[WIP] hipTENSOR support in TensorOperations

Project.toml

@@ -41,7 +41,7 @@ TensorOperationscuTENSORExt = "cuTENSOR"
 TensorOperationsJLArraysExt = "JLArrays"
 
 [compat]
-AMDGPU = "2"
+AMDGPU = "2.8"
 Aqua = "0.6, 0.7, 0.8"
 Adapt = "4"
 Bumper = "0.6, 0.7"

docs/src/index.md

@@ -60,6 +60,10 @@ The TensorOperations.jl package is centered around the following features:
     [CUDA.jl and cuTENSOR.jl](https://github.com/JuliaGPU/CUDA.jl), by relying on (and thus
     providing a high level interface into) NVidia's
     [cuTENSOR](https://developer.nvidia.com/cutensor) library.
+  - Support for `ROCArray` objects if used together with
+    [AMDGPU.jl and hipTENSOR](https://github.com/JuliaGPU/AMDGPU.jl), by relying on (and thus
+    providing a high level interface into) AMD's
+    [hipTENSOR](https://rocm.docs.amd.com/projects/hipTensor/en/latest/index.html) library.
   - Fall-back support for `AbstractArray` instances which are not strided, by using a
     general implementation using only constructions from Julia's `Base` and `LinearAlgebra`
     modules (`reshape`, `permutedims`, `mul!`, etc.) and some indexing in the case of

docs/src/man/indexnotation.md

@@ -471,16 +471,17 @@ simple wrapper over the cuTENSOR library of NVidia, and will only be loaded when
 `cuTENSOR.jl` package is loaded. The `@tensor` macro will then automatically work for
 operations between GPU arrays.
 
-Mixed operations between host arrays (e.g. `Array`) and device arrays (e.g. `CuArray`) will
-fail. However, if one wants to harness the computing power of the GPU to perform all tensor
-operations, there is a dedicated macro `@cutensor`. This will transfer all host arrays to
-the GPU before performing the requested operations. If the output is an existing host array,
-the result will be copied back. If a new result array is created (i.e. using `:=`), it will
-remain on the GPU device and it is up to the user to transfer it back. Arrays are transfered
-to the GPU just before they are first used, and in a complicated tensor expression, this
-might have the benefit that transer of the later arrays overlaps with computation of earlier
-operations.
+Mixed operations between host arrays (e.g. `Array`) and device arrays (e.g. `CuArray` or `ROCArray`)
+will fail. However, if one wants to harness the computing power of the GPU to perform all tensor
+operations, there is a dedicated macro `@cutensor` for CUDA devices, or `@hiptensor` for AMD.
+This will transfer all host arrays to the GPU before performing the requested operations.
+If the output is an existing host array, the result will be copied back.
+If a new result array is created (i.e. using `:=`), it will remain on the GPU device and it
+is up to the user to transfer it back. Arrays are transfered to the GPU just before they are
+first used, and in a complicated tensor expression, this might have the benefit that transer
+of the later arrays overlaps with computation of earlier operations.
 
 ```@docs
 @cutensor
+@hiptensor
 ```