HP Forum Archive 17

[ Return to Index | Top of Index ]

35S Benchmark test Message #1 Posted by Ralph on 27 July 2007, 1:40 a.m. I ran the benchmark tests from the page on this forum here. and got some odd results. The 33S program fitted to the 35S ran very slow, about 471 seconds on my stopwatch. I read there is probably some internal work because of the vector math function added. It might add some overhead. OK I can accept this, it makes sense to me. Then for grins I converted the 41/42 program to the 35S. 323 seconds. Slow but not as slow!?! I dug out the 33S and it runs it's test in 129 seconds. All to my stop watch and skills. From the chart on the forum page, there are the following times: 42S 425 sec. 32S 262 sec. 32SII 344 sec. And my timer: 33S 129 sec. 35S/33S 417 sec. 35S/42S 323 sec. Next is to put the 42S version on the 33S. I'm guessing it should be faster if it fits.

Re: 35S Benchmark test Message #2 Posted by Xerxes on 27 July 2007, 6:30 a.m., in response to message #1 by Ralph Hi Ralph, the 41/42 version of the benchmark was already tested on the 33S by Gerson, with some modificatios of course, with 251 seconds. The main reason for the different versions is the way how the calculators handle indirect addressing and the fastest way to increment or decrement variables. Considering the new concept of the 35S, I guess it is necessary to make an implementation especially for this calculator. As soon as the 35S is available in Germany, I will try it out and add the result to the data base. Are you sure with the 42S result? In the table you will find the 42S with 732 sec (fast mode: 362 sec). Edited: 29 July 2007, 6:29 a.m. after one or more responses were posted

35S Benchmark test Message #3 Posted by Ralph on 27 July 2007, 9:12 a.m., in response to message #2 by Xerxes Thanks. I didn't find the conversion but it was getting late last night. You are correct on the time for the 42S. I didn't search the page, just eyeballed it. And misread or typed, common for me. I know it's blazing fast on my HP4705 Pocket PC. I think the use of the two indirect register variables, I&J plus using the ISG & DSE commands over the math make the 42 version run faster on the 35S. Given the calculator is said to check whether the math is conventional or vector. The 42 code went in almost as written. I put a little bit in the front to set the index range so to not have to do it manually. Probably added a second or two. I find these tests fascinating. For my use, my program runs at the same speed on the 33 or 35 as far as I can tell.

Re: 35S Benchmark test Message #4 Posted by Gerson W. Barbosa on 27 July 2007, 9:19 a.m., in response to message #1 by Ralph Ralph, Can you post both your 35s versions, or at least the fastest one? Just for reference. Perhaps you'd like to take a look at these old threads: http://www.hpmuseum.org/cgi-sys/cgiwrap/hpmuseum/archv016.cgi?read=106865 http://www.hpmuseum.org/cgi-sys/cgiwrap/hpmuseum/archv016.cgi?read=107447 I've noticed Xerxes's article include a Lua version, but I wasn't able to run it in HP-Lua. Has anyone tried it? Regards, Gerson. Edited: 27 July 2007, 9:28 a.m.

35S Benchmark test Message #5 Posted by Ralph on 27 July 2007, 10:59 p.m., in response to message #4 by Gerson W. Barbosa Here is the 42 version placed on the 35S T003-T008 make sure the indirect registers are allocated in case I have reset things previously. 35S LN=28 CK=53FC T001 LBL T T002 CLVAR T003 9 T004 STO I T005 1 T006 STO (I) T007 0 T008 STO I T009 8 T010 STO R LN=24 CK=FF6C U001 LBL U U002 RCL I U003 RCL R U004 X=Y? U005 GTO Y001 U006 ISG I U007 DEG U008 STO (I) LN=15 CK=7ECA V001 LBL V V002 ISG S V003 DEG V004 RCL I V005 STO J LN=51 CK=8CB6 W001 LBL W W002 DSE J W003 DEG W004 RCL J W005 X=0? W006 GTO U001 W007 RCL (I) W008 RCL (J) W009 - W010 X=0? W011 GTO X001 W012 ABS W013 RCL I W014 RCL J W015 - W016 X<>Y? W017 GTO W001 LN=15 CK=BB97 X001 LBL X X002 DSE (I) X003 GTO V001 X004 DSE I X005 GTO X001 LN=9 CK=DC18 Y001 LBL Y Y002 RCL S Y003 RTN

Re: 35S Benchmark test Message #6 Posted by Xerxes on 29 July 2007, 9:05 a.m., in response to message #5 by Ralph I have added your program to the list as the best solution so far. Thank you for testing. Edited: 29 July 2007, 4:27 p.m.

Re: 35S Benchmark test Message #7 Posted by Xerxes on 29 July 2007, 6:48 a.m., in response to message #4 by Gerson W. Barbosa Hi Gerson, Can you explain the problem with HPLua please? I led from that the Lua version is usable for all lua interpreters.

Re: 35S Benchmark test Message #8 Posted by Gerson W. Barbosa on 29 July 2007, 9:13 a.m., in response to message #7 by Xerxes Hi Xerxes, By what I can remember, I reset the calculator after waiting for one minute or so. I'll try again later. Regards, Gerson.

Re: 35S Benchmark test Message #9 Posted by Xerxes on 29 July 2007, 4:39 p.m., in response to message #8 by Gerson W. Barbosa The Lua version on the Casio ClassPad takes 110 seconds if you use the 100 iterations in the original listing. On what calculator did you test it?

Re: 35S Benchmark test Message #10 Posted by Gerson W. Barbosa on 31 July 2007, 7:50 p.m., in response to message #9 by Xerxes Hi Xerxes, HP-50g (HPLua version 0.4) -> 1000 iterations in 4 min 3.5 sec. That's 0.2435 seconds per iteration. Regards, Gerson.

Re: 35S Benchmark test Message #11 Posted by Xerxes on 1 Aug 2007, 5:44 a.m., in response to message #10 by Gerson W. Barbosa Thanks Gerson.

[ Return to Index | Top of Index ]

Go back to the main exhibit hall