public void Allocate1D <T>(T _)
where T : unmanaged
{
using var buffer = new MemoryBufferCache(Accelerator);
// Allocate several elements one after another
for (int i = 1; i <= 1024; i <<= 1)
{
var view = buffer.Allocate <T>(i);
Assert.Equal(view.Length, i);
Assert.Equal(view.LengthInBytes, Interop.SizeOf <T>() * i);
Assert.True(buffer.CacheSizeInBytes >= Interop.SizeOf <T>() * i);
Assert.True(buffer.GetCacheSize <T>() == i);
}
// "Allocate" the same elements in reverse order
for (int i = 1024; i >= 1; i >>= 1)
{
var view = buffer.Allocate <T>(i);
Assert.Equal(view.Length, i);
Assert.Equal(view.LengthInBytes, Interop.SizeOf <T>() * i);
Assert.True(buffer.CacheSizeInBytes >= Interop.SizeOf <T>() * i);
Assert.True(buffer.GetCacheSize <T>() >= i);
}
}
/// <summary>
/// Constructs a new CPU-based runtime context for parallel processing.
/// </summary>
/// <param name="accelerator">The target CPU accelerator.</param>
public CPURuntimeGroupContext(CPUAccelerator accelerator)
{
Debug.Assert(accelerator != null, "Invalid accelerator");
Accelerator = accelerator;
groupBarrier = new Barrier(0);
sharedMemoryBuffer = new MemoryBufferCache(accelerator);
broadcastBuffer = accelerator.Allocate <byte>(BroadcastBufferSize);
}
internal RadixSortPairsProvider(
Accelerator accelerator,
RadixSortPairsProviderImplementation implementation)
: base(accelerator)
{
bufferCache = new MemoryBufferCache(Accelerator);
this.implementation = implementation;
}
/// <summary>
/// Constructs a new CPU-based runtime context for parallel processing.
/// </summary>
/// <param name="multiprocessor">The target CPU multiprocessor.</param>
protected CPURuntimeContext(CPUMultiprocessor multiprocessor)
{
Multiprocessor = multiprocessor
?? throw new ArgumentNullException(nameof(multiprocessor));
broadcastBuffer = new MemoryBufferCache(multiprocessor.Accelerator);
broadcastBuffer.Allocate <int>(16);
}
/// <summary>
/// Constructs a new CPU-based runtime context for parallel processing.
/// </summary>
/// <param name="accelerator">The target CPU accelerator.</param>
/// <param name="numThreadsPerWarp">The number of threads per warp.</param>
public CPURuntimeWarpContext(CPUAccelerator accelerator, int numThreadsPerWarp)
: base(accelerator)
{
WarpSize = numThreadsPerWarp;
shuffleBuffer = new MemoryBufferCache(accelerator);
shuffleBuffer.Allocate <int>(2 * sizeof(int) * numThreadsPerWarp);
}
/// <summary>
/// Constructs a new CPU-based runtime context for parallel processing.
/// </summary>
/// <param name="accelerator">The target CPU accelerator.</param>
protected CPURuntimeContext(CPUAccelerator accelerator)
{
Accelerator = accelerator
?? throw new ArgumentNullException(nameof(accelerator));
barrier = new Barrier(0);
broadcastBuffer = new MemoryBufferCache(accelerator);
broadcastBuffer.Allocate <int>(16);
}
/// <summary>
/// Constructs a new allocation operation.
/// </summary>
public GetShuffleMemory(MemoryBufferCache shuffleBuffer, int warpSize)
{
ShuffleBuffer = shuffleBuffer;
WarpSize = warpSize;
}
internal ScanProvider(Accelerator accelerator)
: base(accelerator)
{
bufferCache = new MemoryBufferCache(Accelerator);
}
internal RadixSortProvider(Accelerator accelerator)
: base(accelerator)
{
bufferCache = new MemoryBufferCache(accelerator);
}
/// <summary>
/// Constructs a new CPU-based runtime context for parallel processing.
/// </summary>
/// <param name="accelerator">The target CPU accelerator.</param>
public CPURuntimeGroupContext(CPUAccelerator accelerator)
{
Debug.Assert(accelerator != null, "Invalid accelerator");
groupBarrier = new Barrier(0);
sharedMemoryBuffer = new MemoryBufferCache(accelerator);
}
internal ScanProvider(Accelerator accelerator, ScanProviderImplementation implementation)
: base(accelerator)
{
bufferCache = new MemoryBufferCache(Accelerator);
this.implementation = implementation;
}