GLM 0.9.2.0 provides many exciting features which first of all is the CUDA copiler support so that GLM can be used within a CUDA kernel. This is possible only by defining GLM_FORCE_CUDA before including glm/glm.hpp in the CUDA Kernel source code.
New experimental extensions are also available. GLM_GTX_noise is based on the work by Stefan Gustavson and Ashima Arts on WebGL-noise which adds perlin and simplex noise functions in a pure computational way. If everything goes well, this extension will be promoted to GLM core as implementation of the GLSL noise functions.
GLM_GTX_ulp provides functions to evaluate the number of ULPs between two floating-point values which gives a way to mesure the accuracy of a function compare to a reference function. In the future, this extension could be used to update the documentation of function and gives the level of accuracy they provide.
Finally, GLM_GTX_matrix_interpolation is an extension developed by Ghenadii Ursachi to extract axis and angle of a 4 by 4 matrix but also directly interpolate between two matrices. This extension works only on linear transformed matrices.
Last but not least: with the version 0.9.2, GLM supports CTest to manage GLM test suite which makes tests much easier to maintain. The number of tests have significantly increase, even if a lot more tests are necessary.
GLM 0.9.2 is not backward compatible with GLM 0.9.1 with only one point: Quaternion SLERP interpolation! In GLM 0.9.1 the function 'mix' always interpolate using the short rotation path but it's no longer the case. The function 'shortMix' has to be called for short path interpolation. 'fastMix' will interpolate two quaternions using a normalized linear quaternion interpolation with non-constant rotation speed.