test/Transforms/LoopVectorize/replicate-region-guard-unknown-prof.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter "br i1|= select |!prof|!llvm.loop|^.*:" --version 6
; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s

; Tail folding masks the body on the header mask, so the guard's frequency is
; unknown.
define void @tail_folded_predicated_store(ptr noalias %a, ptr noalias %b, i64 %n) optsize !prof !0 {
; CHECK-LABEL: define void @tail_folded_predicated_store(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] !prof [[PROF0:![0-9]+]] {
; CHECK:  [[ENTRY:.*:]]
; CHECK:  [[VECTOR_PH:.*:]]
; CHECK:  [[VECTOR_BODY:.*]]:
; CHECK:    br i1 [[TMP5:%.*]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
; CHECK:  [[PRED_LOAD_IF]]:
; CHECK:  [[PRED_LOAD_CONTINUE]]:
; CHECK:    br i1 [[TMP10:%.*]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2:.*]]
; CHECK:  [[PRED_LOAD_IF1]]:
; CHECK:  [[PRED_LOAD_CONTINUE2]]:
; CHECK:    br i1 [[TMP15:%.*]], label %[[PRED_LOAD_IF3:.*]], label %[[PRED_LOAD_CONTINUE4:.*]]
; CHECK:  [[PRED_LOAD_IF3]]:
; CHECK:  [[PRED_LOAD_CONTINUE4]]:
; CHECK:    br i1 [[TMP20:%.*]], label %[[PRED_LOAD_IF5:.*]], label %[[PRED_LOAD_CONTINUE6:.*]]
; CHECK:  [[PRED_LOAD_IF5]]:
; CHECK:  [[PRED_LOAD_CONTINUE6]]:
; CHECK:    [[TMP26:%.*]] = select <4 x i1> [[TMP1:%.*]], <4 x i1> [[TMP25:%.*]], <4 x i1> zeroinitializer
; CHECK:    br i1 [[TMP27:%.*]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]], !prof [[PROF1:![0-9]+]]
; CHECK:  [[PRED_STORE_IF]]:
; CHECK:  [[PRED_STORE_CONTINUE]]:
; CHECK:    br i1 [[TMP30:%.*]], label %[[PRED_STORE_IF7:.*]], label %[[PRED_STORE_CONTINUE8:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF7]]:
; CHECK:  [[PRED_STORE_CONTINUE8]]:
; CHECK:    br i1 [[TMP33:%.*]], label %[[PRED_STORE_IF9:.*]], label %[[PRED_STORE_CONTINUE10:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF9]]:
; CHECK:  [[PRED_STORE_CONTINUE10]]:
; CHECK:    br i1 [[TMP36:%.*]], label %[[PRED_STORE_IF11:.*]], label %[[PRED_STORE_CONTINUE12:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF11]]:
; CHECK:  [[PRED_STORE_CONTINUE12]]:
; CHECK:    br i1 [[TMP39:%.*]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !prof [[PROF2:![0-9]+]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK:  [[MIDDLE_BLOCK]]:
; CHECK:  [[EXIT:.*:]]
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %l = load i32, ptr %gep.b, align 4
  %c = icmp sgt i32 %l, 0
  br i1 %c, label %then, label %loop.latch, !prof !1

then:
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
  store i32 %l, ptr %gep.a, align 4
  br label %loop.latch

loop.latch:
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, %n
  br i1 %ec, label %exit, label %loop.header, !prof !2

exit:
  ret void
}

; Same without tail folding, where the guard keeps the original weights.
define void @predicated_store(ptr noalias %a, ptr noalias %b, i64 %n) !prof !0 {
; CHECK-LABEL: define void @predicated_store(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) !prof [[PROF0]] {
; CHECK:  [[ENTRY:.*:]]
; CHECK:    br i1 [[MIN_ITERS_CHECK:%.*]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]], !prof [[PROF7:![0-9]+]]
; CHECK:  [[VECTOR_PH]]:
; CHECK:  [[VECTOR_BODY:.*]]:
; CHECK:    br i1 [[TMP3:%.*]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF]]:
; CHECK:  [[PRED_STORE_CONTINUE]]:
; CHECK:    br i1 [[TMP6:%.*]], label %[[PRED_STORE_IF1:.*]], label %[[PRED_STORE_CONTINUE2:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF1]]:
; CHECK:  [[PRED_STORE_CONTINUE2]]:
; CHECK:    br i1 [[TMP10:%.*]], label %[[PRED_STORE_IF3:.*]], label %[[PRED_STORE_CONTINUE4:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF3]]:
; CHECK:  [[PRED_STORE_CONTINUE4]]:
; CHECK:    br i1 [[TMP14:%.*]], label %[[PRED_STORE_IF5:.*]], label %[[PRED_STORE_CONTINUE6:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF5]]:
; CHECK:  [[PRED_STORE_CONTINUE6]]:
; CHECK:    br i1 [[TMP18:%.*]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !prof [[PROF2]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK:  [[MIDDLE_BLOCK]]:
; CHECK:    br i1 [[CMP_N:%.*]], label %[[EXIT:.*]], label %[[SCALAR_PH]], !prof [[PROF1]]
; CHECK:  [[SCALAR_PH]]:
; CHECK:  [[LOOP_HEADER:.*]]:
; CHECK:    br i1 [[C:%.*]], label %[[THEN:.*]], label %[[LOOP_LATCH:.*]], !prof [[PROF1]]
; CHECK:  [[THEN]]:
; CHECK:  [[LOOP_LATCH]]:
; CHECK:    br i1 [[EC:%.*]], label %[[EXIT]], label %[[LOOP_HEADER]], !prof [[PROF9:![0-9]+]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK:  [[EXIT]]:
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %l = load i32, ptr %gep.b, align 4
  %c = icmp sgt i32 %l, 0
  br i1 %c, label %then, label %loop.latch, !prof !1

then:
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
  store i32 %l, ptr %gep.a, align 4
  br label %loop.latch

loop.latch:
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, %n
  br i1 %ec, label %exit, label %loop.header, !prof !2

exit:
  ret void
}

define void @tail_folded_predicated_store_not_profiled(ptr noalias %a, ptr noalias %b, i64 %n) optsize {
; CHECK-LABEL: define void @tail_folded_predicated_store_not_profiled(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; CHECK:  [[ENTRY:.*:]]
; CHECK:  [[VECTOR_PH:.*:]]
; CHECK:  [[VECTOR_BODY:.*]]:
; CHECK:    br i1 [[TMP5:%.*]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
; CHECK:  [[PRED_LOAD_IF]]:
; CHECK:  [[PRED_LOAD_CONTINUE]]:
; CHECK:    br i1 [[TMP10:%.*]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2:.*]]
; CHECK:  [[PRED_LOAD_IF1]]:
; CHECK:  [[PRED_LOAD_CONTINUE2]]:
; CHECK:    br i1 [[TMP15:%.*]], label %[[PRED_LOAD_IF3:.*]], label %[[PRED_LOAD_CONTINUE4:.*]]
; CHECK:  [[PRED_LOAD_IF3]]:
; CHECK:  [[PRED_LOAD_CONTINUE4]]:
; CHECK:    br i1 [[TMP20:%.*]], label %[[PRED_LOAD_IF5:.*]], label %[[PRED_LOAD_CONTINUE6:.*]]
; CHECK:  [[PRED_LOAD_IF5]]:
; CHECK:  [[PRED_LOAD_CONTINUE6]]:
; CHECK:    [[TMP26:%.*]] = select <4 x i1> [[TMP1:%.*]], <4 x i1> [[TMP25:%.*]], <4 x i1> zeroinitializer
; CHECK:    br i1 [[TMP27:%.*]], label %[[PRED_STORE_IF:.*]], label %[[PRED_STORE_CONTINUE:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF]]:
; CHECK:  [[PRED_STORE_CONTINUE]]:
; CHECK:    br i1 [[TMP30:%.*]], label %[[PRED_STORE_IF7:.*]], label %[[PRED_STORE_CONTINUE8:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF7]]:
; CHECK:  [[PRED_STORE_CONTINUE8]]:
; CHECK:    br i1 [[TMP33:%.*]], label %[[PRED_STORE_IF9:.*]], label %[[PRED_STORE_CONTINUE10:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF9]]:
; CHECK:  [[PRED_STORE_CONTINUE10]]:
; CHECK:    br i1 [[TMP36:%.*]], label %[[PRED_STORE_IF11:.*]], label %[[PRED_STORE_CONTINUE12:.*]], !prof [[PROF1]]
; CHECK:  [[PRED_STORE_IF11]]:
; CHECK:  [[PRED_STORE_CONTINUE12]]:
; CHECK:    br i1 [[TMP39:%.*]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !prof [[PROF2]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK:  [[MIDDLE_BLOCK]]:
; CHECK:  [[EXIT:.*:]]
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %l = load i32, ptr %gep.b, align 4
  %c = icmp sgt i32 %l, 0
  br i1 %c, label %then, label %loop.latch, !prof !1

then:
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %iv
  store i32 %l, ptr %gep.a, align 4
  br label %loop.latch

loop.latch:
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, %n
  br i1 %ec, label %exit, label %loop.header, !prof !2

exit:
  ret void
}

!0 = !{!"function_entry_count", i64 1000}
!1 = !{!"branch_weights", i32 1, i32 3}
!2 = !{!"branch_weights", i32 1, i32 99}
;.
; CHECK: [[PROF0]] = !{!"function_entry_count", i64 1000}
; CHECK: [[PROF1]] = !{!"branch_weights", i32 1, i32 3}
; CHECK: [[PROF2]] = !{!"branch_weights", i32 1, i32 24}
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META4:![0-9]+]], [[META5:![0-9]+]], [[META6:![0-9]+]]}
; CHECK: [[META4]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK: [[META5]] = !{!"llvm.loop.unroll.runtime.disable"}
; CHECK: [[META6]] = !{!"llvm.loop.estimated_trip_count", i32 25}
; CHECK: [[PROF7]] = !{!"branch_weights", i32 1, i32 127}
; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META4]], [[META5]], [[META6]]}
; CHECK: [[PROF9]] = !{!"branch_weights", i32 1, i32 0}
; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META5]], [[META4]], [[META11:![0-9]+]]}
; CHECK: [[META11]] = !{!"llvm.loop.estimated_trip_count", i32 0}
; CHECK: [[LOOP12]] = distinct !{[[LOOP12]], [[META4]], [[META5]], [[META6]]}
;.