[LoopFuse] Ensure inner loops are in loop simplified form under new PM

[Narutoworld]

Summary

This patch tries to fix issue.

LoopInfo doesn't give all loops in a loop nest, it gives top level loops only. While isLoopSimplifyForm() only checkes for the outter most loop of a loop nest. As a result, inner loops that are not in simplied form can not be simplified with the original code.

This patch tries to run simplifyLoop() on loop nests anyway and let it simplify each loop in a loop nest recursively.

[uabelho]

Test case?

[Narutoworld]

testcase added

[uabelho]

I've verified that this solves the original problem I saw.
The fix looks reasonable to me but I'd like someone who knows this code to accept it.

[aeubanks]

description nit: LI doesn't give you loop nests, it gives you top level loops

lg with test and description fixes

llvm/test/Transforms/LoopFusion/ensure_loop_simplify_form.ll

37

there's already a RUN at the top

66

this is UB, can you replace these with a constant like true?

[Narutoworld]

update testcase based on comments

[Narutoworld]

resolve comments

llvm/lib/Transforms/Scalar/LoopFuse.cpp

PreservedAnalyses LoopFusePass::run(Function &F, FunctionAnalysisManager &AM) {
   const TargetTransformInfo &TTI = AM.getResult<TargetIRAnalysis>(F);
   const DataLayout &DL = F.getParent()->getDataLayout();
 
   // Ensure loops are in simplifed form which is a pre-requisite for loop fusion
   // pass. Added only for new PM since the legacy PM has already added
   // LoopSimplify pass as a dependency.
   bool Changed = false;
   for (auto &L : LI) {
-    if (!L->isLoopSimplifyForm())
-      Changed |= simplifyLoop(L, &DT, &LI, &SE, &AC, nullptr,
-                              false /* PreserveLCSSA */);
+    Changed |=
+        simplifyLoop(L, &DT, &LI, &SE, &AC, nullptr, false /* PreserveLCSSA */);
   }
   if (Changed)
     PDT.recalculate(F);
 
   LoopFuser LF(LI, DT, DI, SE, PDT, ORE, DL, AC, TTI);
   Changed |= LF.fuseLoops(F);
   if (!Changed)
     return PreservedAnalyses::all();

llvm/test/Transforms/LoopFusion/ensure_loop_simplify_form.ll

 
 for.cond.cleanup5:                                ; preds = %for.cond.cleanup
   ret void
 
 for.body6:                                        ; preds = %for.body6, %for.cond.cleanup
   %v_loop_2.0.v_loop_2.0.v_loop_2.0.18 = load volatile i32, i32* %dummy, align 1
   br label %for.body6
 }
+
+define void @f943(i8* %dummy) {
+; CHECK-LABEL: @f943(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_COND32_I_I:%.*]]
+; CHECK:       for.cond32.i.i:
+; CHECK-NEXT:    br label [[FOR_BODY37_I_I:%.*]]
+; CHECK:       for.cond42.preheader.i.i:
+; CHECK-NEXT:    br i1 true, label [[FOR_BODY44_I_I_PREHEADER:%.*]], label [[FOR_END47_I_I:%.*]]
+; CHECK:       for.body44.i.i.preheader:
+; CHECK-NEXT:    br label [[FOR_BODY44_I_I:%.*]]
+; CHECK:       for.body37.i.i:
+; CHECK-NEXT:    br i1 true, label [[FOR_BODY37_I_I]], label [[FOR_COND42_PREHEADER_I_I:%.*]]
+; CHECK:       for.body44.i.i:
+; CHECK-NEXT:    store volatile i8 poison, i8* [[DUMMY:%.*]], align 1
+; CHECK-NEXT:    br i1 true, label [[FOR_BODY44_I_I]], label [[FOR_END47_I_I_LOOPEXIT:%.*]]
+; CHECK:       for.end47.i.i.loopexit:
+; CHECK-NEXT:    br label [[FOR_END47_I_I]]
+; CHECK:       for.end47.i.i:
+; CHECK-NEXT:    br label [[FOR_COND32_I_I]]
+;
+entry:
+  br label %for.cond32.i.i
+
+for.cond32.i.i:                                   ; preds = %for.end47.i.i, %entry
+  br label %for.body37.i.i
+
+for.cond42.preheader.i.i:                         ; preds = %for.body37.i.i
+  br i1 true, label %for.body44.i.i, label %for.end47.i.i
+
+for.body37.i.i:                                   ; preds = %for.body37.i.i, %for.cond32.i.i
+  br i1 true, label %for.body37.i.i, label %for.cond42.preheader.i.i
+
+for.body44.i.i:                                   ; preds = %for.body44.i.i, %for.cond42.preheader.i.i
+  store volatile i8 poison, i8* %dummy, align 1
+  br i1 true, label %for.body44.i.i, label %for.end47.i.i
+
+for.end47.i.i:                                    ; preds = %for.body44.i.i, %for.cond42.preheader.i.i
+  br label %for.cond32.i.i
+}
+