void DeadCodeEliminationPass::runOnKernel(ir::IRKernel& k)
{
report("Running dead code elimination on kernel " << k.name);
reportE(REPORT_PTX, k);
Analysis* dfgAnalysis = getAnalysis(Analysis::DataflowGraphAnalysis);
assert(dfgAnalysis != 0);
analysis::DataflowGraph& dfg =
*static_cast<analysis::DataflowGraph*>(dfgAnalysis);
assert(dfg.ssa() != analysis::DataflowGraph::SsaType::None);
BlockSet blocks;
report(" Starting by scanning all basic blocks");
for(iterator block = dfg.begin(); block != dfg.end(); ++block)
{
report(" Queueing up BB_" << block->id());
blocks.insert(block);
}
while(!blocks.empty())
{
iterator block = *blocks.begin();
blocks.erase(blocks.begin());
eliminateDeadInstructions(dfg, blocks, block);
}
report("Finished running dead code elimination on kernel " << k.name);
reportE(REPORT_PTX, k);
}
Shape *MakeLoop(BlockSet &Blocks, BlockSet& Entries, BlockSet &NextEntries) {
// Find the inner blocks in this loop. Proceed backwards from the entries until
// you reach a seen block, collecting as you go.
BlockSet InnerBlocks;
BlockSet Queue = Entries;
while (Queue.size() > 0) {
Block *Curr = *(Queue.begin());
Queue.erase(Queue.begin());
if (InnerBlocks.find(Curr) == InnerBlocks.end()) {
// This element is new, mark it as inner and remove from outer
InnerBlocks.insert(Curr);
Blocks.erase(Curr);
// Add the elements prior to it
for (BlockBranchMap::iterator iter = Curr->BranchesIn.begin(); iter != Curr->BranchesIn.end(); iter++) {
Queue.insert(iter->first);
}
}
}
assert(InnerBlocks.size() > 0);
for (BlockSet::iterator iter = InnerBlocks.begin(); iter != InnerBlocks.end(); iter++) {
Block *Curr = *iter;
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *Possible = iter->first;
if (InnerBlocks.find(Possible) == InnerBlocks.end() &&
NextEntries.find(Possible) == NextEntries.find(Possible)) {
NextEntries.insert(Possible);
}
}
}
PrintDebug("creating loop block:\n");
DebugDump(InnerBlocks, " inner blocks:");
DebugDump(Entries, " inner entries:");
DebugDump(Blocks, " outer blocks:");
DebugDump(NextEntries, " outer entries:");
// TODO: Optionally hoist additional blocks into the loop
LoopShape *Loop = new LoopShape();
Notice(Loop);
// Solipsize the loop, replacing with break/continue and marking branches as Processed (will not affect later calculations)
// A. Branches to the loop entries become a continue to this shape
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
Solipsize(*iter, Branch::Continue, Loop, InnerBlocks);
}
// B. Branches to outside the loop (a next entry) become breaks on this shape
for (BlockSet::iterator iter = NextEntries.begin(); iter != NextEntries.end(); iter++) {
Solipsize(*iter, Branch::Break, Loop, InnerBlocks);
}
// Finish up
Shape *Inner = Process(InnerBlocks, Entries, NULL);
Loop->Inner = Inner;
return Loop;
}
// If a block has multiple entries but no exits, and it is small enough, it is useful to split it.
// A common example is a C++ function where everything ends up at a final exit block and does some
// RAII cleanup. Without splitting, we will be forced to introduce labelled loops to allow
// reaching the final block
void SplitDeadEnds() {
unsigned TotalCodeSize = 0;
for (BlockSet::iterator iter = Live.begin(); iter != Live.end(); iter++) {
Block *Curr = *iter;
TotalCodeSize += strlen(Curr->Code);
}
BlockSet Splits;
BlockSet Removed;
//DebugDump(Live, "before");
for (BlockSet::iterator iter = Live.begin(); iter != Live.end(); iter++) {
Block *Original = *iter;
if (Original->BranchesIn.size() <= 1 || Original->BranchesOut.size() > 0) continue; // only dead ends, for now
if (contains(Original->BranchesOut, Original)) continue; // cannot split a looping node
if (strlen(Original->Code)*(Original->BranchesIn.size()-1) > TotalCodeSize/5) continue; // if splitting increases raw code size by a significant amount, abort
// Split the node (for simplicity, we replace all the blocks, even though we could have reused the original)
PrintDebug("Splitting block %d\n", Original->Id);
for (BlockSet::iterator iter = Original->BranchesIn.begin(); iter != Original->BranchesIn.end(); iter++) {
Block *Prior = *iter;
Block *Split = new Block(Original->Code, Original->BranchVar);
Parent->AddBlock(Split);
PrintDebug(" to %d\n", Split->Id);
Split->BranchesIn.insert(Prior);
Branch *Details = Prior->BranchesOut[Original];
Prior->BranchesOut[Split] = new Branch(Details->Condition, Details->Code);
Prior->BranchesOut.erase(Original);
for (BlockBranchMap::iterator iter = Original->BranchesOut.begin(); iter != Original->BranchesOut.end(); iter++) {
Block *Post = iter->first;
Branch *Details = iter->second;
Split->BranchesOut[Post] = new Branch(Details->Condition, Details->Code);
Post->BranchesIn.insert(Split);
}
Splits.insert(Split);
Removed.insert(Original);
}
for (BlockBranchMap::iterator iter = Original->BranchesOut.begin(); iter != Original->BranchesOut.end(); iter++) {
Block *Post = iter->first;
Post->BranchesIn.erase(Original);
}
//DebugDump(Live, "mid");
}
for (BlockSet::iterator iter = Splits.begin(); iter != Splits.end(); iter++) {
Live.insert(*iter);
}
for (BlockSet::iterator iter = Removed.begin(); iter != Removed.end(); iter++) {
Live.erase(*iter);
}
//DebugDump(Live, "after");
}
Udm::Object MatLabUdmChart::distinguish( Udm::Object udmParent ) {
SLSF::State state;
static boost::regex stateflowRegex( "stateflow", boost::regex_constants::perl | boost::regex_constants::icase );
boost::match_results<std::string::const_iterator> results;
SLSF::Subsystem subsystemParent = SLSF::Subsystem::Cast( udmParent );
BlockSet blockSet = subsystemParent.Block_kind_children();
Udm::Object chartParent = udmParent;
for( BlockSet::iterator blsItr = blockSet.begin() ; blsItr != blockSet.end() ; ++blsItr ) {
Block block = *blsItr;
std::string tag( block.Tag() );
if ( regex_search( tag, results, stateflowRegex ) ) {
chartParent = block;
break;
}
}
#if PARADIGM == CyberComposition_PARADIGM
state = SLSF::State::Create( chartParent );
#else
state = SLSF::State::Create( UdmEngine::get_singleton().getTopLevelState() );
SLSF::ConnectorRef connectorRef = SLSF::ConnectorRef::Create( chartParent );
connectorRef.ref() = state;
#endif
return state;
}
Layer Layer::getSubgraphConnectedToTheseOutputs(
const NeuronSet& outputs) const
{
typedef std::set<size_t> BlockSet;
BlockSet blocks;
// TODO: eliminate the reundant inserts
for(auto& output : outputs)
{
size_t block = (output / getOutputBlockingFactor()) % this->blocks();
blocks.insert(block);
}
Layer layer(blocks.size(), getInputBlockingFactor(),
getOutputBlockingFactor(), blockStep());
for(auto& block : blocks)
{
size_t blockIndex = block - *blocks.begin();
layer[blockIndex] = (*this)[block];
layer.at_bias(blockIndex) = at_bias(block);
}
return layer;
}
// If a block has multiple entries but no exits, and it is small enough, it is useful to split it.
// A common example is a C++ function where everything ends up at a final exit block and does some
// RAII cleanup. Without splitting, we will be forced to introduce labelled loops to allow
// reaching the final block
void SplitDeadEnds() {
int TotalCodeSize = 0;
for (BlockSet::iterator iter = Live.begin(); iter != Live.end(); iter++) {
Block *Curr = *iter;
TotalCodeSize += strlen(Curr->Code);
}
for (BlockSet::iterator iter = Live.begin(); iter != Live.end(); iter++) {
Block *Original = *iter;
if (Original->BranchesIn.size() <= 1 || Original->BranchesOut.size() > 0) continue;
if (strlen(Original->Code)*(Original->BranchesIn.size()-1) > TotalCodeSize/5) continue; // if splitting increases raw code size by a significant amount, abort
// Split the node (for simplicity, we replace all the blocks, even though we could have reused the original)
for (BlockBranchMap::iterator iter = Original->BranchesIn.begin(); iter != Original->BranchesIn.end(); iter++) {
Block *Prior = iter->first;
Block *Split = new Block(Original->Code);
Split->BranchesIn[Prior] = new Branch(NULL);
Prior->BranchesOut[Split] = new Branch(Prior->BranchesOut[Original]->Condition, Prior->BranchesOut[Original]->Code);
Prior->BranchesOut.erase(Original);
Parent->AddBlock(Split);
Live.insert(Split);
}
}
}
void TextAutosizer::FingerprintMapper::assertMapsAreConsistent()
{
// For each fingerprint -> block mapping in m_blocksForFingerprint we should have an associated
// map from block -> fingerprint in m_fingerprints.
ReverseFingerprintMap::iterator end = m_blocksForFingerprint.end();
for (ReverseFingerprintMap::iterator fingerprintIt = m_blocksForFingerprint.begin(); fingerprintIt != end; ++fingerprintIt) {
Fingerprint fingerprint = fingerprintIt->key;
BlockSet* blocks = fingerprintIt->value.get();
for (BlockSet::iterator blockIt = blocks->begin(); blockIt != blocks->end(); ++blockIt) {
const LayoutBlock* block = (*blockIt);
ASSERT(m_fingerprints.get(block) == fingerprint);
}
}
}
void ConstantPropagationPass::runOnKernel(ir::IRKernel& k)
{
report("Running constant propagation on kernel " << k.name);
Analysis* dfgAnalysis = getAnalysis("DataflowGraphAnalysis");
assert(dfgAnalysis != 0);
analysis::DataflowGraph& dfg =
*static_cast<analysis::DataflowGraph*>(dfgAnalysis);
dfg.convertToSSAType(analysis::DataflowGraph::Minimal);
assert(dfg.ssa() == analysis::DataflowGraph::Minimal);
BlockSet blocks;
report(" Starting by scanning all basic blocks");
for(iterator block = dfg.begin(); block != dfg.end(); ++block)
{
report(" Queueing up BB_" << block->id());
blocks.insert(block);
}
while(!blocks.empty())
{
iterator block = *blocks.begin();
blocks.erase(blocks.begin());
eliminateRedundantInstructions(dfg, blocks, block);
}
report("Finished running constant propagation on kernel " << k.name);
reportE(REPORT_PTX, k);
}
Shape *MakeMultiple(BlockSet &Blocks, BlockSet& Entries, BlockBlockSetMap& IndependentGroups, Shape *Prev, BlockSet &NextEntries) {
PrintDebug("creating multiple block with %d inner groups\n", IndependentGroups.size());
bool Fused = !!(Shape::IsSimple(Prev));
MultipleShape *Multiple = new MultipleShape();
Notice(Multiple);
BlockSet CurrEntries;
for (BlockBlockSetMap::iterator iter = IndependentGroups.begin(); iter != IndependentGroups.end(); iter++) {
Block *CurrEntry = iter->first;
BlockSet &CurrBlocks = iter->second;
PrintDebug(" multiple group with entry %d:\n", CurrEntry->Id);
DebugDump(CurrBlocks, " ");
// Create inner block
CurrEntries.clear();
CurrEntries.insert(CurrEntry);
for (BlockSet::iterator iter = CurrBlocks.begin(); iter != CurrBlocks.end(); iter++) {
Block *CurrInner = *iter;
// Remove the block from the remaining blocks
Blocks.erase(CurrInner);
// Find new next entries and fix branches to them
for (BlockBranchMap::iterator iter = CurrInner->BranchesOut.begin(); iter != CurrInner->BranchesOut.end();) {
Block *CurrTarget = iter->first;
BlockBranchMap::iterator Next = iter;
Next++;
if (CurrBlocks.find(CurrTarget) == CurrBlocks.end()) {
NextEntries.insert(CurrTarget);
Solipsize(CurrTarget, Branch::Break, Multiple, CurrBlocks);
}
iter = Next; // increment carefully because Solipsize can remove us
}
}
Multiple->InnerMap[CurrEntry] = Process(CurrBlocks, CurrEntries, NULL);
// If we are not fused, then our entries will actually be checked
if (!Fused) {
CurrEntry->IsCheckedMultipleEntry = true;
}
}
DebugDump(Blocks, " remaining blocks after multiple:");
// Add entries not handled as next entries, they are deferred
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
Block *Entry = *iter;
if (IndependentGroups.find(Entry) == IndependentGroups.end()) {
NextEntries.insert(Entry);
}
}
return Multiple;
}
ControlFlowGraph::ConstBlockPointerVector
ControlFlowGraph::executable_sequence() const {
typedef std::unordered_set<const_iterator> BlockSet;
ConstBlockPointerVector sequence;
BlockSet unscheduled;
for(const_iterator i = begin(); i != end(); ++i) {
unscheduled.insert(i);
}
report("Getting executable sequence.");
sequence.push_back(get_entry_block());
unscheduled.erase(get_entry_block());
report(" added " << get_entry_block()->label());
while (!unscheduled.empty()) {
if (sequence.back()->has_fallthrough_edge()) {
const_edge_iterator fallthroughEdge
= sequence.back()->get_fallthrough_edge();
sequence.push_back(fallthroughEdge->tail);
unscheduled.erase(fallthroughEdge->tail);
}
else {
// find a new block, favor branch targets over random blocks
const_iterator next = *unscheduled.begin();
for(const_edge_pointer_iterator
edge = sequence.back()->out_edges.begin();
edge != sequence.back()->out_edges.end(); ++edge)
{
if(unscheduled.count((*edge)->tail) != 0)
{
next = (*edge)->tail;
}
}
// rewind through fallthrough edges to find the beginning of the
// next chain of fall throughs
report(" restarting at " << next->label());
bool rewinding = true;
while (rewinding) {
rewinding = false;
for (const_edge_pointer_iterator
edge = next->in_edges.begin();
edge != next->in_edges.end(); ++edge) {
if ((*edge)->type == Edge::FallThrough) {
assertM(unscheduled.count((*edge)->head) != 0,
(*edge)->head->label()
<< " has multiple fallthrough branches.");
next = (*edge)->head;
report(" rewinding to " << next->label());
rewinding = true;
break;
}
}
}
sequence.push_back(next);
unscheduled.erase(next);
}
report(" added " << sequence.back()->label());
}
return sequence;
}
// Main function.
// Process a set of blocks with specified entries, returns a shape
// The Make* functions receive a NextEntries. If they fill it with data, those are the entries for the
// ->Next block on them, and the blocks are what remains in Blocks (which Make* modify). In this way
// we avoid recursing on Next (imagine a long chain of Simples, if we recursed we could blow the stack).
Shape *Process(BlockSet &Blocks, BlockSet& InitialEntries, Shape *Prev) {
PrintDebug("Process() called\n");
BlockSet *Entries = &InitialEntries;
BlockSet TempEntries[2];
int CurrTempIndex = 0;
BlockSet *NextEntries;
Shape *Ret = NULL;
#define Make(call) \
Shape *Temp = call; \
if (Prev) Prev->Next = Temp; \
if (!Ret) Ret = Temp; \
if (!NextEntries->size()) { PrintDebug("Process() returning\n"); return Ret; } \
Prev = Temp; \
Entries = NextEntries; \
continue;
while (1) {
PrintDebug("Process() running\n");
DebugDump(Blocks, " blocks : ");
DebugDump(*Entries, " entries: ");
CurrTempIndex = 1-CurrTempIndex;
NextEntries = &TempEntries[CurrTempIndex];
NextEntries->clear();
if (Entries->size() == 0) return Ret;
if (Entries->size() == 1) {
Block *Curr = *(Entries->begin());
if (Curr->BranchesIn.size() == 0) {
// One entry, no looping ==> Simple
Make(MakeSimple(Blocks, Curr, *NextEntries));
}
// One entry, looping ==> Loop
Make(MakeLoop(Blocks, *Entries, *NextEntries));
}
// More than one entry, try to eliminate through a Multiple groups of
// independent blocks from an entry/ies. It is important to remove through
// multiples as opposed to looping since the former is more performant.
BlockBlockSetMap IndependentGroups;
FindIndependentGroups(Blocks, *Entries, IndependentGroups);
PrintDebug("Independent groups: %d\n", IndependentGroups.size());
if (IndependentGroups.size() > 0) {
// We can handle a group in a multiple if its entry cannot be reached by another group.
// Note that it might be reachable by itself - a loop. But that is fine, we will create
// a loop inside the multiple block (which is the performant order to do it).
for (BlockBlockSetMap::iterator iter = IndependentGroups.begin(); iter != IndependentGroups.end();) {
Block *Entry = iter->first;
BlockSet &Group = iter->second;
BlockBlockSetMap::iterator curr = iter++; // iterate carefully, we may delete
for (BlockBranchMap::iterator iterBranch = Entry->BranchesIn.begin(); iterBranch != Entry->BranchesIn.end(); iterBranch++) {
Block *Origin = iterBranch->first;
if (Group.find(Origin) == Group.end()) {
// Reached from outside the group, so we cannot handle this
PrintDebug("Cannot handle group with entry %d because of incoming branch from %d\n", Entry->Id, Origin->Id);
IndependentGroups.erase(curr);
break;
}
}
}
// As an optimization, if we have 2 independent groups, and one is a small dead end, we can handle only that dead end.
// The other then becomes a Next - without nesting in the code and recursion in the analysis.
// TODO: if the larger is the only dead end, handle that too
// TODO: handle >2 groups
// TODO: handle not just dead ends, but also that do not branch to the NextEntries. However, must be careful
// there since we create a Next, and that Next can prevent eliminating a break (since we no longer
// naturally reach the same place), which may necessitate a one-time loop, which makes the unnesting
// pointless.
if (IndependentGroups.size() == 2) {
// Find the smaller one
BlockBlockSetMap::iterator iter = IndependentGroups.begin();
Block *SmallEntry = iter->first;
int SmallSize = iter->second.size();
iter++;
Block *LargeEntry = iter->first;
int LargeSize = iter->second.size();
if (SmallSize != LargeSize) { // ignore the case where they are identical - keep things symmetrical there
if (SmallSize > LargeSize) {
Block *Temp = SmallEntry;
SmallEntry = LargeEntry;
LargeEntry = Temp; // Note: we did not flip the Sizes too, they are now invalid. TODO: use the smaller size as a limit?
}
// Check if dead end
bool DeadEnd = true;
BlockSet &SmallGroup = IndependentGroups[SmallEntry];
for (BlockSet::iterator iter = SmallGroup.begin(); iter != SmallGroup.end(); iter++) {
Block *Curr = *iter;
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *Target = iter->first;
if (SmallGroup.find(Target) == SmallGroup.end()) {
DeadEnd = false;
break;
}
}
if (!DeadEnd) break;
}
if (DeadEnd) {
PrintDebug("Removing nesting by not handling large group because small group is dead end\n");
IndependentGroups.erase(LargeEntry);
}
}
}
PrintDebug("Handleable independent groups: %d\n", IndependentGroups.size());
if (IndependentGroups.size() > 0) {
// Some groups removable ==> Multiple
Make(MakeMultiple(Blocks, *Entries, IndependentGroups, Prev, *NextEntries));
}
}
// No independent groups, must be loopable ==> Loop
Make(MakeLoop(Blocks, *Entries, *NextEntries));
}
}
// For each entry, find the independent group reachable by it. The independent group is
// the entry itself, plus all the blocks it can reach that cannot be directly reached by another entry. Note that we
// ignore directly reaching the entry itself by another entry.
void FindIndependentGroups(BlockSet &Blocks, BlockSet &Entries, BlockBlockSetMap& IndependentGroups) {
typedef std::map<Block*, Block*> BlockBlockMap;
struct HelperClass {
BlockBlockSetMap& IndependentGroups;
BlockBlockMap Ownership; // For each block, which entry it belongs to. We have reached it from there.
HelperClass(BlockBlockSetMap& IndependentGroupsInit) : IndependentGroups(IndependentGroupsInit) {}
void InvalidateWithChildren(Block *New) { // TODO: rename New
BlockList ToInvalidate; // Being in the list means you need to be invalidated
ToInvalidate.push_back(New);
while (ToInvalidate.size() > 0) {
Block *Invalidatee = ToInvalidate.front();
ToInvalidate.pop_front();
Block *Owner = Ownership[Invalidatee];
if (IndependentGroups.find(Owner) != IndependentGroups.end()) { // Owner may have been invalidated, do not add to IndependentGroups!
IndependentGroups[Owner].erase(Invalidatee);
}
if (Ownership[Invalidatee]) { // may have been seen before and invalidated already
Ownership[Invalidatee] = NULL;
for (BlockBranchMap::iterator iter = Invalidatee->BranchesOut.begin(); iter != Invalidatee->BranchesOut.end(); iter++) {
Block *Target = iter->first;
BlockBlockMap::iterator Known = Ownership.find(Target);
if (Known != Ownership.end()) {
Block *TargetOwner = Known->second;
if (TargetOwner) {
ToInvalidate.push_back(Target);
}
}
}
}
}
}
};
HelperClass Helper(IndependentGroups);
// We flow out from each of the entries, simultaneously.
// When we reach a new block, we add it as belonging to the one we got to it from.
// If we reach a new block that is already marked as belonging to someone, it is reachable by
// two entries and is not valid for any of them. Remove it and all it can reach that have been
// visited.
BlockList Queue; // Being in the queue means we just added this item, and we need to add its children
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
Block *Entry = *iter;
Helper.Ownership[Entry] = Entry;
IndependentGroups[Entry].insert(Entry);
Queue.push_back(Entry);
}
while (Queue.size() > 0) {
Block *Curr = Queue.front();
Queue.pop_front();
Block *Owner = Helper.Ownership[Curr]; // Curr must be in the ownership map if we are in the queue
if (!Owner) continue; // we have been invalidated meanwhile after being reached from two entries
// Add all children
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *New = iter->first;
BlockBlockMap::iterator Known = Helper.Ownership.find(New);
if (Known == Helper.Ownership.end()) {
// New node. Add it, and put it in the queue
Helper.Ownership[New] = Owner;
IndependentGroups[Owner].insert(New);
Queue.push_back(New);
continue;
}
Block *NewOwner = Known->second;
if (!NewOwner) continue; // We reached an invalidated node
if (NewOwner != Owner) {
// Invalidate this and all reachable that we have seen - we reached this from two locations
Helper.InvalidateWithChildren(New);
}
// otherwise, we have the same owner, so do nothing
}
}
// Having processed all the interesting blocks, we remain with just one potential issue:
// If a->b, and a was invalidated, but then b was later reached by someone else, we must
// invalidate b. To check for this, we go over all elements in the independent groups,
// if an element has a parent which does *not* have the same owner, we must remove it
// and all its children.
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
BlockSet &CurrGroup = IndependentGroups[*iter];
BlockList ToInvalidate;
for (BlockSet::iterator iter = CurrGroup.begin(); iter != CurrGroup.end(); iter++) {
Block *Child = *iter;
for (BlockBranchMap::iterator iter = Child->BranchesIn.begin(); iter != Child->BranchesIn.end(); iter++) {
Block *Parent = iter->first;
if (Helper.Ownership[Parent] != Helper.Ownership[Child]) {
ToInvalidate.push_back(Child);
}
}
}
while (ToInvalidate.size() > 0) {
Block *Invalidatee = ToInvalidate.front();
ToInvalidate.pop_front();
Helper.InvalidateWithChildren(Invalidatee);
}
}
// Remove empty groups
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
if (IndependentGroups[*iter].size() == 0) {
IndependentGroups.erase(*iter);
}
}
#if DEBUG
PrintDebug("Investigated independent groups:\n");
for (BlockBlockSetMap::iterator iter = IndependentGroups.begin(); iter != IndependentGroups.end(); iter++) {
DebugDump(iter->second, " group: ");
}
#endif
}
Shape *MakeLoop(BlockSet &Blocks, BlockSet& Entries, BlockSet &NextEntries) {
// Find the inner blocks in this loop. Proceed backwards from the entries until
// you reach a seen block, collecting as you go.
BlockSet InnerBlocks;
BlockSet Queue = Entries;
while (Queue.size() > 0) {
Block *Curr = *(Queue.begin());
Queue.erase(Queue.begin());
if (!contains(InnerBlocks, Curr)) {
// This element is new, mark it as inner and remove from outer
InnerBlocks.insert(Curr);
Blocks.erase(Curr);
// Add the elements prior to it
for (BlockSet::iterator iter = Curr->BranchesIn.begin(); iter != Curr->BranchesIn.end(); iter++) {
Queue.insert(*iter);
}
#if 0
// Add elements it leads to, if they are dead ends. There is no reason not to hoist dead ends
// into loops, as it can avoid multiple entries after the loop
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *Target = iter->first;
if (Target->BranchesIn.size() <= 1 && Target->BranchesOut.size() == 0) {
Queue.insert(Target);
}
}
#endif
}
}
assert(InnerBlocks.size() > 0);
for (BlockSet::iterator iter = InnerBlocks.begin(); iter != InnerBlocks.end(); iter++) {
Block *Curr = *iter;
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *Possible = iter->first;
if (!contains(InnerBlocks, Possible)) {
NextEntries.insert(Possible);
}
}
}
#if 0
// We can avoid multiple next entries by hoisting them into the loop.
if (NextEntries.size() > 1) {
BlockBlockSetMap IndependentGroups;
FindIndependentGroups(NextEntries, IndependentGroups, &InnerBlocks);
while (IndependentGroups.size() > 0 && NextEntries.size() > 1) {
Block *Min = NULL;
int MinSize = 0;
for (BlockBlockSetMap::iterator iter = IndependentGroups.begin(); iter != IndependentGroups.end(); iter++) {
Block *Entry = iter->first;
BlockSet &Blocks = iter->second;
if (!Min || Blocks.size() < MinSize) { // TODO: code size, not # of blocks
Min = Entry;
MinSize = Blocks.size();
}
}
// check how many new entries this would cause
BlockSet &Hoisted = IndependentGroups[Min];
bool abort = false;
for (BlockSet::iterator iter = Hoisted.begin(); iter != Hoisted.end() && !abort; iter++) {
Block *Curr = *iter;
for (BlockBranchMap::iterator iter = Curr->BranchesOut.begin(); iter != Curr->BranchesOut.end(); iter++) {
Block *Target = iter->first;
if (Hoisted.find(Target) == Hoisted.end() && NextEntries.find(Target) == NextEntries.end()) {
// abort this hoisting
abort = true;
break;
}
}
}
if (abort) {
IndependentGroups.erase(Min);
continue;
}
// hoist this entry
PrintDebug("hoisting %d into loop\n", Min->Id);
NextEntries.erase(Min);
for (BlockSet::iterator iter = Hoisted.begin(); iter != Hoisted.end(); iter++) {
Block *Curr = *iter;
InnerBlocks.insert(Curr);
Blocks.erase(Curr);
}
IndependentGroups.erase(Min);
}
}
#endif
PrintDebug("creating loop block:\n");
DebugDump(InnerBlocks, " inner blocks:");
DebugDump(Entries, " inner entries:");
DebugDump(Blocks, " outer blocks:");
DebugDump(NextEntries, " outer entries:");
LoopShape *Loop = new LoopShape();
Notice(Loop);
// Solipsize the loop, replacing with break/continue and marking branches as Processed (will not affect later calculations)
// A. Branches to the loop entries become a continue to this shape
for (BlockSet::iterator iter = Entries.begin(); iter != Entries.end(); iter++) {
Solipsize(*iter, Branch::Continue, Loop, InnerBlocks);
}
// B. Branches to outside the loop (a next entry) become breaks on this shape
for (BlockSet::iterator iter = NextEntries.begin(); iter != NextEntries.end(); iter++) {
Solipsize(*iter, Branch::Break, Loop, InnerBlocks);
}
// Finish up
Shape *Inner = Process(InnerBlocks, Entries, NULL);
Loop->Inner = Inner;
return Loop;
}