OpenCL kernel to multiply matrices when only the input matrix RHS (src1) has been reshaped. More...

#include <ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h>

Collaboration diagram for ClGemmMatrixMultiplyReshapedOnlyRhsKernel:

Public Member Functions
	ClGemmMatrixMultiplyReshapedOnlyRhsKernel ()

	ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE (ClGemmMatrixMultiplyReshapedOnlyRhsKernel)

void	configure (const ClCompileContext &compile_context, const ITensorInfo src0, const ITensorInfo src1, const ITensorInfo src2, ITensorInfo dst, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
	Initialise the kernel's input and output. More...

void	run_op (ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override
	Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue. More...

Public Member Functions inherited from ICLKernel
	ICLKernel ()
	Constructor. More...

cl::Kernel &	kernel ()
	Returns a reference to the OpenCL kernel of this object. More...

CLKernelType	type () const
	Returns the CL kernel type. More...

template<typename T >
void	add_1D_array_argument (unsigned int &idx, const ICLArray< T > *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
	Add the passed 1D array's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_1D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_1D_tensor_argument_if (bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx if the condition is true. More...

void	add_2D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_2D_tensor_argument_if (bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx if the condition is true. More...

void	add_3D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 3D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_4D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 4D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_5D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 5D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_3d_tensor_nhw_argument (unsigned int &idx, const ICLTensor *tensor)
	Add the passed NHW 3D tensor's parameters to the object's kernel's arguments by passing strides, dimensions and the offset to the first valid element in bytes. More...

void	add_4d_tensor_nhwc_argument (unsigned int &idx, const ICLTensor *tensor)
	Add the passed NHWC 4D tensor's parameters to the object's kernel's arguments by passing strides, dimensions and the offset to the first valid element in bytes. More...

virtual void	run (const Window &window, cl::CommandQueue &queue)
	Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue. More...

virtual void	run_composite_op (ITensorPack &tensors, const Window &window, cl::CommandQueue &queue, const experimental::dynamic_fusion::ClExecutionDescriptor &exec_desc)
	The execution is carried out through run_op method. But the run_op method needs to be extended to include ClExecutionDescriptor as now LWS GWS tuning will be separated from the IKernel. More...

template<typename T >
void	add_argument (unsigned int &idx, T value)
	Add the passed parameters to the object's kernel's arguments starting from the index idx. More...

void	set_lws_hint (const cl::NDRange &lws_hint)
	Set the Local-Workgroup-Size hint. More...

cl::NDRange	lws_hint () const
	Return the Local-Workgroup-Size hint. More...

void	set_wbsm_hint (const cl_int &wbsm_hint)
	Set the workgroup batch size modifier hint. More...

cl_int	wbsm_hint () const
	Return the workgroup batch size modifier hint. More...

const std::string &	config_id () const
	Get the configuration ID. More...

void	set_target (GPUTarget target)
	Set the targeted GPU architecture. More...

void	set_target (cl::Device &device)
	Set the targeted GPU architecture according to the CL device. More...

GPUTarget	get_target () const
	Get the targeted GPU architecture. More...

size_t	get_max_workgroup_size ()
	Get the maximum workgroup size for the device the CLKernelLibrary uses. More...

template<unsigned int dimension_size>
void	add_tensor_argument (unsigned &idx, const ICLTensor *tensor, const Window &window)

template<typename T , unsigned int dimension_size>
void	add_array_argument (unsigned &idx, const ICLArray< T > *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
	Add the passed array's parameters to the object's kernel's arguments starting from the index idx. 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...

Static Public Member Functions
static Status	validate (const ITensorInfo src0, const ITensorInfo src1, const ITensorInfo src2, const ITensorInfo dst, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
	Static function to check if given info will lead to a valid configuration. More...

Static Public Member Functions inherited from ICLKernel
static constexpr unsigned int	num_arguments_per_3d_tensor_nhw ()
	Returns the number of arguments enqueued per NHW 3D Tensor object. More...

static constexpr unsigned int	num_arguments_per_4d_tensor_nhwc ()
	Returns the number of arguments enqueued per NHWC 4D Tensor object. More...

static constexpr unsigned int	num_arguments_per_1D_array ()
	Returns the number of arguments enqueued per 1D array object. More...

static constexpr unsigned int	num_arguments_per_1D_tensor ()
	Returns the number of arguments enqueued per 1D tensor object. More...

static constexpr unsigned int	num_arguments_per_2D_tensor ()
	Returns the number of arguments enqueued per 2D tensor object. More...

static constexpr unsigned int	num_arguments_per_3D_tensor ()
	Returns the number of arguments enqueued per 3D tensor object. More...

static constexpr unsigned int	num_arguments_per_4D_tensor ()
	Returns the number of arguments enqueued per 4D tensor object. More...

static cl::NDRange	gws_from_window (const Window &window)
	Get the global work size given an execution window. More...

Detailed Description

OpenCL kernel to multiply matrices when only the input matrix RHS (src1) has been reshaped.

Note: The input matrix src1 must be reshaped through ClGemmReshapeRhsMatrixKernel

Definition at line 43 of file ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h.

Constructor & Destructor Documentation

◆ ClGemmMatrixMultiplyReshapedOnlyRhsKernel()

ClGemmMatrixMultiplyReshapedOnlyRhsKernel ( )

Definition at line 182 of file ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp.

References arm_compute::GEMM.

 {
     _type = CLKernelType::GEMM;
 }

Member Function Documentation

◆ ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE()

ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE ( ClGemmMatrixMultiplyReshapedOnlyRhsKernel )

◆ configure()

void configure	(	const ClCompileContext &	compile_context,
		const ITensorInfo *	src0,
		const ITensorInfo *	src1,
		const ITensorInfo *	src2,
		ITensorInfo *	dst,
		float	alpha,
		float	beta,
		const GEMMLHSMatrixInfo &	lhs_info,
		const GEMMRHSMatrixInfo &	rhs_info,
		const GEMMKernelInfo &	gemm_info
	)

Initialise the kernel's input and output.

Note

If rhs_info.export_to_cl_image = true, this OpenCL kernel will fetch the RHS data using the OpenCL read_image built-in function. Reading from the OpenCL image object can increase the performance. However, since the OpenCL image object is created importing the OpenCL buffer, the following conditions are required:

rhs_info.n0 can only be 4, 8 and 16
rhs_info.k0 can only be 4, 8 and 16
Data type can only be F32
The platform should support the OpenCL cl_khr_image2d_from_buffer extension
The stride Y for the src1 should satisfy the OpenCL pitch alignment requirement
src1 width should be less or equal to (CL_DEVICE_IMAGE2D_MAX_WIDTH * 4)
src1 (height * depth) should be less or equal to CL_DEVICE_IMAGE2D_MAX_HEIGHT

Parameters

[in]	compile_context	The compile context to be used.
[in]	src0	Input tensor containing the LHS matrix. Data type supported: F16/F32 (only F32 if rhs_info.export_to_cl_image = true). The number of dimensions for the LHS matrix must be less or equal than 4.
[in]	src1	Input tensor containing the RHS reshaped matrix. Data type supported: same as `src0`. The number of dimensions for the RHS matrix must be less or equal than 3.
[in]	src2	Input tensor containing the bias matrix. Data type supported: same as `src0`.
[out]	dst	Output tensor to store the result of matrix multiplication. Data type supported: same as `src0`
[in]	alpha	Weight of the matrix product
[in]	beta	Weight of the matrix bias
[in]	lhs_info	LHS matrix information used to retrieve the number of rows to be processed by each thread. Only the following values are supported: lhs_info.m0: 1,2,3,4,5,6,7,8
[in]	rhs_info	RHS matrix information used for reshaping the src1 tensor. Only the following values are supported: rhs_info.k0: 2,3,4,8,16 rhs_info.n0: 2,3,4,8,16 rhs_info.transpose: true,false
[in]	gemm_info	GEMM information used to retrieve the original dimensions of the input matrices

Definition at line 187 of file ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp.

References CLBuildOptions::add_option(), CLBuildOptions::add_option_if(), ARM_COMPUTE_ERROR_ON, ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_THROW_ON, arm_compute::auto_init_if_empty(), ICloneable< T >::clone(), arm_compute::misc::shape_calculator::compute_mm_shape(), arm_compute::create_kernel(), ITensorInfo::data_type(), GEMMKernelInfo::depth_output_gemm3d, ITensorInfo::dimension(), GEMMRHSMatrixInfo::export_to_cl_image, arm_compute::float_to_string_with_full_precision(), arm_compute::test::validation::gemm_info, CLKernelLibrary::get(), arm_compute::get_cl_type_from_data_type(), arm_compute::get_padding_info(), GEMMKernelInfo::has_pad_y, arm_compute::has_padding_changed(), arm_compute::helpers::float_ops::is_one(), kernel_name, arm_compute::test::validation::lhs_info, arm_compute::lower_string(), GEMMLHSMatrixInfo::m0, ITensorInfo::num_dimensions(), CLBuildOptions::options(), GEMMKernelInfo::post_ops, arm_compute::preferred_dummy_work_items_support(), GEMMKernelInfo::reinterpret_input_as_3d, arm_compute::test::validation::rhs_info, arm_compute::string_from_activation_func(), arm_compute::string_from_data_type(), arm_compute::support::cpp11::to_string(), arm_compute::upper_string(), arm_compute::cpu::kernels::validate_and_configure_window(), and arm_compute::cpu::kernels::validate_arguments().

 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src0, src1, dst);
 
     // dst tensor auto initialization if not yet initialized
     auto_init_if_empty(*dst, src0->clone()->set_tensor_shape(misc::shape_calculator::compute_mm_shape(*src0, *src1, gemm_info)));
 
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src0, src1, src2, dst, alpha, beta, lhs_info, rhs_info, gemm_info));
 
     _reinterpret_input_as_3d  = gemm_info.reinterpret_input_as_3d;
     _reinterpret_output_as_3d = gemm_info.depth_output_gemm3d != 0;
     _use_dummy_work_items     = preferred_dummy_work_items_support(CLKernelLibrary::get().get_device());
     _add_bias                 = src2 != nullptr;
     _export_to_cl_image       = rhs_info.export_to_cl_image;
     _has_pad_y                = gemm_info.has_pad_y;
     _num_post_op_args         = gemm_info.post_ops.total_num_arguments();
 
     auto padding_info = get_padding_info({ src0, src1, src2, dst });
 
     // In case both input and dst have to be reinterpreted as 3D tensors,
     // force reinterpret_input_as_3d and reinterpret_output_as_3d to be false.
     if((_reinterpret_input_as_3d == _reinterpret_output_as_3d) && _has_pad_y)
     {
         _reinterpret_input_as_3d  = false;
         _reinterpret_output_as_3d = false;
     }
 
     // Check if we need to slide the matrix B
     const unsigned int num_dimensions_src0 = src0->num_dimensions();
     _slide_matrix_b                        = (src1->num_dimensions() >= num_dimensions_src0);
 
     ElementsProcessed num_elements_processed{};
 
     // Configure kernel window
     Window win = validate_and_configure_window(src0->clone().get(), src1->clone().get(), (src2 != nullptr) ? src2->clone().get() : nullptr, dst->clone().get(), lhs_info, rhs_info, gemm_info,
                                                num_elements_processed);
     ICLKernel::configure_internal(win);
 
     // If _reinterpret_input_as_3d = reinterpret_output_as_3d = true,
     // we will dispatch a batched-GEMM to reduce the complexity of the address calculation within the OpenCL kernel.
     // This means that the actual m used by the kernel is given by dst->dimension(1) and not by gemm_info.m
     const unsigned int internal_m = _reinterpret_output_as_3d ? gemm_info.m : dst->dimension(1);
 
     // These variables are used only if gemm_info.has_pad_y == true
     const unsigned int h_gemm_3d = _reinterpret_output_as_3d ? dst->dimension(1) : src0->dimension(1);
     const unsigned int d_gemm_3d = _reinterpret_output_as_3d ? dst->dimension(2) : src0->dimension(2);
 
     // Shrink M0 to be always <= M (internal_m) to prevent out-of-bounds reads.
     // NOTE: This might have implications on heuristics and performance
     const unsigned int internal_m0 = std::min(internal_m, lhs_info.m0);
 
     // Calculate partial (store instead of load) M0 and partial N0 for the partial blocks at the end of a row/column if any. This is to avoid padding.
     const unsigned int partial_store_m0 = internal_m % internal_m0;
     const unsigned int partial_store_n0 = gemm_info.n % rhs_info.n0;
     _m                                  = internal_m;
     _n                                  = gemm_info.n;
     _k                                  = gemm_info.k;
     // Create build options
     CLBuildOptions build_opts;
     build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src0->data_type()));
     build_opts.add_option_if(!(helpers::float_ops::is_one(alpha)), "-DALPHA=" + float_to_string_with_full_precision(alpha));
     build_opts.add_option_if(src2 != nullptr, "-DBETA=" + float_to_string_with_full_precision(beta));
     build_opts.add_option_if(helpers::float_ops::is_one(beta), "-DUNIT_BETA");
     build_opts.add_option_if(gemm_info.broadcast_bias, "-DBROADCAST_BIAS");
     build_opts.add_option_if(!_slide_matrix_b, "-DMATRIX_B_DEPTH=" + support::cpp11::to_string(src1->dimension(2)));
     build_opts.add_option_if(rhs_info.interleave, "-DRHS_INTERLEAVE");
     build_opts.add_option_if(_use_dummy_work_items, "-DDUMMY_WORK_ITEMS");
     build_opts.add_option_if(rhs_info.export_to_cl_image, "-DOPENCL_IMAGE_SUPPORT");
     build_opts.add_option("-DRHS_HEIGHT=" + support::cpp11::to_string(src1->dimension(1)));
     build_opts.add_option("-DM0=" + support::cpp11::to_string(internal_m0));
     build_opts.add_option("-DN0=" + support::cpp11::to_string(rhs_info.n0));
     build_opts.add_option("-DK0=" + support::cpp11::to_string(rhs_info.k0));
     build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0));
     build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string(partial_store_m0));
     build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(partial_store_n0));
     if(_has_pad_y)
     {
         build_opts.add_option_if(_reinterpret_input_as_3d, "-DREINTERPRET_INPUT_AS_3D");
         build_opts.add_option_if(_reinterpret_output_as_3d, "-DREINTERPRET_OUTPUT_AS_3D");
         build_opts.add_option_if(_reinterpret_input_as_3d || _reinterpret_output_as_3d, "-DHEIGHT_GEMM3D=" + support::cpp11::to_string(h_gemm_3d));
         build_opts.add_option_if(_reinterpret_input_as_3d || _reinterpret_output_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(d_gemm_3d));
     }
     // If post_ops are used, then we disable the use of gemm_info.activation_info
     if(gemm_info.post_ops.size() > 0)
     {
         post_op_utils.set_post_ops_cl_build_options(build_opts, gemm_info.post_ops);
     }
     else
     {
         build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(gemm_info.activation_info.activation())));
         build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.a()));
         build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.b()));
     }
 
     std::string kernel_name("gemm_mm_reshaped_only_rhs_");
     kernel_name += rhs_info.transpose ? "t" : "nt";
     kernel_name += rhs_info.export_to_cl_image ? "_texture" : "";
     post_op_utils.set_post_ops_cl_kernel_name(kernel_name, gemm_info.post_ops);
 
     // A macro guard to compile ONLY the kernel of interest
     build_opts.add_option("-D" + upper_string(kernel_name));
 
     // Create kernel
     _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
 
     // Set config_id for enabling LWS tuning
     _config_id = kernel_name;
     _config_id += "_";
     _config_id += (_has_pad_y ? "" : "no_pad_y_");
     _config_id += (_add_bias ? "add_bias_" : "");
     _config_id += (gemm_info.broadcast_bias ? "broadcast_bias_" : "");
     _config_id += (_reinterpret_input_as_3d ? "3di_" : "");
     _config_id += (_reinterpret_output_as_3d ? "3do_" : "");
     _config_id += (gemm_info.activation_info.enabled() ? "fused_activation_" : "");
     _config_id += lower_string(string_from_data_type(src0->data_type()));
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(1));
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(0));
     _config_id += "_";
     _config_id += support::cpp11::to_string(gemm_info.k);
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(2));
     _config_id += "_";
     _config_id += support::cpp11::to_string(lhs_info.m0);
     _config_id += "_";
     _config_id += support::cpp11::to_string(rhs_info.n0);
     _config_id += "_";
     _config_id += support::cpp11::to_string(rhs_info.k0);
     _config_id += "_";
     _config_id += support::cpp11::to_string(rhs_info.h0);
     _config_id += "_";
     _config_id += support::cpp11::to_string(rhs_info.interleave);
 
     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
 }

◆ run_op()

void run_op	(	ITensorPack &	tensors,
		const Window &	window,
		cl::CommandQueue &	queue
	)

overridevirtual

Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.

Note: The queue is not flushed by this method, and therefore the kernel will not have been executed by the time this method returns.

Parameters

[in]	tensors	A vector containing the tensors to operato on.
[in]	window	Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
[in,out]	queue	Command queue on which to enqueue the kernel.

Reimplemented from ICLKernel.

Definition at line 334 of file ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp.

References arm_compute::ACL_DST, arm_compute::ACL_SRC_0, arm_compute::ACL_SRC_1, arm_compute::ACL_SRC_2, ICLKernel::add_2D_tensor_argument(), ICLKernel::add_2D_tensor_argument_if(), ARM_COMPUTE_ERROR_ON, ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW, ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL, arm_compute::create_image2d_from_buffer(), Window::DimX, Window::DimY, arm_compute::enqueue(), Window::first_slice_window_3D(), CLKernelLibrary::get(), ITensorPack::get_const_tensor(), arm_compute::experimental::get_post_op_arg_type(), ITensorPack::get_tensor(), ICLKernel::lws_hint(), Window::set(), arm_compute::test::validation::reference::slice(), Window::slide_window_slice_3D(), and IKernel::window().

 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
     const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
     const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
     const auto src2 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_2));
     auto       dst  = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
 
     ARM_COMPUTE_ERROR_ON_NULLPTR(src0, src1, dst);
     ARM_COMPUTE_ERROR_ON(_add_bias && src2 == nullptr);
 
     if(src1->info()->num_dimensions() < 3)
     {
         // The stride_z for matrix B must be zero if we do not slice
         ARM_COMPUTE_ERROR_ON(src1->info()->strides_in_bytes()[3] != 0);
     }
 
     const size_t lhs_idx_batch_size = _reinterpret_input_as_3d && !_has_pad_y ? 3u : 2u;
     const size_t rhs_idx_batch_size = 2u;
     const size_t bia_idx_batch_size = 2u;
     const size_t out_idx_batch_size = _reinterpret_output_as_3d && !_has_pad_y ? 3u : 2u;
 
     Window slice          = window.first_slice_window_3D();
     Window slice_matrix_b = slice;
 
     slice_matrix_b.set(Window::DimX, Window::Dimension(0, 1, 1));
     slice_matrix_b.set(Window::DimY, Window::Dimension(0, 1, 1));
 
     // Get cross plane pads
     const unsigned int total_cross_plane_pad_lhs = src0->info()->padding().top + src0->info()->padding().bottom;
     const unsigned int total_cross_plane_pad_out = dst->info()->padding().top + dst->info()->padding().bottom;
 
     // The execution should fail if we try to run with has_pad_y = false but we have padding in either the LHS or DST tensor
     ARM_COMPUTE_ERROR_ON(!_has_pad_y && ((total_cross_plane_pad_lhs != 0) || (total_cross_plane_pad_out != 0)));
 
     cl::Image2D src1_image2d;
 
     if(_export_to_cl_image)
     {
         const TensorShape shape2d(src1->info()->dimension(0) / 4, src1->info()->dimension(1) * src1->info()->dimension(2));
         const size_t      image_row_pitch = src1->info()->strides_in_bytes()[1];
 
         src1_image2d = create_image2d_from_buffer(CLKernelLibrary::get().context(), src1->cl_buffer(), shape2d, src1->info()->data_type(), image_row_pitch);
     }
 
     do
     {
         Window slice_b = slice;
         // Don't slice matrix B along the z dimension if matrix B has just 2 dimensions and matrix A more than 2
         // This scenario can happen when the matrix multiplication is used to perform a convolution operation
         if(!_slide_matrix_b)
         {
             slice_b = slice_matrix_b;
         }
 
         unsigned int idx = 0;
 
         // LHS buffer
         add_2D_tensor_argument(idx, src0, slice);
 
         // RHS buffer or RHS OpenCL image (_export_to_cl_image == true)
         if(_export_to_cl_image)
         {
             _kernel.setArg(idx++, src1_image2d);
         }
         else
         {
             add_2D_tensor_argument(idx, src1, slice_b);
         }
 
         // Bias buffer (_add_bias == true)
         add_2D_tensor_argument_if(_add_bias, idx, src2, slice);
 
         // dst buffer
         add_2D_tensor_argument(idx, dst, slice);
 
         // post op argument buffers
         for(size_t i = 0; i < _num_post_op_args; ++i)
         {
             const auto post_op_arg = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(experimental::get_post_op_arg_type(i)));
             add_2D_tensor_argument(idx, post_op_arg, slice);
         }
 
         // LHS stride_z
         _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src0->info()->strides_in_bytes()[lhs_idx_batch_size]));
 
         // RHS stride_z (not used if _export_to_cl_image == true)
         _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src1->info()->strides_in_bytes()[rhs_idx_batch_size]));
 
         // Bias stride_z (if _add_bias == true)
         if(_add_bias)
         {
             _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src2->info()->strides_in_bytes()[bia_idx_batch_size]));
         }
 
         // dst stride_z
         _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(dst->info()->strides_in_bytes()[out_idx_batch_size]));
         // post op argument stride_z
         for(size_t i = 0; i < _num_post_op_args; ++i)
         {
             const auto post_op_arg = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(experimental::get_post_op_arg_type(i)));
             _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(post_op_arg->info()->strides_in_bytes()[2]));
         }
 
         // Cross-plan padding (if _reinterpret_input_as_3d = true)
         if(_reinterpret_input_as_3d && _has_pad_y)
         {
             _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(total_cross_plane_pad_lhs));
         }
 
         // Cross-plan padding (if reinterpret_output_as_3d = true)
         if(_reinterpret_output_as_3d && _has_pad_y)
         {
             _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(total_cross_plane_pad_out));
         }
 
         // Pass m, n and k at runtime as signed ints, to ensure results of any subractions they could be operand in, would still be signed.
         _kernel.setArg<cl_int>(idx++, _m);
         _kernel.setArg<cl_int>(idx++, _n);
         _kernel.setArg<cl_int>(idx++, _k);
 
         enqueue(queue, *this, slice, lws_hint(), _use_dummy_work_items);
     }
     while(window.slide_window_slice_3D(slice));
 }

◆ validate()

Status validate	(	const ITensorInfo *	src0,
		const ITensorInfo *	src1,
		const ITensorInfo *	src2,
		const ITensorInfo *	dst,
		float	alpha,
		float	beta,
		const GEMMLHSMatrixInfo &	lhs_info,
		const GEMMRHSMatrixInfo &	rhs_info,
		const GEMMKernelInfo &	gemm_info
	)

static

Static function to check if given info will lead to a valid configuration.

Returns: a status

Definition at line 326 of file ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp.

References ARM_COMPUTE_RETURN_ON_ERROR, and arm_compute::cpu::kernels::validate_arguments().

 {
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src0, src1, src2, dst, alpha, beta, lhs_info, rhs_info, gemm_info));
     return Status{};
 }

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

src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h
src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.cpp

Public Member Functions

Static Public Member Functions