On 11/11/24 7:12 AM, Robin Dapp wrote:
Hi,
instead of loading the permutation indices and using vmslt in order to
determine which elements belong to which source vector we can compute
the proper mask at compile time. That way we can emit vlm instead of
vle + vmslt.
Regtested on rv64gcv.
Re
Hi,
instead of loading the permutation indices and using vmslt in order to
determine which elements belong to which source vector we can compute
the proper mask at compile time. That way we can emit vlm instead of
vle + vmslt.
Regtested on rv64gcv.
Regards
Robin
gcc/ChangeLog:
* conf
