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

stillwater-sc / universal / 42174431513
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 06 Nov 2026 06:42AM UTC
Jobs 1
Files 861
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

06 Nov 2026 06:26AM UTC coverage: 88.592% (-0.006%) from 88.598%
42174431513

push

github

web-flow
feat(numeric): compensated exp, exp2, exp10, expm1, log, log2, log10 and log1p for faithful (#1959)

Part 2 of #1952. They are evaluated in faithful arithmetic, not through
the FloatingPointType, which would keep half the pair's precision.

- exp(x) = 2^m exp(r)^512 with x = m ln2 + 512 r. x - m ln2 is computed
  exactly in the exact expansion from #1954, with ln 2 split into four
  Real parts. A Taylor kernel gives exp(r) - 1, then nine squarings
  2s + s^2 keep the error relative to s.
- exp2 reduces by the nearest integer. exp10 forms x ln10 - m ln2
  exactly in the expansion; a pair product x * ln10 was 78 units off at
  x = 30. expm1 squares a scaled kernel for |x| < 1/2.
- log(x) = k ln2 + 2 atanh((b - 1) / (b + 1)), with b in
  [sqrt(1/2), sqrt(2)), so k ln2 + log(b) never cancels and log(1 - tiny)
  keeps its precision. log2 is exact for a power of two, log10 divides by
  ln 10, and log1p uses the atanh form for |x| < 1/4.
- Errors follow <cmath>: ERANGE for exp overflow and underflow and for
  log(0), EDOM for a negative log argument. With
  FAITHFUL_THROW_ARITHMETIC_EXCEPTION set, the log domain errors throw
  the new faithful_negative_log_arg, as sqrt throws
  faithful_negative_sqrt_arg; range results do not throw. expm1 of a
  large negative argument does not leave its inner exp's ERANGE.

math/constants.hpp holds ln 2 and ln 10 as four doubles, peeled exactly
into Real parts, so a float host gets them to about 100 bits.

numeric/faithful/exponential.cpp measures each function over 20000
arguments per range, against dd for faithful<float> and qd for
faithful<double>. The worst error is 2.9 and 3.7 units of epsilon(),
against a bound of 5. A long double reference failed on MSVC and macOS
ARM64, where long double is 53 bits and cannot hold a pair such as
(1, 2^-40). Evaluating through Real is 10^8 to 10^27 units off, and the
test fails it. numeric/faithful/math_exceptions.cpp builds with the
throw switch set.

Resolves #1955
Relates to #1952... (continued)

338 of 387 new or added lines in 6 files covered. (87.34%)

1 existing line in 1 file now uncovered.

58443 of 65969 relevant lines covered (88.59%)

9020689.26 hits per line

Uncovered Changes

Lines Coverage ∆ File
23
61.02
numeric/faithful/math_exceptions.cpp
22
83.7
numeric/faithful/exponential.cpp
4
96.3
include/sw/universal/number/faithful/math/functions/exponent.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 42174431513.1 06 Nov 2026 06:42AM UTC 861
88.59
GitHub Action Run
Source Files on build 42174431513
  • Tree
  • List 861
  • Changed 3
  • Source Changed 0
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42174431513
  • a1274613 on github
  • Prev Build on main (#42166969421)
  • Next Build on main (#42180824817)
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