This also refactors cpuid reading so we can pass it specific eax input values, and implements detection for these features accordingly. I actually wonder if it wouldn't make more sense for Cranelift to do this itself? In the meanwhile, that's an easy thing to implement in Spidermonkey. It's a bit sad we have a few implementations of ReadCPUInfo in tree, but the other ones are part of embedded libraries if I understand correctly.

Comment on attachment 9023607 [details] [diff] [review] fix.patch Review of attachment 9023607 [details] [diff] [review]: ----------------------------------------------------------------- Looks good!

Cranelift has cranelift-native which can do CPU feature detection, but using SpiderMonkey's has the advantage of using SpiderMonkey's existing command-line flags for enabling/disabling use of various features, like --no-sse3.

Pushed by bbouvier@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/9e6f82773bdc Implement CPU detection of LZCNT, BMI1, BMI2; r=sunfish