mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

delete ENABLE_TUPLE_UNFOLD

This commit is contained in:
wYann 2023-03-03 10:28:13 +08:00
parent 390f6e35ba
commit b294db6b05
8 changed files with 0 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cmake/options.cmake
View File

@ -76,8 +76,6 @@ if(ENABLE_ASAN)
endif()
endif()
add_compile_definitions(ENABLE_TUPLE_UNFOLD)
if(DEBUG_MODE)
set(CMAKE_BUILD_TYPE "Debug")
add_compile_definitions(MEM_REUSE_DEBUG)

5
mindspore/ccsrc/backend/common/optimizer/common_backend_optimization.cc
View File

@ -55,14 +55,9 @@ PassManagerPtr GetBackendCommonOptimizationPassManagerPtr(const FuncGraphPtr &gr
common_pm->AddPass(std::make_shared<ConvertTupleOutputToMaketuple>());
common_pm->AddPass(std::make_shared<ConvertUnusedTupleParaToMakeTuple>());
common_pm->AddPass(std::make_shared<ConvertConstScalarToTensor>());
#ifdef ENABLE_TUPLE_UNFOLD
MS_LOG(INFO) << "Enable tuple unfold.";
if (graph->has_flag(kAttrMutableKernel) || graph->has_flag(kFlagEnableRunGraphBySingleOp)) {
common_pm->AddPass(std::make_shared<ConvertTupleInputToDynamicInput>());
}
#else
common_pm->AddPass(std::make_shared<ConvertTupleInputToDynamicInput>());
#endif
common_pm->AddPass(std::make_shared<FlattenConcatFission>());
common_pm->AddPass(std::make_shared<AddDropoutAttrs>());
return common_pm;

7
mindspore/ccsrc/backend/common/pass/convert_tuple_output_to_maketuple.cc
View File

@ -49,25 +49,18 @@ AnfNodePtr ConvertTupleInputToMakeTuple(const FuncGraphPtr &graph, const AnfNode
return make_tuple;
}
#ifdef ENABLE_TUPLE_UNFOLD
bool IsKerenlGraphOutput(const FuncGraphPtr &func_graph, const AnfNodePtr &node) {
const auto &outputs =
common::AnfAlgo::GetAllOutputIndexByReturnTypes(func_graph->output(), {prim::kPrimTupleGetItem});
return std::find_if(outputs.begin(), outputs.end(), [&node](const auto &output) { return output.first == node; }) !=
outputs.end();
}
#endif
bool IsNeedConvert(const FuncGraphPtr &func_graph, const AnfNodePtr &input) {
#ifdef ENABLE_TUPLE_UNFOLD
MS_EXCEPTION_IF_NULL(input);
return (input->Type() != nullptr && AnfUtils::IsRealKernel(input) && common::AnfAlgo::IsTupleOutput(input) &&
!common::AnfAlgo::CheckPrimitiveType(input, prim::kPrimCall) &&
(input->isa<Parameter>() || input->isa<ValueNode>() || IsKerenlGraphOutput(func_graph, input)));
#else
return (input->Type() != nullptr && AnfUtils::IsRealKernel(input) && common::AnfAlgo::IsTupleOutput(input) &&
!common::AnfAlgo::CheckPrimitiveType(input, prim::kPrimCall));
#endif
}
} // namespace

10
mindspore/ccsrc/backend/common/session/anf_runtime_algorithm.cc
View File

@ -139,7 +139,6 @@ tensor::TensorPtr GetForwardOutputTensor(const AnfNodePtr &node) {
return nullptr;
}
#ifdef ENABLE_TUPLE_UNFOLD
size_t GetOutputTensorNumByKernelInfo(const AnfNodePtr &node) {
MS_EXCEPTION_IF_NULL(node);
MS_EXCEPTION_IF_NULL(node->kernel_info());
@ -149,7 +148,6 @@ size_t GetOutputTensorNumByKernelInfo(const AnfNodePtr &node) {
MS_EXCEPTION_IF_NULL(build_info);
return build_info->GetAllOutputDeviceTypes().size();
}
#endif
} // namespace
AnfNodePtr AnfRuntimeAlgorithm::MakeMonadValueNode(const KernelGraphPtr &kg) {
@ -196,7 +194,6 @@ void AnfRuntimeAlgorithm::KeepOrder(const KernelGraphPtr &kg, const AnfNodePtr &
}
size_t AnfRuntimeAlgorithm::GetOutputTensorNum(const AnfNodePtr &node) {
#ifdef ENABLE_TUPLE_UNFOLD
MS_EXCEPTION_IF_NULL(node);
size_t res;
TypePtr type = node->Type();
@ -225,13 +222,9 @@ size_t AnfRuntimeAlgorithm::GetOutputTensorNum(const AnfNodePtr &node) {
res = 1;
}
return res;
#else
return AnfUtils::GetOutputTensorNum(node);
#endif
}
size_t AnfRuntimeAlgorithm::GetOutputNumWithoutKernelInfo(const AnfNodePtr &node) {
#ifdef ENABLE_TUPLE_UNFOLD
MS_EXCEPTION_IF_NULL(node);
const auto &kernel_info = node->kernel_info();
if (kernel_info != nullptr) {
@ -261,9 +254,6 @@ size_t AnfRuntimeAlgorithm::GetOutputNumWithoutKernelInfo(const AnfNodePtr &node
res = 1;
}
return res;
#else
return AnfUtils::GetOutputTensorNum(node);
#endif
}
namespace {

10
mindspore/ccsrc/plugin/device/cpu/hal/device/kernel_select_cpu.cc
View File

@ -82,13 +82,11 @@ void GetOutputDtypes(const CNodePtr &kernel_node, std::vector<TypeId> *output_ty
}
}
#ifdef ENABLE_TUPLE_UNFOLD
// Real tuple isn't expanded.
void GetOutputDtypesForRealTuple(const CNodePtr &kernel_node, std::vector<TypeId> *output_types) {
TypeId dtype = common::AnfAlgo::GetOutputInferDataType(kernel_node, 0);
(void)output_types->emplace_back(dtype);
}
#endif
void GetOutputFormat(const CNodePtr &kernel_node, std::vector<std::string> *output_formats) {
size_t output_num = AnfAlgo::GetOutputElementNum(kernel_node);
@ -544,7 +542,6 @@ bool SelectKernel(const CNodePtr &kernel_node, kernel::KernelAttr *selected_kern
bool input_matched = false;
for (auto kernel_attr : kernel_attrs) {
output_types.clear();
#ifdef ENABLE_TUPLE_UNFOLD
// The real tuple and allsame don't fold the tuple.
if (kernel_attr.GetAllSame() ||
(kernel_attr.GetOutputSize() != 0 && kernel_attr.GetOutputAttr(0).object_type == kObjectTypeTuple)) {
@ -552,9 +549,6 @@ bool SelectKernel(const CNodePtr &kernel_node, kernel::KernelAttr *selected_kern
} else {
GetOutputDtypes(kernel_node, &output_types);
}
#else
GetOutputDtypes(kernel_node, &output_types);
#endif
MS_LOG(DEBUG) << "Select kernel for op: " << kernel_node->fullname_with_scope() << ", input types:" << input_types
<< ", output types:" << output_types;
@ -647,7 +641,6 @@ std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kerne
// First select the kernel object types.
std::vector<kernel::KernelAttr> object_selected_kernel_attrs;
const auto &kernel_attrs = kernel::NativeCpuKernelMod::GetCpuSupportedList(op_name);
#ifdef ENABLE_TUPLE_UNFOLD
if (kernel_attrs.empty()) {
return KernelNotSupportWarning(kernel_node, false);
} else if (kernel_attrs[0].GetSkipCheck()) {
@ -656,9 +649,6 @@ std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kerne
!kernel::SelectKernelByObjectType(kernel_node, kernel_attrs, &object_selected_kernel_attrs, false)) {
return kernel::KernelObjectTypeNotSupportWarning(kernel_node);
}
#else
object_selected_kernel_attrs = kernel_attrs;
#endif
// Second select the matched kernel attr.
kernel::KernelAttr selected_kernel_attr;

2
mindspore/ccsrc/plugin/device/cpu/hal/hardware/cpu_device_context.cc
View File

@ -236,9 +236,7 @@ void CPUKernelExecutor::OptimizeGraphImpl(const KernelGraphPtr &graph) const {
MS_EXCEPTION_IF_NULL(graph);
auto optimizer = std::make_shared<opt::GraphOptimizer>();
auto pm = std::make_shared<opt::PassManager>();
#ifdef ENABLE_TUPLE_UNFOLD
pm->AddPass(std::make_shared<opt::InsertTypeTransformOp>("insert_type_transform_op"));
#endif
pm->AddPass(std::make_shared<opt::InsertFormatTransformOpCPU>("insert_format_transform_op_cpu"));
pm->AddPass(std::make_shared<opt::AllReduceFusion>());
pm->AddPass(std::make_shared<opt::InsertCastCPU>("insert_cast"));

10
mindspore/ccsrc/plugin/device/gpu/hal/device/kernel_info_setter.cc
View File

@ -593,7 +593,6 @@ bool GetSelectKernelResult(const CNodePtr &kernel_node,
return result;
}
#ifdef ENABLE_TUPLE_UNFOLD
bool GetSelectKernelObjectTypeResult(const CNodePtr &kernel_node, KernelType kernel_type) {
auto kernel_name = common::AnfAlgo::GetCNodeName(kernel_node);
// Only the kernel nodes that register kernel attr can support the backoff.
@ -631,7 +630,6 @@ bool GetSelectKernelObjectTypeResult(const CNodePtr &kernel_node, KernelType ker
kernel::SetKernelObjectTypeWithSelectedAttr(kernel_node, object_selected_kernel_attrs[0]);
return true;
}
#endif
std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kernel_node, KernelType kernel_type) {
MS_EXCEPTION_IF_NULL(kernel_node);
@ -643,12 +641,10 @@ std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kerne
}
auto builder = std::make_shared<KernelBuildInfo::KernelBuildInfoBuilder>();
AnfAlgo::SetSelectKernelBuildInfo(builder->Build(), kernel_node.get());
#ifdef ENABLE_TUPLE_UNFOLD
bool selected = GetSelectKernelObjectTypeResult(kernel_node, kernel_type);
if (!selected) {
return kernel::KernelObjectTypeNotSupportWarning(kernel_node);
}
#endif
std::vector<std::string> inputs_format;
std::vector<TypeId> inputs_type;
size_t input_num = common::AnfAlgo::GetInputTensorNum(kernel_node);
@ -659,21 +655,17 @@ std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kerne
std::vector<std::string> outputs_format;
std::vector<TypeId> outputs_type;
#ifdef ENABLE_TUPLE_UNFOLD
auto output_kernel_object_types = builder->Build()->GetAllOutputKernelObjectTypes();
if (output_kernel_object_types.size() == 1 && output_kernel_object_types[0] == kernel::KernelObjectType::TUPLE) {
outputs_type = {common::AnfAlgo::GetOutputInferDataType(kernel_node, 0)};
outputs_format = {kOpFormat_DEFAULT};
} else {
#endif
size_t output_num = AnfAlgo::GetOutputElementNum(kernel_node);
for (size_t output_index = 0; output_index < output_num; ++output_index) {
outputs_format.emplace_back(kOpFormat_DEFAULT);
outputs_type.push_back(common::AnfAlgo::GetOutputInferDataType(kernel_node, output_index));
}
#ifdef ENABLE_TUPLE_UNFOLD
}
#endif
std::string origin_data_format = kOpFormat_DEFAULT;
if (IsNeedProcessFormatInfo(kernel_node, inputs_type)) {
UpdateKernelFormatInfo(kernel_node, inputs_type, &inputs_format, &outputs_format, &origin_data_format);
@ -683,12 +675,10 @@ std::pair<std::string, ExceptionType> SetKernelInfoWithMsg(const CNodePtr &kerne
builder->SetInputsDeviceType(inputs_type);
builder->SetOutputsFormat(outputs_format);
builder->SetOutputsDeviceType(outputs_type);
#ifdef ENABLE_TUPLE_UNFOLD
kernel::UnfoldKernelBuildInfo(kernel_node);
if (!common::AnfAlgo::HasNodeAttr(kAttrDynInputSizes, kernel_node)) {
kernel::SetDynamicInputSizeAttr(kernel_node);
}
#endif
MS_LOG(INFO) << kernel_node->fullname_with_scope() << " kernel attr info: "
<< kernel::FetchPrintInfoByKernelAttr(kernel::GetKernelAttrFromBuildInfo(builder->Build()));

2
mindspore/ccsrc/plugin/device/gpu/hal/hardware/gpu_device_context.cc
View File

@ -356,9 +356,7 @@ void GPUKernelExecutor::OptimizeGraphWithDeviceInfo(const KernelGraphPtr &graph)
// Graph optimization relevant to device data format
auto optimizer = std::make_shared<opt::GraphOptimizer>();
auto pm = std::make_shared<opt::PassManager>();
#ifdef ENABLE_TUPLE_UNFOLD
pm->AddPass(std::make_shared<opt::InsertTypeTransformOp>("insert_type_transform_op"));
#endif
// ReplaceAddNFusion depends on the input expansion of AddN, so must be after the operator select.
pm->AddPass(std::make_shared<opt::ReplaceAddNFusion>());
// PrintReduceFusion depends on the input expansion of Print, so must be after the operator select.