lkml.org 
[lkml]   [2016]   [Aug]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH 2/3] x86: support REP MOVSB copy_page()
Microbenchmark shows that "REP MOVSB" copy_page() is faster
than "REP MOVSQ" version on Intel i5-something Haswell
REP_GOOD/ERMS capable CPU.

N=1<<27
rep movsq: 6.758841901 ± 0.04%
rep movsb: 6.253927309 ± 0.02%
-----------------------------------
-7.5%

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

arch/x86/lib/copy_page_64.S | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)

--- a/arch/x86/lib/copy_page_64.S
+++ b/arch/x86/lib/copy_page_64.S
@@ -12,12 +12,21 @@
*/
ALIGN
ENTRY(copy_page)
- ALTERNATIVE "jmp copy_page_regs", "", X86_FEATURE_REP_GOOD
+ ALTERNATIVE_2 "jmp copy_page_regs", \
+ "", X86_FEATURE_REP_GOOD, \
+ "jmp copy_page_rep_movsb", X86_FEATURE_ERMS
+
movl $4096/8, %ecx
rep movsq
ret
ENDPROC(copy_page)

+ENTRY(copy_page_rep_movsb)
+ mov $4096, %ecx
+ rep movsb
+ ret
+ENDPROC(copy_page_rep_movsb)
+
ENTRY(copy_page_regs)
subq $2*8, %rsp
movq %rbx, (%rsp)
\
 
 \ /
  Last update: 2016-09-17 09:57    [W:0.066 / U:0.116 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site