Specialise univariate transforms

[penelopeysm]

Sort of along the lines of #1142.

I suspect that this small patch might provide most of the perf benefits.

[penelopeysm]

i mean isn't this just annoying, this is what i mean when i say bijectors is a pain to deal with

[penelopeysm]

and also distributions to a lesser extent

size(dist) and Bijectors.output_size are not well-defined for things like LKJCholesky or ProductNamedTupleDistribution

that's fine because their use cases are (possibly) different, but dynamicppl has its own needs, and the correct solution is to have a clean api that generalises across all of these cases

[github-actions]

Benchmark Report

• this PR's head: a302b9d9ece3f21468ebda39c769f79789efe2bc
• base branch: 052bc1950df3a42e14b56eae51af236881092f90

Computer Information

Julia Version 1.11.7
Commit f2b3dbda30a (2025-09-08 12:10 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

┌───────────────────────┬───────┬─────────────┬───────────────────┬────────┬─────────────────────────────────┬────────────────────────────┐
│                       │       │             │                   │        │        t(eval) / t(ref)         │     t(grad) / t(eval)      │
│                       │       │             │                   │        │ ──────────┬───────────┬──────── │ ───────┬─────────┬──────── │
│                 Model │   Dim │  AD Backend │           VarInfo │ Linked │      base │   this PR │ speedup │   base │ this PR │ speedup │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼────────┼─────────┼─────────┤
│               Dynamic │    10 │    mooncake │             typed │   true │    398.57 │    435.63 │    0.91 │  11.64 │   10.36 │    1.12 │
│                   LDA │    12 │ reversediff │             typed │   true │   2761.55 │   2927.77 │    0.94 │   2.07 │    2.15 │    0.96 │
│   Loop univariate 10k │ 10000 │    mooncake │             typed │   true │ 151539.66 │ 127439.34 │    1.19 │   5.61 │    5.22 │    1.07 │
│    Loop univariate 1k │  1000 │    mooncake │             typed │   true │  15455.07 │  12951.41 │    1.19 │   5.12 │    5.75 │    0.89 │
│      Multivariate 10k │ 10000 │    mooncake │             typed │   true │  31126.99 │  32278.50 │    0.96 │   9.70 │    9.79 │    0.99 │
│       Multivariate 1k │  1000 │    mooncake │             typed │   true │   4217.05 │   3800.45 │    1.11 │   7.47 │    8.60 │    0.87 │
│ Simple assume observe │     1 │ forwarddiff │             typed │  false │     16.78 │     21.48 │    0.78 │   1.87 │    1.50 │    1.24 │
│           Smorgasbord │   201 │      enzyme │             typed │   true │   2544.57 │   3175.27 │    0.80 │   4.72 │    4.03 │    1.17 │
│           Smorgasbord │   201 │ forwarddiff │       simple_dict │   true │  23330.22 │  24952.11 │    0.93 │  24.52 │   24.78 │    0.99 │
│           Smorgasbord │   201 │ forwarddiff │ simple_namedtuple │   true │   1029.85 │   1073.88 │    0.96 │ 173.45 │   78.50 │    2.21 │
│           Smorgasbord │   201 │ forwarddiff │             typed │  false │   2520.19 │   2675.93 │    0.94 │  47.85 │   85.01 │    0.56 │
│           Smorgasbord │   201 │ forwarddiff │      typed_vector │   true │   2646.54 │   3131.72 │    0.85 │  44.60 │   40.51 │    1.10 │
│           Smorgasbord │   201 │ forwarddiff │           untyped │   true │   2229.92 │   2822.88 │    0.79 │  48.70 │   42.68 │    1.14 │
│           Smorgasbord │   201 │ forwarddiff │    untyped_vector │   true │   2284.56 │   2851.46 │    0.80 │  45.70 │   42.56 │    1.07 │
│           Smorgasbord │   201 │    mooncake │             typed │   true │   2549.02 │   3160.30 │    0.81 │   5.73 │    5.59 │    1.02 │
│           Smorgasbord │   201 │ reversediff │             typed │   true │   2656.72 │   3110.77 │    0.85 │  54.96 │   50.19 │    1.09 │
│              Submodel │     1 │    mooncake │             typed │   true │     24.96 │     26.31 │    0.95 │   5.42 │    5.48 │    0.99 │
└───────────────────────┴───────┴─────────────┴───────────────────┴────────┴───────────┴───────────┴─────────┴────────┴─────────┴─────────┘

[github-actions]

DynamicPPL.jl documentation for PR #1145 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1145/

[codecov]

Codecov Report

❌ Patch coverage is 85.71429% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.47%. Comparing base (052bc19) to head (a302b9d).

Files with missing lines	Patch %	Lines
src/utils.jl	85.71%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1145       +/-   ##
===========================================
+ Coverage   37.18%   81.47%   +44.29%     
===========================================
  Files          40       40               
  Lines        3819     3839       +20     
===========================================
+ Hits         1420     3128     +1708     
+ Misses       2399      711     -1688     

☔ 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.

[penelopeysm]

What a shocker, it doesn't. Maybe it's only noticeable for FastLDF, which is what I ran the benchmarks in #1142 on. We can shelve this for now and come back to it later.