Skip to content

Commit

Permalink
Merge pull request #206 from qubic/develop
Browse files Browse the repository at this point in the history 
Release/v1.224.2 to main
  • Loading branch information
@philippwerner
philippwerner authored Oct 24, 2024
2 parents 765a9d4 + e946fb8 commit bd249ac
Show file tree
Hide file tree
Showing 3 changed files with 85 additions and 45 deletions.
2 changes: 1 addition & 1 deletion src/public_settings.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,7 @@

#define VERSION_A 1
#define VERSION_B 224
#define VERSION_C 1
#define VERSION_C 2

// Epoch and initial tick for node startup
#define EPOCH 132
Expand Down
127 changes: 84 additions & 43 deletions src/score.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,12 +27,15 @@ template<
>
struct ScoreFunction
{
static constexpr unsigned long long signOffset = 64;
static constexpr unsigned long long allNeuronsCount = dataLength + numberOfHiddenNeurons + dataLength;
static constexpr unsigned long long computeNeuronsCount = numberOfHiddenNeurons + dataLength;
static constexpr unsigned long long synapseSignsCount = (dataLength + numberOfHiddenNeurons + dataLength) * numberOfNeighborNeurons / 64;
static constexpr unsigned long long synapseInputCount = synapseSignsCount + maxDuration;

static_assert(allNeuronsCount < 0xFFFFFFFF, "Current implementation only support MAX_UINT32 neuron");
static_assert(numberOfNeighborNeurons < 0x7FFFFFFF, "Current implementation only support MAX_UINT32 number of neighbors");
static_assert((allNeuronsCount* numberOfNeighborNeurons) % 64 == 0, "numberOfNeighborNeurons * allNeuronsCount must dividable by 64");

long long miningData[dataLength];

struct K12EngineX1 {
Expand All @@ -43,6 +46,8 @@ struct ScoreFunction
unsigned long long Asa, Ase, Asi, Aso, Asu;
unsigned long long scatteredStates[25];
int leftByteInCurrentState;
unsigned char* _pPoolBuffer;
unsigned int _x;
private:
void _scatterFromVector() {
copyToStateScalar(scatteredStates)
Expand All @@ -58,7 +63,8 @@ struct ScoreFunction
}
public:
K12EngineX1() {}
void initState(const unsigned long long* comp_u64, const unsigned long long* nonce_u64) {
void initState(const unsigned long long* comp_u64, const unsigned long long* nonce_u64, unsigned char* pPoolBuffer)
{
Aba = comp_u64[0];
Abe = comp_u64[1];
Abi = comp_u64[2];
Expand All @@ -69,8 +75,14 @@ struct ScoreFunction
Agi = nonce_u64[3];
Ago = Agu = Aka = Ake = Aki = Ako = Aku = Ama = Ame = Ami = Amo = Amu = Asa = Ase = Asi = Aso = Asu = 0;
leftByteInCurrentState = 0;

_x = 0;
_pPoolBuffer = pPoolBuffer;
write(_pPoolBuffer, RANDOM2_POOL_SIZE);
}
void write(unsigned char* out0, int size) {

void write(unsigned char* out0, int size)
{
unsigned char* s0 = (unsigned char*)scatteredStates;
if (leftByteInCurrentState) {
int copySize = size < leftByteInCurrentState ? size : leftByteInCurrentState;
Expand All @@ -91,6 +103,16 @@ struct ScoreFunction
}
}

unsigned int random2FromPrecomputedPool(unsigned char* output, unsigned long long outputSize)
{
for (unsigned long long i = 0; i < outputSize; i += 8)
{
*((unsigned long long*) & output[i]) = *((unsigned long long*) & _pPoolBuffer[_x & (RANDOM2_POOL_ACTUAL_SIZE - 1)]);
_x = _x * 1664525 + 1013904223;// https://en.wikipedia.org/wiki/Linear_congruential_generator#Parameters_in_common_use
}
return _x;
}

void scatterFromVector() {
_scatterFromVector();
}
Expand All @@ -112,26 +134,27 @@ struct ScoreFunction
}
};

struct PoolSynapseData
{
unsigned int neuronIndex;
unsigned int supplierIndexWithSign;
};

struct computeBuffer {
struct Neuron
{
char* input;
};
struct Synapse
{
//unsigned long long signs[(dataLength + numberOfHiddenNeurons + dataLength) * numberOfNeighborNeurons / 64];
//unsigned long long sequence[maxDuration];
unsigned long long* data;

// Pointer to data
unsigned long long* signs;
unsigned long long* sequence;
};
K12EngineX1 k12;
Neuron _neurons;
Synapse _synapses;

unsigned char* _poolBuffer;
PoolSynapseData* _poolSynapseData;
unsigned char* _poolRandom2Buffer;

} *_computeBuffer = nullptr;
m256i currentRandomSeed;
Expand Down Expand Up @@ -167,20 +190,20 @@ struct ScoreFunction
{
for (unsigned int i = 0; i < solutionBufferCount; i++)
{
if (_computeBuffer[i]._poolBuffer)
if (_computeBuffer[i]._poolRandom2Buffer)
{
freePool(_computeBuffer[i]._poolBuffer);
freePool(_computeBuffer[i]._poolRandom2Buffer);
}

if (_computeBuffer[i]._neurons.input)
{
freePool(_computeBuffer[i]._neurons.input);
_computeBuffer[i]._neurons.input = nullptr;
}
if (_computeBuffer[i]._synapses.data)
if (_computeBuffer[i]._synapses.signs)
{
freePool(_computeBuffer[i]._synapses.data);
_computeBuffer[i]._synapses.data = nullptr;
freePool(_computeBuffer[i]._synapses.signs);
_computeBuffer[i]._synapses.signs = nullptr;
}
}

Expand All @@ -203,7 +226,7 @@ struct ScoreFunction
{
auto& cb = _computeBuffer[bufId];

if (!allocatePool(RANDOM2_POOL_SIZE, (void**)&(cb._poolBuffer)))
if (!allocatePool(RANDOM2_POOL_SIZE, (void**)&(cb._poolRandom2Buffer)))
{
logToConsole(L"Failed to allocate memory for score pool buffer!");
return false;
Expand All @@ -219,30 +242,35 @@ struct ScoreFunction
return false;
}

if (!allocatePool(synapseInputCount * sizeof(unsigned long long), (void**)&(cb._synapses.data)))
if (!allocatePool(synapseSignsCount * sizeof(unsigned long long), (void**)&(cb._synapses.signs)))
{
CHAR16 log[256];
setText(log, L"Failed to allocate memory for synapses! Try to allocated ");
appendNumber(log, synapseInputCount * sizeof(unsigned long long) / 1024, true);
appendNumber(log, synapseSignsCount * sizeof(unsigned long long) / 1024, true);
appendText(log, L" KB");
logToConsole(log);
return false;
}

if (!allocatePool(RANDOM2_POOL_SIZE * sizeof(PoolSynapseData), (void**)&(cb._poolSynapseData)))
{
CHAR16 log[256];
setText(log, L"Failed to allocate memory for pool synapse data! Try to allocated ");
appendNumber(log, RANDOM2_POOL_SIZE * sizeof(PoolSynapseData) / 1024, true);
appendText(log, L" KB");
logToConsole(log);
return false;
}
}
}

for (int i = 0; i < solutionBufferCount; i++) {
setMem(_computeBuffer[i]._synapses.data, sizeof(_computeBuffer[i]._synapses.data[0]) * synapseInputCount, 0);
_computeBuffer[i]._synapses.signs = _computeBuffer[i]._synapses.data;
_computeBuffer[i]._synapses.sequence = _computeBuffer[i]._synapses.data + synapseSignsCount;

setMem(_computeBuffer[i]._synapses.signs, sizeof(_computeBuffer[i]._synapses.signs[0]) * synapseSignsCount, 0);
setMem(_computeBuffer[i]._poolSynapseData, sizeof(_computeBuffer[i]._poolSynapseData[0]) * RANDOM2_POOL_SIZE, 0);
setMem(_computeBuffer[i]._neurons.input, sizeof(_computeBuffer[i]._neurons.input[0]) * allNeuronsCount, 0);
solutionEngineLock[i] = 0;
}

static_assert(synapseInputCount * sizeof(*(_computeBuffer->_synapses.data)) % 8 == 0, "Random2 require output size is a multiplier of 8");

#if USE_SCORE_CACHE
scoreCacheLock = 0;
setMem(&scoreCache, sizeof(scoreCache), 0);
Expand Down Expand Up @@ -298,7 +326,7 @@ struct ScoreFunction

void generateSynapse(computeBuffer& cb, int solutionBufIdx, const m256i& publicKey, const m256i& nonce)
{
random2(&publicKey.m256i_u8[0], &nonce.m256i_u8[0], (unsigned char*)(cb._synapses.data), synapseInputCount * sizeof(cb._synapses.data[0]), cb._poolBuffer);
random2(&publicKey.m256i_u8[0], &nonce.m256i_u8[0], (unsigned char*)(cb._synapses.data), synapseInputCount * sizeof(cb._synapses.data[0]), cb._poolRandom2Buffer);
}

bool isValidScore(unsigned int solutionScore)
Expand All @@ -316,23 +344,29 @@ struct ScoreFunction
const int solutionBufIdx = (int)(processor_Number % solutionBufferCount);

unsigned int score = 0;
auto& cb = _computeBuffer[solutionBufIdx];
auto& synapses = cb._synapses;
unsigned int random2XVal = 0;
computeBuffer& cb = _computeBuffer[solutionBufIdx];
unsigned long long* pSynapseSigns = cb._synapses.signs;
PoolSynapseData* pPoolSynapseData = cb._poolSynapseData;
auto& neurons = cb._neurons;

setMem(neurons.input, sizeof(neurons.input[0]) * allNeuronsCount, 0);

generateSynapse(cb, solutionBufIdx, publicKey, nonce);

for (int i = 0; i < dataLength; i++)
{
neurons.input[i] = (char)miningData[i];
}

for (long long tick = 0; tick < maxDuration; tick++)
//generateSynapse(cb, solutionBufIdx, publicKey, nonce);
cb.k12.initState(&publicKey.m256i_u64[0], &nonce.m256i_u64[0], cb._poolRandom2Buffer);

random2XVal = cb.k12.random2FromPrecomputedPool((unsigned char*)pSynapseSigns, synapseSignsCount * sizeof(unsigned long long));

for (unsigned int i = 0; i < RANDOM2_POOL_SIZE; i++)
{
const unsigned long long neuronIndex = dataLength + synapses.sequence[tick] % computeNeuronsCount;
const unsigned long long neighborNeuronIndex = (synapses.sequence[tick] / computeNeuronsCount) % numberOfNeighborNeurons;
const unsigned int poolIdx = i & (RANDOM2_POOL_ACTUAL_SIZE - 1);
const unsigned long long poolValue = *((unsigned long long*) & cb._poolRandom2Buffer[poolIdx]);
const unsigned long long neuronIndex = dataLength + poolValue % computeNeuronsCount;
const unsigned long long neighborNeuronIndex = (poolValue / computeNeuronsCount) % numberOfNeighborNeurons;
unsigned long long supplierNeuronIndex;
if (neighborNeuronIndex < numberOfNeighborNeurons / 2)
{
Expand All @@ -344,18 +378,25 @@ struct ScoreFunction
}
const unsigned long long offset = neuronIndex * numberOfNeighborNeurons + neighborNeuronIndex;

char nnV = 0;
//if (!(synapses.signs[offset / 64] & (1ULL << (offset % 64))))
if (!(synapses.signs[offset >> 6] & (1ULL << (offset & 63ULL))))
{
nnV = neurons.input[supplierNeuronIndex];
}
else
{
nnV = -neurons.input[supplierNeuronIndex];
}
unsigned int isPositive = !(pSynapseSigns[offset >> 6] & (1ULL << (offset & 63ULL))) ? 1 : 0;

pPoolSynapseData[i].neuronIndex = (unsigned int)neuronIndex;
pPoolSynapseData[i].supplierIndexWithSign = ((unsigned int)supplierNeuronIndex << 1) | isPositive;
}

for (long long tick = 0; tick < maxDuration; tick++)
{
PoolSynapseData data = pPoolSynapseData[random2XVal & (RANDOM2_POOL_ACTUAL_SIZE - 1)];
unsigned int neuronIndex = data.neuronIndex;
unsigned int supplierNeuronIndex = (data.supplierIndexWithSign >> 1);
unsigned int sign = (data.supplierIndexWithSign & 1U);

char nnV = neurons.input[supplierNeuronIndex];
nnV = sign ? nnV : -nnV;
neurons.input[neuronIndex] += nnV;
clampNeuron(neurons.input[neuronIndex]);

random2XVal = random2XVal * 1664525 + 1013904223;
}

score = 0;
Expand Down
1 change: 0 additions & 1 deletion test/score_params.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,6 @@ static constexpr unsigned long long kSettings[][MAX_PARAM_TYPE] = {
{6000, 832, 52158792ULL},
{6080, 2792, 45545677ULL},
{6024, 4784, 35247639ULL},
{7192, 6475, 17517712ULL},
{7192, 5184, 32472324ULL},
{12088, 7864, 22188546ULL},

Expand Down

0 comments on commit bd249ac

Please sign in to comment.