On 11/19/2015 12:02 PM, Manuel López-Ibáñez wrote:
On 19 November 2015 at 17:54, Jeff Law <l...@redhat.com> wrote:
But there were a couple of patches from you some time ago, for
example: http://permalink.gmane.org/gmane.comp.gcc.patches/343476
What happened with those?
On hold pending fixing the type-limits warning placement. Essentially that
has to be untangled first.
Could you elaborate on this? Or point me to some previous email
thread? (I don't have enough free time to follow the mailing list
anymore, sorry).
I don't have the thread handy, but essentially the operand shortening
code has warning bits inside it. As a result pulling out functionality
(such as operand canonicalization) and putting into match.pd results in
missing a warning -- ie a regression.
So we have to detangle the operand shortening from warning detection.
Kai's idea was to first make the shortening code "pure" in the sense
that it would have no side effects other than to generate the warnings.
Canonicalization and other transformations would still occur
internally, but not be reflected in the IL.
That was the overall plan and he posted a patch for that. But that
patch didn't do the due diligence to verify that once the shortening
code was made "pure" that we didn't regress on the quality of the code
we generated.
jeff
