## Introduction

This is the first in what I am hoping is a series of articles describing some of my coding for Cryptography and Prime Numbers. You can find all of this
on the Internet (somewhere) but I am trying to give a newbie a jumping off point. One of the first things you will need for Cryptography and Primality Testing
is some prime numbers. A thousand is a good start but a million or more is better. And once the bugs are out, I trust the PC and code more than my own ability
to perform perfect data entry for a million Primes. And that is a lot of work. The other option is to buy such a list but I am cheap. And there are free lists available
but there are format issues there. (Plus, there might be an error in the file. Unlikely but possible.)

Originally I thought I could just start at 7, take its square root, check for divisors from 2 up to that square root. If no divisors found, the number
is prime and put it into a `List`

. If a divisor is found, skip to the next number and repeat. This is fine for a single number that is not too large or for a list
of a couple thousand Primes (I suppose), but it gets slow waiting for a lot of numbers to be checked in this way.

Late at night, I read on some Cryptography board a very smart idea someone described. Set an upper bound for the highest number
you want to go to, and then starting at the lowest Prime, mark the square of that number as being not Prime and march up to the highest number by increments
of twice the Prime. When that lowest Prime has finished, go to the next unmarked Prime and repeat the squaring and setting loop. (If that seems complicated, look at the pictures below.
Pictures are how I make sense out of complex problems.)

As for what you get for your speed of running... it will depend on your PC (CPU speed and memory size and speed, how much you have running
at the same time, etc.). If the code runs too slow, lower the iMax number.

## Background

Avoid testing and setting all the numbers that are cleanly divisible by 2, 3, and 5. This can be done by taking the Modulus 30 of each number. (And it reduces the amount
of numbers that need to be examined to 8 out of 30.) A nice graphic of this idea can be seen in Rick Oden's article (here). Imagine a table of 30 columns reduced to remove all columns with even numbers as well
as those divisible by by 3 or 5. The first 6 rows of this are:

Every one of those numbers is potentially Prime. For the sake of argument, skip the number "1". I just put it into the table because otherwise
there is an empty space there and the top row looks aesthetically pleasing because it holds "N % 30" for every cell beneath it. (Some smart person is going
to object: but "1" is not Prime! I say, chill. It's a special case. 1 is the Identity for Multiplication. Just let it slide for now.)

So starting with the number "7", square that number to get "49". We mark that as not Prime. And we also advance by 7 (which is even
and not on the table) to "56". Skipping that, we go 7 more and reach "77". And to save time, we will not try to advance by 7. (An odd plus an odd
number is always an even (and not on our table). An odd number and twice the Prime in question will be odd and may be on the table.) We add 14 to get to the next
number that has a factor of 7, which is "91"... and on we go doing that, gaps of 14, and marking that number as not Prime. If a number is not on our table,
we keep going anyway until we reach the maximum number. And our table will look like this:

The next unmarked number in the table is "11". Its square is "121". Mark it not Prime. Add 22 and get to "143",
mark it not Prime. And on and on by steps of 22...

If we advance to a marked cell, it's composite and can be skipped. ("49" is our first such cell. And we skip it. The next cell after it is 53,
square it and set that, and increment by 106 and it's back to business as usual.)

To code this, allocate a large array of bytes to work in. Bytes_Allocated * 30 = N = Maximum Prime Number to march up to. Each byte in the array corresponds
to N / 30. Each bit in a byte corresponds to the 8 possible values of N % 30. (1, 7, 11, 13, 17, 19, 23, 29) for 8 possible values. 1 per bit. So, if we allocate
a 1000 bytes, we can cover the Primes up to the number 30000. When the code finishes, any bit left unset will represent a Prime.
In this case, I am adding each found Prime to a `List`

.

## Using the Code

The following C# code can be cut and paste into Visual Studio. Tie it into your own button click. After that, all the usual - use the code at your own risk
and I am not responsible for your possibly getting hooked on chasing Prime Numbers or Primality Testing.

There are no fancy classes here. Just going for speed with some globals, passing by reference, and the optimizations described above for what to even consider Prime.

List<UInt32> listPrimes = new List<UInt32>();
private void button8_Click(object sender, EventArgs e)
{
UInt32 iBump = 2;
UInt32 iPrime = 0;
UInt32 iPrime2 = 0;
UInt32 iPrimeSqr = 0;
UInt32 iMax = 20000; iMax = 200000; iMax = 2000000; iMax = 20000000; UInt32 iMax8 = iMax * 8;
UInt64 iiMax8 = iMax * 8;
if ( iiMax8 > UInt32.MaxValue )
{
System.Diagnostics.Debug.WriteLine (
"Too many numbers requested for this data type!");
return;
}
byte[] aPrimesBool = new byte [ iMax ];
UInt32 iFactor = 0;
DateTime dt0 = DateTime.Now;
DateTime dt1 = DateTime.Now;
listPrimes.Clear();
listPrimes.Add ( 2 );
listPrimes.Add ( 3 );
listPrimes.Add ( 5 );
for ( iFactor = 0; iFactor < iMax; iFactor += 1 )
{
aPrimesBool [ iFactor ] = 0xff;
}
dt1 = DateTime.Now;
System.Diagnostics.Debug.WriteLine ( "start after : " +
( dt1 - dt0).TotalSeconds + " sec");
_iGW = 0; _iGR = 7; for ( iPrime = 7; iPrime < iMax8; iPrime += iBump )
{
if ( iPrime % 1000000 == 1 )
{
dt1 = DateTime.Now;
System.Diagnostics.Debug.WriteLine ( "reached: " +
iPrime + " after : " + ( dt1 - dt0).TotalSeconds + " sec");
}
if ( GetCodePrime ( ref aPrimesBool ) ) {
listPrimes.Add ( iPrime );
iPrime2 = iPrime * 2;
iPrimeSqr = iPrime * iPrime;
for ( iFactor = iPrimeSqr; iFactor < iMax8; iFactor += iPrime2 )
{
ClrCodePrime ( ref iFactor, ref aPrimesBool );
}
}
if ( iBump == 2 )
iBump = 4;
else
iBump = 2;
if ( _iGR < 29 )
{
_iGR += iBump; }
else
{
_iGR = 1;
_iGW++;
}
}
dt1 = DateTime.Now;
System.Diagnostics.Debug.WriteLine ( "done: " + iPrime );
System.Diagnostics.Debug.WriteLine ( "count = " + listPrimes.Count );
aPrimesBool = null;
System.GC.WaitForPendingFinalizers();
}
UInt32 _iCR = 0; UInt32 _iCW = 0; private void ClrCodePrime ( ref UInt32 iNum, ref byte[] bA )
{
_iCR = iNum % 30;
switch ( _iCR )
{
case 1:
_iCW = iNum / 30; bA [ _iCW ] &= 0xfe; return;
case 7:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xfd; return;
case 11:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xfb; return;
case 13:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xf7; return;
case 17:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xef; return;
case 19:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xdf; return;
case 23:
_iCW = iNum / 30;
bA [ _iCW ] &= 0xbf; return;
case 29:
_iCW = iNum / 30;
bA [ _iCW ] &= 0x7f; return;
}
return;
}
UInt32 _iGR = 0; UInt32 _iGW = 0; private bool GetCodePrime ( ref byte[] bA )
{
switch ( _iGR )
{
case 1: return ( bA[_iGW] & 0x01 ) == 0x01;
case 7: return ( bA[_iGW] & 0x02 ) == 0x02;
case 11: return ( bA[_iGW] & 0x04 ) == 0x04;
case 13: return ( bA[_iGW] & 0x08 ) == 0x08;
case 17: return ( bA[_iGW] & 0x10 ) == 0x10;
case 19: return ( bA[_iGW] & 0x20 ) == 0x20;
case 23: return ( bA[_iGW] & 0x40 ) == 0x40;
case 29: return ( bA[_iGW] & 0x80 ) == 0x80;
}
return false;
}

I have shied away from multi-threading to make it a little easier to understand. If you find a way to make this code run faster, by all means please share what you changed.

## Points of Interest

Some cute bit clearing is going on in the `ClrCodePrime`

routine. And because it is not good classes, be careful when changing stuff. It is very possible
to break something accidentally. (Part of the whole thing with why classes are better than this dinosaur style of code. I maintain that the old lizard runs faster.
But if you have a faster class, share and we can all decide if it is better.)

With generating a lot of Primes, it is less efficient to compute each Prime individually (and look beneath it to see that it has no Factors)
and more efficient for each Prime to set all of the factorable Composites above it to some requested Maximum.

## History

- Initial article = 9/16/2011.