private void InvokeLauncher <T>(
T dimension,
AcceleratorStream stream,
object[] args,
int[] dynSharedMemArraySizes)
where T : IIndex
{
var numSharedMemVars = dynSharedMemArraySizes == null ? 0 : dynSharedMemArraySizes.Length;
var entryPoint = CompiledKernel.EntryPoint;
if (entryPoint.NumDynamicallySizedSharedMemoryVariables != numSharedMemVars)
{
throw new ArgumentException(RuntimeErrorMessages.InvalidNumberOfDynamicallySharedMemoryVariableArgs);
}
if (entryPoint.NumUniformVariables != args.Length)
{
throw new ArgumentException(RuntimeErrorMessages.InvalidNumberOfUniformArgs);
}
var reflectionArgs = new object[KernelParameterOffset + args.Length + numSharedMemVars];
reflectionArgs[KernelInstanceParamIdx] = this;
reflectionArgs[KernelStreamParamIdx] = stream ?? throw new ArgumentNullException(nameof(stream));
reflectionArgs[KernelParamDimensionIdx] = dimension;
args.CopyTo(reflectionArgs, KernelParameterOffset);
if (numSharedMemVars > 0)
{
dynSharedMemArraySizes.CopyTo(reflectionArgs, KernelParameterOffset + args.Length);
}
Launcher.Invoke(null, reflectionArgs);
}
/// <summary>
/// Copies the contents to this buffer to the given array.
/// </summary>
/// <param name="stream">The used accelerator stream.</param>
/// <param name="target">The target array.</param>
/// <param name="sourceOffset">The source offset.</param>
/// <param name="targetOffset">The target offset.</param>
/// <param name="extent">The length.</param>
/// <remarks>
/// Note that the output array will contain the data as a transposed array to
/// match the source layout.
/// </remarks>
public void CopyTo(
AcceleratorStream stream,
T[,,] target,
LongIndex3 sourceOffset,
LongIndex3 targetOffset,
LongIndex3 extent)
{
if (target == null)
{
throw new ArgumentNullException(nameof(target));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 || sourceOffset.Z < 0 ||
sourceOffset.X >= Extent.X ||
sourceOffset.Y >= Extent.Y ||
sourceOffset.Z >= Extent.Z)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
if (targetOffset.X < 0 || targetOffset.Y < 0 || targetOffset.Z < 0 ||
targetOffset.X >= target.GetLongLength(0) ||
targetOffset.Y >= target.GetLongLength(1) ||
targetOffset.Z >= target.GetLongLength(2))
{
throw new ArgumentOutOfRangeException(nameof(targetOffset));
}
if (extent.X < 0 || extent.Y < 0 || extent.Z < 0 ||
sourceOffset.X + extent.X > Extent.X ||
sourceOffset.Y + extent.Y > Extent.Y ||
sourceOffset.Z + extent.Z > Extent.Z ||
targetOffset.X + extent.X > target.GetLongLength(0) ||
targetOffset.Y + extent.Y > target.GetLongLength(1) ||
targetOffset.Z + extent.Z > target.GetLongLength(2))
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var tempBuffer = new T[extent.Size];
buffer.CopyTo(stream, tempBuffer, sourceOffset, 0, extent);
for (long i = 0; i < extent.X; ++i)
{
for (long j = 0; j < extent.Y; ++j)
{
for (long k = 0; k < extent.Z; ++k)
{
var sourceIdx = new LongIndex3(i, j, k).
ComputeLinearIndex(extent);
target[
i + targetOffset.X,
j + targetOffset.Y,
k + targetOffset.Z] = tempBuffer[sourceIdx];
}
}
}
}
/// <summary>
/// Copies data from the associated accelerator into CPU memory.
/// </summary>
/// <param name="stream">The stream to use.</param>
public void CopyFromAccelerator(AcceleratorStream stream)
{
if (stream == null)
{
throw new ArgumentNullException(nameof(stream));
}
CopyToView(stream, CPUView.BaseView, 0);
}
/// <summary>
/// Copies data from CPU memory to the associated accelerator.
/// </summary>
/// <param name="stream">The stream to use.</param>
/// <param name="acceleratorMemoryOffset">The target memory offset.</param>
public void CopyToAccelerator(
AcceleratorStream stream,
TIndex acceleratorMemoryOffset) =>
CopyToAccelerator(
stream,
default,
acceleratorMemoryOffset,
CPUView.Length);
/// <summary>
/// Copies data from CPU memory to the associated accelerator.
/// </summary>
/// <param name="stream">The stream to use.</param>
public void CopyToAccelerator(AcceleratorStream stream)
{
if (stream == null)
{
throw new ArgumentNullException(nameof(stream));
}
CopyFromView(stream, CPUView.AsLinearView(), 0);
}
/// <summary>
/// Copies data from the associated accelerator into CPU memory.
/// </summary>
/// <param name="stream">The stream to use.</param>
/// <param name="cpuMemoryOffset">the target memory offset.</param>
public void CopyFromAccelerator(
AcceleratorStream stream,
TIndex cpuMemoryOffset) =>
CopyFromAccelerator(
stream,
default,
cpuMemoryOffset,
Length);
public void CopyFrom(
AcceleratorStream stream,
T[][][] source,
Index3 sourceOffset,
Index3 targetOffset,
Index3 extent)
{
if (extent.X < 0 || extent.Y < 0 || extent.Z < 0 ||
extent.X > source.Length)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 || sourceOffset.Z < 0 ||
sourceOffset.X >= extent.X ||
sourceOffset.Y >= extent.Y ||
sourceOffset.Z >= extent.Z)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
var tempBuffer = new T[extent.Size];
for (int i = 0; i < extent.X; ++i)
{
var subData = source[i + sourceOffset.X];
if (subData == null)
{
continue;
}
for (int j = 0; j < extent.Y; ++j)
{
var subSubData = subData[j + sourceOffset.Y];
if (subSubData == null)
{
continue;
}
// Skip entries that are out of bounds
for (
int k = 0, e = IntrinsicMath.Min(subSubData.Length, extent.Z);
k < e;
++k)
{
var targetIdx = new Index3(i, j, k).ComputeLinearIndex(extent);
tempBuffer[targetIdx] = subSubData[k + sourceOffset.Z];
}
}
}
buffer.CopyFrom(
stream,
tempBuffer,
0,
targetOffset,
extent.Size);
}
/// <summary>
/// Copies data from CPU memory to the associated accelerator.
/// </summary>
/// <param name="stream">The stream to use.</param>
/// <param name="cpuMemoryOffset">the source memory offset.</param>
/// <param name="acceleratorMemoryOffset">The target memory offset.</param>
public void CopyToAccelerator(
AcceleratorStream stream,
TIndex cpuMemoryOffset,
TIndex acceleratorMemoryOffset) =>
CopyToAccelerator(
stream,
cpuMemoryOffset,
acceleratorMemoryOffset,
CPUArrayView.Length);
public void Launch <TIndex>(
TIndex dimension,
AcceleratorStream stream,
int[] dynSharedMemArraySizes,
params object[] args)
where TIndex : struct, IGroupedIndex
{
InvokeLauncher(dimension, stream, args, dynSharedMemArraySizes);
}
public unsafe void CopyFrom <T>(AcceleratorStream stream, T source, Index sourceIndex)
where T : struct
{
using (var wrapper = ViewPointerWrapper.Create(ref source))
{
var view = new ArrayView <T>(wrapper, 0, 1);
cache.CopyFromView(stream, view.Cast <byte>(), sourceIndex);
}
}
public void CopyTo(
AcceleratorStream stream,
T[][][] target,
Index3 sourceOffset,
Index3 targetOffset,
Index3 extent)
{
if (target == null)
{
throw new ArgumentNullException(nameof(target));
}
if (extent.X < 0 || extent.Y < 0 || extent.Z < 0 ||
extent.X > target.Length)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 || sourceOffset.Z < 0 ||
sourceOffset.X >= Extent.X || sourceOffset.Y >= Extent.Y ||
sourceOffset.Z >= Extent.Z)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
var tempBuffer = new T[extent.Size];
buffer.CopyTo(
stream,
tempBuffer,
sourceOffset,
0,
extent);
for (int i = 0; i < extent.X; ++i)
{
var subData = target[i + targetOffset.X];
if (subData == null)
{
continue;
}
for (int j = 0; j < extent.Y; ++j)
{
var subSubData = subData[j + targetOffset.Y];
if (subSubData == null)
{
continue;
}
for (int k = 0; k < extent.Z; ++k)
{
var sourceIdx = new Index3(i, j, k).ComputeLinearIndex(extent);
subSubData[k + targetOffset.Z] = tempBuffer[sourceIdx];
}
}
}
}
/// <summary>
/// Sets the contents of the current buffer to the given byte value.
/// </summary>
/// <param name="stream">The used accelerator stream.</param>
/// <param name="value">The value to write into the memory buffer.</param>
/// <param name="offsetInBytes">The raw offset in bytes.</param>
/// <param name="lengthInBytes">The raw length in bytes.</param>
protected internal override void MemSetInternal(
AcceleratorStream stream,
byte value,
long offsetInBytes,
long lengthInBytes) =>
Buffer.MemSetInternal(
stream,
value,
offsetInBytes,
lengthInBytes);
public unsafe void CopyTo <T>(AcceleratorStream stream, out T target, Index targetIndex)
where T : struct
{
target = default;
using (var wrapper = ViewPointerWrapper.Create(ref target))
{
var view = new ArrayView <T>(wrapper, 0, 1);
cache.CopyToView(stream, view.Cast <byte>(), targetIndex);
}
}
/// <summary>
/// Copies the contents to this buffer from the given array.
/// </summary>
/// <param name="stream">The used accelerator stream.</param>
/// <param name="source">The source array.</param>
/// <param name="sourceOffset">The source offset.</param>
/// <param name="targetOffset">The target offset.</param>
/// <param name="extent">The length.</param>
/// <remarks>
/// Note that the input array will stored as a transposed array to match the
/// target layout.
/// </remarks>
public void CopyFrom(
AcceleratorStream stream,
T[,] source,
LongIndex2 sourceOffset,
LongIndex2 targetOffset,
LongIndex2 extent)
{
if (source == null)
{
throw new ArgumentNullException(nameof(source));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 ||
sourceOffset.X >= source.GetLongLength(0) ||
sourceOffset.Y >= source.GetLongLength(1))
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
if (targetOffset.X < 0 || targetOffset.Y < 0 ||
targetOffset.X >= Extent.X ||
targetOffset.Y >= Extent.Y)
{
throw new ArgumentOutOfRangeException(nameof(targetOffset));
}
if (extent.X < 0 || extent.Y < 0 ||
sourceOffset.X + extent.X > source.GetLongLength(0) ||
sourceOffset.Y + extent.Y > source.GetLongLength(1) ||
targetOffset.X + extent.X > Extent.X ||
targetOffset.Y + extent.Y > Extent.Y)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var tempBuffer = new T[extent.Size];
for (long i = 0; i < extent.X; ++i)
{
for (long j = 0; j < extent.Y; ++j)
{
var targetIdx = new LongIndex2(i, j).
ComputeLinearIndex(extent);
tempBuffer[targetIdx] =
source[i + sourceOffset.X, j + sourceOffset.Y];
}
}
buffer.CopyFrom(
stream,
tempBuffer,
0,
targetOffset,
extent.Size);
}
public TDelegate CreateStreamLauncherDelegate <TDelegate>(AcceleratorStream stream)
where TDelegate : class
{
if (stream == null)
{
throw new ArgumentNullException(nameof(stream));
}
Debug.Assert(implicitStreamLauncher != null, "Invalid implicit stream launcher");
return((implicitStreamLauncher.CreateDelegate(typeof(TDelegate),
new ImplicitKernelLauncherArgument(this, stream)) as object) as TDelegate);
}
public void CopyTo(
AcceleratorStream stream,
T[][] target,
LongIndex2 sourceOffset,
LongIndex2 targetOffset,
LongIndex2 extent)
{
if (target == null)
{
throw new ArgumentNullException(nameof(target));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 ||
sourceOffset.X >= Extent.X ||
sourceOffset.Y >= Extent.Y)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
if (targetOffset.X < 0 || targetOffset.Y < 0 ||
targetOffset.X >= target.LongLength)
{
throw new ArgumentOutOfRangeException(nameof(targetOffset));
}
if (extent.X < 0 || extent.Y < 0 ||
sourceOffset.X + extent.X > Extent.X ||
sourceOffset.Y + extent.Y > Extent.Y ||
targetOffset.X + extent.X > target.LongLength)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var tempBuffer = new T[extent.Size];
buffer.CopyTo(stream, tempBuffer, sourceOffset, 0, extent);
for (long i = 0; i < extent.X; ++i)
{
var subData = target[i + targetOffset.X];
if (subData == null)
{
continue;
}
for (long j = 0; j < extent.Y; ++j)
{
var sourceIdx = new LongIndex2(i, j).
ComputeLinearIndex(extent);
subData[j + targetOffset.Y] = tempBuffer[sourceIdx];
}
}
}
/// <summary>
/// Loads the given kernel and returns a launcher delegate that is associated
/// with the given accelerator stream. Consequently, the resulting delegate
/// cannot receive other accelerator streams.
/// </summary>
/// <param name="method">The method to compile into a kernel.</param>
/// <param name="stream">The accelerator stream to use.</param>
/// <returns>The loaded kernel-launcher delegate.</returns>
public TDelegate LoadAutoGroupedStreamKernel <TDelegate>(MethodInfo method, AcceleratorStream stream)
where TDelegate : class
{
var loader = AutoKernelLoader.Default;
var launcher = new StreamLauncherProvider(stream);
return(LoadGenericKernel <TDelegate, AutoKernelLoader, StreamLauncherProvider>(
method,
KernelSpecialization.Empty,
ref loader,
ref launcher));
}
/// <summary>
/// Loads the given kernel and returns a launcher delegate that is associated
/// with the given accelerator stream. Consequently, the resulting delegate
/// cannot receive other accelerator streams.
/// </summary>
/// <typeparam name="TDelegate">The delegate type.</typeparam>
/// <param name="method">The method to compile into a kernel.</param>
/// <param name="stream">The accelerator stream to use.</param>
/// <param name="specialization">The kernel specialization.</param>
/// <returns>The loaded kernel-launcher delegate.</returns>
/// <remarks>
/// Note that implictly-grouped kernels will be launched with a group size
/// of the current warp size of the accelerator.
/// </remarks>
public TDelegate LoadStreamKernel <TDelegate>(MethodInfo method, AcceleratorStream stream, KernelSpecialization specialization)
where TDelegate : class
{
var loader = DefaultKernelLoader.Default;
var launcher = new StreamLauncherProvider(stream);
return(LoadGenericKernel <TDelegate, DefaultKernelLoader, StreamLauncherProvider>(
method,
specialization,
ref loader,
ref launcher));
}
public T[,,] GetAs3DArray(AcceleratorStream stream, Index3 offset, Index3 extent)
{
if (extent.X < 1 || extent.Y < 1 || extent.Z < 1)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var result = new T[extent.X, extent.Y, extent.Z];
CopyTo(stream, result, offset, Index3.Zero, extent);
return(result);
}
public void MemSet(
AcceleratorStream stream,
byte value,
long targetOffsetInBytes,
long length)
{
if (length == 0)
{
return;
}
var targetView = AsRawArrayView(targetOffsetInBytes, length);
MemSet(stream, value, targetView);
}
public void CopyTo(
AcceleratorStream stream,
T[,] target,
Index2 sourceOffset,
Index2 targetOffset,
Index2 extent)
{
if (target == null)
{
throw new ArgumentNullException(nameof(target));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 ||
sourceOffset.X >= Extent.X ||
sourceOffset.Y >= Extent.Y)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
if (targetOffset.X < 0 || targetOffset.Y < 0 ||
targetOffset.X >= target.GetLength(0) ||
targetOffset.Y >= target.GetLength(1))
{
throw new ArgumentOutOfRangeException(nameof(targetOffset));
}
if (extent.X < 0 || extent.Y < 0 ||
sourceOffset.X + extent.X > Extent.X ||
sourceOffset.Y + extent.Y > Extent.Y ||
targetOffset.X + extent.X > target.GetLength(0) ||
targetOffset.Y + extent.Y > target.GetLength(1))
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var tempBuffer = new T[extent.Size];
buffer.CopyTo(stream, tempBuffer, sourceOffset, 0, extent);
for (int i = 0; i < extent.X; ++i)
{
for (int j = 0; j < extent.Y; ++j)
{
var sourceIdx = new Index2(i, j).ComputeLinearIndex(extent);
target[i + targetOffset.X, j + targetOffset.Y] =
tempBuffer[sourceIdx];
}
}
}
/// <summary>
/// Copies data from the associated accelerator into CPU memory.
/// </summary>
/// <param name="stream">The stream to use.</param>
/// <param name="offset">The element offset.</param>
/// <param name="length">The length (number of elements).</param>
public void CopyFromAcceleratorAsync(
AcceleratorStream stream,
long offset,
long length)
{
if (stream == null)
{
throw new ArgumentNullException(nameof(stream));
}
var sourceView = GPUView.SubView(offset, length);
var targetView = CPUView.SubView(offset, length);
sourceView.CopyTo(stream, targetView);
}
/// <summary>
/// Loads the given kernel and returns a launcher delegate that is associated
/// with the given accelerator stream. Consequently, the resulting delegate
/// cannot receive other accelerator streams.
/// </summary>
/// <param name="method">The method to compile into a kernel.</param>
/// <param name="customGroupSize">The custom group size to use.</param>
/// <param name="stream">The accelerator stream to use.</param>
/// <returns>The loaded kernel-launcher delegate.</returns>
/// <remarks>
/// Note that implictly-grouped kernel will be launched with the given
/// group size.
/// </remarks>
public TDelegate LoadImplicitlyGroupedStreamKernel <TDelegate>(
MethodInfo method,
int customGroupSize,
AcceleratorStream stream)
where TDelegate : class
{
var loader = new GroupedKernelLoader(customGroupSize);
var launcher = new StreamLauncherProvider(stream);
return(LoadGenericKernel <TDelegate, GroupedKernelLoader, StreamLauncherProvider>(
method,
new KernelSpecialization(customGroupSize, null),
ref loader,
ref launcher));
}
/// <summary>
/// Copies the current contents into a new 2D array.
/// </summary>
/// <param name="stream">The used accelerator stream.</param>
/// <param name="offset">The offset.</param>
/// <param name="extent">The extent (number of elements).</param>
/// <returns>A new array holding the requested contents.</returns>
public T[,] GetAs2DArray(
AcceleratorStream stream,
LongIndex2 offset,
LongIndex2 extent)
{
if (extent.X < 1 || extent.Y < 1)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var result = new T[extent.X, extent.Y];
CopyTo(stream, result, offset, LongIndex2.Zero, extent);
return(result);
}
private void InvokeLauncher <T>(T dimension, AcceleratorStream stream, object[] args)
where T : IIndex
{
if (NumParameters != args.Length)
{
throw new ArgumentException(RuntimeErrorMessages.InvalidNumberOfUniformArgs);
}
var reflectionArgs = new object[KernelParameterOffset + args.Length];
reflectionArgs[KernelInstanceParamIdx] = this;
reflectionArgs[KernelStreamParamIdx] = stream ?? throw new ArgumentNullException(nameof(stream));
reflectionArgs[KernelParamDimensionIdx] = dimension;
args.CopyTo(reflectionArgs, KernelParameterOffset);
Launcher.Invoke(null, reflectionArgs);
}
public void CopyFrom(
AcceleratorStream stream,
T[,,] source,
Index3 sourceOffset,
Index3 targetOffset,
Index3 extent)
{
if (extent.X < 0 || extent.Y < 0 || extent.Z < 0 ||
extent.X > source.GetLength(0) ||
extent.Y > source.GetLength(1) ||
extent.Z > source.GetLength(2))
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
if (sourceOffset.X < 0 || sourceOffset.Y < 0 || sourceOffset.Z < 0 ||
sourceOffset.X >= extent.X ||
sourceOffset.Y >= extent.Y ||
sourceOffset.Z >= extent.Z)
{
throw new ArgumentOutOfRangeException(nameof(sourceOffset));
}
var tempBuffer = new T[extent.Size];
for (int i = 0; i < extent.X; ++i)
{
for (int j = 0; j < extent.Y; ++j)
{
for (int k = 0; k < extent.Z; ++k)
{
var targetIdx = new Index3(i, j, k).ComputeLinearIndex(extent);
tempBuffer[targetIdx] = source[
i + sourceOffset.X,
j + sourceOffset.Y,
k + sourceOffset.Z];
}
}
}
buffer.CopyFrom(
stream,
tempBuffer,
0,
targetOffset,
extent.Size);
}
/// <summary>
/// Loads the given kernel and returns a launcher delegate that is associated
/// with the given accelerator stream. Consequently, the resulting delegate
/// cannot receive other accelerator streams.
/// </summary>
/// <param name="method">The method to compile into a kernel.</param>
/// <param name="stream">The accelerator stream to use.</param>
/// <param name="groupSize">The estimated group size to gain maximum occupancy on this device.</param>
/// <param name="minGridSize">The minimum grid size to gain maximum occupancy on this device.</param>
/// <returns>The loaded kernel-launcher delegate.</returns>
public TDelegate LoadAutoGroupedStreamKernel <TDelegate>(
MethodInfo method,
AcceleratorStream stream,
out int groupSize,
out int minGridSize)
where TDelegate : class
{
var loader = AutoKernelLoader.Default;
var launcher = new StreamLauncherProvider(stream);
var result = LoadGenericKernel <TDelegate, AutoKernelLoader, StreamLauncherProvider>(
method,
KernelSpecialization.Empty,
ref loader,
ref launcher);
groupSize = loader.GroupSize;
minGridSize = loader.MinGridSize;
return(result);
}
/// <summary>
/// Sets the contents of the current buffer to the given byte value.
/// </summary>
/// <param name="stream">The used accelerator stream.</param>
/// <param name="value">The value to write into the memory buffer.</param>
/// <param name="offsetInBytes">The raw offset in bytes.</param>
/// <param name="lengthInBytes">The raw length in bytes.</param>
public void MemSet(
AcceleratorStream stream,
byte value,
long offsetInBytes,
long lengthInBytes)
{
if (lengthInBytes < 0)
{
throw new ArgumentOutOfRangeException(nameof(lengthInBytes));
}
if (offsetInBytes < 0 || offsetInBytes + lengthInBytes > LengthInBytes)
{
throw new ArgumentOutOfRangeException(nameof(offsetInBytes));
}
MemSetInternal(
stream,
value,
offsetInBytes,
lengthInBytes);
}
/// <summary>
/// Copies data from the associated accelerator into CPU memory.
/// </summary>
/// <param name="stream">The stream to use.</param>
/// <param name="acceleratorMemoryOffset">The source memory offset.</param>
/// <param name="cpuMemoryOffset">the target memory offset.</param>
/// <param name="extent">The extent (number of elements).</param>
public void CopyFromAccelerator(
AcceleratorStream stream,
TIndex acceleratorMemoryOffset,
TIndex cpuMemoryOffset,
LongIndex1 extent)
{
if (stream == null)
{
throw new ArgumentNullException(nameof(stream));
}
if (!cpuMemoryOffset.InBounds(CPUView.Extent))
{
throw new ArgumentOutOfRangeException(nameof(cpuMemoryOffset));
}
if (!acceleratorMemoryOffset.InBounds(Extent))
{
throw new ArgumentOutOfRangeException(nameof(acceleratorMemoryOffset));
}
if (extent < 1 || extent > Length)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
var linearSourceIndex =
acceleratorMemoryOffset.ComputeLinearIndex(Extent);
var linearTargetIndex =
cpuMemoryOffset.ComputeLinearIndex(CPUView.Extent);
if (linearSourceIndex + extent > Length ||
linearTargetIndex + extent > CPUView.Length)
{
throw new ArgumentOutOfRangeException(nameof(extent));
}
CopyToView(
stream,
CPUView.GetSubView(cpuMemoryOffset, extent),
linearSourceIndex);
}
/// <summary>
/// Loads the given kernel and returns a launcher delegate that is associated
/// with the given accelerator stream. Consequently, the resulting delegate
/// cannot receive other accelerator streams.
/// </summary>
/// <typeparam name="TDelegate">The delegate type.</typeparam>
/// <param name="method">The method to compile into a kernel.</param>
/// <param name="stream">The accelerator stream to use.</param>
/// <returns>The loaded kernel-launcher delegate.</returns>
/// <remarks>
/// Note that implictly-grouped kernels will be launched with a group size
/// of the current warp size of the accelerator.
/// </remarks>
public TDelegate LoadStreamKernel <TDelegate>(MethodInfo method, AcceleratorStream stream)
where TDelegate : class
{
return(LoadStreamKernel <TDelegate>(method, stream, KernelSpecialization.Empty));
}
