This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Use -frounding-math for math testsuite

From: Roland McGrath <>
Date: Mon,  7 May 2012 16:00:31 -0700 (PDT)

> Since this exact sequence:
>> +	sethi	%hi(_GLOBAL_OFFSET_TABLE_-4), %##reg;			\
>>  	call	__sparc_get_pc_thunk.reg;				\
>>  	 or	%##reg, %lo(_GLOBAL_OFFSET_TABLE_+4), %##reg;		\
> must appear together for subtle reasons, it seems wise to put it
> into its own macro with a comment saying that it's important that
> it not be broken up (and why)
> Conversely, you could just litter the code with some local labels and use
> -(1f-0b) sorts of calculations instead of -4.  Then you could schedule the
> instructions however you want without breaking the semantics.  The use of
> the labels would also serve as implicit documentation about what the magic
> constant is needed for.  Seems better that way to me.

Ok, I'll just abstract it behind a macro with a big comment, thanks.

Trying to schedule these instructions is what got me into trouble in
the first place, so I'll take a pass on the -(1f-0b) stuff for now.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]