test/Transforms/SCCP/with.overflow.ll

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

declare { i8, i1 } @llvm.uadd.with.overflow.i8(i8, i8)
declare { i8, i1 } @llvm.usub.with.overflow.i8(i8, i8)
declare { i8, i1 } @llvm.umul.with.overflow.i8(i8, i8)
declare { i8, i1 } @llvm.sadd.with.overflow.i8(i8, i8)
declare { i8, i1 } @llvm.ssub.with.overflow.i8(i8, i8)
declare { i8, i1 } @llvm.smul.with.overflow.i8(i8, i8)
declare { <2 x i8>, <2 x i1> } @llvm.uadd.with.overflow.v2i8(<2 x i8>, <2 x i8>)
declare void @use.i1(i1)

define void @unsigned_overflow(ptr %p) {
; CHECK-LABEL: @unsigned_overflow(
; CHECK-NEXT:    [[V0_100:%.*]] = load i8, ptr [[P:%.*]], align 1, !range [[RNG0:![0-9]+]]
; CHECK-NEXT:    [[V0_155:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG1:![0-9]+]]
; CHECK-NEXT:    [[V0_156:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG2:![0-9]+]]
; CHECK-NEXT:    [[V100_255:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG3:![0-9]+]]
; CHECK-NEXT:    [[V99_255:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG4:![0-9]+]]
; CHECK-NEXT:    [[V1_2:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG5:![0-9]+]]
; CHECK-NEXT:    [[V1_3:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG6:![0-9]+]]
; CHECK-NEXT:    [[WO1:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[V0_100]], i8 [[V0_155]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO2:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[V0_100]], i8 [[V0_156]])
; CHECK-NEXT:    [[OV2:%.*]] = extractvalue { i8, i1 } [[WO2]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV2]])
; CHECK-NEXT:    [[WO3:%.*]] = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 [[V100_255]], i8 [[V0_100]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO4:%.*]] = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 [[V99_255]], i8 [[V0_100]])
; CHECK-NEXT:    [[OV4:%.*]] = extractvalue { i8, i1 } [[WO4]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV4]])
; CHECK-NEXT:    [[WO5:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[V0_100]], i8 [[V1_2]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO6:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[V0_100]], i8 [[V1_3]])
; CHECK-NEXT:    [[OV6:%.*]] = extractvalue { i8, i1 } [[WO6]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV6]])
; CHECK-NEXT:    ret void
;
  %v0_100 = load i8, ptr %p, !range !{i8 0, i8 101}
  %v0_155 = load i8, ptr %p, !range !{i8 0, i8 156}
  %v0_156 = load i8, ptr %p, !range !{i8 0, i8 157}
  %v100_255 = load i8, ptr %p, !range !{i8 100, i8 0}
  %v99_255 = load i8, ptr %p, !range !{i8 99, i8 0}
  %v1_2 = load i8, ptr %p, !range !{i8 1, i8 3}
  %v1_3 = load i8, ptr %p, !range !{i8 1, i8 4}

  %wo1 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 %v0_100, i8 %v0_155)
  %ov1 = extractvalue { i8, i1 } %wo1, 1
  call void @use.i1(i1 %ov1)

  %wo2 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 %v0_100, i8 %v0_156)
  %ov2 = extractvalue { i8, i1 } %wo2, 1
  call void @use.i1(i1 %ov2)

  %wo3 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 %v100_255, i8 %v0_100)
  %ov3 = extractvalue { i8, i1 } %wo3, 1
  call void @use.i1(i1 %ov3)

  %wo4 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 %v99_255, i8 %v0_100)
  %ov4 = extractvalue { i8, i1 } %wo4, 1
  call void @use.i1(i1 %ov4)

  %wo5 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 %v0_100, i8 %v1_2)
  %ov5 = extractvalue { i8, i1 } %wo5, 1
  call void @use.i1(i1 %ov5)

  %wo6 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 %v0_100, i8 %v1_3)
  %ov6 = extractvalue { i8, i1 } %wo6, 1
  call void @use.i1(i1 %ov6)
  ret void
}

define void @signed_overflow(ptr %p) {
; CHECK-LABEL: @signed_overflow(
; CHECK-NEXT:    [[V0_100:%.*]] = load i8, ptr [[P:%.*]], align 1, !range [[RNG0]]
; CHECK-NEXT:    [[V0_27:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG7:![0-9]+]]
; CHECK-NEXT:    [[V0_28:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG8:![0-9]+]]
; CHECK-NEXT:    [[VM27_0:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG9:![0-9]+]]
; CHECK-NEXT:    [[VM28_0:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG10:![0-9]+]]
; CHECK-NEXT:    [[V1_4:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG11:![0-9]+]]
; CHECK-NEXT:    [[V1_5:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG12:![0-9]+]]
; CHECK-NEXT:    [[WO1:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[V0_100]], i8 [[V0_27]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO2:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[V0_100]], i8 [[V0_28]])
; CHECK-NEXT:    [[OV2:%.*]] = extractvalue { i8, i1 } [[WO2]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV2]])
; CHECK-NEXT:    [[WO3:%.*]] = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 [[V0_100]], i8 [[VM27_0]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO4:%.*]] = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 [[V0_100]], i8 [[VM28_0]])
; CHECK-NEXT:    [[OV4:%.*]] = extractvalue { i8, i1 } [[WO4]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV4]])
; CHECK-NEXT:    [[WO5:%.*]] = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 [[V0_27]], i8 [[V1_4]])
; CHECK-NEXT:    call void @use.i1(i1 false)
; CHECK-NEXT:    [[WO6:%.*]] = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 [[V0_27]], i8 [[V1_5]])
; CHECK-NEXT:    [[OV6:%.*]] = extractvalue { i8, i1 } [[WO6]], 1
; CHECK-NEXT:    call void @use.i1(i1 [[OV6]])
; CHECK-NEXT:    ret void
;
  %v0_100 = load i8, ptr %p, !range !{i8 0, i8 101}
  %v0_27 = load i8, ptr %p, !range !{i8 0, i8 28}
  %v0_28 = load i8, ptr %p, !range !{i8 0, i8 29}
  %vm27_0 = load i8, ptr %p, !range !{i8 -27, i8 0}
  %vm28_0 = load i8, ptr %p, !range !{i8 -28, i8 0}
  %v1_4 = load i8, ptr %p, !range !{i8 1, i8 5}
  %v1_5 = load i8, ptr %p, !range !{i8 1, i8 6}

  %wo1 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %v0_100, i8 %v0_27)
  %ov1 = extractvalue { i8, i1 } %wo1, 1
  call void @use.i1(i1 %ov1)

  %wo2 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %v0_100, i8 %v0_28)
  %ov2 = extractvalue { i8, i1 } %wo2, 1
  call void @use.i1(i1 %ov2)

  %wo3 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %v0_100, i8 %vm27_0)
  %ov3 = extractvalue { i8, i1 } %wo3, 1
  call void @use.i1(i1 %ov3)

  %wo4 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %v0_100, i8 %vm28_0)
  %ov4 = extractvalue { i8, i1 } %wo4, 1
  call void @use.i1(i1 %ov4)

  %wo5 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 %v0_27, i8 %v1_4)
  %ov5 = extractvalue { i8, i1 } %wo5, 1
  call void @use.i1(i1 %ov5)

  %wo6 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 %v0_27, i8 %v1_5)
  %ov6 = extractvalue { i8, i1 } %wo6, 1
  call void @use.i1(i1 %ov6)
  ret void
}

define void @unsigned_result(ptr %p) {
; CHECK-LABEL: @unsigned_result(
; CHECK-NEXT:    [[V0_20:%.*]] = load i8, ptr [[P:%.*]], align 1, !range [[RNG13:![0-9]+]]
; CHECK-NEXT:    [[V20_40:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG14:![0-9]+]]
; CHECK-NEXT:    [[V0_10:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG15:![0-9]+]]
; CHECK-NEXT:    [[V2_3:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG16:![0-9]+]]
; CHECK-NEXT:    [[WO1:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[V0_20]], i8 [[V20_40]])
; CHECK-NEXT:    [[RES1:%.*]] = extractvalue { i8, i1 } [[WO1]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ugt i8 [[RES1]], 20
; CHECK-NEXT:    call void @use.i1(i1 [[CMP1_3]])
; CHECK-NEXT:    [[CMP1_4:%.*]] = icmp ult i8 [[RES1]], 60
; CHECK-NEXT:    call void @use.i1(i1 [[CMP1_4]])
; CHECK-NEXT:    [[WO2:%.*]] = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 [[V0_10]], i8 [[V20_40]])
; CHECK-NEXT:    [[RES2:%.*]] = extractvalue { i8, i1 } [[WO2]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP2_3:%.*]] = icmp ugt i8 [[RES2]], -40
; CHECK-NEXT:    call void @use.i1(i1 [[CMP2_3]])
; CHECK-NEXT:    [[CMP2_4:%.*]] = icmp ult i8 [[RES2]], -10
; CHECK-NEXT:    call void @use.i1(i1 [[CMP2_4]])
; CHECK-NEXT:    [[WO3:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[V20_40]], i8 [[V2_3]])
; CHECK-NEXT:    [[RES3:%.*]] = extractvalue { i8, i1 } [[WO3]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP3_3:%.*]] = icmp ugt i8 [[RES3]], 40
; CHECK-NEXT:    call void @use.i1(i1 [[CMP3_3]])
; CHECK-NEXT:    [[CMP3_4:%.*]] = icmp ult i8 [[RES3]], 120
; CHECK-NEXT:    call void @use.i1(i1 [[CMP3_4]])
; CHECK-NEXT:    ret void
;
  %v0_20 = load i8, ptr %p, !range !{i8 0, i8 21}
  %v20_40 = load i8, ptr %p, !range !{i8 20, i8 41}
  %v0_10 = load i8, ptr %p, !range !{i8 0, i8 11}
  %v2_3 = load i8, ptr %p, !range !{i8 2, i8 4}
  %wo1 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 %v0_20, i8 %v20_40)
  %res1 = extractvalue { i8, i1 } %wo1, 0
  %cmp1.1 = icmp uge i8 %res1, 20
  call void @use.i1(i1 %cmp1.1)
  %cmp1.2 = icmp ule i8 %res1, 60
  call void @use.i1(i1 %cmp1.2)
  %cmp1.3 = icmp ugt i8 %res1, 20
  call void @use.i1(i1 %cmp1.3)
  %cmp1.4 = icmp ult i8 %res1, 60
  call void @use.i1(i1 %cmp1.4)

  ; This case actually does overflow, but we can still determine the range.
  %wo2 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 %v0_10, i8 %v20_40)
  %res2 = extractvalue { i8, i1 } %wo2, 0
  %cmp2.1 = icmp uge i8 %res2, -40
  call void @use.i1(i1 %cmp2.1)
  %cmp2.2 = icmp ule i8 %res2, -10
  call void @use.i1(i1 %cmp2.2)
  %cmp2.3 = icmp ugt i8 %res2, -40
  call void @use.i1(i1 %cmp2.3)
  %cmp2.4 = icmp ult i8 %res2, -10
  call void @use.i1(i1 %cmp2.4)

  %wo3 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 %v20_40, i8 %v2_3)
  %res3 = extractvalue { i8, i1 } %wo3, 0
  %cmp3.1 = icmp uge i8 %res3, 40
  call void @use.i1(i1 %cmp3.1)
  %cmp3.2 = icmp ule i8 %res3, 120
  call void @use.i1(i1 %cmp3.2)
  %cmp3.3 = icmp ugt i8 %res3, 40
  call void @use.i1(i1 %cmp3.3)
  %cmp3.4 = icmp ult i8 %res3, 120
  call void @use.i1(i1 %cmp3.4)
  ret void
}

define void @signed_result(ptr %p) {
; CHECK-LABEL: @signed_result(
; CHECK-NEXT:    [[V0_20:%.*]] = load i8, ptr [[P:%.*]], align 1, !range [[RNG13]]
; CHECK-NEXT:    [[V20_40:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG14]]
; CHECK-NEXT:    [[V0_10:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG15]]
; CHECK-NEXT:    [[V2_3:%.*]] = load i8, ptr [[P]], align 1, !range [[RNG16]]
; CHECK-NEXT:    [[WO1:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[V0_20]], i8 [[V20_40]])
; CHECK-NEXT:    [[RES1:%.*]] = extractvalue { i8, i1 } [[WO1]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ugt i8 [[RES1]], 20
; CHECK-NEXT:    call void @use.i1(i1 [[CMP1_3]])
; CHECK-NEXT:    [[CMP1_4:%.*]] = icmp ult i8 [[RES1]], 60
; CHECK-NEXT:    call void @use.i1(i1 [[CMP1_4]])
; CHECK-NEXT:    [[WO2:%.*]] = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 [[V0_10]], i8 [[V20_40]])
; CHECK-NEXT:    [[RES2:%.*]] = extractvalue { i8, i1 } [[WO2]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP2_3:%.*]] = icmp ugt i8 [[RES2]], -40
; CHECK-NEXT:    call void @use.i1(i1 [[CMP2_3]])
; CHECK-NEXT:    [[CMP2_4:%.*]] = icmp ult i8 [[RES2]], -10
; CHECK-NEXT:    call void @use.i1(i1 [[CMP2_4]])
; CHECK-NEXT:    [[WO3:%.*]] = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 [[V20_40]], i8 [[V2_3]])
; CHECK-NEXT:    [[RES3:%.*]] = extractvalue { i8, i1 } [[WO3]], 0
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    call void @use.i1(i1 true)
; CHECK-NEXT:    [[CMP3_3:%.*]] = icmp ugt i8 [[RES3]], 40
; CHECK-NEXT:    call void @use.i1(i1 [[CMP3_3]])
; CHECK-NEXT:    [[CMP3_4:%.*]] = icmp ult i8 [[RES3]], 120
; CHECK-NEXT:    call void @use.i1(i1 [[CMP3_4]])
; CHECK-NEXT:    ret void
;
  %v0_20 = load i8, ptr %p, !range !{i8 0, i8 21}
  %v20_40 = load i8, ptr %p, !range !{i8 20, i8 41}
  %v0_10 = load i8, ptr %p, !range !{i8 0, i8 11}
  %v2_3 = load i8, ptr %p, !range !{i8 2, i8 4}
  %wo1 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %v0_20, i8 %v20_40)
  %res1 = extractvalue { i8, i1 } %wo1, 0
  %cmp1.1 = icmp uge i8 %res1, 20
  call void @use.i1(i1 %cmp1.1)
  %cmp1.2 = icmp ule i8 %res1, 60
  call void @use.i1(i1 %cmp1.2)
  %cmp1.3 = icmp ugt i8 %res1, 20
  call void @use.i1(i1 %cmp1.3)
  %cmp1.4 = icmp ult i8 %res1, 60
  call void @use.i1(i1 %cmp1.4)

  %wo2 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 %v0_10, i8 %v20_40)
  %res2 = extractvalue { i8, i1 } %wo2, 0
  %cmp2.1 = icmp uge i8 %res2, -40
  call void @use.i1(i1 %cmp2.1)
  %cmp2.2 = icmp ule i8 %res2, -10
  call void @use.i1(i1 %cmp2.2)
  %cmp2.3 = icmp ugt i8 %res2, -40
  call void @use.i1(i1 %cmp2.3)
  %cmp2.4 = icmp ult i8 %res2, -10
  call void @use.i1(i1 %cmp2.4)

  %wo3 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 %v20_40, i8 %v2_3)
  %res3 = extractvalue { i8, i1 } %wo3, 0
  %cmp3.1 = icmp uge i8 %res3, 40
  call void @use.i1(i1 %cmp3.1)
  %cmp3.2 = icmp ule i8 %res3, 120
  call void @use.i1(i1 %cmp3.2)
  %cmp3.3 = icmp ugt i8 %res3, 40
  call void @use.i1(i1 %cmp3.3)
  %cmp3.4 = icmp ult i8 %res3, 120
  call void @use.i1(i1 %cmp3.4)
  ret void
}

; SCCP doesn't really support vector ranges yet, just make sure we don't crash.
define <2 x i1> @vec(<2 x i8> %v1, <2 x i8> %v2) {
; CHECK-LABEL: @vec(
; CHECK-NEXT:    [[WO:%.*]] = call { <2 x i8>, <2 x i1> } @llvm.uadd.with.overflow.v2i8(<2 x i8> [[V1:%.*]], <2 x i8> [[V2:%.*]])
; CHECK-NEXT:    [[OV:%.*]] = extractvalue { <2 x i8>, <2 x i1> } [[WO]], 1
; CHECK-NEXT:    ret <2 x i1> [[OV]]
;
  %wo = call { <2 x i8>, <2 x i1> } @llvm.uadd.with.overflow.v2i8(<2 x i8> %v1, <2 x i8> %v2)
  %ov = extractvalue { <2 x i8>, <2 x i1> } %wo, 1
  ret <2 x i1> %ov
}