Test the performance effects of LLVM D80707

[cuviper]

Proposed LLVM change: https://reviews.llvm.org/D80707

[rust-highfive]

@cuviper: no appropriate reviewer found, use r? to override

[rust-highfive]

⚠️ Warning ⚠️

• These commits modify submodules.

[cuviper]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 7bb8b6f with merge 5f884a00ab7292844b781231b742372b81bcd4ef...

[cuviper]

I couldn't reproduce the initial 30% (!) performance claim in that change, and neither could Serge when he tried again, so he has now updated the numbers with just a modest 1% reduction in instruction count. Still, if that little bit shows in rustc too, it may be worthwhile to apply.

[bors]

☀️ Try build successful - checks-azure
Build commit: 5f884a00ab7292844b781231b742372b81bcd4ef (5f884a00ab7292844b781231b742372b81bcd4ef)

[rust-timer]

Queued 5f884a00ab7292844b781231b742372b81bcd4ef with parent 4512721, future comparison URL.

[rust-timer]

Finished benchmarking try commit 5f884a00ab7292844b781231b742372b81bcd4ef, comparison URL.

[cuviper]

Serge has found that a lot of passes don't actually set that "changed" flag correctly. While he is chasing those down, the cumulative fixes will probably be too much to backport. So, we'll await the change in some future LLVM rebase. :)