[clang] Bfi precision (PR #66285)

[Nikita Popov via cfe-commits]

nikic wrote:

It looks like this has made clang 0.7% slower (in the sense that the clang stage2 binary is optimized worse -- there is no negative impact on stage1). The binary is also 0.7% larger. It looks like this change has significant impact on inlining heuristics.

Not really sure what to make of that, I assume you determined that this ends up being faster on average, so maybe clang is just unlucky.

https://github.com/llvm/llvm-project/pull/66285