CuArrays.jl
CuArrays.jl copied to clipboard
JuliaGPU
A Curious Cumulation of CUDA Cuisine
Porting codes from CuGammaFuns.jll; see https://github.com/xukai92/CuGammaFuns.jl/issues/1 Currently I - add a new folder called `special` and put `gamma.jl` inside - defines the new diff. rules in `forwarddiff.jl` - link implemented...
Fixes https://github.com/FluxML/Flux.jl/issues/1114 The context here is that on the first call to the layer (also simulated via calling the `Flux.reset!` on the structure), the gradients for the hidden layer were...
Massively speeds up host operations, without losing `assertscalar`-like functionality. Doesn't properly works right now, due to the coherency requirements of unified memory on pre-sm_60 hardware. Basically, we'd need to synchronize...
## Current Issues: - [ ] Line 47 in upsample.jl is not a safe operation without atomic addition. It results in incorrect and inconsistent answers in the backward pass. @maleadt...
Some of the commonly used kernels (conv and maxpool) are not deterministic by default. This hurts reproducibility a lot - when fix random seed (e.g. by `seed!(1)`) users cannot reproduce...
In fact, there are more algorithms than we include now: https://docs.nvidia.com/cuda/cutensor/api/types.html#cutensoralgo-t `values >= 0 correspond to certain sub-algorithms of GETT` I checked and currently the maximum number of algorithms function...
as reported in https://github.com/JuliaGPU/CuArrays.jl/issues/629 one could still add an additional test.
hi there, I really enjoyed the tutorial on starting out with GPU and I thought I'd give it a go for a very specific use case. value function iteration is...
Thanks to `IRTools.jl`, we can do some nifty things with Julia IR. Like using a `dynamo` to walk through the deep IR and offload sensible ops to the GPU. ```julia...
DiffEqBase.jl has been carrying an ldiv! overload to make it work for awhile (https://github.com/JuliaDiffEq/DiffEqBase.jl/blob/master/src/init.jl#L148-L152), and I think it might be a good time to upstream it.
