mirror of
https://github.com/KhronosGroup/SPIRV-Tools
synced 2024-11-26 21:30:07 +00:00
538512e8e8
The management of equation facts suffered from two problems: (1) The processing of an equation fact required the data descriptors used in the equation to be in canonical form. However, during fact processing it can be deduced that certain data descriptors are equivalent, causing their equivalence classes to be merged, and that could cause previously canonical data descriptors to no longer be canonical. (2) Related to this, if id equations were known about a canonical data descriptor dd1, and other id equations known about a different canonical data descriptor dd2, the equation facts about these data descriptors were not being merged in the event that dd1 and dd2 were deduced to be equivalent. This changes solves (1) by not requiring equation facts to be in canonical form while processing them, but instead always checking whether (not necessary canonical) data descriptors are equivalent when looking for corollaries of equation facts, rather than comparing them using ==. Problem (2) is solved by adding logic to merge sets of equations when data descriptors are made equivalent. In addition, the change also requires elements to be registered in an equivalence relation before they can be made equivalent, rather than being added (if not already present) at the point of being made equivalent.
246 lines
9.5 KiB
C++
246 lines
9.5 KiB
C++
// Copyright (c) 2019 Google LLC
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#ifndef SOURCE_FUZZ_EQUIVALENCE_RELATION_H_
|
|
#define SOURCE_FUZZ_EQUIVALENCE_RELATION_H_
|
|
|
|
#include <memory>
|
|
#include <unordered_map>
|
|
#include <unordered_set>
|
|
#include <vector>
|
|
|
|
#include "source/util/make_unique.h"
|
|
|
|
namespace spvtools {
|
|
namespace fuzz {
|
|
|
|
// A class for representing an equivalence relation on objects of type |T|,
|
|
// which should be a value type. The type |T| is required to have a copy
|
|
// constructor, and |PointerHashT| and |PointerEqualsT| must be functors
|
|
// providing hashing and equality testing functionality for pointers to objects
|
|
// of type |T|.
|
|
//
|
|
// A disjoint-set (a.k.a. union-find or merge-find) data structure is used to
|
|
// represent the equivalence relation. Path compression is used. Union by
|
|
// rank/size is not used.
|
|
//
|
|
// Each disjoint set is represented as a tree, rooted at the representative
|
|
// of the set.
|
|
//
|
|
// Getting the representative of a value simply requires chasing parent pointers
|
|
// from the value until you reach the root.
|
|
//
|
|
// Checking equivalence of two elements requires checking that the
|
|
// representatives are equal.
|
|
//
|
|
// Traversing the tree rooted at a value's representative visits the value's
|
|
// equivalence class.
|
|
//
|
|
// |PointerHashT| and |PointerEqualsT| are used to define *equality* between
|
|
// values, and otherwise are *not* used to define the equivalence relation
|
|
// (except that equal values are equivalent). The equivalence relation is
|
|
// constructed by repeatedly adding pairs of (typically non-equal) values that
|
|
// are deemed to be equivalent.
|
|
//
|
|
// For example in an equivalence relation on integers, 1 and 5 might be added
|
|
// as equivalent, so that IsEquivalent(1, 5) holds, because they represent
|
|
// IDs in a SPIR-V binary that are known to contain the same value at run time,
|
|
// but clearly 1 != 5. Since 1 and 1 are equal, IsEquivalent(1, 1) will also
|
|
// hold.
|
|
//
|
|
// Each unique (up to equality) value added to the relation is copied into
|
|
// |owned_values_|, so there is one canonical memory address per unique value.
|
|
// Uniqueness is ensured by storing (and checking) a set of pointers to these
|
|
// values in |value_set_|, which uses |PointerHashT| and |PointerEqualsT|.
|
|
//
|
|
// |parent_| and |children_| encode the equivalence relation, i.e., the trees.
|
|
template <typename T, typename PointerHashT, typename PointerEqualsT>
|
|
class EquivalenceRelation {
|
|
public:
|
|
// Requires that |value1| and |value2| are already registered in the
|
|
// equivalence relation. Merges the equivalence classes associated with
|
|
// |value1| and |value2|.
|
|
void MakeEquivalent(const T& value1, const T& value2) {
|
|
assert(Exists(value1) &&
|
|
"Precondition: value1 must already be registered.");
|
|
assert(Exists(value2) &&
|
|
"Precondition: value2 must already be registered.");
|
|
|
|
// Look up canonical pointers to each of the values in the value pool.
|
|
const T* value1_ptr = *value_set_.find(&value1);
|
|
const T* value2_ptr = *value_set_.find(&value2);
|
|
|
|
// If the values turn out to be identical, they are already in the same
|
|
// equivalence class so there is nothing to do.
|
|
if (value1_ptr == value2_ptr) {
|
|
return;
|
|
}
|
|
|
|
// Find the representative for each value's equivalence class, and if they
|
|
// are not already in the same class, make one the parent of the other.
|
|
const T* representative1 = Find(value1_ptr);
|
|
const T* representative2 = Find(value2_ptr);
|
|
assert(representative1 && "Representatives should never be null.");
|
|
assert(representative2 && "Representatives should never be null.");
|
|
if (representative1 != representative2) {
|
|
parent_[representative1] = representative2;
|
|
children_[representative2].push_back(representative1);
|
|
}
|
|
}
|
|
|
|
// Requires that |value| is not known to the equivalence relation. Registers
|
|
// it in its own equivalence class and returns a pointer to the equivalence
|
|
// class representative.
|
|
const T* Register(const T& value) {
|
|
assert(!Exists(value));
|
|
|
|
// This relies on T having a copy constructor.
|
|
auto unique_pointer_to_value = MakeUnique<T>(value);
|
|
auto pointer_to_value = unique_pointer_to_value.get();
|
|
owned_values_.push_back(std::move(unique_pointer_to_value));
|
|
value_set_.insert(pointer_to_value);
|
|
|
|
// Initially say that the value is its own parent and that it has no
|
|
// children.
|
|
assert(pointer_to_value && "Representatives should never be null.");
|
|
parent_[pointer_to_value] = pointer_to_value;
|
|
children_[pointer_to_value] = std::vector<const T*>();
|
|
|
|
return pointer_to_value;
|
|
}
|
|
|
|
// Returns exactly one representative per equivalence class.
|
|
std::vector<const T*> GetEquivalenceClassRepresentatives() const {
|
|
std::vector<const T*> result;
|
|
for (auto& value : owned_values_) {
|
|
if (parent_[value.get()] == value.get()) {
|
|
result.push_back(value.get());
|
|
}
|
|
}
|
|
return result;
|
|
}
|
|
|
|
// Returns pointers to all values in the equivalence class of |value|, which
|
|
// must already be part of the equivalence relation.
|
|
std::vector<const T*> GetEquivalenceClass(const T& value) const {
|
|
assert(Exists(value));
|
|
|
|
std::vector<const T*> result;
|
|
|
|
// Traverse the tree of values rooted at the representative of the
|
|
// equivalence class to which |value| belongs, and collect up all the values
|
|
// that are encountered. This constitutes the whole equivalence class.
|
|
std::vector<const T*> stack;
|
|
stack.push_back(Find(*value_set_.find(&value)));
|
|
while (!stack.empty()) {
|
|
const T* item = stack.back();
|
|
result.push_back(item);
|
|
stack.pop_back();
|
|
for (auto child : children_[item]) {
|
|
stack.push_back(child);
|
|
}
|
|
}
|
|
return result;
|
|
}
|
|
|
|
// Returns true if and only if |value1| and |value2| are in the same
|
|
// equivalence class. Both values must already be known to the equivalence
|
|
// relation.
|
|
bool IsEquivalent(const T& value1, const T& value2) const {
|
|
return Find(&value1) == Find(&value2);
|
|
}
|
|
|
|
// Returns all values known to be part of the equivalence relation.
|
|
std::vector<const T*> GetAllKnownValues() const {
|
|
std::vector<const T*> result;
|
|
for (auto& value : owned_values_) {
|
|
result.push_back(value.get());
|
|
}
|
|
return result;
|
|
}
|
|
|
|
// Returns true if and only if |value| is known to be part of the equivalence
|
|
// relation.
|
|
bool Exists(const T& value) const {
|
|
return value_set_.find(&value) != value_set_.end();
|
|
}
|
|
|
|
// Returns the representative of the equivalence class of |value|, which must
|
|
// already be known to the equivalence relation. This is the 'Find' operation
|
|
// in a classic union-find data structure.
|
|
const T* Find(const T* value) const {
|
|
assert(Exists(*value));
|
|
|
|
// Get the canonical pointer to the value from the value pool.
|
|
const T* known_value = *value_set_.find(value);
|
|
assert(parent_[known_value] && "Every known value should have a parent.");
|
|
|
|
// Compute the result by chasing parents until we find a value that is its
|
|
// own parent.
|
|
const T* result = known_value;
|
|
while (parent_[result] != result) {
|
|
result = parent_[result];
|
|
}
|
|
assert(result && "Representatives should never be null.");
|
|
|
|
// At this point, |result| is the representative of the equivalence class.
|
|
// Now perform the 'path compression' optimization by doing another pass up
|
|
// the parent chain, setting the parent of each node to be the
|
|
// representative, and rewriting children correspondingly.
|
|
const T* current = known_value;
|
|
while (parent_[current] != result) {
|
|
const T* next = parent_[current];
|
|
parent_[current] = result;
|
|
children_[result].push_back(current);
|
|
auto child_iterator =
|
|
std::find(children_[next].begin(), children_[next].end(), current);
|
|
assert(child_iterator != children_[next].end() &&
|
|
"'next' is the parent of 'current', so 'current' should be a "
|
|
"child of 'next'");
|
|
children_[next].erase(child_iterator);
|
|
current = next;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
private:
|
|
// Maps every value to a parent. The representative of an equivalence class
|
|
// is its own parent. A value's representative can be found by walking its
|
|
// chain of ancestors.
|
|
//
|
|
// Mutable because the intuitively const method, 'Find', performs path
|
|
// compression.
|
|
mutable std::unordered_map<const T*, const T*> parent_;
|
|
|
|
// Stores the children of each value. This allows the equivalence class of
|
|
// a value to be calculated by traversing all descendents of the class's
|
|
// representative.
|
|
//
|
|
// Mutable because the intuitively const method, 'Find', performs path
|
|
// compression.
|
|
mutable std::unordered_map<const T*, std::vector<const T*>> children_;
|
|
|
|
// The values known to the equivalence relation are allocated in
|
|
// |owned_values_|, and |value_pool_| provides (via |PointerHashT| and
|
|
// |PointerEqualsT|) a means for mapping a value of interest to a pointer
|
|
// into an equivalent value in |owned_values_|.
|
|
std::unordered_set<const T*, PointerHashT, PointerEqualsT> value_set_;
|
|
std::vector<std::unique_ptr<T>> owned_values_;
|
|
};
|
|
|
|
} // namespace fuzz
|
|
} // namespace spvtools
|
|
|
|
#endif // SOURCE_FUZZ_EQUIVALENCE_RELATION_H_
|