34S SUM PROD change request ccccc >= fff

[Egan Ford]

Could you please change "ccccc > fff" to "ccccc >= fff" so that ccccc can count down to and include 0.

Thanks.

[Paul Dale]

This is somewhat limited by the definition of DSE unfortunately.

Here is the sum and product code as it currently exists in the firmware image. It is just a plain old keystroke program. Feel free to submit a patch :-)

The stack will always be four levels deep when this code is run and registers 0-4 and flags 0-14 are freely available.

Then again, I could look into adding DSL and ISE as instructions.

- Pauli

/**************************************************************************/

/* Sigma and products

 * Register use:

 * 0	I

 * 1	product/sum

 * 2	carry for sum

 * 3	saved I

 */

		LBL 99					/* Entry: SUMMATION */

			XEQ entry

			SPEC?

				JMP sum_product_okay

			STO 00

			STO 03				/* Save for LastX*/

			IP				/* First function call is separate*/

			FILL				/* to avoid Kahan summing from zero*/

			XEQUSR				/* six extra instructions save nine*/

			SPEC?				/* from executing*/

				JMP sum_product_nan

			STO 01

			iC 0

			STO 02

			JMP sum_entry

sum_loop::		RCL 00

			IP

			FILL

			XEQUSR

			SPEC?

				JMP sum_product_nan

			RCL- 02

			ENTER[^]

			RCL+ 01

			ENTER[^]

			RCL- 01

			RCL- Z

			x[<->]y

			[cmplx]STO 01

sum_entry::		DSE 00

				JMP sum_loop

			JMP sum_product_okay



		LBL 98					/* Entry: PRODUCT */

			XEQ entry

			SPEC?

				JMP sum_product_okay

			STO 00

			STO 03

			IP				/* First function call is separate*/

			FILL				/* to avoid a multiply*/

			XEQUSR

			SPEC?

				JMP sum_product_nan

			STO 01

			JMP product_entry

product_loop::		RCL 00

			IP

			FILL

			XEQUSR

			SPEC?

				JMP sum_product_nan

			STO[times] 01

product_entry::		DSE 00

				JMP product_loop

sum_product_okay::	RCL 03

			STO L

			iC 0

			FILL

			RCL 01

			JMP exit

sum_product_nan::	RCL 03

sum_product_error::	STO L

			iC 0

			FILL

			# NaN

			JMP exit

[Paul Dale]

DSL and ISE are now present :-)

- Pauli

[Walter B]

... and documented in the manual :-)

Walter

[Marcus von Cube, Germany]

Pauli, what is the reasoning behind the complicated summation. It must have to do with increasing accuracy but how does it work?

If you want to make the code shorter, the two loops may easily be merged and the operation selected by a flag inside the loop.

[Paul Dale]

The summation is indeed an attempt to improve precision. The algorithm used is known as the Kahan sum.

Essentially, it maintains the sum and some of the digits lost off the end in two registers. The lost digits act as a correction term as each new item is added in.

Yes, the two loops could be merged. I didn't investigate this option since neither is very large.

- Pauli

[Marcus von Cube, Germany]

We will certainly be fighting for a handful of bytes again...