https://bugs.winehq.org/show_bug.cgi?id=53213
Zeb Figura z.figura12@gmail.com changed:
What |Removed |Added ---------------------------------------------------------------------------- CC| |z.figura12@gmail.com
--- Comment #2 from Zeb Figura z.figura12@gmail.com --- The given test uses a sample with a POINT mip filtering modes with explicit LOD, and specifies that LOD to be exactly 0.5. llvmpipe, when interpreting LOD, rounds to even (by default), yielding 0.0.
GL 3.2, if my reading is correct, specifies rounding to -∞ [it specifically says ceil(lod + 1/2)]. Somewhere between there and 4.6 allowance is made for a non-preferred "alternative" of rounding to +∞ (it specifically says floor(lod + 1/2)]. The latter wording is carried forward to Vulkan.
The Direct3D spec, unusually, actually gives a ruling on this: rounding towards +∞ [again, specified in terms of floor(lod + 1/2)].
radeonsi, at least, consistently rounds toward +∞. llvmpipe doesn't follow any spec consistently. That said, do we care?