Releases: ROCm/rocALUTION
Releases · ROCm/rocALUTION
rocALUTION 3.2.1 for ROCm 6.3.0
Added
- Support for gfx1200, gfx1201, and gfx1151.
Changed
- Changed the default compiler from
hipcc
toamdclang
in the installation script and cmake files. - Changed the address sanitizer build targets. Now only
gfx908:xnack+
,gfx90a:xnack+
,gfx940:xnack+
,gfx941:xnack+
, andgfx942:xnack+
are built withBUILD_ADDRESS_SANITIZER=ON
.
Resolved issues
- Fix hang in
RS-AMG
for Navi on some specific matrix sparsity patterns. - Fix wrong results in
Apply
on multi-GPU setups.
rocALUTION 3.2.1 for ROCm 6.2.4
Additions
- GFX1151 Support
rocALUTION 3.2.0 for ROCm 6.2.2
rocALUTION code for ROCm 6.2.2 did not change. The library was rebuilt for the updated ROCm 6.2.2 stack.
rocALUTION 3.2.0 for ROCm 6.2.1
rocALUTION code for ROCm 6.2.1 did not change. The library was rebuilt for the updated ROCm 6.2.1 stack.
rocALUTION 3.2.0 for ROCm 6.2.0
Additions
- New file I/O based on rocsparse I/O format
GetConvergenceHistory
for ItILU0 preconditioner
Deprecations
LocalMatrix::ReadFileCSR
LocalMatrix::WriteFileCSR
GlobalMatrix::ReadFileCSR
GlobalMatrix::WriteFileCSR
rocALUTION 3.1.1 for ROCm 6.1.2
rocALUTION code for ROCm 6.1.2 did not change. The library was rebuilt for the updated ROCm 6.1.2 stack.
rocALUTION 3.1.1 for ROCm 6.1.1
rocALUTION code for ROCm 6.1.1 did not change. The library was rebuilt for the updated ROCm 6.1.1 stack.
rocALUTION 3.1.1 for ROCm 6.1.0
Additions
TripleMatrixProduct
functionality forGlobalMatrix
- Multi-Node/GPU support for
UA-AMG
,SA-AMG
andRS-AMG
- Iterative ILU0 preconditioner
ItILU0
- Iterative triangular solve, selectable via
SolverDecr
class
Deprecations
LocalMatrix::AMGConnect
LocalMatrix::AMGAggregate
LocalMatrix::AMGPMISAggregate
LocalMatrix::AMGSmoothedAggregation
LocalMatrix::AMGAggregation
PairwiseAMG
Known Issues
PairwiseAMG
does currently not support matrix sizes that exceed int32 rangePairwiseAMG
might fail building the hierarchy on certain input matrices
rocALUTION 3.0.3 for ROCm 6.0.2
rocALUTION code for ROCm 6.0.2 did not change. The library was rebuilt for the updated ROCm 6.0.2 stack.
rocALUTION 3.0.3 for ROCm 6.0.0
Added
- Added support for 64bit integer vectors
- Added inclusive and exclusive sum functionality for Vector classes
- Added Transpose functionality for Global/LocalMatrix
- Added TripleMatrixProduct functionality LocalMatrix
- Added Sort() function for LocalVector class
- Added multiple stream support to the HIP backend
Optimized
- GlobalMatrix::Apply() now uses multiple streams to better hide communication
Changed
- Matrix dimensions and number of non-zeros are now stored using 64bit integers
- Improved ILUT preconditioner
Removed
- Removed LocalVector::GetIndexValues(ValueType*)
- Removed LocalVector::SetIndexValues(const ValueType*)
- Removed LocalMatrix::RSDirectInterpolation(const LocalVector&, const LocalVector&, LocalMatrix*, LocalMatrix*)
- Removed LocalMatrix::RSExtPIInterpolation(const LocalVector&, const LocalVector&, bool, float, LocalMatrix*, LocalMatrix*)
- Removed LocalMatrix::RugeStueben()
- Removed LocalMatrix::AMGSmoothedAggregation(ValueType, const LocalVector&, const LocalVector&, LocalMatrix*, LocalMatrix*, int)
- Removed LocalMatrix::AMGAggregation(const LocalVector&, LocalMatrix*, LocalMatrix*)
Fixed
- Unit tests do not ignore BCSR block dimension anymore
- Fixed typos in the documentation
- Fixed a bug in multicoloring for non-symmetric matrix patterns