Delete most source code and instead rely on NDTensorsCUDAExt

Project.toml

@@ -1,33 +1,15 @@
 name = "ITensorGPU"
 uuid = "d89171c1-af8f-46b3-badf-d2a472317c15"
 authors = ["Katharine Hyatt", "Matthew Fishman <[email protected]>"]
-version = "0.1.7"
+version = "0.2.0"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
-Combinatorics = "861a8166-3701-5b0c-9a16-15d98fcdc6aa"
-Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
 ITensors = "9136182c-28ba-11e9-034c-db9fb085ebd5"
-LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
-NDTensors = "23ae76d9-e61a-49c4-8f12-3f1a16adf9cf"
-Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
-SimpleTraits = "699a6c99-e7fa-54fc-8d76-47d257e15c1d"
-StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
-Strided = "5e0ebb24-38b0-5f93-81fe-25c709ecae67"
-TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
-cuTENSOR = "011b41b2-24ef-40a8-b3eb-fa098493e9e1"
 
 [compat]
-Adapt = "3.5, 4"
-CUDA = "4.0"
-Combinatorics = "1.0.2"
-Functors = "0.2, 0.3, 0.4"
-ITensors = "= 0.3.37"
-NDTensors = "0.1.50"
-SimpleTraits = "0.9.4"
-StaticArrays = "1.2.13"
-Strided = "1.1.2, 2"
-TimerOutputs = "0.5.13"
-cuTENSOR = "1.1.0"
-julia = "1.6 - 1.9"
+Adapt = "3, 4"
+CUDA = "4, 5"
+ITensors = "0.3, 0.4, 0.5"
+julia = "1.6"

README.md

@@ -1,50 +1,5 @@
-# ITensorGPU: Intelligent Tensors with GPU acceleration
-
-[![codecov](https://codecov.io/gh/ITensor/ITensorGPU.jl/branch/main/graph/badge.svg)](https://codecov.io/gh/ITensor/ITensorGPU.jl)
-
-[![gitlab-ci](https://gitlab.com/JuliaGPU/ITensorGPU-jl/badges/master/pipeline.svg)](https://gitlab.com/JuliaGPU/ITensorGPU-jl/commits/master)
-
-This package extends the functionality of [ITensors.jl](https://github.com/ITensor/ITensors.jl) to make use of CUDA-enabled GPUs to accelerate tensor contractions and factorizations. It sits on top of the wonderful [CUDA.jl](https://github.com/JuliaGPU/CUDA.jl) package and uses NVIDIA's [cuTENSOR](https://developer.nvidia.com/cutensor) library for high-performance tensor operations.
-
-## Installing ITensorGPU.jl
-
-Dependencies:
-  - [Julia 1.3 or later](https://julialang.org/downloads/)
-  - [CUDA 10.1 or later](https://developer.nvidia.com/cuda-downloads) -- Currently only NVIDIA GPUs are supported. NVIDIA drivers are required so that Julia can make use of the NVIDIA GPU on your system.
-  - [cuTENSOR v1.0.0 or later](https://developer.nvidia.com/cutensor) -- A specialized library for perfoming permutation-free tensor contractions on the GPU. `libcutensor.so` needs to be in your `LD_LIBRARY_PATH` so that `CUDA.jl` will be able to find it.
-  - [ITensors.jl](https://itensor.github.io/ITensors.jl/stable/#Installation-1)
-
-To properly install CUDA with Julia, it may be helpful to first follow the [CUDA.jl installation instructions](https://juliagpu.github.io/CUDA.jl/stable/installation/overview/) and test that you have that installed properly and that it is able to use `cuTENSOR`. You can run the commands:
-```julia
-julia> using CUDA.CUTENSOR
-
-julia> CUTENSOR.has_cutensor()
-true
-
-julia> CUTENSOR.version()
-v"1.2.1"
-```
-to check that `CUDA.jl` can see the version of `cuTENSOR` you have installed.
-
-Once you have all of the dependencies installed, you can then go ahead and install `ITensorGPU.jl` with the following command:
-```
-julia> ]
-
-pkg> add ITensorGPU
-```
-
-To check if this has all worked, you can run the package tests using:
-```julia
-julia> ]
-
-pkg> test ITensorGPU
-```
-
-## Examples
-
-Take a look at the `examples/` directory for examples of running ITensor calculations on the GPU.
-
-For an application of `ITensorGPU.jl` to more sophisticated tensor network calculations, take a look at [PEPS.jl](https://github.com/ITensor/PEPS.jl).
-
-For some background on the development and design of this package, you can take a look at [this blog post](https://kshyatt.github.io/post/itensorsgpu/) by Katie Hyatt, original author of the `ITensorGPU.jl` package.
+| :warning: WARNING          |
+|:---------------------------|
+| The ITensorGPU.jl package is deprecated and only provided for backwards compatibility. For an alternative, see the ITensors.jl documentation about [running ITensor on GPUs](https://itensor.github.io/ITensors.jl/dev/RunningOnGPUs.html). |
 
+TEST

src/ITensorGPU.jl

@@ -1,114 +1,38 @@
 module ITensorGPU
-using NDTensors
-
-using Adapt
-using CUDA
-using CUDA.CUBLAS
-using CUDA.CUSOLVER
-using Functors
-using ITensors
-using LinearAlgebra
-using Random
-using SimpleTraits
-using StaticArrays
-using Strided
-using TimerOutputs
-using cuTENSOR
-
-using NDTensors: setdata, setstorage, cpu, IsWrappedArray, parenttype
-
-import Adapt: adapt_structure
-import Base: *, permutedims!
-import CUDA: CuArray, CuMatrix, CuVector, cu
-import CUDA.Mem: pin
-import ITensors:
-  randn!,
-  compute_contraction_labels,
-  eigen,
-  tensor,
-  scale!,
-  unioninds,
-  array,
-  matrix,
-  vector,
-  polar,
-  tensors,
-  truncate!,
-  leftlim,
-  rightlim,
-  permute,
-  BroadcastStyle,
-  Indices
-import NDTensors:
-  Atrans,
-  Btrans,
-  CombinerTensor,
-  ContractionProperties,
-  Combiner,
-  Ctrans,
-  Diag,
-  DiagTensor,
-  Dense,
-  DenseTensor,
-  NonuniformDiag,
-  NonuniformDiagTensor,
-  Tensor,
-  UniformDiag,
-  UniformDiagTensor,
-  _contract!!,
-  _contract!,
-  _contract_scalar!,
-  _contract_scalar_noperm!,
-  can_contract,
-  compute_contraction_properties!,
-  contract!!,
-  contract!,
-  contract,
-  contraction_output,
-  contraction_output_type,
-  data,
-  getperm,
-  ind,
-  is_trivial_permutation,
-  outer!,
-  outer!!,
-  permutedims!!,
-  set_eltype,
-  set_ndims,
-  similartype,
-  zero_contraction_output
-import cuTENSOR: cutensorContractionPlan_t, cutensorAlgo_t
-
-#const ContractionPlans = Dict{String, Tuple{cutensorAlgo_t, cutensorContractionPlan_t}}()
-const ContractionPlans = Dict{String,cutensorAlgo_t}()
-
-include("cuarray/set_types.jl")
-include("traits.jl")
-include("adapt.jl")
-include("tensor/cudense.jl")
-include("tensor/dense.jl")
-include("tensor/culinearalgebra.jl")
-include("tensor/cutruncate.jl")
-include("tensor/cucombiner.jl")
-include("tensor/cudiag.jl")
-include("cuitensor.jl")
-include("mps/cumps.jl")
-
-export cu,
-  cpu, cuITensor, randomCuITensor, cuMPS, randomCuMPS, productCuMPS, randomCuMPO, cuMPO
-
-## TODO: Is this needed?
-## const devs = Ref{Vector{CUDAdrv.CuDevice}}()
-## const dev_rows = Ref{Int}(0)
-## const dev_cols = Ref{Int}(0)
-## function __init__()
-##   voltas    = filter(dev->occursin("V100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
-##   pascals    = filter(dev->occursin("P100", CUDAdrv.name(dev)), collect(CUDAdrv.devices()))
-##   devs[] = voltas[1:1]
-##   #devs[] = pascals[1:2]
-##   CUBLASMG.cublasMgDeviceSelect(CUBLASMG.mg_handle(), length(devs[]), devs[])
-##   dev_rows[] = 1
-##   dev_cols[] = 1
-## end
-
-end #module
+using Adapt: adapt
+using CUDA: CUDA, CuArray, cu
+export cu
+using ITensors: cpu
+export cpu
+
+using ITensors: ITensor, cpu, cu, randomITensor
+function cuITensor(args...; kwargs...)
+  return adapt(CuArray, ITensor(args...; kwargs...))
+end
+function randomCuITensor(args...; kwargs...)
+  return adapt(CuArray, randomITensor(args...; kwargs...))
+end
+export cuITensor, randomCuITensor
+
+# TODO: Change over to `using ITensorMPS`
+# once it is registered.
+using ITensors.ITensorMPS: MPO, MPS, randomMPO, randomMPS
+function cuMPS(args...; kwargs...)
+  return adapt(CuArray, MPS(args...; kwargs...))
+end
+cuMPS(tn::MPS) = adapt(CuArray, tn)
+function productCuMPS(args...; kwargs...)
+  return adapt(CuArray, MPS(args...; kwargs...))
+end
+function randomCuMPS(args...; kwargs...)
+  return adapt(CuArray, randomMPS(args...; kwargs...))
+end
+function cuMPO(args...; kwargs...)
+  return adapt(CuArray, MPO(args...; kwargs...))
+end
+cuMPO(tn::MPO) = adapt(CuArray, tn)
+function randomCuMPO(args...; kwargs...)
+  return adapt(CuArray, randomMPO(args...; kwargs...))
+end
+export cuMPO, cuMPS, productCuMPS, randomCuMPO, randomCuMPS
+end