stillwater-sc / universal / 36142964455
87%
master: 84%

LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran
Jobs 1
Files 676
Run time 3min
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: 85.977% (-0.01%) from 85.99%
36142964455

push

github

web-flow 
fix(headers): exact decimal digit and exponent traits across the remaining types (#1607)

* fix(headers): exact decimal digit and exponent traits across the remaining types

Closes the digits10 work in one change: #1601's remainder, #1602, #1603 and
#1604. After this there is no `/ 3.3` left in any numeric_limits header.

#1601, the last three types. is_exact settles which pair applies, so the
"contract needs deciding" note on the issue was over-cautious -- the types
already declare what they are. fixpnt reports is_exact false and takes the
floating-point pair; quire and rational report is_exact true and take the
integer pair, with max_digits10 0 as the native integers have it. rational was
returning digits10 + 1 for a type that has no decimal round trip to size. The
base8/base10/base16 rational specializations express digits in base-N digits
rather than bits and are left alone.

#1602, lns and dbns. digits was derived from the exponent RANGE, so lns<32,8>
reported 4194312 significand bits and 1271003 decimal digits in a 32-bit type.
An lns stores a fixed-point exponent and its precision is uniform in log space,
so there is no IEEE significand to count; adjacent values differ by a factor
2^(2^-rbits), which is an equivalent significand width of rbits + 0.5 at every
configuration measured, making rbits + 1 the honest round-up. lns<32,8> now
reports 9 digits and 2 decimal digits.

#1603, bfloat16. digits was 7, the fraction width, where the convention
everywhere else counts the implicit bit too: float is 24 for 23, microfloat is
fbits + 1. It could not be folded into #1605 because the two defects interact --
the exact formula on digits=7 gives digits10=1, worse than the accidentally
correct 2 that digits/3.3 produced. Fixing digits first makes it 2 by the
formula.

#1604, min_exponent10 and max_exponent10. The same approximation, on a pair
that additionally relied on static_cast truncating a negative quotient toward
zero to stand in for ceil. duble repo... (continued)

45506 of 52928 relevant lines covered (85.98%)

9402821.36 hits per line

Coverage Regressions

Lines Coverage ∆ File
8
95.41
-0.45% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 36142964455.1 676
85.98
GitHub Action Run
Source Files on build 36142964455