|
template<typename T > |
std::pair< T, T > | get_activation_layer_test_bounds (ActivationLayerInfo::ActivationFunction activation, DataType data_type) |
| Helper function to get the testing range for each activation layer. More...
|
|
template<typename T > |
SimpleTensor< float > | convert_from_asymmetric (const SimpleTensor< T > &src) |
| Convert an asymmetric quantized simple tensor into float using tensor quantization information. More...
|
|
template<typename T > |
SimpleTensor< float > | convert_from_symmetric (const SimpleTensor< T > &src) |
| Convert quantized simple tensor into float using tensor quantization information. More...
|
|
template<typename T > |
void | matrix_multiply (const SimpleTensor< T > &a, const SimpleTensor< T > &b, SimpleTensor< T > &out) |
| Matrix multiply between 2 float simple tensors. More...
|
|
template<typename T > |
void | transpose_matrix (const SimpleTensor< T > &in, SimpleTensor< T > &out) |
| Transpose matrix. More...
|
|
template<typename T > |
void | get_tile (const SimpleTensor< T > &in, SimpleTensor< T > &tile, const Coordinates &coord) |
| Get a 2D tile from a tensor. More...
|
|
template<typename T > |
void | zeros (SimpleTensor< T > &in, const Coordinates &anchor, const TensorShape &shape) |
| Fill with zeros the input tensor in the area defined by anchor and shape. More...
|
|
std::pair< int, int > | get_quantized_bounds (const QuantizationInfo &quant_info, float min, float max) |
| Helper function to compute quantized min and max bounds. More...
|
|
std::pair< int, int > | get_quantized_qasymm8_signed_bounds (const QuantizationInfo &quant_info, float min, float max) |
| Helper function to compute asymmetric quantized signed min and max bounds. More...
|
|
std::pair< int, int > | get_symm_quantized_per_channel_bounds (const QuantizationInfo &quant_info, float min, float max, size_t channel_id=0) |
| Helper function to compute symmetric quantized min and max bounds. More...
|
|
void | add_padding_x (std::initializer_list< ITensor * > tensors, const DataLayout &data_layout=DataLayout::NHWC, bool only_right_pad=false) |
| Add random padding along the X axis (between 1 and 16 columns per side) to all the input tensors. More...
|
|
QuantizationHint | suggest_conv_dst_q_info_and_bias (const QuantizationInfo &in_q_info, const QuantizationInfo &weight_q_info, int32_t height, int32_t width, int32_t channels, DataType data_type, float bias_fraction) |
| For 2d convolution, given the Lhs/Rhs matrix quantization informations and the convolution dimension, calculate a suitable output quantization and suggested bias range for obtaining non-saturated outputs with high probability. More...
|
|
QuantizationHint | suggest_matmul_dst_q_info_and_bias (const QuantizationInfo &lhs_q_info, const QuantizationInfo &rhs_q_info, int32_t m, int32_t n, int32_t k, DataType data_type, float bias_fraction) |
| For a matrix multiplication, given the Lhs/Rhs matrix quantization informations and the matrix multiplication dimensions, calculate a suitable output quantization and suggested bias range for obtaining non-saturated outputs with high probability. More...
|
|
QuantizationHint | suggest_mac_dst_q_info_and_bias (const QuantizationInfo &lhs_q_info, const QuantizationInfo &rhs_q_info, int32_t k, DataType data_type, float bias_fraction, int num_sd=2) |
| For a multiply-accumulate (mac), given the Lhs/Rhs vector quantization informations and the dot product dimensions, calculate a suitable output quantization and suggested bias range for obtaining non-saturated outputs with high probability. More...
|
|