hipBLAS 0.48.0 for ROCm 4.5.0

Added

• Added HIPBLAS_STATUS_UNKNOWN for unsupported backend status codes
• Added more support for hipblas-bench

Fixed

• Avoid large offset overflow for gemv and hemv in hipblas-test

Changed

• Packaging split into a runtime package called hipblas and a development package called hipblas-devel. The development package depends on runtime. The runtime package suggests the development package for all supported OSes except CentOS 7 to aid in the transition. The suggests feature in packaging is introduced as a deprecated feature and will be removed in a future rocm release.

hipBLAS 0.45.0 for ROCm 4.3.1

No changes made for ROCm 4.3.1.

hipBLAS 0.45.0 for ROCm 4.3.0

Added

• Added hipblasStatusToString

Fixed

• Added catch() blocks around API calls to prevent the leak of C++ exceptions

hipBLAS-0.44.0 for ROCm 4.2.0

Added

• Made necessary changes to work with rocBLAS' gemm_ex changes. When using rocBLAS backend, hipBLAS will query the preferable
  layout of int8 data to be passed to gemm_ex, and will pass in the resulting flag. Users must be sure to use the preferable
  data format when calling gemm_ex with a rocBLAS backend.
• Added hipblas-bench with support for:
  • copy, swap, scal

hipBLAS-0.42.0 for ROCm 4.1.0

Added
Added the following functions. All added functions include batched and strided-batched support with rocBLAS backend:
axpy_ex
dot_ex
nrm2_ex
rot_ex
scal_ex

Fixed
Fixed complex unit test bug caused by incorrect caxpy and zaxpy function signatures

Known Issues

• None

hipBLAS-0.38.0 for ROCm 4.0.0

New Features

• No new features

Known Issues

• None

hipBLAS-0.38.0 for ROCm 3.10.0

New Features

• Added hipblasSetAtomicsMode and hipblasGetAtomicsMode
• No longer look for CUDA backend unless --cuda build flag is passed

Known Issues

• None

hipBLAS-0.36.0 for ROCm 3.9.0

New Features

Known Issues

• None

hipBLAS-0.34.0 for ROCm 3.8.0

New Features

• No new features

Known Issues

• None

hipBLAS-0.32.0 for ROCm 3.7.0

New Features

• Improvements to rocblas_Xgemm_batched performance for small m, n, k.
• Improvements to rocblas_Xgemv_batched and rocblas_Xgemv_strided_batched performance for small m (QMCPACK use).
• Improvements to rocblas_Xdot (batched and non-batched) performance when both incx and incy are 1
• FP32 ONNX BERT MI50 performance improved 28%
• FP32 BDAS MI50/MI60 Performance improved significantly
• Added substitution method for small trsm sizes with m <= 64 && n <= 64. Increases performance drastically for small batched trsm.
• Add Fortran interface for BLAS 1, BLAS 2, BLAS 3
• Add tbsv, tbsv_batched, and tbsv_strided_batched
• Add hemm, hemm_batched, and hemm_strided_batched
• Add symm, symm_batched, and symm_strided_batched
• Add complex versions of geam, along with geam_batched and geam_strided_batched
• Add gemm_batched_ex and gemm_strided_batched_ex

Known Issues

• None