Can tensor cores be used on non-ideal matrices? and some other questions about tensor cores

From what i’ve read from about tensor cores and cublas they only work with matrices with dimension in multiples of 8. Therefore if I have 2 matrices A[18393][663] B[663][40] , and I wish to do matrix matrix multiplication AxB what is the best way to optimize this. Should I pad the matrices with 0 till they are multiples of 8 and use gemmEx()? or should i use wmma to do it part by part .
Also cublasXtGemmEx() does not support tensor cores right?