[nvidia target] Startup delay on certain platforms when host-device memory migration is used
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 1.1k
- Forks
- 456
- Avg merge
- 1d 22h
- Merged PRs (30d)
- 165
Description
On some platforms, when host-device migration is used on the nvidia target (via setting CUDAQ_MAX_CPU_MEMORY_GB), the simulation startup time is delay.
This can be checked by enabling CUDAQ logging, export CUDAQ_LOG_LEVEL=info.
The simulation is stalled after a Create zero initial state line, e.g.,
[2024-11-21 05:55:05.391] [info] [CircuitSimulator.h:913] Allocating 34 new qubits.
[2024-11-21 05:55:05.391] [info] [CusvsimCircuitSimulatorSingleProcess.h:38] [CusvsimSingleProcessCircuitSimulator] Create zero initial state of 34 qubits.
It will eventually finish.
This seems to only occur on certain (platform + driver version) combination.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Reproduce with CUDAQ_MAX_CPU_MEMORY_GB set on an affected platform and CUDAQ_LOG_LEVEL=info, using the log point in CusvsimCircuitSimulatorSingleProcess.h and CircuitSimulator.h as the starting trace. Compare startup behavior around “Create zero initial state” and verify that host-device migration no longer causes the reported delay on the affected platform and driver combination.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- backend, performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100