From: Evan Cheng <evan.cheng@apple.com>
Date: Tue, 20 Oct 2009 01:31:09 +0000 (+0000)
Subject: If the physical register being spilled does not have an interval, spill its sub-regis... 
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=0222a8cfb8f8f3f67e4a07164eb1ecf9c44e6f64;p=oota-llvm.git

If the physical register being spilled does not have an interval, spill its sub-registers instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@84586 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/LiveIntervalAnalysis.cpp b/lib/CodeGen/LiveIntervalAnalysis.cpp
index 93d3d4c8389..51d6bb1b9ad 100644
--- a/lib/CodeGen/LiveIntervalAnalysis.cpp
+++ b/lib/CodeGen/LiveIntervalAnalysis.cpp
@@ -2603,7 +2603,19 @@ bool LiveIntervals::spillPhysRegAroundRegDefsUses(const LiveInterval &li,
            tri_->isSuperRegister(*AS, SpillReg));
 
   bool Cut = false;
-  LiveInterval &pli = getInterval(SpillReg);
+  SmallVector<unsigned, 4> PRegs;
+  if (hasInterval(SpillReg))
+    PRegs.push_back(SpillReg);
+  else {
+    SmallSet<unsigned, 4> Added;
+    for (const unsigned* AS = tri_->getSubRegisters(SpillReg); *AS; ++AS)
+      if (Added.insert(*AS) && hasInterval(*AS)) {
+        PRegs.push_back(*AS);
+        for (const unsigned* ASS = tri_->getSubRegisters(*AS); *ASS; ++ASS)
+          Added.insert(*ASS);
+      }
+  }
+
   SmallPtrSet<MachineInstr*, 8> SeenMIs;
   for (MachineRegisterInfo::reg_iterator I = mri_->reg_begin(li.reg),
          E = mri_->reg_end(); I != E; ++I) {
@@ -2613,8 +2625,12 @@ bool LiveIntervals::spillPhysRegAroundRegDefsUses(const LiveInterval &li,
       continue;
     SeenMIs.insert(MI);
     LiveIndex Index = getInstructionIndex(MI);
-    if (pli.liveAt(Index)) {
-      vrm.addEmergencySpill(SpillReg, MI);
+    for (unsigned i = 0, e = PRegs.size(); i != e; ++i) {
+      unsigned PReg = PRegs[i];
+      LiveInterval &pli = getInterval(PReg);
+      if (!pli.liveAt(Index))
+        continue;
+      vrm.addEmergencySpill(PReg, MI);
       LiveIndex StartIdx = getLoadIndex(Index);
       LiveIndex EndIdx = getNextSlot(getStoreIndex(Index));
       if (pli.isInOneLiveRange(StartIdx, EndIdx)) {
@@ -2626,12 +2642,12 @@ bool LiveIntervals::spillPhysRegAroundRegDefsUses(const LiveInterval &li,
         Msg << "Ran out of registers during register allocation!";
         if (MI->getOpcode() == TargetInstrInfo::INLINEASM) {
           Msg << "\nPlease check your inline asm statement for invalid "
-               << "constraints:\n";
+              << "constraints:\n";
           MI->print(Msg, tm_);
         }
         llvm_report_error(Msg.str());
       }
-      for (const unsigned* AS = tri_->getSubRegisters(SpillReg); *AS; ++AS) {
+      for (const unsigned* AS = tri_->getSubRegisters(PReg); *AS; ++AS) {
         if (!hasInterval(*AS))
           continue;
         LiveInterval &spli = getInterval(*AS);
diff --git a/test/CodeGen/X86/2009-10-19-EmergencySpill.ll b/test/CodeGen/X86/2009-10-19-EmergencySpill.ll
new file mode 100644
index 00000000000..ba44a2e64fe
--- /dev/null
+++ b/test/CodeGen/X86/2009-10-19-EmergencySpill.ll
@@ -0,0 +1,54 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -disable-fp-elim
+; rdar://7291624
+
+%union.RtreeCoord = type { float }
+%struct.RtreeCell = type { i64, [10 x %union.RtreeCoord] }
+%struct.Rtree = type { i32, i32*, i32, i32, i32, i32, i8*, i8* }
+%struct.RtreeNode = type { i32*, i64, i32, i32, i8*, i32* }
+
+define fastcc void @nodeOverwriteCell(%struct.Rtree* nocapture %pRtree, %struct.RtreeNode* nocapture %pNode, %struct.RtreeCell* nocapture %pCell, i32 %iCell) nounwind ssp {
+entry:
+  %0 = load i8** undef, align 8                   ; <i8*> [#uses=2]
+  %1 = load i32* undef, align 8                   ; <i32> [#uses=1]
+  %2 = mul i32 %1, %iCell                         ; <i32> [#uses=1]
+  %3 = add nsw i32 %2, 4                          ; <i32> [#uses=1]
+  %4 = sext i32 %3 to i64                         ; <i64> [#uses=2]
+  %5 = load i64* null, align 8                    ; <i64> [#uses=2]
+  %6 = lshr i64 %5, 48                            ; <i64> [#uses=1]
+  %7 = trunc i64 %6 to i8                         ; <i8> [#uses=1]
+  store i8 %7, i8* undef, align 1
+  %8 = lshr i64 %5, 8                             ; <i64> [#uses=1]
+  %9 = trunc i64 %8 to i8                         ; <i8> [#uses=1]
+  %.sum4 = add i64 %4, 6                          ; <i64> [#uses=1]
+  %10 = getelementptr inbounds i8* %0, i64 %.sum4 ; <i8*> [#uses=1]
+  store i8 %9, i8* %10, align 1
+  %11 = getelementptr inbounds %struct.Rtree* %pRtree, i64 0, i32 3 ; <i32*> [#uses=1]
+  br i1 undef, label %bb.nph, label %bb2
+
+bb.nph:                                           ; preds = %entry
+  %tmp25 = add i64 %4, 11                         ; <i64> [#uses=1]
+  br label %bb
+
+bb:                                               ; preds = %bb, %bb.nph
+  %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %bb ] ; <i64> [#uses=3]
+  %scevgep = getelementptr %struct.RtreeCell* %pCell, i64 0, i32 1, i64 %indvar ; <%union.RtreeCoord*> [#uses=1]
+  %scevgep12 = bitcast %union.RtreeCoord* %scevgep to i32* ; <i32*> [#uses=1]
+  %tmp = shl i64 %indvar, 2                       ; <i64> [#uses=1]
+  %tmp26 = add i64 %tmp, %tmp25                   ; <i64> [#uses=1]
+  %scevgep27 = getelementptr i8* %0, i64 %tmp26   ; <i8*> [#uses=1]
+  %12 = load i32* %scevgep12, align 4             ; <i32> [#uses=1]
+  %13 = lshr i32 %12, 24                          ; <i32> [#uses=1]
+  %14 = trunc i32 %13 to i8                       ; <i8> [#uses=1]
+  store i8 %14, i8* undef, align 1
+  store i8 undef, i8* %scevgep27, align 1
+  %15 = load i32* %11, align 4                    ; <i32> [#uses=1]
+  %16 = shl i32 %15, 1                            ; <i32> [#uses=1]
+  %17 = icmp sgt i32 %16, undef                   ; <i1> [#uses=1]
+  %indvar.next = add i64 %indvar, 1               ; <i64> [#uses=1]
+  br i1 %17, label %bb, label %bb2
+
+bb2:                                              ; preds = %bb, %entry
+  %18 = getelementptr inbounds %struct.RtreeNode* %pNode, i64 0, i32 3 ; <i32*> [#uses=1]
+  store i32 1, i32* %18, align 4
+  ret void
+}