Larry Gritz
Larry Gritz
Let's bring up at the TSC meeting. Maybe people from NVIDIA are involved in LLVM's PTX back end, or have connections who are?
Haha, it turns out we DID implement this in our renderer! But I didn't document it in the OSL spec. And it has different names than I picked here. I'm...
Up to each app. It's fine to be fully detailed, or just use "gaffer" for the app/renderer name and if you need to distinguish contexts, you can always have a...
After mulling it over, I prefer "renderer:name" and "renderer:version", though I'm going to change the version from an int[4] to a single int, and let the renderers decide how their...
Um... I think fell through the cracks as we were debating what the names ought to be. I will revive it next week.
@tgrant-nv were you building with USE_PYTHON=OFF? Python is needed for the python bindings to the oslquery library, so there is a find_package(Python) already happening at some point. And also python...
fmod(-0.666667, 0.500000) = 0.333333 fmod(-0.666667, 0.500000) = -0.166667 Whoa, it's different with the same input values and only differing in lockgeom?
It means we're doing the calculation differently for the constant-folded case versus the runtime case.
Even if we get a fix in NVPTX in the very next major version (13), it will probably be years before that's the minimum required LLVM version for OSL. So...