Migrate from LinearAlgebraMPI to HPCLinearAlgebra

Update MultiGridBarrierMPI to use the renamed HPCLinearAlgebra package
with its new backend-based API:

- Update imports to use HPCLinearAlgebra and new type names (HPCVector,
  HPCMatrix, HPCSparseMatrix)
- Update constructors to use backend parameter: HPCMatrix(data, backend)
- Add _backend_instance_from_type() to support CUDA/Metal backends in
  amgb_zeros for vectors
- Make GPU backend creation lazy in test_utils.jl to avoid calling
  backend_cuda_mpi() before MPI.Init()
- Add test_2d.jl for 2D integration tests with CUDA support
- Fix runtests.jl to use 2 ranks for 2D tests (NCCL fails when ranks
  exceed available GPUs)
- Update documentation references

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: Sebastien Loisel

Project.toml

@@ -1,24 +1,39 @@
 name = "MultiGridBarrierMPI"
 uuid = "abf18f27-d12f-4566-94ed-07bf0c385f70"
-version = "0.1.1"
 authors = ["S. Loisel"]
+version = "0.1.1"
 
 [deps]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+CUDSS_jll = "4889d778-9329-5762-9fec-0578a5d30366"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
-LinearAlgebraMPI = "5bdd2be4-ae34-42ef-8b36-f4c85d48f377"
+HPCLinearAlgebra = "537374f1-5608-4525-82fb-641dce542540"
 MPI = "da04e1cc-30fd-572f-bb4f-1f8673147195"
+MPIPreferences = "3da0fdf6-3ccc-4f1b-acd9-58baa6c99267"
+Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
 MultiGridBarrier = "9e2c1f1d-9131-4ad4-b32f-bd2a0b0ecd1e"
+NCCL_jll = "4d6d38e4-5b87-5e63-912a-873ff2d649b7"
 PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
+[sources]
+HPCLinearAlgebra = {path = "../LinearAlgebraMPI.jl"}
+
 [compat]
 BenchmarkTools = "1.6"
-LinearAlgebraMPI = "0.1"
+CUDA = "5.9.6"
+CUDSS_jll = "0.7.1"
+HPCLinearAlgebra = "0.1"
 MPI = "0.20"
+MPIPreferences = "0.1.11"
+Metal = "1.9.1"
 MultiGridBarrier = "0.11"
+NCCL_jll = "2"
 PrecompileTools = "1"
+StaticArrays = "1.9.16"
 julia = "1.10"
 
 [extras]

README.md

@@ -7,17 +7,17 @@
 
 **Author:** S. Loisel
 
-A Julia package that bridges MultiGridBarrier.jl and LinearAlgebraMPI.jl for distributed multigrid barrier computations using native MPI types.
+A Julia package that bridges MultiGridBarrier.jl and HPCLinearAlgebra.jl for distributed multigrid barrier computations using native MPI types.
 
 ## Overview
 
-MultiGridBarrierMPI.jl extends the MultiGridBarrier.jl package to work with LinearAlgebraMPI.jl's distributed matrix and vector types. This enables efficient parallel computation of multigrid barrier methods across multiple MPI ranks without requiring PETSc.
+MultiGridBarrierMPI.jl extends the MultiGridBarrier.jl package to work with HPCLinearAlgebra.jl's distributed matrix and vector types. This enables efficient parallel computation of multigrid barrier methods across multiple MPI ranks without requiring PETSc.
 
 ## Key Features
 
 - **1D, 2D, and 3D Support**: Full support for 1D, 2D triangular, and 3D hexahedral finite elements
 - **Seamless Integration**: Drop-in replacement for MultiGridBarrier's native types
-- **Pure Julia MPI**: Uses LinearAlgebraMPI.jl for distributed linear algebra (no external libraries required)
+- **Pure Julia MPI**: Uses HPCLinearAlgebra.jl for distributed linear algebra (no external libraries required)
 - **Type Conversion**: Easy conversion between native Julia arrays and MPI distributed types
 - **MPI-Aware**: All operations correctly handle MPI collective requirements
 - **MUMPS Solver**: Uses MUMPS direct solver for accurate Newton iterations
@@ -31,7 +31,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # Solve with MPI distributed types (L=3 refinement levels)
 sol_mpi = fem2d_mpi_solve(Float64; L=3, p=1.0, verbose=false)
@@ -92,7 +92,7 @@ julia --project make.jl
 This package is part of a larger ecosystem:
 
 - **[MultiGridBarrier.jl](https://github.com/sloisel/MultiGridBarrier.jl)**: Core multigrid barrier method implementation
-- **[LinearAlgebraMPI.jl](https://github.com/sloisel/LinearAlgebraMPI.jl)**: Pure Julia distributed linear algebra with MPI
+- **[HPCLinearAlgebra.jl](https://github.com/sloisel/HPCLinearAlgebra.jl)**: Pure Julia distributed linear algebra with MPI
 - **MPI.jl**: Julia MPI bindings for distributed computing
 
 ## Requirements

docs/Project.toml

@@ -1,7 +1,7 @@
 [deps]
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 DocumenterTools = "35a29f4d-8980-5a13-9543-d66fff28ecb8"
-LinearAlgebraMPI = "5bdd2be4-ae34-42ef-8b36-f4c85d48f377"
+HPCLinearAlgebra = "537374f1-5608-4525-82fb-641dce542540"
 MultiGridBarrier = "9e2c1f1d-9131-4ad4-b32f-bd2a0b0ecd1e"
 MultiGridBarrierMPI = "abf18f27-d12f-4566-94ed-07bf0c385f70"
 

docs/src/api.md

@@ -48,19 +48,19 @@ When converting from native Julia types to MPI distributed types:
 
 | Native Type | MPI Type | Usage |
 |-------------|----------|-------|
-| `Matrix{T}` | `MatrixMPI{T}` | Geometry coordinates, dense data |
-| `Vector{T}` | `VectorMPI{T}` | Weights, dense vectors |
-| `SparseMatrixCSC{T,Int}` | `SparseMatrixMPI{T}` | Sparse operators, subspace matrices |
+| `Matrix{T}` | `HPCMatrix{T}` | Geometry coordinates, dense data |
+| `Vector{T}` | `HPCVector{T}` | Weights, dense vectors |
+| `SparseMatrixCSC{T,Int}` | `HPCSparseMatrix{T}` | Sparse operators, subspace matrices |
 
 ### MPI to Native Conversions
 
 When converting from MPI distributed types back to native Julia types:
 
 | MPI Type | Native Type |
 |----------|-------------|
-| `MatrixMPI{T}` | `Matrix{T}` |
-| `VectorMPI{T}` | `Vector{T}` |
-| `SparseMatrixMPI{T}` | `SparseMatrixCSC{T,Int}` |
+| `HPCMatrix{T}` | `Matrix{T}` |
+| `HPCVector{T}` | `Vector{T}` |
+| `HPCSparseMatrix{T}` | `SparseMatrixCSC{T,Int}` |
 
 ## Geometry Structure
 
@@ -73,14 +73,14 @@ Geometry{T, Matrix{T}, Vector{T}, SparseMatrixCSC{T,Int}, Discretization}
 
 **MPI Geometry:**
 ```julia
-Geometry{T, MatrixMPI{T}, VectorMPI{T}, SparseMatrixMPI{T}, Discretization}
+Geometry{T, HPCMatrix{T}, HPCVector{T}, HPCSparseMatrix{T}, Discretization}
 ```
 
 ### Fields
 
 - **`discretization`**: Discretization information (domain, mesh, etc.)
-- **`x`**: Geometry coordinates (Matrix or MatrixMPI)
-- **`w`**: Quadrature weights (Vector or VectorMPI)
+- **`x`**: Geometry coordinates (Matrix or HPCMatrix)
+- **`w`**: Quadrature weights (Vector or HPCVector)
 - **`operators`**: Dictionary of operators (id, dx, dy, etc.)
 - **`subspaces`**: Dictionary of subspace projection matrices
 - **`refine`**: Vector of refinement matrices (coarse -> fine)
@@ -104,12 +104,12 @@ The `AMGBSOL` type from MultiGridBarrier contains the complete solution:
 
 ## MPI and IO Utilities
 
-### LinearAlgebraMPI.io0()
+### HPCLinearAlgebra.io0()
 
 Returns an IO stream that only writes on rank 0:
 
 ```julia
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 println(io0(), "This prints once from rank 0")
 ```
@@ -132,7 +132,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 using LinearAlgebra
 
@@ -163,7 +163,7 @@ id_native = g_native.operators[:id]  # SparseMatrixCSC
 
 # MPI geometry
 g_mpi = native_to_mpi(g_native)
-id_mpi = g_mpi.operators[:id]  # SparseMatrixMPI
+id_mpi = g_mpi.operators[:id]  # HPCSparseMatrix
 
 # Convert back if needed
 id_back = SparseMatrixCSC(id_mpi)  # SparseMatrixCSC

docs/src/guide.md

@@ -30,7 +30,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 
 # Step 1: Solve with MPI distributed types
@@ -100,9 +100,9 @@ g_mpi = native_to_mpi(g_native)
 
 | Native Type | MPI Type | Description |
 |-------------|----------|-------------|
-| `Matrix{T}` | `MatrixMPI{T}` | Dense distributed matrix |
-| `Vector{T}` | `VectorMPI{T}` | Dense distributed vector |
-| `SparseMatrixCSC{T,Int}` | `SparseMatrixMPI{T}` | Sparse distributed matrix |
+| `Matrix{T}` | `HPCMatrix{T}` | Dense distributed matrix |
+| `Vector{T}` | `HPCVector{T}` | Dense distributed vector |
+| `SparseMatrixCSC{T,Int}` | `HPCSparseMatrix{T}` | Sparse distributed matrix |
 
 ### MPI to Native
 
@@ -135,7 +135,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 
 # 1. Create native geometry with specific parameters
@@ -168,7 +168,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 using LinearAlgebra
 
@@ -193,10 +193,10 @@ end
 
 ### Printing from One Rank
 
-Use `io0()` from LinearAlgebraMPI to print from rank 0 only:
+Use `io0()` from HPCLinearAlgebra to print from rank 0 only:
 
 ```julia
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # This prints once (from rank 0)
 println(io0(), "Hello from rank 0!")
@@ -268,7 +268,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # Solve a 1D problem with 4 multigrid levels (2^4 = 16 elements)
 sol = fem1d_mpi_solve(Float64; L=4, p=1.0, verbose=true)
@@ -297,7 +297,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # Solve a 2D problem
 sol = fem2d_mpi_solve(Float64; L=2, p=1.0, verbose=true)
@@ -327,7 +327,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # Solve a 3D problem with Q3 elements and 2 multigrid levels
 sol = fem3d_mpi_solve(Float64; L=2, k=3, p=1.0, verbose=true)
@@ -357,7 +357,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 
 # Create MPI geometry
@@ -393,7 +393,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 sol = fem2d_mpi_solve(Float64; L=3, p=1.0)
 sol_native = mpi_to_native(sol)

docs/src/index.md

@@ -10,17 +10,17 @@ v = string(pkgversion(MultiGridBarrierMPI))
 md"# MultiGridBarrierMPI.jl $v"
 ```
 
-**A Julia package that bridges MultiGridBarrier.jl and LinearAlgebraMPI.jl for distributed multigrid barrier computations.**
+**A Julia package that bridges MultiGridBarrier.jl and HPCLinearAlgebra.jl for distributed multigrid barrier computations.**
 
 ## Overview
 
-MultiGridBarrierMPI.jl extends the MultiGridBarrier.jl package to work with LinearAlgebraMPI.jl's distributed matrix and vector types. This enables efficient parallel computation of multigrid barrier methods across multiple MPI ranks using pure Julia distributed types (no PETSc required).
+MultiGridBarrierMPI.jl extends the MultiGridBarrier.jl package to work with HPCLinearAlgebra.jl's distributed matrix and vector types. This enables efficient parallel computation of multigrid barrier methods across multiple MPI ranks using pure Julia distributed types (no PETSc required).
 
 ## Key Features
 
 - **1D, 2D, and 3D Support**: Full support for 1D elements, 2D triangular, and 3D hexahedral finite elements
 - **Seamless Integration**: Drop-in replacement for MultiGridBarrier's native types
-- **Pure Julia MPI**: Uses LinearAlgebraMPI.jl for distributed linear algebra
+- **Pure Julia MPI**: Uses HPCLinearAlgebra.jl for distributed linear algebra
 - **Type Conversion**: Easy conversion between native Julia arrays and MPI distributed types
 - **MPI-Aware**: All operations correctly handle MPI collective requirements
 - **MUMPS Solver**: Uses MUMPS direct solver for accurate Newton iterations
@@ -34,7 +34,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 using MultiGridBarrier
 
 # Solve with MPI distributed types (L=3 refinement levels)
@@ -71,7 +71,7 @@ Depth = 2
 This package is part of a larger ecosystem:
 
 - **[MultiGridBarrier.jl](https://github.com/sloisel/MultiGridBarrier.jl)**: Core multigrid barrier method implementation (1D, 2D, and 3D)
-- **[LinearAlgebraMPI.jl](https://github.com/sloisel/LinearAlgebraMPI.jl)**: Pure Julia distributed linear algebra with MPI
+- **[HPCLinearAlgebra.jl](https://github.com/sloisel/HPCLinearAlgebra.jl)**: Pure Julia distributed linear algebra with MPI
 - **MPI.jl**: Julia MPI bindings for distributed computing
 
 ## Requirements

docs/src/installation.md

@@ -10,7 +10,7 @@ For HPC environments, you may want to configure MPI.jl to use your system's MPI
 
 ### MUMPS
 
-The package uses MUMPS for sparse direct solves through LinearAlgebraMPI.jl. MUMPS is typically available through your system's package manager or HPC module system.
+The package uses MUMPS for sparse direct solves through HPCLinearAlgebra.jl. MUMPS is typically available through your system's package manager or HPC module system.
 
 ## Package Installation
 
@@ -75,7 +75,7 @@ using MPI
 MPI.Init()
 
 using MultiGridBarrierMPI
-using LinearAlgebraMPI
+using HPCLinearAlgebra
 
 # Your parallel code here
 sol = fem2d_mpi_solve(Float64; L=3, p=1.0)
@@ -89,7 +89,7 @@ mpiexec -n 4 julia --project my_program.jl
 ```
 
 !!! tip "Output from Rank 0 Only"
-    Use `io0()` from LinearAlgebraMPI for output to avoid duplicate messages:
+    Use `io0()` from HPCLinearAlgebra for output to avoid duplicate messages:
     ```julia
     println(io0(), "This prints once from rank 0")
     ```
@@ -106,7 +106,7 @@ using Pkg; Pkg.build("MPI")
 
 ### MUMPS Issues
 
-If MUMPS fails to load, ensure it's properly installed on your system and that LinearAlgebraMPI.jl can find it.
+If MUMPS fails to load, ensure it's properly installed on your system and that HPCLinearAlgebra.jl can find it.
 
 ### Test Failures