diff options
Diffstat (limited to 'arm_compute/core/CL/CLHelpers.h')
| -rw-r--r-- | arm_compute/core/CL/CLHelpers.h | 98 |
1 files changed, 76 insertions, 22 deletions
diff --git a/arm_compute/core/CL/CLHelpers.h b/arm_compute/core/CL/CLHelpers.h index fc3f4d5db0..1a639e47f9 100644 --- a/arm_compute/core/CL/CLHelpers.h +++ b/arm_compute/core/CL/CLHelpers.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2020 ARM Limited. + * Copyright (c) 2016-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -26,13 +26,13 @@ #include "arm_compute/core/CL/CLTypes.h" #include "arm_compute/core/CL/OpenCL.h" +#include "arm_compute/core/Types.h" #include <set> #include <string> namespace arm_compute { -class CLCoreRuntimeContext; class CLCompileContext; class CLBuildOptions; @@ -41,6 +41,9 @@ enum class DataType; /** Max vector width of an OpenCL vector */ static constexpr unsigned int max_cl_vector_width = 16; +/** Max number of manual loop unrolling */ +static constexpr int max_manual_loop_unrolling = 128; + /** Translates a tensor data type to the appropriate OpenCL type. * * @param[in] dt @ref DataType to be translated to OpenCL type. @@ -97,14 +100,6 @@ std::string get_cl_dot8_acc_type_from_data_type(const DataType &dt); */ std::string get_data_size_from_data_type(const DataType &dt); -/** Translates fixed point tensor data type to the underlying OpenCL type. - * - * @param[in] dt @ref DataType to be translated to OpenCL type. - * - * @return The string specifying the underlying OpenCL type to be used. - */ -std::string get_underlying_cl_type_from_data_type(const DataType &dt); - /** Helper function to get the GPU target from CL device * * @param[in] device A CL device @@ -129,6 +124,14 @@ CLVersion get_cl_version(const cl::Device &device); */ size_t get_cl_image_pitch_alignment(const cl::Device &device); +/** Helper function to check whether non-uniform work group is supported + * + * @param[in] device A CL device + * + * @return True if the feature is supported + */ +bool get_cl_non_uniform_work_group_supported(const cl::Device &device); + /** Helper function to check whether a given extension is supported * * @param[in] device A CL device @@ -176,7 +179,9 @@ bool dot8_acc_supported(const cl::Device &device); * * @return True if the configuration is supported */ -bool cl_winograd_convolution_layer_supported(const Size2D &output_tile, const Size2D &kernel_size, DataLayout data_layout); +bool cl_winograd_convolution_layer_supported(const Size2D &output_tile, + const Size2D &kernel_size, + DataLayout data_layout); /** Helper function to get the preferred native vector width size for built-in scalar types that can be put into vectors * @@ -204,16 +209,6 @@ bool preferred_dummy_work_items_support(const cl::Device &device); */ bool image2d_from_buffer_supported(const cl::Device &device); -/** Creates an opencl kernel - * - * @param[in] ctx A context to be used to create the opencl kernel. - * @param[in] kernel_name The kernel name. - * @param[in] build_opts The build options to be used for the opencl kernel compilation. - * - * @return An opencl kernel - */ -cl::Kernel create_opencl_kernel(CLCoreRuntimeContext *ctx, const std::string &kernel_name, const CLBuildOptions &build_opts); - /** Creates an opencl kernel using a compile context * * @param[in] ctx A compile context to be used to create the opencl kernel. @@ -222,7 +217,9 @@ cl::Kernel create_opencl_kernel(CLCoreRuntimeContext *ctx, const std::string &ke * * @return An opencl kernel */ -cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set<std::string> &build_opts = std::set<std::string>()); +cl::Kernel create_kernel(const CLCompileContext &ctx, + const std::string &kernel_name, + const std::set<std::string> &build_opts = std::set<std::string>()); /** Creates a suitable LWS hint object for parallel implementations. Sets the number of WG based on the input size. * If input width is smaller than 128 we can use fewer threads than 8. @@ -234,5 +231,62 @@ cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_ */ cl::NDRange create_lws_hint_parallel_implementations(unsigned int input_dimension, unsigned int vector_size); +/* Helper function to check if the workgroup batch size modifier parameter is supported on the cl device + * + * @param[in] device cl device to check for support + * + * @return true if the workgroup batch size modifier parameter is supported, false otherwise + */ +bool get_wbsm_support_info(const cl::Device &device); + +/* Helper function to set the workgroup batch size modifier parameter in the kernel + * + * @param[in] kernel cl kernel to set the workgroup batch size modifier parameter + * @param[in] wbsm_hint workgroup batch size modifier to use + */ +void set_wbsm(cl::Kernel &kernel, cl_int wbsm_hint); + +/* Helper function to check if we can export the tensor to cl_image + * + * @param[in] input tensor + * + * @return true if we can export the tensor to cl_image + */ +bool export_to_cl_image(const ITensorInfo *tensor); + +/* Helper function to force unroll with pragma when any of the input values (iterations) are greater than @ref max_manual_loop_unrolling + * + * This function passes UNROLL_WITH_PRAGMA at compile time when any of the input values are greater than @ref max_manual_loop_unrolling + * + * @param[in] built_opts OpenCL kernel build options + * @param[in] values Input values (iterations) + * + */ +void set_unroll_with_pragma(CLBuildOptions &built_opts, std::initializer_list<int> values); + +/** Helper function to check whether the cl_arm_matrix_multiply extension is supported + * + * @param[in] device A CL device + * + * @return True if the extension is supported + */ +bool arm_matrix_multiply_supported(const cl::Device &device); + +/** Check whether cl_khr_command_buffer extension is supported by the specified CL device. + * + * @param[in] device The CL device + * + * @return True if the extension is supported by the CL device. + */ +bool command_buffer_supported(const cl::Device &device); + +/** Check whether cl_khr_command_buffer_mutable_dispatch extension is supported by the specified CL device. + * + * @param[in] device The CL device + * + * @return True if the extension is supported by the CL device. + */ +bool command_buffer_mutable_dispatch_supported(const cl::Device &device); + } // namespace arm_compute #endif /* ARM_COMPUTE_CLHELPERS_H */ |