llvm/test/Transforms/LoadStoreVectorizer/X86/vectorize-i8-nested-add.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -o - -S -load-store-vectorizer -dce %s | FileCheck %s

; Make sure LoadStoreVectorizer vectorizes the loads below.
; In order to prove that the vectorization is safe, it tries to
; match nested adds and find an expression that adds a constant
; value to an existing index and the result doesn't overflow.

target triple = "x86_64--"

define void @ld_v4i8_add_nsw(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_nsw(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = add nsw i32 [[V0:%.*]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add nsw i32 [[V1:%.*]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP133:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP184:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP41]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP82]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP133]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP184]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %tmp = add nsw i32 %v0, -1
  %tmp1 = add nsw i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add nsw i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nsw i32 %v0, 1
  %tmp10 = add nsw i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nsw i32 %v0, 2
  %tmp15 = add nsw i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

define void @ld_v4i8_add_nuw(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_nuw(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = add nuw i32 [[V0:%.*]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add nuw i32 [[V1:%.*]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP133:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP184:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP41]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP82]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP133]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP184]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %tmp = add nuw i32 %v0, -1
  %tmp1 = add nuw i32 %v1, %tmp
  %tmp2 = zext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add nuw i32 %v1, %v0
  %tmp6 = zext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nuw i32 %v0, 1
  %tmp10 = add nuw i32 %v1, %tmp9
  %tmp11 = zext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nuw i32 %v0, 2
  %tmp15 = add nuw i32 %v1, %tmp14
  %tmp16 = zext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

; Apply different operand orders for the nested add sequences
define void @ld_v4i8_add_nsw_operand_orders(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_nsw_operand_orders(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = add nsw i32 [[V0:%.*]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add nsw i32 [[V1:%.*]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP133:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP184:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP41]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP82]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP133]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP184]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %tmp = add nsw i32 %v0, -1
  %tmp1 = add nsw i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add nsw i32 %v0, %v1
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nsw i32 %v0, 1
  %tmp10 = add nsw i32 %tmp9, %v1
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nsw i32 %v0, 2
  %tmp15 = add nsw i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

; Apply different operand orders for the nested add sequences
define void @ld_v4i8_add_nuw_operand_orders(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_nuw_operand_orders(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = add nuw i32 [[V0:%.*]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add nuw i32 [[V1:%.*]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP133:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP184:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP41]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP82]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP133]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP184]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %tmp = add nuw i32 %v0, -1
  %tmp1 = add nuw i32 %v1, %tmp
  %tmp2 = zext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add nuw i32 %v0, %v1
  %tmp6 = zext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nuw i32 %v0, 1
  %tmp10 = add nuw i32 %tmp9, %v1
  %tmp11 = zext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nuw i32 %v0, 2
  %tmp15 = add nuw i32 %v1, %tmp14
  %tmp16 = zext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

define void @ld_v4i8_add_known_bits(i32 %ind0, i32 %ind1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_known_bits(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 4
; CHECK-NEXT:    [[TMP:%.*]] = add i32 [[V0]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[V1]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i8, i8* [[TMP3]], align 1
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <3 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <3 x i8>, <3 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <3 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <3 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP183:%.*]] = extractelement <3 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP4]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP132]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP183]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 4
  %tmp = add i32 %v0, -1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

define void @ld_v4i8_add_known_bits1(i32 %ind0, i32 %ind1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_known_bits1(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 4
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP183:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP44]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP132]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP183]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 4
  %tmp = add i32 %v0, 3
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

define void @ld_v4i8_add_known_bits_by_assume(i32 %ind0, i32 %ind1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_known_bits_by_assume(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 3
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 3
; CHECK-NEXT:    [[AND_I:%.*]] = and i32 [[V0]], 3
; CHECK-NEXT:    [[CMP_I:%.*]] = icmp eq i32 [[AND_I]], 0
; CHECK-NEXT:    [[AND_I_1:%.*]] = and i32 [[V1]], 3
; CHECK-NEXT:    [[CMP_I_1:%.*]] = icmp eq i32 [[AND_I_1]], 0
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I]])
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I_1]])
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP183:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP44]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP132]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP183]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 3
  %v1 = mul i32 %ind1, 3
  %and.i = and i32 %v0, 3
  %cmp.i = icmp eq i32 %and.i, 0
  %and.i.1 = and i32 %v1, 3
  %cmp.i.1 = icmp eq i32 %and.i.1, 0
  call void @llvm.assume(i1 %cmp.i)
  call void @llvm.assume(i1 %cmp.i.1)
  %tmp = add i32 %v0, 3
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

declare void @llvm.assume(i1)

define void @ld_v4i8_add_assume_on_arg(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_assume_on_arg(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[AND_I:%.*]] = and i32 [[V0:%.*]], 3
; CHECK-NEXT:    [[CMP_I:%.*]] = icmp eq i32 [[AND_I]], 0
; CHECK-NEXT:    [[AND_I_1:%.*]] = and i32 [[V1:%.*]], 3
; CHECK-NEXT:    [[CMP_I_1:%.*]] = icmp eq i32 [[AND_I_1]], 0
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I]])
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I_1]])
; CHECK-NEXT:    [[TMP:%.*]] = add nsw i32 [[V0]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[V1]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i8, i8* [[TMP3]], align 1
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <3 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <3 x i8>, <3 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <3 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <3 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP183:%.*]] = extractelement <3 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP4]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP132]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP183]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %and.i = and i32 %v0, 3
  %cmp.i = icmp eq i32 %and.i, 0
  %and.i.1 = and i32 %v1, 3
  %cmp.i.1 = icmp eq i32 %and.i.1, 0
  call void @llvm.assume(i1 %cmp.i)
  call void @llvm.assume(i1 %cmp.i.1)
  %tmp = add nsw i32 %v0, -1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nsw i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nsw i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

define void @ld_v4i8_add_assume_on_arg1(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_assume_on_arg1(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[AND_I:%.*]] = and i32 [[V0:%.*]], 3
; CHECK-NEXT:    [[CMP_I:%.*]] = icmp eq i32 [[AND_I]], 0
; CHECK-NEXT:    [[AND_I_1:%.*]] = and i32 [[V1:%.*]], 3
; CHECK-NEXT:    [[CMP_I_1:%.*]] = icmp eq i32 [[AND_I_1]], 0
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I]])
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP_I_1]])
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <4 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i8>, <4 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <4 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <4 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP183:%.*]] = extractelement <4 x i8> [[TMP1]], i32 2
; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <4 x i8> [[TMP1]], i32 3
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP44]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP132]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP183]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %and.i = and i32 %v0, 3
  %cmp.i = icmp eq i32 %and.i, 0
  %and.i.1 = and i32 %v1, 3
  %cmp.i.1 = icmp eq i32 %and.i.1, 0
  call void @llvm.assume(i1 %cmp.i)
  call void @llvm.assume(i1 %cmp.i.1)
  %tmp = add nsw i32 %v0, 3
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nsw i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nsw i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}

; Address computations are partly separated by control flow and with llvm.assume placed
; in the second basic block

define void @ld_v2i8_add_different_contexts(i32 %ind0, i32 %ind1, i8* %src, <2 x i8>* %dst) {
; CHECK-LABEL: @ld_v2i8_add_different_contexts(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 3
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[BIT_COND:%.*]] = icmp eq i32 [[V1]], 0
; CHECK-NEXT:    br i1 [[BIT_COND]], label [[BB_LOADS:%.*]], label [[BB_SKIP:%.*]]
; CHECK:       bb.loads:
; CHECK-NEXT:    call void @llvm.assume(i1 [[BIT_COND]])
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <2 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <2 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <2 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <2 x i8> undef, i8 [[TMP42]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    store <2 x i8> [[TMP20]], <2 x i8>* [[DST:%.*]]
; CHECK-NEXT:    br label [[BB_SKIP]]
; CHECK:       bb.skip:
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 3
  %tmp5 = add i32 %v1, %v0
  %bit_cond = icmp eq i32 %v1, 0
  br i1 %bit_cond, label %bb.loads, label %bb.skip

bb.loads:
  call void @llvm.assume(i1 %bit_cond)
  %tmp = add nsw i32 %v0, 1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp19 = insertelement <2 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <2 x i8> %tmp19, i8 %tmp8, i32 1
  store <2 x i8> %tmp20, <2 x i8>* %dst
  br label %bb.skip

bb.skip:
  ret void
}

; Same as ld_v2i8_add_different_contexts but with llvm.assume placed between loads

define void @ld_v2i8_add_different_contexts1(i32 %ind0, i32 %ind1, i8* %src, <2 x i8>* %dst) {
; CHECK-LABEL: @ld_v2i8_add_different_contexts1(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 3
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[BIT_COND:%.*]] = icmp eq i32 [[V1]], 0
; CHECK-NEXT:    br i1 [[BIT_COND]], label [[BB_LOADS:%.*]], label [[BB_SKIP:%.*]]
; CHECK:       bb.loads:
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <2 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <2 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <2 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    call void @llvm.assume(i1 [[BIT_COND]])
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <2 x i8> undef, i8 [[TMP42]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    store <2 x i8> [[TMP20]], <2 x i8>* [[DST:%.*]]
; CHECK-NEXT:    br label [[BB_SKIP]]
; CHECK:       bb.skip:
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 3
  %tmp5 = add i32 %v1, %v0
  %bit_cond = icmp eq i32 %v1, 0
  br i1 %bit_cond, label %bb.loads, label %bb.skip

bb.loads:
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  call void @llvm.assume(i1 %bit_cond)
  %tmp = add nsw i32 %v0, 1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp19 = insertelement <2 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <2 x i8> %tmp19, i8 %tmp8, i32 1
  store <2 x i8> %tmp20, <2 x i8>* %dst
  br label %bb.skip

bb.skip:
  ret void
}

; llvm.assume is placed between loads in a single basic block

define void @ld_v2i8_add_context(i32 %ind0, i32 %ind1, i8* %src, <2 x i8>* %dst) {
; CHECK-LABEL: @ld_v2i8_add_context(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 3
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <2 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <2 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <2 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[BIT_COND:%.*]] = icmp eq i32 [[TMP5]], 0
; CHECK-NEXT:    call void @llvm.assume(i1 [[BIT_COND]])
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <2 x i8> undef, i8 [[TMP42]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    store <2 x i8> [[TMP20]], <2 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 3
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %bit_cond = icmp eq i32 %tmp5, 0
  call void @llvm.assume(i1 %bit_cond)
  %tmp = add nsw i32 %v0, 1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp19 = insertelement <2 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <2 x i8> %tmp19, i8 %tmp8, i32 1
  store <2 x i8> %tmp20, <2 x i8>* %dst
  ret void
}

; Placing llvm.assume after all the loads and stores in the basic block still works

define void @ld_v2i8_add_context1(i32 %ind0, i32 %ind1, i8* %src, <2 x i8>* %dst) {
; CHECK-LABEL: @ld_v2i8_add_context1(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[V0:%.*]] = mul i32 [[IND0:%.*]], 4
; CHECK-NEXT:    [[V1:%.*]] = mul i32 [[IND1:%.*]], 3
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[TMP7]] to <2 x i8>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* [[TMP0]], align 1
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <2 x i8> [[TMP1]], i32 0
; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <2 x i8> [[TMP1]], i32 1
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <2 x i8> undef, i8 [[TMP42]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x i8> [[TMP19]], i8 [[TMP81]], i32 1
; CHECK-NEXT:    store <2 x i8> [[TMP20]], <2 x i8>* [[DST:%.*]]
; CHECK-NEXT:    [[BIT_COND:%.*]] = icmp eq i32 [[TMP5]], 0
; CHECK-NEXT:    call void @llvm.assume(i1 [[BIT_COND]])
; CHECK-NEXT:    ret void
;
bb:
  %v0 = mul i32 %ind0, 4
  %v1 = mul i32 %ind1, 3
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp = add nsw i32 %v0, 1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp19 = insertelement <2 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <2 x i8> %tmp19, i8 %tmp8, i32 1
  store <2 x i8> %tmp20, <2 x i8>* %dst
  %bit_cond = icmp eq i32 %tmp5, 0
  call void @llvm.assume(i1 %bit_cond)
  ret void
}

; Make sure we don't vectorize the loads below because the source of
; sext instructions doesn't have the nsw flag or known bits allowing
; to apply the vectorization.

define void @ld_v4i8_add_not_safe(i32 %v0, i32 %v1, i8* %src, <4 x i8>* %dst) {
; CHECK-LABEL: @ld_v4i8_add_not_safe(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = add nsw i32 [[V0:%.*]], -1
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[V1:%.*]], [[TMP]]
; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[SRC:%.*]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i8, i8* [[TMP3]], align 1
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[V1]], [[V0]]
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[SRC]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load i8, i8* [[TMP7]], align 1
; CHECK-NEXT:    [[TMP9:%.*]] = add nsw i32 [[V0]], 1
; CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[V1]], [[TMP9]]
; CHECK-NEXT:    [[TMP11:%.*]] = sext i32 [[TMP10]] to i64
; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, i8* [[SRC]], i64 [[TMP11]]
; CHECK-NEXT:    [[TMP13:%.*]] = load i8, i8* [[TMP12]], align 1
; CHECK-NEXT:    [[TMP14:%.*]] = add nsw i32 [[V0]], 2
; CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[V1]], [[TMP14]]
; CHECK-NEXT:    [[TMP16:%.*]] = sext i32 [[TMP15]] to i64
; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i8, i8* [[SRC]], i64 [[TMP16]]
; CHECK-NEXT:    [[TMP18:%.*]] = load i8, i8* [[TMP17]], align 1
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i8> undef, i8 [[TMP4]], i32 0
; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i8> [[TMP19]], i8 [[TMP8]], i32 1
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP13]], i32 2
; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP18]], i32 3
; CHECK-NEXT:    store <4 x i8> [[TMP22]], <4 x i8>* [[DST:%.*]]
; CHECK-NEXT:    ret void
;
bb:
  %tmp = add nsw i32 %v0, -1
  %tmp1 = add i32 %v1, %tmp
  %tmp2 = sext i32 %tmp1 to i64
  %tmp3 = getelementptr inbounds i8, i8* %src, i64 %tmp2
  %tmp4 = load i8, i8* %tmp3, align 1
  %tmp5 = add i32 %v1, %v0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i8, i8* %src, i64 %tmp6
  %tmp8 = load i8, i8* %tmp7, align 1
  %tmp9 = add nsw i32 %v0, 1
  %tmp10 = add i32 %v1, %tmp9
  %tmp11 = sext i32 %tmp10 to i64
  %tmp12 = getelementptr inbounds i8, i8* %src, i64 %tmp11
  %tmp13 = load i8, i8* %tmp12, align 1
  %tmp14 = add nsw i32 %v0, 2
  %tmp15 = add i32 %v1, %tmp14
  %tmp16 = sext i32 %tmp15 to i64
  %tmp17 = getelementptr inbounds i8, i8* %src, i64 %tmp16
  %tmp18 = load i8, i8* %tmp17, align 1
  %tmp19 = insertelement <4 x i8> undef, i8 %tmp4, i32 0
  %tmp20 = insertelement <4 x i8> %tmp19, i8 %tmp8, i32 1
  %tmp21 = insertelement <4 x i8> %tmp20, i8 %tmp13, i32 2
  %tmp22 = insertelement <4 x i8> %tmp21, i8 %tmp18, i32 3
  store <4 x i8> %tmp22, <4 x i8>* %dst
  ret void
}