CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput > Class Template Reference

This class is a wrapper for the assembly kernels. More...

#include <CpuGemmAssemblyWrapperKernel.h>

Collaboration diagram for CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput >:

Public Member Functions
	CpuGemmAssemblyWrapperKernel ()
	Constructor. More...
	CpuGemmAssemblyWrapperKernel (CpuGemmAssemblyWrapperKernel &)=delete
	CpuGemmAssemblyWrapperKernel (CpuGemmAssemblyWrapperKernel &&)=default
CpuGemmAssemblyWrapperKernel &	operator= (CpuGemmAssemblyWrapperKernel &)=delete
const char *	name () const override
	Name of the kernel. More...
void	run (const Window &window, const ThreadInfo &info) override
	Execute the kernel on the passed window. More...
void	run_nd (const Window &window, const ThreadInfo &info, const Window &thread_locator) override
	legacy compatibility layer for implemantions which do not support thread_locator In these cases we simply narrow the interface down the legacy version More...
void	configure (arm_gemm::GemmCommon< TypeInput, TypeOutput > *kernel, std::string kernel_name_tag)
	Initialise the kernel's input and output. More...
size_t	get_mws (const CPUInfo &platform, size_t thread_count) const override
	Return minimum workload size of the relevant kernel. More...
Public Member Functions inherited from ICPPKernel
virtual	~ICPPKernel ()=default
	Default destructor. More...
virtual void	run_op (ITensorPack &tensors, const Window &window, const ThreadInfo &info)
	Execute the kernel on the passed window. More...
Public Member Functions inherited from IKernel
	IKernel ()
	Constructor. More...
virtual	~IKernel ()=default
	Destructor. More...
virtual bool	is_parallelisable () const
	Indicates whether or not the kernel is parallelisable. More...
virtual BorderSize	border_size () const
	The size of the border for that kernel. More...
const Window &	window () const
	The maximum window the kernel can be executed on. More...
bool	is_window_configured () const
	Function to check if the embedded window of this kernel has been configured. More...

Additional Inherited Members
Static Public Attributes inherited from ICPPKernel
static constexpr size_t	default_mws = 1

Detailed Description

template<typename TypeInput, typename TypeOutput>
class arm_compute::cpu::kernel::CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput >

This class is a wrapper for the assembly kernels.

Some kernels were written in assembly and highly optimised for specific CPUs like A53 or A55. This class works as a wrapper for these assembly kernels. The arm compute library creates an instance of CpuGemmAssemblyWrapperKernel and other auxiliary data structures to execute a single assembly kernel in the context of an NEFunctions.

The type T is the type of the actual kernel implemented in assembly which is of type template<typename To, typename Tr> class GemmCommon

Definition at line 55 of file CpuGemmAssemblyWrapperKernel.h.

Constructor & Destructor Documentation

CpuGemmAssemblyWrapperKernel() [1/3]

CpuGemmAssemblyWrapperKernel ( )

inline

Constructor.

Definition at line 60 of file CpuGemmAssemblyWrapperKernel.h.

        : _kernel(nullptr), _name("CpuGemmAssemblyWrapperKernel")
    {
    }

CpuGemmAssemblyWrapperKernel() [2/3]

CpuGemmAssemblyWrapperKernel ( CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput > &  )

delete

CpuGemmAssemblyWrapperKernel() [3/3]

CpuGemmAssemblyWrapperKernel ( CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput > &&  )

default

Member Function Documentation

configure()

void configure ( arm_gemm::GemmCommon< TypeInput, TypeOutput > *  kernel, std::string  kernel_name_tag  )

inline

Initialise the kernel's input and output.

Parameters

[in]	kernel	Pointer to an assembly kernel implementation.
[in]	kernel_name_tag	Tag to be attacehd to the kernel's name.

Definition at line 104 of file CpuGemmAssemblyWrapperKernel.h.

    {
        ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(kernel)));
        _kernel = kernel;
 
        Window win = to_window(kernel->get_window_size());
 
        INEKernel::configure(win);
 
        if(!kernel_name_tag.empty())
        {
            _name += "/" + kernel_name_tag;
        }
    }

References ARM_COMPUTE_ERROR_ON_NULLPTR, IGemmCommon::get_window_size(), and arm_gemm::to_window().

get_mws()

size_t get_mws ( const CPUInfo &  platform, size_t  thread_count  ) const

inlineoverridevirtual

Return minimum workload size of the relevant kernel.

Parameters

[in]	platform	The CPU platform used to create the context.
[in]	thread_count	Number of threads in the execution.

Returns

[out] small_network_mws Minimum workload size for requested configuration.

Reimplemented from ICPPKernel.

Definition at line 125 of file CpuGemmAssemblyWrapperKernel.h.

    {
        ARM_COMPUTE_UNUSED(thread_count);
        ARM_COMPUTE_UNUSED(platform);
 
        return ICPPKernel::default_mws;
    }

References ARM_COMPUTE_UNUSED, and ICPPKernel::default_mws.

name()

const char* name ( ) const

inlineoverridevirtual

Name of the kernel.

Returns

Kernel name

Implements ICPPKernel.

Definition at line 69 of file CpuGemmAssemblyWrapperKernel.h.

    {
        return _name.c_str();
    }

operator=()

CpuGemmAssemblyWrapperKernel& operator= ( CpuGemmAssemblyWrapperKernel< TypeInput, TypeOutput > &  )

delete

run()

void run ( const Window &  window, const ThreadInfo &  info  )

inlineoverridevirtual

Execute the kernel on the passed window.

Warning

If is_parallelisable() returns false then the passed window must be equal to window()

Note

The window has to be a region within the window returned by the window() method

The width of the window has to be a multiple of num_elems_processed_per_iteration().

Parameters

[in]	window	Region on which to execute the kernel. (Must be a region of the window returned by window())
[in]	info	Info about executing thread and CPU.

Reimplemented from ICPPKernel.

Definition at line 74 of file CpuGemmAssemblyWrapperKernel.h.

    {
        ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(_kernel)));
        ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
 
        auto win = arm_gemm::to_ndcoord(window);
 
        arm_gemm::ndcoord_t thread_locator{};
 
        _kernel->execute(win, thread_locator, info.thread_id);
    }

References ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL, IGemmCommon::execute(), arm_compute::test::validation::info, arm_gemm::to_ndcoord(), and IKernel::window().

run_nd()

void run_nd ( const Window &  window, const ThreadInfo &  info, const Window &  thread_locator  )

inlineoverridevirtual

legacy compatibility layer for implemantions which do not support thread_locator In these cases we simply narrow the interface down the legacy version

Parameters

[in]	window	Region on which to execute the kernel. (Must be a region of the window returned by window())
[in]	info	Info about executing thread and CPU.
[in]	thread_locator	Specifies "where" the current thread is in the multi-dimensional space

Reimplemented from ICPPKernel.

Definition at line 87 of file CpuGemmAssemblyWrapperKernel.h.

    {
        ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(_kernel)));
        ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
 
        //convert between arm_compute and arm_gemm types
        auto ndc_win = arm_gemm::to_ndcoord(window);
        auto ndc_tlc = arm_gemm::to_ndcoord(thread_locator);
 
        _kernel->execute(ndc_win, ndc_tlc, info.thread_id);
    }

References ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL, IGemmCommon::execute(), arm_compute::test::validation::info, arm_gemm::to_ndcoord(), and IKernel::window().

---

The documentation for this class was generated from the following file:

• src/cpu/kernels/assembly/CpuGemmAssemblyWrapperKernel.h