const TypePtr* tptr; | ||||||
| 268 | if ((UseCompressedOops || UseCompressedClassPointers) && | ||||||
| 269 | (CompressedOops::shift() == 0 || CompressedKlassPointers::shift() == 0)) { | ||||||
| 270 | // 32-bits narrow oop can be the base of address expressions | ||||||
| 271 | tptr = base->get_ptr_type(); | ||||||
| 272 | } else { | ||||||
| 273 | // only regular oops are expected here | ||||||
| 274 | tptr = base->bottom_type()->is_ptr(); | ||||||
| 275 | } | ||||||
| 276 | // Give up if offset is not a compile-time constant. | ||||||
| 277 | if (offset == Type::OffsetBot || tptr->_offset == Type::OffsetBot) | ||||||
| 278 | continue; | ||||||
| 279 | offset += tptr->_offset; // correct if base is offseted | ||||||
| 280 | // Give up if reference is beyond page size. | ||||||
| 281 | if (MacroAssembler::needs_explicit_null_check(offset)) | ||||||
| 282 | continue; | ||||||
| 283 | // Give up if base is a decode node and the heap base is not protected. | ||||||
| 284 | if (base->is_Mach() && base->as_Mach()->ideal_Opcode() == Op_DecodeN && | ||||||
| 285 | !CompressedOops::use_implicit_null_checks()) | ||||||
| 286 | continue; | ||||||
| 287 | } | ||||||
| 288 | } | ||||||
| 289 | |||||||
| 290 | // Check ctrl input to see if the null-check dominates the memory op | ||||||
| 291 | Block *cb = get_block_for_node(mach); | ||||||
| 292 | cb = cb->_idom; // Always hoist at least 1 block | ||||||
| 293 | if( !was_store ) { // Stores can be hoisted only one block | ||||||
| 294 | while( cb->_dom_depth > (block->_dom_depth + 1)) | ||||||
| 295 | cb = cb->_idom; // Hoist loads as far as we want | ||||||
| 296 | // The non-null-block should dominate the memory op, too. Live | ||||||
| 297 | // range spilling will insert a spill in the non-null-block if it is | ||||||
| 298 | // needs to spill the memory op for an implicit null check. | ||||||
| 299 | if (cb->_dom_depth == (block->_dom_depth + 1)) { | ||||||
| 300 | if (cb != not_null_block) continue; | ||||||
| 301 | cb = cb->_idom; | ||||||
| 302 | } | ||||||
| 303 | } | ||||||
| 304 | if( cb != block ) continue; | ||||||
| 305 | |||||||
| 306 | // Found a memory user; see if it can be hoisted to check-block | ||||||
| 307 | uint vidx = 0; // Capture index of value into memop | ||||||
| 308 | uint j; | ||||||
| 309 | for( j = mach->req()-1; j > 0; j-- ) { | ||||||
| 310 | if( mach->in(j) == val ) { | ||||||
| 311 | vidx = j; | ||||||
| 312 | // Ignore DecodeN val which could be hoisted to where needed. | ||||||
| 313 | if( is_decoden ) continue; | ||||||
| 314 | } | ||||||
| 315 | // Block of memory-op input | ||||||
| 316 | Block *inb = get_block_for_node(mach->in(j)); | ||||||
| 317 | Block *b = block; // Start from nul check | ||||||
| 318 | while( b != inb && b->_dom_depth > inb->_dom_depth ) | ||||||
| 319 | b = b->_idom; // search upwards for input | ||||||
| 320 | // See if input dominates null check | ||||||
| 321 | if( b != inb ) | ||||||
| 322 | break; | ||||||
| 323 | } | ||||||
| 324 | if( j > 0 ) | ||||||
| 325 | continue; | ||||||
| 326 | Block *mb = get_block_for_node(mach); | ||||||
| 327 | // Hoisting stores requires more checks for the anti-dependence case. | ||||||
| 328 | // Give up hoisting if we have to move the store past any load. | ||||||
| 329 | if( was_store ) { | ||||||
| 330 | Block *b = mb; // Start searching here for a local load | ||||||
| 331 | // mach use (faulting) trying to hoist | ||||||
| 332 | // n might be blocker to hoisting | ||||||
| 333 | while( b != block ) { | ||||||
| 334 | uint k; | ||||||
| 335 | for( k = 1; k < b->number_of_nodes(); k++ ) { | ||||||
| 336 | Node *n = b->get_node(k); | ||||||
| 337 | if( n->needs_anti_dependence_check() && | ||||||
| 338 | n->in(LoadNode::Memory) == mach->in(StoreNode::Memory) ) | ||||||
| 339 | break; // Found anti-dependent load | ||||||
| 340 | } | ||||||
| 341 | if( k < b->number_of_nodes() ) | ||||||
| 342 | break; // Found anti-dependent load | ||||||
| 343 | // Make sure control does not do a merge (would have to check allpaths) | ||||||
| 344 | if( b->num_preds() != 2 ) break; | ||||||
| 345 | b = get_block_for_node(b->pred(1)); // Move up to predecessor block | ||||||
| 346 | } | ||||||
| 347 | if( b != block ) continue; | ||||||
| 348 | } | ||||||
| 349 | |||||||
| 350 | // Make sure this memory op is not already being used for a NullCheck | ||||||
| 351 | Node *e = mb->end(); | ||||||
| 352 | if( e->is_MachNullCheck() && e->in(1) == mach ) | ||||||
| 353 | continue; // Already being used as a NULL check | ||||||
| 354 | |||||||
| 355 | // Found a candidate! Pick one with least dom depth - the highest | ||||||
| 356 | // in the dom tree should be closest to the null check. | ||||||
| 357 | if (best == NULL__null || get_block_for_node(mach)->_dom_depth < get_block_for_node(best)->_dom_depth) { | ||||||
| 358 | best = mach; | ||||||
| 359 | bidx = vidx; | ||||||
| 360 | } | ||||||
| 361 | } | ||||||
| 362 | // No candidate! | ||||||
| 363 | if (best == NULL__null) { | ||||||
| 364 | return; | ||||||
| 365 | } | ||||||
| 366 | |||||||
| 367 | // ---- Found an implicit null check | ||||||
| 368 | #ifndef PRODUCT | ||||||
| 369 | extern int implicit_null_checks; | ||||||
| 370 | implicit_null_checks++; | ||||||
| 371 | #endif | ||||||
| 372 | |||||||
| 373 | if( is_decoden ) { | ||||||
| 374 | // Check if we need to hoist decodeHeapOop_not_null first. | ||||||
| 375 | Block *valb = get_block_for_node(val); | ||||||
| 376 | if( block != valb && block->_dom_depth < valb->_dom_depth ) { | ||||||
| 377 | // Hoist it up to the end of the test block together with its inputs if they exist. | ||||||
| 378 | for (uint i = 2; i < val->req(); i++) { | ||||||
| 379 | // DecodeN has 2 regular inputs + optional MachTemp or load Base inputs. | ||||||
| 380 | Node *temp = val->in(i); | ||||||
| 381 | Block *tempb = get_block_for_node(temp); | ||||||
| 382 | if (!tempb->dominates(block)) { | ||||||
| 383 | assert(block->dominates(tempb), "sanity check: temp node placement")do { if (!(block->dominates(tempb))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 383, "assert(" "block->dominates(tempb)" ") failed", "sanity check: temp node placement" ); ::breakpoint(); } } while (0); | ||||||
| 384 | // We only expect nodes without further inputs, like MachTemp or load Base. | ||||||
| 385 | assert(temp->req() == 0 || (temp->req() == 1 && temp->in(0) == (Node*)C->root()),do { if (!(temp->req() == 0 || (temp->req() == 1 && temp->in(0) == (Node*)C->root()))) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 386, "assert(" "temp->req() == 0 || (temp->req() == 1 && temp->in(0) == (Node*)C->root())" ") failed", "need for recursive hoisting not expected"); ::breakpoint (); } } while (0) | ||||||
| 386 | "need for recursive hoisting not expected")do { if (!(temp->req() == 0 || (temp->req() == 1 && temp->in(0) == (Node*)C->root()))) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 386, "assert(" "temp->req() == 0 || (temp->req() == 1 && temp->in(0) == (Node*)C->root())" ") failed", "need for recursive hoisting not expected"); ::breakpoint (); } } while (0); | ||||||
| 387 | tempb->find_remove(temp); | ||||||
| 388 | block->add_inst(temp); | ||||||
| 389 | map_node_to_block(temp, block); | ||||||
| 390 | } | ||||||
| 391 | } | ||||||
| 392 | valb->find_remove(val); | ||||||
| 393 | block->add_inst(val); | ||||||
| 394 | map_node_to_block(val, block); | ||||||
| 395 | // DecodeN on x86 may kill flags. Check for flag-killing projections | ||||||
| 396 | // that also need to be hoisted. | ||||||
| 397 | for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) { | ||||||
| 398 | Node* n = val->fast_out(j); | ||||||
| 399 | if( n->is_MachProj() ) { | ||||||
| 400 | get_block_for_node(n)->find_remove(n); | ||||||
| 401 | block->add_inst(n); | ||||||
| 402 | map_node_to_block(n, block); | ||||||
| 403 | } | ||||||
| 404 | } | ||||||
| 405 | } | ||||||
| 406 | } | ||||||
| 407 | // Hoist the memory candidate up to the end of the test block. | ||||||
| 408 | Block *old_block = get_block_for_node(best); | ||||||
| 409 | old_block->find_remove(best); | ||||||
| 410 | block->add_inst(best); | ||||||
| 411 | map_node_to_block(best, block); | ||||||
| 412 | |||||||
| 413 | // Move the control dependence if it is pinned to not-null block. | ||||||
| 414 | // Don't change it in other cases: NULL or dominating control. | ||||||
| 415 | Node* ctrl = best->in(0); | ||||||
| 416 | if (ctrl != NULL__null && get_block_for_node(ctrl) == not_null_block) { | ||||||
| 417 | // Set it to control edge of null check. | ||||||
| 418 | best->set_req(0, proj->in(0)->in(0)); | ||||||
| 419 | } | ||||||
| 420 | |||||||
| 421 | // Check for flag-killing projections that also need to be hoisted | ||||||
| 422 | // Should be DU safe because no edge updates. | ||||||
| 423 | for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) { | ||||||
| 424 | Node* n = best->fast_out(j); | ||||||
| 425 | if( n->is_MachProj() ) { | ||||||
| 426 | get_block_for_node(n)->find_remove(n); | ||||||
| 427 | block->add_inst(n); | ||||||
| 428 | map_node_to_block(n, block); | ||||||
| 429 | } | ||||||
| 430 | } | ||||||
| 431 | |||||||
| 432 | // proj==Op_True --> ne test; proj==Op_False --> eq test. | ||||||
| 433 | // One of two graph shapes got matched: | ||||||
| 434 | // (IfTrue (If (Bool NE (CmpP ptr NULL)))) | ||||||
| 435 | // (IfFalse (If (Bool EQ (CmpP ptr NULL)))) | ||||||
| 436 | // NULL checks are always branch-if-eq. If we see a IfTrue projection | ||||||
| 437 | // then we are replacing a 'ne' test with a 'eq' NULL check test. | ||||||
| 438 | // We need to flip the projections to keep the same semantics. | ||||||
| 439 | if( proj->Opcode() == Op_IfTrue ) { | ||||||
| 440 | // Swap order of projections in basic block to swap branch targets | ||||||
| 441 | Node *tmp1 = block->get_node(block->end_idx()+1); | ||||||
| 442 | Node *tmp2 = block->get_node(block->end_idx()+2); | ||||||
| 443 | block->map_node(tmp2, block->end_idx()+1); | ||||||
| 444 | block->map_node(tmp1, block->end_idx()+2); | ||||||
| 445 | Node *tmp = new Node(C->top()); // Use not NULL input | ||||||
| 446 | tmp1->replace_by(tmp); | ||||||
| 447 | tmp2->replace_by(tmp1); | ||||||
| 448 | tmp->replace_by(tmp2); | ||||||
| 449 | tmp->destruct(NULL__null); | ||||||
| 450 | } | ||||||
| 451 | |||||||
| 452 | // Remove the existing null check; use a new implicit null check instead. | ||||||
| 453 | // Since schedule-local needs precise def-use info, we need to correct | ||||||
| 454 | // it as well. | ||||||
| 455 | Node *old_tst = proj->in(0); | ||||||
| 456 | MachNode *nul_chk = new MachNullCheckNode(old_tst->in(0),best,bidx); | ||||||
| 457 | block->map_node(nul_chk, block->end_idx()); | ||||||
| 458 | map_node_to_block(nul_chk, block); | ||||||
| 459 | // Redirect users of old_test to nul_chk | ||||||
| 460 | for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2) | ||||||
| 461 | old_tst->last_out(i2)->set_req(0, nul_chk); | ||||||
| 462 | // Clean-up any dead code | ||||||
| 463 | for (uint i3 = 0; i3 < old_tst->req(); i3++) { | ||||||
| 464 | Node* in = old_tst->in(i3); | ||||||
| 465 | old_tst->set_req(i3, NULL__null); | ||||||
| 466 | if (in->outcnt() == 0) { | ||||||
| 467 | // Remove dead input node | ||||||
| 468 | in->disconnect_inputs(C); | ||||||
| 469 | block->find_remove(in); | ||||||
| 470 | } | ||||||
| 471 | } | ||||||
| 472 | |||||||
| 473 | latency_from_uses(nul_chk); | ||||||
| 474 | latency_from_uses(best); | ||||||
| 475 | |||||||
| 476 | // insert anti-dependences to defs in this block | ||||||
| 477 | if (! best->needs_anti_dependence_check()) { | ||||||
| 478 | for (uint k = 1; k < block->number_of_nodes(); k++) { | ||||||
| 479 | Node *n = block->get_node(k); | ||||||
| 480 | if (n->needs_anti_dependence_check() && | ||||||
| 481 | n->in(LoadNode::Memory) == best->in(StoreNode::Memory)) { | ||||||
| 482 | // Found anti-dependent load | ||||||
| 483 | insert_anti_dependences(block, n); | ||||||
| 484 | } | ||||||
| 485 | } | ||||||
| 486 | } | ||||||
| 487 | } | ||||||
| 488 | |||||||
| 489 | |||||||
| 490 | //------------------------------select----------------------------------------- | ||||||
| 491 | // Select a nice fellow from the worklist to schedule next. If there is only | ||||||
| 492 | // one choice, then use it. Projections take top priority for correctness | ||||||
| 493 | // reasons - if I see a projection, then it is next. There are a number of | ||||||
| 494 | // other special cases, for instructions that consume condition codes, et al. | ||||||
| 495 | // These are chosen immediately. Some instructions are required to immediately | ||||||
| 496 | // precede the last instruction in the block, and these are taken last. Of the | ||||||
| 497 | // remaining cases (most), choose the instruction with the greatest latency | ||||||
| 498 | // (that is, the most number of pseudo-cycles required to the end of the | ||||||
| 499 | // routine). If there is a tie, choose the instruction with the most inputs. | ||||||
| 500 | Node* PhaseCFG::select( | ||||||
| 501 | Block* block, | ||||||
| 502 | Node_List &worklist, | ||||||
| 503 | GrowableArray<int> &ready_cnt, | ||||||
| 504 | VectorSet &next_call, | ||||||
| 505 | uint sched_slot, | ||||||
| 506 | intptr_t* recalc_pressure_nodes) { | ||||||
| 507 | |||||||
| 508 | // If only a single entry on the stack, use it | ||||||
| 509 | uint cnt = worklist.size(); | ||||||
| 510 | if (cnt == 1) { | ||||||
| 511 | Node *n = worklist[0]; | ||||||
| 512 | worklist.map(0,worklist.pop()); | ||||||
| 513 | return n; | ||||||
| 514 | } | ||||||
| 515 | |||||||
| 516 | uint choice = 0; // Bigger is most important | ||||||
| 517 | uint latency = 0; // Bigger is scheduled first | ||||||
| 518 | uint score = 0; // Bigger is better | ||||||
| 519 | int idx = -1; // Index in worklist | ||||||
| 520 | int cand_cnt = 0; // Candidate count | ||||||
| 521 | bool block_size_threshold_ok = (recalc_pressure_nodes != NULL__null) && (block->number_of_nodes() > 10); | ||||||
| 522 | |||||||
| 523 | for( uint i=0; i<cnt; i++ ) { // Inspect entire worklist | ||||||
| 524 | // Order in worklist is used to break ties. | ||||||
| 525 | // See caller for how this is used to delay scheduling | ||||||
| 526 | // of induction variable increments to after the other | ||||||
| 527 | // uses of the phi are scheduled. | ||||||
| 528 | Node *n = worklist[i]; // Get Node on worklist | ||||||
| 529 | |||||||
| 530 | int iop = n->is_Mach() ? n->as_Mach()->ideal_Opcode() : 0; | ||||||
| 531 | if( n->is_Proj() || // Projections always win | ||||||
| 532 | n->Opcode()== Op_Con || // So does constant 'Top' | ||||||
| 533 | iop == Op_CreateEx || // Create-exception must start block | ||||||
| 534 | iop == Op_CheckCastPP | ||||||
| 535 | ) { | ||||||
| 536 | worklist.map(i,worklist.pop()); | ||||||
| 537 | return n; | ||||||
| 538 | } | ||||||
| 539 | |||||||
| 540 | // Final call in a block must be adjacent to 'catch' | ||||||
| 541 | Node *e = block->end(); | ||||||
| 542 | if( e->is_Catch() && e->in(0)->in(0) == n ) | ||||||
| 543 | continue; | ||||||
| 544 | |||||||
| 545 | // Memory op for an implicit null check has to be at the end of the block | ||||||
| 546 | if( e->is_MachNullCheck() && e->in(1) == n ) | ||||||
| 547 | continue; | ||||||
| 548 | |||||||
| 549 | // Schedule IV increment last. | ||||||
| 550 | if (e->is_Mach() && e->as_Mach()->ideal_Opcode() == Op_CountedLoopEnd) { | ||||||
| 551 | // Cmp might be matched into CountedLoopEnd node. | ||||||
| 552 | Node *cmp = (e->in(1)->ideal_reg() == Op_RegFlags) ? e->in(1) : e; | ||||||
| 553 | if (cmp->req() > 1 && cmp->in(1) == n && n->is_iteratively_computed()) { | ||||||
| 554 | continue; | ||||||
| 555 | } | ||||||
| 556 | } | ||||||
| 557 | |||||||
| 558 | uint n_choice = 2; | ||||||
| 559 | |||||||
| 560 | // See if this instruction is consumed by a branch. If so, then (as the | ||||||
| 561 | // branch is the last instruction in the basic block) force it to the | ||||||
| 562 | // end of the basic block | ||||||
| 563 | if ( must_clone[iop] ) { | ||||||
| 564 | // See if any use is a branch | ||||||
| 565 | bool found_machif = false; | ||||||
| 566 | |||||||
| 567 | for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { | ||||||
| 568 | Node* use = n->fast_out(j); | ||||||
| 569 | |||||||
| 570 | // The use is a conditional branch, make them adjacent | ||||||
| 571 | if (use->is_MachIf() && get_block_for_node(use) == block) { | ||||||
| 572 | found_machif = true; | ||||||
| 573 | break; | ||||||
| 574 | } | ||||||
| 575 | |||||||
| 576 | // More than this instruction pending for successor to be ready, | ||||||
| 577 | // don't choose this if other opportunities are ready | ||||||
| 578 | if (ready_cnt.at(use->_idx) > 1) | ||||||
| 579 | n_choice = 1; | ||||||
| 580 | } | ||||||
| 581 | |||||||
| 582 | // loop terminated, prefer not to use this instruction | ||||||
| 583 | if (found_machif) | ||||||
| 584 | continue; | ||||||
| 585 | } | ||||||
| 586 | |||||||
| 587 | // See if this has a predecessor that is "must_clone", i.e. sets the | ||||||
| 588 | // condition code. If so, choose this first | ||||||
| 589 | for (uint j = 0; j < n->req() ; j++) { | ||||||
| 590 | Node *inn = n->in(j); | ||||||
| 591 | if (inn) { | ||||||
| 592 | if (inn->is_Mach() && must_clone[inn->as_Mach()->ideal_Opcode()] ) { | ||||||
| 593 | n_choice = 3; | ||||||
| 594 | break; | ||||||
| 595 | } | ||||||
| 596 | } | ||||||
| 597 | } | ||||||
| 598 | |||||||
| 599 | // MachTemps should be scheduled last so they are near their uses | ||||||
| 600 | if (n->is_MachTemp()) { | ||||||
| 601 | n_choice = 1; | ||||||
| 602 | } | ||||||
| 603 | |||||||
| 604 | uint n_latency = get_latency_for_node(n); | ||||||
| 605 | uint n_score = n->req(); // Many inputs get high score to break ties | ||||||
| 606 | |||||||
| 607 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 608 | if (recalc_pressure_nodes[n->_idx] == 0x7fff7fff) { | ||||||
| 609 | _regalloc->_scratch_int_pressure.init(_regalloc->_sched_int_pressure.high_pressure_limit()); | ||||||
| 610 | _regalloc->_scratch_float_pressure.init(_regalloc->_sched_float_pressure.high_pressure_limit()); | ||||||
| 611 | // simulate the notion that we just picked this node to schedule | ||||||
| 612 | n->add_flag(Node::Flag_is_scheduled); | ||||||
| 613 | // now caculate its effect upon the graph if we did | ||||||
| 614 | adjust_register_pressure(n, block, recalc_pressure_nodes, false); | ||||||
| 615 | // return its state for finalize in case somebody else wins | ||||||
| 616 | n->remove_flag(Node::Flag_is_scheduled); | ||||||
| 617 | // now save the two final pressure components of register pressure, limiting pressure calcs to short size | ||||||
| 618 | short int_pressure = (short)_regalloc->_scratch_int_pressure.current_pressure(); | ||||||
| 619 | short float_pressure = (short)_regalloc->_scratch_float_pressure.current_pressure(); | ||||||
| 620 | recalc_pressure_nodes[n->_idx] = int_pressure; | ||||||
| 621 | recalc_pressure_nodes[n->_idx] |= (float_pressure << 16); | ||||||
| 622 | } | ||||||
| 623 | |||||||
| 624 | if (_scheduling_for_pressure) { | ||||||
| 625 | latency = n_latency; | ||||||
| 626 | if (n_choice != 3) { | ||||||
| 627 | // Now evaluate each register pressure component based on threshold in the score. | ||||||
| 628 | // In general the defining register type will dominate the score, ergo we will not see register pressure grow on both banks | ||||||
| 629 | // on a single instruction, but we might see it shrink on both banks. | ||||||
| 630 | // For each use of register that has a register class that is over the high pressure limit, we build n_score up for | ||||||
| 631 | // live ranges that terminate on this instruction. | ||||||
| 632 | if (_regalloc->_sched_int_pressure.current_pressure() > _regalloc->_sched_int_pressure.high_pressure_limit()) { | ||||||
| 633 | short int_pressure = (short)recalc_pressure_nodes[n->_idx]; | ||||||
| 634 | n_score = (int_pressure < 0) ? ((score + n_score) - int_pressure) : (int_pressure > 0) ? 1 : n_score; | ||||||
| 635 | } | ||||||
| 636 | if (_regalloc->_sched_float_pressure.current_pressure() > _regalloc->_sched_float_pressure.high_pressure_limit()) { | ||||||
| 637 | short float_pressure = (short)(recalc_pressure_nodes[n->_idx] >> 16); | ||||||
| 638 | n_score = (float_pressure < 0) ? ((score + n_score) - float_pressure) : (float_pressure > 0) ? 1 : n_score; | ||||||
| 639 | } | ||||||
| 640 | } else { | ||||||
| 641 | // make sure we choose these candidates | ||||||
| 642 | score = 0; | ||||||
| 643 | } | ||||||
| 644 | } | ||||||
| 645 | } | ||||||
| 646 | |||||||
| 647 | // Keep best latency found | ||||||
| 648 | cand_cnt++; | ||||||
| 649 | if (choice < n_choice || | ||||||
| 650 | (choice == n_choice && | ||||||
| 651 | ((StressLCM && C->randomized_select(cand_cnt)) || | ||||||
| 652 | (!StressLCM && | ||||||
| 653 | (latency < n_latency || | ||||||
| 654 | (latency == n_latency && | ||||||
| 655 | (score < n_score))))))) { | ||||||
| 656 | choice = n_choice; | ||||||
| 657 | latency = n_latency; | ||||||
| 658 | score = n_score; | ||||||
| 659 | idx = i; // Also keep index in worklist | ||||||
| 660 | } | ||||||
| 661 | } // End of for all ready nodes in worklist | ||||||
| 662 | |||||||
| 663 | guarantee(idx >= 0, "index should be set")do { if (!(idx >= 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 663, "guarantee(" "idx >= 0" ") failed", "index should be set" ); ::breakpoint(); } } while (0); | ||||||
| 664 | Node *n = worklist[(uint)idx]; // Get the winner | ||||||
| 665 | |||||||
| 666 | worklist.map((uint)idx, worklist.pop()); // Compress worklist | ||||||
| 667 | return n; | ||||||
| 668 | } | ||||||
| 669 | |||||||
| 670 | //-------------------------adjust_register_pressure---------------------------- | ||||||
| 671 | void PhaseCFG::adjust_register_pressure(Node* n, Block* block, intptr_t* recalc_pressure_nodes, bool finalize_mode) { | ||||||
| 672 | PhaseLive* liveinfo = _regalloc->get_live(); | ||||||
| 673 | IndexSet* liveout = liveinfo->live(block); | ||||||
| 674 | // first adjust the register pressure for the sources | ||||||
| 675 | for (uint i = 1; i < n->req(); i++) { | ||||||
| 676 | bool lrg_ends = false; | ||||||
| 677 | Node *src_n = n->in(i); | ||||||
| 678 | if (src_n == NULL__null) continue; | ||||||
| 679 | if (!src_n->is_Mach()) continue; | ||||||
| 680 | uint src = _regalloc->_lrg_map.find(src_n); | ||||||
| 681 | if (src == 0) continue; | ||||||
| 682 | LRG& lrg_src = _regalloc->lrgs(src); | ||||||
| 683 | // detect if the live range ends or not | ||||||
| 684 | if (liveout->member(src) == false) { | ||||||
| 685 | lrg_ends = true; | ||||||
| 686 | for (DUIterator_Fast jmax, j = src_n->fast_outs(jmax); j < jmax; j++) { | ||||||
| 687 | Node* m = src_n->fast_out(j); // Get user | ||||||
| 688 | if (m == n) continue; | ||||||
| 689 | if (!m->is_Mach()) continue; | ||||||
| 690 | MachNode *mach = m->as_Mach(); | ||||||
| 691 | bool src_matches = false; | ||||||
| 692 | int iop = mach->ideal_Opcode(); | ||||||
| 693 | |||||||
| 694 | switch (iop) { | ||||||
| 695 | case Op_StoreB: | ||||||
| 696 | case Op_StoreC: | ||||||
| 697 | case Op_StoreCM: | ||||||
| 698 | case Op_StoreD: | ||||||
| 699 | case Op_StoreF: | ||||||
| 700 | case Op_StoreI: | ||||||
| 701 | case Op_StoreL: | ||||||
| 702 | case Op_StoreP: | ||||||
| 703 | case Op_StoreN: | ||||||
| 704 | case Op_StoreVector: | ||||||
| 705 | case Op_StoreVectorMasked: | ||||||
| 706 | case Op_StoreVectorScatter: | ||||||
| 707 | case Op_StoreVectorScatterMasked: | ||||||
| 708 | case Op_StoreNKlass: | ||||||
| 709 | for (uint k = 1; k < m->req(); k++) { | ||||||
| 710 | Node *in = m->in(k); | ||||||
| 711 | if (in == src_n) { | ||||||
| 712 | src_matches = true; | ||||||
| 713 | break; | ||||||
| 714 | } | ||||||
| 715 | } | ||||||
| 716 | break; | ||||||
| 717 | |||||||
| 718 | default: | ||||||
| 719 | src_matches = true; | ||||||
| 720 | break; | ||||||
| 721 | } | ||||||
| 722 | |||||||
| 723 | // If we have a store as our use, ignore the non source operands | ||||||
| 724 | if (src_matches == false) continue; | ||||||
| 725 | |||||||
| 726 | // Mark every unscheduled use which is not n with a recalculation | ||||||
| 727 | if ((get_block_for_node(m) == block) && (!m->is_scheduled())) { | ||||||
| 728 | if (finalize_mode && !m->is_Phi()) { | ||||||
| 729 | recalc_pressure_nodes[m->_idx] = 0x7fff7fff; | ||||||
| 730 | } | ||||||
| 731 | lrg_ends = false; | ||||||
| 732 | } | ||||||
| 733 | } | ||||||
| 734 | } | ||||||
| 735 | // if none, this live range ends and we can adjust register pressure | ||||||
| 736 | if (lrg_ends) { | ||||||
| 737 | if (finalize_mode) { | ||||||
| 738 | _regalloc->lower_pressure(block, 0, lrg_src, NULL__null, _regalloc->_sched_int_pressure, _regalloc->_sched_float_pressure); | ||||||
| 739 | } else { | ||||||
| 740 | _regalloc->lower_pressure(block, 0, lrg_src, NULL__null, _regalloc->_scratch_int_pressure, _regalloc->_scratch_float_pressure); | ||||||
| 741 | } | ||||||
| 742 | } | ||||||
| 743 | } | ||||||
| 744 | |||||||
| 745 | // now add the register pressure from the dest and evaluate which heuristic we should use: | ||||||
| 746 | // 1.) The default, latency scheduling | ||||||
| 747 | // 2.) Register pressure scheduling based on the high pressure limit threshold for int or float register stacks | ||||||
| 748 | uint dst = _regalloc->_lrg_map.find(n); | ||||||
| 749 | if (dst != 0) { | ||||||
| 750 | LRG& lrg_dst = _regalloc->lrgs(dst); | ||||||
| 751 | if (finalize_mode) { | ||||||
| 752 | _regalloc->raise_pressure(block, lrg_dst, _regalloc->_sched_int_pressure, _regalloc->_sched_float_pressure); | ||||||
| 753 | // check to see if we fall over the register pressure cliff here | ||||||
| 754 | if (_regalloc->_sched_int_pressure.current_pressure() > _regalloc->_sched_int_pressure.high_pressure_limit()) { | ||||||
| 755 | _scheduling_for_pressure = true; | ||||||
| 756 | } else if (_regalloc->_sched_float_pressure.current_pressure() > _regalloc->_sched_float_pressure.high_pressure_limit()) { | ||||||
| 757 | _scheduling_for_pressure = true; | ||||||
| 758 | } else { | ||||||
| 759 | // restore latency scheduling mode | ||||||
| 760 | _scheduling_for_pressure = false; | ||||||
| 761 | } | ||||||
| 762 | } else { | ||||||
| 763 | _regalloc->raise_pressure(block, lrg_dst, _regalloc->_scratch_int_pressure, _regalloc->_scratch_float_pressure); | ||||||
| 764 | } | ||||||
| 765 | } | ||||||
| 766 | } | ||||||
| 767 | |||||||
| 768 | //------------------------------set_next_call---------------------------------- | ||||||
| 769 | void PhaseCFG::set_next_call(Block* block, Node* n, VectorSet& next_call) { | ||||||
| 770 | if( next_call.test_set(n->_idx) ) return; | ||||||
| 771 | for( uint i=0; i<n->len(); i++ ) { | ||||||
| 772 | Node *m = n->in(i); | ||||||
| 773 | if( !m ) continue; // must see all nodes in block that precede call | ||||||
| 774 | if (get_block_for_node(m) == block) { | ||||||
| 775 | set_next_call(block, m, next_call); | ||||||
| 776 | } | ||||||
| 777 | } | ||||||
| 778 | } | ||||||
| 779 | |||||||
| 780 | //------------------------------needed_for_next_call--------------------------- | ||||||
| 781 | // Set the flag 'next_call' for each Node that is needed for the next call to | ||||||
| 782 | // be scheduled. This flag lets me bias scheduling so Nodes needed for the | ||||||
| 783 | // next subroutine call get priority - basically it moves things NOT needed | ||||||
| 784 | // for the next call till after the call. This prevents me from trying to | ||||||
| 785 | // carry lots of stuff live across a call. | ||||||
| 786 | void PhaseCFG::needed_for_next_call(Block* block, Node* this_call, VectorSet& next_call) { | ||||||
| 787 | // Find the next control-defining Node in this block | ||||||
| 788 | Node* call = NULL__null; | ||||||
| 789 | for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) { | ||||||
| 790 | Node* m = this_call->fast_out(i); | ||||||
| 791 | if (get_block_for_node(m) == block && // Local-block user | ||||||
| 792 | m != this_call && // Not self-start node | ||||||
| 793 | m->is_MachCall()) { | ||||||
| 794 | call = m; | ||||||
| 795 | break; | ||||||
| 796 | } | ||||||
| 797 | } | ||||||
| 798 | if (call == NULL__null) return; // No next call (e.g., block end is near) | ||||||
| 799 | // Set next-call for all inputs to this call | ||||||
| 800 | set_next_call(block, call, next_call); | ||||||
| 801 | } | ||||||
| 802 | |||||||
| 803 | //------------------------------add_call_kills------------------------------------- | ||||||
| 804 | // helper function that adds caller save registers to MachProjNode | ||||||
| 805 | static void add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe) { | ||||||
| 806 | // Fill in the kill mask for the call | ||||||
| 807 | for( OptoReg::Name r = OptoReg::Name(0); r < _last_Mach_Reg; r=OptoReg::add(r,1) ) { | ||||||
| 808 | if( !regs.Member(r) ) { // Not already defined by the call | ||||||
| 809 | // Save-on-call register? | ||||||
| 810 | if ((save_policy[r] == 'C') || | ||||||
| |||||||
| 811 | (save_policy[r] == 'A') || | ||||||
| 812 | ((save_policy[r] == 'E') && exclude_soe)) { | ||||||
| 813 | proj->_rout.Insert(r); | ||||||
| 814 | } | ||||||
| 815 | } | ||||||
| 816 | } | ||||||
| 817 | } | ||||||
| 818 | |||||||
| 819 | |||||||
| 820 | //------------------------------sched_call------------------------------------- | ||||||
| 821 | uint PhaseCFG::sched_call(Block* block, uint node_cnt, Node_List& worklist, GrowableArray<int>& ready_cnt, MachCallNode* mcall, VectorSet& next_call) { | ||||||
| 822 | RegMask regs; | ||||||
| 823 | |||||||
| 824 | // Schedule all the users of the call right now. All the users are | ||||||
| 825 | // projection Nodes, so they must be scheduled next to the call. | ||||||
| 826 | // Collect all the defined registers. | ||||||
| 827 | for (DUIterator_Fast imax, i = mcall->fast_outs(imax); i < imax; i++) { | ||||||
| |||||||
| 828 | Node* n = mcall->fast_out(i); | ||||||
| 829 | assert( n->is_MachProj(), "" )do { if (!(n->is_MachProj())) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 829, "assert(" "n->is_MachProj()" ") failed", ""); ::breakpoint (); } } while (0); | ||||||
| 830 | int n_cnt = ready_cnt.at(n->_idx)-1; | ||||||
| 831 | ready_cnt.at_put(n->_idx, n_cnt); | ||||||
| 832 | assert( n_cnt == 0, "" )do { if (!(n_cnt == 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 832, "assert(" "n_cnt == 0" ") failed", ""); ::breakpoint() ; } } while (0); | ||||||
| 833 | // Schedule next to call | ||||||
| 834 | block->map_node(n, node_cnt++); | ||||||
| 835 | // Collect defined registers | ||||||
| 836 | regs.OR(n->out_RegMask()); | ||||||
| 837 | // Check for scheduling the next control-definer | ||||||
| 838 | if( n->bottom_type() == Type::CONTROL ) | ||||||
| 839 | // Warm up next pile of heuristic bits | ||||||
| 840 | needed_for_next_call(block, n, next_call); | ||||||
| 841 | |||||||
| 842 | // Children of projections are now all ready | ||||||
| 843 | for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { | ||||||
| 844 | Node* m = n->fast_out(j); // Get user | ||||||
| 845 | if(get_block_for_node(m) != block) { | ||||||
| 846 | continue; | ||||||
| 847 | } | ||||||
| 848 | if( m->is_Phi() ) continue; | ||||||
| 849 | int m_cnt = ready_cnt.at(m->_idx) - 1; | ||||||
| 850 | ready_cnt.at_put(m->_idx, m_cnt); | ||||||
| 851 | if( m_cnt == 0 ) | ||||||
| 852 | worklist.push(m); | ||||||
| 853 | } | ||||||
| 854 | |||||||
| 855 | } | ||||||
| 856 | |||||||
| 857 | // Act as if the call defines the Frame Pointer. | ||||||
| 858 | // Certainly the FP is alive and well after the call. | ||||||
| 859 | regs.Insert(_matcher.c_frame_pointer()); | ||||||
| 860 | |||||||
| 861 | // Set all registers killed and not already defined by the call. | ||||||
| 862 | uint r_cnt = mcall->tf()->range()->cnt(); | ||||||
| 863 | int op = mcall->ideal_Opcode(); | ||||||
| 864 | MachProjNode *proj = new MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj ); | ||||||
| 865 | map_node_to_block(proj, block); | ||||||
| 866 | block->insert_node(proj, node_cnt++); | ||||||
| 867 | |||||||
| 868 | // Select the right register save policy. | ||||||
| 869 | const char *save_policy = NULL__null; | ||||||
| 870 | switch (op) { | ||||||
| 871 | case Op_CallRuntime: | ||||||
| 872 | case Op_CallLeaf: | ||||||
| 873 | case Op_CallLeafNoFP: | ||||||
| 874 | case Op_CallLeafVector: | ||||||
| 875 | // Calling C code so use C calling convention | ||||||
| 876 | save_policy = _matcher._c_reg_save_policy; | ||||||
| 877 | break; | ||||||
| 878 | |||||||
| 879 | case Op_CallStaticJava: | ||||||
| 880 | case Op_CallDynamicJava: | ||||||
| 881 | // Calling Java code so use Java calling convention | ||||||
| 882 | save_policy = _matcher._register_save_policy; | ||||||
| 883 | break; | ||||||
| 884 | case Op_CallNative: | ||||||
| 885 | // We use the c reg save policy here since Foreign Linker | ||||||
| 886 | // only supports the C ABI currently. | ||||||
| 887 | // TODO compute actual save policy based on nep->abi | ||||||
| 888 | save_policy = _matcher._c_reg_save_policy; | ||||||
| 889 | break; | ||||||
| 890 | |||||||
| 891 | default: | ||||||
| 892 | ShouldNotReachHere()do { (*g_assert_poison) = 'X';; report_should_not_reach_here( "/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 892); ::breakpoint(); } while (0); | ||||||
| 893 | } | ||||||
| 894 | |||||||
| 895 | // When using CallRuntime mark SOE registers as killed by the call | ||||||
| 896 | // so values that could show up in the RegisterMap aren't live in a | ||||||
| 897 | // callee saved register since the register wouldn't know where to | ||||||
| 898 | // find them. CallLeaf and CallLeafNoFP are ok because they can't | ||||||
| 899 | // have debug info on them. Strictly speaking this only needs to be | ||||||
| 900 | // done for oops since idealreg2debugmask takes care of debug info | ||||||
| 901 | // references but there no way to handle oops differently than other | ||||||
| 902 | // pointers as far as the kill mask goes. | ||||||
| 903 | // | ||||||
| 904 | // Also, native callees can not save oops, so we kill the SOE registers | ||||||
| 905 | // here in case a native call has a safepoint. This doesn't work for | ||||||
| 906 | // RBP though, which seems to be special-cased elsewhere to always be | ||||||
| 907 | // treated as alive, so we instead manually save the location of RBP | ||||||
| 908 | // before doing the native call (see NativeInvokerGenerator::generate). | ||||||
| 909 | bool exclude_soe = op
| ||||||
| 910 | || (op
| ||||||
| 911 | |||||||
| 912 | // If the call is a MethodHandle invoke, we need to exclude the | ||||||
| 913 | // register which is used to save the SP value over MH invokes from | ||||||
| 914 | // the mask. Otherwise this register could be used for | ||||||
| 915 | // deoptimization information. | ||||||
| 916 | if (op
| ||||||
| 917 | MachCallStaticJavaNode* mcallstaticjava = (MachCallStaticJavaNode*) mcall; | ||||||
| 918 | if (mcallstaticjava->_method_handle_invoke) | ||||||
| 919 | proj->_rout.OR(Matcher::method_handle_invoke_SP_save_mask()); | ||||||
| 920 | } | ||||||
| 921 | |||||||
| 922 | add_call_kills(proj, regs, save_policy, exclude_soe); | ||||||
| 923 | |||||||
| 924 | return node_cnt; | ||||||
| 925 | } | ||||||
| 926 | |||||||
| 927 | |||||||
| 928 | //------------------------------schedule_local--------------------------------- | ||||||
| 929 | // Topological sort within a block. Someday become a real scheduler. | ||||||
| 930 | bool PhaseCFG::schedule_local(Block* block, GrowableArray<int>& ready_cnt, VectorSet& next_call, intptr_t *recalc_pressure_nodes) { | ||||||
| 931 | // Already "sorted" are the block start Node (as the first entry), and | ||||||
| 932 | // the block-ending Node and any trailing control projections. We leave | ||||||
| 933 | // these alone. PhiNodes and ParmNodes are made to follow the block start | ||||||
| 934 | // Node. Everything else gets topo-sorted. | ||||||
| 935 | |||||||
| 936 | #ifndef PRODUCT | ||||||
| 937 | if (trace_opto_pipelining()) { | ||||||
| 938 | tty->print_cr("# --- schedule_local B%d, before: ---", block->_pre_order); | ||||||
| 939 | for (uint i = 0;i < block->number_of_nodes(); i++) { | ||||||
| 940 | tty->print("# "); | ||||||
| 941 | block->get_node(i)->fast_dump(); | ||||||
| 942 | } | ||||||
| 943 | tty->print_cr("#"); | ||||||
| 944 | } | ||||||
| 945 | #endif | ||||||
| 946 | |||||||
| 947 | // RootNode is already sorted | ||||||
| 948 | if (block->number_of_nodes() == 1) { | ||||||
| 949 | return true; | ||||||
| 950 | } | ||||||
| 951 | |||||||
| 952 | bool block_size_threshold_ok = (recalc_pressure_nodes != NULL__null) && (block->number_of_nodes() > 10); | ||||||
| 953 | |||||||
| 954 | // We track the uses of local definitions as input dependences so that | ||||||
| 955 | // we know when a given instruction is avialable to be scheduled. | ||||||
| 956 | uint i; | ||||||
| 957 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 958 | for (i = 1; i < block->number_of_nodes(); i++) { // setup nodes for pressure calc | ||||||
| 959 | Node *n = block->get_node(i); | ||||||
| 960 | n->remove_flag(Node::Flag_is_scheduled); | ||||||
| 961 | if (!n->is_Phi()) { | ||||||
| 962 | recalc_pressure_nodes[n->_idx] = 0x7fff7fff; | ||||||
| 963 | } | ||||||
| 964 | } | ||||||
| 965 | } | ||||||
| 966 | |||||||
| 967 | // Move PhiNodes and ParmNodes from 1 to cnt up to the start | ||||||
| 968 | uint node_cnt = block->end_idx(); | ||||||
| 969 | uint phi_cnt = 1; | ||||||
| 970 | for( i = 1; i<node_cnt; i++ ) { // Scan for Phi | ||||||
| 971 | Node *n = block->get_node(i); | ||||||
| 972 | if( n->is_Phi() || // Found a PhiNode or ParmNode | ||||||
| 973 | (n->is_Proj() && n->in(0) == block->head()) ) { | ||||||
| 974 | // Move guy at 'phi_cnt' to the end; makes a hole at phi_cnt | ||||||
| 975 | block->map_node(block->get_node(phi_cnt), i); | ||||||
| 976 | block->map_node(n, phi_cnt++); // swap Phi/Parm up front | ||||||
| 977 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 978 | // mark n as scheduled | ||||||
| 979 | n->add_flag(Node::Flag_is_scheduled); | ||||||
| 980 | } | ||||||
| 981 | } else { // All others | ||||||
| 982 | // Count block-local inputs to 'n' | ||||||
| 983 | uint cnt = n->len(); // Input count | ||||||
| 984 | uint local = 0; | ||||||
| 985 | for( uint j=0; j<cnt; j++ ) { | ||||||
| 986 | Node *m = n->in(j); | ||||||
| 987 | if( m && get_block_for_node(m) == block && !m->is_top() ) | ||||||
| 988 | local++; // One more block-local input | ||||||
| 989 | } | ||||||
| 990 | ready_cnt.at_put(n->_idx, local); // Count em up | ||||||
| 991 | |||||||
| 992 | #ifdef ASSERT1 | ||||||
| 993 | if (UseG1GC) { | ||||||
| 994 | if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_StoreCM ) { | ||||||
| 995 | // Check the precedence edges | ||||||
| 996 | for (uint prec = n->req(); prec < n->len(); prec++) { | ||||||
| 997 | Node* oop_store = n->in(prec); | ||||||
| 998 | if (oop_store != NULL__null) { | ||||||
| 999 | assert(get_block_for_node(oop_store)->_dom_depth <= block->_dom_depth, "oop_store must dominate card-mark")do { if (!(get_block_for_node(oop_store)->_dom_depth <= block->_dom_depth)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 999, "assert(" "get_block_for_node(oop_store)->_dom_depth <= block->_dom_depth" ") failed", "oop_store must dominate card-mark"); ::breakpoint (); } } while (0); | ||||||
| 1000 | } | ||||||
| 1001 | } | ||||||
| 1002 | } | ||||||
| 1003 | } | ||||||
| 1004 | #endif | ||||||
| 1005 | |||||||
| 1006 | // A few node types require changing a required edge to a precedence edge | ||||||
| 1007 | // before allocation. | ||||||
| 1008 | if( n->is_Mach() && n->req() > TypeFunc::Parms && | ||||||
| 1009 | (n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire || | ||||||
| 1010 | n->as_Mach()->ideal_Opcode() == Op_MemBarVolatile) ) { | ||||||
| 1011 | // MemBarAcquire could be created without Precedent edge. | ||||||
| 1012 | // del_req() replaces the specified edge with the last input edge | ||||||
| 1013 | // and then removes the last edge. If the specified edge > number of | ||||||
| 1014 | // edges the last edge will be moved outside of the input edges array | ||||||
| 1015 | // and the edge will be lost. This is why this code should be | ||||||
| 1016 | // executed only when Precedent (== TypeFunc::Parms) edge is present. | ||||||
| 1017 | Node *x = n->in(TypeFunc::Parms); | ||||||
| 1018 | if (x != NULL__null && get_block_for_node(x) == block && n->find_prec_edge(x) != -1) { | ||||||
| 1019 | // Old edge to node within same block will get removed, but no precedence | ||||||
| 1020 | // edge will get added because it already exists. Update ready count. | ||||||
| 1021 | int cnt = ready_cnt.at(n->_idx); | ||||||
| 1022 | assert(cnt > 1, "MemBar node %d must not get ready here", n->_idx)do { if (!(cnt > 1)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1022, "assert(" "cnt > 1" ") failed", "MemBar node %d must not get ready here" , n->_idx); ::breakpoint(); } } while (0); | ||||||
| 1023 | ready_cnt.at_put(n->_idx, cnt-1); | ||||||
| 1024 | } | ||||||
| 1025 | n->del_req(TypeFunc::Parms); | ||||||
| 1026 | n->add_prec(x); | ||||||
| 1027 | } | ||||||
| 1028 | } | ||||||
| 1029 | } | ||||||
| 1030 | for(uint i2=i; i2< block->number_of_nodes(); i2++ ) // Trailing guys get zapped count | ||||||
| 1031 | ready_cnt.at_put(block->get_node(i2)->_idx, 0); | ||||||
| 1032 | |||||||
| 1033 | // All the prescheduled guys do not hold back internal nodes | ||||||
| 1034 | uint i3; | ||||||
| 1035 | for (i3 = 0; i3 < phi_cnt; i3++) { // For all pre-scheduled | ||||||
| 1036 | Node *n = block->get_node(i3); // Get pre-scheduled | ||||||
| 1037 | for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { | ||||||
| 1038 | Node* m = n->fast_out(j); | ||||||
| 1039 | if (get_block_for_node(m) == block) { // Local-block user | ||||||
| 1040 | int m_cnt = ready_cnt.at(m->_idx)-1; | ||||||
| 1041 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 1042 | // mark m as scheduled | ||||||
| 1043 | if (m_cnt < 0) { | ||||||
| 1044 | m->add_flag(Node::Flag_is_scheduled); | ||||||
| 1045 | } | ||||||
| 1046 | } | ||||||
| 1047 | ready_cnt.at_put(m->_idx, m_cnt); // Fix ready count | ||||||
| 1048 | } | ||||||
| 1049 | } | ||||||
| 1050 | } | ||||||
| 1051 | |||||||
| 1052 | Node_List delay; | ||||||
| 1053 | // Make a worklist | ||||||
| 1054 | Node_List worklist; | ||||||
| 1055 | for(uint i4=i3; i4<node_cnt; i4++ ) { // Put ready guys on worklist | ||||||
| 1056 | Node *m = block->get_node(i4); | ||||||
| 1057 | if( !ready_cnt.at(m->_idx) ) { // Zero ready count? | ||||||
| 1058 | if (m->is_iteratively_computed()) { | ||||||
| 1059 | // Push induction variable increments last to allow other uses | ||||||
| 1060 | // of the phi to be scheduled first. The select() method breaks | ||||||
| 1061 | // ties in scheduling by worklist order. | ||||||
| 1062 | delay.push(m); | ||||||
| 1063 | } else if (m->is_Mach() && m->as_Mach()->ideal_Opcode() == Op_CreateEx) { | ||||||
| 1064 | // Force the CreateEx to the top of the list so it's processed | ||||||
| 1065 | // first and ends up at the start of the block. | ||||||
| 1066 | worklist.insert(0, m); | ||||||
| 1067 | } else { | ||||||
| 1068 | worklist.push(m); // Then on to worklist! | ||||||
| 1069 | } | ||||||
| 1070 | } | ||||||
| 1071 | } | ||||||
| 1072 | while (delay.size()) { | ||||||
| 1073 | Node* d = delay.pop(); | ||||||
| 1074 | worklist.push(d); | ||||||
| 1075 | } | ||||||
| 1076 | |||||||
| 1077 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 1078 | // To stage register pressure calculations we need to examine the live set variables | ||||||
| 1079 | // breaking them up by register class to compartmentalize the calculations. | ||||||
| 1080 | _regalloc->_sched_int_pressure.init(Matcher::int_pressure_limit()); | ||||||
| 1081 | _regalloc->_sched_float_pressure.init(Matcher::float_pressure_limit()); | ||||||
| 1082 | _regalloc->_scratch_int_pressure.init(Matcher::int_pressure_limit()); | ||||||
| 1083 | _regalloc->_scratch_float_pressure.init(Matcher::float_pressure_limit()); | ||||||
| 1084 | |||||||
| 1085 | _regalloc->compute_entry_block_pressure(block); | ||||||
| 1086 | } | ||||||
| 1087 | |||||||
| 1088 | // Warm up the 'next_call' heuristic bits | ||||||
| 1089 | needed_for_next_call(block, block->head(), next_call); | ||||||
| 1090 | |||||||
| 1091 | #ifndef PRODUCT | ||||||
| 1092 | if (trace_opto_pipelining()) { | ||||||
| 1093 | for (uint j=0; j< block->number_of_nodes(); j++) { | ||||||
| 1094 | Node *n = block->get_node(j); | ||||||
| 1095 | int idx = n->_idx; | ||||||
| 1096 | tty->print("# ready cnt:%3d ", ready_cnt.at(idx)); | ||||||
| 1097 | tty->print("latency:%3d ", get_latency_for_node(n)); | ||||||
| 1098 | tty->print("%4d: %s\n", idx, n->Name()); | ||||||
| 1099 | } | ||||||
| 1100 | } | ||||||
| 1101 | #endif | ||||||
| 1102 | |||||||
| 1103 | uint max_idx = (uint)ready_cnt.length(); | ||||||
| 1104 | // Pull from worklist and schedule | ||||||
| 1105 | while( worklist.size() ) { // Worklist is not ready | ||||||
| 1106 | |||||||
| 1107 | #ifndef PRODUCT | ||||||
| 1108 | if (trace_opto_pipelining()) { | ||||||
| 1109 | tty->print("# ready list:"); | ||||||
| 1110 | for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist | ||||||
| 1111 | Node *n = worklist[i]; // Get Node on worklist | ||||||
| 1112 | tty->print(" %d", n->_idx); | ||||||
| 1113 | } | ||||||
| 1114 | tty->cr(); | ||||||
| 1115 | } | ||||||
| 1116 | #endif | ||||||
| 1117 | |||||||
| 1118 | // Select and pop a ready guy from worklist | ||||||
| 1119 | Node* n = select(block, worklist, ready_cnt, next_call, phi_cnt, recalc_pressure_nodes); | ||||||
| 1120 | block->map_node(n, phi_cnt++); // Schedule him next | ||||||
| 1121 | |||||||
| 1122 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 1123 | n->add_flag(Node::Flag_is_scheduled); | ||||||
| 1124 | |||||||
| 1125 | // Now adjust the resister pressure with the node we selected | ||||||
| 1126 | if (!n->is_Phi()) { | ||||||
| 1127 | adjust_register_pressure(n, block, recalc_pressure_nodes, true); | ||||||
| 1128 | } | ||||||
| 1129 | } | ||||||
| 1130 | |||||||
| 1131 | #ifndef PRODUCT | ||||||
| 1132 | if (trace_opto_pipelining()) { | ||||||
| 1133 | tty->print("# select %d: %s", n->_idx, n->Name()); | ||||||
| 1134 | tty->print(", latency:%d", get_latency_for_node(n)); | ||||||
| 1135 | n->dump(); | ||||||
| 1136 | if (Verbose) { | ||||||
| 1137 | tty->print("# ready list:"); | ||||||
| 1138 | for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist | ||||||
| 1139 | Node *n = worklist[i]; // Get Node on worklist | ||||||
| 1140 | tty->print(" %d", n->_idx); | ||||||
| 1141 | } | ||||||
| 1142 | tty->cr(); | ||||||
| 1143 | } | ||||||
| 1144 | } | ||||||
| 1145 | |||||||
| 1146 | #endif | ||||||
| 1147 | if( n->is_MachCall() ) { | ||||||
| 1148 | MachCallNode *mcall = n->as_MachCall(); | ||||||
| 1149 | phi_cnt = sched_call(block, phi_cnt, worklist, ready_cnt, mcall, next_call); | ||||||
| 1150 | continue; | ||||||
| 1151 | } | ||||||
| 1152 | |||||||
| 1153 | if (n->is_Mach() && n->as_Mach()->has_call()) { | ||||||
| 1154 | RegMask regs; | ||||||
| 1155 | regs.Insert(_matcher.c_frame_pointer()); | ||||||
| 1156 | regs.OR(n->out_RegMask()); | ||||||
| 1157 | |||||||
| 1158 | MachProjNode *proj = new MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj ); | ||||||
| 1159 | map_node_to_block(proj, block); | ||||||
| 1160 | block->insert_node(proj, phi_cnt++); | ||||||
| 1161 | |||||||
| 1162 | add_call_kills(proj, regs, _matcher._c_reg_save_policy, false); | ||||||
| 1163 | } | ||||||
| 1164 | |||||||
| 1165 | // Children are now all ready | ||||||
| 1166 | for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) { | ||||||
| 1167 | Node* m = n->fast_out(i5); // Get user | ||||||
| 1168 | if (get_block_for_node(m) != block) { | ||||||
| 1169 | continue; | ||||||
| 1170 | } | ||||||
| 1171 | if( m->is_Phi() ) continue; | ||||||
| 1172 | if (m->_idx >= max_idx) { // new node, skip it | ||||||
| 1173 | assert(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call(), "unexpected node types")do { if (!(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call())) { (*g_assert_poison) = 'X'; ; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1173, "assert(" "m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call()" ") failed", "unexpected node types"); ::breakpoint(); } } while (0); | ||||||
| 1174 | continue; | ||||||
| 1175 | } | ||||||
| 1176 | int m_cnt = ready_cnt.at(m->_idx) - 1; | ||||||
| 1177 | ready_cnt.at_put(m->_idx, m_cnt); | ||||||
| 1178 | if( m_cnt == 0 ) | ||||||
| 1179 | worklist.push(m); | ||||||
| 1180 | } | ||||||
| 1181 | } | ||||||
| 1182 | |||||||
| 1183 | if( phi_cnt != block->end_idx() ) { | ||||||
| 1184 | // did not schedule all. Retry, Bailout, or Die | ||||||
| 1185 | if (C->subsume_loads() == true && !C->failing()) { | ||||||
| 1186 | // Retry with subsume_loads == false | ||||||
| 1187 | // If this is the first failure, the sentinel string will "stick" | ||||||
| 1188 | // to the Compile object, and the C2Compiler will see it and retry. | ||||||
| 1189 | C->record_failure(C2Compiler::retry_no_subsuming_loads()); | ||||||
| 1190 | } else { | ||||||
| 1191 | assert(false, "graph should be schedulable")do { if (!(false)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1191, "assert(" "false" ") failed", "graph should be schedulable" ); ::breakpoint(); } } while (0); | ||||||
| 1192 | } | ||||||
| 1193 | // assert( phi_cnt == end_idx(), "did not schedule all" ); | ||||||
| 1194 | return false; | ||||||
| 1195 | } | ||||||
| 1196 | |||||||
| 1197 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 1198 | _regalloc->compute_exit_block_pressure(block); | ||||||
| 1199 | block->_reg_pressure = _regalloc->_sched_int_pressure.final_pressure(); | ||||||
| 1200 | block->_freg_pressure = _regalloc->_sched_float_pressure.final_pressure(); | ||||||
| 1201 | } | ||||||
| 1202 | |||||||
| 1203 | #ifndef PRODUCT | ||||||
| 1204 | if (trace_opto_pipelining()) { | ||||||
| 1205 | tty->print_cr("#"); | ||||||
| 1206 | tty->print_cr("# after schedule_local"); | ||||||
| 1207 | for (uint i = 0;i < block->number_of_nodes();i++) { | ||||||
| 1208 | tty->print("# "); | ||||||
| 1209 | block->get_node(i)->fast_dump(); | ||||||
| 1210 | } | ||||||
| 1211 | tty->print_cr("# "); | ||||||
| 1212 | |||||||
| 1213 | if (OptoRegScheduling && block_size_threshold_ok) { | ||||||
| 1214 | tty->print_cr("# pressure info : %d", block->_pre_order); | ||||||
| 1215 | _regalloc->print_pressure_info(_regalloc->_sched_int_pressure, "int register info"); | ||||||
| 1216 | _regalloc->print_pressure_info(_regalloc->_sched_float_pressure, "float register info"); | ||||||
| 1217 | } | ||||||
| 1218 | tty->cr(); | ||||||
| 1219 | } | ||||||
| 1220 | #endif | ||||||
| 1221 | |||||||
| 1222 | return true; | ||||||
| 1223 | } | ||||||
| 1224 | |||||||
| 1225 | //--------------------------catch_cleanup_fix_all_inputs----------------------- | ||||||
| 1226 | static void catch_cleanup_fix_all_inputs(Node *use, Node *old_def, Node *new_def) { | ||||||
| 1227 | for (uint l = 0; l < use->len(); l++) { | ||||||
| 1228 | if (use->in(l) == old_def) { | ||||||
| 1229 | if (l < use->req()) { | ||||||
| 1230 | use->set_req(l, new_def); | ||||||
| 1231 | } else { | ||||||
| 1232 | use->rm_prec(l); | ||||||
| 1233 | use->add_prec(new_def); | ||||||
| 1234 | l--; | ||||||
| 1235 | } | ||||||
| 1236 | } | ||||||
| 1237 | } | ||||||
| 1238 | } | ||||||
| 1239 | |||||||
| 1240 | //------------------------------catch_cleanup_find_cloned_def------------------ | ||||||
| 1241 | Node* PhaseCFG::catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) { | ||||||
| 1242 | assert( use_blk != def_blk, "Inter-block cleanup only")do { if (!(use_blk != def_blk)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1242, "assert(" "use_blk != def_blk" ") failed", "Inter-block cleanup only" ); ::breakpoint(); } } while (0); | ||||||
| 1243 | |||||||
| 1244 | // The use is some block below the Catch. Find and return the clone of the def | ||||||
| 1245 | // that dominates the use. If there is no clone in a dominating block, then | ||||||
| 1246 | // create a phi for the def in a dominating block. | ||||||
| 1247 | |||||||
| 1248 | // Find which successor block dominates this use. The successor | ||||||
| 1249 | // blocks must all be single-entry (from the Catch only; I will have | ||||||
| 1250 | // split blocks to make this so), hence they all dominate. | ||||||
| 1251 | while( use_blk->_dom_depth > def_blk->_dom_depth+1 ) | ||||||
| 1252 | use_blk = use_blk->_idom; | ||||||
| 1253 | |||||||
| 1254 | // Find the successor | ||||||
| 1255 | Node *fixup = NULL__null; | ||||||
| 1256 | |||||||
| 1257 | uint j; | ||||||
| 1258 | for( j = 0; j < def_blk->_num_succs; j++ ) | ||||||
| 1259 | if( use_blk == def_blk->_succs[j] ) | ||||||
| 1260 | break; | ||||||
| 1261 | |||||||
| 1262 | if( j == def_blk->_num_succs ) { | ||||||
| 1263 | // Block at same level in dom-tree is not a successor. It needs a | ||||||
| 1264 | // PhiNode, the PhiNode uses from the def and IT's uses need fixup. | ||||||
| 1265 | Node_Array inputs = new Node_List(); | ||||||
| 1266 | for(uint k = 1; k < use_blk->num_preds(); k++) { | ||||||
| 1267 | Block* block = get_block_for_node(use_blk->pred(k)); | ||||||
| 1268 | inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, n_clone_idx)); | ||||||
| 1269 | } | ||||||
| 1270 | |||||||
| 1271 | // Check to see if the use_blk already has an identical phi inserted. | ||||||
| 1272 | // If it exists, it will be at the first position since all uses of a | ||||||
| 1273 | // def are processed together. | ||||||
| 1274 | Node *phi = use_blk->get_node(1); | ||||||
| 1275 | if( phi->is_Phi() ) { | ||||||
| 1276 | fixup = phi; | ||||||
| 1277 | for (uint k = 1; k < use_blk->num_preds(); k++) { | ||||||
| 1278 | if (phi->in(k) != inputs[k]) { | ||||||
| 1279 | // Not a match | ||||||
| 1280 | fixup = NULL__null; | ||||||
| 1281 | break; | ||||||
| 1282 | } | ||||||
| 1283 | } | ||||||
| 1284 | } | ||||||
| 1285 | |||||||
| 1286 | // If an existing PhiNode was not found, make a new one. | ||||||
| 1287 | if (fixup == NULL__null) { | ||||||
| 1288 | Node *new_phi = PhiNode::make(use_blk->head(), def); | ||||||
| 1289 | use_blk->insert_node(new_phi, 1); | ||||||
| 1290 | map_node_to_block(new_phi, use_blk); | ||||||
| 1291 | for (uint k = 1; k < use_blk->num_preds(); k++) { | ||||||
| 1292 | new_phi->set_req(k, inputs[k]); | ||||||
| 1293 | } | ||||||
| 1294 | fixup = new_phi; | ||||||
| 1295 | } | ||||||
| 1296 | |||||||
| 1297 | } else { | ||||||
| 1298 | // Found the use just below the Catch. Make it use the clone. | ||||||
| 1299 | fixup = use_blk->get_node(n_clone_idx); | ||||||
| 1300 | } | ||||||
| 1301 | |||||||
| 1302 | return fixup; | ||||||
| 1303 | } | ||||||
| 1304 | |||||||
| 1305 | //--------------------------catch_cleanup_intra_block-------------------------- | ||||||
| 1306 | // Fix all input edges in use that reference "def". The use is in the same | ||||||
| 1307 | // block as the def and both have been cloned in each successor block. | ||||||
| 1308 | static void catch_cleanup_intra_block(Node *use, Node *def, Block *blk, int beg, int n_clone_idx) { | ||||||
| 1309 | |||||||
| 1310 | // Both the use and def have been cloned. For each successor block, | ||||||
| 1311 | // get the clone of the use, and make its input the clone of the def | ||||||
| 1312 | // found in that block. | ||||||
| 1313 | |||||||
| 1314 | uint use_idx = blk->find_node(use); | ||||||
| 1315 | uint offset_idx = use_idx - beg; | ||||||
| 1316 | for( uint k = 0; k < blk->_num_succs; k++ ) { | ||||||
| 1317 | // Get clone in each successor block | ||||||
| 1318 | Block *sb = blk->_succs[k]; | ||||||
| 1319 | Node *clone = sb->get_node(offset_idx+1); | ||||||
| 1320 | assert( clone->Opcode() == use->Opcode(), "" )do { if (!(clone->Opcode() == use->Opcode())) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1320, "assert(" "clone->Opcode() == use->Opcode()" ") failed" , ""); ::breakpoint(); } } while (0); | ||||||
| 1321 | |||||||
| 1322 | // Make use-clone reference the def-clone | ||||||
| 1323 | catch_cleanup_fix_all_inputs(clone, def, sb->get_node(n_clone_idx)); | ||||||
| 1324 | } | ||||||
| 1325 | } | ||||||
| 1326 | |||||||
| 1327 | //------------------------------catch_cleanup_inter_block--------------------- | ||||||
| 1328 | // Fix all input edges in use that reference "def". The use is in a different | ||||||
| 1329 | // block than the def. | ||||||
| 1330 | void PhaseCFG::catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) { | ||||||
| 1331 | if( !use_blk ) return; // Can happen if the use is a precedence edge | ||||||
| 1332 | |||||||
| 1333 | Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, n_clone_idx); | ||||||
| 1334 | catch_cleanup_fix_all_inputs(use, def, new_def); | ||||||
| 1335 | } | ||||||
| 1336 | |||||||
| 1337 | //------------------------------call_catch_cleanup----------------------------- | ||||||
| 1338 | // If we inserted any instructions between a Call and his CatchNode, | ||||||
| 1339 | // clone the instructions on all paths below the Catch. | ||||||
| 1340 | void PhaseCFG::call_catch_cleanup(Block* block) { | ||||||
| 1341 | |||||||
| 1342 | // End of region to clone | ||||||
| 1343 | uint end = block->end_idx(); | ||||||
| 1344 | if( !block->get_node(end)->is_Catch() ) return; | ||||||
| 1345 | // Start of region to clone | ||||||
| 1346 | uint beg = end; | ||||||
| 1347 | while(!block->get_node(beg-1)->is_MachProj() || | ||||||
| 1348 | !block->get_node(beg-1)->in(0)->is_MachCall() ) { | ||||||
| 1349 | beg--; | ||||||
| 1350 | assert(beg > 0,"Catch cleanup walking beyond block boundary")do { if (!(beg > 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1350, "assert(" "beg > 0" ") failed", "Catch cleanup walking beyond block boundary" ); ::breakpoint(); } } while (0); | ||||||
| 1351 | } | ||||||
| 1352 | // Range of inserted instructions is [beg, end) | ||||||
| 1353 | if( beg == end ) return; | ||||||
| 1354 | |||||||
| 1355 | // Clone along all Catch output paths. Clone area between the 'beg' and | ||||||
| 1356 | // 'end' indices. | ||||||
| 1357 | for( uint i = 0; i < block->_num_succs; i++ ) { | ||||||
| 1358 | Block *sb = block->_succs[i]; | ||||||
| 1359 | // Clone the entire area; ignoring the edge fixup for now. | ||||||
| 1360 | for( uint j = end; j > beg; j-- ) { | ||||||
| 1361 | Node *clone = block->get_node(j-1)->clone(); | ||||||
| 1362 | sb->insert_node(clone, 1); | ||||||
| 1363 | map_node_to_block(clone, sb); | ||||||
| 1364 | if (clone->needs_anti_dependence_check()) { | ||||||
| 1365 | insert_anti_dependences(sb, clone); | ||||||
| 1366 | } | ||||||
| 1367 | } | ||||||
| 1368 | } | ||||||
| 1369 | |||||||
| 1370 | |||||||
| 1371 | // Fixup edges. Check the def-use info per cloned Node | ||||||
| 1372 | for(uint i2 = beg; i2 < end; i2++ ) { | ||||||
| 1373 | uint n_clone_idx = i2-beg+1; // Index of clone of n in each successor block | ||||||
| 1374 | Node *n = block->get_node(i2); // Node that got cloned | ||||||
| 1375 | // Need DU safe iterator because of edge manipulation in calls. | ||||||
| 1376 | Unique_Node_List* out = new Unique_Node_List(); | ||||||
| 1377 | for (DUIterator_Fast j1max, j1 = n->fast_outs(j1max); j1 < j1max; j1++) { | ||||||
| 1378 | out->push(n->fast_out(j1)); | ||||||
| 1379 | } | ||||||
| 1380 | uint max = out->size(); | ||||||
| 1381 | for (uint j = 0; j < max; j++) {// For all users | ||||||
| 1382 | Node *use = out->pop(); | ||||||
| 1383 | Block *buse = get_block_for_node(use); | ||||||
| 1384 | if( use->is_Phi() ) { | ||||||
| 1385 | for( uint k = 1; k < use->req(); k++ ) | ||||||
| 1386 | if( use->in(k) == n ) { | ||||||
| 1387 | Block* b = get_block_for_node(buse->pred(k)); | ||||||
| 1388 | Node *fixup = catch_cleanup_find_cloned_def(b, n, block, n_clone_idx); | ||||||
| 1389 | use->set_req(k, fixup); | ||||||
| 1390 | } | ||||||
| 1391 | } else { | ||||||
| 1392 | if (block == buse) { | ||||||
| 1393 | catch_cleanup_intra_block(use, n, block, beg, n_clone_idx); | ||||||
| 1394 | } else { | ||||||
| 1395 | catch_cleanup_inter_block(use, buse, n, block, n_clone_idx); | ||||||
| 1396 | } | ||||||
| 1397 | } | ||||||
| 1398 | } // End for all users | ||||||
| 1399 | |||||||
| 1400 | } // End of for all Nodes in cloned area | ||||||
| 1401 | |||||||
| 1402 | // Remove the now-dead cloned ops | ||||||
| 1403 | for(uint i3 = beg; i3 < end; i3++ ) { | ||||||
| 1404 | block->get_node(beg)->disconnect_inputs(C); | ||||||
| 1405 | block->remove_node(beg); | ||||||
| 1406 | } | ||||||
| 1407 | |||||||
| 1408 | // If the successor blocks have a CreateEx node, move it back to the top | ||||||
| 1409 | for (uint i4 = 0; i4 < block->_num_succs; i4++) { | ||||||
| 1410 | Block *sb = block->_succs[i4]; | ||||||
| 1411 | uint new_cnt = end - beg; | ||||||
| 1412 | // Remove any newly created, but dead, nodes by traversing their schedule | ||||||
| 1413 | // backwards. Here, a dead node is a node whose only outputs (if any) are | ||||||
| 1414 | // unused projections. | ||||||
| 1415 | for (uint j = new_cnt; j > 0; j--) { | ||||||
| 1416 | Node *n = sb->get_node(j); | ||||||
| 1417 | // Individual projections are examined together with all siblings when | ||||||
| 1418 | // their parent is visited. | ||||||
| 1419 | if (n->is_Proj()) { | ||||||
| 1420 | continue; | ||||||
| 1421 | } | ||||||
| 1422 | bool dead = true; | ||||||
| 1423 | for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { | ||||||
| 1424 | Node* out = n->fast_out(i); | ||||||
| 1425 | // n is live if it has a non-projection output or a used projection. | ||||||
| 1426 | if (!out->is_Proj() || out->outcnt() > 0) { | ||||||
| 1427 | dead = false; | ||||||
| 1428 | break; | ||||||
| 1429 | } | ||||||
| 1430 | } | ||||||
| 1431 | if (dead) { | ||||||
| 1432 | // n's only outputs (if any) are unused projections scheduled next to n | ||||||
| 1433 | // (see PhaseCFG::select()). Remove these projections backwards. | ||||||
| 1434 | for (uint k = j + n->outcnt(); k > j; k--) { | ||||||
| 1435 | Node* proj = sb->get_node(k); | ||||||
| 1436 | assert(proj->is_Proj() && proj->in(0) == n,do { if (!(proj->is_Proj() && proj->in(0) == n) ) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1437, "assert(" "proj->is_Proj() && proj->in(0) == n" ") failed", "projection should correspond to dead node"); :: breakpoint(); } } while (0) | ||||||
| 1437 | "projection should correspond to dead node")do { if (!(proj->is_Proj() && proj->in(0) == n) ) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/lcm.cpp" , 1437, "assert(" "proj->is_Proj() && proj->in(0) == n" ") failed", "projection should correspond to dead node"); :: breakpoint(); } } while (0); | ||||||
| 1438 | proj->disconnect_inputs(C); | ||||||
| 1439 | sb->remove_node(k); | ||||||
| 1440 | new_cnt--; | ||||||
| 1441 | } | ||||||
| 1442 | // Now remove the node itself. | ||||||
| 1443 | n->disconnect_inputs(C); | ||||||
| 1444 | sb->remove_node(j); | ||||||
| 1445 | new_cnt--; | ||||||
| 1446 | } | ||||||
| 1447 | } | ||||||
| 1448 | // If any newly created nodes remain, move the CreateEx node to the top | ||||||
| 1449 | if (new_cnt > 0) { | ||||||
| 1450 | Node *cex = sb->get_node(1+new_cnt); | ||||||
| 1451 | if( cex->is_Mach() && cex->as_Mach()->ideal_Opcode() == Op_CreateEx ) { | ||||||
| 1452 | sb->remove_node(1+new_cnt); | ||||||
| 1453 | sb->insert_node(cex, 1); | ||||||
| 1454 | } | ||||||
| 1455 | } | ||||||
| 1456 | } | ||||||
| 1457 | } |
| 1 | /* |
| 2 | * Copyright (c) 1997, 2021, 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 SHARE_OPTO_NODE_HPP |
| 26 | #define SHARE_OPTO_NODE_HPP |
| 27 | |
| 28 | #include "libadt/vectset.hpp" |
| 29 | #include "opto/compile.hpp" |
| 30 | #include "opto/type.hpp" |
| 31 | #include "utilities/copy.hpp" |
| 32 | |
| 33 | // Portions of code courtesy of Clifford Click |
| 34 | |
| 35 | // Optimization - Graph Style |
| 36 | |
| 37 | |
| 38 | class AbstractLockNode; |
| 39 | class AddNode; |
| 40 | class AddPNode; |
| 41 | class AliasInfo; |
| 42 | class AllocateArrayNode; |
| 43 | class AllocateNode; |
| 44 | class ArrayCopyNode; |
| 45 | class BaseCountedLoopNode; |
| 46 | class BaseCountedLoopEndNode; |
| 47 | class BlackholeNode; |
| 48 | class Block; |
| 49 | class BoolNode; |
| 50 | class BoxLockNode; |
| 51 | class CMoveNode; |
| 52 | class CallDynamicJavaNode; |
| 53 | class CallJavaNode; |
| 54 | class CallLeafNode; |
| 55 | class CallLeafNoFPNode; |
| 56 | class CallNode; |
| 57 | class CallRuntimeNode; |
| 58 | class CallNativeNode; |
| 59 | class CallStaticJavaNode; |
| 60 | class CastFFNode; |
| 61 | class CastDDNode; |
| 62 | class CastVVNode; |
| 63 | class CastIINode; |
| 64 | class CastLLNode; |
| 65 | class CatchNode; |
| 66 | class CatchProjNode; |
| 67 | class CheckCastPPNode; |
| 68 | class ClearArrayNode; |
| 69 | class CmpNode; |
| 70 | class CodeBuffer; |
| 71 | class ConstraintCastNode; |
| 72 | class ConNode; |
| 73 | class CompareAndSwapNode; |
| 74 | class CompareAndExchangeNode; |
| 75 | class CountedLoopNode; |
| 76 | class CountedLoopEndNode; |
| 77 | class DecodeNarrowPtrNode; |
| 78 | class DecodeNNode; |
| 79 | class DecodeNKlassNode; |
| 80 | class EncodeNarrowPtrNode; |
| 81 | class EncodePNode; |
| 82 | class EncodePKlassNode; |
| 83 | class FastLockNode; |
| 84 | class FastUnlockNode; |
| 85 | class HaltNode; |
| 86 | class IfNode; |
| 87 | class IfProjNode; |
| 88 | class IfFalseNode; |
| 89 | class IfTrueNode; |
| 90 | class InitializeNode; |
| 91 | class JVMState; |
| 92 | class JumpNode; |
| 93 | class JumpProjNode; |
| 94 | class LoadNode; |
| 95 | class LoadStoreNode; |
| 96 | class LoadStoreConditionalNode; |
| 97 | class LockNode; |
| 98 | class LongCountedLoopNode; |
| 99 | class LongCountedLoopEndNode; |
| 100 | class LoopNode; |
| 101 | class LShiftNode; |
| 102 | class MachBranchNode; |
| 103 | class MachCallDynamicJavaNode; |
| 104 | class MachCallJavaNode; |
| 105 | class MachCallLeafNode; |
| 106 | class MachCallNode; |
| 107 | class MachCallNativeNode; |
| 108 | class MachCallRuntimeNode; |
| 109 | class MachCallStaticJavaNode; |
| 110 | class MachConstantBaseNode; |
| 111 | class MachConstantNode; |
| 112 | class MachGotoNode; |
| 113 | class MachIfNode; |
| 114 | class MachJumpNode; |
| 115 | class MachNode; |
| 116 | class MachNullCheckNode; |
| 117 | class MachProjNode; |
| 118 | class MachReturnNode; |
| 119 | class MachSafePointNode; |
| 120 | class MachSpillCopyNode; |
| 121 | class MachTempNode; |
| 122 | class MachMergeNode; |
| 123 | class MachMemBarNode; |
| 124 | class Matcher; |
| 125 | class MemBarNode; |
| 126 | class MemBarStoreStoreNode; |
| 127 | class MemNode; |
| 128 | class MergeMemNode; |
| 129 | class MoveNode; |
| 130 | class MulNode; |
| 131 | class MultiNode; |
| 132 | class MultiBranchNode; |
| 133 | class NeverBranchNode; |
| 134 | class Opaque1Node; |
| 135 | class OuterStripMinedLoopNode; |
| 136 | class OuterStripMinedLoopEndNode; |
| 137 | class Node; |
| 138 | class Node_Array; |
| 139 | class Node_List; |
| 140 | class Node_Stack; |
| 141 | class OopMap; |
| 142 | class ParmNode; |
| 143 | class PCTableNode; |
| 144 | class PhaseCCP; |
| 145 | class PhaseGVN; |
| 146 | class PhaseIterGVN; |
| 147 | class PhaseRegAlloc; |
| 148 | class PhaseTransform; |
| 149 | class PhaseValues; |
| 150 | class PhiNode; |
| 151 | class Pipeline; |
| 152 | class ProjNode; |
| 153 | class RangeCheckNode; |
| 154 | class RegMask; |
| 155 | class RegionNode; |
| 156 | class RootNode; |
| 157 | class SafePointNode; |
| 158 | class SafePointScalarObjectNode; |
| 159 | class StartNode; |
| 160 | class State; |
| 161 | class StoreNode; |
| 162 | class SubNode; |
| 163 | class SubTypeCheckNode; |
| 164 | class Type; |
| 165 | class TypeNode; |
| 166 | class UnlockNode; |
| 167 | class VectorNode; |
| 168 | class LoadVectorNode; |
| 169 | class LoadVectorMaskedNode; |
| 170 | class StoreVectorMaskedNode; |
| 171 | class LoadVectorGatherNode; |
| 172 | class StoreVectorNode; |
| 173 | class StoreVectorScatterNode; |
| 174 | class VectorMaskCmpNode; |
| 175 | class VectorUnboxNode; |
| 176 | class VectorSet; |
| 177 | class VectorReinterpretNode; |
| 178 | class ShiftVNode; |
| 179 | |
| 180 | // The type of all node counts and indexes. |
| 181 | // It must hold at least 16 bits, but must also be fast to load and store. |
| 182 | // This type, if less than 32 bits, could limit the number of possible nodes. |
| 183 | // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.) |
| 184 | typedef unsigned int node_idx_t; |
| 185 | |
| 186 | |
| 187 | #ifndef OPTO_DU_ITERATOR_ASSERT1 |
| 188 | #ifdef ASSERT1 |
| 189 | #define OPTO_DU_ITERATOR_ASSERT1 1 |
| 190 | #else |
| 191 | #define OPTO_DU_ITERATOR_ASSERT1 0 |
| 192 | #endif |
| 193 | #endif //OPTO_DU_ITERATOR_ASSERT |
| 194 | |
| 195 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 196 | class DUIterator; |
| 197 | class DUIterator_Fast; |
| 198 | class DUIterator_Last; |
| 199 | #else |
| 200 | typedef uint DUIterator; |
| 201 | typedef Node** DUIterator_Fast; |
| 202 | typedef Node** DUIterator_Last; |
| 203 | #endif |
| 204 | |
| 205 | // Node Sentinel |
| 206 | #define NodeSentinel(Node*)-1 (Node*)-1 |
| 207 | |
| 208 | // Unknown count frequency |
| 209 | #define COUNT_UNKNOWN(-1.0f) (-1.0f) |
| 210 | |
| 211 | //------------------------------Node------------------------------------------- |
| 212 | // Nodes define actions in the program. They create values, which have types. |
| 213 | // They are both vertices in a directed graph and program primitives. Nodes |
| 214 | // are labeled; the label is the "opcode", the primitive function in the lambda |
| 215 | // calculus sense that gives meaning to the Node. Node inputs are ordered (so |
| 216 | // that "a-b" is different from "b-a"). The inputs to a Node are the inputs to |
| 217 | // the Node's function. These inputs also define a Type equation for the Node. |
| 218 | // Solving these Type equations amounts to doing dataflow analysis. |
| 219 | // Control and data are uniformly represented in the graph. Finally, Nodes |
| 220 | // have a unique dense integer index which is used to index into side arrays |
| 221 | // whenever I have phase-specific information. |
| 222 | |
| 223 | class Node { |
| 224 | friend class VMStructs; |
| 225 | |
| 226 | // Lots of restrictions on cloning Nodes |
| 227 | NONCOPYABLE(Node)Node(Node const&) = delete; Node& operator=(Node const &) = delete; |
| 228 | |
| 229 | public: |
| 230 | friend class Compile; |
| 231 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 232 | friend class DUIterator_Common; |
| 233 | friend class DUIterator; |
| 234 | friend class DUIterator_Fast; |
| 235 | friend class DUIterator_Last; |
| 236 | #endif |
| 237 | |
| 238 | // Because Nodes come and go, I define an Arena of Node structures to pull |
| 239 | // from. This should allow fast access to node creation & deletion. This |
| 240 | // field is a local cache of a value defined in some "program fragment" for |
| 241 | // which these Nodes are just a part of. |
| 242 | |
| 243 | inline void* operator new(size_t x) throw() { |
| 244 | Compile* C = Compile::current(); |
| 245 | Node* n = (Node*)C->node_arena()->AmallocWords(x); |
| 246 | return (void*)n; |
| 247 | } |
| 248 | |
| 249 | // Delete is a NOP |
| 250 | void operator delete( void *ptr ) {} |
| 251 | // Fancy destructor; eagerly attempt to reclaim Node numberings and storage |
| 252 | void destruct(PhaseValues* phase); |
| 253 | |
| 254 | // Create a new Node. Required is the number is of inputs required for |
| 255 | // semantic correctness. |
| 256 | Node( uint required ); |
| 257 | |
| 258 | // Create a new Node with given input edges. |
| 259 | // This version requires use of the "edge-count" new. |
| 260 | // E.g. new (C,3) FooNode( C, NULL, left, right ); |
| 261 | Node( Node *n0 ); |
| 262 | Node( Node *n0, Node *n1 ); |
| 263 | Node( Node *n0, Node *n1, Node *n2 ); |
| 264 | Node( Node *n0, Node *n1, Node *n2, Node *n3 ); |
| 265 | Node( Node *n0, Node *n1, Node *n2, Node *n3, Node *n4 ); |
| 266 | Node( Node *n0, Node *n1, Node *n2, Node *n3, Node *n4, Node *n5 ); |
| 267 | Node( Node *n0, Node *n1, Node *n2, Node *n3, |
| 268 | Node *n4, Node *n5, Node *n6 ); |
| 269 | |
| 270 | // Clone an inherited Node given only the base Node type. |
| 271 | Node* clone() const; |
| 272 | |
| 273 | // Clone a Node, immediately supplying one or two new edges. |
| 274 | // The first and second arguments, if non-null, replace in(1) and in(2), |
| 275 | // respectively. |
| 276 | Node* clone_with_data_edge(Node* in1, Node* in2 = NULL__null) const { |
| 277 | Node* nn = clone(); |
| 278 | if (in1 != NULL__null) nn->set_req(1, in1); |
| 279 | if (in2 != NULL__null) nn->set_req(2, in2); |
| 280 | return nn; |
| 281 | } |
| 282 | |
| 283 | private: |
| 284 | // Shared setup for the above constructors. |
| 285 | // Handles all interactions with Compile::current. |
| 286 | // Puts initial values in all Node fields except _idx. |
| 287 | // Returns the initial value for _idx, which cannot |
| 288 | // be initialized by assignment. |
| 289 | inline int Init(int req); |
| 290 | |
| 291 | //----------------- input edge handling |
| 292 | protected: |
| 293 | friend class PhaseCFG; // Access to address of _in array elements |
| 294 | Node **_in; // Array of use-def references to Nodes |
| 295 | Node **_out; // Array of def-use references to Nodes |
| 296 | |
| 297 | // Input edges are split into two categories. Required edges are required |
| 298 | // for semantic correctness; order is important and NULLs are allowed. |
| 299 | // Precedence edges are used to help determine execution order and are |
| 300 | // added, e.g., for scheduling purposes. They are unordered and not |
| 301 | // duplicated; they have no embedded NULLs. Edges from 0 to _cnt-1 |
| 302 | // are required, from _cnt to _max-1 are precedence edges. |
| 303 | node_idx_t _cnt; // Total number of required Node inputs. |
| 304 | |
| 305 | node_idx_t _max; // Actual length of input array. |
| 306 | |
| 307 | // Output edges are an unordered list of def-use edges which exactly |
| 308 | // correspond to required input edges which point from other nodes |
| 309 | // to this one. Thus the count of the output edges is the number of |
| 310 | // users of this node. |
| 311 | node_idx_t _outcnt; // Total number of Node outputs. |
| 312 | |
| 313 | node_idx_t _outmax; // Actual length of output array. |
| 314 | |
| 315 | // Grow the actual input array to the next larger power-of-2 bigger than len. |
| 316 | void grow( uint len ); |
| 317 | // Grow the output array to the next larger power-of-2 bigger than len. |
| 318 | void out_grow( uint len ); |
| 319 | |
| 320 | public: |
| 321 | // Each Node is assigned a unique small/dense number. This number is used |
| 322 | // to index into auxiliary arrays of data and bit vectors. |
| 323 | // The field _idx is declared constant to defend against inadvertent assignments, |
| 324 | // since it is used by clients as a naked field. However, the field's value can be |
| 325 | // changed using the set_idx() method. |
| 326 | // |
| 327 | // The PhaseRenumberLive phase renumbers nodes based on liveness information. |
| 328 | // Therefore, it updates the value of the _idx field. The parse-time _idx is |
| 329 | // preserved in _parse_idx. |
| 330 | const node_idx_t _idx; |
| 331 | DEBUG_ONLY(const node_idx_t _parse_idx;)const node_idx_t _parse_idx; |
| 332 | // IGV node identifier. Two nodes, possibly in different compilation phases, |
| 333 | // have the same IGV identifier if (and only if) they are the very same node |
| 334 | // (same memory address) or one is "derived" from the other (by e.g. |
| 335 | // renumbering or matching). This identifier makes it possible to follow the |
| 336 | // entire lifetime of a node in IGV even if its C2 identifier (_idx) changes. |
| 337 | NOT_PRODUCT(node_idx_t _igv_idx;)node_idx_t _igv_idx; |
| 338 | |
| 339 | // Get the (read-only) number of input edges |
| 340 | uint req() const { return _cnt; } |
| 341 | uint len() const { return _max; } |
| 342 | // Get the (read-only) number of output edges |
| 343 | uint outcnt() const { return _outcnt; } |
| 344 | |
| 345 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 346 | // Iterate over the out-edges of this node. Deletions are illegal. |
| 347 | inline DUIterator outs() const; |
| 348 | // Use this when the out array might have changed to suppress asserts. |
| 349 | inline DUIterator& refresh_out_pos(DUIterator& i) const; |
| 350 | // Does the node have an out at this position? (Used for iteration.) |
| 351 | inline bool has_out(DUIterator& i) const; |
| 352 | inline Node* out(DUIterator& i) const; |
| 353 | // Iterate over the out-edges of this node. All changes are illegal. |
| 354 | inline DUIterator_Fast fast_outs(DUIterator_Fast& max) const; |
| 355 | inline Node* fast_out(DUIterator_Fast& i) const; |
| 356 | // Iterate over the out-edges of this node, deleting one at a time. |
| 357 | inline DUIterator_Last last_outs(DUIterator_Last& min) const; |
| 358 | inline Node* last_out(DUIterator_Last& i) const; |
| 359 | // The inline bodies of all these methods are after the iterator definitions. |
| 360 | #else |
| 361 | // Iterate over the out-edges of this node. Deletions are illegal. |
| 362 | // This iteration uses integral indexes, to decouple from array reallocations. |
| 363 | DUIterator outs() const { return 0; } |
| 364 | // Use this when the out array might have changed to suppress asserts. |
| 365 | DUIterator refresh_out_pos(DUIterator i) const { return i; } |
| 366 | |
| 367 | // Reference to the i'th output Node. Error if out of bounds. |
| 368 | Node* out(DUIterator i) const { assert(i < _outcnt, "oob")do { if (!(i < _outcnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 368, "assert(" "i < _outcnt" ") failed", "oob"); ::breakpoint (); } } while (0); return _out[i]; } |
| 369 | // Does the node have an out at this position? (Used for iteration.) |
| 370 | bool has_out(DUIterator i) const { return i < _outcnt; } |
| 371 | |
| 372 | // Iterate over the out-edges of this node. All changes are illegal. |
| 373 | // This iteration uses a pointer internal to the out array. |
| 374 | DUIterator_Fast fast_outs(DUIterator_Fast& max) const { |
| 375 | Node** out = _out; |
| 376 | // Assign a limit pointer to the reference argument: |
| 377 | max = out + (ptrdiff_t)_outcnt; |
| 378 | // Return the base pointer: |
| 379 | return out; |
| 380 | } |
| 381 | Node* fast_out(DUIterator_Fast i) const { return *i; } |
| 382 | // Iterate over the out-edges of this node, deleting one at a time. |
| 383 | // This iteration uses a pointer internal to the out array. |
| 384 | DUIterator_Last last_outs(DUIterator_Last& min) const { |
| 385 | Node** out = _out; |
| 386 | // Assign a limit pointer to the reference argument: |
| 387 | min = out; |
| 388 | // Return the pointer to the start of the iteration: |
| 389 | return out + (ptrdiff_t)_outcnt - 1; |
| 390 | } |
| 391 | Node* last_out(DUIterator_Last i) const { return *i; } |
| 392 | #endif |
| 393 | |
| 394 | // Reference to the i'th input Node. Error if out of bounds. |
| 395 | Node* in(uint i) const { assert(i < _max, "oob: i=%d, _max=%d", i, _max)do { if (!(i < _max)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 395, "assert(" "i < _max" ") failed", "oob: i=%d, _max=%d" , i, _max); ::breakpoint(); } } while (0); return _in[i]; } |
| 396 | // Reference to the i'th input Node. NULL if out of bounds. |
| 397 | Node* lookup(uint i) const { return ((i < _max) ? _in[i] : NULL__null); } |
| 398 | // Reference to the i'th output Node. Error if out of bounds. |
| 399 | // Use this accessor sparingly. We are going trying to use iterators instead. |
| 400 | Node* raw_out(uint i) const { assert(i < _outcnt,"oob")do { if (!(i < _outcnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 400, "assert(" "i < _outcnt" ") failed", "oob"); ::breakpoint (); } } while (0); return _out[i]; } |
| 401 | // Return the unique out edge. |
| 402 | Node* unique_out() const { assert(_outcnt==1,"not unique")do { if (!(_outcnt==1)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 402, "assert(" "_outcnt==1" ") failed", "not unique"); ::breakpoint (); } } while (0); return _out[0]; } |
| 403 | // Delete out edge at position 'i' by moving last out edge to position 'i' |
| 404 | void raw_del_out(uint i) { |
| 405 | assert(i < _outcnt,"oob")do { if (!(i < _outcnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 405, "assert(" "i < _outcnt" ") failed", "oob"); ::breakpoint (); } } while (0); |
| 406 | assert(_outcnt > 0,"oob")do { if (!(_outcnt > 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 406, "assert(" "_outcnt > 0" ") failed", "oob"); ::breakpoint (); } } while (0); |
| 407 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 408 | // Record that a change happened here. |
| 409 | debug_only(_last_del = _out[i]; ++_del_tick)_last_del = _out[i]; ++_del_tick; |
| 410 | #endif |
| 411 | _out[i] = _out[--_outcnt]; |
| 412 | // Smash the old edge so it can't be used accidentally. |
| 413 | debug_only(_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef)_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef; |
| 414 | } |
| 415 | |
| 416 | #ifdef ASSERT1 |
| 417 | bool is_dead() const; |
| 418 | #define is_not_dead(n)((n) == __null || !VerifyIterativeGVN || !((n)->is_dead()) ) ((n) == NULL__null || !VerifyIterativeGVN || !((n)->is_dead())) |
| 419 | bool is_reachable_from_root() const; |
| 420 | #endif |
| 421 | // Check whether node has become unreachable |
| 422 | bool is_unreachable(PhaseIterGVN &igvn) const; |
| 423 | |
| 424 | // Set a required input edge, also updates corresponding output edge |
| 425 | void add_req( Node *n ); // Append a NEW required input |
| 426 | void add_req( Node *n0, Node *n1 ) { |
| 427 | add_req(n0); add_req(n1); } |
| 428 | void add_req( Node *n0, Node *n1, Node *n2 ) { |
| 429 | add_req(n0); add_req(n1); add_req(n2); } |
| 430 | void add_req_batch( Node* n, uint m ); // Append m NEW required inputs (all n). |
| 431 | void del_req( uint idx ); // Delete required edge & compact |
| 432 | void del_req_ordered( uint idx ); // Delete required edge & compact with preserved order |
| 433 | void ins_req( uint i, Node *n ); // Insert a NEW required input |
| 434 | void set_req( uint i, Node *n ) { |
| 435 | assert( is_not_dead(n), "can not use dead node")do { if (!(((n) == __null || !VerifyIterativeGVN || !((n)-> is_dead())))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 435, "assert(" "((n) == __null || !VerifyIterativeGVN || !((n)->is_dead()))" ") failed", "can not use dead node"); ::breakpoint(); } } while (0); |
| 436 | assert( i < _cnt, "oob: i=%d, _cnt=%d", i, _cnt)do { if (!(i < _cnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 436, "assert(" "i < _cnt" ") failed", "oob: i=%d, _cnt=%d" , i, _cnt); ::breakpoint(); } } while (0); |
| 437 | assert( !VerifyHashTableKeys || _hash_lock == 0,do { if (!(!VerifyHashTableKeys || _hash_lock == 0)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 438, "assert(" "!VerifyHashTableKeys || _hash_lock == 0" ") failed" , "remove node from hash table before modifying it"); ::breakpoint (); } } while (0) |
| 438 | "remove node from hash table before modifying it")do { if (!(!VerifyHashTableKeys || _hash_lock == 0)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 438, "assert(" "!VerifyHashTableKeys || _hash_lock == 0" ") failed" , "remove node from hash table before modifying it"); ::breakpoint (); } } while (0); |
| 439 | Node** p = &_in[i]; // cache this._in, across the del_out call |
| 440 | if (*p != NULL__null) (*p)->del_out((Node *)this); |
| 441 | (*p) = n; |
| 442 | if (n != NULL__null) n->add_out((Node *)this); |
| 443 | Compile::current()->record_modified_node(this); |
| 444 | } |
| 445 | // Light version of set_req() to init inputs after node creation. |
| 446 | void init_req( uint i, Node *n ) { |
| 447 | assert( i == 0 && this == n ||do { if (!(i == 0 && this == n || ((n) == __null || ! VerifyIterativeGVN || !((n)->is_dead())))) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 448, "assert(" "i == 0 && this == n || ((n) == __null || !VerifyIterativeGVN || !((n)->is_dead()))" ") failed", "can not use dead node"); ::breakpoint(); } } while (0) |
| 448 | is_not_dead(n), "can not use dead node")do { if (!(i == 0 && this == n || ((n) == __null || ! VerifyIterativeGVN || !((n)->is_dead())))) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 448, "assert(" "i == 0 && this == n || ((n) == __null || !VerifyIterativeGVN || !((n)->is_dead()))" ") failed", "can not use dead node"); ::breakpoint(); } } while (0); |
| 449 | assert( i < _cnt, "oob")do { if (!(i < _cnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 449, "assert(" "i < _cnt" ") failed", "oob"); ::breakpoint (); } } while (0); |
| 450 | assert( !VerifyHashTableKeys || _hash_lock == 0,do { if (!(!VerifyHashTableKeys || _hash_lock == 0)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 451, "assert(" "!VerifyHashTableKeys || _hash_lock == 0" ") failed" , "remove node from hash table before modifying it"); ::breakpoint (); } } while (0) |
| 451 | "remove node from hash table before modifying it")do { if (!(!VerifyHashTableKeys || _hash_lock == 0)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 451, "assert(" "!VerifyHashTableKeys || _hash_lock == 0" ") failed" , "remove node from hash table before modifying it"); ::breakpoint (); } } while (0); |
| 452 | assert( _in[i] == NULL, "sanity")do { if (!(_in[i] == __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 452, "assert(" "_in[i] == __null" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 453 | _in[i] = n; |
| 454 | if (n != NULL__null) n->add_out((Node *)this); |
| 455 | Compile::current()->record_modified_node(this); |
| 456 | } |
| 457 | // Find first occurrence of n among my edges: |
| 458 | int find_edge(Node* n); |
| 459 | int find_prec_edge(Node* n) { |
| 460 | for (uint i = req(); i < len(); i++) { |
| 461 | if (_in[i] == n) return i; |
| 462 | if (_in[i] == NULL__null) { |
| 463 | DEBUG_ONLY( while ((++i) < len()) assert(_in[i] == NULL, "Gap in prec edges!"); )while ((++i) < len()) do { if (!(_in[i] == __null)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 463, "assert(" "_in[i] == __null" ") failed", "Gap in prec edges!" ); ::breakpoint(); } } while (0); |
| 464 | break; |
| 465 | } |
| 466 | } |
| 467 | return -1; |
| 468 | } |
| 469 | int replace_edge(Node* old, Node* neww, PhaseGVN* gvn = NULL__null); |
| 470 | int replace_edges_in_range(Node* old, Node* neww, int start, int end, PhaseGVN* gvn); |
| 471 | // NULL out all inputs to eliminate incoming Def-Use edges. |
| 472 | void disconnect_inputs(Compile* C); |
| 473 | |
| 474 | // Quickly, return true if and only if I am Compile::current()->top(). |
| 475 | bool is_top() const { |
| 476 | assert((this == (Node*) Compile::current()->top()) == (_out == NULL), "")do { if (!((this == (Node*) Compile::current()->top()) == ( _out == __null))) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 476, "assert(" "(this == (Node*) Compile::current()->top()) == (_out == __null)" ") failed", ""); ::breakpoint(); } } while (0); |
| 477 | return (_out == NULL__null); |
| 478 | } |
| 479 | // Reaffirm invariants for is_top. (Only from Compile::set_cached_top_node.) |
| 480 | void setup_is_top(); |
| 481 | |
| 482 | // Strip away casting. (It is depth-limited.) |
| 483 | Node* uncast(bool keep_deps = false) const; |
| 484 | // Return whether two Nodes are equivalent, after stripping casting. |
| 485 | bool eqv_uncast(const Node* n, bool keep_deps = false) const { |
| 486 | return (this->uncast(keep_deps) == n->uncast(keep_deps)); |
| 487 | } |
| 488 | |
| 489 | // Find out of current node that matches opcode. |
| 490 | Node* find_out_with(int opcode); |
| 491 | // Return true if the current node has an out that matches opcode. |
| 492 | bool has_out_with(int opcode); |
| 493 | // Return true if the current node has an out that matches any of the opcodes. |
| 494 | bool has_out_with(int opcode1, int opcode2, int opcode3, int opcode4); |
| 495 | |
| 496 | private: |
| 497 | static Node* uncast_helper(const Node* n, bool keep_deps); |
| 498 | |
| 499 | // Add an output edge to the end of the list |
| 500 | void add_out( Node *n ) { |
| 501 | if (is_top()) return; |
| 502 | if( _outcnt == _outmax ) out_grow(_outcnt); |
| 503 | _out[_outcnt++] = n; |
| 504 | } |
| 505 | // Delete an output edge |
| 506 | void del_out( Node *n ) { |
| 507 | if (is_top()) return; |
| 508 | Node** outp = &_out[_outcnt]; |
| 509 | // Find and remove n |
| 510 | do { |
| 511 | assert(outp > _out, "Missing Def-Use edge")do { if (!(outp > _out)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 511, "assert(" "outp > _out" ") failed", "Missing Def-Use edge" ); ::breakpoint(); } } while (0); |
| 512 | } while (*--outp != n); |
| 513 | *outp = _out[--_outcnt]; |
| 514 | // Smash the old edge so it can't be used accidentally. |
| 515 | debug_only(_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef)_out[_outcnt] = (Node *)(uintptr_t)0xdeadbeef; |
| 516 | // Record that a change happened here. |
| 517 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 518 | debug_only(_last_del = n; ++_del_tick)_last_del = n; ++_del_tick; |
| 519 | #endif |
| 520 | } |
| 521 | // Close gap after removing edge. |
| 522 | void close_prec_gap_at(uint gap) { |
| 523 | assert(_cnt <= gap && gap < _max, "no valid prec edge")do { if (!(_cnt <= gap && gap < _max)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 523, "assert(" "_cnt <= gap && gap < _max" ") failed" , "no valid prec edge"); ::breakpoint(); } } while (0); |
| 524 | uint i = gap; |
| 525 | Node *last = NULL__null; |
| 526 | for (; i < _max-1; ++i) { |
| 527 | Node *next = _in[i+1]; |
| 528 | if (next == NULL__null) break; |
| 529 | last = next; |
| 530 | } |
| 531 | _in[gap] = last; // Move last slot to empty one. |
| 532 | _in[i] = NULL__null; // NULL out last slot. |
| 533 | } |
| 534 | |
| 535 | public: |
| 536 | // Globally replace this node by a given new node, updating all uses. |
| 537 | void replace_by(Node* new_node); |
| 538 | // Globally replace this node by a given new node, updating all uses |
| 539 | // and cutting input edges of old node. |
| 540 | void subsume_by(Node* new_node, Compile* c) { |
| 541 | replace_by(new_node); |
| 542 | disconnect_inputs(c); |
| 543 | } |
| 544 | void set_req_X(uint i, Node *n, PhaseIterGVN *igvn); |
| 545 | void set_req_X(uint i, Node *n, PhaseGVN *gvn); |
| 546 | // Find the one non-null required input. RegionNode only |
| 547 | Node *nonnull_req() const; |
| 548 | // Add or remove precedence edges |
| 549 | void add_prec( Node *n ); |
| 550 | void rm_prec( uint i ); |
| 551 | |
| 552 | // Note: prec(i) will not necessarily point to n if edge already exists. |
| 553 | void set_prec( uint i, Node *n ) { |
| 554 | assert(i < _max, "oob: i=%d, _max=%d", i, _max)do { if (!(i < _max)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 554, "assert(" "i < _max" ") failed", "oob: i=%d, _max=%d" , i, _max); ::breakpoint(); } } while (0); |
| 555 | assert(is_not_dead(n), "can not use dead node")do { if (!(((n) == __null || !VerifyIterativeGVN || !((n)-> is_dead())))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 555, "assert(" "((n) == __null || !VerifyIterativeGVN || !((n)->is_dead()))" ") failed", "can not use dead node"); ::breakpoint(); } } while (0); |
| 556 | assert(i >= _cnt, "not a precedence edge")do { if (!(i >= _cnt)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 556, "assert(" "i >= _cnt" ") failed", "not a precedence edge" ); ::breakpoint(); } } while (0); |
| 557 | // Avoid spec violation: duplicated prec edge. |
| 558 | if (_in[i] == n) return; |
| 559 | if (n == NULL__null || find_prec_edge(n) != -1) { |
| 560 | rm_prec(i); |
| 561 | return; |
| 562 | } |
| 563 | if (_in[i] != NULL__null) _in[i]->del_out((Node *)this); |
| 564 | _in[i] = n; |
| 565 | n->add_out((Node *)this); |
| 566 | } |
| 567 | |
| 568 | // Set this node's index, used by cisc_version to replace current node |
| 569 | void set_idx(uint new_idx) { |
| 570 | const node_idx_t* ref = &_idx; |
| 571 | *(node_idx_t*)ref = new_idx; |
| 572 | } |
| 573 | // Swap input edge order. (Edge indexes i1 and i2 are usually 1 and 2.) |
| 574 | void swap_edges(uint i1, uint i2) { |
| 575 | debug_only(uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH)uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH; |
| 576 | // Def-Use info is unchanged |
| 577 | Node* n1 = in(i1); |
| 578 | Node* n2 = in(i2); |
| 579 | _in[i1] = n2; |
| 580 | _in[i2] = n1; |
| 581 | // If this node is in the hash table, make sure it doesn't need a rehash. |
| 582 | assert(check_hash == NO_HASH || check_hash == hash(), "edge swap must preserve hash code")do { if (!(check_hash == NO_HASH || check_hash == hash())) { ( *g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 582, "assert(" "check_hash == NO_HASH || check_hash == hash()" ") failed", "edge swap must preserve hash code"); ::breakpoint (); } } while (0); |
| 583 | } |
| 584 | |
| 585 | // Iterators over input Nodes for a Node X are written as: |
| 586 | // for( i = 0; i < X.req(); i++ ) ... X[i] ... |
| 587 | // NOTE: Required edges can contain embedded NULL pointers. |
| 588 | |
| 589 | //----------------- Other Node Properties |
| 590 | |
| 591 | // Generate class IDs for (some) ideal nodes so that it is possible to determine |
| 592 | // the type of a node using a non-virtual method call (the method is_<Node>() below). |
| 593 | // |
| 594 | // A class ID of an ideal node is a set of bits. In a class ID, a single bit determines |
| 595 | // the type of the node the ID represents; another subset of an ID's bits are reserved |
| 596 | // for the superclasses of the node represented by the ID. |
| 597 | // |
| 598 | // By design, if A is a supertype of B, A.is_B() returns true and B.is_A() |
| 599 | // returns false. A.is_A() returns true. |
| 600 | // |
| 601 | // If two classes, A and B, have the same superclass, a different bit of A's class id |
| 602 | // is reserved for A's type than for B's type. That bit is specified by the third |
| 603 | // parameter in the macro DEFINE_CLASS_ID. |
| 604 | // |
| 605 | // By convention, classes with deeper hierarchy are declared first. Moreover, |
| 606 | // classes with the same hierarchy depth are sorted by usage frequency. |
| 607 | // |
| 608 | // The query method masks the bits to cut off bits of subclasses and then compares |
| 609 | // the result with the class id (see the macro DEFINE_CLASS_QUERY below). |
| 610 | // |
| 611 | // Class_MachCall=30, ClassMask_MachCall=31 |
| 612 | // 12 8 4 0 |
| 613 | // 0 0 0 0 0 0 0 0 1 1 1 1 0 |
| 614 | // | | | | |
| 615 | // | | | Bit_Mach=2 |
| 616 | // | | Bit_MachReturn=4 |
| 617 | // | Bit_MachSafePoint=8 |
| 618 | // Bit_MachCall=16 |
| 619 | // |
| 620 | // Class_CountedLoop=56, ClassMask_CountedLoop=63 |
| 621 | // 12 8 4 0 |
| 622 | // 0 0 0 0 0 0 0 1 1 1 0 0 0 |
| 623 | // | | | |
| 624 | // | | Bit_Region=8 |
| 625 | // | Bit_Loop=16 |
| 626 | // Bit_CountedLoop=32 |
| 627 | |
| 628 | #define DEFINE_CLASS_ID(cl, supcl, subn) \ |
| 629 | Bit_##cl = (Class_##supcl == 0) ? 1 << subn : (Bit_##supcl) << (1 + subn) , \ |
| 630 | Class_##cl = Class_##supcl + Bit_##cl , \ |
| 631 | ClassMask_##cl = ((Bit_##cl << 1) - 1) , |
| 632 | |
| 633 | // This enum is used only for C2 ideal and mach nodes with is_<node>() methods |
| 634 | // so that its values fit into 32 bits. |
| 635 | enum NodeClasses { |
| 636 | Bit_Node = 0x00000000, |
| 637 | Class_Node = 0x00000000, |
| 638 | ClassMask_Node = 0xFFFFFFFF, |
| 639 | |
| 640 | DEFINE_CLASS_ID(Multi, Node, 0) |
| 641 | DEFINE_CLASS_ID(SafePoint, Multi, 0) |
| 642 | DEFINE_CLASS_ID(Call, SafePoint, 0) |
| 643 | DEFINE_CLASS_ID(CallJava, Call, 0) |
| 644 | DEFINE_CLASS_ID(CallStaticJava, CallJava, 0) |
| 645 | DEFINE_CLASS_ID(CallDynamicJava, CallJava, 1) |
| 646 | DEFINE_CLASS_ID(CallRuntime, Call, 1) |
| 647 | DEFINE_CLASS_ID(CallLeaf, CallRuntime, 0) |
| 648 | DEFINE_CLASS_ID(CallLeafNoFP, CallLeaf, 0) |
| 649 | DEFINE_CLASS_ID(Allocate, Call, 2) |
| 650 | DEFINE_CLASS_ID(AllocateArray, Allocate, 0) |
| 651 | DEFINE_CLASS_ID(AbstractLock, Call, 3) |
| 652 | DEFINE_CLASS_ID(Lock, AbstractLock, 0) |
| 653 | DEFINE_CLASS_ID(Unlock, AbstractLock, 1) |
| 654 | DEFINE_CLASS_ID(ArrayCopy, Call, 4) |
| 655 | DEFINE_CLASS_ID(CallNative, Call, 5) |
| 656 | DEFINE_CLASS_ID(MultiBranch, Multi, 1) |
| 657 | DEFINE_CLASS_ID(PCTable, MultiBranch, 0) |
| 658 | DEFINE_CLASS_ID(Catch, PCTable, 0) |
| 659 | DEFINE_CLASS_ID(Jump, PCTable, 1) |
| 660 | DEFINE_CLASS_ID(If, MultiBranch, 1) |
| 661 | DEFINE_CLASS_ID(BaseCountedLoopEnd, If, 0) |
| 662 | DEFINE_CLASS_ID(CountedLoopEnd, BaseCountedLoopEnd, 0) |
| 663 | DEFINE_CLASS_ID(LongCountedLoopEnd, BaseCountedLoopEnd, 1) |
| 664 | DEFINE_CLASS_ID(RangeCheck, If, 1) |
| 665 | DEFINE_CLASS_ID(OuterStripMinedLoopEnd, If, 2) |
| 666 | DEFINE_CLASS_ID(NeverBranch, MultiBranch, 2) |
| 667 | DEFINE_CLASS_ID(Start, Multi, 2) |
| 668 | DEFINE_CLASS_ID(MemBar, Multi, 3) |
| 669 | DEFINE_CLASS_ID(Initialize, MemBar, 0) |
| 670 | DEFINE_CLASS_ID(MemBarStoreStore, MemBar, 1) |
| 671 | |
| 672 | DEFINE_CLASS_ID(Mach, Node, 1) |
| 673 | DEFINE_CLASS_ID(MachReturn, Mach, 0) |
| 674 | DEFINE_CLASS_ID(MachSafePoint, MachReturn, 0) |
| 675 | DEFINE_CLASS_ID(MachCall, MachSafePoint, 0) |
| 676 | DEFINE_CLASS_ID(MachCallJava, MachCall, 0) |
| 677 | DEFINE_CLASS_ID(MachCallStaticJava, MachCallJava, 0) |
| 678 | DEFINE_CLASS_ID(MachCallDynamicJava, MachCallJava, 1) |
| 679 | DEFINE_CLASS_ID(MachCallRuntime, MachCall, 1) |
| 680 | DEFINE_CLASS_ID(MachCallLeaf, MachCallRuntime, 0) |
| 681 | DEFINE_CLASS_ID(MachCallNative, MachCall, 2) |
| 682 | DEFINE_CLASS_ID(MachBranch, Mach, 1) |
| 683 | DEFINE_CLASS_ID(MachIf, MachBranch, 0) |
| 684 | DEFINE_CLASS_ID(MachGoto, MachBranch, 1) |
| 685 | DEFINE_CLASS_ID(MachNullCheck, MachBranch, 2) |
| 686 | DEFINE_CLASS_ID(MachSpillCopy, Mach, 2) |
| 687 | DEFINE_CLASS_ID(MachTemp, Mach, 3) |
| 688 | DEFINE_CLASS_ID(MachConstantBase, Mach, 4) |
| 689 | DEFINE_CLASS_ID(MachConstant, Mach, 5) |
| 690 | DEFINE_CLASS_ID(MachJump, MachConstant, 0) |
| 691 | DEFINE_CLASS_ID(MachMerge, Mach, 6) |
| 692 | DEFINE_CLASS_ID(MachMemBar, Mach, 7) |
| 693 | |
| 694 | DEFINE_CLASS_ID(Type, Node, 2) |
| 695 | DEFINE_CLASS_ID(Phi, Type, 0) |
| 696 | DEFINE_CLASS_ID(ConstraintCast, Type, 1) |
| 697 | DEFINE_CLASS_ID(CastII, ConstraintCast, 0) |
| 698 | DEFINE_CLASS_ID(CheckCastPP, ConstraintCast, 1) |
| 699 | DEFINE_CLASS_ID(CastLL, ConstraintCast, 2) |
| 700 | DEFINE_CLASS_ID(CastFF, ConstraintCast, 3) |
| 701 | DEFINE_CLASS_ID(CastDD, ConstraintCast, 4) |
| 702 | DEFINE_CLASS_ID(CastVV, ConstraintCast, 5) |
| 703 | DEFINE_CLASS_ID(CMove, Type, 3) |
| 704 | DEFINE_CLASS_ID(SafePointScalarObject, Type, 4) |
| 705 | DEFINE_CLASS_ID(DecodeNarrowPtr, Type, 5) |
| 706 | DEFINE_CLASS_ID(DecodeN, DecodeNarrowPtr, 0) |
| 707 | DEFINE_CLASS_ID(DecodeNKlass, DecodeNarrowPtr, 1) |
| 708 | DEFINE_CLASS_ID(EncodeNarrowPtr, Type, 6) |
| 709 | DEFINE_CLASS_ID(EncodeP, EncodeNarrowPtr, 0) |
| 710 | DEFINE_CLASS_ID(EncodePKlass, EncodeNarrowPtr, 1) |
| 711 | DEFINE_CLASS_ID(Vector, Type, 7) |
| 712 | DEFINE_CLASS_ID(VectorMaskCmp, Vector, 0) |
| 713 | DEFINE_CLASS_ID(VectorUnbox, Vector, 1) |
| 714 | DEFINE_CLASS_ID(VectorReinterpret, Vector, 2) |
| 715 | DEFINE_CLASS_ID(ShiftV, Vector, 3) |
| 716 | |
| 717 | DEFINE_CLASS_ID(Proj, Node, 3) |
| 718 | DEFINE_CLASS_ID(CatchProj, Proj, 0) |
| 719 | DEFINE_CLASS_ID(JumpProj, Proj, 1) |
| 720 | DEFINE_CLASS_ID(IfProj, Proj, 2) |
| 721 | DEFINE_CLASS_ID(IfTrue, IfProj, 0) |
| 722 | DEFINE_CLASS_ID(IfFalse, IfProj, 1) |
| 723 | DEFINE_CLASS_ID(Parm, Proj, 4) |
| 724 | DEFINE_CLASS_ID(MachProj, Proj, 5) |
| 725 | |
| 726 | DEFINE_CLASS_ID(Mem, Node, 4) |
| 727 | DEFINE_CLASS_ID(Load, Mem, 0) |
| 728 | DEFINE_CLASS_ID(LoadVector, Load, 0) |
| 729 | DEFINE_CLASS_ID(LoadVectorGather, LoadVector, 0) |
| 730 | DEFINE_CLASS_ID(LoadVectorMasked, LoadVector, 1) |
| 731 | DEFINE_CLASS_ID(Store, Mem, 1) |
| 732 | DEFINE_CLASS_ID(StoreVector, Store, 0) |
| 733 | DEFINE_CLASS_ID(StoreVectorScatter, StoreVector, 0) |
| 734 | DEFINE_CLASS_ID(StoreVectorMasked, StoreVector, 1) |
| 735 | DEFINE_CLASS_ID(LoadStore, Mem, 2) |
| 736 | DEFINE_CLASS_ID(LoadStoreConditional, LoadStore, 0) |
| 737 | DEFINE_CLASS_ID(CompareAndSwap, LoadStoreConditional, 0) |
| 738 | DEFINE_CLASS_ID(CompareAndExchangeNode, LoadStore, 1) |
| 739 | |
| 740 | DEFINE_CLASS_ID(Region, Node, 5) |
| 741 | DEFINE_CLASS_ID(Loop, Region, 0) |
| 742 | DEFINE_CLASS_ID(Root, Loop, 0) |
| 743 | DEFINE_CLASS_ID(BaseCountedLoop, Loop, 1) |
| 744 | DEFINE_CLASS_ID(CountedLoop, BaseCountedLoop, 0) |
| 745 | DEFINE_CLASS_ID(LongCountedLoop, BaseCountedLoop, 1) |
| 746 | DEFINE_CLASS_ID(OuterStripMinedLoop, Loop, 2) |
| 747 | |
| 748 | DEFINE_CLASS_ID(Sub, Node, 6) |
| 749 | DEFINE_CLASS_ID(Cmp, Sub, 0) |
| 750 | DEFINE_CLASS_ID(FastLock, Cmp, 0) |
| 751 | DEFINE_CLASS_ID(FastUnlock, Cmp, 1) |
| 752 | DEFINE_CLASS_ID(SubTypeCheck,Cmp, 2) |
| 753 | |
| 754 | DEFINE_CLASS_ID(MergeMem, Node, 7) |
| 755 | DEFINE_CLASS_ID(Bool, Node, 8) |
| 756 | DEFINE_CLASS_ID(AddP, Node, 9) |
| 757 | DEFINE_CLASS_ID(BoxLock, Node, 10) |
| 758 | DEFINE_CLASS_ID(Add, Node, 11) |
| 759 | DEFINE_CLASS_ID(Mul, Node, 12) |
| 760 | DEFINE_CLASS_ID(ClearArray, Node, 14) |
| 761 | DEFINE_CLASS_ID(Halt, Node, 15) |
| 762 | DEFINE_CLASS_ID(Opaque1, Node, 16) |
| 763 | DEFINE_CLASS_ID(Move, Node, 17) |
| 764 | DEFINE_CLASS_ID(LShift, Node, 18) |
| 765 | |
| 766 | _max_classes = ClassMask_Move |
| 767 | }; |
| 768 | #undef DEFINE_CLASS_ID |
| 769 | |
| 770 | // Flags are sorted by usage frequency. |
| 771 | enum NodeFlags { |
| 772 | Flag_is_Copy = 1 << 0, // should be first bit to avoid shift |
| 773 | Flag_rematerialize = 1 << 1, |
| 774 | Flag_needs_anti_dependence_check = 1 << 2, |
| 775 | Flag_is_macro = 1 << 3, |
| 776 | Flag_is_Con = 1 << 4, |
| 777 | Flag_is_cisc_alternate = 1 << 5, |
| 778 | Flag_is_dead_loop_safe = 1 << 6, |
| 779 | Flag_may_be_short_branch = 1 << 7, |
| 780 | Flag_avoid_back_to_back_before = 1 << 8, |
| 781 | Flag_avoid_back_to_back_after = 1 << 9, |
| 782 | Flag_has_call = 1 << 10, |
| 783 | Flag_is_reduction = 1 << 11, |
| 784 | Flag_is_scheduled = 1 << 12, |
| 785 | Flag_has_vector_mask_set = 1 << 13, |
| 786 | Flag_is_expensive = 1 << 14, |
| 787 | Flag_is_predicated_vector = 1 << 15, |
| 788 | Flag_for_post_loop_opts_igvn = 1 << 16, |
| 789 | _last_flag = Flag_for_post_loop_opts_igvn |
| 790 | }; |
| 791 | |
| 792 | class PD; |
| 793 | |
| 794 | private: |
| 795 | juint _class_id; |
| 796 | juint _flags; |
| 797 | |
| 798 | static juint max_flags(); |
| 799 | |
| 800 | protected: |
| 801 | // These methods should be called from constructors only. |
| 802 | void init_class_id(juint c) { |
| 803 | _class_id = c; // cast out const |
| 804 | } |
| 805 | void init_flags(uint fl) { |
| 806 | assert(fl <= max_flags(), "invalid node flag")do { if (!(fl <= max_flags())) { (*g_assert_poison) = 'X'; ; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 806, "assert(" "fl <= max_flags()" ") failed", "invalid node flag" ); ::breakpoint(); } } while (0); |
| 807 | _flags |= fl; |
| 808 | } |
| 809 | void clear_flag(uint fl) { |
| 810 | assert(fl <= max_flags(), "invalid node flag")do { if (!(fl <= max_flags())) { (*g_assert_poison) = 'X'; ; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 810, "assert(" "fl <= max_flags()" ") failed", "invalid node flag" ); ::breakpoint(); } } while (0); |
| 811 | _flags &= ~fl; |
| 812 | } |
| 813 | |
| 814 | public: |
| 815 | const juint class_id() const { return _class_id; } |
| 816 | |
| 817 | const juint flags() const { return _flags; } |
| 818 | |
| 819 | void add_flag(juint fl) { init_flags(fl); } |
| 820 | |
| 821 | void remove_flag(juint fl) { clear_flag(fl); } |
| 822 | |
| 823 | // Return a dense integer opcode number |
| 824 | virtual int Opcode() const; |
| 825 | |
| 826 | // Virtual inherited Node size |
| 827 | virtual uint size_of() const; |
| 828 | |
| 829 | // Other interesting Node properties |
| 830 | #define DEFINE_CLASS_QUERY(type) \ |
| 831 | bool is_##type() const { \ |
| 832 | return ((_class_id & ClassMask_##type) == Class_##type); \ |
| 833 | } \ |
| 834 | type##Node *as_##type() const { \ |
| 835 | assert(is_##type(), "invalid node class: %s", Name())do { if (!(is_##type())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 835, "assert(" "is_##type()" ") failed", "invalid node class: %s" , Name()); ::breakpoint(); } } while (0); \ |
| 836 | return (type##Node*)this; \ |
| 837 | } \ |
| 838 | type##Node* isa_##type() const { \ |
| 839 | return (is_##type()) ? as_##type() : NULL__null; \ |
| 840 | } |
| 841 | |
| 842 | DEFINE_CLASS_QUERY(AbstractLock) |
| 843 | DEFINE_CLASS_QUERY(Add) |
| 844 | DEFINE_CLASS_QUERY(AddP) |
| 845 | DEFINE_CLASS_QUERY(Allocate) |
| 846 | DEFINE_CLASS_QUERY(AllocateArray) |
| 847 | DEFINE_CLASS_QUERY(ArrayCopy) |
| 848 | DEFINE_CLASS_QUERY(BaseCountedLoop) |
| 849 | DEFINE_CLASS_QUERY(BaseCountedLoopEnd) |
| 850 | DEFINE_CLASS_QUERY(Bool) |
| 851 | DEFINE_CLASS_QUERY(BoxLock) |
| 852 | DEFINE_CLASS_QUERY(Call) |
| 853 | DEFINE_CLASS_QUERY(CallNative) |
| 854 | DEFINE_CLASS_QUERY(CallDynamicJava) |
| 855 | DEFINE_CLASS_QUERY(CallJava) |
| 856 | DEFINE_CLASS_QUERY(CallLeaf) |
| 857 | DEFINE_CLASS_QUERY(CallLeafNoFP) |
| 858 | DEFINE_CLASS_QUERY(CallRuntime) |
| 859 | DEFINE_CLASS_QUERY(CallStaticJava) |
| 860 | DEFINE_CLASS_QUERY(Catch) |
| 861 | DEFINE_CLASS_QUERY(CatchProj) |
| 862 | DEFINE_CLASS_QUERY(CheckCastPP) |
| 863 | DEFINE_CLASS_QUERY(CastII) |
| 864 | DEFINE_CLASS_QUERY(CastLL) |
| 865 | DEFINE_CLASS_QUERY(ConstraintCast) |
| 866 | DEFINE_CLASS_QUERY(ClearArray) |
| 867 | DEFINE_CLASS_QUERY(CMove) |
| 868 | DEFINE_CLASS_QUERY(Cmp) |
| 869 | DEFINE_CLASS_QUERY(CountedLoop) |
| 870 | DEFINE_CLASS_QUERY(CountedLoopEnd) |
| 871 | DEFINE_CLASS_QUERY(DecodeNarrowPtr) |
| 872 | DEFINE_CLASS_QUERY(DecodeN) |
| 873 | DEFINE_CLASS_QUERY(DecodeNKlass) |
| 874 | DEFINE_CLASS_QUERY(EncodeNarrowPtr) |
| 875 | DEFINE_CLASS_QUERY(EncodeP) |
| 876 | DEFINE_CLASS_QUERY(EncodePKlass) |
| 877 | DEFINE_CLASS_QUERY(FastLock) |
| 878 | DEFINE_CLASS_QUERY(FastUnlock) |
| 879 | DEFINE_CLASS_QUERY(Halt) |
| 880 | DEFINE_CLASS_QUERY(If) |
| 881 | DEFINE_CLASS_QUERY(RangeCheck) |
| 882 | DEFINE_CLASS_QUERY(IfProj) |
| 883 | DEFINE_CLASS_QUERY(IfFalse) |
| 884 | DEFINE_CLASS_QUERY(IfTrue) |
| 885 | DEFINE_CLASS_QUERY(Initialize) |
| 886 | DEFINE_CLASS_QUERY(Jump) |
| 887 | DEFINE_CLASS_QUERY(JumpProj) |
| 888 | DEFINE_CLASS_QUERY(LongCountedLoop) |
| 889 | DEFINE_CLASS_QUERY(LongCountedLoopEnd) |
| 890 | DEFINE_CLASS_QUERY(Load) |
| 891 | DEFINE_CLASS_QUERY(LoadStore) |
| 892 | DEFINE_CLASS_QUERY(LoadStoreConditional) |
| 893 | DEFINE_CLASS_QUERY(Lock) |
| 894 | DEFINE_CLASS_QUERY(Loop) |
| 895 | DEFINE_CLASS_QUERY(LShift) |
| 896 | DEFINE_CLASS_QUERY(Mach) |
| 897 | DEFINE_CLASS_QUERY(MachBranch) |
| 898 | DEFINE_CLASS_QUERY(MachCall) |
| 899 | DEFINE_CLASS_QUERY(MachCallNative) |
| 900 | DEFINE_CLASS_QUERY(MachCallDynamicJava) |
| 901 | DEFINE_CLASS_QUERY(MachCallJava) |
| 902 | DEFINE_CLASS_QUERY(MachCallLeaf) |
| 903 | DEFINE_CLASS_QUERY(MachCallRuntime) |
| 904 | DEFINE_CLASS_QUERY(MachCallStaticJava) |
| 905 | DEFINE_CLASS_QUERY(MachConstantBase) |
| 906 | DEFINE_CLASS_QUERY(MachConstant) |
| 907 | DEFINE_CLASS_QUERY(MachGoto) |
| 908 | DEFINE_CLASS_QUERY(MachIf) |
| 909 | DEFINE_CLASS_QUERY(MachJump) |
| 910 | DEFINE_CLASS_QUERY(MachNullCheck) |
| 911 | DEFINE_CLASS_QUERY(MachProj) |
| 912 | DEFINE_CLASS_QUERY(MachReturn) |
| 913 | DEFINE_CLASS_QUERY(MachSafePoint) |
| 914 | DEFINE_CLASS_QUERY(MachSpillCopy) |
| 915 | DEFINE_CLASS_QUERY(MachTemp) |
| 916 | DEFINE_CLASS_QUERY(MachMemBar) |
| 917 | DEFINE_CLASS_QUERY(MachMerge) |
| 918 | DEFINE_CLASS_QUERY(Mem) |
| 919 | DEFINE_CLASS_QUERY(MemBar) |
| 920 | DEFINE_CLASS_QUERY(MemBarStoreStore) |
| 921 | DEFINE_CLASS_QUERY(MergeMem) |
| 922 | DEFINE_CLASS_QUERY(Move) |
| 923 | DEFINE_CLASS_QUERY(Mul) |
| 924 | DEFINE_CLASS_QUERY(Multi) |
| 925 | DEFINE_CLASS_QUERY(MultiBranch) |
| 926 | DEFINE_CLASS_QUERY(Opaque1) |
| 927 | DEFINE_CLASS_QUERY(OuterStripMinedLoop) |
| 928 | DEFINE_CLASS_QUERY(OuterStripMinedLoopEnd) |
| 929 | DEFINE_CLASS_QUERY(Parm) |
| 930 | DEFINE_CLASS_QUERY(PCTable) |
| 931 | DEFINE_CLASS_QUERY(Phi) |
| 932 | DEFINE_CLASS_QUERY(Proj) |
| 933 | DEFINE_CLASS_QUERY(Region) |
| 934 | DEFINE_CLASS_QUERY(Root) |
| 935 | DEFINE_CLASS_QUERY(SafePoint) |
| 936 | DEFINE_CLASS_QUERY(SafePointScalarObject) |
| 937 | DEFINE_CLASS_QUERY(Start) |
| 938 | DEFINE_CLASS_QUERY(Store) |
| 939 | DEFINE_CLASS_QUERY(Sub) |
| 940 | DEFINE_CLASS_QUERY(SubTypeCheck) |
| 941 | DEFINE_CLASS_QUERY(Type) |
| 942 | DEFINE_CLASS_QUERY(Vector) |
| 943 | DEFINE_CLASS_QUERY(VectorMaskCmp) |
| 944 | DEFINE_CLASS_QUERY(VectorUnbox) |
| 945 | DEFINE_CLASS_QUERY(VectorReinterpret); |
| 946 | DEFINE_CLASS_QUERY(LoadVector) |
| 947 | DEFINE_CLASS_QUERY(LoadVectorGather) |
| 948 | DEFINE_CLASS_QUERY(StoreVector) |
| 949 | DEFINE_CLASS_QUERY(StoreVectorScatter) |
| 950 | DEFINE_CLASS_QUERY(ShiftV) |
| 951 | DEFINE_CLASS_QUERY(Unlock) |
| 952 | |
| 953 | #undef DEFINE_CLASS_QUERY |
| 954 | |
| 955 | // duplicate of is_MachSpillCopy() |
| 956 | bool is_SpillCopy () const { |
| 957 | return ((_class_id & ClassMask_MachSpillCopy) == Class_MachSpillCopy); |
| 958 | } |
| 959 | |
| 960 | bool is_Con () const { return (_flags & Flag_is_Con) != 0; } |
| 961 | // The data node which is safe to leave in dead loop during IGVN optimization. |
| 962 | bool is_dead_loop_safe() const; |
| 963 | |
| 964 | // is_Copy() returns copied edge index (0 or 1) |
| 965 | uint is_Copy() const { return (_flags & Flag_is_Copy); } |
| 966 | |
| 967 | virtual bool is_CFG() const { return false; } |
| 968 | |
| 969 | // If this node is control-dependent on a test, can it be |
| 970 | // rerouted to a dominating equivalent test? This is usually |
| 971 | // true of non-CFG nodes, but can be false for operations which |
| 972 | // depend for their correct sequencing on more than one test. |
| 973 | // (In that case, hoisting to a dominating test may silently |
| 974 | // skip some other important test.) |
| 975 | virtual bool depends_only_on_test() const { assert(!is_CFG(), "")do { if (!(!is_CFG())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 975, "assert(" "!is_CFG()" ") failed", ""); ::breakpoint(); } } while (0); return true; }; |
| 976 | |
| 977 | // When building basic blocks, I need to have a notion of block beginning |
| 978 | // Nodes, next block selector Nodes (block enders), and next block |
| 979 | // projections. These calls need to work on their machine equivalents. The |
| 980 | // Ideal beginning Nodes are RootNode, RegionNode and StartNode. |
| 981 | bool is_block_start() const { |
| 982 | if ( is_Region() ) |
| 983 | return this == (const Node*)in(0); |
| 984 | else |
| 985 | return is_Start(); |
| 986 | } |
| 987 | |
| 988 | // The Ideal control projection Nodes are IfTrue/IfFalse, JumpProjNode, Root, |
| 989 | // Goto and Return. This call also returns the block ending Node. |
| 990 | virtual const Node *is_block_proj() const; |
| 991 | |
| 992 | // The node is a "macro" node which needs to be expanded before matching |
| 993 | bool is_macro() const { return (_flags & Flag_is_macro) != 0; } |
| 994 | // The node is expensive: the best control is set during loop opts |
| 995 | bool is_expensive() const { return (_flags & Flag_is_expensive) != 0 && in(0) != NULL__null; } |
| 996 | |
| 997 | // An arithmetic node which accumulates a data in a loop. |
| 998 | // It must have the loop's phi as input and provide a def to the phi. |
| 999 | bool is_reduction() const { return (_flags & Flag_is_reduction) != 0; } |
| 1000 | |
| 1001 | bool is_predicated_vector() const { return (_flags & Flag_is_predicated_vector) != 0; } |
| 1002 | |
| 1003 | // The node is a CountedLoopEnd with a mask annotation so as to emit a restore context |
| 1004 | bool has_vector_mask_set() const { return (_flags & Flag_has_vector_mask_set) != 0; } |
| 1005 | |
| 1006 | // Used in lcm to mark nodes that have scheduled |
| 1007 | bool is_scheduled() const { return (_flags & Flag_is_scheduled) != 0; } |
| 1008 | |
| 1009 | bool for_post_loop_opts_igvn() const { return (_flags & Flag_for_post_loop_opts_igvn) != 0; } |
| 1010 | |
| 1011 | //----------------- Optimization |
| 1012 | |
| 1013 | // Get the worst-case Type output for this Node. |
| 1014 | virtual const class Type *bottom_type() const; |
| 1015 | |
| 1016 | // If we find a better type for a node, try to record it permanently. |
| 1017 | // Return true if this node actually changed. |
| 1018 | // Be sure to do the hash_delete game in the "rehash" variant. |
| 1019 | void raise_bottom_type(const Type* new_type); |
| 1020 | |
| 1021 | // Get the address type with which this node uses and/or defs memory, |
| 1022 | // or NULL if none. The address type is conservatively wide. |
| 1023 | // Returns non-null for calls, membars, loads, stores, etc. |
| 1024 | // Returns TypePtr::BOTTOM if the node touches memory "broadly". |
| 1025 | virtual const class TypePtr *adr_type() const { return NULL__null; } |
| 1026 | |
| 1027 | // Return an existing node which computes the same function as this node. |
| 1028 | // The optimistic combined algorithm requires this to return a Node which |
| 1029 | // is a small number of steps away (e.g., one of my inputs). |
| 1030 | virtual Node* Identity(PhaseGVN* phase); |
| 1031 | |
| 1032 | // Return the set of values this Node can take on at runtime. |
| 1033 | virtual const Type* Value(PhaseGVN* phase) const; |
| 1034 | |
| 1035 | // Return a node which is more "ideal" than the current node. |
| 1036 | // The invariants on this call are subtle. If in doubt, read the |
| 1037 | // treatise in node.cpp above the default implemention AND TEST WITH |
| 1038 | // +VerifyIterativeGVN! |
| 1039 | virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
| 1040 | |
| 1041 | // Some nodes have specific Ideal subgraph transformations only if they are |
| 1042 | // unique users of specific nodes. Such nodes should be put on IGVN worklist |
| 1043 | // for the transformations to happen. |
| 1044 | bool has_special_unique_user() const; |
| 1045 | |
| 1046 | // Skip Proj and CatchProj nodes chains. Check for Null and Top. |
| 1047 | Node* find_exact_control(Node* ctrl); |
| 1048 | |
| 1049 | // Check if 'this' node dominates or equal to 'sub'. |
| 1050 | bool dominates(Node* sub, Node_List &nlist); |
| 1051 | |
| 1052 | protected: |
| 1053 | bool remove_dead_region(PhaseGVN *phase, bool can_reshape); |
| 1054 | public: |
| 1055 | |
| 1056 | // See if there is valid pipeline info |
| 1057 | static const Pipeline *pipeline_class(); |
| 1058 | virtual const Pipeline *pipeline() const; |
| 1059 | |
| 1060 | // Compute the latency from the def to this instruction of the ith input node |
| 1061 | uint latency(uint i); |
| 1062 | |
| 1063 | // Hash & compare functions, for pessimistic value numbering |
| 1064 | |
| 1065 | // If the hash function returns the special sentinel value NO_HASH, |
| 1066 | // the node is guaranteed never to compare equal to any other node. |
| 1067 | // If we accidentally generate a hash with value NO_HASH the node |
| 1068 | // won't go into the table and we'll lose a little optimization. |
| 1069 | static const uint NO_HASH = 0; |
| 1070 | virtual uint hash() const; |
| 1071 | virtual bool cmp( const Node &n ) const; |
| 1072 | |
| 1073 | // Operation appears to be iteratively computed (such as an induction variable) |
| 1074 | // It is possible for this operation to return false for a loop-varying |
| 1075 | // value, if it appears (by local graph inspection) to be computed by a simple conditional. |
| 1076 | bool is_iteratively_computed(); |
| 1077 | |
| 1078 | // Determine if a node is a counted loop induction variable. |
| 1079 | // NOTE: The method is defined in "loopnode.cpp". |
| 1080 | bool is_cloop_ind_var() const; |
| 1081 | |
| 1082 | // Return a node with opcode "opc" and same inputs as "this" if one can |
| 1083 | // be found; Otherwise return NULL; |
| 1084 | Node* find_similar(int opc); |
| 1085 | |
| 1086 | // Return the unique control out if only one. Null if none or more than one. |
| 1087 | Node* unique_ctrl_out() const; |
| 1088 | |
| 1089 | // Set control or add control as precedence edge |
| 1090 | void ensure_control_or_add_prec(Node* c); |
| 1091 | |
| 1092 | //----------------- Code Generation |
| 1093 | |
| 1094 | // Ideal register class for Matching. Zero means unmatched instruction |
| 1095 | // (these are cloned instead of converted to machine nodes). |
| 1096 | virtual uint ideal_reg() const; |
| 1097 | |
| 1098 | static const uint NotAMachineReg; // must be > max. machine register |
| 1099 | |
| 1100 | // Do we Match on this edge index or not? Generally false for Control |
| 1101 | // and true for everything else. Weird for calls & returns. |
| 1102 | virtual uint match_edge(uint idx) const; |
| 1103 | |
| 1104 | // Register class output is returned in |
| 1105 | virtual const RegMask &out_RegMask() const; |
| 1106 | // Register class input is expected in |
| 1107 | virtual const RegMask &in_RegMask(uint) const; |
| 1108 | // Should we clone rather than spill this instruction? |
| 1109 | bool rematerialize() const; |
| 1110 | |
| 1111 | // Return JVM State Object if this Node carries debug info, or NULL otherwise |
| 1112 | virtual JVMState* jvms() const; |
| 1113 | |
| 1114 | // Print as assembly |
| 1115 | virtual void format( PhaseRegAlloc *, outputStream* st = tty ) const; |
| 1116 | // Emit bytes starting at parameter 'ptr' |
| 1117 | // Bump 'ptr' by the number of output bytes |
| 1118 | virtual void emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const; |
| 1119 | // Size of instruction in bytes |
| 1120 | virtual uint size(PhaseRegAlloc *ra_) const; |
| 1121 | |
| 1122 | // Convenience function to extract an integer constant from a node. |
| 1123 | // If it is not an integer constant (either Con, CastII, or Mach), |
| 1124 | // return value_if_unknown. |
| 1125 | jint find_int_con(jint value_if_unknown) const { |
| 1126 | const TypeInt* t = find_int_type(); |
| 1127 | return (t != NULL__null && t->is_con()) ? t->get_con() : value_if_unknown; |
| 1128 | } |
| 1129 | // Return the constant, knowing it is an integer constant already |
| 1130 | jint get_int() const { |
| 1131 | const TypeInt* t = find_int_type(); |
| 1132 | guarantee(t != NULL, "must be con")do { if (!(t != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1132, "guarantee(" "t != NULL" ") failed", "must be con"); :: breakpoint(); } } while (0); |
| 1133 | return t->get_con(); |
| 1134 | } |
| 1135 | // Here's where the work is done. Can produce non-constant int types too. |
| 1136 | const TypeInt* find_int_type() const; |
| 1137 | const TypeInteger* find_integer_type(BasicType bt) const; |
| 1138 | |
| 1139 | // Same thing for long (and intptr_t, via type.hpp): |
| 1140 | jlong get_long() const { |
| 1141 | const TypeLong* t = find_long_type(); |
| 1142 | guarantee(t != NULL, "must be con")do { if (!(t != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1142, "guarantee(" "t != NULL" ") failed", "must be con"); :: breakpoint(); } } while (0); |
| 1143 | return t->get_con(); |
| 1144 | } |
| 1145 | jlong find_long_con(jint value_if_unknown) const { |
| 1146 | const TypeLong* t = find_long_type(); |
| 1147 | return (t != NULL__null && t->is_con()) ? t->get_con() : value_if_unknown; |
| 1148 | } |
| 1149 | const TypeLong* find_long_type() const; |
| 1150 | |
| 1151 | jlong get_integer_as_long(BasicType bt) const { |
| 1152 | const TypeInteger* t = find_integer_type(bt); |
| 1153 | guarantee(t != NULL, "must be con")do { if (!(t != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1153, "guarantee(" "t != NULL" ") failed", "must be con"); :: breakpoint(); } } while (0); |
| 1154 | return t->get_con_as_long(bt); |
| 1155 | } |
| 1156 | const TypePtr* get_ptr_type() const; |
| 1157 | |
| 1158 | // These guys are called by code generated by ADLC: |
| 1159 | intptr_t get_ptr() const; |
| 1160 | intptr_t get_narrowcon() const; |
| 1161 | jdouble getd() const; |
| 1162 | jfloat getf() const; |
| 1163 | |
| 1164 | // Nodes which are pinned into basic blocks |
| 1165 | virtual bool pinned() const { return false; } |
| 1166 | |
| 1167 | // Nodes which use memory without consuming it, hence need antidependences |
| 1168 | // More specifically, needs_anti_dependence_check returns true iff the node |
| 1169 | // (a) does a load, and (b) does not perform a store (except perhaps to a |
| 1170 | // stack slot or some other unaliased location). |
| 1171 | bool needs_anti_dependence_check() const; |
| 1172 | |
| 1173 | // Return which operand this instruction may cisc-spill. In other words, |
| 1174 | // return operand position that can convert from reg to memory access |
| 1175 | virtual int cisc_operand() const { return AdlcVMDeps::Not_cisc_spillable; } |
| 1176 | bool is_cisc_alternate() const { return (_flags & Flag_is_cisc_alternate) != 0; } |
| 1177 | |
| 1178 | // Whether this is a memory-writing machine node. |
| 1179 | bool is_memory_writer() const { return is_Mach() && bottom_type()->has_memory(); } |
| 1180 | |
| 1181 | //----------------- Printing, etc |
| 1182 | #ifndef PRODUCT |
| 1183 | private: |
| 1184 | int _indent; |
| 1185 | |
| 1186 | public: |
| 1187 | void set_indent(int indent) { _indent = indent; } |
| 1188 | |
| 1189 | private: |
| 1190 | static bool add_to_worklist(Node* n, Node_List* worklist, Arena* old_arena, VectorSet* old_space, VectorSet* new_space); |
| 1191 | public: |
| 1192 | Node* find(int idx, bool only_ctrl = false); // Search the graph for the given idx. |
| 1193 | Node* find_ctrl(int idx); // Search control ancestors for the given idx. |
| 1194 | void dump() const { dump("\n"); } // Print this node. |
| 1195 | void dump(const char* suffix, bool mark = false, outputStream *st = tty) const; // Print this node. |
| 1196 | void dump(int depth) const; // Print this node, recursively to depth d |
| 1197 | void dump_ctrl(int depth) const; // Print control nodes, to depth d |
| 1198 | void dump_comp() const; // Print this node in compact representation. |
| 1199 | // Print this node in compact representation. |
| 1200 | void dump_comp(const char* suffix, outputStream *st = tty) const; |
| 1201 | virtual void dump_req(outputStream *st = tty) const; // Print required-edge info |
| 1202 | virtual void dump_prec(outputStream *st = tty) const; // Print precedence-edge info |
| 1203 | virtual void dump_out(outputStream *st = tty) const; // Print the output edge info |
| 1204 | virtual void dump_spec(outputStream *st) const {}; // Print per-node info |
| 1205 | // Print compact per-node info |
| 1206 | virtual void dump_compact_spec(outputStream *st) const { dump_spec(st); } |
| 1207 | void dump_related() const; // Print related nodes (depends on node at hand). |
| 1208 | // Print related nodes up to given depths for input and output nodes. |
| 1209 | void dump_related(uint d_in, uint d_out) const; |
| 1210 | void dump_related_compact() const; // Print related nodes in compact representation. |
| 1211 | // Collect related nodes. |
| 1212 | virtual void related(GrowableArray<Node*> *in_rel, GrowableArray<Node*> *out_rel, bool compact) const; |
| 1213 | // Collect nodes starting from this node, explicitly including/excluding control and data links. |
| 1214 | void collect_nodes(GrowableArray<Node*> *ns, int d, bool ctrl, bool data) const; |
| 1215 | |
| 1216 | // Node collectors, to be used in implementations of Node::rel(). |
| 1217 | // Collect the entire data input graph. Include control inputs if requested. |
| 1218 | void collect_nodes_in_all_data(GrowableArray<Node*> *ns, bool ctrl) const; |
| 1219 | // Collect the entire control input graph. Include data inputs if requested. |
| 1220 | void collect_nodes_in_all_ctrl(GrowableArray<Node*> *ns, bool data) const; |
| 1221 | // Collect the entire output graph until hitting and including control nodes. |
| 1222 | void collect_nodes_out_all_ctrl_boundary(GrowableArray<Node*> *ns) const; |
| 1223 | |
| 1224 | void verify_edges(Unique_Node_List &visited); // Verify bi-directional edges |
| 1225 | static void verify(int verify_depth, VectorSet& visited, Node_List& worklist); |
| 1226 | |
| 1227 | // This call defines a class-unique string used to identify class instances |
| 1228 | virtual const char *Name() const; |
| 1229 | |
| 1230 | void dump_format(PhaseRegAlloc *ra) const; // debug access to MachNode::format(...) |
| 1231 | // RegMask Print Functions |
| 1232 | void dump_in_regmask(int idx) { in_RegMask(idx).dump(); } |
| 1233 | void dump_out_regmask() { out_RegMask().dump(); } |
| 1234 | static bool in_dump() { return Compile::current()->_in_dump_cnt > 0; } |
| 1235 | void fast_dump() const { |
| 1236 | tty->print("%4d: %-17s", _idx, Name()); |
| 1237 | for (uint i = 0; i < len(); i++) |
| 1238 | if (in(i)) |
| 1239 | tty->print(" %4d", in(i)->_idx); |
| 1240 | else |
| 1241 | tty->print(" NULL"); |
| 1242 | tty->print("\n"); |
| 1243 | } |
| 1244 | #endif |
| 1245 | #ifdef ASSERT1 |
| 1246 | void verify_construction(); |
| 1247 | bool verify_jvms(const JVMState* jvms) const; |
| 1248 | int _debug_idx; // Unique value assigned to every node. |
| 1249 | int debug_idx() const { return _debug_idx; } |
| 1250 | void set_debug_idx( int debug_idx ) { _debug_idx = debug_idx; } |
| 1251 | |
| 1252 | Node* _debug_orig; // Original version of this, if any. |
| 1253 | Node* debug_orig() const { return _debug_orig; } |
| 1254 | void set_debug_orig(Node* orig); // _debug_orig = orig |
| 1255 | void dump_orig(outputStream *st, bool print_key = true) const; |
| 1256 | |
| 1257 | int _hash_lock; // Barrier to modifications of nodes in the hash table |
| 1258 | void enter_hash_lock() { ++_hash_lock; assert(_hash_lock < 99, "in too many hash tables?")do { if (!(_hash_lock < 99)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1258, "assert(" "_hash_lock < 99" ") failed", "in too many hash tables?" ); ::breakpoint(); } } while (0); } |
| 1259 | void exit_hash_lock() { --_hash_lock; assert(_hash_lock >= 0, "mispaired hash locks")do { if (!(_hash_lock >= 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1259, "assert(" "_hash_lock >= 0" ") failed", "mispaired hash locks" ); ::breakpoint(); } } while (0); } |
| 1260 | |
| 1261 | static void init_NodeProperty(); |
| 1262 | |
| 1263 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 1264 | const Node* _last_del; // The last deleted node. |
| 1265 | uint _del_tick; // Bumped when a deletion happens.. |
| 1266 | #endif |
| 1267 | #endif |
| 1268 | }; |
| 1269 | |
| 1270 | inline bool not_a_node(const Node* n) { |
| 1271 | if (n == NULL__null) return true; |
| 1272 | if (((intptr_t)n & 1) != 0) return true; // uninitialized, etc. |
| 1273 | if (*(address*)n == badAddress((address)::badAddressVal)) return true; // kill by Node::destruct |
| 1274 | return false; |
| 1275 | } |
| 1276 | |
| 1277 | //----------------------------------------------------------------------------- |
| 1278 | // Iterators over DU info, and associated Node functions. |
| 1279 | |
| 1280 | #if OPTO_DU_ITERATOR_ASSERT1 |
| 1281 | |
| 1282 | // Common code for assertion checking on DU iterators. |
| 1283 | class DUIterator_Common { |
| 1284 | #ifdef ASSERT1 |
| 1285 | protected: |
| 1286 | bool _vdui; // cached value of VerifyDUIterators |
| 1287 | const Node* _node; // the node containing the _out array |
| 1288 | uint _outcnt; // cached node->_outcnt |
| 1289 | uint _del_tick; // cached node->_del_tick |
| 1290 | Node* _last; // last value produced by the iterator |
| 1291 | |
| 1292 | void sample(const Node* node); // used by c'tor to set up for verifies |
| 1293 | void verify(const Node* node, bool at_end_ok = false); |
| 1294 | void verify_resync(); |
| 1295 | void reset(const DUIterator_Common& that); |
| 1296 | |
| 1297 | // The VDUI_ONLY macro protects code conditionalized on VerifyDUIterators |
| 1298 | #define I_VDUI_ONLY(i,x) { if ((i)._vdui) { x; } } |
| 1299 | #else |
| 1300 | #define I_VDUI_ONLY(i,x) { } |
| 1301 | #endif //ASSERT |
| 1302 | }; |
| 1303 | |
| 1304 | #define VDUI_ONLY(x) I_VDUI_ONLY(*this, x) |
| 1305 | |
| 1306 | // Default DU iterator. Allows appends onto the out array. |
| 1307 | // Allows deletion from the out array only at the current point. |
| 1308 | // Usage: |
| 1309 | // for (DUIterator i = x->outs(); x->has_out(i); i++) { |
| 1310 | // Node* y = x->out(i); |
| 1311 | // ... |
| 1312 | // } |
| 1313 | // Compiles in product mode to a unsigned integer index, which indexes |
| 1314 | // onto a repeatedly reloaded base pointer of x->_out. The loop predicate |
| 1315 | // also reloads x->_outcnt. If you delete, you must perform "--i" just |
| 1316 | // before continuing the loop. You must delete only the last-produced |
| 1317 | // edge. You must delete only a single copy of the last-produced edge, |
| 1318 | // or else you must delete all copies at once (the first time the edge |
| 1319 | // is produced by the iterator). |
| 1320 | class DUIterator : public DUIterator_Common { |
| 1321 | friend class Node; |
| 1322 | |
| 1323 | // This is the index which provides the product-mode behavior. |
| 1324 | // Whatever the product-mode version of the system does to the |
| 1325 | // DUI index is done to this index. All other fields in |
| 1326 | // this class are used only for assertion checking. |
| 1327 | uint _idx; |
| 1328 | |
| 1329 | #ifdef ASSERT1 |
| 1330 | uint _refresh_tick; // Records the refresh activity. |
| 1331 | |
| 1332 | void sample(const Node* node); // Initialize _refresh_tick etc. |
| 1333 | void verify(const Node* node, bool at_end_ok = false); |
| 1334 | void verify_increment(); // Verify an increment operation. |
| 1335 | void verify_resync(); // Verify that we can back up over a deletion. |
| 1336 | void verify_finish(); // Verify that the loop terminated properly. |
| 1337 | void refresh(); // Resample verification info. |
| 1338 | void reset(const DUIterator& that); // Resample after assignment. |
| 1339 | #endif |
| 1340 | |
| 1341 | DUIterator(const Node* node, int dummy_to_avoid_conversion) |
| 1342 | { _idx = 0; debug_only(sample(node))sample(node); } |
| 1343 | |
| 1344 | public: |
| 1345 | // initialize to garbage; clear _vdui to disable asserts |
| 1346 | DUIterator() |
| 1347 | { /*initialize to garbage*/ debug_only(_vdui = false)_vdui = false; } |
| 1348 | |
| 1349 | DUIterator(const DUIterator& that) |
| 1350 | { _idx = that._idx; debug_only(_vdui = false; reset(that))_vdui = false; reset(that); } |
| 1351 | |
| 1352 | void operator++(int dummy_to_specify_postfix_op) |
| 1353 | { _idx++; VDUI_ONLY(verify_increment()); } |
| 1354 | |
| 1355 | void operator--() |
| 1356 | { VDUI_ONLY(verify_resync()); --_idx; } |
| 1357 | |
| 1358 | ~DUIterator() |
| 1359 | { VDUI_ONLY(verify_finish()); } |
| 1360 | |
| 1361 | void operator=(const DUIterator& that) |
| 1362 | { _idx = that._idx; debug_only(reset(that))reset(that); } |
| 1363 | }; |
| 1364 | |
| 1365 | DUIterator Node::outs() const |
| 1366 | { return DUIterator(this, 0); } |
| 1367 | DUIterator& Node::refresh_out_pos(DUIterator& i) const |
| 1368 | { I_VDUI_ONLY(i, i.refresh()); return i; } |
| 1369 | bool Node::has_out(DUIterator& i) const |
| 1370 | { I_VDUI_ONLY(i, i.verify(this,true));return i._idx < _outcnt; } |
| 1371 | Node* Node::out(DUIterator& i) const |
| 1372 | { I_VDUI_ONLY(i, i.verify(this)); return debug_only(i._last=)i._last= _out[i._idx]; } |
| 1373 | |
| 1374 | |
| 1375 | // Faster DU iterator. Disallows insertions into the out array. |
| 1376 | // Allows deletion from the out array only at the current point. |
| 1377 | // Usage: |
| 1378 | // for (DUIterator_Fast imax, i = x->fast_outs(imax); i < imax; i++) { |
| 1379 | // Node* y = x->fast_out(i); |
| 1380 | // ... |
| 1381 | // } |
| 1382 | // Compiles in product mode to raw Node** pointer arithmetic, with |
| 1383 | // no reloading of pointers from the original node x. If you delete, |
| 1384 | // you must perform "--i; --imax" just before continuing the loop. |
| 1385 | // If you delete multiple copies of the same edge, you must decrement |
| 1386 | // imax, but not i, multiple times: "--i, imax -= num_edges". |
| 1387 | class DUIterator_Fast : public DUIterator_Common { |
| 1388 | friend class Node; |
| 1389 | friend class DUIterator_Last; |
| 1390 | |
| 1391 | // This is the pointer which provides the product-mode behavior. |
| 1392 | // Whatever the product-mode version of the system does to the |
| 1393 | // DUI pointer is done to this pointer. All other fields in |
| 1394 | // this class are used only for assertion checking. |
| 1395 | Node** _outp; |
| 1396 | |
| 1397 | #ifdef ASSERT1 |
| 1398 | void verify(const Node* node, bool at_end_ok = false); |
| 1399 | void verify_limit(); |
| 1400 | void verify_resync(); |
| 1401 | void verify_relimit(uint n); |
| 1402 | void reset(const DUIterator_Fast& that); |
| 1403 | #endif |
| 1404 | |
| 1405 | // Note: offset must be signed, since -1 is sometimes passed |
| 1406 | DUIterator_Fast(const Node* node, ptrdiff_t offset) |
| 1407 | { _outp = node->_out + offset; debug_only(sample(node))sample(node); } |
| 1408 | |
| 1409 | public: |
| 1410 | // initialize to garbage; clear _vdui to disable asserts |
| 1411 | DUIterator_Fast() |
| 1412 | { /*initialize to garbage*/ debug_only(_vdui = false)_vdui = false; } |
| 1413 | |
| 1414 | DUIterator_Fast(const DUIterator_Fast& that) |
| 1415 | { _outp = that._outp; debug_only(_vdui = false; reset(that))_vdui = false; reset(that); } |
| 1416 | |
| 1417 | void operator++(int dummy_to_specify_postfix_op) |
| 1418 | { _outp++; VDUI_ONLY(verify(_node, true)); } |
| 1419 | |
| 1420 | void operator--() |
| 1421 | { VDUI_ONLY(verify_resync()); --_outp; } |
| 1422 | |
| 1423 | void operator-=(uint n) // applied to the limit only |
| 1424 | { _outp -= n; VDUI_ONLY(verify_relimit(n)); } |
| 1425 | |
| 1426 | bool operator<(DUIterator_Fast& limit) { |
| 1427 | I_VDUI_ONLY(*this, this->verify(_node, true)); |
| 1428 | I_VDUI_ONLY(limit, limit.verify_limit()); |
| 1429 | return _outp < limit._outp; |
| 1430 | } |
| 1431 | |
| 1432 | void operator=(const DUIterator_Fast& that) |
| 1433 | { _outp = that._outp; debug_only(reset(that))reset(that); } |
| 1434 | }; |
| 1435 | |
| 1436 | DUIterator_Fast Node::fast_outs(DUIterator_Fast& imax) const { |
| 1437 | // Assign a limit pointer to the reference argument: |
| 1438 | imax = DUIterator_Fast(this, (ptrdiff_t)_outcnt); |
| 1439 | // Return the base pointer: |
| 1440 | return DUIterator_Fast(this, 0); |
| 1441 | } |
| 1442 | Node* Node::fast_out(DUIterator_Fast& i) const { |
| 1443 | I_VDUI_ONLY(i, i.verify(this)); |
| 1444 | return debug_only(i._last=)i._last= *i._outp; |
| 1445 | } |
| 1446 | |
| 1447 | |
| 1448 | // Faster DU iterator. Requires each successive edge to be removed. |
| 1449 | // Does not allow insertion of any edges. |
| 1450 | // Usage: |
| 1451 | // for (DUIterator_Last imin, i = x->last_outs(imin); i >= imin; i -= num_edges) { |
| 1452 | // Node* y = x->last_out(i); |
| 1453 | // ... |
| 1454 | // } |
| 1455 | // Compiles in product mode to raw Node** pointer arithmetic, with |
| 1456 | // no reloading of pointers from the original node x. |
| 1457 | class DUIterator_Last : private DUIterator_Fast { |
| 1458 | friend class Node; |
| 1459 | |
| 1460 | #ifdef ASSERT1 |
| 1461 | void verify(const Node* node, bool at_end_ok = false); |
| 1462 | void verify_limit(); |
| 1463 | void verify_step(uint num_edges); |
| 1464 | #endif |
| 1465 | |
| 1466 | // Note: offset must be signed, since -1 is sometimes passed |
| 1467 | DUIterator_Last(const Node* node, ptrdiff_t offset) |
| 1468 | : DUIterator_Fast(node, offset) { } |
| 1469 | |
| 1470 | void operator++(int dummy_to_specify_postfix_op) {} // do not use |
| 1471 | void operator<(int) {} // do not use |
| 1472 | |
| 1473 | public: |
| 1474 | DUIterator_Last() { } |
| 1475 | // initialize to garbage |
| 1476 | |
| 1477 | DUIterator_Last(const DUIterator_Last& that) = default; |
| 1478 | |
| 1479 | void operator--() |
| 1480 | { _outp--; VDUI_ONLY(verify_step(1)); } |
| 1481 | |
| 1482 | void operator-=(uint n) |
| 1483 | { _outp -= n; VDUI_ONLY(verify_step(n)); } |
| 1484 | |
| 1485 | bool operator>=(DUIterator_Last& limit) { |
| 1486 | I_VDUI_ONLY(*this, this->verify(_node, true)); |
| 1487 | I_VDUI_ONLY(limit, limit.verify_limit()); |
| 1488 | return _outp >= limit._outp; |
| 1489 | } |
| 1490 | |
| 1491 | DUIterator_Last& operator=(const DUIterator_Last& that) = default; |
| 1492 | }; |
| 1493 | |
| 1494 | DUIterator_Last Node::last_outs(DUIterator_Last& imin) const { |
| 1495 | // Assign a limit pointer to the reference argument: |
| 1496 | imin = DUIterator_Last(this, 0); |
| 1497 | // Return the initial pointer: |
| 1498 | return DUIterator_Last(this, (ptrdiff_t)_outcnt - 1); |
| 1499 | } |
| 1500 | Node* Node::last_out(DUIterator_Last& i) const { |
| 1501 | I_VDUI_ONLY(i, i.verify(this)); |
| 1502 | return debug_only(i._last=)i._last= *i._outp; |
| 1503 | } |
| 1504 | |
| 1505 | #endif //OPTO_DU_ITERATOR_ASSERT |
| 1506 | |
| 1507 | #undef I_VDUI_ONLY |
| 1508 | #undef VDUI_ONLY |
| 1509 | |
| 1510 | // An Iterator that truly follows the iterator pattern. Doesn't |
| 1511 | // support deletion but could be made to. |
| 1512 | // |
| 1513 | // for (SimpleDUIterator i(n); i.has_next(); i.next()) { |
| 1514 | // Node* m = i.get(); |
| 1515 | // |
| 1516 | class SimpleDUIterator : public StackObj { |
| 1517 | private: |
| 1518 | Node* node; |
| 1519 | DUIterator_Fast i; |
| 1520 | DUIterator_Fast imax; |
| 1521 | public: |
| 1522 | SimpleDUIterator(Node* n): node(n), i(n->fast_outs(imax)) {} |
| 1523 | bool has_next() { return i < imax; } |
| 1524 | void next() { i++; } |
| 1525 | Node* get() { return node->fast_out(i); } |
| 1526 | }; |
| 1527 | |
| 1528 | |
| 1529 | //----------------------------------------------------------------------------- |
| 1530 | // Map dense integer indices to Nodes. Uses classic doubling-array trick. |
| 1531 | // Abstractly provides an infinite array of Node*'s, initialized to NULL. |
| 1532 | // Note that the constructor just zeros things, and since I use Arena |
| 1533 | // allocation I do not need a destructor to reclaim storage. |
| 1534 | class Node_Array : public ResourceObj { |
| 1535 | friend class VMStructs; |
| 1536 | protected: |
| 1537 | Arena* _a; // Arena to allocate in |
| 1538 | uint _max; |
| 1539 | Node** _nodes; |
| 1540 | void grow( uint i ); // Grow array node to fit |
| 1541 | public: |
| 1542 | Node_Array(Arena* a, uint max = OptoNodeListSize) : _a(a), _max(max) { |
| 1543 | _nodes = NEW_ARENA_ARRAY(a, Node*, max)(Node**) (a)->Amalloc((max) * sizeof(Node*)); |
| 1544 | clear(); |
| 1545 | } |
| 1546 | |
| 1547 | Node_Array(Node_Array* na) : _a(na->_a), _max(na->_max), _nodes(na->_nodes) {} |
| 1548 | Node *operator[] ( uint i ) const // Lookup, or NULL for not mapped |
| 1549 | { return (i<_max) ? _nodes[i] : (Node*)NULL__null; } |
| 1550 | Node* at(uint i) const { assert(i<_max,"oob")do { if (!(i<_max)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1550, "assert(" "i<_max" ") failed", "oob"); ::breakpoint (); } } while (0); return _nodes[i]; } |
| 1551 | Node** adr() { return _nodes; } |
| 1552 | // Extend the mapping: index i maps to Node *n. |
| 1553 | void map( uint i, Node *n ) { if( i>=_max ) grow(i); _nodes[i] = n; } |
| 1554 | void insert( uint i, Node *n ); |
| 1555 | void remove( uint i ); // Remove, preserving order |
| 1556 | // Clear all entries in _nodes to NULL but keep storage |
| 1557 | void clear() { |
| 1558 | Copy::zero_to_bytes(_nodes, _max * sizeof(Node*)); |
| 1559 | } |
| 1560 | |
| 1561 | uint Size() const { return _max; } |
| 1562 | void dump() const; |
| 1563 | }; |
| 1564 | |
| 1565 | class Node_List : public Node_Array { |
| 1566 | friend class VMStructs; |
| 1567 | uint _cnt; |
| 1568 | public: |
| 1569 | Node_List(uint max = OptoNodeListSize) : Node_Array(Thread::current()->resource_area(), max), _cnt(0) {} |
| 1570 | Node_List(Arena *a, uint max = OptoNodeListSize) : Node_Array(a, max), _cnt(0) {} |
| 1571 | bool contains(const Node* n) const { |
| 1572 | for (uint e = 0; e < size(); e++) { |
| 1573 | if (at(e) == n) return true; |
| 1574 | } |
| 1575 | return false; |
| 1576 | } |
| 1577 | void insert( uint i, Node *n ) { Node_Array::insert(i,n); _cnt++; } |
| 1578 | void remove( uint i ) { Node_Array::remove(i); _cnt--; } |
| 1579 | void push( Node *b ) { map(_cnt++,b); } |
| 1580 | void yank( Node *n ); // Find and remove |
| 1581 | Node *pop() { return _nodes[--_cnt]; } |
| 1582 | void clear() { _cnt = 0; Node_Array::clear(); } // retain storage |
| 1583 | void copy(const Node_List& from) { |
| 1584 | if (from._max > _max) { |
| 1585 | grow(from._max); |
| 1586 | } |
| 1587 | _cnt = from._cnt; |
| 1588 | Copy::conjoint_words_to_higher((HeapWord*)&from._nodes[0], (HeapWord*)&_nodes[0], from._max * sizeof(Node*)); |
| 1589 | } |
| 1590 | |
| 1591 | uint size() const { return _cnt; } |
| 1592 | void dump() const; |
| 1593 | void dump_simple() const; |
| 1594 | }; |
| 1595 | |
| 1596 | //------------------------------Unique_Node_List------------------------------- |
| 1597 | class Unique_Node_List : public Node_List { |
| 1598 | friend class VMStructs; |
| 1599 | VectorSet _in_worklist; |
| 1600 | uint _clock_index; // Index in list where to pop from next |
| 1601 | public: |
| 1602 | Unique_Node_List() : Node_List(), _clock_index(0) {} |
| 1603 | Unique_Node_List(Arena *a) : Node_List(a), _in_worklist(a), _clock_index(0) {} |
| 1604 | |
| 1605 | void remove( Node *n ); |
| 1606 | bool member( Node *n ) { return _in_worklist.test(n->_idx) != 0; } |
| 1607 | VectorSet& member_set(){ return _in_worklist; } |
| 1608 | |
| 1609 | void push(Node* b) { |
| 1610 | if( !_in_worklist.test_set(b->_idx) ) |
| 1611 | Node_List::push(b); |
| 1612 | } |
| 1613 | Node *pop() { |
| 1614 | if( _clock_index >= size() ) _clock_index = 0; |
| 1615 | Node *b = at(_clock_index); |
| 1616 | map( _clock_index, Node_List::pop()); |
| 1617 | if (size() != 0) _clock_index++; // Always start from 0 |
| 1618 | _in_worklist.remove(b->_idx); |
| 1619 | return b; |
| 1620 | } |
| 1621 | Node *remove(uint i) { |
| 1622 | Node *b = Node_List::at(i); |
| 1623 | _in_worklist.remove(b->_idx); |
| 1624 | map(i,Node_List::pop()); |
| 1625 | return b; |
| 1626 | } |
| 1627 | void yank(Node *n) { |
| 1628 | _in_worklist.remove(n->_idx); |
| 1629 | Node_List::yank(n); |
| 1630 | } |
| 1631 | void clear() { |
| 1632 | _in_worklist.clear(); // Discards storage but grows automatically |
| 1633 | Node_List::clear(); |
| 1634 | _clock_index = 0; |
| 1635 | } |
| 1636 | |
| 1637 | // Used after parsing to remove useless nodes before Iterative GVN |
| 1638 | void remove_useless_nodes(VectorSet& useful); |
| 1639 | |
| 1640 | bool contains(const Node* n) const { |
| 1641 | fatal("use faster member() instead")do { (*g_assert_poison) = 'X';; report_fatal(INTERNAL_ERROR, "/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1641, "use faster member() instead"); ::breakpoint(); } while (0); |
| 1642 | return false; |
| 1643 | } |
| 1644 | |
| 1645 | #ifndef PRODUCT |
| 1646 | void print_set() const { _in_worklist.print(); } |
| 1647 | #endif |
| 1648 | }; |
| 1649 | |
| 1650 | // Inline definition of Compile::record_for_igvn must be deferred to this point. |
| 1651 | inline void Compile::record_for_igvn(Node* n) { |
| 1652 | _for_igvn->push(n); |
| 1653 | } |
| 1654 | |
| 1655 | //------------------------------Node_Stack------------------------------------- |
| 1656 | class Node_Stack { |
| 1657 | friend class VMStructs; |
| 1658 | protected: |
| 1659 | struct INode { |
| 1660 | Node *node; // Processed node |
| 1661 | uint indx; // Index of next node's child |
| 1662 | }; |
| 1663 | INode *_inode_top; // tos, stack grows up |
| 1664 | INode *_inode_max; // End of _inodes == _inodes + _max |
| 1665 | INode *_inodes; // Array storage for the stack |
| 1666 | Arena *_a; // Arena to allocate in |
| 1667 | void grow(); |
| 1668 | public: |
| 1669 | Node_Stack(int size) { |
| 1670 | size_t max = (size > OptoNodeListSize) ? size : OptoNodeListSize; |
| 1671 | _a = Thread::current()->resource_area(); |
| 1672 | _inodes = NEW_ARENA_ARRAY( _a, INode, max )(INode*) (_a)->Amalloc((max) * sizeof(INode)); |
| 1673 | _inode_max = _inodes + max; |
| 1674 | _inode_top = _inodes - 1; // stack is empty |
| 1675 | } |
| 1676 | |
| 1677 | Node_Stack(Arena *a, int size) : _a(a) { |
| 1678 | size_t max = (size > OptoNodeListSize) ? size : OptoNodeListSize; |
| 1679 | _inodes = NEW_ARENA_ARRAY( _a, INode, max )(INode*) (_a)->Amalloc((max) * sizeof(INode)); |
| 1680 | _inode_max = _inodes + max; |
| 1681 | _inode_top = _inodes - 1; // stack is empty |
| 1682 | } |
| 1683 | |
| 1684 | void pop() { |
| 1685 | assert(_inode_top >= _inodes, "node stack underflow")do { if (!(_inode_top >= _inodes)) { (*g_assert_poison) = 'X' ;; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1685, "assert(" "_inode_top >= _inodes" ") failed", "node stack underflow" ); ::breakpoint(); } } while (0); |
| 1686 | --_inode_top; |
| 1687 | } |
| 1688 | void push(Node *n, uint i) { |
| 1689 | ++_inode_top; |
| 1690 | if (_inode_top >= _inode_max) grow(); |
| 1691 | INode *top = _inode_top; // optimization |
| 1692 | top->node = n; |
| 1693 | top->indx = i; |
| 1694 | } |
| 1695 | Node *node() const { |
| 1696 | return _inode_top->node; |
| 1697 | } |
| 1698 | Node* node_at(uint i) const { |
| 1699 | assert(_inodes + i <= _inode_top, "in range")do { if (!(_inodes + i <= _inode_top)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1699, "assert(" "_inodes + i <= _inode_top" ") failed", "in range" ); ::breakpoint(); } } while (0); |
| 1700 | return _inodes[i].node; |
| 1701 | } |
| 1702 | uint index() const { |
| 1703 | return _inode_top->indx; |
| 1704 | } |
| 1705 | uint index_at(uint i) const { |
| 1706 | assert(_inodes + i <= _inode_top, "in range")do { if (!(_inodes + i <= _inode_top)) { (*g_assert_poison ) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1706, "assert(" "_inodes + i <= _inode_top" ") failed", "in range" ); ::breakpoint(); } } while (0); |
| 1707 | return _inodes[i].indx; |
| 1708 | } |
| 1709 | void set_node(Node *n) { |
| 1710 | _inode_top->node = n; |
| 1711 | } |
| 1712 | void set_index(uint i) { |
| 1713 | _inode_top->indx = i; |
| 1714 | } |
| 1715 | uint size_max() const { return (uint)pointer_delta(_inode_max, _inodes, sizeof(INode)); } // Max size |
| 1716 | uint size() const { return (uint)pointer_delta((_inode_top+1), _inodes, sizeof(INode)); } // Current size |
| 1717 | bool is_nonempty() const { return (_inode_top >= _inodes); } |
| 1718 | bool is_empty() const { return (_inode_top < _inodes); } |
| 1719 | void clear() { _inode_top = _inodes - 1; } // retain storage |
| 1720 | |
| 1721 | // Node_Stack is used to map nodes. |
| 1722 | Node* find(uint idx) const; |
| 1723 | }; |
| 1724 | |
| 1725 | |
| 1726 | //-----------------------------Node_Notes-------------------------------------- |
| 1727 | // Debugging or profiling annotations loosely and sparsely associated |
| 1728 | // with some nodes. See Compile::node_notes_at for the accessor. |
| 1729 | class Node_Notes { |
| 1730 | friend class VMStructs; |
| 1731 | JVMState* _jvms; |
| 1732 | |
| 1733 | public: |
| 1734 | Node_Notes(JVMState* jvms = NULL__null) { |
| 1735 | _jvms = jvms; |
| 1736 | } |
| 1737 | |
| 1738 | JVMState* jvms() { return _jvms; } |
| 1739 | void set_jvms(JVMState* x) { _jvms = x; } |
| 1740 | |
| 1741 | // True if there is nothing here. |
| 1742 | bool is_clear() { |
| 1743 | return (_jvms == NULL__null); |
| 1744 | } |
| 1745 | |
| 1746 | // Make there be nothing here. |
| 1747 | void clear() { |
| 1748 | _jvms = NULL__null; |
| 1749 | } |
| 1750 | |
| 1751 | // Make a new, clean node notes. |
| 1752 | static Node_Notes* make(Compile* C) { |
| 1753 | Node_Notes* nn = NEW_ARENA_ARRAY(C->comp_arena(), Node_Notes, 1)(Node_Notes*) (C->comp_arena())->Amalloc((1) * sizeof(Node_Notes )); |
| 1754 | nn->clear(); |
| 1755 | return nn; |
| 1756 | } |
| 1757 | |
| 1758 | Node_Notes* clone(Compile* C) { |
| 1759 | Node_Notes* nn = NEW_ARENA_ARRAY(C->comp_arena(), Node_Notes, 1)(Node_Notes*) (C->comp_arena())->Amalloc((1) * sizeof(Node_Notes )); |
| 1760 | (*nn) = (*this); |
| 1761 | return nn; |
| 1762 | } |
| 1763 | |
| 1764 | // Absorb any information from source. |
| 1765 | bool update_from(Node_Notes* source) { |
| 1766 | bool changed = false; |
| 1767 | if (source != NULL__null) { |
| 1768 | if (source->jvms() != NULL__null) { |
| 1769 | set_jvms(source->jvms()); |
| 1770 | changed = true; |
| 1771 | } |
| 1772 | } |
| 1773 | return changed; |
| 1774 | } |
| 1775 | }; |
| 1776 | |
| 1777 | // Inlined accessors for Compile::node_nodes that require the preceding class: |
| 1778 | inline Node_Notes* |
| 1779 | Compile::locate_node_notes(GrowableArray<Node_Notes*>* arr, |
| 1780 | int idx, bool can_grow) { |
| 1781 | assert(idx >= 0, "oob")do { if (!(idx >= 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1781, "assert(" "idx >= 0" ") failed", "oob"); ::breakpoint (); } } while (0); |
| 1782 | int block_idx = (idx >> _log2_node_notes_block_size); |
| 1783 | int grow_by = (block_idx - (arr == NULL__null? 0: arr->length())); |
| 1784 | if (grow_by >= 0) { |
| 1785 | if (!can_grow) return NULL__null; |
| 1786 | grow_node_notes(arr, grow_by + 1); |
| 1787 | } |
| 1788 | if (arr == NULL__null) return NULL__null; |
| 1789 | // (Every element of arr is a sub-array of length _node_notes_block_size.) |
| 1790 | return arr->at(block_idx) + (idx & (_node_notes_block_size-1)); |
| 1791 | } |
| 1792 | |
| 1793 | inline bool |
| 1794 | Compile::set_node_notes_at(int idx, Node_Notes* value) { |
| 1795 | if (value == NULL__null || value->is_clear()) |
| 1796 | return false; // nothing to write => write nothing |
| 1797 | Node_Notes* loc = locate_node_notes(_node_note_array, idx, true); |
| 1798 | assert(loc != NULL, "")do { if (!(loc != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1798, "assert(" "loc != __null" ") failed", ""); ::breakpoint (); } } while (0); |
| 1799 | return loc->update_from(value); |
| 1800 | } |
| 1801 | |
| 1802 | |
| 1803 | //------------------------------TypeNode--------------------------------------- |
| 1804 | // Node with a Type constant. |
| 1805 | class TypeNode : public Node { |
| 1806 | protected: |
| 1807 | virtual uint hash() const; // Check the type |
| 1808 | virtual bool cmp( const Node &n ) const; |
| 1809 | virtual uint size_of() const; // Size is bigger |
| 1810 | const Type* const _type; |
| 1811 | public: |
| 1812 | void set_type(const Type* t) { |
| 1813 | assert(t != NULL, "sanity")do { if (!(t != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1813, "assert(" "t != __null" ") failed", "sanity"); ::breakpoint (); } } while (0); |
| 1814 | debug_only(uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH)uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH; |
| 1815 | *(const Type**)&_type = t; // cast away const-ness |
| 1816 | // If this node is in the hash table, make sure it doesn't need a rehash. |
| 1817 | assert(check_hash == NO_HASH || check_hash == hash(), "type change must preserve hash code")do { if (!(check_hash == NO_HASH || check_hash == hash())) { ( *g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1817, "assert(" "check_hash == NO_HASH || check_hash == hash()" ") failed", "type change must preserve hash code"); ::breakpoint (); } } while (0); |
| 1818 | } |
| 1819 | const Type* type() const { assert(_type != NULL, "sanity")do { if (!(_type != __null)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1819, "assert(" "_type != __null" ") failed", "sanity"); :: breakpoint(); } } while (0); return _type; }; |
| 1820 | TypeNode( const Type *t, uint required ) : Node(required), _type(t) { |
| 1821 | init_class_id(Class_Type); |
| 1822 | } |
| 1823 | virtual const Type* Value(PhaseGVN* phase) const; |
| 1824 | virtual const Type *bottom_type() const; |
| 1825 | virtual uint ideal_reg() const; |
| 1826 | #ifndef PRODUCT |
| 1827 | virtual void dump_spec(outputStream *st) const; |
| 1828 | virtual void dump_compact_spec(outputStream *st) const; |
| 1829 | #endif |
| 1830 | }; |
| 1831 | |
| 1832 | #include "opto/opcodes.hpp" |
| 1833 | |
| 1834 | #define Op_IL(op)inline int Op_op(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1834, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_opI ; } return Op_opL; } \ |
| 1835 | inline int Op_ ## op(BasicType bt) { \ |
| 1836 | assert(bt == T_INT || bt == T_LONG, "only for int or longs")do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1836, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); \ |
| 1837 | if (bt == T_INT) { \ |
| 1838 | return Op_## op ## I; \ |
| 1839 | } \ |
| 1840 | return Op_## op ## L; \ |
| 1841 | } |
| 1842 | |
| 1843 | Op_IL(Add)inline int Op_Add(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1843, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_AddI ; } return Op_AddL; } |
| 1844 | Op_IL(Sub)inline int Op_Sub(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1844, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_SubI ; } return Op_SubL; } |
| 1845 | Op_IL(Mul)inline int Op_Mul(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1845, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_MulI ; } return Op_MulL; } |
| 1846 | Op_IL(URShift)inline int Op_URShift(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error( "/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1846, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_URShiftI ; } return Op_URShiftL; } |
| 1847 | Op_IL(LShift)inline int Op_LShift(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error( "/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1847, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_LShiftI ; } return Op_LShiftL; } |
| 1848 | Op_IL(Xor)inline int Op_Xor(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1848, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_XorI ; } return Op_XorL; } |
| 1849 | Op_IL(Cmp)inline int Op_Cmp(BasicType bt) { do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1849, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); if (bt == T_INT) { return Op_CmpI ; } return Op_CmpL; } |
| 1850 | |
| 1851 | inline int Op_Cmp_unsigned(BasicType bt) { |
| 1852 | assert(bt == T_INT || bt == T_LONG, "only for int or longs")do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1852, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); |
| 1853 | if (bt == T_INT) { |
| 1854 | return Op_CmpU; |
| 1855 | } |
| 1856 | return Op_CmpUL; |
| 1857 | } |
| 1858 | |
| 1859 | inline int Op_Cast(BasicType bt) { |
| 1860 | assert(bt == T_INT || bt == T_LONG, "only for int or longs")do { if (!(bt == T_INT || bt == T_LONG)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/node.hpp" , 1860, "assert(" "bt == T_INT || bt == T_LONG" ") failed", "only for int or longs" ); ::breakpoint(); } } while (0); |
| 1861 | if (bt == T_INT) { |
| 1862 | return Op_CastII; |
| 1863 | } |
| 1864 | return Op_CastLL; |
| 1865 | } |
| 1866 | |
| 1867 | #endif // SHARE_OPTO_NODE_HPP |
| 1 | /* |
| 2 | * Copyright (c) 1997, 2021, 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 SHARE_OPTO_REGMASK_HPP |
| 26 | #define SHARE_OPTO_REGMASK_HPP |
| 27 | |
| 28 | #include "code/vmreg.hpp" |
| 29 | #include "opto/optoreg.hpp" |
| 30 | #include "utilities/count_leading_zeros.hpp" |
| 31 | #include "utilities/count_trailing_zeros.hpp" |
| 32 | |
| 33 | class LRG; |
| 34 | |
| 35 | //-------------Non-zero bit search methods used by RegMask--------------------- |
| 36 | // Find lowest 1, undefined if empty/0 |
| 37 | static unsigned int find_lowest_bit(uintptr_t mask) { |
| 38 | return count_trailing_zeros(mask); |
| 39 | } |
| 40 | // Find highest 1, undefined if empty/0 |
| 41 | static unsigned int find_highest_bit(uintptr_t mask) { |
| 42 | return count_leading_zeros(mask) ^ (BitsPerWord - 1U); |
| 43 | } |
| 44 | |
| 45 | //------------------------------RegMask---------------------------------------- |
| 46 | // The ADL file describes how to print the machine-specific registers, as well |
| 47 | // as any notion of register classes. We provide a register mask, which is |
| 48 | // just a collection of Register numbers. |
| 49 | |
| 50 | // The ADLC defines 2 macros, RM_SIZE and FORALL_BODY. |
| 51 | // RM_SIZE is the size of a register mask in 32-bit words. |
| 52 | // FORALL_BODY replicates a BODY macro once per word in the register mask. |
| 53 | // The usage is somewhat clumsy and limited to the regmask.[h,c]pp files. |
| 54 | // However, it means the ADLC can redefine the unroll macro and all loops |
| 55 | // over register masks will be unrolled by the correct amount. |
| 56 | |
| 57 | class RegMask { |
| 58 | |
| 59 | friend class RegMaskIterator; |
| 60 | |
| 61 | // The RM_SIZE is aligned to 64-bit - assert that this holds |
| 62 | LP64_ONLY(STATIC_ASSERT(is_aligned(RM_SIZE, 2)))static_assert((is_aligned(RM_SIZE, 2)), "is_aligned(RM_SIZE, 2)" ); |
| 63 | |
| 64 | static const unsigned int _WordBitMask = BitsPerWord - 1U; |
| 65 | static const unsigned int _LogWordBits = LogBitsPerWord; |
| 66 | static const unsigned int _RM_SIZE = LP64_ONLY(RM_SIZE >> 1)RM_SIZE >> 1 NOT_LP64(RM_SIZE); |
| 67 | static const unsigned int _RM_MAX = _RM_SIZE - 1U; |
| 68 | |
| 69 | union { |
| 70 | // Array of Register Mask bits. This array is large enough to cover |
| 71 | // all the machine registers and all parameters that need to be passed |
| 72 | // on the stack (stack registers) up to some interesting limit. Methods |
| 73 | // that need more parameters will NOT be compiled. On Intel, the limit |
| 74 | // is something like 90+ parameters. |
| 75 | int _RM_I[RM_SIZE]; |
| 76 | uintptr_t _RM_UP[_RM_SIZE]; |
| 77 | }; |
| 78 | |
| 79 | // The low and high water marks represents the lowest and highest word |
| 80 | // that might contain set register mask bits, respectively. We guarantee |
| 81 | // that there are no bits in words outside this range, but any word at |
| 82 | // and between the two marks can still be 0. |
| 83 | unsigned int _lwm; |
| 84 | unsigned int _hwm; |
| 85 | |
| 86 | public: |
| 87 | enum { CHUNK_SIZE = _RM_SIZE * BitsPerWord }; |
| 88 | |
| 89 | // SlotsPerLong is 2, since slots are 32 bits and longs are 64 bits. |
| 90 | // Also, consider the maximum alignment size for a normally allocated |
| 91 | // value. Since we allocate register pairs but not register quads (at |
| 92 | // present), this alignment is SlotsPerLong (== 2). A normally |
| 93 | // aligned allocated register is either a single register, or a pair |
| 94 | // of adjacent registers, the lower-numbered being even. |
| 95 | // See also is_aligned_Pairs() below, and the padding added before |
| 96 | // Matcher::_new_SP to keep allocated pairs aligned properly. |
| 97 | // If we ever go to quad-word allocations, SlotsPerQuad will become |
| 98 | // the controlling alignment constraint. Note that this alignment |
| 99 | // requirement is internal to the allocator, and independent of any |
| 100 | // particular platform. |
| 101 | enum { SlotsPerLong = 2, |
| 102 | SlotsPerVecA = 8, |
| 103 | SlotsPerVecS = 1, |
| 104 | SlotsPerVecD = 2, |
| 105 | SlotsPerVecX = 4, |
| 106 | SlotsPerVecY = 8, |
| 107 | SlotsPerVecZ = 16, |
| 108 | SlotsPerRegVectMask = X86_ONLY(2)2 NOT_X86(1) |
| 109 | }; |
| 110 | |
| 111 | // A constructor only used by the ADLC output. All mask fields are filled |
| 112 | // in directly. Calls to this look something like RM(1,2,3,4); |
| 113 | RegMask( |
| 114 | # define BODY(I) int a##I, |
| 115 | FORALL_BODYBODY(0) BODY(1) BODY(2) BODY(3) BODY(4) BODY(5) BODY(6) BODY( 7) BODY(8) BODY(9) BODY(10) BODY(11) BODY(12) BODY(13) BODY(14 ) BODY(15) BODY(16) BODY(17) BODY(18) BODY(19) BODY(20) BODY( 21) |
| 116 | # undef BODY |
| 117 | int dummy = 0) { |
| 118 | #if defined(VM_LITTLE_ENDIAN1) || !defined(_LP641) |
| 119 | # define BODY(I) _RM_I[I] = a##I; |
| 120 | #else |
| 121 | // We need to swap ints. |
| 122 | # define BODY(I) _RM_I[I ^ 1] = a##I; |
| 123 | #endif |
| 124 | FORALL_BODYBODY(0) BODY(1) BODY(2) BODY(3) BODY(4) BODY(5) BODY(6) BODY( 7) BODY(8) BODY(9) BODY(10) BODY(11) BODY(12) BODY(13) BODY(14 ) BODY(15) BODY(16) BODY(17) BODY(18) BODY(19) BODY(20) BODY( 21) |
| 125 | # undef BODY |
| 126 | _lwm = 0; |
| 127 | _hwm = _RM_MAX; |
| 128 | while (_hwm > 0 && _RM_UP[_hwm] == 0) _hwm--; |
| 129 | while ((_lwm < _hwm) && _RM_UP[_lwm] == 0) _lwm++; |
| 130 | assert(valid_watermarks(), "post-condition")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 130, "assert(" "valid_watermarks()" ") failed", "post-condition" ); ::breakpoint(); } } while (0); |
| 131 | } |
| 132 | |
| 133 | // Handy copying constructor |
| 134 | RegMask(RegMask *rm) { |
| 135 | _hwm = rm->_hwm; |
| 136 | _lwm = rm->_lwm; |
| 137 | for (unsigned i = 0; i < _RM_SIZE; i++) { |
| 138 | _RM_UP[i] = rm->_RM_UP[i]; |
| 139 | } |
| 140 | assert(valid_watermarks(), "post-condition")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 140, "assert(" "valid_watermarks()" ") failed", "post-condition" ); ::breakpoint(); } } while (0); |
| 141 | } |
| 142 | |
| 143 | // Construct an empty mask |
| 144 | RegMask() : _RM_UP(), _lwm(_RM_MAX), _hwm(0) { |
| 145 | assert(valid_watermarks(), "post-condition")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 145, "assert(" "valid_watermarks()" ") failed", "post-condition" ); ::breakpoint(); } } while (0); |
| 146 | } |
| 147 | |
| 148 | // Construct a mask with a single bit |
| 149 | RegMask(OptoReg::Name reg) : RegMask() { |
| 150 | Insert(reg); |
| 151 | } |
| 152 | |
| 153 | // Check for register being in mask |
| 154 | bool Member(OptoReg::Name reg) const { |
| 155 | assert(reg < CHUNK_SIZE, "")do { if (!(reg < CHUNK_SIZE)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 155, "assert(" "reg < CHUNK_SIZE" ") failed", ""); ::breakpoint (); } } while (0); |
| 156 | |
| 157 | unsigned r = (unsigned)reg; |
| 158 | return _RM_UP[r >> _LogWordBits] & (uintptr_t(1) << (r & _WordBitMask)); |
| 159 | } |
| 160 | |
| 161 | // The last bit in the register mask indicates that the mask should repeat |
| 162 | // indefinitely with ONE bits. Returns TRUE if mask is infinite or |
| 163 | // unbounded in size. Returns FALSE if mask is finite size. |
| 164 | bool is_AllStack() const { |
| 165 | return (_RM_UP[_RM_MAX] & (uintptr_t(1) << _WordBitMask)) != 0; |
| 166 | } |
| 167 | |
| 168 | void set_AllStack() { |
| 169 | _RM_UP[_RM_MAX] |= (uintptr_t(1) << _WordBitMask); |
| 170 | } |
| 171 | |
| 172 | // Test for being a not-empty mask. |
| 173 | bool is_NotEmpty() const { |
| 174 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 174, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 175 | uintptr_t tmp = 0; |
| 176 | for (unsigned i = _lwm; i <= _hwm; i++) { |
| 177 | tmp |= _RM_UP[i]; |
| 178 | } |
| 179 | return tmp; |
| 180 | } |
| 181 | |
| 182 | // Find lowest-numbered register from mask, or BAD if mask is empty. |
| 183 | OptoReg::Name find_first_elem() const { |
| 184 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 184, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 185 | for (unsigned i = _lwm; i <= _hwm; i++) { |
| 186 | uintptr_t bits = _RM_UP[i]; |
| 187 | if (bits) { |
| 188 | return OptoReg::Name((i << _LogWordBits) + find_lowest_bit(bits)); |
| 189 | } |
| 190 | } |
| 191 | return OptoReg::Name(OptoReg::Bad); |
| 192 | } |
| 193 | |
| 194 | // Get highest-numbered register from mask, or BAD if mask is empty. |
| 195 | OptoReg::Name find_last_elem() const { |
| 196 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 196, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 197 | // Careful not to overflow if _lwm == 0 |
| 198 | unsigned i = _hwm + 1; |
| 199 | while (i > _lwm) { |
| 200 | uintptr_t bits = _RM_UP[--i]; |
| 201 | if (bits) { |
| 202 | return OptoReg::Name((i << _LogWordBits) + find_highest_bit(bits)); |
| 203 | } |
| 204 | } |
| 205 | return OptoReg::Name(OptoReg::Bad); |
| 206 | } |
| 207 | |
| 208 | // Clear out partial bits; leave only aligned adjacent bit pairs. |
| 209 | void clear_to_pairs(); |
| 210 | |
| 211 | #ifdef ASSERT1 |
| 212 | // Verify watermarks are sane, i.e., within bounds and that no |
| 213 | // register words below or above the watermarks have bits set. |
| 214 | bool valid_watermarks() const { |
| 215 | assert(_hwm < _RM_SIZE, "_hwm out of range: %d", _hwm)do { if (!(_hwm < _RM_SIZE)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 215, "assert(" "_hwm < _RM_SIZE" ") failed", "_hwm out of range: %d" , _hwm); ::breakpoint(); } } while (0); |
| 216 | assert(_lwm < _RM_SIZE, "_lwm out of range: %d", _lwm)do { if (!(_lwm < _RM_SIZE)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 216, "assert(" "_lwm < _RM_SIZE" ") failed", "_lwm out of range: %d" , _lwm); ::breakpoint(); } } while (0); |
| 217 | for (unsigned i = 0; i < _lwm; i++) { |
| 218 | assert(_RM_UP[i] == 0, "_lwm too high: %d regs at: %d", _lwm, i)do { if (!(_RM_UP[i] == 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 218, "assert(" "_RM_UP[i] == 0" ") failed", "_lwm too high: %d regs at: %d" , _lwm, i); ::breakpoint(); } } while (0); |
| 219 | } |
| 220 | for (unsigned i = _hwm + 1; i < _RM_SIZE; i++) { |
| 221 | assert(_RM_UP[i] == 0, "_hwm too low: %d regs at: %d", _hwm, i)do { if (!(_RM_UP[i] == 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 221, "assert(" "_RM_UP[i] == 0" ") failed", "_hwm too low: %d regs at: %d" , _hwm, i); ::breakpoint(); } } while (0); |
| 222 | } |
| 223 | return true; |
| 224 | } |
| 225 | #endif // !ASSERT |
| 226 | |
| 227 | // Test that the mask contains only aligned adjacent bit pairs |
| 228 | bool is_aligned_pairs() const; |
| 229 | |
| 230 | // mask is a pair of misaligned registers |
| 231 | bool is_misaligned_pair() const; |
| 232 | // Test for single register |
| 233 | bool is_bound1() const; |
| 234 | // Test for a single adjacent pair |
| 235 | bool is_bound_pair() const; |
| 236 | // Test for a single adjacent set of ideal register's size. |
| 237 | bool is_bound(uint ireg) const; |
| 238 | |
| 239 | // Check that whether given reg number with size is valid |
| 240 | // for current regmask, where reg is the highest number. |
| 241 | bool is_valid_reg(OptoReg::Name reg, const int size) const; |
| 242 | |
| 243 | // Find the lowest-numbered register set in the mask. Return the |
| 244 | // HIGHEST register number in the set, or BAD if no sets. |
| 245 | // Assert that the mask contains only bit sets. |
| 246 | OptoReg::Name find_first_set(LRG &lrg, const int size) const; |
| 247 | |
| 248 | // Clear out partial bits; leave only aligned adjacent bit sets of size. |
| 249 | void clear_to_sets(const unsigned int size); |
| 250 | // Smear out partial bits to aligned adjacent bit sets. |
| 251 | void smear_to_sets(const unsigned int size); |
| 252 | // Test that the mask contains only aligned adjacent bit sets |
| 253 | bool is_aligned_sets(const unsigned int size) const; |
| 254 | |
| 255 | // Test for a single adjacent set |
| 256 | bool is_bound_set(const unsigned int size) const; |
| 257 | |
| 258 | static bool is_vector(uint ireg); |
| 259 | static int num_registers(uint ireg); |
| 260 | static int num_registers(uint ireg, LRG &lrg); |
| 261 | |
| 262 | // Fast overlap test. Non-zero if any registers in common. |
| 263 | bool overlap(const RegMask &rm) const { |
| 264 | assert(valid_watermarks() && rm.valid_watermarks(), "sanity")do { if (!(valid_watermarks() && rm.valid_watermarks( ))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 264, "assert(" "valid_watermarks() && rm.valid_watermarks()" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 265 | unsigned hwm = MIN2(_hwm, rm._hwm); |
| 266 | unsigned lwm = MAX2(_lwm, rm._lwm); |
| 267 | uintptr_t result = 0; |
| 268 | for (unsigned i = lwm; i <= hwm; i++) { |
| 269 | result |= _RM_UP[i] & rm._RM_UP[i]; |
| 270 | } |
| 271 | return result; |
| 272 | } |
| 273 | |
| 274 | // Special test for register pressure based splitting |
| 275 | // UP means register only, Register plus stack, or stack only is DOWN |
| 276 | bool is_UP() const; |
| 277 | |
| 278 | // Clear a register mask |
| 279 | void Clear() { |
| 280 | _lwm = _RM_MAX; |
| 281 | _hwm = 0; |
| 282 | memset(_RM_UP, 0, sizeof(uintptr_t) * _RM_SIZE); |
| 283 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 283, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 284 | } |
| 285 | |
| 286 | // Fill a register mask with 1's |
| 287 | void Set_All() { |
| 288 | _lwm = 0; |
| 289 | _hwm = _RM_MAX; |
| 290 | memset(_RM_UP, 0xFF, sizeof(uintptr_t) * _RM_SIZE); |
| 291 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 291, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 292 | } |
| 293 | |
| 294 | // Insert register into mask |
| 295 | void Insert(OptoReg::Name reg) { |
| 296 | assert(reg != OptoReg::Bad, "sanity")do { if (!(reg != OptoReg::Bad)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 296, "assert(" "reg != OptoReg::Bad" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 297 | assert(reg != OptoReg::Special, "sanity")do { if (!(reg != OptoReg::Special)) { (*g_assert_poison) = 'X' ;; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 297, "assert(" "reg != OptoReg::Special" ") failed", "sanity" ); ::breakpoint(); } } while (0); |
| 298 | assert(reg < CHUNK_SIZE, "sanity")do { if (!(reg < CHUNK_SIZE)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 298, "assert(" "reg < CHUNK_SIZE" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 299 | assert(valid_watermarks(), "pre-condition")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 299, "assert(" "valid_watermarks()" ") failed", "pre-condition" ); ::breakpoint(); } } while (0); |
| 300 | unsigned r = (unsigned)reg; |
| 301 | unsigned index = r >> _LogWordBits; |
| 302 | if (index > _hwm) _hwm = index; |
| 303 | if (index < _lwm) _lwm = index; |
| 304 | _RM_UP[index] |= (uintptr_t(1) << (r & _WordBitMask)); |
| 305 | assert(valid_watermarks(), "post-condition")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 305, "assert(" "valid_watermarks()" ") failed", "post-condition" ); ::breakpoint(); } } while (0); |
| 306 | } |
| 307 | |
| 308 | // Remove register from mask |
| 309 | void Remove(OptoReg::Name reg) { |
| 310 | assert(reg < CHUNK_SIZE, "")do { if (!(reg < CHUNK_SIZE)) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 310, "assert(" "reg < CHUNK_SIZE" ") failed", ""); ::breakpoint (); } } while (0); |
| 311 | unsigned r = (unsigned)reg; |
| 312 | _RM_UP[r >> _LogWordBits] &= ~(uintptr_t(1) << (r & _WordBitMask)); |
| 313 | } |
| 314 | |
| 315 | // OR 'rm' into 'this' |
| 316 | void OR(const RegMask &rm) { |
| 317 | assert(valid_watermarks() && rm.valid_watermarks(), "sanity")do { if (!(valid_watermarks() && rm.valid_watermarks( ))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 317, "assert(" "valid_watermarks() && rm.valid_watermarks()" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 318 | // OR widens the live range |
| 319 | if (_lwm > rm._lwm) _lwm = rm._lwm; |
| 320 | if (_hwm < rm._hwm) _hwm = rm._hwm; |
| 321 | for (unsigned i = _lwm; i <= _hwm; i++) { |
| 322 | _RM_UP[i] |= rm._RM_UP[i]; |
| 323 | } |
| 324 | assert(valid_watermarks(), "sanity")do { if (!(valid_watermarks())) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 324, "assert(" "valid_watermarks()" ") failed", "sanity"); :: breakpoint(); } } while (0); |
| 325 | } |
| 326 | |
| 327 | // AND 'rm' into 'this' |
| 328 | void AND(const RegMask &rm) { |
| 329 | assert(valid_watermarks() && rm.valid_watermarks(), "sanity")do { if (!(valid_watermarks() && rm.valid_watermarks( ))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 329, "assert(" "valid_watermarks() && rm.valid_watermarks()" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 330 | // Do not evaluate words outside the current watermark range, as they are |
| 331 | // already zero and an &= would not change that |
| 332 | for (unsigned i = _lwm; i <= _hwm; i++) { |
| 333 | _RM_UP[i] &= rm._RM_UP[i]; |
| 334 | } |
| 335 | // Narrow the watermarks if &rm spans a narrower range. |
| 336 | // Update after to ensure non-overlapping words are zeroed out. |
| 337 | if (_lwm < rm._lwm) _lwm = rm._lwm; |
| 338 | if (_hwm > rm._hwm) _hwm = rm._hwm; |
| 339 | } |
| 340 | |
| 341 | // Subtract 'rm' from 'this' |
| 342 | void SUBTRACT(const RegMask &rm) { |
| 343 | assert(valid_watermarks() && rm.valid_watermarks(), "sanity")do { if (!(valid_watermarks() && rm.valid_watermarks( ))) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 343, "assert(" "valid_watermarks() && rm.valid_watermarks()" ") failed", "sanity"); ::breakpoint(); } } while (0); |
| 344 | unsigned hwm = MIN2(_hwm, rm._hwm); |
| 345 | unsigned lwm = MAX2(_lwm, rm._lwm); |
| 346 | for (unsigned i = lwm; i <= hwm; i++) { |
| 347 | _RM_UP[i] &= ~rm._RM_UP[i]; |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | // Compute size of register mask: number of bits |
| 352 | uint Size() const; |
| 353 | |
| 354 | #ifndef PRODUCT |
| 355 | void print() const { dump(); } |
| 356 | void dump(outputStream *st = tty) const; // Print a mask |
| 357 | #endif |
| 358 | |
| 359 | static const RegMask Empty; // Common empty mask |
| 360 | static const RegMask All; // Common all mask |
| 361 | |
| 362 | static bool can_represent(OptoReg::Name reg) { |
| 363 | // NOTE: -1 in computation reflects the usage of the last |
| 364 | // bit of the regmask as an infinite stack flag and |
| 365 | // -7 is to keep mask aligned for largest value (VecZ). |
| 366 | return (int)reg < (int)(CHUNK_SIZE - 1); |
| 367 | } |
| 368 | static bool can_represent_arg(OptoReg::Name reg) { |
| 369 | // NOTE: -SlotsPerVecZ in computation reflects the need |
| 370 | // to keep mask aligned for largest value (VecZ). |
| 371 | return (int)reg < (int)(CHUNK_SIZE - SlotsPerVecZ); |
| 372 | } |
| 373 | }; |
| 374 | |
| 375 | class RegMaskIterator { |
| 376 | private: |
| 377 | uintptr_t _current_bits; |
| 378 | unsigned int _next_index; |
| 379 | OptoReg::Name _reg; |
| 380 | const RegMask& _rm; |
| 381 | public: |
| 382 | RegMaskIterator(const RegMask& rm) : _current_bits(0), _next_index(rm._lwm), _reg(OptoReg::Bad), _rm(rm) { |
| 383 | // Calculate the first element |
| 384 | next(); |
| 385 | } |
| 386 | |
| 387 | bool has_next() { |
| 388 | return _reg != OptoReg::Bad; |
| 389 | } |
| 390 | |
| 391 | // Get the current element and calculate the next |
| 392 | OptoReg::Name next() { |
| 393 | OptoReg::Name r = _reg; |
| 394 | |
| 395 | // This bit shift scheme, borrowed from IndexSetIterator, |
| 396 | // shifts the _current_bits down by the number of trailing |
| 397 | // zeros - which leaves the "current" bit on position zero, |
| 398 | // then subtracts by 1 to clear it. This quirk avoids the |
| 399 | // undefined behavior that could arise if trying to shift |
| 400 | // away the bit with a single >> (next_bit + 1) shift when |
| 401 | // next_bit is 31/63. It also keeps number of shifts and |
| 402 | // arithmetic ops to a minimum. |
| 403 | |
| 404 | // We have previously found bits at _next_index - 1, and |
| 405 | // still have some left at the same index. |
| 406 | if (_current_bits != 0) { |
| 407 | unsigned int next_bit = find_lowest_bit(_current_bits); |
| 408 | assert(_reg != OptoReg::Bad, "can't be in a bad state")do { if (!(_reg != OptoReg::Bad)) { (*g_assert_poison) = 'X'; ; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 408, "assert(" "_reg != OptoReg::Bad" ") failed", "can't be in a bad state" ); ::breakpoint(); } } while (0); |
| 409 | assert(next_bit > 0, "must be")do { if (!(next_bit > 0)) { (*g_assert_poison) = 'X';; report_vm_error ("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 409, "assert(" "next_bit > 0" ") failed", "must be"); :: breakpoint(); } } while (0); |
| 410 | assert(((_current_bits >> next_bit) & 0x1) == 1, "lowest bit must be set after shift")do { if (!(((_current_bits >> next_bit) & 0x1) == 1 )) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 410, "assert(" "((_current_bits >> next_bit) & 0x1) == 1" ") failed", "lowest bit must be set after shift"); ::breakpoint (); } } while (0); |
| 411 | _current_bits = (_current_bits >> next_bit) - 1; |
| 412 | _reg = OptoReg::add(_reg, next_bit); |
| 413 | return r; |
| 414 | } |
| 415 | |
| 416 | // Find the next word with bits |
| 417 | while (_next_index <= _rm._hwm) { |
| 418 | _current_bits = _rm._RM_UP[_next_index++]; |
| 419 | if (_current_bits != 0) { |
| 420 | // Found a word. Calculate the first register element and |
| 421 | // prepare _current_bits by shifting it down and clearing |
| 422 | // the lowest bit |
| 423 | unsigned int next_bit = find_lowest_bit(_current_bits); |
| 424 | assert(((_current_bits >> next_bit) & 0x1) == 1, "lowest bit must be set after shift")do { if (!(((_current_bits >> next_bit) & 0x1) == 1 )) { (*g_assert_poison) = 'X';; report_vm_error("/home/daniel/Projects/java/jdk/src/hotspot/share/opto/regmask.hpp" , 424, "assert(" "((_current_bits >> next_bit) & 0x1) == 1" ") failed", "lowest bit must be set after shift"); ::breakpoint (); } } while (0); |
| 425 | _current_bits = (_current_bits >> next_bit) - 1; |
| 426 | _reg = OptoReg::Name(((_next_index - 1) << RegMask::_LogWordBits) + next_bit); |
| 427 | return r; |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | // No more bits |
| 432 | _reg = OptoReg::Name(OptoReg::Bad); |
| 433 | return r; |
| 434 | } |
| 435 | }; |
| 436 | |
| 437 | // Do not use this constant directly in client code! |
| 438 | #undef RM_SIZE |
| 439 | |
| 440 | #endif // SHARE_OPTO_REGMASK_HPP |