This changes the definitions of the matrix load/store intrinsics to load/store from/to a pointer, and not to a vector, as discussed in D83477. I will commit this and D83477 as 1 commit, put am putting this new bit up for review separately because that makes this part easier to review.
LGTM, thank you very much! This has been discussed already in D83477. Passing in a pointer to a vector is confusing, as it seems to suggest we only access elements of the vector pointer, but we may access other elements, if the stride > rows.