Enamex
December 25, 2019, 12:09am
1
UPDATE : Tensor Comprehensions (the C++ library and the Python wrapper) has been discontinued .
See: https://github.com/facebookresearch/TensorComprehensions/issues/610#issuecomment-439503312
Tensor Comprehensions:
They’re currently Linux x64 only (maybe even practically Ubuntu only). I can’t even try them.
No recurrence support yet? Should be fixable with a jitted outer loop over a TCed cell.
No news for more than a year now. Is it abandoned?
ptrblck
December 25, 2019, 8:52pm
2
The GitHub repo doesn’t seem to get much activity in the last year, so I assume the development is currently stopped.
Enamex
December 25, 2019, 11:44pm
3
opened 02:57PM - 08 Nov 18 UTC
First, let me say I love, love, love this project. It makes creating fast kernel… s a real breeze, unlike anything else I've seen.
I am running into a bug (I need parens around a generated if_then_else macro), and I see that there is a fix for it in master. But I'm having trouble building the code. Maybe someone who has the build system set up could update the Conda package? It looks like the project has stabilized (no commits in 2 months), so it seems like a good time to do it?
Hello @larspars
My apologies for the long delay, I am sorry to say that the core TC team has moved on to other things.
As far as I know the project is not maintained anymore but someone from Facebook should voice in, I am not in a position to speak for them.
Best,
N
Seems that way, unfortunately.
Still: What are the difference between JIT and tensor_comprehensions for custom kernels?
For a discontinued project, I’m curious what we’re missing. What can’t the next nearest tool do (here JIT, I think?)? Is that limitation fundamental or just till it finds a place on the roadmap?