Wilco (wilco.dijkstra.delete@this.ntlworld.com) on February 23, 2021 3:48 am wrote:
> Anon (no.delete@this.spam.com) on February 23, 2021 3:05 am wrote:
> > anon2 (anon.delete@this.anon.com) on February 22, 2021 7:17 pm wrote:
> > > I don't think memcpy, memset instructions are bad per se, though I still don't understand
> > > the fascination with them, unless their proponents are going to move on to do-daxpy, route-ip-packet,
> > > gzip-memory, etc instructions when/if one day Intel's rep ; mov finally doesn't suck. But
> > > I digress, the point was not a totally open-ended "ISA does not matter".
> >
> > memcpy is quite common, easy to implement in hardware and very inefficient to implement in software.
>
> You forgot the sarcasm tag :-)
>
> bench-memcpy-random in GLIBC shows just how "efficient" rep movsb (__memcpy_erms) is on my 3700X:
>
>
> "timing_type": "hp_timing",
> "functions": {
> "memcpy": {
> "bench-variant": "random",
> "ifuncs": ["__memcpy_avx_unaligned", "__memcpy_avx_unaligned_erms", "__memcpy_ssse3_back", "__memcpy_ssse3",
> "__memcpy_sse2_unaligned", "__memcpy_sse2_unaligned_erms", "__memcpy_erms"],
> "results": [
> {
> "max-size": 4096,
> "timings": [201553, 200959, 272047, 251099, 164108, 165561, 428758]
> },
> {
> "max-size": 8192,
> "timings": [199125, 198071, 263701, 247578, 159813, 159340, 421392]
> },
> {
> "max-size": 16384,
> "timings": [199538, 196131, 267008, 250593, 161848, 163724, 421873]
> },
> {
> "max-size": 32768,
> "timings": [203553, 194081, 268296, 253730, 169461, 170411, 419616]
> },
> {
> "max-size": 65536,
> "timings": [204701, 200065, 267463, 255599, 163664, 167617, 424114]
> },
> {
> "max-size": 131072,
> "timings": [199415, 205714, 272646, 255835, 165658, 172426, 425134]
> },
> {
> "max-size": 262144,
> "timings": [197246, 203843, 269340, 265621, 159263, 161571, 449099]
> },
> {
> "max-size": 524288,
> "timings": [205213, 211435, 277131, 285266, 170522, 170597, 612529]
> }]
> }
> }
>
>
> Wilco

I don't think Zen 2 supports ERMS.