[llvm] 8b45614 - LAA: add missed swap when inverting src, sink (#122254)

[via llvm-commits]

Author: Ramkumar Ramachandra
Date: 2025-01-13T13:07:19Z
New Revision: 8b4561467e828c3755ad2942715fac00de2be4a7

URL: https://github.com/llvm/llvm-project/commit/8b4561467e828c3755ad2942715fac00de2be4a7
DIFF: https://github.com/llvm/llvm-project/commit/8b4561467e828c3755ad2942715fac00de2be4a7.diff

LOG: LAA: add missed swap when inverting src, sink (#122254)

When inverting source and sink on a negative induction step, the types
of the source and sink should also be swapped. This fixes a bug in the
code that follows, that computes properties based on these types. With
234cc40 ([LAA] Limit no-overlap check to at least one loop-invariant
accesses.), that code is guarded by a loop-invariant condition: however,
the commit did not add any new tests exercising the guarded code, and
hence the bugfix in this patch requires additional tests to exercise
that guarded codepath.

Added: 
    

Modified: 
    llvm/lib/Analysis/LoopAccessAnalysis.cpp
    llvm/test/Analysis/LoopAccessAnalysis/depend_diff_types.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index 38e9145826c08e..2a68979add666d 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -1921,6 +1921,7 @@ MemoryDepChecker::getDependenceDistanceStrideAndSize(
   if (StrideAPtr && *StrideAPtr < 0) {
     std::swap(Src, Sink);
     std::swap(AInst, BInst);
+    std::swap(ATy, BTy);
     std::swap(StrideAPtr, StrideBPtr);
   }
 

diff  --git a/llvm/test/Analysis/LoopAccessAnalysis/depend_
diff _types.ll b/llvm/test/Analysis/LoopAccessAnalysis/depend_
diff _types.ll
index 0bdcc357901487..e855578e794fa9 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/depend_
diff _types.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/depend_
diff _types.ll
@@ -194,3 +194,79 @@ loop:
 exit:
   ret void
 }
+
+; In the following test, the sink is loop-invariant.
+
+define void @type_size_equivalence_sink_loopinv(ptr nocapture %vec, i64 %n) {
+; CHECK-LABEL: 'type_size_equivalence_sink_loopinv'
+; CHECK-NEXT:    loop:
+; CHECK-NEXT:      Memory dependences are safe
+; CHECK-NEXT:      Dependences:
+; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT:      SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Expressions re-written:
+;
+entry:
+  %gep.n = getelementptr inbounds i64, ptr %vec, i64 %n
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+
+  %gep.iv = getelementptr i64, ptr %vec, i64 %iv
+  %ld.i64 = load i64, ptr %gep.iv, align 8
+
+  %ld.i64.i32 = trunc i64 %ld.i64 to i32
+  store i32 %ld.i64.i32, ptr %gep.n, align 8
+
+  %iv.next = add nuw nsw i64 %iv, 1
+  %cond = icmp eq i64 %iv.next, %n
+  br i1 %cond, label %exit, label %loop
+
+exit:
+  ret void
+}
+
+; Variant of the above, with a negative induction step and a gep exposing
+; type-mismtach.
+
+define void @type_size_equivalence_sink_loopinv_negind(ptr nocapture %vec, i64 %n) {
+; CHECK-LABEL: 'type_size_equivalence_sink_loopinv_negind'
+; CHECK-NEXT:    loop:
+; CHECK-NEXT:      Memory dependences are safe
+; CHECK-NEXT:      Dependences:
+; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT:      SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Expressions re-written:
+;
+entry:
+  %minus.n = sub nsw i64 0, %n
+  %gep.minus.n = getelementptr inbounds i64, ptr %vec, i64 %minus.n
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+
+  %minus.iv = sub nsw i64 0, %iv
+  %gep.minus.iv = getelementptr i64, ptr %vec, i64 %minus.iv
+  %gep.minus.iv.4 = getelementptr i8, ptr %gep.minus.iv, i64 -4
+  %ld.i64 = load i64, ptr %gep.minus.iv.4, align 8
+
+  %ld.i64.i32 = trunc i64 %ld.i64 to i32
+  store i32 %ld.i64.i32, ptr %gep.minus.n, align 8
+
+  %iv.next = add nuw nsw i64 %iv, 1
+  %cond = icmp eq i64 %iv.next, %n
+  br i1 %cond, label %exit, label %loop
+
+exit:
+  ret void
+}