[PATCH] D31310: Fix trellis layout when there is triangle.

[Dehao Chen via Phabricator via llvm-commits]

danielcdh created this revision.
Herald added a subscriber: nemanjai.

In the trellis CFG layout, the best 2 edges found are only within the trellis. But if there is hot succ edge out of the trellis, the 2nd best edge may not not be optimal. This patch disables the layout of the 2nd best edge if hot succ edge out of the trelis is found.


https://reviews.llvm.org/D31310

Files:
  lib/CodeGen/MachineBlockPlacement.cpp
  test/CodeGen/PowerPC/tail-dup-layout.ll


Index: test/CodeGen/PowerPC/tail-dup-layout.ll
===================================================================
--- test/CodeGen/PowerPC/tail-dup-layout.ll
+++ test/CodeGen/PowerPC/tail-dup-layout.ll
@@ -474,6 +474,52 @@
   ret void
 }
 
+; Verify that in the triangular trellis case, if there is hot successor
+; outside of trellis, we should be able to connect them by not connecting
+; the triagular edges.
+; CHECK-LABEL: trellis_triangle
+; CHECK: # %entry
+; CHECK: # %b
+; CHECK: # %d
+; CHECK: # %ret
+; CHECK: # %c
+; CHECK: # %e
+define void @trellis_no_triangle(i32 %tag) {
+entry:
+  br label %a
+a:
+  call void @a()
+  call void @a()
+  %tagbits.a = and i32 %tag, 3
+  %tagbits.a.eq0 = icmp eq i32 %tagbits.a, 0
+  br i1 %tagbits.a.eq0, label %b, label %c, !prof !8 ; 98 to 2
+b:
+  call void @b()
+  call void @b()
+  %tagbits.b = and i32 %tag, 12
+  %tagbits.b.eq1 = icmp eq i32 %tagbits.b, 8
+  br i1 %tagbits.b.eq1, label %d, label %e, !prof !9 ; 97 to 1
+d:
+  call void @d()
+  call void @d()
+  %tagbits.d = and i32 %tag, 48
+  %tagbits.d.eq1 = icmp eq i32 %tagbits.d, 32
+  br i1 %tagbits.d.eq1, label %ret, label %e, !prof !10 ; 96 to 2
+c:
+  call void @c()
+  call void @c()
+  %tagbits.c = and i32 %tag, 12
+  %tagbits.c.eq0 = icmp eq i32 %tagbits.c, 0
+  br i1 %tagbits.c.eq0, label %d, label %e, !prof !2 ; 1 to 1
+e:
+  call void @e()
+  call void @e()
+  br label %ret
+ret:
+  call void @f()
+  ret void
+}
+
 declare void @a()
 declare void @b()
 declare void @c()
@@ -492,3 +538,6 @@
 !5 = !{!"branch_weights", i32 2, i32 8}
 !6 = !{!"branch_weights", i32 3, i32 4}
 !7 = !{!"branch_weights", i32 4, i32 2}
+!8 = !{!"branch_weights", i32 98, i32 2}
+!9 = !{!"branch_weights", i32 97, i32 1}
+!10 = !{!"branch_weights", i32 96, i32 2}
Index: lib/CodeGen/MachineBlockPlacement.cpp
===================================================================
--- lib/CodeGen/MachineBlockPlacement.cpp
+++ lib/CodeGen/MachineBlockPlacement.cpp
@@ -957,6 +957,7 @@
     return Result;
   }
 
+  bool ComputeBestB = true;
   // Did we pick the triangle edge? If tail-duplication is profitable, do
   // that instead. Otherwise merge the triangle edge now while we know it is
   // optimal.
@@ -978,10 +979,18 @@
       Result.ShouldTailDup = true;
       return Result;
     }
+    // If any of the BestA.Dest's successor is outside of the Trellis, we should
+    // not use BestB as it may not be optimal.
+    for (MachineBasicBlock *Succ : BestA.Dest->successors())
+      if (!Successors.count(Succ)) {
+        ComputeBestB = false;
+        break;
+      }
   }
-  // We have already computed the optimal edge for the other side of the
-  // trellis.
-  ComputedEdges[BestB.Src] = { BestB.Dest, false };
+  if (ComputeBestB)
+    // We have already computed the optimal edge for the other side of the
+    // trellis.
+    ComputedEdges[BestB.Src] = { BestB.Dest, false };
 
   auto TrellisSucc = BestA.Dest;
   DEBUG(BranchProbability SuccProb = getAdjustedProbability(


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D31310.92868.patch
Type: text/x-patch
Size: 2998 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170323/9edf630a/attachment.bin>