IdTech4 Open Source
From The DarkMod Wiki
id Software has a long history making the code to their engines open source after a while. Even after being sold to Zenimax, this trend continued. The source code can be found on the officiel FTP server from id software.
idTech4 (the engine behind Doom 3 and Quake 4) has been open sourced. For The Dark Mod this means that we can fix a long list of issues that are currently not possible.
Here is a small, incomplete list, in no particular order. Some of these items have already been accomplished.
See also: Untethered_TDM.
ToDo
| Category | Issue | Possible outcome
|
|---|---|---|
| Graphic | No instancing. Multiple instances of the same model are all rendered one by one, instead of uploading it once and then telling the graphic card to render them all in one go. | Performance increase.
|
| Coding | Single-threaded coding. It may be possible to divide the sub-systems into separate execution threads for processing on separate cores once the source code is available. The game Prey has implemented this type of change to Id Tech 4. On the extreme side of this concept, OpenCL could be used to perform AI path-finding, or physics calculations. | Performance increase. Note: Some work has already been done by Revelator to port BFG's threading and to use Vector Fabric's renderer optimization via OpenMP. Currently unstable for Windows. |
| Graphic | Shadow Volumes are calculated on the CPU. Moving to a GPU calculated Shadow Volume or a Shadow Mapping technique like Parallel Split Variance Shadow Maps would allow more on-screen geometry with active lighting applied. Soft shadows would also be MUCH less expensive than the current Z-Buffer render and Post-Process method. | Performance increase. Higher poly-counts. More active lights. Soft shadows. |
| Graphic | No access to Vertex positions. For vertex shader or geometry shader (DX10+ hardware) operations like R2VB, Vertex Texturing or Tessellation the correct Vertex locations will be needed. Fur shaders would also be possible... | Performance increase, Fur shaders, Higher poly-counts. Note: Some of this functionality was probably already exposed but no ARB programmers could use it because it
would have increased the length of the entire interaction shader for anything "lit" though I don't believe anyone's tried calling a VFP from a material that includes the whole light interaction chain (which might've also worked. Nonetheless this remains open for improvement with robust native implementations. |
| Mapping | ||
| Graphic | Support for Light-Maps or Deluxe-Maps. While static lighting methods are not really an ideal usage for this engine, these features can have their utility. You could bake ambient bounced lights for areas that are predominantly statically lit. To make these methods more Doom 3 compatible, these could be implemented as an Ambient Light (3D texture support?) as a more robust extension of the Strombine method that is currently available (but very cumbersome). Maps made in other engines (without game specific assets) could be more easily migrated? The unused megatexture code is very similar to lightmap code. We could use the megatexture mod's reverse engineered vfp as a starting point and add PCX support. We'd want to correct the faulty UV space issues since this was built for terrains initially and stretches UV's vertically (etc). | Performance increase. Graphic enhancement. Migrated content? |
| Graphic | Local Cubemaps. (See HL2 env_cubemap.) Granting the ability to have reflective materials more accurately reflect their environment. It may be possible to setup scripted material fades (shaderparm?) to implement this with the current SDK framework but it would be yet another CPU dependent operation. The Skin changing in the LOD system could also be used perhaps? This one is definitely questionable about requiring Open Source, but is may be beneficial to set this up in the renderer. | More accurate reflective materials. Gold loot (etc). |
| Animation | GPU Skinning. Having the GPU stretch and transform the mesh would be tremendous performance enhancement for animations and would allow for extremely detailed facial animations like Team Fortress 2. | Thus far, this has proven to be the single largest performance jump from Vanilla Doom 3 to Doom 3 BFG. There is an open bounty to get this ported:
https://www.bountysource.com/issues/2883499-port-gpu-skinning |
| GUI | Font rendering is buggy (only alpha map, misses character 0xFF, no Unicode etc.). | Better looking GUI and easier translations. |
| GUI | The GUI is 4:3 (640x480), but covers always the entire screen and is thus distorted on 16:9. | GUI looks better on 16:9 and on multi-monitor setups. |
| Sound | Determine why "no_dups" doesn't always work and correct it. Remove the SDK workaround for this problem. | Better sound randomization. |
| Entities | Correct the occasional bad "trace" the engine performs. | Eliminate spatial relationship problems. |
| Mapping | Eliminate as many hard-coded limits as possible. | Reduce mapper frustration. |
| Coding | Grant the ability to create NEW material keywords | Allow per-material effects that do not require every surface to include the keyword stage. Example: Fur Shaders, Anisotropic Material, etc. Keywords are already possible, the intended design would allow defined material fragment programs to modify the surface interaction. A simple workaround is to do it the way Prey
does it and include the whole light chain in the vfp. I believe this should be possible even now. Changes to allow surface shaders to define the light interaction that renders against them are in https://github.com/nbohr1more/Doom3_divide (see shaderSpectrum keyword changes) |
| Graphic | Add the ability to have multiple dynamic LightFalloffImage stages | Use Texture Sheets to render multiple light passes and maximize the work allocated to each light render. (Less waste.) Animated light texture effects can have all 3 axis of movement. |
Done
| Category | Issue | Explanation |
|---|---|---|
| Sound | Vis-portals do not diminish sound while opened. | Open vis-portals would also diminish sound. Sound loss would scale with the size of the portal up to a certain size, where it would then be large enough to have not require sound loss. |
| GUI | Language support (via 'sys_lang') is limited to a fixed set of languages. Also the GUI refues to use non-ASCII strings for choices or values. | Unicode support, better i18n support and UI translateable to more languages. |
| Graphic | The lightgem needs to render to a texture and transfer this texture back to the CPU via a complicated in-memory file scheme. | Performance increase. |
| Graphic | No access to the Z-Buffer. Without access to this data, any depth-based fragment shader effects like SSAO or Depth of Field require an extremely expensive z-pass render. | Performance increase. More effect options. |
| Sound | EAX does not work under Linux. | Works there, too. (uncommited branch by Serpentine) |
| Graphic | LATC2 Normal Map compression support (AKA OpenGL 3Dc) See: http://www.opengl.org/registry/specs/EXT/texture_compression_latc.txt ... Reduced Normal Map memory requirements both on the storage volume and in graphic RAM. | Performance increase. Graphic enhancement. Bigger maps for less RAM. Code is committed, Normal Maps were not converted due to concerns about Intel compatibility. Serpentine was last looking into RGTC as an alternative. |
| Graphic | Add keywords for Cubemap light method and 3D texture light methods | Extend the Doom 3 light method to more projection types without the need to replace or upgrade existing maps and assets. Example implementation already worked-out in tracker 3881 with a generalized fix to allow any arbitrary interaction ARB to run per-light based on the material def |
