disk | ||
blank_prontodos.dsk | ||
build.sh | ||
ca65_fixes.inc | ||
README.MD | ||
rpm_ca65.s | ||
rpm_m32.s | ||
rpm_oop.js | ||
rpm_proc.js | ||
rpm.bas | ||
rpm.c | ||
rpm.debug.bas | ||
rpm.dsk | ||
tasc_disasm_clean.s | ||
tasc_disasm_org.s |
Russian Peasant Multiplication
From Assembly to Basic to C to Javascript!
Here are my implementations of Russian Peasant Multiplication implemented in various languages:
- 6502 Assembly Language (Both ca65 and merlin32 sources)
- Applesoft BASIC
- JavaScript (Procedural version)
- JavaScript (OOP version)
A .dsk image has been provided as an convenience.
To see how much faster the Assembly version is then the BASIC version:
RUN RPM.BAS
BRUN RPM.BIN
And enter in 123456789
* 987654321
respectively for A and B ...
Version | Time |
---|---|
Applesoft | 33 s |
Assembly | ~1 s |
So what the heck is it?
An alternative algorithm to implement multiplication using only:
- bit-shifts (left and right), and
- addition.
Algorithm
- Initialize Sum <- zero. In C nomenclature:
Sum = 0;
- If B is odd then add A to Sum. In C nomenclature:
if( B & 1 ) Sum += A;
- Multiply A by 2 -- that is, Shift A left by one. In C nomenclature:
A <<= 1;
- Divide B by 2 -- that is, Shift B right by one. In C nomenclature:
B >>= 1;
- If B is zero then STOP.
while( b ) { ... }
- Goto step 2
Paste the following program into an online C compiler
#include <stdio.h>
int RPM( int a, int b )
{
int sum = 0;
while( b )
{
if( b & 1 )
sum += a;
a <<= 1;
b >>= 1;
}
return sum;
}
int main()
{
return printf( "%d\n", RPM( 86, 57 ) );
}
Examples
Example of "traditional" multiplication:
In base 10:
86
x 57
----
602
430
====
4902
In base 2:
01010110 (86)
00111001 (57)
--------
01010110 (86 * 2^0 = 86)
00000000 (86 * 2^1 = 172) <- wasted work, partial sum = 0
00000000 (86 * 2^2 = 344) <- wasted work, partial sum = 0
01010110 (86 * 2^3 = 688)
01010110 (86 * 2^4 = 1376)
01010110 (86 * 2^5 = 2752)
==============
01001100100110 (4902 = 86*2^0 + 86*2^3 + 86*2^4 + 86*2^5)
Example of Russian Peasant multiplication:
In Base 10:
A B B Odd? Sum = 0
86 57 Yes + A = 86
x 2 = 172 / 2 = 28 No = 86
x 2 = 344 / 2 = 14 No = 86
x 2 = 688 / 2 = 7 Yes + A = 774
x 2 = 1376 / 2 = 3 Yes + A = 2150
x 2 = 2752 / 2 = 1 Yes + A = 4902
In Base 2:
A B B Odd? Sum = 0
01010110 00111001 Yes + A = 00000001010110
010101100 00011100 No = 00000001010110
0101011000 00001110 No = 00000001010110
01010110000 00000111 Yes + A = 00001100000110
010101100000 00000011 Yes + A = 00100001100110
0101011000000 00000001 Yes + A = 01001100100110
In Base 8:
A B B Odd? Sum = 0
126 71 Yes + A = 126
x 2 = 254 / 2 = 34 No = 126
x 2 = 530 / 2 = 16 No = 126
x 2 = 1260 / 2 = 7 Yes + A = 1406
x 2 = 2540 / 2 = 3 Yes + A = 4146
x 2 = 5300 / 2 = 1 Yes + A = 11446
In Base 16:
A B B Odd? Sum = 0
56 39 Yes + A = 56
x 2 = AC / 2 = 1C No = 56
x 2 = 158 / 2 = E No = 56
x 2 = 2B0 / 2 = 7 Yes + A = 306
x 2 = 560 / 2 = 3 Yes + A = 866
x 2 = AC0 / 2 = 1 Yes + A = 1326
Bases
Does this algorithm work in other bases such as 2, 8, or 16?
Consider the question:
Q. Does multipling by 2 work in other bases? A. Yes.
Q. Why? A. When we write a number in a different base we have the same representation but a different presentation.
Adding, Multiplying, Dividing all function the same regardless of which base we use.
Efficiency
For a "BigInt" or "BigNumber" library this is NOT the most efficient (*) way to multiply numbers as it doesn't scale (**). However, it is rather trivial to implement. You only need a few functions:
isEven()
isZero()
Shl()
Shr()
AddTo()
Notes:
(*) Almost everyone uses FFT (Fast Fourier Transforms), Toom, and/or Karatsuba for multiplication. For example GMP, GNU Multiple Precision arithmetic library, uses seven different multiplication algorithms!
- Basecase
- Karatsuba
- Toom-3
- Toom-4
- Toom-6.5
- Toom-8.5
- FFT
(**) What do we mean by "Doesn't scale"? As the input numbers uses more bits we end up doing more work other other algorithms.