Incorrect suggestion for `clippy::suboptimal_flops`

[bsdrks]

Summary

Clippy suggests that I change 1_f64 - (x - 1.0).powi(2) into (x - 1.0).mul_add(-x - 1.0, 1_f64), but these expressions do not yield identical results.

Lint Name

clippy::suboptimal_flops

Reproducer

I tried this code:

1_f64 - (x - 1.0).powi(2)

I saw this happen:

multiply and add expressions can be calculated more efficiently and accurately
for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#suboptimal_flopsclippy[Click for full compiler diagnostic](rust-analyzer-diagnostics-view:/diagnostic%20message?2#file[...]%2Fsrc%2Feasing%2Ffunction.rs)
function.rs(70, 5): consider using: `(x - 1.0).mul_add(-x - 1.0, 1_f64)`

The suggested expression is not equivalent to the original. Some examples:

1_f64 - (x - 1.0).powi(2), where x = 0.25, gives 0.4375
(x - 1_f64).mul_add(-x - 1_f64, 1_f64), where x = 0.25, gives 1.9375

1_f64 - (x - 1.0).powi(2), where x = 0.5, gives 0.75
(x - 1_f64).mul_add(-x - 1_f64, 1_f64), where x = 0.5, gives 1.75

1_f64 - (x - 1.0).powi(2), where x = 0.75, gives 0.9375
(x - 1_f64).mul_add(-x - 1_f64, 1_f64), where x = 0.75, gives 1.4375

1_f64 - (x - 1.0).powi(2), where x = 1, gives 1
(x - 1_f64).mul_add(-x - 1_f64, 1_f64), where x = 1, gives 1

1_f64 - (x - 1.0).powi(2), where x = 0, gives 0
(x - 1_f64).mul_add(-x - 1_f64, 1_f64), where x = 0, gives 2

Version

rustc 1.67.0-nightly (1eb62b123 2022-11-27)
binary: rustc
commit-hash: 1eb62b1235fd77200e6bd967d70e83c0f2497233
commit-date: 2022-11-27
host: x86_64-unknown-linux-gnu
release: 1.67.0-nightly
LLVM version: 15.0.4

Related

Probably related to #4735 and #4751, but this is fairly dangerous.

Additional Labels

No response

[ericwu17]

Thanks for the report! I found that the reason why clippy was giving this suggestion was because it was trying to negate the expression x - 1.0 and did so by simply adding a negative sign in front. Obviously, this did not achieve the desired effect. I'm about to open a PR for this issue.