Loading textures -> slow!
BlitzMax Forums/OpenGL Module/Loading textures -> slow!
| ||
| hi! just a question: i am loading textures in my openGL engine by opening the image file and putting each texel into the bank like that: Function LoadTexture(file$, flags = 0) img = LoadImage(file$) w = ImageWidth(img) h = ImageHeight(img) nw = GetWidth(w) nh = GetWidth(h) t.texture = New texture t\bank = CreateBank(12) t\w = nw / BLE_TextureQuality t\h = nh / BLE_TextureQuality t\sx# = 1 t\sy# = 1 bank_size = t\w * t\h * 3 bank_img = CreateBank(bank_size) ib = ImageBuffer(img) LockBuffer ib For y = 0 To t\h - 1 off = bank_size - (y + 1) * t\w * 3 For x = 0 To t\w - 1 rgb = ReadPixelFast(w * x * BLE_TextureQuality / nw, h * y * BLE_TextureQuality / nh, ib) PokeByte bank_img, off, (rgb And $FF0000) Shr 16 PokeByte bank_img, off + 1, (rgb And $FF00) Shr 8 PokeByte bank_img, off + 2, rgb And $FF off = off + 3 Next Next UnlockBuffer ib FreeImage img glGenTextures 3, t\bank glBindTexture GL_TEXTURE_2D, PeekInt(t\bank, 0) glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST glTexImage2D GL_TEXTURE_2D, 0, 3, t\w, t\h, 0, GL_RGB, GL_UNSIGNED_BYTE, bank_img glBindTexture GL_TEXTURE_2D, PeekInt(t\bank, 4) glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR glTexImage2D GL_TEXTURE_2D, 0, 3, t\w, t\h, 0, GL_RGB, GL_UNSIGNED_BYTE, bank_img glBindTexture GL_TEXTURE_2D, PeekInt(t\bank, 8) glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR glTexParameteri GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR gluBuild2DMipmaps GL_TEXTURE_2D, 3, t\w, t\h, GL_RGB, GL_UNSIGNED_BYTE, bank_img FreeBank bank_img Return Handle(t) End Function this is extremely slow because i am using ~20 textures about 512*512 on a little demo map. isn't there any way to make it work faster? thanks. |
| ||
| Maybe this one could be usefull for you. It is the loadtexture method off 'minib3d'. |
| ||
| nope, sorry. my problem is, that pushing each of the (256^2)*20 texels into data banks is very slow. even on debuger off. a c++ library would be useful, but i don't know how. so i think i have to improve it in blitz. |
| ||
| You don't need to load image. Load image means loads the data into main memory, upload it to the graphics card as a texture, then you're also doing al lock image which then has to download the image from the graphics card and turn it into a pixmap, then you're doing additional processing on it. Use LoadPixmap. Also do not use PokeByte or ReadPixelFast, use pointer and treat the data like an array e.g. Data[Offset]. If you use PokeByte for every pixel you are causing it to look up the bank object, read the bank's base pointer, you have to calculate x and y offsets turned into an address, and then access the pixel. Very slow. Just have a source pointer and dest pointer and read whole Int's at a time where possible and use binary operations to shift stuff around and get the values where you want them. To make it even faster you can load RAW graphics data directly into the bank, perhaps a line at a time if you need a modulo, which is faster yet then decompressing an image. I do that for 20 512x512 alpha channel images and it takes about a second to load. |
| ||
| oh well :0. i don't quite understand. is there a code snippet out there somewhere? edit: oh and i am programming in bb, not bmx. is there still a way? |
| ||
| I do this by loading my own texture format straight into a pixelmap, then setting the gl texture pixels with this. I still haven't figured out how to store pixel data in the compressed format, and load the compressed data straight into OpenGL...it would be a lot faster. |
| ||
| i know. the stf format... it has the same size as your textures would have in bmp. i tested it. 2 shorts for w/h in the file and then for each pixel 3 bytes in the texture format file. this is not very faster :( edit: 9500 ms loading jpg. 6800 ms loading own texture format |
| ||
| Instead of reading a byte for each pixel, just copy the whole pixel data at once with ReadBytes(). |
| ||
| neither slow. |
   |