<div dir="ltr">That's what the code comment would have us believe, but... :)<br><br>define i8* @isa_impl_wrap(i8** %x) {<br> %t2 = alloca i8*<br> %t4 = load i8*, i8** %x, !nonnull !0<br> store i8* %t4, i8** %t2<br> %t6 = load i8*, i8** %t2<br> ret i8* %t6<br>}<br>!0 = !{}<br><br>$ ./opt -instcombine cse_nonnull.ll -S<br>...<br>define i8* @isa_impl_wrap(i8** %x) {<br> %t4 = load i8*, i8** %x, align 8, !nonnull !0 <--- nonnull preserved when earlier instruction has it<br> ret i8* %t4<br>}<br><br></div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Jan 3, 2017 at 8:53 AM, Davide Italiano <span dir="ltr"><<a href="mailto:davide@freebsd.org" target="_blank">davide@freebsd.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On Tue, Jan 3, 2017 at 7:21 AM, Sanjay Patel <<a href="mailto:spatel@rotateright.com">spatel@rotateright.com</a>> wrote:<br>
><br>
><br>
> On Tue, Jan 3, 2017 at 4:29 AM, Davide Italiano <<a href="mailto:davide@freebsd.org">davide@freebsd.org</a>> wrote:<br>
>><br>
>> On Mon, Jan 2, 2017 at 3:25 PM, Sanjay Patel via llvm-commits<br>
>> <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a>> wrote:<br>
>> > Author: spatel<br>
>> > Date: Mon Jan 2 17:25:28 2017<br>
>> > New Revision: 290844<br>
>> ><br>
>> > URL: <a href="http://llvm.org/viewvc/llvm-project?rev=290844&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project?rev=290844&view=rev</a><br>
>> > Log:<br>
>> > [InstCombine] use combineMetadataForCSE instead of copying it; NFCI<br>
>> ><br>
>> > Modified:<br>
>> > llvm/trunk/lib/Transforms/<wbr>InstCombine/<wbr>InstCombineLoadStoreAlloca.cpp<br>
>> ><br>
>> > Modified:<br>
>> > llvm/trunk/lib/Transforms/<wbr>InstCombine/<wbr>InstCombineLoadStoreAlloca.cpp<br>
>> > URL:<br>
>> > <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp?rev=290844&r1=290843&r2=290844&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project/llvm/trunk/lib/<wbr>Transforms/InstCombine/<wbr>InstCombineLoadStoreAlloca.<wbr>cpp?rev=290844&r1=290843&r2=<wbr>290844&view=diff</a><br>
>> ><br>
>> > ==============================<wbr>==============================<wbr>==================<br>
>> > --- llvm/trunk/lib/Transforms/<wbr>InstCombine/<wbr>InstCombineLoadStoreAlloca.cpp<br>
>> > (original)<br>
>> > +++ llvm/trunk/lib/Transforms/<wbr>InstCombine/<wbr>InstCombineLoadStoreAlloca.cpp<br>
>> > Mon Jan 2 17:25:28 2017<br>
>> > @@ -850,20 +850,10 @@ Instruction *InstCombiner::visitLoadInst<br>
>> > // separated by a few arithmetic operations.<br>
>> > BasicBlock::iterator BBI(LI);<br>
>> > bool IsLoadCSE = false;<br>
>> > - if (Value *AvailableVal =<br>
>> > - FindAvailableLoadedValue(&LI, LI.getParent(), BBI,<br>
>> > - DefMaxInstsToScan, AA, &IsLoadCSE)) {<br>
>> > - if (IsLoadCSE) {<br>
>> > - LoadInst *NLI = cast<LoadInst>(AvailableVal);<br>
>> > - unsigned KnownIDs[] = {<br>
>> > - LLVMContext::MD_tbaa, LLVMContext::MD_alias_scope,<br>
>> > - LLVMContext::MD_noalias, LLVMContext::MD_range,<br>
>> > - LLVMContext::MD_invariant_<wbr>load, LLVMContext::MD_nonnull,<br>
>> > - LLVMContext::MD_invariant_<wbr>group, LLVMContext::MD_align,<br>
>> > - LLVMContext::MD_<wbr>dereferenceable,<br>
>> > - LLVMContext::MD_<wbr>dereferenceable_or_null};<br>
>> > - combineMetadata(NLI, &LI, KnownIDs);<br>
>> > - };<br>
>> > + if (Value *AvailableVal = FindAvailableLoadedValue(<br>
>> > + &LI, LI.getParent(), BBI, DefMaxInstsToScan, AA, &IsLoadCSE))<br>
>> > {<br>
>> > + if (IsLoadCSE)<br>
>> > + combineMetadataForCSE(cast<<wbr>LoadInst>(AvailableVal), &LI);<br>
>> ><br>
>><br>
>> We have still a lot of places where we have fundamentally the same'ish<br>
>> version of this code in the tree<br>
>> (GVN/GVNHoist/SimplifyCFG/<wbr>BBVectorize), is it possible to convert to<br>
>> `combineMetadataForCSE` (or there's some caveat here I'm missing)?<br>
><br>
><br>
><br>
> Add VectorUtils.cpp -> llvm::propagateMetadata() to the list; that's used by<br>
> the Loop and SLP vectorizers. I've never looked at this closely (I'm here in<br>
> the context of PR28430), so I'm not sure what, if any, differences these<br>
> callers expect. It's also seems like combineMetadataForCSE is not behaving<br>
> ideally. For example, I got here because:<br>
><br>
> define i8* @isa_impl_wrap(i8** %x) {<br>
> %t2 = alloca i8*<br>
> %t4 = load i8*, i8** %x<br>
> store i8* %t4, i8** %t2<br>
> %t6 = load i8*, i8** %t2, !nonnull !0 <--- we should preserve that to<br>
> optimize dyn_cast<br>
> ret i8* %t6<br>
> }<br>
> !0 = !{}<br>
><br>
> $ ./opt -instcombine cse_nonnull.ll -S<br>
> ...<br>
> define i8* @isa_impl_wrap(i8** %x) {<br>
> %t4 = load i8*, i8** %x, align 8 <--- but it's gone now<br>
> ret i8* %t4<br>
> }<br>
><br>
<br>
</div></div>Isn't this because we're combining %t4 and %t6 and we can set<br>
`!nonnull` if both instructions have the !nonnull metadata?<br>
<br>
>From `llvm::combineMetadata()`:<br>
case LLVMContext::MD_nonnull:<br>
// Only set the !nonnull if it is present in both instructions.<br>
K->setMetadata(Kind, JMD);<br>
<div class="HOEnZb"><div class="h5"><br>
--<br>
Davide<br>
<br>
"There are no solved problems; there are only problems that are more<br>
or less solved" -- Henri Poincare<br>
</div></div></blockquote></div><br></div>