Assume pixel format is FLOAT for all channels.

* relative_eta here is eta_t / eta_i */

* Reference: [Sampling the GGX Distribution of Visible Normals, Unity: Heitz ; 2018] */

* Sample the distribution anisotropic GGX of visible normals using * the spherical caps formulation which is slightly faster than the traditional *

* The sampled direction is returned in the local shading frame of the basis used for 'local_view_direction' */

* References: * [1] [Practical multiple scattering compensation for microfacet models, Turquin, 2019] [Main implementation] * [2] [Revisiting Physic

* Reference: https://stackoverflow.com/questions/8507885/shift-hue-of-an-rgb-color */

* Returns false if the given 'point_on_light' is occluded from the point of view of the * representative point of the grid cell given by 'hash_grid_c

TODO try just passing LightPresamplingParameters in there instead of everything individually

marker output is also

* Converts an XYZ value to the sRGB color space (linear, no gamma correction is applied) */

* pixel_converged is set to true if the given pixel has reached the noise * threshold given in render_data.render_settings.stop_pixel_percentage_conv

* Reference: * * [1] [Generalization of Adobe's Fresnel Model, Hoffman, 2023] * [2] [Adobe Standard Material, Technical Documentation, Kutz, Hasan,

Pixels per degree (PPD).

* Reference: https://en.wikipedia.org/wiki/Pairing_function */

* Returns true if the color has a NaN or INF component. * False otherwise */

* Returns true if the color has a negative component. * False otherwise */

Performs spatial filtering (and clamps the results) on both the reference and test image at the same time (for better performance). Filtering has been

This includes convolution (using separable filtering) of grayRefImage and grayTestImage for both edge and point filtering. In addition, it computes th

* Once the objects have been inserted in the hierarchy, this function computes * the bounding volume of all the node in the hierarchy

* Counts how many neighbors are eligible for reuse. * This is needed for proper normalization by pairwise MIS weights. * * A neighbor is not eligib

Main function for computing (the image metric called) FLIP between a reference image and a test image. * See FLIP-tool.cpp for usage of this fun

TODO make some simplification assuming that ReSTIR DI is never inside a surface (the camera being inside a surface may be an annoying case to handle)

* Finds the position of least signigicant bit set to 1 in a 32 bit unsigned integer. * Returs a value between 0 and 32 inclusive. * * Returns 0

* Flips the surface normals if necessary such that they are facing us. * * The normals are only flipped if some conditions are met, read the * c