Clear heuristic candidate list without try-finally

This commit is contained in:
Marek Lemańczyk committed 2025-10-15 23:57:46 +02:00
1 parent 51730bcba3
commit 0a23502441
1 file changed
+99 -103
+99 -103
View File
@@ -407,131 +407,127 @@ internal static partial class PrimeOrderCalculator
int capacity = config.MaxPowChecks <= 0 ? 64 : config.MaxPowChecks * 4;
List<ulong> candidates = AcquireHeuristicCandidateList(capacity);
try
FactorEntry[] factorArray = orderFactors.Factors!;
// DebugLog("Building candidates list");
BuildCandidates(order, factorArray, orderFactors.Count, candidates, capacity);
if (candidates.Count == 0)
{
FactorEntry[] factorArray = orderFactors.Factors!;
// DebugLog("Building candidates list");
BuildCandidates(order, factorArray, orderFactors.Count, candidates, capacity);
if (candidates.Count == 0)
candidates.Clear();
return false;
}
// DebugLog("Sorting candidates");
SortCandidates(prime, previousOrder, candidates);
int powBudget = config.MaxPowChecks <= 0 ? candidates.Count : config.MaxPowChecks;
int powUsed = 0;
int candidateCount = candidates.Count;
bool allowGpuBatch = true;
Span<ulong> candidateSpan = CollectionsMarshal.AsSpan(candidates);
// DebugLog(() => $"Checking candidates ({candidateCount} candidates, {powBudget} pow budget)");
int index = 0;
const int MaxGpuBatchSize = 256;
const int StackGpuBatchSize = 64;
Span<ulong> stackGpuRemainders = stackalloc ulong[StackGpuBatchSize];
ArrayPool<ulong> pool = ThreadStaticPools.UlongPool;
while (index < candidateCount && powUsed < powBudget)
{
int remaining = candidateCount - index;
int budgetRemaining = powBudget - powUsed;
int batchSize = Math.Min(remaining, Math.Min(budgetRemaining, MaxGpuBatchSize));
if (batchSize <= 0)
{
return false;
break;
}
// DebugLog("Sorting candidates");
SortCandidates(prime, previousOrder, candidates);
int powBudget = config.MaxPowChecks <= 0 ? candidates.Count : config.MaxPowChecks;
int powUsed = 0;
int candidateCount = candidates.Count;
bool allowGpuBatch = true;
Span<ulong> candidateSpan = CollectionsMarshal.AsSpan(candidates);
// DebugLog(() => $"Checking candidates ({candidateCount} candidates, {powBudget} pow budget)");
int index = 0;
const int MaxGpuBatchSize = 256;
const int StackGpuBatchSize = 64;
Span<ulong> stackGpuRemainders = stackalloc ulong[StackGpuBatchSize];
ArrayPool<ulong> pool = ThreadStaticPools.UlongPool;
while (index < candidateCount && powUsed < powBudget)
ReadOnlySpan<ulong> batch = candidateSpan.Slice(index, batchSize);
ulong[]? gpuPool = null;
Span<ulong> pooledGpuRemainders = default;
bool gpuSuccess = false;
bool gpuStackRemainders = false;
GpuPow2ModStatus status = GpuPow2ModStatus.Unavailable;
if (allowGpuBatch && IsGpuPow2Allowed)
{
int remaining = candidateCount - index;
int budgetRemaining = powBudget - powUsed;
int batchSize = Math.Min(remaining, Math.Min(budgetRemaining, MaxGpuBatchSize));
if (batchSize <= 0)
if (batchSize <= StackGpuBatchSize)
{
break;
}
ReadOnlySpan<ulong> batch = candidateSpan.Slice(index, batchSize);
ulong[]? gpuPool = null;
Span<ulong> pooledGpuRemainders = default;
bool gpuSuccess = false;
bool gpuStackRemainders = false;
GpuPow2ModStatus status = GpuPow2ModStatus.Unavailable;
if (allowGpuBatch && IsGpuPow2Allowed)
{
if (batchSize <= StackGpuBatchSize)
Span<ulong> localRemainders = stackGpuRemainders.Slice(0, batchSize);
status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, localRemainders, divisorData);
if (status == GpuPow2ModStatus.Success)
{
Span<ulong> localRemainders = stackGpuRemainders.Slice(0, batchSize);
status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, localRemainders, divisorData);
if (status == GpuPow2ModStatus.Success)
{
gpuSuccess = true;
gpuStackRemainders = true;
}
}
else
{
gpuPool = pool.Rent(batchSize);
Span<ulong> pooledRemainders = gpuPool.AsSpan(0, batchSize);
status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, pooledRemainders, divisorData);
if (status == GpuPow2ModStatus.Success)
{
pooledGpuRemainders = pooledRemainders;
gpuSuccess = true;
}
else
{
pool.Return(gpuPool, clearArray: false);
gpuPool = null;
}
}
if (!gpuSuccess && (status == GpuPow2ModStatus.Overflow || status == GpuPow2ModStatus.Unavailable))
{
allowGpuBatch = false;
gpuSuccess = true;
gpuStackRemainders = true;
}
}
for (int i = 0; i < batchSize && powUsed < powBudget; i++)
else
{
ulong candidate = batch[i];
powUsed++;
bool equalsOne;
if (gpuSuccess)
gpuPool = pool.Rent(batchSize);
Span<ulong> pooledRemainders = gpuPool.AsSpan(0, batchSize);
status = PrimeOrderGpuHeuristics.TryPow2ModBatch(batch, prime, pooledRemainders, divisorData);
if (status == GpuPow2ModStatus.Success)
{
ulong remainderValue = gpuStackRemainders ? stackGpuRemainders[i] : pooledGpuRemainders[i];
equalsOne = remainderValue == 1UL;
pooledGpuRemainders = pooledRemainders;
gpuSuccess = true;
}
else
{
equalsOne = Pow2EqualsOneCpu(candidate, prime, divisorData);
}
if (!equalsOne)
{
continue;
}
if (!TryConfirmCandidateCpu(prime, candidate, divisorData, config, ref powUsed, powBudget))
{
continue;
}
if (gpuPool is not null)
{
pool.Return(gpuPool, clearArray: false);
gpuPool = null;
}
result = candidate;
return true;
}
if (!gpuSuccess && (status == GpuPow2ModStatus.Overflow || status == GpuPow2ModStatus.Unavailable))
{
allowGpuBatch = false;
}
}
for (int i = 0; i < batchSize && powUsed < powBudget; i++)
{
ulong candidate = batch[i];
powUsed++;
bool equalsOne;
if (gpuSuccess)
{
ulong remainderValue = gpuStackRemainders ? stackGpuRemainders[i] : pooledGpuRemainders[i];
equalsOne = remainderValue == 1UL;
}
else
{
equalsOne = Pow2EqualsOneCpu(candidate, prime, divisorData);
}
if (!equalsOne)
{
continue;
}
if (!TryConfirmCandidateCpu(prime, candidate, divisorData, config, ref powUsed, powBudget))
{
continue;
}
if (gpuPool is not null)
{
pool.Return(gpuPool, clearArray: false);
}
index += batchSize;
candidates.Clear();
result = candidate;
return true;
}
if (gpuPool is not null)
{
pool.Return(gpuPool, clearArray: false);
}
index += batchSize;
}
finally
{
candidates.Clear();
}
// DebugLog("No candidate confirmed");
candidates.Clear();
// DebugLog("No candidate confirmed");
return false;
}
finally