From: Steen Schmidt on 4 Sep 2006 13:11
Paul Schlyter wrote:
> > What would you do when you have calculated 10000! ? Would you
> > display the 35660 digits on the screen for the user to decipher one
> > at a time?
> :-) ...an interesting suggestion. If one new digit is displayed each
> second, it would take almost 10 hours to display the result
So, what would you do? Save it in binary form in flash for further
study when you get old?
> > What if the result was to be used by a subsequent calculation? Or
> > let's say you coded a new symbolic integration engine in C - how
> > would you output the integral of 'Exp(X^2),X'?
> By turning on the appropriate pixels in the display, of course!
So you code your own pretty-print display routine in C++ with scrolling
etc.? And then what? Not much use in displaying a result if you can't
save it or see it again later (what if the calculation took an hour?
You don't just repeat it...) or use it for further calculation.
> > The result is
> > '1/2*Sqrt(pi)*Erfi(X)', but Erfi(X) isn't defined in the TI AMS. You
> > could code Erfi(X) in C, but the result had to reside temporarily in
> > the TI OS environment,
> WHAT "TI OS" ???? I was discussing a C or C++ program without any
And I'm pointing out the futility of your point. Why don't you take a
stand on my questions? A calculator without storage facility is ok for
a 4-banger (and there it's also frustrating, since you often want to
save the result AND work with it further), but not for something as
complex as the calculators we're debating here. If you can't somehow
integrate your data with the existing OS, you can't use the data for
much. You can display it (if you code a viewer) and nothing else.
Doesn't have to be that complex before that's not practical. So we're
back to wasting our time, which means games.
> Coding without any OS support is indeed a lot of work - if your
> program is non-trivial.
Describe a trivial program worth anything, that you wouldn't just code
in a native (i.e. built-in) programming language. Do my examples fit in
From: Michael Kuyumcu on 4 Sep 2006 14:19
Tak! for the input and the link. Now I can write a program that was
dwelling in my mind for a long time. I just had been too lazy to
implement the arbitrary precision math... Great!
Steen Schmidt wrote:
> Michael Kuyumcu wrote:
> > Hey, that's terrific! Thanks for the info. Will I get the longfloat
> > info at hpcalc.org?
> Yes, you can find the latest longfloat lib here:
From: Yao Konan on 4 Sep 2006 18:55
Yes it is probably buggy but we should try on another calculator.
Unfortunately i just have my TI92+ actually.
Have you tried on the HP49G+ ?
Michael Kuyumcu a écrit :
> The same result in approx mode, but what a difference in speed! Is this
> buggy, or what?
> Michael Kuyumcu
From: Yao Konan on 4 Sep 2006 19:04
Steen Schmidt a écrit :
> Yao Konan wrote:
> > However it is more than sure that some limitations remain.
> Most of the limitations it seems. Directory structure for one, as well
> as all the math limitations.
With the concept of documents and of problems,i would say that the
memory structure has completely changed.
> The C program probably executes in 0.01 seconds or less - the rest is
> the RPL wrapper. The FIBNUM function executes out of the emulator
> completely transparent to the user - i.e. it works as any other
> function in the Saturn environment.
Interesting,too bad i am not interested by the HP50G.
> But for your analogy to work we assume the NSpire emulates the 68k like
> the HP49G+/50G emulates the Saturn? Since I expect the OS to be
> recompiled for the processor used in the NSpire, I'd expected the
> performance to be nearer what we achieve in C on the HPs. I'd have
> expected something like FIBNUM(10000) to run in 0.5 seconds or
Well,for this to be true we should assume that the TI code is not
bloated when it is.Moreover the For loop bring a significant overhead
otherwise the TI-NSpire will take less
than 4 s or even 3 s to compete FIBNUM(1000).
> But in UserRPL the recursive Fibonacci formula traverses to
> FIBNUM(1000) in 4 seconds on a HP50G.
So as fast as the TI-NSpire though the HP50G doesn't have the overhead
of TI-Basic For loop.
From: Veli-Pekka Nousiainen on 5 Sep 2006 05:01
Yao Konan wrote:
> Steen Schmidt a ?crit :
> Interesting,too bad i am not interested by the HP50G.
>> But for your analogy to work we assume the NSpire emulates the 68k
>> like the HP49G+/50G emulates the Saturn? Since I expect the OS to be
>> recompiled for the processor used in the NSpire, I'd expected the
>> performance to be nearer what we achieve in C on the HPs. I'd have
>> expected something like FIBNUM(10000) to run in 0.5 seconds or
> Well,for this to be true we should assume that the TI code is not
> bloated when it is.Moreover the For loop bring a significant overhead
> otherwise the TI-NSpire will take less
> than 4 s or even 3 s to compete FIBNUM(1000).
>> But in UserRPL the recursive Fibonacci formula traverses to
>> FIBNUM(1000) in 4 seconds on a HP50G.
You mean a loop...for the recursion you need a 50G in USB...
oe an emulator on a PC