[PATCH] Fix unresolved symbols on non-C99 platforms

[Eric Botcazou]

> Sorry for the breakage.

No problem.

> I don't understand how using a compiler built-in can put requirements on
> target libraries, but I'll try to be more careful. 

A compiler built-in ultimately relies on a generic or back-end specific 
expander.  For trunc*, it's a back-end expander, whose associated pattern is 
named 'btrunc'.  See builtins.c:expand_builtin_mathfn:

    case BUILT_IN_TRUNC:
    case BUILT_IN_TRUNCF:
    case BUILT_IN_TRUNCL:
      builtin_optab = btrunc_optab; break;

It's a relatively new pattern:

2004-09-04  Uros Bizjak  <uros@kss-loka.si>

	* builtins.c (expand_builtin_mathfn): Handle BUILT_IN_RINT{,F,L}
	using rint_optab.
	(expand_builtin): Expand BUILT_IN_RINT{,F,L} using
	expand_builtin_mathfn.
	* genopinit.c (optabs): Rename trunc_optab to btrunc_optab. Use
	btrunc?f patterns for btrunc_optab. Implement rint_optab using
	rint?f patterns.
	* optabs.c (init_optabs): Initialize rint_optab.
	* optabs.h (enum optab_index): Rename OTI_trunc to OTI_btrunc.
	Add new OTI_rint.
	(btrunc_optab): Rename macro from trunc_optab.
	(rint_optab): Define corresponding macro.

Since the SPARC back-end doesn't define this pattern, the built-in piggybacks 
on the library function.  A quick grep shows that only x86 defines 'btrunc'.


[Uros, would you mind documenting the new standard patterns you have added?]

-- 
Eric Botcazou