FW: [llvm] r181753 - This is the first of three patches which creates stubs used for

Doug Gilmore Doug.Gilmore at imgtec.com
Tue May 14 16:37:14 PDT 2013


Resending -- sorry for any duplicate messages.

Doug
________________________________________
From: Doug Gilmore
Sent: Tuesday, May 14, 2013 4:28 PM
To: llvm-commits at cs.uiuc.edu
Subject: [llvm] r181753 - This is the first of three patches which creates stubs used for

> Hi
>
> Hi Reed,
>
> I’m confused. There have been multiple very strong objections to
> having the compiler generate inline asm nodes like this. Did I miss
> the discussion where that got resolved?
>
> -Jim
Hi Jim,

I went along with this approach of using inline asm generation since
it allowed us to generate optimal code for the stub functions.

The functions being called contain very little code so it is important
to generate these stub functions as efficiently as possible.

Having this functionality working sooner rather than later speeds up
testing since tracking down regressions using only soft-float can be
very time consuming.  We would really like to attend to the test-suite
regressions so that we start getting clean test-suite runs from our
build-bots for MIPS16.

Do you think we really need to pull and rework the code?  Or can we
put a FIXME comment in the code and attend to it after we fix all of
the MIPS16 test-suite regressions?

Regards,

Doug
>
> On May 13, 2013, at 7:00 PM, Reed Kotler <rkotler at mips.com> wrote:
>
> > Author: rkotler
> > Date: Mon May 13 21:00:24 2013
> > New Revision: 181753
> >
> > URL: http://llvm.org/viewvc/llvm-project?rev=181753&view=rev
> > Log:
> > This is the first of three patches which creates stubs used for
> > Mips16/32 floating point interoperability.





More information about the llvm-commits mailing list