Andrej
Do we not have to include it? Do we need to change the .cu as well?
wow the diff on this PR. This is what is needed to compile C on Windows?
The const changes look good even as a general thing independent of Windows, I'd merge that regardless of Windows considerations. The makefile echo I'm not a huge fan of. Ideally...
Yeah I think this is a good idea, I have a janky uncommitted version of this but I think it makes sense to merge to master in some form.
I cannot reproduce this speedup. Also I don't fully understand it.
But which kernel is this re-utilizing from? The previous kernel is a cuBLAS matmul.
merged in 315af5f, slightly modified with comments ty. it does help about 1ms/iter on average from me, from 193ms -> 192ms. might be more on other GPUs
So I want to do this, I'm just not sure when :)
I certainly don't want to go down the IFDEF switching hell.
Hey David, sorry. recurrentjs was not really meant for production or cleanliness, it's a "are you neural nets expert? ok here's some dump of code you might like" kind of...