Idea and original implementation by Stephane Nicolet
7 threads 15+0.05
ELO: 3.54 +-2.9 (95%) LOS: 99.2%
Total: 17971 W: 2976 L: 2793 D: 12202
There is no functional change in single thread mode
if ( sp->allSlavesSearching
&& sp->slavesMask.count() < MAX_SLAVES_PER_SPLITPOINT)
{
if ( sp->allSlavesSearching
&& sp->slavesMask.count() < MAX_SLAVES_PER_SPLITPOINT)
{
while ( sp.slavesMask.count() < MAX_SLAVES_PER_SPLITPOINT
&& (slave = Threads.available_slave(&sp)) != nullptr)
{
while ( sp.slavesMask.count() < MAX_SLAVES_PER_SPLITPOINT
&& (slave = Threads.available_slave(&sp)) != nullptr)
{
+ slave->allocMutex.lock();
if (slave->can_join(activeSplitPoint))
{
activeSplitPoint->slavesMask.set(slave->idx);
slave->activeSplitPoint = activeSplitPoint;
slave->searching = true;
if (slave->can_join(activeSplitPoint))
{
activeSplitPoint->slavesMask.set(slave->idx);
slave->activeSplitPoint = activeSplitPoint;
slave->searching = true;
- slave->sleepCondition.notify_one(); // Could be sleeping
+ slave->allocMutex.unlock();
+
+ slave->notify_one(); // Could be sleeping
}
// Everything is set up. The master thread enters the idle loop, from which
}
// Everything is set up. The master thread enters the idle loop, from which
const size_t MAX_SPLITPOINTS_PER_THREAD = 8;
const size_t MAX_SLAVES_PER_SPLITPOINT = 4;
const size_t MAX_SPLITPOINTS_PER_THREAD = 8;
const size_t MAX_SLAVES_PER_SPLITPOINT = 4;
+class Spinlock {
+ std::atomic_int _lock;
+
+public:
+ Spinlock() { _lock = 1; } // Init here to workaround a bug with MSVC 2013
+ void lock() {
+ while (_lock.fetch_sub(1, std::memory_order_acquire) != 1)
+ for (int cnt = 0; _lock.load(std::memory_order_relaxed) <= 0; ++cnt)
+ if (cnt >= 10000) std::this_thread::yield(); // Be nice to hyperthreading
+ }
+ void unlock() { _lock.store(1, std::memory_order_release); }
+};
+
/// SplitPoint struct stores information shared by the threads searching in
/// parallel below the same split point. It is populated at splitting time.
/// SplitPoint struct stores information shared by the threads searching in
/// parallel below the same split point. It is populated at splitting time.
SplitPoint* parentSplitPoint;
// Shared variable data
SplitPoint* parentSplitPoint;
// Shared variable data
std::bitset<MAX_THREADS> slavesMask;
volatile bool allSlavesSearching;
volatile uint64_t nodes;
std::bitset<MAX_THREADS> slavesMask;
volatile bool allSlavesSearching;
volatile uint64_t nodes;
SplitPoint* volatile activeSplitPoint;
volatile size_t splitPointsSize;
volatile bool searching;
SplitPoint* volatile activeSplitPoint;
volatile size_t splitPointsSize;
volatile bool searching;