BNNSGather(_:_:_:_:_:)
Gathers the elements of a tensor along a single axis.
Declaration
func BNNSGather(_ axis: Int, _ input: UnsafePointer<BNNSNDArrayDescriptor>, _ indices: UnsafePointer<BNNSNDArrayDescriptor>, _ output: UnsafeMutablePointer<BNNSNDArrayDescriptor>, _ filter_params: UnsafePointer<BNNSFilterParameters>?) -> Int32Parameters
- axis:
The axis along which the operation gathers the indices.
- input:
A pointer to the input descriptor.
- indices:
A pointer to the indices descriptor.
- output:
A pointer to the output descriptor.
- filter_params:
The filter runtime parameters.
Discussion
Use BNNSGather(_:_:_:_:_:) to gather elements — that you specify by index — into an output tensor.
In the simplest case, use BNNSGather(_:_:_:_:_:) to gather elements from a 1D vector with indices defined as a 1D vector. The following code gathers the four elements at indices [1, 3, 7, 5]:
let values: [Float] = [10, 20, 30, 40, 50, 60, 70, 80]
var inputDescriptor = BNNSNDArrayDescriptor.allocate(
initializingFrom: values,
shape: .vector(values.count))
let indices: [Int32] = [1, 3, 7, 5]
var indicesDescriptor = BNNSNDArrayDescriptor.allocate(
initializingFrom: indices,
shape: .vector(indices.count))
var outputDescriptor = BNNSNDArrayDescriptor.allocateUninitialized(
scalarType: Float.self,
shape: indicesDescriptor.shape)
let error = BNNSGather(0,
&inputDescriptor,
&indicesDescriptor,
&outputDescriptor,
nil)
On return, outputDescriptor contains the values [20.0, 40.0, 80.0, 60.0].
BNNSGather(_:_:_:_:_:) supports gathering from a tensor with two or more dimensions using indices defined as a 1D vector. In this case, the indices correspond to the values along an entire axis and the input and output shapes must match.
The following code generates a 3 x 4 matrix from the rows of a 3 x 3 matrix:
let values: [Float] = [10, 20, 30,
40, 50, 60,
70, 80, 90]
var inputDescriptor = BNNSNDArrayDescriptor.allocate(
initializingFrom: values,
shape: .matrixRowMajor(3, 3))
let indices: [Int32] = [0, 1, // Elements `0, 1` from row `0` = `10, 20`
2, 0, // Elements `2, 0` from row `1` = `60, 40`
1, 1] // Elements `1, 1` from row `2` = `80, 80`
var indicesDescriptor = BNNSNDArrayDescriptor.allocate(
initializingFrom: indices,
shape: .matrixRowMajor(2, 3))
var outputDescriptor = BNNSNDArrayDescriptor.allocateUninitialized(
scalarType: Float.self,
shape: indicesDescriptor.shape)
let error = BNNSGather(1, // axis
&inputDescriptor,
&indicesDescriptor,
&outputDescriptor,
nil)On return, outputDescriptor contains the following values:
[ 10.0, 20.0,
60.0, 40.0,
80.0, 80.0 ]The function returns an error if any of the indices are out of range.
See Also
Gather and scatter operations
Calculating the dominant colors in an imagegather(input:indices:output:axis:filterParameters:)gatherND(input:indices:output:filterParameters:)scatter(input:indices:output:axis:reductionFunction:filterParameters:)scatterND(input:indices:output:reductionFunction:filterParameters:)BNNSGatherND(_:_:_:_:)BNNSScatter(_:_:_:_:_:_:)BNNSScatterND(_:_:_:_:_:)