internal Gpu4DTensor(CudaProvider provider, IReadOnlyList <I3DTensor> tensorList)
{
var first = tensorList.First();
_cuda = provider;
_rows = first.RowCount;
_columns = first.ColumnCount;
_depth = first.Depth;
_count = tensorList.Count;
provider.Register(this);
_data = _cuda.CreateZeroMatrix(_rows * _columns * _depth, _count);
_subVector = new Lazy <List <GpuVector[]> >(_GetSubVectors);
_tensorInfo = new Lazy <TensorInput>(_GetInput);
for (var i = 0; i < _count; i++)
{
_data.Column(i).AddInPlace(tensorList[i].ConvertToVector());
}
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public GpuMatrix(CudaProvider cuda, int rows, int columns, Func <int, int, float> init)
{
_cuda = cuda;
_rows = rows;
_columns = columns;
var count = rows * columns;
var data = new float[count];
for (var j = 0; j < columns; j++)
{
for (var i = 0; i < rows; i++)
{
data[j * rows + i] = init(i, j);
}
}
_data = cuda.Allocate(count);
_data.CopyToDevice(data);
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public Gpu4DTensor(CudaProvider provider, IReadOnlyList <FloatTensor> data)
{
var firstTensor = data.First();
var firstMatrix = firstTensor.Matrix.First();
_cuda = provider;
_rows = firstMatrix.RowCount;
_columns = firstMatrix.ColumnCount;
_depth = firstTensor.Matrix.Length;
_count = data.Count;
var matrixSize = _rows * _columns;
_data = provider.CreateMatrix(matrixSize * _depth, _count, (index, k) => {
var z = index / matrixSize;
var rem = index % matrixSize;
var i = rem / _rows;
var j = rem % _rows;
return(data[k].Matrix[z].Row[j].Data[i]);
});
provider.Register(this);
_subVector = new Lazy <List <GpuVector[]> >(_GetSubVectors);
_tensorInfo = new Lazy <TensorInput>(_GetInput);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public Gpu3DTensor(CudaProvider provider, int rows, int columns, int depth, IReadOnlyList <GpuMatrix> data)
{
_cuda = provider;
_rows = rows;
_columns = columns;
_depth = depth;
_data = data;
}
public GpuVector(CudaProvider cuda, IDeviceMemoryPtr data, bool isOwner)
{
_cuda = cuda;
_data = data;
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
internal GpuMatrix(CudaProvider cuda, int rows, int columns, IDeviceMemoryPtr gpuData)
{
_cuda = cuda;
_rows = rows;
_columns = columns;
_data = gpuData;
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public GpuMatrix(CudaProvider cuda, int rows, int columns, IDeviceMemoryPtr data, bool isOwner)
{
Debug.Assert(rows * columns == data.Size);
_cuda = cuda;
_rows = rows;
_columns = columns;
_data = data;
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public Gpu3DTensor(CudaProvider provider, int rows, int columns, int depth, IReadOnlyList <GpuMatrix> data)
{
_cuda = provider;
_rows = rows;
_columns = columns;
_depth = depth;
_data = data;
_tensorInfo = new Lazy <TensorInput>(() => new TensorInput(rows, columns, new[] { data.Select(m => m.Memory).ToList() }));
provider.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
internal GpuVector(CudaProvider cuda, int size, CudaDeviceVariable <float> data, bool shouldRegister = true)
{
_cuda = cuda;
_data = data;
_size = size;
if (shouldRegister)
{
cuda.Register(this);
}
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
internal GpuMatrix(CudaProvider cuda, int rows, int columns, CudaDeviceVariable <float> gpuData, bool shouldRegister = true)
{
_cuda = cuda;
_rows = rows;
_columns = columns;
_data = gpuData;
if (shouldRegister)
{
cuda.Register(this);
}
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public Gpu3DTensor(CudaProvider provider, int rows, int columns, int depth, IDeviceMemoryPtr data, bool isOwner)
{
Debug.Assert(rows * columns * depth == data.Size);
_cuda = provider;
_rows = rows;
_columns = columns;
_depth = depth;
_data = data;
_blockSize = rows * columns;
provider.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public GpuVector(CudaProvider cuda, int size, Func <int, float> init)
{
_cuda = cuda;
var data = new float[size];
for (var i = 0; i < size; i++)
{
data[i] = init(i);
}
_data = cuda.Allocate(size);
_data.CopyToDevice(data);
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
internal Gpu4DTensor(CudaProvider provider, IMatrix data, int rows, int columns, int depth)
{
_cuda = provider;
_data = data;
_rows = rows;
_columns = columns;
_depth = depth;
_count = data.ColumnCount;
provider.Register(this);
_subVector = new Lazy <List <GpuVector[]> >(_GetSubVectors);
_tensorInfo = new Lazy <TensorInput>(_GetInput);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public GpuVector(CudaProvider cuda, int size, Func <int, float> init)
{
_cuda = cuda;
var data = new float[size];
for (var i = 0; i < size; i++)
{
data[i] = init(i);
}
_data = new CudaDeviceVariable <float>(cuda.Context.AllocateMemory(size * sizeof(float)));
_data.CopyToDevice(data);
_size = size;
cuda.Register(this);
#if DEBUG
if (_id == _badAlloc)
{
Debugger.Break();
}
#endif
}
public GpuMatrix(CudaProvider cuda, IIndexableMatrix matrix) : this(cuda, matrix.RowCount, matrix.ColumnCount, (j, k) => matrix[j, k])
{
}
public GpuVector(CudaProvider cuda, IIndexableVector vector) : this(cuda, vector.Count, i => vector[i])
{
}
