Path: utzoo!attcan!uunet!lll-winken!lll-lcc!ames!necntc!dandelion!ulowell!cg-atla!bradlee
From: bradlee@cg-atla.UUCP (Rob Bradlee X5153)
Newsgroups: comp.arch
Subject: Re: getting rid of branches
Message-ID: <4957@cg-atla.UUCP>
Date: 5 Jul 88 20:42:56 GMT
References: <12258@mimsy.UUCP> <28200173@urbsdc>
Reply-To: bradlee@cg-atla.UUCP (Rob Bradlee X5153)
Organization: Compugraphic Corp. Wilmington, MA
Lines: 47

To: ulowell!bbn!mit-eddie!bloom-beacon!tut.cis.ohio-state.edu!mailrus!iuvax!pur-ee!a.cs.uiuc.edu!uxc.cso.uiuc.edu!urbsdc!aglew
Subject: Re: getting rid of branches
Newsgroups: comp.arch
In-Reply-To: <28200173@urbsdc>
References: <12258@mimsy.UUCP>
Organization: Compugraphic Corp. Wilmington, MA
Cc: 
Bcc: 

In article <28200173@urbsdc> you write:
>
>
>This comes perilosly close to something I'm trying.
>Davidson at the UIll, now UMich, proposed a partitioned
>access/execute architecture a while back, where you basically
>have two processors, one for address and memory computations,
>one for (mainly floating point?) calculation.
>They run independently, with FIFOs between them for passage of
>variables back and forth, condition codes, etc. They basically
>run two versions of exactly the same program, one with all the
>FP taken out, and the other with all the memory references
>replaced by "get the next value from the memory unit".
>
>Anyone else doing similar?

This sounds quite similar to the architecture of array processors built
by CSP, Inc of Billerica, MA.  They use an IPU (integer processor) and
an APU (arithmetic processor) connected by FIFOs.  The IPU would 
handle computations to determine the next piece of data to get from
memory and send it over to the APU for processing. The APU would put
the result in another FIFO for return to the IPU which could then
determine where to put it into memory.  The best part was that the
two processors operated asynchronously waiting on their respective
input queues if one got ahead of the other.  Unfortunately, to do
anything really useful a program had to be done by hand in assembler.
However, they did have a FORTRAN compiler which generated sometimes adequate
code.

-- 
Rob Bradlee  w:(617)-658-5600 X5153  h:(617)-944-5595
Compugraphic Corp.	    ...!{ima,ulowell,ism780c}!cg-atla!bradlee
200 Ballardvale St.
Wilmington, Mass. 01887           The Nordic Way: Ski till it hurts!
-- 
Rob Bradlee  w:(617)-658-5600 X5153  h:(617)-944-5595
Compugraphic Corp.	    ...!{ima,ulowell,ism780c}!cg-atla!bradlee
200 Ballardvale St.
Wilmington, Mass. 01887           The Nordic Way: Ski till it hurts!