[InstCombine] Shift amount reassociation in bittest: trunc-of-shl (PR42399)

[lebedev.ri]

Summary

This is continuation of D63829 / https://bugs.llvm.org/show_bug.cgi?id=42399

I thought naive pattern would solve my issue, but nope, it involved truncation,
thus more folds needed.. This isn't really the fold i'm interested in,
i need trunc-of-lshr, but i'we decided to start with shl because it's simpler.

In this case, no extra legality checks are needed:
https://rise4fun.com/Alive/CAb

We should be careful about not increasing instruction count,
since we need to produce zext because and is done in wider type.

[lebedev.ri]

bump

[spatel]

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

3315–3321 ↗

(On Diff #214538)

Can we make the change to use Instruction*/m_Instruction independently (ie as a pre-commit) of this patch?
(I'd be fine without requiring a weird ConstExpr test to prove the diff if that was a possibility.)

3372 ↗

(On Diff #214538)

Extra period at end.

[lebedev.ri]

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

3315–3321 ↗

(On Diff #214538)

I actually already did that in rL368554 to fix https://bugs.llvm.org/show_bug.cgi?id=42962
I just need to rebase this, hold on..

[lebedev.ri]

Rebased, thanks for taking a look!

[spatel]

llvm/test/Transforms/InstCombine/shift-amount-reassociation-in-bittest-with-truncation-shl.ll

414 ↗

(On Diff #215443)

Please explain and/or add a comment here.
If we leave the add constant as 1, this gets simplified away completely with this patch? Why doesn't it happen with 32?
Do we already have a test that covers the case for the original test (with add of 1)?

[lebedev.ri]

Thanks for taking a look!

llvm/test/Transforms/InstCombine/shift-amount-reassociation-in-bittest-with-truncation-shl.ll

414 ↗

(On Diff #215443)

With %t2 = add i32 %len, 1 the total shift is ((32-len)+(len+1))==33.
The widest type here is i64, and 33 is a valid shift amount there.
But since one of the values being shifted is i32, the result can be constant-folded to false
since you will effectively shift-out all the bits.
https://rise4fun.com/Alive/kk1

With 32 here, the total shift amount will be 64, which is not okay for i64.
But that would already imply UB, and we should be folding it into undef
https://rise4fun.com/Alive/Tzn

I'm trying to keep changes reviewable, so i did not get to that part yet.

The @t9_oneuse5 gets constant-folded too, so there is some coverage.

[spatel]

LGTM

llvm/test/Transforms/InstCombine/shift-amount-reassociation-in-bittest-with-truncation-shl.ll

414 ↗

(On Diff #215443)

Ok - thanks for the explanation.

[lebedev.ri]

Thank you for the review.
I'll follow-up here with a patch for trunc-of-lshr (finally, the one i need).

llvm/trunk/include/llvm/IR/PatternMatch.h

 }
 
 /// Matches Trunc.
 template <typename OpTy>
 inline CastClass_match<OpTy, Instruction::Trunc> m_Trunc(const OpTy &Op) {
   return CastClass_match<OpTy, Instruction::Trunc>(Op);
 }
 
+template <typename OpTy>
+inline match_combine_or<CastClass_match<OpTy, Instruction::Trunc>, OpTy>
+m_TruncOrSelf(const OpTy &Op) {
+  return m_CombineOr(m_Trunc(Op), Op);
+}
+
 /// Matches SExt.
 template <typename OpTy>
 inline CastClass_match<OpTy, Instruction::SExt> m_SExt(const OpTy &Op) {
   return CastClass_match<OpTy, Instruction::SExt>(Op);
 }
 
 /// Matches ZExt.
 template <typename OpTy>

llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp

foldICmpWithTruncSignExtendedVal(ICmpInst &I,
   return T1;
 }
 
 // Given pattern:
 //   icmp eq/ne (and ((x shift Q), (y oppositeshift K))), 0
 // we should move shifts to the same hand of 'and', i.e. rewrite as
 //   icmp eq/ne (and (x shift (Q+K)), y), 0  iff (Q+K) u< bitwidth(x)
 // We are only interested in opposite logical shifts here.
+// One of the shifts can be truncated. For now, it can only be 'shl'.
 // If we can, we want to end up creating 'lshr' shift.
 static Value *
 foldShiftIntoShiftInAnotherHandOfAndInICmp(ICmpInst &I, const SimplifyQuery SQ,
                                            InstCombiner::BuilderTy &Builder) {
   if (!I.isEquality() || !match(I.getOperand(1), m_Zero()) ||
       !I.getOperand(0)->hasOneUse())
     return nullptr;
 
   auto m_AnyLogicalShift = m_LogicalShift(m_Value(), m_Value());
-  auto m_AnyLShr = m_LShr(m_Value(), m_Value());
 
-  // Look for an 'and' of two (opposite) logical shifts.
-  // Pick the single-use shift as XShift.
-  Instruction *XShift, *YShift;
-  if (!match(I.getOperand(0),
-             m_c_And(m_CombineAnd(m_AnyLogicalShift, m_Instruction(XShift)),
-                     m_CombineAnd(m_AnyLogicalShift, m_Instruction(YShift)))))
+  // Look for an 'and' of two logical shifts, one of which may be truncated.
+  // We use m_TruncOrSelf() on the RHS to correctly handle commutative case.
+  Instruction *XShift, *MaybeTruncation, *YShift;
+  if (!match(
+          I.getOperand(0),
+                  m_CombineAnd(m_TruncOrSelf(m_CombineAnd(
+                                   m_AnyLogicalShift, m_Instruction(YShift))),
+                               m_Instruction(MaybeTruncation)))))
+    return nullptr;
+
+  Instruction *UntruncatedShift = XShift;
+
+  // We potentially looked past 'trunc', but only when matching YShift,
+  // therefore YShift must have the widest type.
+  Type *WidestTy = YShift->getType();
+  assert(XShift->getType() == I.getOperand(0)->getType() &&
+         "We did not look past any shifts while matching XShift though.");
+  bool HadTrunc = WidestTy != I.getOperand(0)->getType();
+
+  if (HadTrunc) {
+    // We did indeed have a truncation. For now, let's only proceed if the 'shl'
+    // was truncated, since that does not require any extra legality checks.
+    // FIXME: trunc-of-lshr.
+    if (!match(YShift, m_Shl(m_Value(), m_Value())))
       return nullptr;
+  }
 
   // If YShift is a 'lshr', swap the shifts around.
-  if (match(YShift, m_AnyLShr))
+  if (match(YShift, m_LShr(m_Value(), m_Value())))
     std::swap(XShift, YShift);
 
   // The shifts must be in opposite directions.
   auto XShiftOpcode = XShift->getOpcode();
   if (XShiftOpcode == YShift->getOpcode())
     return nullptr; // Do not care about same-direction shifts here.
 
   Value *X, *XShAmt, *Y, *YShAmt;
-  match(XShift, m_BinOp(m_Value(X), m_Value(XShAmt)));
-  match(YShift, m_BinOp(m_Value(Y), m_Value(YShAmt)));
+  match(XShift, m_BinOp(m_Value(X), m_ZExtOrSelf(m_Value(XShAmt))));
+  match(YShift, m_BinOp(m_Value(Y), m_ZExtOrSelf(m_Value(YShAmt))));
 
   // If one of the values being shifted is a constant, then we will end with
-  // and+icmp, and shift instr will be constant-folded. If they are not,
+  // and+icmp, and [zext+]shift instrs will be constant-folded. If they are not,
   // however, we will need to ensure that we won't increase instruction count.
   if (!isa<Constant>(X) && !isa<Constant>(Y)) {
     // At least one of the hands of the 'and' should be one-use shift.
     if (!match(I.getOperand(0),
                m_c_And(m_OneUse(m_AnyLogicalShift), m_Value())))
       return nullptr;
+    if (HadTrunc) {
+      // Due to the 'trunc', we will need to widen X. For that either the old
+      // 'trunc' or the shift amt in the non-truncated shift should be one-use.
+      if (!MaybeTruncation->hasOneUse() &&
+          !UntruncatedShift->getOperand(1)->hasOneUse())
+        return nullptr;
+    }
   }
 
+  // We have two shift amounts from two different shifts. The types of those
+  // shift amounts may not match. If that's the case let's bailout now.
+  if (XShAmt->getType() != YShAmt->getType())
+    return nullptr;
+
   // Can we fold (XShAmt+YShAmt) ?
-  Value *NewShAmt = SimplifyAddInst(XShAmt, YShAmt, /*IsNSW=*/false,
-                                    /*IsNUW=*/false, SQ.getWithInstruction(&I));
+  auto *NewShAmt = dyn_cast_or_null<Constant>(
+      SimplifyAddInst(XShAmt, YShAmt, /*isNSW=*/false,
+                      /*isNUW=*/false, SQ.getWithInstruction(&I)));
   if (!NewShAmt)
     return nullptr;
   // Is the new shift amount smaller than the bit width?
   // FIXME: could also rely on ConstantRange.
-  unsigned BitWidth = X->getType()->getScalarSizeInBits();
-  if (!match(NewShAmt, m_SpecificInt_ICMP(ICmpInst::Predicate::ICMP_ULT,
-                                          APInt(BitWidth, BitWidth))))
-    return nullptr;
-  // All good, we can do this fold. The shift is the same that was for X.
+  if (!match(NewShAmt, m_SpecificInt_ICMP(
+                           ICmpInst::Predicate::ICMP_ULT,
+                           APInt(NewShAmt->getType()->getScalarSizeInBits(),
+                                 WidestTy->getScalarSizeInBits()))))
+    return nullptr;
+  // All good, we can do this fold.
+  NewShAmt = ConstantExpr::getZExtOrBitCast(NewShAmt, WidestTy);
+  X = Builder.CreateZExt(X, WidestTy);
+  // The shift is the same that was for X.
   Value *T0 = XShiftOpcode == Instruction::BinaryOps::LShr
                   ? Builder.CreateLShr(X, NewShAmt)
                   : Builder.CreateShl(X, NewShAmt);
   Value *T1 = Builder.CreateAnd(T0, Y);
   return Builder.CreateICmp(I.getPredicate(), T1,
-                            Constant::getNullValue(X->getType()));
+                            Constant::getNullValue(WidestTy));
 }
 
 /// Try to fold icmp (binop), X or icmp X, (binop).
 /// TODO: A large part of this logic is duplicated in InstSimplify's
 /// simplifyICmpWithBinOp(). We should be able to share that and avoid the code
 /// duplication.
 Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) {
   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

llvm/trunk/test/Transforms/InstCombine/shift-amount-reassociation-in-bittest-with-truncation-shl.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt %s -instcombine -S | FileCheck %s
 
 ; Given pattern:
 ;   icmp eq/ne (and ((x shift Q), (y oppositeshift K))), 0
 ; we should move shifts to the same hand of 'and', i.e. e.g. rewrite as
 ;   icmp eq/ne (and (((x shift Q) shift K), y)), 0
 ; We are only interested in opposite logical shifts here.
+; We still can handle the case where there is a truncation between a shift
+; and an 'and', but for now only if it's 'shl' - simpler legality check.
 
 ;-------------------------------------------------------------------------------
 ; Basic scalar tests
 ;-------------------------------------------------------------------------------
 
 define i1 @t0_const_after_fold_lshr_shl_ne(i32 %x, i64 %y, i32 %len) {
 ; CHECK-LABEL: @t0_const_after_fold_lshr_shl_ne(
-; CHECK-NEXT:    [[T0:%.*]] = sub i32 32, [[LEN:%.*]]
-; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[X:%.*]], [[T0]]
-; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], -1
-; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
-; CHECK-NEXT:    [[T3:%.*]] = shl i64 [[Y:%.*]], [[T2_WIDE]]
-; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
-; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
-; CHECK-NEXT:    ret i1 [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr i32 [[X:%.*]], 31
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
+; CHECK-NEXT:    [[TMP3:%.*]] = and i64 [[TMP2]], [[Y:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    ret i1 [[TMP4]]
 ;
   %t0 = sub i32 32, %len
   %t1 = lshr i32 %x, %t0
   %t2 = add i32 %len, -1
   %t2_wide = zext i32 %t2 to i64
   %t3 = shl i64 %y, %t2_wide
   %t3_trunc = trunc i64 %t3 to i32
   %t4 = and i32 %t1, %t3_trunc
   %t5 = icmp ne i32 %t4, 0
   ret i1 %t5
 }
 
 ;-------------------------------------------------------------------------------
 ; Very basic vector tests
 ;-------------------------------------------------------------------------------
 
 define <2 x i1> @t1_vec_splat(<2 x i32> %x, <2 x i64> %y, <2 x i32> %len) {
 ; CHECK-LABEL: @t1_vec_splat(
-; CHECK-NEXT:    [[T0:%.*]] = sub <2 x i32> <i32 32, i32 32>, [[LEN:%.*]]
-; CHECK-NEXT:    [[T1:%.*]] = lshr <2 x i32> [[X:%.*]], [[T0]]
-; CHECK-NEXT:    [[T2:%.*]] = add <2 x i32> [[LEN]], <i32 -1, i32 -1>
-; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext <2 x i32> [[T2]] to <2 x i64>
-; CHECK-NEXT:    [[T3:%.*]] = shl <2 x i64> [[Y:%.*]], [[T2_WIDE]]
-; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc <2 x i64> [[T3]] to <2 x i32>
-; CHECK-NEXT:    [[T4:%.*]] = and <2 x i32> [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne <2 x i32> [[T4]], zeroinitializer
-; CHECK-NEXT:    ret <2 x i1> [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr <2 x i32> [[X:%.*]], <i32 31, i32 31>
+; CHECK-NEXT:    [[TMP2:%.*]] = zext <2 x i32> [[TMP1]] to <2 x i64>
+; CHECK-NEXT:    [[TMP3:%.*]] = and <2 x i64> [[TMP2]], [[Y:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <2 x i64> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    ret <2 x i1> [[TMP4]]
 ;
   %t0 = sub <2 x i32> <i32 32, i32 32>, %len
   %t1 = lshr <2 x i32> %x, %t0
   %t2 = add <2 x i32> %len, <i32 -1, i32 -1>
   %t2_wide = zext <2 x i32> %t2 to <2 x i64>
   %t3 = shl <2 x i64> %y, %t2_wide
   %t3_trunc = trunc <2 x i64> %t3 to <2 x i32>
   %t4 = and <2 x i32> %t1, %t3_trunc
   %t5 = icmp ne <2 x i32> %t4, <i32 0, i32 0>
   ret <2 x i1> %t5
 }
 
 define <2 x i1> @t2_vec_nonsplat(<2 x i32> %x, <2 x i64> %y, <2 x i32> %len) {
 ; CHECK-LABEL: @t2_vec_nonsplat(
-; CHECK-NEXT:    [[T0:%.*]] = sub <2 x i32> <i32 30, i32 32>, [[LEN:%.*]]
-; CHECK-NEXT:    [[T1:%.*]] = lshr <2 x i32> [[X:%.*]], [[T0]]
-; CHECK-NEXT:    [[T2:%.*]] = add <2 x i32> [[LEN]], <i32 1, i32 -2>
-; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext <2 x i32> [[T2]] to <2 x i64>
-; CHECK-NEXT:    [[T3:%.*]] = shl <2 x i64> [[Y:%.*]], [[T2_WIDE]]
-; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc <2 x i64> [[T3]] to <2 x i32>
-; CHECK-NEXT:    [[T4:%.*]] = and <2 x i32> [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne <2 x i32> [[T4]], zeroinitializer
-; CHECK-NEXT:    ret <2 x i1> [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = zext <2 x i32> [[X:%.*]] to <2 x i64>
+; CHECK-NEXT:    [[TMP2:%.*]] = lshr <2 x i64> [[TMP1]], <i64 31, i64 30>
+; CHECK-NEXT:    [[TMP3:%.*]] = and <2 x i64> [[TMP2]], [[Y:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <2 x i64> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    ret <2 x i1> [[TMP4]]
 ;
   %t0 = sub <2 x i32> <i32 30, i32 32>, %len
   %t1 = lshr <2 x i32> %x, %t0
   %t2 = add <2 x i32> %len, <i32 1, i32 -2>
   %t2_wide = zext <2 x i32> %t2 to <2 x i64>
   %t3 = shl <2 x i64> %y, %t2_wide
   %t3_trunc = trunc <2 x i64> %t3 to <2 x i32>
   %t4 = and <2 x i32> %t1, %t3_trunc
;
   ret i1 %t5
 }
 
 ; Ok, trunc has no extra uses.
 define i1 @t6_oneuse3(i32 %x, i64 %y, i32 %len) {
 ; CHECK-LABEL: @t6_oneuse3(
 ; CHECK-NEXT:    [[T0:%.*]] = sub i32 32, [[LEN:%.*]]
 ; CHECK-NEXT:    call void @use32(i32 [[T0]])
-; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[X:%.*]], [[T0]]
 ; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], -1
 ; CHECK-NEXT:    call void @use32(i32 [[T2]])
 ; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
 ; CHECK-NEXT:    call void @use64(i64 [[T2_WIDE]])
 ; CHECK-NEXT:    [[T3:%.*]] = shl i64 [[Y:%.*]], [[T2_WIDE]]
 ; CHECK-NEXT:    call void @use64(i64 [[T3]])
-; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
-; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
-; CHECK-NEXT:    ret i1 [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr i32 [[X:%.*]], 31
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
+; CHECK-NEXT:    [[TMP3:%.*]] = and i64 [[TMP2]], [[Y]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    ret i1 [[TMP4]]
 ;
   %t0 = sub i32 32, %len
   call void @use32(i32 %t0)
   %t1 = lshr i32 %x, %t0 ; no extra uses
   %t2 = add i32 %len, -1
   call void @use32(i32 %t2)
   %t2_wide = zext i32 %t2 to i64
   call void @use64(i64 %t2_wide)
   %t3 = shl i64 %y, %t2_wide
   call void @use64(i64 %t3)
   %t3_trunc = trunc i64 %t3 to i32 ; no extra uses
   %t4 = and i32 %t1, %t3_trunc ; no extra uses
   %t5 = icmp ne i32 %t4, 0
   ret i1 %t5
 }
 
 ; Ok, shift amount of non-truncated shift has no extra uses;
 define i1 @t7_oneuse4(i32 %x, i64 %y, i32 %len) {
 ; CHECK-LABEL: @t7_oneuse4(
-; CHECK-NEXT:    [[T0:%.*]] = sub i32 32, [[LEN:%.*]]
-; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[X:%.*]], [[T0]]
-; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], -1
+; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN:%.*]], -1
 ; CHECK-NEXT:    call void @use32(i32 [[T2]])
 ; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
 ; CHECK-NEXT:    call void @use64(i64 [[T2_WIDE]])
 ; CHECK-NEXT:    [[T3:%.*]] = shl i64 [[Y:%.*]], [[T2_WIDE]]
 ; CHECK-NEXT:    call void @use64(i64 [[T3]])
 ; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
 ; CHECK-NEXT:    call void @use32(i32 [[T3_TRUNC]])
-; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
-; CHECK-NEXT:    ret i1 [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = lshr i32 [[X:%.*]], 31
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
+; CHECK-NEXT:    [[TMP3:%.*]] = and i64 [[TMP2]], [[Y]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    ret i1 [[TMP4]]
 ;
   %t0 = sub i32 32, %len ; no extra uses
   %t1 = lshr i32 %x, %t0 ; no extra uses
   %t2 = add i32 %len, -1
   call void @use32(i32 %t2)
   %t2_wide = zext i32 %t2 to i64
   call void @use64(i64 %t2_wide)
   %t3 = shl i64 %y, %t2_wide

 ; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], -1
 ; CHECK-NEXT:    call void @use32(i32 [[T2]])
 ; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
 ; CHECK-NEXT:    call void @use64(i64 [[T2_WIDE]])
 ; CHECK-NEXT:    [[T3:%.*]] = shl i64 [[Y:%.*]], [[T2_WIDE]]
 ; CHECK-NEXT:    call void @use64(i64 [[T3]])
 ; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
 ; CHECK-NEXT:    call void @use32(i32 [[T3_TRUNC]])
-; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
-; CHECK-NEXT:    ret i1 [[T5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = and i64 [[Y]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    ret i1 [[TMP2]]
 ;
   %t0 = sub i32 32, %len
   call void @use32(i32 %t0)
   %t1 = lshr i32 4242424242, %t0 ; shift-of-constant
   call void @use32(i32 %t1)
   %t2 = add i32 %len, -1
   call void @use32(i32 %t2)
   %t2_wide = zext i32 %t2 to i64

 ; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], -1
 ; CHECK-NEXT:    call void @use32(i32 [[T2]])
 ; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
 ; CHECK-NEXT:    call void @use64(i64 [[T2_WIDE]])
 ; CHECK-NEXT:    [[T3:%.*]] = shl i64 4242424242, [[T2_WIDE]]
 ; CHECK-NEXT:    call void @use64(i64 [[T3]])
 ; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
 ; CHECK-NEXT:    call void @use32(i32 [[T3_TRUNC]])
-; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
-; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
-; CHECK-NEXT:    ret i1 [[T5]]
+; CHECK-NEXT:    ret i1 false
 ;
   %t0 = sub i32 32, %len
   call void @use32(i32 %t0)
   %t1 = lshr i32 %x, %t0 ; shift-of-constant
   call void @use32(i32 %t1)
   %t2 = add i32 %len, -1
   call void @use32(i32 %t2)
   %t2_wide = zext i32 %t2 to i64

 ;-------------------------------------------------------------------------------
 ; Negative tests
 ;-------------------------------------------------------------------------------
 
 define i1 @n13_overshift(i32 %x, i64 %y, i32 %len) {
 ; CHECK-LABEL: @n13_overshift(
 ; CHECK-NEXT:    [[T0:%.*]] = sub i32 32, [[LEN:%.*]]
 ; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[X:%.*]], [[T0]]
-; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], 1
+; CHECK-NEXT:    [[T2:%.*]] = add i32 [[LEN]], 32
 ; CHECK-NEXT:    [[T2_WIDE:%.*]] = zext i32 [[T2]] to i64
 ; CHECK-NEXT:    [[T3:%.*]] = shl i64 [[Y:%.*]], [[T2_WIDE]]
 ; CHECK-NEXT:    [[T3_TRUNC:%.*]] = trunc i64 [[T3]] to i32
 ; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T1]], [[T3_TRUNC]]
 ; CHECK-NEXT:    [[T5:%.*]] = icmp ne i32 [[T4]], 0
 ; CHECK-NEXT:    ret i1 [[T5]]
 ;
   %t0 = sub i32 32, %len
   %t1 = lshr i32 %x, %t0
-  %t2 = add i32 %len, 1 ; too much
+  %t2 = add i32 %len, 32 ; too much
   %t2_wide = zext i32 %t2 to i64
   %t3 = shl i64 %y, %t2_wide
   %t3_trunc = trunc i64 %t3 to i32
   %t4 = and i32 %t1, %t3_trunc
   %t5 = icmp ne i32 %t4, 0
   ret i1 %t5
 }