Add Enzyme rules for factorizations by kshyatt · Pull Request #464 · QuantumKitHub/TensorKit.jl

kshyatt · 2026-06-26T00:48:35Z

For some reason svd_compact and svd_trunc_no_error don't play nicely with Enzyme here using the MAK rules, I think because of the DiagonalTensorMap output. I can try to investigate further if preferred. I also added some additional logic in pullbacks to match what the MAK pullbacks actually kick back.

kshyatt · 2026-06-26T00:53:39Z

Oh crap lol I was working on top of a dev-ed MAK

codecov · 2026-06-26T01:34:06Z

Codecov Report

❌ Patch coverage is 0% with 50 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
ext/TensorKitEnzymeExt/factorizations.jl	0.00%	27 Missing ⚠️
src/factorizations/pullbacks.jl	0.00%	21 Missing ⚠️
ext/TensorKitEnzymeExt/utility.jl	0.00%	1 Missing ⚠️
src/factorizations/diagonal.jl	0.00%	1 Missing ⚠️

Files with missing lines	Coverage Δ
ext/TensorKitEnzymeExt/TensorKitEnzymeExt.jl	`100.00% <ø> (ø)`
ext/TensorKitEnzymeExt/utility.jl	`18.18% <0.00%> (-0.21%)`	⬇️
src/factorizations/diagonal.jl	`67.64% <0.00%> (-2.05%)`	⬇️
src/factorizations/pullbacks.jl	`47.69% <0.00%> (-22.77%)`	⬇️
ext/TensorKitEnzymeExt/factorizations.jl	`0.00% <0.00%> (ø)`

... and 6 files with indirect coverage changes

🚀 New features to boost your workflow:

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

kshyatt · 2026-06-26T14:36:48Z

Weirdly the eig failures seem to go away on MatrixAlgebraKit master, let me try with some [sources] and see if it's reproducible outside of my laptop

github-actions · 2026-06-29T19:32:18Z

Your PR no longer requires formatting changes. Thank you for your contribution!

kshyatt · 2026-07-16T14:23:27Z

This map business with block is not the nicest but I thought it was better than letting block(nothing, c) go through silently in other cases

kshyatt · 2026-09-30T18:26:46Z

Gonna make this a draft since 1.10 is going to fail anyway and the forward mode stuff needs some love

kshyatt · 2026-10-13T13:07:41Z

I think the only method definitions that could be collapsed are the qr, polar and lq ones since the _vals ones are already handled by MAK. I'll do that then.

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

Jutho

I think I've checked most things and approve. I haven't read all the tests. Out of curiosity, are these all hand-written or did you get 🤖 help?

kshyatt · 2026-10-13T19:48:17Z

I wrote the tests but pulled a lot from the existing Mooncake and ChainRules ones

kshyatt · 2026-10-14T09:37:50Z

the gpu fail is expected, i'm going to (finally!) merge this

kshyatt requested a review from lkdvos June 26, 2026 00:48

kshyatt marked this pull request as draft June 26, 2026 14:10

kshyatt force-pushed the ksh/enz_fact branch from 4f513a2 to b05080c Compare June 26, 2026 14:39

kshyatt force-pushed the ksh/enz_fact branch 4 times, most recently from e0a9f0c to 9742ab7 Compare July 6, 2026 18:14

kshyatt force-pushed the ksh/enz_fact branch 3 times, most recently from 049c1a0 to a60a6ca Compare July 15, 2026 14:21

kshyatt force-pushed the ksh/enz_fact branch from a60a6ca to d6bc397 Compare July 16, 2026 07:03

kshyatt force-pushed the ksh/enz_fact branch from 700017d to f503e3a Compare July 17, 2026 09:00

kshyatt mentioned this pull request Jul 25, 2026

Very large compilation burden (time & memory) for test_reverse/autodiff_thunk with complicated types EnzymeAD/Enzyme.jl#3392

Open

kshyatt force-pushed the ksh/enz_fact branch 3 times, most recently from 215e3f0 to 78d8a98 Compare August 5, 2026 09:50

kshyatt force-pushed the ksh/enz_fact branch 2 times, most recently from a938fca to 17e422c Compare August 5, 2026 22:06

kshyatt marked this pull request as ready for review August 6, 2026 07:58

kshyatt force-pushed the ksh/enz_fact branch from 17e422c to c4d0f3d Compare September 14, 2026 12:46

kshyatt force-pushed the ksh/enz_fact branch from e72c78c to 47a62e8 Compare September 24, 2026 09:11

vchuravy mentioned this pull request Sep 29, 2026

Only write printed type names into the IR when asked to EnzymeAD/Enzyme.jl#3711

Merged

lkdvos reviewed Sep 30, 2026

View reviewed changes

Comment thread src/factorizations/pullbacks.jl Outdated

kshyatt marked this pull request as draft September 30, 2026 18:26

lkdvos reviewed Oct 13, 2026

View reviewed changes

Comment thread src/factorizations/pushforwards.jl

lkdvos approved these changes Oct 13, 2026

View reviewed changes

kshyatt and others added 5 commits October 13, 2026 15:33

Add Enzyme rules for factorizations

cdd8236

Get rid of unneeded projector rules

2d1d9e4

Update src/factorizations/pullbacks.jl

8ce9b20

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

Get rid of some Nothing specializations

ea21098

Update src/factorizations/pushforwards.jl

135bcde

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

kshyatt force-pushed the ksh/enz_fact branch from ea9ffb5 to 135bcde Compare October 13, 2026 13:33

kshyatt enabled auto-merge (squash) October 13, 2026 13:33