OpenBLAS v0.3.9 Release Notes
Release Date: 2020-03-01 // about 4 years ago-
common:
- ๐ Fixed a miscompilation of the GETRF functions with CMAKE
- The size of the memory buffer used for splitting GEMM tasks across
๐ท multiple threads can now be configured in the build system. - ๐ Imported bugfix 390 from LAPACK (missing NaN propagation in xCOMBSSQ)
POWER:
- ๐ fixed several compilation problems related to endianness and
๐ ELF version support on POWER8 and POWER9. - ๐ fixed misuse of the absolute value IAMIN/IAMAX in place of IMIN/IMAX
- ๐ fixed a race condition in the level3 blas code
MIPS64:
- ๐ fixed misuse of the absolute value IAMIN/IAMAX in place of IMIN/IMAX
ARMV7:
- ๐ fixed a race condition in the level3 blas code
- ๐ fixed a compilation problem on Android
ARMV8:
- โ Added support for Ampere EMAG8180
- โ Added support for Neoverse N1
- ๐ improved performance of the blas_lock function
- ๐ fixed a race condition in the level3 blas code
- ๐ Fixed a performance regression on TSV110 servers
x86_64:
- ๐ Fixed a long-standing error with undeclared register clobbers in
the DSCAL microkernel for Haswell,SkylakeX and Zen exposed by gcc9.2 - ๐ Fixed a long-standing bug in the SSE implementation of the IAMAX functions
- Fixed a cmake build failure with DYNAMIC_ARCH on x86_64
- ๐ Fixed an oversight in the cpu detection code for Intel Goldmont+,
Cannon Lake and Ice Lake - ๐ Fixed compile failure on OSX when the compiler name contains a dash
(e.g. gcc-9) - ๐ Fixed compilation with MinGW on SkylakeX
- ๐ Improved speed of the AVX512 GEMM3M code, added an AVX512 kernel for
๐ STRMM and improved performance of the AVX2 GEMM kernels
IBM Z:
- ๐ fixed compilation of the DYNAMIC_ARCH code