Dears,
first this is my first post on the forum, so hi to everyone!
Well, now to the problem. I cannot step into the OpenCL kernel using CodeXL, I usually insert a breakpoint just before clEnqueueNDRangeKernel() function and then, when I try to step into the kernel the following message appears:
"The Process was suspended before a kernel enqueued for debug has started executing.
Disable all API function breakpoints and resume debugging (F5) to continue into the kernel."
This happens with the Teapot example and other codes (I am able to run it, but only on the host side, if I try to step into the kernel I obtain the error). If I follow the recommendation and remove all the breakpoints happens the same, so I have no clue of what is going on. If I put a breakpoint within the kernel function I obtain the same result
In the Event properties dialog appears a SIGSEGV error:
Reason SIGSEGV
Address 0000000000000000
Details Invalid memory reference
Thanks for your time, my system is
CentOS 6.4
AMD Catalyst 13.4 (installed from the website)
CodeXL 1.1
AMD HD 6700 series GPU
I also attached an OpenCL code that triggers the error.
Solved! Go to Solution.
Hi,
CodeXL v1.2 was released two days ago. Please give it a try and see if it solves the problem.
I successfully built and debugged the sources you provided on CentOS 6.2 + Radeon HD 6800 + CodeXL 1.2 + Catalyst 13.1 (driver 9.012). No SIGSEGV error was observed.
I don't think the newer driver and OS version should make a difference, although anything is possible.
Please post back your results with CodeXL v1.2.
Hi,
CodeXL v1.2 was released two days ago. Please give it a try and see if it solves the problem.
I successfully built and debugged the sources you provided on CentOS 6.2 + Radeon HD 6800 + CodeXL 1.2 + Catalyst 13.1 (driver 9.012). No SIGSEGV error was observed.
I don't think the newer driver and OS version should make a difference, although anything is possible.
Please post back your results with CodeXL v1.2.
Dear Doron,
I installed CodeXL v1.2 and now everything is working flawlessly. Thanks for your suggestion!.
Hello, everyone.
Sorry to reopen this thread, but I'm using the latest version of CodeXL (1.4.5728.0) and I'm still not able to debug even a single/stupid kernel:
__kernel void cleanVector(__global int* vector_Integers_1, __const int iLength)
{
__private int iGlobalID = get_global_id(0),
iGlobalSize = get_global_size(0);
while (iGlobalID < iLength)
{
vector_Integers_1[iGlobalID] = 0;
iGlobalID += iGlobalSize;
}
}
It gives me the same enigmatic (at least for me) error of this thread's opening message. I'm using a 5870M. Can anyone help me?
Thanks in advance.
Walid
I've exactly the same problem. I'm working on Ubuntu 13.10 using codexl 1.4 and the last beta drivers (14.4) installed through the official installer (i.e. run installer->build debs->install with dpkg). Trying to debug the teapot example produces the above mentioned problem: after an enqueueNDRange function gets hit, the step-inside terminates the application with a SIGPIPE and shows the error message. You can find the logs attached to the message, even if it seems there's no useful information ...
Sincerely,
Daniele
OK, my fault As stated in the release notes ( http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2013/12/AMD_CodeXL_Release_Notes.pdf ) new driver versions > 14.10 are not supported. Installing back the driver packaged in the official ubuntu repositories (13.101) kernel debugging is working again