|
| template<TensorDataType TComputePrecision, TensorDataType TParameterPrecision = TComputePrecision> |
| CublasLtLinearPlan< TComputePrecision, TParameterPrecision > | Mila::Dnn::Compute::Cuda::build_linear_plan (cublasLtHandle_t handle, int outer_size, int in_features, int out_features, bool has_bias, cublasComputeType_t compute_type, cudaDataType_t scale_type, const float *weight_scale=nullptr) |
| | Build a cuBLASLt plan for a Linear matmul.
|
| template<TensorDataType TComputePrecision, TensorDataType TParameterPrecision = TComputePrecision> |
| void | Mila::Dnn::Compute::Cuda::execute_linear_plan (cublasLtHandle_t handle, const CublasLtLinearPlan< TComputePrecision, TParameterPrecision > &plan, const float *alpha, const void *A, const void *B, const float *beta, typename CublasLtLinearPlan< TComputePrecision, TParameterPrecision >::ActivationType *C, const typename CublasLtLinearPlan< TComputePrecision, TParameterPrecision >::ActivationType *bias, const float *weight_scale, cudaStream_t stream, void *workspace=nullptr, size_t workspace_size=0) |
| | Execute a previously-built CublasLtLinearPlan.
|