trsm is not actually implemented yet

[vchuravy]

Partial fix to #2813

The crux is that Julia used a different function and while trsm is defined in Enzyme proper there aren't currently any rules attached https://github.com/EnzymeAD/Enzyme/blob/a13f632e2fbaed1f971de015f15e8f4b353e66cb/enzyme/Enzyme/BlasDerivatives.td#L705-L713

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.79%. Comparing base (30e0519) to head (013ba70).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2814      +/-   ##
==========================================
- Coverage   67.79%   67.79%   -0.01%     
==========================================
  Files          58       58              
  Lines       20723    20723              
==========================================
- Hits        14050    14049       -1     
- Misses       6673     6674       +1     

☔ View full report in Codecov by Sentry.
📢 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.

[github-actions]

Benchmark Results

	main	013ba70...	main / 013ba70...
basics/make_zero/namedtuple	0.0528 ± 0.0031 μs	0.0526 ± 0.0026 μs	1 ± 0.076
basics/make_zero/struct	0.255 ± 0.009 μs	0.253 ± 0.0057 μs	1.01 ± 0.042
basics/overhead	4.03 ± 0.001 ns	4.95 ± 0.01 ns	0.814 ± 0.0017
basics/remake_zero!/namedtuple	0.235 ± 0.008 μs	0.242 ± 0.0083 μs	0.971 ± 0.047
basics/remake_zero!/struct	0.234 ± 0.01 μs	0.233 ± 0.0084 μs	1 ± 0.056
fold_broadcast/multidim_sum_bcast/1D	10.3 ± 0.64 μs	10.3 ± 0.24 μs	1 ± 0.067
fold_broadcast/multidim_sum_bcast/2D	12.1 ± 0.29 μs	12.1 ± 0.27 μs	0.997 ± 0.033
time_to_load	1.26 ± 0.012 s	1.28 ± 0.019 s	0.987 ± 0.017

Benchmark Plots

A plot of the benchmark results has been uploaded as an artifact at https://github.com/EnzymeAD/Enzyme.jl/actions/runs/19671170824/artifacts/4673611025.

[wsmoses]

incidentally that was actually reverted: JuliaLang/LinearAlgebra.jl#1239

but apparently not backported yet

[wsmoses]

sorry quick post commit review, can this be limited to 1.12+.

It will otherwise slow down other code by making it non-blas, if not differentiated

[vchuravy]

Oof, that's frustrating, shouldn't we ideally not do the blas replacement when a function is not being differentiated? Otherwise this problem occurs for many more functions?

[wsmoses]

it does indeed occur for more functions (which is why we should be somewhat sparing on the blas replacements).

The issue is that at the time of replacement, we haven't run activity analysis so we don't know what is needed and what isn't