invalid memory access and then seg fault on iMac with Radeon HD 4850

[ochafik]

From @hakuliu on November 13, 2013 17:8

I was having trouble getting anything to run on this particular hardware, and try to distill is to some reproducible cases...

I made a little test suite that can be accessed in the following link:
https://drive.google.com/file/d/0Byprex-Adx53c1V0UWN0a09ZdU0/edit?usp=sharing

the package includes a simple test in c which gets all platform and devices and displays some information of those devices. It then runs a very simple kernel which takes a float * input and adds 1 to each element to produce the output. This simple test case, of course, ran just fine on the machine in question.

There is another test in the package which performs the same basic test, but using the JavaCL RC3 package. running this with ant gives me a segmentation fault. (only sometimes...I think the first run is okay and second is bad....which is interesting)

I have included the output of both runs
https://gist.github.com/hakuliu/7452553

I know the native code ran from JavaCL is probably not the same as the one as I have in my C code. I'll have to possibly look through the native code called from JavaCL... I'll continue investigating this error later as I know it's hard to get hardware of specific configuration...

Copied from original issue: nativelibs4java/nativelibs4java#458

[ochafik]

From @hakuliu on November 18, 2013 21:48

an update: I was trying to try to reproduce the code from CLProgram.createKernel by calling OpenCLLibrary.clCreateKernel() directly, so i can get a more granular look at the line that's causing the segfault from before, but I'm having a little trouble figuring out how to get the long for the cl_program... here's a gist of what i'm having troubles with....

https://gist.github.com/hakuliu/7535564

I'm doing this on a computer that i'm positive does NOT have an error creating the kernel...but with the following code, i get a SIGSEGV...which means I probably messed up the pointer to the cl_program...

(calling clCreateKernel(0,0,0) will return 0, as it should...i think)