mirror of
https://github.com/openjdk/jdk.git
synced 2025-08-28 07:14:30 +02:00
8022284: Hide internal data structure in PhaseCFG
Hide private node to block mapping using public interface Reviewed-by: kvn, roland
This commit is contained in:
parent
7107241178
commit
bfe8385fa9
19 changed files with 327 additions and 280 deletions
|
@ -44,7 +44,7 @@ public class PhaseCFG extends Phase {
|
||||||
Type type = db.lookupType("PhaseCFG");
|
Type type = db.lookupType("PhaseCFG");
|
||||||
numBlocksField = new CIntField(type.getCIntegerField("_num_blocks"), 0);
|
numBlocksField = new CIntField(type.getCIntegerField("_num_blocks"), 0);
|
||||||
blocksField = type.getAddressField("_blocks");
|
blocksField = type.getAddressField("_blocks");
|
||||||
bbsField = type.getAddressField("_bbs");
|
bbsField = type.getAddressField("_node_to_block_mapping");
|
||||||
brootField = type.getAddressField("_broot");
|
brootField = type.getAddressField("_broot");
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
@ -221,7 +221,7 @@ bool Block::has_uncommon_code() const {
|
||||||
//------------------------------is_uncommon------------------------------------
|
//------------------------------is_uncommon------------------------------------
|
||||||
// True if block is low enough frequency or guarded by a test which
|
// True if block is low enough frequency or guarded by a test which
|
||||||
// mostly does not go here.
|
// mostly does not go here.
|
||||||
bool Block::is_uncommon( Block_Array &bbs ) const {
|
bool Block::is_uncommon(PhaseCFG* cfg) const {
|
||||||
// Initial blocks must never be moved, so are never uncommon.
|
// Initial blocks must never be moved, so are never uncommon.
|
||||||
if (head()->is_Root() || head()->is_Start()) return false;
|
if (head()->is_Root() || head()->is_Start()) return false;
|
||||||
|
|
||||||
|
@ -238,7 +238,7 @@ bool Block::is_uncommon( Block_Array &bbs ) const {
|
||||||
uint uncommon_for_freq_preds = 0;
|
uint uncommon_for_freq_preds = 0;
|
||||||
|
|
||||||
for( uint i=1; i<num_preds(); i++ ) {
|
for( uint i=1; i<num_preds(); i++ ) {
|
||||||
Block* guard = bbs[pred(i)->_idx];
|
Block* guard = cfg->get_block_for_node(pred(i));
|
||||||
// Check to see if this block follows its guard 1 time out of 10000
|
// Check to see if this block follows its guard 1 time out of 10000
|
||||||
// or less.
|
// or less.
|
||||||
//
|
//
|
||||||
|
@ -285,11 +285,11 @@ void Block::dump_bidx(const Block* orig, outputStream* st) const {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
void Block::dump_pred(const Block_Array *bbs, Block* orig, outputStream* st) const {
|
void Block::dump_pred(const PhaseCFG* cfg, Block* orig, outputStream* st) const {
|
||||||
if (is_connector()) {
|
if (is_connector()) {
|
||||||
for (uint i=1; i<num_preds(); i++) {
|
for (uint i=1; i<num_preds(); i++) {
|
||||||
Block *p = ((*bbs)[pred(i)->_idx]);
|
Block *p = cfg->get_block_for_node(pred(i));
|
||||||
p->dump_pred(bbs, orig, st);
|
p->dump_pred(cfg, orig, st);
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
dump_bidx(orig, st);
|
dump_bidx(orig, st);
|
||||||
|
@ -297,7 +297,7 @@ void Block::dump_pred(const Block_Array *bbs, Block* orig, outputStream* st) con
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
void Block::dump_head( const Block_Array *bbs, outputStream* st ) const {
|
void Block::dump_head(const PhaseCFG* cfg, outputStream* st) const {
|
||||||
// Print the basic block
|
// Print the basic block
|
||||||
dump_bidx(this, st);
|
dump_bidx(this, st);
|
||||||
st->print(": #\t");
|
st->print(": #\t");
|
||||||
|
@ -311,26 +311,28 @@ void Block::dump_head( const Block_Array *bbs, outputStream* st ) const {
|
||||||
if( head()->is_block_start() ) {
|
if( head()->is_block_start() ) {
|
||||||
for (uint i=1; i<num_preds(); i++) {
|
for (uint i=1; i<num_preds(); i++) {
|
||||||
Node *s = pred(i);
|
Node *s = pred(i);
|
||||||
if (bbs) {
|
if (cfg != NULL) {
|
||||||
Block *p = (*bbs)[s->_idx];
|
Block *p = cfg->get_block_for_node(s);
|
||||||
p->dump_pred(bbs, p, st);
|
p->dump_pred(cfg, p, st);
|
||||||
} else {
|
} else {
|
||||||
while (!s->is_block_start())
|
while (!s->is_block_start())
|
||||||
s = s->in(0);
|
s = s->in(0);
|
||||||
st->print("N%d ", s->_idx );
|
st->print("N%d ", s->_idx );
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
} else
|
} else {
|
||||||
st->print("BLOCK HEAD IS JUNK ");
|
st->print("BLOCK HEAD IS JUNK ");
|
||||||
|
}
|
||||||
|
|
||||||
// Print loop, if any
|
// Print loop, if any
|
||||||
const Block *bhead = this; // Head of self-loop
|
const Block *bhead = this; // Head of self-loop
|
||||||
Node *bh = bhead->head();
|
Node *bh = bhead->head();
|
||||||
if( bbs && bh->is_Loop() && !head()->is_Root() ) {
|
|
||||||
|
if ((cfg != NULL) && bh->is_Loop() && !head()->is_Root()) {
|
||||||
LoopNode *loop = bh->as_Loop();
|
LoopNode *loop = bh->as_Loop();
|
||||||
const Block *bx = (*bbs)[loop->in(LoopNode::LoopBackControl)->_idx];
|
const Block *bx = cfg->get_block_for_node(loop->in(LoopNode::LoopBackControl));
|
||||||
while (bx->is_connector()) {
|
while (bx->is_connector()) {
|
||||||
bx = (*bbs)[bx->pred(1)->_idx];
|
bx = cfg->get_block_for_node(bx->pred(1));
|
||||||
}
|
}
|
||||||
st->print("\tLoop: B%d-B%d ", bhead->_pre_order, bx->_pre_order);
|
st->print("\tLoop: B%d-B%d ", bhead->_pre_order, bx->_pre_order);
|
||||||
// Dump any loop-specific bits, especially for CountedLoops.
|
// Dump any loop-specific bits, especially for CountedLoops.
|
||||||
|
@ -349,29 +351,32 @@ void Block::dump_head( const Block_Array *bbs, outputStream* st ) const {
|
||||||
st->print_cr("");
|
st->print_cr("");
|
||||||
}
|
}
|
||||||
|
|
||||||
void Block::dump() const { dump(NULL); }
|
void Block::dump() const {
|
||||||
|
dump(NULL);
|
||||||
|
}
|
||||||
|
|
||||||
void Block::dump( const Block_Array *bbs ) const {
|
void Block::dump(const PhaseCFG* cfg) const {
|
||||||
dump_head(bbs);
|
dump_head(cfg);
|
||||||
uint cnt = _nodes.size();
|
for (uint i=0; i< _nodes.size(); i++) {
|
||||||
for( uint i=0; i<cnt; i++ )
|
|
||||||
_nodes[i]->dump();
|
_nodes[i]->dump();
|
||||||
|
}
|
||||||
tty->print("\n");
|
tty->print("\n");
|
||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
//=============================================================================
|
//=============================================================================
|
||||||
//------------------------------PhaseCFG---------------------------------------
|
//------------------------------PhaseCFG---------------------------------------
|
||||||
PhaseCFG::PhaseCFG( Arena *a, RootNode *r, Matcher &m ) :
|
PhaseCFG::PhaseCFG(Arena* arena, RootNode* root, Matcher& matcher)
|
||||||
Phase(CFG),
|
: Phase(CFG)
|
||||||
_bbs(a),
|
, _block_arena(arena)
|
||||||
_root(r),
|
, _node_to_block_mapping(arena)
|
||||||
_node_latency(NULL)
|
, _root(root)
|
||||||
|
, _node_latency(NULL)
|
||||||
#ifndef PRODUCT
|
#ifndef PRODUCT
|
||||||
, _trace_opto_pipelining(TraceOptoPipelining || C->method_has_option("TraceOptoPipelining"))
|
, _trace_opto_pipelining(TraceOptoPipelining || C->method_has_option("TraceOptoPipelining"))
|
||||||
#endif
|
#endif
|
||||||
#ifdef ASSERT
|
#ifdef ASSERT
|
||||||
, _raw_oops(a)
|
, _raw_oops(arena)
|
||||||
#endif
|
#endif
|
||||||
{
|
{
|
||||||
ResourceMark rm;
|
ResourceMark rm;
|
||||||
|
@ -380,13 +385,13 @@ PhaseCFG::PhaseCFG( Arena *a, RootNode *r, Matcher &m ) :
|
||||||
// Node on demand.
|
// Node on demand.
|
||||||
Node *x = new (C) GotoNode(NULL);
|
Node *x = new (C) GotoNode(NULL);
|
||||||
x->init_req(0, x);
|
x->init_req(0, x);
|
||||||
_goto = m.match_tree(x);
|
_goto = matcher.match_tree(x);
|
||||||
assert(_goto != NULL, "");
|
assert(_goto != NULL, "");
|
||||||
_goto->set_req(0,_goto);
|
_goto->set_req(0,_goto);
|
||||||
|
|
||||||
// Build the CFG in Reverse Post Order
|
// Build the CFG in Reverse Post Order
|
||||||
_num_blocks = build_cfg();
|
_num_blocks = build_cfg();
|
||||||
_broot = _bbs[_root->_idx];
|
_broot = get_block_for_node(_root);
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------build_cfg--------------------------------------
|
//------------------------------build_cfg--------------------------------------
|
||||||
|
@ -440,9 +445,9 @@ uint PhaseCFG::build_cfg() {
|
||||||
// 'p' now points to the start of this basic block
|
// 'p' now points to the start of this basic block
|
||||||
|
|
||||||
// Put self in array of basic blocks
|
// Put self in array of basic blocks
|
||||||
Block *bb = new (_bbs._arena) Block(_bbs._arena,p);
|
Block *bb = new (_block_arena) Block(_block_arena, p);
|
||||||
_bbs.map(p->_idx,bb);
|
map_node_to_block(p, bb);
|
||||||
_bbs.map(x->_idx,bb);
|
map_node_to_block(x, bb);
|
||||||
if( x != p ) { // Only for root is x == p
|
if( x != p ) { // Only for root is x == p
|
||||||
bb->_nodes.push((Node*)x);
|
bb->_nodes.push((Node*)x);
|
||||||
}
|
}
|
||||||
|
@ -473,16 +478,16 @@ uint PhaseCFG::build_cfg() {
|
||||||
// Check if it the fist node pushed on stack at the beginning.
|
// Check if it the fist node pushed on stack at the beginning.
|
||||||
if (idx == 0) break; // end of the build
|
if (idx == 0) break; // end of the build
|
||||||
// Find predecessor basic block
|
// Find predecessor basic block
|
||||||
Block *pb = _bbs[x->_idx];
|
Block *pb = get_block_for_node(x);
|
||||||
// Insert into nodes array, if not already there
|
// Insert into nodes array, if not already there
|
||||||
if( !_bbs.lookup(proj->_idx) ) {
|
if (!has_block(proj)) {
|
||||||
assert( x != proj, "" );
|
assert( x != proj, "" );
|
||||||
// Map basic block of projection
|
// Map basic block of projection
|
||||||
_bbs.map(proj->_idx,pb);
|
map_node_to_block(proj, pb);
|
||||||
pb->_nodes.push(proj);
|
pb->_nodes.push(proj);
|
||||||
}
|
}
|
||||||
// Insert self as a child of my predecessor block
|
// Insert self as a child of my predecessor block
|
||||||
pb->_succs.map(pb->_num_succs++, _bbs[np->_idx]);
|
pb->_succs.map(pb->_num_succs++, get_block_for_node(np));
|
||||||
assert( pb->_nodes[ pb->_nodes.size() - pb->_num_succs ]->is_block_proj(),
|
assert( pb->_nodes[ pb->_nodes.size() - pb->_num_succs ]->is_block_proj(),
|
||||||
"too many control users, not a CFG?" );
|
"too many control users, not a CFG?" );
|
||||||
}
|
}
|
||||||
|
@ -511,15 +516,15 @@ void PhaseCFG::insert_goto_at(uint block_no, uint succ_no) {
|
||||||
RegionNode* region = new (C) RegionNode(2);
|
RegionNode* region = new (C) RegionNode(2);
|
||||||
region->init_req(1, proj);
|
region->init_req(1, proj);
|
||||||
// setup corresponding basic block
|
// setup corresponding basic block
|
||||||
Block* block = new (_bbs._arena) Block(_bbs._arena, region);
|
Block* block = new (_block_arena) Block(_block_arena, region);
|
||||||
_bbs.map(region->_idx, block);
|
map_node_to_block(region, block);
|
||||||
C->regalloc()->set_bad(region->_idx);
|
C->regalloc()->set_bad(region->_idx);
|
||||||
// add a goto node
|
// add a goto node
|
||||||
Node* gto = _goto->clone(); // get a new goto node
|
Node* gto = _goto->clone(); // get a new goto node
|
||||||
gto->set_req(0, region);
|
gto->set_req(0, region);
|
||||||
// add it to the basic block
|
// add it to the basic block
|
||||||
block->_nodes.push(gto);
|
block->_nodes.push(gto);
|
||||||
_bbs.map(gto->_idx, block);
|
map_node_to_block(gto, block);
|
||||||
C->regalloc()->set_bad(gto->_idx);
|
C->regalloc()->set_bad(gto->_idx);
|
||||||
// hook up successor block
|
// hook up successor block
|
||||||
block->_succs.map(block->_num_succs++, out);
|
block->_succs.map(block->_num_succs++, out);
|
||||||
|
@ -570,7 +575,7 @@ void PhaseCFG::convert_NeverBranch_to_Goto(Block *b) {
|
||||||
gto->set_req(0, b->head());
|
gto->set_req(0, b->head());
|
||||||
Node *bp = b->_nodes[end_idx];
|
Node *bp = b->_nodes[end_idx];
|
||||||
b->_nodes.map(end_idx,gto); // Slam over NeverBranch
|
b->_nodes.map(end_idx,gto); // Slam over NeverBranch
|
||||||
_bbs.map(gto->_idx, b);
|
map_node_to_block(gto, b);
|
||||||
C->regalloc()->set_bad(gto->_idx);
|
C->regalloc()->set_bad(gto->_idx);
|
||||||
b->_nodes.pop(); // Yank projections
|
b->_nodes.pop(); // Yank projections
|
||||||
b->_nodes.pop(); // Yank projections
|
b->_nodes.pop(); // Yank projections
|
||||||
|
@ -613,7 +618,7 @@ bool PhaseCFG::move_to_next(Block* bx, uint b_index) {
|
||||||
// If the previous block conditionally falls into bx, return false,
|
// If the previous block conditionally falls into bx, return false,
|
||||||
// because moving bx will create an extra jump.
|
// because moving bx will create an extra jump.
|
||||||
for(uint k = 1; k < bx->num_preds(); k++ ) {
|
for(uint k = 1; k < bx->num_preds(); k++ ) {
|
||||||
Block* pred = _bbs[bx->pred(k)->_idx];
|
Block* pred = get_block_for_node(bx->pred(k));
|
||||||
if (pred == _blocks[bx_index-1]) {
|
if (pred == _blocks[bx_index-1]) {
|
||||||
if (pred->_num_succs != 1) {
|
if (pred->_num_succs != 1) {
|
||||||
return false;
|
return false;
|
||||||
|
@ -682,7 +687,7 @@ void PhaseCFG::remove_empty() {
|
||||||
|
|
||||||
// Look for uncommon blocks and move to end.
|
// Look for uncommon blocks and move to end.
|
||||||
if (!C->do_freq_based_layout()) {
|
if (!C->do_freq_based_layout()) {
|
||||||
if( b->is_uncommon(_bbs) ) {
|
if (b->is_uncommon(this)) {
|
||||||
move_to_end(b, i);
|
move_to_end(b, i);
|
||||||
last--; // No longer check for being uncommon!
|
last--; // No longer check for being uncommon!
|
||||||
if( no_flip_branch(b) ) { // Fall-thru case must follow?
|
if( no_flip_branch(b) ) { // Fall-thru case must follow?
|
||||||
|
@ -870,28 +875,31 @@ void PhaseCFG::_dump_cfg( const Node *end, VectorSet &visited ) const {
|
||||||
} while( !p->is_block_start() );
|
} while( !p->is_block_start() );
|
||||||
|
|
||||||
// Recursively visit
|
// Recursively visit
|
||||||
for( uint i=1; i<p->req(); i++ )
|
for (uint i = 1; i < p->req(); i++) {
|
||||||
_dump_cfg(p->in(i),visited);
|
_dump_cfg(p->in(i), visited);
|
||||||
|
}
|
||||||
|
|
||||||
// Dump the block
|
// Dump the block
|
||||||
_bbs[p->_idx]->dump(&_bbs);
|
get_block_for_node(p)->dump(this);
|
||||||
}
|
}
|
||||||
|
|
||||||
void PhaseCFG::dump( ) const {
|
void PhaseCFG::dump( ) const {
|
||||||
tty->print("\n--- CFG --- %d BBs\n",_num_blocks);
|
tty->print("\n--- CFG --- %d BBs\n",_num_blocks);
|
||||||
if( _blocks.size() ) { // Did we do basic-block layout?
|
if (_blocks.size()) { // Did we do basic-block layout?
|
||||||
for( uint i=0; i<_num_blocks; i++ )
|
for (uint i = 0; i < _num_blocks; i++) {
|
||||||
_blocks[i]->dump(&_bbs);
|
_blocks[i]->dump(this);
|
||||||
|
}
|
||||||
} else { // Else do it with a DFS
|
} else { // Else do it with a DFS
|
||||||
VectorSet visited(_bbs._arena);
|
VectorSet visited(_block_arena);
|
||||||
_dump_cfg(_root,visited);
|
_dump_cfg(_root,visited);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
void PhaseCFG::dump_headers() {
|
void PhaseCFG::dump_headers() {
|
||||||
for( uint i = 0; i < _num_blocks; i++ ) {
|
for( uint i = 0; i < _num_blocks; i++ ) {
|
||||||
if( _blocks[i] == NULL ) continue;
|
if (_blocks[i]) {
|
||||||
_blocks[i]->dump_head(&_bbs);
|
_blocks[i]->dump_head(this);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -904,7 +912,7 @@ void PhaseCFG::verify( ) const {
|
||||||
uint j;
|
uint j;
|
||||||
for (j = 0; j < cnt; j++) {
|
for (j = 0; j < cnt; j++) {
|
||||||
Node *n = b->_nodes[j];
|
Node *n = b->_nodes[j];
|
||||||
assert( _bbs[n->_idx] == b, "" );
|
assert(get_block_for_node(n) == b, "");
|
||||||
if (j >= 1 && n->is_Mach() &&
|
if (j >= 1 && n->is_Mach() &&
|
||||||
n->as_Mach()->ideal_Opcode() == Op_CreateEx) {
|
n->as_Mach()->ideal_Opcode() == Op_CreateEx) {
|
||||||
assert(j == 1 || b->_nodes[j-1]->is_Phi(),
|
assert(j == 1 || b->_nodes[j-1]->is_Phi(),
|
||||||
|
@ -913,13 +921,12 @@ void PhaseCFG::verify( ) const {
|
||||||
for (uint k = 0; k < n->req(); k++) {
|
for (uint k = 0; k < n->req(); k++) {
|
||||||
Node *def = n->in(k);
|
Node *def = n->in(k);
|
||||||
if (def && def != n) {
|
if (def && def != n) {
|
||||||
assert(_bbs[def->_idx] || def->is_Con(),
|
assert(get_block_for_node(def) || def->is_Con(), "must have block; constants for debug info ok");
|
||||||
"must have block; constants for debug info ok");
|
|
||||||
// Verify that instructions in the block is in correct order.
|
// Verify that instructions in the block is in correct order.
|
||||||
// Uses must follow their definition if they are at the same block.
|
// Uses must follow their definition if they are at the same block.
|
||||||
// Mostly done to check that MachSpillCopy nodes are placed correctly
|
// Mostly done to check that MachSpillCopy nodes are placed correctly
|
||||||
// when CreateEx node is moved in build_ifg_physical().
|
// when CreateEx node is moved in build_ifg_physical().
|
||||||
if (_bbs[def->_idx] == b &&
|
if (get_block_for_node(def) == b &&
|
||||||
!(b->head()->is_Loop() && n->is_Phi()) &&
|
!(b->head()->is_Loop() && n->is_Phi()) &&
|
||||||
// See (+++) comment in reg_split.cpp
|
// See (+++) comment in reg_split.cpp
|
||||||
!(n->jvms() != NULL && n->jvms()->is_monitor_use(k))) {
|
!(n->jvms() != NULL && n->jvms()->is_monitor_use(k))) {
|
||||||
|
|
|
@ -48,13 +48,12 @@ class Block_Array : public ResourceObj {
|
||||||
friend class VMStructs;
|
friend class VMStructs;
|
||||||
uint _size; // allocated size, as opposed to formal limit
|
uint _size; // allocated size, as opposed to formal limit
|
||||||
debug_only(uint _limit;) // limit to formal domain
|
debug_only(uint _limit;) // limit to formal domain
|
||||||
|
Arena *_arena; // Arena to allocate in
|
||||||
protected:
|
protected:
|
||||||
Block **_blocks;
|
Block **_blocks;
|
||||||
void grow( uint i ); // Grow array node to fit
|
void grow( uint i ); // Grow array node to fit
|
||||||
|
|
||||||
public:
|
public:
|
||||||
Arena *_arena; // Arena to allocate in
|
|
||||||
|
|
||||||
Block_Array(Arena *a) : _arena(a), _size(OptoBlockListSize) {
|
Block_Array(Arena *a) : _arena(a), _size(OptoBlockListSize) {
|
||||||
debug_only(_limit=0);
|
debug_only(_limit=0);
|
||||||
_blocks = NEW_ARENA_ARRAY( a, Block *, OptoBlockListSize );
|
_blocks = NEW_ARENA_ARRAY( a, Block *, OptoBlockListSize );
|
||||||
|
@ -284,15 +283,15 @@ class Block : public CFGElement {
|
||||||
// helper function that adds caller save registers to MachProjNode
|
// helper function that adds caller save registers to MachProjNode
|
||||||
void add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe);
|
void add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe);
|
||||||
// Schedule a call next in the block
|
// Schedule a call next in the block
|
||||||
uint sched_call(Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call);
|
uint sched_call(Matcher &matcher, PhaseCFG* cfg, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call);
|
||||||
|
|
||||||
// Perform basic-block local scheduling
|
// Perform basic-block local scheduling
|
||||||
Node *select(PhaseCFG *cfg, Node_List &worklist, GrowableArray<int> &ready_cnt, VectorSet &next_call, uint sched_slot);
|
Node *select(PhaseCFG *cfg, Node_List &worklist, GrowableArray<int> &ready_cnt, VectorSet &next_call, uint sched_slot);
|
||||||
void set_next_call( Node *n, VectorSet &next_call, Block_Array &bbs );
|
void set_next_call( Node *n, VectorSet &next_call, PhaseCFG* cfg);
|
||||||
void needed_for_next_call(Node *this_call, VectorSet &next_call, Block_Array &bbs);
|
void needed_for_next_call(Node *this_call, VectorSet &next_call, PhaseCFG* cfg);
|
||||||
bool schedule_local(PhaseCFG *cfg, Matcher &m, GrowableArray<int> &ready_cnt, VectorSet &next_call);
|
bool schedule_local(PhaseCFG *cfg, Matcher &m, GrowableArray<int> &ready_cnt, VectorSet &next_call);
|
||||||
// Cleanup if any code lands between a Call and his Catch
|
// Cleanup if any code lands between a Call and his Catch
|
||||||
void call_catch_cleanup(Block_Array &bbs, Compile *C);
|
void call_catch_cleanup(PhaseCFG* cfg, Compile *C);
|
||||||
// Detect implicit-null-check opportunities. Basically, find NULL checks
|
// Detect implicit-null-check opportunities. Basically, find NULL checks
|
||||||
// with suitable memory ops nearby. Use the memory op to do the NULL check.
|
// with suitable memory ops nearby. Use the memory op to do the NULL check.
|
||||||
// I can generate a memory op if there is not one nearby.
|
// I can generate a memory op if there is not one nearby.
|
||||||
|
@ -331,15 +330,15 @@ class Block : public CFGElement {
|
||||||
|
|
||||||
// Use frequency calculations and code shape to predict if the block
|
// Use frequency calculations and code shape to predict if the block
|
||||||
// is uncommon.
|
// is uncommon.
|
||||||
bool is_uncommon( Block_Array &bbs ) const;
|
bool is_uncommon(PhaseCFG* cfg) const;
|
||||||
|
|
||||||
#ifndef PRODUCT
|
#ifndef PRODUCT
|
||||||
// Debugging print of basic block
|
// Debugging print of basic block
|
||||||
void dump_bidx(const Block* orig, outputStream* st = tty) const;
|
void dump_bidx(const Block* orig, outputStream* st = tty) const;
|
||||||
void dump_pred(const Block_Array *bbs, Block* orig, outputStream* st = tty) const;
|
void dump_pred(const PhaseCFG* cfg, Block* orig, outputStream* st = tty) const;
|
||||||
void dump_head( const Block_Array *bbs, outputStream* st = tty ) const;
|
void dump_head(const PhaseCFG* cfg, outputStream* st = tty) const;
|
||||||
void dump() const;
|
void dump() const;
|
||||||
void dump( const Block_Array *bbs ) const;
|
void dump(const PhaseCFG* cfg) const;
|
||||||
#endif
|
#endif
|
||||||
};
|
};
|
||||||
|
|
||||||
|
@ -349,6 +348,12 @@ class Block : public CFGElement {
|
||||||
class PhaseCFG : public Phase {
|
class PhaseCFG : public Phase {
|
||||||
friend class VMStructs;
|
friend class VMStructs;
|
||||||
private:
|
private:
|
||||||
|
// Arena for the blocks to be stored in
|
||||||
|
Arena* _block_arena;
|
||||||
|
|
||||||
|
// Map nodes to owning basic block
|
||||||
|
Block_Array _node_to_block_mapping;
|
||||||
|
|
||||||
// Build a proper looking cfg. Return count of basic blocks
|
// Build a proper looking cfg. Return count of basic blocks
|
||||||
uint build_cfg();
|
uint build_cfg();
|
||||||
|
|
||||||
|
@ -371,22 +376,42 @@ class PhaseCFG : public Phase {
|
||||||
|
|
||||||
Block* insert_anti_dependences(Block* LCA, Node* load, bool verify = false);
|
Block* insert_anti_dependences(Block* LCA, Node* load, bool verify = false);
|
||||||
void verify_anti_dependences(Block* LCA, Node* load) {
|
void verify_anti_dependences(Block* LCA, Node* load) {
|
||||||
assert(LCA == _bbs[load->_idx], "should already be scheduled");
|
assert(LCA == get_block_for_node(load), "should already be scheduled");
|
||||||
insert_anti_dependences(LCA, load, true);
|
insert_anti_dependences(LCA, load, true);
|
||||||
}
|
}
|
||||||
|
|
||||||
public:
|
public:
|
||||||
PhaseCFG( Arena *a, RootNode *r, Matcher &m );
|
PhaseCFG(Arena* arena, RootNode* root, Matcher& matcher);
|
||||||
|
|
||||||
uint _num_blocks; // Count of basic blocks
|
uint _num_blocks; // Count of basic blocks
|
||||||
Block_List _blocks; // List of basic blocks
|
Block_List _blocks; // List of basic blocks
|
||||||
RootNode *_root; // Root of whole program
|
RootNode *_root; // Root of whole program
|
||||||
Block_Array _bbs; // Map Nodes to owning Basic Block
|
|
||||||
Block *_broot; // Basic block of root
|
Block *_broot; // Basic block of root
|
||||||
uint _rpo_ctr;
|
uint _rpo_ctr;
|
||||||
CFGLoop* _root_loop;
|
CFGLoop* _root_loop;
|
||||||
float _outer_loop_freq; // Outmost loop frequency
|
float _outer_loop_freq; // Outmost loop frequency
|
||||||
|
|
||||||
|
|
||||||
|
// set which block this node should reside in
|
||||||
|
void map_node_to_block(const Node* node, Block* block) {
|
||||||
|
_node_to_block_mapping.map(node->_idx, block);
|
||||||
|
}
|
||||||
|
|
||||||
|
// removes the mapping from a node to a block
|
||||||
|
void unmap_node_from_block(const Node* node) {
|
||||||
|
_node_to_block_mapping.map(node->_idx, NULL);
|
||||||
|
}
|
||||||
|
|
||||||
|
// get the block in which this node resides
|
||||||
|
Block* get_block_for_node(const Node* node) const {
|
||||||
|
return _node_to_block_mapping[node->_idx];
|
||||||
|
}
|
||||||
|
|
||||||
|
// does this node reside in a block; return true
|
||||||
|
bool has_block(const Node* node) const {
|
||||||
|
return (_node_to_block_mapping.lookup(node->_idx) != NULL);
|
||||||
|
}
|
||||||
|
|
||||||
// Per node latency estimation, valid only during GCM
|
// Per node latency estimation, valid only during GCM
|
||||||
GrowableArray<uint> *_node_latency;
|
GrowableArray<uint> *_node_latency;
|
||||||
|
|
||||||
|
@ -405,7 +430,7 @@ class PhaseCFG : public Phase {
|
||||||
void Estimate_Block_Frequency();
|
void Estimate_Block_Frequency();
|
||||||
|
|
||||||
// Global Code Motion. See Click's PLDI95 paper. Place Nodes in specific
|
// Global Code Motion. See Click's PLDI95 paper. Place Nodes in specific
|
||||||
// basic blocks; i.e. _bbs now maps _idx for all Nodes to some Block.
|
// basic blocks; i.e. _node_to_block_mapping now maps _idx for all Nodes to some Block.
|
||||||
void GlobalCodeMotion( Matcher &m, uint unique, Node_List &proj_list );
|
void GlobalCodeMotion( Matcher &m, uint unique, Node_List &proj_list );
|
||||||
|
|
||||||
// Compute the (backwards) latency of a node from the uses
|
// Compute the (backwards) latency of a node from the uses
|
||||||
|
@ -454,7 +479,7 @@ class PhaseCFG : public Phase {
|
||||||
// Insert a node into a block, and update the _bbs
|
// Insert a node into a block, and update the _bbs
|
||||||
void insert( Block *b, uint idx, Node *n ) {
|
void insert( Block *b, uint idx, Node *n ) {
|
||||||
b->_nodes.insert( idx, n );
|
b->_nodes.insert( idx, n );
|
||||||
_bbs.map( n->_idx, b );
|
map_node_to_block(n, b);
|
||||||
}
|
}
|
||||||
|
|
||||||
#ifndef PRODUCT
|
#ifndef PRODUCT
|
||||||
|
@ -543,7 +568,7 @@ class CFGLoop : public CFGElement {
|
||||||
_child(NULL),
|
_child(NULL),
|
||||||
_exit_prob(1.0f) {}
|
_exit_prob(1.0f) {}
|
||||||
CFGLoop* parent() { return _parent; }
|
CFGLoop* parent() { return _parent; }
|
||||||
void push_pred(Block* blk, int i, Block_List& worklist, Block_Array& node_to_blk);
|
void push_pred(Block* blk, int i, Block_List& worklist, PhaseCFG* cfg);
|
||||||
void add_member(CFGElement *s) { _members.push(s); }
|
void add_member(CFGElement *s) { _members.push(s); }
|
||||||
void add_nested_loop(CFGLoop* cl);
|
void add_nested_loop(CFGLoop* cl);
|
||||||
Block* head() {
|
Block* head() {
|
||||||
|
|
|
@ -426,14 +426,16 @@ static void do_liveness( PhaseRegAlloc *regalloc, PhaseCFG *cfg, Block_List *wor
|
||||||
}
|
}
|
||||||
memset( live, 0, cfg->_num_blocks * (max_reg_ints<<LogBytesPerInt) );
|
memset( live, 0, cfg->_num_blocks * (max_reg_ints<<LogBytesPerInt) );
|
||||||
// Push preds onto worklist
|
// Push preds onto worklist
|
||||||
for( uint i=1; i<root->req(); i++ )
|
for (uint i = 1; i < root->req(); i++) {
|
||||||
worklist->push(cfg->_bbs[root->in(i)->_idx]);
|
Block* block = cfg->get_block_for_node(root->in(i));
|
||||||
|
worklist->push(block);
|
||||||
|
}
|
||||||
|
|
||||||
// ZKM.jar includes tiny infinite loops which are unreached from below.
|
// ZKM.jar includes tiny infinite loops which are unreached from below.
|
||||||
// If we missed any blocks, we'll retry here after pushing all missed
|
// If we missed any blocks, we'll retry here after pushing all missed
|
||||||
// blocks on the worklist. Normally this outer loop never trips more
|
// blocks on the worklist. Normally this outer loop never trips more
|
||||||
// than once.
|
// than once.
|
||||||
while( 1 ) {
|
while (1) {
|
||||||
|
|
||||||
while( worklist->size() ) { // Standard worklist algorithm
|
while( worklist->size() ) { // Standard worklist algorithm
|
||||||
Block *b = worklist->rpop();
|
Block *b = worklist->rpop();
|
||||||
|
@ -537,8 +539,10 @@ static void do_liveness( PhaseRegAlloc *regalloc, PhaseCFG *cfg, Block_List *wor
|
||||||
for( l=0; l<max_reg_ints; l++ )
|
for( l=0; l<max_reg_ints; l++ )
|
||||||
old_live[l] = tmp_live[l];
|
old_live[l] = tmp_live[l];
|
||||||
// Push preds onto worklist
|
// Push preds onto worklist
|
||||||
for( l=1; l<(int)b->num_preds(); l++ )
|
for (l = 1; l < (int)b->num_preds(); l++) {
|
||||||
worklist->push(cfg->_bbs[b->pred(l)->_idx]);
|
Block* block = cfg->get_block_for_node(b->pred(l));
|
||||||
|
worklist->push(block);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -629,10 +633,9 @@ void Compile::BuildOopMaps() {
|
||||||
// pred to this block. Otherwise we have to grab a new OopFlow.
|
// pred to this block. Otherwise we have to grab a new OopFlow.
|
||||||
OopFlow *flow = NULL; // Flag for finding optimized flow
|
OopFlow *flow = NULL; // Flag for finding optimized flow
|
||||||
Block *pred = (Block*)0xdeadbeef;
|
Block *pred = (Block*)0xdeadbeef;
|
||||||
uint j;
|
|
||||||
// Scan this block's preds to find a done predecessor
|
// Scan this block's preds to find a done predecessor
|
||||||
for( j=1; j<b->num_preds(); j++ ) {
|
for (uint j = 1; j < b->num_preds(); j++) {
|
||||||
Block *p = _cfg->_bbs[b->pred(j)->_idx];
|
Block* p = _cfg->get_block_for_node(b->pred(j));
|
||||||
OopFlow *p_flow = flows[p->_pre_order];
|
OopFlow *p_flow = flows[p->_pre_order];
|
||||||
if( p_flow ) { // Predecessor is done
|
if( p_flow ) { // Predecessor is done
|
||||||
assert( p_flow->_b == p, "cross check" );
|
assert( p_flow->_b == p, "cross check" );
|
||||||
|
|
|
@ -295,7 +295,7 @@ void PhaseChaitin::new_lrg(const Node *x, uint lrg) {
|
||||||
|
|
||||||
|
|
||||||
bool PhaseChaitin::clone_projs_shared(Block *b, uint idx, Node *con, Node *copy, uint max_lrg_id) {
|
bool PhaseChaitin::clone_projs_shared(Block *b, uint idx, Node *con, Node *copy, uint max_lrg_id) {
|
||||||
Block *bcon = _cfg._bbs[con->_idx];
|
Block* bcon = _cfg.get_block_for_node(con);
|
||||||
uint cindex = bcon->find_node(con);
|
uint cindex = bcon->find_node(con);
|
||||||
Node *con_next = bcon->_nodes[cindex+1];
|
Node *con_next = bcon->_nodes[cindex+1];
|
||||||
if (con_next->in(0) != con || !con_next->is_MachProj()) {
|
if (con_next->in(0) != con || !con_next->is_MachProj()) {
|
||||||
|
@ -306,7 +306,7 @@ bool PhaseChaitin::clone_projs_shared(Block *b, uint idx, Node *con, Node *copy,
|
||||||
Node *kills = con_next->clone();
|
Node *kills = con_next->clone();
|
||||||
kills->set_req(0, copy);
|
kills->set_req(0, copy);
|
||||||
b->_nodes.insert(idx, kills);
|
b->_nodes.insert(idx, kills);
|
||||||
_cfg._bbs.map(kills->_idx, b);
|
_cfg.map_node_to_block(kills, b);
|
||||||
new_lrg(kills, max_lrg_id);
|
new_lrg(kills, max_lrg_id);
|
||||||
return true;
|
return true;
|
||||||
}
|
}
|
||||||
|
@ -962,8 +962,7 @@ void PhaseChaitin::gather_lrg_masks( bool after_aggressive ) {
|
||||||
// AggressiveCoalesce. This effectively pre-virtual-splits
|
// AggressiveCoalesce. This effectively pre-virtual-splits
|
||||||
// around uncommon uses of common defs.
|
// around uncommon uses of common defs.
|
||||||
const RegMask &rm = n->in_RegMask(k);
|
const RegMask &rm = n->in_RegMask(k);
|
||||||
if( !after_aggressive &&
|
if (!after_aggressive && _cfg.get_block_for_node(n->in(k))->_freq > 1000 * b->_freq) {
|
||||||
_cfg._bbs[n->in(k)->_idx]->_freq > 1000*b->_freq ) {
|
|
||||||
// Since we are BEFORE aggressive coalesce, leave the register
|
// Since we are BEFORE aggressive coalesce, leave the register
|
||||||
// mask untrimmed by the call. This encourages more coalescing.
|
// mask untrimmed by the call. This encourages more coalescing.
|
||||||
// Later, AFTER aggressive, this live range will have to spill
|
// Later, AFTER aggressive, this live range will have to spill
|
||||||
|
@ -1709,16 +1708,15 @@ Node *PhaseChaitin::find_base_for_derived( Node **derived_base_map, Node *derive
|
||||||
// set control to _root and place it into Start block
|
// set control to _root and place it into Start block
|
||||||
// (where top() node is placed).
|
// (where top() node is placed).
|
||||||
base->init_req(0, _cfg._root);
|
base->init_req(0, _cfg._root);
|
||||||
Block *startb = _cfg._bbs[C->top()->_idx];
|
Block *startb = _cfg.get_block_for_node(C->top());
|
||||||
startb->_nodes.insert(startb->find_node(C->top()), base );
|
startb->_nodes.insert(startb->find_node(C->top()), base );
|
||||||
_cfg._bbs.map( base->_idx, startb );
|
_cfg.map_node_to_block(base, startb);
|
||||||
assert(_lrg_map.live_range_id(base) == 0, "should not have LRG yet");
|
assert(_lrg_map.live_range_id(base) == 0, "should not have LRG yet");
|
||||||
}
|
}
|
||||||
if (_lrg_map.live_range_id(base) == 0) {
|
if (_lrg_map.live_range_id(base) == 0) {
|
||||||
new_lrg(base, maxlrg++);
|
new_lrg(base, maxlrg++);
|
||||||
}
|
}
|
||||||
assert(base->in(0) == _cfg._root &&
|
assert(base->in(0) == _cfg._root && _cfg.get_block_for_node(base) == _cfg.get_block_for_node(C->top()), "base NULL should be shared");
|
||||||
_cfg._bbs[base->_idx] == _cfg._bbs[C->top()->_idx], "base NULL should be shared");
|
|
||||||
derived_base_map[derived->_idx] = base;
|
derived_base_map[derived->_idx] = base;
|
||||||
return base;
|
return base;
|
||||||
}
|
}
|
||||||
|
@ -1754,12 +1752,12 @@ Node *PhaseChaitin::find_base_for_derived( Node **derived_base_map, Node *derive
|
||||||
base->as_Phi()->set_type(t);
|
base->as_Phi()->set_type(t);
|
||||||
|
|
||||||
// Search the current block for an existing base-Phi
|
// Search the current block for an existing base-Phi
|
||||||
Block *b = _cfg._bbs[derived->_idx];
|
Block *b = _cfg.get_block_for_node(derived);
|
||||||
for( i = 1; i <= b->end_idx(); i++ ) {// Search for matching Phi
|
for( i = 1; i <= b->end_idx(); i++ ) {// Search for matching Phi
|
||||||
Node *phi = b->_nodes[i];
|
Node *phi = b->_nodes[i];
|
||||||
if( !phi->is_Phi() ) { // Found end of Phis with no match?
|
if( !phi->is_Phi() ) { // Found end of Phis with no match?
|
||||||
b->_nodes.insert( i, base ); // Must insert created Phi here as base
|
b->_nodes.insert( i, base ); // Must insert created Phi here as base
|
||||||
_cfg._bbs.map( base->_idx, b );
|
_cfg.map_node_to_block(base, b);
|
||||||
new_lrg(base,maxlrg++);
|
new_lrg(base,maxlrg++);
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
|
@ -1815,8 +1813,8 @@ bool PhaseChaitin::stretch_base_pointer_live_ranges(ResourceArea *a) {
|
||||||
if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_CmpI ) {
|
if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_CmpI ) {
|
||||||
Node *phi = n->in(1);
|
Node *phi = n->in(1);
|
||||||
if( phi->is_Phi() && phi->as_Phi()->region()->is_Loop() ) {
|
if( phi->is_Phi() && phi->as_Phi()->region()->is_Loop() ) {
|
||||||
Block *phi_block = _cfg._bbs[phi->_idx];
|
Block *phi_block = _cfg.get_block_for_node(phi);
|
||||||
if( _cfg._bbs[phi_block->pred(2)->_idx] == b ) {
|
if (_cfg.get_block_for_node(phi_block->pred(2)) == b) {
|
||||||
const RegMask *mask = C->matcher()->idealreg2spillmask[Op_RegI];
|
const RegMask *mask = C->matcher()->idealreg2spillmask[Op_RegI];
|
||||||
Node *spill = new (C) MachSpillCopyNode( phi, *mask, *mask );
|
Node *spill = new (C) MachSpillCopyNode( phi, *mask, *mask );
|
||||||
insert_proj( phi_block, 1, spill, maxlrg++ );
|
insert_proj( phi_block, 1, spill, maxlrg++ );
|
||||||
|
@ -1870,7 +1868,7 @@ bool PhaseChaitin::stretch_base_pointer_live_ranges(ResourceArea *a) {
|
||||||
if ((_lrg_map.live_range_id(base) >= _lrg_map.max_lrg_id() || // (Brand new base (hence not live) or
|
if ((_lrg_map.live_range_id(base) >= _lrg_map.max_lrg_id() || // (Brand new base (hence not live) or
|
||||||
!liveout.member(_lrg_map.live_range_id(base))) && // not live) AND
|
!liveout.member(_lrg_map.live_range_id(base))) && // not live) AND
|
||||||
(_lrg_map.live_range_id(base) > 0) && // not a constant
|
(_lrg_map.live_range_id(base) > 0) && // not a constant
|
||||||
_cfg._bbs[base->_idx] != b) { // base not def'd in blk)
|
_cfg.get_block_for_node(base) != b) { // base not def'd in blk)
|
||||||
// Base pointer is not currently live. Since I stretched
|
// Base pointer is not currently live. Since I stretched
|
||||||
// the base pointer to here and it crosses basic-block
|
// the base pointer to here and it crosses basic-block
|
||||||
// boundaries, the global live info is now incorrect.
|
// boundaries, the global live info is now incorrect.
|
||||||
|
@ -1993,8 +1991,8 @@ void PhaseChaitin::dump(const Node *n) const {
|
||||||
tty->print("\n");
|
tty->print("\n");
|
||||||
}
|
}
|
||||||
|
|
||||||
void PhaseChaitin::dump( const Block * b ) const {
|
void PhaseChaitin::dump(const Block *b) const {
|
||||||
b->dump_head( &_cfg._bbs );
|
b->dump_head(&_cfg);
|
||||||
|
|
||||||
// For all instructions
|
// For all instructions
|
||||||
for( uint j = 0; j < b->_nodes.size(); j++ )
|
for( uint j = 0; j < b->_nodes.size(); j++ )
|
||||||
|
@ -2299,7 +2297,7 @@ void PhaseChaitin::dump_lrg( uint lidx, bool defs_only ) const {
|
||||||
if (_lrg_map.find_const(n) == lidx) {
|
if (_lrg_map.find_const(n) == lidx) {
|
||||||
if (!dump_once++) {
|
if (!dump_once++) {
|
||||||
tty->cr();
|
tty->cr();
|
||||||
b->dump_head( &_cfg._bbs );
|
b->dump_head(&_cfg);
|
||||||
}
|
}
|
||||||
dump(n);
|
dump(n);
|
||||||
continue;
|
continue;
|
||||||
|
@ -2314,7 +2312,7 @@ void PhaseChaitin::dump_lrg( uint lidx, bool defs_only ) const {
|
||||||
if (_lrg_map.find_const(m) == lidx) {
|
if (_lrg_map.find_const(m) == lidx) {
|
||||||
if (!dump_once++) {
|
if (!dump_once++) {
|
||||||
tty->cr();
|
tty->cr();
|
||||||
b->dump_head(&_cfg._bbs);
|
b->dump_head(&_cfg);
|
||||||
}
|
}
|
||||||
dump(n);
|
dump(n);
|
||||||
}
|
}
|
||||||
|
|
|
@ -52,7 +52,7 @@ void PhaseCoalesce::dump() const {
|
||||||
// Print a nice block header
|
// Print a nice block header
|
||||||
tty->print("B%d: ",b->_pre_order);
|
tty->print("B%d: ",b->_pre_order);
|
||||||
for( j=1; j<b->num_preds(); j++ )
|
for( j=1; j<b->num_preds(); j++ )
|
||||||
tty->print("B%d ", _phc._cfg._bbs[b->pred(j)->_idx]->_pre_order);
|
tty->print("B%d ", _phc._cfg.get_block_for_node(b->pred(j))->_pre_order);
|
||||||
tty->print("-> ");
|
tty->print("-> ");
|
||||||
for( j=0; j<b->_num_succs; j++ )
|
for( j=0; j<b->_num_succs; j++ )
|
||||||
tty->print("B%d ",b->_succs[j]->_pre_order);
|
tty->print("B%d ",b->_succs[j]->_pre_order);
|
||||||
|
@ -208,7 +208,7 @@ void PhaseAggressiveCoalesce::insert_copy_with_overlap( Block *b, Node *copy, ui
|
||||||
copy->set_req(idx,tmp);
|
copy->set_req(idx,tmp);
|
||||||
// Save source in temp early, before source is killed
|
// Save source in temp early, before source is killed
|
||||||
b->_nodes.insert(kill_src_idx,tmp);
|
b->_nodes.insert(kill_src_idx,tmp);
|
||||||
_phc._cfg._bbs.map( tmp->_idx, b );
|
_phc._cfg.map_node_to_block(tmp, b);
|
||||||
last_use_idx++;
|
last_use_idx++;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -286,7 +286,7 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
|
||||||
Node *m = n->in(j);
|
Node *m = n->in(j);
|
||||||
uint src_name = _phc._lrg_map.find(m);
|
uint src_name = _phc._lrg_map.find(m);
|
||||||
if (src_name != phi_name) {
|
if (src_name != phi_name) {
|
||||||
Block *pred = _phc._cfg._bbs[b->pred(j)->_idx];
|
Block *pred = _phc._cfg.get_block_for_node(b->pred(j));
|
||||||
Node *copy;
|
Node *copy;
|
||||||
assert(!m->is_Con() || m->is_Mach(), "all Con must be Mach");
|
assert(!m->is_Con() || m->is_Mach(), "all Con must be Mach");
|
||||||
// Rematerialize constants instead of copying them
|
// Rematerialize constants instead of copying them
|
||||||
|
@ -305,7 +305,7 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
|
||||||
}
|
}
|
||||||
// Insert the copy in the use-def chain
|
// Insert the copy in the use-def chain
|
||||||
n->set_req(j, copy);
|
n->set_req(j, copy);
|
||||||
_phc._cfg._bbs.map( copy->_idx, pred );
|
_phc._cfg.map_node_to_block(copy, pred);
|
||||||
// Extend ("register allocate") the names array for the copy.
|
// Extend ("register allocate") the names array for the copy.
|
||||||
_phc._lrg_map.extend(copy->_idx, phi_name);
|
_phc._lrg_map.extend(copy->_idx, phi_name);
|
||||||
} // End of if Phi names do not match
|
} // End of if Phi names do not match
|
||||||
|
@ -343,13 +343,13 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
|
||||||
n->set_req(idx, copy);
|
n->set_req(idx, copy);
|
||||||
// Extend ("register allocate") the names array for the copy.
|
// Extend ("register allocate") the names array for the copy.
|
||||||
_phc._lrg_map.extend(copy->_idx, name);
|
_phc._lrg_map.extend(copy->_idx, name);
|
||||||
_phc._cfg._bbs.map( copy->_idx, b );
|
_phc._cfg.map_node_to_block(copy, b);
|
||||||
}
|
}
|
||||||
|
|
||||||
} // End of is two-adr
|
} // End of is two-adr
|
||||||
|
|
||||||
// Insert a copy at a debug use for a lrg which has high frequency
|
// Insert a copy at a debug use for a lrg which has high frequency
|
||||||
if (b->_freq < OPTO_DEBUG_SPLIT_FREQ || b->is_uncommon(_phc._cfg._bbs)) {
|
if (b->_freq < OPTO_DEBUG_SPLIT_FREQ || b->is_uncommon(&_phc._cfg)) {
|
||||||
// Walk the debug inputs to the node and check for lrg freq
|
// Walk the debug inputs to the node and check for lrg freq
|
||||||
JVMState* jvms = n->jvms();
|
JVMState* jvms = n->jvms();
|
||||||
uint debug_start = jvms ? jvms->debug_start() : 999999;
|
uint debug_start = jvms ? jvms->debug_start() : 999999;
|
||||||
|
@ -391,7 +391,7 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
|
||||||
uint max_lrg_id = _phc._lrg_map.max_lrg_id();
|
uint max_lrg_id = _phc._lrg_map.max_lrg_id();
|
||||||
_phc.new_lrg(copy, max_lrg_id);
|
_phc.new_lrg(copy, max_lrg_id);
|
||||||
_phc._lrg_map.set_max_lrg_id(max_lrg_id + 1);
|
_phc._lrg_map.set_max_lrg_id(max_lrg_id + 1);
|
||||||
_phc._cfg._bbs.map(copy->_idx, b);
|
_phc._cfg.map_node_to_block(copy, b);
|
||||||
//tty->print_cr("Split a debug use in Aggressive Coalesce");
|
//tty->print_cr("Split a debug use in Aggressive Coalesce");
|
||||||
} // End of if high frequency use/def
|
} // End of if high frequency use/def
|
||||||
} // End of for all debug inputs
|
} // End of for all debug inputs
|
||||||
|
@ -437,7 +437,10 @@ void PhaseAggressiveCoalesce::coalesce( Block *b ) {
|
||||||
Block *bs = b->_succs[i];
|
Block *bs = b->_succs[i];
|
||||||
// Find index of 'b' in 'bs' predecessors
|
// Find index of 'b' in 'bs' predecessors
|
||||||
uint j=1;
|
uint j=1;
|
||||||
while( _phc._cfg._bbs[bs->pred(j)->_idx] != b ) j++;
|
while (_phc._cfg.get_block_for_node(bs->pred(j)) != b) {
|
||||||
|
j++;
|
||||||
|
}
|
||||||
|
|
||||||
// Visit all the Phis in successor block
|
// Visit all the Phis in successor block
|
||||||
for( uint k = 1; k<bs->_nodes.size(); k++ ) {
|
for( uint k = 1; k<bs->_nodes.size(); k++ ) {
|
||||||
Node *n = bs->_nodes[k];
|
Node *n = bs->_nodes[k];
|
||||||
|
@ -510,9 +513,9 @@ void PhaseConservativeCoalesce::union_helper( Node *lr1_node, Node *lr2_node, ui
|
||||||
if( bindex < b->_fhrp_index ) b->_fhrp_index--;
|
if( bindex < b->_fhrp_index ) b->_fhrp_index--;
|
||||||
|
|
||||||
// Stretched lr1; add it to liveness of intermediate blocks
|
// Stretched lr1; add it to liveness of intermediate blocks
|
||||||
Block *b2 = _phc._cfg._bbs[src_copy->_idx];
|
Block *b2 = _phc._cfg.get_block_for_node(src_copy);
|
||||||
while( b != b2 ) {
|
while( b != b2 ) {
|
||||||
b = _phc._cfg._bbs[b->pred(1)->_idx];
|
b = _phc._cfg.get_block_for_node(b->pred(1));
|
||||||
_phc._live->live(b)->insert(lr1);
|
_phc._live->live(b)->insert(lr1);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -532,7 +535,7 @@ uint PhaseConservativeCoalesce::compute_separating_interferences(Node *dst_copy,
|
||||||
bindex2--; // Chain backwards 1 instruction
|
bindex2--; // Chain backwards 1 instruction
|
||||||
while( bindex2 == 0 ) { // At block start, find prior block
|
while( bindex2 == 0 ) { // At block start, find prior block
|
||||||
assert( b2->num_preds() == 2, "cannot double coalesce across c-flow" );
|
assert( b2->num_preds() == 2, "cannot double coalesce across c-flow" );
|
||||||
b2 = _phc._cfg._bbs[b2->pred(1)->_idx];
|
b2 = _phc._cfg.get_block_for_node(b2->pred(1));
|
||||||
bindex2 = b2->end_idx()-1;
|
bindex2 = b2->end_idx()-1;
|
||||||
}
|
}
|
||||||
// Get prior instruction
|
// Get prior instruction
|
||||||
|
@ -676,8 +679,8 @@ bool PhaseConservativeCoalesce::copy_copy(Node *dst_copy, Node *src_copy, Block
|
||||||
|
|
||||||
if (UseFPUForSpilling && rm.is_AllStack() ) {
|
if (UseFPUForSpilling && rm.is_AllStack() ) {
|
||||||
// Don't coalesce when frequency difference is large
|
// Don't coalesce when frequency difference is large
|
||||||
Block *dst_b = _phc._cfg._bbs[dst_copy->_idx];
|
Block *dst_b = _phc._cfg.get_block_for_node(dst_copy);
|
||||||
Block *src_def_b = _phc._cfg._bbs[src_def->_idx];
|
Block *src_def_b = _phc._cfg.get_block_for_node(src_def);
|
||||||
if (src_def_b->_freq > 10*dst_b->_freq )
|
if (src_def_b->_freq > 10*dst_b->_freq )
|
||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
@ -690,7 +693,7 @@ bool PhaseConservativeCoalesce::copy_copy(Node *dst_copy, Node *src_copy, Block
|
||||||
// Another early bail-out test is when we are double-coalescing and the
|
// Another early bail-out test is when we are double-coalescing and the
|
||||||
// 2 copies are separated by some control flow.
|
// 2 copies are separated by some control flow.
|
||||||
if( dst_copy != src_copy ) {
|
if( dst_copy != src_copy ) {
|
||||||
Block *src_b = _phc._cfg._bbs[src_copy->_idx];
|
Block *src_b = _phc._cfg.get_block_for_node(src_copy);
|
||||||
Block *b2 = b;
|
Block *b2 = b;
|
||||||
while( b2 != src_b ) {
|
while( b2 != src_b ) {
|
||||||
if( b2->num_preds() > 2 ){// Found merge-point
|
if( b2->num_preds() > 2 ){// Found merge-point
|
||||||
|
@ -701,7 +704,7 @@ bool PhaseConservativeCoalesce::copy_copy(Node *dst_copy, Node *src_copy, Block
|
||||||
//record_bias( _phc._lrgs, lr1, lr2 );
|
//record_bias( _phc._lrgs, lr1, lr2 );
|
||||||
return false; // To hard to find all interferences
|
return false; // To hard to find all interferences
|
||||||
}
|
}
|
||||||
b2 = _phc._cfg._bbs[b2->pred(1)->_idx];
|
b2 = _phc._cfg.get_block_for_node(b2->pred(1));
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -786,8 +789,9 @@ bool PhaseConservativeCoalesce::copy_copy(Node *dst_copy, Node *src_copy, Block
|
||||||
// Conservative (but pessimistic) copy coalescing of a single block
|
// Conservative (but pessimistic) copy coalescing of a single block
|
||||||
void PhaseConservativeCoalesce::coalesce( Block *b ) {
|
void PhaseConservativeCoalesce::coalesce( Block *b ) {
|
||||||
// Bail out on infrequent blocks
|
// Bail out on infrequent blocks
|
||||||
if( b->is_uncommon(_phc._cfg._bbs) )
|
if (b->is_uncommon(&_phc._cfg)) {
|
||||||
return;
|
return;
|
||||||
|
}
|
||||||
// Check this block for copies.
|
// Check this block for copies.
|
||||||
for( uint i = 1; i<b->end_idx(); i++ ) {
|
for( uint i = 1; i<b->end_idx(); i++ ) {
|
||||||
// Check for actual copies on inputs. Coalesce a copy into its
|
// Check for actual copies on inputs. Coalesce a copy into its
|
||||||
|
|
|
@ -2262,7 +2262,7 @@ void Compile::dump_asm(int *pcs, uint pc_limit) {
|
||||||
tty->print("%3.3x ", pcs[n->_idx]);
|
tty->print("%3.3x ", pcs[n->_idx]);
|
||||||
else
|
else
|
||||||
tty->print(" ");
|
tty->print(" ");
|
||||||
b->dump_head( &_cfg->_bbs );
|
b->dump_head(_cfg);
|
||||||
if (b->is_connector()) {
|
if (b->is_connector()) {
|
||||||
tty->print_cr(" # Empty connector block");
|
tty->print_cr(" # Empty connector block");
|
||||||
} else if (b->num_preds() == 2 && b->pred(1)->is_CatchProj() && b->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) {
|
} else if (b->num_preds() == 2 && b->pred(1)->is_CatchProj() && b->pred(1)->as_CatchProj()->_con == CatchProjNode::fall_through_index) {
|
||||||
|
@ -3525,7 +3525,7 @@ void Compile::ConstantTable::add(Constant& con) {
|
||||||
}
|
}
|
||||||
|
|
||||||
Compile::Constant Compile::ConstantTable::add(MachConstantNode* n, BasicType type, jvalue value) {
|
Compile::Constant Compile::ConstantTable::add(MachConstantNode* n, BasicType type, jvalue value) {
|
||||||
Block* b = Compile::current()->cfg()->_bbs[n->_idx];
|
Block* b = Compile::current()->cfg()->get_block_for_node(n);
|
||||||
Constant con(type, value, b->_freq);
|
Constant con(type, value, b->_freq);
|
||||||
add(con);
|
add(con);
|
||||||
return con;
|
return con;
|
||||||
|
|
|
@ -105,8 +105,8 @@ void PhaseCFG::Dominators( ) {
|
||||||
|
|
||||||
// Step 2:
|
// Step 2:
|
||||||
Node *whead = w->_block->head();
|
Node *whead = w->_block->head();
|
||||||
for( uint j=1; j < whead->req(); j++ ) {
|
for (uint j = 1; j < whead->req(); j++) {
|
||||||
Block *b = _bbs[whead->in(j)->_idx];
|
Block* b = get_block_for_node(whead->in(j));
|
||||||
Tarjan *vx = &tarjan[b->_pre_order];
|
Tarjan *vx = &tarjan[b->_pre_order];
|
||||||
Tarjan *u = vx->EVAL();
|
Tarjan *u = vx->EVAL();
|
||||||
if( u->_semi < w->_semi )
|
if( u->_semi < w->_semi )
|
||||||
|
|
|
@ -66,7 +66,7 @@
|
||||||
// are in b also.
|
// are in b also.
|
||||||
void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) {
|
void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) {
|
||||||
// Set basic block of n, Add n to b,
|
// Set basic block of n, Add n to b,
|
||||||
_bbs.map(n->_idx, b);
|
map_node_to_block(n, b);
|
||||||
b->add_inst(n);
|
b->add_inst(n);
|
||||||
|
|
||||||
// After Matching, nearly any old Node may have projections trailing it.
|
// After Matching, nearly any old Node may have projections trailing it.
|
||||||
|
@ -75,11 +75,12 @@ void PhaseCFG::schedule_node_into_block( Node *n, Block *b ) {
|
||||||
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
|
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
|
||||||
Node* use = n->fast_out(i);
|
Node* use = n->fast_out(i);
|
||||||
if (use->is_Proj()) {
|
if (use->is_Proj()) {
|
||||||
Block* buse = _bbs[use->_idx];
|
Block* buse = get_block_for_node(use);
|
||||||
if (buse != b) { // In wrong block?
|
if (buse != b) { // In wrong block?
|
||||||
if (buse != NULL)
|
if (buse != NULL) {
|
||||||
buse->find_remove(use); // Remove from wrong block
|
buse->find_remove(use); // Remove from wrong block
|
||||||
_bbs.map(use->_idx, b); // Re-insert in this block
|
}
|
||||||
|
map_node_to_block(use, b);
|
||||||
b->add_inst(use);
|
b->add_inst(use);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -97,7 +98,7 @@ void PhaseCFG::replace_block_proj_ctrl( Node *n ) {
|
||||||
if (p != NULL && p != n) { // Control from a block projection?
|
if (p != NULL && p != n) { // Control from a block projection?
|
||||||
assert(!n->pinned() || n->is_MachConstantBase(), "only pinned MachConstantBase node is expected here");
|
assert(!n->pinned() || n->is_MachConstantBase(), "only pinned MachConstantBase node is expected here");
|
||||||
// Find trailing Region
|
// Find trailing Region
|
||||||
Block *pb = _bbs[in0->_idx]; // Block-projection already has basic block
|
Block *pb = get_block_for_node(in0); // Block-projection already has basic block
|
||||||
uint j = 0;
|
uint j = 0;
|
||||||
if (pb->_num_succs != 1) { // More then 1 successor?
|
if (pb->_num_succs != 1) { // More then 1 successor?
|
||||||
// Search for successor
|
// Search for successor
|
||||||
|
@ -127,14 +128,15 @@ void PhaseCFG::schedule_pinned_nodes( VectorSet &visited ) {
|
||||||
while ( spstack.is_nonempty() ) {
|
while ( spstack.is_nonempty() ) {
|
||||||
Node *n = spstack.pop();
|
Node *n = spstack.pop();
|
||||||
if( !visited.test_set(n->_idx) ) { // Test node and flag it as visited
|
if( !visited.test_set(n->_idx) ) { // Test node and flag it as visited
|
||||||
if( n->pinned() && !_bbs.lookup(n->_idx) ) { // Pinned? Nail it down!
|
if( n->pinned() && !has_block(n)) { // Pinned? Nail it down!
|
||||||
assert( n->in(0), "pinned Node must have Control" );
|
assert( n->in(0), "pinned Node must have Control" );
|
||||||
// Before setting block replace block_proj control edge
|
// Before setting block replace block_proj control edge
|
||||||
replace_block_proj_ctrl(n);
|
replace_block_proj_ctrl(n);
|
||||||
Node *input = n->in(0);
|
Node *input = n->in(0);
|
||||||
while( !input->is_block_start() )
|
while (!input->is_block_start()) {
|
||||||
input = input->in(0);
|
input = input->in(0);
|
||||||
Block *b = _bbs[input->_idx]; // Basic block of controlling input
|
}
|
||||||
|
Block *b = get_block_for_node(input); // Basic block of controlling input
|
||||||
schedule_node_into_block(n, b);
|
schedule_node_into_block(n, b);
|
||||||
}
|
}
|
||||||
for( int i = n->req() - 1; i >= 0; --i ) { // For all inputs
|
for( int i = n->req() - 1; i >= 0; --i ) { // For all inputs
|
||||||
|
@ -149,7 +151,7 @@ void PhaseCFG::schedule_pinned_nodes( VectorSet &visited ) {
|
||||||
// Assert that new input b2 is dominated by all previous inputs.
|
// Assert that new input b2 is dominated by all previous inputs.
|
||||||
// Check this by by seeing that it is dominated by b1, the deepest
|
// Check this by by seeing that it is dominated by b1, the deepest
|
||||||
// input observed until b2.
|
// input observed until b2.
|
||||||
static void assert_dom(Block* b1, Block* b2, Node* n, Block_Array &bbs) {
|
static void assert_dom(Block* b1, Block* b2, Node* n, const PhaseCFG* cfg) {
|
||||||
if (b1 == NULL) return;
|
if (b1 == NULL) return;
|
||||||
assert(b1->_dom_depth < b2->_dom_depth, "sanity");
|
assert(b1->_dom_depth < b2->_dom_depth, "sanity");
|
||||||
Block* tmp = b2;
|
Block* tmp = b2;
|
||||||
|
@ -162,7 +164,7 @@ static void assert_dom(Block* b1, Block* b2, Node* n, Block_Array &bbs) {
|
||||||
for (uint j=0; j<n->len(); j++) { // For all inputs
|
for (uint j=0; j<n->len(); j++) { // For all inputs
|
||||||
Node* inn = n->in(j); // Get input
|
Node* inn = n->in(j); // Get input
|
||||||
if (inn == NULL) continue; // Ignore NULL, missing inputs
|
if (inn == NULL) continue; // Ignore NULL, missing inputs
|
||||||
Block* inb = bbs[inn->_idx];
|
Block* inb = cfg->get_block_for_node(inn);
|
||||||
tty->print("B%d idom=B%d depth=%2d ",inb->_pre_order,
|
tty->print("B%d idom=B%d depth=%2d ",inb->_pre_order,
|
||||||
inb->_idom ? inb->_idom->_pre_order : 0, inb->_dom_depth);
|
inb->_idom ? inb->_idom->_pre_order : 0, inb->_dom_depth);
|
||||||
inn->dump();
|
inn->dump();
|
||||||
|
@ -174,20 +176,20 @@ static void assert_dom(Block* b1, Block* b2, Node* n, Block_Array &bbs) {
|
||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
static Block* find_deepest_input(Node* n, Block_Array &bbs) {
|
static Block* find_deepest_input(Node* n, const PhaseCFG* cfg) {
|
||||||
// Find the last input dominated by all other inputs.
|
// Find the last input dominated by all other inputs.
|
||||||
Block* deepb = NULL; // Deepest block so far
|
Block* deepb = NULL; // Deepest block so far
|
||||||
int deepb_dom_depth = 0;
|
int deepb_dom_depth = 0;
|
||||||
for (uint k = 0; k < n->len(); k++) { // For all inputs
|
for (uint k = 0; k < n->len(); k++) { // For all inputs
|
||||||
Node* inn = n->in(k); // Get input
|
Node* inn = n->in(k); // Get input
|
||||||
if (inn == NULL) continue; // Ignore NULL, missing inputs
|
if (inn == NULL) continue; // Ignore NULL, missing inputs
|
||||||
Block* inb = bbs[inn->_idx];
|
Block* inb = cfg->get_block_for_node(inn);
|
||||||
assert(inb != NULL, "must already have scheduled this input");
|
assert(inb != NULL, "must already have scheduled this input");
|
||||||
if (deepb_dom_depth < (int) inb->_dom_depth) {
|
if (deepb_dom_depth < (int) inb->_dom_depth) {
|
||||||
// The new inb must be dominated by the previous deepb.
|
// The new inb must be dominated by the previous deepb.
|
||||||
// The various inputs must be linearly ordered in the dom
|
// The various inputs must be linearly ordered in the dom
|
||||||
// tree, or else there will not be a unique deepest block.
|
// tree, or else there will not be a unique deepest block.
|
||||||
DEBUG_ONLY(assert_dom(deepb, inb, n, bbs));
|
DEBUG_ONLY(assert_dom(deepb, inb, n, cfg));
|
||||||
deepb = inb; // Save deepest block
|
deepb = inb; // Save deepest block
|
||||||
deepb_dom_depth = deepb->_dom_depth;
|
deepb_dom_depth = deepb->_dom_depth;
|
||||||
}
|
}
|
||||||
|
@ -243,7 +245,7 @@ bool PhaseCFG::schedule_early(VectorSet &visited, Node_List &roots) {
|
||||||
++i;
|
++i;
|
||||||
if (in == NULL) continue; // Ignore NULL, missing inputs
|
if (in == NULL) continue; // Ignore NULL, missing inputs
|
||||||
int is_visited = visited.test_set(in->_idx);
|
int is_visited = visited.test_set(in->_idx);
|
||||||
if (!_bbs.lookup(in->_idx)) { // Missing block selection?
|
if (!has_block(in)) { // Missing block selection?
|
||||||
if (is_visited) {
|
if (is_visited) {
|
||||||
// assert( !visited.test(in->_idx), "did not schedule early" );
|
// assert( !visited.test(in->_idx), "did not schedule early" );
|
||||||
return false;
|
return false;
|
||||||
|
@ -265,9 +267,9 @@ bool PhaseCFG::schedule_early(VectorSet &visited, Node_List &roots) {
|
||||||
// any projections which depend on them.
|
// any projections which depend on them.
|
||||||
if (!n->pinned()) {
|
if (!n->pinned()) {
|
||||||
// Set earliest legal block.
|
// Set earliest legal block.
|
||||||
_bbs.map(n->_idx, find_deepest_input(n, _bbs));
|
map_node_to_block(n, find_deepest_input(n, this));
|
||||||
} else {
|
} else {
|
||||||
assert(_bbs[n->_idx] == _bbs[n->in(0)->_idx], "Pinned Node should be at the same block as its control edge");
|
assert(get_block_for_node(n) == get_block_for_node(n->in(0)), "Pinned Node should be at the same block as its control edge");
|
||||||
}
|
}
|
||||||
|
|
||||||
if (nstack.is_empty()) {
|
if (nstack.is_empty()) {
|
||||||
|
@ -313,8 +315,8 @@ Block* Block::dom_lca(Block* LCA) {
|
||||||
// The definition must dominate the use, so move the LCA upward in the
|
// The definition must dominate the use, so move the LCA upward in the
|
||||||
// dominator tree to dominate the use. If the use is a phi, adjust
|
// dominator tree to dominate the use. If the use is a phi, adjust
|
||||||
// the LCA only with the phi input paths which actually use this def.
|
// the LCA only with the phi input paths which actually use this def.
|
||||||
static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, Block_Array &bbs) {
|
static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, const PhaseCFG* cfg) {
|
||||||
Block* buse = bbs[use->_idx];
|
Block* buse = cfg->get_block_for_node(use);
|
||||||
if (buse == NULL) return LCA; // Unused killing Projs have no use block
|
if (buse == NULL) return LCA; // Unused killing Projs have no use block
|
||||||
if (!use->is_Phi()) return buse->dom_lca(LCA);
|
if (!use->is_Phi()) return buse->dom_lca(LCA);
|
||||||
uint pmax = use->req(); // Number of Phi inputs
|
uint pmax = use->req(); // Number of Phi inputs
|
||||||
|
@ -329,7 +331,7 @@ static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, Block_Array
|
||||||
// more than once.
|
// more than once.
|
||||||
for (uint j=1; j<pmax; j++) { // For all inputs
|
for (uint j=1; j<pmax; j++) { // For all inputs
|
||||||
if (use->in(j) == def) { // Found matching input?
|
if (use->in(j) == def) { // Found matching input?
|
||||||
Block* pred = bbs[buse->pred(j)->_idx];
|
Block* pred = cfg->get_block_for_node(buse->pred(j));
|
||||||
LCA = pred->dom_lca(LCA);
|
LCA = pred->dom_lca(LCA);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -342,8 +344,7 @@ static Block* raise_LCA_above_use(Block* LCA, Node* use, Node* def, Block_Array
|
||||||
// which are marked with the given index. Return the LCA (in the dom tree)
|
// which are marked with the given index. Return the LCA (in the dom tree)
|
||||||
// of all marked blocks. If there are none marked, return the original
|
// of all marked blocks. If there are none marked, return the original
|
||||||
// LCA.
|
// LCA.
|
||||||
static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
|
static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark, Block* early, const PhaseCFG* cfg) {
|
||||||
Block* early, Block_Array &bbs) {
|
|
||||||
Block_List worklist;
|
Block_List worklist;
|
||||||
worklist.push(LCA);
|
worklist.push(LCA);
|
||||||
while (worklist.size() > 0) {
|
while (worklist.size() > 0) {
|
||||||
|
@ -366,7 +367,7 @@ static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
|
||||||
} else {
|
} else {
|
||||||
// Keep searching through this block's predecessors.
|
// Keep searching through this block's predecessors.
|
||||||
for (uint j = 1, jmax = mid->num_preds(); j < jmax; j++) {
|
for (uint j = 1, jmax = mid->num_preds(); j < jmax; j++) {
|
||||||
Block* mid_parent = bbs[ mid->pred(j)->_idx ];
|
Block* mid_parent = cfg->get_block_for_node(mid->pred(j));
|
||||||
worklist.push(mid_parent);
|
worklist.push(mid_parent);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -384,7 +385,7 @@ static Block* raise_LCA_above_marks(Block* LCA, node_idx_t mark,
|
||||||
// be earlier (at a shallower dom_depth) than the true schedule_early
|
// be earlier (at a shallower dom_depth) than the true schedule_early
|
||||||
// point of the node. We compute this earlier block as a more permissive
|
// point of the node. We compute this earlier block as a more permissive
|
||||||
// site for anti-dependency insertion, but only if subsume_loads is enabled.
|
// site for anti-dependency insertion, but only if subsume_loads is enabled.
|
||||||
static Block* memory_early_block(Node* load, Block* early, Block_Array &bbs) {
|
static Block* memory_early_block(Node* load, Block* early, const PhaseCFG* cfg) {
|
||||||
Node* base;
|
Node* base;
|
||||||
Node* index;
|
Node* index;
|
||||||
Node* store = load->in(MemNode::Memory);
|
Node* store = load->in(MemNode::Memory);
|
||||||
|
@ -412,12 +413,12 @@ static Block* memory_early_block(Node* load, Block* early, Block_Array &bbs) {
|
||||||
Block* deepb = NULL; // Deepest block so far
|
Block* deepb = NULL; // Deepest block so far
|
||||||
int deepb_dom_depth = 0;
|
int deepb_dom_depth = 0;
|
||||||
for (int i = 0; i < mem_inputs_length; i++) {
|
for (int i = 0; i < mem_inputs_length; i++) {
|
||||||
Block* inb = bbs[mem_inputs[i]->_idx];
|
Block* inb = cfg->get_block_for_node(mem_inputs[i]);
|
||||||
if (deepb_dom_depth < (int) inb->_dom_depth) {
|
if (deepb_dom_depth < (int) inb->_dom_depth) {
|
||||||
// The new inb must be dominated by the previous deepb.
|
// The new inb must be dominated by the previous deepb.
|
||||||
// The various inputs must be linearly ordered in the dom
|
// The various inputs must be linearly ordered in the dom
|
||||||
// tree, or else there will not be a unique deepest block.
|
// tree, or else there will not be a unique deepest block.
|
||||||
DEBUG_ONLY(assert_dom(deepb, inb, load, bbs));
|
DEBUG_ONLY(assert_dom(deepb, inb, load, cfg));
|
||||||
deepb = inb; // Save deepest block
|
deepb = inb; // Save deepest block
|
||||||
deepb_dom_depth = deepb->_dom_depth;
|
deepb_dom_depth = deepb->_dom_depth;
|
||||||
}
|
}
|
||||||
|
@ -488,14 +489,14 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
|
||||||
// and other inputs are first available. (Computed by schedule_early.)
|
// and other inputs are first available. (Computed by schedule_early.)
|
||||||
// For normal loads, 'early' is the shallowest place (dom graph wise)
|
// For normal loads, 'early' is the shallowest place (dom graph wise)
|
||||||
// to look for anti-deps between this load and any store.
|
// to look for anti-deps between this load and any store.
|
||||||
Block* early = _bbs[load_index];
|
Block* early = get_block_for_node(load);
|
||||||
|
|
||||||
// If we are subsuming loads, compute an "early" block that only considers
|
// If we are subsuming loads, compute an "early" block that only considers
|
||||||
// memory or address inputs. This block may be different than the
|
// memory or address inputs. This block may be different than the
|
||||||
// schedule_early block in that it could be at an even shallower depth in the
|
// schedule_early block in that it could be at an even shallower depth in the
|
||||||
// dominator tree, and allow for a broader discovery of anti-dependences.
|
// dominator tree, and allow for a broader discovery of anti-dependences.
|
||||||
if (C->subsume_loads()) {
|
if (C->subsume_loads()) {
|
||||||
early = memory_early_block(load, early, _bbs);
|
early = memory_early_block(load, early, this);
|
||||||
}
|
}
|
||||||
|
|
||||||
ResourceArea *area = Thread::current()->resource_area();
|
ResourceArea *area = Thread::current()->resource_area();
|
||||||
|
@ -619,7 +620,7 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
|
||||||
// or else observe that 'store' is all the way up in the
|
// or else observe that 'store' is all the way up in the
|
||||||
// earliest legal block for 'load'. In the latter case,
|
// earliest legal block for 'load'. In the latter case,
|
||||||
// immediately insert an anti-dependence edge.
|
// immediately insert an anti-dependence edge.
|
||||||
Block* store_block = _bbs[store->_idx];
|
Block* store_block = get_block_for_node(store);
|
||||||
assert(store_block != NULL, "unused killing projections skipped above");
|
assert(store_block != NULL, "unused killing projections skipped above");
|
||||||
|
|
||||||
if (store->is_Phi()) {
|
if (store->is_Phi()) {
|
||||||
|
@ -637,7 +638,7 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
|
||||||
for (uint j = PhiNode::Input, jmax = store->req(); j < jmax; j++) {
|
for (uint j = PhiNode::Input, jmax = store->req(); j < jmax; j++) {
|
||||||
if (store->in(j) == mem) { // Found matching input?
|
if (store->in(j) == mem) { // Found matching input?
|
||||||
DEBUG_ONLY(found_match = true);
|
DEBUG_ONLY(found_match = true);
|
||||||
Block* pred_block = _bbs[store_block->pred(j)->_idx];
|
Block* pred_block = get_block_for_node(store_block->pred(j));
|
||||||
if (pred_block != early) {
|
if (pred_block != early) {
|
||||||
// If any predecessor of the Phi matches the load's "early block",
|
// If any predecessor of the Phi matches the load's "early block",
|
||||||
// we do not need a precedence edge between the Phi and 'load'
|
// we do not need a precedence edge between the Phi and 'load'
|
||||||
|
@ -711,7 +712,7 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
|
||||||
// preventing the load from sinking past any block containing
|
// preventing the load from sinking past any block containing
|
||||||
// a store that may invalidate the memory state required by 'load'.
|
// a store that may invalidate the memory state required by 'load'.
|
||||||
if (must_raise_LCA)
|
if (must_raise_LCA)
|
||||||
LCA = raise_LCA_above_marks(LCA, load->_idx, early, _bbs);
|
LCA = raise_LCA_above_marks(LCA, load->_idx, early, this);
|
||||||
if (LCA == early) return LCA;
|
if (LCA == early) return LCA;
|
||||||
|
|
||||||
// Insert anti-dependence edges from 'load' to each store
|
// Insert anti-dependence edges from 'load' to each store
|
||||||
|
@ -720,7 +721,7 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
|
||||||
if (LCA->raise_LCA_mark() == load_index) {
|
if (LCA->raise_LCA_mark() == load_index) {
|
||||||
while (non_early_stores.size() > 0) {
|
while (non_early_stores.size() > 0) {
|
||||||
Node* store = non_early_stores.pop();
|
Node* store = non_early_stores.pop();
|
||||||
Block* store_block = _bbs[store->_idx];
|
Block* store_block = get_block_for_node(store);
|
||||||
if (store_block == LCA) {
|
if (store_block == LCA) {
|
||||||
// add anti_dependence from store to load in its own block
|
// add anti_dependence from store to load in its own block
|
||||||
assert(store != load->in(0), "dependence cycle found");
|
assert(store != load->in(0), "dependence cycle found");
|
||||||
|
@ -754,7 +755,7 @@ private:
|
||||||
|
|
||||||
public:
|
public:
|
||||||
// Constructor for the iterator
|
// Constructor for the iterator
|
||||||
Node_Backward_Iterator(Node *root, VectorSet &visited, Node_List &stack, Block_Array &bbs);
|
Node_Backward_Iterator(Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg);
|
||||||
|
|
||||||
// Postincrement operator to iterate over the nodes
|
// Postincrement operator to iterate over the nodes
|
||||||
Node *next();
|
Node *next();
|
||||||
|
@ -762,12 +763,12 @@ public:
|
||||||
private:
|
private:
|
||||||
VectorSet &_visited;
|
VectorSet &_visited;
|
||||||
Node_List &_stack;
|
Node_List &_stack;
|
||||||
Block_Array &_bbs;
|
PhaseCFG &_cfg;
|
||||||
};
|
};
|
||||||
|
|
||||||
// Constructor for the Node_Backward_Iterator
|
// Constructor for the Node_Backward_Iterator
|
||||||
Node_Backward_Iterator::Node_Backward_Iterator( Node *root, VectorSet &visited, Node_List &stack, Block_Array &bbs )
|
Node_Backward_Iterator::Node_Backward_Iterator( Node *root, VectorSet &visited, Node_List &stack, PhaseCFG &cfg)
|
||||||
: _visited(visited), _stack(stack), _bbs(bbs) {
|
: _visited(visited), _stack(stack), _cfg(cfg) {
|
||||||
// The stack should contain exactly the root
|
// The stack should contain exactly the root
|
||||||
stack.clear();
|
stack.clear();
|
||||||
stack.push(root);
|
stack.push(root);
|
||||||
|
@ -797,8 +798,8 @@ Node *Node_Backward_Iterator::next() {
|
||||||
_visited.set(self->_idx);
|
_visited.set(self->_idx);
|
||||||
|
|
||||||
// Now schedule all uses as late as possible.
|
// Now schedule all uses as late as possible.
|
||||||
uint src = self->is_Proj() ? self->in(0)->_idx : self->_idx;
|
const Node* src = self->is_Proj() ? self->in(0) : self;
|
||||||
uint src_rpo = _bbs[src]->_rpo;
|
uint src_rpo = _cfg.get_block_for_node(src)->_rpo;
|
||||||
|
|
||||||
// Schedule all nodes in a post-order visit
|
// Schedule all nodes in a post-order visit
|
||||||
Node *unvisited = NULL; // Unvisited anti-dependent Node, if any
|
Node *unvisited = NULL; // Unvisited anti-dependent Node, if any
|
||||||
|
@ -814,7 +815,7 @@ Node *Node_Backward_Iterator::next() {
|
||||||
|
|
||||||
// do not traverse backward control edges
|
// do not traverse backward control edges
|
||||||
Node *use = n->is_Proj() ? n->in(0) : n;
|
Node *use = n->is_Proj() ? n->in(0) : n;
|
||||||
uint use_rpo = _bbs[use->_idx]->_rpo;
|
uint use_rpo = _cfg.get_block_for_node(use)->_rpo;
|
||||||
|
|
||||||
if ( use_rpo < src_rpo )
|
if ( use_rpo < src_rpo )
|
||||||
continue;
|
continue;
|
||||||
|
@ -852,7 +853,7 @@ void PhaseCFG::ComputeLatenciesBackwards(VectorSet &visited, Node_List &stack) {
|
||||||
tty->print("\n#---- ComputeLatenciesBackwards ----\n");
|
tty->print("\n#---- ComputeLatenciesBackwards ----\n");
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
Node_Backward_Iterator iter((Node *)_root, visited, stack, _bbs);
|
Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
|
||||||
Node *n;
|
Node *n;
|
||||||
|
|
||||||
// Walk over all the nodes from last to first
|
// Walk over all the nodes from last to first
|
||||||
|
@ -883,7 +884,7 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
|
||||||
|
|
||||||
uint nlen = n->len();
|
uint nlen = n->len();
|
||||||
uint use_latency = _node_latency->at_grow(n->_idx);
|
uint use_latency = _node_latency->at_grow(n->_idx);
|
||||||
uint use_pre_order = _bbs[n->_idx]->_pre_order;
|
uint use_pre_order = get_block_for_node(n)->_pre_order;
|
||||||
|
|
||||||
for ( uint j=0; j<nlen; j++ ) {
|
for ( uint j=0; j<nlen; j++ ) {
|
||||||
Node *def = n->in(j);
|
Node *def = n->in(j);
|
||||||
|
@ -903,7 +904,7 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
// If the defining block is not known, assume it is ok
|
// If the defining block is not known, assume it is ok
|
||||||
Block *def_block = _bbs[def->_idx];
|
Block *def_block = get_block_for_node(def);
|
||||||
uint def_pre_order = def_block ? def_block->_pre_order : 0;
|
uint def_pre_order = def_block ? def_block->_pre_order : 0;
|
||||||
|
|
||||||
if ( (use_pre_order < def_pre_order) ||
|
if ( (use_pre_order < def_pre_order) ||
|
||||||
|
@ -931,10 +932,11 @@ void PhaseCFG::partial_latency_of_defs(Node *n) {
|
||||||
// Compute the latency of a specific use
|
// Compute the latency of a specific use
|
||||||
int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
|
int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
|
||||||
// If self-reference, return no latency
|
// If self-reference, return no latency
|
||||||
if (use == n || use->is_Root())
|
if (use == n || use->is_Root()) {
|
||||||
return 0;
|
return 0;
|
||||||
|
}
|
||||||
|
|
||||||
uint def_pre_order = _bbs[def->_idx]->_pre_order;
|
uint def_pre_order = get_block_for_node(def)->_pre_order;
|
||||||
uint latency = 0;
|
uint latency = 0;
|
||||||
|
|
||||||
// If the use is not a projection, then it is simple...
|
// If the use is not a projection, then it is simple...
|
||||||
|
@ -946,7 +948,7 @@ int PhaseCFG::latency_from_use(Node *n, const Node *def, Node *use) {
|
||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
uint use_pre_order = _bbs[use->_idx]->_pre_order;
|
uint use_pre_order = get_block_for_node(use)->_pre_order;
|
||||||
|
|
||||||
if (use_pre_order < def_pre_order)
|
if (use_pre_order < def_pre_order)
|
||||||
return 0;
|
return 0;
|
||||||
|
@ -1018,7 +1020,7 @@ Block* PhaseCFG::hoist_to_cheaper_block(Block* LCA, Block* early, Node* self) {
|
||||||
uint start_latency = _node_latency->at_grow(LCA->_nodes[0]->_idx);
|
uint start_latency = _node_latency->at_grow(LCA->_nodes[0]->_idx);
|
||||||
uint end_latency = _node_latency->at_grow(LCA->_nodes[LCA->end_idx()]->_idx);
|
uint end_latency = _node_latency->at_grow(LCA->_nodes[LCA->end_idx()]->_idx);
|
||||||
bool in_latency = (target <= start_latency);
|
bool in_latency = (target <= start_latency);
|
||||||
const Block* root_block = _bbs[_root->_idx];
|
const Block* root_block = get_block_for_node(_root);
|
||||||
|
|
||||||
// Turn off latency scheduling if scheduling is just plain off
|
// Turn off latency scheduling if scheduling is just plain off
|
||||||
if (!C->do_scheduling())
|
if (!C->do_scheduling())
|
||||||
|
@ -1126,12 +1128,12 @@ void PhaseCFG::schedule_late(VectorSet &visited, Node_List &stack) {
|
||||||
tty->print("\n#---- schedule_late ----\n");
|
tty->print("\n#---- schedule_late ----\n");
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
Node_Backward_Iterator iter((Node *)_root, visited, stack, _bbs);
|
Node_Backward_Iterator iter((Node *)_root, visited, stack, *this);
|
||||||
Node *self;
|
Node *self;
|
||||||
|
|
||||||
// Walk over all the nodes from last to first
|
// Walk over all the nodes from last to first
|
||||||
while (self = iter.next()) {
|
while (self = iter.next()) {
|
||||||
Block* early = _bbs[self->_idx]; // Earliest legal placement
|
Block* early = get_block_for_node(self); // Earliest legal placement
|
||||||
|
|
||||||
if (self->is_top()) {
|
if (self->is_top()) {
|
||||||
// Top node goes in bb #2 with other constants.
|
// Top node goes in bb #2 with other constants.
|
||||||
|
@ -1179,7 +1181,7 @@ void PhaseCFG::schedule_late(VectorSet &visited, Node_List &stack) {
|
||||||
for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
|
for (DUIterator_Fast imax, i = self->fast_outs(imax); i < imax; i++) {
|
||||||
// For all uses, find LCA
|
// For all uses, find LCA
|
||||||
Node* use = self->fast_out(i);
|
Node* use = self->fast_out(i);
|
||||||
LCA = raise_LCA_above_use(LCA, use, self, _bbs);
|
LCA = raise_LCA_above_use(LCA, use, self, this);
|
||||||
}
|
}
|
||||||
} // (Hide defs of imax, i from rest of block.)
|
} // (Hide defs of imax, i from rest of block.)
|
||||||
|
|
||||||
|
@ -1187,7 +1189,7 @@ void PhaseCFG::schedule_late(VectorSet &visited, Node_List &stack) {
|
||||||
// requirement for correctness but it reduces useless
|
// requirement for correctness but it reduces useless
|
||||||
// interference between temps and other nodes.
|
// interference between temps and other nodes.
|
||||||
if (mach != NULL && mach->is_MachTemp()) {
|
if (mach != NULL && mach->is_MachTemp()) {
|
||||||
_bbs.map(self->_idx, LCA);
|
map_node_to_block(self, LCA);
|
||||||
LCA->add_inst(self);
|
LCA->add_inst(self);
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
@ -1262,10 +1264,10 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
|
||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
// Initialize the bbs.map for things on the proj_list
|
// Initialize the node to block mapping for things on the proj_list
|
||||||
uint i;
|
for (uint i = 0; i < proj_list.size(); i++) {
|
||||||
for( i=0; i < proj_list.size(); i++ )
|
unmap_node_from_block(proj_list[i]);
|
||||||
_bbs.map(proj_list[i]->_idx, NULL);
|
}
|
||||||
|
|
||||||
// Set the basic block for Nodes pinned into blocks
|
// Set the basic block for Nodes pinned into blocks
|
||||||
Arena *a = Thread::current()->resource_area();
|
Arena *a = Thread::current()->resource_area();
|
||||||
|
@ -1333,7 +1335,7 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
|
||||||
for( int i= matcher._null_check_tests.size()-2; i>=0; i-=2 ) {
|
for( int i= matcher._null_check_tests.size()-2; i>=0; i-=2 ) {
|
||||||
Node *proj = matcher._null_check_tests[i ];
|
Node *proj = matcher._null_check_tests[i ];
|
||||||
Node *val = matcher._null_check_tests[i+1];
|
Node *val = matcher._null_check_tests[i+1];
|
||||||
_bbs[proj->_idx]->implicit_null_check(this, proj, val, allowed_reasons);
|
get_block_for_node(proj)->implicit_null_check(this, proj, val, allowed_reasons);
|
||||||
// The implicit_null_check will only perform the transformation
|
// The implicit_null_check will only perform the transformation
|
||||||
// if the null branch is truly uncommon, *and* it leads to an
|
// if the null branch is truly uncommon, *and* it leads to an
|
||||||
// uncommon trap. Combined with the too_many_traps guards
|
// uncommon trap. Combined with the too_many_traps guards
|
||||||
|
@ -1353,7 +1355,7 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
|
||||||
uint max_idx = C->unique();
|
uint max_idx = C->unique();
|
||||||
GrowableArray<int> ready_cnt(max_idx, max_idx, -1);
|
GrowableArray<int> ready_cnt(max_idx, max_idx, -1);
|
||||||
visited.Clear();
|
visited.Clear();
|
||||||
for (i = 0; i < _num_blocks; i++) {
|
for (uint i = 0; i < _num_blocks; i++) {
|
||||||
if (!_blocks[i]->schedule_local(this, matcher, ready_cnt, visited)) {
|
if (!_blocks[i]->schedule_local(this, matcher, ready_cnt, visited)) {
|
||||||
if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
|
if (!C->failure_reason_is(C2Compiler::retry_no_subsuming_loads())) {
|
||||||
C->record_method_not_compilable("local schedule failed");
|
C->record_method_not_compilable("local schedule failed");
|
||||||
|
@ -1364,8 +1366,9 @@ void PhaseCFG::GlobalCodeMotion( Matcher &matcher, uint unique, Node_List &proj_
|
||||||
|
|
||||||
// If we inserted any instructions between a Call and his CatchNode,
|
// If we inserted any instructions between a Call and his CatchNode,
|
||||||
// clone the instructions on all paths below the Catch.
|
// clone the instructions on all paths below the Catch.
|
||||||
for( i=0; i < _num_blocks; i++ )
|
for (uint i = 0; i < _num_blocks; i++) {
|
||||||
_blocks[i]->call_catch_cleanup(_bbs, C);
|
_blocks[i]->call_catch_cleanup(this, C);
|
||||||
|
}
|
||||||
|
|
||||||
#ifndef PRODUCT
|
#ifndef PRODUCT
|
||||||
if (trace_opto_pipelining()) {
|
if (trace_opto_pipelining()) {
|
||||||
|
@ -1392,7 +1395,7 @@ void PhaseCFG::Estimate_Block_Frequency() {
|
||||||
Block_List worklist;
|
Block_List worklist;
|
||||||
Block* root_blk = _blocks[0];
|
Block* root_blk = _blocks[0];
|
||||||
for (uint i = 1; i < root_blk->num_preds(); i++) {
|
for (uint i = 1; i < root_blk->num_preds(); i++) {
|
||||||
Block *pb = _bbs[root_blk->pred(i)->_idx];
|
Block *pb = get_block_for_node(root_blk->pred(i));
|
||||||
if (pb->has_uncommon_code()) {
|
if (pb->has_uncommon_code()) {
|
||||||
worklist.push(pb);
|
worklist.push(pb);
|
||||||
}
|
}
|
||||||
|
@ -1401,7 +1404,7 @@ void PhaseCFG::Estimate_Block_Frequency() {
|
||||||
Block* uct = worklist.pop();
|
Block* uct = worklist.pop();
|
||||||
if (uct == _broot) continue;
|
if (uct == _broot) continue;
|
||||||
for (uint i = 1; i < uct->num_preds(); i++) {
|
for (uint i = 1; i < uct->num_preds(); i++) {
|
||||||
Block *pb = _bbs[uct->pred(i)->_idx];
|
Block *pb = get_block_for_node(uct->pred(i));
|
||||||
if (pb->_num_succs == 1) {
|
if (pb->_num_succs == 1) {
|
||||||
worklist.push(pb);
|
worklist.push(pb);
|
||||||
} else if (pb->num_fall_throughs() == 2) {
|
} else if (pb->num_fall_throughs() == 2) {
|
||||||
|
@ -1430,7 +1433,7 @@ void PhaseCFG::Estimate_Block_Frequency() {
|
||||||
Block_List worklist;
|
Block_List worklist;
|
||||||
Block* root_blk = _blocks[0];
|
Block* root_blk = _blocks[0];
|
||||||
for (uint i = 1; i < root_blk->num_preds(); i++) {
|
for (uint i = 1; i < root_blk->num_preds(); i++) {
|
||||||
Block *pb = _bbs[root_blk->pred(i)->_idx];
|
Block *pb = get_block_for_node(root_blk->pred(i));
|
||||||
if (pb->has_uncommon_code()) {
|
if (pb->has_uncommon_code()) {
|
||||||
worklist.push(pb);
|
worklist.push(pb);
|
||||||
}
|
}
|
||||||
|
@ -1439,7 +1442,7 @@ void PhaseCFG::Estimate_Block_Frequency() {
|
||||||
Block* uct = worklist.pop();
|
Block* uct = worklist.pop();
|
||||||
uct->_freq = PROB_MIN;
|
uct->_freq = PROB_MIN;
|
||||||
for (uint i = 1; i < uct->num_preds(); i++) {
|
for (uint i = 1; i < uct->num_preds(); i++) {
|
||||||
Block *pb = _bbs[uct->pred(i)->_idx];
|
Block *pb = get_block_for_node(uct->pred(i));
|
||||||
if (pb->_num_succs == 1 && pb->_freq > PROB_MIN) {
|
if (pb->_num_succs == 1 && pb->_freq > PROB_MIN) {
|
||||||
worklist.push(pb);
|
worklist.push(pb);
|
||||||
}
|
}
|
||||||
|
@ -1499,7 +1502,7 @@ CFGLoop* PhaseCFG::create_loop_tree() {
|
||||||
Block* loop_head = b;
|
Block* loop_head = b;
|
||||||
assert(loop_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
|
assert(loop_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
|
||||||
Node* tail_n = loop_head->pred(LoopNode::LoopBackControl);
|
Node* tail_n = loop_head->pred(LoopNode::LoopBackControl);
|
||||||
Block* tail = _bbs[tail_n->_idx];
|
Block* tail = get_block_for_node(tail_n);
|
||||||
|
|
||||||
// Defensively filter out Loop nodes for non-single-entry loops.
|
// Defensively filter out Loop nodes for non-single-entry loops.
|
||||||
// For all reasonable loops, the head occurs before the tail in RPO.
|
// For all reasonable loops, the head occurs before the tail in RPO.
|
||||||
|
@ -1514,13 +1517,13 @@ CFGLoop* PhaseCFG::create_loop_tree() {
|
||||||
loop_head->_loop = nloop;
|
loop_head->_loop = nloop;
|
||||||
// Add to nloop so push_pred() will skip over inner loops
|
// Add to nloop so push_pred() will skip over inner loops
|
||||||
nloop->add_member(loop_head);
|
nloop->add_member(loop_head);
|
||||||
nloop->push_pred(loop_head, LoopNode::LoopBackControl, worklist, _bbs);
|
nloop->push_pred(loop_head, LoopNode::LoopBackControl, worklist, this);
|
||||||
|
|
||||||
while (worklist.size() > 0) {
|
while (worklist.size() > 0) {
|
||||||
Block* member = worklist.pop();
|
Block* member = worklist.pop();
|
||||||
if (member != loop_head) {
|
if (member != loop_head) {
|
||||||
for (uint j = 1; j < member->num_preds(); j++) {
|
for (uint j = 1; j < member->num_preds(); j++) {
|
||||||
nloop->push_pred(member, j, worklist, _bbs);
|
nloop->push_pred(member, j, worklist, this);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -1557,9 +1560,9 @@ CFGLoop* PhaseCFG::create_loop_tree() {
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------push_pred--------------------------------------
|
//------------------------------push_pred--------------------------------------
|
||||||
void CFGLoop::push_pred(Block* blk, int i, Block_List& worklist, Block_Array& node_to_blk) {
|
void CFGLoop::push_pred(Block* blk, int i, Block_List& worklist, PhaseCFG* cfg) {
|
||||||
Node* pred_n = blk->pred(i);
|
Node* pred_n = blk->pred(i);
|
||||||
Block* pred = node_to_blk[pred_n->_idx];
|
Block* pred = cfg->get_block_for_node(pred_n);
|
||||||
CFGLoop *pred_loop = pred->_loop;
|
CFGLoop *pred_loop = pred->_loop;
|
||||||
if (pred_loop == NULL) {
|
if (pred_loop == NULL) {
|
||||||
// Filter out blocks for non-single-entry loops.
|
// Filter out blocks for non-single-entry loops.
|
||||||
|
@ -1580,7 +1583,7 @@ void CFGLoop::push_pred(Block* blk, int i, Block_List& worklist, Block_Array& no
|
||||||
Block* pred_head = pred_loop->head();
|
Block* pred_head = pred_loop->head();
|
||||||
assert(pred_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
|
assert(pred_head->num_preds() - 1 == 2, "loop must have 2 predecessors");
|
||||||
assert(pred_head != head(), "loop head in only one loop");
|
assert(pred_head != head(), "loop head in only one loop");
|
||||||
push_pred(pred_head, LoopNode::EntryControl, worklist, node_to_blk);
|
push_pred(pred_head, LoopNode::EntryControl, worklist, cfg);
|
||||||
} else {
|
} else {
|
||||||
assert(pred_loop->_parent == this && _parent == NULL, "just checking");
|
assert(pred_loop->_parent == this && _parent == NULL, "just checking");
|
||||||
}
|
}
|
||||||
|
|
|
@ -413,9 +413,9 @@ void IdealGraphPrinter::visit_node(Node *n, bool edges, VectorSet* temp_set) {
|
||||||
print_prop("debug_idx", node->_debug_idx);
|
print_prop("debug_idx", node->_debug_idx);
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
if(C->cfg() != NULL) {
|
if (C->cfg() != NULL) {
|
||||||
Block *block = C->cfg()->_bbs[node->_idx];
|
Block* block = C->cfg()->get_block_for_node(node);
|
||||||
if(block == NULL) {
|
if (block == NULL) {
|
||||||
print_prop("block", C->cfg()->_blocks[0]->_pre_order);
|
print_prop("block", C->cfg()->_blocks[0]->_pre_order);
|
||||||
} else {
|
} else {
|
||||||
print_prop("block", block->_pre_order);
|
print_prop("block", block->_pre_order);
|
||||||
|
|
|
@ -565,7 +565,7 @@ uint PhaseChaitin::build_ifg_physical( ResourceArea *a ) {
|
||||||
lrgs(r)._def = 0;
|
lrgs(r)._def = 0;
|
||||||
}
|
}
|
||||||
n->disconnect_inputs(NULL, C);
|
n->disconnect_inputs(NULL, C);
|
||||||
_cfg._bbs.map(n->_idx,NULL);
|
_cfg.unmap_node_from_block(n);
|
||||||
n->replace_by(C->top());
|
n->replace_by(C->top());
|
||||||
// Since yanking a Node from block, high pressure moves up one
|
// Since yanking a Node from block, high pressure moves up one
|
||||||
hrp_index[0]--;
|
hrp_index[0]--;
|
||||||
|
@ -607,7 +607,7 @@ uint PhaseChaitin::build_ifg_physical( ResourceArea *a ) {
|
||||||
if( n->is_SpillCopy()
|
if( n->is_SpillCopy()
|
||||||
&& lrgs(r).is_singledef() // MultiDef live range can still split
|
&& lrgs(r).is_singledef() // MultiDef live range can still split
|
||||||
&& n->outcnt() == 1 // and use must be in this block
|
&& n->outcnt() == 1 // and use must be in this block
|
||||||
&& _cfg._bbs[n->unique_out()->_idx] == b ) {
|
&& _cfg.get_block_for_node(n->unique_out()) == b ) {
|
||||||
// All single-use MachSpillCopy(s) that immediately precede their
|
// All single-use MachSpillCopy(s) that immediately precede their
|
||||||
// use must color early. If a longer live range steals their
|
// use must color early. If a longer live range steals their
|
||||||
// color, the spill copy will split and may push another spill copy
|
// color, the spill copy will split and may push another spill copy
|
||||||
|
|
|
@ -237,7 +237,7 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
}
|
}
|
||||||
|
|
||||||
// Check ctrl input to see if the null-check dominates the memory op
|
// Check ctrl input to see if the null-check dominates the memory op
|
||||||
Block *cb = cfg->_bbs[mach->_idx];
|
Block *cb = cfg->get_block_for_node(mach);
|
||||||
cb = cb->_idom; // Always hoist at least 1 block
|
cb = cb->_idom; // Always hoist at least 1 block
|
||||||
if( !was_store ) { // Stores can be hoisted only one block
|
if( !was_store ) { // Stores can be hoisted only one block
|
||||||
while( cb->_dom_depth > (_dom_depth + 1))
|
while( cb->_dom_depth > (_dom_depth + 1))
|
||||||
|
@ -262,7 +262,7 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
if( is_decoden ) continue;
|
if( is_decoden ) continue;
|
||||||
}
|
}
|
||||||
// Block of memory-op input
|
// Block of memory-op input
|
||||||
Block *inb = cfg->_bbs[mach->in(j)->_idx];
|
Block *inb = cfg->get_block_for_node(mach->in(j));
|
||||||
Block *b = this; // Start from nul check
|
Block *b = this; // Start from nul check
|
||||||
while( b != inb && b->_dom_depth > inb->_dom_depth )
|
while( b != inb && b->_dom_depth > inb->_dom_depth )
|
||||||
b = b->_idom; // search upwards for input
|
b = b->_idom; // search upwards for input
|
||||||
|
@ -272,7 +272,7 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
}
|
}
|
||||||
if( j > 0 )
|
if( j > 0 )
|
||||||
continue;
|
continue;
|
||||||
Block *mb = cfg->_bbs[mach->_idx];
|
Block *mb = cfg->get_block_for_node(mach);
|
||||||
// Hoisting stores requires more checks for the anti-dependence case.
|
// Hoisting stores requires more checks for the anti-dependence case.
|
||||||
// Give up hoisting if we have to move the store past any load.
|
// Give up hoisting if we have to move the store past any load.
|
||||||
if( was_store ) {
|
if( was_store ) {
|
||||||
|
@ -291,7 +291,7 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
break; // Found anti-dependent load
|
break; // Found anti-dependent load
|
||||||
// Make sure control does not do a merge (would have to check allpaths)
|
// Make sure control does not do a merge (would have to check allpaths)
|
||||||
if( b->num_preds() != 2 ) break;
|
if( b->num_preds() != 2 ) break;
|
||||||
b = cfg->_bbs[b->pred(1)->_idx]; // Move up to predecessor block
|
b = cfg->get_block_for_node(b->pred(1)); // Move up to predecessor block
|
||||||
}
|
}
|
||||||
if( b != this ) continue;
|
if( b != this ) continue;
|
||||||
}
|
}
|
||||||
|
@ -303,15 +303,15 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
|
|
||||||
// Found a candidate! Pick one with least dom depth - the highest
|
// Found a candidate! Pick one with least dom depth - the highest
|
||||||
// in the dom tree should be closest to the null check.
|
// in the dom tree should be closest to the null check.
|
||||||
if( !best ||
|
if (best == NULL || cfg->get_block_for_node(mach)->_dom_depth < cfg->get_block_for_node(best)->_dom_depth) {
|
||||||
cfg->_bbs[mach->_idx]->_dom_depth < cfg->_bbs[best->_idx]->_dom_depth ) {
|
|
||||||
best = mach;
|
best = mach;
|
||||||
bidx = vidx;
|
bidx = vidx;
|
||||||
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
// No candidate!
|
// No candidate!
|
||||||
if( !best ) return;
|
if (best == NULL) {
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
// ---- Found an implicit null check
|
// ---- Found an implicit null check
|
||||||
extern int implicit_null_checks;
|
extern int implicit_null_checks;
|
||||||
|
@ -319,29 +319,29 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
|
|
||||||
if( is_decoden ) {
|
if( is_decoden ) {
|
||||||
// Check if we need to hoist decodeHeapOop_not_null first.
|
// Check if we need to hoist decodeHeapOop_not_null first.
|
||||||
Block *valb = cfg->_bbs[val->_idx];
|
Block *valb = cfg->get_block_for_node(val);
|
||||||
if( this != valb && this->_dom_depth < valb->_dom_depth ) {
|
if( this != valb && this->_dom_depth < valb->_dom_depth ) {
|
||||||
// Hoist it up to the end of the test block.
|
// Hoist it up to the end of the test block.
|
||||||
valb->find_remove(val);
|
valb->find_remove(val);
|
||||||
this->add_inst(val);
|
this->add_inst(val);
|
||||||
cfg->_bbs.map(val->_idx,this);
|
cfg->map_node_to_block(val, this);
|
||||||
// DecodeN on x86 may kill flags. Check for flag-killing projections
|
// DecodeN on x86 may kill flags. Check for flag-killing projections
|
||||||
// that also need to be hoisted.
|
// that also need to be hoisted.
|
||||||
for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) {
|
for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) {
|
||||||
Node* n = val->fast_out(j);
|
Node* n = val->fast_out(j);
|
||||||
if( n->is_MachProj() ) {
|
if( n->is_MachProj() ) {
|
||||||
cfg->_bbs[n->_idx]->find_remove(n);
|
cfg->get_block_for_node(n)->find_remove(n);
|
||||||
this->add_inst(n);
|
this->add_inst(n);
|
||||||
cfg->_bbs.map(n->_idx,this);
|
cfg->map_node_to_block(n, this);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
// Hoist the memory candidate up to the end of the test block.
|
// Hoist the memory candidate up to the end of the test block.
|
||||||
Block *old_block = cfg->_bbs[best->_idx];
|
Block *old_block = cfg->get_block_for_node(best);
|
||||||
old_block->find_remove(best);
|
old_block->find_remove(best);
|
||||||
add_inst(best);
|
add_inst(best);
|
||||||
cfg->_bbs.map(best->_idx,this);
|
cfg->map_node_to_block(best, this);
|
||||||
|
|
||||||
// Move the control dependence
|
// Move the control dependence
|
||||||
if (best->in(0) && best->in(0) == old_block->_nodes[0])
|
if (best->in(0) && best->in(0) == old_block->_nodes[0])
|
||||||
|
@ -352,9 +352,9 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) {
|
for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) {
|
||||||
Node* n = best->fast_out(j);
|
Node* n = best->fast_out(j);
|
||||||
if( n->is_MachProj() ) {
|
if( n->is_MachProj() ) {
|
||||||
cfg->_bbs[n->_idx]->find_remove(n);
|
cfg->get_block_for_node(n)->find_remove(n);
|
||||||
add_inst(n);
|
add_inst(n);
|
||||||
cfg->_bbs.map(n->_idx,this);
|
cfg->map_node_to_block(n, this);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -385,7 +385,7 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
|
||||||
Node *old_tst = proj->in(0);
|
Node *old_tst = proj->in(0);
|
||||||
MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx);
|
MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx);
|
||||||
_nodes.map(end_idx(),nul_chk);
|
_nodes.map(end_idx(),nul_chk);
|
||||||
cfg->_bbs.map(nul_chk->_idx,this);
|
cfg->map_node_to_block(nul_chk, this);
|
||||||
// Redirect users of old_test to nul_chk
|
// Redirect users of old_test to nul_chk
|
||||||
for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2)
|
for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2)
|
||||||
old_tst->last_out(i2)->set_req(0, nul_chk);
|
old_tst->last_out(i2)->set_req(0, nul_chk);
|
||||||
|
@ -468,7 +468,7 @@ Node *Block::select(PhaseCFG *cfg, Node_List &worklist, GrowableArray<int> &read
|
||||||
Node* use = n->fast_out(j);
|
Node* use = n->fast_out(j);
|
||||||
|
|
||||||
// The use is a conditional branch, make them adjacent
|
// The use is a conditional branch, make them adjacent
|
||||||
if (use->is_MachIf() && cfg->_bbs[use->_idx]==this ) {
|
if (use->is_MachIf() && cfg->get_block_for_node(use) == this) {
|
||||||
found_machif = true;
|
found_machif = true;
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
|
@ -529,13 +529,14 @@ Node *Block::select(PhaseCFG *cfg, Node_List &worklist, GrowableArray<int> &read
|
||||||
|
|
||||||
|
|
||||||
//------------------------------set_next_call----------------------------------
|
//------------------------------set_next_call----------------------------------
|
||||||
void Block::set_next_call( Node *n, VectorSet &next_call, Block_Array &bbs ) {
|
void Block::set_next_call( Node *n, VectorSet &next_call, PhaseCFG* cfg) {
|
||||||
if( next_call.test_set(n->_idx) ) return;
|
if( next_call.test_set(n->_idx) ) return;
|
||||||
for( uint i=0; i<n->len(); i++ ) {
|
for( uint i=0; i<n->len(); i++ ) {
|
||||||
Node *m = n->in(i);
|
Node *m = n->in(i);
|
||||||
if( !m ) continue; // must see all nodes in block that precede call
|
if( !m ) continue; // must see all nodes in block that precede call
|
||||||
if( bbs[m->_idx] == this )
|
if (cfg->get_block_for_node(m) == this) {
|
||||||
set_next_call( m, next_call, bbs );
|
set_next_call(m, next_call, cfg);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -545,12 +546,12 @@ void Block::set_next_call( Node *n, VectorSet &next_call, Block_Array &bbs ) {
|
||||||
// next subroutine call get priority - basically it moves things NOT needed
|
// next subroutine call get priority - basically it moves things NOT needed
|
||||||
// for the next call till after the call. This prevents me from trying to
|
// for the next call till after the call. This prevents me from trying to
|
||||||
// carry lots of stuff live across a call.
|
// carry lots of stuff live across a call.
|
||||||
void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, Block_Array &bbs) {
|
void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, PhaseCFG* cfg) {
|
||||||
// Find the next control-defining Node in this block
|
// Find the next control-defining Node in this block
|
||||||
Node* call = NULL;
|
Node* call = NULL;
|
||||||
for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) {
|
for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) {
|
||||||
Node* m = this_call->fast_out(i);
|
Node* m = this_call->fast_out(i);
|
||||||
if( bbs[m->_idx] == this && // Local-block user
|
if(cfg->get_block_for_node(m) == this && // Local-block user
|
||||||
m != this_call && // Not self-start node
|
m != this_call && // Not self-start node
|
||||||
m->is_MachCall() )
|
m->is_MachCall() )
|
||||||
call = m;
|
call = m;
|
||||||
|
@ -558,7 +559,7 @@ void Block::needed_for_next_call(Node *this_call, VectorSet &next_call, Block_Ar
|
||||||
}
|
}
|
||||||
if (call == NULL) return; // No next call (e.g., block end is near)
|
if (call == NULL) return; // No next call (e.g., block end is near)
|
||||||
// Set next-call for all inputs to this call
|
// Set next-call for all inputs to this call
|
||||||
set_next_call(call, next_call, bbs);
|
set_next_call(call, next_call, cfg);
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------add_call_kills-------------------------------------
|
//------------------------------add_call_kills-------------------------------------
|
||||||
|
@ -578,7 +579,7 @@ void Block::add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_p
|
||||||
|
|
||||||
|
|
||||||
//------------------------------sched_call-------------------------------------
|
//------------------------------sched_call-------------------------------------
|
||||||
uint Block::sched_call( Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
|
uint Block::sched_call( Matcher &matcher, PhaseCFG* cfg, uint node_cnt, Node_List &worklist, GrowableArray<int> &ready_cnt, MachCallNode *mcall, VectorSet &next_call ) {
|
||||||
RegMask regs;
|
RegMask regs;
|
||||||
|
|
||||||
// Schedule all the users of the call right now. All the users are
|
// Schedule all the users of the call right now. All the users are
|
||||||
|
@ -597,12 +598,14 @@ uint Block::sched_call( Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_
|
||||||
// Check for scheduling the next control-definer
|
// Check for scheduling the next control-definer
|
||||||
if( n->bottom_type() == Type::CONTROL )
|
if( n->bottom_type() == Type::CONTROL )
|
||||||
// Warm up next pile of heuristic bits
|
// Warm up next pile of heuristic bits
|
||||||
needed_for_next_call(n, next_call, bbs);
|
needed_for_next_call(n, next_call, cfg);
|
||||||
|
|
||||||
// Children of projections are now all ready
|
// Children of projections are now all ready
|
||||||
for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
|
for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
|
||||||
Node* m = n->fast_out(j); // Get user
|
Node* m = n->fast_out(j); // Get user
|
||||||
if( bbs[m->_idx] != this ) continue;
|
if(cfg->get_block_for_node(m) != this) {
|
||||||
|
continue;
|
||||||
|
}
|
||||||
if( m->is_Phi() ) continue;
|
if( m->is_Phi() ) continue;
|
||||||
int m_cnt = ready_cnt.at(m->_idx)-1;
|
int m_cnt = ready_cnt.at(m->_idx)-1;
|
||||||
ready_cnt.at_put(m->_idx, m_cnt);
|
ready_cnt.at_put(m->_idx, m_cnt);
|
||||||
|
@ -620,7 +623,7 @@ uint Block::sched_call( Matcher &matcher, Block_Array &bbs, uint node_cnt, Node_
|
||||||
uint r_cnt = mcall->tf()->range()->cnt();
|
uint r_cnt = mcall->tf()->range()->cnt();
|
||||||
int op = mcall->ideal_Opcode();
|
int op = mcall->ideal_Opcode();
|
||||||
MachProjNode *proj = new (matcher.C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
|
MachProjNode *proj = new (matcher.C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj );
|
||||||
bbs.map(proj->_idx,this);
|
cfg->map_node_to_block(proj, this);
|
||||||
_nodes.insert(node_cnt++, proj);
|
_nodes.insert(node_cnt++, proj);
|
||||||
|
|
||||||
// Select the right register save policy.
|
// Select the right register save policy.
|
||||||
|
@ -708,7 +711,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
uint local = 0;
|
uint local = 0;
|
||||||
for( uint j=0; j<cnt; j++ ) {
|
for( uint j=0; j<cnt; j++ ) {
|
||||||
Node *m = n->in(j);
|
Node *m = n->in(j);
|
||||||
if( m && cfg->_bbs[m->_idx] == this && !m->is_top() )
|
if( m && cfg->get_block_for_node(m) == this && !m->is_top() )
|
||||||
local++; // One more block-local input
|
local++; // One more block-local input
|
||||||
}
|
}
|
||||||
ready_cnt.at_put(n->_idx, local); // Count em up
|
ready_cnt.at_put(n->_idx, local); // Count em up
|
||||||
|
@ -720,7 +723,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
for (uint prec = n->req(); prec < n->len(); prec++) {
|
for (uint prec = n->req(); prec < n->len(); prec++) {
|
||||||
Node* oop_store = n->in(prec);
|
Node* oop_store = n->in(prec);
|
||||||
if (oop_store != NULL) {
|
if (oop_store != NULL) {
|
||||||
assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
|
assert(cfg->get_block_for_node(oop_store)->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -753,7 +756,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
Node *n = _nodes[i3]; // Get pre-scheduled
|
Node *n = _nodes[i3]; // Get pre-scheduled
|
||||||
for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
|
for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
|
||||||
Node* m = n->fast_out(j);
|
Node* m = n->fast_out(j);
|
||||||
if( cfg->_bbs[m->_idx] ==this ) { // Local-block user
|
if (cfg->get_block_for_node(m) == this) { // Local-block user
|
||||||
int m_cnt = ready_cnt.at(m->_idx)-1;
|
int m_cnt = ready_cnt.at(m->_idx)-1;
|
||||||
ready_cnt.at_put(m->_idx, m_cnt); // Fix ready count
|
ready_cnt.at_put(m->_idx, m_cnt); // Fix ready count
|
||||||
}
|
}
|
||||||
|
@ -786,7 +789,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
}
|
}
|
||||||
|
|
||||||
// Warm up the 'next_call' heuristic bits
|
// Warm up the 'next_call' heuristic bits
|
||||||
needed_for_next_call(_nodes[0], next_call, cfg->_bbs);
|
needed_for_next_call(_nodes[0], next_call, cfg);
|
||||||
|
|
||||||
#ifndef PRODUCT
|
#ifndef PRODUCT
|
||||||
if (cfg->trace_opto_pipelining()) {
|
if (cfg->trace_opto_pipelining()) {
|
||||||
|
@ -837,7 +840,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
#endif
|
#endif
|
||||||
if( n->is_MachCall() ) {
|
if( n->is_MachCall() ) {
|
||||||
MachCallNode *mcall = n->as_MachCall();
|
MachCallNode *mcall = n->as_MachCall();
|
||||||
phi_cnt = sched_call(matcher, cfg->_bbs, phi_cnt, worklist, ready_cnt, mcall, next_call);
|
phi_cnt = sched_call(matcher, cfg, phi_cnt, worklist, ready_cnt, mcall, next_call);
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -847,7 +850,7 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
regs.OR(n->out_RegMask());
|
regs.OR(n->out_RegMask());
|
||||||
|
|
||||||
MachProjNode *proj = new (matcher.C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj );
|
MachProjNode *proj = new (matcher.C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj );
|
||||||
cfg->_bbs.map(proj->_idx,this);
|
cfg->map_node_to_block(proj, this);
|
||||||
_nodes.insert(phi_cnt++, proj);
|
_nodes.insert(phi_cnt++, proj);
|
||||||
|
|
||||||
add_call_kills(proj, regs, matcher._c_reg_save_policy, false);
|
add_call_kills(proj, regs, matcher._c_reg_save_policy, false);
|
||||||
|
@ -856,7 +859,9 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, GrowableArray<int> &
|
||||||
// Children are now all ready
|
// Children are now all ready
|
||||||
for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) {
|
for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) {
|
||||||
Node* m = n->fast_out(i5); // Get user
|
Node* m = n->fast_out(i5); // Get user
|
||||||
if( cfg->_bbs[m->_idx] != this ) continue;
|
if (cfg->get_block_for_node(m) != this) {
|
||||||
|
continue;
|
||||||
|
}
|
||||||
if( m->is_Phi() ) continue;
|
if( m->is_Phi() ) continue;
|
||||||
if (m->_idx >= max_idx) { // new node, skip it
|
if (m->_idx >= max_idx) { // new node, skip it
|
||||||
assert(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call(), "unexpected node types");
|
assert(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call(), "unexpected node types");
|
||||||
|
@ -914,7 +919,7 @@ static void catch_cleanup_fix_all_inputs(Node *use, Node *old_def, Node *new_def
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------catch_cleanup_find_cloned_def------------------
|
//------------------------------catch_cleanup_find_cloned_def------------------
|
||||||
static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
|
static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
|
||||||
assert( use_blk != def_blk, "Inter-block cleanup only");
|
assert( use_blk != def_blk, "Inter-block cleanup only");
|
||||||
|
|
||||||
// The use is some block below the Catch. Find and return the clone of the def
|
// The use is some block below the Catch. Find and return the clone of the def
|
||||||
|
@ -940,7 +945,8 @@ static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def
|
||||||
// PhiNode, the PhiNode uses from the def and IT's uses need fixup.
|
// PhiNode, the PhiNode uses from the def and IT's uses need fixup.
|
||||||
Node_Array inputs = new Node_List(Thread::current()->resource_area());
|
Node_Array inputs = new Node_List(Thread::current()->resource_area());
|
||||||
for(uint k = 1; k < use_blk->num_preds(); k++) {
|
for(uint k = 1; k < use_blk->num_preds(); k++) {
|
||||||
inputs.map(k, catch_cleanup_find_cloned_def(bbs[use_blk->pred(k)->_idx], def, def_blk, bbs, n_clone_idx));
|
Block* block = cfg->get_block_for_node(use_blk->pred(k));
|
||||||
|
inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, cfg, n_clone_idx));
|
||||||
}
|
}
|
||||||
|
|
||||||
// Check to see if the use_blk already has an identical phi inserted.
|
// Check to see if the use_blk already has an identical phi inserted.
|
||||||
|
@ -962,7 +968,7 @@ static Node *catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def
|
||||||
if (fixup == NULL) {
|
if (fixup == NULL) {
|
||||||
Node *new_phi = PhiNode::make(use_blk->head(), def);
|
Node *new_phi = PhiNode::make(use_blk->head(), def);
|
||||||
use_blk->_nodes.insert(1, new_phi);
|
use_blk->_nodes.insert(1, new_phi);
|
||||||
bbs.map(new_phi->_idx, use_blk);
|
cfg->map_node_to_block(new_phi, use_blk);
|
||||||
for (uint k = 1; k < use_blk->num_preds(); k++) {
|
for (uint k = 1; k < use_blk->num_preds(); k++) {
|
||||||
new_phi->set_req(k, inputs[k]);
|
new_phi->set_req(k, inputs[k]);
|
||||||
}
|
}
|
||||||
|
@ -1002,17 +1008,17 @@ static void catch_cleanup_intra_block(Node *use, Node *def, Block *blk, int beg,
|
||||||
//------------------------------catch_cleanup_inter_block---------------------
|
//------------------------------catch_cleanup_inter_block---------------------
|
||||||
// Fix all input edges in use that reference "def". The use is in a different
|
// Fix all input edges in use that reference "def". The use is in a different
|
||||||
// block than the def.
|
// block than the def.
|
||||||
static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, Block_Array &bbs, int n_clone_idx) {
|
static void catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, PhaseCFG* cfg, int n_clone_idx) {
|
||||||
if( !use_blk ) return; // Can happen if the use is a precedence edge
|
if( !use_blk ) return; // Can happen if the use is a precedence edge
|
||||||
|
|
||||||
Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, bbs, n_clone_idx);
|
Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, cfg, n_clone_idx);
|
||||||
catch_cleanup_fix_all_inputs(use, def, new_def);
|
catch_cleanup_fix_all_inputs(use, def, new_def);
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------call_catch_cleanup-----------------------------
|
//------------------------------call_catch_cleanup-----------------------------
|
||||||
// If we inserted any instructions between a Call and his CatchNode,
|
// If we inserted any instructions between a Call and his CatchNode,
|
||||||
// clone the instructions on all paths below the Catch.
|
// clone the instructions on all paths below the Catch.
|
||||||
void Block::call_catch_cleanup(Block_Array &bbs, Compile* C) {
|
void Block::call_catch_cleanup(PhaseCFG* cfg, Compile* C) {
|
||||||
|
|
||||||
// End of region to clone
|
// End of region to clone
|
||||||
uint end = end_idx();
|
uint end = end_idx();
|
||||||
|
@ -1037,7 +1043,7 @@ void Block::call_catch_cleanup(Block_Array &bbs, Compile* C) {
|
||||||
// since clones dominate on each path.
|
// since clones dominate on each path.
|
||||||
Node *clone = _nodes[j-1]->clone();
|
Node *clone = _nodes[j-1]->clone();
|
||||||
sb->_nodes.insert( 1, clone );
|
sb->_nodes.insert( 1, clone );
|
||||||
bbs.map(clone->_idx,sb);
|
cfg->map_node_to_block(clone, sb);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -1054,18 +1060,19 @@ void Block::call_catch_cleanup(Block_Array &bbs, Compile* C) {
|
||||||
uint max = out->size();
|
uint max = out->size();
|
||||||
for (uint j = 0; j < max; j++) {// For all users
|
for (uint j = 0; j < max; j++) {// For all users
|
||||||
Node *use = out->pop();
|
Node *use = out->pop();
|
||||||
Block *buse = bbs[use->_idx];
|
Block *buse = cfg->get_block_for_node(use);
|
||||||
if( use->is_Phi() ) {
|
if( use->is_Phi() ) {
|
||||||
for( uint k = 1; k < use->req(); k++ )
|
for( uint k = 1; k < use->req(); k++ )
|
||||||
if( use->in(k) == n ) {
|
if( use->in(k) == n ) {
|
||||||
Node *fixup = catch_cleanup_find_cloned_def(bbs[buse->pred(k)->_idx], n, this, bbs, n_clone_idx);
|
Block* block = cfg->get_block_for_node(buse->pred(k));
|
||||||
|
Node *fixup = catch_cleanup_find_cloned_def(block, n, this, cfg, n_clone_idx);
|
||||||
use->set_req(k, fixup);
|
use->set_req(k, fixup);
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
if (this == buse) {
|
if (this == buse) {
|
||||||
catch_cleanup_intra_block(use, n, this, beg, n_clone_idx);
|
catch_cleanup_intra_block(use, n, this, beg, n_clone_idx);
|
||||||
} else {
|
} else {
|
||||||
catch_cleanup_inter_block(use, buse, n, this, bbs, n_clone_idx);
|
catch_cleanup_inter_block(use, buse, n, this, cfg, n_clone_idx);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
} // End for all users
|
} // End for all users
|
||||||
|
|
|
@ -101,7 +101,7 @@ void PhaseLive::compute(uint maxlrg) {
|
||||||
for( uint k=1; k<cnt; k++ ) {
|
for( uint k=1; k<cnt; k++ ) {
|
||||||
Node *nk = n->in(k);
|
Node *nk = n->in(k);
|
||||||
uint nkidx = nk->_idx;
|
uint nkidx = nk->_idx;
|
||||||
if( _cfg._bbs[nkidx] != b ) {
|
if (_cfg.get_block_for_node(nk) != b) {
|
||||||
uint u = _names[nkidx];
|
uint u = _names[nkidx];
|
||||||
use->insert( u );
|
use->insert( u );
|
||||||
DEBUG_ONLY(def_outside->insert( u );)
|
DEBUG_ONLY(def_outside->insert( u );)
|
||||||
|
@ -121,7 +121,7 @@ void PhaseLive::compute(uint maxlrg) {
|
||||||
|
|
||||||
// Push these live-in things to predecessors
|
// Push these live-in things to predecessors
|
||||||
for( uint l=1; l<b->num_preds(); l++ ) {
|
for( uint l=1; l<b->num_preds(); l++ ) {
|
||||||
Block *p = _cfg._bbs[b->pred(l)->_idx];
|
Block *p = _cfg.get_block_for_node(b->pred(l));
|
||||||
add_liveout( p, use, first_pass );
|
add_liveout( p, use, first_pass );
|
||||||
|
|
||||||
// PhiNode uses go in the live-out set of prior blocks.
|
// PhiNode uses go in the live-out set of prior blocks.
|
||||||
|
@ -142,8 +142,10 @@ void PhaseLive::compute(uint maxlrg) {
|
||||||
assert( delta->count(), "missing delta set" );
|
assert( delta->count(), "missing delta set" );
|
||||||
|
|
||||||
// Add new-live-in to predecessors live-out sets
|
// Add new-live-in to predecessors live-out sets
|
||||||
for( uint l=1; l<b->num_preds(); l++ )
|
for (uint l = 1; l < b->num_preds(); l++) {
|
||||||
add_liveout( _cfg._bbs[b->pred(l)->_idx], delta, first_pass );
|
Block* block = _cfg.get_block_for_node(b->pred(l));
|
||||||
|
add_liveout(block, delta, first_pass);
|
||||||
|
}
|
||||||
|
|
||||||
freeset(b);
|
freeset(b);
|
||||||
} // End of while-worklist-not-empty
|
} // End of while-worklist-not-empty
|
||||||
|
|
|
@ -42,7 +42,6 @@ class AliasInfo;
|
||||||
class AllocateArrayNode;
|
class AllocateArrayNode;
|
||||||
class AllocateNode;
|
class AllocateNode;
|
||||||
class Block;
|
class Block;
|
||||||
class Block_Array;
|
|
||||||
class BoolNode;
|
class BoolNode;
|
||||||
class BoxLockNode;
|
class BoxLockNode;
|
||||||
class CMoveNode;
|
class CMoveNode;
|
||||||
|
|
|
@ -68,7 +68,6 @@ void Compile::Output() {
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
// Make sure I can find the Start Node
|
// Make sure I can find the Start Node
|
||||||
Block_Array& bbs = _cfg->_bbs;
|
|
||||||
Block *entry = _cfg->_blocks[1];
|
Block *entry = _cfg->_blocks[1];
|
||||||
Block *broot = _cfg->_broot;
|
Block *broot = _cfg->_broot;
|
||||||
|
|
||||||
|
@ -77,8 +76,8 @@ void Compile::Output() {
|
||||||
// Replace StartNode with prolog
|
// Replace StartNode with prolog
|
||||||
MachPrologNode *prolog = new (this) MachPrologNode();
|
MachPrologNode *prolog = new (this) MachPrologNode();
|
||||||
entry->_nodes.map( 0, prolog );
|
entry->_nodes.map( 0, prolog );
|
||||||
bbs.map( prolog->_idx, entry );
|
_cfg->map_node_to_block(prolog, entry);
|
||||||
bbs.map( start->_idx, NULL ); // start is no longer in any block
|
_cfg->unmap_node_from_block(start); // start is no longer in any block
|
||||||
|
|
||||||
// Virtual methods need an unverified entry point
|
// Virtual methods need an unverified entry point
|
||||||
|
|
||||||
|
@ -117,8 +116,7 @@ void Compile::Output() {
|
||||||
if( m->is_Mach() && m->as_Mach()->ideal_Opcode() != Op_Halt ) {
|
if( m->is_Mach() && m->as_Mach()->ideal_Opcode() != Op_Halt ) {
|
||||||
MachEpilogNode *epilog = new (this) MachEpilogNode(m->as_Mach()->ideal_Opcode() == Op_Return);
|
MachEpilogNode *epilog = new (this) MachEpilogNode(m->as_Mach()->ideal_Opcode() == Op_Return);
|
||||||
b->add_inst( epilog );
|
b->add_inst( epilog );
|
||||||
bbs.map(epilog->_idx, b);
|
_cfg->map_node_to_block(epilog, b);
|
||||||
//_regalloc->set_bad(epilog->_idx); // Already initialized this way.
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -252,7 +250,7 @@ void Compile::Insert_zap_nodes() {
|
||||||
if (insert) {
|
if (insert) {
|
||||||
Node *zap = call_zap_node(n->as_MachSafePoint(), i);
|
Node *zap = call_zap_node(n->as_MachSafePoint(), i);
|
||||||
b->_nodes.insert( j, zap );
|
b->_nodes.insert( j, zap );
|
||||||
_cfg->_bbs.map( zap->_idx, b );
|
_cfg->map_node_to_block(zap, b);
|
||||||
++j;
|
++j;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -1234,7 +1232,7 @@ void Compile::fill_buffer(CodeBuffer* cb, uint* blk_starts) {
|
||||||
#ifdef ASSERT
|
#ifdef ASSERT
|
||||||
if (!b->is_connector()) {
|
if (!b->is_connector()) {
|
||||||
stringStream st;
|
stringStream st;
|
||||||
b->dump_head(&_cfg->_bbs, &st);
|
b->dump_head(_cfg, &st);
|
||||||
MacroAssembler(cb).block_comment(st.as_string());
|
MacroAssembler(cb).block_comment(st.as_string());
|
||||||
}
|
}
|
||||||
jmp_target[i] = 0;
|
jmp_target[i] = 0;
|
||||||
|
@ -1310,7 +1308,7 @@ void Compile::fill_buffer(CodeBuffer* cb, uint* blk_starts) {
|
||||||
MachNode *nop = new (this) MachNopNode(nops_cnt);
|
MachNode *nop = new (this) MachNopNode(nops_cnt);
|
||||||
b->_nodes.insert(j++, nop);
|
b->_nodes.insert(j++, nop);
|
||||||
last_inst++;
|
last_inst++;
|
||||||
_cfg->_bbs.map( nop->_idx, b );
|
_cfg->map_node_to_block(nop, b);
|
||||||
nop->emit(*cb, _regalloc);
|
nop->emit(*cb, _regalloc);
|
||||||
cb->flush_bundle(true);
|
cb->flush_bundle(true);
|
||||||
current_offset = cb->insts_size();
|
current_offset = cb->insts_size();
|
||||||
|
@ -1395,7 +1393,7 @@ void Compile::fill_buffer(CodeBuffer* cb, uint* blk_starts) {
|
||||||
if (needs_padding && replacement->avoid_back_to_back()) {
|
if (needs_padding && replacement->avoid_back_to_back()) {
|
||||||
MachNode *nop = new (this) MachNopNode();
|
MachNode *nop = new (this) MachNopNode();
|
||||||
b->_nodes.insert(j++, nop);
|
b->_nodes.insert(j++, nop);
|
||||||
_cfg->_bbs.map(nop->_idx, b);
|
_cfg->map_node_to_block(nop, b);
|
||||||
last_inst++;
|
last_inst++;
|
||||||
nop->emit(*cb, _regalloc);
|
nop->emit(*cb, _regalloc);
|
||||||
cb->flush_bundle(true);
|
cb->flush_bundle(true);
|
||||||
|
@ -1549,7 +1547,7 @@ void Compile::fill_buffer(CodeBuffer* cb, uint* blk_starts) {
|
||||||
if( padding > 0 ) {
|
if( padding > 0 ) {
|
||||||
MachNode *nop = new (this) MachNopNode(padding / nop_size);
|
MachNode *nop = new (this) MachNopNode(padding / nop_size);
|
||||||
b->_nodes.insert( b->_nodes.size(), nop );
|
b->_nodes.insert( b->_nodes.size(), nop );
|
||||||
_cfg->_bbs.map( nop->_idx, b );
|
_cfg->map_node_to_block(nop, b);
|
||||||
nop->emit(*cb, _regalloc);
|
nop->emit(*cb, _regalloc);
|
||||||
current_offset = cb->insts_size();
|
current_offset = cb->insts_size();
|
||||||
}
|
}
|
||||||
|
@ -1737,7 +1735,6 @@ uint Scheduling::_total_instructions_per_bundle[Pipeline::_max_instrs_per_cycle+
|
||||||
Scheduling::Scheduling(Arena *arena, Compile &compile)
|
Scheduling::Scheduling(Arena *arena, Compile &compile)
|
||||||
: _arena(arena),
|
: _arena(arena),
|
||||||
_cfg(compile.cfg()),
|
_cfg(compile.cfg()),
|
||||||
_bbs(compile.cfg()->_bbs),
|
|
||||||
_regalloc(compile.regalloc()),
|
_regalloc(compile.regalloc()),
|
||||||
_reg_node(arena),
|
_reg_node(arena),
|
||||||
_bundle_instr_count(0),
|
_bundle_instr_count(0),
|
||||||
|
@ -2085,8 +2082,9 @@ void Scheduling::DecrementUseCounts(Node *n, const Block *bb) {
|
||||||
if( def->is_Proj() ) // If this is a machine projection, then
|
if( def->is_Proj() ) // If this is a machine projection, then
|
||||||
def = def->in(0); // propagate usage thru to the base instruction
|
def = def->in(0); // propagate usage thru to the base instruction
|
||||||
|
|
||||||
if( _bbs[def->_idx] != bb ) // Ignore if not block-local
|
if(_cfg->get_block_for_node(def) != bb) { // Ignore if not block-local
|
||||||
continue;
|
continue;
|
||||||
|
}
|
||||||
|
|
||||||
// Compute the latency
|
// Compute the latency
|
||||||
uint l = _bundle_cycle_number + n->latency(i);
|
uint l = _bundle_cycle_number + n->latency(i);
|
||||||
|
@ -2358,9 +2356,10 @@ void Scheduling::ComputeUseCount(const Block *bb) {
|
||||||
Node *inp = n->in(k);
|
Node *inp = n->in(k);
|
||||||
if (!inp) continue;
|
if (!inp) continue;
|
||||||
assert(inp != n, "no cycles allowed" );
|
assert(inp != n, "no cycles allowed" );
|
||||||
if( _bbs[inp->_idx] == bb ) { // Block-local use?
|
if (_cfg->get_block_for_node(inp) == bb) { // Block-local use?
|
||||||
if( inp->is_Proj() ) // Skip through Proj's
|
if (inp->is_Proj()) { // Skip through Proj's
|
||||||
inp = inp->in(0);
|
inp = inp->in(0);
|
||||||
|
}
|
||||||
++_uses[inp->_idx]; // Count 1 block-local use
|
++_uses[inp->_idx]; // Count 1 block-local use
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -2643,7 +2642,7 @@ void Scheduling::anti_do_def( Block *b, Node *def, OptoReg::Name def_reg, int is
|
||||||
return;
|
return;
|
||||||
|
|
||||||
Node *pinch = _reg_node[def_reg]; // Get pinch point
|
Node *pinch = _reg_node[def_reg]; // Get pinch point
|
||||||
if( !pinch || _bbs[pinch->_idx] != b || // No pinch-point yet?
|
if ((pinch == NULL) || _cfg->get_block_for_node(pinch) != b || // No pinch-point yet?
|
||||||
is_def ) { // Check for a true def (not a kill)
|
is_def ) { // Check for a true def (not a kill)
|
||||||
_reg_node.map(def_reg,def); // Record def/kill as the optimistic pinch-point
|
_reg_node.map(def_reg,def); // Record def/kill as the optimistic pinch-point
|
||||||
return;
|
return;
|
||||||
|
@ -2669,7 +2668,7 @@ void Scheduling::anti_do_def( Block *b, Node *def, OptoReg::Name def_reg, int is
|
||||||
_cfg->C->record_method_not_compilable("too many D-U pinch points");
|
_cfg->C->record_method_not_compilable("too many D-U pinch points");
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
_bbs.map(pinch->_idx,b); // Pretend it's valid in this block (lazy init)
|
_cfg->map_node_to_block(pinch, b); // Pretend it's valid in this block (lazy init)
|
||||||
_reg_node.map(def_reg,pinch); // Record pinch-point
|
_reg_node.map(def_reg,pinch); // Record pinch-point
|
||||||
//_regalloc->set_bad(pinch->_idx); // Already initialized this way.
|
//_regalloc->set_bad(pinch->_idx); // Already initialized this way.
|
||||||
if( later_def->outcnt() == 0 || later_def->ideal_reg() == MachProjNode::fat_proj ) { // Distinguish def from kill
|
if( later_def->outcnt() == 0 || later_def->ideal_reg() == MachProjNode::fat_proj ) { // Distinguish def from kill
|
||||||
|
@ -2713,9 +2712,9 @@ void Scheduling::anti_do_use( Block *b, Node *use, OptoReg::Name use_reg ) {
|
||||||
return;
|
return;
|
||||||
Node *pinch = _reg_node[use_reg]; // Get pinch point
|
Node *pinch = _reg_node[use_reg]; // Get pinch point
|
||||||
// Check for no later def_reg/kill in block
|
// Check for no later def_reg/kill in block
|
||||||
if( pinch && _bbs[pinch->_idx] == b &&
|
if ((pinch != NULL) && _cfg->get_block_for_node(pinch) == b &&
|
||||||
// Use has to be block-local as well
|
// Use has to be block-local as well
|
||||||
_bbs[use->_idx] == b ) {
|
_cfg->get_block_for_node(use) == b) {
|
||||||
if( pinch->Opcode() == Op_Node && // Real pinch-point (not optimistic?)
|
if( pinch->Opcode() == Op_Node && // Real pinch-point (not optimistic?)
|
||||||
pinch->req() == 1 ) { // pinch not yet in block?
|
pinch->req() == 1 ) { // pinch not yet in block?
|
||||||
pinch->del_req(0); // yank pointer to later-def, also set flag
|
pinch->del_req(0); // yank pointer to later-def, also set flag
|
||||||
|
@ -2895,7 +2894,7 @@ void Scheduling::garbage_collect_pinch_nodes() {
|
||||||
int trace_cnt = 0;
|
int trace_cnt = 0;
|
||||||
for (uint k = 0; k < _reg_node.Size(); k++) {
|
for (uint k = 0; k < _reg_node.Size(); k++) {
|
||||||
Node* pinch = _reg_node[k];
|
Node* pinch = _reg_node[k];
|
||||||
if (pinch != NULL && pinch->Opcode() == Op_Node &&
|
if ((pinch != NULL) && pinch->Opcode() == Op_Node &&
|
||||||
// no predecence input edges
|
// no predecence input edges
|
||||||
(pinch->req() == pinch->len() || pinch->in(pinch->req()) == NULL) ) {
|
(pinch->req() == pinch->len() || pinch->in(pinch->req()) == NULL) ) {
|
||||||
cleanup_pinch(pinch);
|
cleanup_pinch(pinch);
|
||||||
|
|
|
@ -96,9 +96,6 @@ private:
|
||||||
// List of nodes currently available for choosing for scheduling
|
// List of nodes currently available for choosing for scheduling
|
||||||
Node_List _available;
|
Node_List _available;
|
||||||
|
|
||||||
// Mapping from node (index) to basic block
|
|
||||||
Block_Array& _bbs;
|
|
||||||
|
|
||||||
// For each instruction beginning a bundle, the number of following
|
// For each instruction beginning a bundle, the number of following
|
||||||
// nodes to be bundled with it.
|
// nodes to be bundled with it.
|
||||||
Bundle *_node_bundling_base;
|
Bundle *_node_bundling_base;
|
||||||
|
|
|
@ -78,11 +78,13 @@ bool PhaseChaitin::may_be_copy_of_callee( Node *def ) const {
|
||||||
// Helper function for yank_if_dead
|
// Helper function for yank_if_dead
|
||||||
int PhaseChaitin::yank( Node *old, Block *current_block, Node_List *value, Node_List *regnd ) {
|
int PhaseChaitin::yank( Node *old, Block *current_block, Node_List *value, Node_List *regnd ) {
|
||||||
int blk_adjust=0;
|
int blk_adjust=0;
|
||||||
Block *oldb = _cfg._bbs[old->_idx];
|
Block *oldb = _cfg.get_block_for_node(old);
|
||||||
oldb->find_remove(old);
|
oldb->find_remove(old);
|
||||||
// Count 1 if deleting an instruction from the current block
|
// Count 1 if deleting an instruction from the current block
|
||||||
if( oldb == current_block ) blk_adjust++;
|
if (oldb == current_block) {
|
||||||
_cfg._bbs.map(old->_idx,NULL);
|
blk_adjust++;
|
||||||
|
}
|
||||||
|
_cfg.unmap_node_from_block(old);
|
||||||
OptoReg::Name old_reg = lrgs(_lrg_map.live_range_id(old)).reg();
|
OptoReg::Name old_reg = lrgs(_lrg_map.live_range_id(old)).reg();
|
||||||
if( regnd && (*regnd)[old_reg]==old ) { // Instruction is currently available?
|
if( regnd && (*regnd)[old_reg]==old ) { // Instruction is currently available?
|
||||||
value->map(old_reg,NULL); // Yank from value/regnd maps
|
value->map(old_reg,NULL); // Yank from value/regnd maps
|
||||||
|
@ -433,7 +435,7 @@ void PhaseChaitin::post_allocate_copy_removal() {
|
||||||
bool missing_some_inputs = false;
|
bool missing_some_inputs = false;
|
||||||
Block *freed = NULL;
|
Block *freed = NULL;
|
||||||
for( j = 1; j < b->num_preds(); j++ ) {
|
for( j = 1; j < b->num_preds(); j++ ) {
|
||||||
Block *pb = _cfg._bbs[b->pred(j)->_idx];
|
Block *pb = _cfg.get_block_for_node(b->pred(j));
|
||||||
// Remove copies along phi edges
|
// Remove copies along phi edges
|
||||||
for( uint k=1; k<phi_dex; k++ )
|
for( uint k=1; k<phi_dex; k++ )
|
||||||
elide_copy( b->_nodes[k], j, b, *blk2value[pb->_pre_order], *blk2regnd[pb->_pre_order], false );
|
elide_copy( b->_nodes[k], j, b, *blk2value[pb->_pre_order], *blk2regnd[pb->_pre_order], false );
|
||||||
|
@ -478,7 +480,7 @@ void PhaseChaitin::post_allocate_copy_removal() {
|
||||||
} else {
|
} else {
|
||||||
if( !freed ) { // Didn't get a freebie prior block
|
if( !freed ) { // Didn't get a freebie prior block
|
||||||
// Must clone some data
|
// Must clone some data
|
||||||
freed = _cfg._bbs[b->pred(1)->_idx];
|
freed = _cfg.get_block_for_node(b->pred(1));
|
||||||
Node_List &f_value = *blk2value[freed->_pre_order];
|
Node_List &f_value = *blk2value[freed->_pre_order];
|
||||||
Node_List &f_regnd = *blk2regnd[freed->_pre_order];
|
Node_List &f_regnd = *blk2regnd[freed->_pre_order];
|
||||||
for( uint k = 0; k < (uint)_max_reg; k++ ) {
|
for( uint k = 0; k < (uint)_max_reg; k++ ) {
|
||||||
|
@ -488,7 +490,7 @@ void PhaseChaitin::post_allocate_copy_removal() {
|
||||||
}
|
}
|
||||||
// Merge all inputs together, setting to NULL any conflicts.
|
// Merge all inputs together, setting to NULL any conflicts.
|
||||||
for( j = 1; j < b->num_preds(); j++ ) {
|
for( j = 1; j < b->num_preds(); j++ ) {
|
||||||
Block *pb = _cfg._bbs[b->pred(j)->_idx];
|
Block *pb = _cfg.get_block_for_node(b->pred(j));
|
||||||
if( pb == freed ) continue; // Did self already via freelist
|
if( pb == freed ) continue; // Did self already via freelist
|
||||||
Node_List &p_regnd = *blk2regnd[pb->_pre_order];
|
Node_List &p_regnd = *blk2regnd[pb->_pre_order];
|
||||||
for( uint k = 0; k < (uint)_max_reg; k++ ) {
|
for( uint k = 0; k < (uint)_max_reg; k++ ) {
|
||||||
|
@ -515,8 +517,9 @@ void PhaseChaitin::post_allocate_copy_removal() {
|
||||||
u = u ? NodeSentinel : x; // Capture unique input, or NodeSentinel for 2nd input
|
u = u ? NodeSentinel : x; // Capture unique input, or NodeSentinel for 2nd input
|
||||||
}
|
}
|
||||||
if( u != NodeSentinel ) { // Junk Phi. Remove
|
if( u != NodeSentinel ) { // Junk Phi. Remove
|
||||||
b->_nodes.remove(j--); phi_dex--;
|
b->_nodes.remove(j--);
|
||||||
_cfg._bbs.map(phi->_idx,NULL);
|
phi_dex--;
|
||||||
|
_cfg.unmap_node_from_block(phi);
|
||||||
phi->replace_by(u);
|
phi->replace_by(u);
|
||||||
phi->disconnect_inputs(NULL, C);
|
phi->disconnect_inputs(NULL, C);
|
||||||
continue;
|
continue;
|
||||||
|
|
|
@ -132,7 +132,7 @@ void PhaseChaitin::insert_proj( Block *b, uint i, Node *spill, uint maxlrg ) {
|
||||||
}
|
}
|
||||||
|
|
||||||
b->_nodes.insert(i,spill); // Insert node in block
|
b->_nodes.insert(i,spill); // Insert node in block
|
||||||
_cfg._bbs.map(spill->_idx,b); // Update node->block mapping to reflect
|
_cfg.map_node_to_block(spill, b); // Update node->block mapping to reflect
|
||||||
// Adjust the point where we go hi-pressure
|
// Adjust the point where we go hi-pressure
|
||||||
if( i <= b->_ihrp_index ) b->_ihrp_index++;
|
if( i <= b->_ihrp_index ) b->_ihrp_index++;
|
||||||
if( i <= b->_fhrp_index ) b->_fhrp_index++;
|
if( i <= b->_fhrp_index ) b->_fhrp_index++;
|
||||||
|
@ -219,7 +219,7 @@ uint PhaseChaitin::split_USE( Node *def, Block *b, Node *use, uint useidx, uint
|
||||||
use->set_req(useidx, def);
|
use->set_req(useidx, def);
|
||||||
} else {
|
} else {
|
||||||
// Block and index where the use occurs.
|
// Block and index where the use occurs.
|
||||||
Block *b = _cfg._bbs[use->_idx];
|
Block *b = _cfg.get_block_for_node(use);
|
||||||
// Put the clone just prior to use
|
// Put the clone just prior to use
|
||||||
int bindex = b->find_node(use);
|
int bindex = b->find_node(use);
|
||||||
// DEF is UP, so must copy it DOWN and hook in USE
|
// DEF is UP, so must copy it DOWN and hook in USE
|
||||||
|
@ -270,7 +270,7 @@ uint PhaseChaitin::split_USE( Node *def, Block *b, Node *use, uint useidx, uint
|
||||||
int bindex;
|
int bindex;
|
||||||
// Phi input spill-copys belong at the end of the prior block
|
// Phi input spill-copys belong at the end of the prior block
|
||||||
if( use->is_Phi() ) {
|
if( use->is_Phi() ) {
|
||||||
b = _cfg._bbs[b->pred(useidx)->_idx];
|
b = _cfg.get_block_for_node(b->pred(useidx));
|
||||||
bindex = b->end_idx();
|
bindex = b->end_idx();
|
||||||
} else {
|
} else {
|
||||||
// Put the clone just prior to use
|
// Put the clone just prior to use
|
||||||
|
@ -335,7 +335,7 @@ Node *PhaseChaitin::split_Rematerialize( Node *def, Block *b, uint insidx, uint
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
|
||||||
Block *b_def = _cfg._bbs[def->_idx];
|
Block *b_def = _cfg.get_block_for_node(def);
|
||||||
int idx_def = b_def->find_node(def);
|
int idx_def = b_def->find_node(def);
|
||||||
Node *in_spill = get_spillcopy_wide( in, def, i );
|
Node *in_spill = get_spillcopy_wide( in, def, i );
|
||||||
if( !in_spill ) return 0; // Bailed out
|
if( !in_spill ) return 0; // Bailed out
|
||||||
|
@ -589,7 +589,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
UPblock[slidx] = true;
|
UPblock[slidx] = true;
|
||||||
// Record following instruction in case 'n' rematerializes and
|
// Record following instruction in case 'n' rematerializes and
|
||||||
// kills flags
|
// kills flags
|
||||||
Block *pred1 = _cfg._bbs[b->pred(1)->_idx];
|
Block *pred1 = _cfg.get_block_for_node(b->pred(1));
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -601,7 +601,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
// Grab predecessor block header
|
// Grab predecessor block header
|
||||||
n1 = b->pred(1);
|
n1 = b->pred(1);
|
||||||
// Grab the appropriate reaching def info for inpidx
|
// Grab the appropriate reaching def info for inpidx
|
||||||
pred = _cfg._bbs[n1->_idx];
|
pred = _cfg.get_block_for_node(n1);
|
||||||
pidx = pred->_pre_order;
|
pidx = pred->_pre_order;
|
||||||
Node **Ltmp = Reaches[pidx];
|
Node **Ltmp = Reaches[pidx];
|
||||||
bool *Utmp = UP[pidx];
|
bool *Utmp = UP[pidx];
|
||||||
|
@ -616,7 +616,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
// Grab predecessor block headers
|
// Grab predecessor block headers
|
||||||
n2 = b->pred(inpidx);
|
n2 = b->pred(inpidx);
|
||||||
// Grab the appropriate reaching def info for inpidx
|
// Grab the appropriate reaching def info for inpidx
|
||||||
pred = _cfg._bbs[n2->_idx];
|
pred = _cfg.get_block_for_node(n2);
|
||||||
pidx = pred->_pre_order;
|
pidx = pred->_pre_order;
|
||||||
Ltmp = Reaches[pidx];
|
Ltmp = Reaches[pidx];
|
||||||
Utmp = UP[pidx];
|
Utmp = UP[pidx];
|
||||||
|
@ -701,7 +701,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
// Grab predecessor block header
|
// Grab predecessor block header
|
||||||
n1 = b->pred(1);
|
n1 = b->pred(1);
|
||||||
// Grab the appropriate reaching def info for k
|
// Grab the appropriate reaching def info for k
|
||||||
pred = _cfg._bbs[n1->_idx];
|
pred = _cfg.get_block_for_node(n1);
|
||||||
pidx = pred->_pre_order;
|
pidx = pred->_pre_order;
|
||||||
Node **Ltmp = Reaches[pidx];
|
Node **Ltmp = Reaches[pidx];
|
||||||
bool *Utmp = UP[pidx];
|
bool *Utmp = UP[pidx];
|
||||||
|
@ -919,7 +919,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
_lrg_map.extend(def->_idx, 0);
|
_lrg_map.extend(def->_idx, 0);
|
||||||
_cfg._bbs.map(def->_idx,b);
|
_cfg.map_node_to_block(def, b);
|
||||||
n->set_req(inpidx, def);
|
n->set_req(inpidx, def);
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
@ -1291,7 +1291,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
for( insidx = 0; insidx < phis->size(); insidx++ ) {
|
for( insidx = 0; insidx < phis->size(); insidx++ ) {
|
||||||
Node *phi = phis->at(insidx);
|
Node *phi = phis->at(insidx);
|
||||||
assert(phi->is_Phi(),"This list must only contain Phi Nodes");
|
assert(phi->is_Phi(),"This list must only contain Phi Nodes");
|
||||||
Block *b = _cfg._bbs[phi->_idx];
|
Block *b = _cfg.get_block_for_node(phi);
|
||||||
// Grab the live range number
|
// Grab the live range number
|
||||||
uint lidx = _lrg_map.find_id(phi);
|
uint lidx = _lrg_map.find_id(phi);
|
||||||
uint slidx = lrg2reach[lidx];
|
uint slidx = lrg2reach[lidx];
|
||||||
|
@ -1315,7 +1315,7 @@ uint PhaseChaitin::Split(uint maxlrg, ResourceArea* split_arena) {
|
||||||
// DEF has the wrong UP/DOWN value.
|
// DEF has the wrong UP/DOWN value.
|
||||||
for( uint i = 1; i < b->num_preds(); i++ ) {
|
for( uint i = 1; i < b->num_preds(); i++ ) {
|
||||||
// Get predecessor block pre-order number
|
// Get predecessor block pre-order number
|
||||||
Block *pred = _cfg._bbs[b->pred(i)->_idx];
|
Block *pred = _cfg.get_block_for_node(b->pred(i));
|
||||||
pidx = pred->_pre_order;
|
pidx = pred->_pre_order;
|
||||||
// Grab reaching def
|
// Grab reaching def
|
||||||
Node *def = Reaches[pidx][slidx];
|
Node *def = Reaches[pidx][slidx];
|
||||||
|
|
|
@ -1098,7 +1098,7 @@ typedef BinaryTreeDictionary<Metablock, FreeList> MetablockTreeDictionary;
|
||||||
\
|
\
|
||||||
c2_nonstatic_field(PhaseCFG, _num_blocks, uint) \
|
c2_nonstatic_field(PhaseCFG, _num_blocks, uint) \
|
||||||
c2_nonstatic_field(PhaseCFG, _blocks, Block_List) \
|
c2_nonstatic_field(PhaseCFG, _blocks, Block_List) \
|
||||||
c2_nonstatic_field(PhaseCFG, _bbs, Block_Array) \
|
c2_nonstatic_field(PhaseCFG, _node_to_block_mapping, Block_Array) \
|
||||||
c2_nonstatic_field(PhaseCFG, _broot, Block*) \
|
c2_nonstatic_field(PhaseCFG, _broot, Block*) \
|
||||||
\
|
\
|
||||||
c2_nonstatic_field(PhaseRegAlloc, _node_regs, OptoRegPair*) \
|
c2_nonstatic_field(PhaseRegAlloc, _node_regs, OptoRegPair*) \
|
||||||
|
|
Loading…
Add table
Add a link
Reference in a new issue