; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64-none-linux-gnu < %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel < %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI

define void @matrix_mul_unsigned(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i16 %val) {
; CHECK-SD-LABEL: matrix_mul_unsigned:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    dup v0.4h, w3
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff8
; CHECK-SD-NEXT:  .LBB0_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #8
; CHECK-SD-NEXT:    ldp d1, d2, [x9]
; CHECK-SD-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-SD-NEXT:    add w0, w0, #8
; CHECK-SD-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q2, [x9]
; CHECK-SD-NEXT:    b.ne .LBB0_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_unsigned:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w8, w3, #0xffff
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    mov w8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff8
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:  .LBB0_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #8
; CHECK-GI-NEXT:    ldp d1, d2, [x9]
; CHECK-GI-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-GI-NEXT:    add w0, w0, #8
; CHECK-GI-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    stp q1, q2, [x9]
; CHECK-GI-NEXT:    b.ne .LBB0_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = zext i16 %val to i32
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 8
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967288
  %broadcast.splatinsert = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> undef, <4 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <4 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <4 x i16>, ptr %9, align 2
  %10 = zext <4 x i16> %wide.load to <4 x i32>
  %11 = zext <4 x i16> %wide.load30 to <4 x i32>
  %12 = mul nuw nsw <4 x i32> %broadcast.splat, %10
  %13 = mul nuw nsw <4 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <4 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 4
  %17 = bitcast ptr %16 to ptr
  store <4 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 8
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}

define void @matrix_mul_signed(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i16 %val) {
; CHECK-SD-LABEL: matrix_mul_signed:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    dup v0.4h, w3
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff8
; CHECK-SD-NEXT:  .LBB1_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, sxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #8
; CHECK-SD-NEXT:    ldp d1, d2, [x9]
; CHECK-SD-NEXT:    add x9, x1, w0, sxtw #2
; CHECK-SD-NEXT:    add w0, w0, #8
; CHECK-SD-NEXT:    smull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    smull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q2, [x9]
; CHECK-SD-NEXT:    b.ne .LBB1_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_signed:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    sxth w8, w3
; CHECK-GI-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    sxtw x8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff8
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:  .LBB1_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, sxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #8
; CHECK-GI-NEXT:    ldp d1, d2, [x9]
; CHECK-GI-NEXT:    add x9, x1, w0, sxtw #2
; CHECK-GI-NEXT:    add w0, w0, #8
; CHECK-GI-NEXT:    smull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    smull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    stp q1, q2, [x9]
; CHECK-GI-NEXT:    b.ne .LBB1_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = sext i16 %val to i32
  %wide.trip.count = sext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 8
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967288
  %broadcast.splatinsert = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> undef, <4 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = sext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <4 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <4 x i16>, ptr %9, align 2
  %10 = sext <4 x i16> %wide.load to <4 x i32>
  %11 = sext <4 x i16> %wide.load30 to <4 x i32>
  %12 = mul nsw <4 x i32> %broadcast.splat, %10
  %13 = mul nsw <4 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <4 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 4
  %17 = bitcast ptr %16 to ptr
  store <4 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 8
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}


define void @matrix_mul_double_shuffle(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i16 %val) {
; CHECK-SD-LABEL: matrix_mul_double_shuffle:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    dup v0.4h, w3
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff8
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 killed $x0 def $x0
; CHECK-SD-NEXT:  .LBB2_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldrh w9, [x2], #16
; CHECK-SD-NEXT:    subs x8, x8, #8
; CHECK-SD-NEXT:    dup v1.4h, w9
; CHECK-SD-NEXT:    ubfiz x9, x0, #2, #32
; CHECK-SD-NEXT:    add w0, w0, #8
; CHECK-SD-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    str q1, [x1, x9]
; CHECK-SD-NEXT:    b.ne .LBB2_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_double_shuffle:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w9, w3, #0xffff
; CHECK-GI-NEXT:    adrp x8, .LCPI2_0
; CHECK-GI-NEXT:    dup v0.4s, w9
; CHECK-GI-NEXT:    mov w9, w0
; CHECK-GI-NEXT:    ldr q1, [x8, :lo12:.LCPI2_0]
; CHECK-GI-NEXT:    and x8, x9, #0xfffffff8
; CHECK-GI-NEXT:  .LBB2_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldrh w9, [x2], #16
; CHECK-GI-NEXT:    subs x8, x8, #8
; CHECK-GI-NEXT:    fmov s2, w9
; CHECK-GI-NEXT:    mov w9, w0
; CHECK-GI-NEXT:    add w0, w0, #8
; CHECK-GI-NEXT:    lsl x9, x9, #2
; CHECK-GI-NEXT:    tbl v2.16b, { v2.16b, v3.16b }, v1.16b
; CHECK-GI-NEXT:    mul v2.4s, v0.4s, v2.4s
; CHECK-GI-NEXT:    str q2, [x1, x9]
; CHECK-GI-NEXT:    b.ne .LBB2_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = zext i16 %val to i32
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 8
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967288
  %broadcast.splatinsert = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %g = getelementptr inbounds i16, ptr %A, i64 %index
  %val1 = load i16, ptr %g
  %splat.input.ext = zext i16 %val1 to i32
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %splat.input.ext, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> %broadcast.splat, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = mul nuw nsw <4 x i32> %broadcast.splat, %broadcast.splat32
  %7 = getelementptr inbounds i32, ptr %C, i64 %5
  %8 = bitcast ptr %7 to ptr
  store <4 x i32> %6, ptr %8, align 4
  %index.next = add i64 %index, 8
  %9 = icmp eq i64 %index.next, %n.vec
  br i1 %9, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}


define void @larger_smull(ptr nocapture noundef readonly %x, i16 noundef %y, ptr noalias nocapture noundef writeonly %s, i32 noundef %n) {
; CHECK-SD-LABEL: larger_smull:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    cmp w3, #1
; CHECK-SD-NEXT:    b.lt .LBB3_8
; CHECK-SD-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-SD-NEXT:    cmp w3, #15
; CHECK-SD-NEXT:    mov w8, w3
; CHECK-SD-NEXT:    b.hi .LBB3_3
; CHECK-SD-NEXT:  // %bb.2:
; CHECK-SD-NEXT:    mov x9, xzr
; CHECK-SD-NEXT:    b .LBB3_6
; CHECK-SD-NEXT:  .LBB3_3: // %vector.ph
; CHECK-SD-NEXT:    dup v0.8h, w1
; CHECK-SD-NEXT:    and x9, x8, #0xfffffff0
; CHECK-SD-NEXT:    add x10, x2, #32
; CHECK-SD-NEXT:    add x11, x0, #16
; CHECK-SD-NEXT:    and x12, x8, #0xfffffff0
; CHECK-SD-NEXT:  .LBB3_4: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldp q1, q2, [x11, #-16]
; CHECK-SD-NEXT:    subs x12, x12, #16
; CHECK-SD-NEXT:    add x11, x11, #32
; CHECK-SD-NEXT:    smull2 v3.4s, v0.8h, v1.8h
; CHECK-SD-NEXT:    smull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    smull2 v4.4s, v0.8h, v2.8h
; CHECK-SD-NEXT:    smull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q3, [x10, #-32]
; CHECK-SD-NEXT:    stp q2, q4, [x10], #64
; CHECK-SD-NEXT:    b.ne .LBB3_4
; CHECK-SD-NEXT:  // %bb.5: // %middle.block
; CHECK-SD-NEXT:    cmp x9, x8
; CHECK-SD-NEXT:    b.eq .LBB3_8
; CHECK-SD-NEXT:  .LBB3_6: // %for.body.preheader1
; CHECK-SD-NEXT:    sxth w10, w1
; CHECK-SD-NEXT:    add x11, x2, x9, lsl #2
; CHECK-SD-NEXT:    add x12, x0, x9, lsl #1
; CHECK-SD-NEXT:    sub x8, x8, x9
; CHECK-SD-NEXT:  .LBB3_7: // %for.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldrsh w9, [x12], #2
; CHECK-SD-NEXT:    subs x8, x8, #1
; CHECK-SD-NEXT:    mul w9, w9, w10
; CHECK-SD-NEXT:    str w9, [x11], #4
; CHECK-SD-NEXT:    b.ne .LBB3_7
; CHECK-SD-NEXT:  .LBB3_8: // %for.cond.cleanup
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: larger_smull:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    cmp w3, #0
; CHECK-GI-NEXT:    b.le .LBB3_7
; CHECK-GI-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-GI-NEXT:    sxth w8, w1
; CHECK-GI-NEXT:    mov x10, xzr
; CHECK-GI-NEXT:    cmp w3, #16
; CHECK-GI-NEXT:    mov w9, w3
; CHECK-GI-NEXT:    b.lo .LBB3_5
; CHECK-GI-NEXT:  // %bb.2: // %vector.ph
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    and x10, x9, #0xfffffff0
; CHECK-GI-NEXT:    add x11, x2, #32
; CHECK-GI-NEXT:    add x12, x0, #16
; CHECK-GI-NEXT:    and x13, x9, #0xfffffff0
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:  .LBB3_3: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldp q1, q2, [x12, #-16]
; CHECK-GI-NEXT:    mov x14, x11
; CHECK-GI-NEXT:    subs x13, x13, #16
; CHECK-GI-NEXT:    add x12, x12, #32
; CHECK-GI-NEXT:    mov d3, v1.d[1]
; CHECK-GI-NEXT:    mov d4, v2.d[1]
; CHECK-GI-NEXT:    smull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    smull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    smull v3.4s, v0.4h, v3.4h
; CHECK-GI-NEXT:    smull v4.4s, v0.4h, v4.4h
; CHECK-GI-NEXT:    stp q1, q3, [x14, #-32]!
; CHECK-GI-NEXT:    stp q2, q4, [x11], #64
; CHECK-GI-NEXT:    b.ne .LBB3_3
; CHECK-GI-NEXT:  // %bb.4: // %middle.block
; CHECK-GI-NEXT:    cmp x10, x9
; CHECK-GI-NEXT:    b.eq .LBB3_7
; CHECK-GI-NEXT:  .LBB3_5: // %for.body.preheader1
; CHECK-GI-NEXT:    add x11, x2, x10, lsl #2
; CHECK-GI-NEXT:    add x12, x0, x10, lsl #1
; CHECK-GI-NEXT:    sub x9, x9, x10
; CHECK-GI-NEXT:  .LBB3_6: // %for.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldrsh w10, [x12], #2
; CHECK-GI-NEXT:    subs x9, x9, #1
; CHECK-GI-NEXT:    mul w10, w10, w8
; CHECK-GI-NEXT:    str w10, [x11], #4
; CHECK-GI-NEXT:    b.ne .LBB3_6
; CHECK-GI-NEXT:  .LBB3_7: // %for.cond.cleanup
; CHECK-GI-NEXT:    ret
entry:
  %conv1 = sext i16 %y to i32
  %cmp8 = icmp sgt i32 %n, 0
  br i1 %cmp8, label %for.body.preheader, label %for.cond.cleanup

for.body.preheader:                               ; preds = %entry
  %wide.trip.count = zext i32 %n to i64
  %min.iters.check = icmp ult i32 %n, 16
  br i1 %min.iters.check, label %for.body.preheader14, label %vector.ph

vector.ph:                                        ; preds = %for.body.preheader
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <8 x i32> poison, i32 %conv1, i64 0
  %broadcast.splat = shufflevector <8 x i32> %broadcast.splatinsert, <8 x i32> poison, <8 x i32> zeroinitializer
  %broadcast.splatinsert12 = insertelement <8 x i32> poison, i32 %conv1, i64 0
  %broadcast.splat13 = shufflevector <8 x i32> %broadcast.splatinsert12, <8 x i32> poison, <8 x i32> zeroinitializer
  br label %vector.body

vector.body:                                      ; preds = %vector.body, %vector.ph
  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
  %0 = getelementptr inbounds i16, ptr %x, i64 %index
  %1 = bitcast ptr %0 to ptr
  %wide.load = load <8 x i16>, ptr %1, align 2
  %2 = getelementptr inbounds i16, ptr %0, i64 8
  %3 = bitcast ptr %2 to ptr
  %wide.load11 = load <8 x i16>, ptr %3, align 2
  %4 = sext <8 x i16> %wide.load to <8 x i32>
  %5 = sext <8 x i16> %wide.load11 to <8 x i32>
  %6 = mul nsw <8 x i32> %broadcast.splat, %4
  %7 = mul nsw <8 x i32> %broadcast.splat13, %5
  %8 = getelementptr inbounds i32, ptr %s, i64 %index
  %9 = bitcast ptr %8 to ptr
  store <8 x i32> %6, ptr %9, align 4
  %10 = getelementptr inbounds i32, ptr %8, i64 8
  %11 = bitcast ptr %10 to ptr
  store <8 x i32> %7, ptr %11, align 4
  %index.next = add nuw i64 %index, 16
  %12 = icmp eq i64 %index.next, %n.vec
  br i1 %12, label %middle.block, label %vector.body

middle.block:                                     ; preds = %vector.body
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader14

for.body.preheader14:                             ; preds = %for.body.preheader, %middle.block
  %indvars.iv.ph = phi i64 [ 0, %for.body.preheader ], [ %n.vec, %middle.block ]
  br label %for.body

for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %entry
  ret void

for.body:                                         ; preds = %for.body.preheader14, %for.body
  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader14 ]
  %arrayidx = getelementptr inbounds i16, ptr %x, i64 %indvars.iv
  %13 = load i16, ptr %arrayidx, align 2
  %conv = sext i16 %13 to i32
  %mul = mul nsw i32 %conv, %conv1
  %arrayidx3 = getelementptr inbounds i32, ptr %s, i64 %indvars.iv
  store i32 %mul, ptr %arrayidx3, align 4
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
}


define void @larger_umull(ptr nocapture noundef readonly %x, i16 noundef %y, ptr noalias nocapture noundef writeonly %s, i32 noundef %n) {
; CHECK-SD-LABEL: larger_umull:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    cmp w3, #1
; CHECK-SD-NEXT:    b.lt .LBB4_8
; CHECK-SD-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-SD-NEXT:    cmp w3, #15
; CHECK-SD-NEXT:    mov w8, w3
; CHECK-SD-NEXT:    b.hi .LBB4_3
; CHECK-SD-NEXT:  // %bb.2:
; CHECK-SD-NEXT:    mov x9, xzr
; CHECK-SD-NEXT:    b .LBB4_6
; CHECK-SD-NEXT:  .LBB4_3: // %vector.ph
; CHECK-SD-NEXT:    dup v0.8h, w1
; CHECK-SD-NEXT:    and x9, x8, #0xfffffff0
; CHECK-SD-NEXT:    add x10, x2, #32
; CHECK-SD-NEXT:    add x11, x0, #16
; CHECK-SD-NEXT:    and x12, x8, #0xfffffff0
; CHECK-SD-NEXT:  .LBB4_4: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldp q1, q2, [x11, #-16]
; CHECK-SD-NEXT:    subs x12, x12, #16
; CHECK-SD-NEXT:    add x11, x11, #32
; CHECK-SD-NEXT:    umull2 v3.4s, v0.8h, v1.8h
; CHECK-SD-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    umull2 v4.4s, v0.8h, v2.8h
; CHECK-SD-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q3, [x10, #-32]
; CHECK-SD-NEXT:    stp q2, q4, [x10], #64
; CHECK-SD-NEXT:    b.ne .LBB4_4
; CHECK-SD-NEXT:  // %bb.5: // %middle.block
; CHECK-SD-NEXT:    cmp x9, x8
; CHECK-SD-NEXT:    b.eq .LBB4_8
; CHECK-SD-NEXT:  .LBB4_6: // %for.body.preheader1
; CHECK-SD-NEXT:    add x10, x2, x9, lsl #2
; CHECK-SD-NEXT:    add x11, x0, x9, lsl #1
; CHECK-SD-NEXT:    and w12, w1, #0xffff
; CHECK-SD-NEXT:    sub x8, x8, x9
; CHECK-SD-NEXT:  .LBB4_7: // %for.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldrh w9, [x11], #2
; CHECK-SD-NEXT:    subs x8, x8, #1
; CHECK-SD-NEXT:    mul w9, w9, w12
; CHECK-SD-NEXT:    str w9, [x10], #4
; CHECK-SD-NEXT:    b.ne .LBB4_7
; CHECK-SD-NEXT:  .LBB4_8: // %for.cond.cleanup
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: larger_umull:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    cmp w3, #0
; CHECK-GI-NEXT:    b.le .LBB4_7
; CHECK-GI-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    cmp w3, #16
; CHECK-GI-NEXT:    mov w9, w3
; CHECK-GI-NEXT:    b.lo .LBB4_5
; CHECK-GI-NEXT:  // %bb.2: // %vector.ph
; CHECK-GI-NEXT:    and x8, x9, #0xfffffff0
; CHECK-GI-NEXT:    add x10, x2, #32
; CHECK-GI-NEXT:    add x11, x0, #16
; CHECK-GI-NEXT:    and x12, x9, #0xfffffff0
; CHECK-GI-NEXT:  .LBB4_3: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    and w13, w1, #0xffff
; CHECK-GI-NEXT:    ldp q1, q2, [x11, #-16]
; CHECK-GI-NEXT:    dup v0.4s, w13
; CHECK-GI-NEXT:    mov x13, x10
; CHECK-GI-NEXT:    subs x12, x12, #16
; CHECK-GI-NEXT:    add x11, x11, #32
; CHECK-GI-NEXT:    mov d3, v1.d[1]
; CHECK-GI-NEXT:    mov d4, v2.d[1]
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    umull v3.4s, v0.4h, v3.4h
; CHECK-GI-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    umull v0.4s, v0.4h, v4.4h
; CHECK-GI-NEXT:    stp q1, q3, [x13, #-32]!
; CHECK-GI-NEXT:    stp q2, q0, [x10], #64
; CHECK-GI-NEXT:    b.ne .LBB4_3
; CHECK-GI-NEXT:  // %bb.4: // %middle.block
; CHECK-GI-NEXT:    cmp x8, x9
; CHECK-GI-NEXT:    b.eq .LBB4_7
; CHECK-GI-NEXT:  .LBB4_5: // %for.body.preheader1
; CHECK-GI-NEXT:    add x10, x2, x8, lsl #2
; CHECK-GI-NEXT:    add x11, x0, x8, lsl #1
; CHECK-GI-NEXT:    and w12, w1, #0xffff
; CHECK-GI-NEXT:    sub x8, x9, x8
; CHECK-GI-NEXT:  .LBB4_6: // %for.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldrh w9, [x11], #2
; CHECK-GI-NEXT:    subs x8, x8, #1
; CHECK-GI-NEXT:    mul w9, w9, w12
; CHECK-GI-NEXT:    str w9, [x10], #4
; CHECK-GI-NEXT:    b.ne .LBB4_6
; CHECK-GI-NEXT:  .LBB4_7: // %for.cond.cleanup
; CHECK-GI-NEXT:    ret
entry:
  %conv1 = zext i16 %y to i32
  %cmp8 = icmp sgt i32 %n, 0
  br i1 %cmp8, label %for.body.preheader, label %for.cond.cleanup

for.body.preheader:                               ; preds = %entry
  %wide.trip.count = zext i32 %n to i64
  %min.iters.check = icmp ult i32 %n, 16
  br i1 %min.iters.check, label %for.body.preheader14, label %vector.ph

vector.ph:                                        ; preds = %for.body.preheader
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <8 x i32> poison, i32 %conv1, i64 0
  %broadcast.splat = shufflevector <8 x i32> %broadcast.splatinsert, <8 x i32> poison, <8 x i32> zeroinitializer
  %broadcast.splatinsert12 = insertelement <8 x i32> poison, i32 %conv1, i64 0
  %broadcast.splat13 = shufflevector <8 x i32> %broadcast.splatinsert12, <8 x i32> poison, <8 x i32> zeroinitializer
  br label %vector.body

vector.body:                                      ; preds = %vector.body, %vector.ph
  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
  %0 = getelementptr inbounds i16, ptr %x, i64 %index
  %1 = bitcast ptr %0 to ptr
  %wide.load = load <8 x i16>, ptr %1, align 2
  %2 = getelementptr inbounds i16, ptr %0, i64 8
  %3 = bitcast ptr %2 to ptr
  %wide.load11 = load <8 x i16>, ptr %3, align 2
  %4 = zext <8 x i16> %wide.load to <8 x i32>
  %5 = zext <8 x i16> %wide.load11 to <8 x i32>
  %6 = mul nuw <8 x i32> %broadcast.splat, %4
  %7 = mul nuw <8 x i32> %broadcast.splat13, %5
  %8 = getelementptr inbounds i32, ptr %s, i64 %index
  %9 = bitcast ptr %8 to ptr
  store <8 x i32> %6, ptr %9, align 4
  %10 = getelementptr inbounds i32, ptr %8, i64 8
  %11 = bitcast ptr %10 to ptr
  store <8 x i32> %7, ptr %11, align 4
  %index.next = add nuw i64 %index, 16
  %12 = icmp eq i64 %index.next, %n.vec
  br i1 %12, label %middle.block, label %vector.body

middle.block:                                     ; preds = %vector.body
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader14

for.body.preheader14:                             ; preds = %for.body.preheader, %middle.block
  %indvars.iv.ph = phi i64 [ 0, %for.body.preheader ], [ %n.vec, %middle.block ]
  br label %for.body

for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %entry
  ret void

for.body:                                         ; preds = %for.body.preheader14, %for.body
  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader14 ]
  %arrayidx = getelementptr inbounds i16, ptr %x, i64 %indvars.iv
  %13 = load i16, ptr %arrayidx, align 2
  %conv = zext i16 %13 to i32
  %mul = mul nuw i32 %conv, %conv1
  %arrayidx3 = getelementptr inbounds i32, ptr %s, i64 %indvars.iv
  store i32 %mul, ptr %arrayidx3, align 4
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
}


define i16 @red_mla_dup_ext_u8_s8_s16(ptr noalias nocapture noundef readonly %A, i8 noundef %B, i32 noundef %n) {
; CHECK-SD-LABEL: red_mla_dup_ext_u8_s8_s16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    cbz w2, .LBB5_3
; CHECK-SD-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-SD-NEXT:    sxtb w9, w1
; CHECK-SD-NEXT:    cmp w2, #15
; CHECK-SD-NEXT:    mov w10, w2
; CHECK-SD-NEXT:    b.hi .LBB5_4
; CHECK-SD-NEXT:  // %bb.2:
; CHECK-SD-NEXT:    mov x11, xzr
; CHECK-SD-NEXT:    mov w8, wzr
; CHECK-SD-NEXT:    b .LBB5_7
; CHECK-SD-NEXT:  .LBB5_3:
; CHECK-SD-NEXT:    mov w8, wzr
; CHECK-SD-NEXT:    mov w0, w8
; CHECK-SD-NEXT:    ret
; CHECK-SD-NEXT:  .LBB5_4: // %vector.ph
; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
; CHECK-SD-NEXT:    and x11, x10, #0xfffffff0
; CHECK-SD-NEXT:    fmov s2, w9
; CHECK-SD-NEXT:    add x8, x0, #8
; CHECK-SD-NEXT:    and x12, x10, #0xfffffff0
; CHECK-SD-NEXT:  .LBB5_5: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldp d3, d4, [x8, #-8]
; CHECK-SD-NEXT:    subs x12, x12, #16
; CHECK-SD-NEXT:    add x8, x8, #16
; CHECK-SD-NEXT:    ushll v3.8h, v3.8b, #0
; CHECK-SD-NEXT:    ushll v4.8h, v4.8b, #0
; CHECK-SD-NEXT:    mla v0.8h, v3.8h, v2.h[0]
; CHECK-SD-NEXT:    mla v1.8h, v4.8h, v2.h[0]
; CHECK-SD-NEXT:    b.ne .LBB5_5
; CHECK-SD-NEXT:  // %bb.6: // %middle.block
; CHECK-SD-NEXT:    add v0.8h, v1.8h, v0.8h
; CHECK-SD-NEXT:    cmp x11, x10
; CHECK-SD-NEXT:    addv h0, v0.8h
; CHECK-SD-NEXT:    fmov w8, s0
; CHECK-SD-NEXT:    b.eq .LBB5_9
; CHECK-SD-NEXT:  .LBB5_7: // %for.body.preheader1
; CHECK-SD-NEXT:    sub x10, x10, x11
; CHECK-SD-NEXT:    add x11, x0, x11
; CHECK-SD-NEXT:  .LBB5_8: // %for.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldrb w12, [x11], #1
; CHECK-SD-NEXT:    subs x10, x10, #1
; CHECK-SD-NEXT:    madd w8, w12, w9, w8
; CHECK-SD-NEXT:    b.ne .LBB5_8
; CHECK-SD-NEXT:  .LBB5_9: // %for.cond.cleanup
; CHECK-SD-NEXT:    mov w0, w8
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: red_mla_dup_ext_u8_s8_s16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    cbz w2, .LBB5_3
; CHECK-GI-NEXT:  // %bb.1: // %for.body.preheader
; CHECK-GI-NEXT:    cmp w2, #16
; CHECK-GI-NEXT:    mov w8, w2
; CHECK-GI-NEXT:    b.hs .LBB5_4
; CHECK-GI-NEXT:  // %bb.2:
; CHECK-GI-NEXT:    movi d0, #0000000000000000
; CHECK-GI-NEXT:    mov x9, xzr
; CHECK-GI-NEXT:    b .LBB5_8
; CHECK-GI-NEXT:  .LBB5_3:
; CHECK-GI-NEXT:    mov w0, wzr
; CHECK-GI-NEXT:    ret
; CHECK-GI-NEXT:  .LBB5_4: // %vector.ph
; CHECK-GI-NEXT:    lsl w9, w1, #8
; CHECK-GI-NEXT:    movi v0.2d, #0000000000000000
; CHECK-GI-NEXT:    movi v1.2d, #0000000000000000
; CHECK-GI-NEXT:    add x10, x0, #8
; CHECK-GI-NEXT:    and x11, x8, #0xfffffff0
; CHECK-GI-NEXT:    sbfx w9, w9, #8, #8
; CHECK-GI-NEXT:    dup v2.8h, w9
; CHECK-GI-NEXT:    and x9, x8, #0xfffffff0
; CHECK-GI-NEXT:  .LBB5_5: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldp d3, d4, [x10, #-8]
; CHECK-GI-NEXT:    subs x11, x11, #16
; CHECK-GI-NEXT:    add x10, x10, #16
; CHECK-GI-NEXT:    ushll v3.8h, v3.8b, #0
; CHECK-GI-NEXT:    ushll v4.8h, v4.8b, #0
; CHECK-GI-NEXT:    mla v0.8h, v2.8h, v3.8h
; CHECK-GI-NEXT:    mla v1.8h, v2.8h, v4.8h
; CHECK-GI-NEXT:    b.ne .LBB5_5
; CHECK-GI-NEXT:  // %bb.6: // %middle.block
; CHECK-GI-NEXT:    add v0.8h, v1.8h, v0.8h
; CHECK-GI-NEXT:    cmp x9, x8
; CHECK-GI-NEXT:    addv h0, v0.8h
; CHECK-GI-NEXT:    b.ne .LBB5_8
; CHECK-GI-NEXT:  // %bb.7:
; CHECK-GI-NEXT:    fmov w0, s0
; CHECK-GI-NEXT:    ret
; CHECK-GI-NEXT:  .LBB5_8: // %for.body.preheader1
; CHECK-GI-NEXT:    sxtb w10, w1
; CHECK-GI-NEXT:    sub x8, x8, x9
; CHECK-GI-NEXT:    add x9, x0, x9
; CHECK-GI-NEXT:  .LBB5_9: // %for.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldrb w11, [x9], #1
; CHECK-GI-NEXT:    fmov w12, s0
; CHECK-GI-NEXT:    subs x8, x8, #1
; CHECK-GI-NEXT:    mul w11, w11, w10
; CHECK-GI-NEXT:    add w0, w11, w12, uxth
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    b.ne .LBB5_9
; CHECK-GI-NEXT:  // %bb.10: // %for.cond.cleanup
; CHECK-GI-NEXT:    ret
entry:
  %conv2 = sext i8 %B to i16
  %cmp10.not = icmp eq i32 %n, 0
  br i1 %cmp10.not, label %for.cond.cleanup, label %for.body.preheader

for.body.preheader:                               ; preds = %entry
  %wide.trip.count = zext i32 %n to i64
  %min.iters.check = icmp ult i32 %n, 16
  br i1 %min.iters.check, label %for.body.preheader17, label %vector.ph

vector.ph:                                        ; preds = %for.body.preheader
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <8 x i16> poison, i16 %conv2, i64 0
  %broadcast.splat = shufflevector <8 x i16> %broadcast.splatinsert, <8 x i16> poison, <8 x i32> zeroinitializer
  %broadcast.splatinsert15 = insertelement <8 x i16> poison, i16 %conv2, i64 0
  %broadcast.splat16 = shufflevector <8 x i16> %broadcast.splatinsert15, <8 x i16> poison, <8 x i32> zeroinitializer
  br label %vector.body

vector.body:                                      ; preds = %vector.body, %vector.ph
  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
  %vec.phi = phi <8 x i16> [ zeroinitializer, %vector.ph ], [ %8, %vector.body ]
  %vec.phi13 = phi <8 x i16> [ zeroinitializer, %vector.ph ], [ %9, %vector.body ]
  %0 = getelementptr inbounds i8, ptr %A, i64 %index
  %1 = bitcast ptr %0 to ptr
  %wide.load = load <8 x i8>, ptr %1, align 1
  %2 = getelementptr inbounds i8, ptr %0, i64 8
  %3 = bitcast ptr %2 to ptr
  %wide.load14 = load <8 x i8>, ptr %3, align 1
  %4 = zext <8 x i8> %wide.load to <8 x i16>
  %5 = zext <8 x i8> %wide.load14 to <8 x i16>
  %6 = mul nsw <8 x i16> %broadcast.splat, %4
  %7 = mul nsw <8 x i16> %broadcast.splat16, %5
  %8 = add <8 x i16> %6, %vec.phi
  %9 = add <8 x i16> %7, %vec.phi13
  %index.next = add nuw i64 %index, 16
  %10 = icmp eq i64 %index.next, %n.vec
  br i1 %10, label %middle.block, label %vector.body

middle.block:                                     ; preds = %vector.body
  %bin.rdx = add <8 x i16> %9, %8
  %11 = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> %bin.rdx)
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader17

for.body.preheader17:                             ; preds = %for.body.preheader, %middle.block
  %indvars.iv.ph = phi i64 [ 0, %for.body.preheader ], [ %n.vec, %middle.block ]
  %s.011.ph = phi i16 [ 0, %for.body.preheader ], [ %11, %middle.block ]
  br label %for.body

for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %entry
  %s.0.lcssa = phi i16 [ 0, %entry ], [ %11, %middle.block ], [ %add, %for.body ]
  ret i16 %s.0.lcssa

for.body:                                         ; preds = %for.body.preheader17, %for.body
  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader17 ]
  %s.011 = phi i16 [ %add, %for.body ], [ %s.011.ph, %for.body.preheader17 ]
  %arrayidx = getelementptr inbounds i8, ptr %A, i64 %indvars.iv
  %12 = load i8, ptr %arrayidx, align 1
  %13 = zext i8 %12 to i16
  %mul = mul nsw i16 %13, %conv2
  %add = add i16 %mul, %s.011
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
}

define i64 @red_mla_dup_ext_u8_s8_s64(ptr noalias noundef readonly captures(none) %A, i8 noundef %B, i32 noundef %n) {
; CHECK-SD-LABEL: red_mla_dup_ext_u8_s8_s64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $w1 killed $w1 def $x1
; CHECK-SD-NEXT:    cbz w2, .LBB6_3
; CHECK-SD-NEXT:  // %bb.1: // %iter.check
; CHECK-SD-NEXT:    cmp w2, #3
; CHECK-SD-NEXT:    mov w9, w2
; CHECK-SD-NEXT:    b.hi .LBB6_4
; CHECK-SD-NEXT:  // %bb.2:
; CHECK-SD-NEXT:    mov x10, xzr
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:    b .LBB6_13
; CHECK-SD-NEXT:  .LBB6_3:
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:    mov x0, x8
; CHECK-SD-NEXT:    ret
; CHECK-SD-NEXT:  .LBB6_4: // %vector.main.loop.iter.check
; CHECK-SD-NEXT:    cmp w2, #16
; CHECK-SD-NEXT:    b.hs .LBB6_6
; CHECK-SD-NEXT:  // %bb.5:
; CHECK-SD-NEXT:    mov x10, xzr
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:    b .LBB6_10
; CHECK-SD-NEXT:  .LBB6_6: // %vector.ph
; CHECK-SD-NEXT:    mov w8, w1
; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v1.2d, #0000000000000000
; CHECK-SD-NEXT:    sxtb x8, w8
; CHECK-SD-NEXT:    movi v3.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v2.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v6.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v4.2d, #0000000000000000
; CHECK-SD-NEXT:    and x11, x9, #0xc
; CHECK-SD-NEXT:    movi v7.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v5.2d, #0000000000000000
; CHECK-SD-NEXT:    and x10, x9, #0xfffffff0
; CHECK-SD-NEXT:    dup v16.4s, w8
; CHECK-SD-NEXT:    mov x8, x0
; CHECK-SD-NEXT:    and x12, x9, #0xfffffff0
; CHECK-SD-NEXT:  .LBB6_7: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr q17, [x8], #16
; CHECK-SD-NEXT:    subs x12, x12, #16
; CHECK-SD-NEXT:    ushll v18.8h, v17.8b, #0
; CHECK-SD-NEXT:    ushll2 v17.8h, v17.16b, #0
; CHECK-SD-NEXT:    ushll2 v19.4s, v18.8h, #0
; CHECK-SD-NEXT:    ushll v20.4s, v17.4h, #0
; CHECK-SD-NEXT:    ushll v18.4s, v18.4h, #0
; CHECK-SD-NEXT:    ushll2 v17.4s, v17.8h, #0
; CHECK-SD-NEXT:    smlal2 v2.2d, v16.4s, v19.4s
; CHECK-SD-NEXT:    smlal2 v4.2d, v16.4s, v20.4s
; CHECK-SD-NEXT:    smlal v6.2d, v16.2s, v20.2s
; CHECK-SD-NEXT:    smlal v3.2d, v16.2s, v19.2s
; CHECK-SD-NEXT:    smlal2 v0.2d, v16.4s, v18.4s
; CHECK-SD-NEXT:    smlal v7.2d, v16.2s, v17.2s
; CHECK-SD-NEXT:    smlal v1.2d, v16.2s, v18.2s
; CHECK-SD-NEXT:    smlal2 v5.2d, v16.4s, v17.4s
; CHECK-SD-NEXT:    b.ne .LBB6_7
; CHECK-SD-NEXT:  // %bb.8: // %middle.block
; CHECK-SD-NEXT:    add v1.2d, v1.2d, v6.2d
; CHECK-SD-NEXT:    add v3.2d, v3.2d, v7.2d
; CHECK-SD-NEXT:    cmp x10, x9
; CHECK-SD-NEXT:    add v0.2d, v0.2d, v4.2d
; CHECK-SD-NEXT:    add v2.2d, v2.2d, v5.2d
; CHECK-SD-NEXT:    add v1.2d, v1.2d, v3.2d
; CHECK-SD-NEXT:    add v0.2d, v0.2d, v2.2d
; CHECK-SD-NEXT:    add v0.2d, v1.2d, v0.2d
; CHECK-SD-NEXT:    addp d0, v0.2d
; CHECK-SD-NEXT:    fmov x8, d0
; CHECK-SD-NEXT:    b.eq .LBB6_15
; CHECK-SD-NEXT:  // %bb.9: // %vec.epilog.iter.check
; CHECK-SD-NEXT:    cbz x11, .LBB6_13
; CHECK-SD-NEXT:  .LBB6_10: // %vec.epilog.ph
; CHECK-SD-NEXT:    mov w11, w1
; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
; CHECK-SD-NEXT:    movi v3.2d, #0x000000000000ff
; CHECK-SD-NEXT:    sxtb x11, w11
; CHECK-SD-NEXT:    fmov d2, x8
; CHECK-SD-NEXT:    dup v1.2s, w11
; CHECK-SD-NEXT:    mov x11, x10
; CHECK-SD-NEXT:    and x10, x9, #0xfffffffc
; CHECK-SD-NEXT:    sub x8, x11, x10
; CHECK-SD-NEXT:    add x11, x0, x11
; CHECK-SD-NEXT:  .LBB6_11: // %vec.epilog.vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr s4, [x11], #4
; CHECK-SD-NEXT:    adds x8, x8, #4
; CHECK-SD-NEXT:    ushll v4.8h, v4.8b, #0
; CHECK-SD-NEXT:    ushll v4.4s, v4.4h, #0
; CHECK-SD-NEXT:    ushll v5.2d, v4.2s, #0
; CHECK-SD-NEXT:    ushll2 v4.2d, v4.4s, #0
; CHECK-SD-NEXT:    and v5.16b, v5.16b, v3.16b
; CHECK-SD-NEXT:    and v4.16b, v4.16b, v3.16b
; CHECK-SD-NEXT:    xtn v5.2s, v5.2d
; CHECK-SD-NEXT:    xtn v4.2s, v4.2d
; CHECK-SD-NEXT:    smlal v0.2d, v1.2s, v4.2s
; CHECK-SD-NEXT:    smlal v2.2d, v1.2s, v5.2s
; CHECK-SD-NEXT:    b.ne .LBB6_11
; CHECK-SD-NEXT:  // %bb.12: // %vec.epilog.middle.block
; CHECK-SD-NEXT:    add v0.2d, v2.2d, v0.2d
; CHECK-SD-NEXT:    cmp x10, x9
; CHECK-SD-NEXT:    addp d0, v0.2d
; CHECK-SD-NEXT:    fmov x8, d0
; CHECK-SD-NEXT:    b.eq .LBB6_15
; CHECK-SD-NEXT:  .LBB6_13: // %for.body.preheader
; CHECK-SD-NEXT:    sxtb x11, w1
; CHECK-SD-NEXT:    sub x9, x9, x10
; CHECK-SD-NEXT:    add x10, x0, x10
; CHECK-SD-NEXT:  .LBB6_14: // %for.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldrb w12, [x10], #1
; CHECK-SD-NEXT:    subs x9, x9, #1
; CHECK-SD-NEXT:    smaddl x8, w12, w11, x8
; CHECK-SD-NEXT:    b.ne .LBB6_14
; CHECK-SD-NEXT:  .LBB6_15: // %for.cond.cleanup
; CHECK-SD-NEXT:    mov x0, x8
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: red_mla_dup_ext_u8_s8_s64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    // kill: def $w1 killed $w1 def $x1
; CHECK-GI-NEXT:    cbz w2, .LBB6_7
; CHECK-GI-NEXT:  // %bb.1: // %iter.check
; CHECK-GI-NEXT:    movi d0, #0000000000000000
; CHECK-GI-NEXT:    mov x10, xzr
; CHECK-GI-NEXT:    cmp w2, #4
; CHECK-GI-NEXT:    mov w9, w2
; CHECK-GI-NEXT:    b.lo .LBB6_12
; CHECK-GI-NEXT:  // %bb.2: // %vector.main.loop.iter.check
; CHECK-GI-NEXT:    movi d0, #0000000000000000
; CHECK-GI-NEXT:    mov x10, xzr
; CHECK-GI-NEXT:    cmp w2, #16
; CHECK-GI-NEXT:    b.lo .LBB6_9
; CHECK-GI-NEXT:  // %bb.3: // %vector.ph
; CHECK-GI-NEXT:    mov w8, w1
; CHECK-GI-NEXT:    movi v0.2d, #0000000000000000
; CHECK-GI-NEXT:    movi v1.2d, #0000000000000000
; CHECK-GI-NEXT:    sxtb x8, w8
; CHECK-GI-NEXT:    movi v2.2d, #0000000000000000
; CHECK-GI-NEXT:    movi v3.2d, #0000000000000000
; CHECK-GI-NEXT:    movi v4.2d, #0000000000000000
; CHECK-GI-NEXT:    movi v6.2d, #0000000000000000
; CHECK-GI-NEXT:    and x10, x9, #0xfffffff0
; CHECK-GI-NEXT:    dup v5.2d, x8
; CHECK-GI-NEXT:    movi v7.2d, #0000000000000000
; CHECK-GI-NEXT:    and x8, x9, #0xc
; CHECK-GI-NEXT:    mov x11, x0
; CHECK-GI-NEXT:    and x12, x9, #0xfffffff0
; CHECK-GI-NEXT:    xtn v16.2s, v5.2d
; CHECK-GI-NEXT:    movi v5.2d, #0000000000000000
; CHECK-GI-NEXT:  .LBB6_4: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr q17, [x11], #16
; CHECK-GI-NEXT:    subs x12, x12, #16
; CHECK-GI-NEXT:    ushll v18.8h, v17.8b, #0
; CHECK-GI-NEXT:    ushll2 v17.8h, v17.16b, #0
; CHECK-GI-NEXT:    ushll v19.4s, v18.4h, #0
; CHECK-GI-NEXT:    ushll2 v18.4s, v18.8h, #0
; CHECK-GI-NEXT:    ushll v20.4s, v17.4h, #0
; CHECK-GI-NEXT:    ushll2 v17.4s, v17.8h, #0
; CHECK-GI-NEXT:    mov d21, v19.d[1]
; CHECK-GI-NEXT:    mov d22, v18.d[1]
; CHECK-GI-NEXT:    mov d23, v20.d[1]
; CHECK-GI-NEXT:    mov d24, v17.d[1]
; CHECK-GI-NEXT:    smlal v0.2d, v16.2s, v19.2s
; CHECK-GI-NEXT:    smlal v2.2d, v16.2s, v18.2s
; CHECK-GI-NEXT:    smlal v4.2d, v16.2s, v20.2s
; CHECK-GI-NEXT:    smlal v6.2d, v16.2s, v17.2s
; CHECK-GI-NEXT:    smlal v1.2d, v16.2s, v21.2s
; CHECK-GI-NEXT:    smlal v3.2d, v16.2s, v22.2s
; CHECK-GI-NEXT:    smlal v5.2d, v16.2s, v23.2s
; CHECK-GI-NEXT:    smlal v7.2d, v16.2s, v24.2s
; CHECK-GI-NEXT:    b.ne .LBB6_4
; CHECK-GI-NEXT:  // %bb.5: // %middle.block
; CHECK-GI-NEXT:    add v0.2d, v0.2d, v1.2d
; CHECK-GI-NEXT:    add v1.2d, v2.2d, v3.2d
; CHECK-GI-NEXT:    cmp x10, x9
; CHECK-GI-NEXT:    add v2.2d, v4.2d, v5.2d
; CHECK-GI-NEXT:    add v3.2d, v6.2d, v7.2d
; CHECK-GI-NEXT:    add v0.2d, v0.2d, v1.2d
; CHECK-GI-NEXT:    add v1.2d, v2.2d, v3.2d
; CHECK-GI-NEXT:    add v0.2d, v0.2d, v1.2d
; CHECK-GI-NEXT:    addp d0, v0.2d
; CHECK-GI-NEXT:    b.ne .LBB6_8
; CHECK-GI-NEXT:  // %bb.6:
; CHECK-GI-NEXT:    fmov x8, d0
; CHECK-GI-NEXT:    mov x0, x8
; CHECK-GI-NEXT:    ret
; CHECK-GI-NEXT:  .LBB6_7:
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    mov x0, x8
; CHECK-GI-NEXT:    ret
; CHECK-GI-NEXT:  .LBB6_8: // %vec.epilog.iter.check
; CHECK-GI-NEXT:    cbz x8, .LBB6_12
; CHECK-GI-NEXT:  .LBB6_9: // %vec.epilog.ph
; CHECK-GI-NEXT:    mov w8, w1
; CHECK-GI-NEXT:    mov v0.d[1], xzr
; CHECK-GI-NEXT:    movi v1.2d, #0000000000000000
; CHECK-GI-NEXT:    sxtb x8, w8
; CHECK-GI-NEXT:    mov x11, x10
; CHECK-GI-NEXT:    and x10, x9, #0xfffffffc
; CHECK-GI-NEXT:    dup v2.2d, x8
; CHECK-GI-NEXT:    sub x8, x11, x10
; CHECK-GI-NEXT:    add x11, x0, x11
; CHECK-GI-NEXT:    xtn v2.2s, v2.2d
; CHECK-GI-NEXT:  .LBB6_10: // %vec.epilog.vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr w12, [x11], #4
; CHECK-GI-NEXT:    adds x8, x8, #4
; CHECK-GI-NEXT:    fmov s3, w12
; CHECK-GI-NEXT:    uxtb w12, w12
; CHECK-GI-NEXT:    mov b4, v3.b[2]
; CHECK-GI-NEXT:    mov b5, v3.b[1]
; CHECK-GI-NEXT:    mov b6, v3.b[3]
; CHECK-GI-NEXT:    fmov s3, w12
; CHECK-GI-NEXT:    fmov w13, s4
; CHECK-GI-NEXT:    fmov w14, s5
; CHECK-GI-NEXT:    fmov w15, s6
; CHECK-GI-NEXT:    uxtb w13, w13
; CHECK-GI-NEXT:    uxtb w14, w14
; CHECK-GI-NEXT:    uxtb w15, w15
; CHECK-GI-NEXT:    fmov s4, w13
; CHECK-GI-NEXT:    mov v3.s[1], w14
; CHECK-GI-NEXT:    mov v4.s[1], w15
; CHECK-GI-NEXT:    smlal v0.2d, v2.2s, v3.2s
; CHECK-GI-NEXT:    smlal v1.2d, v2.2s, v4.2s
; CHECK-GI-NEXT:    b.ne .LBB6_10
; CHECK-GI-NEXT:  // %bb.11: // %vec.epilog.middle.block
; CHECK-GI-NEXT:    add v0.2d, v0.2d, v1.2d
; CHECK-GI-NEXT:    cmp x10, x9
; CHECK-GI-NEXT:    addp d0, v0.2d
; CHECK-GI-NEXT:    fmov x8, d0
; CHECK-GI-NEXT:    b.eq .LBB6_14
; CHECK-GI-NEXT:  .LBB6_12: // %for.body.preheader
; CHECK-GI-NEXT:    sxtb x11, w1
; CHECK-GI-NEXT:    sub x9, x9, x10
; CHECK-GI-NEXT:    add x10, x0, x10
; CHECK-GI-NEXT:  .LBB6_13: // %for.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldrb w8, [x10], #1
; CHECK-GI-NEXT:    fmov x12, d0
; CHECK-GI-NEXT:    subs x9, x9, #1
; CHECK-GI-NEXT:    madd x8, x8, x11, x12
; CHECK-GI-NEXT:    fmov d0, x8
; CHECK-GI-NEXT:    b.ne .LBB6_13
; CHECK-GI-NEXT:  .LBB6_14: // %for.cond.cleanup
; CHECK-GI-NEXT:    mov x0, x8
; CHECK-GI-NEXT:    ret
entry:
  %cmp5.not = icmp eq i32 %n, 0
  br i1 %cmp5.not, label %for.cond.cleanup, label %iter.check

iter.check:                                       ; preds = %entry
  %conv1 = sext i8 %B to i64
  %wide.trip.count = zext i32 %n to i64
  %min.iters.check = icmp ult i32 %n, 4
  br i1 %min.iters.check, label %for.body.preheader, label %vector.main.loop.iter.check

vector.main.loop.iter.check:                      ; preds = %iter.check
  %min.iters.check9 = icmp ult i32 %n, 16
  br i1 %min.iters.check9, label %vec.epilog.ph, label %vector.ph

vector.ph:                                        ; preds = %vector.main.loop.iter.check
  %n.mod.vf = and i64 %wide.trip.count, 12
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <16 x i64> poison, i64 %conv1, i64 0
  %broadcast.splat = shufflevector <16 x i64> %broadcast.splatinsert, <16 x i64> poison, <16 x i32> zeroinitializer
  br label %vector.body

vector.body:                                      ; preds = %vector.body, %vector.ph
  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
  %vec.phi = phi <16 x i64> [ zeroinitializer, %vector.ph ], [ %3, %vector.body ]
  %0 = getelementptr inbounds nuw i8, ptr %A, i64 %index
  %wide.load = load <16 x i8>, ptr %0, align 1
  %1 = zext <16 x i8> %wide.load to <16 x i64>
  %2 = mul nsw <16 x i64> %broadcast.splat, %1
  %3 = add <16 x i64> %2, %vec.phi
  %index.next = add nuw i64 %index, 16
  %4 = icmp eq i64 %index.next, %n.vec
  br i1 %4, label %middle.block, label %vector.body

middle.block:                                     ; preds = %vector.body
  %5 = tail call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %3)
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br i1 %cmp.n, label %for.cond.cleanup, label %vec.epilog.iter.check

vec.epilog.iter.check:                            ; preds = %middle.block
  %min.epilog.iters.check = icmp eq i64 %n.mod.vf, 0
  br i1 %min.epilog.iters.check, label %for.body.preheader, label %vec.epilog.ph

vec.epilog.ph:                                    ; preds = %vector.main.loop.iter.check, %vec.epilog.iter.check
  %vec.epilog.resume.val = phi i64 [ %n.vec, %vec.epilog.iter.check ], [ 0, %vector.main.loop.iter.check ]
  %bc.merge.rdx = phi i64 [ %5, %vec.epilog.iter.check ], [ 0, %vector.main.loop.iter.check ]
  %n.vec11 = and i64 %wide.trip.count, 4294967292
  %6 = insertelement <4 x i64> <i64 poison, i64 0, i64 0, i64 0>, i64 %bc.merge.rdx, i64 0
  %broadcast.splatinsert12 = insertelement <4 x i64> poison, i64 %conv1, i64 0
  %broadcast.splat13 = shufflevector <4 x i64> %broadcast.splatinsert12, <4 x i64> poison, <4 x i32> zeroinitializer
  br label %vec.epilog.vector.body

vec.epilog.vector.body:                           ; preds = %vec.epilog.vector.body, %vec.epilog.ph
  %index14 = phi i64 [ %vec.epilog.resume.val, %vec.epilog.ph ], [ %index.next17, %vec.epilog.vector.body ]
  %vec.phi15 = phi <4 x i64> [ %6, %vec.epilog.ph ], [ %10, %vec.epilog.vector.body ]
  %7 = getelementptr inbounds nuw i8, ptr %A, i64 %index14
  %wide.load16 = load <4 x i8>, ptr %7, align 1
  %8 = zext <4 x i8> %wide.load16 to <4 x i64>
  %9 = mul nsw <4 x i64> %broadcast.splat13, %8
  %10 = add <4 x i64> %9, %vec.phi15
  %index.next17 = add nuw i64 %index14, 4
  %11 = icmp eq i64 %index.next17, %n.vec11
  br i1 %11, label %vec.epilog.middle.block, label %vec.epilog.vector.body

vec.epilog.middle.block:                          ; preds = %vec.epilog.vector.body
  %12 = tail call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> %10)
  %cmp.n18 = icmp eq i64 %n.vec11, %wide.trip.count
  br i1 %cmp.n18, label %for.cond.cleanup, label %for.body.preheader

for.body.preheader:                               ; preds = %iter.check, %vec.epilog.iter.check, %vec.epilog.middle.block
  %indvars.iv.ph = phi i64 [ 0, %iter.check ], [ %n.vec, %vec.epilog.iter.check ], [ %n.vec11, %vec.epilog.middle.block ]
  %s.06.ph = phi i64 [ 0, %iter.check ], [ %5, %vec.epilog.iter.check ], [ %12, %vec.epilog.middle.block ]
  br label %for.body

for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %vec.epilog.middle.block, %entry
  %s.0.lcssa = phi i64 [ 0, %entry ], [ %5, %middle.block ], [ %12, %vec.epilog.middle.block ], [ %add, %for.body ]
  ret i64 %s.0.lcssa

for.body:                                         ; preds = %for.body.preheader, %for.body
  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader ]
  %s.06 = phi i64 [ %add, %for.body ], [ %s.06.ph, %for.body.preheader ]
  %arrayidx = getelementptr inbounds nuw i8, ptr %A, i64 %indvars.iv
  %13 = load i8, ptr %arrayidx, align 1
  %conv = zext i8 %13 to i64
  %mul = mul nsw i64 %conv, %conv1
  %add = add nsw i64 %mul, %s.06
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
}

define void @sink_v2z64_1(ptr %p, ptr %d, i64 %n, <2 x i32> %a) {
; CHECK-SD-LABEL: sink_v2z64_1:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:  .LBB7_1: // %loop
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr d1, [x0]
; CHECK-SD-NEXT:    subs x2, x2, #8
; CHECK-SD-NEXT:    add x8, x8, #8
; CHECK-SD-NEXT:    umull v1.2d, v1.2s, v0.s[1]
; CHECK-SD-NEXT:    shrn v1.2s, v1.2d, #15
; CHECK-SD-NEXT:    str d1, [x0], #32
; CHECK-SD-NEXT:    b.ne .LBB7_1
; CHECK-SD-NEXT:  // %bb.2: // %exit
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sink_v2z64_1:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    dup v0.2d, v0.d[1]
; CHECK-GI-NEXT:    xtn v0.2s, v0.2d
; CHECK-GI-NEXT:  .LBB7_1: // %loop
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr d1, [x0]
; CHECK-GI-NEXT:    subs x2, x2, #8
; CHECK-GI-NEXT:    add x8, x8, #8
; CHECK-GI-NEXT:    umull v1.2d, v1.2s, v0.2s
; CHECK-GI-NEXT:    shrn v1.2s, v1.2d, #15
; CHECK-GI-NEXT:    str d1, [x0], #32
; CHECK-GI-NEXT:    b.ne .LBB7_1
; CHECK-GI-NEXT:  // %bb.2: // %exit
; CHECK-GI-NEXT:    ret
entry:
  %ext = zext <2 x i32> %a to <2 x i64>
  %broadcast.splat = shufflevector <2 x i64> %ext, <2 x i64> poison, <2 x i32> <i32 1, i32 1>
  br label %loop

loop:
  %index = phi i64 [ 0, %entry ], [ %index.next, %loop ]
  %g = getelementptr inbounds i32, ptr %p, i64 %index
  %gb = bitcast ptr %g to ptr
  %l = load <2 x i32>, ptr %gb, align 4
  %e = zext <2 x i32> %l to <2 x i64>
  %m = mul <2 x i64> %e, %broadcast.splat
  %s = ashr <2 x i64> %m, <i64 15, i64 15>
  %t = trunc <2 x i64> %s to <2 x i32>
  %h = getelementptr inbounds i32, ptr %d, i64 %index
  %hb = bitcast ptr %g to ptr
  store <2 x i32> %t, ptr %hb, align 4
  %index.next = add nuw i64 %index, 8
  %c = icmp eq i64 %index.next, %n
  br i1 %c, label %exit, label %loop

exit:
  ret void
}

define void @sink_v4i64_1(ptr %p, ptr %d, i64 %n, <2 x i32> %a) {
; CHECK-SD-LABEL: sink_v4i64_1:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:  .LBB8_1: // %loop
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr q1, [x0]
; CHECK-SD-NEXT:    subs x2, x2, #8
; CHECK-SD-NEXT:    add x8, x8, #8
; CHECK-SD-NEXT:    smull v2.2d, v1.2s, v0.s[1]
; CHECK-SD-NEXT:    smull2 v1.2d, v1.4s, v0.s[1]
; CHECK-SD-NEXT:    shrn v2.2s, v2.2d, #15
; CHECK-SD-NEXT:    shrn2 v2.4s, v1.2d, #15
; CHECK-SD-NEXT:    str q2, [x0], #32
; CHECK-SD-NEXT:    b.ne .LBB8_1
; CHECK-SD-NEXT:  // %bb.2: // %exit
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sink_v4i64_1:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    dup v0.2d, v0.d[1]
; CHECK-GI-NEXT:    xtn v0.2s, v0.2d
; CHECK-GI-NEXT:  .LBB8_1: // %loop
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr q1, [x0]
; CHECK-GI-NEXT:    subs x2, x2, #8
; CHECK-GI-NEXT:    add x8, x8, #8
; CHECK-GI-NEXT:    mov d2, v1.d[1]
; CHECK-GI-NEXT:    smull v1.2d, v1.2s, v0.2s
; CHECK-GI-NEXT:    smull v2.2d, v2.2s, v0.2s
; CHECK-GI-NEXT:    shrn v1.2s, v1.2d, #15
; CHECK-GI-NEXT:    shrn2 v1.4s, v2.2d, #15
; CHECK-GI-NEXT:    str q1, [x0], #32
; CHECK-GI-NEXT:    b.ne .LBB8_1
; CHECK-GI-NEXT:  // %bb.2: // %exit
; CHECK-GI-NEXT:    ret
entry:
  %ext = sext <2 x i32> %a to <2 x i64>
  %broadcast.splat = shufflevector <2 x i64> %ext, <2 x i64> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  br label %loop

loop:
  %index = phi i64 [ 0, %entry ], [ %index.next, %loop ]
  %g = getelementptr inbounds i32, ptr %p, i64 %index
  %gb = bitcast ptr %g to ptr
  %l = load <4 x i32>, ptr %gb, align 4
  %e = sext <4 x i32> %l to <4 x i64>
  %m = mul <4 x i64> %e, %broadcast.splat
  %s = ashr <4 x i64> %m, <i64 15, i64 15, i64 15, i64 15>
  %t = trunc <4 x i64> %s to <4 x i32>
  %h = getelementptr inbounds i32, ptr %d, i64 %index
  %hb = bitcast ptr %g to ptr
  store <4 x i32> %t, ptr %hb, align 4
  %index.next = add nuw i64 %index, 8
  %c = icmp eq i64 %index.next, %n
  br i1 %c, label %exit, label %loop

exit:
  ret void
}

define void @sink_v8z16_0(ptr %p, ptr %d, i64 %n, <16 x i8> %a) {
; CHECK-SD-LABEL: sink_v8z16_0:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    dup v0.8b, v0.b[0]
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:  .LBB9_1: // %loop
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr d1, [x0]
; CHECK-SD-NEXT:    subs x2, x2, #8
; CHECK-SD-NEXT:    add x8, x8, #8
; CHECK-SD-NEXT:    umull v1.8h, v1.8b, v0.8b
; CHECK-SD-NEXT:    cmlt v1.8h, v1.8h, #0
; CHECK-SD-NEXT:    xtn v1.8b, v1.8h
; CHECK-SD-NEXT:    str d1, [x0], #32
; CHECK-SD-NEXT:    b.ne .LBB9_1
; CHECK-SD-NEXT:  // %bb.2: // %exit
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sink_v8z16_0:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    dup v0.8h, v0.h[0]
; CHECK-GI-NEXT:    xtn v0.8b, v0.8h
; CHECK-GI-NEXT:  .LBB9_1: // %loop
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr d1, [x0]
; CHECK-GI-NEXT:    subs x2, x2, #8
; CHECK-GI-NEXT:    add x8, x8, #8
; CHECK-GI-NEXT:    umull v1.8h, v1.8b, v0.8b
; CHECK-GI-NEXT:    cmlt v1.8h, v1.8h, #0
; CHECK-GI-NEXT:    xtn v1.8b, v1.8h
; CHECK-GI-NEXT:    str d1, [x0], #32
; CHECK-GI-NEXT:    b.ne .LBB9_1
; CHECK-GI-NEXT:  // %bb.2: // %exit
; CHECK-GI-NEXT:    ret
entry:
  %ext = zext <16 x i8> %a to <16 x i16>
  %broadcast.splat = shufflevector <16 x i16> %ext, <16 x i16> poison, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  br label %loop

loop:
  %index = phi i64 [ 0, %entry ], [ %index.next, %loop ]
  %g = getelementptr inbounds i32, ptr %p, i64 %index
  %gb = bitcast ptr %g to ptr
  %l = load <8 x i8>, ptr %gb, align 4
  %e = zext <8 x i8> %l to <8 x i16>
  %m = mul <8 x i16> %e, %broadcast.splat
  %s = ashr <8 x i16> %m, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
  %t = trunc <8 x i16> %s to <8 x i8>
  %h = getelementptr inbounds i32, ptr %d, i64 %index
  %hb = bitcast ptr %g to ptr
  store <8 x i8> %t, ptr %hb, align 4
  %index.next = add nuw i64 %index, 8
  %c = icmp eq i64 %index.next, %n
  br i1 %c, label %exit, label %loop

exit:
  ret void
}

define void @sink_v16s16_8(ptr %p, ptr %d, i64 %n, <16 x i8> %a) {
; CHECK-SD-LABEL: sink_v16s16_8:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    dup v0.16b, v0.b[10]
; CHECK-SD-NEXT:    mov x8, xzr
; CHECK-SD-NEXT:  .LBB10_1: // %loop
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    ldr q1, [x0]
; CHECK-SD-NEXT:    subs x2, x2, #8
; CHECK-SD-NEXT:    add x8, x8, #8
; CHECK-SD-NEXT:    smull v2.8h, v1.8b, v0.8b
; CHECK-SD-NEXT:    smull2 v1.8h, v1.16b, v0.16b
; CHECK-SD-NEXT:    cmlt v1.8h, v1.8h, #0
; CHECK-SD-NEXT:    cmlt v2.8h, v2.8h, #0
; CHECK-SD-NEXT:    uzp1 v1.16b, v2.16b, v1.16b
; CHECK-SD-NEXT:    str q1, [x0], #32
; CHECK-SD-NEXT:    b.ne .LBB10_1
; CHECK-SD-NEXT:  // %bb.2: // %exit
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sink_v16s16_8:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    mov x8, xzr
; CHECK-GI-NEXT:    dup v0.8h, v0.h[2]
; CHECK-GI-NEXT:    xtn v0.8b, v0.8h
; CHECK-GI-NEXT:  .LBB10_1: // %loop
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    ldr q1, [x0]
; CHECK-GI-NEXT:    subs x2, x2, #8
; CHECK-GI-NEXT:    add x8, x8, #8
; CHECK-GI-NEXT:    mov d2, v1.d[1]
; CHECK-GI-NEXT:    smull v1.8h, v1.8b, v0.8b
; CHECK-GI-NEXT:    smull v2.8h, v2.8b, v0.8b
; CHECK-GI-NEXT:    cmlt v1.8h, v1.8h, #0
; CHECK-GI-NEXT:    cmlt v2.8h, v2.8h, #0
; CHECK-GI-NEXT:    uzp1 v1.16b, v1.16b, v2.16b
; CHECK-GI-NEXT:    str q1, [x0], #32
; CHECK-GI-NEXT:    b.ne .LBB10_1
; CHECK-GI-NEXT:  // %bb.2: // %exit
; CHECK-GI-NEXT:    ret
entry:
  %ext = sext <16 x i8> %a to <16 x i16>
  %broadcast.splat = shufflevector <16 x i16> %ext, <16 x i16> poison, <16 x i32> <i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10>
  br label %loop

loop:
  %index = phi i64 [ 0, %entry ], [ %index.next, %loop ]
  %g = getelementptr inbounds i32, ptr %p, i64 %index
  %gb = bitcast ptr %g to ptr
  %l = load <16 x i8>, ptr %gb, align 4
  %e = sext <16 x i8> %l to <16 x i16>
  %m = mul <16 x i16> %e, %broadcast.splat
  %s = ashr <16 x i16> %m, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
  %t = trunc <16 x i16> %s to <16 x i8>
  %h = getelementptr inbounds i32, ptr %d, i64 %index
  %hb = bitcast ptr %g to ptr
  store <16 x i8> %t, ptr %hb, align 4
  %index.next = add nuw i64 %index, 8
  %c = icmp eq i64 %index.next, %n
  br i1 %c, label %exit, label %loop

exit:
  ret void
}

define void @matrix_mul_unsigned_and(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i32 %val) {
; CHECK-SD-LABEL: matrix_mul_unsigned_and:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    dup v0.4h, w3
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff8
; CHECK-SD-NEXT:  .LBB11_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #8
; CHECK-SD-NEXT:    ldp d1, d2, [x9]
; CHECK-SD-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-SD-NEXT:    add w0, w0, #8
; CHECK-SD-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q2, [x9]
; CHECK-SD-NEXT:    b.ne .LBB11_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_unsigned_and:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w8, w3, #0xffff
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    mov w8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff8
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:  .LBB11_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #8
; CHECK-GI-NEXT:    ldp d1, d2, [x9]
; CHECK-GI-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-GI-NEXT:    add w0, w0, #8
; CHECK-GI-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    stp q1, q2, [x9]
; CHECK-GI-NEXT:    b.ne .LBB11_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = and i32 %val, 65535
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 8
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967288
  %broadcast.splatinsert = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> undef, <4 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <4 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <4 x i16>, ptr %9, align 2
  %10 = zext <4 x i16> %wide.load to <4 x i32>
  %11 = zext <4 x i16> %wide.load30 to <4 x i32>
  %12 = mul nuw nsw <4 x i32> %broadcast.splat, %10
  %13 = mul nuw nsw <4 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <4 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 4
  %17 = bitcast ptr %16 to ptr
  store <4 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 8
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}

define void @matrix_mul_unsigned_and_double(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i32 %val) {
; CHECK-SD-LABEL: matrix_mul_unsigned_and_double:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    dup v0.8h, w3
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff0
; CHECK-SD-NEXT:  .LBB12_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #16
; CHECK-SD-NEXT:    ldr q1, [x9]
; CHECK-SD-NEXT:    ldur q2, [x9, #8]
; CHECK-SD-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-SD-NEXT:    add w0, w0, #16
; CHECK-SD-NEXT:    umull2 v3.4s, v0.8h, v1.8h
; CHECK-SD-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-SD-NEXT:    umull2 v4.4s, v0.8h, v2.8h
; CHECK-SD-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-SD-NEXT:    stp q1, q3, [x9]
; CHECK-SD-NEXT:    stp q2, q4, [x9, #32]
; CHECK-SD-NEXT:    b.ne .LBB12_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_unsigned_and_double:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w8, w3, #0xffff
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    mov w8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff0
; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
; CHECK-GI-NEXT:  .LBB12_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #16
; CHECK-GI-NEXT:    ldr q1, [x9]
; CHECK-GI-NEXT:    ldur q2, [x9, #8]
; CHECK-GI-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-GI-NEXT:    add w0, w0, #16
; CHECK-GI-NEXT:    mov d3, v1.d[1]
; CHECK-GI-NEXT:    mov d4, v2.d[1]
; CHECK-GI-NEXT:    umull v1.4s, v0.4h, v1.4h
; CHECK-GI-NEXT:    umull v2.4s, v0.4h, v2.4h
; CHECK-GI-NEXT:    umull v3.4s, v0.4h, v3.4h
; CHECK-GI-NEXT:    umull v4.4s, v0.4h, v4.4h
; CHECK-GI-NEXT:    stp q1, q3, [x9]
; CHECK-GI-NEXT:    stp q2, q4, [x9, #32]!
; CHECK-GI-NEXT:    b.ne .LBB12_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = and i32 %val, 65535
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 16
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <8 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <8 x i32> %broadcast.splatinsert, <8 x i32> undef, <8 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <8 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <8 x i32> %broadcast.splatinsert31, <8 x i32> undef, <8 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <8 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <8 x i16>, ptr %9, align 2
  %10 = zext <8 x i16> %wide.load to <8 x i32>
  %11 = zext <8 x i16> %wide.load30 to <8 x i32>
  %12 = mul nuw nsw <8 x i32> %broadcast.splat, %10
  %13 = mul nuw nsw <8 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <8 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 8
  %17 = bitcast ptr %16 to ptr
  store <8 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 16
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}

define void @matrix_mul_signed_and(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i32 %val) {
; CHECK-SD-LABEL: matrix_mul_signed_and:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    and w8, w3, #0xffff
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    fmov s0, w8
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff8
; CHECK-SD-NEXT:  .LBB13_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #8
; CHECK-SD-NEXT:    ldp d1, d2, [x9]
; CHECK-SD-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-SD-NEXT:    add w0, w0, #8
; CHECK-SD-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    mul v1.4s, v1.4s, v0.s[0]
; CHECK-SD-NEXT:    mul v2.4s, v2.4s, v0.s[0]
; CHECK-SD-NEXT:    stp q1, q2, [x9]
; CHECK-SD-NEXT:    b.ne .LBB13_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_signed_and:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w8, w3, #0xffff
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    mov w8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff8
; CHECK-GI-NEXT:  .LBB13_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #8
; CHECK-GI-NEXT:    ldp d1, d2, [x9]
; CHECK-GI-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-GI-NEXT:    add w0, w0, #8
; CHECK-GI-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll v2.4s, v2.4h, #0
; CHECK-GI-NEXT:    mul v1.4s, v0.4s, v1.4s
; CHECK-GI-NEXT:    mul v2.4s, v0.4s, v2.4s
; CHECK-GI-NEXT:    stp q1, q2, [x9]
; CHECK-GI-NEXT:    b.ne .LBB13_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = and i32 %val, 65535
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 8
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967288
  %broadcast.splatinsert = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> undef, <4 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <4 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <4 x i16>, ptr %9, align 2
  %10 = sext <4 x i16> %wide.load to <4 x i32>
  %11 = sext <4 x i16> %wide.load30 to <4 x i32>
  %12 = mul nuw nsw <4 x i32> %broadcast.splat, %10
  %13 = mul nuw nsw <4 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <4 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 4
  %17 = bitcast ptr %16 to ptr
  store <4 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 8
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}

define void @matrix_mul_signed_and_double(i32 %N, ptr nocapture %C, ptr nocapture readonly %A, i32 %val) {
; CHECK-SD-LABEL: matrix_mul_signed_and_double:
; CHECK-SD:       // %bb.0: // %vector.header
; CHECK-SD-NEXT:    and w8, w3, #0xffff
; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
; CHECK-SD-NEXT:    fmov s0, w8
; CHECK-SD-NEXT:    and x8, x0, #0xfffffff0
; CHECK-SD-NEXT:  .LBB14_1: // %vector.body
; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-SD-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-SD-NEXT:    subs x8, x8, #16
; CHECK-SD-NEXT:    ldr q1, [x9]
; CHECK-SD-NEXT:    ldur q2, [x9, #8]
; CHECK-SD-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-SD-NEXT:    add w0, w0, #16
; CHECK-SD-NEXT:    sshll2 v3.4s, v1.8h, #0
; CHECK-SD-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll2 v4.4s, v2.8h, #0
; CHECK-SD-NEXT:    sshll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    mul v3.4s, v3.4s, v0.s[0]
; CHECK-SD-NEXT:    mul v1.4s, v1.4s, v0.s[0]
; CHECK-SD-NEXT:    mul v4.4s, v4.4s, v0.s[0]
; CHECK-SD-NEXT:    mul v2.4s, v2.4s, v0.s[0]
; CHECK-SD-NEXT:    stp q1, q3, [x9]
; CHECK-SD-NEXT:    stp q2, q4, [x9, #32]
; CHECK-SD-NEXT:    b.ne .LBB14_1
; CHECK-SD-NEXT:  // %bb.2: // %for.end12
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: matrix_mul_signed_and_double:
; CHECK-GI:       // %bb.0: // %vector.header
; CHECK-GI-NEXT:    and w8, w3, #0xffff
; CHECK-GI-NEXT:    dup v0.4s, w8
; CHECK-GI-NEXT:    mov w8, w0
; CHECK-GI-NEXT:    and x8, x8, #0xfffffff0
; CHECK-GI-NEXT:  .LBB14_1: // %vector.body
; CHECK-GI-NEXT:    // =>This Inner Loop Header: Depth=1
; CHECK-GI-NEXT:    add x9, x2, w0, uxtw #1
; CHECK-GI-NEXT:    subs x8, x8, #16
; CHECK-GI-NEXT:    ldr q1, [x9]
; CHECK-GI-NEXT:    ldur q2, [x9, #8]
; CHECK-GI-NEXT:    add x9, x1, w0, uxtw #2
; CHECK-GI-NEXT:    add w0, w0, #16
; CHECK-GI-NEXT:    sshll v3.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v4.4s, v2.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    mul v3.4s, v0.4s, v3.4s
; CHECK-GI-NEXT:    mul v1.4s, v0.4s, v1.4s
; CHECK-GI-NEXT:    mul v4.4s, v0.4s, v4.4s
; CHECK-GI-NEXT:    mul v2.4s, v0.4s, v2.4s
; CHECK-GI-NEXT:    stp q3, q1, [x9]
; CHECK-GI-NEXT:    stp q4, q2, [x9, #32]!
; CHECK-GI-NEXT:    b.ne .LBB14_1
; CHECK-GI-NEXT:  // %bb.2: // %for.end12
; CHECK-GI-NEXT:    ret
vector.header:
  %conv4 = and i32 %val, 65535
  %wide.trip.count = zext i32 %N to i64
  %0 = add nsw i64 %wide.trip.count, -1
  %min.iters.check = icmp ult i32 %N, 16
  %1 = trunc i64 %0 to i32
  %2 = icmp ugt i64 %0, 4294967295
  %n.vec = and i64 %wide.trip.count, 4294967280
  %broadcast.splatinsert = insertelement <8 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat = shufflevector <8 x i32> %broadcast.splatinsert, <8 x i32> undef, <8 x i32> zeroinitializer
  %broadcast.splatinsert31 = insertelement <8 x i32> undef, i32 %conv4, i32 0
  %broadcast.splat32 = shufflevector <8 x i32> %broadcast.splatinsert31, <8 x i32> undef, <8 x i32> zeroinitializer
  %cmp.n = icmp eq i64 %n.vec, %wide.trip.count
  br label %vector.body

vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %3 = trunc i64 %index to i32
  %4 = add i32 %N, %3
  %5 = zext i32 %4 to i64
  %6 = getelementptr inbounds i16, ptr %A, i64 %5
  %7 = bitcast ptr %6 to ptr
  %wide.load = load <8 x i16>, ptr %7, align 2
  %8 = getelementptr inbounds i16, ptr %6, i64 4
  %9 = bitcast ptr %8 to ptr
  %wide.load30 = load <8 x i16>, ptr %9, align 2
  %10 = sext <8 x i16> %wide.load to <8 x i32>
  %11 = sext <8 x i16> %wide.load30 to <8 x i32>
  %12 = mul nuw nsw <8 x i32> %broadcast.splat, %10
  %13 = mul nuw nsw <8 x i32> %broadcast.splat32, %11
  %14 = getelementptr inbounds i32, ptr %C, i64 %5
  %15 = bitcast ptr %14 to ptr
  store <8 x i32> %12, ptr %15, align 4
  %16 = getelementptr inbounds i32, ptr %14, i64 8
  %17 = bitcast ptr %16 to ptr
  store <8 x i32> %13, ptr %17, align 4
  %index.next = add i64 %index, 16
  %18 = icmp eq i64 %index.next, %n.vec
  br i1 %18, label %for.end12, label %vector.body

for.end12:                                        ; preds = %vector.body
  ret void
}

define noundef <8 x i16> @cmplx_mul_combined_re_im(<8 x i16> noundef %a, i64 %scale.coerce) {
; CHECK-SD-LABEL: cmplx_mul_combined_re_im:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    lsr x9, x0, #16
; CHECK-SD-NEXT:    adrp x8, .LCPI15_0
; CHECK-SD-NEXT:    dup v4.8h, w0
; CHECK-SD-NEXT:    ldr q3, [x8, :lo12:.LCPI15_0]
; CHECK-SD-NEXT:    dup v2.8h, w9
; CHECK-SD-NEXT:    sqneg v1.8h, v2.8h
; CHECK-SD-NEXT:    tbl v1.16b, { v1.16b, v2.16b }, v3.16b
; CHECK-SD-NEXT:    rev32 v2.8h, v0.8h
; CHECK-SD-NEXT:    sqdmull v3.4s, v0.4h, v4.4h
; CHECK-SD-NEXT:    sqdmull2 v0.4s, v0.8h, v4.8h
; CHECK-SD-NEXT:    sqdmlal v3.4s, v2.4h, v1.4h
; CHECK-SD-NEXT:    sqdmlal2 v0.4s, v2.8h, v1.8h
; CHECK-SD-NEXT:    uzp2 v0.8h, v3.8h, v0.8h
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: cmplx_mul_combined_re_im:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    lsr w9, w0, #16
; CHECK-GI-NEXT:    adrp x8, .LCPI15_0
; CHECK-GI-NEXT:    rev32 v4.8h, v0.8h
; CHECK-GI-NEXT:    dup v1.8h, w9
; CHECK-GI-NEXT:    fmov s3, w9
; CHECK-GI-NEXT:    sqneg v2.8h, v1.8h
; CHECK-GI-NEXT:    ldr q1, [x8, :lo12:.LCPI15_0]
; CHECK-GI-NEXT:    tbl v1.16b, { v2.16b, v3.16b }, v1.16b
; CHECK-GI-NEXT:    mov d2, v0.d[1]
; CHECK-GI-NEXT:    dup v3.8h, w0
; CHECK-GI-NEXT:    sqdmull v2.4s, v2.4h, v3.4h
; CHECK-GI-NEXT:    sqdmull v5.4s, v4.4h, v1.4h
; CHECK-GI-NEXT:    sqdmlal v5.4s, v0.4h, v3.4h
; CHECK-GI-NEXT:    sqdmlal2 v2.4s, v4.8h, v1.8h
; CHECK-GI-NEXT:    uzp2 v0.8h, v5.8h, v2.8h
; CHECK-GI-NEXT:    ret
entry:
  %scale.sroa.0.0.extract.trunc = trunc i64 %scale.coerce to i16
  %scale.sroa.2.0.extract.shift23 = lshr i64 %scale.coerce, 16
  %scale.sroa.2.0.extract.trunc = trunc i64 %scale.sroa.2.0.extract.shift23 to i16
  %shuffle.i = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
  %vecinit.i24 = insertelement <8 x i16> poison, i16 %scale.sroa.0.0.extract.trunc, i64 0
  %vecinit.i = insertelement <8 x i16> poison, i16 %scale.sroa.2.0.extract.trunc, i64 0
  %vecinit7.i = shufflevector <8 x i16> %vecinit.i, <8 x i16> poison, <8 x i32> zeroinitializer
  %vqnegq_v1.i = tail call noundef <8 x i16> @llvm.aarch64.neon.sqneg.v8i16(<8 x i16> %vecinit7.i)
  %vbsl5.i = shufflevector <8 x i16> %vqnegq_v1.i, <8 x i16> %vecinit.i, <8 x i32> <i32 0, i32 8, i32 2, i32 8, i32 4, i32 8, i32 6, i32 8>
  %shuffle.i40 = shufflevector <8 x i16> %a, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %shuffle.i39 = shufflevector <8 x i16> %vecinit.i24, <8 x i16> poison, <4 x i32> zeroinitializer
  %vqdmull_v2.i36 = tail call noundef <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> %shuffle.i40, <4 x i16> %shuffle.i39)
  %shuffle.i44 = shufflevector <8 x i16> %a, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %vqdmull_v2.i = tail call noundef <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> %shuffle.i44, <4 x i16> %shuffle.i39)
  %shuffle.i38 = shufflevector <8 x i16> %shuffle.i, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %shuffle.i37 = shufflevector <8 x i16> %vbsl5.i, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %vqdmlal2.i45 = tail call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> %shuffle.i38, <4 x i16> %shuffle.i37)
  %vqdmlal_v3.i46 = tail call noundef <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> %vqdmull_v2.i36, <4 x i32> %vqdmlal2.i45)
  %shuffle.i42 = shufflevector <8 x i16> %shuffle.i, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %shuffle.i41 = shufflevector <8 x i16> %vbsl5.i, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %vqdmlal2.i = tail call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> %shuffle.i42, <4 x i16> %shuffle.i41)
  %vqdmlal_v3.i = tail call noundef <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> %vqdmull_v2.i, <4 x i32> %vqdmlal2.i)
  %0 = bitcast <4 x i32> %vqdmlal_v3.i46 to <8 x i16>
  %1 = bitcast <4 x i32> %vqdmlal_v3.i to <8 x i16>
  %shuffle.i35 = shufflevector <8 x i16> %0, <8 x i16> %1, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
  ret <8 x i16> %shuffle.i35
}


;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK: {{.*}}