[ARM CPU] Add ACL FC executor for FP32/FP16 precision #24123
+969
−127
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,134 @@ | ||
| // Copyright (C) 2024 Intel Corporation | ||
| // SPDX-License-Identifier: Apache-2.0 | ||
| // | ||
|
|
||
| #include "acl_common_executor.hpp" | ||
| #include "acl_utils.hpp" | ||
| #include "nodes/executors/memory_arguments.hpp" | ||
| #include "utils/debug_capabilities.h" | ||
|
|
||
| namespace ov { | ||
| namespace intel_cpu { | ||
|
|
||
| static const std::unordered_map<int, ACLArgs> argConvert = { | ||
| {ARG_SRC_0, ACL_SRC_0}, | ||
| {ARG_SRC_1, ACL_SRC_1}, | ||
| {ARG_SRC_2, ACL_SRC_2}, | ||
| {ARG_BIAS, ACL_BIAS}, | ||
| {ARG_WEI, ACL_WEI}, | ||
| {ARG_DST, ACL_DST}, | ||
| }; | ||
|
|
||
| using ACLTypes = std::array<arm_compute::DataType, ACLArgs::COUNT_OF_ARGS>; | ||
| using ACLLayouts = std::array<arm_compute::DataLayout, ACLArgs::COUNT_OF_ARGS>; | ||
|
|
||
| static void initACLTensorParams(const MemoryPtr& memoryPtr, | ||
| const ACLTensorAttrs& attrs, | ||
| arm_compute::TensorShape& tensorShape, | ||
| arm_compute::DataType& dataType, | ||
| arm_compute::DataLayout& dataLayout) { | ||
| dataType = precisionToAclDataType(memoryPtr->getPrecision()); | ||
| dataLayout = getAclDataLayoutByMemoryDesc(memoryPtr->getDescPtr()); | ||
| if (dataType != arm_compute::DataType::UNKNOWN) { | ||
| auto collapsed_dims = collapse_dims_to_max_rank(memoryPtr->getStaticDims(), attrs.maxDimsShape); | ||
| tensorShape = shapeCast(collapsed_dims); | ||
| if (attrs.hasLayoutTypeNHWC) { | ||
| changeLayoutToNH_C({&tensorShape}); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| static std::shared_ptr<arm_compute::TensorInfo> initTensorInfo(const arm_compute::TensorShape& tensorShape, | ||
| const arm_compute::DataType& dataType, | ||
| const arm_compute::DataLayout& dataLayout) { | ||
| std::shared_ptr<arm_compute::TensorInfo> aclMemoryInfo = nullptr; | ||
| if (dataType != arm_compute::DataType::UNKNOWN) { | ||
| aclMemoryInfo = std::make_shared<arm_compute::TensorInfo>( | ||
| tensorShape, 1, | ||
| dataType, | ||
| dataLayout); | ||
| } | ||
| return aclMemoryInfo; | ||
| } | ||
|
|
||
| static std::shared_ptr<arm_compute::Tensor> initTensor(const std::shared_ptr<arm_compute::TensorInfo>& aclMemoryInfo) { | ||
| std::shared_ptr<arm_compute::Tensor> aclMemory = nullptr; | ||
| if (aclMemoryInfo) { | ||
| aclMemory = std::make_shared<arm_compute::Tensor>(); | ||
| aclMemory->allocator()->init(*aclMemoryInfo); | ||
| } | ||
| return aclMemory; | ||
| } | ||
|
|
||
| ACLCommonExecutor::ACLCommonExecutor() { | ||
| for (int i = 0; i < ACLArgs::COUNT_OF_ARGS; ++i) { | ||
| aclTensorAttrs.memoryUsageIndicator[i] = false; | ||
| } | ||
| } | ||
|
|
||
| bool ACLCommonExecutor::update(const MemoryArgs &memory) { | ||
| // Initialize ACL tensors params | ||
| ACLShapes aclMemoryShapes; | ||
| ACLTypes aclDataType{}; | ||
| ACLLayouts aclDataLayout{}; | ||
| for (auto& cpu_mem_ptr : memory) { | ||
| const ACLArgs index = argConvert.at(cpu_mem_ptr.first); | ||
| initACLTensorParams(cpu_mem_ptr.second, aclTensorAttrs, | ||
| aclMemoryShapes[index], | ||
| aclDataType[index], | ||
| aclDataLayout[index]); | ||
| } | ||
|
|
||
| // Update ACL tensors shapes | ||
| updateTensorsShapes(aclMemoryShapes); | ||
|
|
||
| // Initialize arm_compute::TensorInfo objects | ||
| ACLInfos aclMemoryInfos; | ||
| for (int i = 0; i < ACLArgs::COUNT_OF_ARGS; i++) { | ||
| aclMemoryInfos[i] = initTensorInfo(aclMemoryShapes[i], aclDataType[i], aclDataLayout[i]); | ||
| } | ||
|
|
||
| // Validate arm_compute::TensorInfo objects for specific ACL function | ||
| auto tensorsInfoValidateStatus = validateTensorsInfo(aclMemoryInfos); | ||
| if (!tensorsInfoValidateStatus) { | ||
| DEBUG_LOG("ACL operator validation failed: ", tensorsInfoValidateStatus.error_description()); | ||
| return false; | ||
| } | ||
|
|
||
| // Initialize arm_compute::Tensor objects | ||
| for (int i = 0; i < ACLArgs::COUNT_OF_ARGS; i++) { | ||
| aclMemoryTensors[i] = initTensor(aclMemoryInfos[i]); | ||
| // Indicate that arm_compute::Tensor object can use import_memory function | ||
| if (aclMemoryTensors[i]) { | ||
| aclTensorAttrs.memoryUsageIndicator[i] = true; | ||
| } | ||
| } | ||
|
|
||
| // Configure arm_compute::IFunction object | ||
| configureThreadSafe([&] { | ||
| iFunction = configureFunction(aclMemoryTensors); | ||
| }); | ||
| return true; | ||
| } | ||
|
|
||
| void ACLCommonExecutor::execute(const MemoryArgs &memory) { | ||
| // TODO: Move import_memory() to update() function - CVS-145871 | ||
| for (auto& cpu_mem_ptr : memory) { | ||
| const ACLArgs index = argConvert.at(cpu_mem_ptr.first); | ||
| if (aclTensorAttrs.memoryUsageIndicator[index]) { | ||
| aclMemoryTensors[index]->allocator()->import_memory(memory.at(cpu_mem_ptr.first)->getData()); | ||
| } | ||
| } | ||
| iFunction->run(); | ||
| } | ||
|
|
||
| ACLCommonExecutor::~ACLCommonExecutor() { | ||
| for (int i = 0; i < ACLArgs::COUNT_OF_ARGS; i++) { | ||
| if (aclTensorAttrs.memoryUsageIndicator[i]) { | ||
| aclMemoryTensors[i]->allocator()->free(); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| } // namespace intel_cpu | ||
| } // namespace ov | ||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,58 @@ | ||
| // Copyright (C) 2018-2024 Intel Corporation | ||
| // SPDX-License-Identifier: Apache-2.0 | ||
| // | ||
|
|
||
| #pragma once | ||
|
|
||
| #include "cpu_memory.h" | ||
| #include "nodes/executors/executor.hpp" | ||
| #include "arm_compute/runtime/NEON/NEFunctions.h" | ||
|
|
||
| namespace ov { | ||
| namespace intel_cpu { | ||
|
|
||
| enum ACLArgs { | ||
| ACL_SRC_0, | ||
| ACL_SRC_1, | ||
| ACL_SRC_2, | ||
| ACL_BIAS, | ||
| ACL_WEI, | ||
| ACL_DST, | ||
| COUNT_OF_ARGS | ||
| }; | ||
|
|
||
| using ACLFunction = std::unique_ptr<arm_compute::IFunction>; | ||
| using ACLShapes = std::array<arm_compute::TensorShape, ACLArgs::COUNT_OF_ARGS>; | ||
| using ACLInfos = std::array<std::shared_ptr<arm_compute::TensorInfo>, ACLArgs::COUNT_OF_ARGS>; | ||
| using ACLTensors = std::array<std::shared_ptr<arm_compute::Tensor>, ACLArgs::COUNT_OF_ARGS>; | ||
|
|
||
| struct ACLTensorAttrs { | ||
| bool hasLayoutTypeNHWC = false; | ||
| size_t maxDimsShape = arm_compute::MAX_DIMS; | ||
| std::array<bool, ACLArgs::COUNT_OF_ARGS> memoryUsageIndicator; | ||
| }; | ||
|
|
||
| class ACLCommonExecutor : public Executor { | ||
| public: | ||
| ACLCommonExecutor(); | ||
| virtual void updateTensorsShapes(ACLShapes& aclMemoryShapes) = 0; | ||
| virtual arm_compute::Status validateTensorsInfo(const ACLInfos& aclMemoryInfos) = 0; | ||
| virtual ACLFunction configureFunction(const ACLTensors& aclMemoryTensors) = 0; | ||
| impl_desc_type implType() const override { | ||
| return impl_desc_type::acl; | ||
| } | ||
| void execute(const MemoryArgs& memory) override; | ||
| bool update(const MemoryArgs& memory) override; | ||
| ~ACLCommonExecutor(); | ||
|
|
||
| protected: | ||
| ACLTensorAttrs aclTensorAttrs; | ||
| private: | ||
| ACLTensors aclMemoryTensors; | ||
| ACLFunction iFunction = nullptr; | ||
| }; | ||
|
|
||
| using ACLCommonExecutorPtr = std::shared_ptr<ACLCommonExecutor>; | ||
|
|
||
| } // namespace intel_cpu | ||
| } // namespace ov |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I propose to leave a todo regarding the fact, that actually it should be enough to import_memory just once in scope of "update()" method, but it is not working for some reason and should be investigated.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@EgorDuplensky added