In hopper, the blockM/blockN can not be changed....? Otherwise will trigger bug:
Wrong answer! 28766439 errors! 48.989%
Average diff = 1255.64
test: ../../../include/common.h:396: void assert_allclose(DType*, DType*, std::vector, float, bool) [with DType = __half]: Assertion `errors == 0' failed.
Aborted (core dumped)