Ok, so I'm probably hurting my income, but sharing some hints for optimizations:
A quick'n'dirty profiling run showed that we spend very much time in memory allocations, so I am trying to eliminate some of them.
-did the headerhash mod outlined earlier
-did the time limit on Weave() from Sunny King
-removed some printf's , especially
printf("MineProbablePrimeChain() : new sieve (%u/%u) ready in %uus\n", psieve->GetCandidateCount(), nMaxSieveSize, (unsigned int) (GetTimeMicros() - nStart));
from prime.cpp since it's the only call to the rather expensive GetCandidateCount
-after creating vPrimes array in prime.cpp, store vPrimes.size() in a static variably and replace all calls to it with the variable
-in prime.cpp, PrimeTableGetNextPrime and PrimeTableGetPreviousPrime: Use upper_bound/lower_bound instead of linear search:
// Get next prime number of p
bool PrimeTableGetNextPrime(unsigned int& p)
{
std::vector<unsigned int>::iterator foundelement=upper_bound(vPrimes.begin(), vPrimes.end(),p);
if (foundelement!=vPrimes.end()){
p=*foundelement;
return true;
}
return false;
}
// Get previous prime number of p
bool PrimeTableGetPreviousPrime(unsigned int& p)
{
std::vector<unsigned int>::iterator foundelement=lower_bound(vPrimes.begin(), vPrimes.end(),p);
if (foundelement!=vPrimes.end()){
p=*--foundelement;
return true;
}
return false;
}
- in prime.cpp, CSieveOfEratosthenes::Weave() :
Split the for loop into two for loops, one going up to nChainLength, the other one going up to 2*nChainLength
Thus, you can safe the first first if+for clause in the first loop and leave away the if, and leave the if+for completely away in the second loop.
Also,
if (((nBiTwinSeq & 1u) == 0))
...
if (((nBiTwinSeq & 1u) == 1u))
can be substituted by
if (((nBiTwinSeq & 1u) == 0))
....
else
.....
-prime.h: Removed nSieveSize and replaced all occurrences with nMaxSieveSize. Removed nSieveSize from the initializer as it's become obsolete.
-prime.h: Changed
< std::vector<bool> vfCompositeCunningham1;
< std::vector<bool> vfCompositeCunningham2;
< std::vector<bool> vfCompositeBiTwin;
---
> bool vfCompositeCunningham1[1000000];
> bool vfCompositeCunningham2[1000000];
> bool vfCompositeBiTwin[1000000];
Thus, we should save on quite a few mallocs.
-prime.h: Removed nSieveSize and replaced all occurrences with nMaxSieveSize. Removed nSieveSize from the initializer as it's become obsolete.
-prime.h: Changed
< std::vector<bool> vfCompositeCunningham1;
< std::vector<bool> vfCompositeCunningham2;
< std::vector<bool> vfCompositeBiTwin;
---
> bool vfCompositeCunningham1[1000000];
> bool vfCompositeCunningham2[1000000];
> bool vfCompositeBiTwin[1000000];
Thus, we should save on quite a few mallocs.
Don't forget to initialize the array by adding the following below in place of the other initialization
for (int i = 0; i<1000000; ++i){
vfCompositeCunningham1[i] = false;
vfCompositeCunningham2[i] = false;
vfCompositeBiTwin[i] = false;
}
I think it'd be ideal to use a 2D array with the three composite values stored sequentially in memory as single bits each
Probably should be a std::bitset datatype too
+1
to the greedy people!? killurselvessry to be an ass but some of you coders are shitting on my big smile i had when i woke up today! and to think the people I am talking about... I have randomly spoken very positive about in different place in the coin community! psh!
