didgpu - GPU-Accelerated, Checkpointed Difference-in-Differences Estimator

A clean-room reimplementation of the heterogeneity-robust dynamic difference-in-differences estimator of de Chaisemartin and D'Haultfoeuille (2024) <doi:10.1162/rest_a_01414>, designed for long-running econometric work: per-cell checkpointing to disk, resumable runs after crash or out-of-memory, parallel cluster bootstrap, configurable backends (pure R, optional 'Rcpp' and 'CUDA'), and bit-for-bit numerical equivalence with the reference 'DIDmultiplegtDYN' package across binary, multivalued, and continuous treatments.