Hi, thank you very mush for helping me! Please, find information obtained with coredumpctl info as txt file attached to this email. I was able to run the interop examples from CUDA-OpenGL, but I was also able to run this:  https://github.com/9prady9/CLGLInterop  Erik Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Wed28 Feb 2018 at 16:29:47, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Hi Erik, Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes: Can you provide a traceback (using gdb, say)? Are those the OpenCL or the CUDA interop examples? Andreas

I am not a Linux expert, I don't know exactly how to trace all differences. I did the following test, which maybe can give you some insights. I wrote a simple test python script (see attached python  file) and made it run on both my mac laptop (where it works fine) and my other linux computer (where I have problems). There is one particular instruction (see comments in the Python file) that makes the script crashing in Linux when executed: ctx = cl.Context(properties = [(context_properties, platform)] + sharing_properties, devices = gpus) In case I modify this instruction this way: ctx = cl.Context(properties = [(context_properties, platform)], devices = gpus) then the script works fine in Linux too, but of course I don't have the CL-GL interop. I also made my code printing something during the execution. This is its output when correctly executing on my mac (with the " + sharing_properties"): dhcp-107:PythonCLGL Erik$ python simple_test_interop.py Have GL:  True My platform is:  Apple My GPU is:  HD Graphics 5000 My context properties:  4228 My sharing properties:  [(268435456, 140680091402240)] If I run the same script (so, with the " + sharing_properties") on the linux computer I get: [cooper@twinpeaks examples]$ python2 simple_test_interop.py  Have GL:  True My platform is:  NVIDIA CUDA My GPU is:  GeForce GT 640 My context properties:  4228 My sharing properties:  [(8200, 899558088), (8202, <OpenGL.raw.GLX._types.LP_struct__XDisplay object at 0x7f2391351290>)] Segmentation fault (core dumped) Sometimes I see some negative numbers in my sharing properties when on linux: [cooper@twinpeaks examples]$ python2 simple_test_interop.py  Have GL:  True My platform is:  NVIDIA CUDA My GPU is:  GeForce GT 640 My context properties:  4228 My sharing properties:  [(8200, -722625272), (8202, <OpenGL.raw.GLX._types.LP_struct__XDisplay object at 0x7f6ebf0ea290>)] Segmentation fault (core dumped) However, if I remove the " + sharing_properties" from my code what I get on linux is: [cooper@twinpeaks examples]$ python2 simple_test_interop.py  Have GL:  True My platform is:  NVIDIA CUDA My GPU is:  GeForce GT 640 My context properties:  4228 My sharing properties:  [(8200, 710605096), (8202, <OpenGL.raw.GLX._types.LP_struct__XDisplay object at 0x7f920f08a290>)] Also, under this condition, I sometimes see some negative numbers in my sharing properties: [cooper@twinpeaks examples]$ python2 simple_test_interop.py  Have GL:  True My platform is:  NVIDIA CUDA My GPU is:  GeForce GT 640 My context properties:  4228 My sharing properties:  [(8200, -173530072), (8202, <OpenGL.raw.GLX._types.LP_struct__XDisplay object at 0x7f2fb3782290>)] In the latter two cases, having removed the " + sharing_properties", the code correctly opens a blank OpenGL window in linux without crashing. Of course I don't have the CL-GL interop in this case! As far as I can see from this simple script, the only difference I can notice is the structure of the output of the instruction: print "My sharing properties: ", sharing_properties On my mac it's always something like: My sharing properties:  [(268435456, 140680091402240)] and never see negative numbers, while on Linux is: My sharing properties:  [(8200, -173530072), (8202, <OpenGL.raw.GLX._types.LP_struct__XDisplay object at 0x7f2fb3782290>)] and sometimes the second number in (8200, -173530072) is positive or negative. My guess is that the function: get_gl_sharing_context_properties() behaves differently in case it is executing on a Mac computer or on a Linux computer. Would this make sense? Best regards, Erik P.S: on my Linux computer python comes in both python 2.X and 3.X versions. For this, when I installed the system it automatically created the alias python2 for python 2.X. The linux computer have a fresh installation (basically of 1 week ago, when I assembled that PC in order to dedicate it as a machine for numerical simulations). Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Thu01 Mar 2018 at 03:02:30, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes: So this looks like it's crashing in context creation (which is very early on). That means the one likely cause is a difference in parameters passed to clCreateContext. Could you trace through both the C example and the Python example, find and highlight any differences and report back? Andreas

Hi, I wrote a similar code in C++ (see code below this email). I compile it on Mac this way: g++ clgl_interop.cpp -o clgl_interop -framework OpenCL -framework OpenGL -lGLEW -lglfw and on Linux this way: g++ clgl_interop.cpp -o clgl_interop -lOpenCL -lOpenGL -lGLEW -lglfw It looks like the culprit of my problem is the way I create the context. On Linux, if I create the context using: clCreateContextFromType than it works fine. Otherwise if in Linux I use: clCreateContext than it crashes with segmentation fault. I remember somewhere I read, I think in the PyOpenCL demo "gl_interop_demo.py", a comment about that but on that time I was not able to understand it. Could you maybe check the latter file and elaborate this point? Thanks _______________________ C++ code: _______________________ #include <stdio.h> #include <stdlib.h> #include <GL/glew.h> #ifdef __WINDOWS__   #define GLFW_EXPOSE_NATIVE_WIN32   #define GLFW_EXPOSE_NATIVE_WGL #endif #ifdef __linux__   #define GLFW_EXPOSE_NATIVE_X11   #define GLFW_EXPOSE_NATIVE_GLX #endif #include <GLFW/glfw3.h> #include <GLFW/glfw3native.h> #ifdef __APPLE__   #include <OpenGL/OpenGL.h>   #include <OpenCL/opencl.h> #else   #include <GL/gl.h>   #include <CL/cl.h>   #include <CL/cl_gl.h> #endif void window_refresh_callback(GLFWwindow* window) { glClear(GL_COLOR_BUFFER_BIT); glfwSwapBuffers(window); } void key_callback(GLFWwindow* window, int key, int scancode, int action, int mods) {     if (key == GLFW_KEY_ESCAPE && action == GLFW_PRESS)         glfwSetWindowShouldClose(window, GL_TRUE); } int main() {   GLFWwindow* window;   int               i;   int               j;   char*             value;   size_t            valueSize;   cl_uint           platformCount;   cl_platform_id*   platforms;   cl_uint           deviceCount;   cl_device_id*     devices;   cl_uint           maxComputeUnits;   cl_context        context;   int               err;   //////////////////////////////////////////////////////////////////////////////// ///////////////////////// Creating OpenGL context... /////////////////////////// ////////////////////////////////////////////////////////////////////////////////   if (glfwInit() != GLFW_TRUE) {   printf("Unable to initialize GLFW!\n"); glfwTerminate(); exit(EXIT_FAILURE); } glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 4); glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 1); glfwWindowHint(GLFW_OPENGL_FORWARD_COMPAT, GL_TRUE); glfwWindowHint(GLFW_OPENGL_PROFILE, GLFW_OPENGL_CORE_PROFILE); glfwWindowHint(GLFW_SAMPLES, 4);   window = glfwCreateWindow(800, 600, "Test interop CPP", NULL, NULL);   if (!window) {     glfwTerminate();     exit(EXIT_FAILURE);   }   glfwMakeContextCurrent(window); glfwSetWindowRefreshCallback(window, window_refresh_callback);   glfwSetKeyCallback(window, key_callback); glewExperimental = GL_TRUE; if (glewInit() != GLEW_OK) {     printf("Unable to initialize GLEW!\n");     exit(EXIT_FAILURE);   }   //////////////////////////////////////////////////////////////////////////////// ///////////////////////// Creating OpenCL context... /////////////////////////// ////////////////////////////////////////////////////////////////////////////////   // get all platforms   clGetPlatformIDs(0, NULL, &platformCount);   platforms = (cl_platform_id*) malloc(sizeof(cl_platform_id) * platformCount);   clGetPlatformIDs(platformCount, platforms, NULL);   for (i = 0; i < platformCount; i++)   {     // print platform name     clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, 0, NULL, &valueSize);     value = (char*) malloc(valueSize);     clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, valueSize, value, NULL);     printf("\n%d Platform: %s\n", i + 1, value);     free(value);     // print platform profile     clGetPlatformInfo(platforms[i], CL_PLATFORM_PROFILE, 0, NULL, &valueSize);     value = (char*) malloc(valueSize);     clGetPlatformInfo(platforms[i], CL_PLATFORM_PROFILE, valueSize, value, NULL);     printf(" %d.%d Platform profile: %s\n", i + 1, 1, value);     free(value);     // print platform version     clGetPlatformInfo(platforms[i], CL_PLATFORM_VERSION, 0, NULL, &valueSize);     value = (char*) malloc(valueSize);     clGetPlatformInfo(platforms[i], CL_PLATFORM_VERSION, valueSize, value, NULL);     printf(" %d.%d Platform version: %s\n", i + 1, 2, value);     free(value);     // print platform vendor     clGetPlatformInfo(platforms[i], CL_PLATFORM_VENDOR, 0, NULL, &valueSize);     value = (char*) malloc(valueSize);     clGetPlatformInfo(platforms[i], CL_PLATFORM_VENDOR, valueSize, value, NULL);     printf(" %d.%d Platform vendor: %s\n", i + 1, 3, value);     free(value);     // print platform extensions     clGetPlatformInfo(platforms[i], CL_PLATFORM_EXTENSIONS, 0, NULL, &valueSize);     value = (char*) malloc(valueSize);     clGetPlatformInfo(platforms[i], CL_PLATFORM_EXTENSIONS, valueSize, value, NULL);     printf(" %d.%d Platform extensions: %s\n", i + 1, 4, value);     free(value);     // get all devices     clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, 0, NULL, &deviceCount);     devices = (cl_device_id*) malloc(sizeof(cl_device_id) * deviceCount);     clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, deviceCount, devices, NULL);     // for each device print critical attributes     for (j = 0; j < deviceCount; j++)     {       // print device name       clGetDeviceInfo(devices[j], CL_DEVICE_NAME, 0, NULL, &valueSize);       value = (char*) malloc(valueSize);       clGetDeviceInfo(devices[j], CL_DEVICE_NAME, valueSize, value, NULL);       printf("\n%d.%d Device: %s\n", i + 1, j + 1, value);       free(value);       // print hardware device version       clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, 0, NULL, &valueSize);       value = (char*) malloc(valueSize);       clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, valueSize, value, NULL);       printf(" %d.%d.%d Hardware version: %s\n", i + 1, j + 1, 1, value);       free(value);       // print software driver version       clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, 0, NULL, &valueSize);       value = (char*) malloc(valueSize);       clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, valueSize, value, NULL);       printf(" %d.%d.%d Software version: %s\n", i + 1, j + 1, 2, value);       free(value);       // print c version supported by compiler for device       clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, 0, NULL, &valueSize);       value = (char*) malloc(valueSize);       clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, valueSize, value, NULL);       printf(" %d.%d.%d OpenCL C version: %s\n", i + 1, j + 1, 3, value);       free(value);       // print parallel compute units       clGetDeviceInfo(devices[j], CL_DEVICE_MAX_COMPUTE_UNITS,                   sizeof(maxComputeUnits), &maxComputeUnits, NULL);       printf(" %d.%d.%d Parallel compute units: %d\n", i + 1, j + 1, 4, maxComputeUnits);     }   }   #ifdef __APPLE__     CGLContextObj     kCGLContext     = CGLGetCurrentContext();     CGLShareGroupObj  kCGLShareGroup  = CGLGetShareGroup(kCGLContext);     cl_context_properties properties[] =     {         CL_CONTEXT_PROPERTY_USE_CGL_SHAREGROUP_APPLE,         (cl_context_properties) kCGLShareGroup,         0     };   #endif   #ifdef __linux__ cl_context_properties properties[] = {     CL_GL_CONTEXT_KHR, (cl_context_properties)glfwGetGLXContext(window),     CL_GLX_DISPLAY_KHR, (cl_context_properties)glfwGetX11Display(),     CL_CONTEXT_PLATFORM, (cl_context_properties)platforms[0],     0 }; #endif #ifdef __WINDOWS__ cl_context_properties properties[] = { CL_GL_CONTEXT_KHR, (cl_context_properties)glfwGetWGLContext(window), CL_WGL_HDC_KHR, (cl_context_properties)GetDC(glfwGetWin32Window(window)), CL_CONTEXT_PLATFORM, (cl_context_properties)platforms[0], 0 }; #endif   ////////////////////////////////////////////////////////////////////////////////   //////////////////////////// HERE IS THE PROBLEM: //////////////////////////////   ////////////////////////////////////////////////////////////////////////////////   // On my Linux machine, this works:   context = clCreateContextFromType(properties, CL_DEVICE_TYPE_GPU, NULL, NULL, &err);   // On my Mac machine, this works but if I use it in Linux than it crashes with segmentation fault:   //context = clCreateContext(properties, 1, &devices[1], NULL, NULL, &err);   ////////////////////////////////////////////////////////////////////////////////   ///////////////////////////// END OF THE PROBLEM: //////////////////////////////   ////////////////////////////////////////////////////////////////////////////////     if (!(context))     {       printf("\nError: Failed creating OpenCL-GL shared context!\n");       exit(err);     }     else     {       printf("\nCreated OpenCL-GL shared context!\n");     }   while (!glfwWindowShouldClose(window)) // Looping here until the window is closed...   {     glClear(GL_COLOR_BUFFER_BIT);     glfwSwapBuffers(window);     glfwPollEvents();   }   glfwDestroyWindow(window); glfwTerminate();   free(devices);   free(platforms);   return 0; } Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Thu01 Mar 2018 at 17:15:56, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Hi Erik, From looking at the code, it seems that the C example you point to and PyOpenCL do equivalent things, at least up to context creation. Could you perhaps print what the C example passes for the context properties and compare the numerical values with the ones passed by PyOpenCL? Andreas Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes:

Not yet, but I can do it as you suggest. However, I believed that Pyopencl, somewhere buried into some C-wrappers, should be able to make a correct use of the clCreateContextFromType the way I am using it in C (which is working for me in Linux) when invoking it from Python like this: ... platform = cl.get_platforms() context =   cl.Context(dev_type = cl.device_type.GPU,                                    properties = [(cl.context_properties.PLATFORM,                                    platform[0])] + get_gl_sharing_context_properties()) ... but this crashes on Linux with the same segmentation fault. I tried to reverse-engineer a bit your code: in "cffi_cl.py", for the class Context, I found this code fragment: ... if devices is not None:             # from device list             if dev_type is not None:                 raise RuntimeError("one of 'devices' or 'dev_type' "                                    "must be None",                                    status_code.INVALID_VALUE, "Context")             _devices, num_devices = _clobj_list(devices)             # TODO parameter order? (for clobj_list)             _handle_error(_lib.create_context(_ctx, c_props,                                               num_devices, _devices))         else:             # from device type             if dev_type is None:                 dev_type = device_type.DEFAULT             _handle_error(_lib.create_context_from_type(_ctx, c_props,                                                         dev_type)) ... as far as I can understand, when invoking the Context class by passing the "dev_type" parameter = "cl.device_type.GPU" this should select the "else" case, hence the wrapper "create_context_from_type" which I found defined in "context.cpp": ... create_context_from_type(clobj_t *_ctx, const cl_context_properties *props,                          cl_device_type dev_type) {     // TODO debug print properties     return c_handle_error([&] {             *_ctx = new context(                 pyopencl_call_guarded(                     clCreateContextFromType,                     const_cast<cl_context_properties*>(props),                     dev_type, nullptr, nullptr), false);         }); } ... Do you think there is any possibility that the latter "pyopencl_call_guarded(clCreateContextFromType, ..etc..)" wrapper in not doing what it is supposed to do? I mean, is it possible that function is not building the correct "clCreateContextFromType" C-function with all necessary arguments? I am speculating this because, in my C test program, the clCreateContextFromType function works fine in both Linux and Mac. I agree with you it is still not clear why the simple clCreateContext in C apparently does not work in Linux (at least on my computer).               Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Tue06 Mar 2018 at 18:23:02, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes: That sounds like an implementation bug then. Have you checked the Nvidia dev forums or tried reporting this issue (along with your reproducer) to them? Andreas

Today I have posted this on the Nvidia forum: https://devtalk.nvidia.com/default/topic/1030737/?comment=5243539 Let's see what they say about it. At the moment I don't know how to proceed: let me know if you have any other idea I can test. Thanks Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Tue06 Mar 2018 at 21:43:16, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes: Thanks for investigating! Given the amount of shared code (with clCreateContext) and compiler type checking present, I'd say it'd be hard for the clCreateContextFromType wrapper to be wrong. But I did notice that the function is untested, and I'd be happy to merge a patch that adds a test. Andreas

...sorry: here is the attachment :) Erik Zorzin address: Viale Miramare 235, 34136 Trieste - ITALY phone: (+39) 3291555703 webpage: http://www.zorzin.com (sent from my MacBook Air, Erik Zorzin - HOME) On Wed28 Feb 2018 at 16:29:47, Andreas Kloeckner (lists(a)informa.tiker.net) wrote: Hi Erik, Erik Zorzin - HOME <erik(a)zorzin.com&gt; writes: Can you provide a traceback (using gdb, say)? Are those the OpenCL or the CUDA interop examples? Andreas