This commit provides useful functionality for path tracing rendering core.
1) Graphic3d_RenderingParams class was extended with additional AdaptiveScreenSampling option (disabled by default).
If this option is enabled, path tracing tries to adjust the number of samples for different screen areas.
In this way, the more complex areas (from the point of light conditions) are sampled more intensively,
while the simple areas are sampled very rarely.
For example, caustics and glossy reflections are typical candidates for more precise sampling.
In general, this allows to equalize image convergence and not to waste resources for already converged areas.
It is also possible to visualize sampling densities by enabling ShowSamplingTiles option
(activating and deactivating this option does not affect on the accumulated image).
2) Mixing OpenGL and ray-tracing output has been changed.
Now blending is performed using OpenGL functionality, while ray-tracing shaders only output correct Z-value.
Test case bugs vis bug27083 has been updated,
since the alpha value is now correctly set by Ray-Tracing to 1, opaque.
In frames of this issue binary BVH tree produced by building algorithms was collapsed into 4-ary BVH (QBVH).
The BVH traversal code in GLSL was modified to process such trees correctly.
This allows to implore thread coherence, decrease BVH memory consumption (~2 times), and use traversal stack of the half size.
As a result, ray tracing scalability is improved, as well as rendering performance. For various setups, speedup is 12-18%.
View-projection matrix was added to raytrace shaders (as uniform) in order to compute correct depth values for OpenGL.
For path tracing the additional depth buffer sampler was added to Display.fs program. It allows propagation of depth values from internal FBO to resulting FBO.
The old approach of mixing of OpenGL and ray-tracing graphics was kept in order to keep correct blending of transparent ray-traced objects with non-transparent OpenGL objects.
Replace 64-bit handles of bindless textures by uvec2 type in GLSL code for compatibility with AMD drivers.
OpenGl_View::initProgram() - fix NULL-dereference.
Backside triangles are handled correctly by implementing two-sided lighting model. Ray-tracing shader was optimized (up to 25% performance increase).
Test case for CR25833
This patch eliminates 3 samplers used in ray-tracing mode:
//! Texture buffer of data records of bottom-level BVH nodes.
Handle(OpenGl_TextureBufferArb) myObjectNodeInfoTexture;
//! Texture buffer of minimum points of bottom-level BVH nodes.
Handle(OpenGl_TextureBufferArb) myObjectMinPointTexture;
//! Texture buffer of maximum points of bottom-level BVH nodes.
Handle(OpenGl_TextureBufferArb) myObjectMaxPointTexture;
Serialized data contained in corresponding texture buffers were added to global scene buffers:
//! Texture buffer of data records of high-level BVH nodes.
Handle(OpenGl_TextureBufferArb) mySceneNodeInfoTexture;
//! Texture buffer of minimum points of high-level BVH nodes.
Handle(OpenGl_TextureBufferArb) mySceneMinPointTexture;
//! Texture buffer of maximum points of high-level BVH nodes.
Handle(OpenGl_TextureBufferArb) mySceneMaxPointTexture;
This modifications leads also to small performance improvement (~2%) due to higher texture cache efficiency. Some modifications in traversal function (GLSL code) also improve performance (~3%).
OpenGl_GraphicDriver::Redraw() - do not recompute structures more than required
OpenGl_Workspace::Raytrace() - bind proper FBO before clearing it
Visual3d_View::Redraw() - perform automatic 2nd redraw on device lost
Refraction index passed to raytrace shader.
For transparent objects implemented ray refraction according to refraction index.
Transparent materials modified to look better with refractions.
