Errors after make: not enough memory

[ochapman-dphil]

After running make and pip install . all seemed well.

However upon testing the error message not enough memory: is the number of qubits too large? always appears. This occurs when the qsimcirq_test.py::test_cirq_qsim_gpu_amplitudes test is run and in general when setting qsimcirq.QSimOptions(use_gpu=True). This occurs regardless of whether gpu_mode is set to 0 or 1.

My system should have ample memory available. Are there any likely fixes?

[sergeisakov]

In this context, the error message means that cudaMalloc fails somehow. Certainly, there should be enough memory on your GPU for just two qubits in qsimcirq_test.py::test_cirq_qsim_gpu_amplitudes. Could you add ErrorCheck(rc); after line 85 in lib/vectorspace_cuda.h (and perhaps add #include "util_cuda.h" at the beginning of that file after line 22) to see the actual CUDA error message?

[ochapman-dphil]

The error message gives CUDA error: the provided PTX was compiled with an unsupported toolchain. /tmp/pip-req-build-8_972xno/pybind_interface/custatevec/../../lib/vectorspace_cuda.h 89

I assume this means that the wrong version of CUDA was used to compile the code. I've discovered that nvidia-smi returns CUDA Version: 11.2 whereas nvcc --version returns Cuda compilation tools, release 11.4, V11.4.100 Build cuda_11.4.r11.4/compiler.30188945_0