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

This matches the legacy pipeline/pass.

Reviewed By: asbirlea, SjoerdMeijer

Differential Revision: https://reviews.llvm.org/D94559

lib/Passes/PassBuilder.cpp

@@ -724,7 +724,8 @@ PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level,
   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());
 

lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

@@ -2906,6 +2906,10 @@ static bool unswitchLoop(Loop &L, DominatorTree &DT, LoopInfo &LI,
   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

test/Transforms/SimpleLoopUnswitch/ARM/nontrivial-unswitch-cost.ll

@@ -111,33 +111,19 @@ loop_exit:
 define void @test_unswitch_minsize(i1* %ptr, i1 %cond) #0 {
 ; CHECK-LABEL: @test_unswitch_minsize(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[ENTRY_SPLIT_US:%.*]], label [[ENTRY_SPLIT:%.*]]
-; CHECK:       entry.split.us:
-; CHECK-NEXT:    br label [[LOOP_BEGIN_US:%.*]]
-; CHECK:       loop_begin.us:
-; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_A_US:%.*]]
-; CHECK:       loop_a.us:
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    br label [[LOOP_LATCH_US:%.*]]
-; CHECK:       loop_latch.us:
-; CHECK-NEXT:    [[V_US:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V_US]], label [[LOOP_BEGIN_US]], label [[LOOP_EXIT_SPLIT_US:%.*]]
-; CHECK:       loop_exit.split.us:
-; CHECK-NEXT:    br label [[LOOP_EXIT:%.*]]
-; CHECK:       entry.split:
 ; CHECK-NEXT:    br label [[LOOP_BEGIN:%.*]]
 ; CHECK:       loop_begin:
 ; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_B:%.*]]
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[LOOP_A:%.*]], label [[LOOP_B:%.*]]
+; CHECK:       loop_a:
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
 ; CHECK:       loop_b:
 ; CHECK-NEXT:    call void @b()
-; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
+; CHECK-NEXT:    br label [[LOOP_LATCH]]
 ; CHECK:       loop_latch:
-; CHECK-NEXT:    [[V:%.*]] = load i1, i1* [[PTR]], align 1
+; CHECK-NEXT:    [[V:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V]], label [[LOOP_BEGIN]], label [[LOOP_EXIT_SPLIT:%.*]]
+; CHECK-NEXT:    br i1 [[V]], label [[LOOP_BEGIN]], label [[LOOP_EXIT:%.*]]
-; CHECK:       loop_exit.split:
-; CHECK-NEXT:    br label [[LOOP_EXIT]]
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    ret void
 ;
@@ -235,39 +221,25 @@ loop_exit:
 define void @test_unswitch_non_dup_code_minsize(i1* %ptr, i1 %cond) #0 {
 ; CHECK-LABEL: @test_unswitch_non_dup_code_minsize(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[ENTRY_SPLIT_US:%.*]], label [[ENTRY_SPLIT:%.*]]
-; CHECK:       entry.split.us:
-; CHECK-NEXT:    br label [[LOOP_BEGIN_US:%.*]]
-; CHECK:       loop_begin.us:
-; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_A_US:%.*]]
-; CHECK:       loop_a.us:
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    br label [[LOOP_LATCH_US:%.*]]
-; CHECK:       loop_latch.us:
-; CHECK-NEXT:    [[V_US:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V_US]], label [[LOOP_BEGIN_US]], label [[LOOP_EXIT_SPLIT_US:%.*]]
-; CHECK:       loop_exit.split.us:
-; CHECK-NEXT:    br label [[LOOP_EXIT:%.*]]
-; CHECK:       entry.split:
 ; CHECK-NEXT:    br label [[LOOP_BEGIN:%.*]]
 ; CHECK:       loop_begin:
 ; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_B:%.*]]
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[LOOP_A:%.*]], label [[LOOP_B:%.*]]
+; CHECK:       loop_a:
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
 ; CHECK:       loop_b:
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
-; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
+; CHECK-NEXT:    br label [[LOOP_LATCH]]
 ; CHECK:       loop_latch:
-; CHECK-NEXT:    [[V:%.*]] = load i1, i1* [[PTR]], align 1
+; CHECK-NEXT:    [[V:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V]], label [[LOOP_BEGIN]], label [[LOOP_EXIT_SPLIT:%.*]]
+; CHECK-NEXT:    br i1 [[V]], label [[LOOP_BEGIN]], label [[LOOP_EXIT:%.*]]
-; CHECK:       loop_exit.split:
-; CHECK-NEXT:    br label [[LOOP_EXIT]]
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    ret void
 ;
@@ -387,45 +359,31 @@ loop_exit:
 define void @test_unswitch_non_dup_code_in_cfg_minsize(i1* %ptr, i1 %cond) #0 {
 ; CHECK-LABEL: @test_unswitch_non_dup_code_in_cfg_minsize(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[ENTRY_SPLIT_US:%.*]], label [[ENTRY_SPLIT:%.*]]
-; CHECK:       entry.split.us:
-; CHECK-NEXT:    br label [[LOOP_BEGIN_US:%.*]]
-; CHECK:       loop_begin.us:
-; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_A_US:%.*]]
-; CHECK:       loop_a.us:
-; CHECK-NEXT:    [[V1_US:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V1_US]], label [[LOOP_A_A_US:%.*]], label [[LOOP_A_B_US:%.*]]
-; CHECK:       loop_a_b.us:
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    br label [[LOOP_LATCH_US:%.*]]
-; CHECK:       loop_a_a.us:
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    br label [[LOOP_LATCH_US]]
-; CHECK:       loop_latch.us:
-; CHECK-NEXT:    [[V3_US:%.*]] = load i1, i1* [[PTR]], align 1
-; CHECK-NEXT:    br i1 [[V3_US]], label [[LOOP_BEGIN_US]], label [[LOOP_EXIT_SPLIT_US:%.*]]
-; CHECK:       loop_exit.split.us:
-; CHECK-NEXT:    br label [[LOOP_EXIT:%.*]]
-; CHECK:       entry.split:
 ; CHECK-NEXT:    br label [[LOOP_BEGIN:%.*]]
 ; CHECK:       loop_begin:
 ; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_B:%.*]]
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[LOOP_A:%.*]], label [[LOOP_B:%.*]]
+; CHECK:       loop_a:
+; CHECK-NEXT:    [[V1:%.*]] = load i1, i1* [[PTR:%.*]], align 1
+; CHECK-NEXT:    br i1 [[V1]], label [[LOOP_A_A:%.*]], label [[LOOP_A_B:%.*]]
+; CHECK:       loop_a_a:
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
+; CHECK:       loop_a_b:
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    br label [[LOOP_LATCH]]
 ; CHECK:       loop_b:
 ; CHECK-NEXT:    [[V2:%.*]] = load i1, i1* [[PTR]], align 1
 ; CHECK-NEXT:    br i1 [[V2]], label [[LOOP_B_A:%.*]], label [[LOOP_B_B:%.*]]
 ; CHECK:       loop_b_a:
 ; CHECK-NEXT:    call void @b()
-; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
+; CHECK-NEXT:    br label [[LOOP_LATCH]]
 ; CHECK:       loop_b_b:
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    br label [[LOOP_LATCH]]
 ; CHECK:       loop_latch:
 ; CHECK-NEXT:    [[V3:%.*]] = load i1, i1* [[PTR]], align 1
-; CHECK-NEXT:    br i1 [[V3]], label [[LOOP_BEGIN]], label [[LOOP_EXIT_SPLIT:%.*]]
+; CHECK-NEXT:    br i1 [[V3]], label [[LOOP_BEGIN]], label [[LOOP_EXIT:%.*]]
-; CHECK:       loop_exit.split:
-; CHECK-NEXT:    br label [[LOOP_EXIT]]
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    ret void
 ;
@@ -666,31 +624,22 @@ loop_exit:
 define void @test_unswitch_dedicated_exiting_minsize(i1* %ptr, i1 %cond) #0 {
 ; CHECK-LABEL: @test_unswitch_dedicated_exiting_minsize(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[ENTRY_SPLIT_US:%.*]], label [[ENTRY_SPLIT:%.*]]
-; CHECK:       entry.split.us:
-; CHECK-NEXT:    br label [[LOOP_BEGIN_US:%.*]]
-; CHECK:       loop_begin.us:
-; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_A_US:%.*]]
-; CHECK:       loop_a.us:
-; CHECK-NEXT:    call void @a()
-; CHECK-NEXT:    br label [[LOOP_LATCH_US:%.*]]
-; CHECK:       loop_latch.us:
-; CHECK-NEXT:    [[V_US:%.*]] = load i1, i1* [[PTR:%.*]], align 1
-; CHECK-NEXT:    br i1 [[V_US]], label [[LOOP_BEGIN_US]], label [[LOOP_EXIT_SPLIT_US:%.*]]
-; CHECK:       loop_exit.split.us:
-; CHECK-NEXT:    br label [[LOOP_EXIT:%.*]]
-; CHECK:       entry.split:
 ; CHECK-NEXT:    br label [[LOOP_BEGIN:%.*]]
 ; CHECK:       loop_begin:
 ; CHECK-NEXT:    call void @x()
-; CHECK-NEXT:    br label [[LOOP_B_EXIT:%.*]]
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[LOOP_A:%.*]], label [[LOOP_B_EXIT:%.*]]
+; CHECK:       loop_a:
+; CHECK-NEXT:    call void @a()
+; CHECK-NEXT:    br label [[LOOP_LATCH:%.*]]
 ; CHECK:       loop_b_exit:
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    call void @b()
 ; CHECK-NEXT:    ret void
+; CHECK:       loop_latch:
+; CHECK-NEXT:    [[V:%.*]] = load i1, i1* [[PTR:%.*]], align 1
+; CHECK-NEXT:    br i1 [[V]], label [[LOOP_BEGIN]], label [[LOOP_EXIT:%.*]]
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    ret void
 ;

test/Transforms/SimpleLoopUnswitch/pipeline.ll

@@ -1,12 +1,15 @@
-; RUN: opt < %s -S -passes="default<O1>" | FileCheck %s -check-prefix=O1
+; RUN: opt < %s -S -passes="default<O1>" | FileCheck %s -check-prefixes=TRIVIAL,CHECK
-; RUN: opt < %s -S -passes="default<O2>" | FileCheck %s -check-prefix=O2
+; 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:
+; TRIVIAL-NOT: loop_begin.us:
-; O2: loop_begin.us:
+; NONTRIVIAL: loop_begin.us:
 
 define i32 @test1(i1* %ptr, i1 %cond1, i1 %cond2) {
 entry:
@@ -37,3 +40,34 @@ latch:
 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
+}