From: Dan Gohman <gohman@apple.com>
Date: Tue, 12 May 2009 01:27:58 +0000 (+0000)
Subject: When forgetting SCEVs for loop PHIs, don't forget SCEVUnknown values.
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=efb9fbfdab6cab88146792a5db79315313270c7f;p=oota-llvm.git

When forgetting SCEVs for loop PHIs, don't forget SCEVUnknown values.
These values aren't analyzable, so they don't care if more information
about the loop trip count can be had. Also, SCEVUnknown is used for
a PHI while the PHI itself is being analyzed, so it needs to be left
in the Scalars map. This fixes a variety of subtle issues.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71533 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index 5cd9739e223..a71db865cbf 100644
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -2309,10 +2309,20 @@ void ScalarEvolution::forgetLoopBackedgeTakenCount(const Loop *L) {
 void ScalarEvolution::forgetLoopPHIs(const Loop *L) {
   BasicBlock *Header = L->getHeader();
 
+  // Push all Loop-header PHIs onto the Worklist stack, except those
+  // that are presently represented via a SCEVUnknown. SCEVUnknown for
+  // a PHI either means that it has an unrecognized structure, or it's
+  // a PHI that's in the progress of being computed by createNodeForPHI.
+  // In the former case, additional loop trip count information isn't
+  // going to change anything. In the later case, createNodeForPHI will
+  // perform the necessary updates on its own when it gets to that point.
   SmallVector<Instruction *, 16> Worklist;
   for (BasicBlock::iterator I = Header->begin();
-       PHINode *PN = dyn_cast<PHINode>(I); ++I)
-    Worklist.push_back(PN);
+       PHINode *PN = dyn_cast<PHINode>(I); ++I) {
+    std::map<SCEVCallbackVH, SCEVHandle>::iterator It = Scalars.find((Value*)I);
+    if (It != Scalars.end() && !isa<SCEVUnknown>(It->second))
+      Worklist.push_back(PN);
+  }
 
   while (!Worklist.empty()) {
     Instruction *I = Worklist.pop_back_val();
diff --git a/test/Transforms/IndVarSimplify/loop_evaluate_5.ll b/test/Transforms/IndVarSimplify/loop_evaluate_5.ll
new file mode 100644
index 00000000000..c6ce4a2122d
--- /dev/null
+++ b/test/Transforms/IndVarSimplify/loop_evaluate_5.ll
@@ -0,0 +1,32 @@
+; RUN: llvm-as < %s | opt -indvars | llvm-dis | grep {120, %bb2.bb3_crit_edge}
+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"
+target triple = "i686-pc-linux-gnu"
+
+; Indvars should be able to compute an exit value for %tmp1.
+
+define i32 @testcase(i5 zeroext %k) nounwind readnone {
+entry:
+	br i1 false, label %bb3, label %bb.nph
+
+bb.nph:		; preds = %entry
+	br label %bb
+
+bb:		; preds = %bb2, %bb.nph
+	%result2 = phi i32 [ %tmp1, %bb2 ], [ 0, %bb.nph ]		; <i32> [#uses=1]
+	%k_01 = phi i5 [ %indvar_next1, %bb2 ], [ 0, %bb.nph ]		; <i5> [#uses=2]
+	%tmp2 = zext i5 %k_01 to i32		; <i32> [#uses=1]
+	%tmp1 = add i32 %tmp2, %result2		; <i32> [#uses=2]
+	%indvar_next1 = add i5 %k_01, 1		; <i5> [#uses=2]
+	br label %bb2
+
+bb2:		; preds = %bb
+	%phitmp = icmp eq i5 %indvar_next1, -16		; <i1> [#uses=1]
+	br i1 %phitmp, label %bb2.bb3_crit_edge, label %bb
+
+bb2.bb3_crit_edge:		; preds = %bb2
+	br label %bb3
+
+bb3:		; preds = %bb2.bb3_crit_edge, %entry
+	%result.lcssa = phi i32 [ %tmp1, %bb2.bb3_crit_edge ], [ 0, %entry ]		; <i32> [#uses=1]
+	ret i32 %result.lcssa
+}