42 using namespace misc::shape_calculator;
46 using ElementsProcessed = Steps;
48 Status
validate_arguments(
const ITensorInfo *input0,
const ITensorInfo *input1,
const ITensorInfo *output,
const GEMMLHSMatrixInfo &lhs_info,
const GEMMRHSMatrixInfo &rhs_info,
49 const GEMMReshapeInfo &gemm_info)
70 const int m = gemm_info.m();
71 const int n = gemm_info.n();
72 const int k = gemm_info.k();
81 if(gemm_info.reinterpret_input_as_3d())
90 if(output->total_size() != 0)
92 const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(
compute_mm_shape(*input0, *input1, gemm_info));
100 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output,
const GEMMLHSMatrixInfo &lhs_info,
const GEMMRHSMatrixInfo &rhs_info,
101 const GEMMReshapeInfo &gemm_info, ElementsProcessed &num_elements_processed)
103 unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0];
104 unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1];
105 bool reinterpret_input_as_3d = gemm_info.reinterpret_input_as_3d();
106 bool reinterpret_output_as_3d = (gemm_info.depth_output_gemm3d() != 0);
109 bool window_changed =
false;
113 if(reinterpret_input_as_3d == reinterpret_output_as_3d)
115 reinterpret_output_as_3d =
false;
121 TensorInfo tmp_info(*output);
123 if(reinterpret_output_as_3d)
127 TensorShape tmp_shape(output->tensor_shape());
128 tmp_shape.collapse(2U, 1U);
129 tmp_info.set_tensor_shape(tmp_shape);
133 num_elems_processed_per_iteration_x = rhs_info.n0;
134 num_elems_processed_per_iteration_y = lhs_info.m0;
136 win =
calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
139 AccessWindowStatic input1_access(input1, 0, 0,
140 ceil_to_multiple(input1->dimension(0), num_elems_processed_per_iteration_x),
141 input1->dimension(1));
147 Window collapsed = win;
148 const unsigned int dimension_to_collapse = std::min(static_cast<unsigned int>(output->num_dimensions()), 2u);
149 collapsed = win.collapse(win, dimension_to_collapse);
152 return std::make_pair(err, collapsed);
157 : _input0(nullptr), _input1(nullptr), _output(nullptr), _slide_matrix_b(true), _reinterpret_input_as_3d(false), _reinterpret_output_as_3d(false), _use_dummy_work_items(false)
187 if(_reinterpret_input_as_3d == _reinterpret_output_as_3d)
189 _reinterpret_input_as_3d =
false;
190 _reinterpret_output_as_3d =
false;
197 ElementsProcessed num_elements_processed{};
200 auto win_config = validate_and_configure_window(input0->
info(), input1->
info(), output->
info(), lhs_info, rhs_info, gemm_info, num_elements_processed);
202 ICLKernel::configure_internal(win_config.second);
207 const unsigned int internal_m = _reinterpret_output_as_3d ? gemm_info.m() : output->
info()->
dimension(1);
209 const unsigned int partial_store_m0 = internal_m % lhs_info.
m0;
210 const unsigned int partial_store_n0 = gemm_info.n() % rhs_info.n0;
214 const unsigned int internal_m0 = std::min(internal_m, lhs_info.
m0);
218 build_opts.
add_option_if(_reinterpret_input_as_3d,
"-DREINTERPRET_INPUT_AS_3D");
219 build_opts.
add_option_if(_reinterpret_output_as_3d,
"-DREINTERPRET_OUTPUT_AS_3D");
223 build_opts.
add_option_if(_use_dummy_work_items,
"-DDUMMY_WORK_ITEMS");
244 _config_id += (_reinterpret_input_as_3d ?
"3di_" :
"");
245 _config_id += (_reinterpret_output_as_3d ?
"3do_" :
"");
266 ElementsProcessed num_elements_processed{};
269 input1->
clone().get(),
270 output->
clone().get(),
274 num_elements_processed)
297 if(_reinterpret_input_as_3d)
302 _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad));
305 if(_reinterpret_output_as_3d)
310 _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad));
320 slice_b = slice_matrix_b;
323 unsigned int idx = 0;
327 _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input0->
info()->
strides_in_bytes()[2]));
328 _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input1->
info()->
strides_in_bytes()[2]));
329 _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_output->
info()->
strides_in_bytes()[2]));
unsigned int top
top of the border
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
bool dot8_supported(const cl::Device &device)
Helper function to check whether the cl_arm_integer_dot_product_int8 extension is supported.
std::unordered_map< const ITensorInfo *, PaddingSize > get_padding_info(std::initializer_list< const ITensorInfo * > infos)
Stores padding information before configuring a kernel.
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint=CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items=false)
Add the kernel to the command queue with the given window.
const StringSet & options() const
Gets the current options list set.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
bool preferred_dummy_work_items_support(const cl::Device &device)
Helper function to check if "dummy work-items" are preferred to have a power of two NDRange In case d...
GEMM reshape information class.
std::string get_cl_dot8_acc_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL dot8 accumulator type.
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
std::string to_string(T &&value)
Convert integer and float values to string.
virtual DataType data_type() const =0
Data type used for each element of the tensor.
TensorShape compute_mm_shape(const ITensorInfo &input0, const ITensorInfo &input1, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info)
Calculate the matrix multiplication output shape of two tensors.
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
GEMM LHS (Left Hand Side) matrix information.
Store the tensor's metadata.
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Describe one of the image's dimensions with a start, end and step.
unsigned int bottom
bottom of the border
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Copyright (c) 2017-2021 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set< std::string > &build_opts=std::set< std::string >())
Creates an opencl kernel using a compile context.
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
GEMM RHS (Right Hand Side) matrix information.
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
quantized, asymmetric fixed-point 8-bit number unsigned
void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
Initialise the kernel's input and output.
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
bool auto_init_if_empty(ITensorInfo &info, const TensorShape &shape, int num_channels, DataType data_type, QuantizationInfo quantization_info=QuantizationInfo())
Auto initialize the tensor info (shape, number of channels and data type) if the current assignment i...
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
virtual PaddingSize padding() const =0
Padding of tensor.
static constexpr unsigned int num_arguments_per_2D_tensor()
Returns the number of arguments enqueued per 2D tensor object.
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
bool has_padding_changed(const std::unordered_map< const ITensorInfo *, PaddingSize > &padding_map)
Check if the previously stored padding info has changed after configuring a kernel.
quantized, symmetric fixed-point 8-bit number
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.
static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpMatrixMultiply...
quantized, symmetric per channel fixed-point 8-bit number
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
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.
Interface for OpenCL tensor.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
CLGEMMLowpMatrixMultiplyNativeKernel()
Default Constructor.
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Wrapper to configure the Khronos OpenCL C++ header.
int depth_output_gemm3d() const
Depth (third dimension) of the output tensor to be used with the GEMM3D kernel.
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
unsigned int k0
Number of partial accumulations performed by the matrix multiplication.
unsigned int m0
Number of rows processed by the matrix multiplication.
quantized, asymmetric fixed-point 8-bit number signed
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
Window first_slice_window_3D() const
First 3D slice of the window.
bool reinterpret_input_as_3d() const
Flag which specifies if the input tensor has to be reinterpreted as 3D.
Describe a multidimensional execution window.
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)