This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PATCH: Enable TFmode for SSE2


On Sun, Jun 29, 2008 at 8:13 AM, Joseph S. Myers
<joseph@codesourcery.com> wrote:
> On Sun, 29 Jun 2008, H.J. Lu wrote:
>
>> I tried to move soft-fp to libgcc. But it doesn't work. The problems are
>>
>> 1. arm and rs6000 also uses soft-fp, which I don't have.
>> 2. gcc/Makefile needs t-soft-fp to generate libgcc.map.
>>
>> Unless we move all libgcc stuff to libgcc, it seems impossible to configure
>> soft-fp in libgcc completely. I will try a hybrid setup. That is configure
>> some of soft-fp in gcc and some in libgcc.
>
> I'm not sure where libgcc.map comes in; as far as I can tell that's
> generated in the libgcc directory (using configuration variables passed
> over from the gcc directory).

It is libgcc.mvars.

>
> I expect that configuring soft-fp in libgcc will require LIB2ADD (=
> LIB2FUNCS_EXTRA) and LIB2FUNCS_EXCLUDE, presently configured in the gcc
> directory with values put in libgcc.mvars, to be moved to the libgcc
> directory, for all targets using either of those variables.  Those are the
> variables modified in soft-fp/t-softfp depending on the target
> configuration that we now wish to depend on the multilib.  It might be
> cleanest to move the soft-fp directory itself, the SFP_MACHINE rules and
> the sfp-machine.h files at the same time.  LIB2FUNCS_EXCLUDE is only used
> by soft-fp, so that part of the move should be straightforward, but lots
> of targets set LIB2FUNCS_EXTRA; converting them should be mechanical if
> tedious.
>

I am leaving soft-fp as is for now. But now I run into

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36667


-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]