math.h / cmath isinf problem.

[obm]

Dear all,
When I am trying to compile magma with intel compilers (17 up 5), the compilation fails, saying math.h isinf and isnan functions are not recognized.
Actually this problem is known: https://software.intel.com/en-us/forums ... pic/638013

It seems a workaround solution is to bake a macro in the source code such that the correct math.h is included.

Maybe you can consider this workaround in your source tree?

[mgates3]

Can you give the specific error output, when doing 'make -j1' (i.e., single-threaded make). That will be helpful to understand where the error occurs.
-mark

[ibnu_syuhada_p3]

I also meet this problem. I am not using intel compiler, but using gnu compiler

[mgates3]

Again, please provide some specifics such as your platform, g++ version, and the output of "make -j1" showing the error, so that we can attempt to reproduce the issue.
-mark

[brianborchers]

This appears to be the same problem that I reported this evening with testing/testing_zlange.cpp. I've given more specific error messages in my report.

[mgates3]

The link on the Intel site seems to be the *opposite* problem, namely that using std::isinf and std::isnan fails when including both math.h and cmath, with Intel icc 16 + gcc 6. That problem I can replicate. It doesn't occur for icc 17 or 18.

I haven't been able to replicate the problem that using isinf and isnan (without std:: prefix) fails. MAGMA includes math.h, so isinf and isnan should be defined.

See further discussion at viewtopic.php?f=2&t=1659

-mark

[mgates3]

This problem should be resolved now in the bitbucket repo. The MAGMA testers include cmath and use std::isinf, std::isnan, and std::copysign.
-mark