Does anyone have a programme to get the mantissa of an extended real on the 50g - the shorter the better?
Here is my attempt at a solution, but I fear the programme is too large & slow:
Code:
::
CK1&Dispatch
BINT63
::
DUP
%%0=
?SEMI
%%ABS
%%10
OVER
DO>STR
DUP
CHR_E
BINT1
POSCHR
#1+LAST$
FPTR2 ^S>Z
FPTR2 ^Z2%%
%%^
%%/
;
;
(10-07-2017 03:24 PM)Gerald H Wrote: [ -> ]Here is my attempt at a solution, but I fear the programme is too large & slow:
...
This is smaller and faster, but there's probably still a better way:
Code:
::
CK1&Dispatch
BINT63 ::
TOTEMPOB
CODEM
A=DAT1 A
AD0EX
D0+5
C=0 A
DAT0=C A
D0+15
D0+5
DAT0=C P
AD0EX
RPL
ENDCODE
;
;
Excellent, DavidM!
Much faster than any programme I have seen to find extended mantissa.
Deals with %%0 & negative powers correctly, only critique is negative numbers remain negative.
(10-07-2017 06:19 PM)Gerald H Wrote: [ -> ]...negative numbers remain negative.
I'll admit that I didn't test this out extensively, but I did try at least one negative number and the sign was changed to positive (as it should have been, due to the DAT0=C P step). What did you test with this that didn't result in a positive mantissa?
Various negatives,eg %% -7.7E1, %% -1E0, %% -1E66, the negative sign is consistently retained.
I haven't found one negative number that shows a positive mantissa.
Sorry, I must have erred in copying the instructions, on recompiling the negatives are dealt with correctly.
(10-07-2017 07:46 PM)Gerald H Wrote: [ -> ]Sorry, I must have erred in copying the instructions, on recompiling the negatives are dealt with correctly.
No problem. You had me really scratching my head for a while there, though.
I should have included the checksum/byte count as a safeguard.
Here's the commented code, in case anyone is interested. I did swap two lines in this one to make the flow more consistent (it doesn't make a difference in the result, of course).
Checksum: FAECh
Bytes: 41
Code:
::
CK1&Dispatch ( must have at least 1 argument on stack )
BINT63 :: ( argument type must be extended real )
TOTEMPOB ( make a duplicate of the extended real for the result )
CODEM % start MASD code block
A=DAT1 A % A.A -> SL1 object (the result)
AD0EX % save D0 in A.A, D0 now points to result object
D0+5 % advance D0 past prologue
C=0 A % C.A = 0
DAT0=C A % zero-out exponent field
D0+5 % advance D0 past exponent field
D0+15 % advance D0 past mantissa field
DAT0=C P % zero-out sign field
AD0EX % restore original D0 from A.A
RPL % return to RPL
ENDCODE ( end MASD code block )
;
;
Quite informative, thank you.
Arno
Some background:
You may remember a library for the 48G
http://www.hpcalc.org/details/1243
called EXMATH which took control of screen & keyboard & created an RPN calculator working with exponents up to 49999.
I have updated the library to work on the 49G ver 1.19-6 & 2.10-7 & on the 50g ver 2.10-7, adding some extra functions & trying to improve performance.
I am confident that my own programming is not optimal & am very grateful for assistance.
Helping with things like that is for the time being way beyond my abilities (and the amount of time I can invest), I am refreshing my URPL which hadn't been used for mor than 10 years, I have recently started looking into the available SRPL books and I like writing little programs. So I usually have a close look at your programs which always seem to be very effective.
Arno
The annotations are very welcome, DavidM, as I am fearful of writing any assembly language code & regard its effect as magical.
Writing programmes in User & Sys RPL requires much less initiation, really just a mechanized form of algebra.
I think, Arno, that the change from User to Sys is not so big a step & the gain in speed is well worth the effort. I look forward to you writing some interesting programmes.
(10-09-2017 10:17 AM)Gerald H Wrote: [ -> ]The annotations are very welcome, DavidM, as I am fearful of writing any assembly language code & regard its effect as magical.
I think of writing Saturn code in a similar fashion to building furniture out of Lego blocks. It can be done, but you better have both a good reason and a good plan or you'll go insane before you finish it.
Although I rarely choose assembly solutions as a first approach, this one seemed like a perfect fit -- simply setting the exponent and sign fields to 0 was all that was needed, and I knew that would be fairly easy (and fast) in assembly.
(10-09-2017 10:17 AM)Gerald H Wrote: [ -> ]I think, Arno, that the change from User to Sys is not so big a step & the gain in speed is well worth the effort. I look forward to you writing some interesting programmes.
I agree!
(10-08-2017 06:42 AM)Gerald H Wrote: [ -> ]...
I have updated the library to work on the 49G ver 1.19-6 & 2.10-7 & on the 50g ver 2.10-7, adding some extra functions & trying to improve performance.
...
where can we find it?
(10-10-2017 06:32 PM)franz.b Wrote: [ -> ] (10-08-2017 06:42 AM)Gerald H Wrote: [ -> ]...
I have updated the library to work on the 49G ver 1.19-6 & 2.10-7 & on the 50g ver 2.10-7, adding some extra functions & trying to improve performance.
...
where can we find it?
Now available at
http://www.hpcalc.org/details/8702
(10-13-2017 05:41 AM)Gerald H Wrote: [ -> ] (10-10-2017 06:32 PM)franz.b Wrote: [ -> ]where can we find it?
Now available at
http://www.hpcalc.org/details/8702
thank you very much! I saw it yesterday in hpcalc news and today in hpforums posts!