N-archive/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu.git synced 2024-11-30 02:32:46 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #2767 from yuriks/quaternion-flip-comment

Browse source 
OpenGL: Update comment on AreQuaternionsOpposite with new information
This commit is contained in:
Yuri Kunde Schlesner 2017-06-12 16:31:55 -07:00 committed by GitHub
commit 791cd14c8d
1 changed files with 11 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -182,19 +182,22 @@ RasterizerOpenGL::RasterizerOpenGL() : shader_dirty(true) {
RasterizerOpenGL::~RasterizerOpenGL() {}
/**
* This is a helper function to resolve an issue with opposite quaternions being interpolated by
* OpenGL. See below for a detailed description of this issue (yuriks):
* This is a helper function to resolve an issue when interpolating opposite quaternions. See below
* for a detailed description of this issue (yuriks):
*
* For any rotation, there are two quaternions Q, and -Q, that represent the same rotation. If you
* interpolate two quaternions that are opposite, instead of going from one rotation to another
* using the shortest path, you'll go around the longest path. You can test if two quaternions are
* opposite by checking if Dot(Q1, W2) < 0. In that case, you can flip either of them, therefore
* making Dot(-Q1, W2) positive.
* opposite by checking if Dot(Q1, Q2) < 0. In that case, you can flip either of them, therefore
* making Dot(Q1, -Q2) positive.
*
* NOTE: This solution corrects this issue per-vertex before passing the quaternions to OpenGL. This
* should be correct for nearly all cases, however a more correct implementation (but less trivial
* and perhaps unnecessary) would be to handle this per-fragment, by interpolating the quaternions
* manually using two Lerps, and doing this correction before each Lerp.
* This solution corrects this issue per-vertex before passing the quaternions to OpenGL. This is
* correct for most cases but can still rotate around the long way sometimes. An implementation
* which did `lerp(lerp(Q1, Q2), Q3)` (with proper weighting), applying the dot product check
* between each step would work for those cases at the cost of being more complex to implement.
*
* Fortunately however, the 3DS hardware happens to also use this exact same logic to work around
* these issues, making this basic implementation actually more accurate to the hardware.
*/
static bool AreQuaternionsOpposite(Math::Vec4<Pica::float24> qa, Math::Vec4<Pica::float24> qb) {
Math::Vec4f a{qa.x.ToFloat32(), qa.y.ToFloat32(), qa.z.ToFloat32(), qa.w.ToFloat32()};