Using find_package(CUDAQ REQUIRED) in CMake does not work

[diehlpk]

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

Using find_package(CUDAQ REQUIRED) in a CMake project results in the following error

-- The CUDA Quantum compiler identification is NVQ++ - /opt/nvidia/cudaq/bin/nvq++
-- Number of NVIDIA GPUs detected: 1
-- CUDA Quantum Target = nvidia
-- Configuring done
CMake Error: Error required internal CMake variable not set, cmake may not be built correctly.
Missing variable is:
CMAKE_CUDAQ_CREATE_SHARED_LIBRARY
CMake Error: Error required internal CMake variable not set, cmake may not be built correctly.
Missing variable is:
CMAKE_CUDAQ_CREATE_SHARED_LIBRARY
-- Generating done
CMake Generate step failed.  Build files cannot be regenerated correctly.

Steps to reproduce the bug

Using the latest image and include include find_package(CUDAQ REQUIRED) in your CMake project.

Expected behavior

I would assume that CMAKE_CUDAQ_CREATE_SHARED_LIBRARY should be set, since I am not sure what provide. It seems that On or Off does not work.

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA-Q version:
• Python version:
• C++ compiler:
• Operating system:

Suggestions

No response

[scothalverson]

I think I have resolved this for Patrick. Needed to add set_target_properties(LIB_NAME PROPERTIES LINKER_LANGUAGE CXX) to CMake to get C++ libraries to link properly. Waiting on Patrick to confirm this works for him.

[diehlpk]

Yes, that works for me.

[diehlpk]

@scothalverson I a use include(FetchContent), I get the following error

-- Number of NVIDIA GPUs detected: 1
-- CUDA Quantum Target = nvidia
-- Configuring done
CMake Error: Error required internal CMake variable not set, cmake may not be built correctly.
Missing variable is:
CMAKE_CUDAQ_ARCHIVE_CREATE
CMake Error: Error required internal CMake variable not set, cmake may not be built correctly.
Missing variable is:
CMAKE_CUDAQ_ARCHIVE_FINISH
-- Generating done
CMake Generate step failed.  Build files cannot be regenerated correctly.

[diehlpk]

It seems that set(CMAKE_CUDAQ_ARCHIVE_CREATE "<CMAKE_AR> qc <TARGET> <LINK_FLAGS> <OBJECTS>") does the trick and I can compile the code again.

[amccaskey]

Hey @diehlpk and @scothalverson .

Try to add the following before find_package(CUDAQ REQUIRED).

set(CUDAQ_LIBRARY_MODE ON)

Historically, there was some early implementation work that tried to expose CUDA-Q as a CMake language (enable_language(CUDAQ)). This was found to be brittle and we've since moved to better approaches. the CUDA_LIBRARY_MODE flag will disable this early work. Ultimately this should become the default IMO. Checkout the https://github.com/nvidia/cudaqx repo for a great reference on how to build off of CUDA-Q with CMake.

An interesting pattern that others will likely want to follow can be found in the cudaqx repo - separate compilation of host and qpu code. Checkout cmake/Modules/CUDA-QX.cmake there for a function that users can leverage to use nvq++ on known quantum kernel source files cudaqx_add_device_code. IMO this should be part of CUDA-Q so others can use it.