Skip to contents

Primitive Generic Reduce

Source: R/primitives.R
prim_reduce.Rd

Reduces an array along the specified axes using a user-supplied associative reducer. reductor and init must satisfy the constraints in the "Associativity Requirement" section below.

Usage

prim_reduce(x, init, axes, drop = TRUE, reductor)

Arguments

x

(arrayish)
The array to reduce. Can be any data type. x and init must have the same data type. An R value among them assumes the data type of the others when it is in its data type category, and its default data type when none of them has one.

init

(arrayish)
Scalar initial value, with no axes. Must be the neutral element w.r.t. reductor, and shares x's data type – see x.

axes

(integer())
Axes to reduce over. Negative values count from the end, i.e. -1 refers to the last axis.

drop

(logical(1))
If TRUE (default) the reduced axes are removed; if FALSE they are kept with size 1.

reductor

(function)
Binary reducer producing a scalar of the same data type as x. Its two arguments are passed by position, so they may carry any names. Must be associative (see "Associativity Requirement").

Value

(arrayish)
Has the input's data type. The shape is the input's with the reduced axes removed (drop = TRUE) or set to 1 (drop = FALSE).

Associativity Requirement

The order in which reductor is applied across the reduction window is implementation-defined. If the reductor is not associative, the result is ill-defined. Furthermore, init must be the neutral element for this reductor. Because arithmetic in a float data type is non-associative, the output of the reduction can differ between backends (GPU, CPU), even if the underlying mathematical function (like +) is associative.

Implemented Rules

  • stablehlo

StableHLO

Lowers to hlo_reduce(), specified under reduce. The body is reductor.

See also

prim_reduce_sum(), prim_reduce_max()

Examples

# `init` shares `x`'s data type, and the reduced axis disappears
x <- nv_array(c(1, 2, 3, 4))
prim_reduce(x, init = nv_scalar(0), axes = 1L, reductor = prim_add)
#> AnvlArray
#>  10
#> [ CPUf32{} ] 
prim_reduce(x, init = nv_scalar(1), axes = 1L, reductor = prim_mul)
#> AnvlArray
#>  24
#> [ CPUf32{} ]