RFA: Some fixes in dfa ppc8450 description

David Edelsohn dje@watson.ibm.com
Wed Apr 2 16:00:00 GMT 2003

>>>>> Vladimir Makarov writes:

Vlad> The most of ppc processors are smart enough and does not require (the
Vlad> first cycle) look ahead.  But this is not case for 8450 (as I understand
Vlad> it from the documentation).

Vlad> Generally speaking, the look ahead has advantages and disadvantages.

	For Power4, one wants to maximize dispatch groups -- pulling less
high priority instructions forward if it will fill unused slots in a
dispatch group, improving function unit utilization and minimizing
resource reservations.  I would expect that lookahead would accomplish
that, but, as I mentioned, enabling it resulted in a degradation in


More information about the Gcc-patches mailing list