[llvm] [LoopInterchange] Relax the legality check to accept more patterns (PR #118267)

[Madhur Amilkanthwar via llvm-commits]

================
@@ -0,0 +1,74 @@
+; REQUIRES: asserts
+; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info \
+; RUN:     -S -debug 2>&1 | FileCheck %s
+
+ at aa = dso_local global [256 x [256 x float]] zeroinitializer, align 64
+ at bb = dso_local global [256 x [256 x float]] zeroinitializer, align 64
+
+declare i32 @dummy(ptr noundef, ptr noundef)
+
+;;  for (int nl=0;nl<100;++nl) {
+;;    for (int i=0;i<256;++i) {
+;;      for (int j=1;j<256;++j)
+;;        aa[j][i] = aa[j-1][i] + bb[j][i];
+;;    }
+;;    dummy(aa, bb);
+;;  }
+;;
+;; The direction vector of `aa` is [S = >]. We can swap the innermost two
+;; loops, The direction vector after interchanging will be [S > =].
+
+; CHECK: Dependency matrix before interchange:
+; CHECK-NEXT: S = >
+; CHECK-NEXT: S = =
+; CHECK-NEXT: Processing InnerLoopId = 2 and OuterLoopId = 1
+; CHECK-NEXT: Checking if loops are tightly nested
+; CHECK-NEXT: Checking instructions in Loop header and Loop latch
+; CHECK-NEXT: Loops are perfectly nested
+; CHECK-NEXT: Loops are legal to interchange
+; CHECK: Dependency matrix after interchange:
+; CHECK-NEXT: S > =
+; CHECK-NEXT: S = =
+
+define void @f() {
+entry:
+  br label %for.cond1.preheader
+
+; Loop:
+for.cond1.preheader:                              ; preds = %entry, %for.cond.cleanup3
+  %nl.036 = phi i32 [ 0, %entry ], [ %inc23, %for.cond.cleanup3 ]
+  br label %for.cond5.preheader
+
+for.cond.cleanup3:                                ; preds = %for.cond.cleanup7
+  %call = tail call i32 @dummy(ptr noundef nonnull @aa, ptr noundef nonnull @bb)
----------------
madhur13490 wrote:

Do you really need function call here?

https://github.com/llvm/llvm-project/pull/118267