Fix loop rerolling pass failure with non-consant loop lower bound

The loop rerolling pass was failing with an assertion failure from a
failed cast on loops like this:

  void foo(int *A, int *B, int m, int n) {
    for (int i = m; i < n; i+=4) {
      A[i+0] = B[i+0] * 4;
      A[i+1] = B[i+1] * 4;
      A[i+2] = B[i+2] * 4;
      A[i+3] = B[i+3] * 4;
    }
  }

The code was casting the SCEV-expanded code for the new
induction variable to a phi-node. When the loop had a non-constant
lower bound, the SCEV expander would end the code expansion with an
add insted of a phi node and the cast would fail.

It looks like the cast to a phi node was only needed to get the
induction variable value coming from the backedge to compute the end
of loop condition. This patch changes the loop reroller to compare
the induction variable to the number of times the backedge is taken
instead of the iteration count of the loop. In other words, we stop
the loop when the current value of the induction variable ==
IterationCount-1. Previously, the comparison was comparing the
induction variable value from the next iteration == IterationCount.

This problem only seems to occur on 32-bit targets. For some reason,
the loop is not rerolled on 64-bit targets.

PR18290


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@198425 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Transforms/Scalar/LoopRerollPass.cpp

@@ -1089,9 +1089,8 @@ bool LoopReroll::reroll(Instruction *IV, Loop *L, BasicBlock *Header,
                            L, SCEV::FlagAnyWrap));
   { // Limit the lifetime of SCEVExpander.
     SCEVExpander Expander(*SE, "reroll");
-    PHINode *NewIV =
+    Value *NewIV = Expander.expandCodeFor(H, IV->getType(), Header->begin());
-      cast<PHINode>(Expander.expandCodeFor(H, IV->getType(),
+
-                                           Header->begin()));
     for (DenseSet<Instruction *>::iterator J = BaseUseSet.begin(),
          JE = BaseUseSet.end(); J != JE; ++J)
       (*J)->replaceUsesOfWith(IV, NewIV);
@@ -1102,20 +1101,23 @@ bool LoopReroll::reroll(Instruction *IV, Loop *L, BasicBlock *Header,
         if (Inc == 1)
           ICSCEV =
             SE->getMulExpr(ICSCEV, SE->getConstant(ICSCEV->getType(), Scale));
-        Value *IC;
+        // Iteration count SCEV minus 1
-        if (isa<SCEVConstant>(ICSCEV)) {
+        const SCEV *ICMinus1SCEV =
-          IC = Expander.expandCodeFor(ICSCEV, NewIV->getType(), BI);
+          SE->getMinusSCEV(ICSCEV, SE->getConstant(ICSCEV->getType(), 1));
+
+        Value *ICMinus1; // Iteration count minus 1
+        if (isa<SCEVConstant>(ICMinus1SCEV)) {
+          ICMinus1 = Expander.expandCodeFor(ICMinus1SCEV, NewIV->getType(), BI);
         } else {
           BasicBlock *Preheader = L->getLoopPreheader();
           if (!Preheader)
             Preheader = InsertPreheaderForLoop(L, this);
 
-          IC = Expander.expandCodeFor(ICSCEV, NewIV->getType(),
+          ICMinus1 = Expander.expandCodeFor(ICMinus1SCEV, NewIV->getType(),
-                                      Preheader->getTerminator());
+                                            Preheader->getTerminator());
         }
  
-        Value *NewIVNext = NewIV->getIncomingValueForBlock(Header); 
+        Value *Cond = new ICmpInst(BI, CmpInst::ICMP_EQ, NewIV, ICMinus1,
-        Value *Cond = new ICmpInst(BI, CmpInst::ICMP_EQ, NewIVNext, IC,
                                    "exitcond");
         BI->setCondition(Cond);
 

test/Transforms/LoopReroll/basic.ll

@@ -33,7 +33,7 @@ for.body:                                         ; preds = %for.body, %entry
 ; CHECK: %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %entry ]
 ; CHECK: %call = tail call i32 @foo(i32 %indvar) #1
 ; CHECK: %indvar.next = add i32 %indvar, 1
-; CHECK: %exitcond1 = icmp eq i32 %indvar.next, 498
+; CHECK: %exitcond1 = icmp eq i32 %indvar, 497
 ; CHECK: br i1 %exitcond1, label %for.end, label %for.body
 
 ; CHECK: ret
@@ -83,7 +83,7 @@ for.body:                                         ; preds = %entry, %for.body
 ; CHECK: %arrayidx = getelementptr inbounds i32* %x, i64 %indvar
 ; CHECK: store i32 %call, i32* %arrayidx, align 4
 ; CHECK: %indvar.next = add i64 %indvar, 1
-; CHECK: %exitcond = icmp eq i64 %indvar.next, 1500
+; CHECK: %exitcond = icmp eq i64 %indvar, 1499
 ; CHECK: br i1 %exitcond, label %for.end, label %for.body
 
 ; CHECK: ret
@@ -131,7 +131,7 @@ for.body:                                         ; preds = %for.body, %entry
 ; CHECK: %arrayidx = getelementptr inbounds i32* %x, i64 %indvars.iv
 ; CHECK: store i32 %call, i32* %arrayidx, align 4
 ; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; CHECK: %exitcond1 = icmp eq i64 %indvars.iv.next, 1500
+; CHECK: %exitcond1 = icmp eq i64 %indvars.iv, 1499
 ; CHECK: br i1 %exitcond1, label %for.end, label %for.body
 
 ; CHECK: ret
@@ -213,7 +213,7 @@ for.body:                                         ; preds = %entry, %for.body
 ; CHECK: %add = fadd float %1, %mul
 ; CHECK: store float %add, float* %arrayidx2, align 4
 ; CHECK: %indvar.next = add i64 %indvar, 1
-; CHECK: %exitcond = icmp eq i64 %indvar.next, 3200
+; CHECK: %exitcond = icmp eq i64 %indvar, 3199
 ; CHECK: br i1 %exitcond, label %for.end, label %for.body
 
 ; CHECK: ret
@@ -313,7 +313,7 @@ for.body:                                         ; preds = %entry, %for.body
 ; CHECK: %add = fadd float %2, %mul
 ; CHECK: store float %add, float* %arrayidx4, align 4
 ; CHECK: %indvar.next = add i64 %indvar, 1
-; CHECK: %exitcond = icmp eq i64 %indvar.next, 3200
+; CHECK: %exitcond = icmp eq i64 %indvar, 3199
 ; CHECK: br i1 %exitcond, label %for.end, label %for.body
 
 ; CHECK: ret

test/Transforms/LoopReroll/nonconst_lb.ll

@@ -0,0 +1,152 @@
+; RUN: opt < %s -loop-reroll -S | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
+target triple = "thumbv7-none-linux"
+
+;void foo(int *A, int *B, int m, int n) {
+;  for (int i = m; i < n; i+=4) {
+;    A[i+0] = B[i+0] * 4;
+;    A[i+1] = B[i+1] * 4;
+;    A[i+2] = B[i+2] * 4;
+;    A[i+3] = B[i+3] * 4;
+;  }
+;}
+define void @foo(i32* nocapture %A, i32* nocapture readonly %B, i32 %m, i32 %n) {
+entry:
+  %cmp34 = icmp slt i32 %m, %n
+  br i1 %cmp34, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.035 = phi i32 [ %add18, %for.body ], [ %m, %entry ]
+  %arrayidx = getelementptr inbounds i32* %B, i32 %i.035
+  %0 = load i32* %arrayidx, align 4
+  %mul = shl nsw i32 %0, 2
+  %arrayidx2 = getelementptr inbounds i32* %A, i32 %i.035
+  store i32 %mul, i32* %arrayidx2, align 4
+  %add3 = add nsw i32 %i.035, 1
+  %arrayidx4 = getelementptr inbounds i32* %B, i32 %add3
+  %1 = load i32* %arrayidx4, align 4
+  %mul5 = shl nsw i32 %1, 2
+  %arrayidx7 = getelementptr inbounds i32* %A, i32 %add3
+  store i32 %mul5, i32* %arrayidx7, align 4
+  %add8 = add nsw i32 %i.035, 2
+  %arrayidx9 = getelementptr inbounds i32* %B, i32 %add8
+  %2 = load i32* %arrayidx9, align 4
+  %mul10 = shl nsw i32 %2, 2
+  %arrayidx12 = getelementptr inbounds i32* %A, i32 %add8
+  store i32 %mul10, i32* %arrayidx12, align 4
+  %add13 = add nsw i32 %i.035, 3
+  %arrayidx14 = getelementptr inbounds i32* %B, i32 %add13
+  %3 = load i32* %arrayidx14, align 4
+  %mul15 = shl nsw i32 %3, 2
+  %arrayidx17 = getelementptr inbounds i32* %A, i32 %add13
+  store i32 %mul15, i32* %arrayidx17, align 4
+  %add18 = add nsw i32 %i.035, 4
+  %cmp = icmp slt i32 %add18, %n
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+; CHECK-LABEL: @foo
+; CHECK: for.body.preheader:                               ; preds = %entry
+; CHECK:   %0 = add i32 %n, -1
+; CHECK:   %1 = sub i32 %0, %m
+; CHECK:   %2 = lshr i32 %1, 2
+; CHECK:   %3 = mul i32 %2, 4
+; CHECK:   %4 = add i32 %m, %3
+; CHECK:   %5 = add i32 %4, 3
+; CHECK:   br label %for.body
+
+; CHECK: for.body:                                         ; preds = %for.body, %for.body.preheader
+; CHECK:   %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %for.body.preheader ]
+; CHECK:   %6 = add i32 %m, %indvar
+; CHECK:   %arrayidx = getelementptr inbounds i32* %B, i32 %6
+; CHECK:   %7 = load i32* %arrayidx, align 4
+; CHECK:   %mul = shl nsw i32 %7, 2
+; CHECK:   %arrayidx2 = getelementptr inbounds i32* %A, i32 %6
+; CHECK:   store i32 %mul, i32* %arrayidx2, align 4
+; CHECK:   %indvar.next = add i32 %indvar, 1
+; CHECK:   %exitcond = icmp eq i32 %6, %5
+; CHECK:   br i1 %exitcond, label %for.end, label %for.body
+
+;void daxpy_ur(int n,float da,float *dx,float *dy)
+;    {
+;    int m = n % 4;
+;    for (int i = m; i < n; i = i + 4)
+;        {
+;        dy[i]   = dy[i]   + da*dx[i];
+;        dy[i+1] = dy[i+1] + da*dx[i+1];
+;        dy[i+2] = dy[i+2] + da*dx[i+2];
+;        dy[i+3] = dy[i+3] + da*dx[i+3];
+;        }
+;    }
+define void @daxpy_ur(i32 %n, float %da, float* nocapture readonly %dx, float* nocapture %dy) {
+entry:
+  %rem = srem i32 %n, 4
+  %cmp55 = icmp slt i32 %rem, %n
+  br i1 %cmp55, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.056 = phi i32 [ %add27, %for.body ], [ %rem, %entry ]
+  %arrayidx = getelementptr inbounds float* %dy, i32 %i.056
+  %0 = load float* %arrayidx, align 4
+  %arrayidx1 = getelementptr inbounds float* %dx, i32 %i.056
+  %1 = load float* %arrayidx1, align 4
+  %mul = fmul float %1, %da
+  %add = fadd float %0, %mul
+  store float %add, float* %arrayidx, align 4
+  %add3 = add nsw i32 %i.056, 1
+  %arrayidx4 = getelementptr inbounds float* %dy, i32 %add3
+  %2 = load float* %arrayidx4, align 4
+  %arrayidx6 = getelementptr inbounds float* %dx, i32 %add3
+  %3 = load float* %arrayidx6, align 4
+  %mul7 = fmul float %3, %da
+  %add8 = fadd float %2, %mul7
+  store float %add8, float* %arrayidx4, align 4
+  %add11 = add nsw i32 %i.056, 2
+  %arrayidx12 = getelementptr inbounds float* %dy, i32 %add11
+  %4 = load float* %arrayidx12, align 4
+  %arrayidx14 = getelementptr inbounds float* %dx, i32 %add11
+  %5 = load float* %arrayidx14, align 4
+  %mul15 = fmul float %5, %da
+  %add16 = fadd float %4, %mul15
+  store float %add16, float* %arrayidx12, align 4
+  %add19 = add nsw i32 %i.056, 3
+  %arrayidx20 = getelementptr inbounds float* %dy, i32 %add19
+  %6 = load float* %arrayidx20, align 4
+  %arrayidx22 = getelementptr inbounds float* %dx, i32 %add19
+  %7 = load float* %arrayidx22, align 4
+  %mul23 = fmul float %7, %da
+  %add24 = fadd float %6, %mul23
+  store float %add24, float* %arrayidx20, align 4
+  %add27 = add nsw i32 %i.056, 4
+  %cmp = icmp slt i32 %add27, %n
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+; CHECK-LABEL: @daxpy_ur
+; CHECK: for.body.preheader:
+; CHECK:   %0 = add i32 %n, -1
+; CHECK:   %1 = sub i32 %0, %rem
+; CHECK:   %2 = lshr i32 %1, 2
+; CHECK:   %3 = mul i32 %2, 4
+; CHECK:   %4 = add i32 %rem, %3
+; CHECK:   %5 = add i32 %4, 3
+; CHECK:   br label %for.body
+
+; CHECK: for.body:
+; CHECK:   %indvar = phi i32 [ %indvar.next, %for.body ], [ 0, %for.body.preheader ]
+; CHECK:   %6 = add i32 %rem, %indvar
+; CHECK:   %arrayidx = getelementptr inbounds float* %dy, i32 %6
+; CHECK:   %7 = load float* %arrayidx, align 4
+; CHECK:   %arrayidx1 = getelementptr inbounds float* %dx, i32 %6
+; CHECK:   %8 = load float* %arrayidx1, align 4
+; CHECK:   %mul = fmul float %8, %da
+; CHECK:   %add = fadd float %7, %mul
+; CHECK:   store float %add, float* %arrayidx, align 4
+; CHECK:   %indvar.next = add i32 %indvar, 1
+; CHECK:   %exitcond = icmp eq i32 %6, %5
+; CHECK:   br i1 %exitcond, label %for.end, label %for.body

test/Transforms/LoopReroll/reduction.ll

@@ -38,7 +38,7 @@ for.body:                                         ; preds = %entry, %for.body
 ; CHECK: %0 = load i32* %arrayidx, align 4
 ; CHECK: %add = add nsw i32 %0, %r.029
 ; CHECK: %indvar.next = add i64 %indvar, 1
-; CHECK: %exitcond = icmp eq i64 %indvar.next, 400
+; CHECK: %exitcond = icmp eq i64 %indvar, 399
 ; CHECK: br i1 %exitcond, label %for.end, label %for.body
 
 ; CHECK: ret
@@ -83,7 +83,7 @@ for.body:                                         ; preds = %entry, %for.body
 ; CHECK: %0 = load float* %arrayidx, align 4
 ; CHECK: %add = fadd float %0, %r.029
 ; CHECK: %indvar.next = add i64 %indvar, 1
-; CHECK: %exitcond = icmp eq i64 %indvar.next, 400
+; CHECK: %exitcond = icmp eq i64 %indvar, 399
 ; CHECK: br i1 %exitcond, label %for.end, label %for.body
 
 ; CHECK: ret