Quake engine

Simplified process of reducing map complexity in Quake

Quake was the first true-3D game to use a special map design system that preprocessed and pre-rendered some elements of the 3D environment, so as to reduce the processing required when playing the game on the 50–75 MHz CPUs of the time. The 3D environment in which the game takes place is referred to as a map, even though it is three-dimensional in nature rather than a flat 2D space. The map editor program uses a number of simple convex 3D geometric objects known as brushes that are sized and rotated to build the environment. The brushes are placed and oriented to create an enclosed, empty, volumetric space, and when the design is complete, the map is run through the rendering preprocessor. The preprocessor is used to locate two types of empty space in the map: the empty space enclosed by brushes where the game will be played and the other empty space outside the brushes that the player will never see. The preprocessor then strips away the back faces of the individual brushes, which are outside the game space, leaving only a few polygons that define the outer perimeter of the enclosed game space.

Generally, once a map has been preprocessed, it cannot be re-edited in a normal fashion because the original brushes have been cut into small pieces. Instead, the original map editor data with the brushes is retained and used to create new versions of the map. But it is possible to edit a processed map by opening it in a special vertex editor and editing the raw vertex data, or to add or remove individual triangle faces. Though difficult, this technique was occasionally used by cheaters to create windows in walls, to see normally hidden enemies approaching from behind doors and walls, and resulted in an anti-cheat mechanism used in recent 3D games that calculates a checksum for each file used in the game, to detect players using potentially hacked map files.

A processed map file can have a much lower polygon count than the original unprocessed map, often by 50–80%. On the 50–75 MHz PCs of the time, it was common for this pruning step to take many hours to complete on a map, often running overnight if the map design was extremely complex.

This preprocessing step cannot work if there are any small holes or "leaks" that interconnect the interior game space with the exterior empty space, and it was common for complex map-building projects to be abandoned because the map designer could not locate the leaks in their map. To prevent leaks, the brushes should overlap and slightly interpenetrate each other; attempting to perfectly align along the edges of unusually shaped brushes on a grid can result in very small gaps that are difficult to locate. Modern level editors feature leak detection, alleviating this issue.

The open sky in Quake maps is in fact not open, but is covered over and enclosed with large brushes, and textured with a special skybox texture, which is programmed to use sphere mapping, and thus always looks the same from any viewing position, giving the illusion of a distant sky.

Quake also incorporated the use of lightmaps and 3D light sources, as opposed to the sector-based static lighting used in games of the past. id Software's innovation has been used for many 3D games released since, particularly first-person shooters, though id Software switched to a Unified lighting and shadowing model for Doom 3 (however, they switched back to a lightmapped or semi-lightmapped method starting with RAGE). After a map had been pruned of excess polygons, a second preprocessing system was used to precalculate and bake the lightmaps into the game map to further reduce load on the CPU when playing the game. However, full light processing could take an extremely long time, so for the initial map design process, lesser-quality light processing could be done, but at the cost of creating a jagged stair-step lightcast around lights.

To further decrease the workload of 3D rendering, a mechanism was developed to section off large regions of the map not currently visible to the player, so the engine would not need to render those unseen spaces. A 3D rendering engine without any such optimizations must draw every part of the world and then attempt to determine which polygons are the closest, then hide all polygons located behind these closest polygons (a technique known as Z-buffering). Just because a polygon is not visible does not mean it is excluded from the scene calculations.

The Quake engine was optimized specifically to obviate this problem. The engine could be told ahead of time to not calculate rendering for all objects in any space out of the player's view, greatly reducing the rendering load on the CPU. This effect is noticeable in the game as small tunnels with sharp 90-degree bends leading from one large space into another. This small tunnel serves to block view into the adjoining unrendered space, and a special type of transparent brush (called a visportal) is placed within it to define the edge of where the engine should stop rendering the adjoining space. It is uncommon in the original Quake to be able to see across the entire length of a map, and outdoor spaces are often very tall and narrow, primarily utilizing distance above into open sky or below into lava to create a low-polygonal illusion of expanse.

A binary space partitioning (BSP) tree is built from the map, simplifying complexity of searching for a given polygon to O(number of polygons). Each leaf creates some area of 3D space (imagine cutting a pie into arbitrary pieces). The leaves of this binary tree have polygons of the original map associated with them, which are then used for computing each area's visibility. For each area, the VSD algorithm finds the parts of the map for which a line of sight exists. This is called the potentially visible set (PVS).[5]

This process uses large amounts of memory, since it should take ${\displaystyle O(n^{2})}$ (where ${\displaystyle n}$ is the number of polygons) bits (only visible/hidden information is needed). John Carmack realized that one area sees just a small fraction of the other areas, so he compressed this information by using run-length encoding (RLE). This is what allowed Quake's complex geometry to be rendered so quickly on the hardware of the time.

To reduce overdraw (rendering a new pixel that hides a previously rendered point, meaning the previous work was useless and wasted), the environment was displayed first, from front to back. To hide parts of walls hidden by other walls, a Global Edge List was sorting edges of already rendered polygons; new polygons were first clipped against previous edges so that only visible parts would get to the frame buffer.

Also while rendering the environment, a z-buffer was filled but never read while rendering the environment, as the BSP tree and Global Edge List ensured that each pixel was rendered only once. The ZBuffer was later used to render correctly characters and other moving objects that were partially hidden by the environment.

The pixel rendering loop was implemented in assembly. The texture coordinates perspective correction and interpolation was done using the floating-point unit, due to the limited number of integer registers; it also allows to compute expensive division operation (part of perspective correction) on the floating-point unit in parallel with integer interpolation (in other words, at no cost).

The base texture and the lightmap of a wall were rendered at the same time: a "surface cache" was creating new surfaces, which are new pre-lighted textures which combines the base and light map textures baked together. Surfaces not used since a few frames were released, while new required Surfaces were dynamically created. Generating the surfaces was consuming less time than a secondary lighting pass would have. To save memory, smaller surfaces using mipmaps of the original texture were generated first for further walls.

Characters were lit by an ambient light and fixed distant light both dependent on the light level of the spot on the floor where the character was standing. Close-by characters were Gouraud shaded and affinely textured, whereas faraway characters were rendered using triangle subdivision.[6]

Quake was one of the first games to support 3D hardware acceleration. While initially released with only software rendering, John Carmack created a version of the Quake executable that took advantage of Rendition's Vérité 1000 graphics chip (VQuake). OpenGL support was added in the form of the GLQuake executable for Windows 95 and higher. Other cards capable of rendering GLQuake were a professional (and very expensive) Intergraph 3D OpenGL card and later, the PowerVR cards.

To optimize the software rendering engine, lightmaps were shared by polygons that were close in space, and in the same leaf of the BSP tree. This means that quite often polygons using the same main texture could not be rendered at the same time with the 3D acceleration, due to the multi-texturing second unit having to be reconfigured with another lightmap. This architecture decision reduced hardware-accelerated rendering performance.

Quake includes cooperative and deathmatch multiplayer modes over LAN or the Internet. Additional multiplayer modes were later added using mods.

Quake uses the client–server model, where a server has control of all game events. All players connect to this server in order to participate, with the server telling the clients what is happening in the game. The server may either be a dedicated server or a Listen Server. Even in the latter situation, Quake still uses the client-server model, as opposed to the peer-to-peer networking used by some other games. Quake thus cannot suffer from de-synchronized network games that could occur from different clients disagreeing with each other, since the server is always the final authority.