Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: Notesfiles $Revision: 1.7.0.10 $; site uokvax.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!inuxc!pur-ee!uiucdcs!okstate.UUCP!uokvax.UUCP!emjej
From: emjej@uokvax.UUCP
Newsgroups: net.micro.6809
Subject: Re: Orphaned Response
Message-ID: <3500144@uokvax.UUCP>
Date: Fri, 11-Apr-86 18:55:00 EST
Article-I.D.: uokvax.3500144
Posted: Fri Apr 11 18:55:00 1986
Date-Received: Sat, 19-Apr-86 13:48:10 EST
References: <801@ihwpt.UUCP>
Lines: 41
Nf-ID: #R:ihwpt.UUCP:801:uokvax.UUCP:3500144:000:1752
Nf-From: uokvax.UUCP!emjej    Apr 11 17:55:00 1986


/* Written  6:03 pm  Apr  2, 1986 by knudsen@ihwpt.UUCP in net.micro.6809 */
I just got the dhrystone C program source from net.sources
and have edited it for the Coco OS9 C.
When/if I get results, they'll show up here the next day.
Anyone else want to try, and compare results?
/* End of text from net.micro.6809 */

Mike mailed me the source--I ran it on a 2 MHz 6809 system (SSB Chieftain,
OS-9/6809 Level II v. 1.2).  Since the getime() doesn't give you closer
resolution than a second anyway, I just conditionally uncompiled all the
timing code and did this:

	cc <options> dhry.c
	load date; load dhry
	date t; dhry; date t

The form in which I got the code provided for conditional compilation with
local variables and formal parameters declared register (though Microware
C honors only one such per function).  I added code to allow conditional
compilation with scalar global variables requested especially to be on the
direct page.  Here's what I got--

regs? DP? 	raw delta t	Dhry/sec
 N     N	   4:02		 207
 Y     N	   3:58		 210
 N     Y	   3:59		 210
 Y     Y	   3:21 (!)	 249

There were few global variables, so I could understand the relatively small
difference DP variables alone would make, but I was REAL startled by the
effect of using BOTH registers and DP variables.

Whether such maneuvers are "kosher" for Dhrystone benchmarking is for others
to decide, though the eighty-eighty-sucks systems listed, with only one ex-
ception, were directed to generate "small model" code, so...But frankly,
who cares?  Just remember it when *you* write C programs for OS-9.  It's
essentially trivial to make "direct" go away when you recompile for OS-9/68000
or other, lesser, systems :-) --so why not take advantage of it?

						James Jones