clinfo | query OpenCL platform and device parameters | GPU library

I would like to make run an old N-body which uses OpenCL.

I have 2 cards NVIDIA A6000 with NVLink, a component which binds from an hardware (and maybe software ?) point of view these 2 GPU cards.

But at the execution, I get the following result:

Here is the kernel code used (I have put pragma that I estimate useful for NVIDIA cards):

I am new to OpenCL and I have read in several places that if/else structures should be avoided, mainly because when the evaluation of threads is different (divergent branching) there is a significant slowdown.

Nonetheless, I used one if(cond) followed by some prints to guarantee that when a forbidden condition is met I can debug what caused it. The point is, whenever the if() is true for a single thread I will kill the process, therefore I am not worried about different threads evaluating the condition differently.

Yet, I found out that even when all threads evaluate false in this if(), there is a massive slowdown when compared to not using the if() -- I commented out the conditional and its body statements to verify.

An observation: I have a kernel (128 workgroups, each one with 128 workitems) that calls a function foo(), and the if/prints are inside foo(). The relevant part of foo() is as follows:

During compilation, I am compiling a large program. One compilation step gives me the following error:

clang++ -o selfdrive/camerad/camerad -Wl,--as-needed -Wl,-rpath=/home/ziyuan/openpilot/phonelibs/snpe/x86_64-linux-clang -Wl,-rpath=/home/ziyuan/openpilot/cereal -Wl,-rpath=/home/ziyuan/openpilot/selfdrive/common selfdrive/camerad/main.o selfdrive/camerad/cameras/camera_common.o selfdrive/camerad/transforms/rgb_to_yuv.o selfdrive/camerad/imgproc/utils.o selfdrive/camerad/cameras/camera_frame_stream.o -Lphonelibs/snpe/x86_64-linux-clang -Lphonelibs/libyuv/x64/lib -Lphonelibs/mapbox-gl-native-qt/x86_64 -Lcereal -Lselfdrive/common -L/usr/lib -L/usr/local/lib -Lcereal -Lphonelibs -Lopendbc/can -Lselfdrive/boardd -Lselfdrive/common -lm -lpthread selfdrive/common/libcommon.a -ljson11 -ljpeg -lOpenCL cereal/libcereal.a cereal/libmessaging.a -lzmq -lcapnp -lkj cereal/libvisionipc.a selfdrive/common/libgpucommon.a -lGL

/usr/bin/ld: selfdrive/camerad/cameras/camera_common.o: in function CameraBuf::init(_cl_device_id*, _cl_context*, CameraState*, VisionIpcServer*, int, VisionStreamType, VisionStreamType, void (*)(void*, int))': /home/ziyuan/openpilot/selfdrive/camerad/cameras/camera_common.cc:92: undefined reference to clCreateCommandQueueWithProperties'

When I check the linker, I got the following:

I'm struggling with getting GL+CL to work together.

I've been following this tutorial. In my code I first call clGetPlatformIDs and retrieve the first (and only) platform. I also get my gl_context from SDL2. Then I want to query the device used by OpenGL with help of clGetGLContextInfoKHR. I successfully obtain this function with clGetExtensionFunctionAddressForPlatform(platform_id, "clGetGLContextInfoKHR") but unfortunately when I call it, I get a segmentation fault. My code is written in Rust but I use low level OpenCL binding, so it looks almost like its C counterpart.

NVIDIA has recently announced OpenCL 3.0 support for their graphics cards. I have NVIDIA Geforce MX150 card and I have updated my CUDA toolkit to version 11.3 with 465.19.01 driver. I have installed clinfo tool that displays all information regarding OpenCL platforms and devices. When I run it, it dislpays:

• Device Version = OpenCL 3.0 CUDA
• Driver Version = 465.19.01
• Device OpenCL C Version = OpenCL C 1.2

I am a bit confused here. What is the difference between Device version & Device OpenCL C version? Will I be able to run OpenCL 3.0 code on my card or it just still support OpenCL 1.2 specs?

The implementation of emulated atomics in openCL following the STREAM blog works nicely for atomic add in 32bit, on CPU as well as NVIDIA and AMD GPUs.

The 64bit equivalent based on the cl_khr_int64_base_atomics extension seems to run properly on (pocl and intel) CPU as well as NVIDIA openCL drivers.

I fail to make 64bit work on AMD GPU cards though -- both on amdgpu-pro and rocm (3.5.0) environments, running on a Radeon VII and a Radeon Instinct MI50, respectively.

The implementation goes as follows:

Post-solution edit: The issue is with the code alone. There is no hardware issue here. Now to the original post:

I'm trying to get a basic OpenCL program to work.

The program simply creates a buffer, writes 42 to the buffer, reads it, and outputs.

Here's the code, written in C:

I currently have a program running all the time using my Nvidia GPU. I would like to run another one aside, which would use OpenCV with OpenCL. I use Ubuntu 18.04 and my processor is an Intel i7-9750H (with UHD Graphics 630).

I run this C++ code to detect the available devices:

I've been using java in IntelliJ as a hobby for quite some time and have decided to start using OpenCL for parallel computing. I'm pretty sure that I've made no mistake downloading and adding the LWJGL library to my project, since it does always compile. The problem is that when I run the program, it produces this output: