also test simd_select_bitmask on arrays for less than 8 elements

[RalfJung]

No description provided.

[RalfJung]

@bors r+

[bors]

📌 Commit 4a7cfac has been approved by RalfJung

It is now in the queue for this repository.

[bors]

⌛ Testing commit 4a7cfac with merge ee7516c...

[bors]

💔 Test failed - checks-actions

[RalfJung]

@workingjubilee @calebzulawski this code is causing a Miri UB error when run on a big-endian target:

    let bitmask = u8x4::from_array([0b00001101, 0, 0, 0]);
    assert_eq!(
        mask32x4::from_bitmask_vector(bitmask),
        mask32x4::from_array([true, false, true, true]),
    );

It seems to be calling simd_select_bitmask with 10110000. My interpretation was that even on big-endian, it is the N least significant bits that carry the mask when N<8, so this is wrong? I think it should be 00001011.

Currently what it does seems inconsistent with simd_bitmask.

This is basically rust-lang/portable-simd#267 again, but for simd_select_bitmask rather than simd_bitmask.

[RalfJung]

Filed an issue: rust-lang/portable-simd#379

I think what Miri does here is right, or at least it is consistent with simd_bitmask and with rust-lang/portable-simd#267. So I'll land this.

@bors r+

[bors]

📌 Commit 2e66bec has been approved by RalfJung

It is now in the queue for this repository.

[bors]

⌛ Testing commit 2e66bec with merge bad98ce...

[bors]

☀️ Test successful - checks-actions
Approved by: RalfJung
Pushing bad98ce to master...