Please sign in to comment.
Change allocation mechanism to use GHC internals
This commit changes the Data.Vector.SIMD.Mutable.MVector allocation mechanism. It no longer uses _mm_malloc bindings through FFI, but the internal newAlignedPinnedByteArray# function and some trickery. Most likely highly non-portable. But brings SIMD vector speed on-par with Storable. Could be improved a little more by not using newForeignPtr_ but GHC.ForeignPtr.PlainPtr if this were exported.
- Loading branch information...
Showing with 18 additions and 26 deletions.