MCSDK HPC 3.x Trouble shooting

HPC (High Performance Computing) Development Tools for MCSDK

Version 3.0

Trouble Shooting

Last updated: 07/28/2014

---

Frequently Asked Questions

Listed below are some Frequently Asked Questions. Please click on the "Expand" link adjacent to any question to see the answer.

• Can I run OpenMPI demos on a single K2H node?

Yes. You can, but the instructions to run the demos will require the following changes:

1. Remove k2hnode2 from "-host k2hnode1,k2hnode2"
2. If demo uses OpenCL (such as the fftdemos), please specify number of processes as 1 (i.e change "-np 2" to "-np 1")

• I see the error "CMEM Error: init: Failed to open /dev/cmem: 'No such file or directory'" when I try to run one of the demos

There could be a few reasons. a) You have not reserved atleast 1536M of memory in U-boot (using mem_reserve variable) b)You could be using an real-time kernel (The pre-built CMEMK module is compatible only with the non-real-time kernel) c)The MCSDK-HPC installation is incomplete or you didn't reboot the EVM after finishing installation. The CMEM module is inserted via udev rules

• How can I check if CMEM module is properly inserted?

Use the following commands:

         dmesg | grep -A 10 CMEMK  ==> this will show if Kernel attempted to insert CMEMK
         cat /dev/mtd1 | grep mem_reserve  ==> This should read 1536M (if not, please change uboot variable)
         cat /dev/mtd1 | grep ddr3a_size  ==> Needs to be set if user has >2 GB Memory

Also, check if the file "/etc/modules-load.d/cmem.conf" exists on the EVM filesystem. If not, you either missed "make install" or the "make install" step didn't succeed

• Do I need sudo access?

User needs sudo access for installation. Once all the packages are installed, for running sample applications, user will not need sudo access"

• Can two users simultaneously use one K2H node? What are the restrictions?

Two or more users can concurrently connect to a K2H node via SSH. However, both users cannot concurrently use shared resources (for e.g. offloading to C66x, allocate memory from CMEM, use OpenCL, OpenMPI etc services"

• I have the environment variables set properly. However, when I run make, I get errors as if environment variables are not set

Please check if you are running 'sudo make'. The environment variables are not auto-exported to sudo. The recommended way is to change the permissions for the root folder to 777 and not use sudo for make"

• Hyperlink shows “failed” for some steps. Is this as expected?

Yes. This is expected, since probing of certain Hyperlink interfaces failed. This means Hyperlink initialization failed since remote peer is not opening that particular interface. For example, using node1 and node4 to run testmpi can have console display as follows:

<syntaxhighlight lang="bash"> [fpc1n1:02505] fpc1n1, hyplnk0 attempt opening [fpc1n4:02529] fpc1n4, hyplnk1 attempt opening Timeout exceeded for connection, exiting... Peripheral setup failed, retVal = 3 Peripheral init failed for: arm-remote-hyplnk-0 hyplnk open failed [fpc1n1:02505] fpc1n1, hyplnk0 open failed [fpc1n1:02505] fpc1n1, hyplnk1 attempt opening Timeout exceeded for connection, exiting... Peripheral setup failed, retVal = 3 Peripheral init failed for: arm-remote-hyplnk-1 hyplnk open failed [fpc1n4:02529] fpc1n4, hyplnk1 open failed [fpc1n4:02529] fpc1n4, hyplnk0 attempt opening [fpc1n1:02505] fpc1n1, hyplnk0=(nil) hyplnk1=0x69570 [fpc1n1:02505] fpc1n1, HLINK turned ON !!! [fpc1n4:02529] fpc1n4, hyplnk0=0x69548 hyplnk1=(nil) [fpc1n4:02529] fpc1n4, HLINK turned ON !!! Hello world from processor fpc1n4, rank 1 out of 2 processors locally obtained hostname fpc1n4 Hello world from processor fpc1n1, rank 0 out of 2 processors locally obtained hostname fpc1n1 </syntaxhighlight>

In this example, hyplnk0 opening failed on node1, since node2 (connected via its port1 to port0 of node1) is not started. The same happens for hyplnk1 on node 4 that is connected to hyplnk1 of node 3 that is also not activated. But hyplnk1 on node1 and hyplnk0 on node4 succeeded – as they are connected to each other and both ports are activated.

• New Question Template

Here is the answer"