This looks like a straightforward oversight. I am building my pure haskell LU solver under FreeBSD. I use the -fllvm compiler option because I need extra numerical performance.

When I build, I get (this is the output from GitLab CI, which is in turn running 'stack build')

Running with gitlab-runner 11.2.0 (35e8515d)
  on stacker.18clay.com a2111f63
Using Shell executor...
Running on stacker.18clay.com...
Fetching changes...
Removing .stack-work/
HEAD is now at f8fe3b4 Use llvm60 on FreeBSD.
From https://gitlab.18clay.com/software/luSolve
   f8fe3b4..c993a26  streamlined -> origin/streamlined
Checking out c993a262 as streamlined...
Skipping Git submodules setup
$ stack build --pedantic
luSolve-0.5: configure (lib)
Configuring luSolve-0.5...
luSolve-0.5: build (lib)
Preprocessing library for luSolve-0.5..
Building library for luSolve-0.5..
[1 of 1] Compiling Numeric.LinearAlgebra.LUSolve ( src/Numeric/LinearAlgebra/LUSolve.hs, .stack-work/dist/x86_64-freebsd/Cabal-2.4.0.1/build/Numeric/LinearAlgebra/LUSolve.o )
ghc: panic! (the 'impossible' happened)
  (GHC version 8.6.1 for x86_64-portbld-freebsd):
	Failed to lookup the datalayout for x86_64-unknown-freebsd; available targets: ["i386-unknown-windows","i686-unknown-windows","x86_64-unknown-windows","arm-unknown-linux-gnueabihf","armv6-unknown-linux-gnueabihf","armv6l-unknown-linux-gnueabihf","armv7-unknown-linux-gnueabihf","armv7a-unknown-linux-gnueabi","armv7l-unknown-linux-gnueabihf","aarch64-unknown-linux-gnu","aarch64-unknown-linux","i386-unknown-linux-gnu","i386-unknown-linux","x86_64-unknown-linux-gnu","x86_64-unknown-linux","armv7-unknown-linux-androideabi","aarch64-unknown-linux-android","powerpc64le-unknown-linux","amd64-portbld-freebsd","arm-unknown-nto-qnx-eabi","i386-apple-darwin","x86_64-apple-darwin","armv7-apple-ios","aarch64-apple-ios","i386-apple-ios","x86_64-apple-ios","aarch64-unknown-freebsd","armv6-unknown-freebsd-gnueabihf","armv7-unknown-freebsd-gnueabihf"]
CallStack (from HasCallStack):
  error, called at compiler/llvmGen/LlvmCodeGen.hs:96:24 in ghc:LlvmCodeGen

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug


--  While building custom Setup.hs for package luSolve-0.5 using:
      /home/gitlab-runner/.stack/setup-exe-cache/x86_64-freebsd/Cabal-simple_mPHDZzAJ_2.4.0.1_ghc-8.6.1 --builddir=.stack-work/dist/x86_64-freebsd/Cabal-2.4.0.1 build lib:luSolve --ghc-options " -ddump-hi -ddump-to-file"
    Process exited with code: ExitFailure 1
ERROR: Job failed: exit status 1

The bug seems simple. The compiler is looking for the datalayout for x86_64-unknown-freebsd; but what is available is amd64-portbld-freebsd. Is this just a simple naming error? The stack resolver was the cutting-edge nightly-2018-10-6; if it is a problem in their packaging rather than ghc itself I will file a report with them (though ghc did panic here, and as the breath ebbed from its ASTs and light faded from the code generator, it pleaded that a bug report be filed).

I've attached the .cabal file in case it helps.