I have clonned v1.13.0 and trying to build pytorch with CUDA 3.5 support.
$ nvidia-smi
Wed Dec 7 18:38:18 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.141.03 Driver Version: 470.141.03 CUDA Version: 11.4 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... Off | 00000000:01:00.0 N/A | N/A |
| N/A 36C P8 N/A / N/A | 152MiB / 2001MiB | N/A Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
$ USE_SYSTEM_LIBS=1 MAX_JOBS=2 TORCH_CUDA_ARCH_LIST=3.5 python setup.py install
...
-- Configuring incomplete, errors occurred!
See also "/tmp/pytorch/build/CMakeFiles/CMakeOutput.log".
See also "/tmp/pytorch/build/CMakeFiles/CMakeError.log".
$ tail build/CMakeFiles/CMakeError.log
Run Build Command(s):/usr/bin/ninja cmTC_8db07 && [1/2] Building CXX object CMakeFiles/cmTC_8db07.dir/src.cxx.o
FAILED: CMakeFiles/cmTC_8db07.dir/src.cxx.o
/usr/bin/c++ -DHAS_FCOLOR_DIAGNOSTICS -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fPIE -fcolor-diagnostics -std=gnu++14 -o CMakeFiles/cmTC_8db07.dir/src.cxx.o -c /tmp/pytorch/build/CMakeFiles/CMakeTmp/src.cxx
c++: error: unrecognized command-line option '-fcolor-diagnostics'
ninja: build stopped: subcommand failed.
Source file was:
int main() { return 0; }
$ c++ --version
c++ (Gentoo 11.3.0 p7) 11.3.0
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
$ ninja --version
1.11.1
$ cmake --version
cmake version 3.24.3
I’m stuck at this point.
UPD: gcc has an option called fdiagnostics-color (not -fcolor-diagnostics), is it cmake bug?
UPD2: looks like problem with Cannot add target-level dependencies to non-existent target "gloo_cuda". Will try master branch…
-- Found uv: 1.38.1 (found version "1.38.1")
Found gloo: /usr/lib64/libgloo.so
CMake Error at cmake/Dependencies.cmake:1500 (add_dependencies):
Cannot add target-level dependencies to non-existent target "gloo_cuda".
The add_dependencies works for top-level logical targets created by the
add_executable, add_library, or add_custom_target commands. If you want to
add file-level dependencies see the DEPENDS option of the add_custom_target
and add_custom_command commands.
Call Stack (most recent call first):
CMakeLists.txt:715 (include)