Hi,
I've been working for quite a while with 1.5 x64 SDK version since it was released.
Just today I've updated to the latest 2012 x64 SDK.
Details:
What I've noticed is that sometimes while Debbuging you OpenCL code (*.cl)
the position of the current execution line does not corresponds with actual line in the file.
Take a look at my example screenshot. I've placed debug break point at the begin of the kernel.
So, debugger stopped at that break point, but actually it is not there. Only after stepping over 7 lines
variable uint gi = get_global_id(0) will be initialized.
The OpenCLJIT says: OpenCLJIT!BSplineDecompositionImageFilter() Line:173
but actually line uint gi = get_global_id(0); located on 166
Also if the OpenCL function is very simple like this:
float compute(const float index,
__global const INPIXELTYPE* in)
{
return 0.0;
}
You will not be able to put exact breakpoint inside. So, what I have to do is to put
more breakpoints below this function hoping some of them will be triggered.
That is crazy way of working.
I' don't know what is the cause of this problem. I've tried to experiment with Linux/Windows file formats,
tabs vs spaces, removing comments, making all parameters of the functions in one line
float compute(const float index, __global const INPIXELTYPE* in)
No effect. Also for some *cl files all the sudden it is working properly.
Also I can just close my solution and open one of the Intel's examples and it is perfectly fine.
I was hoping that 2012 SDK will solve it, but not :/
This problem still exist both for VisualStudio 9 and 10 in 2012 SDK and 1.5 SDK.
Do you have any ideas what could trigger this behaviour in Intel OpenCL debbuger?
I've attached screenshot from VisualStudio 10
Thanks, Denis



