<div dir="ltr"><div><div>Thanks, Silviu. For reference, Benjamin patched the bug (thanks!) with:<br><a href="http://reviews.llvm.org/rL269757">http://reviews.llvm.org/rL269757</a><br><br></div>I got a bunch of stage 2 ubsan bot fail mails overnight though, so there may be another problem introduced with:<br><a href="http://reviews.llvm.org/rL269728">http://reviews.llvm.org/rL269728</a><br><br></div>I'll try to repro the failure now.<br><br></div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, May 17, 2016 at 5:59 AM, Silviu Baranga <span dir="ltr"><<a href="mailto:Silviu.Baranga@arm.com" target="_blank">Silviu.Baranga@arm.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div link="blue" vlink="purple" lang="EN-GB">
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">Hi Sanjay,<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">Looks like we have another failure related to this:<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><a href="https://llvm.org/bugs/show_bug.cgi?id=27786" target="_blank">https://llvm.org/bugs/show_bug.cgi?id=27786</a><u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">Could you please have a look?<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">Cheers,<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d">Silviu<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1f497d"><u></u> <u></u></span></p>
<div style="border:none;border-left:solid blue 1.5pt;padding:0cm 0cm 0cm 4.0pt">
<div>
<div style="border:none;border-top:solid #b5c4df 1.0pt;padding:3.0pt 0cm 0cm 0cm">
<p class="MsoNormal"><b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif"" lang="EN-US">From:</span></b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif"" lang="EN-US"> llvm-commits [mailto:<a href="mailto:llvm-commits-bounces@lists.llvm.org" target="_blank">llvm-commits-bounces@lists.llvm.org</a>]
<b>On Behalf Of </b>Sanjay Patel via llvm-commits<br>
<b>Sent:</b> 16 May 2016 16:16<br>
<b>To:</b> James Molloy<br>
<b>Cc:</b> llvm-commits<br>
<b>Subject:</b> Re: [llvm] r269426 - [InstCombine] canonicalize* LE/GE vector integer comparisons to LT/GT (PR26701, PR26819)<u></u><u></u></span></p>
</div>
</div><div><div class="h5">
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal">Thanks, James. Taking a look now.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal">On Mon, May 16, 2016 at 8:48 AM, James Molloy <<a href="mailto:james@jamesmolloy.co.uk" target="_blank">james@jamesmolloy.co.uk</a>> wrote:<u></u><u></u></p>
<div>
<p class="MsoNormal">Hi Sanjay,<u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Our internal random builders have found a failing testcase that bisects to this commit. Could you please take a look?<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">To reproduce: opt -instcombine test.ll -disable-output<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal">opt: /work/llvm-test/slave2/continuous/build/llvm/include/llvm/Support/Casting.h:237: typename cast_retty<X, Y *>::ret_type llvm::cast(Y *) [X = llvm::ConstantInt, Y = llvm::Constant]: Assertion `isa<X>(Val) && "cast<Ty>() argument of incompatible
type!"' failed.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">The testcase is available at <a href="http://llvm.org/pr27756" target="_blank">
http://llvm.org/pr27756</a> .<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Cheers,<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">James<u></u><u></u></p>
</div>
</div>
<div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div>
<p class="MsoNormal">On Fri, 13 May 2016 at 16:16 Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>> wrote:<u></u><u></u></p>
</div>
<blockquote style="border:none;border-left:solid #cccccc 1.0pt;padding:0cm 0cm 0cm 6.0pt;margin-left:4.8pt;margin-right:0cm">
<p class="MsoNormal">Author: spatel<br>
Date: Fri May 13 10:10:46 2016<br>
New Revision: 269426<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=269426&view=rev" target="_blank">
http://llvm.org/viewvc/llvm-project?rev=269426&view=rev</a><br>
Log:<br>
[InstCombine] canonicalize* LE/GE vector integer comparisons to LT/GT (PR26701, PR26819)<br>
<br>
*We don't currently handle the edge case constants (min/max values), so it's not a complete<br>
canonicalization.<br>
<br>
To fully solve the motivating bugs, we need to enhance this to recognize a zero vector<br>
too because that's a ConstantAggregateZero which is a ConstantData, not a ConstantVector<br>
or a ConstantDataVector.<br>
<br>
Differential Revision: <a href="http://reviews.llvm.org/D17859" target="_blank">http://reviews.llvm.org/D17859</a><br>
<br>
<br>
<br>
Added:<br>
llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll<br>
Modified:<br>
llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp<br>
llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll<br>
<br>
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=269426&r1=269425&r2=269426&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=269426&r1=269425&r2=269426&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp (original)<br>
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Fri May 13 10:10:46 2016<br>
@@ -3125,11 +3125,64 @@ static ICmpInst *canonicalizeCmpWithCons<br>
assert(!Op1C->isMinValue(true)); // A >=s MIN -> TRUE<br>
return new ICmpInst(ICmpInst::ICMP_SGT, Op0, Builder.getInt(Op1Val - 1));<br>
default:<br>
- break;<br>
+ return nullptr;<br>
}<br>
}<br>
<br>
- // TODO: Handle vectors.<br>
+ // The usual vector types are ConstantDataVector. Exotic vector types are<br>
+ // ConstantVector. They both derive from Constant.<br>
+ if (isa<ConstantDataVector>(Op1) || isa<ConstantVector>(Op1)) {<br>
+ Constant *Op1C = cast<Constant>(Op1);<br>
+ Type *Op1Type = Op1->getType();<br>
+ unsigned NumElts = Op1Type->getVectorNumElements();<br>
+<br>
+ // Set the new comparison predicate and splat a vector of 1 or -1 to<br>
+ // increment or decrement the vector constants. But first, check that no<br>
+ // elements of the constant vector would overflow/underflow when we<br>
+ // increment/decrement the constants.<br>
+ //<br>
+ // TODO? If the edge cases for vectors were guaranteed to be handled as they<br>
+ // are for scalar, we could remove the min/max checks here. However, to do<br>
+ // that, we would have to use insertelement/shufflevector to replace edge<br>
+ // values.<br>
+<br>
+ CmpInst::Predicate NewPred;<br>
+ Constant *OneOrNegOne = nullptr;<br>
+ switch (I.getPredicate()) {<br>
+ case ICmpInst::ICMP_ULE:<br>
+ for (unsigned i = 0; i != NumElts; ++i)<br>
+ if (cast<ConstantInt>(Op1C->getAggregateElement(i))->isMaxValue(false))<br>
+ return nullptr;<br>
+ NewPred = ICmpInst::ICMP_ULT;<br>
+ OneOrNegOne = ConstantInt::get(Op1Type, 1);<br>
+ break;<br>
+ case ICmpInst::ICMP_SLE:<br>
+ for (unsigned i = 0; i != NumElts; ++i)<br>
+ if (cast<ConstantInt>(Op1C->getAggregateElement(i))->isMaxValue(true))<br>
+ return nullptr;<br>
+ NewPred = ICmpInst::ICMP_SLT;<br>
+ OneOrNegOne = ConstantInt::get(Op1Type, 1);<br>
+ break;<br>
+ case ICmpInst::ICMP_UGE:<br>
+ for (unsigned i = 0; i != NumElts; ++i)<br>
+ if (cast<ConstantInt>(Op1C->getAggregateElement(i))->isMinValue(false))<br>
+ return nullptr;<br>
+ NewPred = ICmpInst::ICMP_UGT;<br>
+ OneOrNegOne = ConstantInt::get(Op1Type, -1);<br>
+ break;<br>
+ case ICmpInst::ICMP_SGE:<br>
+ for (unsigned i = 0; i != NumElts; ++i)<br>
+ if (cast<ConstantInt>(Op1C->getAggregateElement(i))->isMinValue(true))<br>
+ return nullptr;<br>
+ NewPred = ICmpInst::ICMP_SGT;<br>
+ OneOrNegOne = ConstantInt::get(Op1Type, -1);<br>
+ break;<br>
+ default:<br>
+ return nullptr;<br>
+ }<br>
+<br>
+ return new ICmpInst(NewPred, Op0, ConstantExpr::getAdd(Op1C, OneOrNegOne));<br>
+ }<br>
<br>
return nullptr;<br>
}<br>
<br>
Added: llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll?rev=269426&view=auto" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll?rev=269426&view=auto</a><br>
==============================================================================<br>
--- llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll (added)<br>
+++ llvm/trunk/test/Transforms/InstCombine/icmp-vec.ll Fri May 13 10:10:46 2016<br>
@@ -0,0 +1,124 @@<br>
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py<br>
+; RUN: opt < %s -instcombine -S | FileCheck %s<br>
+<br>
+; Canonicalize vector ge/le comparisons with constants to gt/lt.<br>
+<br>
+; Normal types are ConstantDataVectors. Test the constant values adjacent to the<br>
+; min/max values that we're not allowed to transform.<br>
+<br>
+define <2 x i1> @sge(<2 x i8> %x) {<br>
+; CHECK-LABEL: @sge(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <2 x i8> %x, <i8 -128, i8 126><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sge <2 x i8> %x, <i8 -127, i8 -129><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @uge(<2 x i8> %x) {<br>
+; CHECK-LABEL: @uge(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <2 x i8> %x, <i8 -2, i8 0><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp uge <2 x i8> %x, <i8 -1, i8 1><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @sle(<2 x i8> %x) {<br>
+; CHECK-LABEL: @sle(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> %x, <i8 127, i8 -127><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sle <2 x i8> %x, <i8 126, i8 128><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @ule(<2 x i8> %x) {<br>
+; CHECK-LABEL: @ule(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> %x, <i8 -1, i8 1><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp ule <2 x i8> %x, <i8 254, i8 0><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+; Weird types are ConstantVectors, not ConstantDataVectors. For an i3 type:<br>
+; Signed min = -4<br>
+; Unsigned min = 0<br>
+; Signed max = 3<br>
+; Unsigned max = 7<br>
+<br>
+define <3 x i1> @sge_weird(<3 x i3> %x) {<br>
+; CHECK-LABEL: @sge_weird(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <3 x i3> %x, <i3 -4, i3 2, i3 -1><br>
+; CHECK-NEXT: ret <3 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sge <3 x i3> %x, <i3 -3, i3 -5, i3 0><br>
+ ret <3 x i1> %cmp<br>
+}<br>
+<br>
+define <3 x i1> @uge_weird(<3 x i3> %x) {<br>
+; CHECK-LABEL: @uge_weird(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <3 x i3> %x, <i3 -2, i3 0, i3 1><br>
+; CHECK-NEXT: ret <3 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp uge <3 x i3> %x, <i3 -1, i3 1, i3 2><br>
+ ret <3 x i1> %cmp<br>
+}<br>
+<br>
+define <3 x i1> @sle_weird(<3 x i3> %x) {<br>
+; CHECK-LABEL: @sle_weird(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt <3 x i3> %x, <i3 3, i3 -3, i3 1><br>
+; CHECK-NEXT: ret <3 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sle <3 x i3> %x, <i3 2, i3 4, i3 0><br>
+ ret <3 x i1> %cmp<br>
+}<br>
+<br>
+define <3 x i1> @ule_weird(<3 x i3> %x) {<br>
+; CHECK-LABEL: @ule_weird(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult <3 x i3> %x, <i3 -1, i3 1, i3 2><br>
+; CHECK-NEXT: ret <3 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp ule <3 x i3> %x, <i3 6, i3 0, i3 1><br>
+ ret <3 x i1> %cmp<br>
+}<br>
+<br>
+; We can't do the transform if any constants are already at the limits.<br>
+<br>
+define <2 x i1> @sge_min(<2 x i3> %x) {<br>
+; CHECK-LABEL: @sge_min(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp sge <2 x i3> %x, <i3 -4, i3 1><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sge <2 x i3> %x, <i3 -4, i3 1><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @uge_min(<2 x i3> %x) {<br>
+; CHECK-LABEL: @uge_min(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp uge <2 x i3> %x, <i3 1, i3 0><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp uge <2 x i3> %x, <i3 1, i3 0><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @sle_max(<2 x i3> %x) {<br>
+; CHECK-LABEL: @sle_max(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp sle <2 x i3> %x, <i3 1, i3 3><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp sle <2 x i3> %x, <i3 1, i3 3><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
+define <2 x i1> @ule_max(<2 x i3> %x) {<br>
+; CHECK-LABEL: @ule_max(<br>
+; CHECK-NEXT: [[CMP:%.*]] = icmp ule <2 x i3> %x, <i3 -1, i3 1><br>
+; CHECK-NEXT: ret <2 x i1> [[CMP]]<br>
+;<br>
+ %cmp = icmp ule <2 x i3> %x, <i3 7, i3 1><br>
+ ret <2 x i1> %cmp<br>
+}<br>
+<br>
<br>
Modified: llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll?rev=269426&r1=269425&r2=269426&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll?rev=269426&r1=269425&r2=269426&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll (original)<br>
+++ llvm/trunk/test/Transforms/LoopVectorize/if-conversion.ll Fri May 13 10:10:46 2016<br>
@@ -73,7 +73,7 @@ for.end:<br>
;CHECK-LABEL: @reduction_func(<br>
;CHECK: load <4 x i32><br>
;CHECK: add <4 x i32><br>
-;CHECK: icmp sle <4 x i32><br>
+;CHECK: icmp slt <4 x i32><br>
;CHECK: select <4 x i1><br>
;CHECK: ret i32<br>
define i32 @reduction_func(i32* nocapture %A, i32 %n) nounwind uwtable readonly ssp {<br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><u></u><u></u></p>
</blockquote>
</div>
</div>
</div>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div></div></div>
</div>
</div>
</blockquote></div><br></div>