mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-01 03:45:06 +00:00
7d4dc66846
### Description <!-- Describe your changes. --> Make EP's member function, GenerateMetaDefId, a standalone function which decouples from EP ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. --> This change is for ExecutionProvider API refactoring, we will make a clean ExecutionProvider API first for later EPv2 work
90 lines
3.3 KiB
C++
90 lines
3.3 KiB
C++
// Copyright (c) Microsoft Corporation. All rights reserved.
|
|
// Licensed under the MIT License.
|
|
|
|
#pragma once
|
|
#include <set>
|
|
#include "core/framework/execution_provider.h"
|
|
#include "core/framework/model_metadef_id_generator.h"
|
|
|
|
namespace onnxruntime {
|
|
namespace internal_testing_ep {
|
|
|
|
class InternalTestingExecutionProvider : public IExecutionProvider {
|
|
public:
|
|
InternalTestingExecutionProvider(const std::unordered_set<std::string>& ops,
|
|
const std::unordered_set<std::string>& stop_ops = {},
|
|
DataLayout preferred_layout = DataLayout::NCHW);
|
|
|
|
virtual ~InternalTestingExecutionProvider();
|
|
|
|
std::vector<std::unique_ptr<ComputeCapability>>
|
|
GetCapability(const onnxruntime::GraphViewer& graph_view,
|
|
const IKernelLookup& /*kernel_lookup*/) const override;
|
|
|
|
common::Status Compile(const std::vector<FusedNodeAndGraph>& fused_nodes,
|
|
std::vector<NodeComputeInfo>& node_compute_funcs) override;
|
|
|
|
DataLayout GetPreferredLayout() const override;
|
|
std::shared_ptr<KernelRegistry> GetKernelRegistry() const override;
|
|
|
|
InternalTestingExecutionProvider& SetDebugOutput(bool debug_output) {
|
|
debug_output_ = debug_output;
|
|
return *this;
|
|
}
|
|
|
|
InternalTestingExecutionProvider& EnableStaticKernels() {
|
|
#if defined(ORT_MINIMAL_BUILD)
|
|
ORT_THROW("Static kernels are not currently supported in a minimal build");
|
|
#else
|
|
enable_static_kernels_ = true;
|
|
return *this;
|
|
|
|
#endif
|
|
}
|
|
|
|
/// <summary>
|
|
/// Request all nodes in GetCapability.
|
|
/// If EnableStaticKernels has been called, use static kernels for all nodes.
|
|
/// Otherwise compile the requested nodes.
|
|
///
|
|
/// NOTE: If using static kernels the graph will not be executable as we don't have the kernel implementations
|
|
/// so this is for testing model initialization components such as optimizers and layout transformation.
|
|
/// </summary>
|
|
InternalTestingExecutionProvider& TakeAllNodes() {
|
|
take_all_nodes_ = true;
|
|
return *this;
|
|
}
|
|
|
|
std::vector<AllocatorPtr> CreatePreferredAllocators() override;
|
|
|
|
private:
|
|
const std::string ep_name_;
|
|
|
|
// List of operators that the EP will claim nodes for.
|
|
// If take_all_nodes_ is true this is ignored.
|
|
const std::unordered_set<std::string> ops_;
|
|
|
|
// operators that we stop processing at.
|
|
// all nodes of an operator in this list and all their downstream nodes will be skipped
|
|
// e.g. NonMaxSuppression is the beginning of post-processing in an SSD model. It's unsupported for NNAPI,
|
|
// so from the NMS node on we want to use the CPU EP as the remaining work to do is far cheaper than
|
|
// the cost of going back to NNAPI.
|
|
const std::unordered_set<std::string> stop_ops_;
|
|
|
|
bool debug_output_{false};
|
|
bool enable_static_kernels_{false};
|
|
|
|
// request all nodes ignoring ops_ and stop_ops_.
|
|
// if enabled_static_kernels_ use static kernels for them, otherwise compile.
|
|
bool take_all_nodes_{false};
|
|
|
|
DataLayout preferred_layout_; // request all nodes
|
|
|
|
// per-instance kernel registry so tests using static kernels don't clash.
|
|
// shared_ptr as required by IExecutionProvider::GetKernelRegistry
|
|
std::shared_ptr<KernelRegistry> kernel_registry_;
|
|
ModelMetadefIdGenerator metadef_id_generator_;
|
|
};
|
|
|
|
} // namespace internal_testing_ep
|
|
} // namespace onnxruntime
|