Chris Fallin

@shamatar I'm going through old open PRs and found this; at this point we now have a new mid-end optimization framework that allows rewrites to be written in a more...

@rylev I'm cleaning up old PRs and came upon this; are you still interested in this change? (Lots of conflicts now but it seems it wouldn't be too hard to...

@abrown I'm cleaning up old PRs and came upon this; do you think there are still improvements here that are useful to salvage? Constant-handling has changed quite a bit in...

Hi @rahulchaphalkar -- it looks like the difference is down to two fundamental factors: - We have explicit stack checks rather than implicit stack probes and reliance on guard pages....

Adding myself as a reviewer; happy to take a look at this at the beginning of next week (thanks!). For context for folks, this comes out of the ISLE-based Cranelift...

(I'll run some benchmarks to see what the compile-time impact is like, and how this benefits non-SpiderMonkey workloads; TBD.)

Benchmarks: - Sightglass measures the following for `spidermonkey.wasm`: ``` compilation :: cycles :: benchmarks/spidermonkey/benchmark.wasm Δ = 1456249767.40 ± 1315409020.02 (confidence = 99%) original.so is 1.01x to 1.17x faster than cache-indirect-calls.so!...

Updated, let me know what you think!

(Oh, and forgot to mention: thanks for the thorough benchmarks on this! I've gotta dig into why Sightglass is so finicky for me...)

@tertsdiepraam thanks for this! The tests are in `cranelift/filetests/filetests/`; you can find midend-optimizer-specific tests that assert on expected simplified IR in `egraph/`, and tests that check execution results in `runtests/`....