Skip to content
This repository was archived by the owner on Apr 28, 2025. It is now read-only.

Add ldexpf16, ldexpf128, scalbnf16, and scalbnf128 #391

Merged
merged 2 commits into from
Feb 5, 2025
Merged

Add ldexpf16, ldexpf128, scalbnf16, and scalbnf128 #391

merged 2 commits into from
Feb 5, 2025

Conversation

tgross35
Copy link
Contributor

@tgross35 tgross35 commented Jan 3, 2025 •
edited
Loading

The scalbn functions are similar enough that they can easily be made generic. Do so and add f16 and f128 versions.

@tgross35
Copy link
Contributor Author

tgross35 commented Jan 3, 2025 •
edited
Loading

f16 and f128 versions are untested, blocked until I can figure out how to do that with MPFR.

Edit: done in #392. I cherry picked that commit here but that one should probably merge first.

@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch 2 times, most recently from 13238e7 to 84597eb Compare January 3, 2025 04:43
@tgross35 tgross35 marked this pull request as ready for review January 3, 2025 05:34
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from 84597eb to 3d70168 Compare January 3, 2025 05:41
@tgross35 tgross35 force-pushed the tgross35/f16-f128-copysign-fabs branch from e6907f5 to 31a58aa Compare January 3, 2025 05:49
@tgross35 tgross35 changed the title Add scalbnf16 and scalbnf128 Add ldexpf16, ldexpf128, scalbnf16, and scalbnf128 Jan 3, 2025
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch 4 times, most recently from 0b59cdc to c4eca7d Compare January 3, 2025 06:27
@tgross35 tgross35 marked this pull request as draft January 3, 2025 06:33
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from c3df336 to b0e4038 Compare January 3, 2025 08:48
@tgross35 tgross35 force-pushed the tgross35/f16-f128-copysign-fabs branch 2 times, most recently from b1b01c5 to 58775fc Compare January 4, 2025 00:25
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from b0e4038 to cd3ef7a Compare January 4, 2025 00:27
@tgross35 tgross35 force-pushed the tgross35/f16-f128-copysign-fabs branch 2 times, most recently from b617507 to 336a6b6 Compare January 6, 2025 09:10
Base automatically changed from tgross35/f16-f128-copysign-fabs to master January 6, 2025 09:25
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from cd3ef7a to b801ddd Compare January 6, 2025 11:54
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch 2 times, most recently from e9d7e6c to 4d58493 Compare January 23, 2025 11:54
@tgross35 tgross35 mentioned this pull request Jan 23, 2025
Merged
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from 4d58493 to c997735 Compare January 23, 2025 23:20
@tgross35
Copy link
Contributor Author

I think the f16 version is running into the same issue as #419 (comment) (among other issues) so I am just going to drop that for now, keeping only scalbnf128.

@tgross35
Copy link
Contributor Author

Oh, interesting; the same problem seems to be showing up for the f128 version:

thread 'mp_edge_case_scalbnf128' panicked at crates/libm-test/tests/multiprecision.rs:17:48:
called `Result::unwrap()` on an `Err` value:
    input:    (0x00000000000000000000000000000003, -2) (0x00000000000000000000000000000003, 0xfffffffe)
    expected: 0x00000000000000000000000000000000 0x00000000000000000000000000000000
    actual:   0x00000000000000000000000000000001 0x00000000000000000000000000000001

Caused by:
    ulp 1 > 0

Testing with the new (unreleased) version of Rug returns 0x00000000000000000000000000000001, so I will just wait for that release.

@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch 2 times, most recently from 7394ee7 to ca979f9 Compare January 25, 2025 01:25
@tgross35 tgross35 mentioned this pull request Jan 25, 2025
Merged
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from ca979f9 to 946f6ec Compare January 25, 2025 01:27
@tgross35
Copy link
Contributor Author

f16 needs way more rounds than the other operations, I am going to check if there is a better algorithm.

@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch 2 times, most recently from 9fe11d3 to 80b1394 Compare January 30, 2025 06:00
@tgross35
Fix hex float trait recursion problem
ac0714b
@tgross35
Add scalbnf16, scalbnf128, ldexpf16, and ldexpf128
9c93e01 
Use the generic `scalbn` to provide `f16` and `f128` versions, which
also work for `ldexp`.

This involves a new algorithm for `f16` because the default does not
converge fast enough with a limited number of rounds.
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from cef07af to 9c93e01 Compare February 5, 2025 13:41
@tgross35 tgross35 marked this pull request as ready for review February 5, 2025 13:52
@tgross35 tgross35 force-pushed the tgross35/f16-f128-scalbn branch from e307755 to 9c93e01 Compare February 5, 2025 14:11
@tgross35 tgross35 enabled auto-merge February 5, 2025 14:16
@tgross35 tgross35 merged commit 3109ebe into master Feb 5, 2025
70 checks passed
@tgross35 tgross35 deleted the tgross35/f16-f128-scalbn branch February 5, 2025 15:00
tgross35 added a commit that referenced this pull request Apr 18, 2025
@tgross35
Merge pull request #391 from rust-lang/tgross35/f16-f128-scalbn
370831d 
Add `ldexpf16`, `ldexpf128`, `scalbnf16`, and `scalbnf128`
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@tgross35