Allow BraidingTensor to have a custom storage type by kshyatt · Pull Request #393 · QuantumKitHub/TensorKit.jl

kshyatt · 2026-04-03T09:27:20Z

This is probably the nicest way to unblock the various MPSKit changes. I found the similarmatrixtype approach the least gross looking way to achieve this, but open to other ideas.

kshyatt · 2026-04-06T13:44:03Z

The only thing missing here is the parser preprocessing for @planar, where we need to extract the storagetype to hand it over to the BraidingTensor constructor.

Jutho · 2026-04-07T21:03:19Z

I will try to review asap, but I first have to catch up on the latest status of the treetransform stuff, as I never managed to review that part of the vectorize fusiontrees PR.

codecov · 2026-04-14T12:47:00Z

Codecov Report

❌ Patch coverage is 83.33333% with 13 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/tensors/braidingtensor.jl	83.92%	9 Missing ⚠️
ext/TensorKitAdaptExt.jl	50.00%	2 Missing ⚠️
ext/TensorKitCUDAExt/TensorKitCUDAExt.jl	77.77%	2 Missing ⚠️

Files with missing lines	Coverage Δ
ext/TensorKitCUDAExt/cutensormap.jl	`76.82% <100.00%> (+2.85%)`	⬆️
ext/TensorKitAdaptExt.jl	`41.66% <50.00%> (-8.34%)`	⬇️
ext/TensorKitCUDAExt/TensorKitCUDAExt.jl	`80.00% <77.77%> (-20.00%)`	⬇️
src/tensors/braidingtensor.jl	`68.07% <83.92%> (+1.18%)`	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

kshyatt · 2026-04-15T13:09:13Z

Anything more need doing here or is this ready to go after #389?

lkdvos · 2026-04-16T14:35:20Z

        # partial construction: only construct rowr and colr when needed
    end
 end
+function BraidingTensor{T, S}(V1::S, V2::S, ::Type{A}, adjoint::Bool = false) where {T, S <: IndexSpace, A}


I think I'm not completely convinced by this constructor syntax, it feels like at this point we might as well just call BraidingTensor{T, S, A} directly, rather than including the Type{A} as an argument. This also just cuts down slightly on the number of constructors we need to have, under the less-is-more strategy :)

lkdvos · 2026-04-16T14:44:29Z

+    m * n == 0  && return data # s ∉ blocksectors(b)
+    fill!(data, zero(eltype(b)))
+
+    if sectortype(b) === Trivial


Mostly out of curiosity, is this an optimization or was this required?

Sorry, what's "this"?

My bad, I meant the split between the trivial and non-trivial implementation.

Oh it was so that both could dispatch to set_sublock (which I could specialize) but do their own setup, and to cleanup a quite long function.

lkdvos · 2026-04-16T14:45:57Z

+        h′ = force_planar(h)
+
+        for alloc in
+            (TensorOperations.DefaultAllocator(), TensorOperations.CUDAAllocator())


I think this might just run the same test twice, (although I might be wrong) does the default allocator not coincide with the CUDAAllocator here?

lkdvos · 2026-04-16T14:47:15Z

 projects = ["test"]

+[extras]
+GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"


This might be a merge error, and probably belongs in a different project.toml?

Ah no this was here because of a GPUArrays bug that's now fixed

lkdvos · 2026-04-16T14:48:39Z

+
+
+function TensorKit.add_kernel_nonthreaded!(
+        ::TensorKit.FusionStyle,
+        tdst::CuTensorMap, tsrc::CuTensorMap, p, transformer::TensorKit.GenericTreeTransformer, α, β, backend...
+    )
+    # preallocate buffers
+    buffers = TensorKit.allocate_buffers(tdst, tsrc, transformer)
+
+    for subtransformer in transformer.data
+        # Special case without intermediate buffers whenever there is only a single block
+        if length(subtransformer[1]) == 1
+            TensorKit._add_transform_single!(tdst, tsrc, p, subtransformer, α, β, backend...)
+        else
+            cu_subtransformer = tuple(CUDA.adapt(CuArray, subtransformer[1]), subtransformer[2:end]...)
+            TensorKit._add_transform_multi!(tdst, tsrc, p, cu_subtransformer, buffers, α, β, backend...)
+        end
+    end
+    return nothing
+end


This probably is something separate from this PR right? (Also, might no longer be necessary of the mul! specializations of StridedViews are correctly handled in the later versions of Strided.jl

No, this was needed for a test to work, let me look up which one in a moment...

Co-authored-by: Lukas Devos <ldevos98@gmail.com>

github-actions · 2026-04-16T16:09:20Z

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic main) to apply these changes.

Click here to view the suggested changes.

diff --git a/src/tensors/braidingtensor.jl b/src/tensors/braidingtensor.jl
index 49d1b77..7d873ed 100644
--- a/src/tensors/braidingtensor.jl
+++ b/src/tensors/braidingtensor.jl
@@ -113,7 +113,7 @@ end
 
 function _set_subblock!(data, val)
     f(I) = ((I[1] == I[4]) & (I[2] == I[3])) * val
-    data .= f.(CartesianIndices(data))
+    return data .= f.(CartesianIndices(data))
 end

kshyatt requested review from Jutho and borisdevos April 3, 2026 09:27

kshyatt mentioned this pull request Apr 6, 2026

Overhaul test infrastructure #389

Merged

kshyatt mentioned this pull request Apr 8, 2026

More tweaks #375

Draft

kshyatt force-pushed the ksh/cubt branch from c69cc40 to c7afa63 Compare April 14, 2026 11:57

Jutho reviewed Apr 14, 2026

View reviewed changes

Comment thread src/tensors/braidingtensor.jl Outdated

Jutho reviewed Apr 14, 2026

View reviewed changes

Comment thread src/tensors/braidingtensor.jl

kshyatt marked this pull request as ready for review April 15, 2026 13:09

kshyatt and others added 10 commits April 15, 2026 14:57

Allow BraidingTensor to have a custom storage type

0190c09

Use planar instead

d2aaf55

Preserve memory type

fc0020f

Many more working things

b7339ec

Fix ambiguitiy

b0aff17

Use approx for some GPU permutes

3653185

Remove unneeded change

fc3edc2

Trim promote_storagetype

26e6959

Remove similarmatrixtype

b8c9d00

Remove dead import

6d0049a

kshyatt force-pushed the ksh/cubt branch from ab02606 to 6d0049a Compare April 15, 2026 19:00

kshyatt added 3 commits April 16, 2026 04:41

Some preamble fixes

6577957

Small fixes

0fd01cf

Protect us from GPUArrays RNG change

99ddd05

lkdvos reviewed Apr 16, 2026

View reviewed changes

Update src/tensors/braidingtensor.jl

a268464

Co-authored-by: Lukas Devos <ldevos98@gmail.com>

Conversation

kshyatt commented Apr 3, 2026

Uh oh!

kshyatt commented Apr 6, 2026

Uh oh!

Jutho commented Apr 7, 2026

Uh oh!

codecov bot commented Apr 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Uh oh!

kshyatt commented Apr 15, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Apr 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov bot commented Apr 14, 2026 •

edited

Loading