NNM is a "No-Nonsense", C++ 17, single-header-only Math library geared towards graphics programming. Think of it as a lightweight, more readable alternative to a library like glm and inspired by the OOP style of Godot's math library.
#define NNM_BOUNDS_CHECK // optional bounds checking
#include <nnm/nnm.hpp>
// Compute transform of an object in 3d space
nnm::Transform3f calculate_object_transform(
const nnm::Vector3f& position, const nnm::Vector2f& rotation)
{
constexpr float aspect_ratio = nnm::Vector2(16.0f, 9.0f).aspect_ratio();
constexpr float fov = nnm::radians(90.0f);
const auto projection
= nnm::Transform3f::from_perspective_left_hand_neg1to1(
fov, aspect_ratio, 0.1f, 100.0f);
const auto view
= nnm::Transform3f::from_translation(position)
.rotate_axis_angle(nnm::Vector3f::axis_y(), rotation.y)
.rotate_axis_angle(nnm::Vector3f::axis_x(), rotation.x);
const auto model = nnm::Transform3f::from_scale(nnm::Vector3f::all(0.5f));
return model.transform(view).transform(projection);
}
A full list of all functions/classes can be found in docs/reference.md.
NNM provides classes for:
-
Vectors
Vector2
Vector2i
Vector3
Vector3i
Vector4
-
Matrices
Matrix2
Matrix3
Matrix4
-
Transformations
Basis2
Transform2
Basis3
Transform3
-
Misc.
Quaternion
where the trailing number represents the "dimension" of the class. Most of these classes, except for the integer
variants, are templated for any Real
value such as float
or double
. Aliases are created for those cases for
example Vector2f
and Vector2d
for float
and double
respectively.
In addition to these classes, NNM provides for a number of standalone math functions such as lerp
, clamp
,
approx_equal
, etc.
Other notable features/qualities of NNM include:
- No template magic, just the basics for your choice of
Real
value. - Extremely readable. There are almost no macros and light use of templates.
- Minimal abbreviations. It's the 21st century, we can afford readable function names!
- Optional bounds checking for accessors with an optional
#define NNM_BOUNDS_CHECK
before including. - No external dependencies! Even minimal use of the standard library to just
<cmath>
,<functional>
,<optional>
, and<stdexcept>
if optional bounds checking is enabled. - No special SIMD instructions or compiler intrinsics. This makes NNM extremely portable. I personally have faith in modern compilers to auto-vectorize when necessary.
std::hash
specializations forVector2i
andVector3i
.begin()
andend()
iterators for looping through classes with a ranged-for loop.
The easiest way would be to just copy the include/nnm/nnm.hpp
file directly into your project and just #include
it
to use!
Another method is to add the project as a CMake submodule by copying the repository to a directory and adding:
add_subdirectory(external/path/to/nnm)
target_link_libraries(your_project PRIVATE nnm)
You might be asking, "What is up with all these variations of perspective/orthographic projection methods? Which one do I use??" This depends on the graphics API you are using as well as the coordinate system you choose for your application. There are two considerations to be aware of, the handedness and the normalization of the graphics API. Here is a short list of common graphics APIs' normalized device coordinate (NDC) conventions:
- OpenGL/WebGL/WebGPU/Metal: left-handed, -1 to 1
- DirectX: left-handed, 0 to 1
- Vulkan: right-handed, 0 to 1
This means if your application and graphics API use the same coordinate system, you just need to choose the method with the correct handedness and normalization corresponding to that graphics API. However, if your application and chosen graphics API have mismatching handedness, you must flip the z-axis. For example, if you are using OpenGL (which is left-handed) but wish to use right-handed coordinates in your application, you would still use the projection method corresponding with OpenGL's convention of left-handed and -1 to 1 normalization but just need to flip the z-axis to account for the switched handedness:
const auto proj = nnm::Transform3f::from_perspective_left_hand_neg1to1(...).scale_local({1, 1, -1});
After cloning the repository, configure the project with CMake ensuring to enable building tests:
cd nnm
cmake -S . -B build -DNNM_BUILD_TESTS=ON
cmake --build build
NNM is licensed under the MIT license. See LICENSE.txt
for full license.