RecordingGpuBackend

Models & Types · AiDotNet.Tensors.Engines.Gpu

No summary documentation available yet.

Inheritance: DelegatingGpuBackend → RecordingGpuBackend

Constructors

Constructor Summary
RecordingGpuBackend(IDirectGpuBackend inner)

Properties

Property Summary
ExecutionGraphBuilder GraphBuilder { get; }
bool ExecuteWhileRecording { get; set; }
bool IsRecording { get; }

Methods

Method Summary
DeferredDownload DownloadBufferDeferred(IGpuBuffer buffer, int size)
IGpuBuffer AllocateBuffer(float[] data)
IGpuBuffer GemmBias(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuBuffer GemmBiasGelu(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuBuffer GemmBiasRelu(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuBuffer GemmBiasSigmoid(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuBuffer GemmBiasTanh(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuBuffer MatMul(IGpuBuffer A, IGpuBuffer B, int M, int N, int K)
float[] DownloadBuffer(IGpuBuffer buffer)
void AdaptiveAvgPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth)
void Add(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void AvgPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW, bool countIncludePad)
void BatchNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer runningMean, IGpuBuffer runningVar, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batch, int channels, int spatialSize, float epsilon, float momentum, bool training)
void BatchedGemm(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, int batchCount, float alpha, float beta)
void BeginRecording(ExecutionGraphBuilder graphBuilder)
void CDist(IGpuBuffer x1, IGpuBuffer x2, IGpuBuffer output, int m, int n, int d, float p)
void Conv2D(IGpuBuffer input, IGpuBuffer kernel, IGpuBuffer output, int batch, int inChannels, int inHeight, int inWidth, int outChannels, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW, int dilationH, int dilationW)
void Conv2DBiasAdd(IGpuBuffer output, IGpuBuffer bias, int batch, int channels, int spatialSize)
void ConvTranspose2D(IGpuBuffer input, IGpuBuffer kernel, IGpuBuffer output, int batch, int inChannels, int inHeight, int inWidth, int outChannels, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW, int outputPadH, int outputPadW)
void Copy(IGpuBuffer source, IGpuBuffer destination, int size)
void Copy(IGpuBuffer source, int sourceOffset, IGpuBuffer destination, int destinationOffset, int length)
void Divide(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void Embedding(IGpuBuffer indices, IGpuBuffer embeddingTable, IGpuBuffer output, int numIndices, int embeddingDim)
void EndRecording()
void FlashAttention(IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer output, IGpuBuffer mask, int batch, int numHeads, int seqLen, int headDim, float scale, bool isCausal)
void Gelu(IGpuBuffer A, IGpuBuffer B, int size)
void Gemm(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta)
void GlobalAvgPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int height, int width)
void GlobalMaxPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int height, int width)
void GroupNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batch, int numGroups, int channels, int spatialSize, float epsilon)
void InstanceNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batch, int channels, int spatialSize, float epsilon)
void Int32ToFloat(IGpuBuffer raw, IGpuBuffer output, int n)
void LayerNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batchSize, int normalizedSize, float epsilon)
void LeakyRelu(IGpuBuffer A, IGpuBuffer B, float alpha, int size)
void MaxPool2D(IGpuBuffer input, IGpuBuffer output, IGpuBuffer indices, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW)
void Multiply(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void NearestNeighborUpsample(IGpuBuffer input, IGpuBuffer output, int batchChannels, int height, int width, int scaleFactor)
void PDist(IGpuBuffer input, IGpuBuffer output, int n, int d, float p)
void Permute(IGpuBuffer input, IGpuBuffer output, int[] shape, int[] permutation)
void Relu(IGpuBuffer A, IGpuBuffer B, int size)
void RmsNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer saveRms, int batchSize, int normalizedSize, float epsilon)
void RopeInterleaved(IGpuBuffer input, IGpuBuffer cos, IGpuBuffer sin, IGpuBuffer output, int rows, int headDim, int seqLen, int startPosition)
void Scale(IGpuBuffer A, IGpuBuffer B, float scalar, int size)
void ScaledDotProductAttention(IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer output, IGpuBuffer attentionWeights, IGpuBuffer mask, int batch, int numHeads, int seqQ, int seqK, int headDim, float scale, bool isCausal, float softcap, int numKVHeads)
void ScatterPositions(IGpuBuffer output, int srcDim, int inner, int n)
void Sigmoid(IGpuBuffer A, IGpuBuffer B, int size)
void Softmax(IGpuBuffer A, IGpuBuffer B, int batchSize, int features)
void Subtract(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void Swish(IGpuBuffer A, IGpuBuffer B, int size)
void Tanh(IGpuBuffer A, IGpuBuffer B, int size)
void Transpose(IGpuBuffer A, IGpuBuffer B, int rows, int cols)