Intel® Math Kernel Library (Intel® MKL) 2017 Bug Fixes list

Intel(R) MKL 2017 Update 4 ( September 2017)

MKLD-3176Fixed the wrong result issue of dlange routine for some specific problem sizes and when GNU OpenMP threading layer is used
MKLD-3089Fixed the problem when Intel MKl Pardiso spends too much time in timer of Intel Pardiso routine
MKLD-3074Added LAPACKE ?larcm function description to the Developer Reference for MKL 2018 - C
MKLD-2999Fixed the problem with LAPACKE_dstein routine when it uses nth element of the off-diagonal array (which is supposed to have n-1 elements)
MKLD-2994Fixed the Unhanding Exception bug in MKL Data Fitting API dfdNewTask1D routine
MKLD-2973Improved the performance of 2D FFT in-place computations for some specific problem size on Intel® Xeon Phi™ product family x200
MKLD-2797Fixed the problem of ?GETRS in the case of a singular matrix
MKLD-2788Fixed the problem when mkl_set_num_threads routine is causing segmentation fault for Intel PARDISO
MKLD-2770Fixed the problem of Intel Pardiso in the case when CBWR mode and for some specific problem sizes
MKLD-2766Improved the problem when Intel Pardiso shows poor multithreading performance when solving multiple RHS and linking against MKL with TBB threading layer
MKLD-2733Fixed the _MSC_VER mismatch again MKL installation problem
MKLD-2704Fixed the issue whit missing variables in stdcall headers in dorm22 and some others Lapack's routines


Intel(R) MKL 2017 Update 3 ( May 2017)

DPD200591548Fixed the MKL 2017 Update 2 Auto Linking issue in Visual Studio 2015 
DPD200591000Fixed the mkl_sparse_s_add issue 
DPD200590797Fixed the Data Fitting Example issue with passing incorrect value of method to DFDInterpolate1D function.
DPD200417141Fixed the LAPACKE_zposv computation solution
DPD200590397Fixed Intel(R) MKL Sparse Solver ( DSS API ) issue with different results call dss_factor_real twice after changing the "values" of the matrix, not the structure.
DPD200589917Fixed the issue with Intel(R) Pardiso hanging at the factorization stage
DPD200576142Improved the performace of functions DORCSD2BY1 routine
DPD200575247Added declaration of PZLAHQR into Intel MKL Reference Manual
DPD200418144Fixed the wrong result issue when calling zgetrf by MKL_DIRECT_CALL
DPD200589953Fixed the issue DGETSLS routine in the case of some specific tall and skinny inputs
DPD200591941Fixed the significant performance degradation of DGETRF routine for some specific problem sizes on AVX2 IA.

Intel(R) MKL 2017 Update 2 ( 22 February 2016)

DPD200590035Fixed the Intel MKL PARDISO Schur complement incorrect results when problem size exceeded 500x500
DPD200589865Improved the description of cpstrf and zpstrf routines
DPD200589508Fixed the mkl_cspblas_dbsrgemv issue when the input matrix contained empty (zeroed) row
DPD200589306Fixed incorrect behavior of Intel MKL PARDISO in parallel mode
DPD200588868Added support of Intel® Threading Building Blocks (Intel® TBB) threading layer for Intel MKL PARDISO at the solving stage
DPD200415105Fixed LAPACKE_zgesdd failure with some specific input problems
DPD200588368Improved the performance of dlange routine with the first argument 'F' - Frobenius norm
DPD200588151Fixed vdrnggaussian routine failure for some specific seed values and when -fpe0 compiler flag has been used
DPD200587834Introduced MKL_NUM_STRIPES environment variable and accompanying Intel MKL support functions to control the 2D partitioning of multithreaded GEMM
DPD200587189Fixed the Parallel Direct Sparse Solver for Clusters failure when 0-based indexed BSR matrix is used
DPD200586069Improved the GEMM performance of small tall-and-skinny matrices on Intel® Xeon Phi™ processor x200
DPD200585963Fixed performance degradation with DGEMM when the leading dimensional  was a multiple of 256
DPD200582562Added support for mkl_progress in p?getrf ScaLAPACK routine
DPD200417039Addressed DZGEMM failures on 64-bit Intel® Advanced Vector Extensions 2 (Intel® AVX2), Intel® Advanced Vector Extensions 512 (Intel® AVX-512) architectures for cases where matrix C had a large leading dimension
DPD200589046Fixed the inconsistency of vslSSNewTask API for Windows* OS and Linux* OS
DPD200585715Fixed the vslLeapfrogStream mode issue with Wichmann Hill RNG
DPD200588183 Improved SGEMM performance on Intel® Advanced Vector Extensions (Intel® AVX)for very small matrix sizes


Intel(R) MKL 2017 Update 1 ( 1 November 2016)

DPD200588182Fixed the Parallel Direct Sparse Solver for Clusters failure when number of processes more than 1
DPD200587997Fixed the VML Sinh problem incorrect results in the case with some specific input problems
DPD200587705Added the application Note to the dfdSearchCells1D() routine behavior
DPD200585902Fixed the Inconsistency between declaration in header files and doc's description for pstrmv routine 
DPD200585716Fixed the LAPACKE_dsyev issue when Automatic Offload mode doesn't work on Intel® Xeon Phi™
DPD200585668Fixed the latest HPCG failure with the small problem size on KNL
DPD200585403Fixed the issue when Outlier detection is hanging in BACON computation
DPD200585222Fixed the Intel MKL PARDISO iterative refinement problem when extended precision ( iparm[8] < 0) is used
DPD200584932Improved the description of Column-major BSR format with zero based indexing
DPD200584617Improved the description of  Variable BSR format (  block compressed sparse row ) for sparse matrixes
DPD200374978Improved the performance of GEMM family for skinny matrix A(n,k) for n/k > 100 on KNL
DPD200412741Fixed cblas_dgemm compilation error when MKL_DIRECT_CALL_SEQ and /DMKL_ILP64 compiler options are used
DPD200412992Improved the performance of PZGETRF routine for some specific problem sizes
DPD200586734Fixed the zfeast_hcsrev issue when routine corrupts the m0 value in some specific cases
DPD200586957Fixed the Intel MKL custom lib build tool issue when tool doesn't work with makefile in a non-default location
DPD200587189Fixed the problem of Intel Parallel Direct Sparse Solver for Cluster routine when 0-based indexed BSR matrix is used
DPD200587659Fixed the issue with ScaLapack pdtran routine which crashes with the message: "MKL_SCALAPACK_ALLOCATE in pdtran_.c is unsucceseful, size = 18446744073510584320"
DPD200587661Fixed the "Conditional jump or move depends on uninitialised value(s)" issue detected valgrind or gdb with Intel MKL Pardiso
DPD200588229Fixed the compile time error when compiling the mkl_sparse_set_memory_hint routine with SPARSE_MEMORY_AGGRESSIVE option 
DPD200556202Optimized the performance of fully distributed reordering step of Parallel Direct Sparse Solver for Clusters
DPD200586299Fixed the problem when v?Powx routine changes MM_GET_FLUSH_ZERO_MODE() behavior


Intel® MKL 2017 ( 6 September 2016) 

DPD200584193Parallel Direct Sparse Solver for Cluster crahes or returns wrong solving comlex linear systems with Hermit Matrixes in the case if number of MPI processes > 2
DPD200583633Fixed the problem when Intel MKL produces the error message "mismatch detected for '_MSC_VER' .... " when staticaly links with mkl_tbb_thread.lib
DPD200583000Added description of LapackE_xerbla routine to the MKL's documentation
DPD200409897Fixed the Intel MKL Pardiso failure when multiple simultaneous  (single-threaded) instances are used in a OpenMp loop
DPD200582626Fixed the Intel MKL Pardiso failures when METIS reodering and huge block on the diagonal 
DPD200578518df?InterpolateEx1D function returns incorrect index of the cell that contains the right boundary of the interpolation interval
DPD200243632Extended version of the Integrate function, df?IntegrateEx1D does not pass the type of integration limit, left or right, into the callback function 
DPD200582431Fixed CP2K code crash in MKLs routines after moving since MKL 11.3.1 to 11.3.2
DPD200584978Fixed Intel MKL ERROR: Parameter 3 was incorrect on entry to DSYTRF
DPD200571698Optimized performance of ?getrf, ?getrs and ?getri for very small cases.
DPD200584139Fixed mkl_lapack.h : const modificator have to be removed (?lacrm,?larcm)
DPD200575462Added support of partial SVD functionality in MKL
DPD200582751Fixed low Automatic Offload performance of mkl_?getrfnpi
DPD200584255Fixed issue with calling LAPACK functions with a workspace query so that unused parameters could be passed in as nullptr.

Intel® MKL 2017 Beta ( 17 Feb 2016) 

DPD200577117 SP2DP custom library sources created and published in online article
DPD200575663 Fixed the problem of SVD computation of very wide matrix performing significantly slower than SVD computation of its transposition
DPD200574694 Introduced support for major-column layout in the results returned by the df?Interpolate1D routine
DPD200576806 Fixed the Intel MKL Pardiso hang problem when called from an OMP critical section
DPD200580172 Fixed the problem with Intel MKL Pardiso's hanging on the reordering stage for real and symmetric indefinite matrix
DPD200576441 Fixed the issue with Intel MKL Cluster FFTW and original FFTW produce the different results in the case of out-of-place computation
DPD200574978 Resolved the segmentation fault resulting from a call to pthread_exit(0)
DPD200571078 Introduced mkl_progress support for the Parallel Direct Sparse Solver for Clusters and fixed the incorrect behavior with mkl_progress routine from Intel MKL Pardiso for SMP
DPD200372223 Improved performance of spline interpolation on multiple threads
DPD200374978 Improved performance of GEMM family routines for skinny matrix A(n,k) for n/k > 100 for Intel(R) Xeon Phi™ ( aka KNL )


For more complete information about compiler optimizations, see our Optimization Notice.