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

Daily update on 20250921-173333 by Nitload Youmu. Additional Message : New design of matrix enumeration, capable for parallel task publication.

Browse Source 
Signed-off-by: Nitload <YoumuLiu_NSI@whv0t.onmicrosoft.com>
This commit is contained in:
Nitload
2025-09-21 17:33:33 +08:00
parent 0404ecf044
commit b8a51c13dd
15 changed files with 463 additions and 304 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Common/HighPerformance/Memory/SlabAllocator.cs
+20 -21
View File
@@ -59,27 +59,26 @@ namespace StgSharp.HighPerformance.Memory
/// Thrown when requesting unsupported configurations.
/// </exception>
/// <remarks>
/// <para><strong> Important Threading Considerations: </strong></para> <para>While SLAB
/// /// allocators themselves may be designed for single-threaded use, allocated objects
/// are /// frequently shared across multiple threads in real-world applications. Due
/// to the /// special implementation of expandable concurrent sequential SLAB
/// allocators, they rely /// on internal expansion locks to ensure that memory buffer
/// expansions (which involve /// copying the entire buffer to a new location) do not
/// occur while other threads are /// reading from allocated regions. </para>
/// <para><strong>Sequential Layout Limitations: ///</strong><br /> Sequential layout SLAB
/// allocators cannot provide a single-threaded /// version because:<list
/// type="bullet"><item><description>They use contiguous memory /// layout that requires
/// buffer expansion when capacity is exceeded
/// ///</description></item><item><description>Buffer expansion involves copying all ///
/// existing data to a new memory location</description></item><item><description> ///
/// Without proper synchronization, concurrent access to allocated objects during ///
/// expansion would cause memory corruption</description></item><item><description>The ///
/// expansion lock mechanism is essential even if allocation itself is single-threaded
/// ///</description></item></list></para> <para><strong> Chunked Layout: </strong><br />
/// /// Chunked layout allocators can optionally provide single-threaded versions since
/// they /// don't require buffer expansion and copying. Each chunk is allocated
/// independently, /// avoiding the memory relocation issues present in sequential
/// layouts. </para>
/// <para><strong> Important Threading Considerations: </strong></para> <para> While SLAB
/// allocators themselves may be designed for single-threaded use, allocated objects are
/// frequently shared across multiple threads in real-world applications. Due to the
/// special implementation of expandable concurrent sequential SLAB allocators, they rely
/// on internal expansion locks to ensure that memory buffer expansions (which involve
/// copying the entire buffer to a new location) do not occur while other threads are
/// reading from allocated regions.</para> <para><strong> Sequential Layout Limitations:
/// </strong><br /> Sequential layout SLAB allocators cannot provide a single-threaded
/// version because:<list type="bullet"><item><description> They use contiguous memory
/// layout that requires buffer expansion when capacity is exceeded
/// </description></item><item><description> Buffer expansion involves copying all
/// existing data to a new memory location </description></item><item><description>
/// Without proper synchronization, concurrent access to allocated objects during
/// expansion would cause memory corruption </description></item><item><description> The
/// expansion lock mechanism is essential even if allocation itself is single-
/// threaded</description></item></list></para> <para><strong> Chunked Layout:
/// </strong><br /> Chunked layout allocators can optionally provide single-threaded
/// versions since they don't require buffer expansion and copying. Each chunk is
/// allocated independently, avoiding the memory relocation issues present in sequential
/// layouts.</para>
/// </remarks>
public static SlabAllocator<T> Create(nuint count, SlabBufferLayout layout, bool concurrentSupport = true)
{