[BUG]: `number/float16/base/to-word` mis-rounds subnormals at the 11-bit shift boundary

未关闭
#16,510 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@harsh-thakkar7 已经在做这个了。

开始于 2026年11月5日。

  • #16511 来自 @harsh-thakkar7 —— 未关闭

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
停滞
领域
backend

调研方向

The issue identifies lib/main.js and the subnormal rounding branch as the starting point. Check the conversion against native Float16Array around the 11-bit shift boundary, including positive and negative values and the exact halfway value. Done means the reported mismatches are fixed and focused regression coverage confirms the expected half-precision words; pull request #16511 is already open.

由索引模型根据 Issue 内容生成。

描述

Description

@stdlib/number/float16/base/to-word returns zero for float32 values just above 2^-25 that should round to the smallest positive half-precision subnormal. Native Float16Array rounds these to 0x0001.

var toWord = require( '@stdlib/number/float16/base/to-word' );

toWord( 2.98038125038147e-8 );  // returns 0x0000, expected 0x0001
toWord( -2.98038125038147e-8 ); // returns 0x8000, expected 0x8001

Comparing against new Float16Array over 20,000 values in [2^-25, 2^-24) shows 10,015 mismatches; the values that should round to 0x0001 are returned as zero. The exact halfway value 2^-25 correctly ties to zero in both implementations.

Root Cause

In lib/main.js, the subnormal rounding path branches on:

if ( shift < 11 ) {

When shift === 11, the code falls into the else branch, which extracts the round bit from the original 23-bit mantissa at the wrong position. With shift = 11 the 11-bit mantissa is shifted out entirely, and the round bit is its lowest bit (bit 10), which the first branch already handles correctly. Values requiring shift > 11 return zero earlier via the f16Exponent < -10 check, so the else branch is only reachable for shift === 11.

Changing the condition to shift <= 11 fixes all mismatches in a 50,000-value sweep.

Environment
  • Version: develop (7cf56423a)
  • Node.js with native Float16Array support
  • The bug is platform-independent and affects the JavaScript path only; the C path returns the bits of an already-encoded stdlib_float16_t.
Checklist
  • Read and understood the Code of Conduct.
  • Searched for existing issues and pull requests.
主要语言
JavaScript
星标
6k
派生
1.3k
平均合并
1 天 21 小时
30 天内合并 PR
224

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

stdlib-js/stdlib 的其他 Issue

查看 stdlib-js/stdlib 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。