pub trait CudaDtype:
AccelDtype
+ DeviceRepr
+ ValidAsZeroBits {
// Required methods
fn cuda_data_type() -> cudaDataType_t;
fn cublas_compute_type() -> cublasComputeType_t;
fn cuda_type_name() -> &'static str;
fn cudnn_data_type() -> cudnnDataType_t;
fn nccl_data_type() -> ncclDataType_t;
}Expand description
CUDA-specific layer over AccelDtype.
The cudarc bounds (DeviceRepr, ValidAsZeroBits) are part of the
supertrait list so dispatch payloads can call stream.alloc_zeros::<T>
behind a single T: CudaDtype bound.
Required Methods§
fn cuda_data_type() -> cudaDataType_t
fn cublas_compute_type() -> cublasComputeType_t
Sourcefn cuda_type_name() -> &'static str
fn cuda_type_name() -> &'static str
CUDA C++ type name for NVRTC source generation.
fn cudnn_data_type() -> cudnnDataType_t
fn nccl_data_type() -> ncclDataType_t
Dyn Compatibility§
This trait is not dyn compatible.
In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.