<div dir="ltr"><a href="http://www.xml.com/ldd/chapter/book/ch03.html">http://www.xml.com/ldd/chapter/book/ch03.html</a><br>Hope this help.<br><br><div class="gmail_quote">On Wed, Oct 8, 2008 at 9:39 PM, Dominik Bozek <span dir="ltr"><domino@mikroswiat.pl></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Hi all,<br>
<br>
I have done a test of memcpy() and __copy_tofrom_user() on the mpc8313.<br>
And the major conclusion is that __copy_tofrom_user is more efficient<br>
than memcpy. Sometimes about 40%.<br>
<br>
If I good understand, the memcpy() just copy the data, while<br>
__copy_tofrom_user() take care if the memory wasn't swapped out. So then<br>
memcpy() shall be faster than __copy_tofrom_user(). Am I right?<br>
Is here anybody, who can confirm such results and maybe is able to<br>
improve the memcpy()?<br>
<br>
<br>
Let talk about the test.<br>
I have prepared two pieces of memory of size 64KB and I make sure that<br>
this memory is not swapped out (necessary for memcpy() later). Then I<br>
run one of the memory copy function to transfer 32MB and I measure the<br>
time. The memory is copied in chunks from 64KB to 8B. I take care about<br>
the cache calling flush_dcache_range() whenever whole 64KB was used.<br>
I know, that memcpy on the kernel level is not intended to copy memory<br>
blocks in userspace and __copy_tofrom_user is not intended to copy data<br>
only between two user blocks, but for the performance test it doesn't<br>
matter.<br>
Bellow you may see the short piece of code in the kernel module.<br>
<br>
#define TEST_BUF_SIZE (64*1024)<br>
int function;<br>
char *buf1, *buf2, *buf1_bis, *buf2_bis;<br>
unsigned int size, cnt;<br>
<br>
get_user(function, &((TEST_ARG*)(arg))->function);<br>
get_user(buf1, &((TEST_ARG*)(arg))->buf1);<br>
get_user(buf2, &((TEST_ARG*)(arg))->buf2);<br>
get_user(size, &((TEST_ARG*)(arg))->size);<br>
<br>
cnt = (32*1024*1024)/size; /* how many repeats of memory copy is needed<br>
to transfer 32MB ? */<br>
buf1_bis = buf1;<br>
buf2_bis = buf2;<br>
<br>
switch (function)<br>
{<br>
case MEMCPY_TEST:<br>
while (cnt-->0)<br>
{<br>
if (buf1_bis >= buf1+TEST_BUF_SIZE)<br>
{<br>
/* need for flusch data cache as seldom as possible */<br>
buf1_bis = buf1;<br>
buf2_bis = buf2;<br>
flush_dcache_range((int)buf1, (int)(buf2+TEST_BUF_SIZE));<br>
}<br>
if (buf1_bis != memcpy(buf1_bis, buf2_bis, size))<br>
break;<br>
buf1_bis += size;<br>
buf2_bis += size;<br>
}<br>
break;<br>
<br>
case COPY_TOFROM_USER_TEST:<br>
while (cnt-->0)<br>
{<br>
if (buf1_bis >= buf1+TEST_BUF_SIZE)<br>
{<br>
/* need for flusch data cache as seldom as possible */<br>
buf1_bis = buf1;<br>
buf2_bis = buf2;<br>
flush_dcache_range((int)buf1, (int)(buf2+TEST_BUF_SIZE));<br>
}<br>
ret = __copy_tofrom_user(buf1_bis, buf2_bis, size);<br>
if (ret != 0)<br>
break;<br>
buf1_bis += size;<br>
buf2_bis += size;<br>
}<br>
break;<br>
}<br>
<br>
<br>
Bellow are the results:<br>
<br>
memcpy()<br>
chunk: 65536 [B] | transfer: 69.2 [MB/s] | time: 1.849727 [s] |<br>
size: 128.000 [MB]<br>
chunk: 32768 [B] | transfer: 69.2 [MB/s] | time: 1.849700 [s] |<br>
size: 128.000 [MB]<br>
chunk: 16384 [B] | transfer: 69.2 [MB/s] | time: 1.849845 [s] |<br>
size: 128.000 [MB]<br>
chunk: 8192 [B] | transfer: 69.2 [MB/s] | time: 1.850535 [s] |<br>
size: 128.000 [MB]<br>
chunk: 4096 [B] | transfer: 69.1 [MB/s] | time: 1.853405 [s] |<br>
size: 128.000 [MB]<br>
chunk: 2048 [B] | transfer: 69.1 [MB/s] | time: 1.852877 [s] |<br>
size: 128.000 [MB]<br>
chunk: 1024 [B] | transfer: 69.2 [MB/s] | time: 1.849963 [s] |<br>
size: 128.000 [MB]<br>
chunk: 512 [B] | transfer: 69.0 [MB/s] | time: 1.853793 [s] |<br>
size: 128.000 [MB]<br>
chunk: 256 [B] | transfer: 68.6 [MB/s] | time: 1.866222 [s] |<br>
size: 128.000 [MB]<br>
chunk: 128 [B] | transfer: 68.0 [MB/s] | time: 1.883002 [s] |<br>
size: 128.000 [MB]<br>
chunk: 64 [B] | transfer: 67.2 [MB/s] | time: 1.904073 [s] |<br>
size: 128.000 [MB]<br>
chunk: 32 [B] | transfer: 64.7 [MB/s] | time: 1.978109 [s] |<br>
size: 128.000 [MB]<br>
chunk: 16 [B] | transfer: 54.5 [MB/s] | time: 2.348682 [s] |<br>
size: 128.000 [MB]<br>
chunk: 8 [B] | transfer: 47.4 [MB/s] | time: 2.698635 [s] |<br>
size: 128.000 [MB]<br>
<br>
<br>
__copy_tofrom_user()<br>
chunk: 65536 [B] | transfer: 97.3 [MB/s] | time: 1.315155 [s] |<br>
size: 128.000 [MB]<br>
chunk: 32768 [B] | transfer: 97.3 [MB/s] | time: 1.315762 [s] |<br>
size: 128.000 [MB]<br>
chunk: 16384 [B] | transfer: 97.2 [MB/s] | time: 1.316946 [s] |<br>
size: 128.000 [MB]<br>
chunk: 8192 [B] | transfer: 96.8 [MB/s] | time: 1.321705 [s] |<br>
size: 128.000 [MB]<br>
chunk: 4096 [B] | transfer: 96.6 [MB/s] | time: 1.325516 [s] |<br>
size: 128.000 [MB]<br>
chunk: 2048 [B] | transfer: 96.6 [MB/s] | time: 1.325570 [s] |<br>
size: 128.000 [MB]<br>
chunk: 1024 [B] | transfer: 96.8 [MB/s] | time: 1.322599 [s] |<br>
size: 128.000 [MB]<br>
chunk: 512 [B] | transfer: 97.8 [MB/s] | time: 1.308186 [s] |<br>
size: 128.000 [MB]<br>
chunk: 256 [B] | transfer: 100.2 [MB/s] | time: 1.277788 [s] |<br>
size: 128.000 [MB]<br>
chunk: 128 [B] | transfer: 91.5 [MB/s] | time: 1.398216 [s] |<br>
size: 128.000 [MB]<br>
chunk: 64 [B] | transfer: 87.0 [MB/s] | time: 1.471784 [s] |<br>
size: 128.000 [MB]<br>
chunk: 32 [B] | transfer: 75.0 [MB/s] | time: 1.706426 [s] |<br>
size: 128.000 [MB]<br>
chunk: 16 [B] | transfer: 47.8 [MB/s] | time: 2.678039 [s] |<br>
size: 128.000 [MB]<br>
chunk: 8 [B] | transfer: 41.5 [MB/s] | time: 3.084689 [s] |<br>
size: 128.000 [MB]<br>
<br>
Regards<br>
Dominik Bozek<br>
<br>
<br>
BTW. The memcpy() maybe optimized as it is on i32 when the size of block<br>
is known at compile time.<br>
<br>
_______________________________________________<br>
Linuxppc-embedded mailing list<br>
<a href="mailto:Linuxppc-embedded@ozlabs.org">Linuxppc-embedded@ozlabs.org</a><br>
<a href="https://ozlabs.org/mailman/listinfo/linuxppc-embedded" target="_blank">https://ozlabs.org/mailman/listinfo/linuxppc-embedded</a><br>
</blockquote></div><br><br clear="all"><br>-- <br>Best regards,<br>Tuan Duong<br>Mobile: 0983349121<br>
</div>