We have noticed a recent regression in module compilation times on the MBP M1/M2. Sometimes it takes over 100 seconds when it used to take less than 15.

Several non-scientific numbers have been reported on various platforms. At first three individual Mac users started reporting extreme numbers, often exceeding +90 seconds, while Windows and Linux users report an average of around 50 seconds, with some reporting more normal numbers below 10 seconds. We have attempted to isolate the problem by manually clearing all caches and see how long it takes to recompile from scratch.

Upon further profiling, it appears that the performance bottleneck is related to the "cranelift" library (see image).

Any tips on how we can help find out the root cause here are appreciated. Are there perhaps benchmarks to run to see if regressions happened?

Some non-scientific data:

47s, 11th Gen Intel® Core™ i7-11800H, linux
15s, linux, i9-12900H (22s without cache)
6s, i9-13900k, windows (14s without cache)
8.2s amd 5950x, linux
45s 9th gen intel i7, windows
56s AMD Ryzen 9 6900HS, windows (36s without cache) 
90s Mac Pro M1/M2, macos

Perhaps introduced in v10.0.0 or v10.0.1, as we have not upgraded to 11 yet, and in 9.0.4 we fixed another module compilation issue and had not experienced any problems yet up till quite recently