dummy-link

InplaceOps

Convenient macros for in-place matrix operations in Julia

Readme

InplaceOps.jl

Build Status

Pkg status v0.4 Pkg status v0.5 Pkg status v0.6

InplaceOps.jl is a Julia package that provides macros that enable a simple syntax for performing in-place (i.e. overwriting) array operations (At_mul_B! and the like, as well as broadcast!) using standard arithmetic operators (*, \, ...; and their broadcasting equivalents .*, ...).

Overwriting an array used in the operation

The macros @in1! and @in2! will overwrite the entries of the first and second arguments respectively:

julia> using InplaceOps

julia> X = randn(5,5);

julia> C = cholfact(X'X)
Cholesky{Float64} with factor:
5x5 Triangular{Float64}:
 2.98062  0.357635  -1.42933  -1.40386     0.371712
 0.0      2.00525   -1.48687   4.4738e-5  -1.66761 
 0.0      0.0        1.37296   1.69426     0.430224
 0.0      0.0        0.0       0.5942      1.93128 
 0.0      0.0        0.0       0.0         1.07208 

julia> v = randn(5)
5-element Array{Float64,1}:
  0.633849 
 -1.92691  
 -0.817877 
  0.608644 
 -0.0069844

julia> pointer(v)
Ptr{Float64} @0x00007fcb313e3f70

julia> u = @in2! C \ v
5-element Array{Float64,1}:
   2.88607
 -47.6234 
 -51.754  
  43.7515 
 -10.5208 

julia> pointer(u)
Ptr{Float64} @0x00007fcb313e3f70

Note that it is not always possible to do operations in-place.

julia> @in2! X * v
ERROR: no method A_mul_B!(Array{Float64,2}, Array{Float64,1})
 in mul! at /Users/simon/.julia/v0.3/InplaceOps/src/InplaceOps.jl:40

Overwriting a different array

If you want to put the output into an array that is not used internally, you can use the @into! macro:

julia> u = Array(Float64,5);

julia> @into! u = X * v
5-element Array{Float64,1}:
  0.919228
 -5.40522 
  9.94378 
 -1.80869 
  5.51495 

Warnings

You should generally not try to assign to an object that appears elsewhere (unless you really know what you're doing). For example,

julia> X = randn(5,5)
5x5 Array{Float64,2}:
 -0.53897   -0.139062  -0.769589   1.35453     0.326885
  0.270242   1.24454   -0.700411   0.256821    0.225506
 -0.503335  -1.09109    1.14986    0.750931    2.58097 
  0.153684   1.33869   -1.07673   -1.55332    -0.596944
  0.996492  -1.12004    0.915189  -0.0312484   0.104354

julia> @into! X = X*X
5x5 Array{Float64,2}:
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0

At the moment, .+ and .* operators are only parsed as binary operators (x.+ y), so n-ary operators (x .+ y .+ z) cannot be computed in one step (see #7368).

First Commit

05/22/2014

Last Touched

16 days ago

Commits

21 commits