[PATCHES] PR18303: Use appropriate Feature flags for encoding instructions

David Woodhouse dwmw2 at infradead.org
Tue Jan 28 15:21:27 PST 2014


On Tue, 2014-01-28 at 12:58 -0800, David Peixotto wrote:
> > I think we *do* have sufficient consensus and if we just fixed the
> > uniquification of the STI -- or showed that the memory usage wasn't
> > a serious concern in practice so it could reasonably be left as a
> > later optimisation 

FWIW I went for the latter option. On X86 the MCSubtargetInfo is
apparently 104 bytes, and the MCRelaxableFragment therefore grows by
about 32% to 424 bytes. There aren't *that* many fragments since this is
a per-module thing, so in a world where 'ld' linking libLLVM.so can grow
to about 1.5GiB I think I can live with that.

That's not to rule out the optimisation we discussed, but just to
suggest that we don't have to hold up the existing patches (and keep
rebasing and retesting them) while we wait for one of us to get round to
it.


> -- I'd be happy with
> > committing it and waiting to see who shouts at me for it :)

> If you want to take up the mantle and get the patches committed you
> have my support :) They really represent your work and insight
> anyway... I just followed it to the logical conclusion suggested by
> Jim.
> 
> I agree with you that I do not recall any dissent. You could try
> uploading to phabricator to get someone to give it a +1. It seems to
> be a good way to get eyes on the larger patches. I've also seen that
> people tend to do more post-commit reviews as well, so as you said,
> you could try committing and waiting for that feedback.

I can't see any way forward that doesn't at least *start* with this set
of patches. Committed in r200345-r200351.

Thanks for the help with this.

-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse at intel.com                              Intel Corporation
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 5745 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140128/dcab6eb0/attachment.bin>


More information about the llvm-commits mailing list