On Mon, Jun 2, 2014 at 11:04 AM, Teresa Johnson <tejohn...@google.com> wrote: > On Mon, Jun 2, 2014 at 10:26 AM, Dehao Chen <de...@google.com> wrote: >> Just tried with Teresa's patch, the ICE in >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61384 is not resolved. > > I will take a look to see why this wasn't fixed by my profile redesign.
Turns out this is a case of garbage-in, garbage-out. Jump threading is producing a block with a frequency >10K, but it is directly due to a block with an invalid frequency after the upstream ccp2 pass. After ccp2: ;; basic block 3, loop depth 1, count 0, freq 9100, maybe hot ;; prev block 2, next block 4, flags: (NEW, REACHABLE) ;; pred: 10 [91.0%] (TRUE_VALUE,EXECUTABLE) ... ;; succ: 5 [50.0%] (TRUE_VALUE,EXECUTABLE) ;; 4 [50.0%] (FALSE_VALUE,EXECUTABLE) ;; basic block 4, loop depth 1, count 0, freq 6825, maybe hot ;; Invalid sum of incoming frequencies 4550, should be 6825 ;; prev block 3, next block 5, flags: (NEW, REACHABLE) ;; pred: 3 [50.0%] (FALSE_VALUE,EXECUTABLE) ;; , discriminator 4 ;; succ: 5 [100.0%] (FALLTHRU,EXECUTABLE) Jump threading introduced BB 13, with predecessors 3 and 4 shown above: ;; basic block 13, loop depth 1, count 0, freq 11375, maybe hot ;; Invalid sum of outgoing probabilities 40.0% ;; prev block 12, next block 1, flags: (NEW, REACHABLE) ;; pred: 4 [100.0%] (FALLTHRU,EXECUTABLE) ;; 3 [50.0%] (TRUE_VALUE,EXECUTABLE) 11375 = 9100/2 + 6825 If bb 4 had the correct frequency of 4550 (9100/2), this block would have gotten the correct count of 9100. BB 13's successor has the correct frequency of 9100. Teresa > > Teresa > >> >> Dehao >> >> On Mon, Jun 2, 2014 at 9:45 AM, Jeff Law <l...@redhat.com> wrote: >>> On 06/02/14 10:17, Dehao Chen wrote: >>>> >>>> We need to rebuild frequency after vrp, otherwise the following code >>>> in tree-ssa-threadupdate.c will make the frequency larger than >>>> upper-bound. >>>> >>>> /* Excessive jump threading may make frequencies large enough >>>> so >>>> the computation overflows. */ >>>> if (rd->dup_blocks[0]->frequency < BB_FREQ_MAX * 2) >>>> rd->dup_blocks[0]->frequency += EDGE_FREQUENCY (e); >>>> >>>> This is referring to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61384 >>> >>> Can you try this with Teresa's revamping of the jump threading frequency >>> updates? It's still in my queue of things to review. >>> >>> Fixing this stuff in the updater would be better than rebuilding the >>> frequencies, IMHO. >>> >>> Jeff >>> > > > > -- > Teresa Johnson | Software Engineer | tejohn...@google.com | 408-460-2413 -- Teresa Johnson | Software Engineer | tejohn...@google.com | 408-460-2413
