Always clear threadToQpuId when we clear platformQPUs

[1tnguyen]

Description

We should clear threadToQpuId when we initialize the platformQPUs list (clear then repopulate).
The runtime has a static quantum platform instance, hence there could be stray info between run.

For example, a previous run with nvidia-mqpu may populate threadToQpuId based on the number of GPUs. Then, a subsequent remote-mqpu run may have a collision when looking up threadToQpuId with thread id => out of bound access to the platformQPUs map.

This is primarily a Python specific bug as we cannot switch targets with C++.

[bmhowe23]

This is a good find, Thien. Do we also need similar changes here?

cuda-quantum/runtime/cudaq/platform/default/DefaultQuantumPlatform.cpp

Line 84 in 2cf6b04

platformQPUs.clear();

cuda-quantum/runtime/cudaq/platform/default/DefaultQuantumPlatform.cpp

Line 123 in 2cf6b04

platformQPUs.clear();

(Those changes also do platformQPUs.clear() and would therefore seem to invalidate threadToQpuId, if it existed.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[1tnguyen]

This is a good find, Thien. Do we also need similar changes here?

cuda-quantum/runtime/cudaq/platform/default/DefaultQuantumPlatform.cpp

Line 84 in 2cf6b04

platformQPUs.clear();

cuda-quantum/runtime/cudaq/platform/default/DefaultQuantumPlatform.cpp

Line 123 in 2cf6b04

platformQPUs.clear();

(Those changes also do platformQPUs.clear() and would therefore seem to invalidate threadToQpuId, if it existed.

Good point. Added in b0b5f4a.

I think for DefaultQuantumPlatform, this will not cause a problem as we only expect a single qpu (qpu_id = 0) but it's good to have both the vector and the map cleared at the same time throughout the code.

[bmhowe23]

👍

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.