[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [patches] Including IBM add-ons in EGLIBC?



I agree with Joseph, it is a good idea to add powerpc-cpu and DFP to
eglibc.


On Thu, 2006-11-16 at 21:17 +0000, Joseph S. Myers wrote:
> On Thu, 16 Nov 2006, Mark Mitchell wrote:
> 
> > Geoff Smith at IBM asked me to forward the following note about possibly
> > functionality to include in EGLIBC.
> > 
> > Perhaps the EGLIBC maintainers could comment?
> 
> I think adding these to EGLIBC 2.5 and EGLIBC trunk is a good idea - 
> subject to all the assignments to the FSF being in place.  As Steve Munroe 
> is a maintainer he can add the powerpc-cpu add-on himself.
> 
> In the case of the DFP add-on, which overrides some files from core glibc, 
> I would encourage patching those files in the libc/ directory directly 
> instead for the EGLIBC version (conditional on appropriate feature test 
> macros for the headers, and with some other condition for the files that 
> get compiled into libc - for example, have SUPPORT_DECIMAL_FLOAT 
> conditionals in libc/stdio-common/vfprintf.c, and have 
> dfp/sysdeps/dfp/stdio-common/vfprintf.c define SUPPORT_DECIMAL_FLOAT then 
> include the file from libc).  These patches to core libc would need 
> posting separately for review.
> 
> > > Possible work -- add optimizations for Freescale's E500 cpus, extend to 
> > > non-powerpc CPUs
> 
> I intend to implement the basic port for E500 (v1 and v2) based on the old 
> SPE add-on (available on savannah for glibc 2.3.4) and reimplement the 
> parts of the add-on not assigned to the FSF - once I have GCC for E500 
> back in a reliable state and have implemented 128-bit long double support.  
> Because much of the port is needed to operate correctly with hardware 
> floating-point on those CPUs (separate <fenv.h> and setjmp/longjmp 
> implementations, for example), I think that should go in the ports 
> collection (for EGLIBC and FSF GLIBC), leaving the spe add-on only for the 
> extra functions defined in the SPE PIM (string-to-fixed-point 
> conversions); anything that is purely an optimization could go in 
> powerpc-cpu, though with a directory for the CPU in ports I'm not sure how 
> much benefit there would be to separating out some files like that.
> 

If there is a directory for the e500 in ports I don't see much benefit
in separating out the SPE PIM functions, on balance it would be better
to keep them together.