Trying to use gpuArray for graph function

Hi,

We are trying to use gpuArray's to speed up our graph function call but after converting 's' and 't' to gpuArray's the graph function has an error which we have been unable to move past.

The code, error and code output are below. Both gpuArray's are 28 rows with no zero's or negative numbers.

Any help is greatly appreciated.

Thanks

Code:

fprintf('t:')
t = gpuArray((table2array(edgedefs(:,5))))
class (t)
%t = gpuArray(t);


s = gpuArray((table2array(edgedefs(:,4))))
class(s)
%s = gpuArray(s)
w = table2array(edgedefs(:,3));

fprintf('Before Graph\n\n')
beforegraph = clock

G = graph(s, t)

Error message:

Error using matlab.internal.graph.MLGraph
Source must be a numeric array of positive integer node indices.

Error in matlab.internal.graph.constructFromEdgeList (line 218)
    G = underlyingCtor(s, t, totalNodes);

Error in graph (line 300)
                matlab.internal.graph.constructFromEdgeList(...

Error in a_2020_05_15_Sc (line 145)
G = graph(s, t)

Output:

>> t:
t =

      398894
      398895
      398896
      398897
      398899
      398900
      398901
      398902
      398903
      398904
      398905
      398906
      398907
      398908
      398909
      398910
      398911
      398912
      398913
      398914
      398915
      398916
      398917
      398918
      398919
      398920
      398921
      398922


ans =

    'gpuArray'


s =

        7754
        7755
        7756
        7757
        7759
        7760
        7761
        7762
        7763
        7764
        7765
        7766
        7767
        7768
        7769
        7770
        7771
        7772
        7773
        7774
        7775
        7776
        7777
        7778
        7779
        7780
        7781
        7782


ans =

    'gpuArray'

Before Graph


beforegraph =

         2020            5           17           20           39         4.14
         

Answer 1

[Edric Ellis]

Unfortunately, graph does not support using gpuArray data in this way. Is there some specific algorithm on your graph that you're looking to speed up using the GPU?

[Chad Davies]

Hi Edric,

Thanks for the info.

These are the three commands that I'm using to display the data. I have about 1 million nodes to plot in my test case. When we process the real data, there will be 1 billion nodes.

The graph and plot functions below execute very quickly, but the layout function takes 6-8 hours. Is there a faster way to run the layout fuction?

G = graph(s, t)
figure;
H = plot(G,'MarkerSize',4)
layout(H,'force3','UseGravity',true,'Iterations',50, 'WeightEffect', 'inverse')

Thank for any help that you can provide.