Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

explain why we can mutate the FPU control word #114813

Merged
merged 1 commit into from
Sep 5, 2023
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions library/core/src/num/dec2flt/fpu.rs
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,17 @@ pub use fpu_precision::set_precision;
// round to 80 bits causing double rounding to happen when values are eventually represented as
// 32/64 bit float values. To overcome this, the FPU control word can be set so that the
// computations are performed in the desired precision.
//
// Note that normally, it is Undefined Behavior to alter the FPU control word while Rust code runs.
// The compiler assumes that the control word is always in its default state. However, in this
// particular case the semantics with the altered control word are actually *more faithful*
// to Rust semantics than the default -- arguably it is all the code that runs *outside* of the scope
// of a `set_precision` guard that is wrong.
// In other words, we are only using this to work around <https://github.com/rust-lang/rust/issues/114479>.
// Sometimes killing UB with UB actually works...
// (If this is used to set 32bit precision, there is still a risk that the compiler moves some 64bit
// operation into the scope of the `set_precision` guard. So it's not like this is totally sound.
// But it's not really any less sound than the default state of 80bit precision...)
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wonder if we can avoid this by relying on double-rounding being okay between f64 and f32. IOW, if we have set_precision always set the precision to 64bit, then it should produce correct results even for 32bit operations -- and if the compiler chooses to reorder some 64bit operations then there's no problem.

#[cfg(all(target_arch = "x86", not(target_feature = "sse2")))]
mod fpu_precision {
use core::arch::asm;
Expand Down
1 change: 1 addition & 0 deletions library/core/src/num/dec2flt/number.rs
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,7 @@ impl Number {
/// There is an exception: disguised fast-path cases, where we can shift
/// powers-of-10 from the exponent to the significant digits.
pub fn try_fast_path<F: RawFloat>(&self) -> Option<F> {
// Here we need to work around <https://github.com/rust-lang/rust/issues/114479>.
// The fast path crucially depends on arithmetic being rounded to the correct number of bits
// without any intermediate rounding. On x86 (without SSE or SSE2) this requires the precision
// of the x87 FPU stack to be changed so that it directly rounds to 64/32 bit.
Expand Down
Loading