/// <summary>
/// Adds two matrices on the GPU
/// </summary>
/// <param name="one">The first matrix</param>
/// <param name="two">The second matrix</param>
/// <returns>The result of the addition</returns>
public BufferedFastMatrix <T> Add(BufferedFastMatrix <T> one, BufferedFastMatrix <T> two)
{
if (one == null || two == null)
{
throw new ArgumentNullException();
}
if ((one.Rows != two.Rows) || (one.Columns != two.Columns))
{
throw new BadDimensionException(one.Rows, one.Columns, two.Rows,
two.Columns);
}
MemoryBuffer2D <T> resultBuffer;
one.CopyToGPU();
two.CopyToGPU();
resultBuffer = accelerator.Allocate <T>(one.Rows, one.Columns);
one.WaitForCopy(); //this function call is currently not required,
//will come up with a better solution later but for now I'm just
//gonna leave it here
two.WaitForCopy();
GPUAddKernel(resultBuffer.Extent, one.buffer.View, two.buffer.View, resultBuffer.View);
accelerator.Synchronize();
var tempArray = resultBuffer.GetAs2DArray();
accelerator.Synchronize();
BufferedFastMatrix <T> returnMatrix = new BufferedFastMatrix <T>(tempArray);
return(returnMatrix);
}
/// <summary>
/// Transposes a matrix on the GPU
/// </summary>
/// <param name="matrix">The matrix</param>
/// <returns>The transposed matrix</returns>
public BufferedFastMatrix <T> Transpose(BufferedFastMatrix <T> matrix)
{
if (matrix == null)
{
throw new ArgumentNullException();
}
Accelerator accelerator;
MemoryBuffer2D <T> resultBuffer;
matrix.CopyToGPU();
accelerator = HardwareAcceleratorManager.GPUAccelerator;
resultBuffer = accelerator.Allocate <T>(matrix.Columns, matrix.Rows);
var kernel = GPUTransposeKernel;
matrix.WaitForCopy();
kernel(resultBuffer.Extent, matrix.buffer.View, resultBuffer.View);
accelerator.Synchronize();
var tempArray = resultBuffer.GetAs2DArray();
accelerator.Synchronize();
BufferedFastMatrix <T> returnMatrix = new BufferedFastMatrix <T>(tempArray);
return(returnMatrix);
}
/// <summary>
/// Subtracts two matrices on the GPU
/// </summary>
/// <param name="one">The first matrix</param>
/// <param name="two">The second matrix</param>
/// <returns>The result of the subtraction (one - two) </returns>
public BufferedFastMatrix <T> Subtract(BufferedFastMatrix <T> one, BufferedFastMatrix <T> two)
{
if (one == null || two == null)
{
throw new ArgumentNullException();
}
if ((one.Rows != two.Rows) || (one.Columns != two.Columns))
{
throw new BadDimensionException(one.Rows, one.Columns, two.Rows,
two.Columns);
}
MemoryBuffer2D <T> resultBuffer;
//start tasks
one.CopyToGPU();
two.CopyToGPU();
resultBuffer = accelerator.Allocate <T>(one.Rows, one.Columns);
one.WaitForCopy();
two.WaitForCopy();
GPUSubKernel(resultBuffer.Extent, one.buffer.View, two.buffer.View, resultBuffer.View);
accelerator.Synchronize();
var tempArray = resultBuffer.GetAs2DArray();
accelerator.Synchronize();
BufferedFastMatrix <T> returnMatrix = new BufferedFastMatrix <T>(tempArray);
return(returnMatrix);
}
public unsafe BufferedFastMatrix <T> AddShared(BufferedFastMatrix <T> one, BufferedFastMatrix <T> two)
{
if (one == null || two == null)
{
throw new ArgumentNullException();
}
if ((one.Rows != two.Rows) || (one.Columns != two.Columns))
{
throw new BadDimensionException(one.Rows, one.Columns, two.Rows,
two.Columns);
}
MemoryBuffer2D <T> resultBuffer;
one.CopyToGPU();
two.CopyToGPU();
resultBuffer = accelerator.Allocate <T>(one.Rows, one.Columns);
one.WaitForCopy();
two.WaitForCopy();
KernelConfig config = new KernelConfig(accelerator.MaxGridSize.X, accelerator.MaxNumThreadsPerGroup);
AddSharedKernel(config, one.buffer.View, two.buffer.View, resultBuffer);
Console.WriteLine(accelerator.MaxGridSize.X);
accelerator.Synchronize();
var tempArray = resultBuffer.GetAs2DArray();
accelerator.Synchronize();
BufferedFastMatrix <T> returnMatrix = new BufferedFastMatrix <T>(tempArray);
return(returnMatrix);
}
