์ต์ ์์ค๋ก pytorch์ CUDA ๋ฒ์ (CUDNN ์ ์ธ)์ ๋น๋ํ์ง ๋ชปํ์ต๋๋ค.
OS: ๋ฐ๋น์ ๋ถ์์ /์คํ์
์ปดํ์ผ๋ฌ: gcc-5, g++-5
CUDA: 8.0.44(๋ฐ๋น์์์ ์ ๊ณตํ๋ ํจํค์ง)
๋น๋๋ก๊ทธ: http://debomatic-amd64.debian.net/distribution#experimental/pytorch -contrib/0.1.7~1/buildlog
-- The C compiler identification is GNU 5.4.1
-- The CXX compiler identification is GNU 5.4.1
-- Check for working C compiler: /usr/bin/gcc-5
-- Check for working C compiler: /usr/bin/gcc-5 -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/g++-5
-- Check for working CXX compiler: /usr/bin/g++-5 -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found CUDA: /usr (found suitable version "8.0", minimum required is "7.0")
-- Configuring done
-- Generating done
-- Build files have been written to: /<<PKGBUILDDIR>>/torch/lib/build/nccl
make[2]: Entering directory '/<<PKGBUILDDIR>>/torch/lib/build/nccl'
make[3]: Entering directory '/<<PKGBUILDDIR>>/torch/lib/build/nccl'
make[4]: Entering directory '/<<PKGBUILDDIR>>/torch/lib/build/nccl'
Scanning dependencies of target nccl
make[4]: Leaving directory '/<<PKGBUILDDIR>>/torch/lib/build/nccl'
make[4]: Entering directory '/<<PKGBUILDDIR>>/torch/lib/build/nccl'
[100%] Generating lib/libnccl.so
make[5]: Entering directory '/<<PKGBUILDDIR>>/torch/lib/nccl'
ls: cannot access '/usr/lib64/libcudart.so.*': No such file or directory
ls: cannot access '/usr/lib64/libcudart.so.*': No such file or directory
Grabbing src/nccl.h > /<<PKGBUILDDIR>>/torch/lib/build/nccl/include/nccl.h
Compiling src/libwrap.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/libwrap.o
Compiling src/core.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/core.o
Compiling src/all_gather.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/all_gather.o
Compiling src/all_reduce.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/all_reduce.o
Compiling src/broadcast.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/broadcast.o
Compiling src/reduce.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/reduce.o
Compiling src/reduce_scatter.cu > /<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/reduce_scatter.o
src/core.cu(724): error: expected an expression
src/core.cu(724): error: expected an expression
2 errors detected in the compilation of "/tmp/tmpxft_00002c02_00000000-13_core.compute_52.cpp1.ii".
Makefile:98: recipe for target '/<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/core.o' failed
make[5]: *** [/<<PKGBUILDDIR>>/torch/lib/build/nccl/obj/core.o] Error 2
๋๋ ์ด๊ฒ์ ๋ํด ์๋ฌด ์๊ฐ์ด ์๋ค ...
์ฌ๊ธฐ ๋ฅผ ๋ณด๋ฉด ls
๊ฐ libcudart.so
๋ฅผ ์ฐพ์ ์ ์๊ธฐ ๋๋ฌธ์ CUDA_MAJOR
๋ฐ CUDA_MINOR
๊ฐ ๋น์ด ์๊ธฐ ๋๋ฌธ์ ์คํจํ๋ ๊ฒ์ ์ ์ ์์ต๋๋ค. ๋ฌธ์์ด. ๋ฐ๋น์ CUDA ํจํค์ง๊ฐ ์ด๋ป๊ฒ ์๊ฒผ๋์ง ๋ชจ๋ฅด๊ฒ ์ง๋ง ์๋ง๋ ๋ฒ์ ํ์ฅ์ด ์๋ libcudart.so
๊ฐ ์์ ๊ฒ์
๋๋ค.
CUDA_VERSION=8.0 pp setup.py build
์ด ๋์์ด ๋ ๊น์?
@apaszke ๊ฐ์ฌํฉ๋๋ค. ์์ ์ ๋ ๊ฐ์ง ํ๊ฒฝ ๋ณ์๋ฅผ ๋ด๋ณด๋ด๋ ๊ฒ์ ๋๋ค.
export CUDA_HOME=/usr
export CUDA_LIB=/usr/lib/$(shell dpkg-architecture -qDEB_HOST_MULTIARCH)
๊ฐ์ฅ ์ ์ฉํ ๋๊ธ
@apaszke ๊ฐ์ฌํฉ๋๋ค. ์์ ์ ๋ ๊ฐ์ง ํ๊ฒฝ ๋ณ์๋ฅผ ๋ด๋ณด๋ด๋ ๊ฒ์ ๋๋ค.