From: Steven Blakeslee <hidden> Date: 2003-02-14 21:24:55
I've done a search of the mailing list but I did not find anything related
to my question. I have found on my companies MPC 8260 board that the first
packet sent to the Ethernet when initialized is never received. At first I
thought the network interface had not come up fully so I put a large delay
between the FCC3 Ethernet init and the reading for the first packet. Also,
I have tried reading packets for several seconds then sending it a packet,
it still does not receive it. It seems to receive every packet after the
first one. Has anyone seen something like this.
Thank you,
Steven Blakeslee
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
Hi
I noticed that the lmw and stmw instructions are not used in copy_page(), clear_page() etc.
Aren't these instructions faster than a bunch of lwz/stw?
I assume they are inappropriate for some reason, but I would like to know why.
Would they work on 8xx?
Jocke
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
From: Dan Malek <hidden> Date: 2003-02-20 04:06:38
Joakim Tjernlund wrote:
I noticed that the lmw and stmw instructions are not used in copy_page(), clear_page() etc.
Aren't these instructions faster than a bunch of lwz/stw?
Motorola doesn't guarantee these are any faster, and most documentation indicates
they are likely to be slower. If you look closely, you may notice that not
too many places can utilize these instructions, and often the load or store
is done to take advantages of some pipeline optimizations with comparisons
to values in well known registers.
-- Dan
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
I noticed that the lmw and stmw instructions are not used in copy_page(), clear_page() etc.
Aren't these instructions faster than a bunch of lwz/stw?
Motorola doesn't guarantee these are any faster, and most documentation indicates
they are likely to be slower. If you look closely, you may notice that not
too many places can utilize these instructions, and often the load or store
is done to take advantages of some pipeline optimizations with comparisons
to values in well known registers.
Did some crude benchmarking using clear_page() as my test function.
I made a version that uses stmw and compared that with the orginal
clear_page().
Result:
The stmw version was much slower. When I increased the number of
bytes to copy per loop to 32, it became as fast as the orginal.
Jocke
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
From: Dan Malek <hidden> Date: 2003-02-21 12:48:39
Joakim Tjernlund wrote:
Did some crude benchmarking using clear_page() as my test function.
I'm sure someone has done it before. What did you use for a
benchmark program? You should also be reading the linuxppc-dev
list, as there are many things like this discussed there. The
embedded stuff is just a subset of the global PowerPC development.
Paul (among many others) has documented many benchmark and optimization
activities.
Thanks.
-- Dan
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
Did some crude benchmarking using clear_page() as my test function.
I'm sure someone has done it before.
I searched this list and the dev list using "lmw" as search key
and it didn't turn up anything relevant.
What did you use for a benchmark program?
I hacked my own. I just executed clear_page() in userspace a few times
with and without lmw and measured time with gettimeofday(). Not a very
good test but since the difference is so big, I don't think futher tests
are needed.
You should also be reading the linuxppc-dev
list, as there are many things like this discussed there. The
embedded stuff is just a subset of the global PowerPC development.
Paul (among many others) has documented many benchmark and optimization
activities.
Where is this documented? I can't find much in the lists.
Jocke
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/