[Passes] Move LoopInterchange into optimization pipeline (#145503)

As mentioned in #145071,
LoopInterchange should be part of the optimization pipeline rather than
the simplification pipeline. This patch moves LoopInterchange into the
optimization pipeline.

More contexts:

- By default, LoopInterchange attempts to improve data locality,
however, it also takes increasing vectorization opportunities into
account. Given that, it is reasonable to run it as close to
vectorization as possible.
- I looked into previous changes related to the placement of
LoopInterchange, but couldn’t find any strong motivation suggesting that
it benefits other simplifications.
- As far as I tried some tests (including llvm-test-suite), removing
LoopInterchange from the simplification pipeline does not affect other
simplifications. Therefore, there doesn't seem to be much value in
keeping it there.
- The new position reduces compile-time for ThinLTO, probably because it
only runs once per function in post-link optimization, rather than both
in pre-link and post-link optimization.

I haven't encountered any cases where the positional difference affects
optimization results, so please feel free to revert if you run into any issues.

Author: kasuga-fj

llvm/lib/Passes/PassBuilderPipelines.cpp

@@ -690,9 +690,6 @@ PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level,
 
   LPM2.addPass(LoopDeletionPass());
 
-  if (PTO.LoopInterchange)
-    LPM2.addPass(LoopInterchangePass());
-
   // Do not enable unrolling in PreLinkThinLTO phase during sample PGO
   // because it changes IR to makes profile annotation in back compile
   // inaccurate. The normal unroller doesn't pay attention to forced full unroll
@@ -1547,6 +1544,10 @@ PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
   //        this may need to be revisited once we run GVN before loop deletion
   //        in the simplification pipeline.
   LPM.addPass(LoopDeletionPass());
+
+  if (PTO.LoopInterchange)
+    LPM.addPass(LoopInterchangePass());
+
   OptimizePM.addPass(createFunctionToLoopPassAdaptor(
       std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false));
 

llvm/test/Transforms/LoopInterchange/position-in-pipeline.ll

@@ -0,0 +1,48 @@
+; RUN: opt -passes='default<O3>' -enable-loopinterchange -disable-output \
+; RUN:     -disable-verify -verify-analysis-invalidation=0 \
+; RUN:     -debug-pass-manager=quiet %s 2>&1 | FileCheck %s
+
+; Test the position of LoopInterchange in the pass pipeline.
+
+; CHECK-NOT:  Running pass: LoopInterchangePass
+; CHECK:      Running pass: ControlHeightReductionPass
+; CHECK-NEXT: Running pass: LoopSimplifyPass
+; CHECK-NEXT: Running pass: LCSSAPass
+; CHECK-NEXT: Running pass: LoopRotatePass
+; CHECK-NEXT: Running pass: LoopDeletionPass
+; CHECK-NEXT: Running pass: LoopRotatePass
+; CHECK-NEXT: Running pass: LoopDeletionPass
+; CHECK-NEXT: Running pass: LoopInterchangePass
+; CHECK-NEXT: Running pass: LoopDistributePass
+; CHECK-NEXT: Running pass: InjectTLIMappings
+; CHECK-NEXT: Running pass: LoopVectorizePass
+
+
+define void @foo(ptr %a, i32 %n) {
+entry:
+  br label %for.i.header
+
+for.i.header:
+  %i = phi i32 [ 0, %entry ], [ %i.next, %for.i.latch ]
+  br label %for.j
+
+for.j:
+  %j = phi i32 [ 0, %for.i.header ], [ %j.next, %for.j ]
+  %tmp = mul i32 %i, %n
+  %offset = add i32 %tmp, %j
+  %idx = getelementptr inbounds i32, ptr %a, i32 %offset
+  %load = load i32, ptr %idx, align 4
+  %inc = add i32 %load, 1
+  store i32 %inc, ptr %idx, align 4
+  %j.next = add i32 %j, 1
+  %j.exit = icmp eq i32 %j.next, %n
+  br i1 %j.exit, label %for.i.latch, label %for.j
+
+for.i.latch:
+  %i.next = add i32 %i, 1
+  %i.exit = icmp eq i32 %i.next, %n
+  br i1 %i.exit, label %for.i.header, label %exit
+
+exit:
+  ret void
+}