paleolimbot commented on code in PR #488:
URL: https://github.com/apache/arrow-nanoarrow/pull/488#discussion_r1618038071
##########
extensions/nanoarrow_device/src/nanoarrow/nanoarrow_device_cuda.c:
##########
@@ -32,67 +71,63 @@ static void ArrowDeviceCudaDeallocator(struct
ArrowBufferAllocator* allocator,
struct ArrowDeviceCudaAllocatorPrivate* allocator_private =
(struct ArrowDeviceCudaAllocatorPrivate*)allocator->private_data;
- int prev_device = 0;
- // Not ideal: we have no place to communicate any errors here
- cudaGetDevice(&prev_device);
- cudaSetDevice((int)allocator_private->device_id);
-
switch (allocator_private->device_type) {
case ARROW_DEVICE_CUDA:
- cudaFree(allocator_private->allocated_ptr);
+ cuMemFree((CUdeviceptr)allocator_private->allocated_ptr);
break;
case ARROW_DEVICE_CUDA_HOST:
- cudaFreeHost(allocator_private->allocated_ptr);
+ cuMemFreeHost(allocator_private->allocated_ptr);
break;
default:
break;
}
- cudaSetDevice(prev_device);
Review Comment:
Yes, device setting is now managed with the push/pop context (the notion of
a "current device" is only available in the runtime API and I don't think we
have another option).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
