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

stillwater-sc / universal / 39877997506
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/issue-2113-followups
DEFAULT BRANCH: master
Ran 23 Oct 2026 12:43AM UTC
Jobs 1
Files 761
Run time 2min
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

23 Oct 2026 12:25AM UTC coverage: 87.984% (-0.002%) from 87.986%
39877997506

push

github

web-flow
perf(efloat): log's atanh series on the fixed-point engine, AGM from 45000 bits (#1796)

* perf(efloat): log's atanh series on the fixed-point engine, AGM from 45000 bits

log_working summed 2 atanh(v), |v| < 2^-10, term by term in efloat, and
handed over to the AGM above 1440 working bits, where the two tied at
2048 bits. Now:

- atanh_working_small(v) = v S(z), S(z) = 1 + z/3 + z^2/5 + ..., z = v^2,
  summed by fixed_series::arctangent_series, which gains a flag for the
  non-alternating series (#1792's engine, generalized rather than
  copied). S is in [1, 1.1), so v S keeps its relative precision as
  x -> 1.
- The AGM switch, log_agm_bits_per_series_bit, moves from 144 to 4500,
  measured: from the table the two cross near 45000 working bits, and
  near 5000 k for x = 1 + 2^-k (k = 20-40, at 65536 and 131072 bits).
- The log(j/512) table entries come from the same routine, 7-10x
  cheaper than their efloat series. With the table now in use at high
  precision, that keeps cold-cache calls ahead of main (mpfr_check's
  log means: 40 us at 2048 bits, main 83; 1.4 ms at 8192, main 2.3).

mpfr_gap, efloat / MPFR, 128 / 512 / 2048 / 8192 bits:
  log    1.1 1.2 2.4 2.2x  ->  0.7 0.5 0.7 1.1x
  log1p  1.1 1.2 2.4 2.2x  ->  0.8 0.5 0.7 1.1x
  asinh  1.3 1.3 2.5 2.4x  ->  0.9 0.6 0.8 1.2x
  atanh  1.2 1.3 2.3 2.2x  ->  0.8 0.5 0.8 1.2x
  pow    0.9 0.9 1.6 1.6x  ->  0.8 0.6 0.7 1.1x

log_working stays 14.6-40 bits inside log_error_bits, as main's series
did (14.1-40). New test elastic/efloat/math/log_kernel.cpp checks the
bound against four exact roots and the plain series at 2P + 256 bits.

Resolves #1795

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* docs(efloat): history row for #1796

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(efloat): log_kernel reference keeps relative precision next to 1

The reference reduced every x by four square roots, whose r - 1 cancels
for x next to 1: above 8192 bits, at P + 256 bits, it was... (continued)

34 of 34 new or added lines in 2 files covered. (100.0%)

4 existing lines in 1 file now uncovered.

53554 of 60868 relevant lines covered (87.98%)

9446634.7 hits per line

Coverage Regressions

Lines Coverage ∆ File
4
84.27
-1.84% include/sw/universal/number/efloat/math/logarithm.hpp
Jobs
ID Job ID Ran Files Coverage
1 39877997506.1 23 Oct 2026 12:43AM UTC 761
87.98
GitHub Action Run
Source Files on build 39877997506
  • Tree
  • List 761
  • Changed 4
  • Source Changed 2
  • Coverage Changed 4
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #39877997506
  • c4564be3 on github
  • Prev Build on main (#39867576832)
  • Next Build on main (#39896143399)
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