Add `f16` and `f128` math functions #127027

tgross35 · 2024-06-27T09:15:48Z

This adds intrinsics and math functions for f16 and f128 floating point types. Support is quite limited and some things are broken so tests don't run on many platforms, but this provides a starting point.

try-job: aarch64-gnu

r? ghost
@rustbot label +F-f16_and_f128

tgross35 · 2024-06-27T09:25:52Z

@rustbot label +rla-silenced

tgross35 · 2024-07-12T08:21:16Z

Still getting a failure on f16 locally (aarch64) that I can't explain, 1.0f16.powi(1) is returning a wrong result. This happens in the unit tests but doesn't seem to replicate when I copy them to a separate file, making this difficult to debug.

May as well see if this shows up in CI.

@bors try

Add `f16` and `f128` math functions WIP This will be pretty finicky among selection failures, the lowering bug, and lack of symbols. But might as well start somewhere. Effectively blocked on const casting. try-job: aarch64-gnu

bors · 2024-07-12T08:22:29Z

⌛ Trying commit 0bca7c3 with merge 4ad38c5...

tgross35 · 2024-07-12T08:50:23Z

Figured out how to reproduce, the incorrect value is only returned if optimizations are on 💀

@bors cancel

tgross35 · 2024-07-12T17:36:30Z

f16 issue on aarch: llvm/llvm-project#98665

Add `f16` and `f128` math functions WIP This will be pretty finicky among selection failures, the lowering bug, and lack of symbols. But might as well start somewhere. Checks will fail until rust-lang#126636 makes it to beta, which should be next week (July 19). try-job: aarch64-gnu

tgross35 · 2024-07-12T17:49:17Z

There are probably some rough edges that I need to smooth out and some tests that need tweaking, but this PR is somewhat large so I think somebody can start taking a look at it. f16 pass on my aarch machine, I can't test f128 locally so I'm going via try jobs.

Note that this is still going to fail the check CI job until nightly clippy becomes beta next week.

r? @dtolnay

Since I know you have reviewed the math functions before. Feel free to reroll if you prefer.

Add `f16` and `f128` math functions This adds intrinsics and math functions for `f16` and `f128` floating point types. Support is quite limited and some things are broken so tests don't run on many platforms, but this provides a starting point. Checks will fail until rust-lang#126636 makes it to beta, which should be next week (July 19). try-job: aarch64-gnu

tgross35 · 2024-07-17T23:33:44Z

Thank you for the review! Yes, the Clippy ICE fix should be be in beta Friday/Saturday so that should (hopefully) fix CI.

edit: update, the bootstrap bump has unfortunately just taken a while #128083.

bors · 2024-07-29T05:17:48Z

☔ The latest upstream changes (presumably #125443) made this pull request unmergeable. Please resolve the merge conflicts.

Since LLVM <https://reviews.llvm.org/D99439> (4c7f820, "Update @llvm.powi to handle different int sizes for the exponent"), the size of the integer can be specified for the `powi` intrinsic. Make use of this so it is more obvious that integer size is consistent across all float types. This feature is available since LLVM 13 (October 2021). Based on bootstrap we currently support >= 17.0, so there should be no support problems.

These already exist in the compiler. Expose them in core so we can add their library functions.

This adds missing functions for math operations on the new float types. Platform support is pretty spotty at this point, since even platforms with generally good support can be missing math functions. `std/build.rs` is updated to reflect this.

`min`, `max`, and similar functions require external math routines. Add these under the same gates as `std` math functions (`reliable_f16_math` and `reliable_f128_math`).

Clarify what makes some operations not safe, and correct comment in the default branch ("not safe" -> "safe").

Due to a LLVM bug, `f128` math functions link successfully but LLVM chooses the wrong symbols (`long double` symbols rather than those for binary128). Since this is a notable problem that may surprise a number of users, add a note about it. Link: llvm/llvm-project#44744

tgross35 · 2024-07-30T22:25:55Z

The stage0 bump took longer than expected but finally merged, so I was able to rebase. Since the last review I removed all the cfg(not(bootstrap)) and added a note in the docs about the x86 math bug, so users are aware of it.

One more try job to make sure nothing got broken:

@bors try

bors · 2024-07-30T22:27:05Z

⌛ Trying commit d64bbb1 with merge 2318850...

tgross35 · 2024-07-30T23:29:56Z

x86 CI passed and it looks like the aarch64 try job has tested the libraries already, so this should be about ready. Changes were trivial since the last look so

@bors r=dtolnay

bors · 2024-07-30T23:29:58Z

📌 Commit d64bbb1 has been approved by dtolnay

It is now in the queue for this repository.

bors · 2024-07-31T00:03:40Z

☀️ Test successful - checks-actions
Approved by: dtolnay
Pushing 2318850 to master...

tgross35 · 2024-07-31T00:08:31Z

What the heck

https://rust-lang.zulipchat.com/#narrow/stream/242791-t-infra/topic/bors.20merged.20PR.20without.20running.20all.20tests

tgross35 · 2024-07-31T00:14:02Z

Mark reset master. I'll open a new PR.

rustbot added O-unix Operating system: Unix-like S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-libs Relevant to the library team, which will review and decide on the PR/issue. F-f16_and_f128 `#![feature(f16)]`, `#![feature(f128)]` labels Jun 27, 2024

This comment has been minimized.

Sign in to view

rustbot added the rla-silenced Silences rust-log-analyzer postings to the PR it's added on. label Jun 27, 2024

tgross35 force-pushed the f16-f128-math branch from 1f04794 to a6f7ac8 Compare July 12, 2024 07:26