[PATCH] [ARM] Use AEABI aligned function variants
will.newton at gmail.com
Wed Mar 4 09:17:46 PST 2015
On Wed, Mar 4, 2015 at 4:51 PM, John Brawn <john.brawn at arm.com> wrote:
>> It may be worth noting that with glibc these functions may be no
>> better and in some cases slower than the standard ISO C functions.
> Trunk glibc aliases __aeabi_memcpy4 to __aeabi_memcpy (and similarly
> with the other functions) so it should be no slower as LLVM already
> translates memcpy to __aeabi_memcpy for AEABI targets through the
> generic Libcall lowering mechanism.
My understanding is the AEABI variants disallow using VFP/NEON
registers (for ABI reasons) hence the glibc AEABI memcpy will be
slower than the ISO C memcpy when VFP/NEON is present.
More information about the llvm-commits