Wasmi 2.0 reports a 2.2x execution speedup and publishes the interpreter design behind it

Sources: primary, discussion
Summary: The release note reports 2.2x over Wasmi 1.0 in geometric mean across the wasmi-benchmarks suite on an Apple M2 Pro, run on three hardware setups including AMD EPYC 7763 and Intel Xeon Platinum 8370C, against Wasm3, WAMR fast-interpreter, Wasmtime Pulley and Makepad Stitch, from a benchmark project the author states is reproducible. The post details the changes: four instruction dispatch modes, three accumulator registers held in hardware registers, and a single contiguous instance allocation that removes the dependent-load chain behind global access. It also records the costs, that the accumulator design requires copy instructions the previous design did not, and a rejected experiment that regressed performance on calls.
Why it matters: Wasmi is the plugin and smart-contract runtime under Typst, Zellij, Soroban and Ripple, and the release publishes its method, its comparison set and its regressions rather than a single number.
Follow-up: Track whether the benchmark suite is run independently on non-Apple hardware.