Skip to content

Frequent cache misses in Li #13

@cgiovanetti

Description

@cgiovanetti

Thermodynamics module occasionally recompiles, e.g.

thermo_model_DNeff = BackgroundModel()

(t_vec_ref, a_vec_ref, rho_g_vec, rho_nu_vec, rho_NP_vec, P_NP_vec, Neff_vec
) = thermo_model_DNeff(jnp.asarray(0.))

followed by

input_dNeff = jnp.linspace(-5, 5, 10,dtype=float) 
for dNeff in tqdm(input_dNeff):
    t_vec, a_vec, rho_g_vec, rho_nu_vec, rho_extra_vec, P_extra_vec, Neff_vec = thermo_model_DNeff(jnp.asarray(dNeff))

leads to cache miss with explanation

WARNING:2025-11-27 14:14:19,214:jax._src.pjit:1372: TRACING CACHE MISS at /LINX/linx/thermo.py:141:34 (n_massless_BE.<locals>.<lambda>) costing 114.678 ms because:
  for Li defined at /Users/caragiovanetti/Research/code_not_GH/Grace_LINX/linx/special_funcs.py:91
  all previously seen cache keys are different. Closest previous key:
  * key with different value of static args:
      now 3 and before 4

Also seen in ABCMB testing. Possible culprits are python for loop and weak typing (see cache misses during compilation)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions