Incorret output of POA with automatic vectorization in QEMU-VM (RISC-V)
POA kernel does not output the correct output when compiled with -mepi
and without -fno-vectorize
(automatic vectorization enabled) in QEMU-VM (RISC-V).
We must identify the loop that is causing the problem. We can use #pragma clang loop vectorize(disable) interleave(disable)
to disable the vectorization of a specific loop and compile with -Rpass=loop-vectorize
to see which loops are vectorized.