Edward Kmett <ekmett@gmail.com> writes:

> Currently if you try to use a DoubleX4# and don't have AVX2 turned on, it
> deliberately crashes out during code generation, no?

I very well be missing something, but I don't believe this is true. This
program compiles just fine with merely -fllvm -msse,

    {-# LANGUAGE MagicHash #-}
    {-# LANGUAGE UnboxedTuples #-}
    module Hi where
    import GHC.Prim
    import GHC.Float

    addIt :: DoubleX4# -> DoubleX4# -> DoubleX4#
    addIt x y = plusDoubleX4# x y
    {-# NOINLINE addIt #-}

It produces the following assembler,,

    movupd 0x10(%rbp),%xmm0
    movupd 0x0(%rbp),%xmm1
    movupd 0x30(%rbp),%xmm2
    movupd 0x20(%rbp),%xmm3
    addpd  %xmm1,%xmm3
    addpd  %xmm0,%xmm2
    movupd %xmm2,0x30(%rbp)
    movupd %xmm3,0x20(%rbp)
    mov    0x40(%rbp),%rax
    lea    0x20(%rbp),%rbp
    jmpq   *%rax

The reason for this is that the LLVM code generator just blindly
translates DoubleX4# to LLVM's <4 x double> type. The LLVM code
generator then does whatever it can to produce the code we ask of it,
even if the target doesn't have support for this vector variety.

Cheers,

- Ben