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

stillwater-sc / universal / 40981456672
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 30 Oct 2026 01:00AM UTC
Jobs 1
Files 771
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

30 Oct 2026 12:43AM UTC coverage: 88.285% (+0.03%) from 88.258%
40981456672

push

github

web-flow
feat(posit): correctly rounded exp, log and hypot families, computed in efloat at any nbits (#1870)

exp, exp2, exp10, expm1, log, log2, log10, log1p and hypot went through
double: at most double's 53 bits for posit<64,3> (59 near 1), double's
libm is not correctly rounded, and posit<128,4> and wider overflowed or
underflowed on the way through.

They are now efloat's correctly rounded functions, rounded once into the
posit by posit_limbs::evaluate, the engine for the rest of #1855.
evaluate() applies Ziv's test on the posit side: y = f(x) at W bits is
within 1/2 ulp of the exact value, so when y - 1 ulp and y + 1 ulp round
to the same posit, that posit is correctly rounded; otherwise W doubles,
up to 8 times the first W (fbits + 31, failing about once in 2^28). An
exact result on a posit tie point never passes, and is rounded as it is
after the last retry, ties to even. evaluate() rounds to nearest
whatever efloat's thread rounding mode, and an RAII guard puts efloat's
mode and status flags back on every exit, an exception included.

Saturation follows the posit rules: exp past maxpos is maxpos and below
minpos is minpos, also past efloat's own 2^(2^62) range; log(0),
log(negative) and log1p(<= -1) are NaR. hypot(x, y, z) (it was under
#ifdef LATER) sums exact squares at 2W and rounds once, so hypot(1, 2, 2)
is 3; hypotf and hypotl forward to hypot.

Cost (ns, posit<16,1> / <32,2> / <64,3> / <128,4> / <256,5>, main ->
now): exp 60/142/264/886/1414 -> 655/662/710/850/1203, log similar,
hypot 125/259/529/1296/2064 -> 191/195/247/322/466. The user chose
efloat as the only engine (no libm fast path), for results that do not
depend on a platform's libm.

Test static/tapered/posit/math/elementary.cpp: every posit up to 10 bits
for the eight unary functions, every pair at 5-6 bits for hypot, and
random posits at 16 to 256 bits, against the same efloat function at
4 (fbits + 1) + 64 bits rounded once by from_efloat (efloat is verified
against MPFR, from_eflo... (continued)

71 of 71 new or added lines in 4 files covered. (100.0%)

3 existing lines in 2 files now uncovered.

55715 of 63108 relevant lines covered (88.29%)

8837488.2 hits per line

Coverage Regressions

Lines Coverage ∆ File
2
96.18
-0.22% include/sw/universal/number/posit/posit_impl.hpp
1
97.04
0.0% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 40981456672.1 30 Oct 2026 01:00AM UTC 771
88.29
GitHub Action Run
Source Files on build 40981456672
  • Tree
  • List 771
  • Changed 7
  • Source Changed 3
  • Coverage Changed 7
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #40981456672
  • 48a2df3b on github
  • Prev Build on main (#40976210318)
  • Next Build on main (#40985620493)
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