test/Transforms/GuardWidening/profile-based-profitability-intrinsics.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
; RUN: opt -S -passes=guard-widening < %s | FileCheck %s

; Function Attrs: nocallback nofree nosync willreturn
declare void @llvm.experimental.guard(i1, ...) #0

; Hot loop, frequently entered, should widen.
define i32 @test_intrinsic_very_profitable(i32 %n, i1 %cond.1, i1 %cond.2) {
; CHECK-LABEL: define i32 @test_intrinsic_very_profitable
; CHECK-SAME: (i32 [[N:%.*]], i1 [[COND_1:%.*]], i1 [[COND_2:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COND_2_GW_FR:%.*]] = freeze i1 [[COND_2]]
; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_2_GW_FR]]
; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[WIDE_CHK]]) [ "deopt"() ]
; CHECK-NEXT:    [[LOOP_PRECONDITION:%.*]] = icmp uge i32 [[N]], 100
; CHECK-NEXT:    br i1 [[LOOP_PRECONDITION]], label [[LOOP:%.*]], label [[FAILED:%.*]], !prof [[PROF0:![0-9]+]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_NEXT]], 100
; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]], !prof [[PROF1:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    ret i32 0
; CHECK:       failed:
; CHECK-NEXT:    ret i32 -1
;
entry:
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.1) [ "deopt"() ]
  %loop.precondition = icmp uge i32 %n, 100
  br i1 %loop.precondition, label %loop, label %failed, !prof !0

loop:                                             ; preds = %loop, %entry
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.2) [ "deopt"() ]
  %iv.next = add nuw nsw i32 %iv, 1
  %loop.cond = icmp ult i32 %iv.next, 100
  br i1 %loop.cond, label %loop, label %exit, !prof !1

exit:                                             ; preds = %loop
  ret i32 0

failed:                                           ; preds = %entry
  ret i32 -1
}

; Even though the loop is rarely entered, it has so many iterations that the widening
; is still profitable.
define i32 @test_intrinsic_profitable(i32 %n, i1 %cond.1, i1 %cond.2) {
; CHECK-LABEL: define i32 @test_intrinsic_profitable
; CHECK-SAME: (i32 [[N:%.*]], i1 [[COND_1:%.*]], i1 [[COND_2:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COND_2_GW_FR:%.*]] = freeze i1 [[COND_2]]
; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_2_GW_FR]]
; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[WIDE_CHK]]) [ "deopt"() ]
; CHECK-NEXT:    [[LOOP_PRECONDITION:%.*]] = icmp uge i32 [[N]], 100
; CHECK-NEXT:    br i1 [[LOOP_PRECONDITION]], label [[LOOP:%.*]], label [[FAILED:%.*]], !prof [[PROF2:![0-9]+]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_NEXT]], 100
; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]], !prof [[PROF1]]
; CHECK:       exit:
; CHECK-NEXT:    ret i32 0
; CHECK:       failed:
; CHECK-NEXT:    ret i32 -1
;
entry:
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.1) [ "deopt"() ]
  %loop.precondition = icmp uge i32 %n, 100
  br i1 %loop.precondition, label %loop, label %failed, !prof !2

loop:                                             ; preds = %loop, %entry
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.2) [ "deopt"() ]
  %iv.next = add nuw nsw i32 %iv, 1
  %loop.cond = icmp ult i32 %iv.next, 100
  br i1 %loop.cond, label %loop, label %exit, !prof !1

exit:                                             ; preds = %loop
  ret i32 0

failed:                                           ; preds = %entry
  ret i32 -1
}

; Loop's hotness compensates rareness of its entrance. We still want to widen, because
; it may open up some optimization opportunities.
define i32 @test_intrinsic_neutral(i32 %n, i1 %cond.1, i1 %cond.2) {
; CHECK-LABEL: define i32 @test_intrinsic_neutral
; CHECK-SAME: (i32 [[N:%.*]], i1 [[COND_1:%.*]], i1 [[COND_2:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COND_2_GW_FR:%.*]] = freeze i1 [[COND_2]]
; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_2_GW_FR]]
; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[WIDE_CHK]]) [ "deopt"() ]
; CHECK-NEXT:    [[LOOP_PRECONDITION:%.*]] = icmp uge i32 [[N]], 100
; CHECK-NEXT:    br i1 [[LOOP_PRECONDITION]], label [[LOOP:%.*]], label [[FAILED:%.*]], !prof [[PROF3:![0-9]+]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_NEXT]], 100
; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]], !prof [[PROF1]]
; CHECK:       exit:
; CHECK-NEXT:    ret i32 0
; CHECK:       failed:
; CHECK-NEXT:    ret i32 -1
;
entry:
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.1) [ "deopt"() ]
  %loop.precondition = icmp uge i32 %n, 100
  br i1 %loop.precondition, label %loop, label %failed, !prof !3

loop:                                             ; preds = %loop, %entry
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.2) [ "deopt"() ]
  %iv.next = add nuw nsw i32 %iv, 1
  %loop.cond = icmp ult i32 %iv.next, 100
  br i1 %loop.cond, label %loop, label %exit, !prof !1

exit:                                             ; preds = %loop
  ret i32 0

failed:                                           ; preds = %entry
  ret i32 -1
}

; FIXME: This loop is so rarely entered, that we don't want to widen here.
define i32 @test_intrinsic_very_unprofitable(i32 %n, i1 %cond.1, i1 %cond.2) {
; CHECK-LABEL: define i32 @test_intrinsic_very_unprofitable
; CHECK-SAME: (i32 [[N:%.*]], i1 [[COND_1:%.*]], i1 [[COND_2:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COND_2_GW_FR:%.*]] = freeze i1 [[COND_2]]
; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_2_GW_FR]]
; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[WIDE_CHK]]) [ "deopt"() ]
; CHECK-NEXT:    [[LOOP_PRECONDITION:%.*]] = icmp uge i32 [[N]], 100
; CHECK-NEXT:    br i1 [[LOOP_PRECONDITION]], label [[LOOP:%.*]], label [[FAILED:%.*]], !prof [[PROF4:![0-9]+]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_NEXT]], 100
; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]], !prof [[PROF1]]
; CHECK:       exit:
; CHECK-NEXT:    ret i32 0
; CHECK:       failed:
; CHECK-NEXT:    ret i32 -1
;
entry:
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.1) [ "deopt"() ]
  %loop.precondition = icmp uge i32 %n, 100
  br i1 %loop.precondition, label %loop, label %failed, !prof !4

loop:                                             ; preds = %loop, %entry
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.2) [ "deopt"() ]
  %iv.next = add nuw nsw i32 %iv, 1
  %loop.cond = icmp ult i32 %iv.next, 100
  br i1 %loop.cond, label %loop, label %exit, !prof !1

exit:                                             ; preds = %loop
  ret i32 0

failed:                                           ; preds = %entry
  ret i32 -1
}

; FIXME: This loop is so rarely entered, that we don't want to widen here.
define i32 @test_intrinsic_unprofitable(i32 %n, i1 %cond.1, i1 %cond.2) {
; CHECK-LABEL: define i32 @test_intrinsic_unprofitable
; CHECK-SAME: (i32 [[N:%.*]], i1 [[COND_1:%.*]], i1 [[COND_2:%.*]]) {
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COND_2_GW_FR:%.*]] = freeze i1 [[COND_2]]
; CHECK-NEXT:    [[WIDE_CHK:%.*]] = and i1 [[COND_1]], [[COND_2_GW_FR]]
; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[WIDE_CHK]]) [ "deopt"() ]
; CHECK-NEXT:    [[LOOP_PRECONDITION:%.*]] = icmp uge i32 [[N]], 100
; CHECK-NEXT:    br i1 [[LOOP_PRECONDITION]], label [[LOOP:%.*]], label [[FAILED:%.*]], !prof [[PROF5:![0-9]+]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp ult i32 [[IV_NEXT]], 100
; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]], !prof [[PROF1]]
; CHECK:       exit:
; CHECK-NEXT:    ret i32 0
; CHECK:       failed:
; CHECK-NEXT:    ret i32 -1
;
entry:
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.1) [ "deopt"() ]
  %loop.precondition = icmp uge i32 %n, 100
  br i1 %loop.precondition, label %loop, label %failed, !prof !5

loop:                                             ; preds = %loop, %entry
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  call void (i1, ...) @llvm.experimental.guard(i1 %cond.2) [ "deopt"() ]
  %iv.next = add nuw nsw i32 %iv, 1
  %loop.cond = icmp ult i32 %iv.next, 100
  br i1 %loop.cond, label %loop, label %exit, !prof !1

exit:                                             ; preds = %loop
  ret i32 0

failed:                                           ; preds = %entry
  ret i32 -1
}

attributes #0 = { nocallback nofree nosync willreturn }

!0 = !{!"branch_weights", i32 1048576, i32 1}
!1 = !{!"branch_weights", i32 99, i32 1}
!2 = !{!"branch_weights", i32 1, i32 10}
!3 = !{!"branch_weights", i32 1, i32 99}
!4 = !{!"branch_weights", i32 1, i32 1048576}
!5 = !{!"branch_weights", i32 1, i32 1000}