Intel Communities: Message List - Running RCCE v1.0.13 emulator issues

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Mon, 10 Jan 2011 20:05:21 GMT

Hi Carsten and Ted,

It seems that Cygwin default stack size is 2048KB and this attribute cannot be changed using commands from the shell.

One can, however, compile with a different stack size, using:

g++ -Wl,--stack,8000000

(for 8MB stack)

I added this option to /common/symbols.in, re-makeall and pingpong worked with the large buffer.

All is clear here now.

Thank you very much.

Ohn

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Mon, 10 Jan 2011 18:22:59 GMT

Ohn,

Is this what you are doing ... running pingpong from within the Makefile?

If I add the rccerun line to the Makefile as follows ...

pingpong: $(PINGPONGOBJS) Makefile
$(CCOMPILE) -o pingpong $(PINGPONGOBJS) $(CFLAGS)
rccerun -nue 2 -f ../../hosts/rc.hosts pingpong

then I get the error

g++ -o pingpong RCCE_pingpong.o /home/tkubasx/sandbox/RCCE_V1.0.13/bin/OMP/libRCCE_bigflags_nongory_nopwrmgmt.a -O3 -fopenmp -I/home/tkubasx/sandbox/RCCE_V1.0.13/include
rccerun -nue 2 -f ../../hosts/rc.hosts pingpong
pingpong 2 1.0 00 01
/home/tkubasx/sandbox/RCCE_V1.0.13/rccerun: line 183: 10934 Segmentation fault $EXE $ARGS
make: *** [pingpong] Error 139

If I take out the rccerun from the Makefile and run it from the command line, it works fine.

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Mon, 10 Jan 2011 07:39:39 GMT

I recently encountered similar issues with pingpong:

Running pingpong from the shell (/bin/bash) succeeded.

Running it from within a makefile failed.

The reason was, that make uses /bin/sh as shell which

(by default) provides different amounts of memory.

The memory for buffer[] is taken from the stack (once per UE).

I don't know how to set maximum stack size for sh

but for bash one can use:

ulimit -s <stack size in kB>

Setting the stack size to (a bit more) than

4MB per UE should be enough to make pingpong run.

The tool ulimit also has the option of setting stack size to "unlimitted"

but that actually appeared to provide less memory

(It let pingpong fail). The default stack size may depend

on some system specific setting: I you are using a different shell

you will probably have to use a different tool

to set the stack size large enough...

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Sat, 08 Jan 2011 22:11:47 GMT

I further investigated my problem:

PINGPONG works for me only when I drastically reduce the size of array "buffer[]" in RCCE_pingpong.c (from the default [1024*1024*4] to let's say [32*32*4]).

Bigger values for "buffer[]" or "nrounds" make the program very slow.

BIG numbers (like the default 1024*1024*4) make pingpong exit instantly after running it.

How do you explain that?

Also, what is buffer[] anyway? Which memory allocates these 4MB?

About the single/double printing of "Hello World" -

The phenomenon has ceased to occur. I don't know what I did that fixed it. I did not touch the .c or .exe.

It's strange. Thanks for trying it for me though.

Ohn

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Wed, 05 Jan 2011 06:52:03 GMT

I redid the example and ensured that I was using the same rccerun. You are correct ... I may have been using a more recent rccerun. But I don't see any errors using the 1.0.13 rccerun ... as shown below. Did you put rccerun in your path?

Of course, this is Linux not Win7/cygwin. We'll set up a Win7/cygwin system here locally to test the emulator on, hopefully by next week. We haven't seen reported problems with the emulator, but then most people use it on Linux.

$export PATH=.:/home/tkubasx/sandbox/RCCE_V1.0.13:$PATH
$echo $PATH
.:/home/tkubasx/sandbox/RCCE_V1.0.13:/home/tkubasx/bin:.:/home/tkubasx/bin:/home/tkubasx/sandboxJF/rcce:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/opt/sccKit/1.3.0/bin
$cd sandbox
$cd RCCE_V1.0.13/
$ls
apps build_stress_test configure hosts makeall man README sourcing utils
bin common COPYING include Makefile rccerun run_stress_test src

$which rccerun
/home/tkubasx/sandbox/RCCE_V1.0.13/rccerun

$rccerun -nue 2 -f ../../hosts/rc.hosts hello
hello 2 1.0 00 01
hello from RCCE
hello from RCCE

I can run this many times and see the same result.

$rccerun -nue 2 -f ../../hosts/rc.hosts pingpong
pingpong 2 1.0 00 01
32 0.000000740
32 0.000000736

2493920 0.001060648
2965792 0.001343287
3526944 0.001673127

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Tue, 04 Jan 2011 09:29:03 GMT

I think we are running different rcceruns.

I'll describe what I do different:

1. As the document suggests - I copied rccerun from /RCCE_V1.0.13/ to the running directory. I can't just type rccerun anywhere and run it.

2. Then I run it with "./rccerun -nue 2 -f ../../hosts/rc.hosts pingpong.exe".

3. I have to state "pingpong.exe" instead of "pingpong".

4. It fails, and I have to go to rccerun, line 155, and add "./" before "$EXE $ARGS". Then it runs and finishes but prints nothing.

If not tried, can you please try using /RCCE_V1.0.13/rccerun , without any "intel" definitions you might have in your login, and see if still no errors?

Thanks Ted, you help a lot!

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Mon, 03 Jan 2011 23:12:59 GMT

Actually not a hang ... I get the following error but it just took a long time.

tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$ rccerun -nue 2 -f ../../hosts/rc.hosts hello
pssh -h PSSH_HOST_FILE.7236 -t -1 -p 2 /home/tkubasx/sandbox/RCCE_V1.0.13/apps/HELLO/mpb.7236 < /dev/null
[1] 15:23:15 [FAILURE] rck00 Exited with error code 255
[2] 15:23:15 [FAILURE] rck01 Exited with error code 255
pssh -h PSSH_HOST_FILE.7236 -t -1 -P -p 2 /home/tkubasx/sandbox/RCCE_V1.0.13/apps/HELLO/hello 2 0.533 00 01 < /dev/null
[1] 15:23:36 [FAILURE] rck00 Exited with error code 255
[2] 15:23:36 [FAILURE] rck01 Exited with error code 255
tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$

Re: Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Mon, 03 Jan 2011 23:11:52 GMT

I tried the emulator for 1.0.13 on Linux with gcc (actually g++ but on my Ubuntu it's the same).

I see it working fine, but I had to add the -emulator switch to rccerun. Now I know that the docs say that -emulator is the default when you buld with the emulator, but rcce 1.0.13 is an old version. The emulator is broken in the trunk, and that is the main thing holding up rcce 1.1.

I have not yet tried the emulator on a non-Linux platform. I understand that you are using Win 7/cygwin. I don't have cygwin installed on my laptop, but if this is still an issue with you, then I will.

Here's how I ran the "simplest Hello World in the world." I'm not seeing that strange issue you've seen with "Hello World" being printed only once.

tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$ rccerun -emulator -nue 2 -f ../../hosts/rc.hosts hello
hello 2 0.533 00 01
hello from RCCE
hello from RCCE
tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$ rccerun -emulator -nue 2 -f ../../hosts/rc.hosts hello
hello 2 0.533 00 01
hello from RCCE
hello from RCCE
tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$ rccerun -emulator -nue 2 -f ../../hosts/rc.hosts hello
hello 2 0.533 00 01
hello from RCCE
hello from RCCE
tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$ rccerun -emulator -nue 2 -f ../../hosts/rc.hosts hello
hello 2 0.533 00 01
hello from RCCE
hello from RCCE

tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/HELLO$

I can also run pinpong.

I built rcce as follows ....

./configure emulator

./makeall

cd apps/PINGPONG

make

The I ran pingpong as

tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/PINGPONG$ rccerun -emulator -nue 2 -f ../../hosts/rc.hosts pingpong
pingpong 2 0.533 00 01
32 0.000000739
32 0.000000735

2965792 0.001312684
3526944 0.001710459
tkubasx@marc101:~/sandbox/RCCE_V1.0.13/apps/PINGPONG$

When I do not put the -emulator switch in, I see a hang.

Running RCCE v1.0.13 emulator issues

webadmin@intel.com — Sun, 02 Jan 2011 19:26:25 GMT

Hi,

Before running my experimental programs I am trying to make the "PINGPONG" example work, as SCC_Platform_Overview suggests.

But I encount problems.

I am not filing a bugzilla (yet) because I don't know if these are bugs or maybe I'm doing something wrong.

The problems:

(1)

I get an error on rccerun, line 155, saying "pingpong.exe command not found".

When I add "./" at the beginning of the line, it runs without errors.

(2)

Running "./rccerun -nue 2 -f ../../hosts/rc.hosts pingpong.exe" does nothing. It just writes the command line and exits.

Debugging a little, I found that having RCCE_send() or RCCE_recv() anywhere in the code - prevents the IO from performing any printf() elsewhere.

(3)

I wrote the simplest Hello World program in the world:

int main(...) {

printf ("Hello World\n");

}

I run it:

./rccerun -nue 2 -f ../../hosts/rc.hosts try.exe

The output:

Hello World

Whenever I run it again, I get:

Hello World

(printed only once!)

It continues happening until I re-build my program.

I hope that anyone can help me. I really want to get over the first stages.

Thanks in advance,

Ohn