CUDA compilation error for SecDecInternalNPow

[tueda]

I'm wondering whether the following build failure is due to something wrong with my environment or a bug in pySecDec.

generate_F1.py:

import pySecDec as psd

if __name__ == "__main__":
    li = psd.LoopIntegralFromGraph(
        internal_lines=[
            ["m", [1, 2]],
            [0, [1, 2]],
            [0, [1, 3]],
            [0, [2, 3]],
        ],
        powerlist=[1, 1, 1, -2],
        external_lines=[["p1", 1], ["p2", 2], ["p3", 3]],
        replacement_rules=[
            ["p1**2", "msq"],
            ["p2**2", "msq"],
            ["p3**2", "s"],
            ["p1*p2", "s/2-msq"],
            ["p2*p3", "-s/2"],
            ["p1*p3", "-s/2"],
            ["m**2", "msq"],
            ["msq", 1],
        ],
    )

    psd.loop_package(
        name="F1",
        loop_integral=li,
        requested_orders=[0],
        real_parameters=["s"],
    )

Build command:

python3 generate_F1.py
make -C F1 disteval CXXFLAGS='-mavx2 -mfma' SECDEC_WITH_CUDA_FLAGS='-arch=sm_80'

Compilation error:

nvcc -std=c++17 -arch=sm_80  -dc -fatbin -o distsrc/sector_4_0.fatbin distsrc/sector_4_0.cu
distsrc/sector_4_0.cu(55): error: calling a __host__ function("T2  ::SecDecInternalNPowTemplate<(unsigned int)5u, double> (const T2 &)") from a __global__ function("F1_integral__sector_4_order_0") is not allowed

distsrc/sector_4_0.cu(55): error: identifier "SecDecInternalNPowTemplate<(unsigned int)5u, double> " is undefined in device code

distsrc/sector_4_0.cu(56): error: calling a __host__ function("T2  ::SecDecInternalNPowTemplate<(unsigned int)6u, double> (const T2 &)") from a __global__ function("F1_integral__sector_4_order_0") is not allowed

distsrc/sector_4_0.cu(56): error: identifier "SecDecInternalNPowTemplate<(unsigned int)6u, double> " is undefined in device code

4 errors detected in the compilation of "distsrc/sector_4_0.cu".

The build succeeds when powerlist is changed to [1, 1, 1, -1] or when SECDEC_WITH_CUDA_FLAGS is omitted.

My environment:

Ubuntu 20.04.6 LTS (x86_64)
gcc/g++ 10.5.0
Python 3.12.4
pySecDec 1.6.4 (installed in a virtual environment)
CUDA Toolkit 11.8.89

[magv]

Hi, Takahiro. This is indeed a pySecDec problem. We've fixed it internally, and should get it out soon.

[tueda]

That's good to know. Thank you!

[spj101]

This should be fixed in our latest release (1.6.5).