Performance regression with tuple tail recursion in v1.7

[jfeist]

I noticed a quite significant performance regression in tuple tail recursion from julia 1.6 to 1.7. Version 1.6 already had a regression for a specific case compared to 1.5, but in 1.7 it's even worse, and shows up for all the cases I checked. Full details in this thread: https://discourse.julialang.org/t/strange-performance-characteristics-and-regression-with-tuple-recursion/64888/2

[KristofferC]

Running a git bisect might be a good way to find the place where it regressed.

[simeonschaub]

Ref #40814. I suspect c2ec70c is the culprit here as well.

[vtjnash]

Seems to be an LLVM optimization prediction heuristic that we cannot really guarantee. But seems to do well at M=2 again

M = 1:   2.729 ns (0 allocations: 0 bytes)
M = 2:   68.319 ns (0 allocations: 0 bytes)
M = 3:   7.410 μs (0 allocations: 0 bytes)
M = 4:   173.518 μs (0 allocations: 0 bytes)