Releases: llnl/FPChecker
Releases · llnl/FPChecker
FPChecker 0.6.0
- Added FP32 rounding-error tracking (shadow-value style), including improved accumulation across operations and files.
- Added line-granularity reporting for accumulated rounding error, including per-line visualizations in reports.
- Reports high-error hotspots that can help guide mixed-precision and lower-precision tuning decisions.
- Improved reporting outputs: better HTML layout, shell-friendly rounding-error tables, and support for multi-file rounding sections.
- Expanded error-analysis coverage for math functions, floating-point return values/calls, and MPI-related paths.
- Added substantial regression tests for FP32 error analysis and report generation.
- Fixed multiple bugs and memory leaks in analysis and reporting components.
FPChecker 0.5.0
- Improved histograms of exponent usage (added to main report). Merged branches.
- Support for CUDA has been dropped
- Design of the report changed
- New tests added
- Fixed several bugs
FPChecker 0.4.0
FPChecker 0.4.0
FPChecker 0.3.5
FPChecker 0.3.5
FPChecker 0.3.4
FPChecker 0.3.4
FPChecker 0.3.3
FPChecker 0.3.3
FPChecker 0.3.2
FPChecker 0.3.2
FPChecker 0.3.1
FPChecker 0.3.1
FPChecker 0.3.0
FPChecker 0.3.0
FPChecker 0.2.1
Version 0.2.1