search

mitsuba-renderer / mitsuba3

Mitsuba 3: A Retargetable Forward and Inverse Renderer
https://www.mitsuba-renderer.org/
Other
2.1k stars 246 forks source link

FAILED: ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o & bind_llvm.cpp.o #1235

Closed joshuaroy873 closed 3 months ago

joshuaroy873 commented 4 months ago

Summary

I'm having issues compiling mitsuba3. I'm a bit new to build and compile troubleshooting, so please bare with me. I followed the following steps (as per the instructions), but I keep getting an error.

Big picture: I'm trying to install the following software: https://github.com/NVlabs/instant-rm Their installation summary is: Instant RM is based on Mitsuba 3 and Dr.Jit. The cuda_ad_mono_polarized variant is required to run Instant RM. As this variant is not included in the version of Mitsuba available on PyPI, Mitsuba should not be installed using pip and you will need to install it from the sources with this variant. Selecting variants can be achieved by adding "cuda_ad_mono_polarized" to the listd named "enabled" in the mitsuba.conf file which is generated when running CMake. Note that compiling Mitsuba also installs Dr.Jit from the source. We recommend using Mitsuba 3.5.2.

System configuration

System information:

OS: Windows 10 - WSL - Ubuntu 22.04 CPU: Intel i7-13700H GPU: None Python version: Python 3.10.12 LLVM version: 14.0.0 (output to llvm-config --version command) CUDA version: None (modified listd named "enabled" in mitsuba.conf to llvm_ad_mono_polarized, so assuming no CUDA required) NVidia driver: None

Mitsuba version: 3.5.2

Description

(nvidia_rm_venv) jpalathi@DESKTOP-2V947MI:~/mitsuba3/build$ ninja
[0/1] Re-running CMake...
-- Mitsuba v3.5.2
-- Mitsuba: targeting the native CPU architecture (specify MI_NATIVE_FLAGS to change this).
-- pybind11 v2.10.1
-- Mitsuba: building the following variants:
--  * llvm_ad_mono_polarized
-- Dr.Jit v0.4.6
-- Dr.Jit: building the CUDA & LLVM JIT backend.
-- Dr.Jit-Core: LLVM will be loaded dynamically at runtime.
-- Dr.Jit-Core: CUDA will be loaded dynamically at runtime.
-- Dr.Jit-Core: OptiX support enabled.
-- Dr.Jit-Core: OptiX debug and validation flags disabled.
-- Dr.Jit: building the autodiff backend.
-- Dr.Jit: building the Python plugin.
-- GNU detected
-- Configure ILMBASE Version: 2.5.5 Lib API: 25.0.4
-- Configure OpenEXR Version: 2.5.5 Lib API: 25.0.4
-- Tests are disabled. Set FASTFLOAT_TEST to ON to run tests.
-- Mitsuba: using Embree for CPU ray tracing.
-- Mitsuba: OptiX debug and validation flags disabled.
-- Could NOT find Sphinx (missing: SPHINX_EXECUTABLE)
-- Configuring done
-- Generating done
-- Build files have been written to: /home/jpalathi/mitsuba3/build
[117/1044] Building CXX object ext/openexr/IlmBase/IlmThread/CMakeFiles/IlmThread.dir/IlmThreadPool.cpp.o
/home/jpalathi/mitsuba3/ext/openexr/IlmBase/IlmThread/IlmThreadPool.cpp: In member function ‘virtual void IlmThread_2_5::{anonymous}::NullThreadPoolProvider::setNumThreads(int)’:
/home/jpalathi/mitsuba3/ext/openexr/IlmBase/IlmThread/IlmThreadPool.cpp:459:37: warning: unused parameter ‘count’ [-Wunused-parameter]
  459 |     virtual void setNumThreads (int count)
      |                                 ~~~~^~~~~
[177/1044] Building CXX object ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o
FAILED: ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o
/usr/bin/c++ -DDRJIT_ENABLE_AUTODIFF=1 -DDRJIT_ENABLE_CUDA=1 -DDRJIT_ENABLE_JIT=1 -DDRJIT_UNROLL=" " -Ddrjit_python_EXPORTS -I/home/jpalathi/mitsuba3/ext/drjit/include -I/home/jpalathi/mitsuba3/ext/drjit/ext/drjit-core/include -I/home/jpalathi/mitsuba3/ext/drjit/ext/drjit-core/ext/nanothread/include -isystem /usr/include/python3.10 -isystem /home/jpalathi/mitsuba3/ext/pybind11/include -fdiagnostics-color=always -O3 -DNDEBUG -fPIC -fvisibility=hidden -march=native -Wall -Wextra -Wno-unused-local-typedefs -fno-math-errno -ffp-contract=fast -fno-trapping-math -flto -fno-fat-lto-objects -Os -fno-strict-aliasing -MD -MT ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o -MF ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o.d -o ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm_ad.cpp.o -c /home/jpalathi/mitsuba3/ext/drjit/src/python/bind_llvm_ad.cpp
c++: fatal error: Killed signal terminated program cc1plus
compilation terminated.
[187/1044] Linking CXX shared library libdrjit-core.so
lto-wrapper: warning: using serial compilation of 12 LTRANS jobs
[195/1044] Building CXX object ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm.cpp.o
FAILED: ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm.cpp.o
/usr/bin/c++ -DDRJIT_ENABLE_AUTODIFF=1 -DDRJIT_ENABLE_CUDA=1 -DDRJIT_ENABLE_JIT=1 -DDRJIT_UNROLL=" " -Ddrjit_python_EXPORTS -I/home/jpalathi/mitsuba3/ext/drjit/include -I/home/jpalathi/mitsuba3/ext/drjit/ext/drjit-core/include -I/home/jpalathi/mitsuba3/ext/drjit/ext/drjit-core/ext/nanothread/include -isystem /usr/include/python3.10 -isystem /home/jpalathi/mitsuba3/ext/pybind11/include -fdiagnostics-color=always -O3 -DNDEBUG -fPIC -fvisibility=hidden -march=native -Wall -Wextra -Wno-unused-local-typedefs -fno-math-errno -ffp-contract=fast -fno-trapping-math -flto -fno-fat-lto-objects -Os -fno-strict-aliasing -MD -MT ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm.cpp.o -MF ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm.cpp.o.d -o ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_llvm.cpp.o -c /home/jpalathi/mitsuba3/ext/drjit/src/python/bind_llvm.cpp
c++: fatal error: Killed signal terminated program cc1plus
compilation terminated.
[198/1044] Building CXX object ext/drjit/src/python/CMakeFiles/drjit-python.dir/bind_cuda_ad.cpp.o
ninja: build stopped: subcommand failed.

Steps to reproduce

1.

python3 -m venv nvidia_rm_venv
source nvidia_rm_venv/bin/activate

sudo apt install cmake ninja-build llvm clang libc++-dev libc++abi-dev libpng-dev libjpeg-dev libpython3-dev python3-distutils python3-pytest python3-pytest-xdist python3-numpy

git clone --recursive  https://github.com/mitsuba-renderer/mitsuba3.git --branch v3.5.2
cd mitsuba3/
mkdir build
cd build/
cmake -G Ninja ..

2. vi mitsuba.conf (modified listd named "enabled" to only llvm_ad_mono_polarized, removing everything else) ninja

njroussel commented 4 months ago

Hi @joshuaroy873

I think this is the important bit of the build log:

[Killed signal terminated program cc1plus](c++: fatal error: Killed signal terminated program cc1plus)

Usually this indicative of a lack of resources during compilation. Can you check your memory usage during builds? If I recall correclty, WSL also has some restrictions on the amount of RAM that it has access to. A simple solution usually is to reduce the parallelism if you don't mind wating a bit longer: ninja -j 4 will use only 4 threads which will require less RAM.

joshuaroy873 commented 4 months ago

Latest update: I tried to recreate everything as given in https://mitsuba.readthedocs.io/en/latest/src/developer_guide/compiling.html. Sorry for changing the environment in the middle of an issue.

I installed Ubuntu 20.04 WSL, installed the packages as directed in the website, and used 4 threads to ensure that it wasn't terminated due to lack of resources.

I am getting the following error:

jpalathi@DESKTOP-2V947MI:~/mitsuba3/build$ ninja -j 4
[2/274] Linking CXX executable mitsuba
FAILED: mitsuba
: && /usr/bin/clang++-10  -stdlib=libc++ -D_LIBCPP_VERSION -fcolor-diagnostics -O3 -DNDEBUG  -stdlib=libc++ src/mitsuba/CMakeFiles/mitsuba-bin.dir/mitsuba.cpp.o  -o mitsuba  -Wl,-rpath,/home/jpalathi/mitsuba3/build:/home/jpalathi/mitsuba3/build/ext/drjit/drjit:  libmitsuba.so  -ldl  libasmjit-mitsuba.so  -ldl  libpng-mitsuba.so  /usr/lib/x86_64-linux-gnu/libz.so  -lm  libpugixml.so  libjpeg-mitsuba.so  libIlmImf-mitsuba.so  libImath-mitsuba.so  libHalf-mitsuba.so  libIexMath-mitsuba.so  libIlmThread-mitsuba.so  libIex-mitsuba.so  -pthread  /usr/lib/x86_64-linux-gnu/libz.so  ext/drjit/drjit/libdrjit-extra.so  ext/drjit/drjit/libdrjit-core.so  libnanothread.so  ext/rgb2spec/librgb2spec.a  libembree3.so && :
/usr/bin/ld: libmitsuba.so: undefined reference to `mitsuba::ReconstructionFilter<float, mitsuba::Color<float, 3ul> >::is_box_filter() const'
/usr/bin/ld: libmitsuba.so: undefined reference to `mitsuba::ReconstructionFilter<float, mitsuba::Color<float, 3ul> >::m_class'
clang: error: linker command failed with exit code 1 (use -v to see invocation)
[5/274] cd /home/jpalathi/mitsuba3/build/ext/embree/kernels && /usr/bin/cmake -D file=/home/jpalathi/mitsuba3/build/ext/embree/libembree_avx512.a -P /home/jpalathi/mitsuba3/ext/embree/common/cmake/check_globals.cmake
ninja: build stopped: subcommand failed.

P.S. I also had to install an extra pip package called typing-extensions as it was required for the compilation.

njroussel commented 4 months ago

It looks like you're using the most recent master source files, you should be using the latest stable release:

git clone --recursive  https://github.com/mitsuba-renderer/mitsuba3.git --branch v3.5.2
joshuaroy873 commented 3 months ago

Looks like your right. After trying the stable build, it is giving the c++: fatal error: Killed signal terminated program cc1plus error. This time I monitored my resources, and it is maxing out. I'll try it on another PC with Ubuntu and 32 GB RAM. Thanks for the response till now.

merlinND commented 3 months ago

Hello @joshuaroy873,

Two things to note:

  1. @njroussel mentioned above that you can reduce the parallelism during the build using e.g. ninja -j 4. This will use a lot less RAM, and you will be able to complete the build. For your use-case it should be a one-time thing, so it should be fine if it takes longer. You don't need to change machines.
  2. Note that, AFAIK, OptiX is not supported on WSL. If you have to use a Windows machine, I think it's better to compile directly for Windows (do you agree @njroussel?)
joshuaroy873 commented 3 months ago

Git it working in my Ubuntu workstation. Now, I'm getting a different error:

[440/1047] Building CXX object ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfHuf.cpp.o
FAILED: ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfHuf.cpp.o 
/usr/bin/c++ -DILMIMF_EXPORTS -DIlmImf_EXPORTS -I/home/common/Desktop/Joshua_Files/mitsuba3/build/ext/openexr/OpenEXR/IlmImf -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf -I/home/common/Desktop/Joshua_Files/mitsuba3/build/ext/openexr/OpenEXR/config -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/IlmBase/Iex -I/home/common/Desktop/Joshua_Files/mitsuba3/build/ext/openexr/IlmBase/config -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/IlmBase/Half -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/IlmBase/Imath -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/IlmBase/IexMath -I/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/IlmBase/IlmThread -fdiagnostics-color=always -O3 -DNDEBUG -fPIC -march=native -Wall -Wextra -Wno-unused-local-typedefs -fno-math-errno -ffp-contract=fast -fno-trapping-math -MD -MT ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfHuf.cpp.o -MF ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfHuf.cpp.o.d -o ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfHuf.cpp.o -c /home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp: In function ‘void Imf_2_5::hufUncompress(const char*, int, short unsigned int*, int)’:
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp:1105:5: error: ‘uint64_t’ was not declared in this scope
 1105 |     uint64_t nBytes = (static_cast<uint64_t>(nBits)+7) / 8 ;
      |     ^~~~~~~~
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp:56:1: note: ‘uint64_t’ is defined in header ‘<cstdint>’; did you forget to ‘#include <cstdint>’?
   55 | #include <algorithm>
  +++ |+#include <cstdint>
   56 | 
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp:1107:16: error: ‘nBytes’ was not declared in this scope; did you mean ‘nBits’?
 1107 |     if ( ptr + nBytes > compressed+nCompressed)
      |                ^~~~~~
      |                nBits
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfHuf.cpp:1124:31: error: ‘nBytes’ was not declared in this scope; did you mean ‘nBits’?
 1124 |         if( ptr-compressed  + nBytes > nCompressed)
      |                               ^~~~~~
      |                               nBits
[443/1047] Building CXX object ext/openexr/OpenEXR/IlmImf/CMakeFiles/IlmImf.dir/ImfOutputFile.cpp.o
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfOutputFile.cpp: In function ‘void Imf_2_5::{anonymous}::convertToXdr(Imf_2_5::OutputFile::Data*, Imf_2_5::Array<char>&, int, int, int)’:
/home/common/Desktop/Joshua_Files/mitsuba3/ext/openexr/OpenEXR/IlmImf/ImfOutputFile.cpp:339:19: warning: unused parameter ‘inSize’ [-Wunused-parameter]
  339 |               int inSize)
      |               ~~~~^~~~~~
[457/1047] Linking CXX shared module python/drjit/drjit_ext.cpython-312-x86_64-linux-gnu.so
lto-wrapper: warning: using serial compilation of 128 LTRANS jobs
lto-wrapper: note: see the ‘-flto’ option documentation for more information
ninja: build stopped: subcommand failed.

Any inputs? Thanks.

njroussel commented 3 months ago

Note that, AFAIK, OptiX is not supported on WSL. If you have to use a Windows machine, I think it's better to compile directly for Windows (do you agree @njroussel?)

Yes, I don't know why I didn't of this. There seems to be some unofficial ways to get OptiX to run in WSL, but I'd strill strongly recommend running and building it natively.

njroussel commented 3 months ago

@joshuaroy873 Looks like you're using gcc/g++ >= 13. With gcc 13, there have been some breaking changes. Are you using Ubuntu 24.04?

In any case, you should build with clang. The following command should isntall it:

sudo apt install clang libc++-dev libc++abi-dev

And then you must set it do be your default compiler for the current shell session with:

export CC=clang export CXX=clang++

Finally you can build it with:

mkdir build
cd build
cmake -GNinja ..
ninja
joshuaroy873 commented 3 months ago

My bad. I forgot to include the system information and command history used in my workstation. Please find it below: System information: OS: Ubuntu 24.04 CPU: Intel i7-10700K GPU: GeForce RTX 2060 Python version: Python 3.12.3 LLVM version: 14.0.0 (output to llvm-config --version command) CUDA version: Cuda compilation tools, release 12.0, V12.0.140 Build cuda_12.0.r12.0/compiler.32267302_0 NVidia driver: I have an error when I run nvidia-smi NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running. I'm planning on reinstalling Ubuntu to mitigate this issue now.

Terminal history:

git clone --recursive  https://github.com/mitsuba-renderer/mitsuba3.git --branch v3.5.2
sudo apt install clang libc++-dev libc++abi-dev cmake ninja-build
sudo apt install libpng-dev libjpeg-dev
sudo apt install libpython3-dev
sudo apt install python3-pytest python3-pytest-xdist python3-numpy
cd mitsuba3/
mkdir build
cd build
cmake -GNinja ..
vi mitsuba.conf (to edit "enabled" : ["cuda_ad_mono_polarized"])
ninja

As you mentioned, running export CC=clang export CXX=clang++ did help in mitigating my previous error. Now I have a new error:


[836/1047] Linking CXX executable mitsuba
FAILED: mitsuba 
: && /usr/bin/clang++ -stdlib=libc++ -D_LIBCPP_VERSION -fcolor-diagnostics -O3 -DNDEBUG -stdlib=libc++ src/mitsuba/CMakeFiles/mitsuba-bin.dir/mitsuba.cpp.o -o mitsuba  -Wl,-rpath,/home/common/Desktop/Joshua_Files/mitsuba3/build:  libmitsuba.so  -ldl  libasmjit-mitsuba.so  -ldl  libpng-mitsuba.so  /usr/lib/x86_64-linux-gnu/libz.so  -lm  libpugixml.so  libjpeg-mitsuba.so  libIlmImf-mitsuba.so  libImath-mitsuba.so  libHalf-mitsuba.so  libIexMath-mitsuba.so  libIlmThread-mitsuba.so  libIex-mitsuba.so  /usr/lib/x86_64-linux-gnu/libz.so  libdrjit-autodiff.so  libdrjit-core.so  libnanothread.so  ext/rgb2spec/librgb2spec.a  libembree3.so && :
/usr/bin/ld: libmitsuba.so: undefined reference to `mitsuba::ReconstructionFilter<float, mitsuba::Color<float, 3ul> >::m_class'
/usr/bin/ld: libmitsuba.so: undefined reference to `mitsuba::ReconstructionFilter<float, mitsuba::Color<float, 3ul> >::is_box_filter() const'
clang++: error: linker command failed with exit code 1 (use -v to see invocation)
[853/1047] Linking CXX shared module python/drjit/drjit_ext.cpython-312-x86_64-linux-gnu.so
ninja: build stopped: subcommand failed.

If this is because of the NVidia driver, I'm assuming this'll be gone once I reinstall my system. If not, then I'll reach reach out again.

Thanks.

njroussel commented 3 months ago

This is unrelated to the Nvidia driver, I believe.

I think you've done everything properly, with maybe the exception of the mitsuba.conf edit. Did you replace it to be: 

"enabled" : ["cuda_ad_mono_polarized"]

?

Unfortunately, I think Mitsuba 3 always needs to at least compile scalar_rgb. To be safe, my recommendation would be to compile all the default variants and add cuda_ad_mono_polarized to it:

"enabled" : ["scalar_rgb", "scalar_spectral", "cuda_ad_rgb", "llvm_ad_rgb", "cuda_ad_mono_polarized"]
joshuaroy873 commented 3 months ago

Yes, you're right. I did only enable "cuda_ad_mono_polarized". Thanks for the response. I'll try it out.

joshuaroy873 commented 3 months ago

This is unrelated to the Nvidia driver, I believe.

I think you've done everything properly, with maybe the exception of the mitsuba.conf edit. Did you replace it to be:

"enabled" : ["cuda_ad_mono_polarized"]

?

Unfortunately, I think Mitsuba 3 always needs to at least compile scalar_rgb. To be safe, my recommendation would be to compile all the default variants and add cuda_ad_mono_polarized to it:

"enabled" : ["scalar_rgb", "scalar_spectral", "cuda_ad_rgb", "llvm_ad_rgb", "cuda_ad_mono_polarized"]

Thanks a bunch. This worked. Really helpful.