How to Marshal a variable sized array of ON_3dPoint to C#

Replies to This Discussion

Permalink Reply by Andrew Heumann on October 25, 2012 at 4:16pm

Two recommendations:

1. use the non-legacy scripting components, so use RhinoCommon types - this means using Point3d instead of ON_3dPoint, and Line instead of ON_Line. 

2. Use the built-in System.Collections types. Rather than going through all the mess above, simply declare  a List<Point3d> and be done with it. 

 

For example:

List<Point3d> MakeSomePoints(Line L, int num){
List<Point3d> results = new List<Point3d>();
for(int i = 0;i <= num;i++){
results.Add(L.PointAt(i / (double) num));
}
return results;
}

Permalink Reply by Ian Chan on October 26, 2012 at 9:55am

I'm trying to marshal some points which are generated by MarchingCube from c++ to c#
//PInvoke in c#:
//static ON_3dPoint* ComputeMeshPoints(int resX, int resY, int resZ, double data[], double iso, ON_3dPoint* minCorner, ON_3dPoint* maxCorner, int* outNum);
[DllImport("SomniumGrasshopper_C.dll", EntryPoint = "MarchingCubeComputeMeshPoints", CallingConvention = CallingConvention.Cdecl)]
public static extern IntPtr MarchingCubeComputeMeshPoints(int resX, int resY, int resZ, IntPtr data, double iso, ref Point3d minCorner, ref Point3d maxCorner, ref int outNum);
_______________
//Here's some of the code in the component:

double[] dataArray = data.ToArray();
fixed (double* dataPtr = dataArray)
{
Point3d minCorner = box.BoundingBox.Min;
Point3d maxCorner = box.BoundingBox.Max;
int num=0;
IntPtr retrievePtr = SomniumSolver.MarchingCubeComputeMeshPoints(resX, resY, resZ, (IntPtr)dataPtr, isoValue, ref minCorner, ref maxCorner, ref num);
IntPtr cleanUpPtr = retrievePtr;
Mesh mesh = new Mesh();
for (int i = 0; i < num; i++)
{
mesh.Vertices.Add((Point3d)Marshal.PtrToStructure(retrievePtr,typeof(Point3d)));
retrievePtr = (IntPtr)((int)retrievePtr + 0x18);
}

SomniumSolver.MarchingCubeDisposePoints(cleanUpPtr);

for(int i =0;i< mesh.Vertices.Count;i+=3)
{
mesh.Faces.AddFace(i, i + 1, i + 2);
}
if (flag)
{
mesh.Vertices.CombineIdentical(true, true);
mesh.FaceNormals.ComputeFaceNormals();
mesh.Normals.ComputeNormals();
}
DA.SetData(0, mesh);
DA.SetData(1, num);
}
____________________
It involves a lot of copying data into functions so that it's slower than doing all in C#.

Permalink Reply by David Rutten on October 26, 2012 at 10:15am

Where is this C++ code running from?

--

David Rutten

david@mcneel.com

Poprad, Slovakia

Permalink Reply by Ian Chan on October 27, 2012 at 3:55am

ON_3dPoint* MarchingCubeComputeMeshPoints(int resX, int resY, int resZ, double data[], double iso, ON_3dPoint* minCorner,ON_3dPoint*maxCorner, int* outNum)

{

int resX_1 = resX+1, resY_1 = resY+1, resZ_1 = resZ+1;

/*if((resX_1)*(resY_1)*(resZ_1) != sizeof(data) / sizeof(data[0]))

{

*outNum = 0;

return NULL;

}*/

int sliceXY = (resX_1)*(resY_1);

double intervalX = 1.0/resX;

double intervalY = 1.0/resY;

double intervalZ = 1.0/resZ;

ON_BoundingBox box = ON_BoundingBox(*minCorner,*maxCorner);

ON_Box refBox = ON_Box(box);

//Get corners of a single voxel

ON_3dPoint corners[8];

corners[0] = refBox.PointAt(0.0,0.0,0.0);

corners[1] = refBox.PointAt(intervalX,0.0,0.0);

corners[2] = refBox.PointAt(intervalX,intervalY,0.0);

corners[3] = refBox.PointAt(0.0,intervalY,0.0);

corners[4] = refBox.PointAt(0.0,0.0,intervalZ);

corners[5] = refBox.PointAt(intervalX,0.0,intervalZ);

corners[6] = refBox.PointAt(intervalX,intervalY,intervalZ);

corners[7] = refBox.PointAt(0.0,intervalY,intervalZ);

double d[8];

ON_3dPoint intersectionPoints[16];

int minCornerIndex= 0;

vector<ON_3dPoint> pts;

for (int z = 0; z < resZ; z++)

    {

        for (int y = 0; y < resY; y++)

        {

            for (int x = 0; x < resX; x++)

            {

minCornerIndex = z * sliceXY + y * resX_1 + x;

                d[0] = data[minCornerIndex];

                d[1] = data[minCornerIndex + 1];

                d[2] = data[minCornerIndex + 1 + resX_1];

                d[3] = data[minCornerIndex + resX_1];

                d[4] = data[minCornerIndex + sliceXY];

                d[5] = data[minCornerIndex + 1 + sliceXY];

                d[6] = data[minCornerIndex + 1 + resX_1 + sliceXY];

                d[7] = data[minCornerIndex + resX_1 + sliceXY];

ON_3dVector translate = refBox.PointAt(x * intervalX, y * intervalY, z * intervalZ) - corners[0];

int num = VoxelBox::GetFaces(corners, d, iso, intersectionPoints);

if (num > 0)

                {

                    for (int i = 0; i < num; i += 3)

                    {

                        pts.push_back(ON_3dPoint(intersectionPoints[i].x + translate.x, intersectionPoints[i].y + translate.y, intersectionPoints[i].z + translate.z));

                        pts.push_back(ON_3dPoint(intersectionPoints[i + 1].x + translate.x, intersectionPoints[i + 1].y + translate.y, intersectionPoints[i + 1].z + translate.z));

                        pts.push_back(ON_3dPoint(intersectionPoints[i + 2].x + translate.x, intersectionPoints[i + 2].y + translate.y, intersectionPoints[i + 2].z + translate.z));

                        *outNum += 3;

                    }

                }

}

}

}

ON_3dPoint* result = new ON_3dPoint[pts.size()]();

for(unsigned int i =0;i<pts.size();i++)

{

result[i].Set(pts[i].x,pts[i].y,pts[i].z);

}

return result;

}

I think it's slow in tow places.

1. using vector<> to hold all the points first, then copy them to the array. 

2. in c#, List<double> .ToArray

The first time I tested the code in Grasshopper, a warning "can''t find entry point blahblah..." came up. Then I did some google, used extern "C"__declspec(dllexport) and a .DEF file. If more code is added in, updating the .DEF file manually will be annoying. 

How is RhinoCommon manage all the exported functions?

 

Permalink Reply by David Rutten on October 27, 2012 at 6:21am

RhinoCommon consists of two parts, rhcommon_c.dll which is a pure C++ library that exposes Rhino SDK features as easily PInvokeable functions and RhinoCommon.dll which is a pure C# library which invokes these methods.

We have a small console application which runs through the rhcommon_c source code and collects all invokable methods, then it creates a giant C# class which exposes all of these in a way that .NET can call.

There are also a bunch of special classes in the Rhino.Runtime.InteropWrappers namespace that help us ferry data from C++ to C# and vice versa.

Things like ON_3dPoint etc. are somewhat special but basically we make sure that the memory layout of Point3d is the same as ON_3dPoint so we can pass a pointer.

Maybe it would be easier if you created arrays of doubles rather than ON_3dPoint? {x0,y0,z0,x1,y1,z1,x2,y2,z2,...}  Then you can just iterate in steps of 3 and build Point3d structs from the coordinates.

--

David Rutten

david@mcneel.com

Poprad, Slovakia

Permalink Reply by Ian Chan on October 28, 2012 at 12:48pm

Thanks, David. Marshalling double instead of Point3d struct can close the gap a little bit. //double[] dataArray = data.ToArray(); double[] dataArray = (double[])typeof(List) .GetField("_items", BindingFlags.NonPublic | BindingFlags.Instance) .GetValue(data); I googled a way to convert List to array without copying, but there is no significant improvement. Reflection is not a good approach.

Permalink Reply by Steve Baer on October 28, 2012 at 4:17pm

Hi Ian,

Here's some sample code that may clarify how I would go about performing the pInvoke.

https://gist.github.com/3970130

Let me know if you have any questions.

Thanks,

-Steve

Permalink Reply by Ian Chan on December 1, 2012 at 5:04pm

Thanks a lot.
Rhino.Runtime.InteropWrappers namespace is very handy!