Gaurav Dhiman wrote:
Well there have been number of mail on usage of __builtin_expect, but can
any one tell how exactly the compiler use this info to optimize the mahchine
code it writes. AFAIK, if .. else statement is converted into some type jmp
statements, depending upon the condition is true or false. By telling the
compiler that we have more probabilty to have this condition true (in case
of likely) or more probabilty to having this condition false (in case of
unlikely), how we are helping the compiler to arrange the corresponding jump
instruction to optimize the execution of code.
It would be great if somebody can tell how compiler internally benifits from
it to arrange the code in optimized manner.
One possibility is to move the body of an if-statement away -- at the end of the
function or outside a loop -- so it rarely goes into the L1 cache):
if (x != 0)
do_foo;
do_bar;
may be translated to:
cmp x, 0
je L1
do_foo
L1:
do_bar
whereas
if (unlikely (x != 0))
do_foo;
do_bar;
may be translated to
cmp x, 0
jne L1
L2:
do_bar
...
ret
L1: do_foo
j L2
And a more complex example here: http://www.gnu.org/software/gcc/news/reorder.html
~velco
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
