[shukebeta]

Adapt.adapt(target, op) and copy(op) now work for GradientOp, also inside ProdOp, normalOperator and DiagOp. Both return a new operator that is rebuilt from the shape and the directions taken at construction, with a new LinearOperator, closures and temporary vector for the target storage type; op is not modified.

G = GradientOp(Float32; shape=(64, 64))   # CPU
G_gpu = Adapt.adapt(CuArray, G)

• GradientOp now returns a GradientOpImpl <: GradientOp{T} that wraps the LinearOperator and keeps shape, dims and the adjoint's temporary vector, like the other wrapper operators. Properties of the LinearOperator are forwarded.
• dims is iterated once at construction and stored as a tuple, or as an Int for a single direction. Rebuilding does not depend on the caller's container: a vector changed later, or a consumed iterator, has no effect.
• The adapted storage must have the operator's element type. An element-typed target such as CuArray{Float32} for a Float64 operator throws an ArgumentError, and so does an operator built with an S of another element type, whose adjoint rounds through its temporary vector. copy keeps such an operator as it is.
• The GPU gradient kernels no longer set cpu = false, so they also run on CPU array types such as JLArray. Their computation is unchanged.
• test/testAdapt.jl moves GradientOp between Array and JLArray (or the configured GPU array type) for Float32, Float64, ComplexF32, ComplexF64 and Int32, and for single, several, repeated and no directions. It compares forward, transpose and adjoint, with 3- and 5-argument mul!, against the source. It also checks that the source's buffers, temporary vector and counters are unchanged, and that copies get their own temporary vector. It further covers a mutated dims vector, a consumed iterator, createLinearOperator, ProdOp, normalOperator, DiagOp, and the element-type errors.

The tests pass on the CPU with Julia 1.10.12. With two threads, the only failure is the existing DiagOp((op, op)) FFT test from #44.

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.23%. Comparing base (454257e) to head (9fe52df).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
ext/LinearOperatorGPUArraysExt/GradientOp.jl	33.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #48      +/-   ##
==========================================
+ Coverage   73.01%   81.23%   +8.22%     
==========================================
  Files          15       15              
  Lines         593      613      +20     
==========================================
+ Hits          433      498      +65     
+ Misses        160      115      -45     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.