<div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, Dec 27, 2013 at 11:44 AM, Kostya Serebryany <span dir="ltr"><<a href="mailto:kcc@google.com" target="_blank">kcc@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">
<div class="im">On Fri, Dec 27, 2013 at 11:39 AM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@gmail.com" target="_blank">chandlerc@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><br><div class="gmail_quote">
<div>On Fri, Dec 27, 2013 at 2:10 AM, Kostya Serebryany <span dir="ltr"><<a href="mailto:kcc@google.com" target="_blank">kcc@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div style="overflow:hidden">+ // This function may be called only a small fixed amount of times per each<br>
+ // invocation, otherwise we do actually have a leak which we want to report.<br>
+ // If this function is called more than kGraveYardMaxSize times, the pointers<br>
+ // will not be properly buried and a leak detector will report a leak, which<br>
+ // is what we want in such case.<br></div></blockquote><div><br></div></div><div>Interesting. I didn't realize it was going to be *that* tightly bounded.</div><div><br></div><div>This makes me wonder if the whole disable free thing should just be removed at this point. </div>
</div></div></div></blockquote></div><div>I am all for it, --disable-free looks like a hack and we did not see any compile-time loss from removing it (measure on bootstrap).</div></div></div></div></blockquote><div><br></div>
<div>I take these my words back. Building 483.xalancbmk with -O0 (make -j32) shows around 0.5%-1% compile-time difference:</div><div>with -disable-free :</div><div><div>TIME: real: 8.140; user: 177.970; system: 13.180</div>
<div>TIME: real: 8.070; user: 177.940; system: 13.090</div><div>TIME: real: 8.103; user: 177.960; system: 13.410</div><div>TIME: real: 8.122; user: 178.110; system: 13.180</div><div>TIME: real: 8.093; user: 177.660; system: 13.210</div>
</div><div>without -disable-free :<br></div><div><div>TIME: real: 8.140; user: 179.750; system: 13.260</div><div>TIME: real: 8.113; user: 178.770; system: 13.710</div><div>TIME: real: 8.101; user: 179.830; system: 13.180</div>
<div>TIME: real: 8.262; user: 180.160; system: 12.760</div><div>TIME: real: 8.125; user: 179.510; system: 13.070</div></div><div><br></div><div>0.5% may be worth this relatively innocent hack. </div><div><br></div><div> </div>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">
<div class="im"><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>Back in the day, we didn't so carefully use a BumpPtrAllocator in the ASTContext. Today, we might be fine to call free 10 times. </div></div></div>
</div></blockquote><div><br></div></div><div>But this is not 10 calls to free. This is 10-ish calls to delete, where the DTOR destructs other objects and so on. It may call much more than 10 frees.</div><div>Still, see above. </div>
<div class="im">
<div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">
<div>Anyways, not trying to shift the goal posts; I mostly wanted to mention it in case someone gets some time to look into removing all of the leak stuff.</div><div>
<div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div style="overflow:hidden">
+ static const size_t kGraveYardMaxSize = 16;<br>
+ static const void *GraveYard[kGraveYardMaxSize];<br>
+ static llvm::sys::cas_flag GraveYardSize;</div></blockquote><div> </div></div></div>Do these being function-local statics defeat the purpose of using a basic atomic increment? I can't recall if we correctly eliminate the cxa_guard in these cases...</div>
</div></blockquote></div><div>We will not have cxa_guard here because there is no dynamic init -- both objects (the size and the array) are linker-initialized.</div><div>Atomic increment is here for the future case when clang becomes multi-threaded. </div>
</div><br></div></div>
</blockquote></div><br></div></div>