Refactoring GPU extensions

[kmp5VT]

Description

In this PR I am working to address some refactoring issues. Such as removing using statements from module files, try to diagnose/fix GPU blocksparse backend failures, update adapt functions to use storagemode like AMDGPU, remove gpu examples from testing based on plan listed in google drive

Checklist:

• Investigate why the AMDGPU and CUDA backends don’t work for BlockSparse storage. Implement resize! for AMDGPUs to get block sparse operations on AMDGPU working (see Add resize! JuliaGPU/Metal.jl#279 for a reference on how it might be implemented).
• Update the adapt functions to have storagemode as the input parameter
  Update the Extensions libraries to move the bare using statements from the NDTensorsXExt.jl file to the files where they are necessary.
• Delete NDTensors/ext/examples.
• Remove examples from NDTensors test
• Define a MtlArrayAdaptor for adapting storage to MtlArray, analagous to CuArrayAdaptor, to be used here: https://github.com/ITensor/ITensors.jl/blob/v0.3.57/NDTensors/ext/NDTensorsMetalExt/adapt.jl
• update the generic_zeros and generic_rand functions for Dense and AbstractArray
• All unittests pass

[mtfishman]

Note that resize! is defined for AMDGPU.ROCVector: https://github.com/JuliaGPU/AMDGPU.jl/blob/v0.8.11/src/array.jl#L266.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 50.23%. Comparing base (e8197ce) to head (9ce1991).
Report is 4 commits behind head on main.

❗ Current head 9ce1991 differs from pull request most recent head 57abe1f. Consider uploading reports for the commit 57abe1f to get more accurate results

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1365       +/-   ##
===========================================
- Coverage   79.59%   50.23%   -29.36%     
===========================================
  Files         114      113        -1     
  Lines        9032     8979       -53     
===========================================
- Hits         7189     4511     -2678     
- Misses       1843     4468     +2625     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kmp5VT]

So far I have updated the CUDA library to remove using statements from the module package. Next I am looking into this

Investigate why the CUDA backends don’t work for BlockSparse storage (there is no append! function) and find a way to fix this issue.

[mtfishman]

Suggested change

	function generic_zeros(StoreT::Type{<:Dense}, dims::Tuple{Integer})
	function generic_zeros(StoreT::Type{<:Dense}, dims::Integer)

[mtfishman]

My proposal was to define both of them in terms of ::Integer, not ::Tuple{Integer}, since it is a simpler interface for new types to overload in general.

[kmp5VT]

My only concern is that if you do something like this generic_zeros(Dense, (2,)) it calls the AbstractArray version of this funciton

[mtfishman]

Right, so then you should rewrite the AbstractArray version so that doesn't happen.

[kmp5VT]

I am not sure I understand what you're suggesting to do. I reverted the code back the original design you suggested which works for generic_zeros(Dense, 2) and generic_zeros(Dense, (2,))

[kmp5VT]

@mtfishman If you have a minute can you please take a look at this PR again? Thank you I appreciate your time!

[mtfishman]

Looks good, thanks!