After switching to the toolchain v5 as default in 341832a we've observed few performance regressions including a 10% regression in ElemMatchLargeMixedInAndOrWithDuplicates causing BF-37502.

It was determined that on AWS Graviton2 under the mentioned workload the branch misprediction rate increased ~2.5 times while the number of branches executed increased ~5%.

The improvement to ByteCode::getField() aims to reduce the number of branches executed and switch to static branch prediction by avoiding taking branches for presumably the most common execution scenario - when the input is a BSON object.

This change seems to reduce the run-time of the workload by ~2%.