luke
/
verilator
mirror of https://github.com/verilator/verilator.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Speed up DfgGraph decomposition algorithms

This commit is contained in:
Geza Lore 2022-10-08 14:34:07 +01:00
parent ba052beccd
commit 2a110c91cf
4 changed files with 579 additions and 427 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/Makefile_obj.in
View File

@ -182,6 +182,7 @@ RAW_OBJS = \
V3Descope.o \ V3Descope.o \
V3Dfg.o \ V3Dfg.o \
V3DfgAstToDfg.o \ V3DfgAstToDfg.o \
V3DfgDecomposition.o \
V3DfgDfgToAst.o \ V3DfgDfgToAst.o \
V3DfgOptimizer.o \ V3DfgOptimizer.o \
V3DfgPasses.o \ V3DfgPasses.o \

428
src/V3Dfg.cpp
View File

@ -21,9 +21,6 @@
#include "V3File.h" #include "V3File.h"
#include <cctype>
#include <type_traits>
#include <unordered_map>
VL_DEFINE_DEBUG_FUNCTIONS; VL_DEFINE_DEBUG_FUNCTIONS;
@ -47,6 +44,7 @@ void DfgGraph::addGraph(DfgGraph& other) {
if (DfgVertex* vtxp = src.begin()) { if (DfgVertex* vtxp = src.begin()) {
vtxp->m_verticesEnt.moveAppend(src, dst, vtxp); vtxp->m_verticesEnt.moveAppend(src, dst, vtxp);
do { do {
vtxp->m_userCnt = 0;
vtxp->m_graphp = this; vtxp->m_graphp = this;
vtxp = vtxp->verticesNext(); vtxp = vtxp->verticesNext();
} while (vtxp); } while (vtxp);
@ -58,430 +56,6 @@ void DfgGraph::addGraph(DfgGraph& other) {
moveVertexList(other.m_opVertices, m_opVertices); moveVertexList(other.m_opVertices, m_opVertices);
} }
std::vector<std::unique_ptr<DfgGraph>> DfgGraph::splitIntoComponents(std::string label) {
size_t componentNumber = 0;
std::unordered_map<const DfgVertex*, unsigned> vertex2component;
forEachVertex([&](const DfgVertex& vtx) {
// If already assigned this vertex to a component, then continue
if (vertex2component.count(&vtx)) return;
// Work queue for depth first traversal starting from this vertex
std::vector<const DfgVertex*> queue{&vtx};
// Depth first traversal
while (!queue.empty()) {
// Pop next work item
const DfgVertex& item = *queue.back();
queue.pop_back();
// Mark vertex as belonging to current component (if it's not marked yet)
const bool isFirstEncounter = vertex2component.emplace(&item, componentNumber).second;
// If we have already visited this vertex during the traversal, then move on.
if (!isFirstEncounter) continue;
// Enqueue all sources and sinks of this vertex.
item.forEachSource([&](const DfgVertex& src) { queue.push_back(&src); });
item.forEachSink([&](const DfgVertex& dst) { queue.push_back(&dst); });
}
// Done with this component
++componentNumber;
});
// Create the component graphs
std::vector<std::unique_ptr<DfgGraph>> results{componentNumber};
const std::string prefix{name() + (label.empty() ? "" : "-") + label + "-component-"};
for (size_t i = 0; i < componentNumber; ++i) {
results[i].reset(new DfgGraph{*m_modulep, prefix + cvtToStr(i)});
}
// Move all vertices under the corresponding component graphs
forEachVertex([&](DfgVertex& vtx) {
this->removeVertex(vtx);
results[vertex2component[&vtx]]->addVertex(vtx);
});
UASSERT(size() == 0, "'this' DfgGraph should have been emptied");
return results;
}
class ExtractCyclicComponents final {
static constexpr size_t UNASSIGNED = std::numeric_limits<size_t>::max();
// TYPES
struct VertexState {
size_t index; // Used by Pearce's algorithm for detecting SCCs
size_t component = UNASSIGNED; // Result component number (0 stays in input graph)
VertexState(size_t index)
: index{index} {}
};
// STATE
//==========================================================================
// Shared state
DfgGraph& m_dfg; // The input graph
const std::string m_prefix; // Component name prefix
std::unordered_map<const DfgVertex*, VertexState> m_state; // Vertex state
size_t m_nonTrivialSCCs = 0; // Number of non-trivial SCCs in the graph
const bool m_doExpensiveChecks = v3Global.opt.debugCheck();
//==========================================================================
// State for Pearce's algorithm for detecting SCCs
size_t m_index = 0; // Visitation index counter
std::vector<DfgVertex*> m_stack; // The stack used by the algorithm
//==========================================================================
// State for merging
std::unordered_set<const DfgVertex*> m_merged; // Marks visited vertices
//==========================================================================
// State for extraction
// The extracted cyclic components
std::vector<std::unique_ptr<DfgGraph>> m_components;
// Map from 'variable vertex' -> 'component index' -> 'clone in that component'
std::unordered_map<const DfgVertexVar*, std::unordered_map<size_t, DfgVertexVar*>> m_clones;
// METHODS
//==========================================================================
// Methods for Pearce's algorithm to detect strongly connected components
void visitColorSCCs(DfgVertex& vtx) {
const auto pair = m_state.emplace(std::piecewise_construct, //
std::forward_as_tuple(&vtx), //
std::forward_as_tuple(m_index));
// If already visited, then nothing to do
if (!pair.second) return;
// Visiting node
const size_t rootIndex = m_index++;
vtx.forEachSink([&](DfgVertex& child) {
// Visit child
visitColorSCCs(child);
auto& childSatate = m_state.at(&child);
// If the child is not in an SCC
if (childSatate.component == UNASSIGNED) {
auto& vtxState = m_state.at(&vtx);
if (vtxState.index > childSatate.index) vtxState.index = childSatate.index;
}
});
auto& vtxState = m_state.at(&vtx);
if (vtxState.index == rootIndex) {
// This is the 'root' of an SCC
// A trivial SCC contains only a single vertex
const bool isTrivial = m_stack.empty() || m_state.at(m_stack.back()).index < rootIndex;
// We also need a separate component for vertices that drive themselves (which can
// happen for input like 'assign a = a'), as we want to extract them (they are cyclic).
const bool drivesSelf = vtx.findSink<DfgVertex>([&vtx](const DfgVertex& sink) { //
return &vtx == &sink;
});
if (!isTrivial || drivesSelf) {
// Allocate new component
++m_nonTrivialSCCs;
vtxState.component = m_nonTrivialSCCs;
while (!m_stack.empty()) {
DfgVertex* const topp = m_stack.back();
auto& topState = m_state.at(topp);
// Only higher nodes belong to the same SCC
if (topState.index < rootIndex) break;
m_stack.pop_back();
topState.component = m_nonTrivialSCCs;
}
} else {
// Trivial SCC (and does not drive itself), so acyclic. Keep it in original graph.
vtxState.component = 0;
}
} else {
// Not the root of an SCC
m_stack.push_back(&vtx);
}
}
void colorSCCs() {
// Implements Pearce's algorithm to color the strongly connected components. For reference
// see "An Improved Algorithm for Finding the Strongly Connected Components of a Directed
// Graph", David J.Pearce, 2005
m_state.reserve(m_dfg.size());
m_dfg.forEachVertex([&](DfgVertex& vtx) { visitColorSCCs(vtx); });
}
//==========================================================================
// Methods for merging
void visitMergeSCCs(const DfgVertex& vtx, size_t targetComponent) {
// Mark visited/move on if already visited
if (!m_merged.insert(&vtx).second) return;
// Assign vertex to the target component
m_state.at(&vtx).component = targetComponent;
// Visit all neighbours. We stop at variable boundaries,
// which is where we will split the graphs
vtx.forEachSource([=](const DfgVertex& other) {
if (other.is<DfgVertexVar>()) return;
visitMergeSCCs(other, targetComponent);
});
vtx.forEachSink([=](const DfgVertex& other) {
if (other.is<DfgVertexVar>()) return;
visitMergeSCCs(other, targetComponent);
});
}
void mergeSCCs() {
// Ensure that component boundaries are always at variables, by merging SCCs
m_merged.reserve(m_dfg.size());
// Merging stops at variable boundaries, so we don't need to iterate variables. Constants
// are reachable from their sinks, or ar unused, so we don't need to iterate them either.
for (DfgVertex *vtxp = m_dfg.opVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
// Start DFS from each vertex that is in a non-trivial SCC, and merge everything
// that is reachable from it into this component.
if (const size_t target = m_state.at(vtxp).component) visitMergeSCCs(*vtxp, target);
}
}
//==========================================================================
// Methods for extraction
// Retrieve clone of vertex in the given component
DfgVertexVar& getClone(DfgVertexVar& vtx, size_t component) {
UASSERT_OBJ(m_state.at(&vtx).component != component, &vtx, "Vertex is in that component");
DfgVertexVar*& clonep = m_clones[&vtx][component];
if (!clonep) {
DfgGraph& dfg = component == 0 ? m_dfg : *m_components[component - 1];
if (DfgVarPacked* const pVtxp = vtx.cast<DfgVarPacked>()) {
clonep = new DfgVarPacked{dfg, pVtxp->varp()};
} else if (DfgVarArray* const aVtxp = vtx.cast<DfgVarArray>()) {
clonep = new DfgVarArray{dfg, aVtxp->varp()};
}
UASSERT_OBJ(clonep, &vtx, "Unhandled 'DfgVertexVar' sub-type");
if (VL_UNLIKELY(m_doExpensiveChecks)) {
// Assign component number of clone for later checks
m_state
.emplace(std::piecewise_construct, std::forward_as_tuple(clonep),
std::forward_as_tuple(0))
.first->second.component
= component;
}
// We need to mark both the original and the clone as having additional references
vtx.setHasModRefs();
clonep->setHasModRefs();
}
return *clonep;
}
// Fix up non-variable sources of a DfgVertexVar that are in a different component,
// using the provided 'relink' callback
template <typename T_Vertex>
void fixSources(T_Vertex& vtx, std::function<void(T_Vertex&, DfgVertex&, size_t)> relink) {
static_assert(std::is_base_of<DfgVertexVar, T_Vertex>::value,
"'Vertex' must be a 'DfgVertexVar'");
const size_t component = m_state.at(&vtx).component;
vtx.forEachSourceEdge([&](DfgEdge& edge, size_t idx) {
DfgVertex& source = *edge.sourcep();
// DfgVertexVar sources are fixed up by `fixSinks` on those sources
if (source.is<DfgVertexVar>()) return;
const size_t sourceComponent = m_state.at(&source).component;
// Same component is OK
if (sourceComponent == component) return;
// Unlink the source edge (source is reconnected by 'relink'
edge.unlinkSource();
// Apply the fixup
DfgVertexVar& clone = getClone(vtx, sourceComponent);
relink(*(clone.as<T_Vertex>()), source, idx);
});
}
// Fix up sinks of given variable vertex that are in a different component
void fixSinks(DfgVertexVar& vtx) {
const size_t component = m_state.at(&vtx).component;
vtx.forEachSinkEdge([&](DfgEdge& edge) {
const size_t sinkComponent = m_state.at(edge.sinkp()).component;
// Same component is OK
if (sinkComponent == component) return;
// Relink the sink to read the clone
edge.relinkSource(&getClone(vtx, sinkComponent));
});
}
// Fix edges that cross components
void fixEdges(DfgVertexVar& vtx) {
if (DfgVarPacked* const vvtxp = vtx.cast<DfgVarPacked>()) {
fixSources<DfgVarPacked>(
*vvtxp, [&](DfgVarPacked& clone, DfgVertex& driver, size_t driverIdx) {
clone.addDriver(vvtxp->driverFileLine(driverIdx), //
vvtxp->driverLsb(driverIdx), &driver);
});
fixSinks(*vvtxp);
return;
}
if (DfgVarArray* const vvtxp = vtx.cast<DfgVarArray>()) {
fixSources<DfgVarArray>( //
*vvtxp, [&](DfgVarArray& clone, DfgVertex& driver, size_t driverIdx) {
clone.addDriver(vvtxp->driverFileLine(driverIdx), //
vvtxp->driverIndex(driverIdx), &driver);
});
fixSinks(*vvtxp);
return;
}
}
static void packSources(DfgGraph& dfg) {
// Remove undriven variable sources
for (DfgVertexVar *vtxp = dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
if (DfgVarPacked* const varp = vtxp->cast<DfgVarPacked>()) {
varp->packSources();
if (!varp->hasSinks() && varp->arity() == 0) {
VL_DO_DANGLING(varp->unlinkDelete(dfg), varp);
}
return;
}
if (DfgVarArray* const varp = vtxp->cast<DfgVarArray>()) {
varp->packSources();
if (!varp->hasSinks() && varp->arity() == 0) {
VL_DO_DANGLING(varp->unlinkDelete(dfg), varp);
}
return;
}
}
}
void checkGraph(DfgGraph& dfg) const {
// Build set of vertices
std::unordered_set<const DfgVertex*> vertices{dfg.size()};
dfg.forEachVertex([&](const DfgVertex& vtx) { vertices.insert(&vtx); });
// Check that:
// - Edges only cross components at variable boundaries
// - Each edge connects to a vertex that is within the same graph
// - Variable vertex sources are all connected.
dfg.forEachVertex([&](const DfgVertex& vtx) {
const size_t component = m_state.at(&vtx).component;
vtx.forEachSource([&](const DfgVertex& src) {
if (!src.is<DfgVertexVar>()) { // OK to cross at variables
UASSERT_OBJ(component == m_state.at(&src).component, &vtx,
"Edge crossing components without variable involvement");
}
UASSERT_OBJ(vertices.count(&src), &vtx, "Source vertex not in graph");
});
vtx.forEachSink([&](const DfgVertex& snk) {
if (!snk.is<DfgVertexVar>()) { // OK to cross at variables
UASSERT_OBJ(component == m_state.at(&snk).component, &vtx,
"Edge crossing components without variable involvement");
}
UASSERT_OBJ(vertices.count(&snk), &snk, "Sink vertex not in graph");
});
if (const DfgVertexVar* const vtxp = vtx.cast<DfgVertexVar>()) {
vtxp->forEachSourceEdge([](const DfgEdge& edge, size_t) {
UASSERT_OBJ(edge.sourcep(), edge.sinkp(), "Missing source on variable vertex");
});
return;
}
});
}
void extractComponents() {
// If the graph was acyclic (which should be the common case), there will be no non-trivial
// SCCs, so we are done.
if (!m_nonTrivialSCCs) return;
// Allocate result graphs
m_components.resize(m_nonTrivialSCCs);
for (size_t i = 0; i < m_nonTrivialSCCs; ++i) {
m_components[i].reset(new DfgGraph{*m_dfg.modulep(), m_prefix + cvtToStr(i)});
}
// Fix up edges crossing components (we can only do this at variable boundaries, and the
// earlier merging of components ensured crossing in fact only happen at variable
// boundaries). Note that fixing up the edges can create clones of variables. Clones are
// added to the correct component, which also means that they might be added to the
// original DFG. Clones do not need fixing up, but also are not necessarily in the m_state
// map (in fact they are only there in debug mode), so we need to check this.
// Also move vertices into their correct component while we are at it.
for (DfgVertexVar *vtxp = m_dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
// It is possible the last vertex (with a nullptr for 'nextp') gets cloned, and hence
// it's 'nextp' would become none nullptr as the clone is added. However, we don't need
// to iterate clones anyway, so it's ok to get the 'nextp' early in the loop.
nextp = vtxp->verticesNext();
// Clones need not be fixed up
if (!m_state.count(vtxp)) return;
// Fix up the edges crossing components
fixEdges(*vtxp);
// Move the vertex to the component graph (leave component 0, which is the
// originally acyclic sub-graph, in the original graph)
if (const size_t component = m_state.at(vtxp).component) {
m_dfg.removeVertex(*vtxp);
m_components[component - 1]->addVertex(*vtxp);
}
}
// Move other vertices to their component graphs
for (DfgConst *vtxp = m_dfg.constVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
if (const size_t component = m_state.at(vtxp).component) {
m_dfg.removeVertex(*vtxp);
m_components[component - 1]->addVertex(*vtxp);
}
}
for (DfgVertex *vtxp = m_dfg.opVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
if (const size_t component = m_state.at(vtxp).component) {
m_dfg.removeVertex(*vtxp);
m_components[component - 1]->addVertex(*vtxp);
}
}
// Pack sources of variables to remove the now undriven inputs
// (cloning might have unlinked some of the inputs),
packSources(m_dfg);
for (const auto& dfgp : m_components) packSources(*dfgp);
if (VL_UNLIKELY(m_doExpensiveChecks)) {
// Check results for consistency
checkGraph(m_dfg);
for (const auto& dfgp : m_components) checkGraph(*dfgp);
}
}
// CONSTRUCTOR - entry point
explicit ExtractCyclicComponents(DfgGraph& dfg, std::string label)
: m_dfg{dfg}
, m_prefix{dfg.name() + (label.empty() ? "" : "-") + label + "-component-"} {
// Find all the non-trivial SCCs (and trivial cycles) in the graph
colorSCCs();
// Ensure that component boundaries are always at variables, by merging SCCs
mergeSCCs();
// Extract the components
extractComponents();
}
public:
static std::vector<std::unique_ptr<DfgGraph>> apply(DfgGraph& dfg, const std::string& label) {
return std::move(ExtractCyclicComponents{dfg, label}.m_components);
}
};
std::vector<std::unique_ptr<DfgGraph>> DfgGraph::extractCyclicComponents(std::string label) {
return ExtractCyclicComponents::apply(*this, label);
}
static const string toDotId(const DfgVertex& vtx) { return '"' + cvtToHex(&vtx) + '"'; } static const string toDotId(const DfgVertex& vtx) { return '"' + cvtToHex(&vtx) + '"'; }
// Dump one DfgVertex in Graphviz format // Dump one DfgVertex in Graphviz format

33
src/V3Dfg.h
View File

@ -184,6 +184,7 @@ public:
void addGraph(DfgGraph& other); void addGraph(DfgGraph& other);
// Split this graph into individual components (unique sub-graphs with no edges between them). // Split this graph into individual components (unique sub-graphs with no edges between them).
// Also removes any vertices that are not weakly connected to any variable.
// Leaves 'this' graph empty. // Leaves 'this' graph empty.
std::vector<std::unique_ptr<DfgGraph>> splitIntoComponents(std::string label); std::vector<std::unique_ptr<DfgGraph>> splitIntoComponents(std::string label);
@ -340,6 +341,7 @@ public:
// The type of this vertex // The type of this vertex
VDfgType type() const { return m_type; } VDfgType type() const { return m_type; }
// Retrieve user data, constructing it fresh on first try.
template <typename T> template <typename T>
T& user() { T& user() {
static_assert(sizeof(T) <= sizeof(UserDataStorage), static_assert(sizeof(T) <= sizeof(UserDataStorage),
@ -357,6 +359,22 @@ public:
return *storagep; return *storagep;
} }
// Retrieve user data, must be current.
template <typename T>
T& getUser() {
static_assert(sizeof(T) <= sizeof(UserDataStorage),
"Size of user data type 'T' is too large for allocated storage");
static_assert(alignof(T) <= alignof(UserDataStorage),
"Alignment of user data type 'T' is larger than allocated storage");
T* const storagep = reinterpret_cast<T*>(&m_userDataStorage);
#if VL_DEBUG
const uint32_t userCurrent = m_graphp->m_userCurrent;
UASSERT_OBJ(userCurrent, this, "DfgVertex user data used without reserving");
UASSERT_OBJ(m_userCnt == userCurrent, this, "DfgVertex user data is stale");
#endif
return *storagep;
}
// Width of result // Width of result
uint32_t width() const { uint32_t width() const {
// This is a hot enough function that this is an expensive check, so in debug build only. // This is a hot enough function that this is an expensive check, so in debug build only.
@ -412,6 +430,10 @@ public:
DfgVertex* verticesNext() const { return m_verticesEnt.nextp(); } DfgVertex* verticesNext() const { return m_verticesEnt.nextp(); }
DfgVertex* verticesPrev() const { return m_verticesEnt.prevp(); } DfgVertex* verticesPrev() const { return m_verticesEnt.prevp(); }
// Calls given function 'f' for each source vertex of this vertex
// Unconnected source edges are not iterated.
inline void forEachSource(std::function<void(DfgVertex&)> f);
// Calls given function 'f' for each source vertex of this vertex // Calls given function 'f' for each source vertex of this vertex
// Unconnected source edges are not iterated. // Unconnected source edges are not iterated.
inline void forEachSource(std::function<void(const DfgVertex&)> f) const; inline void forEachSource(std::function<void(const DfgVertex&)> f) const;
@ -545,6 +567,7 @@ void DfgGraph::addVertex(DfgVertex& vtx) {
} else { } else {
vtx.m_verticesEnt.pushBack(m_opVertices, &vtx); vtx.m_verticesEnt.pushBack(m_opVertices, &vtx);
} }
vtx.m_userCnt = 0;
vtx.m_graphp = this; vtx.m_graphp = this;
} }
@ -558,6 +581,7 @@ void DfgGraph::removeVertex(DfgVertex& vtx) {
} else { } else {
vtx.m_verticesEnt.unlink(m_opVertices, &vtx); vtx.m_verticesEnt.unlink(m_opVertices, &vtx);
} }
vtx.m_userCnt = 0;
vtx.m_graphp = nullptr; vtx.m_graphp = nullptr;
} }
@ -588,6 +612,15 @@ void DfgGraph::forEachVertex(std::function<void(const DfgVertex&)> f) const {
} }
} }
void DfgVertex::forEachSource(std::function<void(DfgVertex&)> f) {
const auto pair = sourceEdges();
const DfgEdge* const edgesp = pair.first;
const size_t arity = pair.second;
for (size_t i = 0; i < arity; ++i) {
if (DfgVertex* const sourcep = edgesp[i].m_sourcep) f(*sourcep);
}
}
void DfgVertex::forEachSource(std::function<void(const DfgVertex&)> f) const { void DfgVertex::forEachSource(std::function<void(const DfgVertex&)> f) const {
const auto pair = sourceEdges(); const auto pair = sourceEdges();
const DfgEdge* const edgesp = pair.first; const DfgEdge* const edgesp = pair.first;

544
src/V3DfgDecomposition.cpp Normal file
View File

@ -0,0 +1,544 @@
// -*- mode: C++; c-file-style: "cc-mode" -*-
//*************************************************************************
// DESCRIPTION: Verilator: DfgGraph decomposition algorithms
//
// Code available from: https://verilator.org
//
//*************************************************************************
//
// Copyright 2003-2022 by Wilson Snyder. This program is free software; you
// can redistribute it and/or modify it under the terms of either the GNU
// Lesser General Public License Version 3 or the Perl Artistic License
// Version 2.0.
// SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
//
//*************************************************************************
//
// Algorithms that take a DfgGraph and decompose it into multiple DfgGraphs.
//
//*************************************************************************
#include "config_build.h"
#include "verilatedos.h"
#include "V3Dfg.h"
#include "V3File.h"
#include <deque>
#include <unordered_map>
#include <vector>
VL_DEFINE_DEBUG_FUNCTIONS;
class SplitIntoComponents final {
// STATE
DfgGraph& m_dfg; // The input graph
const std::string m_prefix; // Component name prefix
std::vector<std::unique_ptr<DfgGraph>> m_components; // The extracted components
// Component counter - starting from 1 as 0 is the default value used as a marker
size_t m_componentCounter = 1;
void colorComponents() {
// Work queue for depth first traversal starting from this vertex
std::vector<DfgVertex*> queue;
queue.reserve(m_dfg.size());
// any sort of interesting logic must involve a variable, so we only need to iterate them
for (DfgVertexVar *vtxp = m_dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
// If already assigned this vertex to a component, then continue
if (vtxp->user<size_t>()) continue;
// Start depth first traversal at this vertex
queue.push_back(vtxp);
// Depth first traversal
do {
// Pop next work item
DfgVertex& item = *queue.back();
queue.pop_back();
// Move on if already visited
if (item.user<size_t>()) continue;
// Assign to current component
item.user<size_t>() = m_componentCounter;
// Enqueue all sources and sinks of this vertex.
item.forEachSource([&](DfgVertex& src) { queue.push_back(&src); });
item.forEachSink([&](DfgVertex& dst) { queue.push_back(&dst); });
} while (!queue.empty());
// Done with this component
++m_componentCounter;
}
}
void moveVertices(DfgVertex* headp) {
for (DfgVertex *vtxp = headp, *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
DfgVertex& vtx = *vtxp;
if (const size_t component = vtx.user<size_t>()) {
m_dfg.removeVertex(vtx);
m_components[component - 1]->addVertex(vtx);
} else {
// This vertex is not connected to a variable and is hence unused, remove here
vtx.unlinkDelete(m_dfg);
}
}
}
SplitIntoComponents(DfgGraph& dfg, std::string label)
: m_dfg{dfg}
, m_prefix{dfg.name() + (label.empty() ? "" : "-") + label + "-component-"} {
// Component number is stored as DfgVertex::user<size_t>()
const auto userDataInUse = m_dfg.userDataInUse();
// Color each component of the graph
colorComponents();
// Allocate the component graphs
m_components.resize(m_componentCounter - 1);
for (size_t i = 1; i < m_componentCounter; ++i) {
m_components[i - 1].reset(new DfgGraph{*m_dfg.modulep(), m_prefix + cvtToStr(i - 1)});
}
// Move the vertices to the component graphs
moveVertices(m_dfg.varVerticesBeginp());
moveVertices(m_dfg.constVerticesBeginp());
moveVertices(m_dfg.opVerticesBeginp());
//
UASSERT(m_dfg.size() == 0, "'this' DfgGraph should have been emptied");
}
public:
static std::vector<std::unique_ptr<DfgGraph>> apply(DfgGraph& dfg, const std::string& label) {
return std::move(SplitIntoComponents{dfg, label}.m_components);
}
};
std::vector<std::unique_ptr<DfgGraph>> DfgGraph::splitIntoComponents(std::string label) {
return SplitIntoComponents::apply(*this, label);
}
class ExtractCyclicComponents final {
static constexpr size_t UNASSIGNED = std::numeric_limits<size_t>::max();
// TYPES
struct VertexState {
size_t index = UNASSIGNED; // Used by Pearce's algorithm for detecting SCCs
size_t component = UNASSIGNED; // Result component number (0 stays in input graph)
bool merged = false; // Visited in the merging pass
VertexState(){};
};
// STATE
//==========================================================================
// Shared state
DfgGraph& m_dfg; // The input graph
std::deque<VertexState> m_stateStorage; // Container for VertexState instances
const std::string m_prefix; // Component name prefix
size_t m_nonTrivialSCCs = 0; // Number of non-trivial SCCs in the graph
const bool m_doExpensiveChecks = v3Global.opt.debugCheck();
//==========================================================================
// State for Pearce's algorithm for detecting SCCs
size_t m_index = 0; // Visitation index counter
std::vector<DfgVertex*> m_stack; // The stack used by the algorithm
//==========================================================================
// State for extraction
// The extracted cyclic components
std::vector<std::unique_ptr<DfgGraph>> m_components;
// Map from 'variable vertex' -> 'component index' -> 'clone in that component'
std::unordered_map<const DfgVertexVar*, std::unordered_map<size_t, DfgVertexVar*>> m_clones;
// METHODS
//==========================================================================
// Shared methods
VertexState& state(DfgVertex& vtx) const { return *vtx.getUser<VertexState*>(); }
VertexState& allocState(DfgVertex& vtx) {
VertexState*& statep = vtx.user<VertexState*>();
UASSERT_OBJ(!statep, &vtx, "Vertex state already allocated " << cvtToHex(statep));
m_stateStorage.emplace_back();
statep = &m_stateStorage.back();
return *statep;
}
VertexState& getOrAllocState(DfgVertex& vtx) {
VertexState*& statep = vtx.user<VertexState*>();
if (!statep) {
m_stateStorage.emplace_back();
statep = &m_stateStorage.back();
}
return *statep;
}
//==========================================================================
// Methods for Pearce's algorithm to detect strongly connected components
void visitColorSCCs(DfgVertex& vtx, VertexState& vtxState) {
UDEBUGONLY(UASSERT_OBJ(vtxState.index == UNASSIGNED, &vtx, "Already visited vertex"););
// Visiting vertex
const size_t rootIndex = vtxState.index = ++m_index;
// Visit children
vtx.forEachSink([&](DfgVertex& child) {
VertexState& childSatate = getOrAllocState(child);
// If the child has not yet been visited, then continue traversal
if (childSatate.index == UNASSIGNED) visitColorSCCs(child, childSatate);
// If the child is not in an SCC
if (childSatate.component == UNASSIGNED) {
if (vtxState.index > childSatate.index) vtxState.index = childSatate.index;
}
});
if (vtxState.index == rootIndex) {
// This is the 'root' of an SCC
// A trivial SCC contains only a single vertex
const bool isTrivial = m_stack.empty() || state(*m_stack.back()).index < rootIndex;
// We also need a separate component for vertices that drive themselves (which can
// happen for input like 'assign a = a'), as we want to extract them (they are cyclic).
const bool drivesSelf = vtx.findSink<DfgVertex>([&vtx](const DfgVertex& sink) { //
return &vtx == &sink;
});
if (!isTrivial || drivesSelf) {
// Allocate new component
++m_nonTrivialSCCs;
vtxState.component = m_nonTrivialSCCs;
while (!m_stack.empty()) {
VertexState& topState = state(*m_stack.back());
// Only higher nodes belong to the same SCC
if (topState.index < rootIndex) break;
m_stack.pop_back();
topState.component = m_nonTrivialSCCs;
}
} else {
// Trivial SCC (and does not drive itself), so acyclic. Keep it in original graph.
vtxState.component = 0;
}
} else {
// Not the root of an SCC
m_stack.push_back(&vtx);
}
}
void colorSCCs() {
// Implements Pearce's algorithm to color the strongly connected components. For reference
// see "An Improved Algorithm for Finding the Strongly Connected Components of a Directed
// Graph", David J.Pearce, 2005.
// We can leverage some properties of the input graph to gain a bit of speed. Firstly, we
// know constant nodes have no in edges, so they cannot be part of a non-trivial SCC. Mark
// them as such without starting a whole traversal.
for (DfgConst *vtxp = m_dfg.constVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
VertexState& vtxState = allocState(*vtxp);
vtxState.index = 0;
vtxState.component = 0;
}
// Next, we know that all SCCs must include a variable (as the input graph was converted
// from an AST, we can only have a cycle by going through a variable), so we only start
// traversals through them, and only if we know they have both in and out edges.
for (DfgVertexVar *vtxp = m_dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
if (vtxp->arity() > 0 && vtxp->hasSinks()) {
VertexState& vtxState = getOrAllocState(*vtxp);
// If not yet visited, start a traversal
if (vtxState.index == UNASSIGNED) visitColorSCCs(*vtxp, vtxState);
} else {
VertexState& vtxState = getOrAllocState(*vtxp);
UDEBUGONLY(UASSERT_OBJ(vtxState.index == UNASSIGNED || vtxState.component == 0,
vtxp, "Non circular variable must be in a trivial SCC"););
vtxState.index = 0;
vtxState.component = 0;
}
}
// Finally, everything we did not visit through the traversal of a variable cannot be in an
// SCC, (otherwise we would have found it from a variable).
for (DfgVertex *vtxp = m_dfg.opVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
VertexState& vtxState = getOrAllocState(*vtxp);
if (vtxState.index == UNASSIGNED) {
vtxState.index = 0;
vtxState.component = 0;
}
}
}
//==========================================================================
// Methods for merging
void visitMergeSCCs(DfgVertex& vtx, size_t targetComponent) {
VertexState& vtxState = state(vtx);
// Move on if already visited
if (vtxState.merged) return;
// Visiting vertex
vtxState.merged = true;
// Assign vertex to the target component
vtxState.component = targetComponent;
// Visit all neighbours. We stop at variable boundaries,
// which is where we will split the graphs
vtx.forEachSource([=](DfgVertex& other) {
if (other.is<DfgVertexVar>()) return;
visitMergeSCCs(other, targetComponent);
});
vtx.forEachSink([=](DfgVertex& other) {
if (other.is<DfgVertexVar>()) return;
visitMergeSCCs(other, targetComponent);
});
}
void mergeSCCs() {
// Ensure that component boundaries are always at variables, by merging SCCs. Merging stops
// at variable boundaries, so we don't need to iterate variables. Constants are reachable
// from their sinks, or ar unused, so we don't need to iterate them either.
for (DfgVertex *vtxp = m_dfg.opVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
DfgVertex& vtx = *vtxp;
// Start DFS from each vertex that is in a non-trivial SCC, and merge everything
// that is reachable from it into this component.
if (const size_t target = state(vtx).component) visitMergeSCCs(vtx, target);
}
}
//==========================================================================
// Methods for extraction
// Retrieve clone of vertex in the given component
DfgVertexVar& getClone(DfgVertexVar& vtx, size_t component) {
UASSERT_OBJ(state(vtx).component != component, &vtx, "Vertex is in that component");
DfgVertexVar*& clonep = m_clones[&vtx][component];
if (!clonep) {
if (DfgVarPacked* const pVtxp = vtx.cast<DfgVarPacked>()) {
clonep = new DfgVarPacked{m_dfg, pVtxp->varp()};
} else if (DfgVarArray* const aVtxp = vtx.cast<DfgVarArray>()) {
clonep = new DfgVarArray{m_dfg, aVtxp->varp()};
}
UASSERT_OBJ(clonep, &vtx, "Unhandled 'DfgVertexVar' sub-type");
VertexState& cloneStatep = allocState(*clonep);
cloneStatep.component = component;
// We need to mark both the original and the clone as having additional references
vtx.setHasModRefs();
clonep->setHasModRefs();
}
return *clonep;
}
// Fix up non-variable sources of a DfgVertexVar that are in a different component,
// using the provided 'relink' callback
template <typename T_Vertex>
void fixSources(T_Vertex& vtx, std::function<void(T_Vertex&, DfgVertex&, size_t)> relink) {
static_assert(std::is_base_of<DfgVertexVar, T_Vertex>::value,
"'Vertex' must be a 'DfgVertexVar'");
const size_t component = state(vtx).component;
vtx.forEachSourceEdge([&](DfgEdge& edge, size_t idx) {
DfgVertex& source = *edge.sourcep();
// DfgVertexVar sources are fixed up by `fixSinks` on those sources
if (source.is<DfgVertexVar>()) return;
const size_t sourceComponent = state(source).component;
// Same component is OK
if (sourceComponent == component) return;
// Unlink the source edge (source is reconnected by 'relink'
edge.unlinkSource();
// Apply the fixup
DfgVertexVar& clone = getClone(vtx, sourceComponent);
relink(*(clone.as<T_Vertex>()), source, idx);
});
}
// Fix up sinks of given variable vertex that are in a different component
void fixSinks(DfgVertexVar& vtx) {
const size_t component = state(vtx).component;
vtx.forEachSinkEdge([&](DfgEdge& edge) {
const size_t sinkComponent = state(*edge.sinkp()).component;
// Same component is OK
if (sinkComponent == component) return;
// Relink the sink to read the clone
edge.relinkSource(&getClone(vtx, sinkComponent));
});
}
// Fix edges that cross components
void fixEdges(DfgVertexVar& vtx) {
if (DfgVarPacked* const vvtxp = vtx.cast<DfgVarPacked>()) {
fixSources<DfgVarPacked>(
*vvtxp, [&](DfgVarPacked& clone, DfgVertex& driver, size_t driverIdx) {
clone.addDriver(vvtxp->driverFileLine(driverIdx), //
vvtxp->driverLsb(driverIdx), &driver);
});
fixSinks(*vvtxp);
return;
}
if (DfgVarArray* const vvtxp = vtx.cast<DfgVarArray>()) {
fixSources<DfgVarArray>( //
*vvtxp, [&](DfgVarArray& clone, DfgVertex& driver, size_t driverIdx) {
clone.addDriver(vvtxp->driverFileLine(driverIdx), //
vvtxp->driverIndex(driverIdx), &driver);
});
fixSinks(*vvtxp);
return;
}
}
static void packSources(DfgGraph& dfg) {
// Remove undriven variable sources
for (DfgVertexVar *vtxp = dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
if (DfgVarPacked* const varp = vtxp->cast<DfgVarPacked>()) {
varp->packSources();
if (!varp->hasSinks() && varp->arity() == 0) {
VL_DO_DANGLING(varp->unlinkDelete(dfg), varp);
}
return;
}
if (DfgVarArray* const varp = vtxp->cast<DfgVarArray>()) {
varp->packSources();
if (!varp->hasSinks() && varp->arity() == 0) {
VL_DO_DANGLING(varp->unlinkDelete(dfg), varp);
}
return;
}
}
}
void moveVertices(DfgVertex* headp) {
for (DfgVertex *vtxp = headp, *nextp; vtxp; vtxp = nextp) {
nextp = vtxp->verticesNext();
DfgVertex& vtx = *vtxp;
if (const size_t component = state(vtx).component) {
m_dfg.removeVertex(vtx);
m_components[component - 1]->addVertex(vtx);
}
}
}
void checkEdges(DfgGraph& dfg) const {
// Check that:
// - Edges only cross components at variable boundaries
// - Variable vertex sources are all connected.
dfg.forEachVertex([&](DfgVertex& vtx) {
const size_t component = state(vtx).component;
vtx.forEachSource([&](DfgVertex& src) {
if (src.is<DfgVertexVar>()) return; // OK to cross at variables
UASSERT_OBJ(component == state(src).component, &vtx,
"Edge crossing components without variable involvement");
});
vtx.forEachSink([&](DfgVertex& snk) {
if (snk.is<DfgVertexVar>()) return; // OK to cross at variables
UASSERT_OBJ(component == state(snk).component, &vtx,
"Edge crossing components without variable involvement");
});
if (const DfgVertexVar* const vtxp = vtx.cast<DfgVertexVar>()) {
vtxp->forEachSourceEdge([](const DfgEdge& edge, size_t) {
UASSERT_OBJ(edge.sourcep(), edge.sinkp(), "Missing source on variable vertex");
});
}
});
}
void checkGraph(DfgGraph& dfg) const {
// Build set of vertices
std::unordered_set<const DfgVertex*> vertices{dfg.size()};
dfg.forEachVertex([&](const DfgVertex& vtx) { vertices.insert(&vtx); });
// Check that each edge connects to a vertex that is within the same graph
dfg.forEachVertex([&](DfgVertex& vtx) {
vtx.forEachSource([&](DfgVertex& src) {
UASSERT_OBJ(vertices.count(&src), &vtx, "Source vertex not in graph");
});
vtx.forEachSink([&](DfgVertex& snk) {
UASSERT_OBJ(vertices.count(&snk), &snk, "Sink vertex not in graph");
});
});
}
void extractComponents() {
// Allocate result graphs
m_components.resize(m_nonTrivialSCCs);
for (size_t i = 0; i < m_nonTrivialSCCs; ++i) {
m_components[i].reset(new DfgGraph{*m_dfg.modulep(), m_prefix + cvtToStr(i)});
}
// Fix up edges crossing components (we can only do this at variable boundaries, and the
// earlier merging of components ensured crossing in fact only happen at variable
// boundaries). Note that fixing up the edges can create clones of variables. Clones do
// not need fixing up, so we do not need to iterate them.
DfgVertex* const lastp = m_dfg.varVerticesRbeginp();
for (DfgVertexVar *vtxp = m_dfg.varVerticesBeginp(), *nextp; vtxp; vtxp = nextp) {
// It is possible the last vertex (with a nullptr for 'nextp') gets cloned, and hence
// it's 'nextp' would become none nullptr as the clone is added. However, we don't need
// to iterate clones anyway, so it's ok to get the 'nextp' early in the loop.
nextp = vtxp->verticesNext();
DfgVertexVar& vtx = *vtxp;
// Fix up the edges crossing components
fixEdges(vtx);
// Don't iterate clones added during this loop
if (vtxp == lastp) break;
}
// Pack sources of variables to remove the now undriven inputs
// (cloning might have unlinked some of the inputs),
packSources(m_dfg);
for (const auto& dfgp : m_components) packSources(*dfgp);
// Check results for consistency
if (VL_UNLIKELY(m_doExpensiveChecks)) {
checkEdges(m_dfg);
for (const auto& dfgp : m_components) checkEdges(*dfgp);
}
// Move other vertices to their component graphs
// After this, vertex states are invalid as we moved the vertices
moveVertices(m_dfg.varVerticesBeginp());
moveVertices(m_dfg.constVerticesBeginp());
moveVertices(m_dfg.opVerticesBeginp());
// Check results for consistency
if (VL_UNLIKELY(m_doExpensiveChecks)) {
checkGraph(m_dfg);
for (const auto& dfgp : m_components) checkGraph(*dfgp);
}
}
// CONSTRUCTOR - entry point
explicit ExtractCyclicComponents(DfgGraph& dfg, std::string label)
: m_dfg{dfg}
, m_prefix{dfg.name() + (label.empty() ? "" : "-") + label + "-component-"} {
// VertexState is stored as user data
const auto userDataInUse = dfg.userDataInUse();
// Find all the non-trivial SCCs (and trivial cycles) in the graph
colorSCCs();
// If the graph was acyclic (which should be the common case),
// there will be no non-trivial SCCs, so we are done.
if (!m_nonTrivialSCCs) return;
// Ensure that component boundaries are always at variables, by merging SCCs
mergeSCCs();
// Extract the components
extractComponents();
}
public:
static std::vector<std::unique_ptr<DfgGraph>> apply(DfgGraph& dfg, const std::string& label) {
return std::move(ExtractCyclicComponents{dfg, label}.m_components);
}
};
std::vector<std::unique_ptr<DfgGraph>> DfgGraph::extractCyclicComponents(std::string label) {
return ExtractCyclicComponents::apply(*this, label);
}