toFloat3SE produces mantissa of 0 for non zero values

[Zvork]

Inputing float values of 0.388235301 in x/y/z produces the correct shared exponant of 14 but a mantissa of 0,0,0 which unpacks to value 0.f
Compared this with results comming from the glm library packing functions.

[castano]

Hmm... I just copied the code from the OpenGL spec, but I see in the revision comments that there were significant errors in that code. Would you mind updating the code in NVTT to match the spec? I won't have much time until next week.

https://www.khronos.org/registry/OpenGL/extensions/EXT/EXT_texture_shared_exponent.txt

[Zvork]

Sure. We can also use the functions in the GLM library, if you're ok for including it in the externs. It's a header only library. See https://github.com/g-truc/glm
That would solve the R11G11B10 issue also in one sweep (yes the R11G11B10 code seems faulty too).