| Age | Commit message (Collapse) | Author |
|
|
|
Merging with the last few months of work at Purdue
|
|
into dev-traces
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Deadlock fix from stream_manager
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Enabled CUDA 10. Added implementations for __cudaPushCallConfiguratio…
|
|
cuda 10; while assuring backwards compatibility for older cuda versions
|
|
Nvbits
|
|
|
|
|
|
__cudaPopCallConfiguration, neither of which are documented in the CUDA API but they preceed any cudaLaunchKernel() call and push the call configuration on a stack which is then popped when cudaLaunchKernel() is called. Currently the max stack size is 1, an assert will be triggered when this is violated.
|
|
Add a new config option to cap the max number of completed CTAs
|
|
|
|
|
|
|
|
|