![]() |
MMALIB User Guide
|
Common definitions | This module consists of definitions (macros, structures, utility functions) that are commonly applicable to all MMALIB kernels |
▼Convolutional Neural Networks (CNN) kernels | This module consists of kernels to implement the core computations occurring in the context of convolutional neural networks |
MMALIB_CNN_convolveBias_row_ixX_ixX_oxX | Kernel for computing dense CNN convolution with row based processing |
▼MMALIB_CNN_convolve_col_smallNo_highPrecision | NOTE: This API is now a wrapper to MMALIB_CNN_convolve_col_smallNo_highPrecision_pointwisePost with the lutValues input argument set to NULL. It is recommended to call MMALIB_CNN_convolve_col_smallNo_highPrecision_pointwisePost directly |
MMALIB_CNN_convolve_col_smallNo_highPrecision_reorderWeights | MMALIB_CNN_convolve_col_smallNo_highPrecision requires that the weights be preprocessed into a specific arrangement. The functions in this module perform that preprocessing and other associated tasks |
▼MMALIB_CNN_convolve_col_smallNo_highPrecision_pointwisePost | Kernel for computing CNN-style 2D convolution using column major data ordering on the input and output feature maps. This approach computes more quickly if filter grouping is chosen such that Ni=No=1, or if filter grouping is chosen such that NiFrFc < MMA_SIZE, otherwise use regular convolution method MMALIB_CNN_convolve_row_ixX_ixX_oxX. This kernel is also referred to as depth-wise convolution |
MMALIB_CNN_convolve_col_smallNo_highPrecision_pointwisePost_reorderWeights | MMALIB_CNN_convolve_col_smallNo_highPrecision_pointwisePost requires that the weights be preprocessed into a specific arrangement. The functions in this module perform that preprocessing and other associated tasks |
▼MMALIB_CNN_convolve_col_smallNo_ixX_ixX_oxX | Kernel for computing CNN-style 2D convolution using column major data ordering on the input and output feature maps. This approach computes more quickly if filter grouping is chosen such that Ni=No=1, or if filter grouping is chosen such that NiFrFc < MMA_SIZE, otherwise use regular convolution method MMALIB_CNN_convolve_row_ixX_ixX_oxX. This kernel is also referred to as depth-wise convolution |
MMALIB_CNN_convolve_col_smallNo_ixX_ixX_oxX_reorderWeights | MMALIB_CNN_convolve_col_smallNo_ixX_ixX_oxX requires that the weights be preprocessed into a specific arrangement. The functions in this module perform that preprocessing and other associated tasks |
MMALIB_CNN_deconvolve_row_ixX_ixX_oxX | Kernel for computing dense CNN deconvolution with row-based processing and matrix-matrix multiplication |
MMALIB_CNN_fullyConnectedBias_ixX_ixX_oxX | Kernel provides compute functionality of Fully Connected Layer: \( Y^T = X^T \times H^T + B^T\) |
MMALIB_CNN_tensor_convert_ixX_oxX | Kernel for converting tensors of various datatypes and formats |
▼Digital Signal Processing (DSP) kernels | This module consists of kernels that implement DSP algorithms |
MMALIB_DSP_firSmall_ixX_ixX_oxX | Kernel for convolving input data with an input filter (filter size <= MMA_SIZE/2). For input filter size > MMA_SIZE/2 refer to MMALIB_DSP_fir_ixX_ixX_oxX |
MMALIB_DSP_fir_ixX_ixX_oxX | Kernel for convolving input data with an input filter (filter size > MMA_SIZE/2) For input filter size <= MMA_SIZE/2 refer to MMALIB_DSP_firSmall_ixX_ixX_oxX |
▼Linear Algebra (LINALG) kernels | This module consists of kernels within the linear algebra scope |
MMALIB_LINALG_matrixMatrixMultiplyAccumulate_ixX_ixX_ixX_oxX | Kernel for multiplying two matrices with an additive term |
MMALIB_LINALG_matrixMatrixMultiplyBias_ixX_ixX_oxX | Kernel for multiplying two matrices with bias, scale and shift |
MMALIB_LINALG_matrixMatrixMultiply_ixX_ixX_oxX | Kernel for multiplying two matrices |
MMALIB_LINALG_matrixTranspose_ixX_oxX | Kernel for computing the transpose of a matrix |
MMALIB_LINALG_pointwiseMatrixMatrixMultiply_ixX_ixX_oxX | Kernel for computing the pointwise multiplication of two matrices |