HP-25 vs. SR-56 - HHC 2016 Follow-up

[Gene]

For those of you who were (un)fortunate enough to sit through this presentation at this year's HHC, I wanted to add a follow up I found from PPC V4N8P21.

Jim Davidson of the HP 25 Library Fame wrote an article about merged code. He reaches a rather astonishing conclusion

"95% of the time, 100 unmerged steps (GW- the SR-56) is better than 50 merged ones."

Of course there is a "all other things being equal" statement, which we know they are not. This is a merged vs. unmerged analysis only.

Link below:

HP-25 vs. SR-56 analysis

[Valentin Albillo]

.
Hi, Gene:

(10-07-2016 08:03 PM)Gene Wrote:  "95% of the time, 100 unmerged steps (GW- the SR-56) is better than 50 merged ones."

Of course there is a "all other things being equal" statement, which we know they are not.

Right you are, they are not.

For instance, the SR-56 allowed you to call subroutines (nested up to 4 levels deep) and use DSZ (decrement and skip on zero) and its inverse, neither of which were available in the HP-25 at all, thus making much better use of those 100 progamming steps, unmerged or not, not to mention the fact that it could also interact with an optional printer to provide printed output and extremely handy program listings. Which, alas, the HP-25 couldn't either.

Best regards.
V.
.

[Gene]

True true.

The HP 25 had the better programming environment (RPN/stack) but the SR-56 had the features you reference.

Take a look sometime at the youtube video of the HHC presentation and see what you think.

Thanks for reading.

[rprosperi]

(10-07-2016 08:03 PM)Gene Wrote:  For those of you who were (un)fortunate enough to sit through this presentation at this year's HHC....

I enjoyed it Gene! Your illuminating looks at TI machines are refreshing, almost totally new (at least to this crowd) and always interesting. One test of "how interesting" is I have since obtained a couple of these TI machines to explore and compare. My last pre-HP calculator was an SR-51A, and I was very happy to get an excellent example recently, complete with box, manual, charger, etc. all for less than one might pay for almost any HP model, even less than most HP-12Cs.

Keep it up. Off to go read the comparison of merged vs. non-merged....

[Dieter]

(10-07-2016 08:03 PM)Gene Wrote:  Jim Davidson of the HP 25 Library Fame wrote an article about merged code. He reaches a rather astonishing conclusion

"95% of the time, 100 unmerged steps (GW- the SR-56) is better than 50 merged ones."

Ah yes, the merged vs. unmerged thing. I remember a book with various mathematical programs both for the HP67/97 and the TI-58/59, i.e. the same algorithms programmed by the same person. Comparing the program lengths yielded something like a 4:7 ratio, i.e. the 224 steps of the 67/97 were comparable to almost 400 in the TIs. So this ratio is roughly the same as Jim's. Of course this is partially because of the merged/unmerged steps, but also due to the more flexible coding scheme in the HPs.

As far as the HP-25 and SR-56 are concerned: The TI appeared a year later than the HP, and this makes a difference. Features like subroutines and DSZ counters became standard, and another year later the HP29C was introduced which of course offered both, as well as 98 merged steps (comparable to 55-60 equivalent steps in the SR-56, if you take the ratio given in the linked PPC article). All this along with other benefits that show what a difference one year of development made in those days.

Dieter

[Gene]

Very good point, Dieter. A year after the HP 67, we had the TI-58/59 with plug in 5000 step roms, so progress did continue to jump forward.

For the record, I much prefer the RPN programming paradigm over the TI AOS one on these older machines.

[rprosperi]

(10-08-2016 06:45 AM)Dieter Wrote:  ...the HP29C was introduced which of course offered both, as well as 98 merged steps (comparable to 55-60 equivalent steps in the SR-56, if you take the ratio given in the linked PPC article).

Shouldn't the 29C's 98 merged steps expand to a much higher (than 98) number of non-merged steps in the SR-56?

[Valentin Albillo]

.
Hi again, Gene;

(10-07-2016 09:58 PM)Gene Wrote:  Take a look sometime at the youtube video of the HHC presentation and see what you think.

Done. Nice presentation you gave, by the way, I hadn't seen it before I posted my reply to your OP and indeed you mention the capabilities I pointed out.

However, I don't think the "SR-56 - not a loser" header is fair. Matter of fact, from a programming capabilities point of view the SR-56 is the clear winner when compared to the HP-25: more room for larger programs, more storage registers, subroutines, dsz, better accuracy (I'll take 13 digits over 10 any day no matter how well fine-tuned those 10 digits allegedly are).

At the end of the day, all programs you can fit in an HP-25 will also fit in an SR-56 with plenty of room to spare while the converse isn't true at all. I can think of tons of useful ones that would fit and run in an SR-56 but would be impossible to fit in an HP-25 no matter how clever the programming and no matter how good the programming environment. Thus the SR-56 wins hands down.

Last but not least, your sample Gamma program seems to me wholly inadequate to make the point you were trying to make. It's mostly a plain-vanilla straight formula which doesn't require any of the advanced SR-56 capabilities such as subroutines, and which, as you admit in your presentation, you programmed hastily and with no attempt at any optimizations or even competent AOS programming (you mention probably redundant parentheses and storing everything because you sorely missed the RPN stack).

I don't think the resulting program is evidence of anything, least of all of much better efficiency and such. In my opinion, bad evidence is worse than no evidence and only serves to mislead.

Anyway, as I said, nice presentation. Congratulations and thanks for bringing it to my attention.

Best regards.
V.
.

[Dieter]

(10-08-2016 02:17 PM)rprosperi Wrote:  

(10-08-2016 06:45 AM)Dieter Wrote:  ...the HP29C was introduced which of course offered both, as well as 98 merged steps (comparable to 55-60 equivalent steps in the SR-56, if you take the ratio given in the linked PPC article).

Shouldn't the 29C's 98 merged steps expand to a much higher (than 98) number of non-merged steps in the SR-56?

Sure. Sorry if I was not clear enough. I wanted to say that the SR-56's 100 unmerged steps are roughly equivalent to 55...60 merged steps in a 29C.
Or if you prefer it the other way round: its 98 merged steps should equal about 170 unmerged SR-56 steps.

Dieter

[Dieter]

(10-08-2016 06:45 AM)Dieter Wrote:  I remember a book with various mathematical programs both for the HP67/97 and the TI-58/59, i.e. the same algorithms programmed by the same person. Comparing the program lengths yielded something like a 4:7 ratio, i.e. the 224 steps of the 67/97 were comparable to almost 400 in the TIs. So this ratio is roughly the same as Jim's.

Guess what, I now found this book and so I compared the number of steps of the equivalent HP67/97 and TI58/59 programs. Here is the result:

Code:

Compared programs:      34
Max. # of HP steps:    224
Max. # of TI steps:    383
Min. ratio TI/HP:    1.476
Max. ratio TI/HP:    2.239
Average ratio:       1.733
Std.dev. of ratio:   0.189

So the result is similar to Jim's (average 1.727 resp. 1.731, standard deviation 0.162).
They almost agree if the two most extreme cases are neglected (average 1.725, standard deviation 0.166).

Dieter

[rprosperi]

(10-08-2016 05:03 PM)Dieter Wrote:  Sure. Sorry if I was not clear enough. I wanted to say that the SR-56's 100 unmerged steps are roughly equivalent to 55...60 merged steps in a 29C.
Or if you prefer it the other way round: its 98 merged steps should equal about 170 unmerged SR-56 steps.

Very good, thanks for clarifying Dieter. I was afraid I was missing something so basic that I almost didn't post my question.

Valentin has pointed out that the SR-56 is clearly superior, not only with more available steps, but also with more flexible program capabilities such as GSB, DSZ, etc. however I've found that my mind no longer works in an AOS way, so even though it's measurably better, it would not be better for me. Still, it would be interesting to play with one.

[Dieter]

(10-08-2016 09:31 PM)rprosperi Wrote:  Valentin has pointed out that the SR-56 is clearly superior, ...

Compared to the HP-25 – sure. I would be surprised if a later calculator model of that era would not outperform an older model in terms of features and memory. So does the 29C compared to the SR-56. The 29C appeared one year after the SR-56, just as the latter is one year newer than the 25. The 29C offers more memory (98 fully merged steps and 30 instead of 10 data registers), DSZ and ISZ, indirect addressing, etc... and, most important, continuous memory. That's what you could expect from the next year's calculator generation – just as you can when the SR-56 is compared to the older HP-25.

Let's not forget: there were merely five years between the introduction of the first programmable calculator ever and the "next gen" HP-41 series.

Dieter

[Gene]

Agreed, Dieter. It was a time of "leap frog" where manufacturers were ahead for a while (1975 - HP 25 ahead of TI) and then behind (1976 - SR-56 ahead of HP-25) and then 1977-79 (TI-58 hard to beat for price/performance/features IMO) etc.

Ah, times of old...

[rprosperi]

(10-09-2016 01:10 AM)Gene Wrote:  Agreed, Dieter. It was a time of "leap frog" where manufacturers were ahead for a while (1975 - HP 25 ahead of TI) and then behind (1976 - SR-56 ahead of HP-25) and then 1977-79 (TI-58 hard to beat for price/performance/features IMO) etc.

Ah, times of old...

The glory days indeed. The best part is TI and HP were bringing out several new machines every year! It's now 3+ years since the Prime's introduction and anywhere from 5 to 15 years since the significant TI models were introduced.

[SlideRule]

(10-08-2016 06:45 AM)Dieter Wrote:  I remember a book with various mathematical programs both for the HP67/97 and the TI-58/59, i.e. the same algorithms programmed by the same person.
...Guess what, I now found this book...

Does this book have a name?

BEST!
SlideRule

[Dieter]

(10-09-2016 03:09 PM)SlideRule Wrote:  Does this book have a name?

Sure: "Numerische Algorithmen auf programmierbaren Taschenrechnern", by Karl Hainer. I think I got mine in 1981 or 82.

Dieter

[SlideRule]

(10-09-2016 05:05 PM)Dieter Wrote:  Numerische Algorithmen auf programmierbaren Taschenrechnern

Danks.

SlideRule

[SlideRule]

I remember my answer in the PPC's to the merged / unmerged debate from many decades ago. Apparently, I am still as confused by the facts now as I was then since I do not believe the argument is framed very well. The proposition seems to imply ANY 100 unmerged program memory locations are better than 50 potentially merged program memory locations. SEE the difference?

OK, let's try again. 100 unmerged locations will only hold 100 keys while 50 potentially merged locations could hold 100 keys. Hmmm...
Granted, we seldom use all 50 locations for merged keys BUT any of the 50 locations could do so.

Given the choice of exceeding the key per location limitation of unmerged memory I will elect the merged every time {of course, the more the better ala hP-29c, etc}.

Am I tracking with anyone else?

BEST!
SlideRule

[Gene]

I think you are on track - IF - the merged steps were all multiple keystrokes that would take > 1 unmerged step.

STO 4 is one step on the HP 25 but two on the SR-56.

What Jim Davidson found after analyzing a lot of pretty amazing HP 25 programs was that on average, there were about 1.6-1.7 keys per step, so that 50 merged was equivalent to about 80-85 steps on the SR-56.

Jim Davidson had said that 95% of the time, 100 unmerged steps was better than 50 merged ones.

[Dieter]

(10-10-2016 02:42 AM)Gene Wrote:  What Jim Davidson found after analyzing a lot of pretty amazing HP 25 programs was that on average, there were about 1.6-1.7 keys per step, so that 50 merged was equivalent to about 80-85 steps on the SR-56.

Please note that we are talking about two slightly different figures in this thread. One thing is the "keys per step" ratio, the other thing is the "HP steps : TI steps" thing.

For instance, STO+1 requires three keys on most HPs and two (SUM 1) on the SR-56, while it takes one step resp. two steps in program memory. So a key-to-step-ratio of 1,6...1,7 in an HP25 program is one thing, and a TI-steps-to-HP-steps-ratio of 1,7...1,8 is another. The latter consists of both the more merged program steps as well as the more effective way things can be done in RPN, e.g. intermediate results do not require an "=" or even "(" and ")", tests can execute or skip any command instead of merely jumps, etc. While on the other hand special commands like the TI59 OP2x and OP3x increment/decrements or not being limited to one single indirect register will change the figure in favor of TI.

So a fair comparison of the two worlds has to consider more than just the more or less merged program steps. For the HP67/TI59 case, I think the 4:7 ratio fits quite well.

Dieter