tensornet-mps backend is slower than tensornet?

[sergei-mironov]

The problem

I have performed some CudaQ tests involving surface codes for quantum error correction using C++ CudaQ frontend. I run kernels for surface code distance 3 codes (17 qubits), code distance 5 (49 qubits) and code distance 7 (97 qubits) using tensornet and tensornet-mps backend, a sample code is attached below. All the backends managed to finish the computation for distances 3 and 5, the fp32 versions were faster than fp64 as expected, but I got surprising results regarding the overall running times: tensornet-mps backend worked slower than the tensornet backend in all tests.

And more: I tweaked CUDAQ_MPS_{MAX_BOND,ABS_CUTOFF,RELATIVE_CUTOFF} parameters specified on the tensornet-mps backend page in a hope to gain some speed for a precision, but saw no improvement.

Expected results

I expected the tensornet-mps backend to work faster than tensornet at the cost of a lower simulation precision. Therefore, I suspect that either I made some obvious configuration error, or tensornet-mps does not work properly. What do you think? What next steps are possible to debug the problem?

Versions

• Base docker image: nvcr.io/nvidia/quantum/cuda-quantum:cu12-0.10.0
• GPU drivers: NVIDIA-SMI 575.64.03 Driver Version: 575.64.03 CUDA Version: 12.9
• GPU model: NVIDIA GeForce RTX 3090
• CudaQ version: CUDA-Q Version cu12-0.10.0 (https://github.com/NVIDIA/cuda-quantum 857dd2ce0a783c32416af8fba8664ff30f9ddc47)

Sample program

Below I provide the sample program for code distance=5 (file circuit.cpp) and the build instructions (file build.sh).

cudaq-sample-d5.zip