I previously thought that the Intel standard for complex return values on windows was the default for the platform, and so disabled the shims to work around it. However, it turns out that gfortran uses a consistent ABI across platforms, and we want to standardize on that, so let's enable this again. This should fix the memory corruption seen when calling MKL's cdotc_64 on Windows x86_64.
@ViralBShah The saga continues, but I think this should fix the MKL issue on Windows x86_64. The failure here on x86_64 is because we have a bad nightly, that should be fixed soon when new nightlies roll out.
I previously thought that the Intel standard for complex return values on windows was the default for the platform, and so disabled the shims to work around it. However, it turns out that gfortran uses a consistent ABI across platforms, and we want to standardize on that, so let's enable this again. This should fix the memory corruption seen when calling MKL's
cdotc_64
on Windows x86_64.