[llvm] r198795 - Conservatively handle multiple MMOs in MIsNeedChainEdge

[Hal Finkel]

Author: hfinkel
Date: Wed Jan  8 15:52:02 2014
New Revision: 198795

URL: http://llvm.org/viewvc/llvm-project?rev=198795&view=rev
Log:
Conservatively handle multiple MMOs in MIsNeedChainEdge

MIsNeedChainEdge, which is used by -enable-aa-sched-mi (AA in misched), had an
llvm_unreachable when -enable-aa-sched-mi is enabled and we reach an
instruction with multiple MMOs. Instead, return a conservative answer. This
allows testing -enable-aa-sched-mi on x86.

Also, this moves the check above the isUnsafeMemoryObject checks.
isUnsafeMemoryObject is currently correct only for instructions with one MMO
(as noted in the comment in isUnsafeMemoryObject):

  // We purposefully do no check for hasOneMemOperand() here
  // in hope to trigger an assert downstream in order to
  // finish implementation.

The problem with this is that, had the candidate edge passed the
"!MIa->mayStore() && !MIb->mayStore()" check, the hoped-for assert would never
happen (which could, in theory, lead to incorrect behavior if one of these
secondary MMOs was volatile, for example).

Added:
    llvm/trunk/test/CodeGen/X86/misched-aa-mmos.ll
Modified:
    llvm/trunk/lib/CodeGen/ScheduleDAGInstrs.cpp

Modified: llvm/trunk/lib/CodeGen/ScheduleDAGInstrs.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/ScheduleDAGInstrs.cpp?rev=198795&r1=198794&r2=198795&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/ScheduleDAGInstrs.cpp (original)
+++ llvm/trunk/lib/CodeGen/ScheduleDAGInstrs.cpp Wed Jan  8 15:52:02 2014
@@ -512,6 +512,10 @@ static bool MIsNeedChainEdge(AliasAnalys
   if (MIa == MIb)
     return false;
 
+  // FIXME: Need to handle multiple memory operands to support all targets.
+  if (!MIa->hasOneMemOperand() || !MIb->hasOneMemOperand())
+    return true;
+
   if (isUnsafeMemoryObject(MIa, MFI) || isUnsafeMemoryObject(MIb, MFI))
     return true;
 
@@ -527,10 +531,6 @@ static bool MIsNeedChainEdge(AliasAnalys
   MachineMemOperand *MMOa = *MIa->memoperands_begin();
   MachineMemOperand *MMOb = *MIb->memoperands_begin();
 
-  // FIXME: Need to handle multiple memory operands to support all targets.
-  if (!MIa->hasOneMemOperand() || !MIb->hasOneMemOperand())
-    llvm_unreachable("Multiple memory operands.");
-
   // The following interface to AA is fashioned after DAGCombiner::isAlias
   // and operates with MachineMemOperand offset with some important
   // assumptions:

Added: llvm/trunk/test/CodeGen/X86/misched-aa-mmos.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/misched-aa-mmos.ll?rev=198795&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/misched-aa-mmos.ll (added)
+++ llvm/trunk/test/CodeGen/X86/misched-aa-mmos.ll Wed Jan  8 15:52:02 2014
@@ -0,0 +1,37 @@
+; RUN: llc -enable-misched -enable-aa-sched-mi < %s
+
+; This generates a decw instruction, which has two MMOs, and an alias SU edge
+; query involving that instruction. Make sure this does not crash.
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%s1 = type { i16, i16, i32 }
+%c1 = type { %s1*, %u1, i16, i8 }
+%u1 = type { i64 }
+
+declare zeroext i1 @bar(i64*, i32) #5
+
+define i32 @foo() #0 align 2 {
+entry:
+  %temp_rhs = alloca %c1, align 8
+  br i1 undef, label %if.else56, label %cond.end.i
+
+cond.end.i:
+  %significand.i18.i = getelementptr inbounds %c1* %temp_rhs, i64 0, i32 1
+  %exponent.i = getelementptr inbounds %c1* %temp_rhs, i64 0, i32 2
+  %0 = load i16* %exponent.i, align 8
+  %sub.i = add i16 %0, -1
+  store i16 %sub.i, i16* %exponent.i, align 8
+  %parts.i.i = bitcast %u1* %significand.i18.i to i64**
+  %1 = load i64** %parts.i.i, align 8
+  %call5.i = call zeroext i1 @bar(i64* %1, i32 undef) #1
+  unreachable
+
+if.else56:
+  unreachable
+}
+
+attributes #0 = { nounwind uwtable }
+attributes #1 = { nounwind }
+