v3.9.0.21

What's Changed

Fix

• Fix: Fix an extremely serious bug that causes incorrect results in GPU RT-TDDFT calculation by @AsTonyshment in #6767
• Fix energy jumps occur in rt-TDDFT calculations by @ESROAMER in #6779
• [Fix] Report and exit for unsupported DSP KPAR configuration by @Cstandardlib in #6771
• [Fix]Enable abacus to be compiled with CXX 17 & CUDA 13 by @ZhouXY-PKU in #6777

Feature

• Add average iterative diagonalization step output by @Cstandardlib in #6795

Refactor

• Refactor and Optimize snap_psibeta_half_tddft by @dzzz2001 in #6765
• [Refactor] Support different CUDA versions in one single cuda_compat.h by @Cstandardlib in #6770

Test

• Test: Update the parameter and reference results of 23_OF_LPS by @sunliang98 in #6774
• Test: Update CI/CD tests for RT-TDDFT and add hybrid gauge and parameter td_dt tests for CPU and GPU by @AsTonyshment in #6776
• [Test] Use WARNING_QUIT instead of exit, Add and refactor unittests for ParaGlobal by @Cstandardlib in #6781

Others

• Build(deps): Bump actions/upload-artifact from 5 to 6 by @dependabot[bot] in #6783
• Update version to v3.9.0.21 by @xuan112358 in #6797

Full Changelog: v3.9.0.20...v3.9.0.21