fix fabs() to not use x87 for float/double

[WalterBright]

This is a partial fix for https://issues.dlang.org/show_bug.cgi?id=19663

A complete fix would be to make it a compiler intrinsic.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + phobos#7561"

[thewilsonator]

is this Endian safe?

[WalterBright]

It should be fine. I added unittests to verify.

[ibuclaw]

The CI wouldn't really catch that though, as it's all LittleEndian.

However... FYI @thewilsonator, all targets I'm aware of use same endianness for both float and integer words, regardless whether the it's a BigEndian or LittleEndian architecture. The one exception to the rule (there's always one) is the PDP-11, and I doubt Phobos will ever support that. :-)

[thewilsonator]

ditto

[WalterWaldron]

Since it's still round tripping through the stack core.bitop.btr is an option, but it's not a big improvement.

    import core.bitop:btr;
    version (BigEndian) btr(cast(ulong*)&d,7);
    else                btr(cast(ulong*)&d,63);
    return d;

Link to compare functions & compilers: https://d.godbolt.org/z/bvd1Ev

[ibuclaw]

Since it's still round tripping through the stack core.bitop.btr is an option, but it's not a big improvement.

Link to compare functions & compilers: https://d.godbolt.org/z/bvd1Ev

To be fair, the compiler can recognize the pattern and emit btr all by itself. https://d.godbolt.org/z/r5nWM4