Re: compiler-rt and softfloat

[Masao Uebayashi <uebayasi%gmail.com@localhost>]

On Wed, Jun 1, 2011 at 5:03 PM, Joerg Sonnenberger
<joerg%britannica.bec.de@localhost> wrote:
> On Wed, Jun 01, 2011 at 10:20:37AM +0900, Masao Uebayashi wrote:
>> I think so.  I also think that pcc should use the one in pcc-libs.
>
> I disagree.
>
>> These helper functions are the internal ideas compilers have - which
>> is hard to understand and emulate by outsiders.  It's compiler's
>> responsibility to provide helper funcs IMO.
>
> No, actually, they are a pretty consistent, if underspecified API. There
> is little reason to reinvent the wheel for every compiler/platform
> combination.

"Underspecified" - yes, but it's much better now than a few years ago.

"Reinvent the wheel" - that is a problem of theirs, not ours.

What I don't like is managing local changes against them.  We're
picking up part of libgcc funcs (to use libc/softfloat), which is hard
to maintain.  This is why I tried to export all functions from libgcc
in matt-nb5-mips64.  Again, my point is that compilers provide helper
funcs, we should use them as is.

If GCC specifies libgcc API clearly, and compiler-rt people implements
100% compatible versions (for all GCC versions), that's simply great.