This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: ppc64 floating point usage
On Fri, Aug 08, 2003 at 11:15:53PM -0400, David Edelsohn wrote:
> The patch to ignore FPR preferences fixed a oversight in the
> original patterns. The register cost model is about tuning and tuning is
> based on some metric. If you want to equate measuring the metric with
> symptom, fine, but the problem is maximizing (optimizing) some function.
OK, I think I'm beginning to make sense of this. In the case where the
result of a fp -> int conversion is live for any appreciable time, we'd
much rather it lived in a gpr reg. It's just a pity that when the
result is written immediately to a stack slot that there is then no
preference for fpr regs, as that prevents me changing the reg alloc
order.
--
Alan Modra
IBM OzLabs - Linux Technology Centre