As popularly requested in user groups.
Test plan:
```
import torch
a = torch.randn(10000)
device = torch.device('cuda:1')
a = a.to(device)
```
Before:
```
Traceback (most recent call last):
File "/data/users/raymo/pytorch/test/cuda.py", line 6, in <module>
a = a.to(device)
^^^^^^^^^^^^
torch.AcceleratorError: CUDA error: invalid device ordinal
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
```
After:
```
Traceback (most recent call last):
File "/data/users/raymo/pytorch/test/cuda.py", line 6, in <module>
a = a.to(device)
^^^^^^^^^^^^
torch.AcceleratorError: CUDA error: invalid device ordinal
GPU device may be out of range, do you have enough GPUs?
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158395
Approved by: https://github.com/aorenste
Co-authored-by: Aaron Orenstein <aorenste@fb.com>
Summary:
This is a first step towards creating context manager that errors out on synchronizing calls.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61889
Reviewed By: albanD
Differential Revision: D29805280
Pulled By: ngimel
fbshipit-source-id: b66400fbe0941b7daa51e6b30abe27b9cccd4e8a
