BSPv38 - Quake 2 BSP loader lightmap problem
glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MAG_FILTER,GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MIN_FILTER,GL_LINEAR);
except it still looks "wrong" for some reason - my texture width/height calculations seem off, but I've followed that guide EXACTLY.
I also don't understand the part about lightmap being max 16x16...I'm getting lightmaps that are much larger than 16x16 from the exact formula he has given on the site, the web page contradicts itself o.O
Thanks,
HyperNewbie
The formula for texture width and height are wrong, and the website has no mention of how lightmap texture coordinates are calculated.
I'll post my findings here, just in case anyone comes across this page on google with the exact same problem I ran into:
This info is for BSPv38, in other words Quake 2 BSP files, regarding lightmaps and loading and rendering them properly. I used openGL to do the rendering.
Getting lightmap dimensions:
Face structure gives the lightmap offset, but for some crazy reason, does not give the lightmap dimensions: you must calculate them yourself by going through every vertex, calculating their texture map U V co-ordinates and do stuff to them.
So, first, you go through every vertex of the face, and calculate their UV co-ordinates (use the formulae on the website http://www.flipcode.com/archives/Quake_2_BSP_File_Format.shtml).
u = x * u_axis.x + y * u_axis.y + z * u_axis.z + u_offset
v = x * v_axis.x + y * v_axis.y + z * v_axis.z + v_offset
Then, you get the results in two floats, u and v. FLOOR u and FLOOR v into two integers, floor_u and floor_v. Then, you check floor_u and floor_v against max_u and max_v and min_u and min_v (4 of them all integers), to get the maximum and minimum floor_u and floor_v variables seen so far. DO NOT GET MAXIMUM/MINIMUM u and v. We need maximum/minimum floor_u and floor_v.
Then, you divide the min_u and max_u and min_v and max_v variables by 16, in FLOAT. If you are using C++, this means you will need to specifically typecast those 4 ints into float before dividing by 16.
Then, you floor the two min_ variables and ceil the two max_ variables, get the difference between the two, add 1, and typecast the result back to an int.
Here is my little bit of C++ code that does this step:
_minU = (float)floor((float)(minU / 16.0));
_maxU = (float)ceil((float)(maxU / 16.0));
_minV = (float)floor((float)(minV / 16.0));
_maxV = (float)ceil((float)(maxV / 16.0));
LIGHTMAPW = (int)(_maxU - _minU + 1);
LIGHTMAPH = (int)(_maxV - _minV + 1);
Getting lightmap texture co-ordinates:
This will generate a texture co-ordinate that's in the range 0.0 to 1.0, for opengl glTexCoord2f usage.
Let's say we need to get the texture co-ordinates for a vertex.
we first calculate the u and v for the texture map, the formula given above.
Then, we get LIGHTMAPW and LIGHTMAPH , also given above
We also need the minU and minV and maxU and maxV from above, the max/min floor_u and floor_v for all vertexes.
Then, simply follow this formula:
float MidPolyU = (minU + maxU) / 2.0;
float MidPolyV = (minV + maxV) / 2.0;
float MidTexU = LIGHTMAPW / 2.0;
float MidTexV = LIGHTMAPH / 2.0;
float lightmap_u = MidTexU + (u - MidPolyU) / 16.0;
float lightmap_v = MidTexV + (v - MidPolyV) / 16.0;
glTexCoord2f(lightmap_u/LIGHTMAPW ,lightmap_v/LIGHTMAPH );
As you can see, we get the co-ordinates and make them relative to the middle of the texture. Or something like that.
Hopefully this will solve the problems for people who are trying to load quake2's BSP files and get stuck at trying to load the lightmaps, because nowhere on the face of the internet is there such a documentation of this and quake 2 source code takes a while to excavate into.
Check six,
HyperNewbie
But I want to make the answer clear for others, because I spent a lot of time determining the source of wierd/broken/rainbow lightmaps.
So here is how to read lightmaps from quake 2 bsp file.
The next code is a bit simplified and there is no lightmap baking into one single texture here.
Only per-face generation of lightmaps just to be clear of all texture coordinate/dimensions calculations.
1. Just get all the data into memory:
int lightmapDataLength = header.lump[LUMP_LIGHTMAPS].length; // /sizeof(unsigned char);
unsigned char lightmapData = new unsigned char[lightmapDataLength];
fseek(file, header.lump[LUMP_LIGHTMAPS].offset, SEEK_SET); // LUMP_LIGHTMAPS is 7
fread(lightmapData, sizeof(unsigned char), lightmapDataLength, file);
Alternatively you can also seek for each lightmap while iterating through bsp faces. Current lightmap for current face would be located at header.lump[LUMP_LIGHTMAPS].offset + currentFace.lightmap_offset
2. Yes, to get lightmap dimensions you must process all faces and get those damn floored min/max of UVs.
First of all you must init the face data like this:
faces[f].fminUV.x = 999999;
faces[f].fminUV.y = 999999;
faces[f].fmaxUV.x = -99999;
faces[f].fmaxUV.y = -99999;
Then, you calculate the texture coordinates exactly like it's written on all q2bsp specs:
faces[f].texcoords[i*2 + j].x = verts[idx].x * texinfo->u_axis.x + verts[idx].y * texinfo->u_axis.y + verts[idx].z * texinfo->u_axis.z + texinfo->u_offset;
faces[f].texcoords[i*2 + j].y = verts[idx].x * texinfo->v_axis.x + verts[idx].y * texinfo->v_axis.y + verts[idx].z * texinfo->v_axis.z + texinfo->v_offset;
And only after that you compute min/max values while iterating through faces:
faces[f].fminUV.x = min(floor(faces[f].texcoords[i*2 + j].x), faces[f].fminUV.x);
faces[f].fminUV.y = min(floor(faces[f].texcoords[i*2 + j].y), faces[f].fminUV.y);
faces[f].fmaxUV.x = max(floor(faces[f].texcoords[i*2 + j].x), faces[f].fmaxUV.x);
faces[f].fmaxUV.y = max(floor(faces[f].texcoords[i*2 + j].y), faces[f].fmaxUV.y);
3. Now you're ready to get lightmap dimensions:
int lightMapWidth = ceil(faces[f].fmaxUV.x/16) - floor(faces[f].fminUV.x/16) +1;
int lightMapHeight = ceil(faces[f].fmaxUV.y/16) - floor(faces[f].fminUV.y/16) +1;
4. Now the important part which is still never indicated by anybody, with a little exceptions.
First of all if you choose to offset into lightmap data array directly, like:
unsigned char *lm_data = &lightmapData[face->lightmap_offset];
And then pass rgb data into opengl you MUST specify alignment!
glPixelStorei(GL_PACK_ALIGNMENT,1);
glPixelStorei(GL_UNPACK_ALIGNMENT,1);
glTexImage2D(GL_TEXTURE_2D, 0, 3, lightMapWidth, lightMapHeight, 0, GL_RGB, GL_UNSIGNED_BYTE, lm_data);
Without this all the people get lightmaps which are diagonaly projected or a lot of rainbow colors.
This deceives most developers and they think like: damn the calculations are wrong, there is something wrong with pointers/data/uvs/etc BUT ITS THE DAMN ALIGNMENT (blame Carmack for all this shit!)
The alternative way to read the data is to read it as rgb and add fourth component with value 255 and upload into rgba texture, like this:
static unsigned char lm_data[32*32*4] = {0};
fseek(file, header.lump[LUMP_LIGHTMAPS].offset + face->lightmap_offset, SEEK_SET);
for(int y = 0; y<lightMapHeight; y++) {
for(int x = 0; x<lightMapWidth; x++) {
unsigned char rgba[4];
fread(rgba, sizeof(unsigned char), 3, file); // read rgb
rgba[3] = 255; // set the 4th component to 255
memcpy(&lm_data[x*4+(y*4)*lightMapWidth], rgba, sizeof(unsigned char)*4); // write as rgba
}
}
glTexImage2D(GL_TEXTURE_2D, 0, 4, lightMapWidth, lightMapHeight, 0, GL_RGBA, GL_UNSIGNED_BYTE, lm_data);
And don't forget the clamping:
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
5. Rendering (resulting texture coordinates generation):
u = face->texcoords[j].x;
v = face->texcoords[j].y;
u -= face->fminUV.x;
v -= face->fminUV.y;
u /= (face->fmaxUV.x - face->fminUV.x);
v /= (face->fmaxUV.y - face->fminUV.y);
glTexCoord2f(u, v);
glVertex3fv(&face->vertices[j].x);
The result... well not totally but almost identical.
The other alternative is to study the GPL Quake 2 source code and see how that does it (by packing lightmap textures into a 128x128 atlas and adjusting texcoords to match so that alignment isn't actually a problem).
srsly, did you try that yourself?
Why there's a tons of broken specs and such a clever person like you still didn't post any explanation anywhere so people won't be having problems?
Study what?...
A lot of non-commented retarded code with thousands of hax and funny comments like '// sigh' or '// stupid quake bug' where static lightmapping mixed with dynamic one with wrapped functions of memory allocation and also complicated with atlasing and hardware specific routines and around 80kb of code (gl_rsurf.c, gl_light.c, gl_model.c) just for this? LOL
Everyone knows that it does pack into atlas. But the alignment is an actual problem ANYWAY even if you will use atlas.
Why? Well, just try to upload lightmaps into your atlas having everything in GL_RGB (hell isn't obvious to use rgb when it's stored in rgb lol) format with TexSubImage and you'll get nice rainbows.
So there is a choice: you either store it in GL_RGBA filling dump alpha channel with 255 or using glPixelStorei with GL_RGB.
It's not because lightmaps are not in atlas, it's because of how the data is stored in binary bsp file...
That's it and no need to 'study' any code where obvious things wrapped around thousand data structures.
I showed per-face texturing just because it's lot easier to understand. After this you can easily make atlas or whatever you want.
sly, did you try that yourself?
As it happens, I did.
Yes - there is deep ugliness in there, but yes - if you want to work with Quake 2 formats then it is the definitive reference. Please note that I'm not sayig "copy/paste this code". What I'm saying is "if you want to know how lightmaps are supposed to work in the Quake 2 engine, this is the place to go".