Re: [PATCH] PPC32 nofpu fixes

[Steven Munroe <munroesj at us dot ibm dot com>]

Roland McGrath wrote:

> > Roland McGrath wrote:
> >
> >    
> >
> > > I can't figure out why this is anything other than:
> > >
> > > #define NO_FPU
> > > #include <sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext.S>
> > >
> > > and the like, with appropriate #ifdef NO_FPU insertions in the trunk files.
> > >
> > >
> > >
> > >      
> > The common code in the trunk has to have versions with/without 
> > VMX/Altivec plus the ENOSYS stub for pre GLIBC_2_3_3. Assuming the 
> > powerpc NO_FPU implementation is a simple subset of the powerpc ABI then 
> > we need to provide the same version symbols as the base powerpc ABI.
> >    
>
> Sorry, I haven't figured out how this answers my question.
>
>  
When we added the VMX support we needed to change the size of ucontext 
which requires versioning with multiple implementations. So 
get|set|swapcontext require a version with VMX for the new version and 
without VMX for the old version (and smaller context). Each version 
requires a unique name (thus __CONTEXT_FUNC_NAME)

So when you asked for a common implementation between fpu and nofpu, I 
took the opportunity to define a common implementation for VMX+FPU+GPR, 
FPU+GPR, and GPR only (nofpu) that could be conditionally compiled (thus 
__CONTEXT_ENABLE_VRS in addition to __CONTEXT_ENABLE_FPRS).