Abstract Multi-threaded unit tests for high-performance thread-safe data structures typically do not test all behaviour, because only a single scheduling of threads is witnessed per invocation of the unit tests. Model checking such unit tests allows to verify all interleavings of threads. These tests could be written in or compiled to LLVM IR. Existing LLVM IR model checkers like divine and Nidhugg, use an LLVM IR interpreter to determine the next state. This paper introduces llmc, a multi-core explicit-state model checker of multi-threaded LLVM IR that translates LLVM IR to LLVM IR that is executed instead of interpreted. A test suite of 24 tests, stressing data structures, shows that on average llmc clearly outperforms the state-of-the-art tools divine and Nidhugg.

Stateful model checking creates numerous states which need to be stored and checked if already visited. One option for such storage is a hash map and this has been used in many model checkers. In particular, we are interested in the performance of concurrent hash maps for use in multi-core model checkers with a variable state vector size. Previous research claimed that open addressing was the best performing method for the parallel speedup of concurrent hash maps. However, here we demonstrate that chaining lends itself perfectly for use in a concurrent setting. We implemented 12 hash map variants, all aiming at multicore efficiency. 8 of our implementations support variable-length key-value pairs. We compare our implementations and 22 other hash maps by means of an extensive test suite. Of these 34 hash maps, we show the representative performance of 11 hash maps. Our implementations not only support state vectors of variable length, but also feature superior scalability compared with competing hash maps. Our benchmarks show that on 96 cores, our best hash map is between 1.3 and 2.6 times faster than competing hash maps, for a load factor under 1. For higher load factors, it is an order of magnitude faster.