MetalBackend

Models & Types · AiDotNet.Tensors.Engines.DirectGpu.Metal

No summary documentation available yet.

Implements: IAdam8BitWeightDecayBackend, IAnnBackend, IAsyncGpuBackend, IAudioBackend, ICtcLossBackend, IDetectionBackend, IExtendedLinalgBackend, IFftBackend, IFrexpBackend, IGeometryBackend, IGeometryBackwardBackend, IGpuFp16ElementwiseBackend, IGpuHalfPrecisionBackend, IGpuMixedPrecisionConvBackend, IGpuPrecisionBackend, IImportanceSamplingBackend, IInstantNgpBackend, ILinalgBackend, IMeshgridBackend, IModeBackend, IMultiTensorGpuOptimizerBackend, INativeGpuCodegenExecutor, INmsBackend, INonzeroBackend, IOneHotIndexBackend, IParity210Backend, IPixelShuffleBackend, IRenderingGeometryBackend, IResidentIndexBackend, IRoiBackend, ISpiralIndicesBackend, IUniqueConsecutiveBackend

Constructors

Constructor Summary
MetalBackend()
MetalBackend(int deviceIndex)

Properties

Property Summary
IGpuMixedPrecisionConvBackend MixedPrecisionConv { get; }
IGpuStream DefaultStream { get; }
KernelTuningBackend NativeCodegenBackend { get; }
TensorDevice DeviceType { get; }
bool ArgMaxIndicesAreBitReinterpreted { get; }
bool Fp16Im2colAvailable { get; }
bool GuardsFiniteOnDevice { get; }
bool IsAvailable { get; }
bool LinalgKernelsReady { get; }
bool SupportsAsyncTransfer { get; }
bool SupportsBatchExecution { get; }
bool SupportsEvents { get; }
bool SupportsFp16FusedBackward { get; }
bool SupportsFp16NativeOps { get; }
bool SupportsGraphCapture { get; }
bool SupportsHgemm { get; }
bool SupportsMultiStream { get; }
double TheoreticalGflops { get; }
int ComputeUnits { get; }
int LinalgEighMaxN { get; }
int MaxConcurrentStreams { get; }
long GlobalMemoryBytes { get; }
long LocalMemoryBytes { get; }
long MaxBufferAllocBytes { get; }
string BackendName { get; }
string DeviceName { get; }
string DeviceVendor { get; }

Methods

Method Summary
CodegenEmitResult EmitCodegenKernel(CodegenGraph graph, CodegenElementType dtype)
GpuSyncPoint CreateSyncPoint()
GpuSyncPoint CreateSyncPoint(IGpuStream stream)
IGpuBuffer AllocateBuffer(float[] data)
IGpuBuffer AllocateBuffer(int size)
IGpuBuffer AllocateBufferAsync(float[] data, IGpuStream stream)
IGpuBuffer AllocateBufferUninitialized(int size)
IGpuBuffer AllocateByteBuffer(int size)
IGpuBuffer AllocateIntBuffer(int size)
IGpuBuffer AllocateIntBuffer(int[] data)
IGpuBuffer AllocateWorkspaceBuffer(int totalElements)
IGpuBuffer DequantGemmFp8E4M3(IGpuBuffer activations, IGpuBuffer weightsFp8Raw, IGpuBuffer scales, int M, int K, int N, int groupSize, int scaleCount)
IGpuBuffer DequantGemmInt(IGpuBuffer activations, IGpuBuffer weightsInt, IGpuBuffer scales, int M, int K, int N, int groupSize, int scaleCount)
IGpuBuffer FlashDecode(IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, int heads, int kvHeads, int headDim, int seqLen, float scale, int splits)
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 GemmBiasLeakyRelu(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K, float alpha)
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 GemmBiasSwish(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)
IGpuBuffer PagedAttentionDecode(IGpuBuffer q, IGpuBuffer kcache, IGpuBuffer vcache, IGpuBuffer blockTable, int heads, int headDim, int blockSize, int seqLen, float scale)
IGpuBuffer PagedAttentionDecodeGqa(IGpuBuffer q, IGpuBuffer kcache, IGpuBuffer vcache, IGpuBuffer blockTable, int heads, int kvHeads, int headDim, int blockSize, int seqLen, float scale)
IGpuBuffer PagedAttentionPrefill(IGpuBuffer q, IGpuBuffer kcache, IGpuBuffer vcache, IGpuBuffer blockTable, int heads, int headDim, int blockSize, int numQueries, int startPos, float scale)
IGpuBuffer PagedAttentionPrefillGqa(IGpuBuffer q, IGpuBuffer kcache, IGpuBuffer vcache, IGpuBuffer blockTable, int heads, int kvHeads, int headDim, int blockSize, int numQueries, int startPos, float scale)
IGpuBuffer SparseGemmBiasRelu(IGpuBuffer sparseAValues, IGpuBuffer sparseAIndices, IGpuBuffer B, IGpuBuffer bias, int M, int N, int K)
IGpuEvent CreateEvent()
IGpuEvent CreateEvent(bool enableTiming)
IGpuStream CreateStream(GpuStreamType streamType)
IGpuStream CreateStream(GpuStreamType streamType, int priority)
IReadOnlyList<GpuPrecisionCapability> GetPrecisionCapabilities(GpuPrecisionOperation operation)
ValueTask ExecuteCodegenKernelAsync(CodegenKernel kernel, IReadOnlyList<IGpuBuffer> inputs, IReadOnlyList<IGpuBuffer> outputs, int elementCount, CancellationToken cancellationToken)
bool CanCategoricalSample(int rows, int classes)
bool CanExecuteCodegenKernel(CodegenKernel kernel)
bool QueryEventComplete(IGpuEvent gpuEvent)
bool QueryStreamComplete(IGpuStream stream)
bool TryCategoricalSample(IGpuBuffer probabilities, IGpuBuffer oneHot, int rows, int classes, ulong seed)
bool TryConv2DBiasActivation(IGpuBuffer input, IGpuBuffer kernel, IGpuBuffer bias, 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, FusedActivationType activation)
bool TryFusedBatchNormActivation(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, FusedActivationType activation)
bool TryFusedBiasDropout(IGpuBuffer input, IGpuBuffer output, IGpuBuffer bias, IGpuBuffer mask, int rows, int cols, float dropoutRate, float scale)
byte[] DownloadByteBuffer(IGpuBuffer buffer, int byteCount)
float BinaryCrossEntropyLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float CategoricalCrossEntropyLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float CharbonnierLoss(IGpuBuffer predictions, IGpuBuffer targets, int size, float epsilon)
float ContrastiveLoss(IGpuBuffer output1, IGpuBuffer output2, IGpuBuffer labels, int batchSize, int embeddingDim, float margin)
float CrossEntropyLoss(IGpuBuffer predictions, IGpuBuffer targets, int batchSize, int numClasses)
float ElasticNetLoss(IGpuBuffer predictions, IGpuBuffer targets, int size, float l1Weight, float l2Weight)
float ExponentialLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float FocalLoss(IGpuBuffer predictions, IGpuBuffer targets, int size, float alpha, float gamma)
float FusedLinearCrossEntropyDense(IGpuBuffer hidden, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer target, int n, int d, int vocab)
float FusedLinearCrossEntropyIndex(IGpuBuffer hidden, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer targetIds, int n, int d, int vocab)
float GetEventElapsedTime(IGpuEvent start, IGpuEvent end)
float HingeLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float HuberLoss(IGpuBuffer predictions, IGpuBuffer targets, int size, float delta)
float L2Norm(IGpuBuffer A, int size)
float LogCoshLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float MaeLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float Max(IGpuBuffer A, int size)
float Min(IGpuBuffer A, int size)
float ModifiedHuberLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float MseLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float PoissonLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float QuantileLoss(IGpuBuffer predictions, IGpuBuffer targets, int size, float quantile)
float SmoothL1Loss(IGpuBuffer predictions, IGpuBuffer targets, int size, float beta)
float SquaredHingeLoss(IGpuBuffer predictions, IGpuBuffer targets, int size)
float StdDev(IGpuBuffer input, int size)
float Sum(IGpuBuffer A, int size)
float TripletLoss(IGpuBuffer anchor, IGpuBuffer positive, IGpuBuffer negative, int batchSize, int embeddingDim, float margin)
float[] DownloadBuffer(IGpuBuffer buffer)
string ToString()
void Abs(IGpuBuffer A, IGpuBuffer B, int size)
void Acos(IGpuBuffer A, IGpuBuffer B, int size)
void Acosh(IGpuBuffer A, IGpuBuffer B, int size)
void AdadeltaUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer accumGrad, IGpuBuffer accumUpdate, float rho, float epsilon, float weightDecay, int size)
void AdagradUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer accumulatedGrad, float learningRate, float epsilon, float weightDecay, int size)
void Adam8BitUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer mQuant, IGpuBuffer vQuant, IGpuBuffer mScales, IGpuBuffer vScales, float learningRate, float beta1, float beta2, float epsilon, float oneMinusBeta1, float oneMinusBeta2, float biasCorrection1, float biasCorrection2, int blockSize, int paramLength, int numBlocks)
void Adam8BitUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer mQuant, IGpuBuffer vQuant, IGpuBuffer mScales, IGpuBuffer vScales, float learningRate, float beta1, float beta2, float epsilon, float oneMinusBeta1, float oneMinusBeta2, float biasCorrection1, float biasCorrection2, int blockSize, int paramLength, int numBlocks, float weightDecay, bool decoupledWeightDecay)
void AdamMultiTensorUpdate(IReadOnlyList<IGpuBuffer> parameters, IReadOnlyList<IGpuBuffer> gradients, IReadOnlyList<IGpuBuffer> firstMoments, IReadOnlyList<IGpuBuffer> secondMoments, IReadOnlyList<int> sizes, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void AdamMultiTensorUpdateGuarded(bool decoupledWeightDecay, IReadOnlyList<IGpuBuffer> parameters, IReadOnlyList<IGpuBuffer> gradients, IReadOnlyList<IGpuBuffer> firstMoments, IReadOnlyList<IGpuBuffer> secondMoments, IReadOnlyList<int> sizes, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, IGpuBuffer finiteGuard)
void AdamUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AdamUpdateBf16(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AdamWMultiTensorUpdate(IReadOnlyList<IGpuBuffer> parameters, IReadOnlyList<IGpuBuffer> gradients, IReadOnlyList<IGpuBuffer> firstMoments, IReadOnlyList<IGpuBuffer> secondMoments, IReadOnlyList<int> sizes, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void AdamWUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AdamWUpdateBf16(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AdamaxUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer u, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AdaptiveAvgPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth)
void AdaptiveMaxPool2D(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 AddGelu(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void AddRelu(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void AddScalar(IGpuBuffer A, IGpuBuffer B, float scalar, int size)
void AddScaled(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, float scaleA, float scaleB, int size)
void AddSigmoid(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void AffineGrid(IGpuBuffer theta, IGpuBuffer grid, int batch, int outputHeight, int outputWidth)
void AffineGrid3D(IGpuBuffer theta, IGpuBuffer grid, int N, int D, int H, int W, bool alignCorners)
void AmplitudeToDB(IGpuBuffer input, IGpuBuffer output, int length, float minAmplitude, float topDb, bool clipTopDb)
void AmsgradUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, IGpuBuffer vMax, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void AnalyticSignalMask(IGpuBuffer specReal, IGpuBuffer specImag, IGpuBuffer outReal, IGpuBuffer outImag, int batch, int fftSize, int binLow, int binHigh)
void ApplyMelFilterbank(IGpuBuffer powerSpec, IGpuBuffer filterbank, IGpuBuffer melSpec, int numFrames, int numFreqs, int nMels)
void ApplyWindow(IGpuBuffer input, IGpuBuffer window, IGpuBuffer output, int n)
void ArgMax(IGpuBuffer A, IGpuBuffer indices, int outerSize, int reduceSize)
void ArgMaxAxis(IGpuBuffer A, IGpuBuffer indices, int outerSize, int reduceSize)
void ArgMin(IGpuBuffer A, IGpuBuffer indices, int outerSize, int reduceSize)
void ArgMinAxis(IGpuBuffer A, IGpuBuffer indices, int outerSize, int reduceSize)
void Asin(IGpuBuffer A, IGpuBuffer B, int size)
void Asinh(IGpuBuffer A, IGpuBuffer B, int size)
void Atan(IGpuBuffer A, IGpuBuffer B, int size)
void Atan2Elementwise(IGpuBuffer real, IGpuBuffer imag, IGpuBuffer output, int n)
void Atanh(IGpuBuffer A, IGpuBuffer B, int size)
void AvgPool1D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inLength, int outLength, int kernelSize, int stride)
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 AvgPool2DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, 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 AvgPool3D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inDepth, int inHeight, int inWidth, int outDepth, int outHeight, int outWidth, int kernelD, int kernelH, int kernelW, int strideD, int strideH, int strideW, int countIncludePad)
void AvgPool3DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int inDepth, int inHeight, int inWidth, int outDepth, int outHeight, int outWidth, int kernelD, int kernelH, int kernelW, int strideD, int strideH, int strideW, int countIncludePad)
void BatchDotProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int bs, int dim)
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 BatchNormBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gamma, IGpuBuffer saveMean, IGpuBuffer saveInvVar, IGpuBuffer gradInput, IGpuBuffer gradGamma, IGpuBuffer gradBeta, int batch, int channels, int spatialSize, float epsilon)
void BatchOuterProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int batchSize, int M, int N)
void BatchedDotProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer result, int batchSize, int vecSize)
void BatchedFFT(IGpuBuffer inputReal, IGpuBuffer inputImag, IGpuBuffer outputReal, IGpuBuffer outputImag, int batch, int n, bool inverse)
void BatchedFFT2D(IGpuBuffer inputReal, IGpuBuffer inputImag, IGpuBuffer outputReal, IGpuBuffer outputImag, int batch, int height, int width, bool inverse)
void BatchedGemm(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, int batchCount, float alpha, float beta)
void BatchedTranspose(IGpuBuffer A, IGpuBuffer B, int batch, int rows, int cols)
void BceLoss(IGpuBuffer p, IGpuBuffer t, IGpuBuffer l, int sz)
void BceWithLogitsBackward(IGpuBuffer gradOutput, IGpuBuffer logits, IGpuBuffer targets, IGpuBuffer gradInput, int size, float invN)
void BceWithLogitsLoss(IGpuBuffer logits, IGpuBuffer targets, IGpuBuffer loss, int size)
void BeginBatch()
void BeginSecondaryStream()
void BiasAdd(IGpuBuffer A, IGpuBuffer bias, IGpuBuffer C, int M, int N)
void BilinearUpsample2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inH, int inW, int outH, int outW)
void BinaryCrossEntropyBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void BispectrumGather(IGpuBuffer specReal, IGpuBuffer specImag, IGpuBuffer outReal, IGpuBuffer outImag, int maxF1, int maxF2)
void BitonicStep(IGpuBuffer values, IGpuBuffer indices, int rowLen, int k, int j, int numRows, int descending)
void BoxArea(IGpuBuffer boxes, IGpuBuffer output, int n)
void BoxConvert(IGpuBuffer boxes, IGpuBuffer output, int n, int fromFormat, int toFormat)
void BoxIou(IGpuBuffer boxesA, IGpuBuffer boxesB, IGpuBuffer output, int n, int m)
void BroadcastAddFirst(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int outerSize, int innerSize)
void BroadcastAddLast(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int os, int isz)
void BroadcastDivLast(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int os, int isz)
void BroadcastMulFirst(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int outerSize, int innerSize)
void BroadcastMulLast(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int os, int isz)
void BroadcastMultiplyFirstAxis(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int outerSize, int innerSize)
void BroadcastMultiplyLastAxis(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int outerSize, int innerSize)
void BroadcastSubLast(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int os, int isz)
void BuildSpectrum(IGpuBuffer mag, IGpuBuffer phase, IGpuBuffer specRe, IGpuBuffer specIm, int batch, int numFreqs, int numFrames, int nFft)
void CDist(IGpuBuffer x1, IGpuBuffer x2, IGpuBuffer output, int m, int n, int d, float p)
void CIoULoss(IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer loss, int numBoxes)
void CIoULossBackward(IGpuBuffer gradOutput, IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer gradPredicted, int numBoxes)
void CapsuleAgreement(IGpuBuffer predictions, IGpuBuffer output, IGpuBuffer agreement, int batchSize, int inputCapsules, int outputCapsules, int capsuleDim)
void CapsulePredictions(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer output, int batchSize, int inputCapsules, int inputDim, int outputCapsules, int outputDim)
void CapsuleTransform(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer output, int batchSize, int inputCapsules, int inputDim, int numCapsules, int capsuleDim)
void CapsuleWeightedSum(IGpuBuffer coupling, IGpuBuffer predictions, IGpuBuffer output, int batchSize, int inputCapsules, int outputCapsules, int capsuleDim)
void CategoricalCrossEntropyBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void CavityBounceInplace(IGpuBuffer workReal, IGpuBuffer workImag, int total, float invN)
void Cbrt(IGpuBuffer A, IGpuBuffer B, int size)
void Ceiling(IGpuBuffer A, IGpuBuffer B, int size)
void CharbonnierBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float epsilon)
void Clamp(IGpuBuffer A, IGpuBuffer B, float min, float max, int size)
void ClassifyFloat(IGpuBuffer A, IGpuBuffer C, int mode, int size)
void ClipByNorm(IGpuBuffer A, IGpuBuffer B, float maxNorm, int size)
void ClipByValue(IGpuBuffer A, IGpuBuffer B, float clipValue, int size)
void ClipKernel(IGpuBuffer i, IGpuBuffer o, float mn, float mx, int sz)
void ClipScaleFromSumOfSquares(IGpuBuffer sumOfSquares, float maxNorm, IGpuBuffer scale)
void CompleteBoxIou(IGpuBuffer boxesA, IGpuBuffer boxesB, IGpuBuffer output, int n, int m)
void ComplexConjugate(IGpuBuffer input, IGpuBuffer output, int numPairs)
void ComplexDiagonalSsmScanForward(IGpuBuffer input, IGpuBuffer transitionReal, IGpuBuffer transitionImag, IGpuBuffer inputMapReal, IGpuBuffer inputMapImag, IGpuBuffer outputMapReal, IGpuBuffer outputMapImag, IGpuBuffer skip, IGpuBuffer output, int batch, int time, int groups, int width, int state)
void ComplexMagnitude(IGpuBuffer input, IGpuBuffer output, int numPairs)
void ComplexMagnitude(IGpuBuffer real, IGpuBuffer imag, IGpuBuffer magnitude, int n)
void ComplexMatVec(IGpuBuffer matReal, IGpuBuffer matImag, IGpuBuffer vecReal, IGpuBuffer vecImag, IGpuBuffer outReal, IGpuBuffer outImag, int batchSize, int dim)
void ComplexMultiply(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int numPairs)
void ComplexPhase(IGpuBuffer real, IGpuBuffer imag, IGpuBuffer phase, int n)
void ComputeDeltas(IGpuBuffer input, IGpuBuffer output, int leading, int timeAxis, int winLength)
void ComputeDistances(IGpuBuffer queries, IGpuBuffer database, IGpuBuffer distances, int numQueries, int numDatabase, int dim, AnnMetric metric)
void ConcatAxis(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int outerSize, int aInnerSize, int bInnerSize)
void ContrastiveBackward(IGpuBuffer output1, IGpuBuffer output2, IGpuBuffer labels, IGpuBuffer gradOutput1, IGpuBuffer gradOutput2, int batchSize, int embeddingDim, float margin)
void Conv1D(IGpuBuffer input, IGpuBuffer kernel, IGpuBuffer output, int batch, int inChannels, int inLength, int outChannels, int outLength, int kernelLength, int stride, int padding, int dilation)
void Conv1DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer kernel, IGpuBuffer gradInput, int batch, int inChannels, int inLength, int outChannels, int outLength, int kernelLength, int stride, int padding, int dilation)
void Conv1DBackwardKernel(IGpuBuffer input, IGpuBuffer gradOutput, IGpuBuffer gradKernel, int batch, int inChannels, int inLength, int outChannels, int outLength, int kernelLength, int stride, int padding, int dilation)
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 Conv2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer kernel, IGpuBuffer gradInput, 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 Conv2DBackwardKernel(IGpuBuffer input, IGpuBuffer gradOutput, IGpuBuffer gradKernel, 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 Conv3D(IGpuBuffer input, IGpuBuffer kernel, IGpuBuffer output, int batch, int inChannels, int inDepth, int inHeight, int inWidth, int outChannels, int outDepth, int outHeight, int outWidth, int kernelD, int kernelH, int kernelW, int strideD, int strideH, int strideW, int padD, int padH, int padW, int dilationD, int dilationH, int dilationW)
void Conv3DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer weights, IGpuBuffer gradInput, int n, int inC, int inD, int inH, int inW, int outC, int outD, int outH, int outW, int kD, int kH, int kW, int strideD, int strideH, int strideW, int padD, int padH, int padW)
void Conv3DBackwardKernel(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradKernel, int n, int inC, int inD, int inH, int inW, int outC, int outD, int outH, int outW, int kD, int kH, int kW, int strideD, int strideH, int strideW, int padD, int padH, int padW)
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 ConvTranspose2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer kernel, IGpuBuffer gradInput, 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 ConvTranspose2DBackwardKernel(IGpuBuffer input, IGpuBuffer gradOutput, IGpuBuffer gradKernel, 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 ConvTranspose3D(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer output, int n, int inC, int iD, int iH, int iW, int outC, int outD, int outH, int outW, int kD, int kH, int kW, int strideD, int strideH, int strideW, int padD, int padH, int padW)
void ConvTranspose3DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer weights, IGpuBuffer gradInput, int n, int inC, int iD, int iH, int iW, int outC, int outD, int outH, int outW, int kD, int kH, int kW, int strideD, int strideH, int strideW, int padD, int padH, int padW)
void ConvTranspose3DBackwardKernel(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradWeights, int n, int inC, int iD, int iH, int iW, int outC, int outD, int outH, int outW, int kD, int kH, int kW, int strideD, int strideH, int strideW, int padD, int padH, int padW)
void ConvertIndicesToInt32(IGpuBuffer numericIndices, IGpuBuffer int32Indices, int length)
void ConvertToFp16(IGpuBuffer input, IGpuBuffer output, int size)
void ConvertToFp32(IGpuBuffer input, IGpuBuffer output, int size)
void Copy(IGpuBuffer source, IGpuBuffer destination, int size)
void Copy(IGpuBuffer source, int srcOffset, IGpuBuffer destination, int destOffset, int size)
void Copy2DStrided(IGpuBuffer source, IGpuBuffer destination, int numRows, int srcCols, int destTotalCols, int destColOffset)
void CopyBlock2D(IGpuBuffer block, IGpuBuffer output, int blockRows, int blockCols, int totalCols, int rowOff, int colOff)
void CopyBuffer(IGpuBuffer source, IGpuBuffer destination, int size)
void CopyBufferAsync(IGpuBuffer source, IGpuBuffer destination, int size, IGpuStream stream)
void CopyRows(IGpuBuffer src, IGpuBuffer dst, int srcRowLen, int dstRowLen, int numRows, int copyLen)
void Cos(IGpuBuffer A, IGpuBuffer B, int size)
void Cosh(IGpuBuffer A, IGpuBuffer B, int size)
void CosineSimilarity(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int batchSize, int dim)
void Crop2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inH, int inW, int outH, int outW, int offsetH, int offsetW)
void Crop2DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int inH, int inW, int outH, int outW, int offsetH, int offsetW)
void Cross3(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int outerSize, int innerSize)
void CrossEntropyBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int batchSize, int numClasses)
void CrossEntropyLoss(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer loss, int batchSize, int numClasses)
void CsrSddmm(IGpuBuffer rowIndices, IGpuBuffer colIndices, IGpuBuffer x, IGpuBuffer y, IGpuBuffer output, int nnz, int innerK)
void CsrSegmentedMax(IGpuBuffer csrColIndices, IGpuBuffer csrRowPointers, IGpuBuffer input, IGpuBuffer output, int M, int K, int N)
void CsrSegmentedMin(IGpuBuffer csrColIndices, IGpuBuffer csrRowPointers, IGpuBuffer input, IGpuBuffer output, int M, int K, int N)
void CsrSegmentedStdDev(IGpuBuffer csrColIndices, IGpuBuffer csrRowPointers, IGpuBuffer input, IGpuBuffer output, int M, int K, int N, float epsilon)
void CsrSpMM(IGpuBuffer csrValues, IGpuBuffer csrColIndices, IGpuBuffer csrRowPointers, IGpuBuffer denseB, IGpuBuffer output, int M, int K, int N, int nnz)
void CsrSpMMBias(IGpuBuffer csrValues, IGpuBuffer csrColIndices, IGpuBuffer csrRowPointers, IGpuBuffer denseB, IGpuBuffer bias, IGpuBuffer output, int M, int K, int N, int nnz)
void CtcLoss(IGpuBuffer logProbs, IGpuBuffer targets, IGpuBuffer inputLengths, IGpuBuffer targetLengths, IGpuBuffer workspace, IGpuBuffer losses, int maxTime, int batchSize, int numClasses, int maxTargetLength, int blank)
void CumSumAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void DIoULoss(IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer loss, int numBoxes)
void DIoULossBackward(IGpuBuffer gradOutput, IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer gradPredicted, int numBoxes)
void DbToPower(IGpuBuffer db, IGpuBuffer power, int n, float refValue)
void Decompress2x4Sparse(IGpuBuffer sparseValues, IGpuBuffer sparseIndices, IGpuBuffer denseOutput, int M, int K)
void DeformableConv2D(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer offsets, IGpuBuffer mask, 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, int groups, int deformGroups)
void DeformableConv2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer weights, IGpuBuffer offsets, IGpuBuffer mask, IGpuBuffer gradInput, 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, int groups, int deformGroups)
void DeformableConv2DBackwardMask(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weights, IGpuBuffer offsets, IGpuBuffer gradMask, 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, int groups, int deformGroups)
void DeformableConv2DBackwardOffset(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weights, IGpuBuffer offsets, IGpuBuffer mask, IGpuBuffer gradOffsets, 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, int groups, int deformGroups)
void DeformableConv2DBackwardWeights(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer offsets, IGpuBuffer mask, IGpuBuffer gradWeights, 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, int groups, int deformGroups)
void DeinterleaveComplex(IGpuBuffer interleaved, IGpuBuffer real, IGpuBuffer imag, int n)
void DepthwiseConv2D(IGpuBuffer input, IGpuBuffer kernel, 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)
void DepthwiseConv2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer kernel, IGpuBuffer gradInput, int n, int inC, int h, int w, int m, int outH, int outW, int kH, int kW, int strideH, int strideW, int padH, int padW)
void DepthwiseConv2DBackwardKernel(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradKernel, int n, int inC, int h, int w, int m, int outH, int outW, int kH, int kW, int strideH, int strideW, int padH, int padW)
void DiagKernel(IGpuBuffer input, IGpuBuffer output, int n)
void Dispose()
void DistanceBoxIou(IGpuBuffer boxesA, IGpuBuffer boxesB, IGpuBuffer output, int n, int m)
void DivScalar(IGpuBuffer input, IGpuBuffer output, float scalar, int size)
void Divide(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void DotProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer result, int size)
void DownloadBuffer(IGpuBuffer buffer, float[] destination)
void DownloadBufferAsync(IGpuBuffer buffer, float[] destination, IGpuStream stream)
void Dropout(IGpuBuffer input, IGpuBuffer output, IGpuBuffer mask, int size, float dropoutRate, ulong seed, bool training)
void DropoutBackward(IGpuBuffer gradOutput, IGpuBuffer mask, IGpuBuffer gradInput, int size, float dropoutRate)
void DropoutMask(IGpuBuffer mask, int size, float keepProb, ulong seed)
void ElasticNetBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float l1Weight, float l2Weight)
void Elu(IGpuBuffer A, IGpuBuffer B, float alpha, int size)
void EluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer output, IGpuBuffer gradInput, float alpha, int size)
void Embedding(IGpuBuffer indices, IGpuBuffer embeddingTable, IGpuBuffer output, int numIndices, int embeddingDim)
void EmbeddingBackward(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer gradEmbedding, int numIndices, int embeddingDim, int vocabSize)
void EndBatch()
void EndSecondaryStream()
void Enforce2x4Sparsity(IGpuBuffer denseInput, IGpuBuffer sparseValues, IGpuBuffer sparseIndices, int M, int K)
void Equal(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void EqualsKernel(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int sz)
void Exp(IGpuBuffer A, IGpuBuffer B, int size)
void Exp10(IGpuBuffer A, IGpuBuffer B, int size)
void Exp2(IGpuBuffer A, IGpuBuffer B, int size)
void ExpM1(IGpuBuffer A, IGpuBuffer B, int size)
void ExponentialBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void ExtractDiagKernel(IGpuBuffer input, IGpuBuffer output, int n, int cols)
void EyeKernel(IGpuBuffer o, int n)
void FFT(IGpuBuffer inputReal, IGpuBuffer inputImag, IGpuBuffer outputReal, IGpuBuffer outputImag, int n, bool inverse)
void FFT2D(IGpuBuffer inputReal, IGpuBuffer inputImag, IGpuBuffer outputReal, IGpuBuffer outputImag, int height, int width, bool inverse)
void Fill(IGpuBuffer buffer, float value, int size)
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 FlashAttentionBackward(IGpuBuffer gradOutput, IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer output, IGpuBuffer softmaxStats, IGpuBuffer gradQuery, IGpuBuffer gradKey, IGpuBuffer gradValue, int batch, int numHeads, int seqQ, int seqK, int headDim, float scale, bool isCausal, IGpuBuffer attentionBias, int biasBatchStride)
void FlashAttentionV2(IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer output, IGpuBuffer softmaxStats, int batch, int numHeads, int seqQ, int seqK, int headDim, float scale, bool isCausal, IGpuBuffer attentionBias, int biasBatchStride)
void Floor(IGpuBuffer A, IGpuBuffer B, int size)
void Fma(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, IGpuBuffer D, int size)
void FocalBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float alpha, float gamma)
void Fold(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int outputH, int outputW, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW)
void Fp16Add(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int n)
void Fp16Gelu(IGpuBuffer input, IGpuBuffer output, int n)
void Fp16LayerNorm(IGpuBuffer input, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer output, IGpuBuffer meanFp32, IGpuBuffer varFp32, int rows, int cols, float eps)
void Fp16Relu(IGpuBuffer input, IGpuBuffer output, int n)
void Fp16Softmax(IGpuBuffer input, IGpuBuffer output, int rows, int cols)
void FracKernel(IGpuBuffer i, IGpuBuffer o, int sz)
void Frexp(IGpuBuffer input, IGpuBuffer mantissa, IGpuBuffer exponent, int length)
void FtrlUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer z, IGpuBuffer n, float learningRate, float l1Reg, float l2Reg, float beta, int size)
void FusedDDIMStep(IGpuBuffer xT, IGpuBuffer epsilonTheta, IGpuBuffer output, int size, float alphaBarT, float alphaBarTMinus1)
void FusedGemmBiasActivationAsync(IGpuBuffer A, IGpuBuffer B, IGpuBuffer bias, IGpuBuffer output, int M, int N, int K, FusedActivationType activation, IGpuStream stream)
void FusedLinearCrossEntropyDense(IGpuBuffer hidden, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer target, IGpuBuffer meanLoss, int n, int d, int vocab)
void FusedLinearCrossEntropyIndex(IGpuBuffer hidden, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer targetIds, IGpuBuffer meanLoss, int n, int d, int vocab)
void FusedLinearGELU(IGpuBuffer input, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inFeatures, int outFeatures)
void FusedLinearGELUBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weight, IGpuBuffer preActivation, IGpuBuffer gradInput, IGpuBuffer gradWeight, IGpuBuffer gradBias, int batchSize, int inFeatures, int outFeatures)
void FusedLinearReLU(IGpuBuffer input, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inFeatures, int outFeatures)
void FusedLinearReLUBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weight, IGpuBuffer preActivation, IGpuBuffer gradInput, IGpuBuffer gradWeight, IGpuBuffer gradBias, int batchSize, int inFeatures, int outFeatures)
void FusedLinearSigmoid(IGpuBuffer input, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inFeatures, int outFeatures)
void FusedLinearSigmoidBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weight, IGpuBuffer output, IGpuBuffer gradInput, IGpuBuffer gradWeight, IGpuBuffer gradBias, int batchSize, int inFeatures, int outFeatures)
void FusedLinearSwish(IGpuBuffer input, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inFeatures, int outFeatures)
void FusedLinearSwishBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weight, IGpuBuffer preActivation, IGpuBuffer gradInput, IGpuBuffer gradWeight, IGpuBuffer gradBias, int batchSize, int inFeatures, int outFeatures)
void FusedLinearTanh(IGpuBuffer input, IGpuBuffer weight, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inFeatures, int outFeatures)
void FusedLinearTanhBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weight, IGpuBuffer output, IGpuBuffer gradInput, IGpuBuffer gradWeight, IGpuBuffer gradBias, int batchSize, int inFeatures, int outFeatures)
void FusedLoRAForward(IGpuBuffer input, IGpuBuffer baseOutput, IGpuBuffer loraA, IGpuBuffer loraB, IGpuBuffer output, int batchSize, int inputFeatures, int rank, int outputFeatures, float scaling)
void FusedSparseLinear(IGpuBuffer input, IGpuBuffer packedCsr, IGpuBuffer sparseValues, IGpuBuffer bias, IGpuBuffer output, int batchSize, int inputFeatures, int outputFeatures, int nnz, int hasBias, int activation)
void GIoULoss(IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer loss, int numBoxes)
void GIoULossBackward(IGpuBuffer gradOutput, IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer gradPredicted, int numBoxes)
void GatedDeltaNetScanForward(IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, IGpuBuffer alpha, IGpuBuffer beta, IGpuBuffer output, int batch, int seqLen, int modelDim, int numHeads, int headDim)
void Gather(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int numIndices, int featureSize)
void GaussianCovariance(IGpuBuffer rotations, IGpuBuffer scales, IGpuBuffer covariances, int numGaussians)
void GaussianNoise(IGpuBuffer output, int size, float mean, float stdDev, ulong seed)
void GeGluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int outerSize, int halfDim)
void GeGluForward(IGpuBuffer i, IGpuBuffer o, int os, int hd)
void Gelu(IGpuBuffer A, IGpuBuffer B, int size)
void GeluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void Gemm(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta)
void GemmAsync(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta, IGpuStream stream)
void GemmFp16ConvBatched(IGpuBuffer weightsFp16, IGpuBuffer colFp16, IGpuBuffer output, int m, int pixels, int k, int batch, bool halfOutput)
void GemmFp16In32fOut(IGpuBuffer aFp16, IGpuBuffer bFp16, IGpuBuffer cFp32, int m, int n, int k)
void GeneralizedBoxIou(IGpuBuffer boxesA, IGpuBuffer boxesB, IGpuBuffer output, int n, int m)
void GenerateRandomNormal(IGpuBuffer output, int size, float mean, float stdDev, ulong seed)
void GenerateRandomUniform(IGpuBuffer output, int size, float min, float max, ulong seed)
void GenerateSecureRandomUniform(IGpuBuffer output, int size, float min, float max)
void GenerateSpiralIndices(IGpuBuffer vertices, IGpuBuffer faces, IGpuBuffer visited, IGpuBuffer currentRing, IGpuBuffer nextRing, IGpuBuffer output, int numVertices, int numFaces, int spiralLength)
void GenerateStatelessDropoutMask(IGpuBuffer output, int size, uint threshold, float scale, uint seed)
void GlaScanBackward(IGpuBuffer dOut, IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, IGpuBuffer gate, IGpuBuffer dQ, IGpuBuffer dK, IGpuBuffer dV, IGpuBuffer dG, int batch, int seqLen, int modelDim, int numHeads, int headDim)
void GlaScanForward(IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, IGpuBuffer gate, IGpuBuffer output, int batch, int seqLen, int modelDim, int numHeads, int headDim)
void GlobalAvgPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int height, int width)
void GlobalAvgPool2DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int height, int width)
void GlobalMaxPool2D(IGpuBuffer input, IGpuBuffer output, IGpuBuffer indices, int batch, int channels, int height, int width)
void GlobalMaxPool2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int height, int width)
void GlobalMaxPool2DBackward(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer gradInput, int batch, int channels, int height, int width)
void GluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int outerSize, int halfDim)
void GluForward(IGpuBuffer i, IGpuBuffer o, int os, int hd)
void GreaterThan(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void GridSample(IGpuBuffer input, IGpuBuffer grid, IGpuBuffer output, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth, int paddingMode, bool alignCorners)
void GridSample2D(IGpuBuffer input, IGpuBuffer grid, IGpuBuffer output, int N, int H, int W, int C, int outH, int outW, int mode, int padding, bool alignCorners)
void GridSample2DBackwardGrid(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer grid, IGpuBuffer gradGrid, int N, int H, int W, int C, int outH, int outW, int mode, int padding, bool alignCorners)
void GridSample2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer grid, IGpuBuffer gradInput, int N, int H, int W, int C, int outH, int outW, int mode, int padding, bool alignCorners)
void GridSampleBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer grid, IGpuBuffer gradInput, IGpuBuffer gradGrid, int batch, int channels, int inHeight, int inWidth, int outHeight, int outWidth, int paddingMode, bool alignCorners)
void GridSampleBackwardGridNhwc(IGpuBuffer gradOut, IGpuBuffer input, IGpuBuffer grid, IGpuBuffer gradGrid, int batch, int h, int w, int c, int outH, int outW)
void GridSampleBackwardInputNhwc(IGpuBuffer gradOut, IGpuBuffer grid, IGpuBuffer gradIn, int batch, int h, int w, int c, int outH, int outW)
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 GroupedQueryAttention(IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer output, IGpuBuffer attentionWeights, int batch, int numQHeads, int numKVHeads, int seqQ, int seqK, int headDim, float scale, bool isCausal)
void GroupedQueryAttentionBackward(IGpuBuffer gradOutput, IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer attentionWeights, IGpuBuffer gradQuery, IGpuBuffer gradKey, IGpuBuffer gradValue, int batch, int numQHeads, int numKVHeads, int seqQ, int seqK, int headDim, float scale, int numQueriesPerKV)
void GruBackwardSequence(IGpuBuffer gradOutput, IGpuBuffer allH, IGpuBuffer cacheGates, IGpuBuffer weightsIh, IGpuBuffer weightsHh, IGpuBuffer input, IGpuBuffer gradInput, IGpuBuffer gradHInit, IGpuBuffer dHBuffer, IGpuBuffer gradWeightsIh, IGpuBuffer gradWeightsHh, IGpuBuffer gradBiasIh, IGpuBuffer gradBiasHh, int seqLen, int batch, int inputSize, int hiddenSize)
void GruCellBackward(IGpuBuffer gradH, IGpuBuffer gateR, IGpuBuffer gateZ, IGpuBuffer gateN, IGpuBuffer prevH, IGpuBuffer weightsHh, IGpuBuffer gradPrevH, IGpuBuffer gradGateR, IGpuBuffer gradGateZ, IGpuBuffer gradGateN, int batch, int hiddenSize)
void GruForwardSequence(IGpuBuffer input, IGpuBuffer hInit, IGpuBuffer weightsIh, IGpuBuffer weightsHh, IGpuBuffer biasIh, IGpuBuffer biasHh, IGpuBuffer output, IGpuBuffer hFinal, IGpuBuffer allH, IGpuBuffer cacheGates, int seqLen, int batch, int inputSize, int hiddenSize)
void GumbelSoftmax(IGpuBuffer logits, IGpuBuffer output, int outerSize, int innerSize, float temperature, ulong seed)
void Hardsigmoid(IGpuBuffer A, IGpuBuffer B, int size)
void HardsigmoidBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void Hardswish(IGpuBuffer A, IGpuBuffer B, int size)
void HardswishBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void Hardtanh(IGpuBuffer A, IGpuBuffer B, float minVal, float maxVal, int size)
void HardtanhBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, float minVal, float maxVal, int size)
void HashGridEncodeLevel(IGpuBuffer positions, IGpuBuffer hashTable, IGpuBuffer output, int numPoints, float scale, float positionOffset, int gridResolution, int tableSize, int featuresPerLevel, int levelOffset, int outputStride)
void HashGridEncodeLevelBackward(IGpuBuffer positions, IGpuBuffer outputGradient, IGpuBuffer tableGradient, int numPoints, float scale, float positionOffset, int gridResolution, int tableSize, int featuresPerLevel, int levelOffset, int outputStride)
void Hgemm(IGpuBuffer aFp16, IGpuBuffer bFp16, IGpuBuffer cFp16, int m, int n, int k)
void HierarchicalSoftmaxPaths(IGpuBuffer acts, IGpuBuffer output, int rows, int treeDepth, int numClasses)
void HingeBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void Histc(IGpuBuffer input, IGpuBuffer hist, int n, int bins, float mn, float mx)
void HistogramDD(IGpuBuffer samples, IGpuBuffer hist, IGpuBuffer bins, IGpuBuffer mins, IGpuBuffer maxs, int n, int d)
void HuberBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float delta)
void HuberLoss(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer loss, int batchSize, int numFeatures, float delta)
void HuberLossBackward(IGpuBuffer gradOutput, IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float invN, float delta)
void HyperbolicLinearBackwardBiases(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradBiases, int batchSize, int inputFeatures, int outputFeatures, float curvature)
void HyperbolicLinearBackwardInput(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weights, IGpuBuffer gradInput, int batchSize, int inputFeatures, int outputFeatures, float curvature)
void HyperbolicLinearBackwardWeights(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradWeights, int batchSize, int inputFeatures, int outputFeatures, float curvature)
void HyperbolicLinearForward(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer biases, IGpuBuffer output, int batchSize, int inputFeatures, int outputFeatures, float curvature, float epsilon)
void HyperbolicLinearForwardFused(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer biases, IGpuBuffer output, int batchSize, int inputFeatures, int outputFeatures, float curvature, float epsilon)
void IRFFT(IGpuBuffer inputReal, IGpuBuffer inputImag, IGpuBuffer output, int n)
void Im2colKNFp16(IGpuBuffer input, IGpuBuffer outputHalf, int batch, int channels, int height, int width, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW, int dilationH, int dilationW)
void ImportanceSampling(IGpuBuffer tValuesCoarse, IGpuBuffer weightsCoarse, IGpuBuffer fineTValues, int numRays, int numCoarseSamples, int numFineSamples, uint seed)
void IndexAdd(IGpuBuffer destination, IGpuBuffer indices, IGpuBuffer source, IGpuBuffer output, int outerSize, int sourceAxis, int destinationAxis, int innerSize)
void IndexSelect(IGpuBuffer input, IGpuBuffer indices, IGpuBuffer output, int numIndices, int innerSize)
void IndexSelect(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int outerSize, int sourceAxis, int indexAxis, int innerSize)
void IndexWrite(IGpuBuffer output, IGpuBuffer indices, IGpuBuffer source, float fillValue, int mode, int outerSize, int idxAxis, int innerSize, int dstAxis)
void InsertBarrier(IGpuBuffer buffer)
void InstanceNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batch, int channels, int spatialSize, float epsilon)
void InstanceNormBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gamma, IGpuBuffer saveMean, IGpuBuffer saveInvVar, IGpuBuffer gradInput, IGpuBuffer gradGamma, IGpuBuffer gradBeta, int batch, int channels, int spatialSize, float epsilon)
void Int32ToFloat(IGpuBuffer raw, IGpuBuffer output, int n)
void InterleaveComplex(IGpuBuffer real, IGpuBuffer imag, IGpuBuffer interleaved, int n)
void Interpolate2D(IGpuBuffer input, IGpuBuffer output, int N, int C, int Hin, int Win, int Hout, int Wout, int mode, bool alignCorners)
void IoULoss(IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer loss, int numBoxes)
void IoULossBackward(IGpuBuffer gradOutput, IGpuBuffer predicted, IGpuBuffer target, IGpuBuffer gradPredicted, int numBoxes)
void IotaPad(IGpuBuffer idx, int l, int p, int numRows)
void IouFamilyBackward(IGpuBuffer gradOutput, IGpuBuffer boxesA, IGpuBuffer boxesB, IGpuBuffer gradA, IGpuBuffer gradB, int n, int m, int variant)
void IsIn(IGpuBuffer elements, IGpuBuffer sortedTest, IGpuBuffer mask, int numElements, int testLen)
void IstftFromSpectrum(IGpuBuffer specRe, IGpuBuffer specIm, IGpuBuffer window, IGpuBuffer result, IGpuBuffer windowSum, int batch, int numFrames, int nFft, int hop, int outputLength, int center)
void IstftNormalize(IGpuBuffer result, IGpuBuffer windowSum, int total)
void IvfAssign(IGpuBuffer vectors, IGpuBuffer centroids, IGpuBuffer assignments, int numVectors, int numCentroids, int dim, AnnMetric metric)
void KlDivLoss(IGpuBuffer input, IGpuBuffer target, IGpuBuffer loss, int size)
void Kron2D(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int am, int an, int bp, int bq)
void L1Loss(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer loss, int batchSize, int numFeatures)
void L1LossBackward(IGpuBuffer gradOutput, IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float invN)
void LambUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size, float maxTrustRatio, bool biasCorrection)
void LarsUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer velocity, float learningRate, float momentum, float weightDecay, float trustCoeff, int size)
void LaunchFft(IGpuBuffer buffer, int batchCount, int n, bool inverse)
void LayerNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer beta, IGpuBuffer saveMean, IGpuBuffer saveInvVar, int batchSize, int normalizedSize, float epsilon)
void LayerNormBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gamma, IGpuBuffer saveMean, IGpuBuffer saveInvVar, IGpuBuffer gradInput, IGpuBuffer gradGamma, IGpuBuffer gradBeta, int batchSize, int normalizedSize, float epsilon)
void Ldexp(IGpuBuffer input, IGpuBuffer exponents, IGpuBuffer output, int size)
void LeakyRelu(IGpuBuffer A, IGpuBuffer B, float alpha, int size)
void LeakyReluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, float alpha, int size)
void Lerp(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, float t, int size)
void LessThan(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void LinalgCholesky(IGpuBuffer input, IGpuBuffer output, IGpuBuffer info, int batchCount, int n, bool upper)
void LinalgEigh(IGpuBuffer input, IGpuBuffer eigenvalues, IGpuBuffer eigenvectors, int batchCount, int n)
void LinalgLuFactor(IGpuBuffer input, IGpuBuffer output, IGpuBuffer pivots, int batchCount, int m, int n)
void LinalgQrReduced(IGpuBuffer input, IGpuBuffer q, IGpuBuffer r, int batchCount, int m, int n)
void LinspaceKernel(IGpuBuffer output, float start, float step, int size)
void LionUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, float learningRate, float beta1, float beta2, float weightDecay, int size)
void LocallyConnectedConv2D(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer bias, IGpuBuffer output, int batch, int inChannels, int inHeight, int inWidth, int outChannels, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW)
void LocallyConnectedConv2DBackwardBias(IGpuBuffer gradOutput, IGpuBuffer gradBias, int batch, int outChannels, int outHeight, int outWidth)
void LocallyConnectedConv2DBackwardInput(IGpuBuffer gradOutput, IGpuBuffer weights, IGpuBuffer gradInput, int batch, int inChannels, int inHeight, int inWidth, int outChannels, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW)
void LocallyConnectedConv2DBackwardWeights(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradWeights, int batch, int inChannels, int inHeight, int inWidth, int outChannels, int outHeight, int outWidth, int kernelH, int kernelW, int strideH, int strideW)
void Log(IGpuBuffer A, IGpuBuffer B, int size)
void Log10(IGpuBuffer A, IGpuBuffer B, int size)
void Log1P(IGpuBuffer A, IGpuBuffer B, int size)
void Log2(IGpuBuffer A, IGpuBuffer B, int size)
void LogCoshBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void LogSoftmax(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void LogSumExpAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void LogSumExpBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer lse, IGpuBuffer gradInput, int outerSize, int reduceSize)
void LogicalNot(IGpuBuffer a, IGpuBuffer output, int n)
void LogicalOp(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int mode, int n)
void LstmBackwardSequence(IGpuBuffer gradOutput, IGpuBuffer allH, IGpuBuffer allC, IGpuBuffer cacheGates, IGpuBuffer hInit, IGpuBuffer cInit, IGpuBuffer weightsIh, IGpuBuffer weightsHh, IGpuBuffer input, IGpuBuffer gradInput, IGpuBuffer gradHInit, IGpuBuffer gradCInit, IGpuBuffer gradWeightsIh, IGpuBuffer gradWeightsHh, IGpuBuffer gradBiasIh, IGpuBuffer gradBiasHh, int seqLen, int batch, int inputSize, int hiddenSize)
void LstmForwardSequence(IGpuBuffer input, IGpuBuffer hInit, IGpuBuffer cInit, IGpuBuffer weightsIh, IGpuBuffer weightsHh, IGpuBuffer biasIh, IGpuBuffer biasHh, IGpuBuffer output, IGpuBuffer hFinal, IGpuBuffer cFinal, IGpuBuffer allH, IGpuBuffer allC, IGpuBuffer cacheGates, int seqLen, int batch, int inputSize, int hiddenSize)
void MaeBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void Mamba2SsdScanForward(IGpuBuffer x, IGpuBuffer delta, IGpuBuffer aLog, IGpuBuffer bParam, IGpuBuffer cParam, IGpuBuffer dParam, IGpuBuffer output, int batch, int seqLen, int innerDim, int numHeads, int headDim, int stateDim)
void MambaSelectiveScanForward(IGpuBuffer x, IGpuBuffer delta, IGpuBuffer aLog, IGpuBuffer bParam, IGpuBuffer cParam, IGpuBuffer dParam, IGpuBuffer output, int batch, int seqLen, int innerDim, int stateDim)
void MaskedFillBackward(IGpuBuffer gradOutput, IGpuBuffer mask, IGpuBuffer gradInput, int size)
void MaskedFillKernel(IGpuBuffer i, IGpuBuffer m, IGpuBuffer o, float fv, int sz)
void MasksToBoxes(IGpuBuffer masks, IGpuBuffer output, int n, int h, int w)
void MatMulBackwardFp16Fused(IGpuBuffer gradCFp16, IGpuBuffer aFp16, IGpuBuffer bFp16, IGpuBuffer gradAOut, IGpuBuffer gradBOut, int m, int n, int k, bool gradOutHalf)
void MatMulTransposed(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta)
void MatMulTransposedA(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta)
void Max(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void MaxAxis(IGpuBuffer A, IGpuBuffer B, int outerSize, int reduceSize)
void MaxPool1D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inLength, int outLength, int kernelSize, int stride)
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 MaxPool2DBackward(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer gradInput, 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 MaxPool3D(IGpuBuffer input, IGpuBuffer output, IGpuBuffer indices, int batch, int channels, int inDepth, int inHeight, int inWidth, int outDepth, int outHeight, int outWidth, int kernelD, int kernelH, int kernelW, int strideD, int strideH, int strideW)
void MaxPool3DBackward(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer gradInput, int batch, int channels, int inDepth, int inHeight, int inWidth, int outDepth, int outHeight, int outWidth)
void MeanAxis(IGpuBuffer A, IGpuBuffer B, int outerSize, int reduceSize)
void MeasurementForward(IGpuBuffer input, IGpuBuffer output, int batchSize, int stateSize)
void MelFilterbankApply(IGpuBuffer powerSpec, IGpuBuffer melFilters, IGpuBuffer melEnergy, int totalSegBatch, int specBins, int melBins)
void MesaScanForward(IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, IGpuBuffer initialWeights, IGpuBuffer regularization, IGpuBuffer output, IGpuBuffer workWeights, IGpuBuffer covariance, int batch, int time, int model, int heads, int headDim)
void Meshgrid2DPair(IGpuBuffer source0, IGpuBuffer source1, IGpuBuffer output0, IGpuBuffer output1, int n0, int n1, bool xy)
void MfccLog1p(IGpuBuffer input, IGpuBuffer output, int n)
void Min(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void Mish(IGpuBuffer A, IGpuBuffer B, int size)
void MishBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void MobiusAdd(IGpuBuffer x, IGpuBuffer y, IGpuBuffer output, int batchSize, int dim, float curvature)
void Mode(IGpuBuffer input, IGpuBuffer output, int length)
void ModifiedHuberBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void MseBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void MseLoss(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer loss, int batchSize, int numFeatures)
void MseLossBackward(IGpuBuffer gradOutput, IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float invN)
void MuLawDecoding(IGpuBuffer input, IGpuBuffer output, int length, int qc)
void MuLawEncoding(IGpuBuffer input, IGpuBuffer output, int length, int qc)
void MultiTensorScaleByDeviceScalar(IReadOnlyList<IGpuBuffer> tensors, IReadOnlyList<int> sizes, IGpuBuffer scale)
void MultiTensorSumOfSquares(IReadOnlyList<IGpuBuffer> tensors, IReadOnlyList<int> sizes, IGpuBuffer sumOfSquares)
void Multiply(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void NadamUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer m, IGpuBuffer v, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step, int size)
void NagUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer velocity, float learningRate, float momentum, float weightDecay, int size)
void NearestNeighborUpsample(IGpuBuffer input, IGpuBuffer output, int batchChannels, int height, int width, int scaleFactor)
void NearestNeighborUpsample3D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inDepth, int inHeight, int inWidth, int scaleD, int scaleH, int scaleW)
void NearestNeighborUpsample3DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int inDepth, int inHeight, int inWidth, int scaleD, int scaleH, int scaleW)
void NearestNeighborUpsampleBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batchChannels, int height, int width, int scaleFactor)
void Negate(IGpuBuffer A, IGpuBuffer B, int size)
void NextAfter(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int size)
void NllLoss(IGpuBuffer logProbs, IGpuBuffer targets, IGpuBuffer loss, int batchSize, int numClasses)
void Nms(IGpuBuffer boxes, IGpuBuffer scores, IGpuBuffer classIds, IGpuBuffer suppressed, IGpuBuffer outputCapacity, IGpuBuffer outputCount, int length, float iouThreshold, int batched)
void Nonzero(IGpuBuffer input, IGpuBuffer strides, IGpuBuffer outputCapacity, IGpuBuffer outputCount, int length, int rank)
void NormAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void NormBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer norm, IGpuBuffer gradInput, int outerSize, int reduceSize)
void NormalizeL2(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void NormalizeProbabilities(IGpuBuffer probabilities, int batchSize, int stateSize)
void NormalizeRowsFused(IGpuBuffer input, IGpuBuffer output, int rows, int cols)
void NotEqualScalar(IGpuBuffer A, IGpuBuffer C, float scalar, int size)
void NotEqualsKernel(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int sz)
void OctonionAdd(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int count)
void OctonionLinearBackwardBiases(IGpuBuffer gradOutput, IGpuBuffer gradBiases, int batchSize, int outputFeatures)
void OctonionLinearBackwardInput(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer weights, IGpuBuffer gradInput, int batchSize, int inputFeatures, int outputFeatures)
void OctonionLinearBackwardWeights(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradWeights, int batchSize, int inputFeatures, int outputFeatures)
void OctonionLinearForward(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer biases, IGpuBuffer output, int batchSize, int inputFeatures, int outputFeatures)
void OctonionLinearForwardFusedReLU(IGpuBuffer input, IGpuBuffer weights, IGpuBuffer biases, IGpuBuffer output, int batchSize, int inputFeatures, int outputFeatures)
void OctonionMultiply(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int count)
void OneHotInt32(IGpuBuffer int32Indices, IGpuBuffer output, int rows, int depth)
void OneHotKernel(IGpuBuffer idx, IGpuBuffer o, int bs, int nc)
void OuterProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int M, int N)
void PDist(IGpuBuffer input, IGpuBuffer output, int n, int d, float p)
void PRelu(IGpuBuffer input, IGpuBuffer alpha, IGpuBuffer output, int size, int alphaSize)
void PReluBackwardAlpha(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradAlpha, int size, int alphaSize)
void PReluBackwardInput(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer alpha, IGpuBuffer gradInput, int size, int alphaSize)
void PacPhaseBinMi(IGpuBuffer thetaPhase, IGpuBuffer gammaAmp, IGpuBuffer output, int batch, int numSamples, int numGammaBands, int gammaIdx)
void Pad2D(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inH, int inW, int outH, int outW, int padTop, int padLeft, float padValue)
void Pad2DBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int inH, int inW, int outH, int outW, int padTop, int padLeft)
void Pad4D(IGpuBuffer input, IGpuBuffer output, int N, int C, int Hin, int Win, int padN0, int padN1, int padC0, int padC1, int padH0, int padH1, int padW0, int padW1, int mode, float padValue)
void PairwiseDistance(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int m, int n, int dim)
void PairwiseDistanceSquared(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int m, int n, int dim)
void PairwiseIou(IGpuBuffer boxes, IGpuBuffer iou, int n)
void Parity210CdistL2(IGpuBuffer x1, IGpuBuffer x2, IGpuBuffer output, int n, int m, int d)
void Parity210ClampMinMax(IGpuBuffer input, IGpuBuffer lo, IGpuBuffer hi, IGpuBuffer output, int size, bool hasLo, bool hasHi)
void Parity210Copysign(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210CosineSimilarityLast(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int n, int d, float eps)
void Parity210CumMax(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210CumMin(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210CumProd(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210CumSum(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210DiagEmbed(IGpuBuffer input, IGpuBuffer output, int batchSize, int diagLen, int matSize, int offset)
void Parity210Digamma(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210Erfc(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210Erfinv(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210FlipAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210FloatPower(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210Fmod(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210Hypot(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210I0(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210I0e(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210I1(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210I1e(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210IndexAdd(IGpuBuffer output, IGpuBuffer indicesInt32, IGpuBuffer source, int outerSize, int dstAxis, int innerSize, int idxLen)
void Parity210IndexCopy(IGpuBuffer output, IGpuBuffer indicesInt32, IGpuBuffer source, int outerSize, int dstAxis, int innerSize, int idxLen)
void Parity210IndexFill(IGpuBuffer output, IGpuBuffer indicesInt32, float fillValue, int outerSize, int dstAxis, int innerSize, int idxLen)
void Parity210IsFinite(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210IsInf(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210IsNan(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210Lgamma(IGpuBuffer input, IGpuBuffer output, int size)
void Parity210LogAddExp(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210LogAddExp2(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210LogCumSumExp(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize)
void Parity210MaskedScatter(IGpuBuffer output, IGpuBuffer maskInt8, IGpuBuffer prefixSumInt32, IGpuBuffer source, int total)
void Parity210NanToNum(IGpuBuffer input, IGpuBuffer output, int size, float nanVal, float posInfVal, float negInfVal)
void Parity210Remainder(IGpuBuffer a, IGpuBuffer b, IGpuBuffer o, int size)
void Parity210Roll1D(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize, int shift)
void Parity210TakeAlongDim(IGpuBuffer input, IGpuBuffer indicesInt32, IGpuBuffer output, int outerSize, int idxAxis, int innerSize, int srcAxis)
void Parity210TakeLinear(IGpuBuffer input, IGpuBuffer indicesInt32, IGpuBuffer output, int outSize, int inputLinearLen)
void Parity210Tril(IGpuBuffer input, IGpuBuffer output, int batchSize, int rows, int cols, int diagonal)
void Parity210Triu(IGpuBuffer input, IGpuBuffer output, int batchSize, int rows, int cols, int diagonal)
void Parity210Xlog1py(IGpuBuffer x, IGpuBuffer y, IGpuBuffer o, int size)
void Parity210Xlogy(IGpuBuffer x, IGpuBuffer y, IGpuBuffer o, int size)
void Permute(IGpuBuffer input, IGpuBuffer output, int[] shape, int[] permutation)
void PhaseVocoder(IGpuBuffer mag, IGpuBuffer phase, IGpuBuffer newMag, IGpuBuffer newPhase, int leading, int numFrames, int numFreqs, int outFrames, float rate)
void PixelShuffle(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int inH, int inW, int scale)
void PixelShuffleBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int batch, int channels, int inH, int inW, int scale)
void PoincareDistance(IGpuBuffer x, IGpuBuffer y, IGpuBuffer output, int batchSize, int dim, float curvature)
void PoincareExpMap(IGpuBuffer basePoint, IGpuBuffer tangentVec, IGpuBuffer output, int batchSize, int dim, float curvature)
void PoincareProject(IGpuBuffer input, IGpuBuffer output, int batchSize, int dim, float curvature, float epsilon)
void PoissonBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void PolarToComplex(IGpuBuffer magnitude, IGpuBuffer phase, IGpuBuffer real, IGpuBuffer imag, int n)
void Polygamma(IGpuBuffer x, IGpuBuffer output, int n, int size)
void PowScalar(IGpuBuffer i, IGpuBuffer o, float ex, int sz)
void Power(IGpuBuffer A, IGpuBuffer B, float exponent, int size)
void PowerToDb(IGpuBuffer power, IGpuBuffer db, int n, float refValue, float minDb)
void PqAdcScan(IGpuBuffer codes, IGpuBuffer tables, IGpuBuffer distances, int numQueries, int numCodes, int m, int ksub)
void PqComputeDistanceTables(IGpuBuffer queries, IGpuBuffer codebooks, IGpuBuffer tables, int numQueries, int m, int ksub, int dsub, AnnMetric metric)
void ProductAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void ProjectGaussians3DTo2D(IGpuBuffer means3D, IGpuBuffer covariances3D, IGpuBuffer means2D, IGpuBuffer covariances2D, IGpuBuffer depths, IGpuBuffer visible, int numGaussians, int covarianceStride, int imageWidth, int imageHeight, float v00, float v01, float v02, float v03, float v10, float v11, float v12, float v13, float v20, float v21, float v22, float v23, float p00, float p11)
void ProximalL1Update(IGpuBuffer param, IGpuBuffer gradient, float learningRate, float l1Strength, int size)
void PsRoIAlign(IGpuBuffer input, IGpuBuffer boxes, IGpuBuffer output, int N, int C, int H, int W, int K, int outH, int outW, int outputChannels, float spatialScale, int samplingRatio)
void PsRoIPool(IGpuBuffer input, IGpuBuffer boxes, IGpuBuffer output, int N, int C, int H, int W, int K, int outH, int outW, int outputChannels, float spatialScale)
void QuantileBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float quantile)
void QuantumMeasurement(IGpuBuffer realPart, IGpuBuffer imagPart, IGpuBuffer probabilities, int batchSize, int stateSize)
void QuantumRotation(IGpuBuffer stateReal, IGpuBuffer stateImag, IGpuBuffer outReal, IGpuBuffer outImag, IGpuBuffer angles, int numQubits, int batchSize)
void RFFT(IGpuBuffer input, IGpuBuffer outputReal, IGpuBuffer outputImag, int n)
void RRelu(IGpuBuffer input, IGpuBuffer noise, IGpuBuffer output, int size)
void RReluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer noise, IGpuBuffer gradInput, int size)
void RbfForward(IGpuBuffer input, IGpuBuffer centers, IGpuBuffer epsilons, IGpuBuffer output, int batchSize, int numCenters, int inputDim)
void ReGluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int outerSize, int halfDim)
void ReGluForward(IGpuBuffer i, IGpuBuffer o, int os, int hd)
void Reciprocal(IGpuBuffer A, IGpuBuffer B, int size)
void ReciprocalBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void RecordEvent(IGpuEvent gpuEvent, IGpuStream stream)
void RectSlice(IGpuBuffer full, IGpuBuffer slice, int[] fullShape, int[] start, int[] length, bool scatter)
void ReduceLogSumExp(IGpuBuffer input, IGpuBuffer output, float maxVal, int size)
void ReduceLogVariance(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void ReduceLogVarianceBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer means, IGpuBuffer variances, IGpuBuffer gradInput, int outerSize, int reduceSize)
void ReduceMaxBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer maxValues, IGpuBuffer gradInput, int outerSize, int reduceSize)
void ReduceMaxMagnitude(IGpuBuffer input, IGpuBuffer output, int size)
void ReduceMean(IGpuBuffer i, IGpuBuffer o, int sz)
void ReduceMinMagnitude(IGpuBuffer input, IGpuBuffer output, int size)
void ReduceNormL2(IGpuBuffer input, IGpuBuffer output, int size)
void ReduceProduct(IGpuBuffer input, IGpuBuffer output, int size)
void ReduceSumBackward(IGpuBuffer gradOutput, IGpuBuffer gradInput, int outerSize, int reduceSize)
void ReduceSumOfSquares(IGpuBuffer input, IGpuBuffer output, int size)
void ReduceVarianceBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer means, IGpuBuffer gradInput, int outerSize, int reduceSize)
void ReflectPad1d(IGpuBuffer input, IGpuBuffer output, int batch, int l, int lp, int pad)
void Relu(IGpuBuffer A, IGpuBuffer B, int size)
void Relu6(IGpuBuffer A, IGpuBuffer B, int size)
void Relu6Backward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void ReluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void ReluDerivative(IGpuBuffer input, IGpuBuffer output, int size)
void RepeatElements(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize, int repeats)
void Resample(IGpuBuffer input, IGpuBuffer output, int leading, int inLen, int outLen, int up, int down, int halfWidth)
void RgLruScanForward(IGpuBuffer value, IGpuBuffer recGate, IGpuBuffer inpGate, IGpuBuffer decay, IGpuBuffer output, int batch, int seqLen, int recDim)
void RmsNorm(IGpuBuffer input, IGpuBuffer output, IGpuBuffer gamma, IGpuBuffer saveRms, int batchSize, int normalizedSize, float epsilon)
void RmsNormBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gamma, IGpuBuffer saveRms, IGpuBuffer gradInput, IGpuBuffer gradGamma, int batchSize, int normalizedSize, float epsilon)
void RmspropUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer squaredAvg, float learningRate, float rho, float epsilon, float weightDecay, int size)
void RoIAlign(IGpuBuffer input, IGpuBuffer boxes, IGpuBuffer output, int N, int C, int H, int W, int K, int outH, int outW, float spatialScale, int samplingRatio, bool aligned)
void RoIPool(IGpuBuffer input, IGpuBuffer boxes, IGpuBuffer output, int N, int C, int H, int W, int K, int outH, int outW, float spatialScale)
void RopeInterleaved(IGpuBuffer input, IGpuBuffer cos, IGpuBuffer sin, IGpuBuffer output, int rows, int headDim, int seqLen, int startPosition)
void Round(IGpuBuffer A, IGpuBuffer B, int size)
void RoutedDiagonalSsmScanForward(IGpuBuffer input, IGpuBuffer activeMask, IGpuBuffer transition, IGpuBuffer inputMap, IGpuBuffer outputMap, IGpuBuffer skip, IGpuBuffer output, IGpuBuffer stateScratch, int batch, int time, int model, int experts, int state)
void RsqrtKernel(IGpuBuffer input, IGpuBuffer output, int size)
void Rwkv4WkvForward(IGpuBuffer r, IGpuBuffer k, IGpuBuffer v, IGpuBuffer timeDecay, IGpuBuffer timeFirst, IGpuBuffer output, int batch, int seqLen, int modelDim)
void Rwkv7Forward(IGpuBuffer r, IGpuBuffer kappa, IGpuBuffer kTilde, IGpuBuffer v, IGpuBuffer decayLogit, IGpuBuffer iclRate, IGpuBuffer output, IGpuBuffer sbuf, int batch, int seqLen, int modelDim, int numHeads, int headDim)
void SampleRaysWithOccupancy(IGpuBuffer rayOrigins, IGpuBuffer rayDirections, IGpuBuffer occupancyBitfield, IGpuBuffer positions, IGpuBuffer directions, IGpuBuffer validMask, IGpuBuffer tValues, int numRays, int occupancyWordCount, int gridSize, int maxSamples, float minX, float minY, float minZ, float maxX, float maxY, float maxZ, float nearBound, float farBound)
void ScalarMinusTensor(IGpuBuffer input, IGpuBuffer output, float scalar, int size)
void Scale(IGpuBuffer A, IGpuBuffer B, float scalar, int size)
void ScaleByDeviceScalar(IGpuBuffer buffer, IGpuBuffer 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 ScaledDotProductAttentionBackward(IGpuBuffer gradOutput, IGpuBuffer query, IGpuBuffer key, IGpuBuffer value, IGpuBuffer attentionWeights, IGpuBuffer gradQuery, IGpuBuffer gradKey, IGpuBuffer gradValue, int batch, int numHeads, int seqQ, int seqK, int headDim, float scale, bool isCausal)
void ScatterAdd(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer destination, int sourceSize, int destSize)
void ScatterAddBackward(IGpuBuffer gradDestination, IGpuBuffer indices, IGpuBuffer gradSource, int numIndices, int featureSize)
void ScatterAddBackwardRows(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer gradSource, int sourceRows, int innerSize, int outputRows)
void ScatterAddEdges(IGpuBuffer input, IGpuBuffer sourceIndices, IGpuBuffer targetIndices, IGpuBuffer edgeValues, IGpuBuffer output, int numNodes, int numEdges, int features)
void ScatterAddRows(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int sourceRows, int innerSize, int outputRows)
void ScatterMaxBackwardRows(IGpuBuffer gradOutput, IGpuBuffer argmax, IGpuBuffer gradSource, int sourceRows, int innerSize, int outputRows)
void ScatterMaxRows(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int srcDimSize, int innerSize, int outDimSize)
void ScatterMaxWithArgmaxRows(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, IGpuBuffer argmax, int sourceRows, int innerSize, int outputRows)
void ScatterMean(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, IGpuBuffer counts, int sourceSize, int outputSize, int featureSize)
void ScatterMeanBackwardRows(IGpuBuffer gradOutput, IGpuBuffer indices, IGpuBuffer counts, IGpuBuffer gradSource, int sourceRows, int innerSize, int outputRows)
void ScatterMeanRows(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int srcDimSize, int innerSize, int outDimSize)
void ScatterMeanRowsWithCounts(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, IGpuBuffer counts, int sourceRows, int innerSize, int outputRows)
void ScatterPositions(IGpuBuffer output, int srcDim, int inner, int n)
void ScatterReduce(IGpuBuffer output, IGpuBuffer source, IGpuBuffer index, int outerSize, int srcDim, int dstDim, int innerSize, int mode)
void ScatterSoftmaxBackwardRows(IGpuBuffer gradOutput, IGpuBuffer output, IGpuBuffer indices, IGpuBuffer gradSource, int sourceRows, int innerSize, int numGroups)
void ScatterSoftmaxRows(IGpuBuffer source, IGpuBuffer indices, IGpuBuffer output, int sourceRows, int innerSize, int numGroups)
void SearchSorted(IGpuBuffer sortedSeq, IGpuBuffer values, IGpuBuffer output, int seqLen, int numValues, int right)
void Selu(IGpuBuffer A, IGpuBuffer B, float alpha, float scale, int size)
void SeluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, float alpha, float scale, int size)
void SetSliceAxis(IGpuBuffer output, IGpuBuffer values, int outerSize, int axisSize, int stride, int index)
void SetSliceLastAxis(IGpuBuffer output, IGpuBuffer values, int outerSize, int outputInnerSize, int start, int sliceSize)
void SgdMomentumUpdate(IGpuBuffer param, IGpuBuffer gradient, IGpuBuffer velocity, float learningRate, float momentum, float weightDecay, int size)
void SgdMultiTensorUpdate(IReadOnlyList<IGpuBuffer> parameters, IReadOnlyList<IGpuBuffer> gradients, IReadOnlyList<int> sizes, float learningRate, float weightDecay)
void SgdMultiTensorUpdateGuarded(IReadOnlyList<IGpuBuffer> parameters, IReadOnlyList<IGpuBuffer> gradients, IReadOnlyList<int> sizes, float learningRate, float weightDecay, IGpuBuffer finiteGuard)
void SgdUpdate(IGpuBuffer param, IGpuBuffer gradient, float learningRate, float weightDecay, int size)
void ShiftedDiff(IGpuBuffer x, IGpuBuffer mask, int n)
void Sigmoid(IGpuBuffer A, IGpuBuffer B, int size)
void SigmoidBackward(IGpuBuffer gradOutput, IGpuBuffer output, IGpuBuffer gradInput, int size)
void SigmoidDerivative(IGpuBuffer sigmoidOutput, IGpuBuffer output, int size)
void Sign(IGpuBuffer A, IGpuBuffer B, int size)
void Silu(IGpuBuffer A, IGpuBuffer B, int size)
void SiluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void Sin(IGpuBuffer A, IGpuBuffer B, int size)
void SinCosKernel(IGpuBuffer input, IGpuBuffer sinOutput, IGpuBuffer cosOutput, int size)
void Sinh(IGpuBuffer A, IGpuBuffer B, int size)
void SliceAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int stride, int index)
void SliceLastAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int inputInnerSize, int start, int sliceSize)
void SmoothL1Backward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size, float beta)
void Softmax(IGpuBuffer A, IGpuBuffer B, int batchSize, int features)
void SoftmaxBackward(IGpuBuffer gradOutput, IGpuBuffer output, IGpuBuffer gradInput, int batchSize, int features)
void SoftmaxRows(IGpuBuffer input, IGpuBuffer output, int rows, int cols)
void Softplus(IGpuBuffer A, IGpuBuffer B, int size)
void SoftplusBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void SparseAdadeltaUpdate(IGpuBuffer param, IGpuBuffer accumGrad, IGpuBuffer accumUpdate, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float rho, float epsilon, float weightDecay)
void SparseAdagradUpdate(IGpuBuffer param, IGpuBuffer accumulatedGrad, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float epsilon, float weightDecay)
void SparseAdamUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer v, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void SparseAdamWUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer v, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void SparseAdamaxUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer u, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void SparseAmsgradUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer v, IGpuBuffer vMax, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void SparseFtrlUpdate(IGpuBuffer param, IGpuBuffer z, IGpuBuffer n, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float l1Reg, float l2Reg, float beta)
void SparseGemm(IGpuBuffer sparseAValues, IGpuBuffer sparseAIndices, IGpuBuffer B, IGpuBuffer C, int M, int N, int K, float alpha, float beta)
void SparseLionUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float weightDecay)
void SparseNadamUpdate(IGpuBuffer param, IGpuBuffer m, IGpuBuffer v, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float beta1, float beta2, float epsilon, float weightDecay, int step)
void SparseNagUpdate(IGpuBuffer param, IGpuBuffer velocity, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float momentum, float weightDecay)
void SparseProximalL1Update(IGpuBuffer param, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float l1Strength)
void SparseRmspropUpdate(IGpuBuffer param, IGpuBuffer squaredAvg, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float rho, float epsilon, float weightDecay)
void SparseSgdMomentumUpdate(IGpuBuffer param, IGpuBuffer velocity, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float momentum, float weightDecay)
void SparseSgdUpdate(IGpuBuffer param, IGpuBuffer sparseIndices, IGpuBuffer sparseValues, int nnz, float learningRate, float weightDecay)
void Sparsemax(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void SpectralFilter(IGpuBuffer inputReal, IGpuBuffer filterReal, IGpuBuffer filterImag, IGpuBuffer outputReal, int batch, int height, int width, int filterSliceCount)
void SphericalHarmonics(IGpuBuffer shCoefficients, IGpuBuffer viewDirections, IGpuBuffer output, int numPoints, int basisCount, int numChannels, int degree, int broadcastDir)
void SphericalHarmonicsBackward(IGpuBuffer shCoefficients, IGpuBuffer viewDirections, IGpuBuffer outputGradient, IGpuBuffer shGrad, int numPoints, int basisCount, int numChannels, int degree, int broadcastDir)
void SphericalSoftmax(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void SpiralConv(IGpuBuffer vertexFeatures, IGpuBuffer spiralIndices, IGpuBuffer weights, IGpuBuffer biases, IGpuBuffer output, int v, int inC, int spiralLength, int outC)
void SpiralConvBackwardInput(IGpuBuffer gradOutput, IGpuBuffer spiralIndices, IGpuBuffer weights, IGpuBuffer gradVertexFeatures, int v, int inC, int spiralLength, int outC)
void SpiralConvBackwardWeights(IGpuBuffer gradOutput, IGpuBuffer vertexFeatures, IGpuBuffer spiralIndices, IGpuBuffer gradWeights, int v, int inC, int spiralLength, int outC)
void SplitComplexAdd(IGpuBuffer aR, IGpuBuffer aI, IGpuBuffer bR, IGpuBuffer bI, IGpuBuffer oR, IGpuBuffer oI, int n)
void SplitComplexConjugate(IGpuBuffer iR, IGpuBuffer iI, IGpuBuffer oR, IGpuBuffer oI, int n)
void SplitComplexCrossSpectral(IGpuBuffer xR, IGpuBuffer xI, IGpuBuffer yR, IGpuBuffer yI, IGpuBuffer oR, IGpuBuffer oI, int n)
void SplitComplexFromPolar(IGpuBuffer m, IGpuBuffer p, IGpuBuffer oR, IGpuBuffer oI, int n)
void SplitComplexHrrBindAccumulate(IGpuBuffer keyCodeReal, IGpuBuffer keyCodeImag, IGpuBuffer valPermCodeReal, IGpuBuffer valPermCodeImag, IGpuBuffer keyIds, IGpuBuffer valIds, IGpuBuffer memoryReal, IGpuBuffer memoryImag, int N, int D)
void SplitComplexMagnitude(IGpuBuffer iR, IGpuBuffer iI, IGpuBuffer o, int n)
void SplitComplexMagnitudeSquared(IGpuBuffer iR, IGpuBuffer iI, IGpuBuffer o, int n)
void SplitComplexMultiply(IGpuBuffer aR, IGpuBuffer aI, IGpuBuffer bR, IGpuBuffer bI, IGpuBuffer oR, IGpuBuffer oI, int n)
void SplitComplexPhase(IGpuBuffer iR, IGpuBuffer iI, IGpuBuffer o, int n)
void SplitComplexPhaseCoherenceDecode(IGpuBuffer codesReal, IGpuBuffer codesImag, IGpuBuffer queryReal, IGpuBuffer queryImag, IGpuBuffer outScores, int V, int D)
void SplitComplexScale(IGpuBuffer iR, IGpuBuffer iI, IGpuBuffer oR, IGpuBuffer oI, float s, int n)
void SplitComplexTopK(IGpuBuffer inReal, IGpuBuffer inImag, IGpuBuffer outReal, IGpuBuffer outImag, int n, int k)
void SplitComplexUnitPhaseCodebook(IGpuBuffer outReal, IGpuBuffer outImag, int seed, int V, int D, bool kPsk, int k)
void Sqrt(IGpuBuffer A, IGpuBuffer B, int size)
void SquaredHingeBackward(IGpuBuffer predictions, IGpuBuffer targets, IGpuBuffer gradInput, int size)
void Squash(IGpuBuffer input, IGpuBuffer output, int numCapsules, int capsuleDim, float epsilon)
void SquashBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int numCapsules, int capsuleDim, float epsilon)
void Stack2(IGpuBuffer a, IGpuBuffer b, IGpuBuffer output, int size)
void StdAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void StdBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer mean, IGpuBuffer std, IGpuBuffer gradInput, int outerSize, int reduceSize)
void StdpUpdate(IGpuBuffer weights, IGpuBuffer preTrace, IGpuBuffer postTrace, IGpuBuffer preSpike, IGpuBuffer postSpike, float ltpRate, float ltdRate, float homeostasisRate, float minWeight, float maxWeight, int numPre, int numPost)
void StftBackwardFrames(IGpuBuffer grad, IGpuBuffer magnitude, IGpuBuffer phase, IGpuBuffer frameGrad, int batch, int nFft, int numFrames, int numFreqs, int phaseTarget)
void StftBackwardSignal(IGpuBuffer frameGrad, IGpuBuffer window, IGpuBuffer gradSignal, int batch, int length, int pad, int nFft, int hop, int numFrames)
void StftBackwardWindow(IGpuBuffer frameGrad, IGpuBuffer signal, IGpuBuffer gradWindow, int batch, int length, int pad, int nFft, int hop, int numFrames)
void StftMagPhase(IGpuBuffer padded, IGpuBuffer window, IGpuBuffer mag, IGpuBuffer phase, int batch, int lp, int nFft, int hop, int numFrames, int numFreqs)
void StreamWaitEvent(IGpuStream stream, IGpuEvent gpuEvent)
void StridedDotProduct(IGpuBuffer a, IGpuBuffer b, IGpuBuffer result, int aSize, int bSize, int bOffset, int bStride)
void StridedGather(IGpuBuffer src, IGpuBuffer dst, int offset, int stride, int count)
void StridedScatter(IGpuBuffer src, IGpuBuffer dst, int offset, int stride, int count)
void SubScalar(IGpuBuffer i, IGpuBuffer o, float sc, int sz)
void Subtract(IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void SumAxis(IGpuBuffer A, IGpuBuffer B, int outerSize, int reduceSize)
void SwiGluBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int outerSize, int halfDim)
void SwiGluForward(IGpuBuffer i, IGpuBuffer o, int os, int hd)
void Swish(IGpuBuffer A, IGpuBuffer B, int size)
void SwishBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, int size)
void Synchronize()
void SynchronizeStream(IGpuStream stream)
void TakeAlongDim(IGpuBuffer input, IGpuBuffer indices, IGpuBuffer output, int outerSize, int axisOut, int innerSize, int axisIn)
void Tan(IGpuBuffer A, IGpuBuffer B, int size)
void Tanh(IGpuBuffer A, IGpuBuffer B, int size)
void TanhBackward(IGpuBuffer gradOutput, IGpuBuffer output, IGpuBuffer gradInput, int size)
void TanhDerivative(IGpuBuffer tanhOutput, IGpuBuffer output, int size)
void TaylorSoftmax(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize)
void Threshold(IGpuBuffer input, IGpuBuffer output, float threshold, float value, int size)
void ThresholdBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer gradInput, float threshold, int size)
void TileAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int axisSize, int innerSize, int repeats)
void TileBatch(IGpuBuffer input, IGpuBuffer output, int repeats, int innerSize)
void TileLastAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int innerSize, int repeats)
void TopK(IGpuBuffer A, IGpuBuffer values, IGpuBuffer indices, int outerSize, int reduceSize, int k, bool sorted)
void Transpose(IGpuBuffer A, IGpuBuffer B, int rows, int cols)
void TriangularMask(IGpuBuffer output, int rows, int cols, int diagonal, float maskValue)
void TrilinearInterpolate(IGpuBuffer grid, IGpuBuffer positions, IGpuBuffer output, int d, int h, int w, int c, int p, float upperEps)
void TrilinearInterpolateBackward(IGpuBuffer gradOutput, IGpuBuffer positions, IGpuBuffer gradGrid, int d, int h, int w, int c, int p, float upperEps)
void TripletLossBackward(IGpuBuffer anchor, IGpuBuffer positive, IGpuBuffer negative, IGpuBuffer gradAnchor, IGpuBuffer gradPositive, IGpuBuffer gradNegative, int batchSize, int embeddingDim, float margin)
void TrispectrumGather(IGpuBuffer specReal, IGpuBuffer specImag, IGpuBuffer outReal, IGpuBuffer outImag, int maxF1, int maxF2, int maxF3)
void Truncate(IGpuBuffer A, IGpuBuffer B, int size)
void Unfold(IGpuBuffer input, IGpuBuffer output, int batch, int channels, int height, int width, int kernelH, int kernelW, int strideH, int strideW, int padH, int padW)
void Unfold(IGpuBuffer src, IGpuBuffer dst, int outerSize, int dimSize, int innerSize, int nWindows, int size, int step)
void UniformMeshLaplacian(IGpuBuffer faces, IGpuBuffer output, int numFaces, int numVertices)
void UniqueConsecutive(IGpuBuffer input, IGpuBuffer outputCapacity, IGpuBuffer outputCount, int length)
void UniqueConsecutiveWithInfo(IGpuBuffer input, IGpuBuffer outputValues, IGpuBuffer outputInverse, IGpuBuffer outputCounts, IGpuBuffer outputCount, int length, bool returnInverse, bool returnCounts)
void UniqueSortedWithInfo(IGpuBuffer sortedInput, IGpuBuffer sortedOriginalIndices, IGpuBuffer outputValues, IGpuBuffer outputInverse, IGpuBuffer outputCounts, IGpuBuffer outputCount, int length, bool returnInverse, bool returnCounts)
void UpdateTraces(IGpuBuffer traces, IGpuBuffer spikes, IGpuBuffer input, float decay, float threshold, int size)
void UploadBufferAsync(ReadOnlySpan<float> data, IGpuBuffer buffer, IGpuStream stream)
void UploadBufferAsync(float[] data, IGpuBuffer buffer, IGpuStream stream)
void UploadBufferInPlace(float[] data, IGpuBuffer buffer)
void UploadByteBuffer(IGpuBuffer buffer, byte[] data)
void UploadIntBufferInPlace(int[] data, IGpuBuffer buffer)
void UploadToBuffer(IGpuBuffer buffer, float[] source)
void VarAxis(IGpuBuffer A, IGpuBuffer B, int outerSize, int reduceSize)
void VarAxis(IGpuBuffer A, IGpuBuffer mean, IGpuBuffer variance, int outerSize, int reduceSize)
void VarBackward(IGpuBuffer gradOutput, IGpuBuffer input, IGpuBuffer mean, IGpuBuffer gradInput, int outerSize, int reduceSize)
void VarianceAxis(IGpuBuffer input, IGpuBuffer output, int outerSize, int reduceSize)
void Where(IGpuBuffer condition, IGpuBuffer A, IGpuBuffer B, IGpuBuffer C, int size)
void WhereBackward(IGpuBuffer gradOutput, IGpuBuffer condition, IGpuBuffer gradX, IGpuBuffer gradY, int size)
void WidebandLogBinPool(IGpuBuffer magBuf, IGpuBuffer output, int totalSegBatch, int fftSize, int numBins, int usable)
void XLstmScanForward(IGpuBuffer q, IGpuBuffer k, IGpuBuffer v, IGpuBuffer iGate, IGpuBuffer fGate, IGpuBuffer oGate, IGpuBuffer output, int batch, int seqLen, int modelDim, int numHeads, int headDim)
void Zeta(IGpuBuffer x, IGpuBuffer q, IGpuBuffer output, int size)