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

stillwater-sc / universal / 43635043952
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 14 Nov 2026 08:18PM UTC
Jobs 1
Files 893
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

14 Nov 2026 07:59PM UTC coverage: 89.13% (+0.01%) from 89.119%
43635043952

push

github

web-flow
fix(posit): fused operators exact and rounded once, NaR propagates, classify on the encoding (#2074)

The QA sweep in #2047 found posit fam not correctly rounded, the fused operators not
compiling for small configurations, and the classification and NaR-handling functions
going through wider types or disagreeing with one another.

- fma, fam and fmma are computed on 64-bit limbs: every operand enters as its exact
  significand and scale, products and sums are exact, and an addend too small to cross
  a rounding boundary collapses to one sticky unit below all of them; then one
  posit_limbs::encode. fam formed a + b in a blocktriple of the operands' width and
  truncated it to fbits + 3 bits with no sticky: posit<6,1> fam(0.09375, 2.5, 0.4375)
  was 1, not 1.25, with 256 / 816 / 87008 / 126384 failures at <6,1> <7,0> <8,1> <8,2>.
  The blocktriples were sized nbits - 3 - es, which wraps, so posit<5,3> did not compile.
- NaR propagates (user decision, option b): min, max, nextafter, nexttoward and copysign
  with a NaR operand are NaR; isnan(NaR) is true and isinf(NaR) false. max(NaR, x) was
  x, copysign(x, NaR) flipped x, nextafter(x, NaR) stepped up or down by x's sign, and
  isinf(NaR) and isnan(NaR) were both true.
- fpclassify and isnormal decide on the encoding: through long double, posit<1024,5>
  maxpos was FP_INFINITE and minpos FP_ZERO.
- nexttoward compares x with its posit<256,5> target exactly; it rounded x into
  posit<256,5>, so posit<512,6>(1 + 2^-300) toward 1 did not move.
- fdp, fdp_stride and fdp_qc return NaR (or make the quire NaR) for a short vector or a
  zero stride; the asserts vanished under NDEBUG, leaving an out-of-bounds read or an
  endless loop.

Tests: arithmetic/fused_ops.cpp is rewritten around an exact integer<> reference with
the posit tie-point rounding (its std::fma and double references could not see the fam
defect, which started at 6 bits, a level CI did not run); level 1 now covers fma/fam to
7 bits, fmma to 5 ... (continued)

127 of 128 new or added lines in 6 files covered. (99.22%)

1 existing line in 1 file now uncovered.

63186 of 70892 relevant lines covered (89.13%)

11685023.47 hits per line

Uncovered Changes

Lines Coverage ∆ File
1
98.98
-1.02% include/sw/universal/number/posit/atomic_fused_operators.hpp

Coverage Regressions

Lines Coverage ∆ File
1
97.8
-0.06% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 43635043952.1 14 Nov 2026 08:18PM UTC 893
89.13
GitHub Action Run
Source Files on build 43635043952
  • Tree
  • List 893
  • Changed 10
  • Source Changed 5
  • Coverage Changed 10
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #43635043952
  • ed780d42 on github
  • Prev Build on main (#43625651943)
  • Next Build on main (#43646807960)
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