leblanc@eecg.toronto.edu (Marcel LeBlanc) (10/23/90)
After installing SAS/C 5.10 on my 2000, I was curious about any speed improvements that might result from compiling programs with 5.10 vs. 5.04. So I decided to compile Dhrystone 2.1 to test the integer/string performance (that's all I'm really concerned about since my primary use for the 2000 is software development). Anyway, I found the results interesting, so here they are for all to see: To compile, I used: lc -DTIME <options> ... To link I used (c.o,$(OBJS),lcm.lib,lc.lib). Here are the options, with dhrystone/sec results (100K loops unless otherwise indicated): OPTIONS Lattice C 5.04 SAS/C 5.10 (none) 781.3 775.2 -cusrf 781.3 781.3 -cusrf -O 934.6 925.9 -cusrf -O -v 1000.5 983.6 (^^ 1e6 loops) (^^ 1e7 loops) -cusrf -O -v -DREG=register 1025.6 990.7 (^^ 1e6 loops) (^^ 1e7 loops) *NOTE: -v means no stack checking. The 5.10 results are only slightly lower than 5.04, so I'm really not concerned. I was never able to get dhrystone to compile properly with registerized parameters (-rr). I created prototypes for all functions (except Proc_1), but I always got the GURU. Also, when linking with lcmr.lib (5.10), blink could not find "_CXFERR". No problem with lcm.lib. One thing that disturbs me is the huge difference between these numbers and those posted by Martin Hunt of Commodore-Amiga: > 2500/30 3000 (16 MHz) 3000 (25 Mhz) 2000 > > Normal 5950 4300 6050 600 > Registers 6110 4400 6225 620 > 68020 6600 4770 6800 > 68020/Reg 6820 4970 7000 600??? 620??? this is ridiculous! He claims to have used "lc -v -O ...". What gives? Does this mean that the numbers for the A3000 are off too? My A3000 is still on order, so I can't verify the claim of 7000 Dhrys/sec. Comments or corrections? Marcel A. LeBlanc -- Electrical Eng. Computer Group, Univ. of Toronto ----------------------------------------------------------------------- leblanc@eecg.toronto.edu else: uunet!utcsri!eecg!leblanc
jmeissen@oregon.oacis.org ( Staff OACIS) (10/23/90)
In article <1990Oct22.175350.13597@jarvis.csri.toronto.edu> leblanc@eecg.toronto.edu (Marcel LeBlanc) writes: >One thing that disturbs me is the huge difference between these numbers and >those posted by Martin Hunt of Commodore-Amiga: > >> 2500/30 3000 (16 MHz) 3000 (25 Mhz) 2000 >> >> Normal 5950 4300 6050 600 >> Registers 6110 4400 6225 620 > >600??? 620??? this is ridiculous! He claims to have used "lc -v -O ...". >What gives? Lattice has had Dhrystone performance on a stock Amiga of greater than 600 since V3.10, when they were quoting a figure of, I believe, 720. It was my understanding that with appropriate options V5.0 was able to hit around 1,200. Obviously something is wrong with Martin's figures. -- John Meissen .............................. Oregon Advanced Computing Institute jmeissen@oacis.org (Internet) | "That's the remarkable thing about life; ..!sequent!oacis!jmeissen (UUCP) | things are never so bad that they can't jmeissen (BIX) | get worse." - Calvin & Hobbes
phil@adam.adelaide.edu.au (Phil Kernick) (10/23/90)
jmeissen@oregon.oacis.org ( Staff OACIS) writes: >Lattice has had Dhrystone performance on a stock Amiga of greater than >600 since V3.10, when they were quoting a figure of, I believe, 720. >It was my understanding that with appropriate options V5.0 was able to >hit around 1,200. I have an A1000, 68010, 2M with Lattice C V5.05 I get a value of Registers: 1041 No Reg: 1020 Phil. -- Phil Kernick EMail: phil@adam.adelaide.edu.au Departmental Engineer Phone: +618 228 5914 Dept. of Psychology Fax: +618 224 0464 University of Adelaide Mail: GPO Box 498 Adelaide SA 5001
martin@cbmvax.commodore.com (Martin Hunt) (10/23/90)
In article <1990Oct22.175350.13597@jarvis.csri.toronto.edu> leblanc@eecg.toronto.edu (Marcel LeBlanc) writes: > >One thing that disturbs me is the huge difference between these numbers and >those posted by Martin Hunt of Commodore-Amiga: > >> 2500/30 3000 (16 MHz) 3000 (25 Mhz) 2000 >> >> Normal 5950 4300 6050 600 >> Registers 6110 4400 6225 620 >> 68020 6600 4770 6800 >> 68020/Reg 6820 4970 7000 > >600??? 620??? this is ridiculous! He claims to have used "lc -v -O ...". >What gives? Does this mean that the numbers for the A3000 are off too? >My A3000 is still on order, so I can't verify the claim of 7000 Dhrys/sec. > >Comments or corrections? The numbers I gave for the 2000 are for a base 2000, no fast RAM. I get 1075 dhrystones for a 2000 with fast RAM. Sorry about the confusion. BTW, the numbers for the 3000 are without static column RAM. If you have SCRAMs, add 10% to the dhrystones. > >Marcel A. LeBlanc -- Electrical Eng. Computer Group, Univ. of Toronto >----------------------------------------------------------------------- >leblanc@eecg.toronto.edu else: uunet!utcsri!eecg!leblanc Martin Hunt martin@cbmvax.commodore.com Network Engineer {uunet|pyramid|rutgers}!cbmvax!martin Commodore-Amiga "Yeah it is really fast, but is it Yugo compatible?"
stefanb@cip-s02.informatik.rwth-aachen.de (Stefan Becker) (10/23/90)
leblanc@eecg.toronto.edu (Marcel LeBlanc) writes: >What gives? Does this mean that the numbers for the A3000 are off too? >My A3000 is still on order, so I can't verify the claim of 7000 Dhrys/sec. I've got 7700 Dhrys/sec. with GCC 1.37.1. So it's true :-) Stefan Mail : Stefan Becker, Holsteinstrasse 9, W-5100 Aachen /// Only Phone : +49-241-505705 FIDO: 2:242/7.6 Germany /// Amiga makes Domain: stefanb@cip-s02.informatik.rwth-aachen.de \\\/// it possible.. Bang : ..mcvax!unido!rwthinf!cip-s02!stefanb \XX/ -->A3000/25<--
jmeissen@oregon.oacis.org ( Staff OACIS) (10/24/90)
In article <phil.656647168@adam.adelaide.edu.au> phil@adam.adelaide.edu.au (Phil Kernick) writes: >I have an A1000, 68010, 2M with Lattice C V5.05 I get a value of >Registers: 1041 >No Reg: 1020 Make sure you include appropriate header files, like "string.h", that aren't in the original source (technically this was a bug in Dhrystone.c) and also use the -O option to invoke the global optimizer. When you run it, boot with a vanilla WorkBench, with no background tasks running (clock. MemWatch, PerfMon, etc). -- John Meissen .............................. Oregon Advanced Computing Institute jmeissen@oacis.org (Internet) | "That's the remarkable thing about life; ..!sequent!oacis!jmeissen (UUCP) | things are never so bad that they can't jmeissen (BIX) | get worse." - Calvin & Hobbes
uzun@pnet01.cts.com (Roger Uzun) (10/24/90)
I get over 8000 dhrystone 2's /sec when I run dhrystone 2.1 on my A3000.
Try
lc -v -cf -m3 -O -Lmt -f8d -l <fname> <fname>
The -l aligns all data to 32 bits, the fp options are just because of the
timing stuff, it has some fp calcs in it.
-Roger (I will run dhry2 tonight and see what I get again, it was either
8333 or 10,000 / sec I forget which)
-Roger
UUCP: {hplabs!hp-sdd ucsd nosc}!crash!pnet01!uzun
ARPA: crash!pnet01!uzun@nosc.mil
INET: uzun@pnet01.cts.comd87sg@efd.lth.se (Svante Gellerstam) (10/24/90)
In article <39@oregon.oacis.org> jmeissen@oregon.oacis.org ( Staff OACIS) writes: >In article <1990Oct22.175350.13597@jarvis.csri.toronto.edu> leblanc@eecg.toronto.edu (Marcel LeBlanc) writes: >>One thing that disturbs me is the huge difference between these numbers and >>those posted by Martin Hunt of Commodore-Amiga: >> >>> 2500/30 3000 (16 MHz) 3000 (25 Mhz) 2000 >>> >>> Normal 5950 4300 6050 600 >>> Registers 6110 4400 6225 620 >> >>600??? 620??? this is ridiculous! He claims to have used "lc -v -O ...". >>What gives? > >Lattice has had Dhrystone performance on a stock Amiga of greater than >600 since V3.10, when they were quoting a figure of, I believe, 720. >John Meissen .............................. Oregon Advanced Computing Institute >jmeissen@oacis.org (Internet) | "That's the remarkable thing about life; >..!sequent!oacis!jmeissen (UUCP) | things are never so bad that they can't >jmeissen (BIX) | get worse." - Calvin & Hobbes In my tests I hvave seem numbers like 1150 - 1250 on an A1000, and more like 8000 on an A3000... This is dhrystone with registers and short ints. On the A3K (25MHz) burst and cache was active. -- 2:200/107.4 Svante Gellerstam (Fido) d87sg@efd.lth.se (InterNet) It's the african anteater ritual! -- Can't Buy Me Love