[MBP] Avoid tail duplication if it can't bring benefit

Current tail duplication integrated in bb layout is designed to increase the fallthrough from a BB's predecessor to its successor, but we have observed cases that duplication doesn't increase fallthrough, or it brings too much size overhead.

To overcome these two issues in function canTailDuplicateUnplacedPreds I add two checks:

  make sure there is at least one duplication in current work set.
  the number of duplication should not exceed the number of successors.

The modification in hasBetterLayoutPredecessor fixes a bug that potential predecessor must be at the bottom of a chain.

Differential Revision: https://reviews.llvm.org/D64376

1 files changed, 12 insertions, 13 deletions

diff --git a/llvm/test/CodeGen/RISCV/remat.ll b/llvm/test/CodeGen/RISCV/remat.ll
index d8267e7a7ee..e440515d632 100644
--- a/llvm/test/CodeGen/RISCV/remat.ll
+++ b/llvm/test/CodeGen/RISCV/remat.ll
@@ -52,20 +52,19 @@ define i32 @test() nounwind {
 ; RV32I-NEXT:    lui s9, %hi(h)
 ; RV32I-NEXT:    lui s10, %hi(c)
 ; RV32I-NEXT:    lui s11, %hi(b)
-; RV32I-NEXT:    lw a1, %lo(l)(s2)
-; RV32I-NEXT:    bnez a1, .LBB0_4
-; RV32I-NEXT:    j .LBB0_5
+; RV32I-NEXT:    j .LBB0_3
 ; RV32I-NEXT:  .LBB0_2: # %for.inc
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(a)(s6)
 ; RV32I-NEXT:    addi a0, a0, -1
 ; RV32I-NEXT:    sw a0, %lo(a)(s6)
 ; RV32I-NEXT:    beqz a0, .LBB0_11
-; RV32I-NEXT:  # %bb.3: # %for.body
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:  .LBB0_3: # %for.body
+; RV32I-NEXT:    # =>This Inner Loop Header: Depth=1
 ; RV32I-NEXT:    lw a1, %lo(l)(s2)
 ; RV32I-NEXT:    beqz a1, .LBB0_5
-; RV32I-NEXT:  .LBB0_4: # %if.then
+; RV32I-NEXT:  # %bb.4: # %if.then
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a1, %lo(b)(s11)
 ; RV32I-NEXT:    lw a2, %lo(c)(s10)
 ; RV32I-NEXT:    lw a3, %lo(d)(s1)
@@ -73,11 +72,11 @@ define i32 @test() nounwind {
 ; RV32I-NEXT:    addi a5, zero, 32
 ; RV32I-NEXT:    call foo
 ; RV32I-NEXT:  .LBB0_5: # %if.end
-; RV32I-NEXT:    # =>This Inner Loop Header: Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(k)(s3)
 ; RV32I-NEXT:    beqz a0, .LBB0_7
 ; RV32I-NEXT:  # %bb.6: # %if.then3
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(b)(s11)
 ; RV32I-NEXT:    lw a1, %lo(c)(s10)
 ; RV32I-NEXT:    lw a2, %lo(d)(s1)
@@ -86,11 +85,11 @@ define i32 @test() nounwind {
 ; RV32I-NEXT:    addi a5, zero, 64
 ; RV32I-NEXT:    call foo
 ; RV32I-NEXT:  .LBB0_7: # %if.end5
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(j)(s4)
 ; RV32I-NEXT:    beqz a0, .LBB0_9
 ; RV32I-NEXT:  # %bb.8: # %if.then7
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(c)(s10)
 ; RV32I-NEXT:    lw a1, %lo(d)(s1)
 ; RV32I-NEXT:    lw a2, %lo(e)(s0)
@@ -99,11 +98,11 @@ define i32 @test() nounwind {
 ; RV32I-NEXT:    addi a5, zero, 32
 ; RV32I-NEXT:    call foo
 ; RV32I-NEXT:  .LBB0_9: # %if.end9
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(i)(s5)
 ; RV32I-NEXT:    beqz a0, .LBB0_2
 ; RV32I-NEXT:  # %bb.10: # %if.then11
-; RV32I-NEXT:    # in Loop: Header=BB0_5 Depth=1
+; RV32I-NEXT:    # in Loop: Header=BB0_3 Depth=1
 ; RV32I-NEXT:    lw a0, %lo(d)(s1)
 ; RV32I-NEXT:    lw a1, %lo(e)(s0)
 ; RV32I-NEXT:    lw a2, %lo(f)(s7)