![]() |
TI Deep Learning Product User Guide
|
Data Fields | |
uint32_t | data_type |
uint32_t | dim_x |
uint32_t | dim_y |
int32_t | stride_y |
uint32_t | dim_z |
int32_t | stride_z |
float32_tidl * | tensorScaleList |
int32_t * | tensorZeroPointList |
uint32_t TIDL_bufParams3D_t::data_type |
Data type of input TIDL_data_type_e
uint32_t TIDL_bufParams3D_t::dim_x |
Width of buffer in X dimension in elements. This corresponds to width of the tensor which is going to be processed in a given kernel call
uint32_t TIDL_bufParams3D_t::dim_y |
Height of buffer in Y dimension in elements. This corresponds to height of the tensor which is going to be processed in a given kernel call
int32_t TIDL_bufParams3D_t::stride_y |
Stride in Y dimension in bytes.
uint32_t TIDL_bufParams3D_t::dim_z |
Depth of patch in Z dimension in elements.This corresponds to number of channels which is going to be processed in a given kernel call
int32_t TIDL_bufParams3D_t::stride_z |
Stride in Z dimension in bytes.
float32_tidl* TIDL_bufParams3D_t::tensorScaleList |
Pointer to the array containing the tensor's tensor scales on a per channel basis, if quantization is per-tensor only one value is populated
int32_t* TIDL_bufParams3D_t::tensorZeroPointList |
Pointer to the array containing the tensor's zero points on a per channel basis, if quantization is per-tensor only one value is populated