|
Message
From: michael scott<mike.scott@j...>
Date: Thu Jul 21 15:55:08 CEST 2005
Subject: [openrisc] MAC instruction (last time!)
Hi Damjan,
I've just run my original code on rel_29
and the problem persists. However, I now believe this
an extreme corner case 'buglet' as for all practical applications
(FIR,IIR filters etc) you would need to wrap the l.mac
instructions in a loop and by placing the l.macrc in the delay slot
following the loop increment you balance the instruction/mac
pipelines to get the correct value in the register when l.macrc
is executed
Attached code gives approx 13 MegaMACS (avg) continuous. Increasing the
number of in-line macs would bring it closer to 16 MegaMACS theoretical
maximum using a 16 mhz clock
Regards,
Mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MacTest.S
Type: application/octet-stream
Size: 2586 bytes
Desc: not available
Url : openrisc/attachments/20050721/be115f32MacTest.obj
|
 |
Browse
