# tensorplay.accelerator

Source: https://www.tensorplay.cn/docs/accelerator.html

TensorPlay can run on CPUs and on accelerators such as GPUs. The
tensorplay.accelerator module exposes a small, device-agnostic surface for
asking which accelerator is currently active, so code that needs to know where
tensors live does not have to special-case each backend.

```
import tensorplay as tp

dev = tp.accelerator.current_accelerator()
print(dev)                      # e.g. 'cuda:0', or None on a CPU-only machine
```

## Functions

| tensorplay.accelerator.current_accelerator |The accelerator device selected at build time, if any. |
| --- | --- |
| tensorplay.accelerator.device_count | Number of devices for the current accelerator, or zero without one. |
| tensorplay.accelerator.is_available | Whether an accelerator was built and at least one device is visible. |
| tensorplay.accelerator.current_device_index | Index of the currently selected accelerator device. |
| tensorplay.accelerator.set_device_index | Select the accelerator device by index; negative indices are no-ops. |
| tensorplay.accelerator.get_device_capability | Capability map for an accelerator device. |
| tensorplay.accelerator.synchronize | Wait for all work on an accelerator device to complete. |
| tensorplay.accelerator.Graph | Capture/replay graph on the current accelerator device. |

## Submodules

| tensorplay.accelerator.memory |Device-agnostic memory queries for the current accelerator. |
| --- | --- |
| tensorplay.accelerator.random | Device-agnostic random-number helpers for the current accelerator. |
| tensorplay.accelerator.graphs | Device-agnostic capture/replay graphs for the current accelerator. |

## Where to go next

- [CUDA semantics](/docs/cuda.html) — the CUDA backend the accelerator is most often backed by, and its device-management functions.

- [Device concepts](/docs/tensor_attributes.html) — how device objects are spelled and compared.