SPIRV-Tools/source/fuzz/fuzzer_pass_apply_id_synonyms.cpp
Alastair Donaldson 67f4838659
spirv-fuzz: Make handling of synonym facts more efficient (#3301)
The fact manager maintains an equivalence relation on data descriptors
that tracks when one data descriptor could be used in place of
another.  An algorithm to compute the closure of such facts allows
deducing new synonym facts from existing facts.  E.g., for two 2D
vectors u and v it is known that u.x is synonymous with v.x and u.y is
synonymous with v.y, it can be deduced that u and v are synonymous.

The closure computation algorithm is very expensive if we get large
equivalence relations.

This change addresses this in three ways:

- The size of equivalence relations is reduced by limiting the extent
  to which the components of a composite are recursively noted as
  being equivalent, so that when we have large synonymous arrays we do
  not record all array elements as being pairwise equivalent.

- When computing the closure of facts, equivalence classes above a
  certain size are simply skipped (which can lead to missed facts)

- The closure computation is performed less frequently - it is invoked
  explicitly before fuzzer passes that will benefit from data synonym
  facts.  A new transformation is used to control its invocation, so
  that fuzzing and replaying do not get out of sync.

The change also tidies up the order in which some getters are declared
in FuzzerContext.
2020-04-20 19:02:49 +01:00

165 lines
7.1 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.
#include "source/fuzz/fuzzer_pass_apply_id_synonyms.h"
#include "source/fuzz/data_descriptor.h"
#include "source/fuzz/fuzzer_util.h"
#include "source/fuzz/id_use_descriptor.h"
#include "source/fuzz/instruction_descriptor.h"
#include "source/fuzz/transformation_composite_extract.h"
#include "source/fuzz/transformation_compute_data_synonym_fact_closure.h"
#include "source/fuzz/transformation_replace_id_with_synonym.h"
namespace spvtools {
namespace fuzz {
FuzzerPassApplyIdSynonyms::FuzzerPassApplyIdSynonyms(
opt::IRContext* ir_context, TransformationContext* transformation_context,
FuzzerContext* fuzzer_context,
protobufs::TransformationSequence* transformations)
: FuzzerPass(ir_context, transformation_context, fuzzer_context,
transformations) {}
FuzzerPassApplyIdSynonyms::~FuzzerPassApplyIdSynonyms() = default;
void FuzzerPassApplyIdSynonyms::Apply() {
// Compute a closure of data synonym facts, to enrich the pool of synonyms
// that are available.
ApplyTransformation(TransformationComputeDataSynonymFactClosure(
GetFuzzerContext()
->GetMaximumEquivalenceClassSizeForDataSynonymFactClosure()));
for (auto id_with_known_synonyms : GetTransformationContext()
->GetFactManager()
->GetIdsForWhichSynonymsAreKnown()) {
// Gather up all uses of |id_with_known_synonym| as a regular id, and
// subsequently iterate over these uses. We use this separation because,
// when considering a given use, we might apply a transformation that will
// invalidate the def-use manager.
std::vector<std::pair<opt::Instruction*, uint32_t>> uses;
GetIRContext()->get_def_use_mgr()->ForEachUse(
id_with_known_synonyms,
[&uses](opt::Instruction* use_inst, uint32_t use_index) -> void {
// We only gather up regular id uses; e.g. we do not include a use of
// the id as the scope for an atomic operation.
if (use_inst->GetOperand(use_index).type == SPV_OPERAND_TYPE_ID) {
uses.emplace_back(
std::pair<opt::Instruction*, uint32_t>(use_inst, use_index));
}
});
for (auto& use : uses) {
auto use_inst = use.first;
auto use_index = use.second;
auto block_containing_use = GetIRContext()->get_instr_block(use_inst);
// The use might not be in a block; e.g. it could be a decoration.
if (!block_containing_use) {
continue;
}
if (!GetFuzzerContext()->ChoosePercentage(
GetFuzzerContext()->GetChanceOfReplacingIdWithSynonym())) {
continue;
}
// |use_index| is the absolute index of the operand. We require
// the index of the operand restricted to input operands only, so
// we subtract the number of non-input operands from |use_index|.
uint32_t use_in_operand_index =
use_index - use_inst->NumOperands() + use_inst->NumInOperands();
if (!TransformationReplaceIdWithSynonym::UseCanBeReplacedWithSynonym(
GetIRContext(), use_inst, use_in_operand_index)) {
continue;
}
std::vector<const protobufs::DataDescriptor*> synonyms_to_try;
for (auto& data_descriptor :
GetTransformationContext()->GetFactManager()->GetSynonymsForId(
id_with_known_synonyms)) {
protobufs::DataDescriptor descriptor_for_this_id =
MakeDataDescriptor(id_with_known_synonyms, {});
if (DataDescriptorEquals()(data_descriptor, &descriptor_for_this_id)) {
// Exclude the fact that the id is synonymous with itself.
continue;
}
synonyms_to_try.push_back(data_descriptor);
}
while (!synonyms_to_try.empty()) {
auto synonym_to_try =
GetFuzzerContext()->RemoveAtRandomIndex(&synonyms_to_try);
// If the synonym's |index_size| is zero, the synonym represents an id.
// Otherwise it represents some element of a composite structure, in
// which case we need to be able to add an extract instruction to get
// that element out.
if (synonym_to_try->index_size() > 0 &&
!fuzzerutil::CanInsertOpcodeBeforeInstruction(SpvOpCompositeExtract,
use_inst) &&
use_inst->opcode() != SpvOpPhi) {
// We cannot insert an extract before this instruction, so this
// synonym is no good.
continue;
}
if (!fuzzerutil::IdIsAvailableAtUse(GetIRContext(), use_inst,
use_in_operand_index,
synonym_to_try->object())) {
continue;
}
// We either replace the use with an id known to be synonymous (when
// the synonym's |index_size| is 0), or an id that will hold the result
// of extracting a synonym from a composite (when the synonym's
// |index_size| is > 0).
uint32_t id_with_which_to_replace_use;
if (synonym_to_try->index_size() == 0) {
id_with_which_to_replace_use = synonym_to_try->object();
} else {
id_with_which_to_replace_use = GetFuzzerContext()->GetFreshId();
opt::Instruction* instruction_to_insert_before = nullptr;
if (use_inst->opcode() != SpvOpPhi) {
instruction_to_insert_before = use_inst;
} else {
auto parent_block_id =
use_inst->GetSingleWordInOperand(use_in_operand_index + 1);
auto parent_block_instruction =
GetIRContext()->get_def_use_mgr()->GetDef(parent_block_id);
auto parent_block =
GetIRContext()->get_instr_block(parent_block_instruction);
instruction_to_insert_before = parent_block->GetMergeInst()
? parent_block->GetMergeInst()
: parent_block->terminator();
}
ApplyTransformation(TransformationCompositeExtract(
MakeInstructionDescriptor(GetIRContext(),
instruction_to_insert_before),
id_with_which_to_replace_use, synonym_to_try->object(),
fuzzerutil::RepeatedFieldToVector(synonym_to_try->index())));
}
ApplyTransformation(TransformationReplaceIdWithSynonym(
MakeIdUseDescriptorFromUse(GetIRContext(), use_inst,
use_in_operand_index),
id_with_which_to_replace_use));
break;
}
}
}
}
} // namespace fuzz
} // namespace spvtools