PowerPC cleanup and Power4

David Edelsohn dje@watson.ibm.com
Sun Jun 9 10:33:00 GMT 2002


>>>>> Geoff Keating writes:

Geoff> How does this interact with processors where mfcr/mtcrf instructions
Geoff> aren't done in the same function unit as other CR logical instructions
Geoff> (like the 750)?

	Slight improvement because the 750 performs them in the complex
integer unit.  Previously the instructions were considered integer and
assumed to dispatch on both simple integer pipelines, which was even
worse.  It's not perfect, but it's a net improvement for the moment.

Geoff> This almost certainly doesn't do what you wanted.  More likely, you
Geoff> want to change PROCESSOR_DEFAULT64.

	Oops, right, fixed.

Geoff> I would also suggest changing PROCESSOR_DEFAULT64 for AIX, since the
Geoff> default is currently rios1.

	PROCESSOR_DEFAULT64 never was defined as rios1.  PROCESSOR_DEFAULT
is defined as rios1 in rs6000.h, but overridden in aix43.h and aix51.h.

David



More information about the Gcc-patches mailing list