Hi,
I have a slightly obscure problem :) For testing purposes we're running opencl inside a linux VM, so the actual gfx hardware is not accessable. We're using the CPU fallback from intel's opencl impl instead. Apparently, this impl is using TBB to get better performance but our environment does not allow for third parties to create any threads. We don't need any performance, just that it works at all. Is there a way (maybe by setting an environment variable) to disable the multi-threading inside intel's CPU based opencl driver?
Thx, Alex
链接已复制
I believe that's not possible. You may be able to play with global and local sizes to restrict the number of threads but the runtime still depends on tbb. I will fwd your question to see if there are workarounds.
Raghu
Hi,
Also please look at Immediate Execution: http://software.intel.com/sites/products/documentation/ioclsdk/2013/UG/Intel_Immediate_Command_Execution_Extension.htm
