[Pixman] [PATCH 3/3] Rev2 of patch: AVX2 versions of OVER and ROVER operators.

[Chris Wilson]

Quoting Raghuveer Devulapalli (2019-01-17 01:00:00)
> +static force_inline void
> +core_combine_over_u_avx2_no_mask (uint32_t *        pd,
> +                                 const uint32_t*    ps,
> +                                 int                w)
> +{
> +    uint32_t s, d;
> +
> +    /* Align dst on a 16-byte boundary */
> +    while (w && ((uintptr_t)pd & 31))

32-byte.

You can reuse the comment from _mask.

/* We operate on 256-bit chunks, aligned to a 32-byte boundary. */

Explains both the alignment phase and main loop.
-Chris