Standard glibc memcpy() vs. RISC-V Vector extension version on the Allwinner D1 (hoult.org) 3 points by archanox 5y ago ↗ HN
[–] brucehoult 5y ago ↗ That's a pretty nice 3x speedup at 128 byte copy size.Remarkable that the execution time is absolutely identical for anything from 0 to 128 bytes.And the code is 1/24th the size. [–] brucehoult 5y ago ↗ I've reduced the time for lengths from 0 to 64 bytes from 46.8 ns to 30.9 ns without adversely affecting longer lengths. I also knocked 2 bytes off the code size, from 26 to 24 bytes.
[–] brucehoult 5y ago ↗ I've reduced the time for lengths from 0 to 64 bytes from 46.8 ns to 30.9 ns without adversely affecting longer lengths. I also knocked 2 bytes off the code size, from 26 to 24 bytes.
[–] brucehoult 5y ago ↗ I've now done the same thing for strcpy(). It's even more dramatic:http://hoult.org/d1_strcpy.txt
3 comments
[ 4.3 ms ] story [ 12.4 ms ] threadRemarkable that the execution time is absolutely identical for anything from 0 to 128 bytes.
And the code is 1/24th the size.
http://hoult.org/d1_strcpy.txt