Cublas active

[Breush]

No description provided.

[Breush]

Changes to this file should be reverted

[Breush]

Include to be removed

[dlesnoff]

I did not achieve to compile the dgemm benchmark with this pull request.

configuration of the library

./configure --prefix=$PWD
[...]
checking for CUDA >= 5.5.0 ... unknown
WARNING: You appear to be cross compiling, so there is no way to determine
whether your CUDA version is new enough. I am assuming it is.
unknown
WARNING: You appear to be cross compiling, so there is no way to determine
whether your CUDA version is new enough. I am assuming it is.
[...]

The CUDA version check is buggy as I am not cross compiling.
I am compiling fflas-ffpack on my desktop computer with CUDA 12.8 installed.
Givaro and OpenBLAS are automatically detected.

Then I compile fflas-ffpack and I have no problem doing so.

make
make install

DGEMM benchmark

cd benchmarks && make benchmark-dgemm
g++ -DHAVE_CONFIG_H -I. -I..    -O2  -Wall -g -I..  -fabi-version=6  -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -mavx -mavx2 -mavx512f -mavx512vl -mavx512dq -mfma    -fopenmp -MT benchmark-dgemm.o -MD -MP -MF .deps/benchmark-dgemm.Tpo -c -o benchmark-dgemm.o benchmark-dgemm.C
In file included from benchmark-dgemm.C:28:
../fflas-ffpack/config-blas.h:49:10: fatal error: cuda.h: No such file or directory
   49 | #include <cuda.h>
      |          ^~~~~~~~
compilation terminated.
make: *** [Makefile:1112: benchmark-dgemm.o] Error 1

I see no -lcuda nor -lcudart in the g++ invocation call by make
(as expected from the flags defined in the automake macro).

(EDIT: remove plenty of additionnal remarks to simplify the post).

Useful information

Stack Overflow: how to get the cuda version
Stack Overflow: how can I compile cuda code then link it to a C project