How to include CuOmmBaking with minimum CMake config?

Gaming and Visualization Technologies Visualization OptiX
1

I am trying to include OmmBaking from optix-toolkit GitHub - NVIDIA/optix-toolkit: Set of utilities supporting workflows common in GPU raytracing applications
However, there is no individual cmake doc for each of them. Without luck, I have to look into the cmake file.
You guys are using 3 different deps for ommBakingSimple example. That make me confused, are they madatory.

target_link_libraries( ommBakingSimple
  OptiXToolkit::CuOmmBaking
  OptiXToolkit::Util
  OptiX::OptiX
)

Then I started to try it in my cmake. The OptiXToolkit:: name space doesn’t work out of box. ALIAS offer me help, this config eventually become valid. While it didn’t work when I want to include <OptiXToolkit/CuOmmBaking/CuOmmBaking.h> in my cpp file.

target_include_directories(pname PRIVATE ${CMAKE_SOURCE_DIR}/projects/optix-toolkit)
target_include_directories(pname PRIVATE ${CMAKE_SOURCE_DIR}/projects/optix-toolkit/CuOmmBaking)

add_library(OptiXToolkit::CuOmmBaking ALIAS CuOmmBaking)
target_link_libraries(pname PRIVATE OptiXToolkit::CuOmmBaking)
2

Hi @iaomw ,

the OptiX OmmBaking sample needs OmmBaking and OptiX.

The OptiXToolkit::Util library is providing something used by the sample, so yes, all three are required for the sample.

What is used from Util I don’t know off hand, but looking at the sample should make it obvious from what headers it is including.

You should add OmmBaking as a subdirectory and the target should all be setup correctly with the proper aliases. I guess this wasn’t done.

3

Hi,
The vcpkg approach requires me to include(ProjectOptions) before my project(root_project). However the optix render is not root_project, this action will probably pollute other sub projects. And otk is trying to use my CMAKE_TOOLCHAIN_FILE searching for deps, ignored it’s own vcpkg.

If the variable CMAKE_TOOLCHAIN_FILE is not set when ProjectOptions is included, it will be set to point to the CMake integration in the toolkit’s vcpkg submodule. A parent project using vcpkg for dependency management may be using its own submodule of vcpkg for toolchain integration and should set CMAKE_TOOLCHAIN_FILE as appropriate before including the toolkit’s ProjectOptions module.

While I don’t understand how to solve this conflict, my CMAKE_TOOLCHAIN_FILEis sent from command line, what is the before under this use case.

The FetchContent approach seems working at first, cmake stops arguing, and I don’t need pollute root_project. Unfortunately it has compile error.

[build] [466/703 66% :: 78.649] Building CUDA object render\otk\OmmBaking\CMakeFiles\CuOmmBaking.dir\src\Texture.cu.obj

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda.h(22252): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda.h(22958): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cub/warp/specializations/warp_exchange_shfl.cuh(1): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda/functional(1): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda.h(22252): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda.h(22958): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cub/warp/specializations/warp_exchange_shfl.cuh(1): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] C:\PROGRA~1\NVIDIA~2\CUDA\v12.6\include\cuda/functional(1): warning C4819: The file contains a character that cannot be represented in the current code page (936). Save the file in Unicode format to prevent data loss

[build] Texture.cu

[build] tmpxft_00005df0_00000000-10_Texture.cudafe1.cpp

[build] ninja: build stopped: subcommand failed.
4

About the ALIAS OptiX::OptiX, it didn’t work in my cmake even under FetchContent approach.
Through cmake log, I am sure otk successfully find my optix 8.1 install.
-external:I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\include" -external:I"C:\ProgramData\NVIDIA Corporation\OptiX SDK 8.1.0\include"

For the render itself, I already have another approach to include optix, are there any possible conflict?

5

I suspect you’re encountering problems involving the way you are incoporating the OptiX Toolkit in your CMake project. Can you provide your CMakeLists.txt?
You shouldn’t have to define the OptiXToolkit::CuOmmBaking alias. It’s defined at line 61 of OmmBaking/CMakeLists.txt.
OptiX::OptiX is not an alias, it’s an interface library that simply records the location of the OptiX and CUDA include directories as target properties. That is defined in CMake/FindOptiX.cmake
You will very likely encounter problems if you attempt to define OptiX::OptiX in some other way, since CMake will not allow multiple definitions of the same library name.

6

Hello, I didn’t define OptiX::OptiX anywhere. I included Optix headers directly in my cmake and nvrtc config, it works without any problem. The OptiXToolkit::CuOmmBaking alias is only for testing and play, it didn’t work with or without.

Here is my cmake config related to otk below

# add_compile_options(-DOTK_USE_VCPKG=OFF)
# add_compile_options(-DOTK_BUILD_TESTS=OFF)
# add_compile_options(-DOTK_BUILD_EXAMPLES=OFF)
# add_compile_options(-DBUILD_SHARED_LIBS:BOOL=OFF)

add_subdirectory(${CMAKE_SOURCE_DIR}/projects/optix-toolkit otk)
# add_subdirectory(${CMAKE_SOURCE_DIR}/projects/optix-toolkit/OmmBaking omm)
# target_link_libraries(render PRIVATE Optix::Optix)
target_link_libraries(render PRIVATE OptiXToolkit::CuOmmBaking OptiXToolkit::Util)

"cmake.configureSettings": {
        "CMAKE_TOOLCHAIN_FILE": "$env{VCPKG_ROOT}/scripts/buildsystems/vcpkg.cmake",
        "VCPKG_TARGET_TRIPLET": "x64-windows",
        "configured": true,
},
"cmake.configureArgs": [
        // "-DBUILD_SHARED_LIBS:BOOL=OFF",
        "-DOTK_USE_VCPKG=OFF",
        "-DOTK_FETCH_CONTENT:BOOL=OFF",
        "-DOTK_BUILD_DOCS=OFF",
        "-DOTK_BUILD_TESTS=OFF",
        "-DOTK_BUILD_EXAMPLES=OFF",
        //"-DOTK_LIBRARIES=OmmBaking",
        "-DImath_DIR:PATH=C:/Github/Imath",
        "-DOpenEXR_DIR:PATH=C:/Github/openexr",
} // vscode cmake config

However, it’s always trying to find openexr under my VCPKG_ROOT even I offered -DOpenEXR_DIR. It’s DemandLoading tyring to find openexr, but I don’t acutally use DemandLoading at current stage.

Then I found out -DOTK_LIBRARIES=OmmBaking with only target_link_libraries(render PRIVATE OptiXToolkit::CuOmmBaking) could compile without any problem. In that case, OptiXToolkit::Util is gone, it seems not belong to any library DemandLoading;Memory;OmmBaking;ShaderUtil

In my understanding, OmmBaking itself don’t acutually need Optix and other bloat. It could be pure C++/cuda code, cook the omm data, then I pass it to Optix APIs. May require some data struct from Optix headers.

7

Great, I’m glad you got it working. Yes, most of the third party dependencies in the OptiX Toolkit are incurred by the DemandLoading library, so setting OTK_LIBRARIES is a good way to simplify things.
OmmBaking does need the OptiX SDK, which is a header-only package. The OptiX API calls are implemented by the driver.
I thought OmmBaking relied on the Util library for trivial stuff like error reporting. That might have changed. I’ll take a look and eliminate that dependency if possible. Thanks for the note.

8

My config work fine for debug build under vscode/cmake. While it failed under release build. I guess both otk and me included CUDA, they have some kind of conflict.
In my cmake, CUDA is included by this 2 lines.

find_package(CUDAToolkit REQUIRED COMPONENTS cudart nvrtc REQUIRED)
target_link_libraries(render PRIVATE CUDA::cudart CUDA::nvrtc)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaCreateChannelDesc already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaCreateTextureObject already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaDeviceGetAttribute already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaDeviceSynchronize already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaDriverGetVersion already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaFree already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaFreeArray already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaFreeAsync already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaFreeHost already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGetDevice already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGetDeviceProperties_v2 already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGetErrorString already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGetLastError already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGraphicsMapResources already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGraphicsResourceGetMappedPointer already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGraphicsUnmapResources already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaGraphicsUnregisterResource already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaHostAlloc already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaHostGetDevicePointer already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMalloc already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMallocArray already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMallocAsync already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMallocManaged already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMemcpy already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMemcpy2DToArray already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMemcpyToArray already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaMemset already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaRuntimeGetVersion already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaSetDevice already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaStreamDestroy already defined in cudart.lib(cudart64_12.dll)

[build] cudart_static.lib(cudart_generated_cuda_runtime_api.obj) : error LNK2005: cudaStreamSynchronize already defined in cudart.lib(cudart64_12.dll)
9

Try making that PUBLIC instead of PRIVATE. I think duplicate target_link_libraries are supposed to be combined, but perhaps not if they’re PRIVATE. The OmmBaking CUDA link dependency is PUBLIC.

10

It turns out to be the conflict betweencudart and cudart_static. Then I also changed intocudart_static, it works flawless under both PUBLIC/PRIVATE.

Then I also test nvrtc_static, it doen’t compile with error LNK2038: mismatch detected for 'RuntimeLibrary': value 'MD_DynamicRelease' doesn't match value 'MT_StaticRelease' possibly conflict with Qt5 from my vcpkg.

For nvrtc_static, I am considering the possibility making the app fully portable no matter CudaSDK is installed or not on user’s computer.

For CuOmmBaking itself, my final config is below:

cmake -G Ninja -B build -DCMAKE_BUILD_TYPE="${b}" `
    -DCMAKE_TOOLCHAIN_FILE="${env:VCPKG_ROOT}/scripts/buildsystems/vcpkg.cmake" `
    -D_MY_Other_Configs `
    -DOTK_USE_VCPKG:BOOL=OFF `
    -DOTK_FETCH_CONTENT:BOOL=OFF `
    -DOTK_BUILD_DOCS:BOOL=OFF `
    -DOTK_BUILD_TESTS:BOOL=OFF `
    -DOTK_BUILD_EXAMPLES:BOOL=OFF `
    -DOTK_LIBRARIES=OmmBaking
# -DBUILD_SHARED_LIBS:BOOL=OFF # Both ON and OFF work

CMakeList.txt for the Optix render

find_package(CUDAToolkit REQUIRED COMPONENTS cudart nvrtc REQUIRED)
target_link_libraries(render_name PRIVATE CUDA::cudart_static CUDA::nvrtc)

add_subdirectory(${CMAKE_SOURCE_DIR}/projects/optix-toolkit otk)
target_link_libraries(render_name PRIVATE OptiXToolkit::CuOmmBaking)
11

Nice. Thanks for following up.