36 namespace cl_direct_conv
50 &ClDirectConvDefaultConfigBifrost::configure_G71_f16,
51 &ClDirectConvDefaultConfigBifrost::configure_G71_u8);
54 &ClDirectConvDefaultConfigBifrost::configure_default_f16,
55 &ClDirectConvDefaultConfigBifrost::configure_G71_u8);
57 ConfigurationFunctionExecutorPtr func =
nullptr;
83 if(output_shape[0] > 16)
107 if(output_shape[0] > 16)
131 if(output_shape[0] > 16)
155 if(output_shape[0] > 16)
179 if(output_shape[0] > 16)
ClDirectConvDefaultConfigBifrost(GPUTarget gpu)
Constructor.
Basic container for the OpenCL direct convolution configuration functions.
virtual DataType data_type() const =0
Data type used for each element of the tensor.
Bifrost based OpenCL direct convolution configuration.
Store the tensor's metadata.
T get_function(DataType data_type)
Method to return the direct convolution configuration function based on data type.
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
bool export_weights_to_cl_image
Flag to export the weights to cl_image.
SimpleTensor< float > src
Copyright (c) 2017-2023 Arm Limited.
int32_t n0
Number of columns to be processed by the kernel.
int32_t k0
Number of partial accumulations to be processed in a single iteration by the kernel.
#define ARM_COMPUTE_ERROR_ON_MSG(cond, msg)
Compute descriptor used by the direct convolution kernel.
Padding and stride information class.
GPUTarget
Available GPU Targets.
Basic interface for the Direct convolution kernel configuration.
Num samples, height, width, channels.
int32_t m0
Number of rows to be processed by the kernel.
bool export_to_cl_image(const ITensorInfo *tensor)
DirectConvComputeKernelInfo configure(const ITensorInfo *src, const ITensorInfo *wei, const PadStrideInfo &conv_info) override
This method returns the DirectConvComputeKernelInfo for the given inputs.
TensorShape compute_deep_convolution_shape(const TensorShape &input_shape, DataLayout input_data_layout, const TensorShape &weights_shape, const PadStrideInfo &conv_info)
Calculate the deep convolution shape output shape of a tensor.
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.