[llvm-commits] patch: use movd for 32-bits and movq for 64-bits consistently

[Nick Lewycky]

Ping!

This is blocking hand-written assembly in OpenSSL from assembling. My
initial reaction was to fix openssl, but upon further investigation it looks
like llvm is wrong and openssl has it right. Please take a look at this, the
patch was attached to this mail:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20110613/122254.html

Nick

On 13 June 2011 21:29, Nick Lewycky <nlewycky at google.com> wrote:

> I don't understand why we had an alternative for movq/movd. As far as I can
> tell on both x86-32 and x86-64 we're supposed to be using movq for
> xmm<->64-bit memory or register and movd for xmm<->32-bit memory or
> register. This patch removes the alternatives from the .td file and updates
> the test.
>
> It seems likely to me that I'm missing something important here, but
> everything seems to just work for me with this patch applied, both in x86-64
> and x86-32 builds.
>
> This is intended to fix PR10134 but for reference PR2677 and PR2108 are
> also about movd and movq so you may want to look at them too. I'd appreciate
> code review, in particular from Evan and Anton who've commented on bugs
> around this issue in the past.
>
> Nick
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20110616/82022fcd/attachment.html>