[lkml]   [1996]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: TCP/IP Checksumming

On Tue, 26 Nov 1996, Heiko Schroeder wrote:

> Hmmm. The values below show quite the opposite. It is not that
> difficult to fold the carries back, and what is more important, it
> is done only once, not for every word accessed.
> Here are the measured values for some of the functions used:
> csum_partial: sum = 1a8a, min cycles 1406
> csum_simple: sum = 1a8a, min cycles 6846
> csum_opt: sum = 1a8a, min cycles 2455
> csum_opt32: sum = 1a8a, min cycles 1434
> csum_c: sum = 1a8a, min cycles 2626
> csum_partial is the original Linux version.
> csum_simple is the simple function using lodsw and loop, csum_opt is the
> function that was best in the MS-DOS tests. csum_opt32 is csum_opt
> changed to use 32-Bit memory accesses. And for comparison, there is
> a C version of the checksumming algorithm in csum_c.

I don't understand. These numbers are not at all like I get in 32-bit
mode, flat-model, writing results to the UART, with no interrupts.

I'll get back to you later. I have to take a work-break.

Dick Johnson
Richard B. Johnson
Project Engineer
Analogic Corporation
Voice : (508) 977-3000 ext. 3754
Fax : (508) 532-6097
Modem : (508) 977-6870
Ftp :
Email :,
Penguin : Linux version 2.1.13 on an i586 machine.
Warning : It's hard to remain at the trailing edge of technology.

 \ /
  Last update: 2005-03-22 13:38    [W:0.063 / U:1.148 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site