Intel® Math Kernel Library 2020 Bug fixes

By Khang T Nguyen, Pamela Harrison

Published:12/03/2019   Last Updated:08/05/2020

Intel® Math Kernel Library (Intel® MKL) 2020 update 2 (July 2020).

MKLD-9780 Fixed issue with Intel® MKL SGEMM (Intel® AVX512) under performing Intel® MKL sgemv (Intel® AVX2).
MKLD-9720 Fixed incorrect Eigensolver examples.
MKLD-9713 Improved Eigensolver examples.
MKLD-9664 Fixed no const keyword for '*beta' of DSYR2K in mkl_blas.h.
MKLD-9570 Fixed unstable Intel® MKL Benchmark above 290 threads.
MKLD-9504 Fixed Intel® MKL ERROR: Parameter 15 was incorrect on entry to PCGEEVX.
MKLD-9492 Improved mkl_link_tool.exe should be return back for classical version of MKL.
MKLD-9485 Fixed Parallel Direct Sparse Solver for Clusters(CLUSTER_SPARSE_SOLVER) crash at the phase 11 with iparm[2] = 10, iparm[10] = iparm[12] = 0 parameters.
MKLD-9180 Fixed mkl_rt.so missed "fftw_alloc_complex" symbols.
MKLD-9066 Fixed segmentation fault with Intel® MKL Pardiso i_malloc.
MKLD-9064 Fixed Parameter error for CBLAS_ZGEMMT with CblasConjTrans.
MKLD-8660 Fixed Data race problem(s) with SpBLAS (mkl_sparse_sp2m).
MKLD-8512 Fixed Data Fitting Data Races reported by Intel® Inspector.
MKLD-8345 Fixed Intel® MKL Verbose mode which doesn't work with ?DTSVB routine.
MKLD-8196 Fixed memory leakage with MKL_SPARSE_D_MV in the case when lined with Intel® TBB layer.
MKLD-8054 Fixed duplicated value in VSL error.
MKLD-9477 Fixed input array isuppz reference even when jobz parameter is not set to 'V' contrary.

Intel® Math Kernel Library (Intel® MKL) 2020 update 1 (March 2020).

MKLD-3773 Fixed issue: Intel® MKL Pardiso - extra RHS allocation - SIZE*NRHS*sizeof(datatype) for each OpenMP thread causing unnecessary memory usage.
MKLD-4084 Fixed issue: Intel® MKL was searching for dlls in the application process path regardless of where they actually were, resulting in dlls not found.
MKLD-8082 Fixed issue: in Intel® MKL 2019.4, dfeast_[s,h]csrgv produced segmentation fault error due to an error in the verifier of positive define matrix in FEAST.
MKLD-8252 Issue fixed: Incorrect initialization for MKL_SPARSE_D_QR_REORDER in the case of a diagonal matrix was causing exit status 2.
MKLD-8253 Issue fixed: MKL_SPARSE_D_QR_REORDER was returning returns SPARSE_STATUS_ALLOC_FAILED when input matrix contained either a zero-row or a zero-column.
MKLD-8424 Docs fixed: Documentation for mkl_?omatcopy2 transpose was incorrect, causing incorrect results.
MKLD-8649 Docs fixed: Documentation for LaPack storage scheme row-major layout formula was incorrect.
MKLD-8757 Issue fixed: ?geev was allocating an extra row/column in the working array.
MKLD-8776 Issue fixed: mkl_zcscmv and mkl_sparse_z_mv produced different results in Intel® MKL 2019.5.
MKLD-9118 Fixed issue: Loading mkl_core.dll from Intel® MKL 2020 resets the DLL search order, breaking any application relying on the DLL search order set by SetDllDirectory.
MKLD-9146 Not a defect: It can appear that avx512_xsgemv takes longer than the intrinsic functions "avx2_sparse_sgemv_accum16" and "avx512_sparse_sgemv_accum16". However, this is due to function overhead when using verbose.

Intel® Math Kernel Library (Intel® MKL) 2020 (December 2019).

MKLD-3727 Fixed the issue "Cluster Intel® MKL Pardiso OOC mode doesn't work for nonsymmetric matrices when IPARM (11) and IPARM(13) are activated."
MKLD-3733 Fixed the issue MKL_SPARSE_D_GV returns SPARSE_STATUS_ALLOC_FAILED=2.
MKLD-3801 Fixed the issue Intel MKL CNR strict support in the case of different number of threads.
MKLD-3857 Fixed the issue Cluster_sparse_solver fails when MPI version of the nested dissection and symbolic factorization algorithms is used.
MKLD-3907 Removed PGI Support Notes.
MKLD-3940 Fixed the unhandle exception issue when using SparseQR.
MKLD-3958 Added Intel® MKL 2019 to the Intel® Link Line Advisor tool.
MKLD-3966 Fixed the issue Intel® MKL Pardiso cannot being interrupted with mkl_progress function.
MKLD-3971 Fixed the issue Windows Loader deadlock when loading Intel(R) MKL.
MKLD-3975 Fixed the issue MKL_SPARSE_D_MV returns "Intel MKL ERROR: CPU 1 is not supported" when MKL_CBWR=COMPATIBLE.
MKLD-3981 Fixed the issue TREVC3 returns Intel MKL ERROR: Parameter 14.
MKLD-3998 Fixed the issue Intel MKL in mxnet hangs on Mac* OS.
MKLD-3999 Fixed the issue LAPACKE_SSYEVD fails when upper triangular part of the matrix is filled with random numbers.
MKLD-4014 Fixed the issue Intel MKL doesn't limit the number of threads in case MKL_NUM_THREADS is set.
MKLD-4033 Fixed the issue "LNK2005: _POWF already defined in libmmt.lib".
MKLD-4044 Fixed the issue DORMQR returns "Intel MKL ERROR: Parameter 12 was incorrect on entry"
MKLD-4045 Fixed the issue DGEEVX crashes on the exit of the application.
MKLD-4052 Fixed the issue The Executable FFT code size increased around of 2x from MKL versions 11.3.3 to 19.1.
MKLD-4069 Fixed the issue Intel® MKL Pardiso, schur complement method accuracy is not enough.
MKLD-4084 Fixed the issue MKL Libraries mkl _ *. Dll have to load their dependence on the absolute path of the folder where the DLL itself is located.
MKLD-4123 Fixed the issue MKL FFTW produces non-identical outputs.
MKLD-4176 Fixed the issue MKL_SPARSE_SPMM returns wrong results when small matrices multiply large matrices.
MKLD-4217 Fixed the issue Intel® MKL Pardiso reports the "error PARDISO: iterative refinement, contraction rate is greater than 0.9" problem.
MKLD-4235 Fixed the issue MKL_SPARSE_?_TRSM didn't check the format of the Input matrix.
MKLD-7629 Fixed the issue Intel® MKL Pardiso returns NaN in theo solution when the input matrix is ill-conditioned.
MKLD-7663 Fixed the issue MKL_SPARSE_ORDER is too slow.
MKLD-7744 Fixed the issue ?SYEVX routines produce wrong results when using threading.
MKLD-7820 Fixed the issue LAPACKE_DSYEV returns wrong results when using the row-major layout.
MKLD-7851 Fixed the issue FFT generates Floating Point Exception.
MKLD-7875 Fixed the issue Intel® MKL Pardiso 2019.3 produces segmentation fault when parallel factorization control is used.
MKLD-7891 Fixed the issue Intel® MKL Pardiso parallel factorization control (iparm(23)) causes segmentation fault error.
MKLD-8114 Fixed the issue DGESVD performance issue when running in 64bit vs ia32 systems with AVX2 code branch.
MKLD-8278 Fixed the issue DGECON shows 2x performance degradation in 2019.4 vs 2018.1.
MKLD-8357 Fixed the issue ?GTSV - erratum into parameter description.
MKLD-8364 Fixed the issue Static MKL Build fails in INtime RTOS.
MKLD-8369 Fixed the issue SPARSE_STATUS_NOT_SUPPORTED exit status in mkl_sparse_sp2m with symmetric sparse matrices.
MKLD-8384 Fixed the issue Inconsistent C definition of ?gelsd function.
MKLD-8498 Fixed the MKL_SPARSE_S_QR parameter order of columns and x.
MKLD-8650 Fixed the MKL_SPARSE_SP2M description.
MKLD-8686 Fixed the MKL_SPARSE_?_ADD - Application Notes.

Product and Performance Information

1

Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Notice revision #20110804