Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) #136324

GrigorenkoPV · 2025-01-31T00:09:24Z

Tracking issue: #136321

try-job: x86_64-gnu-aux

rustbot · 2025-01-31T00:09:32Z

rustbot has assigned @Amanieu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

tgross35 · 2025-01-31T03:51:20Z

r? tgross35

tgross35 · 2025-01-31T04:27:15Z

1. Should the functions for `f16` & `f128` be gated under the `float_erf` flag or under `f16` & `f128` respectively?

Could you gate behind f16 and f128 but add a commented out // #[unstable(feature = "float_erf", issue = "136321")]? We do this for most methods on these types (like minimum), looks like I forgot to do that for gamma though.

2. A doctest for `erfc` is obvious: check that it's within `fN::EPSILON` from `1 - erf`. But what about `erf`? I currently check that `erf(1.0)` is within `fN::EPSILON` from a hardcoded constant. But the docs say:
   > The precision of this function is non-deterministic. This means it varies by platform, Rust version, and can even differ within the same execution from one invocation to the next.
   > Is there a better way to test `erf`?

Afaik there aren't any "convenient" values, it's probably fine to pick some random numbers. For reference:

> erf.(big.([-1.0, -0.5, 0.0, 0.5, 1.0]))
5-element Vector{BigFloat}:
 -0.8427007929497148693412206350826092592960669979663029084599378978347172540960087
 -0.5204998778130465376827466538919645287364515757579637000588057256471935217168533
  0.0
  0.5204998778130465376827466538919645287364515757579637000588057256471935217168533
  0.8427007929497148693412206350826092592960669979663029084599378978347172540960087

3. `compiler-builtins` has no `erf*` functions exposed [[src](https://github.com/rust-lang/compiler-builtins/blob/a233db5491360c01af0698784940ec4508738f33/src/math.rs)]. Is that an issue? Libm [does provide implementations](https://docs.rs/libm/latest/libm/?search=erf) though.

If you don't mind waiting a week or so (updating builtins is temporarily stuck), I think it would be easiest to expose them from libm before this merges. Otherwise the doctests will need to be gated to

4. Not sure if I should `#[cfg(not(msvc/aix))]` out any things, but I guess CI will tell.

They mention these functions in their docs https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/erf-erff-erfl-erfc-erfcf-erfcl?view=msvc-170 but the docs also mention atanf which is one of the symbols we shim https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/acos-acosf-acosl?view=msvc-170. Maybe they are only available for more recent versions? @ChrisDenton may know better.

In any case we can run a try job for these two targets.

library/std/src/f128.rs

ChrisDenton · 2025-01-31T06:05:28Z

They mention these functions in their docs https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/erf-erff-erfl-erfc-erfcf-erfcl?view=msvc-170 but the docs also mention atanf which is one of the symbols we shim https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/acos-acosf-acosl?view=msvc-170. Maybe they are only available for more recent versions? @ChrisDenton may know better.

In 32-bit x86 MSVC, some UCRT functions are header-only but the DLL does indeed export erf, erfc, erfcf, erfcl, erff and erfl.

GrigorenkoPV · 2025-01-31T22:56:17Z

Could you gate behind f16 and f128 but add a commented out // #[unstable(feature = "float_erf", issue = "136321")]?

Done.

looks like I forgot to do that for gamma though.

Added those as well. I think it's fine to not do it in a separate PR, since it's just a comment, not to mention float_gamma and float_erf are somewhat related.

Afaik there aren't any "convenient" values, it's probably fine to pick some random numbers.

Is one number enough or should I add more?

Also, out of curiosity, what language was that example in?

If you don't mind waiting a week or so (updating builtins is temporarily stuck), I think it would be easiest to expose them from libm before this merges. Otherwise the doctests will need to be gated too

I will file a PR in the compiler-builtins repo.

UPD: Done. rust-lang/compiler-builtins#753

BTW, under what conditions will the doctests fail if the functions are not exposed in compiler-builtins? I'm not entirely familiar with how these things work.

Also add ```rust // #[unstable(feature = "float_gamma", issue = "99842")] ``` to `gamma`-function-related methods on `f16` & `f128`, as per rust-lang#136324 (comment)

tgross35 · 2025-02-01T00:23:42Z

Added those as well. I think it's fine to not do it in a separate PR, since it's just a comment, not to mention float_gamma and float_erf are somewhat related.

Thanks!

Afaik there aren't any "convenient" values, it's probably fine to pick some random numbers.

Is one number enough or should I add more?

🤷 up to you, it's more of a demonstration than a test. I guess if you are looking for ideas you could use something based on the emperical rule:

/// The error function relates what percent of a normal distribution lies
/// within `x` standard deviations (scaled by `1/sqrt(2)`).
fn within_standard_deviations(x: f32) -> f32 {
    (x as f32 * std::f32::consts::FRAC_1_SQRT_2).erf() * 100.0
}

// 68% of a normal distribution is within one standard deviation
assert!((within_standard_deviations(1.0) - 68.269).abs() < 0.01);
// 95% of a normal distribution is within two standard deviations
assert!((within_standard_deviations(2.0) - 95.450).abs() < 0.01);
// 99.7% of a normal distribution is within three standard deviations
assert!((within_standard_deviations(3.0) - 99.730).abs() < 0.01);

Also, out of curiosity, what language was that example in?

Julia, with using SpecialFunctions to get erf. The repl is kind of handy for scratch math, especially since bigfloats are easy to use.

If you don't mind waiting a week or so (updating builtins is temporarily stuck), I think it would be easiest to expose them from libm before this merges. Otherwise the doctests will need to be gated too

I will file a PR in the compiler-builtins repo.

UPD: Done. rust-lang/compiler-builtins#753

Left a comment there but that change LGTM, But as noted there,

@rustbot blocked

BTW, under what conditions will the doctests fail if the functions are not exposed in compiler-builtins? I'm not entirely familiar with how these things work.

Most everything uses the system's C libm if it is available, but compiler-builtins provides the math operations for platforms that don't have it. At least WASM and UEFI would fail without it, I'm not sure exactly what else. You may have seen it already but the config for what platforms get exposed is here https://github.com/rust-lang/compiler-builtins/blob/c038174b9099e85c3c30c4e9ba85c5f4ae8f7d50/src/lib.rs#L44-L60.

tgross35 · 2025-02-08T11:05:30Z

I just reverted the thing that was blocking the builtins update, it will be merged in #136714. Once that merges I'll run a couple try jobs.

This looks pretty reasonable, I think you wanted to update the examples still.

@rustbot author

tgross35 · 2025-02-08T11:07:38Z

library/std/src/f16.rs

+    /// # Unspecified precision
+    ///
+    /// The precision of this function is non-deterministic. This means it varies by platform, Rust version, and
+    /// can even differ within the same execution from one invocation to the next.
+    /// This function currently corresponds to the `erff` from libc on Unix
+    /// and Windows. Note that this might change in the future.


If you're updating the docs, mind copying this section from the f128 or f16 files? The f32/f64 ones just have really random wrapping.

Also add ```rust // #[unstable(feature = "float_gamma", issue = "99842")] ``` to `gamma`-function-related methods on `f16` & `f128`, as per rust-lang#136324 (comment)

GrigorenkoPV · 2025-02-12T13:46:02Z

@rustbot ready

RalfJung · 2025-02-14T15:16:05Z

src/tools/miri/src/shims/foreign_items.rs

@@ -742,6 +742,8 @@ trait EvalContextExtPriv<'tcx>: crate::MiriInterpCxExt<'tcx> {
            | "log1pf"
            | "expm1f"
            | "tgammaf"
+            | "erff"
+            | "erfcf"


Please also add calls to this function in src/tools/miri/tests/pass/float.rs.

RalfJung · 2025-02-14T16:11:40Z

Miri changes LGTM. :) But let's ensure they pass CI.

tgross35 · 2025-02-14T20:42:12Z

@bors try

bors · 2025-02-14T20:43:24Z

⌛ Trying commit 18b9ede with merge 0b51376...

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) Tracking issue: rust-lang#136321 try-job: x86_64-gnu-aux

bors · 2025-02-14T22:33:58Z

☀️ Try build successful - checks-actions
Build commit: 0b51376 (0b513760eb662d9fbdc0af2ea925d36f6cd5a087)

tgross35 · 2025-02-15T01:17:19Z

@bors r+

bors · 2025-02-15T01:17:22Z

📌 Commit 18b9ede has been approved by tgross35

It is now in the queue for this repository.

bors · 2025-02-15T04:52:53Z

⌛ Testing commit 18b9ede with merge f77247a...

bors · 2025-02-15T07:50:57Z

☀️ Test successful - checks-actions
Approved by: tgross35
Pushing f77247a to master...

rust-timer · 2025-02-15T09:08:42Z

Finished benchmarking commit (f77247a): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results (secondary -2.1%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.1%	[-2.1%, -2.1%]	1
All ❌✅ (primary)	-	-	0

Cycles

Results (secondary 9.2%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	9.2%	[8.8%, 10.0%]	6
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 790.529s -> 788.977s (-0.20%)
Artifact size: 347.33 MiB -> 347.33 MiB (-0.00%)

… r=tgross35 tests: Also gate `f16::erfc()` doctest with `reliable_f16_math` cfg In rust-lang#136324 the doctest for `f16::erf()` was gated with `reliable_f16_math`. Add the same gate on `f16::erfc()` to avoid: rust_out.71e2e529d20ea47d-cgu.0:\ (.text._ZN8rust_out4main43_doctest_main_library_std_src_f16_rs_1321_017h485f3ffe6bf2a981E+0x38): \ undefined reference to `__gnu_h2f_ieee' on MIPS (and maybe other architectures). r? tgross35

Rollup merge of rust-lang#137167 - martn3:reliable_f16_math-f16-erfc, r=tgross35 tests: Also gate `f16::erfc()` doctest with `reliable_f16_math` cfg In rust-lang#136324 the doctest for `f16::erf()` was gated with `reliable_f16_math`. Add the same gate on `f16::erfc()` to avoid: rust_out.71e2e529d20ea47d-cgu.0:\ (.text._ZN8rust_out4main43_doctest_main_library_std_src_f16_rs_1321_017h485f3ffe6bf2a981E+0x38): \ undefined reference to `__gnu_h2f_ieee' on MIPS (and maybe other architectures). r? tgross35

Also add ```rust // #[unstable(feature = "float_gamma", issue = "99842")] ``` to `gamma`-function-related methods on `f16` & `f128`, as per rust-lang#136324 (comment)

rustbot assigned Amanieu Jan 31, 2025

rustbot added 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. labels Jan 31, 2025

GrigorenkoPV changed the title ~~Implement f{16,32,64,128}::{erf,erfc} (float_erf)~~ Implement f{16,32,64,128}::{erf,erfc} (#![feature(float_erf)]) Jan 31, 2025

rustbot assigned tgross35 and unassigned Amanieu Jan 31, 2025

tgross35 reviewed Jan 31, 2025

View reviewed changes

library/std/src/f128.rs Outdated Show resolved Hide resolved

tgross35 mentioned this pull request Jan 30, 2025

Tracking Issue for erf and erfc for f16, f32, f64, f128 #136321

Open

4 tasks

GrigorenkoPV force-pushed the erf branch from c7b1278 to 84e15e7 Compare January 31, 2025 22:56

GrigorenkoPV mentioned this pull request Jan 31, 2025

Expose erf{,c}{,f} from libm rust-lang/compiler-builtins#753

Merged

rustbot added the S-blocked Status: Blocked on something else such as an RFC or other implementation work. label Jan 31, 2025

tgross35 reviewed Feb 8, 2025

View reviewed changes

GrigorenkoPV force-pushed the erf branch from 84e15e7 to 0e7a88a Compare February 12, 2025 13:45

GrigorenkoPV force-pushed the erf branch from 0e7a88a to 37006eb Compare February 12, 2025 13:45

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Feb 12, 2025

RalfJung reviewed Feb 14, 2025

View reviewed changes

miri: shims for erf & friends

18b9ede

GrigorenkoPV force-pushed the erf branch from 5b0c80a to 18b9ede Compare February 14, 2025 15:48

bors added a commit to rust-lang-ci/rust that referenced this pull request Feb 14, 2025

Auto merge of rust-lang#136324 - GrigorenkoPV:erf, r=<try>

0b51376

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) Tracking issue: rust-lang#136321 try-job: x86_64-gnu-aux

bors mentioned this pull request Feb 14, 2025

add x86-sse2 (32bit) ABI that requires SSE2 target feature #137037

Merged

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 15, 2025

bors added the merged-by-bors This PR was explicitly merged by bors. label Feb 15, 2025

bors merged commit f77247a into rust-lang:master Feb 15, 2025
7 checks passed

rustbot added this to the 1.86.0 milestone Feb 15, 2025

GrigorenkoPV deleted the erf branch February 15, 2025 14:00

martn3 mentioned this pull request Feb 17, 2025

tests: Also gate f16::erfc() doctest with reliable_f16_math cfg #137167

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) #136324

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) #136324

GrigorenkoPV commented Jan 31, 2025 •

edited by tgross35

Loading

rustbot commented Jan 31, 2025

tgross35 commented Jan 31, 2025

tgross35 commented Jan 31, 2025

ChrisDenton commented Jan 31, 2025 •

edited

Loading

GrigorenkoPV commented Jan 31, 2025 •

edited

Loading

tgross35 commented Feb 1, 2025

tgross35 commented Feb 8, 2025

tgross35 Feb 8, 2025

GrigorenkoPV commented Feb 12, 2025

RalfJung Feb 14, 2025

RalfJung commented Feb 14, 2025

tgross35 commented Feb 14, 2025

bors commented Feb 14, 2025

bors commented Feb 14, 2025

tgross35 commented Feb 15, 2025

bors commented Feb 15, 2025

bors commented Feb 15, 2025

bors commented Feb 15, 2025

rust-timer commented Feb 15, 2025

Implement f{16,32,64,128}::{erf,erfc} (#![feature(float_erf)]) #136324

Implement f{16,32,64,128}::{erf,erfc} (#![feature(float_erf)]) #136324

Conversation

GrigorenkoPV commented Jan 31, 2025 • edited by tgross35 Loading

rustbot commented Jan 31, 2025

tgross35 commented Jan 31, 2025

tgross35 commented Jan 31, 2025

ChrisDenton commented Jan 31, 2025 • edited Loading

GrigorenkoPV commented Jan 31, 2025 • edited Loading

tgross35 commented Feb 1, 2025

tgross35 commented Feb 8, 2025

tgross35 Feb 8, 2025

Choose a reason for hiding this comment

GrigorenkoPV commented Feb 12, 2025

RalfJung Feb 14, 2025

Choose a reason for hiding this comment

RalfJung commented Feb 14, 2025

tgross35 commented Feb 14, 2025

bors commented Feb 14, 2025

bors commented Feb 14, 2025

tgross35 commented Feb 15, 2025

bors commented Feb 15, 2025

bors commented Feb 15, 2025

bors commented Feb 15, 2025

rust-timer commented Feb 15, 2025

Overall result: no relevant changes - no action needed

Instruction count

Max RSS (memory usage)

Cycles

Binary size

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) #136324

Implement `f{16,32,64,128}::{erf,erfc}` (`#![feature(float_erf)]`) #136324

GrigorenkoPV commented Jan 31, 2025 •

edited by tgross35

Loading

ChrisDenton commented Jan 31, 2025 •

edited

Loading

GrigorenkoPV commented Jan 31, 2025 •

edited

Loading