Run a Function at Given Data Types

with_dtypes() wraps f into a function that works at the data types dtypes names: on each call every array argument of a category dtypes names is converted to that data type, the defaults (see default_dtypes()) are set to the float / int entries for the duration of the call, and every returned array of a named category is converted as well.

nv_add_f64 <- with_dtypes(nv_add, c(float = "f64"))

A category dtypes does not name is left alone, in the arguments, in the body and in the result.

Usage

with_dtypes(f, dtypes)

Arguments

f

(function)
The function to wrap.

dtypes

(named character() | named list())
A mapping of the data type categories (float, int and uint) to data types, e.g. c(float = "f64", int = "i64"). Each may be a string or a DataType. A category it does not name is left as it is.

Value

A function with the same arguments as f, a JitFunction if f was one.

Details

Note that f itself can also change the default data types, which overrides the defaults configured by with_dtypes().

See also

default_dtypes(), local_default_dtypes()

Examples

add_f64 <- with_dtypes(nv_add, c(float = "f64"))
# an `f32` argument is converted, and the result comes back as `f64`
dtype(add_f64(nv_array(1, dtype = "f32"), 2.5))
#> <f64>
# A category that is not named is untouched
dtype(add_f64(nv_array(1L, dtype = "i32"), 2L))
#> <i32>
# `uint` is converted too, but sets no default
dtype(with_dtypes(nv_add, c(uint = "ui32"))(nv_array(1L, dtype = "ui8"), 2L))
#> <ui32>