Hi!

I want to use EigensBlas implementation with SuperLU.
So in CMake I configure and generate with Visual Studio as generator. During configuration I get the massage:
"No fortran compiler has been detected, the blas build will be incomplete."
I tried to specify the fortran compiler in CMake giving Path to C:/MinGW/bin/ andalso added this folder to the environment-variables. Got the same error message. Nevertheless the generation as Visual-Studio-Solution works and I get a eigen_blas_static.lib around 16.6 kb by compiling with VS. Is this a fully working fastBlas-library for my purpose, or do I have to get fortran compiler working and how do I do this?

Greetings, medic

Okay got it. It turns out, that CMake has very limited support for exsisting Fortran Compilers. I downloaded a Testversion of IFort and the error disappeared. Now Eigen Blas compiled in two libraries. A eigen_blas.dll(19kb) and a eigen_blas_static(16mb). Are they both needed. Or does the dll only include the static library? How is it intended to use these two?

you need to use only one, the dll is for dynamic linking (and sharing), the other for static linking only.

btw, even without a fortran compiler the produced BLAS is complete enough for most use cases.

I have g++ and gfortran. Is it better to translate Fortran code into C++ or to leave it alone?

Which fortran code?

ggael wrote:Which fortran code?

Like srotm.f, srotmg.f, drotm.f, drotmg.f. They seems to have no counterparts in Eigen. And sorry, I don't understand some points in the BLAS:

• Both xDOTC and xDOTU are to perform op(x)*y i.e. y.dot(x). However, in the code is x.dot(y) and passes the test.
• Why return int for subroutines and what do the return values mean? Some of them return 1(0) for success(failure), but the others always return 0.

ah! I think there is no real need to translate them to c++ as long as there us no need of them for Eigen. However, regarding the band and packed formats, we clear do want to have optimized version in Eigen, but that requires more work.