Suggest if you are happy for any contributions also, creating a to do list on your API. Happy to contribute and grow it.

I know your time is limited and the community would be happy to do add to your code base.

Feel free to close once reviewed.

Hi,

Using the unproject to determine a raycast into the screen (0, 1) range.

I get values starting near 0 to a very large value. The implementation differs from the Sharpdx version where I call

Vector3 screenSpacePoint = new Vector3(mousex, mousey, 0);

 Matrix4x4.Invert(camera.WVPMatrix, out Matrix4x4 inverse);

Vector3 point = Unproject(camera.ViewPort, screenSpacePoint, inverse);

I tested this against the vortice implementation and getting very different result.

The underlying code for the unproject is:

      public static Vector3 Unproject(Viewport vp,   Vector3 source,  Matrix4x4 matrix)
        {
            Vector3 vector_ = new Vector3();

            vector_.X = (((source.X - vp.X) / (vp.Width)) * 2f) - 1f;
            vector_.Y = -((((source.Y - vp.Y) / (vp.Height)) * 2f) - 1f);
            vector_.Z = (source.Z - vp.MinDepth) / (vp.MaxDepth - vp.MinDepth);

            float a = (((vector_.X * matrix.M14) + (vector_.Y * matrix.M24)) + (vector_.Z * matrix.M34)) + matrix.M44;
            //Vector3.Transform(ref vector, ref matrix, out vector);

            vector_ = Vector3.Transform(vector_, matrix);

            if (!IsOne(a))
            {
                vector_ = (vector_ / a);
            }

            return vector_;
        }

        public const float ZeroTolerance = 1e-6f; // Value a 8x higher than 1.19209290E-07F

        public static bool IsOne(float a)
        {
            return IsZero(a - 1.0f);
        }

        public static bool IsZero(float a)
        {
            return Math.Abs(a) < ZeroTolerance;
        }

Happy to provide more samples, but this maybe down to my use fundamentally of the current unproject call.

Should have field offset of 4 and 6 for Z and W respectively.

Great library so far! Having fun converting from sharpdx

+-*/~%^&| << >> Length() AsBytes(Span) ToArray(Span) xyzw get;set;

A Matrix5x4 instance cannot be created because the constructor does not exist.
Just like Matrix3x3 and other Matrix structs, I want the following constructors.
Matrix5x4(float value) Matrix5x4(float m11, float m12, ...)

Like this constructor:

Line 68 of BoundingBox.cs should read _min = value;

This lead to some hilarious bugs on my end 😆

I suspect the Box members are reversed from what they should be. Height should be Bottom - Top, not Top - Bottom. And Depth should be Back - Front, not Front - Back.

At least if you want to match the D3D box structure: https://docs.microsoft.com/en-us/windows/win32/api/d3d12/ns-d3d12-d3d12_box

Description (optional)

MCG0017: MCG0017:NotWinRTType Cannot marshal generic Windows Runtime types with a non Windows Runtime type as a generic type argument. Check the generic arguments of type 'System.Nullable`1<Vortice.Mathematics.Color4>'

Actual Behavior

It runs in debug mode, but won't compile in release mode

Hi, for functional parity, is there consideration of pulling the Ray class from Sharpdx across into Vortice Mathematics?

I've been doing a fair few hours work around why unproject seems to lose accuracy as you get further from 0. The obvious answer is floating point accuracy.

I did play around with double, but essentially the issue is that due to the Matrices also being float, you simply can't just create data out of thin air. (float is substantially more accurate closer to 0 anyway obviously)

I did though figure out a novel approach that will work in most circumstances that drastically improved the accuracy.

What the approach does below is translate the point to the origin, maintaining the orientation (scale other than 1 will not work atm).

Once the unproject is complete, I add the translation back on, which improves the accuracy substantially.

The view matrix needs to be generated without translations, and you can't eliminate from the WVP obviously.

Proposal, that a new unproject in the view class is created (I can do the pull request) that takes in Projection, View and ScreenPosition and returns the world position.

The accuracy is improved by about 90% once your world coordinates get > 1000. I've tested it multiple times and found that the ray projection resolution is now closer to the screen resolution, before, its sub 10% at high world coordinates.

///camera function

        public Matrix4x4 GenerateNonTranslatedWVP()
        {
            Matrix4x4 viewMatrix  = Matrix4x4.CreateLookAt(_position, _targetPosition, Constants.s_upVector);

            viewMatrix.Translation = Vector3.Zero;

            Matrix4x4 projectionMatrix = Matrix4x4.CreatePerspectiveFieldOfView(c_gameFov, _viewport.AspectRatio, 0.01f, _unitDepth);

            return viewMatrix * projectionMatrix;
        }

/// test code

     public static Vector3 Determine3dPointFrom2d(Camera3d camera, in Vector3 screenSpacePoint)
        {
            Matrix4x4 nonTranslated = camera.GenerateNonTranslatedWVP();

            Matrix4x4.Invert(nonTranslated, out Matrix4x4 inverse);

            Vector3 nonTrans_ = camera.ViewPort.Unproject(screenSpacePoint, inverse);

            nonTrans_ += camera.Position;

            return nonTrans_;// Unproject(camera.ViewPort, screenSpacePoint, inverse);
        }

Thoughts?

Error CS0433: The type 'Half' exists in both 'Vortice.Mathematics, Version=1.4.0.0, Culture=neutral, PublicKeyToken=5431ec61a7e925da' and 'System.Runtime, Version=5.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a'

We've noticed that your package on NuGet.org is missing a README file.

Why READMEs are Important

Our customer research indicates that one of the top problems that package consumers face is insufficient package documentation such as README files. Adding a README file to your package will help users quickly understand what the package is and what it does. Since your README will be the first impression for users when they view your package on NuGet.org, it is crucial for authors to write and include high-quality READMEs for their packages.

Get Started with READMEs

How to Add a README

If you're new to NuGet READMEs, follow guidance on how to add one to your package.

How to write high quality of READMEs

Follow a blog post with README best practices and a template file to help you get started.

We Value Your Feedback

We value your feedback. If you encounter any issues or have suggestions, please reply to this issue.

Thank you for your contribution to the NuGet community.

Vortice.Mathematics uses System.Runtime.CompilerServices.Unsafe in version 6.0.0 as a dependency. Would it be possible to downgrade the version to 4.7.0?

This will increase compatibility with other products that also use System.Runtime.CompilerServices.Unsafe as a dependency.

In the versions <= 1.9.143 the library System.Runtime.CompilerServices.Unsafe was not used at all. There the compatibility had been higher.

Description (optional)

Reproduction Steps

Rect rect = new ();
Console.WriteLine(rect);

Expected Behavior

Rect { X = 0, Y = 0 }, { Width = 0, Height = 0 }

Actual Behavior

Rect { X = 0, { Y = 0, { Width = 0, Height = 0 }

Hi, long time since we spoke 😄!

Wanted to share this small nitpick with you.

The method public readonly bool Intersects( in Vector3 v0, in Vector3 v1, in Vector3 v2, out Vector3 pointInTriangle) on the Ray struct has the following disclaimer:

This ONLY works on CW wound triangles.

But what is clock-wise depends on the unit system the user is using. As a DirectX user, I assumed Z- is forward. But it looks like this code uses the OpenGL convention in which Z+ is forward. Which means that every triangle that I believe to be clockwise, is actually counter-clockwise.

A small proof:

    [Fact]
    public void RayTriangleIntersectionTest()
    {
        var position = new Vector3(2.75f, 1000.0f, 1.5f);
        var direction = new Vector3(0.0f, -1.0f, 0.0f);
        var ray = new Ray(position, direction);

        // nw---ne
        //  \    |
        //   \   |
        //    \ se

        var nw = new Vector3(2, 18, 1);
        var ne = new Vector3(3, 18, 1);
        var se = new Vector3(3, 18, 2);

        var hitCW = ray.Intersects(nw, ne, se, out var pitCW);
        var hitCCW = ray.Intersects(se, ne, nw, out var pitCCW);
        True(hitCW);
        False(hitCCW);
    }

It would be nice to add a small clarification, so that users double check what convention they are using.

Hi,

I found a trivial mistake in Color.cs.

        /// <summary>
        /// The alpha component of the color.
        /// </summary>
        [FieldOffset(2)] // <- 3?
        public byte A;

Thank you for your great work.