[NewPM] Only non-trivially loop unswitch at -O3 and for non-optsize functions

[aeubanks]

Summary

This matches the legacy pipeline/pass.

[SjoerdMeijer]

To avoid code-size bloat, this is indeed the behaviour that we want.
Thanks, LGTM.

[asbirlea]

Thank you for confirming @SjoerdMeijer.

[aeubanks]

fix ARM unswitch test: looks like minsize functions are becoming smaller, which is good

llvm/lib/Passes/PassBuilder.cpp

PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level,
   // implications on the outer loop.
   LPM1.addPass(LoopInstSimplifyPass());
   LPM1.addPass(LoopSimplifyCFGPass());
 
   // Disable header duplication in loop rotation at -Oz.
   LPM1.addPass(LoopRotatePass(Level != OptimizationLevel::Oz));
   // TODO: Investigate promotion cap for O1.
   LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap));
-  LPM1.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ true));
+  LPM1.addPass(
+      SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3));
   LPM2.addPass(LoopIdiomRecognizePass());
   LPM2.addPass(IndVarSimplifyPass());
 
   for (auto &C : LateLoopOptimizationsEPCallbacks)
     C(LPM2, Level);
 
   LPM2.addPass(LoopDeletionPass());
   // Do not enable unrolling in PreLinkThinLTO phase during sample PGO

llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

static bool unswitchLoop(Loop &L, DominatorTree &DT, LoopInfo &LI,
   }
 
   // If we're not doing non-trivial unswitching, we're done. We both accept
   // a parameter but also check a local flag that can be used for testing
   // a debugging.
   if (!NonTrivial && !EnableNonTrivialUnswitch)
     return false;
 
+  // Skip non-trivial unswitching for optsize functions.
+  if (L.getHeader()->getParent()->hasOptSize())
+    return false;
+
   // For non-trivial unswitching, because it often creates new loops, we rely on
   // the pass manager to iterate on the loops rather than trying to immediately
   // reach a fixed point. There is no substantial advantage to iterating
   // internally, and if any of the new loops are simplified enough to contain
   // trivial unswitching we want to prefer those.
 
   // Try to unswitch the best invariant condition. We prefer this full unswitch to
   // a partial unswitch when possible below the threshold.

llvm/test/Transforms/SimpleLoopUnswitch/pipeline.ll

-; RUN: opt < %s -S -passes="default<O1>" | FileCheck %s -check-prefix=O1
-; RUN: opt < %s -S -passes="default<O2>" | FileCheck %s -check-prefix=O2
+; RUN: opt < %s -S -passes="default<O1>" | FileCheck %s -check-prefixes=TRIVIAL,CHECK
+; RUN: opt < %s -S -passes="default<O2>" | FileCheck %s -check-prefixes=TRIVIAL,CHECK
+; RUN: opt < %s -S -passes="default<O3>" | FileCheck %s -check-prefixes=NONTRIVIAL,CHECK
+; RUN: opt < %s -S -passes="default<Os>" | FileCheck %s -check-prefixes=TRIVIAL,CHECK
+; RUN: opt < %s -S -passes="default<Oz>" | FileCheck %s -check-prefixes=TRIVIAL,CHECK
 
 declare i32 @a()
 declare i32 @b()
 declare i32 @c()
 
-; O1-NOT: loop_begin.us:
-; O2: loop_begin.us:
+; TRIVIAL-NOT: loop_begin.us:
+; NONTRIVIAL: loop_begin.us:
 
 define i32 @test1(i1* %ptr, i1 %cond1, i1 %cond2) {
 entry:
   br label %loop_begin
 
 loop_begin:
   br i1 %cond1, label %loop_a, label %loop_b
 

 
 latch:
   %v = load i1, i1* %ptr
   br i1 %v, label %loop_begin, label %loop_exit
 
 loop_exit:
   ret i32 0
 }
+
+; CHECK-NOT: loop2_begin.us:
+define i32 @test2(i1* %ptr, i1 %cond1, i1 %cond2) optsize {
+entry:
+  br label %loop2_begin
+
+loop2_begin:
+  br i1 %cond1, label %loop2_a, label %loop2_b
+
+loop2_a:
+  call i32 @a()
+  br label %latch2
+
+loop2_b:
+  br i1 %cond2, label %loop2_b_a, label %loop2_b_b
+
+loop2_b_a:
+  call i32 @b()
+  br label %latch2
+
+loop2_b_b:
+  call i32 @c()
+  br label %latch2
+
+latch2:
+  %v = load i1, i1* %ptr
+  br i1 %v, label %loop2_begin, label %loop2_exit
+
+loop2_exit:
+  ret i32 0
+}