Add FastLapackInterface as dependency to LinearSolveAutotune

- FastLapackInterface is now a direct dependency of LinearSolveAutotune
- Simplified the algorithm detection logic since it's always available
- Updated documentation to reflect that users don't need to load it separately
- The include_fastlapack parameter still defaults to false for backward compatibility

Author: ChrisRackauckas

lib/LinearSolveAutotune/Project.toml

@@ -10,6 +10,7 @@ CPUSummary = "2a0fbf3d-bb9c-48f3-b0a9-814d99fd7ab9"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
+FastLapackInterface = "29a986be-02c6-4525-aec4-84b980013641"
 GitHub = "bc5e4493-9b4d-5f90-b8aa-2b2bcaad7a26"
 LAPACK_jll = "51474c39-65e3-53ba-86ba-03b1b862ec14"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
@@ -36,6 +37,7 @@ CPUSummary = "0.2"
 CUDA = "5"
 DataFrames = "1"
 Dates = "1"
+FastLapackInterface = "2"
 GitHub = "5"
 LAPACK_jll = "3"
 LinearAlgebra = "1"

lib/LinearSolveAutotune/src/LinearSolveAutotune.jl

@@ -36,6 +36,7 @@ using blis_jll
 using LAPACK_jll
 using CUDA
 using Metal
+using FastLapackInterface
 
 
 # Optional dependencies for telemetry and plotting
@@ -180,7 +181,7 @@ Run a comprehensive benchmark of all available LU factorization methods and opti
   - `seconds::Float64 = 0.5`: Maximum time per benchmark
   - `eltypes = (Float32, Float64, ComplexF32, ComplexF64)`: Element types to benchmark
   - `skip_missing_algs::Bool = false`: If false, error when expected algorithms are missing; if true, warn instead
-  - `include_fastlapack::Bool = false`: If true and FastLapackInterface.jl is loaded, includes FastLUFactorization in benchmarks
+  - `include_fastlapack::Bool = false`: If true, includes FastLUFactorization in benchmarks
 
 # Returns
 
@@ -201,8 +202,7 @@ results = autotune_setup(sizes = [:small, :medium, :large, :big])
 # Large matrices only
 results = autotune_setup(sizes = [:large, :big], samples = 10, seconds = 1.0)
 
-# Include FastLapackInterface.jl algorithms if available
-using FastLapackInterface
+# Include FastLapackInterface.jl algorithms
 results = autotune_setup(include_fastlapack = true)
 
 # After running autotune, share results (requires gh CLI or GitHub token)

lib/LinearSolveAutotune/src/algorithms.jl

@@ -5,7 +5,7 @@
 
 Returns a list of available LU factorization algorithms based on the system and loaded packages.
 If skip_missing_algs=false, errors when expected algorithms are missing; if true, warns instead.
-If include_fastlapack=true and FastLapackInterface is loaded, includes FastLUFactorization in benchmarks.
+If include_fastlapack=true, includes FastLUFactorization in benchmarks.
 """
 function get_available_algorithms(; skip_missing_algs::Bool = false, include_fastlapack::Bool = false)
     algs = []
@@ -69,21 +69,10 @@ function get_available_algorithms(; skip_missing_algs::Bool = false, include_fas
     push!(algs, SimpleLUFactorization())
     push!(alg_names, "SimpleLUFactorization")
 
-    # FastLapackInterface LU if requested and available
+    # FastLapackInterface LU if requested (always available as dependency)
     if include_fastlapack
-        try
-            # Try to create a FastLUFactorization to see if the extension is loaded
-            test_alg = FastLUFactorization()
-            push!(algs, test_alg)
-            push!(alg_names, "FastLUFactorization")
-            @info "FastLUFactorization included in benchmarks"
-        catch e
-            if occursin("UndefVarError", string(e))
-                @warn "FastLUFactorization requested but FastLapackInterface.jl not loaded. Load it with: using FastLapackInterface"
-            else
-                @warn "FastLUFactorization requested but not available: $e"
-            end
-        end
+        push!(algs, FastLUFactorization())
+        push!(alg_names, "FastLUFactorization")
     end
 
     return algs, alg_names