Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site mplvax.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!ihnp4!houxm!whuxl!whuxlm!akgua!sdcsvax!sdcc3!mplvax!cdl
From: cdl@mplvax.UUCP (Carl Lowenstein)
Newsgroups: net.micro
Subject: Re: DEC PRO-350 Floating Point
Message-ID: <174@mplvax.UUCP>
Date: Mon, 11-Mar-85 12:00:57 EST
Article-I.D.: mplvax.174
Posted: Mon Mar 11 12:00:57 1985
Date-Received: Wed, 13-Mar-85 04:04:54 EST
References: <2809@ncsu.UUCP>
Reply-To: cdl@mplvax.UUCP (Carl Lowenstein)
Organization: Marine Physical Laborator of SIO at UCSD
Lines: 31
Summary: 

In article <2809@ncsu.UUCP> pls@ncsu.UUCP (Phillip L. Shaffer) writes:
>I am trying to find out execution times for floating-point instructions
>for a DEC PRO-350.  This is needed for analysis of some experiments
>we conducted to compared predicted and actual execution times of some
>signal-processing type algorithms on microprocessors.  
	As background information, the PRO-350 has the same CPU chip set
(J-11) as the 11/23.  One can add to this micro-coded floating-point
support, the KEF11.  This gives the same (46) added instructions as the
hardware floating-point co-processor (FPF11) but is not nearly so fast,
expensive, or power-consuming.  The times are indeed as you quote them:
>	ADDD add double		42.45 microsec
>	SUBD subtract double	43.35 microsec
>	MULD multiply double   193.05 microsec
>	DIVD divide double     239.25 microsec.
	checked in Microcomputers and Memories 1982 (EB-20912-20)
Note also that times go up if operands are in processor memory, rather
than in FPU registers.
>There appears to be no logarithm instruction
	True.  It's floating-point firmware support to emulate the hardware
processors that have been around since 1972 on 11/45's and up.
>We were surprised that the PC/XT times were 20-50% FASTER than the PRO!
	Look for inefficiencies in the code generator of your C compiler.
In a similar situation, code generated by the DEC Fortran IV compiler
(not optimized for this instruction set) is nearly 3x slower than code
generated by the DEC Fortran-77 compiler. (optimized)
	Given the tendency of C to do things in double precision at the drop
of a hat, there may be also a lot of un-needed type conversion going on.

-- 
	carl lowenstein		marine physical lab	u.c. san diego
	{ihnp4|decvax|akgua|dcdwest|ucbvax}	!sdcsvax!mplvax!cdl