Porting it from rvv 0.7.1 to rvv 1.0 should be pretty much one to one. The harder part is that this and other gemm implementations often aren't vector length ag

Porting it from rvv 0.7.1 to rvv 1.0 should be pretty much one to one. The harder part is that this and other gemm implementations often aren't vector length agnostic, this is mostly due to gemm requiering specific knowlage about the cpu to get good performance anyways. Btw, with -O3 -march=rv64gcv both gcc and clang can autovectorize both of your example functions. I'd use -march=rv64gcv_zba_zbb_zbs until profiles are supported in the compilers, then we could so something like -march=rva23.