Re: [dpdk-dev] [PATCH 1/2] ixgbe: add "cold" attribute to setup/teardown fns

[Bruce Richardson <bruce.richardson@intel.com>]

On Fri, Jul 03, 2015 at 09:57:26PM +0200, Thomas Monjalon wrote:
> 2015-07-03 16:56, Bruce Richardson:
> > On Fri, Jul 03, 2015 at 05:45:34PM +0200, Thomas Monjalon wrote:
> > > Hi Bruce,
> > > 
> > > 2015-07-03 16:40, Bruce Richardson:
> > > > As well as the fast-path functions in the rxtx code, there are also
> > > > functions which set up and tear down the descriptor rings. Since these
> > > > are not performance critical functions, there is no need to have them
> > > > extensively optimized, so we add __attribute__((cold)) to their
> > > > definitions. This has the side-effect of making debugging them easier as
> > > > the compiler does not optimize them as heavily, so more variables are
> > > > accessible by default in gdb.
> > > 
> > > What is the benefit, compared to -O0?
> > 
> > First off, it's per function, rather than having to use -O0 globally. Secondly,
> > it doesn't disable optimization, it just tells the compiler that the code is
> > not on the hotpath - whether or not the compiler optimizes it is up to the 
> > compiler itself. From GCC documentation: 
> > https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#Common-Function-Attributes
> > 
> > "The cold attribute on functions is used to inform the compiler that the 
> > function is unlikely to be executed. The function is optimized for size rather 
> > than speed and on many targets it is placed into a special subsection of the 
> > text section so all cold functions appear close together, improving code 
> > locality of non-cold parts of program. The paths leading to calls of cold
> > functions within code are marked as unlikely by the branch prediction mechanism.
> > It is thus useful to mark functions used to handle unlikely conditions, such as
> > perror, as cold to improve optimization of hot functions that do call marked
> > functions in rare occasions."
> 
> I know it may provide some optimization of the hot path.
> I was asking compared to -O0 because you were justifying this change for debug.
> In other words, for debugging, -O0 is probably better. So the reason of this
> change should be the optimization. And it would be interesting to know if you
> have seen some performance improvement.

For some cases, O0 will be necessary, but the advantage of this change is that
for debugging of code that is not in the fast-path, the use of -O0 may be 
unnecessary - which is useful, since you don't always need to do a special debug
build.

As for performance impact: no, I have not seen any performance impact from this
change. Personally, I view this as a low impact change that doesn't really have
any negatives. Is there some concern in particular you have about it? It's really
just providing some extra hints to the compiler.

/Bruce