<div dir="ltr"><div>Ah, I think I see it now. Before, we would call SimplifyAsssociativeOrCommutative() and then try to SimplifyAndInst() which would return if anything simplified and protected all of the later code from easy optimizations. Now, we call SimplifyAndInst() first (which always returns if it made a change), but we then call SimplifyAsssociativeOrCommutative() and continue through visitAnd() even though it may have exposed simplifications.</div><div><br></div><div>The easy solution will be to return immediately from SimplifyAsssociativeOrCommutative() rather than carrying on. This will give SimplifyAndInst() a chance to run again before the other transforms. We'll need to make this change in all places that have this pattern to be safe.<br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Jul 12, 2018 at 6:19 PM, Sanjay Patel <span dir="ltr"><<a href="mailto:spatel@rotateright.com" target="_blank">spatel@rotateright.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>Thanks for letting me know. Not sure how it would've changed, but I'll take a look.</div><div><br></div><div> I see that bugpoint doesn't reduce this further. Any chance that you did that manually?</div><div><br></div><div>My first guess is that we are missing something in InstSimplify because InstCombine is assuming the easy combination of cmps are already gone.<br></div></div><div class="HOEnZb"><div class="h5"><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Jul 12, 2018 at 3:52 PM, Mikhail Zolotukhin <span dir="ltr"><<a href="mailto:mzolotukhin@apple.com" target="_blank">mzolotukhin@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word;line-break:after-white-space">Hi Sanjay,<div><br></div><div>Our fuzzer-bot detected a failure after this change. Here is a reproducer:</div><div><br></div><blockquote style="margin:0 0 0 40px;border:none;padding:0px"><div><font size="1" face="Menlo">$ cat fuzz.ll</font></div><div><font size="1" face="Menlo">source_filename = "M"</font></div><div><font size="1" face="Menlo">target triple = "x86_64-apple-darwin17.6.0"</font></div><div><font size="1" face="Menlo"><br></font></div><div><font size="1" face="Menlo">define void @f() {</font></div><div><font size="1" face="Menlo">BB:</font></div><div><font size="1" face="Menlo"> %A8 = alloca i16</font></div><div><font size="1" face="Menlo"> %L7 = load i16, i16* %A8</font></div><div><font size="1" face="Menlo"> %G21 = getelementptr i16, i16* %A8, i8 -1</font></div><div><font size="1" face="Menlo"> %B11 = udiv i16 %L7, -1</font></div><div><font size="1" face="Menlo"> %G4 = getelementptr i16, i16* %A8, i16 %B11</font></div><div><font size="1" face="Menlo"> %L2 = load i16, i16* %G4</font></div><div><font size="1" face="Menlo"> %L = load i16, i16* %G4</font></div><div><font size="1" face="Menlo"> %B23 = mul i16 %B11, %B11</font></div><div><font size="1" face="Menlo"> %L4 = load i16, i16* %A8</font></div><div><font size="1" face="Menlo"> %B21 = sdiv i16 %L7, %L4</font></div><div><font size="1" face="Menlo"> %B7 = sub i16 0, %B21</font></div><div><font size="1" face="Menlo"> %B18 = mul i16 %B23, %B7</font></div><div><font size="1" face="Menlo"> %C10 = icmp ugt i16 %L, %B11</font></div><div><font size="1" face="Menlo"> %B20 = and i16 %L7, %L2</font></div><div><font size="1" face="Menlo"> %B1 = mul i1 %C10, true</font></div><div><font size="1" face="Menlo"> %C5 = icmp sle i16 %B21, %L</font></div><div><font size="1" face="Menlo"> %C11 = icmp ule i16 %B21, %L</font></div><div><font size="1" face="Menlo"> %C7 = icmp slt i16 %B20, 0</font></div><div><font size="1" face="Menlo"> %B29 = srem i16 %L4, %B18</font></div><div><font size="1" face="Menlo"> %B15 = add i1 %C7, %C10</font></div><div><font size="1" face="Menlo"> %B19 = add i1 %C11, %B15</font></div><div><font size="1" face="Menlo"> %C6 = icmp sge i1 %C11, %B19</font></div><div><font size="1" face="Menlo"> %B33 = or i16 %B29, %L4</font></div><div><font size="1" face="Menlo"> %C13 = icmp uge i1 %C5, %B1</font></div><div><font size="1" face="Menlo"> %C3 = icmp ult i1 %C13, %C6</font></div><div><font size="1" face="Menlo"> store i16 undef, i16* %G21</font></div><div><font size="1" face="Menlo"> %C18 = icmp ule i1 %C10, %C7</font></div><div><font size="1" face="Menlo"> %G26 = getelementptr i1, i1* null, i1 %C3</font></div><div><font size="1" face="Menlo"> store i16 %B33, i16* undef</font></div><div><font size="1" face="Menlo"> store i1 %C18, i1* undef</font></div><div><font size="1" face="Menlo"> store i1* %G26, i1** undef</font></div><div><font size="1" face="Menlo"> ret void</font></div><div><font size="1" face="Menlo">}</font></div><div><font size="1" face="Menlo"><br></font></div><div><font size="1" face="Menlo">$ bin/opt -instcombine fuzz.ll -disable-output</font></div><div><font size="1" face="Menlo">Unknown integer condition code!</font></div><div><font size="1" face="Menlo">UNREACHABLE executed at /Users/mvzolotu/devel/mono/llv<wbr>m-project/llvm/lib/Transforms/<wbr>InstCombine/InstCombineAndOrXo<wbr>r.cpp:1034!</font></div></blockquote><div><br></div><div>I took a look at the code of foldAndOfICmps, and indeed we don’t have a code to handle %I1=%X < 0, %I2=%X < 1, %I3 = %I1 && %I2.</div><div>What would be the best way to handle it (and why didn’t we hit it before)?</div><div><br></div><div>Thanks,</div><div>Michael</div><div><div class="m_1244000893323331575h5"><div><div><br><blockquote type="cite"><div>On Jun 21, 2018, at 10:06 AM, Sanjay Patel via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>> wrote:</div><br class="m_1244000893323331575m_5844902724887122003Apple-interchange-newline"><div><div>Author: spatel<br>Date: Thu Jun 21 10:06:36 2018<br>New Revision: 335258<br><br>URL: <a href="http://llvm.org/viewvc/llvm-project?rev=335258&view=rev" target="_blank">http://llvm.org/viewvc/llvm-pr<wbr>oject?rev=335258&view=rev</a><br>Log:<br>[InstCombine] simplify binops before trying other folds<br><br>This is outwardly NFC from what I can tell, but it should be more efficient <br>to simplify first (despite the name, SimplifyAssociativeOrCommutati<wbr>ve does<br>not actually simplify as InstSimplify does - it creates/morphs instructions).<br><br>This should make it easier to refactor duplicated code that runs for all binops.<br><br>Modified:<br> llvm/trunk/lib/Transforms/I<wbr>nstCombine/InstCombineAddSub.c<wbr>pp<br> llvm/trunk/lib/Transforms/I<wbr>nstCombine/InstCombineAndOrXor<wbr>.cpp<br> llvm/trunk/lib/Transforms/I<wbr>nstCombine/InstCombineMulDivRe<wbr>m.cpp<br> llvm/trunk/lib/Transforms/I<wbr>nstCombine/InstCombineShifts.c<wbr>pp<br><br>Modified: llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAddSub.cpp<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp?rev=335258&r1=335257&r2=335258&view=diff" target="_blank">http://llvm.org/viewvc/llvm-pr<wbr>oject/llvm/trunk/lib/Transform<wbr>s/InstCombine/InstCombineAddSu<wbr>b.cpp?rev=335258&r1=335257&r2=<wbr>335258&view=diff</a><br>==============================<wbr>==============================<wbr>==================<br>--- llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAddSub.cpp (original)<br>+++ llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAddSub.cpp Thu Jun 21 10:06:36 2018<br>@@ -1121,13 +1121,12 @@ static Instruction *canonicalizeLowbitMa<br> }<br><br> Instruction *InstCombiner::visitAdd(Binary<wbr>Operator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);<br>- if (Value *V =<br>- SimplifyAddInst(LHS, RHS, I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>- SQ.ge<wbr>tWithInstruction(&I)))<br>+ if (Value *V = SimplifyAddInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -1140,6 +1139,7 @@ Instruction *InstCombiner::visitAdd(Bina<br><br> // FIXME: This should be moved into the above helper function to allow these<br> // transforms for general constant or constant splat vectors.<br>+ Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);<br> Type *Ty = I.getType();<br> if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {<br> Value *XorLHS = nullptr; ConstantInt *XorRHS = nullptr;<br>@@ -1378,18 +1378,19 @@ Instruction *InstCombiner::visitAdd(Bina<br> }<br><br> Instruction *InstCombiner::visitFAdd(Binar<wbr>yOperator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);<br>- if (Value *V = SimplifyFAddInst(LHS, RHS, I.getFastMathFlags(),<br>+ if (Value *V = SimplifyFAddInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> I.getFastMathFlags(),<br> <wbr> SQ.getWithInstruction(&I))<wbr>)<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br> if (Instruction *FoldedFAdd = foldBinOpIntoSelectOrPhi(I))<br> return FoldedFAdd;<br><br>+ Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);<br> Value *X;<br> // (-X) + Y --> Y - X<br> if (match(LHS, m_FNeg(m_Value(X))))<br>@@ -1555,10 +1556,9 @@ Value *InstCombiner::OptimizePointer<wbr>Diff<br> }<br><br> Instruction *InstCombiner::visitSub(Binary<wbr>Operator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V =<br>- SimplifySubInst(Op0, Op1, I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>- SQ.ge<wbr>tWithInstruction(&I)))<br>+ if (Value *V = SimplifySubInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -1569,6 +1569,7 @@ Instruction *InstCombiner::visitSub(Bina<br> return replaceInstUsesWith(I, V);<br><br> // If this is a 'B = x-(-A)', change to B = x+A.<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (Value *V = dyn_castNegVal(Op1)) {<br> BinaryOperator *Res = BinaryOperator::CreateAdd(Op0, V);<br><br>@@ -1808,8 +1809,8 @@ Instruction *InstCombiner::visitSub(Bina<br> }<br><br> Instruction *InstCombiner::visitFSub(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyFSubInst(Op0, Op1, I.getFastMathFlags(),<br>+ if (Value *V = SimplifyFSubInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> I.getFastMathFlags(),<br> <wbr> SQ.getWithInstruction(&I))<wbr>)<br> return replaceInstUsesWith(I, V);<br><br>@@ -1818,6 +1819,7 @@ Instruction *InstCombiner::visitFSub(Bin<br><br> // Subtraction from -0.0 is the canonical form of fneg.<br> // fsub nsz 0, X ==> fsub nsz -0.0, X<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (I.hasNoSignedZeros() && match(Op0, m_PosZeroFP()))<br> return BinaryOperator::CreateFNegFMF(<wbr>Op1, &I);<br><br><br>Modified: llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAndOrXor.<wbr>cpp<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp?rev=335258&r1=335257&r2=335258&view=diff" target="_blank">http://llvm.org/viewvc/llvm-pr<wbr>oject/llvm/trunk/lib/Transform<wbr>s/InstCombine/InstCombineAndOr<wbr>Xor.cpp?rev=335258&r1=335257&<wbr>r2=335258&view=diff</a><br>==============================<wbr>==============================<wbr>==================<br>--- llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAndOrXor.<wbr>cpp (original)<br>+++ llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineAndOrXor.<wbr>cpp Thu Jun 21 10:06:36 2018<br>@@ -1401,11 +1401,11 @@ Instruction *InstCombiner::narrowMaskedB<br> // here. We should standardize that construct where it is needed or choose some<br> // other way to ensure that commutated variants of patterns are not missed.<br> Instruction *InstCombiner::visitAnd(Binary<wbr>Operator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyAndInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyAndInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -1425,6 +1425,7 @@ Instruction *InstCombiner::visitAnd(Bina<br> if (Value *V = SimplifyBSwap(I, Builder))<br> return replaceInstUsesWith(I, V);<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> const APInt *C;<br> if (match(Op1, m_APInt(C))) {<br> Value *X, *Y;<br>@@ -2015,11 +2016,11 @@ Value *InstCombiner::foldOrOfICmps(I<wbr>CmpI<br> // here. We should standardize that construct where it is needed or choose some<br> // other way to ensure that commutated variants of patterns are not missed.<br> Instruction *InstCombiner::visitOr(BinaryO<wbr>perator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyOrInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyOrInst(I.getOperand(0)<wbr>, I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -2046,6 +2047,7 @@ Instruction *InstCombiner::visitOr(Binar<br> if (Instruction *BSwap = MatchBSwap(I))<br> return BSwap;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> {<br> Value *A;<br> const APInt *C;<br>@@ -2468,11 +2470,11 @@ static Instruction *visitMaskedMerge(Bin<br> // here. We should standardize that construct where it is needed or choose some<br> // other way to ensure that commutated variants of patterns are not missed.<br> Instruction *InstCombiner::visitXor(Binary<wbr>Operator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyXorInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyXorInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -2492,6 +2494,7 @@ Instruction *InstCombiner::visitXor(Bina<br> return replaceInstUsesWith(I, V);<br><br> // A^B --> A|B iff A and B have no bits set in common.<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (haveNoCommonBitsSet(Op0, Op1, DL, &AC, &I, &DT))<br> return BinaryOperator::CreateOr(Op0, Op1);<br><br><br>Modified: llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineMulDivRem.<wbr>cpp<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp?rev=335258&r1=335257&r2=335258&view=diff" target="_blank">http://llvm.org/viewvc/llvm-pr<wbr>oject/llvm/trunk/lib/Transform<wbr>s/InstCombine/InstCombineMulDi<wbr>vRem.cpp?rev=335258&r1=335257&<wbr>r2=335258&view=diff</a><br>==============================<wbr>==============================<wbr>==================<br>--- llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineMulDivRem.<wbr>cpp (original)<br>+++ llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineMulDivRem.<wbr>cpp Thu Jun 21 10:06:36 2018<br>@@ -126,11 +126,11 @@ static Constant *getLogBase2(Type *Ty, C<br> }<br><br> Instruction *InstCombiner::visitMul(Binary<wbr>Operator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyMulInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyMulInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -138,6 +138,7 @@ Instruction *InstCombiner::visitMul(Bina<br> return replaceInstUsesWith(I, V);<br><br> // X * -1 == 0 - X<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (match(Op1, m_AllOnes())) {<br> BinaryOperator *BO = BinaryOperator::CreateNeg(Op0, I.getName());<br> if (I.hasNoSignedWrap())<br>@@ -406,12 +407,12 @@ Instruction *InstCombiner::visitMul(Bina<br> }<br><br> Instruction *InstCombiner::visitFMul(Binar<wbr>yOperator &I) {<br>- bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyFMulInst(Op0, Op1, I.getFastMathFlags(),<br>+ if (Value *V = SimplifyFMulInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> I.getFastMathFlags(),<br> <wbr> SQ.getWithInstruction(&I))<wbr>)<br> return replaceInstUsesWith(I, V);<br><br>+ bool Changed = SimplifyAssociativeOrCommutati<wbr>ve(I);<br> if (Instruction *X = foldShuffledBinop(I))<br> return X;<br><br>@@ -419,6 +420,7 @@ Instruction *InstCombiner::visitFMul(Bin<br> return FoldedMul;<br><br> // X * -1.0 --> -X<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (match(Op1, m_SpecificFP(-1.0)))<br> return BinaryOperator::CreateFNegFMF(<wbr>Op0, &I);<br><br>@@ -936,8 +938,8 @@ static Instruction *narrowUDivURem(Binar<br> }<br><br> Instruction *InstCombiner::visitUDiv(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyUDivInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyUDivInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -948,6 +950,7 @@ Instruction *InstCombiner::visitUDiv(Bin<br> return Common;<br><br> // (x lshr C1) udiv C2 --> x udiv (C2 << C1)<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> {<br> Value *X;<br> const APInt *C1, *C2;<br>@@ -1004,8 +1007,8 @@ Instruction *InstCombiner::visitUDiv(Bin<br> }<br><br> Instruction *InstCombiner::visitSDiv(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifySDivInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifySDivInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -1015,6 +1018,7 @@ Instruction *InstCombiner::visitSDiv(Bin<br> if (Instruction *Common = commonIDivTransforms(I))<br> return Common;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> const APInt *Op1C;<br> if (match(Op1, m_APInt(Op1C))) {<br> // sdiv X, -1 == -X<br>@@ -1147,8 +1151,8 @@ static Instruction *foldFDivConstantDivi<br> }<br><br> Instruction *InstCombiner::visitFDiv(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyFDivInst(Op0, Op1, I.getFastMathFlags(),<br>+ if (Value *V = SimplifyFDivInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> I.getFastMathFlags(),<br> <wbr> SQ.getWithInstruction(&I))<wbr>)<br> return replaceInstUsesWith(I, V);<br><br>@@ -1161,6 +1165,7 @@ Instruction *InstCombiner::visitFDiv(Bin<br> if (Instruction *R = foldFDivConstantDividend(I))<br> return R;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (isa<Constant>(Op0))<br> if (SelectInst *SI = dyn_cast<SelectInst>(Op1))<br> if (Instruction *R = FoldOpIntoSelect(I, SI))<br>@@ -1276,8 +1281,8 @@ Instruction *InstCombiner::commonIRemTra<br> }<br><br> Instruction *InstCombiner::visitURem(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyURemInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyURemInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -1290,6 +1295,7 @@ Instruction *InstCombiner::visitURem(Bin<br> return NarrowRem;<br><br> // X urem Y -> X and Y-1, where Y is a power of 2,<br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> if (isKnownToBeAPowerOfTwo(Op1, /*OrZero*/ true, 0, &I)) {<br> Constant *N1 = Constant::getAllOnesValue(I.ge<wbr>tType());<br> Value *Add = Builder.CreateAdd(Op1, N1);<br>@@ -1314,8 +1320,8 @@ Instruction *InstCombiner::visitURem(Bin<br> }<br><br> Instruction *InstCombiner::visitSRem(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifySRemInst(Op0, Op1, SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifySRemInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -1325,6 +1331,7 @@ Instruction *InstCombiner::visitSRem(Bin<br> if (Instruction *Common = commonIRemTransforms(I))<br> return Common;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> {<br> const APInt *Y;<br> // X % -Y -> X % Y<br>@@ -1386,8 +1393,8 @@ Instruction *InstCombiner::visitSRem(Bin<br> }<br><br> Instruction *InstCombiner::visitFRem(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V = SimplifyFRemInst(Op0, Op1, I.getFastMathFlags(),<br>+ if (Value *V = SimplifyFRemInst(I.getOperand(<wbr>0), I.getOperand(1),<br>+ <wbr> I.getFastMathFlags(),<br> <wbr> SQ.getWithInstruction(&I))<wbr>)<br> return replaceInstUsesWith(I, V);<br><br><br>Modified: llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineShifts.cpp<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp?rev=335258&r1=335257&r2=335258&view=diff" target="_blank">http://llvm.org/viewvc/llvm-pr<wbr>oject/llvm/trunk/lib/Transform<wbr>s/InstCombine/InstCombineShift<wbr>s.cpp?rev=335258&r1=335257&r2=<wbr>335258&view=diff</a><br>==============================<wbr>==============================<wbr>==================<br>--- llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineShifts.cpp (original)<br>+++ llvm/trunk/lib/Transforms/Inst<wbr>Combine/InstCombineShifts.cpp Thu Jun 21 10:06:36 2018<br>@@ -588,10 +588,9 @@ Instruction *InstCombiner::FoldShiftByCo<br> }<br><br> Instruction *InstCombiner::visitShl(Binary<wbr>Operator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V =<br>- SimplifyShlInst(Op0, Op1, I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>- SQ.ge<wbr>tWithInstruction(&I)))<br>+ if (Value *V = SimplifyShlInst(I.getOperand(0<wbr>), I.getOperand(1),<br>+ <wbr> I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -600,11 +599,12 @@ Instruction *InstCombiner::visitShl(Bina<br> if (Instruction *V = commonShiftTransforms(I))<br> return V;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> Type *Ty = I.getType();<br> const APInt *ShAmtAPInt;<br> if (match(Op1, m_APInt(ShAmtAPInt))) {<br> unsigned ShAmt = ShAmtAPInt->getZExtValue();<br>- unsigned BitWidth = I.getType()->getScalarSizeInBi<wbr>ts();<br>+ unsigned BitWidth = Ty->getScalarSizeInBits();<br><br> // shl (zext X), ShAmt --> zext (shl X, ShAmt)<br> // This is only valid if X would have zeros shifted out.<br>@@ -693,9 +693,8 @@ Instruction *InstCombiner::visitShl(Bina<br> }<br><br> Instruction *InstCombiner::visitLShr(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V =<br>- SimplifyLShrInst(Op0, Op1, I.isExact(), SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyLShrInst(I.getOperand(<wbr>0), I.getOperand(1), I.isExact(),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -704,6 +703,7 @@ Instruction *InstCombiner::visitLShr(Bin<br> if (Instruction *R = commonShiftTransforms(I))<br> return R;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> Type *Ty = I.getType();<br> const APInt *ShAmtAPInt;<br> if (match(Op1, m_APInt(ShAmtAPInt))) {<br>@@ -821,9 +821,8 @@ Instruction *InstCombiner::visitLShr(Bin<br> }<br><br> Instruction *InstCombiner::visitAShr(Binar<wbr>yOperator &I) {<br>- Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br>- if (Value *V =<br>- SimplifyAShrInst(Op0, Op1, I.isExact(), SQ.getWithInstruction(&I)))<br>+ if (Value *V = SimplifyAShrInst(I.getOperand(<wbr>0), I.getOperand(1), I.isExact(),<br>+ <wbr> SQ.getWithInstruction(&I)))<br> return replaceInstUsesWith(I, V);<br><br> if (Instruction *X = foldShuffledBinop(I))<br>@@ -832,6 +831,7 @@ Instruction *InstCombiner::visitAShr(Bin<br> if (Instruction *R = commonShiftTransforms(I))<br> return R;<br><br>+ Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);<br> Type *Ty = I.getType();<br> unsigned BitWidth = Ty->getScalarSizeInBits();<br> const APInt *ShAmtAPInt;<br><br><br>______________________________<wbr>_________________<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/<wbr>mailman/listinfo/llvm-commits</a><br></div></div></blockquote></div><br></div></div></div></div></blockquote></div><br></div>
</div></div></blockquote></div><br></div>