Hi Jakub & Andrew,

on 2023/12/12 22:42, Jakub Jelinek wrote:
> On Tue, Dec 12, 2023 at 09:33:38AM -0500, Andrew MacLeod wrote:
>> I leave this for the release managers, but I am not opposed to it for this
>> release... It would be nice to remove it for the next release
> 
> I can live with it for GCC 14, so ok, but it is very ugly.

Thanks, pushed as r14-6478-gfda8e2f8292a90.

And yes, I strongly agree that we should get rid of this in next release.

> 
> We should fix it in a better way for GCC 15+.
> I think we shouldn't lie, both on the mode precisions and on type
> precisions.  The middle-end already contains some hacks to make it
> work to some extent on 2 different modes with same precision (for BFmode vs.
> HFmode), on the FE side if we need a target hook the C/C++ FE will use
> to choose type ranks and/or the type for binary operations, so be it.
> It would be also great if rs6000 backend had just 2 modes for 128-bit
> floats, one for IBM double double, one for IEEE quad, not 3 as it has now,
> perhaps with TFmode being a macro that conditionally expands to one or the
> other.  Or do some tweaks in target hooks to keep backwards compatibility
> with mode attribute and similar.

Thanks for all the insightful suggestions, I just filed PR112993 for
further tracking and self-assigned it.

BR,
Kewen

Reply via email to