Use an environment variable to request MKL

.gitignore

@@ -8,3 +8,4 @@ deps/src
 deps/builds
 docs/build
 docs/site
+.build_settings

README.md

@@ -4,6 +4,15 @@ This package provides Julia bindings to the [FFTW](http://www.fftw.org/) library
 fast Fourier transforms, as well as functionality useful for signal processing.
 These functions were formerly a part of Base Julia.
 
+Users with a build of Julia based on Intel's Math Kernel Library (MKL) can take use MKL
+for FFTs by setting an environment variable `JULIA_FFTW_PROVIDER` to `MKL` and running
+`Pkg.build("FFTW")`.
+Setting this environment variable only needs to be done for the first build of the package;
+after that, the package will remember to use MKL when building and updating.
+Note however that MKL provides only a subset of the functionality provided by FFTW. See
+Intel's [documentation](https://software.intel.com/en-us/mkl-developer-reference-c-using-fftw3-wrappers)
+for more information about potential differences or gaps in functionality.
+
 The FFTW library will be downloaded on versions of Julia where it is no longer distributed
 as part of Julia.
 Note that FFTW is licensed under GPLv2 or higher (see

deps/build.jl

@@ -1,7 +1,17 @@
-# If BLAS was compiled with MKL, we want to use MKL for FFTs as well. Thus
-# we have to do this little dance to get around having to use BinDeps for
-# a library that's already linked to Julia.
-if Base.BLAS.vendor() === :mkl
+# If BLAS was compiled with MKL and the user wants MKL-based FFTs, we'll oblige.
+# In that case, we have to do this little dance to get around having to use BinDeps
+# for a library that's already linked to Julia.
+settings = joinpath(@__DIR__, "..", ".build_settings")
+if haskey(ENV, "JULIA_FFTW_PROVIDER")
+    provider = ENV["JULIA_FFTW_PROVIDER"]
+    open(f -> println(f, provider), settings, "w")
+elseif isfile(settings)
+    provider = readchomp(settings)
+else
+    provider = "FFTW"
+    open(f -> println(f, provider), settings, "w")
+end
+if provider == "MKL" && Base.BLAS.vendor() === :mkl
     mklpath = Libdl.dlpath("libmkl_rt")
     depsfile = joinpath(@__DIR__, "deps.jl")
     isfile(depsfile) && rm(depsfile, force=true)
@@ -16,6 +26,10 @@ if Base.BLAS.vendor() === :mkl
             const libfftwf = "$mklpath"
         """)
     end
+elseif provider == "MKL"
+    error("MKL build requested for FFTW but Julia was not built with MKL.\n",
+          "To fix this, set ENV[\"JULIA_FFTW_PROVIDER\"] = \"FFTW\" and \n",
+          "rerun Pkg.build(\"FFTW\").")
 else
     include("build_fftw.jl")
 end

docs/src/index.md

@@ -14,6 +14,15 @@ Pkg.add("FFTW")
 
 from the Julia REPL.
 
+Users with a build of Julia based on Intel's Math Kernel Library (MKL) can take use MKL
+for FFTs by setting an environment variable `JULIA_FFTW_PROVIDER` to `MKL` and running
+`Pkg.build("FFTW")`.
+Setting this environment variable only needs to be done for the first build of the package;
+after that, the package will remember to use MKL when building and updating.
+Note however that MKL provides only a subset of the functionality provided by FFTW. See
+Intel's [documentation](https://software.intel.com/en-us/mkl-developer-reference-c-using-fftw3-wrappers)
+for more information about potential differences or gaps in functionality.
+
 ## Note
 
 These functions were formerly part of Base Julia.

test/runtests.jl

@@ -2,7 +2,12 @@
 using FFTW
 using FFTW: fftw_vendor
 using AbstractFFTs: Plan, plan_inv
-using Base.Test
+
+if isdefined(Base, :Test) && !Base.isdeprecated(Base, :Test)
+    using Base.Test
+else
+    using Test
+end
 
 # Base Julia issue #19892
 # (test this first to make sure it happens before set_num_threads)