Bug#725649: nvcc fatal : Path to libdevice library not specified
Dnia 2013-10-10, czw o godzinie 23:30 +0200, Andreas Beckmann pisze: On 2013-10-08 19:11, Tomasz Rybak wrote: It seems error lies with nvidia-cuda-toolkit. I've downloaded and tried to compile NVIDIA examples and got: $ make make[1]: Wejście do katalogu to get English output use LC_ALL=C make in the future, no need to rerun it this time so PyCUDA is just victim of this error. PyOpenCL works without problems so it seems that fault lies with nvcc, not with libraries responsible for compilation of kernels. There is probably some setting missing in /etc/nvcc.* (has the upstream config file some new entries in 5.5 that I forgot to add to the Debian customized one?) You were right. There is new entry NVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice after adding line NVVMIR_LIBRARY_DIR = /usr/lib/nvidia-cuda-toolkit/libdevice to nvcc.profile PyCUDA works without any problems. How can I easily reproduce the problem? BTW - please add Conflicts: libcuinj64-5.0 to libcuinj64-5.5, as both provide the same file which file? probably the .so link, so needs to have Breaks+Replaces libcuinj64.so: Selecting previously unselected package libcuinj64-5.0:amd64. (Reading database ... 720002 files and directories currently installed.) Unpacking libcuinj64-5.0:amd64 (from .../libcuinj64-5.0_5.0.35-7_amd64.deb) ... dpkg: error processing /var/cache/apt/archives/libcuinj64-5.0_5.0.35-7_amd64.deb (--unpack): trying to overwrite '/usr/lib/x86_64-linux-gnu/libcuinj64.so', which is also in package libcuinj64-5.5:amd64 5.5.22-1 Regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak signature.asc Description: This is a digitally signed message part
Bug#725649: nvcc fatal : Path to libdevice library not specified
On 2013-10-08 19:11, Tomasz Rybak wrote: It seems error lies with nvidia-cuda-toolkit. I've downloaded and tried to compile NVIDIA examples and got: $ make make[1]: Wejście do katalogu to get English output use LC_ALL=C make in the future, no need to rerun it this time so PyCUDA is just victim of this error. PyOpenCL works without problems so it seems that fault lies with nvcc, not with libraries responsible for compilation of kernels. There is probably some setting missing in /etc/nvcc.* (has the upstream config file some new entries in 5.5 that I forgot to add to the Debian customized one?) How can I easily reproduce the problem? BTW - please add Conflicts: libcuinj64-5.0 to libcuinj64-5.5, as both provide the same file which file? probably the .so link, so needs to have Breaks+Replaces and I had to manually remove the former and unpack the latter during installation of CUDA 5.5 I'll further investigate these issues next week (after holidays). Andreas -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#725649: nvcc fatal : Path to libdevice library not specified
It seems error lies with nvidia-cuda-toolkit. I've downloaded and tried to compile NVIDIA examples and got: $ make make[1]: Wejście do katalogu `/home/tomus/Programs/CUDA/0_Simple/simpleCubemapTexture' /usr/bin/nvcc -ccbin g++ -I../../common/inc -m64-gencode arch=compute_20,code=sm_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=\sm_35,compute_35\ -o simpleCubemapTexture.o -c simpleCubemapTexture.cu nvcc fatal : Path to libdevice library not specified make[1]: *** [simpleCubemapTexture.o] Błąd 255 make[1]: Opuszczenie katalogu `/home/tomus/Programs/CUDA/0_Simple/simpleCubemapTexture' make: *** [0_Simple/simpleCubemapTexture/Makefile.ph_build] Błąd 2 so PyCUDA is just victim of this error. PyOpenCL works without problems so it seems that fault lies with nvcc, not with libraries responsible for compilation of kernels. BTW - please add Conflicts: libcuinj64-5.0 to libcuinj64-5.5, as both provide the same file and I had to manually remove the former and unpack the latter during installation of CUDA 5.5 Best regards. -- Tomasz Rybak GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak signature.asc Description: This is a digitally signed message part
Bug#725649: nvcc fatal : Path to libdevice library not specified
Package:nvidia-cuda-toolkit Version: 5.5.22-1 Running any example with PyCUDA results in: % python hello_gpu.py Traceback (most recent call last): File hello_gpu.py, line 12, in module mod = SourceModule(open(hello_gpu.cu).read()) File /usr/lib/python2.7/dist-packages/pycuda/compiler.py, line 262, in __init__ arch, code, cache_dir, include_dirs) File /usr/lib/python2.7/dist-packages/pycuda/compiler.py, line 252, in compile return compile_plain(source, options, keep, nvcc, cache_dir) File /usr/lib/python2.7/dist-packages/pycuda/compiler.py, line 134, in compile_plain cmdline, stdout=stdout.decode(utf-8), stderr=stderr.decode(utf-8)) pycuda.driver.CompileError: nvcc compilation of /tmp/tmpDX0YZu/kernel.cu failed [command: nvcc --cubin -arch sm_21 -I/usr/lib/python2.7/dist-packages/pycuda/cuda kernel.cu] [stderr: nvcc fatal : Path to libdevice library not specified ] nvidia-cuda toolkit 5.0.35-7 works fine. I'm not sure if this is on PyCUDA or nvidia-cuda-toolkit's side. Apologies in advance if it's PyCUDA. Regards, Vedran -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
