<div dir="ltr">Here is my data from the last night's run of SPEC ref (the smaller the better): <div> <br><div><div><span style="font-family:'courier new',monospace"> w/o weights, with weights, diff</span><br>
</div><div><span style="font-family:'courier new',monospace"> 400.perlbench, 1449.00, 1440.00, 0.99</span><span style="font-family:'courier new',monospace"><br></span></div><div><font face="courier new, monospace"> 401.bzip2, 905.00, 834.00, 0.92</font></div>
<div><font face="courier new, monospace"> 403.gcc, 743.00, 742.00, 1.00</font></div><div><font face="courier new, monospace"> 429.mcf, 315.00, 338.00, 1.07</font></div>
<div><font face="courier new, monospace"> 445.gobmk, 973.00, 882.00, 0.91</font></div><div><font face="courier new, monospace"> 456.hmmer, 1067.00, 1007.00, 0.94</font></div>
<div><font face="courier new, monospace"> 458.sjeng, 1004.00, 1044.00, 1.04</font></div><div><font face="courier new, monospace"> 462.libquantum, 344.00, 398.00, 1.16</font></div>
<div><font face="courier new, monospace"> 464.h264ref, 1485.00, 1277.00, 0.86</font></div><div><font face="courier new, monospace"> 471.omnetpp, 580.00, 609.00, 1.05</font></div>
<div><font face="courier new, monospace"> 473.astar, 575.00, 646.00, 1.12</font></div><div><font face="courier new, monospace"> 483.xalancbmk, 538.00, 510.00, 0.95</font></div>
<div><font face="courier new, monospace"> 433.milc, 650.00, 585.00, 0.90</font></div><div><font face="courier new, monospace"> 444.namd, 592.00, 601.00, 1.02</font></div>
<div><font face="courier new, monospace"> 447.dealII, 625.00, 640.00, 1.02</font></div><div><font face="courier new, monospace"> 450.soplex, 333.00, 348.00, 1.05</font></div>
<div><font face="courier new, monospace"> 453.povray, 493.00, 481.00, 0.98</font></div><div><font face="courier new, monospace"> 470.lbm, 370.00, 380.00, 1.03</font></div>
<div><font face="courier new, monospace"> 482.sphinx3, 868.00, 850.00, 0.98</font></div></div><div><br></div></div><div>It's kind of mixed.... </div><div>I've just migrated to a new machine so I don't know if the results are stable here. </div>
<div>Let me ponder on this change for a few days (after the weekend)... </div><div><br></div><div>--kcc </div><div><br></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, Aug 29, 2014 at 9:06 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:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">I am actually surprised too :) <div>We've used branch weights in a couple of other places but not here. </div>
<div>One can easily create a benchmark where these weights will lead to a performance loss,</div>
<div>but the average SPEC numbers clearly show the overal benefit. </div><div>Let me submit this patch once I confirm the SPEC/ref numbers (the SPEC/test numbers I saw yesterday are very nice)</div><div><br></div><div>
--kcc </div>
</div><div class="HOEnZb"><div class="h5"><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, Aug 29, 2014 at 1:57 AM, Chandler Carruth <span dir="ltr"><<a href="mailto:chandlerc@google.com" target="_blank">chandlerc@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div>Adding ASan folks...</div><div><br></div>Didn't we already look at this Kostya? A bit surprised this wasn't already the case.</div>
<div><div><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, Aug 29, 2014 at 1:24 AM, Jonas Wagner <span dir="ltr"><<a href="mailto:jonas.wagner@epfl.ch" target="_blank">jonas.wagner@epfl.ch</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Experiments on the SPEC benchmarks show that the slow path is rarely<br>
taken, and that branch weights reduce branch misses and overall runtime.<br>
---<br>
lib/Transforms/Instrumentation/AddressSanitizer.cpp | 5 ++++-<br>
1 file changed, 4 insertions(+), 1 deletion(-)<br>
<br>
diff --git a/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/lib/Transforms/Instrumentation/AddressSanitizer.cpp<br>
index 357b5d5..3ff0a61 100644<br>
--- a/lib/Transforms/Instrumentation/AddressSanitizer.cpp<br>
+++ b/lib/Transforms/Instrumentation/AddressSanitizer.cpp<br>
@@ -870,8 +870,11 @@ void AddressSanitizer::instrumentAddress(Instruction *OrigIns,<br>
TerminatorInst *CrashTerm = nullptr;<br>
<br>
if (ClAlwaysSlowPath || (TypeSize < 8 * Granularity)) {<br>
+ // We use branch weights for the slow path check, to indicate that the slow<br>
+ // path is rarely taken. This seems to be the case for SPEC benchmarks.<br>
TerminatorInst *CheckTerm =<br>
- SplitBlockAndInsertIfThen(Cmp, InsertBefore, false);<br>
+ SplitBlockAndInsertIfThen(Cmp, InsertBefore, false,<br>
+ MDBuilder(*C).createBranchWeights(1, 100000));<br>
assert(dyn_cast<BranchInst>(CheckTerm)->isUnconditional());<br>
BasicBlock *NextBB = CheckTerm->getSuccessor(0);<br>
IRB.SetInsertPoint(CheckTerm);<br>
<span><font color="#888888">--<br>
2.1.0<br>
</font></span><div><div><br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu" target="_blank">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
</div></div></blockquote></div><br></div>
</div></div></blockquote></div><br></div>
</div></div></blockquote></div><br></div>