<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta name="Generator" content="Microsoft Word 15 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0in;
margin-bottom:.0001pt;
font-size:12.0pt;
font-family:"Times New Roman",serif;}
a:link, span.MsoHyperlink
{mso-style-priority:99;
color:blue;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:purple;
text-decoration:underline;}
span.EmailStyle17
{mso-style-type:personal-reply;
font-family:"Calibri",sans-serif;
color:#1F497D;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri",sans-serif;}
@page WordSection1
{size:8.5in 11.0in;
margin:56.7pt 42.5pt 56.7pt 85.05pt;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]-->
</head>
<body lang="EN-US" link="blue" vlink="purple">
<div class="WordSection1">
<p class="MsoNormal"><a name="_MailEndCompose"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D">Hi Chandler,<o:p></o:p></span></a></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D"><o:p> </o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D">Thank you for pointing to this. We have reverted the patch for now (r306815). Hopefully, we will be able to re-commit it next week.<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D"><o:p> </o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#525252">-Nikolai<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D"><o:p> </o:p></span></p>
<p class="MsoNormal"><b><span style="font-size:11.0pt;font-family:"Calibri",sans-serif">From:</span></b><span style="font-size:11.0pt;font-family:"Calibri",sans-serif"> Chandler Carruth [mailto:chandlerc@google.com]
<br>
<b>Sent:</b> Friday, June 30, 2017 10:46 AM<br>
<b>To:</b> Bozhenov, Nikolai <nikolai.bozhenov@intel.com>; llvm-commits@lists.llvm.org<br>
<b>Subject:</b> Re: [llvm] r306525 - [InstCombine] Canonicalize clamp of float types to minmax in fast mode.<o:p></o:p></span></p>
<p class="MsoNormal"><o:p> </o:p></p>
<div>
<p class="MsoNormal">This introduces a read of an uninitialized value. It should show up with msan, but even shows up with UBSan because you end up with a crazy value for an enum:<o:p></o:p></p>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<div>
<p class="MsoNormal">% ./asan/bin/llvm-lit -sv ../test/Transforms/InstCombine<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">FAIL: LLVM :: Transforms/InstCombine/clamp-to-minmax.ll (292 of 708)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">******************** TEST 'LLVM :: Transforms/InstCombine/clamp-to-minmax.ll' FAILED ********************<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">Script:<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">--<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">/home/chandlerc/src/llvm.git/build/asan/./bin/opt < /home/chandlerc/src/llvm.git/test/Transforms/InstCombine/clamp-to-minmax.ll -instcombine -S | /home/chandlerc/src/llvm.git/build/asan/./bin/FileCheck /home/chandlerc/src/llvm.git/test/Transforms/InstCombine/clamp-to-minmax.ll<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">--<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">Exit Code: 2<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal">Command Output (stderr):<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">--<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">../../lib/Transforms/InstCombine/InstCombineSelect.cpp:1400:13: runtime error: load of value 2679211552, which is not a valid value for type 'Instruction::CastOps'<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior ../../lib/Transforms/InstCombine/InstCombineSelect.cpp:1400:13 in<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">FileCheck error: '-' is empty.<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">FileCheck command line: /home/chandlerc/src/llvm.git/build/asan/./bin/FileCheck /home/chandlerc/src/llvm.git/test/Transforms/InstCombine/clamp-to-minmax.ll<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal">--<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal">********************<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">Testing Time: 8.91s<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">********************<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal">Failing Tests (1):<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal"> LLVM :: Transforms/InstCombine/clamp-to-minmax.ll<o:p></o:p></p>
</div>
</div>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal">Please fix or revert.<o:p></o:p></p>
</div>
</div>
<p class="MsoNormal"><o:p> </o:p></p>
<div>
<div>
<p class="MsoNormal">On Wed, Jun 28, 2017 at 2:26 AM Nikolai Bozhenov via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a>> wrote:<o:p></o:p></p>
</div>
<blockquote style="border:none;border-left:solid #CCCCCC 1.0pt;padding:0in 0in 0in 6.0pt;margin-left:4.8pt;margin-right:0in">
<p class="MsoNormal">Author: n.bozhenov<br>
Date: Wed Jun 28 02:26:20 2017<br>
New Revision: 306525<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=306525&view=rev" target="_blank">
http://llvm.org/viewvc/llvm-project?rev=306525&view=rev</a><br>
Log:<br>
[InstCombine] Canonicalize clamp of float types to minmax in fast mode.<br>
<br>
Summary:<br>
This commit allows matchSelectPattern to recognize clamp of float<br>
arguments in the presence of FMF the same way as already done for<br>
integers.<br>
<br>
This case is a little different though. With integers, given the<br>
min/max pattern is recognized, DAGBuilder starts selecting MIN/MAX<br>
"automatically". That is not the case for float, because for them only<br>
full FMINNAN/FMINNUM/FMAXNAN/FMAXNUM ISD nodes exist and they do care<br>
about NaNs. On the other hand, some backends (e.g. X86) have only<br>
FMIN/FMAX nodes that do not care about NaNS and the former NAN/NUM<br>
nodes are illegal thus selection is not happening. So I decided to do<br>
such kind of transformation in IR (InstCombiner) instead of<br>
complicating the logic in the backend.<br>
<br>
Reviewers: spatel, jmolloy, majnemer, efriedma, craig.topper<br>
<br>
Reviewed By: efriedma<br>
<br>
Subscribers: hiraditya, javed.absar, n.bozhenov, llvm-commits<br>
<br>
Patch by Andrei Elovikov <<a href="mailto:andrei.elovikov@intel.com" target="_blank">andrei.elovikov@intel.com</a>><br>
<br>
Differential Revision: <a href="https://reviews.llvm.org/D33186" target="_blank">
https://reviews.llvm.org/D33186</a><br>
<br>
Modified:<br>
llvm/trunk/include/llvm/IR/PatternMatch.h<br>
llvm/trunk/lib/Analysis/ValueTracking.cpp<br>
llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp<br>
llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll<br>
<br>
Modified: llvm/trunk/include/llvm/IR/PatternMatch.h<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/PatternMatch.h?rev=306525&r1=306524&r2=306525&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/PatternMatch.h?rev=306525&r1=306524&r2=306525&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/include/llvm/IR/PatternMatch.h (original)<br>
+++ llvm/trunk/include/llvm/IR/PatternMatch.h Wed Jun 28 02:26:20 2017<br>
@@ -195,11 +195,35 @@ struct apint_match {<br>
return false;<br>
}<br>
};<br>
+// Either constexpr if or renaming ConstantFP::getValueAPF to<br>
+// ConstantFP::getValue is needed to do it via single template<br>
+// function for both apint/apfloat.<br>
+struct apfloat_match {<br>
+ const APFloat *&Res;<br>
+ apfloat_match(const APFloat *&R) : Res(R) {}<br>
+ template <typename ITy> bool match(ITy *V) {<br>
+ if (auto *CI = dyn_cast<ConstantFP>(V)) {<br>
+ Res = &CI->getValueAPF();<br>
+ return true;<br>
+ }<br>
+ if (V->getType()->isVectorTy())<br>
+ if (const auto *C = dyn_cast<Constant>(V))<br>
+ if (auto *CI = dyn_cast_or_null<ConstantFP>(C->getSplatValue())) {<br>
+ Res = &CI->getValueAPF();<br>
+ return true;<br>
+ }<br>
+ return false;<br>
+ }<br>
+};<br>
<br>
/// \brief Match a ConstantInt or splatted ConstantVector, binding the<br>
/// specified pointer to the contained APInt.<br>
inline apint_match m_APInt(const APInt *&Res) { return Res; }<br>
<br>
+/// \brief Match a ConstantFP or splatted ConstantVector, binding the<br>
+/// specified pointer to the contained APFloat.<br>
+inline apfloat_match m_APFloat(const APFloat *&Res) { return Res; }<br>
+<br>
template <int64_t Val> struct constantint_match {<br>
template <typename ITy> bool match(ITy *V) {<br>
if (const auto *CI = dyn_cast<ConstantInt>(V)) {<br>
<br>
Modified: llvm/trunk/lib/Analysis/ValueTracking.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ValueTracking.cpp?rev=306525&r1=306524&r2=306525&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ValueTracking.cpp?rev=306525&r1=306524&r2=306525&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Analysis/ValueTracking.cpp (original)<br>
+++ llvm/trunk/lib/Analysis/ValueTracking.cpp Wed Jun 28 02:26:20 2017<br>
@@ -3933,6 +3933,62 @@ static bool isKnownNonZero(const Value *<br>
return false;<br>
}<br>
<br>
+/// Match clamp pattern for float types without care about NaNs or signed zeros.<br>
+/// Given non-min/max outer cmp/select from the clamp pattern this<br>
+/// function recognizes if it can be substitued by a "canonical" min/max<br>
+/// pattern.<br>
+static SelectPatternResult matchFastFloatClamp(CmpInst::Predicate Pred,<br>
+ Value *CmpLHS, Value *CmpRHS,<br>
+ Value *TrueVal, Value *FalseVal,<br>
+ Value *&LHS, Value *&RHS) {<br>
+ // Try to match<br>
+ // X < C1 ? C1 : Min(X, C2) --> Max(C1, Min(X, C2))<br>
+ // X > C1 ? C1 : Max(X, C2) --> Min(C1, Max(X, C2))<br>
+ // and return description of the outer Max/Min.<br>
+<br>
+ // First, check if select has inverse order:<br>
+ if (CmpRHS == FalseVal) {<br>
+ std::swap(TrueVal, FalseVal);<br>
+ Pred = CmpInst::getInversePredicate(Pred);<br>
+ }<br>
+<br>
+ // Assume success now. If there's no match, callers should not use these anyway.<br>
+ LHS = TrueVal;<br>
+ RHS = FalseVal;<br>
+<br>
+ const APFloat *FC1;<br>
+ if (CmpRHS != TrueVal || !match(CmpRHS, m_APFloat(FC1)) || !FC1->isFinite())<br>
+ return {SPF_UNKNOWN, SPNB_NA, false};<br>
+<br>
+ const APFloat *FC2;<br>
+ switch (Pred) {<br>
+ case CmpInst::FCMP_OLT:<br>
+ case CmpInst::FCMP_OLE:<br>
+ case CmpInst::FCMP_ULT:<br>
+ case CmpInst::FCMP_ULE:<br>
+ if (match(FalseVal,<br>
+ m_CombineOr(m_OrdFMin(m_Specific(CmpLHS), m_APFloat(FC2)),<br>
+ m_UnordFMin(m_Specific(CmpLHS), m_APFloat(FC2)))) &&<br>
+ FC1->compare(*FC2) == APFloat::cmpResult::cmpLessThan)<br>
+ return {SPF_FMAXNUM, SPNB_RETURNS_ANY, false};<br>
+ break;<br>
+ case CmpInst::FCMP_OGT:<br>
+ case CmpInst::FCMP_OGE:<br>
+ case CmpInst::FCMP_UGT:<br>
+ case CmpInst::FCMP_UGE:<br>
+ if (match(FalseVal,<br>
+ m_CombineOr(m_OrdFMax(m_Specific(CmpLHS), m_APFloat(FC2)),<br>
+ m_UnordFMax(m_Specific(CmpLHS), m_APFloat(FC2)))) &&<br>
+ FC1->compare(*FC2) == APFloat::cmpResult::cmpGreaterThan)<br>
+ return {SPF_FMINNUM, SPNB_RETURNS_ANY, false};<br>
+ break;<br>
+ default:<br>
+ break;<br>
+ }<br>
+<br>
+ return {SPF_UNKNOWN, SPNB_NA, false};<br>
+}<br>
+<br>
/// Match non-obvious integer minimum and maximum sequences.<br>
static SelectPatternResult matchMinMax(CmpInst::Predicate Pred,<br>
Value *CmpLHS, Value *CmpRHS,<br>
@@ -4140,7 +4196,18 @@ static SelectPatternResult matchSelectPa<br>
}<br>
}<br>
<br>
- return matchMinMax(Pred, CmpLHS, CmpRHS, TrueVal, FalseVal, LHS, RHS);<br>
+ if (CmpInst::isIntPredicate(Pred))<br>
+ return matchMinMax(Pred, CmpLHS, CmpRHS, TrueVal, FalseVal, LHS, RHS);<br>
+<br>
+ // According to (IEEE 754-2008 5.3.1), minNum(0.0, -0.0) and similar<br>
+ // may return either -0.0 or 0.0, so fcmp/select pair has stricter<br>
+ // semantics than minNum. Be conservative in such case.<br>
+ if (NaNBehavior != SPNB_RETURNS_ANY ||<br>
+ (!FMF.noSignedZeros() && !isKnownNonZero(CmpLHS) &&<br>
+ !isKnownNonZero(CmpRHS)))<br>
+ return {SPF_UNKNOWN, SPNB_NA, false};<br>
+<br>
+ return matchFastFloatClamp(Pred, CmpLHS, CmpRHS, TrueVal, FalseVal, LHS, RHS);<br>
}<br>
<br>
static Value *lookThroughCast(CmpInst *CmpI, Value *V1, Value *V2,<br>
<br>
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp?rev=306525&r1=306524&r2=306525&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp?rev=306525&r1=306524&r2=306525&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp (original)<br>
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineSelect.cpp Wed Jun 28 02:26:20 2017<br>
@@ -1374,9 +1374,16 @@ Instruction *InstCombiner::visitSelectIn<br>
auto SPF = SPR.Flavor;<br>
<br>
if (SelectPatternResult::isMinOrMax(SPF)) {<br>
- // Canonicalize so that type casts are outside select patterns.<br>
- if (LHS->getType()->getPrimitiveSizeInBits() !=<br>
- SelType->getPrimitiveSizeInBits()) {<br>
+ // Canonicalize so that<br>
+ // - type casts are outside select patterns.<br>
+ // - float clamp is transformed to min/max pattern<br>
+ Value *CmpLHS = cast<CmpInst>(CondVal)->getOperand(0);<br>
+ Value *CmpRHS = cast<CmpInst>(CondVal)->getOperand(1);<br>
+ if ((LHS->getType()->getPrimitiveSizeInBits() !=<br>
+ SelType->getPrimitiveSizeInBits()) ||<br>
+ (LHS->getType()->isFPOrFPVectorTy() &&<br>
+ ((CmpLHS != LHS && CmpLHS != RHS) ||<br>
+ (CmpRHS != LHS && CmpRHS != RHS)))) {<br>
CmpInst::Predicate Pred = getCmpPredicateForMinMax(SPF, SPR.Ordered);<br>
<br>
Value *Cmp;<br>
<br>
Modified: llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll?rev=306525&r1=306524&r2=306525&view=diff" target="_blank">
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll?rev=306525&r1=306524&r2=306525&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll (original)<br>
+++ llvm/trunk/test/Transforms/InstCombine/clamp-to-minmax.ll Wed Jun 28 02:26:20 2017<br>
@@ -7,9 +7,9 @@ define float @clamp_float_fast_ordered_s<br>
; CHECK-LABEL: @clamp_float_fast_ordered_strict_maxmin(<br>
; CHECK-NEXT: [[CMP2:%.*]] = fcmp fast olt float [[X:%.*]], 2.550000e+02<br>
; CHECK-NEXT: [[MIN:%.*]] = select i1 [[CMP2]], float [[X]], float 2.550000e+02<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast olt float [[X]], 1.000000e+00<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 1.000000e+00, float [[MIN]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast oge float [[MIN]], 1.000000e+00<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MIN]], float 1.000000e+00<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast olt float %x, 255.0<br>
%min = select i1 %cmp2, float %x, float 255.0<br>
@@ -24,9 +24,9 @@ define float @clamp_float_fast_ordered_n<br>
; CHECK-LABEL: @clamp_float_fast_ordered_nonstrict_maxmin(<br>
; CHECK-NEXT: [[CMP2:%.*]] = fcmp fast olt float [[X:%.*]], 2.550000e+02<br>
; CHECK-NEXT: [[MIN:%.*]] = select i1 [[CMP2]], float [[X]], float 2.550000e+02<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast ole float [[X]], 1.000000e+00<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 1.000000e+00, float [[MIN]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast oge float [[MIN]], 1.000000e+00<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MIN]], float 1.000000e+00<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast olt float %x, 255.0<br>
%min = select i1 %cmp2, float %x, float 255.0<br>
@@ -41,9 +41,9 @@ define float @clamp_float_fast_ordered_s<br>
; CHECK-LABEL: @clamp_float_fast_ordered_strict_minmax(<br>
; CHECK-NEXT: [[CMP2:%.*]] = fcmp fast ogt float [[X:%.*]], 1.000000e+00<br>
; CHECK-NEXT: [[MAX:%.*]] = select i1 [[CMP2]], float [[X]], float 1.000000e+00<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast ogt float [[X]], 2.550000e+02<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 2.550000e+02, float [[MAX]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast ole float [[MAX]], 2.550000e+02<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MAX]], float 2.550000e+02<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ogt float %x, 1.0<br>
%max = select i1 %cmp2, float %x, float 1.0<br>
@@ -58,9 +58,9 @@ define float @clamp_float_fast_ordered_n<br>
; CHECK-LABEL: @clamp_float_fast_ordered_nonstrict_minmax(<br>
; CHECK-NEXT: [[CMP2:%.*]] = fcmp fast ogt float [[X:%.*]], 1.000000e+00<br>
; CHECK-NEXT: [[MAX:%.*]] = select i1 [[CMP2]], float [[X]], float 1.000000e+00<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast oge float [[X]], 2.550000e+02<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 2.550000e+02, float [[MAX]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast ole float [[MAX]], 2.550000e+02<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MAX]], float 2.550000e+02<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ogt float %x, 1.0<br>
%max = select i1 %cmp2, float %x, float 1.0<br>
@@ -78,9 +78,9 @@ define float @clamp_float_fast_unordered<br>
; CHECK-LABEL: @clamp_float_fast_unordered_strict_maxmin(<br>
; CHECK-NEXT: [[CMP2_INV:%.*]] = fcmp fast oge float [[X:%.*]], 2.550000e+02<br>
; CHECK-NEXT: [[MIN:%.*]] = select i1 [[CMP2_INV]], float 2.550000e+02, float [[X]]<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast ult float [[X]], 1.000000e+00<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 1.000000e+00, float [[MIN]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast oge float [[MIN]], 1.000000e+00<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MIN]], float 1.000000e+00<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ult float %x, 255.0<br>
%min = select i1 %cmp2, float %x, float 255.0<br>
@@ -95,9 +95,9 @@ define float @clamp_float_fast_unordered<br>
; CHECK-LABEL: @clamp_float_fast_unordered_nonstrict_maxmin(<br>
; CHECK-NEXT: [[CMP2_INV:%.*]] = fcmp fast oge float [[X:%.*]], 2.550000e+02<br>
; CHECK-NEXT: [[MIN:%.*]] = select i1 [[CMP2_INV]], float 2.550000e+02, float [[X]]<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast ule float [[X]], 1.000000e+00<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 1.000000e+00, float [[MIN]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast oge float [[MIN]], 1.000000e+00<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MIN]], float 1.000000e+00<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ult float %x, 255.0<br>
%min = select i1 %cmp2, float %x, float 255.0<br>
@@ -112,9 +112,9 @@ define float @clamp_float_fast_unordered<br>
; CHECK-LABEL: @clamp_float_fast_unordered_strict_minmax(<br>
; CHECK-NEXT: [[CMP2_INV:%.*]] = fcmp fast ole float [[X:%.*]], 1.000000e+00<br>
; CHECK-NEXT: [[MAX:%.*]] = select i1 [[CMP2_INV]], float 1.000000e+00, float [[X]]<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast ugt float [[X]], 2.550000e+02<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 2.550000e+02, float [[MAX]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast ole float [[MAX]], 2.550000e+02<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MAX]], float 2.550000e+02<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ugt float %x, 1.0<br>
%max = select i1 %cmp2, float %x, float 1.0<br>
@@ -129,9 +129,9 @@ define float @clamp_float_fast_unordered<br>
; CHECK-LABEL: @clamp_float_fast_unordered_nonstrict_minmax(<br>
; CHECK-NEXT: [[CMP2_INV:%.*]] = fcmp fast ole float [[X:%.*]], 1.000000e+00<br>
; CHECK-NEXT: [[MAX:%.*]] = select i1 [[CMP2_INV]], float 1.000000e+00, float [[X]]<br>
-; CHECK-NEXT: [[CMP1:%.*]] = fcmp fast uge float [[X]], 2.550000e+02<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[CMP1]], float 2.550000e+02, float [[MAX]]<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast ole float [[MAX]], 2.550000e+02<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[MAX]], float 2.550000e+02<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%cmp2 = fcmp fast ugt float %x, 1.0<br>
%max = select i1 %cmp2, float %x, float 1.0<br>
@@ -143,13 +143,14 @@ define float @clamp_float_fast_unordered<br>
; Some more checks with fast<br>
<br>
; (X > 1.0) ? min(x, 255.0) : 1.0<br>
+; That did not match because select was in inverse order.<br>
define float @clamp_test_1(float %x) {<br>
; CHECK-LABEL: @clamp_test_1(<br>
; CHECK-NEXT: [[INNER_CMP_INV:%.*]] = fcmp fast oge float [[X:%.*]], 2.550000e+02<br>
; CHECK-NEXT: [[INNER_SEL:%.*]] = select i1 [[INNER_CMP_INV]], float 2.550000e+02, float [[X]]<br>
-; CHECK-NEXT: [[OUTER_CMP:%.*]] = fcmp fast ugt float [[X]], 1.000000e+00<br>
-; CHECK-NEXT: [[R:%.*]] = select i1 [[OUTER_CMP]], float [[INNER_SEL]], float 1.000000e+00<br>
-; CHECK-NEXT: ret float [[R]]<br>
+; CHECK-NEXT: [[DOTINV:%.*]] = fcmp fast oge float [[INNER_SEL]], 1.000000e+00<br>
+; CHECK-NEXT: [[R1:%.*]] = select i1 [[DOTINV]], float [[INNER_SEL]], float 1.000000e+00<br>
+; CHECK-NEXT: ret float [[R1]]<br>
;<br>
%inner_cmp = fcmp fast ult float %x, 255.0<br>
%inner_sel = select i1 %inner_cmp, float %x, float 255.0<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><o:p></o:p></p>
</blockquote>
</div>
</div>
</body>
</html>