Bitcast Conversion

Reinterprets the bits of an array as a different data type without modifying the underlying data. If the target type is narrower, an extra trailing axis is added; if wider, the last axis is consumed.

Usage

nv_bitcast_convert(x, dtype)

Arguments

x

(arrayish)
One input. Can be any data type except bool. An R value materializes at its default data type.

dtype

(character(1) | DataType)
Any target data type except bool. One of the same bit width as the input's leaves the shape unchanged; a narrower one adds a leading axis holding the pieces; a wider one consumes the first axis, whose size must equal the ratio of the two widths. The pieces of one element are therefore adjacent in the column-major element order nv_flatten() reads, and a narrowing conversion lays the bytes out the way as_raw() writes them.

Value

(arrayish)
Has the given dtype, and the shape described under dtype.

See also

prim_bitcast_convert(), which this is an alias of, and nv_convert() for value-preserving type conversion.

Examples

# the bits of one i32 reread as four i8, in a new trailing axis
x <- nv_array(1L, dtype = "i32")
nv_bitcast_convert(x, dtype = "i8")
#> AnvlArray
#>  1
#>  0
#>  0
#>  0
#> [ CPUi8{4,1} ]