Function calcNrThreads

src/library/blas/gens/gbmv.cpp:189–254

}

static void
calcNrThreads(
    size_t threads[2],
    const SubproblemDim *subdims,
    const PGranularity *pgran,
    const void *args,
    const void *_extra)
{
	int BLOCKSIZE = pgran->wgSize[0] * pgran->wgSize[1]; // 1D Block
	size_t fM, fN;

    const CLBlasKargs *kargs = (const CLBlasKargs *)args;
	const CLBLASKernExtra *extra = ( CLBLASKernExtra *)_extra;

	clblasOrder order = ( extra->flags & KEXTRA_COLUMN_MAJOR) ? clblasColumnMajor: clblasRowMajor;
	clblasTranspose trans = ( extra->flags & KEXTRA_TRANS_A) ? clblasTrans :
								(( extra->flags & KEXTRA_CONJUGATE_A) ? clblasConjTrans: clblasNoTrans);

    fM = kargs->M;
    fN = kargs->N;
    if ( order == clblasColumnMajor )
    {
        order = clblasRowMajor;
        fM = kargs->N;
        fN = kargs->M;
        if ( trans == clblasNoTrans)
        {
            trans = clblasTrans;
        }
        else if ( trans == clblasTrans )
        {
            trans = clblasNoTrans;
        }
        else // clblasConjTrans
        {
            trans = clblasNoTrans;
        }
    }
    if( (kargs->pigFuncID == CLBLAS_SBMV) || (kargs->pigFuncID == CLBLAS_HBMV) )    // Only NT kernel is used
    {
        trans = clblasNoTrans;
    }

	size_t blocks;
	size_t H = subdims->x;
	size_t TARGET_ROWS =  BLOCKSIZE / H;

	if( trans == clblasNoTrans )
	{
    	blocks = ((fM - 1)/ TARGET_ROWS) + 1;
    }
    else {
        blocks = ((fN - 1)/ H) + 1;
    }

	threads[0] = blocks * BLOCKSIZE;
	threads[1] = 1;