clEnqueueAcquireGLObjects high CPU usage

[Saru]

Hello, I am looking for help about the cl_khr_gl_sharing.

My C program use OpenCL to compute some maths and then display the result by sharing an openGL PBO between CL/GL. In most case, those results are displayed up to 60 fps to the user.

Everything seemed ok, my code works, but I noticed an usual CPU usage with clEnqueueAcquireGLObjects.

Actually even if I run a test program with clEnqueueAcquireGLObjects alone in a loop, i can observe the thing taking all my CPU.

What I dont understand is, why does clEnqueueAcquireGLObjects have so much things to do with the HOST ?
because everything should be computed on my GPU anyway. I thought clEnqueueAcquireGLObjects was just acting like a fence for the interop.

I tried others things like using a texture instead of a PBO, change nothing. and on both windows/linux
my setup use a NVIDIA GPU (GTX 1060), with up to date drivers. had no problem with anything with my hardware until now.

Maybe there is a problem with drivers ? or this is just me who doesnt understand how the cl_khr_gl_sharing should be used.

Here is a simpliflied sample of the relevent code in my app.

// this come after all glx/glew/x11/etc.  initialisation

// the buffer is like this
glGenBuffers(1, &gLPBO);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, gLPBO);
glBufferData(GL_PIXEL_UNPACK_BUFFER, size, NULL, GL_DYNAMIC_DRAW);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);


//this is how i create my context
cl_context_properties properties[] = {
    CL_GL_CONTEXT_KHR, (cl_context_properties)glxctx,
    CL_GLX_DISPLAY_KHR, (cl_context_properties)X11display,
    // CL_CONTEXT_PLATFORM, (cl_context_properties)myPlat,
    0
};

//and m command queue
clctx = clCreateContext(properties,1, &prefDevice, NULL, NULL, &err);
clque = clCreateCommandQueueWithProperties(clctx, prefDevice, 0, &err);


// the fonction to share a buffer between GL/CL
cLPBO = clCreateFromGLBuffer(clctx, CL_MEM_WRITE_ONLY, gLPBO, &err);


// a simple loop
while (1) {

    this_thread::sleep_for(chrono::milliseconds(16));

    err = clEnqueueAcquireGLObjects(clque, 1, &cLPBO, 0, 0, 0);

    // err = clEnqueueNDRangeKernel(clque, mykernel, 1, NULL, globWorkSize, NULL, 0, NULL, NULL);

    err = clEnqueueReleaseGLObjects(clque, 1, &cLPBO, 0, 0, 0);

    // glBindBuffer(GL_PIXEL_UNPACK_BUFFER, gLPBO);
    // glRasterPos2i(-1, -1);
    // glDrawPixels(wndWidth, wndHeight, GL_RGBA, GL_UNSIGNED_BYTE, 0);
    // glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
    // glXSwapBuffers(X11display, X11window);
    // glFinish();

}