Why no Whetstone benchmark for calculators

[StephenG1CMZ]

I have been searching around for some calculator benchmarks, and have been surprised that none of them are listing a Whetstone result - this was quite a common benchmark for PC's at one time.

I am wondering why?

Is it because calculators are so different to PC's that such results are not considered useful?
Is it because standard C (etc.) Source cannot be used?

I have produced a PPL interpretation of the Whetstone benchmark, and it seems to be simple enough to program on a calculator (though it is not verified - can anyone suggest where I can find correct outputs listed, rather than timings?)

[pier4r]

as far as I know whetstone and drystone should be compact enough to fill the cache of a computer CPU of tens of years ago.

Just people focused on simple stuff or stuff involving different operations. Like the SAVAGE benchmark. Moreover if the code is not so easy for ohter machines, you may compare just 2 systems.

I filled the wiki4hp.com with some benchmark results.

[rprosperi]

(04-14-2017 08:58 PM)StephenG1CMZ Wrote:  Is it because standard C (etc.) Source cannot be used?

One reason is very few (not enough to matter) calculators use a procedural language like C or PPL, so porting to each machine is fairly involved to claim that this specific port was not 'tuned' to optimize the apparent results for this machine. One reason the Whetstone (and others) is a meaningful comparison is because the code generally does not vary when going from one machine to another as they all support the common languages.

For example, there are many ways of implementing the Whetstone suite in RPL on a 50g and I'd guess that the different solutions vary by as much 200-300% in performance (maybe more), based on the port and particular commands used. So while it may give you a broad placement of the machine's performance in some larger spectrum, I think it would miss the point of comparing one machine to another.

[Paul Dale]

I imagine the Whetstone benchmark was too large for the early handheld calculators. The first HP calculators that might have had enough memory would have been the 41's (RPN) or 28's (RPL). The BASIC handhelds too. They'd have been very slow.


Pauli

[StephenG1CMZ]

I have just been looking at this benchmark again, and have noticed what seems to be a discrepancy in the on-line pdf spec for the Whetstone benchmark:
http://www.roylongbottom.org.uk/whetston...nchorStart

If I am reading this right, the relevant code is:

Code:


N1:=0
//Module 1
X4:=-1
For 1 to N1... (Not executed?)
End for
Print X4

Which one would expect to deliver -1.
But the descriptive text refers to a convergence to 1.

Am I misunderstanding the code or text?
Should X2,X3 and X4 be 1, -1 or something else on the first output?
Does the on-line pdf match the originals (I don't have the paper version to hand).
Are their errata?