By and by, this bug will accumulate some benchmark programs for microbenchmarking call_indirect, and the results of running them on various hardware (maybe even in various engines).

Emerging microbenchmark suite, culled from previous work: https://github.com/lars-t-hansen/call-ubench. (Prior benchmarking work starts at bug 1639153 comment 138.)

Mostly we care about running time, for now; we've decided to just eat the space cost of the indirect stubs, this will be reevaluated later.

For Firefox, interesting mozilla-central revisions are (this will be updated as we add further optimizations):

Before indirect-stub: d4bd94bc7b58345d02f59b22f35ba6269d8fd2b0
After indirect-stub: 54664793b10f51a867ab1de9dde8c21bbf81d43c
After indirect-stub code placement optimization: 85d024b6a4c8174fe5bf6f219496bb9af61122ac
After null-check optimization: e6529d3eb27e5ba3cecd665c2dee8ec28072feee
After table reorg and code pointer load optimization: TBD
After bounds check load / constant inlining optimization: TBD