Subject: Re: build aborts in regress/lib/libpthread/sem
To: None <port-dreamcast@netbsd.org>
From: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
List: port-dreamcast
Date: 03/01/2003 02:43:49
In article <yf97kbrkm06.fsf@mc.pp.se>
marcus@mc.pp.se wrote:

> > Ok, then how about the attached patch (with some cosmetic changes)?
> > 
> > I'm not sure if unrolling multi/region ops is really worth, though.
> > (it's optional in this patch for now)
> 
> Looks fine.  Some measurements on the unrolling thing might be nice; I
> suspect the entire transfer would complete quite fast anyway in most
> cases.

It looks the unrolling version is a bit slower.
(Maybe due to calling intr suspend/resume functions)
I didn't check difference in interrupt response, though.

> Typically a transfer would be 1500 bytes (MTU) or less, right?

I think so (ETHER_MAX_LEN) if we only have network drivers..
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp