[IndVarSimplify] Widen loop compare instructions.

This improves other optimizations such as LSR.  A sext may be added to the
compare's other operand, but this can often be hoisted outside of the loop.

llvm-svn: 217953

4 files changed, 208 insertions, 8 deletions

diff --git a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
index 27e5ec8cf40..532666ea7c1 100644
--- a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -762,6 +762,8 @@ protected:
 
   Instruction *WidenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter);
 
+  bool WidenLoopCompare(NarrowIVDefUse DU);
+
   void pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef);
 };
 } // anonymous namespace
@@ -926,6 +928,32 @@ static void truncateIVUse(NarrowIVDefUse DU, DominatorTree *DT) {
   DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, Trunc);
 }
 
+/// If the narrow use is a compare instruction, then widen the compare
+//  (and possibly the other operand).  The extend operation is hoisted into the
+// loop preheader as far as possible.
+bool WidenIV::WidenLoopCompare(NarrowIVDefUse DU) {
+  ICmpInst *Cmp = dyn_cast<ICmpInst>(DU.NarrowUse);
+  if (!Cmp)
+    return false;
+
+  Value *Op = Cmp->getOperand(Cmp->getOperand(0) == DU.NarrowDef ? 1 : 0);
+  unsigned CastWidth = SE->getTypeSizeInBits(Op->getType());
+  unsigned IVWidth = SE->getTypeSizeInBits(WideType);
+  assert (CastWidth <= IVWidth && "Unexpected width while widening compare.");
+
+  // Widen the compare instruction.
+  IRBuilder<> Builder(getInsertPointForUses(DU.NarrowUse, DU.NarrowDef, DT));
+  DU.NarrowUse->replaceUsesOfWith(DU.NarrowDef, DU.WideDef);
+
+  // Widen the other operand of the compare, if necessary.
+  if (CastWidth < IVWidth) {
+    bool IsSigned = CmpInst::isSigned(Cmp->getPredicate());
+    Value *ExtOp = getExtend(Op, WideType, IsSigned, Cmp);
+    DU.NarrowUse->replaceUsesOfWith(Op, ExtOp);
+  }
+  return true;
+}
+
 /// WidenIVUse - Determine whether an individual user of the narrow IV can be
 /// widened. If so, return the wide clone of the user.
 Instruction *WidenIV::WidenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter) {
@@ -993,10 +1021,15 @@ Instruction *WidenIV::WidenIVUse(NarrowIVDefUse DU, SCEVExpander &Rewriter) {
 
   // Does this user itself evaluate to a recurrence after widening?
   const SCEVAddRecExpr *WideAddRec = GetWideRecurrence(DU.NarrowUse);
+  if (!WideAddRec)
+    WideAddRec = GetExtendedOperandRecurrence(DU);
+
   if (!WideAddRec) {
-      WideAddRec = GetExtendedOperandRecurrence(DU);
-  }
-  if (!WideAddRec) {
+    // If use is a loop condition, try to promote the condition instead of
+    // truncating the IV first.
+    if (WidenLoopCompare(DU))
+      return nullptr;
+
     // This user does not evaluate to a recurence after widening, so don't
     // follow it. Instead insert a Trunc to kill off the original use,
     // eventually isolating the original narrow IV so it can be removed.
diff --git a/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll b/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
index 057669277cc..a7023f24314 100644
--- a/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
+++ b/llvm/test/Transforms/IndVarSimplify/no-iv-rewrite.ll
@@ -229,10 +229,11 @@ entry:
 ; loop and the OR instruction is replaced by an ADD keeping the result
 ; equivalent.
 ;
+; CHECK: sext
 ; CHECK: loop:
 ; CHECK: phi i64
 ; CHECK-NOT: sext
-; CHECK: icmp slt i32
+; CHECK: icmp slt i64
 ; CHECK: exit:
 ; CHECK: add i64
 loop:
diff --git a/llvm/test/Transforms/IndVarSimplify/verify-scev.ll b/llvm/test/Transforms/IndVarSimplify/verify-scev.ll
index 019f5830d52..b9ce3d63ec9 100644
--- a/llvm/test/Transforms/IndVarSimplify/verify-scev.ll
+++ b/llvm/test/Transforms/IndVarSimplify/verify-scev.ll
@@ -380,11 +380,11 @@ for.body48:                                       ; preds = %for.inc221, %for.bo
 
 for.body65.lr.ph:                                 ; preds = %for.body48
   %0 = load i32* undef, align 4
+  %1 = sext i32 %0 to i64
   br label %for.body65.us
 
 for.body65.us:                                    ; preds = %for.inc219.us, %for.body65.lr.ph
-  %k.09.us = phi i32 [ %inc.us, %for.inc219.us ], [ 1, %for.body65.lr.ph ]
-  %idxprom66.us = sext i32 %k.09.us to i64
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc219.us ], [ 1, %for.body65.lr.ph ]
   br i1 undef, label %for.inc219.us, label %if.end72.us
 
 if.end72.us:                                      ; preds = %for.body65.us
@@ -406,8 +406,8 @@ for.cond152.us:                                   ; preds = %for.cond152.us, %fo
   br i1 undef, label %for.cond139.loopexit.us, label %for.cond152.us
 
 for.inc219.us:                                    ; preds = %for.cond139.loopexit.us, %if.end110.us, %if.end93.us, %for.body65.us
-  %inc.us = add nsw i32 %k.09.us, 1
-  %cmp64.us = icmp sgt i32 %inc.us, %0
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %cmp64.us = icmp sgt i64 %indvars.iv.next, %1
   br i1 %cmp64.us, label %for.inc221, label %for.body65.us
 
 for.cond139.loopexit.us:                          ; preds = %for.cond152.us
diff --git a/llvm/test/Transforms/IndVarSimplify/widen-loop-comp.ll b/llvm/test/Transforms/IndVarSimplify/widen-loop-comp.ll
new file mode 100644
index 00000000000..84738ebe239
--- /dev/null
+++ b/llvm/test/Transforms/IndVarSimplify/widen-loop-comp.ll
@@ -0,0 +1,166 @@
+; RUN: opt < %s -indvars -S | FileCheck %s
+target triple = "aarch64--linux-gnu"
+
+; Check the loop exit i32 compare instruction and operand are widened to i64
+; instead of truncating IV before its use in the i32 compare instruction.
+
+@idx = common global i32 0, align 4
+@e = common global i32 0, align 4
+@ptr = common global i32* null, align 8
+
+; CHECK-LABEL: @test1
+; CHECK: for.body.lr.ph:
+; CHECK: sext i32
+; CHECK: for.cond:
+; CHECK: icmp slt i64
+; CHECK: for.body:
+; CHECK: phi i64
+
+define i32 @test1() {
+entry:
+  store i32 -1, i32* @idx, align 4
+  %0 = load i32* @e, align 4
+  %cmp4 = icmp slt i32 %0, 0
+  br i1 %cmp4, label %for.end.loopexit, label %for.body.lr.ph
+
+for.body.lr.ph:
+  %1 = load i32** @ptr, align 8
+  %2 = load i32* @e, align 4
+  br label %for.body
+
+for.cond:
+  %inc = add nsw i32 %i.05, 1
+  %cmp = icmp slt i32 %i.05, %2
+  br i1 %cmp, label %for.body, label %for.cond.for.end.loopexit_crit_edge
+
+for.body:
+  %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.cond ]
+  %idxprom = sext i32 %i.05 to i64
+  %arrayidx = getelementptr inbounds i32* %1, i64 %idxprom
+  %3 = load i32* %arrayidx, align 4
+  %tobool = icmp eq i32 %3, 0
+  br i1 %tobool, label %if.then, label %for.cond
+
+if.then:
+  %i.05.lcssa = phi i32 [ %i.05, %for.body ]
+  store i32 %i.05.lcssa, i32* @idx, align 4
+  br label %for.end
+
+for.cond.for.end.loopexit_crit_edge:
+  br label %for.end.loopexit
+
+for.end.loopexit:
+  br label %for.end
+
+for.end:
+  %4 = load i32* @idx, align 4
+  ret i32 %4
+}
+
+; CHECK-LABEL: @test2
+; CHECK: for.body4.us
+; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+; CHECK: %cmp2.us = icmp slt i64
+; CHECK-NOT: %2 = trunc i64 %indvars.iv.next to i32
+; CHECK-NOT: %cmp2.us = icmp slt i32
+
+define void @test2([8 x i8]* %a, i8* %b, i8 %limit) {
+entry:
+  %conv = zext i8 %limit to i32
+  %cmp23 = icmp eq i8 %limit, 0
+  br i1 %cmp23, label %for.cond1.preheader, label %for.cond1.preheader.us
+
+for.cond1.preheader.us:
+  %storemerge5.us = phi i32 [ 0, %entry ], [ %inc14.us, %for.inc13.us ]
+  br i1 true, label %for.body4.lr.ph.us, label %for.inc13.us
+
+for.inc13.us:
+  %inc14.us = add nsw i32 %storemerge5.us, 1
+  %cmp.us = icmp slt i32 %inc14.us, 4
+  br i1 %cmp.us, label %for.cond1.preheader.us, label %for.end
+
+for.body4.us:
+  %storemerge14.us = phi i32 [ 0, %for.body4.lr.ph.us ], [ %inc.us, %for.body4.us ]
+  %idxprom.us = sext i32 %storemerge14.us to i64
+  %arrayidx6.us = getelementptr inbounds [8 x i8]* %a, i64 %idxprom5.us, i64 %idxprom.us
+  %0 = load i8* %arrayidx6.us, align 1
+  %idxprom7.us = zext i8 %0 to i64
+  %arrayidx8.us = getelementptr inbounds i8* %b, i64 %idxprom7.us
+  %1 = load i8* %arrayidx8.us, align 1
+  store i8 %1, i8* %arrayidx6.us, align 1
+  %inc.us = add nsw i32 %storemerge14.us, 1
+  %cmp2.us = icmp slt i32 %inc.us, %conv
+  br i1 %cmp2.us, label %for.body4.us, label %for.inc13.us
+
+for.body4.lr.ph.us:
+  %idxprom5.us = sext i32 %storemerge5.us to i64
+  br label %for.body4.us
+
+for.cond1.preheader:
+  %storemerge5 = phi i32 [ 0, %entry ], [ %inc14, %for.inc13 ]
+  br i1 false, label %for.inc13, label %for.inc13
+
+for.inc13:
+  %inc14 = add nsw i32 %storemerge5, 1
+  %cmp = icmp slt i32 %inc14, 4
+  br i1 %cmp, label %for.cond1.preheader, label %for.end
+
+for.end:
+  ret void
+}
+
+; CHECK-LABEL: @test3
+; CHECK: sext i32 %b
+; CHECK: for.cond:
+; CHECK: phi i64
+; CHECK: icmp slt i64
+
+define i32 @test3(i32* %a, i32 %b) {
+entry:
+  br label %for.cond
+
+for.cond:
+  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
+  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+  %cmp = icmp slt i32 %i.0, %b
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+  %idxprom = sext i32 %i.0 to i64
+  %arrayidx = getelementptr inbounds i32* %a, i64 %idxprom
+  %0 = load i32* %arrayidx, align 4
+  %add = add nsw i32 %sum.0, %0
+  %inc = add nsw i32 %i.0, 1
+  br label %for.cond
+
+for.end:
+  ret i32 %sum.0
+}
+
+; CHECK-LABEL: @test4
+; CHECK: zext i32 %b
+; CHECK: for.cond:
+; CHECK: phi i64
+; CHECK: icmp ule i64
+
+define i32 @test4(i32* %a, i32 %b) {
+entry:
+  br label %for.cond
+
+for.cond:
+  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
+  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+  %cmp = icmp ule i32 %i.0, %b
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+  %idxprom = sext i32 %i.0 to i64
+  %arrayidx = getelementptr inbounds i32* %a, i64 %idxprom
+  %0 = load i32* %arrayidx, align 4
+  %add = add nsw i32 %sum.0, %0
+  %inc = add nsw i32 %i.0, 1
+  br label %for.cond
+
+for.end:
+  ret i32 %sum.0
+}