Intel® MKL 11.2 Bug Fixes

Intel® MKL 11.2 Update 4  (20 Sep 2015)

DPD200568644 

vsldConvExec (computes convolution or correlation for multidimensional case) produces the wrong results with MKL 11.2 update2 on AMD*, 64-bit

DPD200569471 

dgelsd (Computes the minimum-norm solution to a linear least squares problem using the singular value decomposition of A and a divide and conquer method) returns the wrong results in the case of GNU* threading 

DPD200571079 Support for heterogeneous nodes in the Parallel Direct Sparse Solver for Clusters
DPD200368814 Apparent memory leak happens in BLAS when using GNU* OpenMP ( libgomp) 
DPD200567191 

Description of mkl_?omatadd routines were improved

DPD200571119 SGEMM performance improvements for the cases when  M=N are about 2K, but  the K size is 10 
DPD200370246 Run-time issue with MKL_Thread_Free_Buffers() call on dynamic-linked Windows when no prior MKL calls made

 

Intel® MKL 11.2 Update 3  (24 Apr 2015)

DPD200563310Removed INCLUDE environment variable from mklvars scripts on Linux and OS X
DPD200568165Fixed the Segmentation fault of Dgemm run in native mode with problem size of 15360x15360
DPD200567263Fixed the problem of mkl_scscmv performance degradation
DPD200543115Fixed the problem of pardiso_getdiag returning pivots in 'wrong' order
DPD200566282Corrected the Error message generated by cluster Pardiso provided with input matrix with the incorrect types
DPD200364328Fixed Pardiso restore handle
DPD200565043Fixed the problem of differing results of Sparse Solvers for Clusters with 1 and 2 threads
DPD200555567Compile time Error which was generated because of conflicting definitions between mkl.fi and BLAS95 fortran module is now fixed
DPD200562025Removed threading restriction for dcsrmultd routine
DPD200529342Fixed Scalability issues on large SGI shared memory systems (1000+ cores)
DPD200552714Intel MKL now gives return code in the case of insufficient memory on the target machine

Intel® MKL 11.2 Update 2  (02 Feb 2015)

DPD200555577Intel(R) Inspector reports P1: Error: Invalid partial memory access in VML
DPD200555092Native C interface for ?geqrf and ?orgqr
DPD200552905Cluster_sparse_solver produces segmentation fault
DPD200552714MKL should give a return code back in the case of insufficient memory on the target
DPD200551706Solving matrix with indefinite pivoting causes incorrect answer
DPD200548107Performance comparison between MKL and IPP on small DGEMM
DPD200547546Cluster_sparse_solver - memory leaks for rank = 0 process
DPD200546215xerbla overrides VML callback
DPD200545682MKL_Free_Buffer Thread Safety
DPD200544895MKL link line advisor produces incorrect link lines (multiple issues)
DPD200542066MKL_DIRECT_CALL for zgemm3 cgemm3
DPD200536540mkl_thread_free_buffers memory leak
DPD200358105MKL access violation with managed unit tests
DPD200534216Code built with FFTW2 wrapper gives seg fault on NHM
DPD200253328FFT convolution performance issue
DPD200509798Summary statistics correlation limited to 6 threads...
DPD200249552Improve DGEMM and DGEMV performance on KNC for 8x8 matrices and oddly-shaped matrices
DPD200385528DSYEVD failure 
DPD200247336Improve DGEMV and DSYMV performance on KNC

 

Intel® MKL 11.2 Update 1  (05 Nov 2014)

DPD200360822Fixed Segmentation fault errors in Intel MKL Scalapack routines - pslaprnt, pclaprin and pzlaprnt
DPD200529405Fixed the problem of LAPACKE_dpptrf with a packed LT row major matrix giving unexpected result
DPD200539888diagonal matrix + pardiso_setenv crashing pardiso - is now resolved
DPD200532954Introduced support for maxfct > 1 in Cluster Pardiso
DPD200541341Pardiso ( SMP version ) reporting the wrong size of memory needed for In-Core computation is now fixed
DPD200541342Fixed the Core dump error with iparm(56) in Pardiso with unsupported matrix type
DPD200249551Improved sequential DGEMM performance (called from parallel region) for MxN matrices (1 < M,N <= 256)
DPD200541048Fixed accuracy issue for OOC mode in Pardiso (where the issue doesn't exist with In-Core mode)
DPD200547167Fixed the problem of "ibmkl_core.so: undefined reference to `__isoc99_fscanf'" with MKL 11.2 on RHEL5
DPD200346477Introduced Const modifier for all Sparse BLAS level 2 routines
DPD200514340Modified RngGaussianMV examples (split existing example by two parts and made each example short)
DPD200360599Introduced examples of sparse / dense matrix operations in MKL, Schur functionality
DPD200536744Fixed the problem of differing results when linked with mkl_gf_intel64 and mkl_intel_intel64
DPD200358204dgetrf no longer spawns multiple threads even for very small matrices (for example, 25x25).Issue is now resolved
DPD200543773Resolved the problem - sscanf() used in MKL service layer causing unexpected termination for ddot()
DPD200542066Implemented MKL_DIRECT_CALL feature for zgemm3 cgemm3
DPD200360822Fixed the problem in ScaLapack: pslaprnt, pclaprin and pzlaprnt producing segmentation fault
DPD200401949Introduced the support of partial factorization and Schur Complement in Pardiso
DPD200359903Resolved - floating invalid/core dumped error in geqrf.f90
DPD200388064Improved Cluster FFT Performance
DPD200362204Some resource files contained old Japanese library names in MKL. These are not corrected
DPD200360966Fixed the warning errors in examples/cblas/common_func.c
DPD200359462Further improved Small DGEMM performance with MKL_DIRECT_CALL feature
DPD200531440Improved the mechanism of reporting incorrect input parameters in ScaLAPACK
DPD200545824Resolved the problem - Cpardiso example failing with 2 MPI procesess
DPD200542630Resolved the problem of cluster_sparse_solver function creating incorrect output

Intel® MKL 11.2  (25 Aug 2014)

DPD200523357In MKL PARDISO C examples like "pardiso_unsym_complex_c.c" and "pardiso_unsym_csc_c.c" column array now allocates elements according to CSR format description
DPD200339806Added a Feature Request Automatic Offload for smaller sizes
DPD200527600Updated the Description content in the file cg_jacobi_precon_c.c
DPD200529078Fixed the DSS crash in dss_reorder function call with CNR mode
DPD200529201Improved output of memory message incase when CPARDISO crash and segfault happens when the size of RHS is very big and memory to work with this case not enough
DPD200346477Added const modifier for ALL CBLAS level 2 routines
DPD200526266Fixed Out of memory error with Cpzgemr2d (addressing the real big tasks to solve in distribution mode)
DPD200532854Fixed the defect of DGEQRF giving wrong results with mkl_set_dynamic(false)
DPD200526266Fixed the defect of DGEQRF giving wrong results with mkl_set_dynamic(false)
DPD200394473Fixed the problem - LAPACKE_dgesdd returns -11
DPD200525377Fixed the problem of matrix inconsistency in cpardiso_unsym_distr_c.c when the initial data ( matrix and rhs) is distributed between several MPI processes
DPD200516177Included a new Feature Request - PARDISO to provide descriptor store/restore capability that can reuse factorization stage result between NASTRAN modules
DPD200516189Introduced support for Direct Sparse Solvers for clusters with single omp thread
DPD200516185Introduced support for Direct Sparse Solvers for clusters with pure MPI mode
DPD200401058Fixed the FFT Error in OOC computation with pardiso
DPD200394086Added mkl_sdiamm example
DPD200515841Fixed DCSRILU0 giving incorrect results when -openmp and -check=stack flags are used
DPD200524415Fixed the request vRngBinomial requiring ntrial >= 0
DPD200530439Removed C++ style comments in the MKL headers
DPD200348654FFT examples (basic_dp_real_dft_3d.c and basic_sp_real_dft_3d.c) description content is should be updated
DPD200521692Fixed the problem of MKL crashing with "Illegal instruction detected" on machines with AMD Jaguar processors
DPD200401462Added Pivot control and diagonal extraction functionality to unsymmetric matrix case and OOC pardiso
DPD200530039Fixed the problem of Pardiso solver hang
DPD200292867Improved performance of Cholesky decomposition on Intel® Many Integrated Core Architecture (Intel® MIC Architecture)
DPD200248270Corrected the problem of DSYGVD on Intel® MIC Architecture returning incorrect results when using selected numbers of threads
For more complete information about compiler optimizations, see our Optimization Notice.