; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -mtriple=thumbv7-unknown-linux-gnueabihf -arm-parallel-dsp -dce %s -S -o - | FileCheck %s

define i32 @first_mul_invalid(ptr nocapture readonly %in, ptr nocapture readonly %b) {
; CHECK-LABEL: @first_mul_invalid(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = load i16, ptr [[IN:%.*]], align 2
; CHECK-NEXT:    [[CONV:%.*]] = sext i16 [[TMP0]] to i32
; CHECK-NEXT:    [[TMP1:%.*]] = load i16, ptr [[B:%.*]], align 2
; CHECK-NEXT:    [[CONV2:%.*]] = sext i16 [[TMP1]] to i32
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 @bar(i32 [[CONV]], i32 [[CONV2]])
; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -1
; CHECK-NEXT:    [[TMP2:%.*]] = load i16, ptr [[ARRAYIDX3]], align 2
; CHECK-NEXT:    [[CONV4:%.*]] = sext i16 [[TMP2]] to i32
; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
; CHECK-NEXT:    [[CONV6:%.*]] = sext i16 [[TMP3]] to i32
; CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[CONV6]], [[CONV4]]
; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[MUL]], [[CALL]]
; CHECK-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -3
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX13]], align 2
; CHECK-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 2
; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX9]], align 2
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.arm.smladx(i32 [[TMP5]], i32 [[TMP7]], i32 [[ADD]])
; CHECK-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -5
; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX25]], align 2
; CHECK-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 4
; CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX21]], align 2
; CHECK-NEXT:    [[TMP13:%.*]] = call i32 @llvm.arm.smladx(i32 [[TMP10]], i32 [[TMP12]], i32 [[TMP8]])
; CHECK-NEXT:    ret i32 [[TMP13]]
;
entry:
  %0 = load i16, ptr %in, align 2
  %conv = sext i16 %0 to i32
  %1 = load i16, ptr %b, align 2
  %conv2 = sext i16 %1 to i32
  %call = tail call i32 @bar(i32 %conv, i32 %conv2)
  %arrayidx3 = getelementptr inbounds i16, ptr %in, i32 -1
  %2 = load i16, ptr %arrayidx3, align 2
  %conv4 = sext i16 %2 to i32
  %arrayidx5 = getelementptr inbounds i16, ptr %b, i32 1
  %3 = load i16, ptr %arrayidx5, align 2
  %conv6 = sext i16 %3 to i32
  %mul = mul nsw i32 %conv6, %conv4
  %add = add i32 %mul, %call
  %arrayidx7 = getelementptr inbounds i16, ptr %in, i32 -2
  %4 = load i16, ptr %arrayidx7, align 2
  %conv8 = sext i16 %4 to i32
  %arrayidx9 = getelementptr inbounds i16, ptr %b, i32 2
  %5 = load i16, ptr %arrayidx9, align 2
  %conv10 = sext i16 %5 to i32
  %mul11 = mul nsw i32 %conv10, %conv8
  %add12 = add i32 %add, %mul11
  %arrayidx13 = getelementptr inbounds i16, ptr %in, i32 -3
  %6 = load i16, ptr %arrayidx13, align 2
  %conv14 = sext i16 %6 to i32
  %arrayidx15 = getelementptr inbounds i16, ptr %b, i32 3
  %7 = load i16, ptr %arrayidx15, align 2
  %conv16 = sext i16 %7 to i32
  %mul17 = mul nsw i32 %conv16, %conv14
  %add18 = add i32 %add12, %mul17
  %arrayidx19 = getelementptr inbounds i16, ptr %in, i32 -4
  %8 = load i16, ptr %arrayidx19, align 2
  %conv20 = sext i16 %8 to i32
  %arrayidx21 = getelementptr inbounds i16, ptr %b, i32 4
  %9 = load i16, ptr %arrayidx21, align 2
  %conv22 = sext i16 %9 to i32
  %mul23 = mul nsw i32 %conv22, %conv20
  %add24 = add i32 %add18, %mul23
  %arrayidx25 = getelementptr inbounds i16, ptr %in, i32 -5
  %10 = load i16, ptr %arrayidx25, align 2
  %conv26 = sext i16 %10 to i32
  %arrayidx27 = getelementptr inbounds i16, ptr %b, i32 5
  %11 = load i16, ptr %arrayidx27, align 2
  %conv28 = sext i16 %11 to i32
  %mul29 = mul nsw i32 %conv28, %conv26
  %add30 = add i32 %add24, %mul29
  ret i32 %add30
}

define i32 @with_no_acc_input(ptr nocapture readonly %in, ptr nocapture readonly %b) {
; CHECK-LABEL: @with_no_acc_input(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i16, ptr [[IN:%.*]], i32 -1
; CHECK-NEXT:    [[LD_2:%.*]] = load i16, ptr [[ARRAYIDX3]], align 2
; CHECK-NEXT:    [[CONV4:%.*]] = sext i16 [[LD_2]] to i32
; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[B:%.*]], i32 1
; CHECK-NEXT:    [[LD_3:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
; CHECK-NEXT:    [[CONV6:%.*]] = sext i16 [[LD_3]] to i32
; CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[CONV6]], [[CONV4]]
; CHECK-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -3
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX13]], align 2
; CHECK-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 2
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX9]], align 2
; CHECK-NEXT:    [[TMP4:%.*]] = call i32 @llvm.arm.smladx(i32 [[TMP1]], i32 [[TMP3]], i32 [[MUL]])
; CHECK-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -5
; CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX25]], align 2
; CHECK-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 4
; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX21]], align 2
; CHECK-NEXT:    [[TMP9:%.*]] = call i32 @llvm.arm.smladx(i32 [[TMP6]], i32 [[TMP8]], i32 [[TMP4]])
; CHECK-NEXT:    ret i32 [[TMP9]]
;
entry:
  %arrayidx3 = getelementptr inbounds i16, ptr %in, i32 -1
  %ld.2 = load i16, ptr %arrayidx3, align 2
  %conv4 = sext i16 %ld.2 to i32
  %arrayidx5 = getelementptr inbounds i16, ptr %b, i32 1
  %ld.3 = load i16, ptr %arrayidx5, align 2
  %conv6 = sext i16 %ld.3 to i32
  %mul = mul nsw i32 %conv6, %conv4
  %arrayidx7 = getelementptr inbounds i16, ptr %in, i32 -2
  %ld.4 = load i16, ptr %arrayidx7, align 2
  %conv8 = sext i16 %ld.4 to i32
  %arrayidx9 = getelementptr inbounds i16, ptr %b, i32 2
  %ld.5 = load i16, ptr %arrayidx9, align 2
  %conv10 = sext i16 %ld.5 to i32
  %mul11 = mul nsw i32 %conv10, %conv8
  %add12 = add i32 %mul, %mul11
  %arrayidx13 = getelementptr inbounds i16, ptr %in, i32 -3
  %ld.6 = load i16, ptr %arrayidx13, align 2
  %conv14 = sext i16 %ld.6 to i32
  %arrayidx15 = getelementptr inbounds i16, ptr %b, i32 3
  %ld.7 = load i16, ptr %arrayidx15, align 2
  %conv16 = sext i16 %ld.7 to i32
  %mul17 = mul nsw i32 %conv16, %conv14
  %add18 = add i32 %add12, %mul17
  %arrayidx19 = getelementptr inbounds i16, ptr %in, i32 -4
  %ld.8 = load i16, ptr %arrayidx19, align 2
  %conv20 = sext i16 %ld.8 to i32
  %arrayidx21 = getelementptr inbounds i16, ptr %b, i32 4
  %ld.9 = load i16, ptr %arrayidx21, align 2
  %conv22 = sext i16 %ld.9 to i32
  %mul23 = mul nsw i32 %conv22, %conv20
  %add24 = add i32 %add18, %mul23
  %arrayidx25 = getelementptr inbounds i16, ptr %in, i32 -5
  %ld.10 = load i16, ptr %arrayidx25, align 2
  %conv26 = sext i16 %ld.10 to i32
  %arrayidx27 = getelementptr inbounds i16, ptr %b, i32 5
  %ld.11 = load i16, ptr %arrayidx27, align 2
  %conv28 = sext i16 %ld.11 to i32
  %mul29 = mul nsw i32 %conv28, %conv26
  %add30 = add i32 %add24, %mul29
  ret i32 %add30
}

define i64 @with_64bit_acc(ptr nocapture readonly %in, ptr nocapture readonly %b) {
; CHECK-LABEL: @with_64bit_acc(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = load i16, ptr [[IN:%.*]], align 2
; CHECK-NEXT:    [[CONV:%.*]] = sext i16 [[TMP0]] to i32
; CHECK-NEXT:    [[TMP1:%.*]] = load i16, ptr [[B:%.*]], align 2
; CHECK-NEXT:    [[CONV2:%.*]] = sext i16 [[TMP1]] to i32
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 @bar(i32 [[CONV]], i32 [[CONV2]])
; CHECK-NEXT:    [[SEXT_0:%.*]] = sext i32 [[CALL]] to i64
; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -1
; CHECK-NEXT:    [[TMP2:%.*]] = load i16, ptr [[ARRAYIDX3]], align 2
; CHECK-NEXT:    [[CONV4:%.*]] = sext i16 [[TMP2]] to i32
; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = load i16, ptr [[ARRAYIDX5]], align 2
; CHECK-NEXT:    [[CONV6:%.*]] = sext i16 [[TMP3]] to i32
; CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[CONV6]], [[CONV4]]
; CHECK-NEXT:    [[SEXT_1:%.*]] = sext i32 [[MUL]] to i64
; CHECK-NEXT:    [[ADD:%.*]] = add i64 [[SEXT_0]], [[SEXT_1]]
; CHECK-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -3
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX13]], align 2
; CHECK-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 2
; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX9]], align 2
; CHECK-NEXT:    [[TMP8:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP5]], i32 [[TMP7]], i64 [[ADD]])
; CHECK-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i16, ptr [[IN]], i32 -5
; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX25]], align 2
; CHECK-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 4
; CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX21]], align 2
; CHECK-NEXT:    [[TMP13:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP10]], i32 [[TMP12]], i64 [[TMP8]])
; CHECK-NEXT:    ret i64 [[TMP13]]
;
entry:
  %0 = load i16, ptr %in, align 2
  %conv = sext i16 %0 to i32
  %1 = load i16, ptr %b, align 2
  %conv2 = sext i16 %1 to i32
  %call = tail call i32 @bar(i32 %conv, i32 %conv2)
  %sext.0 = sext i32 %call to i64
  %arrayidx3 = getelementptr inbounds i16, ptr %in, i32 -1
  %2 = load i16, ptr %arrayidx3, align 2
  %conv4 = sext i16 %2 to i32
  %arrayidx5 = getelementptr inbounds i16, ptr %b, i32 1
  %3 = load i16, ptr %arrayidx5, align 2
  %conv6 = sext i16 %3 to i32
  %mul = mul nsw i32 %conv6, %conv4
  %sext.1 = sext i32 %mul to i64
  %add = add i64 %sext.0, %sext.1
  %arrayidx7 = getelementptr inbounds i16, ptr %in, i32 -2
  %4 = load i16, ptr %arrayidx7, align 2
  %conv8 = sext i16 %4 to i32
  %arrayidx9 = getelementptr inbounds i16, ptr %b, i32 2
  %5 = load i16, ptr %arrayidx9, align 2
  %conv10 = sext i16 %5 to i32
  %mul11 = mul nsw i32 %conv10, %conv8
  %sext.2 = sext i32 %mul11 to i64
  %add12 = add i64 %add, %sext.2
  %arrayidx13 = getelementptr inbounds i16, ptr %in, i32 -3
  %6 = load i16, ptr %arrayidx13, align 2
  %conv14 = sext i16 %6 to i32
  %arrayidx15 = getelementptr inbounds i16, ptr %b, i32 3
  %7 = load i16, ptr %arrayidx15, align 2
  %conv16 = sext i16 %7 to i32
  %mul17 = mul nsw i32 %conv16, %conv14
  %sext.3 = sext i32 %mul17 to i64
  %add18 = add i64 %add12, %sext.3
  %arrayidx19 = getelementptr inbounds i16, ptr %in, i32 -4
  %8 = load i16, ptr %arrayidx19, align 2
  %conv20 = sext i16 %8 to i32
  %arrayidx21 = getelementptr inbounds i16, ptr %b, i32 4
  %9 = load i16, ptr %arrayidx21, align 2
  %conv22 = sext i16 %9 to i32
  %mul23 = mul nsw i32 %conv22, %conv20
  %sext.4 = sext i32 %mul23 to i64
  %add24 = add i64 %add18, %sext.4
  %arrayidx25 = getelementptr inbounds i16, ptr %in, i32 -5
  %10 = load i16, ptr %arrayidx25, align 2
  %conv26 = sext i16 %10 to i32
  %arrayidx27 = getelementptr inbounds i16, ptr %b, i32 5
  %11 = load i16, ptr %arrayidx27, align 2
  %conv28 = sext i16 %11 to i32
  %mul29 = mul nsw i32 %conv28, %conv26
  %sext.5 = sext i32 %mul29 to i64
  %add30 = add i64 %add24, %sext.5
  ret i64 %add30
}

define i64 @with_64bit_add_acc(ptr nocapture readonly %px.10756.unr, ptr nocapture readonly %py.8757.unr, i32 %acc) {
; CHECK-LABEL: @with_64bit_add_acc(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[SUM_3758_UNR:%.*]] = sext i32 [[ACC:%.*]] to i64
; CHECK-NEXT:    br label [[BB_1:%.*]]
; CHECK:       bb.1:
; CHECK-NEXT:    [[INCDEC_PTR184_EPIL:%.*]] = getelementptr inbounds i16, ptr [[PX_10756_UNR:%.*]], i32 1
; CHECK-NEXT:    [[TMP216:%.*]] = load i16, ptr [[PX_10756_UNR]], align 2
; CHECK-NEXT:    [[CONV185_EPIL:%.*]] = sext i16 [[TMP216]] to i32
; CHECK-NEXT:    [[INCDEC_PTR186_EPIL:%.*]] = getelementptr inbounds i16, ptr [[PY_8757_UNR:%.*]], i32 -1
; CHECK-NEXT:    [[TMP217:%.*]] = load i16, ptr [[PY_8757_UNR]], align 2
; CHECK-NEXT:    [[CONV187_EPIL:%.*]] = sext i16 [[TMP217]] to i32
; CHECK-NEXT:    [[MUL_EPIL:%.*]] = mul nsw i32 [[CONV187_EPIL]], [[CONV185_EPIL]]
; CHECK-NEXT:    [[CONV188_EPIL:%.*]] = sext i32 [[MUL_EPIL]] to i64
; CHECK-NEXT:    [[ADD189_EPIL:%.*]] = add nsw i64 [[SUM_3758_UNR]], [[CONV188_EPIL]]
; CHECK-NEXT:    [[INCDEC_PTR190_EPIL:%.*]] = getelementptr inbounds i16, ptr [[PX_10756_UNR]], i32 2
; CHECK-NEXT:    [[TMP218:%.*]] = load i16, ptr [[INCDEC_PTR184_EPIL]], align 2
; CHECK-NEXT:    [[CONV191_EPIL:%.*]] = sext i16 [[TMP218]] to i32
; CHECK-NEXT:    [[TMP219:%.*]] = load i16, ptr [[INCDEC_PTR186_EPIL]], align 2
; CHECK-NEXT:    [[CONV193_EPIL:%.*]] = sext i16 [[TMP219]] to i32
; CHECK-NEXT:    [[MUL194_EPIL:%.*]] = mul nsw i32 [[CONV193_EPIL]], [[CONV191_EPIL]]
; CHECK-NEXT:    [[TMP0:%.*]] = sext i32 [[MUL194_EPIL]] to i64
; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[TMP0]], [[ADD189_EPIL]]
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[INCDEC_PTR190_EPIL]], align 2
; CHECK-NEXT:    [[INCDEC_PTR199_EPIL:%.*]] = getelementptr inbounds i16, ptr [[PY_8757_UNR]], i32 -3
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[INCDEC_PTR199_EPIL]], align 2
; CHECK-NEXT:    [[TMP6:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP5]], i32 [[TMP3]], i64 [[TMP1]])
; CHECK-NEXT:    ret i64 [[TMP6]]
;
entry:
  %sum.3758.unr = sext i32 %acc to i64
  br label %bb.1

bb.1:
  %incdec.ptr184.epil = getelementptr inbounds i16, ptr %px.10756.unr, i32 1
  %tmp216 = load i16, ptr %px.10756.unr, align 2
  %conv185.epil = sext i16 %tmp216 to i32
  %incdec.ptr186.epil = getelementptr inbounds i16, ptr %py.8757.unr, i32 -1
  %tmp217 = load i16, ptr %py.8757.unr, align 2
  %conv187.epil = sext i16 %tmp217 to i32
  %mul.epil = mul nsw i32 %conv187.epil, %conv185.epil
  %conv188.epil = sext i32 %mul.epil to i64
  %add189.epil = add nsw i64 %sum.3758.unr, %conv188.epil
  %incdec.ptr190.epil = getelementptr inbounds i16, ptr %px.10756.unr, i32 2
  %tmp218 = load i16, ptr %incdec.ptr184.epil, align 2
  %conv191.epil = sext i16 %tmp218 to i32
  %incdec.ptr192.epil = getelementptr inbounds i16, ptr %py.8757.unr, i32 -2
  %tmp219 = load i16, ptr %incdec.ptr186.epil, align 2
  %conv193.epil = sext i16 %tmp219 to i32
  %mul194.epil = mul nsw i32 %conv193.epil, %conv191.epil
  %conv195.epil = sext i32 %mul194.epil to i64
  %add196.epil = add nsw i64 %add189.epil, %conv195.epil
  %incdec.ptr197.epil = getelementptr inbounds i16, ptr %px.10756.unr, i32 3
  %tmp220 = load i16, ptr %incdec.ptr190.epil, align 2
  %conv198.epil = sext i16 %tmp220 to i32
  %incdec.ptr199.epil = getelementptr inbounds i16, ptr %py.8757.unr, i32 -3
  %tmp221 = load i16, ptr %incdec.ptr192.epil, align 2
  %conv200.epil = sext i16 %tmp221 to i32
  %mul201.epil = mul nsw i32 %conv200.epil, %conv198.epil
  %conv202.epil = sext i32 %mul201.epil to i64
  %add203.epil = add nsw i64 %add196.epil, %conv202.epil
  %tmp222 = load i16, ptr %incdec.ptr197.epil, align 2
  %conv205.epil = sext i16 %tmp222 to i32
  %tmp223 = load i16, ptr %incdec.ptr199.epil, align 2
  %conv207.epil = sext i16 %tmp223 to i32
  %mul208.epil = mul nsw i32 %conv207.epil, %conv205.epil
  %conv209.epil = sext i32 %mul208.epil to i64
  %add210.epil = add nsw i64 %add203.epil, %conv209.epil
  ret i64 %add210.epil
}

declare dso_local i32 @bar(i32, i32) local_unnamed_addr