[PATCH] D91942: [SCEV] Fix incorrect treatment of max taken count. PR48225

[Max Kazantsev via Phabricator via llvm-commits]

mkazantsev created this revision.
mkazantsev added reviewers: nikic, lebedev.ri, reames, asbirlea.
Herald added subscribers: llvm-commits, javed.absar, hiraditya.
Herald added a project: LLVM.
mkazantsev requested review of this revision.

SCEV makes a logical mistake when handling EitherMayExit in
case when both conditions must be met to exit the loop. The
mistake looks like follows: "if condition `A` fails within at most `X` first
iterations, and `B` fails within at most `Y` first iterations, then `A & B`
fails at most within `min (X, Y)` first iterations". This is wrong, because
both of them must fail at the same time.

Simple example illustrating this is following: we have an IV with step 1,
condition `A` = "IV is even", condition `B` = "IV is odd". Both `A` and `B`
will fail within first two iterations. But it doesn't mean that both of them
will fail within first two first iterations at the same time, which would mean
that IV is neither even nor odd at the same time within first 2 iterations.

We can only do so for known exact BE counts, but not for max.


https://reviews.llvm.org/D91942

Files:
  llvm/lib/Analysis/ScalarEvolution.cpp
  llvm/test/Analysis/ScalarEvolution/pr48225.ll


Index: llvm/test/Analysis/ScalarEvolution/pr48225.ll
===================================================================
--- llvm/test/Analysis/ScalarEvolution/pr48225.ll
+++ llvm/test/Analysis/ScalarEvolution/pr48225.ll
@@ -4,7 +4,6 @@
 
 ; Tests demonstrate the bug reported as PR48225 by Congzhe Cao.
 
-; FIXME: This test demonstrates a bug in max backedge taken count computation.
 ; When %boolcond = false and %cond = 0:
 ; - %cond.false.on.first.iter is false on 1st iteration;
 ; - %cond.false.on.second.iter is false on 2nd iteration;
@@ -17,16 +16,16 @@
 ; CHECK-NEXT:    %conv = zext i1 %boolcond to i32
 ; CHECK-NEXT:    --> (zext i1 %boolcond to i32) U: [0,2) S: [0,2)
 ; CHECK-NEXT:    %iv = phi i32 [ 0, %entry ], [ %inc, %backedge ]
-; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%loop> U: [0,2) S: [0,2) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%loop> U: [0,3) S: [0,3) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %or.cond = and i1 %cond.false.on.first.iter, %cond.false.on.second.iter
 ; CHECK-NEXT:    --> %or.cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Variant }
 ; CHECK-NEXT:    %inc = add nuw nsw i32 %iv, 1
-; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%loop> U: [1,3) S: [1,3) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%loop> U: [1,4) S: [1,4) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @test_and
 ; CHECK-NEXT:  Loop %loop: <multiple exits> Unpredictable backedge-taken count.
 ; CHECK-NEXT:    exit count for loop: 2
 ; CHECK-NEXT:    exit count for backedge: ***COULDNOTCOMPUTE***
-; CHECK-NEXT:  Loop %loop: max backedge-taken count is 1
+; CHECK-NEXT:  Loop %loop: max backedge-taken count is 2
 ; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.
 ;
 entry:
@@ -52,7 +51,6 @@
   ret void
 }
 
-; FIXME: This test demonstrates a bug in max backedge taken count computation.
 ; When %boolcond = false and %cond = 0:
 ; - %cond.true.on.first.iter is true on 1st iteration;
 ; - %cond.true.on.second.iter is true on 2nd iteration;
@@ -65,16 +63,16 @@
 ; CHECK-NEXT:    %conv = zext i1 %boolcond to i32
 ; CHECK-NEXT:    --> (zext i1 %boolcond to i32) U: [0,2) S: [0,2)
 ; CHECK-NEXT:    %iv = phi i32 [ 0, %entry ], [ %inc, %backedge ]
-; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%loop> U: [0,2) S: [0,2) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%loop> U: [0,3) S: [0,3) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %or.cond = or i1 %cond.true.on.first.iter, %cond.true.on.second.iter
 ; CHECK-NEXT:    --> %or.cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Variant }
 ; CHECK-NEXT:    %inc = add nuw nsw i32 %iv, 1
-; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%loop> U: [1,3) S: [1,3) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%loop> U: [1,4) S: [1,4) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @test_or
 ; CHECK-NEXT:  Loop %loop: <multiple exits> Unpredictable backedge-taken count.
 ; CHECK-NEXT:    exit count for loop: 2
 ; CHECK-NEXT:    exit count for backedge: ***COULDNOTCOMPUTE***
-; CHECK-NEXT:  Loop %loop: max backedge-taken count is 1
+; CHECK-NEXT:  Loop %loop: max backedge-taken count is 2
 ; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.
 ;
 entry:
Index: llvm/lib/Analysis/ScalarEvolution.cpp
===================================================================
--- llvm/lib/Analysis/ScalarEvolution.cpp
+++ llvm/lib/Analysis/ScalarEvolution.cpp
@@ -7611,8 +7611,6 @@
   } else {
     // Both conditions must be same at the same time for the loop to exit.
     // For now, be conservative.
-    if (EL0.MaxNotTaken == EL1.MaxNotTaken)
-      MaxBECount = EL0.MaxNotTaken;
     if (EL0.ExactNotTaken == EL1.ExactNotTaken)
       BECount = EL0.ExactNotTaken;
   }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D91942.306962.patch
Type: text/x-patch
Size: 4130 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20201123/f378d655/attachment.bin>