lwip-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lwip-devel] bug entry #27352 code changes


From: Leon Woestenberg
Subject: Re: [lwip-devel] bug entry #27352 code changes
Date: Mon, 13 Jun 2011 22:21:10 +0200

Hello,

On Mon, Jun 13, 2011 at 8:36 PM, Bill Auerbach
<address@hidden> wrote:
> I spent months characterizing and optimizing lwIP - I wouldn't have made all
> of the suggestions and patches if any of them degraded performance.
>
> Perhaps the flip side is to show us where and/or why you think any of the
> changes are dubious and I'll recheck my work.
>
I see copying introduced from packed to unpacked structures which
introduces extra cycles/fetches. In my experience this overhead does
not outweigh the overhead of the processor working on packed
structures (packed in this case meaning we sometimes access a 32-bit
word which has non-natural alignment, but never worse than 16-bit
alignment.)

I'm not claiming I have counter benchmarks yet (on NIOS II for example).

On other projects I work on changes to optimizations are fine if
proven by a benchmark that demonstrated the worst case and average
performance is both equal or better. I was just asking if these were
available.

Regards,
-- 
Leon



reply via email to

[Prev in Thread] Current Thread [Next in Thread]