base:optimal_sort_8-bit_elements

by Mats Rosengren

The bubble sort description from “6502 Software Design” is well written and perfectly fills its purpose of being a pleasant and easy to read tutorial. I'll here try to use the same style to explain a more efficient algorithm that is just as short and equally easy to understand.

I'll use the same example, i.e.:

05 03 04 01 02

My algorithm uses an “outer loop” that is run through as many time as there are elements (i.e. in this case 5 times) and an “inner loop” that gets shorter and shorter when the outer loop progresses. The easiest way to explain the algorithm should be to simply apply it to this short sequence.

The first step in the outer loop will be to find the largest element and to directly put it at its correct place which is at the very end. The first action is therefore to copy the value which is at the last position to another place as this last position will be overwritten by the largest value. In the code below this other place is the byte given the label “WORK3”. In our example it is consequently 02 that is copied to WORK3.

Now starts the inner loop!

The last element, in this case 02, is again copied to another place. This second place is the byte given the label WORK2 in the code. A byte with the label WORK1 is then used as a “pointer” to the place from where the present value in WORK2 was copied. WORK1 is therefore initially given the value 5 which is the last position.

The value in WORK2 is then compared with the other elements of the sequence.

First it is compared with the value 01. As 02 is larger then 01 no action is taken and the inner loop continues to the next element which is 04.

As 04 is larger then the present value in WORK2 this larger value 04 is written to WORK2 and the pointer WORK1 is given the value 3 which is the position of 04.

The inner loop then continues, the only change is that the byte WORK2 now has the larger value 04. WORK2 will then again be updated when the value 05 is found and the pointer in WORK1 will be updated to corresponding index 1.

As the last element now has been reached the first run through the inner loop is finished. The value in WORK2 is the largest value and this is now written to the last position. The value 02 that originally was in this last position is available in WORK3 and the pointer WORK1 shows were the “free spot” is where this value can be put.

When this has been done the new sequence is

02 03 04 01 05

It is known that the largest value now is at the very end and what remain to be sorted is

02 03 04 01

The second step in the outer loop is therefore to find the largest element in this new shorter sequence and to put this largest element at the end of this new sequence using exactly the same inner loop. As this new sequence only have 4 elements the inner loop only has to use 4 steps. After this second step of the outer loop the total sequence is

02 03 01 04 05

Next step in the outer loop will then produce the sequence

02 01 03 04 05

and the final step

01 02 03 04 05

The 6502 code for this algorithm is as follows:

The subroutine (SORT) sorts unordered lists that are comprised of 8-bit elements. The starting address is contained in location ZPADD (low-address byte) and ZPADD + 1 (high-address byte). These bytes must be in the zero page. The length of the list is contained in byte NVAL. Since a byte is 8 bits wide, the list can contain up to 255 elements.

;SORTING SUBROUTINE CODED BY MATS ROSENGREN (MATS.ROSENGREN@ESA.INT) ; ; INPUT: ;ZPADD - START ADDRESS OF SEQUENCE TO BE SORTED SHALL BE PUT IN ZPADD, ZPADD+1 (ZERO PAGE) ; SHOULD POINT TO THE BYTE JUST BEFORE THE FIRST BYTE TO BE SORTED ; ( "LDA (ZPADD),Y" WITH 1<=Y<=255) ;NVAL - NUMBER OF VALUES, 1<= NVAL <= 255 ; VALUE WILL BE DESTROYED (SET TO ZERO) ; ZPADD = $30 ;2 BYTE POINTER IN PAGE ZERO. SET BY CALLING PROGRAM NVAL = $32 ;SET BY CALLING PROGRAM WORK1 = $33 ;3 BYTES USED AS WORKING AREA WORK2 = $34 WORK3 = $35 *=$6000 ;CODE ANYWHERE IN RAM OR ROM SORT LDY NVAL ;START OF SUBROUTINE SORT LDA (ZPADD),Y ;LAST VALUE IN (WHAT IS LEFT OF) SEQUENCE TO BE SORTED STA WORK3 ;SAVE VALUE. WILL BE OVER-WRITTEN BY LARGEST NUMBER BRA L2 L1 DEY BEQ L3 LDA (ZPADD),Y CMP WORK2 BCC L1 L2 STY WORK1 ;INDEX OF POTENTIALLY LARGEST VALUE STA WORK2 ;POTENTIALLY LARGEST VALUE BRA L1 L3 LDY NVAL ;WHERE THE LARGEST VALUE SHALL BE PUT LDA WORK2 ;THE LARGEST VALUE STA (ZPADD),Y ;PUT LARGEST VALUE IN PLACE LDY WORK1 ;INDEX OF FREE SPACE LDA WORK3 ;THE OVER-WRITTEN VALUE STA (ZPADD),Y ;PUT THE OVER-WRITTEN VALUE IN THE FREE SPACE DEC NVAL ;END OF THE SHORTER SEQUENCE STILL LEFT BNE SORT ;START WORKING WITH THE SHORTER SEQUENCE RTS

NVAL that initially is the number of values to be sorted is used as a pointer to the last element of the subsequence that is not yet sorted. This is the loop variable for the outer loop. The outer loop is finished when this loop variable takes the value zero using the conditional branch “BNE”.

The loop variable for the inner loop is register Y that runs from NVAL to 1, again using “BNE” as end condition.

To test the efficiency (and correctness) of this algorithm I generated a totally random permutation of the numbers 1 - 255 using a random number generator. This sequence was successfully sorted using subroutine SORT above. I measured the number of T1H counts of the W65C22S timer 1 during this sorting, it was 2006. As T1H decrements every 256 PHI2 pulses and my W6502 runs at 8 megahertz this corresponds to 0.0642 seconds. Sorting exactly the same sequence with the “Bubble Sort” routine I got 7276 T1H decrements corresponding to 0.233 seconds, i.e 3.6 times as much time was needed.

base/optimal_sort_8-bit_elements.txt · Last modified: 2015-04-17 04:33 (external edit)