Performance was not improved directly, gcc was terrible. Gcc development had actually stagnated for basically all of the early 90s and there was considerable di

Performance was not improved directly, gcc was terrible. Gcc development had actually stagnated for basically all of the early 90s and there was considerable discontent leading to a long forgotten but high profile fork: egcs. This work eventually went back into gcc for the 3.0 major version and the development model evolved considerably but this was well after the development of Quake. Short story Watcom had superior codegen throughout this time and for years later. The performance critical parts of quake were written in Pentium micro optimized assembly. Using DJGPP gave other advantages, the id team developed on Unix/Next boxes. I think they also had some commercial Unix servers for running the expensive level (PVS) compilers. This gave them had a compiler that could target Unixen, DOS and windows. They could ship the compiler if they wanted (though ultimately mods were done through quake C). With source available they were able to and did in fact modify the compiler and DOS extender for their needs as well https://en.m.wikipedia.org/wiki/CWSDPMI https://en.m.wikipedia.org/wiki/CWSDPMI (See CWSDPR0)