Teach ScalarEvolution to consider loop preheaders in the search for

an if statement that guards a loop, to allow indvars to avoid smax
operations in more situations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56232 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index db9792e3b3e4535ca44eaea239359e7414efcf05..60980d26aafaff1bd930594448a5e003bb1a82fa 100644 (file)
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -1480,6 +1480,12 @@ namespace {
     SCEVHandle HowManyLessThans(SCEV *LHS, SCEV *RHS, const Loop *L,
                                 bool isSigned);
 
+    /// getPredecessorWithUniqueSuccessorForBB - Return a predecessor of BB
+    /// (which may not be an immediate predecessor) which has exactly one
+    /// successor from which BB is reachable, or null if no such block is
+    /// found.
+    BasicBlock* getPredecessorWithUniqueSuccessorForBB(BasicBlock *BB);
+
     /// executesAtLeastOnce - Test whether entry to the loop is protected by
     /// a conditional between LHS and RHS.
     bool executesAtLeastOnce(const Loop *L, bool isSigned, SCEV *LHS, SCEV *RHS);
@@ -1825,6 +1831,11 @@ SCEVHandle ScalarEvolutionsImpl::createSCEV(Value *V) {
   case Instruction::Select:
     // This could be a smax or umax that was lowered earlier.
     // Try to recover it.
+    //
+    // FIXME: This doesn't recognize code like this:
+    //    %t = icmp sgt i32 %n, -1
+    //    %max = select i1 %t, i32 %n, i32 0
+    //
     if (ICmpInst *ICI = dyn_cast<ICmpInst>(U->getOperand(0))) {
       Value *LHS = ICI->getOperand(0);
       Value *RHS = ICI->getOperand(1);
@@ -2703,6 +2714,28 @@ SCEVHandle ScalarEvolutionsImpl::HowFarToNonZero(SCEV *V, const Loop *L) {
   return UnknownValue;
 }
 
+/// getPredecessorWithUniqueSuccessorForBB - Return a predecessor of BB
+/// (which may not be an immediate predecessor) which has exactly one
+/// successor from which BB is reachable, or null if no such block is
+/// found.
+///
+BasicBlock *
+ScalarEvolutionsImpl::getPredecessorWithUniqueSuccessorForBB(BasicBlock *BB) {
+  // If the block has a unique predecessor, the predecessor must have
+  // no other successors from which BB is reachable.
+  if (BasicBlock *Pred = BB->getSinglePredecessor())
+    return Pred;
+
+  // A loop's header is defined to be a block that dominates the loop.
+  // If the loop has a preheader, it must be a block that has exactly
+  // one successor that can reach BB. This is slightly more strict
+  // than necessary, but works if critical edges are split.
+  if (Loop *L = LI.getLoopFor(BB))
+    return L->getLoopPreheader();
+
+  return 0;
+}
+
 /// executesAtLeastOnce - Test whether entry to the loop is protected by
 /// a conditional between LHS and RHS.
 bool ScalarEvolutionsImpl::executesAtLeastOnce(const Loop *L, bool isSigned,
@@ -2711,14 +2744,11 @@ bool ScalarEvolutionsImpl::executesAtLeastOnce(const Loop *L, bool isSigned,
   BasicBlock *PreheaderDest = L->getHeader();
 
   // Starting at the preheader, climb up the predecessor chain, as long as
-  // there are unique predecessors, looking for a conditional branch that
-  // protects the loop.
-  // 
-  // This is a conservative apporoximation of a climb of the
-  // control-dependence predecessors.
-
-  for (; Preheader; PreheaderDest = Preheader,
-                    Preheader = Preheader->getSinglePredecessor()) {
+  // there are predecessors that can be found that have unique successors
+  // leading to the original header.
+  for (; Preheader;
+       PreheaderDest = Preheader,
+       Preheader = getPredecessorWithUniqueSuccessorForBB(Preheader)) {
 
     BranchInst *LoopEntryPredicate =
       dyn_cast<BranchInst>(Preheader->getTerminator());

diff --git a/test/Analysis/ScalarEvolution/avoid-smax-0.ll b/test/Analysis/ScalarEvolution/avoid-smax-0.ll
new file mode 100644 (file)
index 0000000..a8c5aba
--- /dev/null
+++ b/test/Analysis/ScalarEvolution/avoid-smax-0.ll
@@ -0,0 +1,35 @@
+; RUN: llvm-as < %s | opt -scalar-evolution -analyze | grep {Loop bb3: (-1 + %n) iterations!}
+
+; We don't want to use a max in the trip count expression in
+; this testcase.
+
+define void @foo(i32 %n, i32* %p, i32* %q) nounwind {
+entry:
+       icmp sgt i32 %n, 0
+       br i1 %0, label %bb, label %return
+
+bb:
+       load i32* %q, align 4
+       icmp eq i32 %1, 0
+       br i1 %2, label %return, label %bb3.preheader
+
+bb3.preheader:
+       br label %bb3
+
+bb3:
+       %i.0 = phi i32 [ %7, %bb3 ], [ 0, %bb3.preheader ]
+       getelementptr i32* %p, i32 %i.0
+       load i32* %3, align 4
+       add i32 %4, 1
+       getelementptr i32* %p, i32 %i.0
+       store i32 %5, i32* %6, align 4
+       add i32 %i.0, 1
+       icmp slt i32 %7, %n
+       br i1 %8, label %bb3, label %return.loopexit
+
+return.loopexit:
+       br label %return
+
+return:
+       ret void
+}

diff --git a/test/Analysis/ScalarEvolution/avoid-smax-1.ll b/test/Analysis/ScalarEvolution/avoid-smax-1.ll
new file mode 100644 (file)
index 0000000..5fcbf62
--- /dev/null
+++ b/test/Analysis/ScalarEvolution/avoid-smax-1.ll
@@ -0,0 +1,234 @@
+; RUN: llvm-as < %s | opt -indvars | llvm-dis > %t
+; RUN: not grep select %t
+; RUN: grep {icmp ne i32.\* %w } %t
+
+; Indvars should be able to insert a canonical induction variable
+; for the bb6 loop without using a maximum calculation (icmp, select)
+; because it should be able to prove that the comparison is guarded
+; by an appropriate conditional branch.
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i386-apple-darwin9"
+
+define void @foo(i8* %r, i32 %s, i32 %w, i32 %x, i8* %j, i32 %d) nounwind {
+entry:
+       %0 = mul i32 %x, %w             ; <i32> [#uses=2]
+       %1 = mul i32 %x, %w             ; <i32> [#uses=1]
+       %2 = sdiv i32 %1, 4             ; <i32> [#uses=1]
+       %.sum2 = add i32 %2, %0         ; <i32> [#uses=2]
+       %cond = icmp eq i32 %d, 1               ; <i1> [#uses=1]
+       br i1 %cond, label %bb29, label %bb10.preheader
+
+bb10.preheader:                ; preds = %entry
+       %3 = icmp sgt i32 %x, 0         ; <i1> [#uses=1]
+       br i1 %3, label %bb.nph9, label %bb18.loopexit
+
+bb.nph7:               ; preds = %bb7.preheader
+       %4 = mul i32 %y.08, %w          ; <i32> [#uses=1]
+       %5 = mul i32 %y.08, %s          ; <i32> [#uses=1]
+       %6 = add i32 %5, 1              ; <i32> [#uses=1]
+       br label %bb6
+
+bb6:           ; preds = %bb7, %bb.nph7
+       %x.06 = phi i32 [ %13, %bb7 ], [ 0, %bb.nph7 ]          ; <i32> [#uses=3]
+       %7 = add i32 %x.06, %4          ; <i32> [#uses=1]
+       %8 = shl i32 %x.06, 1           ; <i32> [#uses=1]
+       %9 = add i32 %6, %8             ; <i32> [#uses=1]
+       %10 = getelementptr i8* %r, i32 %9              ; <i8*> [#uses=1]
+       %11 = load i8* %10, align 1             ; <i8> [#uses=1]
+       %12 = getelementptr i8* %j, i32 %7              ; <i8*> [#uses=1]
+       store i8 %11, i8* %12, align 1
+       %13 = add i32 %x.06, 1          ; <i32> [#uses=2]
+       br label %bb7
+
+bb7:           ; preds = %bb6
+       %14 = icmp slt i32 %13, %w              ; <i1> [#uses=1]
+       br i1 %14, label %bb6, label %bb7.bb9_crit_edge
+
+bb7.bb9_crit_edge:             ; preds = %bb7
+       br label %bb9
+
+bb9:           ; preds = %bb7.preheader, %bb7.bb9_crit_edge
+       %15 = add i32 %y.08, 1          ; <i32> [#uses=2]
+       br label %bb10
+
+bb10:          ; preds = %bb9
+       %16 = icmp slt i32 %15, %x              ; <i1> [#uses=1]
+       br i1 %16, label %bb7.preheader, label %bb10.bb18.loopexit_crit_edge
+
+bb10.bb18.loopexit_crit_edge:          ; preds = %bb10
+       br label %bb10.bb18.loopexit_crit_edge.split
+
+bb10.bb18.loopexit_crit_edge.split:            ; preds = %bb.nph9, %bb10.bb18.loopexit_crit_edge
+       br label %bb18.loopexit
+
+bb.nph9:               ; preds = %bb10.preheader
+       %17 = icmp sgt i32 %w, 0                ; <i1> [#uses=1]
+       br i1 %17, label %bb.nph9.split, label %bb10.bb18.loopexit_crit_edge.split
+
+bb.nph9.split:         ; preds = %bb.nph9
+       br label %bb7.preheader
+
+bb7.preheader:         ; preds = %bb.nph9.split, %bb10
+       %y.08 = phi i32 [ %15, %bb10 ], [ 0, %bb.nph9.split ]           ; <i32> [#uses=3]
+       br i1 true, label %bb.nph7, label %bb9
+
+bb.nph5:               ; preds = %bb18.loopexit
+       %18 = sdiv i32 %w, 2            ; <i32> [#uses=1]
+       %19 = icmp slt i32 %w, 2                ; <i1> [#uses=1]
+       %20 = sdiv i32 %x, 2            ; <i32> [#uses=1]
+       br i1 %19, label %bb18.bb20_crit_edge.split, label %bb.nph5.split
+
+bb.nph5.split:         ; preds = %bb.nph5
+       br label %bb13
+
+bb13:          ; preds = %bb18, %bb.nph5.split
+       %y.14 = phi i32 [ %42, %bb18 ], [ 0, %bb.nph5.split ]           ; <i32> [#uses=4]
+       %21 = mul i32 %18, %y.14                ; <i32> [#uses=2]
+       %22 = shl i32 %y.14, 1          ; <i32> [#uses=1]
+       %23 = srem i32 %y.14, 2         ; <i32> [#uses=1]
+       %24 = add i32 %23, %22          ; <i32> [#uses=1]
+       %25 = mul i32 %24, %s           ; <i32> [#uses=2]
+       br i1 true, label %bb.nph3, label %bb17
+
+bb.nph3:               ; preds = %bb13
+       %26 = add i32 %21, %0           ; <i32> [#uses=1]
+       %27 = add i32 %21, %.sum2               ; <i32> [#uses=1]
+       %28 = sdiv i32 %w, 2            ; <i32> [#uses=1]
+       br label %bb14
+
+bb14:          ; preds = %bb15, %bb.nph3
+       %x.12 = phi i32 [ %40, %bb15 ], [ 0, %bb.nph3 ]         ; <i32> [#uses=5]
+       %29 = shl i32 %x.12, 2          ; <i32> [#uses=1]
+       %30 = add i32 %29, %25          ; <i32> [#uses=1]
+       %31 = getelementptr i8* %r, i32 %30             ; <i8*> [#uses=1]
+       %32 = load i8* %31, align 1             ; <i8> [#uses=1]
+       %.sum = add i32 %26, %x.12              ; <i32> [#uses=1]
+       %33 = getelementptr i8* %j, i32 %.sum           ; <i8*> [#uses=1]
+       store i8 %32, i8* %33, align 1
+       %34 = shl i32 %x.12, 2          ; <i32> [#uses=1]
+       %35 = or i32 %34, 2             ; <i32> [#uses=1]
+       %36 = add i32 %35, %25          ; <i32> [#uses=1]
+       %37 = getelementptr i8* %r, i32 %36             ; <i8*> [#uses=1]
+       %38 = load i8* %37, align 1             ; <i8> [#uses=1]
+       %.sum6 = add i32 %27, %x.12             ; <i32> [#uses=1]
+       %39 = getelementptr i8* %j, i32 %.sum6          ; <i8*> [#uses=1]
+       store i8 %38, i8* %39, align 1
+       %40 = add i32 %x.12, 1          ; <i32> [#uses=2]
+       br label %bb15
+
+bb15:          ; preds = %bb14
+       %41 = icmp sgt i32 %28, %40             ; <i1> [#uses=1]
+       br i1 %41, label %bb14, label %bb15.bb17_crit_edge
+
+bb15.bb17_crit_edge:           ; preds = %bb15
+       br label %bb17
+
+bb17:          ; preds = %bb15.bb17_crit_edge, %bb13
+       %42 = add i32 %y.14, 1          ; <i32> [#uses=2]
+       br label %bb18
+
+bb18.loopexit:         ; preds = %bb10.bb18.loopexit_crit_edge.split, %bb10.preheader
+       %43 = icmp slt i32 %x, 2                ; <i1> [#uses=1]
+       br i1 %43, label %bb20, label %bb.nph5
+
+bb18:          ; preds = %bb17
+       %44 = icmp sgt i32 %20, %42             ; <i1> [#uses=1]
+       br i1 %44, label %bb13, label %bb18.bb20_crit_edge
+
+bb18.bb20_crit_edge:           ; preds = %bb18
+       br label %bb18.bb20_crit_edge.split
+
+bb18.bb20_crit_edge.split:             ; preds = %bb18.bb20_crit_edge, %bb.nph5
+       br label %bb20
+
+bb20:          ; preds = %bb18.bb20_crit_edge.split, %bb18.loopexit
+       switch i32 %d, label %return [
+               i32 3, label %bb22
+               i32 1, label %bb29
+       ]
+
+bb22:          ; preds = %bb20
+       %45 = mul i32 %x, %w            ; <i32> [#uses=1]
+       %46 = sdiv i32 %45, 4           ; <i32> [#uses=1]
+       %.sum3 = add i32 %46, %.sum2            ; <i32> [#uses=2]
+       %47 = add i32 %x, 15            ; <i32> [#uses=1]
+       %48 = and i32 %47, -16          ; <i32> [#uses=1]
+       %49 = add i32 %w, 15            ; <i32> [#uses=1]
+       %50 = and i32 %49, -16          ; <i32> [#uses=1]
+       %51 = mul i32 %48, %s           ; <i32> [#uses=1]
+       %52 = icmp sgt i32 %x, 0                ; <i1> [#uses=1]
+       br i1 %52, label %bb.nph, label %bb26
+
+bb.nph:                ; preds = %bb22
+       br label %bb23
+
+bb23:          ; preds = %bb24, %bb.nph
+       %y.21 = phi i32 [ %57, %bb24 ], [ 0, %bb.nph ]          ; <i32> [#uses=3]
+       %53 = mul i32 %y.21, %50                ; <i32> [#uses=1]
+       %.sum1 = add i32 %53, %51               ; <i32> [#uses=1]
+       %54 = getelementptr i8* %r, i32 %.sum1          ; <i8*> [#uses=1]
+       %55 = mul i32 %y.21, %w         ; <i32> [#uses=1]
+       %.sum5 = add i32 %55, %.sum3            ; <i32> [#uses=1]
+       %56 = getelementptr i8* %j, i32 %.sum5          ; <i8*> [#uses=1]
+       tail call void @llvm.memcpy.i32(i8* %56, i8* %54, i32 %w, i32 1)
+       %57 = add i32 %y.21, 1          ; <i32> [#uses=2]
+       br label %bb24
+
+bb24:          ; preds = %bb23
+       %58 = icmp slt i32 %57, %x              ; <i1> [#uses=1]
+       br i1 %58, label %bb23, label %bb24.bb26_crit_edge
+
+bb24.bb26_crit_edge:           ; preds = %bb24
+       br label %bb26
+
+bb26:          ; preds = %bb24.bb26_crit_edge, %bb22
+       %59 = mul i32 %x, %w            ; <i32> [#uses=1]
+       %.sum4 = add i32 %.sum3, %59            ; <i32> [#uses=1]
+       %60 = getelementptr i8* %j, i32 %.sum4          ; <i8*> [#uses=1]
+       %61 = mul i32 %x, %w            ; <i32> [#uses=1]
+       %62 = sdiv i32 %61, 2           ; <i32> [#uses=1]
+       tail call void @llvm.memset.i32(i8* %60, i8 -128, i32 %62, i32 1)
+       ret void
+
+bb29:          ; preds = %bb20, %entry
+       %63 = add i32 %w, 15            ; <i32> [#uses=1]
+       %64 = and i32 %63, -16          ; <i32> [#uses=1]
+       %65 = icmp sgt i32 %x, 0                ; <i1> [#uses=1]
+       br i1 %65, label %bb.nph11, label %bb33
+
+bb.nph11:              ; preds = %bb29
+       br label %bb30
+
+bb30:          ; preds = %bb31, %bb.nph11
+       %y.310 = phi i32 [ %70, %bb31 ], [ 0, %bb.nph11 ]               ; <i32> [#uses=3]
+       %66 = mul i32 %y.310, %64               ; <i32> [#uses=1]
+       %67 = getelementptr i8* %r, i32 %66             ; <i8*> [#uses=1]
+       %68 = mul i32 %y.310, %w                ; <i32> [#uses=1]
+       %69 = getelementptr i8* %j, i32 %68             ; <i8*> [#uses=1]
+       tail call void @llvm.memcpy.i32(i8* %69, i8* %67, i32 %w, i32 1)
+       %70 = add i32 %y.310, 1         ; <i32> [#uses=2]
+       br label %bb31
+
+bb31:          ; preds = %bb30
+       %71 = icmp slt i32 %70, %x              ; <i1> [#uses=1]
+       br i1 %71, label %bb30, label %bb31.bb33_crit_edge
+
+bb31.bb33_crit_edge:           ; preds = %bb31
+       br label %bb33
+
+bb33:          ; preds = %bb31.bb33_crit_edge, %bb29
+       %72 = mul i32 %x, %w            ; <i32> [#uses=1]
+       %73 = getelementptr i8* %j, i32 %72             ; <i8*> [#uses=1]
+       %74 = mul i32 %x, %w            ; <i32> [#uses=1]
+       %75 = sdiv i32 %74, 2           ; <i32> [#uses=1]
+       tail call void @llvm.memset.i32(i8* %73, i8 -128, i32 %75, i32 1)
+       ret void
+
+return:                ; preds = %bb20
+       ret void
+}
+
+declare void @llvm.memcpy.i32(i8*, i8*, i32, i32) nounwind
+
+declare void @llvm.memset.i32(i8*, i8, i32, i32) nounwind

diff --git a/test/Analysis/ScalarEvolution/avoid-smax.ll b/test/Analysis/ScalarEvolution/avoid-smax.ll
deleted file mode 100644 (file)
index a8c5aba..0000000
--- a/test/Analysis/ScalarEvolution/avoid-smax.ll
+++ /dev/null
@@ -1,35 +0,0 @@
-; RUN: llvm-as < %s | opt -scalar-evolution -analyze | grep {Loop bb3: (-1 + %n) iterations!}
-
-; We don't want to use a max in the trip count expression in
-; this testcase.
-
-define void @foo(i32 %n, i32* %p, i32* %q) nounwind {
-entry:
-       icmp sgt i32 %n, 0
-       br i1 %0, label %bb, label %return
-
-bb:
-       load i32* %q, align 4
-       icmp eq i32 %1, 0
-       br i1 %2, label %return, label %bb3.preheader
-
-bb3.preheader:
-       br label %bb3
-
-bb3:
-       %i.0 = phi i32 [ %7, %bb3 ], [ 0, %bb3.preheader ]
-       getelementptr i32* %p, i32 %i.0
-       load i32* %3, align 4
-       add i32 %4, 1
-       getelementptr i32* %p, i32 %i.0
-       store i32 %5, i32* %6, align 4
-       add i32 %i.0, 1
-       icmp slt i32 %7, %n
-       br i1 %8, label %bb3, label %return.loopexit
-
-return.loopexit:
-       br label %return
-
-return:
-       ret void
-}