rfc and [patch]: Unify -mfpu and .fpu handling, let .fpu toggle features

Renato Golin renato.golin at linaro.org
Fri Aug 15 01:22:01 PDT 2014

On 14 August 2014 19:16, Nico Weber <thakis at chromium.org> wrote:
> Do folks think that this is ok as an incremental improvement than what's
> currently in the tree? On the bug the suggestion was made to just duplicate
> the fpu->feature mapping in two places, which would make this patch much
> smaller (and which would undo the .fpu name fixes), but that seems like a
> worse alternative to me.


You're trying to solve two problems in one patch. The first problem is
an .fpu directive parsing issue, just has to set the bit flag, like
.arch, and the second is sharing the parser and bit flags with clang.

Sometimes, solving one problem not in the best way, and then solving
the other problem later is better from a VCS point of view. Also, they
allow people to discuss the best implementation of the second problem
*after* the original problem was fixed.

I don't think anyone will agree that this is the *right* fix, but as
an interim state, while we discuss a better patter for the second
problem, it's acceptable, with a big FIXME, so that we clean that up
once we agreed on the second problem, that is much greater than the

Of course, if you're ok with the first problem waiting until we have a
better solution for the second problem, than let's not rush ourselves
and wait for a proper fix of both problems.

But my argument so far has never been not to fix the second problem,
just that the fix you propose is not good enough and will create other
problems. And since this discussion touches all other targets
(eventually), we'll have to have a major discussion on the
sub-architecture parsing and representation with other architecture
developers before any decision is taken.


More information about the cfe-commits mailing list