stillwater-sc / universal / 23619415008

85%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

feat(numeric): cross-type conversion infrastructure (#613)

* feat(convert): add cross-type conversion infrastructure for issue #197

Add universal_cast<>, mixed-type operators, and UniversalNumber concept
to enable automatic conversions between posit, cfloat, fixpnt, lns, and
integer types without module cycles.

* fix(test): clean up malformed comments in regression test guards

* fix(numeric): correct misleading long double comment in universal_convert

* test(convert): add exhaustive cfloat conversion regression tests

Add the regression tests from the PR test plan:

  LEVEL_1:
  - Exhaustive cfloat<12,4> -> cfloat<8,3>  (4096 encodings, narrowing)
  - Exhaustive cfloat<8,2>  -> cfloat<16,5> (256 encodings,  widening)
  - Exhaustive cfloat<8,3>  -> cfloat<12,4> (256 encodings,  widening)
  - Spot checks cfloat<32,8> <-> cfloat<16,5> special values
    (zero, +inf, -inf, qNaN)

  LEVEL_2:
  - Exhaustive cfloat<16,5,ttt> -> cfloat<8,3,ttt> (65536 encodings,
    saturating + subnormals + max-exp narrowing)

Also fix the regression level guards to follow the standard cmake
REGRESSION_LEVEL_OVERRIDE pattern used throughout the rest of the
test suite, replacing the hardcoded LEVEL_1=1 / LEVEL_2-4=0 block.

* fix(convert): use precision-aware intermediate to avoid double bottleneck

Replace the fixed double intermediate in universal_cast and universal_assign
with a compile-time selection via if constexpr:
  - precision <= 52 bits: use double (exact, common case)
  - precision >  52 bits: use long double (80-bit on x86/x87)

This addresses the concern that marshalling through double silently loses
precision for Universal types wider than double's 52-bit significand.
The long double path covers types up to 63 fraction bits exactly on x86;
on MSVC/ARM (where long double == double) the precision is unchanged.

Types wider than long double require an efloat/ereal intermediate or a
proper expression-template layer for AST-level type-promotion analysis.

* fix(... (continued)

Coverage Stats

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

5 existing lines in 2 files now uncovered.

44415 of 52625 relevant lines covered (84.4%)

6131075.32 hits per line