Using CUDA.jl

[navidcy]

• Drops CuArrays.jl and CUDAapi.jl and uses CUDA.jl instead
• Disallows scalar operations on GPU

The fact that I had to use @CUDA.allowscalar so many times (especially in the MultiLaywerQG) probably means something. It seems like @views use getindex on the GPU; I was getting errors if I didn't include @CUDA.allowscalar in front of expressions with @views.

!! This should only be merged after FourierFlows/FourierFlows.jl#198 is in master. !!

Closes #96

[codecov]

Codecov Report

Merging #98 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #98   +/-   ##
=======================================
  Coverage   91.89%   91.89%           
=======================================
  Files          20       20           
  Lines        1506     1506           
=======================================
  Hits         1384     1384           
  Misses        122      122           

Impacted Files	Coverage Δ
src/GeophysicalFlows.jl	100.00% <ø> (ø)
src/barotropicqg.jl	100.00% <100.00%> (ø)
src/barotropicqgql.jl	100.00% <100.00%> (ø)
src/multilayerqg.jl	100.00% <100.00%> (ø)
src/twodnavierstokes.jl	100.00% <100.00%> (ø)
src/utils.jl	100.00% <100.00%> (ø)
test/runtests.jl	89.70% <100.00%> (ø)
test/test_barotropicqg.jl	100.00% <100.00%> (ø)
test/test_barotropicqgql.jl	100.00% <100.00%> (ø)
test/test_multilayerqg.jl	100.00% <100.00%> (ø)
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 17443e2...3b7fc63. Read the comment docs.

[glwagner]

Using @views is certainly just a convenience. We'll get better performance writing kernels. This is also true even in cases where we do not use @views (kernels seem to be faster than naive broadcasting, at least right now, which may be an inefficiency in broadcasting that could be fixed sometime in the future). Note that if we write kernels and use KernelAbstractions we also will get multithreaded speed up even on algebraic operations (not just FFTs).

[navidcy]

Using @views is certainly just a convenience. We'll get better performance writing kernels. This is also true even in cases where we do not use @views (kernels seem to be faster than naive broadcasting, at least right now, which may be an inefficiency in broadcasting that could be fixed sometime in the future). Note that if we write kernels and use KernelAbstractions we also will get multithreaded speed up even on algebraic operations (not just FFTs).

OK, I see your point. It takes a bit away from the "Julia can seamlessly be GPU-ready" idea that I had in mind. But, what you are saying is that if the developers work a bit harder here (writing kernels) then the experience will be seamless for the users, right?
But then what's then this is the same for PyCUDA and things like that, no?

[navidcy]

@glwagner this PR is ready, but don't merge yet. First FourierFlows/FourierFlows.jl#198 needs to be merged, make a new release of FourierFlows.jl and then we should remove Manifest.toml from here and add a [compat] entry in Project.toml to force GeophysicalFlows.jl use the latest release of FourierFlows.jl.

[glwagner]

OK, I see your point. It takes a bit away from the "Julia can seamlessly be GPU-ready" idea that I had in mind. But, what you are saying is that if the developers work a bit harder here (writing kernels) then the experience will be seamless for the users, right?
But then what's then this is the same for PyCUDA and things like that, no?

I think I was confusing. If we write kernels, they will be multithreaded and possibly will have slightly better GPU performance than broadcasting. Multithreading is not a GPU concept; multithreading will speed up CPU computations.

Note that fused broadcasts could become multithreaded in the future. They just aren't right now. I think GPU broadcasting could be improved, perhaps.

I am not implying that hand-written kernels are necessary for performant code. I apologize if I implied that.