On Mon, 2006-01-23 at 09:05 -0600, Josh Boyer wrote:
> On Mon, 2006-01-23 at 07:53 -0600, Josh Boyer wrote:
>> > On Mon, 2006-01-23 at 11:30 +0900, Warren Togami wrote:
>
>> > Going off at a tangent....what would people think of the idea of
>> > switching to -Os in $RPM_OPT_FLAGS _after_ FC5 is released?
>>
>> If it results in some form of improvement in performance, I'm all for
>> it.
>
> The "s" in -Os stands for "size", not for "speed". So
if -Os compiled
> code-execution is faster than the default CFLAGS, something must be
> broken elsewhere.
That's not true. Yes, -Os stands for size. However code size _can_
translate to performance increase.
True, but this would mean -Os rsp. RPM_OPT_FLAGS
having missed the
objectives they have been designed for.
But note that I referred to code execution speed, not application
performance.
Think about cache footprint, etc.
Smaller code means more can fit in the cache. It also results in a
smaller DRAM footprint for thinks like shared libraries, etc.
Right, there are
other components interfering but the compiler's code
generation (prelink, shared-libraries, kernel, suboptimal code, cpu
performance, memory performance etc., etc.) on application performance
Identifying and keeping these issues apart, however is difficult.
Ralf