● Field notes — 12 posts on CUDA, C++, benchmarking & AI tooling03 years shipping · CMAP · ONERA · CEA
−9 %fewer mispredictions01Branch predictionAccurate, but late: branch predictors at equal budget+18%bubble operand waits02BranchesBranches, measured: operand waits, prediction and the return stack12×stage threads vs ISS03SystemCA CPU in SystemC, without RTLAgent-firstlanguage vs Dijkstra04AgentsZeroLang Field Notes: An Agent-First Language Meets DijkstraTiny msgswhat drives overhead05MPITiny MPI Messages: The Cost of Convenienceup to 7×faster integer pow06C++ PerfBeating std::pow: Up to 7x Faster Exponentiation in C++Rerununtil stable07BenchmarksYour Benchmarks Are WrongLLM vsweeks of tuning08LLM × C++Can an LLM Optimize My Code Better Than I Can?Branchlessbeat my AVX209SIMDThe Fastest std::find: How Doing More Work Makes You Faster3 layoutsAOS · SOA · AoSoA10Memory layoutAOS vs SOA vs AoSoA: The Answer Is Not What You ThinkEvolvedcircle packing11EvolutionEvolving Algorithms With My Claude Code Plan: ShinkaEvolve + Meridian80×faster Python N-body12N-bodyDoes Shinka Rediscover My SoA Tricks? A Python N-body Warmup