Runtime of OpenCL command queue creation

Raphael_H_ — Fri, 13 Nov 2015 15:53:59 GMT

Hello,

while using the Intel Xeon Phi with OpenCL (currently version 1.2) I stumbled upon some runtime issues on my Linux system. Further measurements showed the creation of the command queue takes around 700ms. This seems very long compared to other tasks. Here are my measurements for the individual calls to the OpenCL API.

clGetDeviceIDs            -->     29.416 ms.
clGetDeviceIDs            -->      0.014 ms.
clCreateContext           -->      3.583 ms.
clCreateCommandQueue      -->    787.205 ms.
clCreateProgramWithSource -->      0.014 ms.
clBuildProgram            -->    209.975 ms.
clCreateKernel            -->      0.473 ms.
clCreateBuffer (in1)      -->      2.127 ms.
clCreateBuffer (in2)      -->      2.042 ms.
clCreateBuffer (out)      -->      0.049 ms.
setKernelArg              -->      0.005 ms.
clEnqueueNDRangeKernel    -->     60.167 ms.
clEnqueueReadBuffer       -->      8.754 ms.

Not sure if this is a result of my implementation or OpenCL the driver. The code for my measurements is attached (requires C++ 11 to compile).

Thanks, Raphael

topic Runtime of OpenCL command queue creation in Software Archive

Runtime of OpenCL command queue creation