Fix machine-cp by having it to check sub-register indicies. e.g.

ecx = mov eax
al  = mov ch
The second copy is not a nop because the sub-indices of ecx,ch is not the
same of that of eax/al.

Re-enabled machine-cp.
PR11940


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151002 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/MachineCopyPropagation.cpp

@@ -99,6 +99,31 @@ static bool NoInterveningSideEffect(const MachineInstr *CopyMI,
   return true;
 }
 
+/// isNopCopy - Return true if the specified copy is really a nop. That is
+/// if the source of the copy is the same of the definition of the copy that
+/// supplied the source. If the source of the copy is a sub-register than it
+/// must check the sub-indices match. e.g.
+/// ecx = mov eax
+/// al  = mov cl
+/// But not
+/// ecx = mov eax
+/// al  = mov ch
+static bool isNopCopy(MachineInstr *CopyMI, unsigned Def, unsigned Src,
+                      const TargetRegisterInfo *TRI) {
+  unsigned SrcSrc = CopyMI->getOperand(1).getReg();
+  if (Def == SrcSrc)
+    return true;
+  if (TRI->isSubRegister(SrcSrc, Def)) {
+    unsigned SrcDef = CopyMI->getOperand(0).getReg();
+    unsigned SubIdx = TRI->getSubRegIndex(SrcSrc, Def);
+    if (!SubIdx)
+      return false;
+    return SubIdx == TRI->getSubRegIndex(SrcDef, Src);
+  }
+
+  return false;
+}
+
 bool MachineCopyPropagation::CopyPropagateBlock(MachineBasicBlock &MBB) {
   SmallSetVector<MachineInstr*, 8> MaybeDeadCopies; // Candidates for deletion
   DenseMap<unsigned, MachineInstr*> AvailCopyMap;   // Def -> available copies map
@@ -122,10 +147,9 @@ bool MachineCopyPropagation::CopyPropagateBlock(MachineBasicBlock &MBB) {
       DenseMap<unsigned, MachineInstr*>::iterator CI = AvailCopyMap.find(Src);
       if (CI != AvailCopyMap.end()) {
         MachineInstr *CopyMI = CI->second;
-        unsigned SrcSrc = CopyMI->getOperand(1).getReg();
         if (!ReservedRegs.test(Def) &&
             (!ReservedRegs.test(Src) || NoInterveningSideEffect(CopyMI, MI)) &&
-            (SrcSrc == Def || TRI->isSubRegister(SrcSrc, Def))) {
+            isNopCopy(CopyMI, Def, Src, TRI)) {
           // The two copies cancel out and the source of the first copy
           // hasn't been overridden, eliminate the second one. e.g.
           //  %ECX<def> = COPY %EAX<kill>

lib/CodeGen/Passes.cpp

@@ -69,7 +69,7 @@ static cl::opt<bool> DisableLSR("disable-lsr", cl::Hidden,
 static cl::opt<bool> DisableCGP("disable-cgp", cl::Hidden,
     cl::desc("Disable Codegen Prepare"));
 static cl::opt<bool> DisableCopyProp("disable-copyprop", cl::Hidden,
-    cl::desc("Disable Copy Propagation pass"), cl::init(true)); // PR11940
+    cl::desc("Disable Copy Propagation pass"));
 static cl::opt<bool> PrintLSR("print-lsr-output", cl::Hidden,
     cl::desc("Print LLVM IR produced by the loop-reduce pass"));
 static cl::opt<bool> PrintISelInput("print-isel-input", cl::Hidden,

test/CodeGen/X86/2012-02-20-MachineCPBug.ll

@@ -0,0 +1,77 @@
+; RUN: llc < %s -mtriple=i386-apple-macosx | FileCheck %s
+; PR11940: Do not optimize away movb %al, %ch
+
+%struct.APInt = type { i64* }
+
+declare noalias i8* @calloc(i32, i32) nounwind
+
+define void @bug(%struct.APInt* noalias nocapture sret %agg.result, %struct.APInt* nocapture %this, i32 %rotateAmt) nounwind align 2 {
+entry:
+; CHECK: bug:
+  %call = tail call i8* @calloc(i32 1, i32 32)
+  %call.i = tail call i8* @calloc(i32 1, i32 32) nounwind
+  %0 = bitcast i8* %call.i to i64*
+  %rem.i = and i32 %rotateAmt, 63
+  %div.i = lshr i32 %rotateAmt, 6
+  %cmp.i = icmp eq i32 %rem.i, 0
+  br i1 %cmp.i, label %for.cond.preheader.i, label %if.end.i
+
+for.cond.preheader.i:                             ; preds = %entry
+  %sub.i = sub i32 4, %div.i
+  %cmp23.i = icmp eq i32 %div.i, 4
+  br i1 %cmp23.i, label %for.body9.lr.ph.i, label %for.body.lr.ph.i
+
+for.body.lr.ph.i:                                 ; preds = %for.cond.preheader.i
+  %pVal.i = getelementptr inbounds %struct.APInt* %this, i32 0, i32 0
+  %.pre5.i = load i64** %pVal.i, align 4
+  br label %for.body.i
+
+for.body.i:                                       ; preds = %for.body.i, %for.body.lr.ph.i
+  %i.04.i = phi i32 [ 0, %for.body.lr.ph.i ], [ %inc.i, %for.body.i ]
+  %add.i = add i32 %i.04.i, %div.i
+  %arrayidx.i = getelementptr inbounds i64* %.pre5.i, i32 %add.i
+  %1 = load i64* %arrayidx.i, align 4
+  %arrayidx3.i = getelementptr inbounds i64* %0, i32 %i.04.i
+  store i64 %1, i64* %arrayidx3.i, align 4
+  %inc.i = add i32 %i.04.i, 1
+  %cmp2.i = icmp ult i32 %inc.i, %sub.i
+  br i1 %cmp2.i, label %for.body.i, label %if.end.i
+
+if.end.i:                                         ; preds = %for.body.i, %entry
+  %cmp81.i = icmp eq i32 %div.i, 3
+  br i1 %cmp81.i, label %_ZNK5APInt4lshrEj.exit, label %for.body9.lr.ph.i
+
+for.body9.lr.ph.i:                                ; preds = %if.end.i, %for.cond.preheader.i
+  %sub58.i = sub i32 3, %div.i
+  %pVal11.i = getelementptr inbounds %struct.APInt* %this, i32 0, i32 0
+  %sh_prom.i = zext i32 %rem.i to i64
+  %sub17.i = sub i32 64, %rem.i
+  %sh_prom18.i = zext i32 %sub17.i to i64
+  %.pre.i = load i64** %pVal11.i, align 4
+  br label %for.body9.i
+
+for.body9.i:                                      ; preds = %for.body9.i, %for.body9.lr.ph.i
+; CHECK: %for.body9.i
+; CHECK: movb %al, %ch
+  %i6.02.i = phi i32 [ 0, %for.body9.lr.ph.i ], [ %inc21.i, %for.body9.i ]
+  %add10.i = add i32 %i6.02.i, %div.i
+  %arrayidx12.i = getelementptr inbounds i64* %.pre.i, i32 %add10.i
+  %2 = load i64* %arrayidx12.i, align 4
+  %shr.i = lshr i64 %2, %sh_prom.i
+  %add14.i = add i32 %add10.i, 1
+  %arrayidx16.i = getelementptr inbounds i64* %.pre.i, i32 %add14.i
+  %3 = load i64* %arrayidx16.i, align 4
+  %shl.i = shl i64 %3, %sh_prom18.i
+  %or.i = or i64 %shl.i, %shr.i
+  %arrayidx19.i = getelementptr inbounds i64* %0, i32 %i6.02.i
+  store i64 %or.i, i64* %arrayidx19.i, align 4
+  %inc21.i = add i32 %i6.02.i, 1
+  %cmp8.i = icmp ult i32 %inc21.i, %sub58.i
+  br i1 %cmp8.i, label %for.body9.i, label %_ZNK5APInt4lshrEj.exit
+
+_ZNK5APInt4lshrEj.exit:                           ; preds = %for.body9.i, %if.end.i
+  %call.i1 = tail call i8* @calloc(i32 1, i32 32) nounwind
+  %4 = getelementptr inbounds %struct.APInt* %agg.result, i32 0, i32 0
+  store i64* %0, i64** %4, align 4
+  ret void
+}

test/CodeGen/X86/machine-cp.ll

@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=x86_64-apple-macosx -mcpu=nocona < %s -disable-copyprop=false | FileCheck %s
+; RUN: llc -mtriple=x86_64-apple-macosx -mcpu=nocona < %s | FileCheck %s
 
 ; After tail duplication, two copies in an early exit BB can be cancelled out.
 ; rdar://10640363