qpms/qpmsblas.h

   1 /*! \file qpmsblas.h
   2  * \brief Naïve implementation of BLAS functions to workaround OpenBLAS bugs.
   3  *
   4  * There is a bug in certain versions of OpenBLAS causing crashes if many
   5  * BLAS functions are run from different threads in parallel.
   6  * Typically, this happens when one calculates some large matrix (such as the
   7  * global translation matrix) in several threads and cblas_zgemm()
   8  * is called on relatively small submatrices.
   9  * Because the submatrices are small, we can use a naïve,
  10  * serial implementation of cblas_zgemm() as a workaround without a substantial
  11  * impact on performance.
  12  *
  13  * If included together with <cblas.h>, this must be include _afterwards_ because of the typedefs!
  14  */
  15 #ifndef QPMSBLAS_H
  16 #define QPMSBLAS_H
  17 #define QPMS_BLAS_INDEX_T long long int
  18
  19 #ifndef CBLAS_H
  20 typedef enum {CblasRowMajor=101, CblasColMajor=102} CBLAS_LAYOUT;
  21 typedef enum {CblasNoTrans=111, CblasTrans=112, CblasConjTrans=113} CBLAS_TRANSPOSE;
  22 typedef enum {CblasUpper=121, CblasLower=122} CBLAS_UPLO;
  23 typedef enum {CblasNonUnit=131, CblasUnit=132} CBLAS_DIAG;
  24 typedef enum {CblasLeft=141, CblasRight=142} CBLAS_SIDE;
  25 #endif
  26
  27 /// Naïve serial reimplementation of cblas_zgemm.
  28 void qpms_zgemm(CBLAS_LAYOUT Order, CBLAS_TRANSPOSE TransA, CBLAS_TRANSPOSE TransB,
  29                 const QPMS_BLAS_INDEX_T M, const QPMS_BLAS_INDEX_T N, const QPMS_BLAS_INDEX_T K,
  30                 const _Complex double *alpha, const _Complex double *A, const QPMS_BLAS_INDEX_T lda,
  31                 const _Complex double *B, const QPMS_BLAS_INDEX_T ldb,
  32                 const _Complex double *beta, _Complex double *C, const QPMS_BLAS_INDEX_T ldc);
  33
  34 #endif //QPMSBLAS_H