I'm trying to minimize the VRAM usage, and since the DXTn compressor is still produces artifacts on higher mip levels, I thought of using 16-bit textures as a fallback. I'm using GL_RGBA4 and GL_RGB5 internal formats to be specific. The problem is that I don't see any difference in VRAM usage and visually the textures look the same, which suggest that they are still in RGB/RGBA. I also tried setting the TGLLOWQTEXTURES environment variable, which used to force every texture into 16-bit, but it didn't have any effect either. My question is, was the 16-bit texture support removed, or is there some bug that prevents the resampling of the texture into the requested format?

It's always good to mention what hardware you're using, as drivers differ between hardware. I'm gonna assume you're using R300-based hardware. The R300 driver doesn't presently support any 16 bit texture formats. The texture formats supported by the R300 driver are:

GL_RGBA,
GL_BGRA,
GL_COMPRESSED_RGBA_S3TC_DXT3_EXT,
GL_COMPRESSED_RGBA_S3TC_DXT5_EXT,
GL_LUMINANCE,
GL_ALPHA,

I'm using R200 (Mac Mini G4 RV280), which used to support 16-bit textures. At least I remember using TGLLOWQTEXTURES with it, but that was probably before the TinGL overhaul. Is it possible to supply my own mipmaps without using gluBuild2DMipmaps? glTexImage2D always tries to set level 0, no matter what argument I pass to it.

[ Edited by BSzili 09.06.2018 - 10:17 ]