|
OpenNN
Open-source neural networks library
|
Process-wide singleton that owns the thread pool and the cuBLAS/cuDNN handles. More...
#include <tensor_utilities.h>
Public Member Functions | |
| ThreadPoolDevice * | get_thread_pool_device () |
| Returns the Eigen ThreadPoolDevice used for CPU tensor evaluations. | |
| void | set_threads_number (int num_threads) |
| Reconfigures the underlying thread pool to use num_threads workers. | |
Static Public Member Functions | |
| static Backend & | instance () |
| Returns the global Backend instance. | |
| static cublasHandle_t | get_cublas_handle () |
| Shared cuBLAS handle for legacy GEMM calls. | |
| static cublasLtHandle_t | get_cublas_lt_handle () |
| Shared cuBLASLt handle for batched/tuned GEMMs. | |
| static cudnnHandle_t | get_cudnn_handle () |
| Shared cuDNN handle. | |
| static cudaStream_t | get_compute_stream () |
| Default CUDA stream used by the compute backend. | |
| static cudnnOpTensorDescriptor_t | get_operator_sum_descriptor () |
| cuDNN op-tensor descriptor configured for elementwise sum. | |
Process-wide singleton that owns the thread pool and the cuBLAS/cuDNN handles.
|
inlinestatic |
Default CUDA stream used by the compute backend.
|
inlinestatic |
Shared cuBLAS handle for legacy GEMM calls.
|
inlinestatic |
Shared cuBLASLt handle for batched/tuned GEMMs.
|
inlinestatic |
Shared cuDNN handle.
|
inlinestatic |
cuDNN op-tensor descriptor configured for elementwise sum.
| ThreadPoolDevice * opennn::Backend::get_thread_pool_device | ( | ) |
Returns the Eigen ThreadPoolDevice used for CPU tensor evaluations.
| void opennn::Backend::set_threads_number | ( | int | num_threads | ) |
Reconfigures the underlying thread pool to use num_threads workers.