AVX2 SAD has wrong return type

[the8472]

https://github.com/rust-lang-nursery/simd/blob/539247c63cd59dc3aac973c02a86845674ee3b6a/src/x86/avx2.rs#L45

fn x86_mm256_sad_epu8(x: u8x32, y: u8x32) -> u8x32
should be
fn x86_mm256_sad_epu8(x: u8x32, y: u8x32) -> u64x4

The intel intrinsics docs state that the result is packed into 4x 64bit values.

I tried fixing it myself but the compiler then complains

error[E0442]: intrinsic return value has wrong type: found vector with length 4, expected length 32
  --> src/x86/avx2.rs:45:5
   |
45 |     fn x86_mm256_sad_epu8(x: u8x32, y: u8x32) -> u64x4;
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[BurntSushi]

The intrinsic definition in LLVM is:

  def int_x86_avx2_psad_bw : GCCBuiltin<"__builtin_ia32_psadbw256">,
              Intrinsic<[llvm_v4i64_ty], [llvm_v32i8_ty,
                         llvm_v32i8_ty], [IntrNoMem, Commutative]>;

Which appears consistent with the Intel intrinsic docs. The next place I'd look is in rustc to see if there's a bug in the corresponding platform-intrinsic. Indeed, the definition appears incorrect, so I think this needs to be filed against rust-lang/rust.

[the8472]

Filed rust-lang/rust#43439