Classes supported by all material representations.

We can index them in a random-access manner, window them in constant time, and use them as targets for a computation.

In particular, delayed arrays are not material as we cannot use them as targets for a computation.

Arrays where the elements that are automatically layed out into some efficient runtime representation.

The implementation uses type families to chose unboxed representations for all elements that can be unboxed. In particular: arrays of unboxed tuples are represented as tuples of unboxed arrays, and nested arrays are represented using a segment descriptor and a single single flat vector containing all the elements.

Layout for dense Foreign arrays.

UNSAFE: Indexing into raw material arrays is not bounds checked. You may want to wrap this with a Checked layout as well.

O(1). Wrap a ForeignPtr as an array.

O(1). Unwrap a ForeignPtr from an array.

O(1). Convert a ByteString to an foreign Array.

O(1). Convert a foreign Vector to a ByteString.

O(1). Convert a storable Vector to a foreign Array

O(1). Convert a foreign array to a storable Vector.

Nested array represented as a flat array of elements, and a segment descriptor that describes how the elements are partitioned into the sub-arrays. Using this representation for multidimentional arrays is significantly more efficient than using a boxed array of arrays, as there is no need to allocate the sub-arrays individually in the heap.

With a nested type like: Array N (Array N (Array U Int)), the concrete representation consists of five flat unboxed vectors: two for each of the segment descriptors associated with each level of nesting, and one unboxed vector to hold all the integer elements.

UNSAFE: Indexing into raw material arrays is not bounds checked. You may want to wrap this with a Checked layout as well.

O(size src) Convert some lists to a nested array.

O(size src) Convert a triply nested list to a triply nested array.

Layout an array as flat vector of boxed elements.

UNSAFE: Indexing into raw material arrays is not bounds checked. You may want to wrap this with a Checked layout as well.

O(1). Wrap a boxed vector as an array.

O(1). Unwrap a boxed vector from an array.

Layout an array as a flat vector of unboxed elements.

This is the most efficient representation for numerical data.

The implementation uses Data.Vector.Unboxed which picks an efficient, specialised representation for every element type. In particular, unboxed vectors of pairs are represented as pairs of unboxed vectors.

UNSAFE: Indexing into raw material arrays is not bounds checked. You may want to wrap this with a Checked layout as well.

O(1). Wrap an unboxed vector as an array.

O(1). Unwrap an unboxed vector from an array.

Apply a function to all the elements of a doubly nested array, preserving the nesting structure.

Scan through an array from front to back. For pairs of successive elements, drop the second one when the given predicate returns true.

This function can be used to remove duplicates from a sorted array.

TODO: generalise to other array types.

O(1). Produce a nested array by taking slices from some array of elements.

This is a constant time operation, as the representation for nested vectors just wraps the starts, lengths and elements vectors.

Take a desired number of segments, and array of key value pairs where the key is the segment number. Partition the values into the stated number of segments, discarding values where the key falls outside the given range.

• This function operates by first allocating a buffer of size (segs * len src) and filling it with a default value. Both the worst case runtime and memory use will be poor for a large number of destination segments.

TODO: we need the pre-init because otherwise unused values in the elems array are undefined. We could avoid this by copying out the used elements after the partition loop finishes. Use a segmented extract function. This would also remove the dependency on the Elt class.

Like partition but use the provided function to compute the segment number for each element.

Like partition but use the provided function to compute the segment number for each element. The function is given the index of the each element, along with the element itself.

Segmented concatenation. Concatenate triply nested vector, producing a doubly nested vector.

• Unlike the plain concat function, this operation is performed entirely on the segment descriptors of the nested arrays, and does not require the inner array elements to be copied.

> import Data.Repa.Nice
> nice $ concats $ fromListss U [["red", "green", "blue"], ["grey", "white"], [], ["black"]]
["red","green","blue","grey","white","black"]

O(len src). Given predicates which detect the start and end of a segment, split an vector into the indicated segments.

O(len src). Given a terminating value, split an vector into segments.

The result segments do not include the terminator.

> import Data.Repa.Nice
> nice $ segmentOn (== ' ') (fromList U "fresh   fried fish  ") 
["fresh "," "," ","fried ","fish "," "]

O(len src). Like segment, but cut the source array twice.

O(len src). Given field and row terminating values, split an array into rows and fields.

For each segment of a nested array, trim elements off the start and end of the segment that match the given predicate.

For each segment of a nested array, trim elements off the end of the segment that match the given predicate.

For each segment of a nested array, trim elements off the start of the segment that match the given predicate.

Ragged transpose of a triply nested array.

• This operation is performed entirely on the segment descriptors of the nested arrays, and does not require the inner array elements to be copied.