Tensor alignStride has now effect

[lrannabauer]

yateto_issue.tar.gz
The attached example seems to ignore the the alignStride=False setting for a matrix U which is the result of a matrix-matrix product SF:

    F     = Tensor('F' , (SWE_DG_DOFS, 3) )
    S     = Tensor('S' , (SWE_DG_DOFS, SWE_DG_DOFS) )
    U     = Tensor('U' , (SWE_DG_DOFS, 3) ,alignStride=False)

kernel:

    example = U["lq"] <= S["lj"] * F["jq"] 

Resulting libxsmm call:
libxsmm_m28_n3_k28_ldA28_ldB28_ldC28_alpha1_beta0_alignedA1_alignedC1_pfsigonly(S, F, U, nullptr, nu
llptr, nullptr);

To run the example set the yateto_dir in runExample.sh and run the script.
Thanks

[uphoffc]

This issue is addressed in the new_alignment_mode branch (PR is open). Please use this branch and pass alignStride=Alignment.Unaligned to force unaligned memory access.

[krenzland]

I've reopened this until the branch is merged into master.