howmany: howmany

Description

How many rows/columns of a matrix can be stored for a given memory size.

Usage

howmany(x, nrow, ncol, out.type = "full", representation = "dense", ...,
  sparsity = 0.05, type = "double", intsize = 4, unit.names = "short")
howmany.par(x, nrow, ncol, out.type = "full", cores = 1, par = "row", ...,
  type = "double", intsize = 4, ICTXT = 0, bldim = c(4, 4))

Arguments

The size of a matrix stored as a memuse class object.

nrow, ncol

Number of (global) rows/columns of the matrix.

out.type

If the full dimensions or a reduced representation should be returned (see Details section below). Options are "full" and "approximate" (with partial matching).

representation

The kind of storage the object would be in, i.e. "dense" or "sparse".

...

Additional arguments.

sparsity

The proportion of sparsity of the matrix if representation="sparse"

type

"double" or "int"; the storage type of the data matrix. If you don't know the type, it is probably stored as a double, so the default value will suffice.

intsize

The size (in bytes) of an integer. Default is 4, but this is platform dependent.

unit.names

string; control for whether the unit names should be printed out or their abbreviation should be used. Options are "long" and "short", respectively. Case is ignored.

cores

The number of cores.

par

The type of parallel distribution.

ICTXT

BLACS ICTXT value.

bldim

Blocking dimension for 2d block-cyclic distribution.

Value

howmany() returns a numeric pair, the dimensions of a matrix.

howmany.par() returns a list (the global and local dimensions), each of which is a numeric pair.

Details

This function provides the maximum dimension of an unallocated, dense, in-core, numeric matrix of known byte size. For example, it will show the largest possible square matrix which is 16 GiB (46340x46340).

If the both nrow and ncol are missing (blank inputs), then the largest square matrix will be returned. If one of nrow or ncol is supplied and the other is missing, then the non-supplied argument (nrow or ncol) will be determined according to the supplied one. If both arguments are supplied, an error is produced --- you probably meant to use howmany().

If out.type="approximate", then a reduced representation of the dimensions will be returned. For example, the reduced representation of the number 1234567890 would be "1.2b", since this number is basically 1.2 billion. Not super useful, but kind of cute, and it arguably enhances readability when fishing for a ballpark figure.

Examples

Run this code

# NOT RUN {
x <- mu(1, "gib")

# largest square matrix that's 1 GiB
howmany(x)
# same, but ballpark figure
howmany(mu(1, "gib"), out.type="approx")
# }
# NOT RUN {
# }