# BNNSFilterCreateFusedLayer(_:_:_:_:) Returns a new fused layer. ``` func BNNSFilterCreateFusedLayer(_ number_of_fused_filters: Int, _ filter_type: UnsafePointer, _ layer_params: UnsafeMutablePointer, _ filter_params: UnsafePointer?) -> BNNSFilter? ``` ## Parameters `number_of_fused_filters` The number of fused filters. `filter_type` Array of fused filter types. `layer_params` Array of fused layer parameter pointers. `filter_params` The filter runtime parameters. ## Discussion Use this function to create an `N` layer fused filter that processes input as: `input → filter0 → filter1 →` … `→ filter N-1 → output`. The first `N - 1` filters must have [`BNNSActivationFunctionIdentity`](/documentation/Accelerate/BNNSActivationFunctionIdentity) activation function. For the first `N - 2` filters, each filter’s output descriptor and its subsequent filter’s input descriptor must have the same size, stride, and data type. [`BNNSFilterCreateFusedLayer(_:_:_:_:)`](/documentation/Accelerate/BNNSFilterCreateFusedLayer(_:_:_:_:)) supports two filters. The first filter must be either [`BNNSConvolution`](/documentation/Accelerate/BNNSConvolution) or [`BNNSFullyConnected`](/documentation/Accelerate/BNNSFullyConnected). The second filter must be one of [`BNNSBatchNorm`](/documentation/Accelerate/BNNSBatchNorm), [`BNNSInstanceNorm`](/documentation/Accelerate/BNNSInstanceNorm), [`BNNSLayerNorm`](/documentation/Accelerate/BNNSLayerNorm), or [`BNNSGroupNorm`](/documentation/Accelerate/BNNSGroupNorm). The following code shows how you pass [`BNNSLayerParametersConvolution`](/documentation/Accelerate/BNNSLayerParametersConvolution) and [`BNNSLayerParametersNormalization`](/documentation/Accelerate/BNNSLayerParametersNormalization) structures to [`BNNSFilterCreateFusedLayer(_:_:_:_:)`](/documentation/Accelerate/BNNSFilterCreateFusedLayer(_:_:_:_:)) to create a fused convolution-layer normalization filter: ```swift // `input`, `output`, and `weights` are single-precision arrays. // `inDescriptor` and `outDescriptor` are the `BNNSNDArrayDescriptor` structures // for the fused layer's input and output. weights.withUnsafeMutableBytes { weightsPtr in let weightsDescriptor = BNNSNDArrayDescriptor(flags: flags, layout: BNNSDataLayoutConvolutionWeightsOIHW, size: (3, 3, 1, 1, 0, 0, 0, 0), stride: (0, 0, 0, 0, 0, 0, 0, 0), data: weightsPtr.baseAddress, data_type: .float, table_data: nil, table_data_type: .float, data_scale: 0, data_bias: 0) let convolutionParameters = BNNSLayerParametersConvolution(i_desc: inDescriptor, w_desc: weightsDescriptor, o_desc: outDescriptor, bias: BNNSNDArrayDescriptor(), activation: .identity, x_stride: 1, y_stride: 1, x_dilation_stride: 0, y_dilation_stride: 0, x_padding: 0, y_padding: 0, groups: 1, pad: (0, 0, 0, 0)) let normalizationParameters = BNNSLayerParametersNormalization(i_desc: outDescriptor, o_desc: outDescriptor, beta_desc: inDescriptor, gamma_desc: inDescriptor, moving_mean_desc: BNNSNDArrayDescriptor(), moving_variance_desc: BNNSNDArrayDescriptor(), momentum: 0, epsilon: 1, activation: .identity, num_groups: 1, normalization_axis: 0) withUnsafeBytes(of: convolutionParameters) { convolutionPtr in withUnsafeBytes(of: normalizationParameters) { normalizationPtr in var parameters = [convolutionPtr.baseAddress!, normalizationPtr.baseAddress!] let filter = BNNSFilterCreateFusedLayer(2, [BNNSConvolution, BNNSLayerNorm], ¶meters, nil) defer { BNNSFilterDestroy(filter) } BNNSFusedFilterApplyBatch(filter, 1, input, input.count, &output, output.count, false) } } } ``` --- Copyright © 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)