This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC 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: More i386/ builtins


> On Mon, Jun 03, 2002 at 12:10:02PM +0100, Neil Booth wrote:
> > ! 	  if (TARGET_MMX)					\
> > ! 	    builtin_define ("__tune_pentium_mmx__");		\
> 
> TARGET_MMX won't be set for -mtune.  You'll need to check
> ix86_cpu_string.  Similar for __tune_athlon_sse__.
> 
> Though really I'm not sure what either of these are supposed
> to mean.  It's not like the code will be any different...
> Maybe your redefinition is sensible -- distinguishing P1 MMX
> from P2 MMX and Athlon SSE from P3 SSE tuning.
> 
> Anyone else have thoughts on this?

The pentium core differs from pentium_mmx core in scheduling and branch
prediction, so theoretically someone may want to have optimized version
for both variants, but I am not quite sure whether it is sane :)

Honza
> 
> > ! 	  if (TARGET_MMX)					\
> > ! 	      builtin_define ("__MMX__");			\
> > ! 	  if (TARGET_SSE)					\
> > ! 	    builtin_define ("__SSE__");				\
> 
> These should be moved down to common code.  They are feature
> specific, not cpu specific.  Any variance from this in the old
> specs was a bug.
> 
> Same for TARGET_SSE2 and "__SSE2__", TARGET_3DNOW and "__3dNOW__",
> TARGET_3DNOW_A and "__3dNOW_A__".
> 
> 
> 
> r~


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