|
/*
* Copyright (c) 1998, 2021, 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 "precompiled.hpp"
#include "compiler/oopMap.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/resourceArea.hpp"
#include "opto/addnode.hpp"
#include "opto/block.hpp"
#include "opto/callnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/chaitin.hpp"
#include "opto/coalesce.hpp"
#include "opto/indexSet.hpp"
#include "opto/machnode.hpp"
#include "opto/memnode.hpp"
#include "opto/opcodes.hpp"
#include <fenv.h>
PhaseIFG::PhaseIFG( Arena *arena ) : Phase(Interference_Graph), _arena(arena) {
}
void PhaseIFG::init( uint maxlrg ) {
_maxlrg = maxlrg;
_yanked = new (_arena) VectorSet(_arena);
_is_square = false;
// Make uninitialized adjacency lists
_adjs = (IndexSet*)_arena->Amalloc(sizeof(IndexSet)*maxlrg);
// Also make empty live range structures
_lrgs = (LRG *)_arena->Amalloc( maxlrg * sizeof(LRG) );
memset((void*)_lrgs,0,sizeof(LRG)*maxlrg);
// Init all to empty
for( uint i = 0; i < maxlrg; i++ ) {
_adjs[i].initialize(maxlrg);
_lrgs[i].Set_All();
}
}
// Add edge between vertices a & b. These are sorted (triangular matrix),
// then the smaller number is inserted in the larger numbered array.
int PhaseIFG::add_edge( uint a, uint b ) {
lrgs(a).invalid_degree();
lrgs(b).invalid_degree();
// Sort a and b, so that a is bigger
assert( !_is_square, "only on triangular" );
if( a < b ) { uint tmp = a; a = b; b = tmp; }
return _adjs[a].insert( b );
}
// Is there an edge between a and b?
int PhaseIFG::test_edge( uint a, uint b ) const {
// Sort a and b, so that a is larger
assert( !_is_square, "only on triangular" );
if( a < b ) { uint tmp = a; a = b; b = tmp; }
return _adjs[a].member(b);
}
// Convert triangular matrix to square matrix
void PhaseIFG::SquareUp() {
assert( !_is_square, "only on triangular" );
// Simple transpose
for(uint i = 0; i < _maxlrg; i++ ) {
if (!_adjs[i].is_empty()) {
IndexSetIterator elements(&_adjs[i]);
uint datum;
while ((datum = elements.next()) != 0) {
_adjs[datum].insert(i);
}
}
}
_is_square = true;
}
// Compute effective degree in bulk
void PhaseIFG::Compute_Effective_Degree() {
assert( _is_square, "only on square" );
for( uint i = 0; i < _maxlrg; i++ )
lrgs(i).set_degree(effective_degree(i));
}
int PhaseIFG::test_edge_sq( uint a, uint b ) const {
assert( _is_square, "only on square" );
// Swap, so that 'a' has the lesser count. Then binary search is on
// the smaller of a's list and b's list.
if( neighbor_cnt(a) > neighbor_cnt(b) ) { uint tmp = a; a = b; b = tmp; }
//return _adjs[a].unordered_member(b);
return _adjs[a].member(b);
}
// Union edges of B into A
void PhaseIFG::Union(uint a, uint b) {
assert( _is_square, "only on square" );
IndexSet *A = &_adjs[a];
if (!_adjs[b].is_empty()) {
IndexSetIterator b_elements(&_adjs[b]);
uint datum;
while ((datum = b_elements.next()) != 0) {
if (A->insert(datum)) {
_adjs[datum].insert(a);
lrgs(a).invalid_degree();
lrgs(datum).invalid_degree();
}
}
}
}
// Yank a Node and all connected edges from the IFG. Return a
// list of neighbors (edges) yanked.
IndexSet *PhaseIFG::remove_node( uint a ) {
assert( _is_square, "only on square" );
assert( !_yanked->test(a), "" );
_yanked->set(a);
// I remove the LRG from all neighbors.
LRG &lrg_a = lrgs(a);
if (!_adjs[a].is_empty()) {
IndexSetIterator elements(&_adjs[a]);
uint datum;
while ((datum = elements.next()) != 0) {
_adjs[datum].remove(a);
lrgs(datum).inc_degree(-lrg_a.compute_degree(lrgs(datum)));
}
}
return neighbors(a);
}
// Re-insert a yanked Node.
void PhaseIFG::re_insert(uint a) {
assert( _is_square, "only on square" );
assert( _yanked->test(a), "" );
_yanked->remove(a);
if (_adjs[a].is_empty()) return;
IndexSetIterator elements(&_adjs[a]);
uint datum;
while ((datum = elements.next()) != 0) {
_adjs[datum].insert(a);
lrgs(datum).invalid_degree();
}
}
// Compute the degree between 2 live ranges. If both live ranges are
// aligned-adjacent powers-of-2 then we use the MAX size. If either is
// mis-aligned (or for Fat-Projections, not-adjacent) then we have to
// MULTIPLY the sizes. Inspect Brigg's thesis on register pairs to see why
// this is so.
int LRG::compute_degree(LRG &l) const {
int tmp;
int num_regs = _num_regs;
int nregs = l.num_regs();
tmp = (_fat_proj || l._fat_proj) // either is a fat-proj?
? (num_regs * nregs) // then use product
: MAX2(num_regs,nregs); // else use max
return tmp;
}
// Compute effective degree for this live range. If both live ranges are
// aligned-adjacent powers-of-2 then we use the MAX size. If either is
// mis-aligned (or for Fat-Projections, not-adjacent) then we have to
// MULTIPLY the sizes. Inspect Brigg's thesis on register pairs to see why
// this is so.
int PhaseIFG::effective_degree(uint lidx) const {
IndexSet *s = neighbors(lidx);
if (s->is_empty()) return 0;
int eff = 0;
int num_regs = lrgs(lidx).num_regs();
int fat_proj = lrgs(lidx)._fat_proj;
IndexSetIterator elements(s);
uint nidx;
while ((nidx = elements.next()) != 0) {
LRG &lrgn = lrgs(nidx);
int nregs = lrgn.num_regs();
eff += (fat_proj || lrgn._fat_proj) // either is a fat-proj?
? (num_regs * nregs) // then use product
: MAX2(num_regs,nregs); // else use max
}
return eff;
}
#ifndef PRODUCT
void PhaseIFG::dump() const {
tty->print_cr("-- Interference Graph --%s--",
_is_square ? "square" : "triangular" );
if (_is_square) {
for (uint i = 0; i < _maxlrg; i++) {
tty->print(_yanked->test(i) ? "XX " : " ");
tty->print("L%d: { ",i);
if (!_adjs[i].is_empty()) {
IndexSetIterator elements(&_adjs[i]);
uint datum;
while ((datum = elements.next()) != 0) {
tty->print("L%d ", datum);
}
}
tty->print_cr("}");
}
return;
}
// Triangular
for( uint i = 0; i < _maxlrg; i++ ) {
uint j;
tty->print(_yanked->test(i) ? "XX " : " ");
tty->print("L%d: { ",i);
for( j = _maxlrg; j > i; j-- )
if( test_edge(j - 1,i) ) {
tty->print("L%d ",j - 1);
}
tty->print("| ");
if (!_adjs[i].is_empty()) {
IndexSetIterator elements(&_adjs[i]);
uint datum;
while ((datum = elements.next()) != 0) {
tty->print("L%d ", datum);
}
}
tty->print("}\n");
}
tty->print("\n");
}
void PhaseIFG::stats() const {
ResourceMark rm;
int *h_cnt = NEW_RESOURCE_ARRAY(int,_maxlrg*2);
memset( h_cnt, 0, sizeof(int)*_maxlrg*2 );
uint i;
for( i = 0; i < _maxlrg; i++ ) {
h_cnt[neighbor_cnt(i)]++;
}
tty->print_cr("--Histogram of counts--");
for( i = 0; i < _maxlrg*2; i++ )
if( h_cnt[i] )
tty->print("%d/%d ",i,h_cnt[i]);
tty->cr();
}
void PhaseIFG::verify( const PhaseChaitin *pc ) const {
// IFG is square, sorted and no need for Find
for( uint i = 0; i < _maxlrg; i++ ) {
assert(!_yanked->test(i) || !neighbor_cnt(i), "Is removed completely" );
IndexSet *set = &_adjs[i];
if (!set->is_empty()) {
IndexSetIterator elements(set);
uint idx;
uint last = 0;
while ((idx = elements.next()) != 0) {
assert(idx != i, "Must have empty diagonal");
assert(pc->_lrg_map.find_const(idx) == idx, "Must not need Find");
assert(_adjs[idx].member(i), "IFG not square");
assert(!_yanked->test(idx), "No yanked neighbors");
assert(last < idx, "not sorted increasing");
last = idx;
}
}
assert(!lrgs(i)._degree_valid || effective_degree(i) == lrgs(i).degree(), "degree is valid but wrong");
}
}
#endif
/*
* Interfere this register with everything currently live.
* Check for interference by checking overlap of regmasks.
* Only interfere if acceptable register masks overlap.
*/
void PhaseChaitin::interfere_with_live(uint lid, IndexSet* liveout) {
if (!liveout->is_empty()) {
LRG& lrg = lrgs(lid);
const RegMask &rm = lrg.mask();
IndexSetIterator elements(liveout);
uint interfering_lid = elements.next();
while (interfering_lid != 0) {
LRG& interfering_lrg = lrgs(interfering_lid);
if (rm.overlap(interfering_lrg.mask())) {
_ifg->add_edge(lid, interfering_lid);
}
interfering_lid = elements.next();
}
}
}
// Actually build the interference graph. Uses virtual registers only, no
// physical register masks. This allows me to be very aggressive when
// coalescing copies. Some of this aggressiveness will have to be undone
// later, but I'd rather get all the copies I can now (since unremoved copies
// at this point can end up in bad places). Copies I re-insert later I have
// more opportunity to insert them in low-frequency locations.
void PhaseChaitin::build_ifg_virtual( ) {
Compile::TracePhase tp("buildIFG_virt", &timers[_t_buildIFGvirtual]);
// For all blocks (in any order) do...
for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
Block* block = _cfg.get_block(i);
IndexSet* liveout = _live->live(block);
// The IFG is built by a single reverse pass over each basic block.
// Starting with the known live-out set, we remove things that get
// defined and add things that become live (essentially executing one
// pass of a standard LIVE analysis). Just before a Node defines a value
// (and removes it from the live-ness set) that value is certainly live.
// The defined value interferes with everything currently live. The
// value is then removed from the live-ness set and it's inputs are
// added to the live-ness set.
for (uint j = block->end_idx() + 1; j > 1; j--) {
Node* n = block->get_node(j - 1);
// Get value being defined
uint r = _lrg_map.live_range_id(n);
// Some special values do not allocate
if (r) {
// Remove from live-out set
liveout->remove(r);
// Copies do not define a new value and so do not interfere.
// Remove the copies source from the liveout set before interfering.
uint idx = n->is_Copy();
if (idx != 0) {
liveout->remove(_lrg_map.live_range_id(n->in(idx)));
}
// Interfere with everything live
interfere_with_live(r, liveout);
}
// Make all inputs live
if (!n->is_Phi()) { // Phi function uses come from prior block
for(uint k = 1; k < n->req(); k++) {
liveout->insert(_lrg_map.live_range_id(n->in(k)));
}
}
// 2-address instructions always have the defined value live
// on entry to the instruction, even though it is being defined
// by the instruction. We pretend a virtual copy sits just prior
// to the instruction and kills the src-def'd register.
// In other words, for 2-address instructions the defined value
// interferes with all inputs.
uint idx;
if( n->is_Mach() && (idx = n->as_Mach()->two_adr()) ) {
const MachNode *mach = n->as_Mach();
// Sometimes my 2-address ADDs are commuted in a bad way.
// We generally want the USE-DEF register to refer to the
// loop-varying quantity, to avoid a copy.
uint op = mach->ideal_Opcode();
// Check that mach->num_opnds() == 3 to ensure instruction is
// not subsuming constants, effectively excludes addI_cin_imm
// Can NOT swap for instructions like addI_cin_imm since it
// is adding zero to yhi + carry and the second ideal-input
// points to the result of adding low-halves.
// Checking req() and num_opnds() does NOT distinguish addI_cout from addI_cout_imm
if( (op == Op_AddI && mach->req() == 3 && mach->num_opnds() == 3) &&
n->in(1)->bottom_type()->base() == Type::Int &&
// See if the ADD is involved in a tight data loop the wrong way
n->in(2)->is_Phi() &&
n->in(2)->in(2) == n ) {
Node *tmp = n->in(1);
n->set_req( 1, n->in(2) );
n->set_req( 2, tmp );
}
// Defined value interferes with all inputs
uint lidx = _lrg_map.live_range_id(n->in(idx));
for (uint k = 1; k < n->req(); k++) {
uint kidx = _lrg_map.live_range_id(n->in(k));
if (kidx != lidx) {
_ifg->add_edge(r, kidx);
}
}
}
} // End of forall instructions in block
} // End of forall blocks
}
#ifdef ASSERT
uint PhaseChaitin::count_int_pressure(IndexSet* liveout) {
if (liveout->is_empty()) {
return 0;
}
IndexSetIterator elements(liveout);
uint lidx = elements.next();
uint cnt = 0;
while (lidx != 0) {
LRG& lrg = lrgs(lidx);
if (lrg.mask_is_nonempty_and_up() &&
!lrg.is_float_or_vector() &&
(lrg.mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ||
(Matcher::has_predicated_vectors() &&
lrg.mask().overlap(*Matcher::idealreg2regmask[Op_RegVectMask])))) {
cnt += lrg.reg_pressure();
}
lidx = elements.next();
}
return cnt;
}
uint PhaseChaitin::count_float_pressure(IndexSet* liveout) {
if (liveout->is_empty()) {
return 0;
}
IndexSetIterator elements(liveout);
uint lidx = elements.next();
uint cnt = 0;
while (lidx != 0) {
LRG& lrg = lrgs(lidx);
if (lrg.mask_is_nonempty_and_up() && lrg.is_float_or_vector()) {
cnt += lrg.reg_pressure();
}
lidx = elements.next();
}
return cnt;
}
#endif
/*
* Adjust register pressure down by 1. Capture last hi-to-low transition,
*/
void PhaseChaitin::lower_pressure(Block* b, uint location, LRG& lrg, IndexSet* liveout, Pressure& int_pressure, Pressure& float_pressure) {
if (lrg.mask_is_nonempty_and_up()) {
if (lrg.is_float_or_vector()) {
float_pressure.lower(lrg, location);
} else {
// Do not count the SP and flag registers
const RegMask& r = lrg.mask();
if (r.overlap(*Matcher::idealreg2regmask[Op_RegI]) ||
(Matcher::has_predicated_vectors() &&
r.overlap(*Matcher::idealreg2regmask[Op_RegVectMask]))) {
int_pressure.lower(lrg, location);
}
}
}
if (_scheduling_info_generated == false) {
assert(int_pressure.current_pressure() == count_int_pressure(liveout), "the int pressure is incorrect");
assert(float_pressure.current_pressure() == count_float_pressure(liveout), "the float pressure is incorrect");
}
}
/* Go to the first non-phi index in a block */
static uint first_nonphi_index(Block* b) {
uint i;
uint end_idx = b->end_idx();
for (i = 1; i < end_idx; i++) {
Node* n = b->get_node(i);
if (!n->is_Phi()) {
break;
}
}
return i;
}
/*
* Spills could be inserted before a CreateEx node which should be the first
* instruction in a block after Phi nodes. If so, move the CreateEx node up.
*/
static void move_exception_node_up(Block* b, uint first_inst, uint last_inst) {
for (uint i = first_inst; i < last_inst; i++) {
Node* ex = b->get_node(i);
if (ex->is_SpillCopy()) {
continue;
}
if (i > first_inst &&
ex->is_Mach() && ex->as_Mach()->ideal_Opcode() == Op_CreateEx) {
b->remove_node(i);
b->insert_node(ex, first_inst);
}
// Stop once a CreateEx or any other node is found
break;
}
}
/*
* When new live ranges are live, we raise the register pressure
*/
void PhaseChaitin::raise_pressure(Block* b, LRG& lrg, Pressure& int_pressure, Pressure& float_pressure) {
if (lrg.mask_is_nonempty_and_up()) {
if (lrg.is_float_or_vector()) {
float_pressure.raise(lrg);
} else {
// Do not count the SP and flag registers
const RegMask& rm = lrg.mask();
if (rm.overlap(*Matcher::idealreg2regmask[Op_RegI]) ||
(Matcher::has_predicated_vectors() &&
rm.overlap(*Matcher::idealreg2regmask[Op_RegVectMask]))) {
int_pressure.raise(lrg);
}
}
}
}
/*
* Computes the initial register pressure of a block, looking at all live
* ranges in the liveout. The register pressure is computed for both float
* and int/pointer registers.
* Live ranges in the liveout are presumed live for the whole block.
* We add the cost for the whole block to the area of the live ranges initially.
* If a live range gets killed in the block, we'll subtract the unused part of
* the block from the area.
*/
void PhaseChaitin::compute_initial_block_pressure(Block* b, IndexSet* liveout, Pressure& int_pressure, Pressure& float_pressure, double cost) {
if (!liveout->is_empty()) {
IndexSetIterator elements(liveout);
uint lid = elements.next();
while (lid != 0) {
LRG &lrg = lrgs(lid);
lrg._area += cost;
raise_pressure(b, lrg, int_pressure, float_pressure);
lid = elements.next();
}
}
assert(int_pressure.current_pressure() == count_int_pressure(liveout), "the int pressure is incorrect");
assert(float_pressure.current_pressure() == count_float_pressure(liveout), "the float pressure is incorrect");
}
/*
* Computes the entry register pressure of a block, looking at all live
* ranges in the livein. The register pressure is computed for both float
* and int/pointer registers.
*/
void PhaseChaitin::compute_entry_block_pressure(Block* b) {
IndexSet *livein = _live->livein(b);
if (!livein->is_empty()) {
IndexSetIterator elements(livein);
uint lid = elements.next();
while (lid != 0) {
LRG &lrg = lrgs(lid);
raise_pressure(b, lrg, _sched_int_pressure, _sched_float_pressure);
lid = elements.next();
}
}
// Now check phis for locally defined inputs
for (uint j = 0; j < b->number_of_nodes(); j++) {
Node* n = b->get_node(j);
if (n->is_Phi()) {
for (uint k = 1; k < n->req(); k++) {
Node* phi_in = n->in(k);
// Because we are talking about phis, raise register pressure once for each
// instance of a phi to account for a single value
if (_cfg.get_block_for_node(phi_in) == b) {
LRG& lrg = lrgs(phi_in->_idx);
raise_pressure(b, lrg, _sched_int_pressure, _sched_float_pressure);
break;
}
}
}
}
_sched_int_pressure.set_start_pressure(_sched_int_pressure.current_pressure());
_sched_float_pressure.set_start_pressure(_sched_float_pressure.current_pressure());
}
/*
* Computes the exit register pressure of a block, looking at all live
* ranges in the liveout. The register pressure is computed for both float
* and int/pointer registers.
*/
void PhaseChaitin::compute_exit_block_pressure(Block* b) {
IndexSet* livein = _live->live(b);
_sched_int_pressure.set_current_pressure(0);
_sched_float_pressure.set_current_pressure(0);
if (!livein->is_empty()) {
IndexSetIterator elements(livein);
uint lid = elements.next();
while (lid != 0) {
LRG &lrg = lrgs(lid);
raise_pressure(b, lrg, _sched_int_pressure, _sched_float_pressure);
lid = elements.next();
}
}
}
/*
* Remove dead node if it's not used.
* We only remove projection nodes if the node "defining" the projection is
* dead, for example on x86, if we have a dead Add node we remove its
* RFLAGS node.
*/
bool PhaseChaitin::remove_node_if_not_used(Block* b, uint location, Node* n, uint lid, IndexSet* liveout) {
Node* def = n->in(0);
if (!n->is_Proj() ||
(_lrg_map.live_range_id(def) && !liveout->member(_lrg_map.live_range_id(def)))) {
if (n->is_MachProj()) {
// Don't remove KILL projections if their "defining" nodes have
// memory effects (have SCMemProj projection node) -
// they are not dead even when their result is not used.
// For example, compareAndSwapL (and other CAS) and EncodeISOArray nodes.
// The method add_input_to_liveout() keeps such nodes alive (put them on liveout list)
// when it sees SCMemProj node in a block. Unfortunately SCMemProj node could be placed
// in block in such order that KILL MachProj nodes are processed first.
if (def->has_out_with(Op_SCMemProj)) {
return false;
}
}
b->remove_node(location);
LRG& lrg = lrgs(lid);
if (lrg._def == n) {
lrg._def = 0;
}
n->disconnect_inputs(C);
_cfg.unmap_node_from_block(n);
n->replace_by(C->top());
return true;
}
return false;
}
/*
* When encountering a fat projection, we might go from a low to high to low
* (since the fat proj only lives at this instruction) going backwards in the
* block. If we find a low to high transition, we record it.
*/
void PhaseChaitin::check_for_high_pressure_transition_at_fatproj(uint& block_reg_pressure, uint location, LRG& lrg, Pressure& pressure, const int op_regtype) {
RegMask mask_tmp = lrg.mask();
mask_tmp.AND(*Matcher::idealreg2regmask[op_regtype]);
pressure.check_pressure_at_fatproj(location, mask_tmp);
}
/*
* Insure high score for immediate-use spill copies so they get a color.
* All single-use MachSpillCopy(s) that immediately precede their
* use must color early. If a longer live range steals their
* color, the spill copy will split and may push another spill copy
* further away resulting in an infinite spill-split-retry cycle.
* Assigning a zero area results in a high score() and a good
* location in the simplify list.
*/
void PhaseChaitin::assign_high_score_to_immediate_copies(Block* b, Node* n, LRG& lrg, uint next_inst, uint last_inst) {
if (n->is_SpillCopy() &&
lrg.is_singledef() && // A multi defined live range can still split
n->outcnt() == 1 && // and use must be in this block
_cfg.get_block_for_node(n->unique_out()) == b) {
Node* single_use = n->unique_out();
assert(b->find_node(single_use) >= next_inst, "Use must be later in block");
// Use can be earlier in block if it is a Phi, but then I should be a MultiDef
// Find first non SpillCopy 'm' that follows the current instruction
// (current_inst - 1) is index for current instruction 'n'
Node* m = n;
for (uint i = next_inst; i <= last_inst && m->is_SpillCopy(); ++i) {
m = b->get_node(i);
}
if (m == single_use) {
lrg._area = 0.0;
}
}
}
/*
* Copies do not define a new value and so do not interfere.
* Remove the copies source from the liveout set before interfering.
*/
void PhaseChaitin::remove_interference_from_copy(Block* b, uint location, uint lid_copy, IndexSet* liveout, double cost, Pressure& int_pressure, Pressure& float_pressure) {
if (liveout->remove(lid_copy)) {
LRG& lrg_copy = lrgs(lid_copy);
lrg_copy._area -= cost;
// Lower register pressure since copy and definition can share the same register
lower_pressure(b, location, lrg_copy, liveout, int_pressure, float_pressure);
}
}
/*
* The defined value must go in a particular register. Remove that register from
* all conflicting parties and avoid the interference.
*/
void PhaseChaitin::remove_bound_register_from_interfering_live_ranges(LRG& lrg, IndexSet* liveout, uint& must_spill) {
if (liveout->is_empty()) return;
// Check for common case
const RegMask& rm = lrg.mask();
int r_size = lrg.num_regs();
// Smear odd bits
IndexSetIterator elements(liveout);
uint l = elements.next();
while (l != 0) {
LRG& interfering_lrg = lrgs(l);
// If 'l' must spill already, do not further hack his bits.
// He'll get some interferences and be forced to spill later.
if (interfering_lrg._must_spill) {
l = elements.next();
continue;
}
// Remove bound register(s) from 'l's choices
RegMask old = interfering_lrg.mask();
uint old_size = interfering_lrg.mask_size();
// Remove the bits from LRG 'rm' from LRG 'l' so 'l' no
// longer interferes with 'rm'. If 'l' requires aligned
// adjacent pairs, subtract out bit pairs.
assert(!interfering_lrg._is_vector || !interfering_lrg._fat_proj, "sanity");
if (interfering_lrg.num_regs() > 1 && !interfering_lrg._fat_proj) {
RegMask r2mask = rm;
// Leave only aligned set of bits.
r2mask.smear_to_sets(interfering_lrg.num_regs());
// It includes vector case.
interfering_lrg.SUBTRACT(r2mask);
interfering_lrg.compute_set_mask_size();
} else if (r_size != 1) {
// fat proj
interfering_lrg.SUBTRACT(rm);
interfering_lrg.compute_set_mask_size();
} else {
// Common case: size 1 bound removal
OptoReg::Name r_reg = rm.find_first_elem();
if (interfering_lrg.mask().Member(r_reg)) {
interfering_lrg.Remove(r_reg);
interfering_lrg.set_mask_size(interfering_lrg.mask().is_AllStack() ? LRG::AllStack_size : old_size - 1);
}
}
// If 'l' goes completely dry, it must spill.
if (interfering_lrg.not_free()) {
// Give 'l' some kind of reasonable mask, so it picks up
// interferences (and will spill later).
interfering_lrg.set_mask(old);
interfering_lrg.set_mask_size(old_size);
must_spill++;
interfering_lrg._must_spill = 1;
interfering_lrg.set_reg(OptoReg::Name(LRG::SPILL_REG));
}
l = elements.next();
}
}
/*
* Start loop at 1 (skip control edge) for most Nodes. SCMemProj's might be the
* sole use of a StoreLConditional. While StoreLConditionals set memory (the
* SCMemProj use) they also def flags; if that flag def is unused the allocator
* sees a flag-setting instruction with no use of the flags and assumes it's
* dead. This keeps the (useless) flag-setting behavior alive while also
* keeping the (useful) memory update effect.
*/
void PhaseChaitin::add_input_to_liveout(Block* b, Node* n, IndexSet* liveout, double cost, Pressure& int_pressure, Pressure& float_pressure) {
JVMState* jvms = n->jvms();
uint debug_start = jvms ? jvms->debug_start() : 999999;
for (uint k = ((n->Opcode() == Op_SCMemProj) ? 0:1); k < n->req(); k++) {
Node* def = n->in(k);
uint lid = _lrg_map.live_range_id(def);
if (!lid) {
continue;
}
LRG& lrg = lrgs(lid);
// No use-side cost for spilling debug info
if (k < debug_start) {
// A USE costs twice block frequency (once for the Load, once
// for a Load-delay). Rematerialized uses only cost once.
lrg._cost += (def->rematerialize() ? b->_freq : (b->_freq * 2));
}
if (liveout->insert(lid)) {
// Newly live things assumed live from here to top of block
lrg._area += cost;
raise_pressure(b, lrg, int_pressure, float_pressure);
assert(int_pressure.current_pressure() == count_int_pressure(liveout), "the int pressure is incorrect");
assert(float_pressure.current_pressure() == count_float_pressure(liveout), "the float pressure is incorrect");
}
assert(lrg._area >= 0.0, "unexpected spill area value %g (rounding mode %x)", lrg._area, fegetround());
}
}
/*
* If we run off the top of the block with high pressure just record that the
* whole block is high pressure. (Even though we might have a transition
* later down in the block)
*/
void PhaseChaitin::check_for_high_pressure_block(Pressure& pressure) {
// current pressure now means the pressure before the first instruction in the block
// (since we have stepped through all instructions backwards)
if (pressure.current_pressure() > pressure.high_pressure_limit()) {
pressure.set_high_pressure_index_to_block_start();
}
}
/*
* Compute high pressure indice; avoid landing in the middle of projnodes
* and set the high pressure index for the block
*/
void PhaseChaitin::adjust_high_pressure_index(Block* b, uint& block_hrp_index, Pressure& pressure) {
uint i = pressure.high_pressure_index();
if (i < b->number_of_nodes() && i < b->end_idx() + 1) {
Node* cur = b->get_node(i);
while (cur->is_Proj() || (cur->is_MachNullCheck()) || cur->is_Catch()) {
cur = b->get_node(--i);
}
}
block_hrp_index = i;
}
void PhaseChaitin::print_pressure_info(Pressure& pressure, const char *str) {
if (str != NULL) {
tty->print_cr("# *** %s ***", str);
}
tty->print_cr("# start pressure is = %d", pressure.start_pressure());
tty->print_cr("# max pressure is = %d", pressure.final_pressure());
tty->print_cr("# end pressure is = %d", pressure.current_pressure());
tty->print_cr("#");
}
/* Build an interference graph:
* That is, if 2 live ranges are simultaneously alive but in their acceptable
* register sets do not overlap, then they do not interfere. The IFG is built
* by a single reverse pass over each basic block. Starting with the known
* live-out set, we remove things that get defined and add things that become
* live (essentially executing one pass of a standard LIVE analysis). Just
* before a Node defines a value (and removes it from the live-ness set) that
* value is certainly live. The defined value interferes with everything
* currently live. The value is then removed from the live-ness set and it's
* inputs are added to the live-ness set.
* Compute register pressure for each block:
* We store the biggest register pressure for each block and also the first
* low to high register pressure transition within the block (if any).
*/
uint PhaseChaitin::build_ifg_physical( ResourceArea *a ) {
Compile::TracePhase tp("buildIFG", &timers[_t_buildIFGphysical]);
uint must_spill = 0;
for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
Block* block = _cfg.get_block(i);
// Clone (rather than smash in place) the liveout info, so it is alive
// for the "collect_gc_info" phase later.
IndexSet liveout(_live->live(block));
uint first_inst = first_nonphi_index(block);
uint last_inst = block->end_idx();
move_exception_node_up(block, first_inst, last_inst);
Pressure int_pressure(last_inst + 1, Matcher::int_pressure_limit());
Pressure float_pressure(last_inst + 1, Matcher::float_pressure_limit());
block->_reg_pressure = 0;
block->_freg_pressure = 0;
int inst_count = last_inst - first_inst;
double cost = (inst_count <= 0) ? 0.0 : block->_freq * double(inst_count);
assert(cost >= 0.0, "negative spill cost" );
compute_initial_block_pressure(block, &liveout, int_pressure, float_pressure, cost);
for (uint location = last_inst; location > 0; location--) {
Node* n = block->get_node(location);
uint lid = _lrg_map.live_range_id(n);
if (lid) {
LRG& lrg = lrgs(lid);
// A DEF normally costs block frequency; rematerialized values are
// removed from the DEF sight, so LOWER costs here.
lrg._cost += n->rematerialize() ? 0 : block->_freq;
if (!liveout.member(lid) && n->Opcode() != Op_SafePoint) {
if (remove_node_if_not_used(block, location, n, lid, &liveout)) {
float_pressure.lower_high_pressure_index();
int_pressure.lower_high_pressure_index();
continue;
}
if (lrg._fat_proj) {
check_for_high_pressure_transition_at_fatproj(block->_reg_pressure, location, lrg, int_pressure, Op_RegI);
check_for_high_pressure_transition_at_fatproj(block->_freg_pressure, location, lrg, float_pressure, Op_RegD);
}
} else {
// A live range ends at its definition, remove the remaining area.
// If the cost is +Inf (which might happen in extreme cases), the lrg area will also be +Inf,
// and +Inf - +Inf = NaN. So let's not do that subtraction.
if (g_isfinite(cost)) {
lrg._area -= cost;
}
assert(lrg._area >= 0.0, "unexpected spill area value %g (rounding mode %x)", lrg._area, fegetround());
assign_high_score_to_immediate_copies(block, n, lrg, location + 1, last_inst);
if (liveout.remove(lid)) {
lower_pressure(block, location, lrg, &liveout, int_pressure, float_pressure);
}
uint copy_idx = n->is_Copy();
if (copy_idx) {
uint lid_copy = _lrg_map.live_range_id(n->in(copy_idx));
remove_interference_from_copy(block, location, lid_copy, &liveout, cost, int_pressure, float_pressure);
}
}
// Since rematerializable DEFs are not bound but the live range is,
// some uses must be bound. If we spill live range 'r', it can
// rematerialize at each use site according to its bindings.
if (lrg.is_bound() && !n->rematerialize() && lrg.mask().is_NotEmpty()) {
remove_bound_register_from_interfering_live_ranges(lrg, &liveout, must_spill);
}
interfere_with_live(lid, &liveout);
}
// Area remaining in the block
inst_count--;
cost = (inst_count <= 0) ? 0.0 : block->_freq * double(inst_count);
if (!n->is_Phi()) {
add_input_to_liveout(block, n, &liveout, cost, int_pressure, float_pressure);
}
}
check_for_high_pressure_block(int_pressure);
check_for_high_pressure_block(float_pressure);
adjust_high_pressure_index(block, block->_ihrp_index, int_pressure);
adjust_high_pressure_index(block, block->_fhrp_index, float_pressure);
// set the final_pressure as the register pressure for the block
block->_reg_pressure = int_pressure.final_pressure();
block->_freg_pressure = float_pressure.final_pressure();
#ifndef PRODUCT
// Gather Register Pressure Statistics
if (PrintOptoStatistics) {
if (block->_reg_pressure > int_pressure.high_pressure_limit() || block->_freg_pressure > float_pressure.high_pressure_limit()) {
_high_pressure++;
} else {
_low_pressure++;
}
}
#endif
}
return must_spill;
}
¤ Dauer der Verarbeitung: 0.29 Sekunden
(vorverarbeitet)
¤
|
Haftungshinweis
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung ist noch experimentell.
|
