FS#62258 - [cuda] CUDA samples don't compile on Arch install by default as Arch doesn't use standard paths

Attached to Project: Community Packages
Opened by Julien Chappuis (Jubijub) - Saturday, 06 April 2019, 18:38 GMT
Last edited by Balló György (City-busz) - Wednesday, 10 April 2019, 08:21 GMT
Task Type Bug Report
Category Packages
Status Closed
Assigned To No-one
Architecture x86_64
Severity Medium
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Description: CUDA samples don compile by default as they expect cuda to be installed in /usr/local/cuda and ignore $PATH and $LD_LIBRARY_PATH variables.


Additional info:
* package version(s) : 10.1.105-6

Steps to reproduce:
* pacman -S cuda
* edit path as instructed, and confirm via nvcc -V that CUDA is installed properly, and that both $PATH and $LD_LIBRARY_PATH both contain the proper /opt/cuda paths.
* copy /opt/cuda/samples somewhere (the copy script in samples/bin won't work either)
* go to 1_Utilities/deviceQuery

Actual result :
* make fails with this error message : make: /usr/local/cuda/bin/nvcc: Command not found
* after investigating, the culprit is the Makefile : they contain a reference to CUDA_PATH
(search for)
# Location of the CUDA toolkit
CUDA_PATH ?= /usr/local/cuda

Workaround :
* changing the cuda path to /opt/cuda and make will compile properly

Expected results :
* Default samples should compile without having to tweak Makefiles (this may warrant a bug upstream, as I don see the point of hardcoding this value. the variable CUDA_PATH is used twice in the Makefile, one to indicate the path to NVCC, the other the path to the lib, so it makes $PATH and $LD_LIBRARY_PATH useless (I suspect they put it because nvcc not found seems to be an immensely popular issue)
This task depends upon

Closed by  Balló György (City-busz)
Wednesday, 10 April 2019, 08:21 GMT
Reason for closing:  Duplicate
Additional comments about closing:   FS#62282 
Comment by Zhen Xi (Mayrixon) - Tuesday, 09 April 2019, 00:07 GMT
I think this bug exists but in the other way. The package cuda is not installed in folder /usr but in /opt. In previous buildings, the variable $PATH and $LD_LIBRARY_PATH would be updated automatically, but not updated this time. I was planning to report this bug when I saw your report.

For your convenience, I think you can add the following commands into your .bashrc or .zshrc temporally:
export PATH=/opt/cuda/bin:$PATH
export LD_LIBRARY_PATH=/opt/cuda/targets/x86_64-linux/lib:$LD_LIBRARY_PATH

Loading...