Compiler Flags

[ranocha]

Do we really want to set all the compiler flags of

toolkitICL/CMakeLists.txt

Line 30 in acb8f5d

SET(CMAKE_CXX_FLAGS " -std=c++11 -O2 -funroll-loops -fstrict-aliasing -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -mavx -mavx2 -fPIC -Wall -mfpmath=sse -Wcast-align -Wl,-E")

In particular, options such as -mavx2 are problematic if the target CPU does not support them. This caused some errors while enabling Mac OS X in #14.

Some reasons for the current choice are probably connected with clusters, where build machines have not the same architecture as the compute nodes. But this could be caught by some other variables passed to cmake. What is your opinion, @philipheinisch?

[philipheinisch]

Good question, one solution would be to automatically detect CPU features and enable appropriate flags automatically – but this could cause problems on compute clusters. Interestingly GCC just ignored unsupported flags for example on ARM. Maybe an internal lookup table for different CPU architectures?

[ranocha]

That might be possible but sounds rather complicated.

[ranocha]

We will check whether the host code is performance critical (probably it is not) and adapt the compiler flags accordingly, e.g. by setting only -march=native.

[ranocha]

Closed in #30.