@ahmadia, @cekees what is the best way to handle this? The -Wa,-q flag needs to be added to every package that is compiled with gcc and with -march=native on OS X. Without it, the linking fails with missing AVX instructions (the flag forces gcc to use the correct OS X linker which understands these instructions):

[bddc] /var/tmp//ccYZ86gg.s:46:no such instruction: `vmovsd (%r14,%rax,8), %xmm0'
[bddc] /var/tmp//ccYZ86gg.s:57:no such instruction: `vmovsd (%r11,%r9,8), %xmm1'
[bddc] /var/tmp//ccYZ86gg.s:61:no such instruction: `vcomisd %xmm0, %xmm1'
[bddc] /var/tmp//ccYZ86gg.s:69:no such instruction: `vmovsd (%r11,%r15,8), %xmm2'
[bddc] /var/tmp//ccYZ86gg.s:74:no such instruction: `vcomisd %xmm2, %xmm0'
[bddc] /var/tmp//ccYZ86gg.s:87:no such instruction: `vmovsd (%r9), %xmm3'
...

So only if gcc and OS X is used, the flag should be set in all CXXFLAGS, CFLAGS and FFLAGS. The problem only manifests if the -march=native option is used, but I think it doesn't hurt to use the flag always.

One way is to set this in the base template yaml for OS X, and then change all Hashstack to only keep appending to CXXFLAGS, CFLAGS and FFLAGS, instead of replacing what was there (the current approach).

I'm not familiar with the -Wa,-q option. If it doesn't hurt anything, maybe we should have an os x profile that adds it and -march=native to the prologue.

Right. Except that we still need to update all Hashstack to keep appending to CXXFLAGS, CFLAGS and FFLAGS as I mentioned above.