running pcsegdist on gpu

Hi, so its said in the documentation of the function pcsegdist, that it supports gpu, but when I run it, if I look into the function in debug mode, then:

isGPU = coder.gpu.internal.isGpuEnabled;

is 0 and it doesn't run on the gpu.

if I run the command:

gpuDevice

it shows that the device is selected:

K>> gpuDevice

ans = 

  CUDADevice with properties:

                      Name: 'NVIDIA GeForce RTX 2080 Ti'
                     Index: 1
         ComputeCapability: '7.5'
            SupportsDouble: 1
             DriverVersion: 11.6000
            ToolkitVersion: 11.2000
        MaxThreadsPerBlock: 1024
          MaxShmemPerBlock: 49152
        MaxThreadBlockSize: [1024 1024 64]
               MaxGridSize: [2.1475e+09 65535 65535]
                 SIMDWidth: 32
               TotalMemory: 1.1811e+10
           AvailableMemory: 1.0273e+10
       MultiprocessorCount: 68
              ClockRateKHz: 1545000
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 1
          CanMapHostMemory: 1
           DeviceSupported: 1
           DeviceAvailable: 1
            DeviceSelected: 1

Anyone knows what could be the issue ?

Thank you in advance.

Answer 1

[Walter Roberson]

The document says that the function is supported by GPU Coder, not that it can run on GPU normally.

There is no support for marking a pointCloud as existing on GPU; you cannot pointCloud() a gpuArray either.

For an example of the flow needed to do pointCloud work on GPU using GPU Coder toolbox, see https://www.mathworks.com/help/gpucoder/ug/lidar-point-cloud-segmentation-using-gpucoder.html

[Ilya K]

From link above, I think mostof the magic is done in this peace of code:

cfg = coder.gpuConfig('mex');
cfg.TargetLang = 'C++';
cfg.DeepLearningConfig = coder.DeepLearningConfig('cudnn');
codegen -config cfg squeezesegv2_predict -args {ones(64,1024,5,'uint8')} -report

especially in the last line. But im not sure how to adjust it to the function

pcsegdist

Should I replace squeezesegv2_predict by pcsegdist and correct the rights args for i (No sure what are the args in my case).

In the example they are also use the help function to translate pointcloud to 2d 5ch image, but im not sure I need that for the clustering.

Could you maybe reffer to the function above and give more details on how can I configure it to run on the GPU through the coder please ?

Thank you in advance.

[Image Analyst]

What's the difference between this clustering method (pcsegdist) and something like dbscan? They sound similar in that points closer than a certain distance are in the same cluster while other points farther away than the specified distance belong to a different cluster? When would you use each?

[Ilya K]

@Image Analyst, pretty mcuh similiar results and also in the documentation of pcsegdist it is said it uses g-dbscan which is a gpu-zed version of dbscan with another method. I think its pretty much the same. The advantage of pcsegdist is it can be used on gpu as much i understand and trying to figure it out in ths post.