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

stillwater-sc / universal / 40131128161
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran 24 Oct 2026 09:04PM UTC
Jobs 1
Files 762
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

24 Oct 2026 08:44PM UTC coverage: 88.161% (+0.05%) from 88.115%
40131128161

push

github

web-flow
perf(efloat): 64-bit least-significant-first limbs in an inline buffer with a per-thread block cache (#1822)

The significand was a std::vector<uint32_t>, most significant limb first, and
every operation packed its operands into 64-bit scratch, rounded back into
32-bit limbs and resized the vector. It is now the efloat_mpn kernels' own
format, 64-bit limbs least significant first, held by efloat_limbs<N>: N limbs
inside the value (min(ceil(nlimbs/2) + 2, EFLOAT_INLINE_LIMBS = 8), 512 bits)
and a heap block past them. Freed blocks go to a cache of 8 per thread, which
the next value or scratch request of their size takes, so c = a op b and
c op= b allocate nothing in a steady state at any precision; the scratch of
*, /, sqrt and the mathlib's kernels comes from the same cache.

- Rounding works on the 64-bit frame (round_into, round_to_precision); the
  rounding shift runs through efloat_mpn::lshift.
- Sums of one sign at the full precision take an mpfr_add1sp-like path;
  every other sum the frame, in two passes instead of five.
- limb_kernels gains full-length sums at every whole-limb gap; storage.cpp
  checks the 32-bit views, copies and moves, and the normalizing trim.
- Products and quotients read the operands' limbs in place; the dividend's
  shift is direct.
- bits(), limb_vector() and setlimb() keep their 32-bit, most significant
  first view, converted; the mathlib reads limb_data() and limb_count(), and
  set_significand() sets the 64-bit limbs.
- The storage works in constant evaluation (#1819): the inline limbs are
  zeroed there and a longer value allocates transiently.

mpfr_gap, i7-12700K, g++ 15.2 -O3 -march=native, against main: a+b
0.40-0.64x, a*b 0.65-0.91x, b/a 0.67-0.91x, a/7 0.74-0.94x, the mathlib
0.75-1.0x; every row within 1.4x MPFR, and every mpmath_gap cell below
gmpy2 2.3.2.

Results are bit-identical to main over 1.2 million old-vs-new results
(arithmetic, integer operands, roots, fma, hypot, truncations, comparisons,
conver... (continued)

578 of 593 new or added lines in 8 files covered. (97.47%)

4 existing lines in 3 files now uncovered.

54457 of 61770 relevant lines covered (88.16%)

9096219.14 hits per line

Uncovered Changes

Lines Coverage ∆ File
11
95.32
1.04% include/sw/universal/number/efloat/efloat_impl.hpp
3
93.63
2.38% include/sw/universal/number/efloat/math/truncate.hpp
1
99.22
include/sw/universal/number/efloat/efloat_limbs.hpp

Coverage Regressions

Lines Coverage ∆ File
2
93.63
2.38% include/sw/universal/number/efloat/math/truncate.hpp
1
97.04
0.0% include/sw/universal/number/cfloat/cfloat_impl.hpp
1
98.13
-0.27% include/sw/universal/number/efloat/math/exponent.hpp
Jobs
ID Job ID Ran Files Coverage
1 40131128161.1 24 Oct 2026 09:04PM UTC 762
88.16
GitHub Action Run
Source Files on build 40131128161
  • Tree
  • List 762
  • Changed 10
  • Source Changed 7
  • Coverage Changed 8
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #40131128161
  • 4a7b457c on github
  • Prev Build on main (#40061163622)
  • Next Build on main (#40136138324)
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