About cupy build error in HIP (AMD ROCm) environment

Hello,I tried to build an environment based on ROCm for Cupy but failed I tried HIP build on “cupy v7.0.0 rc1” but it failed in some environments. I have built and tried multiple environments and reproduced a similar error.

Env1 OS Ubuntu 18.04.3 Xeon E5-2603 v4 gcc 7.4.0 GPU RadeonⅦ ROCm version 2.9.6 MiOpen Version: 2.1.0.7487-rocm-rel-2.9-6-a5a4129

Env2 Run on Env1 ROCm Docker based env The same build error was reproduced in the following Dockerfile https://gist.github.com/KeigoTakamura/188295b7a5a07215d90348c769af3dbc

Env3

OS ubuntu 18.04.3 eGPU Base Dynabook U63 CPU Corei5 7300U GPU Vega56 ROCm version 2.9.6

The following build errors were confirmed in the above environment

cupy/cuda/runtime.cpp: In function 'int __pyx_f_4cupy_4cuda_7runtime_deviceGetByPCIBusId(PyObject*, int)':
cupy/cuda/runtime.cpp:3966:83: error: 'cudaDeviceGetByPCIBusId' was not declared in this scope
   __pyx_v_status = cudaDeviceGetByPCIBusId((&__pyx_v_device), __pyx_v_c_pci_bus_id);
                                                                                   ^
cupy/cuda/runtime.cpp: In function 'PyObject* __pyx_f_4cupy_4cuda_7runtime_deviceGetPCIBusId(int, int)':
cupy/cuda/runtime.cpp:4076:80: error: 'cudaDeviceGetPCIBusId' was not declared in this scope
   __pyx_v_status = cudaDeviceGetPCIBusId(__pyx_v_pci_bus_id, 13, __pyx_v_device);
                                                                                ^
error: command 'x86_64-linux-gnu-gcc' failed with exit status 1

This error could not be reproduced in some environments of ubuntu16.04 LST, but this error was confirmed in multiple environments and machines

[KeigoTakamura]

Finished processing dependencies for cupy==7.0.0rc1

Build succeeded Thank you for a good job!!!

[leofang]

@KeigoTakamura Could you try again the patch from the same PR? I added a stub for libraryPropertyType_t. (Sorry I don’t have an AMD card to test it myself…)