Separate implementations of AMAX and IAMAX on arm

[martin-frbg]

As noted in #1912 and comment on #2142, the combined implementation happens to "do the right thing" on hardfp, but cannot return both value and index on softfp where they would have to share the return register

[ashwinyes]

The implementation amax_vfp.S will not work for double precision or complex cases (both double and single).

[martin-frbg]

Right, so far this is only an attempt to fix SAMAX for #1912 without breaking ISAMAX. So how would I handle double precision, would a single "vmov r0 r1 d0" be sufficient ? (I have even less of an idea how to handle the complex case, on my own I would probably drop them back to their generic C implementations)

[ashwinyes]

For Double, see nrm2_vfp.S. "vmov r0, r1, d0" should be enough.

For Complex case, it would be better to write a file name izamax_vfp.S. For now move back to generic C implementation.

[martin-frbg]

Thank you - that is actually where I got the idea from. It will still be some time until I manage to write any meaningful assembly from scratch.