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 |
|