PPC970: fix scal

[barracuda156]

@martin-frbg This seems to work for py-scipy. Does it look reasonable? Can I test it more specifically, besides scipy building?

Disclaimer: I cannot test this right now for Darwin ppc64, so that may be or not be broken. The fix is specific to Darwin ppc (32-bit ABI).

[barracuda156]

P. S. I do not know what is NeoverseN1-OMP in CI, but it uses ARM flags, so arguable cannot be affected by a PowerPC-specific change.

[barracuda156]

@martin-frbg I will do a clean rebuild from the tip of the develop now to confirm again this works.

[martin-frbg]

Thanks, looks convincing enough for me. The failed CI job happened to be dropped by the cloud provider shortly before completion and should reschedule soon

[barracuda156]

@martin-frbg A bit on a side note, is it possible to add a CI for Linux ppc32 (or FreeBSD / OpenBSD /NetBSD)? Would be useful, IMO.

[martin-frbg]

I can look into this, but it would certainly have to be emulated by qemu

[barracuda156]

py312-scipy build works normally with OpenBLAS from 6f9b9e2 plus my commits from this PR, so I think this is good to go. May not be perfect, but apparently fixes the immediate problem.