1. Compile Legacy 2. Know bugs 3. Explanation of the code 3.1 The memory model 3.2 Hardware Texture model 1. Compile Legacy ================= Sources can be found here: http://sourceforge.net/projects/doomlegacy/ DOS --- need: - djgpp 2.02 (http://www.delorie.com) - allegro 3.12 (http://www.talula.demon.co.uk/allegro/) - libsocket 0.7.4 (beta 4) or better (http://www.phekda.freeserve.co.uk/richdawe/lsck/lsck.htm) - bcd 1.03 (inlcude in this package) - gcc 2.95.2 is recommended - nasm 0.98 (or better) (http://www.cryogen.com/Nasm) compile: make debug: when craching legacy will return eip type make asm, then you will have a 8 megs doom3.s (a assembler file) the you can find the faulting instruction and function ------------------------------------------------------------------------ Linux ----- need: - you can compile X or GGI version just modify the makefile for GGI you will need GGI library - tested with gcc 2.7 and 2.95. - Mesa (or other OpenGL implementation) (optional: for OpenGL renderer) - nasm 0.98 (or better) (http://www.cryogen.com/Nasm) compile make LINUX=1 debug: gdb ? ------------------------------------------------------------------------ Win32 ----- need : - glide 3.x sdk (optional) (http://www.3dfx.com) - directx6 sdk (or higher) (http://www.micosoft.com/directx) - nasm 0.98 (or better) (http://www.cryogen.com/Nasm) - VC6 should also work with VC5 use the file in the VC6 dir and copy it in to the main directory debug: press on "step over" with the release version (there is some debug info on it). Then change the eip in the regster window when you will type enter the edi will go to the faulting instruction. don't forget that you can follow the stack for calls. You can also use trace with the debug version but add -win and -nomouse. 2. Know bugs/issues =================== - opengl/glide lighting may be slow (we have some project to use the bsp to optimize all the stuff) - splats on opengl/glide have seriouse z-problem and is not clipped to the seg - the splats schould be in segs and not in linedef, so we can clip it and render only when needed (actualy there is overdraw in opengl/3dfx) - the sky in opengl/3dfx is not well maped and we have big problem with large sky like in tnt map01 - when clientprediction2 is enabled (see doomdef.h) the spirit mobj have some serious z problem when quiting/entering moving platform - monsters/sprite have feet(head) in ground, caused by z-buffer, solution : trace all view using back-to-front rendering and use clipping like in software for spite (lot of work here) also debug the clipwall in harware, this will also fix the water and transparent wall problems - there is white dots around some wall/floors/ceiling, caused by : - T-intersection in walls, since wall height is define in sector - maybe because the bsp is in fixed and we use float (not sure) - maybe hwr_bsp.c buged :( - midi under win32 don't work properly - palette problem under winNT4 - there is no splats on floors and ceiling, fab have begon but haven't finish - 3dfx splitscren have seriouse projection probleme - sprite that have transparent region (torch, lost soul...) are full transparent in 3dfx/opengl - and some others surprise :) 3. Explanation of the code ========================== 3.1 The memory model (original) (z_zone.c) (by BP) -------------------- Doom allocate a heap of 6 megs at begining and provide a Z_Malloc function to allocate in this heap. Z_Malloc( int size,int tag,void* user ) size is the size in byte tag can be : PU_STATIC allocated static (like malloc do) call Z_Free to free it PU_LEVEL same as static but the zone is "tagged" with the tag PU_LEVEL, when calling Z_FreeTag (PU_LEVEL, PU_LEVEL) all zone tagged with PU_LEVEL are freed (at the end of the level) PU_CACHE this one _can_ be freed automatiquely by one other call to Z_Malloc. the *user point to a pointer to this zone so when freed automatiquely the pointer is set to NULL so eatch time you use it you must check if the pointer is not NULL and reload it. (...) 3.2 Hardware Texture model (by BP) -------------------------- Eatch texture/patch/flats/pic in legacy are converted to hardware texture at runtime (the GlideMipmap_s structure (hw_data.h)). I will call hardware texture a gr_texture so there is no confusion. To remind you : - Texture are set of patch and are associate to linedefs (walls) can be upper, lower or middle texture. It can have hole on it. - patch are sprites (the doom patch are run of vertical lines) - flats are used for floors and ceiling of sectors and have size of 64x64 it can't have hole on it - pic are new legacy format for picture, it can only handle plain texture like flats it is now used for hud in full screen for the main picture of legacy and for coronas (the format was extended to handle 32 bit color or intensity + alpha, not all are implemented at this time) Since patch, flat and pic are basic structure represented by only one lump in the wad, the wad loader allocate for eatch lump a GlideMipmap_s (cache3Dfx) and init data field to NULL. Since the data structure is allocated in PU_3DFXCACHE (like PU_CACHE) the data will be initilised when needed (hw_cache.c). The GlideMipmap_s structures for textures are initialized on HWR_PrepLevelCache (hw_cache.c) it is called in P_SetupLevel (load level) the number of textures is computed with TEXTURE1, TEXTURE2 lumps since this can be changed in runtime in legacy (load a wad while runing) it must be reallocated. Well, at this time it is realloceted at eatch level start. We can do better, since numtextures change only when a wad is loaded. The 3dfx driver use glide3, it load gr_texture in gr_texture memory of the card in fifo order when there is no more place it remove the first gr_texture, the downloaded field of GlideMipmap_s go to false and when needed it is reloaded in gr_texture memory. In OpenGl, since OpenGl keep texture in there own memory and handle gr_texture memory of the card we no more need to redownload it but if we not free time to time gr_texture memory in opengl, it will get alot of memory, so the gr_texture memory is cleared at eatch new level (same time of texture reallocation). Opengl and 3dfx link the loaded gr_texture with the nextmipmap field of GlideMipmap_s so before clear textures of the heap we MUST free gr_texture memory of OpenGl or 3dfx ! Legacy can also draw patch with a differant colormap (thanks to Hurdler). When needed it create the same gr_texture but just with a differant colormap. This one is linked with the original in the GlideMipmap_s with the nextcolormap field. So when a polygone with a gr_texture must be drawn, first we check if the gr_textures is not allready loaded in hadware memory (downloaded field) if not then we check if gr_texture data is there (not grabbed by z_malloc function) if not we must recompute it eatch type of gr_texture (texture, patch, flat, pic have there own methode) the we can send the gr_texture to 3dfx or OpenGl.