On Sun, 19 Nov 2023, Kito Cheng wrote: > Just one minor comment, I think we don't really need to check rv64 or > rv32 for those compiled without any header file test, but I am fine > with that.
We need to set `-march=' differently depending on whether verifying for rv64 or rv32, and then we don't want to limit testing to one of these subarchitectures only, because different code paths are covered in some cases depending on the machine word size. This is especially clear with the cases that succeed for rv64 and currently fail for rv32, but it is also true elsewhere, e.g. where a SImode intermediate SCC result is widened to DImode with rv64, but kept intact with rv64. Maciej
