stillwater-sc / universal / 35406798454

86%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

test(cfloat): make the saturating arithmetic suites test saturation (#1560)

None of the 16 suites under static/float/cfloat/arithmetic/saturating/ could
fail CI on saturation: eight set isSaturating = false (copied from the
non-saturating tree) and eight ran with MANUAL_TESTING 1, which always
returns success. All 16 now test the saturating encoding and gate at
level 1, with labels, tags and suite names derived from their directory.

Turning them on exposed three problems in the tests:
- normal/, subnormal/ and maxexpvalue/addition.cpp listed es = 1
  configurations, which cfloat's static_assert forbids without both
  subnormals and max-exponent values; manual mode meant they never
  compiled. Removed (es = 1 stays covered in subsuper/).
- test_suite_randoms.hpp's executeBinary converted double's result straight
  into the test type, so an overflow of finite operands gave an inf
  reference where a saturating type correctly gives maxpos: 234 of 2,000
  random cfloat<64,11,fft> products were reported as failures. The
  reference now clamps the way a saturating type does. The random sweeps in
  normal/multiplication and normal/division, which ran 0 cases, now run
  1,000 for c16, c32 and c64 (c48 double-rounds through the reference).
- cfloat<14..16,11> with max-exponent values reach 2^1024, past DBL_MAX,
  so the double reference reads them as inf; removed from the level-4
  sweeps in maxexpvalue/ and subsuper/.

subsuper/multiplication's 9-bit sweeps (11.6 s) move to level 2.

Resolves #1557

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Coverage Stats

4 of 4 new or added lines in 1 file covered. (100.0%)

64 existing lines in 5 files now uncovered.

44860 of 52285 relevant lines covered (85.8%)

9197876.12 hits per line