Details
-
Improvement
-
Status: Resolved
-
Minor
-
Resolution: Fixed
-
0.9.7
-
None
-
None
Description
Issue raised in thread containing:
http://nagoya.apache.org/eyebrowse/ReadMsg?listName=log4cxx-user@logging.apache.org&msgNo=310
Content-Type: text/plain; charset=US-ASCII; format=flowed
From: Curt Arnold <carnold@apache.org>
Subject: LOG4CXX macros
Date: Thu, 12 Aug 2004 11:14:10 -0500
In GCC, you can use __builtin_expect to hint to the optimizer the
expected value of a expression
(http://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html#Other-Builtins).
So if LOG4CXX_DEBUG evaluated to:
if (__builtin_expect(log.isDebugEnabled(), 0)) {
log.forceLog(Level::DEBUG, msg);
}
Then the compiler would know that it would be relatively unlikely for
the body of the if to be executed and could optimize appropriately. If
it has an impact on the performance, it should make the code run faster
when the debug messages are disabled and slower when they are enabled.
__builtin_expect doesn't actually call anything.
I'd probably implement this by defining a LOG4CXX_UNLIKELY
macro
that would be used in LOG4CXX_DEBUG macro. If the compiler did not
support branch hints, then the definition of LOG4CXX_UNLIKELY would be:
#define LOG4CXX_UNLIKELY x
For GCC it would be:
#define LOG4CXX_UNLIKELY __builtin_expect(x, 0)
On Aug 12, 2004, at 10:04 AM, Christophe de VIENNE wrote:
> Curt Arnold wrote:
>
>> Several compilers including gcc have a mechanism to hint that one
>> branch of an if statement is more likely to be executed. It may be
>> desirable that the LOG4CXX_DEBUG hint that isDebugEnabled will likely
>> return false which would allow better optimization for the normal
>> case. Any comments on the benefits or consequences of providing
>> branch hints?
>
> I'm not sure of what you mean by branch hint. Do you have a small and
> concrete example ?
>
>
> Christophe
>