Thread: Optimizing hot section of code.

I managed to "optimize" it even further into this stripped down version.
Problem is, there was no change in performance. I can only assume this is what
the compiler ended up with. flp1969 is right in that compilers have gotten really
really good at optimizing code. Really.

The only way I can see to make this faster is to eliminate the right shift operation;
but then I would have to make the array humongus to accomodate the 64 bit indeces.
I assume that alone would bring negative performance effects to other sections of
of the program that work just fine now because they fit nicely in the cache.

TBH, I didn't bother to test it. As it is now, I'm at quite close to Openssl's DES
performance. And my base64 encoding is over 100% faster. Considering my code
-although by no means perfect- is actually readable, and the unholy mess that
the Openssl code base is, case and point, I'd say I'm pretty happy with the result.

Code:

static inline uint32_t    compress(const uint64_t block)
{
    register uint32_t    compressed;

    compressed = 0;
    compressed |= (uint32_t)g_sboxes[0][block >> 58] << 28;
    compressed |= (uint32_t)g_sboxes[1][(block >> 52) & 0x3f] << 24;
    compressed |= (uint32_t)g_sboxes[2][(block >> 46) & 0x3f] << 20;
    compressed |= (uint32_t)g_sboxes[3][(block >> 40) & 0x3f] << 16;
    compressed |= (uint32_t)g_sboxes[4][(block >> 34) & 0x3f] << 12;
    compressed |= (uint32_t)g_sboxes[5][(block >> 28) & 0x3f] << 8;
    compressed |= (uint32_t)g_sboxes[6][(block >> 22) & 0x3f] << 4;
    compressed |= (uint32_t)g_sboxes[7][(block >> 18) & 0x3f];
    return (compressed);
}

Next I have to implement the RSA section, which entails recreating the genrsa,
rsa and rsautl Openssl commands with all their options, as well as creating my own
random number generator. I fear I'm not quite up to the task

FYI, this is a fun website to play with: http://quick-bench.com/AolB5Vfwcex4OYDbGoAF8rZwo54

gg