[CodeGen] Allow partial tail duplication in Machine Block Placement.

[hoyFB]

Summary

A count profile may affect tail duplication's heuristic causing a block to be duplicated in only a part of its predecessors. This is not allowed in the Machine Block Placement pass where an assert will go off. I'm removing the assert and making the optimization bail out when such case happens.

[Carrot]

LGTM.
Thanks for the fixing.

llvm/lib/CodeGen/MachineBlockPlacement.cpp

Removed = maybeTailDuplicateBlock(BB, LPred, Chain, BlockFilter,
                                     DuplicatedToLPred);
   if (!Removed)
     return false;
   DuplicatedToOriginalLPred = DuplicatedToLPred;
   // Iteratively try to duplicate again. It can happen that a block that is
   // duplicated into is still small enough to be duplicated again.
   // No need to call markBlockSuccessors in this case, as the blocks being
   // duplicated from here on are already scheduled.
-  // Note that DuplicatedToLPred always implies Removed.
-  while (DuplicatedToLPred) {
-    assert(Removed && "Block must have been removed to be duplicated into its "
-           "layout predecessor.");
+  while (DuplicatedToLPred && Removed) {
     MachineBasicBlock *DupBB, *DupPred;
     // The removal callback causes Chain.end() to be updated when a block is
     // removed. On the first pass through the loop, the chain end should be the
     // same as it was on function entry. On subsequent passes, because we are
     // duplicating the block at the end of the chain, if it is removed the
     // chain will have shrunk by one block.
     BlockChain::iterator ChainEnd = Chain.end();
     DupBB = *(--ChainEnd);

 /// \p Chain - Chain to which \p LPred belongs, and \p BB will belong.
 /// \p BlockFilter - Set of blocks that belong to the loop being laid out.
 ///                  Used to identify which blocks to update predecessor
 ///                  counts.
 /// \p PrevUnplacedBlockIt - Iterator pointing to the last block that was
 ///                          chosen in the given order due to unnatural CFG
 ///                          only needed if \p BB is removed and
 ///                          \p PrevUnplacedBlockIt pointed to \p BB.
-/// \p DuplicatedToLPred - True if the block was duplicated into LPred. Will
-///                        only be true if the block was removed.
+/// \p DuplicatedToLPred - True if the block was duplicated into LPred.
 /// \return  - True if the block was duplicated into all preds and removed.
 bool MachineBlockPlacement::maybeTailDuplicateBlock(
     MachineBasicBlock *BB, MachineBasicBlock *LPred,
     BlockChain &Chain, BlockFilterSet *BlockFilter,
     MachineFunction::iterator &PrevUnplacedBlockIt,
     bool &DuplicatedToLPred) {
   DuplicatedToLPred = false;
   if (!shouldTailDuplicate(BB))

llvm/test/CodeGen/X86/tail-dup-partial.ll

+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu  -O3 | FileCheck %s
+
+; Function Attrs: uwtable
+; When tail-duplicating during placement, we work backward from blocks with
+; multiple successors. In this case, the block dup1 gets duplicated into dup2
+; and if.then64, and then the block dup2 only gets duplicated into land.lhs.true.
+
+define void @partial_tail_dup(i1 %a1, i1 %a2, i32* %a4, i32* %a5, i8* %a6, i32 %a7) #0 align 2  !prof !1 {
+; CHECK-LABEL: partial_tail_dup:
+; CHECK:        # %bb.0: # %entry
+; CHECK-NEXT:   .p2align 4, 0x90
+; CHECK-NEXT:   .LBB0_1: # %for.cond
+; CHECK-NEXT:   # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:   testb $1, %dil
+; CHECK-NEXT:   je  .LBB0_3
+; CHECK-NEXT:   # %bb.2: # %land.lhs.true
+; CHECK-NEXT:   # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:   movl  $10, (%rdx)
+; CHECK-NEXT:   movl  $2, (%rcx)
+; CHECK-NEXT:   testl %r9d, %r9d
+; CHECK-NEXT:   je  .LBB0_1
+; CHECK-NEXT:   jmp .LBB0_8
+; CHECK-NEXT:   .p2align  4, 0x90
+; CHECK-NEXT:   .LBB0_6: # %dup2
+; CHECK-NEXT:   # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:   movl  $2, (%rcx)
+; CHECK-NEXT:   testl %r9d, %r9d
+; CHECK-NEXT:   je  .LBB0_1
+; CHECK-NEXT:   jmp .LBB0_8
+; CHECK-NEXT:    .p2align 4, 0x90
+; CHECK-NEXT:  .LBB0_3: # %if.end56
+; CHECK-NEXT:    # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:    testb $1, %sil
+; CHECK-NEXT:    je .LBB0_5
+; CHECK-NEXT:  # %bb.4: # %if.then64
+; CHECK-NEXT:    # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:    movb $1, (%r8)
+; CHECK-NEXT:    testl %r9d, %r9d
+; CHECK-NEXT:    je .LBB0_1
+; CHECK-NEXT:    jmp .LBB0_8
+; CHECK-NEXT:  .LBB0_5: # %if.end70
+; CHECK-NEXT:    # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:    movl $12, (%rdx)
+; CHECK-NEXT:    jne .LBB0_6  
+; CHECK-NEXT:  .LBB0_8: # %for.end
+; CHECK-NEXT:    retq
+entry:
+  br label %for.cond
+
+for.cond:                                      
+  br i1 %a1, label %land.lhs.true, label %if.end56
+
+land.lhs.true:                                   
+  store i32 10, i32* %a4, align 8
+  br label %dup2
+
+if.end56:                                        
+  br i1 %a2, label %if.then64, label %if.end70, !prof !2
+
+if.then64:                                       
+  store i8 1, i8* %a6, align 1
+  br label %dup1
+
+if.end70:                                        
+  store i32 12, i32* %a4, align 8
+  br i1 %a2, label %dup2, label %for.end
+
+dup2:                                            
+  store i32 2, i32* %a5, align 4
+  br label %dup1
+
+dup1:                                            
+  %val = load i32, i32* %a4, align 8
+  %switch = icmp ult i32 %a7, 1
+  br i1 %switch, label %for.cond, label %for.end, !prof !3
+
+for.end:                                         
+  ret void
+}
+
+attributes #0 = { uwtable }
+
+!1 = !{!"function_entry_count", i64 2}
+!2 = !{!"branch_weights", i32 5, i32 1}
+!3 = !{!"branch_weights", i32 5, i32 1}