On 11/20/13 03:02, Richard Biener wrote:
Note that this, the intrusiveness of the feature and the questionable
gain makes me question whether GCC should have support for this
feature (and whether we really should rush this in this late).
Thus, I hereby formally ask to push back this feature to 4.10.
Sigh. I'd hoped we were making progress and Ilya could have things
wrapped up in a reasonable amount of time. But I certainly see your
point of view and I have some concerns about the semantics of the
builtins now that we're getting deeper into the bits.
The patches were posted long ago (back in mid Sept) and received
little/no feedback at that time. Ilya played by the rules and it was
our failing as maintainers that caused things to back up. Thus I
believe the code should be given fair consideration for inclusion into 4.9.
--
I suspect the hardware implementation and ABI are largely set by the
need to interoperate with uninstrumented code. Where I think the
patchset falls down is in implementation details.
Anyway, if you're going to stick with your formal request to postpone
until after 4.9, I'm not going to push hard from the other direction.
Given that, we should probably pull out the half-dozen preparatory
patches that went in.
jeff