Add dominance check for the instruction being hoisted.

For example, MachineLICM should not hoist a load that is not guaranteed to be executed.
Radar 10254254.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141569 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/MachineLICM.cpp

@@ -168,6 +168,11 @@ namespace {
     /// 
     bool IsLoopInvariantInst(MachineInstr &I);
 
+    /// IsGuaranteedToExecute - check to make sure that the MI dominates
+    /// all of the exit blocks.  If it doesn't, then there is a path out of the
+    /// loop which does not execute this instruction, so we can't hoist it.
+    bool IsGuaranteedToExecute(MachineInstr *MI);
+
     /// HasAnyPHIUse - Return true if the specified register is used by any
     /// phi node.
     bool HasAnyPHIUse(unsigned Reg) const;
@@ -1129,6 +1134,29 @@ bool MachineLICM::EliminateCSE(MachineInstr *MI,
   return false;
 }
 
+/// IsGuaranteedToExecute - check to make sure that the instruction dominates
+/// all of the exit blocks.  If it doesn't, then there is a path out of the loop
+/// which does not execute this instruction, so we can't hoist it.
+bool MachineLICM::IsGuaranteedToExecute(MachineInstr *MI) {
+  // If the instruction is in the header block for the loop (which is very
+  // common), it is always guaranteed to dominate the exit blocks.  Since this
+  // is a common case, and can save some work, check it now.
+  if (MI->getParent() == CurLoop->getHeader())
+    return true;
+
+  // Get the exit blocks for the current loop.
+  SmallVector<MachineBasicBlock*, 8> ExitBlocks;
+  CurLoop->getExitingBlocks(ExitBlocks);
+
+  // Verify that the block dominates each of the exit blocks of the loop.
+  for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
+    if (ExitBlocks[i] != CurLoop->getHeader() &&
+	!DT->dominates(MI->getParent(), ExitBlocks[i]))
+      return false;
+
+  return true;
+}
+
 /// Hoist - When an instruction is found to use only loop invariant operands
 /// that are safe to hoist, this instruction is called to do the dirty work.
 ///
@@ -1139,6 +1167,8 @@ bool MachineLICM::Hoist(MachineInstr *MI, MachineBasicBlock *Preheader) {
     MI = ExtractHoistableLoad(MI);
     if (!MI) return false;
   }
+  if (!IsGuaranteedToExecute(MI))
+    return false;
 
   // Now move the instructions to the predecessor, inserting it before any
   // terminator instructions.

test/CodeGen/ARM/lsr-unfolded-offset.ll

@@ -4,12 +4,11 @@
 ; register pressure and therefore spilling. There is more room for improvement
 ; here.
 
-; CHECK: sub sp, #{{32|28|24}}
+; CHECK: sub sp, #{{40|32|28|24}}
 
 ; CHECK: %for.inc
 ; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, #
 ; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, #
-; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, #
 ; CHECK: add
 
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32"

test/CodeGen/X86/licm-dominance.ll

@@ -0,0 +1,56 @@
+; RUN: llc < %s | FileCheck %s
+
+; MachineLICM should check dominance before hoisting instructions.
+; CHECK:	xorb	%cl, %cl
+; CHECK-NEXT:	testb	%cl, %cl
+
+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-macosx10.7.2"
+
+define void @CMSColorWorldCreateParametricData() nounwind uwtable optsize ssp {
+entry:
+  br label %for.body.i
+
+for.body.i:                                       ; preds = %entry
+  br i1 undef, label %for.inc.i, label %land.lhs.true21.i
+
+land.lhs.true21.i:                                ; preds = %for.body.i
+  br i1 undef, label %if.then26.i, label %for.inc.i
+
+if.then26.i:                                      ; preds = %land.lhs.true21.i
+  br i1 undef, label %if.else.i.i, label %if.then.i.i
+
+if.then.i.i:                                      ; preds = %if.then26.i
+  unreachable
+
+if.else.i.i:                                      ; preds = %if.then26.i
+  br i1 undef, label %lor.lhs.false.i.i, label %if.then116.i.i
+
+lor.lhs.false.i.i:                                ; preds = %if.else.i.i
+  br i1 undef, label %lor.lhs.false104.i.i, label %if.then116.i.i
+
+lor.lhs.false104.i.i:                             ; preds = %lor.lhs.false.i.i
+  br i1 undef, label %lor.lhs.false108.i.i, label %if.then116.i.i
+
+lor.lhs.false108.i.i:                             ; preds = %lor.lhs.false104.i.i
+  br i1 undef, label %lor.lhs.false112.i.i, label %if.then116.i.i
+
+lor.lhs.false112.i.i:                             ; preds = %lor.lhs.false108.i.i
+  br i1 undef, label %if.else232.i.i, label %if.then116.i.i
+
+if.then116.i.i:                                   ; preds = %lor.lhs.false112.i.i, %lor.lhs.false108.i.i, %lor.lhs.false104.i.i, %lor.lhs.false.i.i, %if.else.i.i
+  unreachable
+
+if.else232.i.i:                                   ; preds = %lor.lhs.false112.i.i
+  br label %for.inc.i
+
+for.inc.i:                                        ; preds = %if.else232.i.i, %land.lhs.true21.i, %for.body.i
+  %cmp17.i = icmp ult i64 undef, undef
+  br i1 %cmp17.i, label %for.body.i, label %if.end28.i
+
+if.end28.i:                                       ; preds = %for.inc.i, %if.then10.i, %if.then6.i
+  unreachable
+
+createTransformParams.exit:                       ; preds = %land.lhs.true3.i, %if.then.i, %land.lhs.true.i, %entry
+  ret void
+}

test/CodeGen/X86/licm-nested.ll

@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=x86_64-apple-darwin -march=x86-64 < %s -o /dev/null -stats -info-output-file - | grep machine-licm | grep 3
+; RUN: llc -mtriple=x86_64-apple-darwin -march=x86-64 < %s -o /dev/null -stats -info-output-file - | grep machine-licm | grep 2
 
 ; MachineLICM should be able to hoist the symbolic addresses out of
 ; the inner loops.