• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

stillwater-sc / universal / 42344713552
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 07 Nov 2026 06:15AM UTC
Jobs 1
Files 868
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst

07 Nov 2026 05:54AM UTC coverage: 88.703% (+0.03%) from 88.67%
42344713552

push

github

web-flow
fix(areal): set the ubit wherever a conversion drops bits, and convert between configurations (#1982)

Part 1 of #1970, from QA sweep #1962. A conversion into areal truncates
and must set the ubit exactly when bits are dropped.

- float and double into a subnormal target of an areal with more
  fraction bits than the source (fbits >= 23 / >= 52): the wider-target
  branch dropped bits with the ubit clear when the subnormal scale took
  the net shift to the right. areal<30,5>(-5.4181703e-07f) came out
  truncated and marked exact. A net right shift of r drops r + 1 bits,
  and the ubit now records them.
- integers above 2^53 went through double when fbits < 53, which
  rounded them first: areal<32,8>(2^60 - 1) was an exact 2^60. Double
  still converts magnitudes below 2^53; larger ones take the native
  path, which keeps the dropped bits.
- the converting constructor areal<a,b>(areal<c,d>) had an empty body.
  It converts as the arithmetic converts its results: normalizeAddition
  into a blocktriple of the source's width, then convert(). No native
  intermediate is involved, so it is exact on every platform, including
  the top binade of areal<64,11>, which overflows double. A NaN keeps
  its kind, -0 stays -0, and an inexact source stays inexact.
- normalizeAddition, normalizeMultiplication and normalizeDivision copied
  the fraction of a subnormal in configurations with 61 or more fraction
  bits without lifting its leading one to the hidden-bit position, so
  those operands were read as 1.f. For an areal<72,4> subnormal x,
  x + x gave 1.7e-18 for 2.3e-10. All three apply the shift now.

Tests:
- New conversion/ubit_oracle.cpp needs no wider type. Clear the
  result's ubit to get t, the source or a truncation of its bits; the
  ubit must be set exactly when double(t) differs from the source. It
  covers:
  - float and double into the 13 configurations the sweep listed;
  - integers above 2^53;
  - the converting constructor over every encoding of f... (continued)

30 of 36 new or added lines in 1 file covered. (83.33%)

2 existing lines in 2 files now uncovered.

59360 of 66920 relevant lines covered (88.7%)

9625564.72 hits per line

Uncovered Changes

Lines Coverage ∆ File
6
95.36
2.01% include/sw/universal/number/areal/areal_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
1
95.36
2.01% include/sw/universal/number/areal/areal_impl.hpp
1
93.16
-0.85% include/sw/universal/verification/correctly_rounded_reference.hpp
Jobs
ID Job ID Ran Files Coverage
1 42344713552.1 07 Nov 2026 06:15AM UTC 868
88.7
GitHub Action Run
Source Files on build 42344713552
  • Tree
  • List 868
  • Changed 3
  • Source Changed 1
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42344713552
  • 43c8d96b on github
  • Prev Build on main (#42337781619)
  • Next Build on main (#42379094381)
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc