<div dir="ltr">OK, I think I understand your suggestion and it sounds good. Are you sending the patch?</div><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Oct 24, 2014 at 9:16 AM, Arnold Schwaighofer <span dir="ltr"><<a href="mailto:aschwaighofer@apple.com" target="_blank">aschwaighofer@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">It is complicated :).<br>
<br>
clang turns on/off vectorization through explicitly setting the PassManager::LoopVectorize/SLPVectorize fields. This overrides the cl:opt flags. Changing the cl::opt default won’t have any effect on clang.<br>
<br>
Changing the default will only effect LTO because libLTO does not explicitly set the PassManager::LoopVectorize flag but that value gets initialized by the cl::opt value.<br>
<br>
I am suggesting changing to:<br>
<br>
<br>
static cl::opt<bool><br>
RunLoopVectorization("vectorize-loops", cl::Hidden, cl::init(true),<br>
<span class=""> cl::desc("Run the Loop vectorization passes"));<br>
<br>
</span>Similar for the slp flag.<br>
<br>
We will have to force disabling loop unrolling in the loop vectorizer so that we get the behavior before your patch:<br>
<br>
PM.add(createLoopVectorizePass(true, LoopVectorize));<br>
<div class="HOEnZb"><div class="h5"><br>
<br>
<br>
> On Oct 24, 2014, at 9:03 AM, JF Bastien <<a href="mailto:jfb@google.com">jfb@google.com</a>> wrote:<br>
><br>
> Yes, cl::opt<bool> defaults to false when there's no init. Changing the default will affect non-LTO too. This may be the right thing to do, but isn't my call. Turning it on only for LTO sounds better IMO, but I'm not sure what you're suggesting: I think there shouldn't be a vectorization flags that are different for LTO and for non-LTO.<br>
><br>
> On Fri, Oct 24, 2014 at 8:57 AM, Arnold Schwaighofer <<a href="mailto:aschwaighofer@apple.com">aschwaighofer@apple.com</a>> wrote:<br>
> JF are you sure that “LoopVectorize” is set to true by default by the PassManager instance of libLTO?<br>
><br>
> The reason why I forced these parameters to true is that this is not the case if I remember correctly.<br>
><br>
> We wanted the default for libLTO to be with vectorization.<br>
><br>
> PassManager.cpp:<br>
><br>
> static cl::opt<bool><br>
> RunLoopVectorization("vectorize-loops", cl::Hidden,<br>
> cl::desc("Run the Loop vectorization passes"));<br>
><br>
> PassManagerBuilder::PassManagerBuilder() {<br>
> OptLevel = 2;<br>
> SizeLevel = 0;<br>
> LibraryInfo = nullptr;<br>
> Inliner = nullptr;<br>
> DisableTailCalls = false;<br>
> DisableUnitAtATime = false;<br>
> DisableUnrollLoops = false;<br>
> BBVectorize = RunBBVectorization;<br>
> SLPVectorize = RunSLPVectorization;<br>
> LoopVectorize = RunLoopVectorization;<br>
> RerollLoops = RunLoopRerolling;<br>
> LoadCombine = RunLoadCombine;<br>
> DisableGVNLoadPRE = false;<br>
> VerifyInput = false;<br>
> VerifyOutput = false;<br>
> StripDebug = false;<br>
> MergeFunctions = false;<br>
> }<br>
><br>
> LTOCodeGenerator.cpp:<br>
><br>
> /// Optimize merged modules using various IPO passes<br>
> bool LTOCodeGenerator::generateObjectFile(raw_ostream &out,<br>
> bool DisableOpt,<br>
> bool DisableInline,<br>
> bool DisableGVNLoadPRE,<br>
> std::string &errMsg) {<br>
> if (!this->determineTarget(errMsg))<br>
> return false;<br>
><br>
> Module *mergedModule = IRLinker.getModule();<br>
><br>
> // Mark which symbols can not be internalized<br>
> this->applyScopeRestrictions();<br>
><br>
> // Instantiate the pass manager to organize the passes.<br>
> PassManager passes;<br>
><br>
> // Add an appropriate DataLayout instance for this module...<br>
> mergedModule->setDataLayout(TargetMach->getSubtargetImpl()->getDataLayout());<br>
><br>
> Triple TargetTriple(TargetMach->getTargetTriple());<br>
> PassManagerBuilder PMB;<br>
> PMB.DisableGVNLoadPRE = DisableGVNLoadPRE;<br>
> if (!DisableInline)<br>
> PMB.Inliner = createFunctionInliningPass();<br>
> PMB.LibraryInfo = new TargetLibraryInfo(TargetTriple);<br>
> if (DisableOpt)<br>
> PMB.OptLevel = 0;<br>
> PMB.VerifyInput = true;<br>
> PMB.VerifyOutput = true;<br>
><br>
> PMB.populateLTOPassManager(passes, TargetMach);<br>
><br>
><br>
><br>
> I think cl::opt<bool> defaults to false and your commit effectively disabled vectorization during LTO. We can recover this by changing the default cl::opt flags 'vectorize-loops’ and 'vectorize-slp' to true. If that does not work (because we make assumption somewhere about the default being false) we can follow the example of “DisableGVNLoadPRE” in LTOCodeGenerator.cpp and a a flag to disable Vectorization during LTO and pass that to the PassManager created in generateObjectFile.<br>
><br>
> PMB.LoopVectorize = !DisableLTOVectorization;<br>
><br>
><br>
> Thanks,<br>
> Arnold<br>
><br>
> > DisableUnrollLoops<br>
> > On Oct 24, 2014, at 5:27 AM, Alexey Volkov <<a href="mailto:avolkov.intel@gmail.com">avolkov.intel@gmail.com</a>> wrote:<br>
> ><br>
> > Hi JF,<br>
> ><br>
> > After your commit I saw a performance regression because of disabled Loop Vectorizer:<br>
> > LV: Not vectorizing: No #pragma vectorize enable.<br>
> > It is really strange since I used -Ofast -flto clang's options to build an application.<br>
> > Before this change loop was successfully vectorized by Loop Vectorizer.<br>
> ><br>
> > Thanks, Alexey.<br>
> ><br>
> > 2014-10-22 3:18 GMT+04:00 JF Bastien <<a href="mailto:jfb@google.com">jfb@google.com</a>>:<br>
> > Author: jfb<br>
> > Date: Tue Oct 21 18:18:21 2014<br>
> > New Revision: 220345<br>
> ><br>
> > URL: <a href="http://llvm.org/viewvc/llvm-project?rev=220345&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=220345&view=rev</a><br>
> > Log:<br>
> > LTO: respect command-line options that disable vectorization.<br>
> ><br>
> > Summary: Patches 202051 and 208013 added calls to LTO's PassManager which unconditionally add LoopVectorizePass and SLPVectorizerPass instead of following the logic in PassManagerBuilder::populateModulePassManager and honoring the -vectorize-loops -run-slp-after-loop-vectorization flags.<br>
> ><br>
> > Reviewers: nadav, aschwaighofer, yijiang<br>
> ><br>
> > Subscribers: llvm-commits<br>
> ><br>
> > Differential Revision: <a href="http://reviews.llvm.org/D5884" target="_blank">http://reviews.llvm.org/D5884</a><br>
> ><br>
> > Modified:<br>
> > llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp<br>
> ><br>
> > Modified: llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp<br>
> > URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=220345&r1=220344&r2=220345&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=220345&r1=220344&r2=220345&view=diff</a><br>
> > ==============================================================================<br>
> > --- llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp (original)<br>
> > +++ llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp Tue Oct 21 18:18:21 2014<br>
> > @@ -440,10 +440,12 @@ void PassManagerBuilder::addLTOOptimizat<br>
> > // More loops are countable; try to optimize them.<br>
> > PM.add(createIndVarSimplifyPass());<br>
> > PM.add(createLoopDeletionPass());<br>
> > - PM.add(createLoopVectorizePass(true, true));<br>
> > + PM.add(createLoopVectorizePass(DisableUnrollLoops, LoopVectorize));<br>
> ><br>
> > // More scalar chains could be vectorized due to more alias information<br>
> > - PM.add(createSLPVectorizerPass()); // Vectorize parallel scalar chains.<br>
> > + if (RunSLPAfterLoopVectorization)<br>
> > + if (SLPVectorize)<br>
> > + PM.add(createSLPVectorizerPass()); // Vectorize parallel scalar chains.<br>
> ><br>
> > // After vectorization, assume intrinsics may tell us more about pointer<br>
> > // alignments.<br>
> ><br>
> ><br>
> > _______________________________________________<br>
> > llvm-commits mailing list<br>
> > <a href="mailto:llvm-commits@cs.uiuc.edu">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>
> ><br>
> ><br>
> ><br>
> > --<br>
> > Alexey Volkov<br>
> > Intel Corporation<br>
> > _______________________________________________<br>
> > llvm-commits mailing list<br>
> > <a href="mailto:llvm-commits@cs.uiuc.edu">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>
><br>
><br>
<br>
</div></div></blockquote></div><br></div>