2011/8/11 Michael Mc Donnell <mich...@mcdonnell.dk>: > > + dst_ptr[0] = src->x < 0.0f ? (SHORT)ceilf(src->x * SHRT_MAX + > 0.5f) :(SHORT)floorf(src->x * SHRT_MAX + 0.5f);
You can use roundf() instead. Actually, notice that maybe what you actually need for correct rounding is rintf() (which essentially matches what D3DXFloat32To16Array does). You should probably test this (even on your own, not necessarily adding those to the testsuite, I'd say). Still on the conversion: going on with this "strategy" is going to require a ton of code to handle every src->dst format combination. You may look into doing that in two steps: first converting src to float4, then float4 into dst. I haven't looked into it, so there may be some detail making this impractical, but still it's worth a try, I guess.
