public CudaStorage(IAllocator allocator, TSCudaContext tsContext, CudaContext context, DType ElementType, long elementCount)
: base(allocator, ElementType, elementCount)
{
TSContext = tsContext;
this.context = context;
bufferHandle = tsContext.AllocatorForDevice(DeviceId).Allocate(ByteLength);
deviceBuffer = bufferHandle.Pointer;
}
public Tensor AddmmBatch(Tensor result, float beta, Tensor src, float alpha, Tensor m1, Tensor m2)
{
TSCudaContext context = CudaHelpers.TSContextForTensor(src);
if (src.ElementType != m1.ElementType || src.ElementType != m2.ElementType || (result != null && result.ElementType != src.ElementType))
{
throw new InvalidOperationException("All tensors must have the same element type");
}
if (result != null && !(result.Storage is CudaStorage))
{
throw new ArgumentException("result must be a CUDA tensor", "result");
}
if (!(m1.Storage is CudaStorage))
{
throw new ArgumentException("m1 must be a CUDA tensor", "m1");
}
if (!(m2.Storage is CudaStorage))
{
throw new ArgumentException("m2 must be a CUDA tensor", "m2");
}
if (src.DimensionCount != 3)
{
throw new ArgumentException("src must be a matrix", "src");
}
if (m1.DimensionCount != 3)
{
throw new ArgumentException("m1 must be a matrix", "m1");
}
if (m2.DimensionCount != 3)
{
throw new ArgumentException("m2 must be a matrix", "m2");
}
if (src.Sizes[1] != m1.Sizes[1] || src.Sizes[2] != m2.Sizes[2] || m1.Sizes[2] != m2.Sizes[1])
{
throw new InvalidOperationException($"Size mismatch, srcSize0 = {src.Sizes[0]}, m1Size0 = {m1.Sizes[0]}, srcSize1 = {src.Sizes[1]}, m2Size1 = {m2.Sizes[1]}, m1Size1 = '{m1.Sizes[1]}', m2Size0 = '{m2.Sizes[0]}'");
}
Tensor writeTarget = TensorResultBuilder.GetWriteTarget(result, src, true, src.Sizes);
if (writeTarget != src)
{
Ops.Copy(writeTarget, src);
}
CudaMatrixMulMM.GemmBatch(context, alpha, m1, m2, beta, writeTarget);
return(writeTarget);
}
public Tensor Dot(Tensor result, Tensor lhs, Tensor rhs)
{
TSCudaContext context = CudaHelpers.TSContextForTensor(lhs);
if (lhs.DimensionCount == 1 && rhs.DimensionCount == 1)
{
return(CudaMatrixMulDot.Dot(context, result, lhs, rhs));
}
else if (lhs.DimensionCount == 2 && rhs.DimensionCount == 1)
{
return(CudaMatrixMulMV.Mul_M_V(context, result, lhs, rhs));
}
else if (lhs.DimensionCount == 2 && rhs.DimensionCount == 2)
{
return(CudaMatrixMulMM.Mul_M_M(context, result, lhs, rhs));
}
else
{
throw new NotSupportedException(string.Format("Multiplication of {0}D with {1}D tensor is not supported"));
}
}
/// <summary>
/// Initializes a new instance of the <see cref="CudaAllocator"/> class.
/// </summary>
/// <param name="context">The context.</param>
/// <param name="deviceId">The device identifier.</param>
public CudaAllocator(TSCudaContext context, int deviceId)
{
this.context = context;
this.deviceId = deviceId;
}
public static Tensor Addmm(Tensor result, float beta, Tensor src, float alpha, Tensor m1, Tensor m2)
{
try
{
TSCudaContext context = CudaHelpers.TSContextForTensor(src);
if (src.ElementType != m1.ElementType || src.ElementType != m2.ElementType || (result != null && result.ElementType != src.ElementType))
{
throw new InvalidOperationException("All tensors must have the same element type");
}
if (result != null && !(result.Storage is CudaStorage))
{
throw new ArgumentException("result must be a CUDA tensor", nameof(result));
}
if (!(m1.Storage is CudaStorage))
{
throw new ArgumentException("m1 must be a CUDA tensor", nameof(m1));
}
if (!(m2.Storage is CudaStorage))
{
throw new ArgumentException("m2 must be a CUDA tensor", nameof(m2));
}
if (src.DimensionCount != 2)
{
throw new ArgumentException("src must be a matrix", nameof(src));
}
if (m1.DimensionCount != 2)
{
throw new ArgumentException("m1 must be a matrix", nameof(m1));
}
if (m2.DimensionCount != 2)
{
throw new ArgumentException("m2 must be a matrix", nameof(m2));
}
if (src.Sizes[0] != m1.Sizes[0] || src.Sizes[1] != m2.Sizes[1] || m1.Sizes[1] != m2.Sizes[0])
{
throw new InvalidOperationException($"Size mismatch, srcSize0 = {src.Sizes[0]}, m1Size0 = {m1.Sizes[0]}, srcSize1 = {src.Sizes[1]}, m2Size1 = {m2.Sizes[1]}, m1Size1 = '{m1.Sizes[1]}', m2Size0 = '{m2.Sizes[0]}'");
}
Tensor writeTarget = TensorResultBuilder.GetWriteTarget(result, src, false, src.Sizes);
if (writeTarget != src)
{
Ops.Copy(writeTarget, src);
}
CudaMatrixMulMM.Gemm(context, alpha, m1, m2, beta, writeTarget);
return(writeTarget);
}
catch (Exception err)
{
Logger.WriteLine($"Exception in Addmm: '{err.Message}'");
Logger.WriteLine($"Call stack: '{err.StackTrace}'");
throw;
}
}
