; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -S -passes=loop-unroll -mattr=+no-default-unroll | FileCheck %s

; Regression test: when a loop has a small MaxTripCount and falls through to
; runtime unrolling, the unroll count should be based on DefaultUnrollRuntimeCount
; (not MaxTripCount). Previously, a quirk would set UP.Count = MaxTripCount (5)
; before the power-of-two reduction, yielding count 2 (5 >> 1). Now the count
; starts at DefaultUnrollRuntimeCount (8) and reduces to 4 (8 >> 1).
;
; This test uses RISC-V because we need a target that sets UP.Force = true
; (via RISCVTTIImpl::getUnrollingPreferences) to trigger the bug.

target triple = "riscv64-unknown-linux-gnu"

define void @test(i32 %start, ptr %p) {
; CHECK-LABEL: @test(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[END:%.*]] = add i32 [[START:%.*]], 5
; CHECK-NEXT:    br label [[LOOP:%.*]]
; CHECK:       loop:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[START]], [[ENTRY:%.*]] ], [ [[IV_NEXT_3:%.*]], [[LOOP_3:%.*]] ]
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, ptr [[P:%.*]], i32 [[IV]]
; CHECK-NEXT:    [[V:%.*]] = load volatile i32, ptr [[GEP]], align 4
; CHECK-NEXT:    store volatile i32 [[V]], ptr [[GEP]], align 4
; CHECK-NEXT:    [[IV_NEXT:%.*]] = add nuw i32 [[IV]], 1
; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[IV_NEXT]], [[END]]
; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP_1:%.*]], label [[EXIT:%.*]]
; CHECK:       loop.1:
; CHECK-NEXT:    [[GEP_1:%.*]] = getelementptr inbounds i32, ptr [[P]], i32 [[IV_NEXT]]
; CHECK-NEXT:    [[V_1:%.*]] = load volatile i32, ptr [[GEP_1]], align 4
; CHECK-NEXT:    store volatile i32 [[V_1]], ptr [[GEP_1]], align 4
; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw i32 [[IV]], 2
; CHECK-NEXT:    [[CMP_1:%.*]] = icmp ult i32 [[IV_NEXT_1]], [[END]]
; CHECK-NEXT:    br i1 [[CMP_1]], label [[LOOP_2:%.*]], label [[EXIT]]
; CHECK:       loop.2:
; CHECK-NEXT:    [[GEP_2:%.*]] = getelementptr inbounds i32, ptr [[P]], i32 [[IV_NEXT_1]]
; CHECK-NEXT:    [[V_2:%.*]] = load volatile i32, ptr [[GEP_2]], align 4
; CHECK-NEXT:    store volatile i32 [[V_2]], ptr [[GEP_2]], align 4
; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw i32 [[IV]], 3
; CHECK-NEXT:    [[CMP_2:%.*]] = icmp ult i32 [[IV_NEXT_2]], [[END]]
; CHECK-NEXT:    br i1 [[CMP_2]], label [[LOOP_3]], label [[EXIT]]
; CHECK:       loop.3:
; CHECK-NEXT:    [[GEP_3:%.*]] = getelementptr inbounds i32, ptr [[P]], i32 [[IV_NEXT_2]]
; CHECK-NEXT:    [[V_3:%.*]] = load volatile i32, ptr [[GEP_3]], align 4
; CHECK-NEXT:    store volatile i32 [[V_3]], ptr [[GEP_3]], align 4
; CHECK-NEXT:    [[IV_NEXT_3]] = add nuw i32 [[IV]], 4
; CHECK-NEXT:    [[CMP_3:%.*]] = icmp ult i32 [[IV_NEXT_3]], [[END]]
; CHECK-NEXT:    br i1 [[CMP_3]], label [[LOOP]], label [[EXIT]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
entry:
  %end = add i32 %start, 5
  br label %loop

loop:
  %iv = phi i32 [ %start, %entry ], [ %iv.next, %loop ]
  %gep = getelementptr inbounds i32, ptr %p, i32 %iv
  %v = load volatile i32, ptr %gep
  %b0 = call i32 @llvm.bswap.i32(i32 %v)
  %b1 = call i32 @llvm.bswap.i32(i32 %b0)
  %b2 = call i32 @llvm.bswap.i32(i32 %b1)
  %b3 = call i32 @llvm.bswap.i32(i32 %b2)
  %b4 = call i32 @llvm.bswap.i32(i32 %b3)
  %b5 = call i32 @llvm.bswap.i32(i32 %b4)
  %b6 = call i32 @llvm.bswap.i32(i32 %b5)
  %b7 = call i32 @llvm.bswap.i32(i32 %b6)
  %b8 = call i32 @llvm.bswap.i32(i32 %b7)
  %b9 = call i32 @llvm.bswap.i32(i32 %b8)
  %b10 = call i32 @llvm.bswap.i32(i32 %b9)
  %b11 = call i32 @llvm.bswap.i32(i32 %b10)
  %b12 = call i32 @llvm.bswap.i32(i32 %b11)
  %b13 = call i32 @llvm.bswap.i32(i32 %b12)
  %b14 = call i32 @llvm.bswap.i32(i32 %b13)
  %b15 = call i32 @llvm.bswap.i32(i32 %b14)
  %b16 = call i32 @llvm.bswap.i32(i32 %b15)
  %b17 = call i32 @llvm.bswap.i32(i32 %b16)
  %b18 = call i32 @llvm.bswap.i32(i32 %b17)
  %b19 = call i32 @llvm.bswap.i32(i32 %b18)
  %b20 = call i32 @llvm.bswap.i32(i32 %b19)
  %b21 = call i32 @llvm.bswap.i32(i32 %b20)
  %b22 = call i32 @llvm.bswap.i32(i32 %b21)
  %b23 = call i32 @llvm.bswap.i32(i32 %b22)
  %b24 = call i32 @llvm.bswap.i32(i32 %b23)
  %b25 = call i32 @llvm.bswap.i32(i32 %b24)
  store volatile i32 %b25, ptr %gep
  %iv.next = add nuw i32 %iv, 1
  %cmp = icmp ult i32 %iv.next, %end
  br i1 %cmp, label %loop, label %exit

exit:
  ret void
}