youmuliu/stgsharp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

optimize para performance of buffered stack

Browse Source
This commit is contained in:
Youmu Liu
2025-11-05 20:33:58 +08:00
parent 9c4c71dc5b
commit ea0e6f3ff6
10 changed files with 516 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Common/HighPerformance/Memory/ConcurrentBufferStack.cs
+71 -53
View File
@@ -65,7 +65,7 @@ namespace StgSharp.HighPerformance.Memory
return new ConcurrentBufferStack<T>(4); // Default small capacity
}
ConcurrentBufferStack<T> stack = new ConcurrentBufferStack<T>(span.Length);
ConcurrentBufferStack<T> stack = new(span.Length);
// Use internal method for efficient bulk initialization
@@ -106,12 +106,20 @@ namespace StgSharp.HighPerformance.Memory
public void Clear()
{
Header* header = (Header*)_buffer;
_lock.EnterExpansionProcess();
try
{
header->Top = 0;
Volatile.Write(ref _count, 0);
_lock.EnterBufferCopy(); // exclusive while mutating Top
try
{
Header* header = (Header*)_buffer;
header->Top = 0;
Volatile.Write(ref _count, 0);
}
finally
{
_lock.ExitBufferCopy();
}
}
finally
{
@@ -127,29 +135,29 @@ namespace StgSharp.HighPerformance.Memory
public T Pop()
{
Header* header = (Header*)_buffer;
while (true)
{
nuint oldTop = Volatile.Read(ref header->Top);
if (oldTop <= 0) {
throw new InvalidOperationException("Stack underflow");
}
if (Interlocked.CompareExchange(ref header->Top, oldTop - 1, oldTop) == oldTop)
_lock.EnterBufferRead(); // prevent buffer being reallocated while reading/modifying Top
try
{
_lock.EnterBufferRead();
try
Header* header = (Header*)_buffer;
nuint oldTop = Volatile.Read(ref header->Top);
if (oldTop <= 0) {
throw new InvalidOperationException("Stack underflow");
}
if (Interlocked.CompareExchange(ref header->Top, oldTop - 1, oldTop) == oldTop)
{
T* dataStart = (T*)(_buffer + sizeof(Header));
_ = Interlocked.Decrement(ref _count);
return dataStart[oldTop - 1];
}
finally
{
_lock.ExitBufferRead();
}
}
Thread.Sleep(0); // Yield to avoid busy waiting
finally
{
_lock.ExitBufferRead();
}
Thread.Sleep(0);
}
}
@@ -157,14 +165,16 @@ namespace StgSharp.HighPerformance.Memory
{
while (true)
{
// Ensure capacity, expanding if needed
_lock.EnterMetaDataRead();
try
{
Header* header = (Header*)_buffer;
nuint oldTop = Volatile.Read(ref header->Top);
nuint topSnapshot = Volatile.Read(ref header->Top);
nuint cap = Volatile.Read(ref header->Capacity);
if (oldTop >= cap)
if (topSnapshot >= cap)
{
// Need expansion
_lock.ExitMetaDataRead();
_lock.EnterExpansionProcess();
try
@@ -193,22 +203,6 @@ namespace StgSharp.HighPerformance.Memory
}
continue;
}
if (Interlocked.CompareExchange(ref header->Top, oldTop + 1, oldTop) == oldTop)
{
_lock.EnterBufferRead();
try
{
T* dataStart = (T*)(_buffer + sizeof(Header));
dataStart[oldTop] = item;
_ = Interlocked.Increment(ref _count);
return;
}
finally
{
_lock.ExitBufferRead();
}
}
}
finally
{
@@ -216,39 +210,63 @@ namespace StgSharp.HighPerformance.Memory
_lock.ExitMetaDataRead();
}
}
Thread.Sleep(0); // Yield to avoid busy waiting
// Reserve and write under exclusive buffer lock to avoid consumers reading unwritten data
_lock.EnterBufferCopy();
try
{
Header* header = (Header*)_buffer;
nuint idx = header->Top; // reserve next slot
if (idx >= header->Capacity)
{
// lost race against expander or other producer; retry
continue;
}
T* dataStart = (T*)(_buffer + sizeof(Header));
dataStart[idx] = item; // write first
header->Top = idx + 1; // then publish by increasing top
_ = Interlocked.Increment(ref _count);
return;
}
finally
{
_lock.ExitBufferCopy();
}
}
}
public bool TryPop(out T value)
{
Header* header = (Header*)_buffer;
while (true)
{
nuint oldTop = Volatile.Read(ref header->Top);
if (oldTop <= 0)
_lock.EnterBufferRead();
try
{
value = default;
return false;
}
Header* header = (Header*)_buffer;
nuint oldTop = Volatile.Read(ref header->Top);
if (oldTop <= 0)
{
value = default;
return false;
}
if (Interlocked.CompareExchange(ref header->Top, oldTop - 1, oldTop) == oldTop)
{
_lock.EnterBufferRead();
try
if (Interlocked.CompareExchange(ref header->Top, oldTop - 1, oldTop) == oldTop)
{
T* dataStart = (T*)(_buffer + sizeof(Header));
value = dataStart[oldTop - 1];
_ = Interlocked.Decrement(ref _count);
return true;
}
finally
{
_lock.ExitBufferRead();
}
return true;
}
Thread.Sleep(0); // Yield to avoid busy waiting
finally
{
_lock.ExitBufferRead();
}
value = default;
return true;
}
Thread.Sleep(0); // Yield to avoid busy waiting
}
// Internal method for efficient bulk initialization