Some programs depend on common CUDA make rule, but CUDA_INSTALL_PATH by default is set to /usr/local/cuda instead of /opt/cuda, thus causing build failures. Reproducible: Always Steps to Reproduce: 1. install dev-util/nvidia-cuda-sdk 2. try installing or compiling by hands some programs dependent on common make rule (e.g. app-crypt/cuda-rarcrypt from pentoo overlay) Actual Results: build failures like "make: /usr/local/cuda/bin/nvcc: Command not found" or "rarcrypt_cuda.cpp:6:26: fatal error: cuda_runtime.h: No such file or directory" However I first noticed this issue on this particular package (app-crypt/cuda-rarcrypt from pentoo overlay), it might affect other packages. trivial patch for this: diff -Naurp common-orig/common.mk common/common.mk --- common-orig/common.mk 2012-01-14 00:10:38.000000000 +0400 +++ common/common.mk 2012-04-22 22:26:09.689872611 +0400 @@ -23,7 +23,7 @@ # Add new SM Versions here as devices with new Compute Capability are released SM_VERSIONS := 10 11 12 13 20 21 -CUDA_INSTALL_PATH ?= /usr/local/cuda +CUDA_INSTALL_PATH ?= /opt/cuda ifdef cuda-install CUDA_INSTALL_PATH := $(cuda-install)
+*nvidia-cuda-sdk-4.2.9-r2 (23 Jan 2013) + + 23 Jan 2013; Justin Lecher <jlec@gentoo.org> + +nvidia-cuda-sdk-4.2.9-r2.ebuild: + Correct installatoion path of cuda in common.mk, #413137 +