diff options
Diffstat (limited to 'src/dynamic_fusion/sketch/gpu/components')
25 files changed, 258 insertions, 295 deletions
diff --git a/src/dynamic_fusion/sketch/gpu/components/GpuKernelComponentFactory.h b/src/dynamic_fusion/sketch/gpu/components/GpuKernelComponentFactory.h index f7f0029618..ee109a7e2b 100644 --- a/src/dynamic_fusion/sketch/gpu/components/GpuKernelComponentFactory.h +++ b/src/dynamic_fusion/sketch/gpu/components/GpuKernelComponentFactory.h @@ -24,8 +24,9 @@ #ifndef SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_GPUKERNELCOMPONENTFACTORY #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_GPUKERNELCOMPONENTFACTORY -#include "Types.h" #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" + +#include "Types.h" #include <memory> namespace arm_compute @@ -49,13 +50,13 @@ public: * @return std::unique_ptr<IGpuKernelComponent> */ template <typename T, typename... Args> - std::unique_ptr<IGpuKernelComponent> create(Args &&... args) + std::unique_ptr<IGpuKernelComponent> create(Args &&...args) { return std::make_unique<T>(_count++, std::forward<Args>(args)...); } private: - ComponentId _count{ 0 }; + ComponentId _count{0}; }; } // namespace dynamic_fusion diff --git a/src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h b/src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h index af766a7ece..4b8eea2f57 100644 --- a/src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h +++ b/src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h @@ -24,11 +24,11 @@ #ifndef ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_IGPUKERNELCOMPONENT #define ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_IGPUKERNELCOMPONENT -#include "Types.h" - #include "src/dynamic_fusion/sketch/ArgumentPack.h" #include "src/dynamic_fusion/sketch/gpu/GpuWorkloadSourceCode.h" +#include "Types.h" + namespace arm_compute { namespace experimental @@ -76,13 +76,8 @@ public: * @param[in] properties Kernel component properties * @param[in] tensors Tensor arguments to the components */ - IGpuKernelComponent( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors) - : _id{ id }, - _properties{ properties }, - _tensors{ tensors } + IGpuKernelComponent(ComponentId id, const Properties &properties, const ArgumentPack<ITensorInfo> &tensors) + : _id{id}, _properties{properties}, _tensors{tensors} { } /** Destructor */ @@ -117,7 +112,7 @@ public: virtual GpuComponentType type() const = 0; private: - ComponentId _id{ -1 }; + ComponentId _id{-1}; Properties _properties{}; ArgumentPack<ITensorInfo> _tensors{}; }; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.cpp index c41257d18c..fdf528a65d 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.cpp @@ -68,17 +68,11 @@ ClComponentActivation::ClComponentActivation(ComponentId const IGpuKernelComponent::Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes) - : IGpuKernelComponent{ id, properties, tensors }, + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<ClTemplateActivation>(id, tensors, attributes) -} + _component_writer{std::make_unique<ClTemplateActivation>(id, tensors, attributes)} #else //ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<GpuCkwActivation>(id, tensors, attributes) -} + _component_writer{std::make_unique<GpuCkwActivation>(id, tensors, attributes)} #endif //ACL_INTERNAL_TEST_CKW_IN_DF { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.h index 9b090af988..02c854356a 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.h @@ -25,9 +25,8 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTACTIVATION #include "arm_compute/function_info/ActivationLayerInfo.h" -#include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" -#include "arm_compute/function_info/ActivationLayerInfo.h" +#include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute { @@ -79,20 +78,17 @@ public: * |F16 |F16 | * |F32 |F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + static Status + validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes); /** Constructor * * Similar to @ref ClComponentActivation::validate() */ - ClComponentActivation( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + ClComponentActivation(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes); /** Destructor */ ~ClComponentActivation() override; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.cpp index 635869f817..b1636795a3 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.cpp @@ -24,6 +24,7 @@ #include "ClComponentCast.h" #include "arm_compute/core/Error.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/ArgumentPack.h" #ifndef ACL_INTERNAL_TEST_CKW_IN_DF @@ -38,11 +39,10 @@ namespace experimental { namespace dynamic_fusion { -Status ClComponentCast::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) +Status ClComponentCast::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) { ARM_COMPUTE_UNUSED(properties, attributes, settings); @@ -53,13 +53,15 @@ Status ClComponentCast::validate( ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(dst); ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst); ARM_COMPUTE_RETURN_ERROR_ON(src == dst); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->data_type() == attributes.data_type(), "input and target data types should be different"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->data_type() == attributes.data_type(), + "input and target data types should be different"); // Validate in case of configured dst - if(dst->total_size() > 0) + if (dst->total_size() > 0) { ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_type() != attributes.data_type(), "dst and target data types should be same"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_type() != attributes.data_type(), + "dst and target data types should be same"); } return Status{}; @@ -69,17 +71,11 @@ ClComponentCast::ClComponentCast(ComponentId id, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes, const Settings &settings) - : IGpuKernelComponent{ id, properties, tensors }, + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<ClTemplateCast>(id, tensors, attributes) -} + _component_writer{std::make_unique<ClTemplateCast>(id, tensors, attributes)} #else //ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<GpuCkwCast>(id, tensors, attributes) -} + _component_writer{std::make_unique<GpuCkwCast>(id, tensors, attributes)} #endif //ACL_INTERNAL_TEST_CKW_IN_DF { ARM_COMPUTE_UNUSED(attributes, settings); diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.h index 37b8cbb6c9..ed77b1203b 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.h @@ -25,6 +25,7 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTCAST #include "arm_compute/dynamic_fusion/sketch/attributes/CastAttributes.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute @@ -93,11 +94,10 @@ public: * |F16 | U8, S8, U16, S16, U32, S32, F32 | * |F32 | U8, S8, U16, S16, U32, S32, F16 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + static Status validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Constructor * diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.cpp index 5626093079..d95e0be1f2 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.cpp @@ -26,6 +26,7 @@ #include "arm_compute/core/CL/CLHelpers.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/dynamic_fusion/sketch/attributes/DepthwiseConv2dAttributes.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.h" @@ -103,11 +104,10 @@ unsigned int Settings::m0() const return _m0; } -Status ClComponentDepthwiseConv2d::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) +Status ClComponentDepthwiseConv2d::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) { ARM_COMPUTE_UNUSED(properties, settings); const auto src = tensors.get_const_tensor(TensorType::ACL_SRC_0); @@ -121,7 +121,7 @@ Status ClComponentDepthwiseConv2d::validate( // Matching data type ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, wei); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, bia); } @@ -129,7 +129,7 @@ Status ClComponentDepthwiseConv2d::validate( // Matching data layout ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, wei); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, bia); } @@ -138,7 +138,7 @@ Status ClComponentDepthwiseConv2d::validate( ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() == 0); ARM_COMPUTE_RETURN_ERROR_ON(wei->tensor_shape().total_size() == 0); ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON(bia->tensor_shape().total_size() == 0); } @@ -148,16 +148,17 @@ Status ClComponentDepthwiseConv2d::validate( const DataLayout data_layout = src->data_layout(); const size_t channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL); - ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) != (src->dimension(channel_idx) * attributes.depth_multiplier())); + ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) != + (src->dimension(channel_idx) * attributes.depth_multiplier())); ARM_COMPUTE_RETURN_ERROR_ON_MSG(wei->num_dimensions() > 3, "Weights can be at most 3 dimensional"); // dst shape is correct - const PadStrideInfo pad_stride_info = PadStrideInfo(attributes.stride().x(), attributes.stride().y(), - attributes.pad().left, attributes.pad().right, - attributes.pad().top, attributes.pad().bottom, - attributes.dimension_rounding_type()); - const ConvolutionInfo conv_info{ pad_stride_info, attributes.depth_multiplier(), ActivationLayerInfo(), attributes.dilation() }; - const TensorShape output_shape = misc::shape_calculator::compute_depthwise_convolution_shape(*src, *wei, conv_info); + const PadStrideInfo pad_stride_info = + PadStrideInfo(attributes.stride().x(), attributes.stride().y(), attributes.pad().left, attributes.pad().right, + attributes.pad().top, attributes.pad().bottom, attributes.dimension_rounding_type()); + const ConvolutionInfo conv_info{pad_stride_info, attributes.depth_multiplier(), ActivationLayerInfo(), + attributes.dilation()}; + const TensorShape output_shape = misc::shape_calculator::compute_depthwise_convolution_shape(*src, *wei, conv_info); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), output_shape); @@ -168,19 +169,22 @@ Status ClComponentDepthwiseConv2d::validate( ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().first > 1 && settings.m0() != 1); ARM_COMPUTE_RETURN_ERROR_ON(conv_info.dilation.x() > 1 && settings.m0() != 1); - if(conv_info.depth_multiplier > 1 && settings.n0() > 1) + if (conv_info.depth_multiplier > 1 && settings.n0() > 1) { ARM_COMPUTE_RETURN_ERROR_ON((conv_info.depth_multiplier % settings.n0()) != 0); } // Check export weights to cl image - ARM_COMPUTE_RETURN_ERROR_ON_MSG((settings.export_weights_to_cl_image() == true) && (export_to_cl_image(wei) == false), "Weights cannot be exported to cl_image!"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG((settings.export_weights_to_cl_image() == true) && + (export_to_cl_image(wei) == false), + "Weights cannot be exported to cl_image!"); ARM_COMPUTE_RETURN_ERROR_ON((settings.export_weights_to_cl_image() == true) && ((settings.n0() % 4) != 0)); - ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) != (src->dimension(channel_idx) * conv_info.depth_multiplier)); + ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) != + (src->dimension(channel_idx) * conv_info.depth_multiplier)); // bia shape is correct - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->dimension(0) != output_shape[channel_idx], "Biases size and number of dst feature maps should match"); @@ -198,14 +202,13 @@ Status ClComponentDepthwiseConv2d::validate( return Status{}; } -ClComponentDepthwiseConv2d::ClComponentDepthwiseConv2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) - : IGpuKernelComponent{ id, properties, tensors }, - _component_writer{ std::make_unique<ClTemplateDepthwiseConv2d>(id, tensors, attributes, settings) } +ClComponentDepthwiseConv2d::ClComponentDepthwiseConv2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) + : IGpuKernelComponent{id, properties, tensors}, + _component_writer{std::make_unique<ClTemplateDepthwiseConv2d>(id, tensors, attributes, settings)} { } ClComponentDepthwiseConv2d::~ClComponentDepthwiseConv2d() diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.h index 0e2b5f14cb..b3e1bd222d 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.h @@ -25,7 +25,9 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTDEPTHWISECONV2D #include "arm_compute/core/Error.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" + #include <memory> namespace arm_compute @@ -77,12 +79,12 @@ public: unsigned int m0() const; private: - bool _export_input_to_cl_image{ false }; /**< Export input to cl_image */ - bool _export_weights_to_cl_image{ false }; /**< Export the weights to cl_image */ - bool _fast_relaxed_math{ true }; /**< Enable/disable -cl-fast-relaxed-math flag */ - bool _is_fma_available{ false }; /**< Is fma instruction available */ - unsigned int _n0{ 0 }; /**< Number of columns processed by each thread */ - unsigned int _m0{ 0 }; /**< Number of rows processed by each thread */ + bool _export_input_to_cl_image{false}; /**< Export input to cl_image */ + bool _export_weights_to_cl_image{false}; /**< Export the weights to cl_image */ + bool _fast_relaxed_math{true}; /**< Enable/disable -cl-fast-relaxed-math flag */ + bool _is_fma_available{false}; /**< Is fma instruction available */ + unsigned int _n0{0}; /**< Number of columns processed by each thread */ + unsigned int _m0{0}; /**< Number of rows processed by each thread */ }; /** Forward declaration */ @@ -127,22 +129,20 @@ public: * |F16 |F16 |F16 |F16 | * |F32 |F32 |F32 |F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + static Status validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Constructor * * Similar to @ref ClComponentDepthwiseConv2d::validate() */ - ClComponentDepthwiseConv2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + ClComponentDepthwiseConv2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Destructor */ ~ClComponentDepthwiseConv2d() override; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.cpp index a713c82003..98f3d6a882 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.cpp @@ -23,8 +23,8 @@ */ #include "ClComponentDirectConv2d.h" -#include "arm_compute/core/Validate.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/core/Validate.h" #include "arm_compute/dynamic_fusion/sketch/attributes/Conv2dAttributes.h" #include "src/core/CL/CLValidate.h" @@ -57,7 +57,8 @@ bool ClComponentDirectConv2dSettings::fast_relaxed_math() const return _fast_relaxed_math; } -ClComponentDirectConv2dSettings &ClComponentDirectConv2dSettings::direct_conv_descriptor(const DirectConvComputeKernelInfo &desc) +ClComponentDirectConv2dSettings & +ClComponentDirectConv2dSettings::direct_conv_descriptor(const DirectConvComputeKernelInfo &desc) { _desc = desc; return *this; @@ -68,11 +69,10 @@ DirectConvComputeKernelInfo ClComponentDirectConv2dSettings::direct_conv_descrip return _desc; } -Status ClComponentDirectConv2d::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) +Status ClComponentDirectConv2d::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) { ARM_COMPUTE_UNUSED(properties); const auto src = tensors.get_const_tensor(TensorType::ACL_SRC_0); @@ -86,7 +86,7 @@ Status ClComponentDirectConv2d::validate( // Matching data type ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, wei); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, bia); } @@ -94,7 +94,7 @@ Status ClComponentDirectConv2d::validate( // Matching data layout ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, wei); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, bia); } @@ -103,7 +103,7 @@ Status ClComponentDirectConv2d::validate( ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() == 0); ARM_COMPUTE_RETURN_ERROR_ON(wei->tensor_shape().total_size() == 0); ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0); - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON(bia->tensor_shape().total_size() == 0); } @@ -112,22 +112,23 @@ Status ClComponentDirectConv2d::validate( // wei shape is correct const DataLayout data_layout = src->data_layout(); const int channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(wei->dimension(channel_idx) != src->dimension(channel_idx), "Weights feature map dimension should match the respective src's one"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(wei->dimension(channel_idx) != src->dimension(channel_idx), + "Weights feature map dimension should match the respective src's one"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(wei->num_dimensions() > 4, "Weights can be at most 4 dimensional"); // dst shape is correct - PadStrideInfo legacy_pad_stride(attributes.stride().x(), attributes.stride().y(), attributes.pad().left, attributes.pad().right, attributes.pad().top, - attributes.pad().bottom, DimensionRoundingType{}); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), - misc::shape_calculator::compute_deep_convolution_shape(*src, *wei, legacy_pad_stride)); + PadStrideInfo legacy_pad_stride(attributes.stride().x(), attributes.stride().y(), attributes.pad().left, + attributes.pad().right, attributes.pad().top, attributes.pad().bottom, + DimensionRoundingType{}); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS( + dst->tensor_shape(), misc::shape_calculator::compute_deep_convolution_shape(*src, *wei, legacy_pad_stride)); // bia shape is correct - if(bia != nullptr) + if (bia != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->dimension(0) != wei->dimension(3), "Biases size and number of dst feature maps should match"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->num_dimensions() > 1, - "Biases should be one dimensional"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->num_dimensions() > 1, "Biases should be one dimensional"); } // 2. Check support level @@ -137,24 +138,25 @@ Status ClComponentDirectConv2d::validate( ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(src, DataLayout::NHWC); const auto desc = settings.direct_conv_descriptor(); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(desc.n0 != 1 && desc.n0 != 2 && desc.n0 != 3 && desc.n0 != 4 && desc.n0 != 8 && desc.n0 != 16, + ARM_COMPUTE_RETURN_ERROR_ON_MSG(desc.n0 != 1 && desc.n0 != 2 && desc.n0 != 3 && desc.n0 != 4 && desc.n0 != 8 && + desc.n0 != 16, "N0 can only be: 1, 2, 3, 4, 8, and 16"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(desc.k0 != 1 && desc.k0 != 2 && desc.k0 != 3 && desc.k0 != 4 && desc.k0 != 8 && desc.k0 != 16, + ARM_COMPUTE_RETURN_ERROR_ON_MSG(desc.k0 != 1 && desc.k0 != 2 && desc.k0 != 3 && desc.k0 != 4 && desc.k0 != 8 && + desc.k0 != 16, "K0 can only be: 1, 2, 3, 4, 8, and 16"); return Status{}; } -ClComponentDirectConv2d::ClComponentDirectConv2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) - : IGpuKernelComponent{ id, properties, tensors }, +ClComponentDirectConv2d::ClComponentDirectConv2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer{ std::make_unique<ClTemplateDirectConv2d>(id, tensors, attributes, settings) } -#else // ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer{ std::make_unique<GpuCkwDirectConv2d>(id, tensors, attributes, settings) } + _component_writer{std::make_unique<ClTemplateDirectConv2d>(id, tensors, attributes, settings)} +#else // ACL_INTERNAL_TEST_CKW_IN_DF + _component_writer{std::make_unique<GpuCkwDirectConv2d>(id, tensors, attributes, settings)} #endif // ACL_INTERNAL_TEST_CKW_IN_DF { } @@ -165,7 +167,7 @@ ClComponentDirectConv2d::~ClComponentDirectConv2d() #ifndef ACL_INTERNAL_TEST_CKW_IN_DF const IGpuTemplateComponentWriter *ClComponentDirectConv2d::template_writer() const -#else // ACL_INTERNAL_TEST_CKW_IN_DF +#else // ACL_INTERNAL_TEST_CKW_IN_DF const IGpuCkwComponentDriver *ClComponentDirectConv2d::ckw_component_driver() const #endif // ACL_INTERNAL_TEST_CKW_IN_DF { diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.h index 24acb1b2c1..d6d9705d3c 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.h @@ -26,7 +26,9 @@ #include "arm_compute/core/Error.h" #include "arm_compute/core/KernelDescriptors.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" + #include <memory> namespace arm_compute @@ -61,7 +63,7 @@ public: DirectConvComputeKernelInfo direct_conv_descriptor() const; private: - bool _fast_relaxed_math{ true }; + bool _fast_relaxed_math{true}; DirectConvComputeKernelInfo _desc{}; // Direct convolution descriptor }; @@ -111,22 +113,20 @@ public: * |F16 |F16 |F16 |F16 | * |F32 |F32 |F32 |F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + static Status validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Constructor * * Similar to @ref ClComponentDirectConv2d::validate() */ - ClComponentDirectConv2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + ClComponentDirectConv2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Destructor */ ~ClComponentDirectConv2d() override; @@ -142,7 +142,7 @@ public: #ifndef ACL_INTERNAL_TEST_CKW_IN_DF const IGpuTemplateComponentWriter *template_writer() const override; #else // ACL_INTERNAL_TEST_CKW_IN_DF - const IGpuCkwComponentDriver *ckw_component_driver() const override; + const IGpuCkwComponentDriver *ckw_component_driver() const override; #endif // ACL_INTERNAL_TEST_CKW_IN_DF /** Get component type */ GpuComponentType type() const override diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp index 88d729170c..5b136427e4 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp @@ -24,6 +24,7 @@ #include "ClComponentElementwiseBinary.h" #include "arm_compute/core/Validate.h" + #include "src/core/CL/CLValidate.h" #ifndef ACL_INTERNAL_TEST_CKW_IN_DF #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.h" @@ -39,56 +40,55 @@ namespace dynamic_fusion { namespace { -std::set<ElementwiseBinaryCommonAttributes::ElementwiseOp> supported_ops -{ - ElementwiseBinaryCommonAttributes::ElementwiseOp::Add, - ElementwiseBinaryCommonAttributes::ElementwiseOp::Sub, - ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul -}; +std::set<ElementwiseBinaryCommonAttributes::ElementwiseOp> supported_ops{ + ElementwiseBinaryCommonAttributes::ElementwiseOp::Add, ElementwiseBinaryCommonAttributes::ElementwiseOp::Sub, + ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul}; } -Status ClComponentElementwiseBinary::validate(const ArgumentPack<ITensorInfo> &tensors, const ElementwiseBinaryCommonAttributes &attributes) +Status ClComponentElementwiseBinary::validate(const ArgumentPack<ITensorInfo> &tensors, + const ElementwiseBinaryCommonAttributes &attributes) { const auto lhs = tensors.get_const_tensor(TensorType::ACL_SRC_0); const auto rhs = tensors.get_const_tensor(TensorType::ACL_SRC_1); const auto dst = tensors.get_const_tensor(TensorType::ACL_DST_0); // Check operator type - ARM_COMPUTE_RETURN_ERROR_ON_MSG(supported_ops.find(attributes.operation()) == supported_ops.end(), "Provided Elementwise operation not supported."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(supported_ops.find(attributes.operation()) == supported_ops.end(), + "Provided Elementwise operation not supported."); // Check validity ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs, dst); //Check data type for different elementwise operators - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F32, DataType::F16, DataType::S32, DataType::S16, DataType::U8); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F32, DataType::F16, DataType::S32, + DataType::S16, DataType::U8); // dst shape is correct const TensorShape out_shape = TensorShape::broadcast_shape(lhs->tensor_shape(), rhs->tensor_shape()); ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst->tensor_shape(), 0), "Wrong shape for dst."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst->tensor_shape(), 0), + "Wrong shape for dst."); const auto &lhs_shape = lhs->tensor_shape(); const auto &rhs_shape = rhs->tensor_shape(); const auto &dst_shape = dst->tensor_shape(); - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - detail::have_different_dimensions(lhs_shape, dst_shape, 0) && detail::have_different_dimensions(rhs_shape, dst_shape, 0), - "Only LHS or RHS can be broadcasting, not both."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(lhs_shape, dst_shape, 0) && + detail::have_different_dimensions(rhs_shape, dst_shape, 0), + "Only LHS or RHS can be broadcasting, not both."); // Dimension Y and Z are collapsed together in the current kernel implementation, // hence they cannot be independently broadcast or non-broadcast. // See: ClTemplateElementwiseBinary::get_window - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - (lhs_shape[1] != dst_shape[1] || rhs_shape[1] != dst_shape[1]) != (lhs_shape[2] != dst_shape[2] || rhs_shape[2] != dst_shape[2]), - "Dimension Y and Z must both be either broadcast or non-broadcast."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG((lhs_shape[1] != dst_shape[1] || rhs_shape[1] != dst_shape[1]) != + (lhs_shape[2] != dst_shape[2] || rhs_shape[2] != dst_shape[2]), + "Dimension Y and Z must both be either broadcast or non-broadcast."); - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - detail::have_different_dimensions(lhs_shape, dst_shape, 3), - "LHS broadcast in dimension 3 or higher is not supported."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(lhs_shape, dst_shape, 3), + "LHS broadcast in dimension 3 or higher is not supported."); - ARM_COMPUTE_RETURN_ERROR_ON_MSG( - detail::have_different_dimensions(rhs_shape, dst_shape, 3), - "RHS broadcast in dimension 3 or higher is not supported."); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(rhs_shape, dst_shape, 3), + "RHS broadcast in dimension 3 or higher is not supported."); // Matching data type ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lhs, rhs); @@ -112,22 +112,15 @@ Status ClComponentElementwiseBinary::validate(const ArgumentPack<ITensorInfo> &t ClComponentElementwiseBinary::~ClComponentElementwiseBinary() { } -ClComponentElementwiseBinary::ClComponentElementwiseBinary( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes) - : IGpuKernelComponent{ id, properties, tensors }, +ClComponentElementwiseBinary::ClComponentElementwiseBinary(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes) + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<ClTemplateElementwiseBinary>(id, tensors, attributes) -} + _component_writer{std::make_unique<ClTemplateElementwiseBinary>(id, tensors, attributes)} #else //ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<GpuCkwElementwiseBinary>(id, tensors, attributes) -} + _component_writer{std::make_unique<GpuCkwElementwiseBinary>(id, tensors, attributes)} #endif //ACL_INTERNAL_TEST_CKW_IN_DF { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.h index f7175903d0..7589b9732c 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.h @@ -82,17 +82,17 @@ public: * |S16 |S16 |S16 | * |U8 |U8 |U8 | */ - static Status validate(const ArgumentPack<ITensorInfo> &tensors, const ElementwiseBinaryCommonAttributes &attributes); + static Status validate(const ArgumentPack<ITensorInfo> &tensors, + const ElementwiseBinaryCommonAttributes &attributes); /** Constructor * * Similar to @ref ClComponentElementwiseBinary::validate() */ - ClComponentElementwiseBinary( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + ClComponentElementwiseBinary(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes); /** Destructor */ ~ClComponentElementwiseBinary() override; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.cpp index 279c77e227..27c13bd654 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.cpp @@ -25,9 +25,10 @@ #include "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.h" #include "arm_compute/core/CL/CLHelpers.h" -#include "arm_compute/core/Validate.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/core/Validate.h" #include "arm_compute/dynamic_fusion/sketch/attributes/SoftmaxAttributes.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateLogits1DMaxShiftExpSum.h" @@ -37,10 +38,9 @@ namespace experimental { namespace dynamic_fusion { -Status ClComponentLogits1DMaxShiftExpSum::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes) +Status ClComponentLogits1DMaxShiftExpSum::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes) { ARM_COMPUTE_UNUSED(properties, attributes); @@ -75,8 +75,8 @@ ClComponentLogits1DMaxShiftExpSum::ClComponentLogits1DMaxShiftExpSum(ComponentId const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes) - : IGpuKernelComponent{ id, properties, tensors }, - _component_writer{ std::make_unique<ClTemplateLogits1DMaxShiftExpSum>(id, tensors, attributes) } + : IGpuKernelComponent{id, properties, tensors}, + _component_writer{std::make_unique<ClTemplateLogits1DMaxShiftExpSum>(id, tensors, attributes)} { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.h index b5db458248..91ab5de3b5 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DMaxShiftExpSum.h @@ -25,6 +25,7 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTLOGITS1DMAXSHIFTEXPSUM #include "arm_compute/dynamic_fusion/sketch/attributes/SoftmaxAttributes.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute @@ -89,10 +90,8 @@ public: * |F16 | F16 | F16 | * |F32 | F32 | F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + static Status + validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes); /** Constructor * diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp index 7864d56d29..fb2544385c 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp @@ -25,9 +25,10 @@ #include "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h" #include "arm_compute/core/CL/CLHelpers.h" -#include "arm_compute/core/Validate.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/core/Validate.h" #include "arm_compute/dynamic_fusion/sketch/attributes/SoftmaxAttributes.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateLogits1DNorm.h" @@ -37,10 +38,9 @@ namespace experimental { namespace dynamic_fusion { -Status ClComponentLogits1DNorm::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes) +Status ClComponentLogits1DNorm::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes) { ARM_COMPUTE_UNUSED(properties, attributes); @@ -77,8 +77,8 @@ ClComponentLogits1DNorm::ClComponentLogits1DNorm(ComponentId const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes) - : IGpuKernelComponent{ id, properties, tensors }, - _component_writer{ std::make_unique<ClTemplateLogits1DNorm>(id, tensors, attributes) } + : IGpuKernelComponent{id, properties, tensors}, + _component_writer{std::make_unique<ClTemplateLogits1DNorm>(id, tensors, attributes)} { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h index 5bd350b9bd..74c0273604 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h @@ -25,6 +25,7 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTLOGITS1DNORM #include "arm_compute/dynamic_fusion/sketch/attributes/SoftmaxAttributes.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute @@ -86,10 +87,8 @@ public: * |F16 | F16 | F16 | * |F32 | F32 | F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + static Status + validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes); /** Constructor * diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.cpp index d415769094..409b191df5 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.cpp @@ -24,13 +24,15 @@ #include "ClComponentPool2d.h" #include "arm_compute/core/Error.h" -#include "arm_compute/core/Validate.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/core/Validate.h" #include "arm_compute/dynamic_fusion/sketch/attributes/Pool2dAttributes.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwPool2d.h" #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplatePool2d.h" #include "src/dynamic_fusion/utils/Utils.h" + #include <memory> namespace arm_compute @@ -39,23 +41,24 @@ namespace experimental { namespace dynamic_fusion { -Status ClComponentPool2d::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) +Status ClComponentPool2d::validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) { ARM_COMPUTE_UNUSED(properties); const auto src = tensors.get_const_tensor(TensorType::ACL_SRC_0); const auto dst = tensors.get_const_tensor(TensorType::ACL_DST_0); ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst); - ARM_COMPUTE_ERROR_ON_MSG((attributes.pool_type() != PoolingType::AVG && attributes.pool_type() != PoolingType::MAX), "Unsupported Pooling type"); + ARM_COMPUTE_ERROR_ON_MSG((attributes.pool_type() != PoolingType::AVG && attributes.pool_type() != PoolingType::MAX), + "Unsupported Pooling type"); // 1. Check validity // Check if pooling is valid - ARM_COMPUTE_RETURN_ERROR_ON_MSG(is_pool_region_entirely_outside_input(convert_pool_attr_to_pool_info(attributes, settings.mixed_precision())), - "Pooling region that is entirely outside input tensor is unsupported"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG( + is_pool_region_entirely_outside_input(convert_pool_attr_to_pool_info(attributes, settings.mixed_precision())), + "Pooling region that is entirely outside input tensor is unsupported"); // Matching data type ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst); @@ -70,8 +73,9 @@ Status ClComponentPool2d::validate( // Device requirements are met ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), - misc::shape_calculator::compute_pool_shape(*src, convert_pool_attr_to_pool_info(attributes, settings.mixed_precision()))); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS( + dst->tensor_shape(), misc::shape_calculator::compute_pool_shape( + *src, convert_pool_attr_to_pool_info(attributes, settings.mixed_precision()))); // 2. Check support level // Data type @@ -83,23 +87,16 @@ Status ClComponentPool2d::validate( return Status{}; } -ClComponentPool2d::ClComponentPool2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings) - : IGpuKernelComponent{ id, properties, tensors }, +ClComponentPool2d::ClComponentPool2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings) + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<ClTemplatePool2d>(id, tensors, attributes, settings) -} + _component_writer{std::make_unique<ClTemplatePool2d>(id, tensors, attributes, settings)} #else //ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<GpuCkwPool2d>(id, tensors, attributes, settings) -} + _component_writer{std::make_unique<GpuCkwPool2d>(id, tensors, attributes, settings)} #endif //ACL_INTERNAL_TEST_CKW_IN_DF { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.h index 6814bf9243..98fed65004 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentPool2d.h @@ -25,6 +25,7 @@ #define ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTPOOL2D_H #include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuPool2d.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute @@ -82,11 +83,10 @@ public: * |F16 |F16 | * |F32 |F32 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + static Status validate(const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Constructor * @@ -96,12 +96,11 @@ public: * @param[in] attributes Component attributes * @param[in] settings Component settings */ - ClComponentPool2d( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes, - const Settings &settings); + ClComponentPool2d(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes, + const Settings &settings); /** Destructor */ ~ClComponentPool2d() override; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.cpp index 66e2ee6956..0ece9de970 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.cpp @@ -22,8 +22,10 @@ * SOFTWARE. */ #include "ClComponentReshape.h" + #include "arm_compute/core/Error.h" #include "arm_compute/core/Validate.h" + #include "src/core/CL/CLValidate.h" #include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateReshape.h" @@ -49,12 +51,10 @@ Status ClComponentReshape::validate(const ArgumentPack<ITensorInfo> &tensors) return Status{}; } -ClComponentReshape::ClComponentReshape( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors) - : IGpuKernelComponent{ id, properties, tensors }, - _component_writer{ std::make_unique<ClTemplateReshape>(id, tensors) } +ClComponentReshape::ClComponentReshape(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors) + : IGpuKernelComponent{id, properties, tensors}, _component_writer{std::make_unique<ClTemplateReshape>(id, tensors)} { } ClComponentReshape::~ClComponentReshape() diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.h index f8d165b4c8..78163d6603 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentReshape.h @@ -73,10 +73,7 @@ public: * @param[in] properties Component properties @ref Properties * @param[in] tensors Tensor arguments to the component */ - ClComponentReshape( - ComponentId id, - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors); + ClComponentReshape(ComponentId id, const Properties &properties, const ArgumentPack<ITensorInfo> &tensors); /** Destructor */ ~ClComponentReshape() override; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.cpp index 6df1d9b3db..b05eb04698 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.cpp @@ -66,7 +66,9 @@ Status ClComponentResize::validate(const IGpuKernelComponent::Properties &proper ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src); // Align corners and sampling policy conformance - ARM_COMPUTE_RETURN_ERROR_ON(attributes.align_corners() && !arm_compute::scale_utils::is_align_corners_allowed_sampling_policy(attributes.sampling_policy())); + ARM_COMPUTE_RETURN_ERROR_ON( + attributes.align_corners() && + !arm_compute::scale_utils::is_align_corners_allowed_sampling_policy(attributes.sampling_policy())); // All tensor infos are initialized ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() == 0); @@ -79,11 +81,11 @@ ClComponentResize::ClComponentResize(ComponentId id, const IGpuKernelComponent::Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const ClComponentResize::Attributes &attributes) - : IGpuKernelComponent{ id, properties, tensors }, + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer{ std::make_unique<ClTemplateResize>(id, tensors, attributes) } -#else // ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer{ std::make_unique<GpuCkwResize>(id, tensors, attributes) } + _component_writer{std::make_unique<ClTemplateResize>(id, tensors, attributes)} +#else // ACL_INTERNAL_TEST_CKW_IN_DF + _component_writer{std::make_unique<GpuCkwResize>(id, tensors, attributes)} #endif // ACL_INTERNAL_TEST_CKW_IN_DF { } @@ -94,7 +96,7 @@ ClComponentResize::~ClComponentResize() #ifndef ACL_INTERNAL_TEST_CKW_IN_DF const IGpuTemplateComponentWriter *ClComponentResize::template_writer() const -#else // ACL_INTERNAL_TEST_CKW_IN_DF +#else // ACL_INTERNAL_TEST_CKW_IN_DF const IGpuCkwComponentDriver *ClComponentResize::ckw_component_driver() const #endif // ACL_INTERNAL_TEST_CKW_IN_DF { diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.h index 474524f8fc..29276c3257 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentResize.h @@ -26,6 +26,7 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTRESIZE #include "arm_compute/dynamic_fusion/sketch/attributes/ResizeAttributes.h" + #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" namespace arm_compute @@ -43,7 +44,7 @@ class ArgumentPack; /** Forward declaration */ #ifndef ACL_INTERNAL_TEST_CKW_IN_DF class ClTemplateResize; -#else // ACL_INTERNAL_TEST_CKW_IN_DF +#else // ACL_INTERNAL_TEST_CKW_IN_DF class GpuCkwResize; #endif // ACL_INTERNAL_TEST_CKW_IN_DF @@ -82,10 +83,8 @@ public: * |U8 |U8 | * |S16 |S16 | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors, - const Attributes &attributes); + static Status + validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors, const Attributes &attributes); /** Constructor * @@ -114,7 +113,7 @@ public: /** Get writer for the component */ #ifndef ACL_INTERNAL_TEST_CKW_IN_DF const IGpuTemplateComponentWriter *template_writer() const override; -#else // ACL_INTERNAL_TEST_CKW_IN_DF +#else // ACL_INTERNAL_TEST_CKW_IN_DF const IGpuCkwComponentDriver *ckw_component_driver() const override; #endif // ACL_INTERNAL_TEST_CKW_IN_DF @@ -127,7 +126,7 @@ public: private: #ifndef ACL_INTERNAL_TEST_CKW_IN_DF std::unique_ptr<ClTemplateResize> _component_writer; -#else // ACL_INTERNAL_TEST_CKW_IN_DF +#else // ACL_INTERNAL_TEST_CKW_IN_DF std::unique_ptr<GpuCkwResize> _component_writer; #endif // ACL_INTERNAL_TEST_CKW_IN_DF }; diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.cpp index 12b81c3d56..dcbecaff35 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.cpp +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.cpp @@ -38,25 +38,19 @@ namespace experimental { namespace dynamic_fusion { -Status ClComponentStore::validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors) +Status ClComponentStore::validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors) { ARM_COMPUTE_UNUSED(properties, tensors); return Status{}; } -ClComponentStore::ClComponentStore(ComponentId id, const Properties &properties, const ArgumentPack<ITensorInfo> &tensors) - : IGpuKernelComponent{ id, properties, tensors }, +ClComponentStore::ClComponentStore(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors) + : IGpuKernelComponent{id, properties, tensors}, #ifndef ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<ClTemplateStore>(id, tensors) -} + _component_writer{std::make_unique<ClTemplateStore>(id, tensors)} #else //ACL_INTERNAL_TEST_CKW_IN_DF - _component_writer -{ - std::make_unique<GpuCkwStore>(id, tensors) -} + _component_writer{std::make_unique<GpuCkwStore>(id, tensors)} #endif //ACL_INTERNAL_TEST_CKW_IN_DF { } diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.h index 853ee39012..948785c480 100644 --- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.h +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.h @@ -25,6 +25,7 @@ #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTSTORE #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" + #include <memory> namespace arm_compute @@ -70,9 +71,7 @@ public: * |:--------------|:--------------| * |All |All | */ - static Status validate( - const Properties &properties, - const ArgumentPack<ITensorInfo> &tensors); + static Status validate(const Properties &properties, const ArgumentPack<ITensorInfo> &tensors); /** Constructor * * Similar to @ref ClComponentStore::validate() diff --git a/src/dynamic_fusion/sketch/gpu/components/utils/type_printer/ElementwiseBinary.h b/src/dynamic_fusion/sketch/gpu/components/utils/type_printer/ElementwiseBinary.h index bc7133f4df..4c3e84e59d 100644 --- a/src/dynamic_fusion/sketch/gpu/components/utils/type_printer/ElementwiseBinary.h +++ b/src/dynamic_fusion/sketch/gpu/components/utils/type_printer/ElementwiseBinary.h @@ -46,18 +46,16 @@ using namespace experimental::dynamic_fusion; */ inline ::std::ostream &operator<<(::std::ostream &os, const ClComponentElementwiseBinary::Attributes::ElementwiseOp &op) { - const std::map<ClComponentElementwiseBinary::Attributes::ElementwiseOp, std::string> op_name = - { - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Add, "add" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Div, "div" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Max, "max" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Min, "min" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Mul, "mul" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Power, "power" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Prelu, "prelu" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::SquaredDiff, "squareddiff" }, - { ClComponentElementwiseBinary::Attributes::ElementwiseOp::Sub, "sub" } - }; + const std::map<ClComponentElementwiseBinary::Attributes::ElementwiseOp, std::string> op_name = { + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Add, "add"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Div, "div"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Max, "max"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Min, "min"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Mul, "mul"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Power, "power"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Prelu, "prelu"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::SquaredDiff, "squareddiff"}, + {ClComponentElementwiseBinary::Attributes::ElementwiseOp::Sub, "sub"}}; os << op_name.at(op); return os; } |