Closed amellnik closed 8 years ago
Titling any issue "quick fix" is just asking for trouble. After correcting the issue about it gives me an error on a system without OpenBLAS on the path:
blas_test.cpp:1:19: fatal error: cblas.h: No such file or directory
#include <cblas.h>
^
compilation terminated.
I'm adding it now to see if it will build with it included.
The problem is that I tested this with a very recent version and it looks like you are testing with 0.4 where the find_library interface has changed. In 0.5, it looks like find_library(ASCIIString) and in 0.4 it expects an array. I see in 0.5 they added a convenience function for just one library check. I will check-in a fix for this.
Maybe I should capture the output so you don't see that cblas error but it isn't really an error, just a way of detecting that the system compiler doesn't support blas.
Ah good catch. Is ParallelAccelerator moving to require 0.5+ or will it continue to support 0.4?
I spoke too soon about OpenBLAS -- I do have it on this system and the folder with cblas.h
is on my path but I'm still seeing that error.
We don't have plans to drop 0.4 support in the immediate future.
We search for OpenBLAS support by looking for libopenblas64_ in the library search path. We test for blas support in the compiler itself by creating a cpp file that includes cblas.h and then trying to compile that. So, OpenBLAS and the cblas.h error are independent.
Closed with 7f7b4bb..6d24eed.
I went to do some benchmarks today but I'm seeing a new error building on master:
I think the names of the libraries here on lines 71 and 72 need to be enclosed in arrays rather than as single strings. I can submit a pull request at lunch if no one beats me to it. Best -A