struct dnnl_matmul_desc_t¶
Overview¶
A descriptor of a matrix multiplication operation. More…
#include <dnnl_types.h> struct dnnl_matmul_desc_t { // fields dnnl_primitive_kind_t primitive_kind; dnnl_memory_desc_t src_desc; dnnl_memory_desc_t weights_desc; dnnl_memory_desc_t bias_desc; dnnl_memory_desc_t dst_desc; dnnl_data_type_t accum_data_type; };
Detailed Documentation¶
A descriptor of a matrix multiplication operation.
2D case: dst[m, n] = src[m, k] * weights[k, n] + bias[m, n]
3D case: dst[mb, m, n] = src[mb, m, k] * weights[mb, k, n] + bias[mb, m, n]
Fields¶
dnnl_primitive_kind_t primitive_kind
The kind of primitive.
Used for self-identifying the primitive descriptor. Must be dnnl_matmul.
dnnl_memory_desc_t src_desc
Source memory descriptor.
dnnl_memory_desc_t weights_desc
Weights memory descriptor.
dnnl_memory_desc_t bias_desc
Bias memory descriptor.
dnnl_memory_desc_t dst_desc
Destination memory descriptor.
dnnl_data_type_t accum_data_type
The accumulator data type. Initialized automatically.