Merge branch 'main' into ksh/cuda_tweaks

Author: kshyatt

.github/workflows/CompatCheck.yml

@@ -49,7 +49,7 @@ jobs:
       matrix:
         downgrade_mode: ['deps']
         group: ${{ fromJSON(needs.setup-matrix.outputs.groups) }}
-        julia-version: ['1', '1.10']
+        julia-version: ['1.10']
     steps:
       - uses: actions/checkout@v6
       - uses: julia-actions/setup-julia@v3

CITATION.cff

@@ -8,9 +8,9 @@ authors:
   given-names: "Jutho"
   orcid: "https://orcid.org/0000-0002-0858-291X"
 title: "TensorKit.jl"
-version: "0.16.3"
+version: "0.16.4"
 doi: "10.5281/zenodo.8421339"
-date-released: "2026-02-22"
+date-released: "2026-04-23"
 url: "https://github.com/QuantumKitHub/TensorKit.jl"
 preferred-citation:
   type: article

Project.toml

@@ -1,6 +1,6 @@
 name = "TensorKit"
 uuid = "07d1fe3e-3e46-537d-9eac-e9e13d0d4cec"
-version = "0.16.3"
+version = "0.17.0"
 authors = ["Jutho Haegeman, Lukas Devos"]
 
 [deps]

docs/src/Changelog.md

@@ -18,22 +18,42 @@ When making changes to this project, please update the "Unreleased" section with
 
 When releasing a new version, move the "Unreleased" changes to a new version section with the release date.
 
-## [Unreleased](https://github.com/QuantumKitHub/TensorKit.jl/compare/v0.16.3...HEAD)
+## [Unreleased](https://github.com/QuantumKitHub/TensorKit.jl/compare/v0.16.4...HEAD)
 
 ### Added
 
-
 ### Changed
 
-
 ### Deprecated
 
-
 ### Removed
 
+### Fixed
+
+### Performance
+
+## [0.16.4](https://github.com/QuantumKitHub/TensorKit.jl/compare/v0.16.3...v0.16.4) - 2026-04-23
+
+### Added
+
+- Partial tensor support for AMDGPU via a new extension ([#341](https://github.com/QuantumKitHub/TensorKit.jl/pull/341))
+- Define `spacetype` for `TruncationSpace` ([#403](https://github.com/QuantumKitHub/TensorKit.jl/pull/403))
+
+### Changed
+
+- Updated MatrixAlgebraKit dependency to v0.6.5 with corresponding API updates ([#390](https://github.com/QuantumKitHub/TensorKit.jl/pull/390))
 
 ### Fixed
 
+- Fix ignored `adjoint` flag in `BraidingTensor` ([#392](https://github.com/QuantumKitHub/TensorKit.jl/pull/392))
+- Fix `MethodError` for certain tensor operations ([#406](https://github.com/QuantumKitHub/TensorKit.jl/pull/406))
+- Add square checks for `project_(anti)hermitian` and eigenvalue decompositions ([#408](https://github.com/QuantumKitHub/TensorKit.jl/pull/408))
+
+### Performance
+
+- Vectorize fusiontree manipulations ([#261](https://github.com/QuantumKitHub/TensorKit.jl/pull/261))
+- Avoid generic matmul fallback in transformation kernel ([#378](https://github.com/QuantumKitHub/TensorKit.jl/pull/378))
+- Reduce cache footprint by decoupling degeneracy-dependent data ([#387](https://github.com/QuantumKitHub/TensorKit.jl/pull/387))
 
 ## [0.16.3](https://github.com/QuantumKitHub/TensorKit.jl/compare/v0.16.2...v0.16.3) - 2026-02-22
 

ext/TensorKitAdaptExt.jl

@@ -15,8 +15,12 @@ function Adapt.adapt_structure(to, x::DiagonalTensorMap)
     data′ = adapt(to, x.data)
     return DiagonalTensorMap(data′, x.domain)
 end
-function Adapt.adapt_structure(::Type{TorA}, x::BraidingTensor) where {TorA <: Union{Number, DenseArray{<:Number}}}
-    return BraidingTensor{scalartype(TorA)}(space(x), x.adjoint)
+function Adapt.adapt_structure(::Type{T}, x::BraidingTensor{T′, S, A}) where {T <: Number, T′, S, A}
+    A′ = TensorKit.similarstoragetype(A, T)
+    return BraidingTensor{T, S, A′}(space(x), x.adjoint)
+end
+function Adapt.adapt_structure(::Type{TA}, x::BraidingTensor{T, S, A}) where {T′, TA <: DenseArray{T′}, T, S, A}
+    return BraidingTensor{T′, S, TA}(space(x), x.adjoint)
 end
 
 end

ext/TensorKitCUDAExt/TensorKitCUDAExt.jl

@@ -3,14 +3,16 @@ module TensorKitCUDAExt
 using CUDA, CUDA.CUBLAS, CUDA.CUSOLVER, LinearAlgebra
 using CUDA: @allowscalar
 using cuTENSOR: cuTENSOR
+using Strided: StridedViews
 import CUDA: rand as curand, rand! as curand!, randn as curandn, randn! as curandn!
+using CUDA.KernelAbstractions: @kernel, @index, get_backend
 
 using TensorKit
 using TensorKit.Factorizations
 using TensorKit.Strided
 using TensorKit.Factorizations: AbstractAlgorithm
 using TensorKit: SectorDict, tensormaptype, scalar, similarstoragetype, AdjointTensorMap, scalartype, project_symmetric_and_check
-import TensorKit: randisometry, rand, randn
+import TensorKit: randisometry, rand, randn, fill_braidingsubblock!
 
 using TensorKit: MatrixAlgebraKit
 
@@ -19,4 +21,18 @@ using Random
 include("cutensormap.jl")
 include("truncation.jl")
 
+function TensorKit.fill_braidingsubblock!(data::TD, val) where {T, TD <: Union{<:CuMatrix{T}, <:StridedViews.StridedView{T, 4, <:CuArray{T}}}}
+    # COV_EXCL_START
+    # kernels are not reachable by coverage
+    @kernel function fill_subblock_kernel!(subblock, val)
+        idx = @index(Global, Cartesian)
+        idx_val = idx[1] == idx[4] && idx[2] == idx[3] ? val : zero(val)
+        @inbounds subblock[idx] = idx_val
+    end
+    # COV_EXCL_STOP
+    kernel = fill_subblock_kernel!(get_backend(data))
+    kernel(data, val; ndrange = size(data))
+    return data
+end
+
 end

ext/TensorKitCUDAExt/cutensormap.jl

@@ -158,3 +158,7 @@ for f in (:sqrt, :log, :asin, :acos, :acosh, :atanh, :acoth)
         return tf
     end
 end
+
+function TensorKit._add_transform_multi!(tdst::CuTensorMap, tsrc, p, (U, structs_dst, structs_src)::Tuple{<:Array, TD, TS}, buffers, alpha, beta, backend...) where {TD, TS}
+    return TensorKit._add_transform_multi!(tdst, tsrc, p, (CUDA.Adapt.adapt(CuArray, U), structs_dst, structs_src), buffers, alpha, beta, backend...)
+end

src/planar/preprocessors.jl

@@ -83,6 +83,23 @@ _add_adjoint(ex) = Expr(TO.prime, ex)
 # spaces from the rest of the expression. Construct the explicit BraidingTensor objects and
 # insert them in the expression.
 function _construct_braidingtensors(ex)
+    function filter_f(expr)
+        if TO.istensor(expr)
+            return _remove_adjoint(TO.decomposetensor(expr)[1]) != :τ
+        elseif TO.istensorexpr(expr)
+            return any(filter_f, expr.args)
+        else
+            return false
+        end
+    end
+    function extract_tensors(tensor_ex)
+        if TO.istensor(tensor_ex)
+            return [TO.decomposetensor(tensor_ex)[1]]
+        elseif TO.istensorexpr(tensor_ex)
+            return collect(Iterators.flatmap(extract_tensors, filter(filter_f, tensor_ex.args)))
+        end
+    end
+    # get storagetype
     ex isa Expr || return ex
     if ex.head == :macrocall && ex.args[1] == Symbol("@notensor")
         return ex
@@ -104,7 +121,9 @@ function _construct_braidingtensors(ex)
                 )
             end
         end
-        newrhs, success = _construct_braidingtensors!(rhs, preargs, indexmap)
+        # if this is a definition, the lhs tensor is NOT yet defined
+        no_τ_ex = reduce(vcat, Iterators.flatmap(extract_tensors, filter(filter_f, rhs.args)); init = Symbol[])
+        newrhs, success = _construct_braidingtensors!(rhs, preargs, indexmap, no_τ_ex)
         success ||
             throw(ArgumentError("cannot determine the spaces of all braiding tensors in $ex"))
         pre = Expr(
@@ -115,7 +134,8 @@ function _construct_braidingtensors(ex)
     elseif TO.istensorexpr(ex)
         preargs = Vector{Any}()
         indexmap = Dict{Any, Any}()
-        newex, success = _construct_braidingtensors!(ex, preargs, indexmap)
+        no_τ_ex = reduce(vcat, Iterators.flatmap(extract_tensors, filter(filter_f, ex.args)); init = Symbol[])
+        newex, success = _construct_braidingtensors!(ex, preargs, indexmap, no_τ_ex)
         success ||
             throw(ArgumentError("cannot determine the spaces of all braiding tensors in $ex"))
         pre = Expr(
@@ -128,7 +148,7 @@ function _construct_braidingtensors(ex)
     end
 end
 
-function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed to be a single tensor expression
+function _construct_braidingtensors!(ex, preargs, indexmap, non_braiding) # ex is guaranteed to be a single tensor expression
     if TO.isscalarexpr(ex)
         # ex could be tensorscalar call with more braiding tensors
         return _construct_braidingtensors(ex), true
@@ -163,7 +183,9 @@ function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed t
             end
             if foundV1 && foundV2
                 s = gensym(:τ)
-                constructex = Expr(:call, GlobalRef(TensorKit, :BraidingTensor), V1, V2)
+                storageex = Expr(:call, GlobalRef(TensorKit, :promote_storagetype), non_braiding...)
+                braidingex = Expr(:call, GlobalRef(TensorKit, :braidingtensortype), V1, V2, storageex)
+                constructex = Expr(:call, braidingex, V1, V2)
                 push!(preargs, Expr(:(=), s, constructex))
                 obj = _is_adjoint(obj) ? _add_adjoint(s) : s
                 success = true
@@ -196,7 +218,7 @@ function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed t
         newargs = Vector{Any}(undef, length(args))
         success = true
         for i in 1:length(ex.args)
-            newargs[i], successa = _construct_braidingtensors!(args[i], preargs, indexmap)
+            newargs[i], successa = _construct_braidingtensors!(args[i], preargs, indexmap, non_braiding)
             success = success && successa
         end
         newex = Expr(ex.head, newargs...)
@@ -212,7 +234,7 @@ function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed t
             for i in 2:length(ex.args)
                 successes[i] && continue
                 newargs[i], successa = _construct_braidingtensors!(
-                    args[i], preargs, indexmap
+                    args[i], preargs, indexmap, non_braiding
                 )
                 successes[i] = successa
             end
@@ -232,7 +254,7 @@ function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed t
         indices = [TO.getindices(arg) for arg in args]
         for i in 2:length(ex.args)
             indexmapa = copy(indexmap)
-            newargs[i], successa = _construct_braidingtensors!(args[i], preargs, indexmapa)
+            newargs[i], successa = _construct_braidingtensors!(args[i], preargs, indexmapa, non_braiding)
             for l in indices[i]
                 if !haskey(indexmap, l) && haskey(indexmapa, l)
                     indexmap[l] = indexmapa[l]
@@ -243,10 +265,10 @@ function _construct_braidingtensors!(ex, preargs, indexmap) # ex is guaranteed t
         newex = Expr(ex.head, newargs...)
         return newex, success
     elseif isexpr(ex, :call) && ex.args[1] == :/ && length(ex.args) == 3
-        newarg, success = _construct_braidingtensors!(ex.args[2], preargs, indexmap)
+        newarg, success = _construct_braidingtensors!(ex.args[2], preargs, indexmap, non_braiding)
         return Expr(:call, :/, newarg, ex.args[3]), success
     elseif isexpr(ex, :call) && ex.args[1] == :\ && length(ex.args) == 3
-        newarg, success = _construct_braidingtensors!(ex.args[3], preargs, indexmap)
+        newarg, success = _construct_braidingtensors!(ex.args[3], preargs, indexmap, non_braiding)
         return Expr(:call, :\, ex.args[2], newarg), success
     else
         error("unexpected expression $ex")

src/spaces/vectorspaces.jl

@@ -409,7 +409,7 @@ while the second will return the spacetype if all types are equal, and throw a [
 """
 check_spacetype(::Type{Bool}, x, y, z...) = _allequal(spacetype, (x, y, z...))
 @noinline function check_spacetype(x, y, z...)
-    check_spacetype(Bool, x, y, z...) || throw(SpaceMismatch("incompatible space types"))
+    check_spacetype(Bool, x, y, z...) || throw(SpaceMismatch(lazy"incompatible space types $(type_repr.(spacetype.((x, y, z...))))"))
     return spacetype(x)
 end