* use static_assert directly. Raze is C++17, no need for that macro shit.
* removed CONSTEXPR - I seriously fail to see the use here, many of the functions marked as CONSTEXPR cannot possibly even be constant evaluated so the declaration makes no sense. Removed most of these and replaced the valid ones with the official constexpr keyword.
* got rid of EDUKE_PREDICT_FALSE - this makes zero sense in script parsing code, at best it will save a few microseconds. Clean code wins.
* replaced Blrintf with xs_CRoundToInt. Shitty name is shitty name, even if derived from POSIX.
* replaced Bstr*casecmp with str*icmp. As these get defined in the CMake project based on actual compiler checks they are preferable here.
* removed lots of other stuff that is not needed with a minimum compiler requirement of C++17.
* Remove fix16.h/cpp and utilise library from m_fixed.h.
* Extend m_fixed.h with two inline functions for int to/from float operations.
* Replace fix16_floor operations with those from xs_Float.h
* Replace multiple Q16.16 conversions from 0 to just be 0.
* Replaced all found in-game bit-shifts and multiplications/divisions with inline functions from m_fixed.h
* Replaced many casts of FRACUNIT as double in SW's panel.cpp as it is converted to double by way of type promotion.
* Fixed missed precision fixes in SW's panel.cpp where some types weren't declared correctly.
* Replaced 100+ `Cos()/Sin() >> 16` operations for Blood with inline functions `CosScale16()/SinScale16()`.
In particular this means to remove the option to disable widescreen aspect ratios. The way this was handled makes no sense with the current render backend.
The aspect ratio code will have to be redone entirely to properly obey the backend's settings.
* Changes performed in 0bd460d9e3 didn't take into account xdimenscale and viewingrangerecip like the days of old and this wasn't picked up in d80a32d379 or d80a32d379, where the applied fixes only appeared to work because they worked for me at 2560x1440p.
* Output from PrintVis() in 9dfd3ddd02 showed resulting global visibility as 0.078125.
* Following 0bd460d9e3, resulting global visibility shown as 0.041667.
* Scaling g_visibility by (8.f / 15.f) (0.533333) restores resulting global visibility to 0.078125.
There are effectively two states - the one in the backend and a local one in the drawer for the render list which is supposed to eliminate some of the more costly repeated calls.
This higher level state was cached globally, which did not work anymore because the real render state could be changed elsewhere without this code realizing it.
All this means that the render list drawer must create a new state cache for each call and also must apply its current pending render state before leaving to ensure that everything is properly reset.
The main reason here is that the scene specific part contains a projection dependent component which would be a problem when transitioning to GZDoom's code. The global viewpoint data already has a field for such a factor so now that gets used.
This also means a significant simplification of the visibility code in Polymost and the removal of several global variables.
Models are inoperable right now anyway so this would never get called, but it does a few things that would cause problems with refactoring the backend code.
They depend on a deleted texture not writing to the depth buffer, but other parts in the engine like ROR surfaces depend on them writing a proper depth buffer value, so for now there is a global variable that allows to exclude a single tile from ever getting rendered.
I have no idea why this needs to be different than in EDuke32, but without always clearing the depth buffer before rendering a scene viewpoint the game will glitch like crazy.
It now picks the minimum of the current formula and the one from before June 2017 - the current one was causing problems with sprites in the distance so now the old one is used as an upper bound.
Since the same word gets used in text messages and local variables in the game code it is easier this way to search for it and facilitate its transition to the translation table management in PaletteContainer.
This removes the last active use case for the software renderer and allows rendering the camera views at a higher resolution.
For Shadow Warrior this necessitated a split of JS_DrawMirrors, because it processed cameras and mirrors in the same loop which cannot be done with the hardware renderer.
This way the view matrix will remain constant per scene and an update of the model matrix is only needed for rendering a model or voxel, reducing the total amount of generated matrices to the point where they can be written to a buffer instead of constantly uploading them as uniforms.
The code needs to be refactored anyway to allow better crosshair control but this also seemed to affect global render state, considering how poorly it was all implemented.
Not tested and probably still not working correctly, the entire feature was a nearly impenetrable mess of convoluted code. This definitely needs verification with multiple mods using it.
