[ofa-general] Re: [PATCH 2 of 2] IB/mlx4: shrinking WQE

[Roland Dreier]

Thanks, applied -- we might as well use this feature I guess.
Although I'm not convinced this is really that useful; the only use I
see for it would be speeding up IPoIB with S/G and checksum offload,
when we have to size send WQEs for the worst case but most packets are
smaller.  But Eli's latest work seems to use selective signaling for
the send queue, so this change doesn't actually help.

I also made a few small changes to places like this:

 > +		ind += DIV_ROUND_UP(size * 16, 1 << qp->sq.wqe_shift);

For this case, the compiler can generate better code if we use "1U <<"
to get an unsigned divisor; in that case it can use a shift instead of
needing to use a divide operation to keep the sign correct.

 - R.