guan404ming commented on code in PR #934:
URL: https://github.com/apache/mahout/pull/934#discussion_r2750487784
##########
qdp/qdp-core/src/lib.rs:
##########
@@ -492,7 +434,9 @@ impl QdpEngine {
{
crate::profile_scope!("GPU::Synchronize");
- sync_cuda_stream(stream, "CUDA stream synchronize
failed")?;
+ self.device.synchronize().map_err(|e| {
Review Comment:
Based on my understadning, `self.device.synchronize()` calls
`cudaDeviceSynchronize`, which blocks all streams on the device. The
`_with_stream` variants should use `sync_cuda_stream(stream, ...)`
(`cudaStreamSynchronize`) to honor the caller's stream. This is a behavioral
regression for multi-stream scenarios.
Could you help confirm this? @ryankert01 @rich7420
--
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]
