[Bug 1443] PowerPC in_cksum() requires optimization

[bugzilla-daemon at rtems.org]

https://www.rtems.org/bugzilla/show_bug.cgi?id=1443

--- Comment #11 from strauman at slac.stanford.edu 2011-09-01 07:19:16 CDT ---
(In reply to comment #10)
> Created attachment 1315 [details]
> Header file derived from current FreeBSD
> 
> This version uses no inline assembler at all and works with -O2 and -O0. 

I must be missing something. This file is packed with inline assembly
(and the Wasabi copyright notice is also present). Doesn't seem to
be all that different from the netbsd version posted by Peter.

However, it's better than what we currently have since it does
not span assembly across multiple __asm__ statements.

However, it still doesn't inform the compiler that the
inline assembly accesses the memory region it checks.
I'd add a "memory" clobber.

-- Till

> A
> simple performance test on a MPC5566 running at 132 MHz shows this (FTP
> transfer to /dev/null and from /dev/zero):
> 
> A = FreeBSD derived with -O2
> B = FreeBSD derived with -O0
> C = current RTEMS with -O2
> D = NetBSD derived with -O2
> 
>   | Upload MB/s | memcpy() % | in_cksum() % |
> --+-------------+------------+--------------+
> A |        6.28 |       11.9 |         16.0 |
> B |        5.34 |       10.6 |         23.4 |
> C |        6.24 |       11.5 |         18.5 |
> D |        6.19 |       11.0 |         18.0 |
> 
>   | Download MB/s | memcpy() % | in_cksum() % |
> --+---------------+------------+--------------+
> A |          4.66 |       14.1 |         9.00 |
> B |          3.97 |       12.4 |         15.5 |
> C |          4.66 |       13.8 |         8.00 |
> D |          4.69 |       14.3 |         7.50 |
> 
> Its hard to tell from this figures what is good or bad.

-- 
Configure bugmail: https://www.rtems.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching all bug changes.