On Thu, Oct 12, 2023 at 03:11:06PM -0700, Darrick J. Wong wrote:
> Hmm, so you want to go from:
>
> union xfs_rtword_ondisk *start, *end, *b;
>
> start = b = xfs_rbmblock_wordptr(bp, startword);
> end = xfs_rbmblock_wordptr(bp, endword);
>
> while (b < end) {
> somevalue = xfs_rtbitmap_getword(mp, b);
> somevalue |= somemask;
> xfs_rtbitmap_setword(mp, b, somevalue);
> b++;
> }
>
> xfs_trans_log_buf(tp, bp, start - bp->b_addr, b - bp->b_addr);
>
> to something like:
>
> for (word = startword; word <= endword; word++) {
> somevalue = xfs_rtbitmap_getword(mp, b);
> somevalue |= somemask;
> xfs_rtbitmap_setword(mp, bp, word, somevalue);
> }
> xfs_rtbitmap_log_buf(tp, bp, startword, endword);
Yes. (although xfs_rtbitmap_log_buf can't just take the words directly
of course, and the xfs_rtbitmap_getword needs word and not the now
not existing b).
> I think that could be done with relatively little churn, though it's
> unfortunate that the second version does 2x(shift + addition) each time
> it goes through the loop instead of the pointer increment that the first
> version does.
I don't really think it matter compared to all the other overhead,
and it keeps a much nicer API.
