Interfacing with C code

The code in MLNanoShaperRunner can be compiled to run as a shared object and ca be interfaced as a shared library.

The first step is to compile the code.

julia --project MLNanoShaperRunner/build/build.jl

Once this is done, we have muliple directory in MLNanoShaperRunner/build/lib.

• include which contains the headers to be included in the C code.
• lib which contains the shared objects that need to be referenced by the code.
• shared which contains the artefacts necessary for the julia code. A copy of the shared directory must be included in the root project of the executable.

Interface

First the Interface code define some structures.

typedef struct {
  float x;
  float y;
  float z;
  float r;
} sphere;

and

typedef struct {
  float x;
  float y;
  float z;
} point;

int load_model(char *path);

Load the model parameters form a serialised training state at absolute path path. Parameters: path - the path to a serialized NamedTyple containing the parameters of the model Return value(int):

• 0: OK
• 1: file not found
• 2: file could not be deserialized properly
• 3: unknow error

int load_atoms(sphere *start, int length);

Load the atoms into the julia model. Start is a pointer to the start of the array of sphere and length is the length of the array

Return an error status:

• 0: OK
• 1: data could not be read
• 2: unknow error

float eval_model(point *start,int length); 

evaluate the model at coordinates start[0],...,start[length -1]

Example

#include "MLNanoShaperRunner.h"
#include "julia_init.h"

int main(int argc,char *argv[]) {
  init_julia(argc, argv);
  load_model("/home/tristan/datasets/models/"
               "angular_dense_2Apf_epoch_10_16451353003083222301");
  sphere data[2]= {{0.,0.,0.,1.},{1.,0.,0.,1.}};
  load_atoms(data,2);
  point x[2] = {{0.,0.,1.},{1.,0.,0.}};
  eval_model(x,2);
  shutdown_julia(0);
  return 0;
}