PengYoun9 commented on code in PR #20362:
URL: https://github.com/apache/tvm/pull/20362#discussion_r4026099257
##########
src/backend/cuda/runtime/cuda_device_api.cc:
##########
@@ -316,6 +317,9 @@ TVM_FFI_STATIC_INIT_BLOCK() {
.def_packed("device_api.cuda",
[](ffi::PackedArgs args, ffi::Any* rv) {
DeviceAPI* ptr = CUDADeviceAPI::Global();
+ // Initialize the pool before contrib thread-local objects
finish construction,
+ // so they return retained workspaces before the pool is
destroyed.
Review Comment:
Yes, you're right. Does this mean I should change the calls at
[cudnn_utils.cc#L168](https://github.com/apache/tvm/blob/main/src/runtime/extra/contrib/cudnn/cudnn_utils.cc#L168)
from `AllocWorkspace` / `FreeWorkspace` to `AllocDataSpace` / `FreeDataSpace`?
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
