Re: init_mov_optab obsolete?

[Zack Weinberg <zack@bitmover.com>]

Jeffrey A Law wrote:
>   In message <199908240535.WAA00508@zack.bitmover.com>you write:
>   > 
>   > If EXTRA_CC_MODES is defined, genemit.c writes out a special function
>   > called init_mov_optab that adds entries to mov_optab for patterns that
>   > move those modes around.  However, it appears to me that
>   > init_all_optabs (generated by genopinit.c) will set up those entries
>   > anyway.
>   > 
>   > Am I right, and if so, can we kill it?  It's not doing any harm, but
>   > it is confusing when you're trying to unravel what the md processors
>   > do.
>   > 
>   > (N.B. No current machine description actually defines patterns to move
>   > extra cc modes around.)
>
> Actually, I do not think the code in genopinit.c works correctly in
> the presence of CC modes.  But I think the code in genemit.c does
> work correctly for the extra CC modes.
> 
> Before killing the code I'd like to have some reasonable expectation
> that the code we leave will actually work.
> 
> Don't let the fact that no ports define patterns to move extra CC
> modes cloud the issue.  We will need it to effectively support some
> stuff in the PA2.0 architecture.  I just haven't written the code
> yet ;-0

I'm pretty sure init_all_optabs will generate either a correct
initialization or no initialization at all.  Can you show me a sample
pattern that should be recognized as an extra-CC-mode move?  Then I
will make sure genopinit.c does the right thing.

zw