Решето Эратосфена (сегмент): медленно работает - как можно ускорить?

typedef UINT64 Number;
 
__device__ Number eratosthenesSingleBlock(const Number from, const Number to)
{
    const Number memorySize = (to - from + 1) / 2;
 
    // initialize
    char* isPrime = new char[memorySize];
    memset(isPrime, 1, memorySize);
 
    for (Number i = 3; i*i <= to; i += 2)
    {
        // skip multiples of three: 9, 15, 21, 27, ...
        if (i >= 3 * 3 && i % 3 == 0)
            continue;
        // skip multiples of five
        if (i >= 5 * 5 && i % 5 == 0)
            continue;
        // skip multiples of seven
        if (i >= 7 * 7 && i % 7 == 0)
            continue;
        // skip multiples of eleven
        if (i >= 11 * 11 && i % 11 == 0)
            continue;
        // skip multiples of thirteen
        if (i >= 13 * 13 && i % 13 == 0)
            continue;
 
        // skip numbers before current slice
        Number minJ = ((from + i - 1) / i)*i;
        if (minJ < i*i)
            minJ = i*i;
 
        // start value must be odd
        if ((minJ & 1) == 0)
            minJ += i;
 
        // find all odd non-primes
        for (Number j = minJ; j <= to; j += 2 * i)
        {
            Number index = j - from;
            isPrime[index / 2] = 0;
        }
    }
 
    // count primes in this block
    UINT32 found = 0;
    for (UINT32 i = 0; i < memorySize; i++)
        found += isPrime[i];
    // 2 is not odd => include on demand
    if (from <= 2)
        found++;
 
    delete[] isPrime;
    return found;
}

D:\aProjects\CudaTest\x64\Release>PrimeSearchCuda.exe
 
Prime search:   4000000000000 .. 4001000000000
 
Accelerator: GeForce GTX 460
Found : 34462887
First prime :   4000000000039
Last prime  :   4000999999987
=============
Exec time : 110 seconds.
Primes/sec : 313663