stillwater-sc / universal / 23769032919
85%
master: 84%

LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran
Jobs 1
Files 630
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
coverage: 84.358% (+0.01%) from 84.347%
23769032919

push

github

web-flow 
feat(ucalc): hfloat hex normalization steps (#681)

* feat(ucalc): hfloat hex normalization steps (#658)

Add IBM System/360 hex float step-by-step decomposition showing:
1. Decompose: sign, hex exponent (16^e), hex fraction (0x0.h1h2h3...)
2. Align hex exponents: shift by whole hex digits (4 bits at a time)
3. Add/subtract hex fractions
4. Hex-normalize: shift left until leading hex digit is non-zero
   Shows the "wobbling precision" -- leading digit 1 has 1 sig bit,
   F has 4 sig bits. Wasted bits = 4 - leading_bits.
5. Truncate (hfloat never rounds up, unlike IEEE)
6. Result in hex notation

Students see why hex normalization causes 0-3 bits of wasted
precision that IEEE binary normalization avoids entirely.

Resolves #658

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix(ucalc): consistent precision and div stub for hfloat steps

Two nitpick fixes:

1. Multiplication result uses std::setprecision(10) via ostringstream
   instead of std::to_string (which gives only 6 digits).

2. Division returns explicit "not yet implemented" step instead of
   empty vector (prevents fallback to IEEE binary decomposition).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix(ucalc): NaN/Inf handling and deduplicate leading-bits in hfloat

Three fixes:

1. decompose_hfloat handles NaN and Inf with early return and
   descriptive hex_str ("NaN", "+Inf", "-Inf").

2. Extract count_leading_bits() helper to deduplicate the
   hex-digit-to-significant-bits calculation used in both
   explain_hfloat_add and explain_hfloat_mul.

3. Both call sites replaced with count_leading_bits(leading_digit).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

44590 of 52858 relevant lines covered (84.36%)

6097133.21 hits per line

Coverage Regressions

Lines Coverage ∆ File
2
89.8
 -0.57% include/sw/universal/number/posito/posito_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 23769032919.1 630
84.36
 GitHub Action Run
Source Files on build 23769032919