Fix large dynamic tape issue

[wsmoses]

Fixes #500

[vchuravy]

That is illegal. NTuple is an immutable so despite it being allocated on the heap we shouldn't mutate it after it has been observed

[wsmoses]

We're not mutating it after being observed, we're initializing it?

[vchuravy]

We are guaranteed to never update values in place? That was my assumption for why it had to be a mutable

[wsmoses]

thinking this through:

This technically overrides the Enzyme internal allocator in its entirety.

We would only have a julia object internally if we needed to cache it for a tape. The tape should be immutable once initialized (technically we may memset it to 0 -- in fact in this very julia_allocator function), but otherwise once its value is set to actually preserve something, it is the same. We even attach invariant.group MD on it

[codecov-commenter]

Codecov Report

Base: 74.87% // Head: 74.61% // Decreases project coverage by -0.26% ⚠️

Coverage data is based on head (2f170b6) compared to base (5bfb57a).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #509      +/-   ##
==========================================
- Coverage   74.87%   74.61%   -0.27%     
==========================================
  Files          17       17              
  Lines        5370     5370              
==========================================
- Hits         4021     4007      -14     
- Misses       1349     1363      +14     

Impacted Files	Coverage Δ
src/compiler.jl	73.75% <100.00%> (-0.03%)	⬇️
src/compiler/orcv2.jl	61.90% <0.00%> (-10.48%)	⬇️
src/compiler/validation.jl	62.06% <0.00%> (-0.63%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.