Hello everyone,
Loiseleur Michel wrote:
Hi *,
Leonard Mada a écrit :
[...]
*Thirteen Prevalent Misconceptions about Floating-Point Arithmetic:*
3. Arithmetic much more precise than the data it operates upon is
needless, and wasteful.
I have concerns that double is not appropriate anymore. It is not
appropriate for Sun, nor for IBM, nor for Google, nor for any
financial institution nor any government.
Is it possible to use greater precision? And still attain reasonable
speed?
The answer is definitely yes.
R, Matlab, other financial tools are orders of magnitude faster than
Calc. And work usually with greater precision than Calc anyway.
Therefore, it should be at least theoretically possible to speed up
Calc to work fast enough with higher precision.
My 2 cents : it's also technically possible. There is, among others, the
libgmp : http://gmplib.org/ .
Thank you for pointing this out. I am really amazed:
both:
1 + gcd(87324,78263148,7896) * (10^1989879887 mod 471!)
and
1 + gcd(9998560,6898880) * (2^1989879887 mod 471!)
computed in 2 ms! That is fast. That is almost (or probably) instantly.
The fact I mentioned R earlier stems from my experience both with R and
Calc: I believe Calc is much too slow, orders of magnitudes slower than R.
I do complex calculations in R, and these are performed almost always
instantly even on my slow PC. Similar calculations take tens of minutes
(and sometimes more than an hour) in Calc. Also, I often see messages
like "Adapting row height", operations which take longer to complete
than the calculation proper. Therefore I am fairly convinced that the
computation time in Calc is only a fraction the total time needed to
complete an operation, and even this could be improved dramatically.
Sincerely,
Leonard
P.S.
Maybe splitting the underlying spreadsheet model in a separate
data-table/matrix and the view-matrix would speed up computations
significantly:
A.) data-matrix: data[x,y] will contain only the values
- computations would be very fast, because:
-- there is NO overhead for other things
-- would iterate only through the values proper
- because I insist on type checking, there would be a type identifier,
too ;)
B.) view-matrix: cell[x,y] contains all other things,
including styles, view-options, formatting, ...
Its job is precisely to adapt storage to the kind of computing needed
and to scale well, whatever the number of digit is.
[...]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
