[RFC] strcpy optimizations

[OndÅej BÃlka <neleai at seznam dot cz>]

Hi, I have several questions about strcpy.

First question: I need to know if it is safe for example to use pblendvb 
on high 8 bytes while other thread modifies lower 8 bytes?

If I could do that then I could create faster strcpy.
As most of time argument is less than 80 I would use it to write 
page at time.

If not then I would use different strategy.

Second question:
Related is that I looked at strcpy implementations and possible improvement
is handle differently copying 1-15 bytes. I put microbenchmark at

kam.mff.cuni.cz/~ondra/strcpy_tail.tar.bz2
It consists from three variants 
rep.s    - using rep stosb
table.s  - current glibc
branch.s - my improvement.

My benchmark branch is fastest(tested on core2,i7,phenomII,fx10).
As it is also smaller fact that I do test with everything in L1 cache
would only help. Did I miss anything?


Third question is what usage is worth optimizing. When I looked how 
firefox(biggest usage on my computer) most of calls are used to copy 
a english dictonary. I do not know what is used strncpy in font 
configuration which also takes considerable portion, typical usage is 
following:

Breakpoint 1, strncpy (x=0x7fffffff72e2 "primary\tGB2312.1980-0:GL\n",
y=0x7ffff6b49004 "imary\tGB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) 
Continuing.

Breakpoint 1, strncpy (x=0x7fffffff72e3 "rimary\tGB2312.1980-0:GL\n",
y=0x7ffff6b49005 "mary\tGB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) 
Continuing.

Breakpoint 1, strncpy (x=0x7fffffff72e4 "imary\tGB2312.1980-0:GL\n",
y=0x7ffff6b49006 "ary\tGB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) 
Continuing.

Breakpoint 1, strncpy (x=0x7fffffff72e5 "mary\tGB2312.1980-0:GL\n",
y=0x7ffff6b49007 "ry\tGB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) 
Continuing.

Breakpoint 1, strncpy (x=0x7fffffff72e6 "ary\tGB2312.1980-0:GL\n",
y=0x7ffff6b49008 "y\tGB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) 
Continuing.

Breakpoint 1, strncpy (x=0x7fffffff72e0 "primary", y=0x7ffff6b4900a
"GB2312.1980-0:GL\n", n=1) at hooks.c:228
228   START_MEASURE(strcpy);
(gdb) bt
#0  strncpy (x=0x7fffffff72e0 "primary", y=0x7ffff6b4900a
"GB2312.1980-0:GL\n", n=1) at hooks.c:228
#1  0x00007ffff14ab7d4 in ?? () from /usr/lib/libX11.so.6
#2  0x00007ffff14ac52b in ?? () from /usr/lib/libX11.so.6
#3  0x00007ffff14aca7c in _XlcCreateLocaleDataBase () from
/usr/lib/libX11.so.6
#4  0x00007ffff14b0d88 in ?? () from /usr/lib/libX11.so.6
#5  0x00007ffff14b0243 in ?? () from /usr/lib/libX11.so.6
#6  0x00007ffff14b0ac6 in _XlcCreateLC () from /usr/lib/libX11.so.6
#7  0x00007ffff14d2d20 in _XlcUtf8Loader () from /usr/lib/libX11.so.6
#8  0x00007ffff14b84d6 in _XOpenLC () from /usr/lib/libX11.so.6
#9  0x00007ffff14b8618 in _XlcCurrentLC () from /usr/lib/libX11.so.6
#10 0x00007ffff14b8649 in XSupportsLocale () from /usr/lib/libX11.so.6
#11 0x00007fffef0e4306 in ?? () from
/usr/lib/x86_64-linux-gnu/libgdk-x11-2.0.so.0
#12 0x00007fffef0e8ca9 in ?? () from
/usr/lib/x86_64-linux-gnu/libgdk-x11-2.0.so.0
#13 0x00007fffef6975ea in ?? () from
/usr/lib/x86_64-linux-gnu/libgtk-x11-2.0.so.0
#14 0x00007ffff00698b8 in g_option_context_parse () from
/lib/x86_64-linux-gnu/libglib-2.0.so.0
#15 0x00007fffef6978a0 in gtk_parse_args () from
/usr/lib/x86_64-linux-gnu/libgtk-x11-2.0.so.0
#16 0x00007ffff396b54b in XRE_main () from
/usr/lib/xulrunner-1.9.1/libxul.so
#17 0x0000000000402465 in ?? ()
#18 0x00007ffff70daead in __libc_start_main (main=<value optimized out>,
argc=<value optimized out>, ubp_av=<value optimized out>, 
    init=<value optimized out>, fini=<value optimized out>,
rtld_fini=<value optimized out>, stack_end=0x7fffffffdb68) at
libc-start.c:228