/* * Copyright (c) 2000, 2009, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include "incls/_precompiled.incl" #include "incls/_reg_split.cpp.incl" //------------------------------Split-------------------------------------- // Walk the graph in RPO and for each lrg which spills, propagate reaching // definitions. During propagation, split the live range around regions of // High Register Pressure (HRP). If a Def is in a region of Low Register // Pressure (LRP), it will not get spilled until we encounter a region of // HRP between it and one of its uses. We will spill at the transition // point between LRP and HRP. Uses in the HRP region will use the spilled // Def. The first Use outside the HRP region will generate a SpillCopy to // hoist the live range back up into a register, and all subsequent uses // will use that new Def until another HRP region is encountered. Defs in // HRP regions will get trailing SpillCopies to push the LRG down into the // stack immediately. // // As a side effect, unlink from (hence make dead) coalesced copies. // static const char out_of_nodes[] = "out of nodes during split"; //------------------------------get_spillcopy_wide----------------------------- // Get a SpillCopy node with wide-enough masks. Use the 'wide-mask', the // wide ideal-register spill-mask if possible. If the 'wide-mask' does // not cover the input (or output), use the input (or output) mask instead. Node *PhaseChaitin::get_spillcopy_wide( Node *def, Node *use, uint uidx ) { // If ideal reg doesn't exist we've got a bad schedule happening // that is forcing us to spill something that isn't spillable. // Bail rather than abort int ireg = def->ideal_reg(); if( ireg == 0 || ireg == Op_RegFlags ) { assert(false, "attempted to spill a non-spillable item"); C->record_method_not_compilable("attempted to spill a non-spillable item"); return NULL; } if (C->check_node_count(NodeLimitFudgeFactor, out_of_nodes)) { return NULL; } const RegMask *i_mask = &def->out_RegMask(); const RegMask *w_mask = C->matcher()->idealreg2spillmask[ireg]; const RegMask *o_mask = use ? &use->in_RegMask(uidx) : w_mask; const RegMask *w_i_mask = w_mask->overlap( *i_mask ) ? w_mask : i_mask; const RegMask *w_o_mask; if( w_mask->overlap( *o_mask ) && // Overlap AND ((ireg != Op_RegL && ireg != Op_RegD // Single use or aligned #ifdef _LP64 && ireg != Op_RegP #endif ) || o_mask->is_aligned_Pairs()) ) { // Don't come here for mis-aligned doubles w_o_mask = w_mask; } else { // wide ideal mask does not overlap with o_mask // Mis-aligned doubles come here and XMM->FPR moves on x86. w_o_mask = o_mask; // Must target desired registers // Does the ideal-reg-mask overlap with o_mask? I.e., can I use // a reg-reg move or do I need a trip across register classes // (and thus through memory)? if( !C->matcher()->idealreg2regmask[ireg]->overlap( *o_mask) && o_mask->is_UP() ) // Here we assume a trip through memory is required. w_i_mask = &C->FIRST_STACK_mask(); } return new (C) MachSpillCopyNode( def, *w_i_mask, *w_o_mask ); } //------------------------------insert_proj------------------------------------ // Insert the spill at chosen location. Skip over any intervening Proj's or // Phis. Skip over a CatchNode and projs, inserting in the fall-through block // instead. Update high-pressure indices. Create a new live range. void PhaseChaitin::insert_proj( Block *b, uint i, Node *spill, uint maxlrg ) { // Skip intervening ProjNodes. Do not insert between a ProjNode and // its definer. while( i < b->_nodes.size() && (b->_nodes[i]->is_Proj() || b->_nodes[i]->is_Phi() ) ) i++; // Do not insert between a call and his Catch if( b->_nodes[i]->is_Catch() ) { // Put the instruction at the top of the fall-thru block. // Find the fall-thru projection while( 1 ) { const CatchProjNode *cp = b->_nodes[++i]->as_CatchProj(); if( cp->_con == CatchProjNode::fall_through_index ) break; } int sidx = i - b->end_idx()-1; b = b->_succs[sidx]; // Switch to successor block i = 1; // Right at start of block } b->_nodes.insert(i,spill); // Insert node in block _cfg._bbs.map(spill->_idx,b); // Update node->block mapping to reflect // Adjust the point where we go hi-pressure if( i <= b->_ihrp_index ) b->_ihrp_index++; if( i <= b->_fhrp_index ) b->_fhrp_index++; // Assign a new Live Range Number to the SpillCopy and grow // the node->live range mapping. new_lrg(spill,maxlrg); } //------------------------------split_DEF-------------------------------------- // There are four categories of Split; UP/DOWN x DEF/USE // Only three of these really occur as DOWN/USE will always color // Any Split with a DEF cannot CISC-Spill now. Thus we need // two helper routines, one for Split DEFS (insert after instruction), // one for Split USES (insert before instruction). DEF insertion // happens inside Split, where the Leaveblock array is updated. uint PhaseChaitin::split_DEF( Node *def, Block *b, int loc, uint maxlrg, Node **Reachblock, Node **debug_defs, GrowableArray splits, int slidx ) { #ifdef ASSERT // Increment the counter for this lrg splits.at_put(slidx, splits.at(slidx)+1); #endif // If we are spilling the memory op for an implicit null check, at the // null check location (ie - null check is in HRP block) we need to do // the null-check first, then spill-down in the following block. // (The implicit_null_check function ensures the use is also dominated // by the branch-not-taken block.) Node *be = b->end(); if( be->is_MachNullCheck() && be->in(1) == def && def == b->_nodes[loc] ) { // Spill goes in the branch-not-taken block b = b->_succs[b->_nodes[b->end_idx()+1]->Opcode() == Op_IfTrue]; loc = 0; // Just past the Region } assert( loc >= 0, "must insert past block head" ); // Get a def-side SpillCopy Node *spill = get_spillcopy_wide(def,NULL,0); // Did we fail to split?, then bail if (!spill) { return 0; } // Insert the spill at chosen location insert_proj( b, loc+1, spill, maxlrg++); // Insert new node into Reaches array Reachblock[slidx] = spill; // Update debug list of reaching down definitions by adding this one debug_defs[slidx] = spill; // return updated count of live ranges return maxlrg; } //------------------------------split_USE-------------------------------------- // Splits at uses can involve redeffing the LRG, so no CISC Spilling there. // Debug uses want to know if def is already stack enabled. uint PhaseChaitin::split_USE( Node *def, Block *b, Node *use, uint useidx, uint maxlrg, bool def_down, bool cisc_sp, GrowableArray splits, int slidx ) { #ifdef ASSERT // Increment the counter for this lrg splits.at_put(slidx, splits.at(slidx)+1); #endif // Some setup stuff for handling debug node uses JVMState* jvms = use->jvms(); uint debug_start = jvms ? jvms->debug_start() : 999999; uint debug_end = jvms ? jvms->debug_end() : 999999; //------------------------------------------- // Check for use of debug info if (useidx >= debug_start && useidx < debug_end) { // Actually it's perfectly legal for constant debug info to appear // just unlikely. In this case the optimizer left a ConI of a 4 // as both inputs to a Phi with only a debug use. It's a single-def // live range of a rematerializable value. The live range spills, // rematerializes and now the ConI directly feeds into the debug info. // assert(!def->is_Con(), "constant debug info already constructed directly"); // Special split handling for Debug Info // If DEF is DOWN, just hook the edge and return // If DEF is UP, Split it DOWN for this USE. if( def->is_Mach() ) { if( def_down ) { // DEF is DOWN, so connect USE directly to the DEF use->set_req(useidx, def); } else { // Block and index where the use occurs. Block *b = _cfg._bbs[use->_idx]; // Put the clone just prior to use int bindex = b->find_node(use); // DEF is UP, so must copy it DOWN and hook in USE // Insert SpillCopy before the USE, which uses DEF as its input, // and defs a new live range, which is used by this node. Node *spill = get_spillcopy_wide(def,use,useidx); // did we fail to split? if (!spill) { // Bail return 0; } // insert into basic block insert_proj( b, bindex, spill, maxlrg++ ); // Use the new split use->set_req(useidx,spill); } // No further split handling needed for this use return maxlrg; } // End special splitting for debug info live range } // If debug info // CISC-SPILLING // Finally, check to see if USE is CISC-Spillable, and if so, // gather_lrg_masks will add the flags bit to its mask, and // no use side copy is needed. This frees up the live range // register choices without causing copy coalescing, etc. if( UseCISCSpill && cisc_sp ) { int inp = use->cisc_operand(); if( inp != AdlcVMDeps::Not_cisc_spillable ) // Convert operand number to edge index number inp = use->as_Mach()->operand_index(inp); if( inp == (int)useidx ) { use->set_req(useidx, def); #ifndef PRODUCT if( TraceCISCSpill ) { tty->print(" set_split: "); use->dump(); } #endif return maxlrg; } } //------------------------------------------- // Insert a Copy before the use // Block and index where the use occurs. int bindex; // Phi input spill-copys belong at the end of the prior block if( use->is_Phi() ) { b = _cfg._bbs[b->pred(useidx)->_idx]; bindex = b->end_idx(); } else { // Put the clone just prior to use bindex = b->find_node(use); } Node *spill = get_spillcopy_wide( def, use, useidx ); if( !spill ) return 0; // Bailed out // Insert SpillCopy before the USE, which uses the reaching DEF as // its input, and defs a new live range, which is used by this node. insert_proj( b, bindex, spill, maxlrg++ ); // Use the spill/clone use->set_req(useidx,spill); // return updated live range count return maxlrg; } //------------------------------split_Rematerialize---------------------------- // Clone a local copy of the def. Node *PhaseChaitin::split_Rematerialize( Node *def, Block *b, uint insidx, uint &maxlrg, GrowableArray splits, int slidx, uint *lrg2reach, Node **Reachblock, bool walkThru ) { // The input live ranges will be stretched to the site of the new // instruction. They might be stretched past a def and will thus // have the old and new values of the same live range alive at the // same time - a definite no-no. Split out private copies of // the inputs. if( def->req() > 1 ) { for( uint i = 1; i < def->req(); i++ ) { Node *in = def->in(i); // Check for single-def (LRG cannot redefined) uint lidx = n2lidx(in); if( lidx >= _maxlrg ) continue; // Value is a recent spill-copy if (lrgs(lidx).is_singledef()) continue; Block *b_def = _cfg._bbs[def->_idx]; int idx_def = b_def->find_node(def); Node *in_spill = get_spillcopy_wide( in, def, i ); if( !in_spill ) return 0; // Bailed out insert_proj(b_def,idx_def,in_spill,maxlrg++); if( b_def == b ) insidx++; def->set_req(i,in_spill); } } Node *spill = def->clone(); if (C->check_node_count(NodeLimitFudgeFactor, out_of_nodes)) { // Check when generating nodes return 0; } // See if any inputs are currently being spilled, and take the // latest copy of spilled inputs. if( spill->req() > 1 ) { for( uint i = 1; i < spill->req(); i++ ) { Node *in = spill->in(i); uint lidx = Find_id(in); // Walk backwards thru spill copy node intermediates if (walkThru) { while ( in->is_SpillCopy() && lidx >= _maxlrg ) { in = in->in(1); lidx = Find_id(in); } if (lidx < _maxlrg && lrgs(lidx).is_multidef()) { // walkThru found a multidef LRG, which is unsafe to use, so // just keep the original def used in the clone. in = spill->in(i); lidx = Find_id(in); } } if( lidx < _maxlrg && lrgs(lidx).reg() >= LRG::SPILL_REG ) { Node *rdef = Reachblock[lrg2reach[lidx]]; if( rdef ) spill->set_req(i,rdef); } } } assert( spill->out_RegMask().is_UP(), "rematerialize to a reg" ); // Rematerialized op is def->spilled+1 set_was_spilled(spill); if( _spilled_once.test(def->_idx) ) set_was_spilled(spill); insert_proj( b, insidx, spill, maxlrg++ ); #ifdef ASSERT // Increment the counter for this lrg splits.at_put(slidx, splits.at(slidx)+1); #endif // See if the cloned def kills any flags, and copy those kills as well uint i = insidx+1; if( clone_projs( b, i, def, spill, maxlrg ) ) { // Adjust the point where we go hi-pressure if( i <= b->_ihrp_index ) b->_ihrp_index++; if( i <= b->_fhrp_index ) b->_fhrp_index++; } return spill; } //------------------------------is_high_pressure------------------------------- // Function to compute whether or not this live range is "high pressure" // in this block - whether it spills eagerly or not. bool PhaseChaitin::is_high_pressure( Block *b, LRG *lrg, uint insidx ) { if( lrg->_was_spilled1 ) return true; // Forced spilling due to conflict? Then split only at binding uses // or defs, not for supposed capacity problems. // CNC - Turned off 7/8/99, causes too much spilling // if( lrg->_is_bound ) return false; // Not yet reached the high-pressure cutoff point, so low pressure uint hrp_idx = lrg->_is_float ? b->_fhrp_index : b->_ihrp_index; if( insidx < hrp_idx ) return false; // Register pressure for the block as a whole depends on reg class int block_pres = lrg->_is_float ? b->_freg_pressure : b->_reg_pressure; // Bound live ranges will split at the binding points first; // Intermediate splits should assume the live range's register set // got "freed up" and that num_regs will become INT_PRESSURE. int bound_pres = lrg->_is_float ? FLOATPRESSURE : INTPRESSURE; // Effective register pressure limit. int lrg_pres = (lrg->get_invalid_mask_size() > lrg->num_regs()) ? (lrg->get_invalid_mask_size() >> (lrg->num_regs()-1)) : bound_pres; // High pressure if block pressure requires more register freedom // than live range has. return block_pres >= lrg_pres; } //------------------------------prompt_use--------------------------------- // True if lidx is used before any real register is def'd in the block bool PhaseChaitin::prompt_use( Block *b, uint lidx ) { if( lrgs(lidx)._was_spilled2 ) return false; // Scan block for 1st use. for( uint i = 1; i <= b->end_idx(); i++ ) { Node *n = b->_nodes[i]; // Ignore PHI use, these can be up or down if( n->is_Phi() ) continue; for( uint j = 1; j < n->req(); j++ ) if( Find_id(n->in(j)) == lidx ) return true; // Found 1st use! if( n->out_RegMask().is_NotEmpty() ) return false; } return false; } //------------------------------Split-------------------------------------- //----------Split Routine---------- // ***** NEW SPLITTING HEURISTIC ***** // DEFS: If the DEF is in a High Register Pressure(HRP) Block, split there. // Else, no split unless there is a HRP block between a DEF and // one of its uses, and then split at the HRP block. // // USES: If USE is in HRP, split at use to leave main LRG on stack. // Else, hoist LRG back up to register only (ie - split is also DEF) // We will compute a new maxlrg as we go uint PhaseChaitin::Split( uint maxlrg ) { NOT_PRODUCT( Compile::TracePhase t3("regAllocSplit", &_t_regAllocSplit, TimeCompiler); ) uint bidx, pidx, slidx, insidx, inpidx, twoidx; uint non_phi = 1, spill_cnt = 0; Node **Reachblock; Node *n1, *n2, *n3; Node_List *defs,*phis; bool *UPblock; bool u1, u2, u3; Block *b, *pred; PhiNode *phi; GrowableArray lidxs; // Array of counters to count splits per live range GrowableArray splits; //----------Setup Code---------- // Create a convenient mapping from lrg numbers to reaches/leaves indices uint *lrg2reach = NEW_RESOURCE_ARRAY( uint, _maxlrg ); // Keep track of DEFS & Phis for later passes defs = new Node_List(); phis = new Node_List(); // Gather info on which LRG's are spilling, and build maps for( bidx = 1; bidx < _maxlrg; bidx++ ) { if( lrgs(bidx).alive() && lrgs(bidx).reg() >= LRG::SPILL_REG ) { assert(!lrgs(bidx).mask().is_AllStack(),"AllStack should color"); lrg2reach[bidx] = spill_cnt; spill_cnt++; lidxs.append(bidx); #ifdef ASSERT // Initialize the split counts to zero splits.append(0); #endif #ifndef PRODUCT if( PrintOpto && WizardMode && lrgs(bidx)._was_spilled1 ) tty->print_cr("Warning, 2nd spill of L%d",bidx); #endif } } // Create side arrays for propagating reaching defs info. // Each block needs a node pointer for each spilling live range for the // Def which is live into the block. Phi nodes handle multiple input // Defs by querying the output of their predecessor blocks and resolving // them to a single Def at the phi. The pointer is updated for each // Def in the block, and then becomes the output for the block when // processing of the block is complete. We also need to track whether // a Def is UP or DOWN. UP means that it should get a register (ie - // it is always in LRP regions), and DOWN means that it is probably // on the stack (ie - it crosses HRP regions). Node ***Reaches = NEW_RESOURCE_ARRAY( Node**, _cfg._num_blocks+1 ); bool **UP = NEW_RESOURCE_ARRAY( bool*, _cfg._num_blocks+1 ); Node **debug_defs = NEW_RESOURCE_ARRAY( Node*, spill_cnt ); VectorSet **UP_entry= NEW_RESOURCE_ARRAY( VectorSet*, spill_cnt ); // Initialize Reaches & UP for( bidx = 0; bidx < _cfg._num_blocks+1; bidx++ ) { Reaches[bidx] = NEW_RESOURCE_ARRAY( Node*, spill_cnt ); UP[bidx] = NEW_RESOURCE_ARRAY( bool, spill_cnt ); Node **Reachblock = Reaches[bidx]; bool *UPblock = UP[bidx]; for( slidx = 0; slidx < spill_cnt; slidx++ ) { UPblock[slidx] = true; // Assume they start in registers Reachblock[slidx] = NULL; // Assume that no def is present } } // Initialize to array of empty vectorsets for( slidx = 0; slidx < spill_cnt; slidx++ ) UP_entry[slidx] = new VectorSet(Thread::current()->resource_area()); //----------PASS 1---------- //----------Propagation & Node Insertion Code---------- // Walk the Blocks in RPO for DEF & USE info for( bidx = 0; bidx < _cfg._num_blocks; bidx++ ) { if (C->check_node_count(spill_cnt, out_of_nodes)) { return 0; } b = _cfg._blocks[bidx]; // Reaches & UP arrays for this block Reachblock = Reaches[b->_pre_order]; UPblock = UP[b->_pre_order]; // Reset counter of start of non-Phi nodes in block non_phi = 1; //----------Block Entry Handling---------- // Check for need to insert a new phi // Cycle through this block's predecessors, collecting Reaches // info for each spilled LRG. If they are identical, no phi is // needed. If they differ, check for a phi, and insert if missing, // or update edges if present. Set current block's Reaches set to // be either the phi's or the reaching def, as appropriate. // If no Phi is needed, check if the LRG needs to spill on entry // to the block due to HRP. for( slidx = 0; slidx < spill_cnt; slidx++ ) { // Grab the live range number uint lidx = lidxs.at(slidx); // Do not bother splitting or putting in Phis for single-def // rematerialized live ranges. This happens alot to constants // with long live ranges. if( lrgs(lidx).is_singledef() && lrgs(lidx)._def->rematerialize() ) { // reset the Reaches & UP entries Reachblock[slidx] = lrgs(lidx)._def; UPblock[slidx] = true; // Record following instruction in case 'n' rematerializes and // kills flags Block *pred1 = _cfg._bbs[b->pred(1)->_idx]; continue; } // Initialize needs_phi and needs_split bool needs_phi = false; bool needs_split = false; bool has_phi = false; // Walk the predecessor blocks to check inputs for that live range // Grab predecessor block header n1 = b->pred(1); // Grab the appropriate reaching def info for inpidx pred = _cfg._bbs[n1->_idx]; pidx = pred->_pre_order; Node **Ltmp = Reaches[pidx]; bool *Utmp = UP[pidx]; n1 = Ltmp[slidx]; u1 = Utmp[slidx]; // Initialize node for saving type info n3 = n1; u3 = u1; // Compare inputs to see if a Phi is needed for( inpidx = 2; inpidx < b->num_preds(); inpidx++ ) { // Grab predecessor block headers n2 = b->pred(inpidx); // Grab the appropriate reaching def info for inpidx pred = _cfg._bbs[n2->_idx]; pidx = pred->_pre_order; Ltmp = Reaches[pidx]; Utmp = UP[pidx]; n2 = Ltmp[slidx]; u2 = Utmp[slidx]; // For each LRG, decide if a phi is necessary if( n1 != n2 ) { needs_phi = true; } // See if the phi has mismatched inputs, UP vs. DOWN if( n1 && n2 && (u1 != u2) ) { needs_split = true; } // Move n2/u2 to n1/u1 for next iteration n1 = n2; u1 = u2; // Preserve a non-NULL predecessor for later type referencing if( (n3 == NULL) && (n2 != NULL) ){ n3 = n2; u3 = u2; } } // End for all potential Phi inputs // check block for appropriate phinode & update edges for( insidx = 1; insidx <= b->end_idx(); insidx++ ) { n1 = b->_nodes[insidx]; // bail if this is not a phi phi = n1->is_Phi() ? n1->as_Phi() : NULL; if( phi == NULL ) { // Keep track of index of first non-PhiNode instruction in block non_phi = insidx; // break out of the for loop as we have handled all phi nodes break; } // must be looking at a phi if( Find_id(n1) == lidxs.at(slidx) ) { // found the necessary phi needs_phi = false; has_phi = true; // initialize the Reaches entry for this LRG Reachblock[slidx] = phi; break; } // end if found correct phi } // end for all phi's // If a phi is needed or exist, check for it if( needs_phi || has_phi ) { // add new phinode if one not already found if( needs_phi ) { // create a new phi node and insert it into the block // type is taken from left over pointer to a predecessor assert(n3,"No non-NULL reaching DEF for a Phi"); phi = new (C, b->num_preds()) PhiNode(b->head(), n3->bottom_type()); // initialize the Reaches entry for this LRG Reachblock[slidx] = phi; // add node to block & node_to_block mapping insert_proj( b, insidx++, phi, maxlrg++ ); non_phi++; // Reset new phi's mapping to be the spilling live range _names.map(phi->_idx, lidx); assert(Find_id(phi) == lidx,"Bad update on Union-Find mapping"); } // end if not found correct phi // Here you have either found or created the Phi, so record it assert(phi != NULL,"Must have a Phi Node here"); phis->push(phi); // PhiNodes should either force the LRG UP or DOWN depending // on its inputs and the register pressure in the Phi's block. UPblock[slidx] = true; // Assume new DEF is UP // If entering a high-pressure area with no immediate use, // assume Phi is DOWN if( is_high_pressure( b, &lrgs(lidx), b->end_idx()) && !prompt_use(b,lidx) ) UPblock[slidx] = false; // If we are not split up/down and all inputs are down, then we // are down if( !needs_split && !u3 ) UPblock[slidx] = false; } // end if phi is needed // Do not need a phi, so grab the reaching DEF else { // Grab predecessor block header n1 = b->pred(1); // Grab the appropriate reaching def info for k pred = _cfg._bbs[n1->_idx]; pidx = pred->_pre_order; Node **Ltmp = Reaches[pidx]; bool *Utmp = UP[pidx]; // reset the Reaches & UP entries Reachblock[slidx] = Ltmp[slidx]; UPblock[slidx] = Utmp[slidx]; } // end else no Phi is needed } // end for all spilling live ranges // DEBUG #ifndef PRODUCT if(trace_spilling()) { tty->print("/`\nBlock %d: ", b->_pre_order); tty->print("Reaching Definitions after Phi handling\n"); for( uint x = 0; x < spill_cnt; x++ ) { tty->print("Spill Idx %d: UP %d: Node\n",x,UPblock[x]); if( Reachblock[x] ) Reachblock[x]->dump(); else tty->print("Undefined\n"); } } #endif //----------Non-Phi Node Splitting---------- // Since phi-nodes have now been handled, the Reachblock array for this // block is initialized with the correct starting value for the defs which // reach non-phi instructions in this block. Thus, process non-phi // instructions normally, inserting SpillCopy nodes for all spill // locations. // Memoize any DOWN reaching definitions for use as DEBUG info for( insidx = 0; insidx < spill_cnt; insidx++ ) { debug_defs[insidx] = (UPblock[insidx]) ? NULL : Reachblock[insidx]; if( UPblock[insidx] ) // Memoize UP decision at block start UP_entry[insidx]->set( b->_pre_order ); } //----------Walk Instructions in the Block and Split---------- // For all non-phi instructions in the block for( insidx = 1; insidx <= b->end_idx(); insidx++ ) { Node *n = b->_nodes[insidx]; // Find the defining Node's live range index uint defidx = Find_id(n); uint cnt = n->req(); if( n->is_Phi() ) { // Skip phi nodes after removing dead copies. if( defidx < _maxlrg ) { // Check for useless Phis. These appear if we spill, then // coalesce away copies. Dont touch Phis in spilling live // ranges; they are busy getting modifed in this pass. if( lrgs(defidx).reg() < LRG::SPILL_REG ) { uint i; Node *u = NULL; // Look for the Phi merging 2 unique inputs for( i = 1; i < cnt; i++ ) { // Ignore repeats and self if( n->in(i) != u && n->in(i) != n ) { // Found a unique input if( u != NULL ) // If it's the 2nd, bail out break; u = n->in(i); // Else record it } } assert( u, "at least 1 valid input expected" ); if( i >= cnt ) { // Found one unique input assert(Find_id(n) == Find_id(u), "should be the same lrg"); n->replace_by(u); // Then replace with unique input n->disconnect_inputs(NULL); b->_nodes.remove(insidx); insidx--; b->_ihrp_index--; b->_fhrp_index--; } } } continue; } assert( insidx > b->_ihrp_index || (b->_reg_pressure < (uint)INTPRESSURE) || b->_ihrp_index > 4000000 || b->_ihrp_index >= b->end_idx() || !b->_nodes[b->_ihrp_index]->is_Proj(), "" ); assert( insidx > b->_fhrp_index || (b->_freg_pressure < (uint)FLOATPRESSURE) || b->_fhrp_index > 4000000 || b->_fhrp_index >= b->end_idx() || !b->_nodes[b->_fhrp_index]->is_Proj(), "" ); // ********** Handle Crossing HRP Boundry ********** if( (insidx == b->_ihrp_index) || (insidx == b->_fhrp_index) ) { for( slidx = 0; slidx < spill_cnt; slidx++ ) { // Check for need to split at HRP boundary - split if UP n1 = Reachblock[slidx]; // bail out if no reaching DEF if( n1 == NULL ) continue; // bail out if live range is 'isolated' around inner loop uint lidx = lidxs.at(slidx); // If live range is currently UP if( UPblock[slidx] ) { // set location to insert spills at // SPLIT DOWN HERE - NO CISC SPILL if( is_high_pressure( b, &lrgs(lidx), insidx ) && !n1->rematerialize() ) { // If there is already a valid stack definition available, use it if( debug_defs[slidx] != NULL ) { Reachblock[slidx] = debug_defs[slidx]; } else { // Insert point is just past last use or def in the block int insert_point = insidx-1; while( insert_point > 0 ) { Node *n = b->_nodes[insert_point]; // Hit top of block? Quit going backwards if( n->is_Phi() ) break; // Found a def? Better split after it. if( n2lidx(n) == lidx ) break; // Look for a use uint i; for( i = 1; i < n->req(); i++ ) if( n2lidx(n->in(i)) == lidx ) break; // Found a use? Better split after it. if( i < n->req() ) break; insert_point--; } maxlrg = split_DEF( n1, b, insert_point, maxlrg, Reachblock, debug_defs, splits, slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; } // This is a new DEF, so update UP UPblock[slidx] = false; #ifndef PRODUCT // DEBUG if( trace_spilling() ) { tty->print("\nNew Split DOWN DEF of Spill Idx "); tty->print("%d, UP %d:\n",slidx,false); n1->dump(); } #endif } } // end if LRG is UP } // end for all spilling live ranges assert( b->_nodes[insidx] == n, "got insidx set incorrectly" ); } // end if crossing HRP Boundry // If the LRG index is oob, then this is a new spillcopy, skip it. if( defidx >= _maxlrg ) { continue; } LRG &deflrg = lrgs(defidx); uint copyidx = n->is_Copy(); // Remove coalesced copy from CFG if( copyidx && defidx == n2lidx(n->in(copyidx)) ) { n->replace_by( n->in(copyidx) ); n->set_req( copyidx, NULL ); b->_nodes.remove(insidx--); b->_ihrp_index--; // Adjust the point where we go hi-pressure b->_fhrp_index--; continue; } #define DERIVED 0 // ********** Handle USES ********** bool nullcheck = false; // Implicit null checks never use the spilled value if( n->is_MachNullCheck() ) nullcheck = true; if( !nullcheck ) { // Search all inputs for a Spill-USE JVMState* jvms = n->jvms(); uint oopoff = jvms ? jvms->oopoff() : cnt; uint old_last = cnt - 1; for( inpidx = 1; inpidx < cnt; inpidx++ ) { // Derived/base pairs may be added to our inputs during this loop. // If inpidx > old_last, then one of these new inputs is being // handled. Skip the derived part of the pair, but process // the base like any other input. if( inpidx > old_last && ((inpidx - oopoff) & 1) == DERIVED ) { continue; // skip derived_debug added below } // Get lidx of input uint useidx = Find_id(n->in(inpidx)); // Not a brand-new split, and it is a spill use if( useidx < _maxlrg && lrgs(useidx).reg() >= LRG::SPILL_REG ) { // Check for valid reaching DEF slidx = lrg2reach[useidx]; Node *def = Reachblock[slidx]; assert( def != NULL, "Using Undefined Value in Split()\n"); // (+++) %%%% remove this in favor of pre-pass in matcher.cpp // monitor references do not care where they live, so just hook if ( jvms && jvms->is_monitor_use(inpidx) ) { // The effect of this clone is to drop the node out of the block, // so that the allocator does not see it anymore, and therefore // does not attempt to assign it a register. def = def->clone(); _names.extend(def->_idx,0); _cfg._bbs.map(def->_idx,b); n->set_req(inpidx, def); if (C->check_node_count(NodeLimitFudgeFactor, out_of_nodes)) { return 0; } continue; } // Rematerializable? Then clone def at use site instead // of store/load if( def->rematerialize() ) { int old_size = b->_nodes.size(); def = split_Rematerialize( def, b, insidx, maxlrg, splits, slidx, lrg2reach, Reachblock, true ); if( !def ) return 0; // Bail out insidx += b->_nodes.size()-old_size; } MachNode *mach = n->is_Mach() ? n->as_Mach() : NULL; // Base pointers and oopmap references do not care where they live. if ((inpidx >= oopoff) || (mach && mach->ideal_Opcode() == Op_AddP && inpidx == AddPNode::Base)) { if (def->rematerialize() && lrgs(useidx)._was_spilled2) { // This def has been rematerialized a couple of times without // progress. It doesn't care if it lives UP or DOWN, so // spill it down now. maxlrg = split_USE(def,b,n,inpidx,maxlrg,false,false,splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; // Reset iterator to skip USE side split } else { // Just hook the def edge n->set_req(inpidx, def); } if (inpidx >= oopoff) { // After oopoff, we have derived/base pairs. We must mention all // derived pointers here as derived/base pairs for GC. If the // derived value is spilling and we have a copy both in Reachblock // (called here 'def') and debug_defs[slidx] we need to mention // both in derived/base pairs or kill one. Node *derived_debug = debug_defs[slidx]; if( ((inpidx - oopoff) & 1) == DERIVED && // derived vs base? mach && mach->ideal_Opcode() != Op_Halt && derived_debug != NULL && derived_debug != def ) { // Actual 2nd value appears // We have already set 'def' as a derived value. // Also set debug_defs[slidx] as a derived value. uint k; for( k = oopoff; k < cnt; k += 2 ) if( n->in(k) == derived_debug ) break; // Found an instance of debug derived if( k == cnt ) {// No instance of debug_defs[slidx] // Add a derived/base pair to cover the debug info. // We have to process the added base later since it is not // handled yet at this point but skip derived part. assert(((n->req() - oopoff) & 1) == DERIVED, "must match skip condition above"); n->add_req( derived_debug ); // this will be skipped above n->add_req( n->in(inpidx+1) ); // this will be processed // Increment cnt to handle added input edges on // subsequent iterations. cnt += 2; } } } continue; } // Special logic for DEBUG info if( jvms && b->_freq > BLOCK_FREQUENCY(0.5) ) { uint debug_start = jvms->debug_start(); // If this is debug info use & there is a reaching DOWN def if ((debug_start <= inpidx) && (debug_defs[slidx] != NULL)) { assert(inpidx < oopoff, "handle only debug info here"); // Just hook it in & move on n->set_req(inpidx, debug_defs[slidx]); // (Note that this can make two sides of a split live at the // same time: The debug def on stack, and another def in a // register. The GC needs to know about both of them, but any // derived pointers after oopoff will refer to only one of the // two defs and the GC would therefore miss the other. Thus // this hack is only allowed for debug info which is Java state // and therefore never a derived pointer.) continue; } } // Grab register mask info const RegMask &dmask = def->out_RegMask(); const RegMask &umask = n->in_RegMask(inpidx); assert(inpidx < oopoff, "cannot use-split oop map info"); bool dup = UPblock[slidx]; bool uup = umask.is_UP(); // Need special logic to handle bound USES. Insert a split at this // bound use if we can't rematerialize the def, or if we need the // split to form a misaligned pair. if( !umask.is_AllStack() && (int)umask.Size() <= lrgs(useidx).num_regs() && (!def->rematerialize() || umask.is_misaligned_Pair())) { // These need a Split regardless of overlap or pressure // SPLIT - NO DEF - NO CISC SPILL maxlrg = split_USE(def,b,n,inpidx,maxlrg,dup,false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; // Reset iterator to skip USE side split continue; } // Here is the logic chart which describes USE Splitting: // 0 = false or DOWN, 1 = true or UP // // Overlap | DEF | USE | Action //------------------------------------------------------- // 0 | 0 | 0 | Copy - mem -> mem // 0 | 0 | 1 | Split-UP - Check HRP // 0 | 1 | 0 | Split-DOWN - Debug Info? // 0 | 1 | 1 | Copy - reg -> reg // 1 | 0 | 0 | Reset Input Edge (no Split) // 1 | 0 | 1 | Split-UP - Check HRP // 1 | 1 | 0 | Split-DOWN - Debug Info? // 1 | 1 | 1 | Reset Input Edge (no Split) // // So, if (dup == uup), then overlap test determines action, // with true being no split, and false being copy. Else, // if DEF is DOWN, Split-UP, and check HRP to decide on // resetting DEF. Finally if DEF is UP, Split-DOWN, with // special handling for Debug Info. if( dup == uup ) { if( dmask.overlap(umask) ) { // Both are either up or down, and there is overlap, No Split n->set_req(inpidx, def); } else { // Both are either up or down, and there is no overlap if( dup ) { // If UP, reg->reg copy // COPY ACROSS HERE - NO DEF - NO CISC SPILL maxlrg = split_USE(def,b,n,inpidx,maxlrg,false,false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; // Reset iterator to skip USE side split } else { // DOWN, mem->mem copy // COPY UP & DOWN HERE - NO DEF - NO CISC SPILL // First Split-UP to move value into Register uint def_ideal = def->ideal_reg(); const RegMask* tmp_rm = Matcher::idealreg2regmask[def_ideal]; Node *spill = new (C) MachSpillCopyNode(def, dmask, *tmp_rm); insert_proj( b, insidx, spill, maxlrg ); // Then Split-DOWN as if previous Split was DEF maxlrg = split_USE(spill,b,n,inpidx,maxlrg,false,false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx += 2; // Reset iterator to skip USE side splits } } // End else no overlap } // End if dup == uup // dup != uup, so check dup for direction of Split else { if( dup ) { // If UP, Split-DOWN and check Debug Info // If this node is already a SpillCopy, just patch the edge // except the case of spilling to stack. if( n->is_SpillCopy() ) { RegMask tmp_rm(umask); tmp_rm.SUBTRACT(Matcher::STACK_ONLY_mask); if( dmask.overlap(tmp_rm) ) { if( def != n->in(inpidx) ) { n->set_req(inpidx, def); } continue; } } // COPY DOWN HERE - NO DEF - NO CISC SPILL maxlrg = split_USE(def,b,n,inpidx,maxlrg,false,false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; // Reset iterator to skip USE side split // Check for debug-info split. Capture it for later // debug splits of the same value if (jvms && jvms->debug_start() <= inpidx && inpidx < oopoff) debug_defs[slidx] = n->in(inpidx); } else { // DOWN, Split-UP and check register pressure if( is_high_pressure( b, &lrgs(useidx), insidx ) ) { // COPY UP HERE - NO DEF - CISC SPILL maxlrg = split_USE(def,b,n,inpidx,maxlrg,true,true, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } insidx++; // Reset iterator to skip USE side split } else { // LRP // COPY UP HERE - WITH DEF - NO CISC SPILL maxlrg = split_USE(def,b,n,inpidx,maxlrg,true,false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } // Flag this lift-up in a low-pressure block as // already-spilled, so if it spills again it will // spill hard (instead of not spilling hard and // coalescing away). set_was_spilled(n->in(inpidx)); // Since this is a new DEF, update Reachblock & UP Reachblock[slidx] = n->in(inpidx); UPblock[slidx] = true; insidx++; // Reset iterator to skip USE side split } } // End else DOWN } // End dup != uup } // End if Spill USE } // End For All Inputs } // End If not nullcheck // ********** Handle DEFS ********** // DEFS either Split DOWN in HRP regions or when the LRG is bound, or // just reset the Reaches info in LRP regions. DEFS must always update // UP info. if( deflrg.reg() >= LRG::SPILL_REG ) { // Spilled? uint slidx = lrg2reach[defidx]; // Add to defs list for later assignment of new live range number defs->push(n); // Set a flag on the Node indicating it has already spilled. // Only do it for capacity spills not conflict spills. if( !deflrg._direct_conflict ) set_was_spilled(n); assert(!n->is_Phi(),"Cannot insert Phi into DEFS list"); // Grab UP info for DEF const RegMask &dmask = n->out_RegMask(); bool defup = dmask.is_UP(); // Only split at Def if this is a HRP block or bound (and spilled once) if( !n->rematerialize() && (((dmask.is_bound1() || dmask.is_bound2() || dmask.is_misaligned_Pair()) && (deflrg._direct_conflict || deflrg._must_spill)) || // Check for LRG being up in a register and we are inside a high // pressure area. Spill it down immediately. (defup && is_high_pressure(b,&deflrg,insidx))) ) { assert( !n->rematerialize(), "" ); assert( !n->is_SpillCopy(), "" ); // Do a split at the def site. maxlrg = split_DEF( n, b, insidx, maxlrg, Reachblock, debug_defs, splits, slidx ); // If it wasn't split bail if (!maxlrg) { return 0; } // Split DEF's Down UPblock[slidx] = 0; #ifndef PRODUCT // DEBUG if( trace_spilling() ) { tty->print("\nNew Split DOWN DEF of Spill Idx "); tty->print("%d, UP %d:\n",slidx,false); n->dump(); } #endif } else { // Neither bound nor HRP, must be LRP // otherwise, just record the def Reachblock[slidx] = n; // UP should come from the outRegmask() of the DEF UPblock[slidx] = defup; // Update debug list of reaching down definitions, kill if DEF is UP debug_defs[slidx] = defup ? NULL : n; #ifndef PRODUCT // DEBUG if( trace_spilling() ) { tty->print("\nNew DEF of Spill Idx "); tty->print("%d, UP %d:\n",slidx,defup); n->dump(); } #endif } // End else LRP } // End if spill def // ********** Split Left Over Mem-Mem Moves ********** // Check for mem-mem copies and split them now. Do not do this // to copies about to be spilled; they will be Split shortly. if( copyidx ) { Node *use = n->in(copyidx); uint useidx = Find_id(use); if( useidx < _maxlrg && // This is not a new split OptoReg::is_stack(deflrg.reg()) && deflrg.reg() < LRG::SPILL_REG ) { // And DEF is from stack LRG &uselrg = lrgs(useidx); if( OptoReg::is_stack(uselrg.reg()) && uselrg.reg() < LRG::SPILL_REG && // USE is from stack deflrg.reg() != uselrg.reg() ) { // Not trivially removed uint def_ideal_reg = Matcher::base2reg[n->bottom_type()->base()]; const RegMask &def_rm = *Matcher::idealreg2regmask[def_ideal_reg]; const RegMask &use_rm = n->in_RegMask(copyidx); if( def_rm.overlap(use_rm) && n->is_SpillCopy() ) { // Bug 4707800, 'n' may be a storeSSL if (C->check_node_count(NodeLimitFudgeFactor, out_of_nodes)) { // Check when generating nodes return 0; } Node *spill = new (C) MachSpillCopyNode(use,use_rm,def_rm); n->set_req(copyidx,spill); n->as_MachSpillCopy()->set_in_RegMask(def_rm); // Put the spill just before the copy insert_proj( b, insidx++, spill, maxlrg++ ); } } } } } // End For All Instructions in Block - Non-PHI Pass // Check if each LRG is live out of this block so as not to propagate // beyond the last use of a LRG. for( slidx = 0; slidx < spill_cnt; slidx++ ) { uint defidx = lidxs.at(slidx); IndexSet *liveout = _live->live(b); if( !liveout->member(defidx) ) { #ifdef ASSERT // The index defidx is not live. Check the liveout array to ensure that // it contains no members which compress to defidx. Finding such an // instance may be a case to add liveout adjustment in compress_uf_map(). // See 5063219. uint member; IndexSetIterator isi(liveout); while ((member = isi.next()) != 0) { assert(defidx != Find_const(member), "Live out member has not been compressed"); } #endif Reachblock[slidx] = NULL; } else { assert(Reachblock[slidx] != NULL,"No reaching definition for liveout value"); } } #ifndef PRODUCT if( trace_spilling() ) b->dump(); #endif } // End For All Blocks //----------PASS 2---------- // Reset all DEF live range numbers here for( insidx = 0; insidx < defs->size(); insidx++ ) { // Grab the def n1 = defs->at(insidx); // Set new lidx for DEF new_lrg(n1, maxlrg++); } //----------Phi Node Splitting---------- // Clean up a phi here, and assign a new live range number // Cycle through this block's predecessors, collecting Reaches // info for each spilled LRG and update edges. // Walk the phis list to patch inputs, split phis, and name phis for( insidx = 0; insidx < phis->size(); insidx++ ) { Node *phi = phis->at(insidx); assert(phi->is_Phi(),"This list must only contain Phi Nodes"); Block *b = _cfg._bbs[phi->_idx]; // Grab the live range number uint lidx = Find_id(phi); uint slidx = lrg2reach[lidx]; // Update node to lidx map new_lrg(phi, maxlrg++); // Get PASS1's up/down decision for the block. int phi_up = !!UP_entry[slidx]->test(b->_pre_order); // Force down if double-spilling live range if( lrgs(lidx)._was_spilled1 ) phi_up = false; // When splitting a Phi we an split it normal or "inverted". // An inverted split makes the splits target the Phi's UP/DOWN // sense inverted; then the Phi is followed by a final def-side // split to invert back. It changes which blocks the spill code // goes in. // Walk the predecessor blocks and assign the reaching def to the Phi. // Split Phi nodes by placing USE side splits wherever the reaching // DEF has the wrong UP/DOWN value. for( uint i = 1; i < b->num_preds(); i++ ) { // Get predecessor block pre-order number Block *pred = _cfg._bbs[b->pred(i)->_idx]; pidx = pred->_pre_order; // Grab reaching def Node *def = Reaches[pidx][slidx]; assert( def, "must have reaching def" ); // If input up/down sense and reg-pressure DISagree if( def->rematerialize() ) { def = split_Rematerialize( def, pred, pred->end_idx(), maxlrg, splits, slidx, lrg2reach, Reachblock, false ); if( !def ) return 0; // Bail out } // Update the Phi's input edge array phi->set_req(i,def); // Grab the UP/DOWN sense for the input u1 = UP[pidx][slidx]; if( u1 != (phi_up != 0)) { maxlrg = split_USE(def, b, phi, i, maxlrg, !u1, false, splits,slidx); // If it wasn't split bail if (!maxlrg) { return 0; } } } // End for all inputs to the Phi } // End for all Phi Nodes // Update _maxlrg to save Union asserts _maxlrg = maxlrg; //----------PASS 3---------- // Pass over all Phi's to union the live ranges for( insidx = 0; insidx < phis->size(); insidx++ ) { Node *phi = phis->at(insidx); assert(phi->is_Phi(),"This list must only contain Phi Nodes"); // Walk all inputs to Phi and Union input live range with Phi live range for( uint i = 1; i < phi->req(); i++ ) { // Grab the input node Node *n = phi->in(i); assert( n, "" ); uint lidx = Find(n); uint pidx = Find(phi); if( lidx < pidx ) Union(n, phi); else if( lidx > pidx ) Union(phi, n); } // End for all inputs to the Phi Node } // End for all Phi Nodes // Now union all two address instructions for( insidx = 0; insidx < defs->size(); insidx++ ) { // Grab the def n1 = defs->at(insidx); // Set new lidx for DEF & handle 2-addr instructions if( n1->is_Mach() && ((twoidx = n1->as_Mach()->two_adr()) != 0) ) { assert( Find(n1->in(twoidx)) < maxlrg,"Assigning bad live range index"); // Union the input and output live ranges uint lr1 = Find(n1); uint lr2 = Find(n1->in(twoidx)); if( lr1 < lr2 ) Union(n1, n1->in(twoidx)); else if( lr1 > lr2 ) Union(n1->in(twoidx), n1); } // End if two address } // End for all defs // DEBUG #ifdef ASSERT // Validate all live range index assignments for( bidx = 0; bidx < _cfg._num_blocks; bidx++ ) { b = _cfg._blocks[bidx]; for( insidx = 0; insidx <= b->end_idx(); insidx++ ) { Node *n = b->_nodes[insidx]; uint defidx = Find(n); assert(defidx < _maxlrg,"Bad live range index in Split"); assert(defidx < maxlrg,"Bad live range index in Split"); } } // Issue a warning if splitting made no progress int noprogress = 0; for( slidx = 0; slidx < spill_cnt; slidx++ ) { if( PrintOpto && WizardMode && splits.at(slidx) == 0 ) { tty->print_cr("Failed to split live range %d", lidxs.at(slidx)); //BREAKPOINT; } else { noprogress++; } } if(!noprogress) { tty->print_cr("Failed to make progress in Split"); //BREAKPOINT; } #endif // Return updated count of live ranges return maxlrg; }