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

stillwater-sc / universal / 40901049687
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran 29 Oct 2026 03:43PM UTC
Jobs 1
Files 768
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

29 Oct 2026 03:23PM UTC coverage: 88.239% (-0.01%) from 88.251%
40901049687

push

github

web-flow
fix(verification): a scaled reference past double's range (#1864)

A cfloat whose range reaches past double's has operands and results that
double holds only as inf, or as zero or a rounded subnormal. With
max-exponent values, cfloat<64,11> has a binade above DBL_MAX; es > 11 also
reaches below DBL_MIN. The random verifier's double reference was inf or
wrong there, so the cfloat subsuper suites kept c64 randoms off, and any
operand double saw as inf went untested.

scaledReference computes those cases at a scale double can hold. Each
operand is m * 2^e, with m in [0.5,1) taken exactly from frexp; the
operation runs on the fractions and the exponent goes back on afterwards.
  - Normal range, where the type's precision does not depend on scale:
    round once at that scale (#1862's correctlyRoundedReference), then
    rescale exactly.
  - Below it, every value is a multiple of the subnormal quantum, so round
    to the nearest multiple directly. A sum there is exact; a product or
    quotient breaks its ties with the sign of its exact error.
  - A type without subnormals flushes a result below its smallest normal
    to zero, as cfloat does in conversion and arithmetic alike.
  - The type's ldexp only encodes a value that is already exact. A value it
    cannot hold finitely and exactly is an overflow: inf, or maxpos for a
    saturating type.

The verifier takes this path, for a type with frexp, ldexp and at most 53
bits, when:
  - an operand or the double result is not finite;
  - the result is below DBL_MIN;
  - an operand does not convert to double exactly.
An operand that only double sees as inf or zero is now tested rather than
skipped. executeBinary now returns whether it produced a reference. The one
case without one is an operand double cannot hold together with a result
the scaled reference declines; the verifier skips it instead of comparing
against a meaningless value. That is about 1 pair in 4000 of
cfloat<60,12> fff, all of them an operand... (continued)

100 of 122 new or added lines in 2 files covered. (81.97%)

1 existing line in 1 file now uncovered.

55566 of 62972 relevant lines covered (88.24%)

8771298.9 hits per line

Uncovered Changes

Lines Coverage ∆ File
17
34.7
1.02% include/sw/universal/verification/test_suite_randoms.hpp
5
94.02
-1.22% include/sw/universal/verification/correctly_rounded_reference.hpp

Coverage Regressions

Lines Coverage ∆ File
1
97.04
0.0% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 40901049687.1 29 Oct 2026 03:43PM UTC 768
88.24
GitHub Action Run
Source Files on build 40901049687
  • Tree
  • List 768
  • Changed 3
  • Source Changed 2
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #40901049687
  • 8702283c on github
  • Prev Build on main (#40888380477)
  • Next Build on main (#40911161556)
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