[llvm] e8d5db2 - [LoopPeeling] Fix weights updating of peeled off branches (#70094)

[via llvm-commits]

Author: Aleksandr Popov
Date: 2023-10-31T14:02:42+01:00
New Revision: e8d5db206c2f5c4afac8288ab96193a5638270ae

URL: https://github.com/llvm/llvm-project/commit/e8d5db206c2f5c4afac8288ab96193a5638270ae
DIFF: https://github.com/llvm/llvm-project/commit/e8d5db206c2f5c4afac8288ab96193a5638270ae.diff

LOG: [LoopPeeling] Fix weights updating of peeled off branches (#70094)

In https://reviews.llvm.org/D64235 a new algorithm has been introduced
for updating the branch weights of latch blocks and their copies.

It increases the probability of going to the exit block for each next
peel iteration, calculating weights by (F - I * E, E), where:
- F is a weight of the edge from latch to header.
- E is a weight of the edge from latch to exit.
- I is a number of peeling iteration.

E.g: Let's say the latch branch weights are (100,300) and the estimated
trip count is 4. If we peel off all 4 iterations the weights of the
copied branches will be:
0: (100,300)
1: (100,200)
2: (100,100)
3: (100,1)

https://godbolt.org/z/93KnoEsT6

So we make the original loop almost unreachable from the 3rd peeled copy
according to the profile data. But that's only true if the profiling
data is accurate.
Underestimated trip count can lead to a performance issues with the
register allocator, which may decide to spill intervals inside the loop
assuming it's unreachable.

Since we don't know how accurate the profiling data is, it seems better
to set neutral 1/1 weights on the last peeled latch branch. After this
change, the weights in the example above will look like this:
0: (100,300)
1: (100,200)
2: (100,100)
3: (100,100)

Co-authored-by: Aleksandr Popov <apopov at azul.com>

Added: 
    

Modified: 
    llvm/lib/Transforms/Utils/LoopPeel.cpp
    llvm/test/Transforms/LoopUnroll/peel-loop-pgo-deopt.ll
    llvm/test/Transforms/LoopUnroll/peel-loop-pgo.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Utils/LoopPeel.cpp b/llvm/lib/Transforms/Utils/LoopPeel.cpp
index 31f065b691f864c..2881444206b0bb5 100644
--- a/llvm/lib/Transforms/Utils/LoopPeel.cpp
+++ b/llvm/lib/Transforms/Utils/LoopPeel.cpp
@@ -636,9 +636,14 @@ static void updateBranchWeights(Instruction *Term, WeightInfo &Info) {
                     MDB.createBranchWeights(Info.Weights));
   for (auto [Idx, SubWeight] : enumerate(Info.SubWeights))
     if (SubWeight != 0)
-      Info.Weights[Idx] = Info.Weights[Idx] > SubWeight
-                              ? Info.Weights[Idx] - SubWeight
-                              : 1;
+      // Don't set the probability of taking the edge from latch to loop header
+      // to less than 1:1 ratio (meaning Weight should not be lower than
+      // SubWeight), as this could significantly reduce the loop's hotness,
+      // which would be incorrect in the case of underestimating the trip count.
+      Info.Weights[Idx] =
+          Info.Weights[Idx] > SubWeight
+              ? std::max(Info.Weights[Idx] - SubWeight, SubWeight)
+              : SubWeight;
 }
 
 /// Initialize the weights for all exiting blocks.

diff  --git a/llvm/test/Transforms/LoopUnroll/peel-loop-pgo-deopt.ll b/llvm/test/Transforms/LoopUnroll/peel-loop-pgo-deopt.ll
index a0dc216133fb7d5..d91cb5bab382782 100644
--- a/llvm/test/Transforms/LoopUnroll/peel-loop-pgo-deopt.ll
+++ b/llvm/test/Transforms/LoopUnroll/peel-loop-pgo-deopt.ll
@@ -21,7 +21,7 @@
 ; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !18
 ; CHECK: [[NEXT2]]:
 ; CHECK: br i1 %c, label %{{.*}}, label %side_exit.loopexit, !prof !15
-; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !19
+; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !18
 
 define i32 @basic(ptr %p, i32 %k, i1 %c) #0 !prof !15 {
 entry:
@@ -85,6 +85,5 @@ attributes #1 = { nounwind optsize }
 ; This is a weights of latch and its copies.
 ;CHECK: !16 = !{!"branch_weights", i32 3001, i32 1001}
 ;CHECK: !17 = !{!"branch_weights", i32 2000, i32 1001}
-;CHECK: !18 = !{!"branch_weights", i32 999, i32 1001}
-;CHECK: !19 = !{!"branch_weights", i32 1, i32 1001}
+;CHECK: !18 = !{!"branch_weights", i32 1001, i32 1001}
 

diff  --git a/llvm/test/Transforms/LoopUnroll/peel-loop-pgo.ll b/llvm/test/Transforms/LoopUnroll/peel-loop-pgo.ll
index cadb6739dbc3fbb..15dce234baee91d 100644
--- a/llvm/test/Transforms/LoopUnroll/peel-loop-pgo.ll
+++ b/llvm/test/Transforms/LoopUnroll/peel-loop-pgo.ll
@@ -24,7 +24,7 @@
 ; CHECK: [[NEXT1]]:
 ; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !17
 ; CHECK: [[NEXT2]]:
-; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !18
+; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !17
 
 define void @basic(ptr %p, i32 %k) #0 !prof !15 {
 entry:
@@ -105,6 +105,5 @@ attributes #1 = { nounwind optsize }
 
 ;CHECK: !15 = !{!"branch_weights", i32 3001, i32 1001}
 ;CHECK: !16 = !{!"branch_weights", i32 2000, i32 1001}
-;CHECK: !17 = !{!"branch_weights", i32 999, i32 1001}
-;CHECK: !18 = !{!"branch_weights", i32 1, i32 1001}
+;CHECK: !17 = !{!"branch_weights", i32 1001, i32 1001}