xfs: convert xfarray insertion sort to heapsort using scratchpad memory
In the previous patch, we created a very basic quicksort implementation for xfile arrays. While the use of an alternate sorting algorithm to avoid quicksort recursion on very small subsets reduces the runtime modestly, we could do better than a load and store-heavy insertion sort, particularly since each load and store requires a page mapping lookup in the xfile. For a small increase in kernel memory requirements, we could instead bulk load the xfarray records into memory, use the kernel's existing heapsort implementation to sort the records, and bulk store the memory buffer back into the xfile. On the author's computer, this reduces the runtime by about 5% on a 500,000 element array. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Kent Overstreet <kent.overstreet@linux.dev> Reviewed-by: Dave Chinner <dchinner@redhat.com>
Showing
Please register or sign in to comment