DirectGpuEngine

Models & Types · AiDotNet.Tensors.Engines.DirectGpu

No summary documentation available yet.

Implements: IDisposable

Constructors

Constructor Summary
DirectGpuEngine()

Properties

Property Summary
IDirectGpuBackend Backend { get; }
KernelFusionManager FusionManager { get; }
bool IsAvailable { get; }
double GlobalMemoryGB { get; }
int ComputeUnits { get; }
int CudaDeviceCount { get; }
long GlobalMemoryBytes { get; }
long LocalMemoryBytes { get; }
string BackendName { get; }
string DeviceName { get; }
string DeviceVendor { get; }

Methods

Method Summary
IDirectGpuBackend CreateBackendForDevice(int deviceIndex)
IGpuBuffer AllocatePersistentBuffer(T[] data)
T[] DenseForwardFused(T[] input, T[] weights, T[] bias, int batchSize, int inputFeatures, int outputFeatures, ActivationType activation)
T[] ExecuteWithFusion(IReadOnlyList<FusableOperation> operations, T[] input, IReadOnlyDictionary<string, T[]> additionalBuffers, IReadOnlyDictionary<string, int> dimensions)
T[] FromFloatArray(float[] data)
T[] MatMul(T[] A, T[] B, int M, int K, int N)
T[] MatMulWithCachedWeights(T[] input, IGpuBuffer cachedWeights, int M, int K, int N)
T[] Relu(T[] input)
T[] Sigmoid(T[] input)
T[] Softmax(T[] input, int batchSize, int features)
bool CanAccessPeer(int device, int peerDevice)
float[] ToFloatArray(Memory<T> data)
float[] ToFloatArray(ReadOnlyMemory<T> data)
float[] ToFloatArray(T[] data)
string GetDiagnostics()
string GetFusionStatistics()
void Dispose()

Fields

Field Summary
bool StrictFp64Fallback