>
>>
>> > +
>> > + default_edge->count = default_count;
>> > + if (count)
>> > + {
>> > + edge e;
>> > + edge_iterator ei;
>> > + FOR_EACH_EDGE (e, ei, stmt_bb->succs)
>> > + e->probability = e->count * REG_BR_PROB_BASE / count;
>> > + }
>>
>> Hmm, this updates origina BB containing the switch statement?
> The out_edges of the original BB.
>> Of course,
>> modulo roundoff errors, this should hold. I wonder where did you got the
>> diferences and why do you need this?
The count of the outgoing edge of BB that jumps to the default label
needs to be updated (to 0 or original_default_count/2, depending on
whether default label is a jump table target). So I need to
redistribute the probabilities of the rest of the edges. That's why I
do this here.
> Originally, I obtained e->probability as e->count * REG_BR_PROB_BASE /
> bb->count. During profile bootstrap, I noticed BBs whose sum of
> outgoing edge counts exceeded the BB's count and hence I normalized
> with the sum of edge counts.
>
>> You are going to output new control flow
>> and find_many_sub_basic_blocks will recompute all the counts/frequencies
>> inside
>> anyway?
> Sorry, I am lost here. I am just updating the probability of
> stmt_bb->succs right?
>
>
