portable-simd: add test for non-power-of-2 bitmask #3655
Conversation
|
Yes, I think that's expected (we reverse the bitmasks in std::simd) FYI, we actually just removed use of that variation of the intrinsic in rust-lang/portable-simd#423 |
The intrinsic is still called in other places though. In which sense is it "removed"? |
Great, thanks. :) |
portable-simd: add test for non-power-of-2 bitmask `@calebzulawski` is that the intended behavior? Specifically for arrays, the bitmask `[1, 0, 0, 1, 0, 0, 1, 0, 1, 0]` becomes - `[0b01001001, 0b01]` on little endian - `[0b10010010, 0b10]` on big endian
We still use the intrinsic, but currently (not yet synced to rust-lang/rust) we only use integer return types |
|
Ah, this seems to fail on big-endian...
I see. If the array support ever gets removed entirely, please ping me. |
|
The docs say
On big-endian, the bitlist Maybe that explains why you had trouble with this in the non-power-of-2 PR? |
|
@bors r+ |
|
☀️ Test successful - checks-actions |
@calebzulawski is that the intended behavior? Specifically for arrays, the bitmask
[1, 0, 0, 1, 0, 0, 1, 0, 1, 0]becomes[0b01001001, 0b01]on little endian[0b10010010, 0b10]on big endian