# A Compiler-First Planetary Compute Engine: Automatic differentiable and performance portable Earth System Modeling (PROPL 2026)

*2026-08-26 — video*


**Position.** A new generation planetary compute engine for climate science needs to overcome two
fundamental limitations of current-generation climate models: (1) it should be end-to-end
auto-differentiable to be able to harness advanced gradient-based algorithms (adjoint and
backpropagation operators) to comprehensively learn from diverse, heterogenous data (observations
and/or high-fidelity simulations); (2) it should be portable across a rapidly diversifying landscape
of compute hardware, from traditional HPC to AI-oriented cloud accelerators. We argue that both
capabilities should be delivered at the compiler level, built around language-agnostic intermediate
representations such as LLVM and MLIR, rather than by rewriting science codes.

**Differentiability as infrastructure.** A "live computational commons" that ingests billions of
observations is, at its core, a massive inverse problem. Extracting actionable information from the
data is achieved via gradient-based parameter calibration, state and boundary-flux estimation,
sensitivity analysis, and uncertainty quantification, each of which becomes tractable if the adjoint
of the physical model is available. The [DJ4Earth](https://dj4earth.github.io/) initiative builds on
the [Enzyme](https://enzyme.mit.edu/) AD compiler operating at LLVM IR, together with MLIR-level
transpilation via Reactant.jl, to produce efficient gradients for the Oceananigans.jl ocean model, a
Julia implementation of the Ice Sheet and Sea level Model (ISSM), and the SpeedyWeather.jl
atmospheric model.

**Portability on the same stack.** Earth system simulation is limited by compute availability, yet
the exaFLOPs being deployed for AI are largely inaccessible to Fortran, C++ and Julia codes targeting
MPI-over-InfiniBand. Compiler-level cross-compilation — running the same Julia ocean and atmosphere
forward and adjoint models on AMD, NVIDIA and Google TPU hardware — offers a more promising path than
a manual rewrite, and lets the commons be retargeted on whatever silicon is cheapest, greenest or
most available.

Talk by William S. Moses (University of Illinois Urbana-Champaign) with Gong Cheng, Valentin Churavy,
Maximilian Gelbrecht, Milan Klöwer, Joseph Kump, Mathieu Morlighem, Sarah Williamson, Dhruv Apte,
Paul Berg, Mosè Giordano, Chris Hill, Nora Loose, Alexis Montoison, Sri Hari Krishna Narayanan, Avik
Pal, Michel Schanen, Simone Silvestri, Greg Wagner and Patrick Heimbach, at the
[PROPL 2026](https://pldi26.sigplan.org/home/propl-2026) workshop at
[PLDI 2026](https://pldi26.sigplan.org/) in Boulder, Colorado on 15th June 2026.
Mirror of <https://www.youtube.com/watch?v=tmXtnRWNRpE>.
Type: Video
URL: https://watch.eeg.cl.cam.ac.uk/videos/watch/87536e7a-9a4c-422f-9f20-446193621cce

## Related

- [The 3rd Programming for the Planet 2026 videos are available](https://anil.recoil.org/notes/pldi26-propl) (note, 2026-08-28)

---
Canonical: https://anil.recoil.org/videos/87536e7a-9a4c-422f-9f20-446193621cce
Type: video
Tags: propl, plancomp, climate, hpc, programming-languages, conference