Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!swrinde!cs.utexas.edu!convex!texsun!newstop!sun!imagen!randy
From: randy@imagen.com (randy scofield)
Newsgroups: comp.arch
Subject: Re: Branch Delay Slots and Main Memory
Message-ID: <1991Apr3.182023.22505@imagen.com>
Date: 3 Apr 91 18:20:23 GMT
References: <28677@dime.cs.umass.edu>
Sender: news@imagen.com
Organization: IMAGEN
Lines: 21

In article <28677@dime.cs.umass.edu> niehaus@tempo.cs.umass.EDU (Douglas Niehaus) writes:
>Ok, so I am reading the RS2000 book by Kane, and I have felt like I understood
>what was going on, but I have a nagging question left.
>
>Branch delay iand load delay slots are one instruction in length, which I can deal with as long
>as what is being fetched from memory is in the cache, but what happens for a cache miss?
>
>I can deduce, I think, that on a miss the who processor must get frozen until the 
>fetch from main memory is finished, or else the single delay slot would not be enough.
>Is this correct? If not, what DOES happen? If so, just how is it done?
>
>Douglas Niehaus
>niehaus@tempo.cs.umass.edu
The processor is stalled on all cache misses.  The R2000 will get the data from memory,
stick it in the cache and then proceed.  All pipe stages are frozen during the stall.
The R3000 will do a block refill of 4/8/16/32 words as oppossed to the single read
of the R2000.  The block refill will hopefully fetch the data that will stop the
next few stalls from occurring.

Hope this answers your question.
Randy