AmpereOne CPU Performance Study List

This is the 2024 version of the Ampere CPU study list, a set of traces used to study microarchitecture features in the AmpereOne custom core, which implements the aarch64 / ARM ISA v8.6. Most of these instruction trace files (.is) were collected for 20M instructions runlength of simulated measurement, in addition to online warmup and cooldown. The tracesets contain complementary memory stream (.ms) and branch stream (.bs) files that provide about 200M instructions worth of memory traffic and branch behavior prior to the start of instruction stream collection, which can be used for offline warmup of the simulator. The 2024-study_list.json file is the manifest and descriptive tracelist for these tests. max_resident_mb and total_cpu_secs specify how much memory and runtime each trace takes to simulate on Panthera; these stats are only meaningful in relative terms for prioritizing batch scheduling of simulation jobs running in a cluster. Suite Traces Workloads CAD 846 19 CloudInfra 727 30 Crypto 194 34 Database 143 15 FSPEC17 165 14 Graphs 224 19 ISPEC17 199 10 Media 288 18 Micros 203 27 Science 596 21 Solver 411 13 SpecJBB 141 14 WebServices 195 19 Total 4332 253 2024-study_list.mm is a mind map depiction of the categories of the workloads, and the two image files are renderings of that map. The auxiliary archive supplements the 2024-study-list trace dataset. When extracted beside the main archive, it populates the same ampere_traces/ hierarchy with companion metadata for the 4,332 study-list instruction traces: memory descriptions (.mem), working-set summaries (.is.wss), instruction footprints (.is.foot), and compressed instruction disassemblies (.is.dis.gz). It contains 4,332 memory files (for wrong path execution), 4,332 memory footprints, 4,332 disassemblies, and 2,516 working-set-size summaries (not all traces had .wss files generated). Use perf-streams to read the traces: https://github.com/AmpereComputing/perf-streams Use the Panthera model to run studies and perform rollups: https://github.com/AmpereComputing/panthera Use the Catscan analyzer to visualize pipelines and debug performance: https://github.com/AmpereComputing/catscan Use the QEMU tracing infrastructure to collect your own traces: https://github.com/AmpereComputing/tracing-qemu