On Thu, Apr 04, 2013 at 04:29:34PM +0200, Marek Polacek wrote:
> On Thu, Apr 04, 2013 at 04:20:36PM +0200, Marek Polacek wrote:
> > On Thu, Apr 04, 2013 at 04:06:35PM +0200, Jakub Jelinek wrote:
> > > In this second hunk the division is done only for flag_guess_branch_prob,
> > > so shouldn't it be
On Thu, Apr 04, 2013 at 04:20:36PM +0200, Marek Polacek wrote:
> On Thu, Apr 04, 2013 at 04:06:35PM +0200, Jakub Jelinek wrote:
> > In this second hunk the division is done only for flag_guess_branch_prob,
> > so shouldn't it be
> > if (flag_guess_branch_prob)
> > {
> > if (PARAM_VALUE
On Thu, Apr 04, 2013 at 04:06:35PM +0200, Jakub Jelinek wrote:
> In this second hunk the division is done only for flag_guess_branch_prob,
> so shouldn't it be
> if (flag_guess_branch_prob)
> {
> if (PARAM_VALUE (HOT_BB_FREQUENCY_FRACTION) == 0
> || edge->frequency <= (CGRAPH_FR
On Thu, Apr 04, 2013 at 04:00:48PM +0200, Marek Polacek wrote:
> --- gcc/predict.c.mp 2013-04-04 15:04:29.925685185 +0200
> +++ gcc/predict.c 2013-04-04 15:04:33.123696281 +0200
> @@ -122,6 +122,8 @@ maybe_hot_frequency_p (struct function *
>if (node->frequency == NODE_FREQUENCY_EXECUTED_O
Another case where we got a SIGFPE if HOT_BB_FREQUENCY_FRACTION
is 0 - in that case we divide by zero.
Regtested/bootstrapped on x86_64-linux, ok for trunk/4.8?
2013-04-04 Marek Polacek
PR tree-optimization/48186
* predict.c (maybe_hot_frequency_p): Return false if
HOT
