Invert the logic of reachesChainWithoutSideEffects(). What we want to check is that there is

NO path to the destination containing side effects, not that SOME path contains no side effects.
In  practice, this only manifests with CombinerAA enabled, because otherwise the chain has little
to no branching, so "any" is effectively equivalent to "all".

llvm-svn: 114268

2 files changed, 33 insertions, 7 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index ad06ebda5b0..77c40afe697 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5478,9 +5478,9 @@ bool SDNode::isOperandOf(SDNode *N) const {
 
 /// reachesChainWithoutSideEffects - Return true if this operand (which must
 /// be a chain) reaches the specified operand without crossing any
-/// side-effecting instructions.  In practice, this looks through token
-/// factors and non-volatile loads.  In order to remain efficient, this only
-/// looks a couple of nodes in, it does not do an exhaustive search.
+/// side-effecting instructions on any chain path.  In practice, this looks 
+/// through token factors and non-volatile loads.  In order to remain efficient, 
+/// this only looks a couple of nodes in, it does not do an exhaustive search.
 bool SDValue::reachesChainWithoutSideEffects(SDValue Dest,
                                                unsigned Depth) const {
   if (*this == Dest) return true;
@@ -5490,12 +5490,12 @@ bool SDValue::reachesChainWithoutSideEffects(SDValue Dest,
   if (Depth == 0) return false;
 
   // If this is a token factor, all inputs to the TF happen in parallel.  If any
-  // of the operands of the TF reach dest, then we can do the xform.
+  // of the operands of the TF does not reach dest, then we cannot do the xform.
   if (getOpcode() == ISD::TokenFactor) {
     for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
-      if (getOperand(i).reachesChainWithoutSideEffects(Dest, Depth-1))
-        return true;
-    return false;
+      if (!getOperand(i).reachesChainWithoutSideEffects(Dest, Depth-1))
+        return false;
+    return true;
   }
 
   // Loads don't have side effects, look through them.
diff --git a/llvm/test/CodeGen/X86/2010-09-17-SideEffectsInChain.ll b/llvm/test/CodeGen/X86/2010-09-17-SideEffectsInChain.ll
new file mode 100644
index 00000000000..35434ffafaf
--- /dev/null
+++ b/llvm/test/CodeGen/X86/2010-09-17-SideEffectsInChain.ll
@@ -0,0 +1,26 @@
+; RUN: llc < %s -combiner-alias-analysis -march=x86-64 | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-darwin10.4"
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+
+define fastcc i32 @cli_magic_scandesc(i8* %in) nounwind ssp {
+entry:
+  %a = alloca [64 x i8]
+  %b = getelementptr inbounds [64 x i8]* %a, i64 0, i32 0
+  %c = getelementptr inbounds [64 x i8]* %a, i64 0, i32 30
+  %d = load i8* %b, align 8
+  %e = load i8* %c, align 8
+  %f = bitcast [64 x i8]* %a to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %f, i8* %in, i64 64, i32 8, i1 false) nounwind
+  store i8 %d, i8* %b, align 8
+  store i8 %e, i8* %c, align 8
+  ret i32 0
+}
+
+; CHECK: movq	___stack_chk_guard@GOTPCREL(%rip), %rax
+; CHECK: movb	(%rsp), %dl
+; CHECK-NEXT: movb	30(%rsp), %sil
+; CHECK: movb	%dl, (%rsp)
+; CHECK-NEXT: movb	%sil, 30(%rsp)
+; CHECK: callq	___stack_chk_fail