Comparing Interpreter performance with Wasm3

[jserv]

Although there have been improvements in the performance of RISC-V instruction interpretation, the reasons why wasm3 outperforms in certain performance metrics, such as CoreMark, remain unclear.

Tested on AMD Ryzen Threadripper 2990WX 32-Core Processor:

• wasm3 (commit d28e14f) + wasm-coremark: 1305.75 iter/s
• rv32emu (commit 97f98e9): 1205.83 iter/s

Comparing a WebAssembly runtime with a RISC-V instruction set simulator isn't straightforward, especially when the same program is compiled for these two distinct architectures. The key difference lies in their virtual machine designs: WebAssembly is stack-based, whereas RISC-V operates on a register-based system. Despite these differences, comparing interpreters is still valuable as it can uncover optimization opportunities for our simulator, particularly in interpreter design. Therefore, analyzing the performance of existing interpreters remains a meaningful exercise.

Action Items:

• Investigate and compare the interpreter behavior between rv32emu and wasm3.
• Identify areas for potential performance improvement.