Re: PATCH: Update for hot/cold partitioning optimization

[Caroline Tice <ctice at apple dot com>]

Thank you for looking at this and getting back to me.  Just for 
clarification, I am assuming from what
you state in your email that (*targetm.have_named_sections) will return 
false for the som architecture.
Is that correct?  If so, then in_section should never receive the value 
"in_unlikely_executed_text", which
means the original macro (below) should be fine whether or not we are 
doing hot/cold partitioning.  Is that
your (better) understanding of the situation?  All I'm trying to do is 
make sure the hot/cold partitioning
stuff doesn't break anything on other architectures.  If you think the 
original macro will work just fine,
then I will pull my modification to it out of the patch; otherwise I 
would appreciate a suggestion as
to the right thing for me to do.  Again, thanks for your input!

-- Caroline Tice
ctice@apple.com

Original macro:

#define ASM_OUTPUT_FUNCTION_PREFIX(FILE, NAME) \
  {                                                                     
\
    const char *name = (*targetm.strip_name_encoding) (NAME);           
\
    if (TARGET_GAS && in_section == in_text)                            
\
      fputs ("\t.NSUBSPA $CODE$,QUAD=0,ALIGN=8,ACCESS=44,CODE_ONLY\n", 
FILE); \
    else if (TARGET_GAS && in_section == in_unlikely_executed_text)     
\
      fputs ("\t.NSUBSPA $CODE$,QUAD=0,ALIGN=8,ACCESS=44,CODE_ONLY\n", 
FILE); \
    else if (TARGET_GAS)                                                
\
      fprintf (FILE,                                                    
\
               "\t.SUBSPA .%s\n", name);                                
\
  }


On Jul 15, 2004, at 5:18 PM, John David Anglin wrote:

> > I would appreciate it if people with experience with the following 
> > config
> > files could make sure I didn't do the wrong thing in there by 
> > accident:
> >
> > 	gcc/config/arm/pe.h
> > 	gcc/config/i386/cygming.h
> > 	gcc/config/mcore/mcore.h
> > 	gcc/config/pa/som.h
>
> The som.h change is not correct.  The macro ASM_OUTPUT_FUNCTION_PREFIX
> is partially a reminant from named sections which were deleted a couple
> of years due to problems with nm and collect2.  This macro essentially
> emits the equivalent of a .section directive before starting to output
> a function.  If you look at varasm.c, you will see it is not suitable
> for switching sections.
>
> The .nsubspa directive creates a new subspace with the name $CODE$.
> The subtle aspect to .nsubspa is that there can be multiple subspaces
> all with the name $CODE$ and they are all unique.  How the linker
> sorts these subspaces is more or less up to it.  The .subspa
> direction can be used to switch to an existing subspace with a given
> name.  However, when you have multiple subspaces with the same name,
> I believe .subspa will put you in the first (default) subspace with
> the specified name.  This is not what's needed for switching back and
> forth between sections.
>
> Currently, we never switch subspaces while emitting the body of a
> function.  We are very dependent on knowning the distance between
> insns for the emissions of branches and the distance to the beginning
> of the function for calls.  Assuming we start outputing text and
> then switch to emitting unlikely text in a new subspace, this is
> going to have lots of implications for branches between the two
> subspaces.  Effectively, these become calls.  GAS currently doesn't
> support calls to alternate entry points.  So, I don't think this
> is going to work with SOM without a lot of careful thought.  It's
> also going to interact with a patch I'm about to submit regarding
> COMDAT support.
>
> Dave
> --
> J. David Anglin                                  
> dave.anglin@nrc-cnrc.gc.ca
> National Research Council of Canada              (613) 990-0752 (FAX: 
> 952-6602)