I wrote a code which I am now testing, but the function clFinish is returning CL_OUT_OF_HOST_MEMORY. I had run smaller examples before and this error did not occur, occuring only for this larger example.
For the larger example, I need to copy 3Mb of information to the GPU and then I launch 81 threads, each one needing approximately more 12Kb to be executed. Adding everything results in about 10Mb, so I don't understand why clFinish returns this error.
On another subject, when I tried to execute this program on a AMD GPU, clBuildProgram returned CL_BUILD_PROGRAM_FAILURE. When I retrieved the error information using clGetProgramBuildInfo, I only obtained warnings and no errors. Why would the code work on a NVIDIA and not on a AMD?