d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

amd64: Stop using REP MOVSB for backward memmove()s.

Browse Source 
Enhanced REP MOVSB feature of CPUs starting from Ivy Bridge makes
REP MOVSB the fastest way to copy memory in most of cases. However
Intel Optimization Reference Manual says: "setting the DF to force
REP MOVSB to copy bytes from high towards low addresses will expe-
rience significant performance degradation". Measurements on Intel
Cascade Lake and Alder Lake, same as on AMD Zen3 show that it can
drop throughput to as low as 2.5-3.5GB/s, comparing to ~10-30GB/s
of REP MOVSQ or hand-rolled loop, used for non-ERMS CPUs.

This patch keeps ERMS use for forward ordered memory copies, but
removes it for backward overlapped moves where it does not work.

Reviewed by:	mjg
MFC after:	2 weeks
This commit is contained in:
Alexander Motin 2022-06-16 13:01:12 -04:00
parent f6379f7fde
commit 6210ac95a1
1 changed files with 0 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
sys/amd64/amd64/support.S
View File

@ -507,13 +507,6 @@ END(memcmp)
ALIGN_TEXT
2256:
std
.if \erms == 1
leaq -1(%rdi,%rcx),%rdi
leaq -1(%rsi,%rcx),%rsi
rep
movsb
cld
.else
leaq -8(%rdi,%rcx),%rdi
leaq -8(%rsi,%rcx),%rsi
shrq $3,%rcx
@ -523,7 +516,6 @@ END(memcmp)
movq %rdx,%rcx
andb $7,%cl
jne 2004b
.endif
\end
ret
.endif