HP Forum Archive 20

[ Return to Index | Top of Index ]

WP34S: An integration wrapper Message #1 Posted by Miguel Toro on 9 Oct 2011, 10:55 p.m. Another week-end little project. As suggested by the WP34s team on a older post, I tried to make a wrapper to improve accuracy for the integration function. A test with sqrt(abs(x-1)), from 0 to 2, gives: WP 34s : 1.30929962 WP 34s with the wrapper (fix 5) : 1.33333672 Expected : 4/3 Using this program: LBL 10 1 - ABS [sqrt] RTN Any comment to improve it as always very welcome. Miguel Edited: 9 Oct 2011, 11:04 p.m.

Re: WP34S: An integration wrapper Message #2 Posted by gene wright on 9 Oct 2011, 11:10 p.m., in response to message #1 by Miguel Toro Hi Miguel! Try it on these expressions from the PPC ROM examples: Dropbox link to IG results

Re: WP34S: An integration wrapper Message #3 Posted by Miguel Toro on 10 Oct 2011, 7:21 a.m., in response to message #2 by gene wright Hi Gene, Here are the results. Not so bad, I think. Function 10, takes too long, so I ran it with FIX 3. Against my own advice, the others where done with FIX 6, getting decent running times. 1: 3.141593 2: 0.666667 3: 0.636620 4: -0.999984 6: 3.141594 8: 0.499995 9: 1.991321 10: 0.785399 FIX 3 (too slow otherwise) 12: 2.203347

Re: WP34S: An integration wrapper Message #4 Posted by Paul Dale on 9 Oct 2011, 11:22 p.m., in response to message #1 by Miguel Toro Nice compact routine. - Pauli

Re: WP34S: An integration wrapper Message #5 Posted by Marcus von Cube, Germany on 10 Oct 2011, 2:36 a.m., in response to message #1 by Miguel Toro A few minor remarks: In your instructions, "cumulatif" should spelled "cumulative". (Neither of us is a native speaker, I assume, so we need to help each other. :-)) Why do you set the stack size to 4? I haven't checked your code but if you don't need A to D for other purposes or rely on the duplication of T just leave the setting alone. If you really need it, use STOM/RCLM to save and restore the mode. The iteration uses the approximate comparison operator which relies on the display setting (Am I right, Pauli?). Just put a comment in that explains how this affects accuracy of the integral. I'd like to see some // comments at the end of the lines. This would help to understand your code and it will not disturb the assembler. Edited: 10 Oct 2011, 2:39 a.m.

Re: WP34S: An integration wrapper Message #6 Posted by Walter B on 10 Oct 2011, 2:40 a.m., in response to message #5 by Marcus von Cube, Germany Quote: The iteration uses the approximate comparison operator which relies on the display setting (I'm I right, Pauli?). Please see page 58/102. You are :-)

Re: WP34S: An integration wrapper Message #7 Posted by Miguel Toro on 10 Oct 2011, 6:54 a.m., in response to message #5 by Marcus von Cube, Germany Got it! I will do the changes. Thanks, Miguel Done! Note: The possibility of using complex instruction makes the code more compact, so i need a four register stack to take advantage of that, so I followed your advice and stored and restored modes. Edited: 10 Oct 2011, 8:23 a.m.

Re: WP34S: An integration wrapper Message #8 Posted by Marcus von Cube, Germany on 10 Oct 2011, 8:38 a.m., in response to message #7 by Miguel Toro Quote: so i need a four register stack to take advantage of that Only if you rely on the duplication of z/t when doing a complex drop or an operation that drops the stack. The way it is now it works and that's fine. You should note in the instructions that J is used for saving the mode. Maybe a limit should be set on the number of iterations. Do you do the halving of intervals over the complete range over and over or do you have a check for each subinterval and divide only if a subinterval is outside the accuracy bounds? The latter would probably need some kind of backtracking which is obviously not possible without recursive programming.

Re: WP34S: An integration wrapper Message #9 Posted by Miguel Toro on 10 Oct 2011, 8:49 a.m., in response to message #8 by Marcus von Cube, Germany Yes. I will think about setting a limit on the number of iteration. Thank you Marcus, Miguel

Re: WP34S: An integration wrapper Message #10 Posted by Paul Dale on 10 Oct 2011, 4:57 p.m., in response to message #9 by Miguel Toro An iteration limit would be a good idea. I tried integrating LN(x) from 0 to 1 and gave up after my laptop had been crunching away for over half an hour. I was using ALL mode sound rounding was to 12 digits not a fixed number of decimals. - Pauli

Re: WP34S: An integration wrapper Message #11 Posted by Miguel Toro on 10 Oct 2011, 9:59 p.m., in response to message #10 by Paul Dale With an iteration limit of 15, LN gives -0.999973 in six seconds. I will leave this limit in the routine.

[ Return to Index | Top of Index ]

Go back to the main exhibit hall