Re-apply 101075 and fix it properly. Just reuse the debug info of the branch instruct...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101162 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index e67bf3d7ad27ece6d6e325e5b12cc5a1cc12e484..66b6562ba5896977bf620a41e9f3132073da5006 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -1684,6 +1684,7 @@ bool X86InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
   // Start from the bottom of the block and work up, examining the
   // terminator instructions.
   MachineBasicBlock::iterator I = MBB.end();
+  MachineBasicBlock::iterator UnCondBrIter = MBB.end();
   while (I != MBB.begin()) {
     --I;
     if (I->isDebugValue())
@@ -1701,6 +1702,8 @@ bool X86InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
 
     // Handle unconditional branches.
     if (I->getOpcode() == X86::JMP_4) {
+      UnCondBrIter = I;
+
       if (!AllowModify) {
         TBB = I->getOperand(0).getMBB();
         continue;
@@ -1718,10 +1721,11 @@ bool X86InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
         TBB = 0;
         I->eraseFromParent();
         I = MBB.end();
+        UnCondBrIter = MBB.end();
         continue;
       }
 
-      // TBB is used to indicate the unconditinal destination.
+      // TBB is used to indicate the unconditional destination.
       TBB = I->getOperand(0).getMBB();
       continue;
     }
@@ -1733,6 +1737,45 @@ bool X86InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
 
     // Working from the bottom, handle the first conditional branch.
     if (Cond.empty()) {
+      MachineBasicBlock *TargetBB = I->getOperand(0).getMBB();
+      if (AllowModify && UnCondBrIter != MBB.end() &&
+          MBB.isLayoutSuccessor(TargetBB)) {
+        // If we can modify the code and it ends in something like:
+        //
+        //     jCC L1
+        //     jmp L2
+        //   L1:
+        //     ...
+        //   L2:
+        //
+        // Then we can change this to:
+        //
+        //     jnCC L2
+        //   L1:
+        //     ...
+        //   L2:
+        //
+        // Which is a bit more efficient.
+        // We conditionally jump to the fall-through block.
+        BranchCode = GetOppositeBranchCondition(BranchCode);
+        unsigned JNCC = GetCondBranchFromCond(BranchCode);
+        MachineBasicBlock::iterator OldInst = I;
+
+        BuildMI(MBB, UnCondBrIter, MBB.findDebugLoc(I), get(JNCC))
+          .addMBB(UnCondBrIter->getOperand(0).getMBB());
+        BuildMI(MBB, UnCondBrIter, MBB.findDebugLoc(I), get(X86::JMP_4))
+          .addMBB(TargetBB);
+        MBB.addSuccessor(TargetBB);
+
+        OldInst->eraseFromParent();
+        UnCondBrIter->eraseFromParent();
+
+        // Restart the analysis.
+        UnCondBrIter = MBB.end();
+        I = MBB.end();
+        continue;
+      }
+
       FBB = TBB;
       TBB = I->getOperand(0).getMBB();
       Cond.push_back(MachineOperand::CreateImm(BranchCode));

diff --git a/test/CodeGen/X86/2010-04-13-AnalyzeBranchCrash.ll b/test/CodeGen/X86/2010-04-13-AnalyzeBranchCrash.ll
new file mode 100644 (file)
index 0000000..fadbd21
--- /dev/null
+++ b/test/CodeGen/X86/2010-04-13-AnalyzeBranchCrash.ll
@@ -0,0 +1,42 @@
+; RUN: llc < %s -mtriple=i386-apple-darwin -mcpu=core2
+; rdar://7857830
+
+%0 = type opaque
+%1 = type opaque
+
+define void @t(%0* %self, i8* nocapture %_cmd, %1* %scroller, i32 %hitPart, float %multiplier) nounwind optsize ssp {
+entry:
+  switch i32 %hitPart, label %if.else [
+    i32 7, label %if.then
+    i32 8, label %if.then
+  ]
+
+if.then:                                          ; preds = %entry, %entry
+  %tmp69 = load float* null, align 4              ; <float> [#uses=1]
+  %cmp19 = icmp eq %1* null, %scroller            ; <i1> [#uses=2]
+  %cond = select i1 %cmp19, float %tmp69, float 0.000000e+00 ; <float> [#uses=1]
+  %call36 = call i64 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i64 (i8*, i8*)*)(i8* undef, i8* undef) nounwind optsize ; <i64> [#uses=2]
+  br i1 %cmp19, label %cond.true32, label %cond.false39
+
+cond.true32:                                      ; preds = %if.then
+  %sroa.store.elt68 = lshr i64 %call36, 32        ; <i64> [#uses=1]
+  %0 = trunc i64 %sroa.store.elt68 to i32         ; <i32> [#uses=1]
+  br label %cond.end47
+
+cond.false39:                                     ; preds = %if.then
+  %1 = trunc i64 %call36 to i32                   ; <i32> [#uses=1]
+  br label %cond.end47
+
+cond.end47:                                       ; preds = %cond.false39, %cond.true32
+  %cond48.in = phi i32 [ %0, %cond.true32 ], [ %1, %cond.false39 ] ; <i32> [#uses=1]
+  %cond48 = bitcast i32 %cond48.in to float       ; <float> [#uses=1]
+  %div = fdiv float %cond, undef                  ; <float> [#uses=1]
+  %div58 = fdiv float %div, %cond48               ; <float> [#uses=1]
+  call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, float)*)(i8* undef, i8* undef, float %div58) nounwind optsize
+  ret void
+
+if.else:                                          ; preds = %entry
+  ret void
+}
+
+declare i8* @objc_msgSend(i8*, i8*, ...)

diff --git a/test/CodeGen/X86/brcond.ll b/test/CodeGen/X86/brcond.ll
index 130483ad841095e73acbddf704f873ad7bb2a0d3..14ee9e50fa8861244f79176fffd2f2c819170b9e 100644 (file)
--- a/test/CodeGen/X86/brcond.ll
+++ b/test/CodeGen/X86/brcond.ll
@@ -67,3 +67,42 @@ return:                                           ; preds = %entry
 ; CHECK-NEXT:  orl     8(%esp), %eax
 ; CHECK-NEXT:  je      LBB3_2
 }
+
+; <rdar://problem/7598384>:
+;
+;    jCC  L1
+;    jmp  L2
+; L1:
+;   ...
+; L2:
+;   ...
+;
+; to:
+; 
+;    jnCC L2
+; L1:
+;   ...
+; L2:
+;   ...
+define float @test4(float %x, float %y) nounwind readnone optsize ssp {
+entry:
+  %0 = fpext float %x to double                   ; <double> [#uses=1]
+  %1 = fpext float %y to double                   ; <double> [#uses=1]
+  %2 = fmul double %0, %1                         ; <double> [#uses=3]
+  %3 = fcmp oeq double %2, 0.000000e+00           ; <i1> [#uses=1]
+  br i1 %3, label %bb2, label %bb1
+
+; CHECK:      jne
+; CHECK-NEXT: jnp
+; CHECK-NOT:  jmp
+; CHECK:      LBB
+
+bb1:                                              ; preds = %entry
+  %4 = fadd double %2, -1.000000e+00              ; <double> [#uses=1]
+  br label %bb2
+
+bb2:                                              ; preds = %entry, %bb1
+  %.0.in = phi double [ %4, %bb1 ], [ %2, %entry ] ; <double> [#uses=1]
+  %.0 = fptrunc double %.0.in to float            ; <float> [#uses=1]
+  ret float %.0
+}