; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple aarch64 -mattr=+sve2 -o - %s | FileCheck %s

define void @test(ptr nocapture noundef readonly %kernel, i32 noundef %kw, float noundef nofpclass(nan inf) %kernel_factor, ptr %call5.i.i.i119) vscale_range(1, 16) {
; CHECK-LABEL: test:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    cmp w1, #1
; CHECK-NEXT:    b.lt .LBB0_6
; CHECK-NEXT:  // %bb.1: // %for.body.lr.ph
; CHECK-NEXT:    rdvl x8, #-2
; CHECK-NEXT:    mov w9, #608 // =0x260
; CHECK-NEXT:    ands x11, x8, x9
; CHECK-NEXT:    b.eq .LBB0_6
; CHECK-NEXT:  // %bb.2: // %for.body.us.preheader
; CHECK-NEXT:    ptrue p0.h
; CHECK-NEXT:    add x11, x2, x11, lsl #1
; CHECK-NEXT:    mov w8, wzr
; CHECK-NEXT:    ptrue p1.b
; CHECK-NEXT:    mov x9, xzr
; CHECK-NEXT:    mov w10, wzr
; CHECK-NEXT:    mov x12, #4 // =0x4
; CHECK-NEXT:    mov x13, #8 // =0x8
; CHECK-NEXT:  .LBB0_3: // %for.body.us
; CHECK-NEXT:    // =>This Loop Header: Depth=1
; CHECK-NEXT:    // Child Loop BB0_4 Depth 2
; CHECK-NEXT:    add x14, x0, x9, lsl #2
; CHECK-NEXT:    sbfiz x15, x8, #1, #32
; CHECK-NEXT:    mov x16, x2
; CHECK-NEXT:    ldp s0, s1, [x14]
; CHECK-NEXT:    add x15, x15, #8
; CHECK-NEXT:    ldp s2, s3, [x14, #8]
; CHECK-NEXT:    ubfiz x14, x8, #1, #32
; CHECK-NEXT:    fcvt h0, s0
; CHECK-NEXT:    fcvt h1, s1
; CHECK-NEXT:    fcvt h2, s2
; CHECK-NEXT:    fcvt h3, s3
; CHECK-NEXT:    mov z0.h, h0
; CHECK-NEXT:    mov z1.h, h1
; CHECK-NEXT:    mov z2.h, h2
; CHECK-NEXT:    mov z3.h, h3
; CHECK-NEXT:  .LBB0_4: // %for.cond.i.preheader.us
; CHECK-NEXT:    // Parent Loop BB0_3 Depth=1
; CHECK-NEXT:    // => This Inner Loop Header: Depth=2
; CHECK-NEXT:    ld1b { z4.b }, p1/z, [x16, x14]
; CHECK-NEXT:    ldr z5, [x16]
; CHECK-NEXT:    add x17, x16, x15
; CHECK-NEXT:    add x18, x16, x14
; CHECK-NEXT:    add x3, x17, #8
; CHECK-NEXT:    add x4, x17, #16
; CHECK-NEXT:    fmad z4.h, p0/m, z0.h, z5.h
; CHECK-NEXT:    ld1b { z5.b }, p1/z, [x16, x15]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z1.h
; CHECK-NEXT:    ld1h { z5.h }, p0/z, [x17, x12, lsl #1]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z2.h
; CHECK-NEXT:    ld1h { z5.h }, p0/z, [x17, x13, lsl #1]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z3.h
; CHECK-NEXT:    ldr z5, [x16, #1, mul vl]
; CHECK-NEXT:    str z4, [x16]
; CHECK-NEXT:    ldr z4, [x18, #1, mul vl]
; CHECK-NEXT:    fmad z4.h, p0/m, z0.h, z5.h
; CHECK-NEXT:    ldr z5, [x17, #1, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z1.h
; CHECK-NEXT:    ldr z5, [x3, #1, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z2.h
; CHECK-NEXT:    ldr z5, [x4, #1, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z3.h
; CHECK-NEXT:    ldr z5, [x16, #2, mul vl]
; CHECK-NEXT:    str z4, [x16, #1, mul vl]
; CHECK-NEXT:    ldr z4, [x18, #2, mul vl]
; CHECK-NEXT:    fmad z4.h, p0/m, z0.h, z5.h
; CHECK-NEXT:    ldr z5, [x17, #2, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z1.h
; CHECK-NEXT:    ldr z5, [x3, #2, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z2.h
; CHECK-NEXT:    ldr z5, [x4, #2, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z3.h
; CHECK-NEXT:    ldr z5, [x16, #3, mul vl]
; CHECK-NEXT:    str z4, [x16, #2, mul vl]
; CHECK-NEXT:    ldr z4, [x18, #3, mul vl]
; CHECK-NEXT:    fmad z4.h, p0/m, z0.h, z5.h
; CHECK-NEXT:    ldr z5, [x17, #3, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z1.h
; CHECK-NEXT:    ldr z5, [x3, #3, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z2.h
; CHECK-NEXT:    ldr z5, [x4, #3, mul vl]
; CHECK-NEXT:    fmla z4.h, p0/m, z5.h, z3.h
; CHECK-NEXT:    str z4, [x16, #3, mul vl]
; CHECK-NEXT:    incb x16, all, mul #4
; CHECK-NEXT:    cmp x16, x11
; CHECK-NEXT:    b.lo .LBB0_4
; CHECK-NEXT:  // %bb.5: // %while.cond.i..exit_crit_edge.us
; CHECK-NEXT:    // in Loop: Header=BB0_3 Depth=1
; CHECK-NEXT:    add w10, w10, #1
; CHECK-NEXT:    add x9, x9, #4
; CHECK-NEXT:    add w8, w8, #16
; CHECK-NEXT:    cmp w10, w1
; CHECK-NEXT:    b.ne .LBB0_3
; CHECK-NEXT:  .LBB0_6: // %exit78
; CHECK-NEXT:    ret
entry:
  ;%call5.i.i.i119 = tail call noalias noundef nonnull dereferenceable(1248) ptr @_Znwm(i64 noundef 1248) #7
  %cmp139 = icmp sgt i32 %kw, 0
  ;tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 2 dereferenceable(1248) %call5.i.i.i119, i8 0, i64 1248, i1 false)
  br i1 %cmp139, label %for.body.lr.ph, label %exit78

for.body.lr.ph:                                   ; preds = %entry
  %0 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
  %vscale = tail call i64 @llvm.vscale.i64()
  %mul5.i = shl nuw nsw i64 %vscale, 5
  %sub.not.i = sub nsw i64 0, %mul5.i
  %sub6.i = and i64 %sub.not.i, 608
  %add.ptr.i = getelementptr inbounds half, ptr %call5.i.i.i119, i64 %sub6.i
  %cmp.i133.not = icmp eq i64 %sub6.i, 0
  %vs2 = shl nuw nsw i64 %vscale, 4
  br i1 %cmp.i133.not, label %exit78, label %for.body.us.preheader

for.body.us.preheader:                            ; preds = %for.body.lr.ph
  %.idx.i.us.2 = shl nuw nsw i64 %vscale, 5
  %.idx.i.us.3 = mul nuw nsw i64 %vscale, 48
  br label %for.body.us

for.body.us:                                      ; preds = %for.body.us.preheader, %while.cond.i..exit_crit_edge.us
  %indvars.iv = phi i64 [ 0, %for.body.us.preheader ], [ %indvars.iv.next, %while.cond.i..exit_crit_edge.us ]
  %i4.0140.us = phi i32 [ 0, %for.body.us.preheader ], [ %inc.us, %while.cond.i..exit_crit_edge.us ]
  %3 = trunc nuw nsw i64 %indvars.iv to i32
  %mul6.us = shl i32 %3, 2
  %idx.ext.us = zext nneg i32 %mul6.us to i64
  %add.ptr.us = getelementptr inbounds half, ptr %call5.i.i.i119, i64 %idx.ext.us
  %mul11.us = or disjoint i32 %mul6.us, 4
  %idx.ext12.us = sext i32 %mul11.us to i64
  %add.ptr13.us = getelementptr inbounds half, ptr %call5.i.i.i119, i64 %idx.ext12.us
  %mul18.us = or disjoint i32 %mul6.us, 8
  %idx.ext19.us = sext i32 %mul18.us to i64
  %add.ptr20.us = getelementptr inbounds half, ptr %call5.i.i.i119, i64 %idx.ext19.us
  %mul25.us = or disjoint i32 %mul6.us, 12
  %idx.ext26.us = sext i32 %mul25.us to i64
  %add.ptr27.us = getelementptr inbounds half, ptr %call5.i.i.i119, i64 %idx.ext26.us
  %add.ptr29.us = getelementptr inbounds float, ptr %kernel, i64 %indvars.iv
  %4 = load float, ptr %add.ptr29.us, align 4
  %5 = fptrunc float %4 to half
  %.splatinsert.i.us = insertelement <vscale x 8 x half> poison, half %5, i64 0
  %6 = shufflevector <vscale x 8 x half> %.splatinsert.i.us, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
  %arrayidx2.i.us = getelementptr inbounds i8, ptr %add.ptr29.us, i64 4
  %7 = load float, ptr %arrayidx2.i.us, align 4
  %8 = fptrunc float %7 to half
  %.splatinsert57.i.us = insertelement <vscale x 8 x half> poison, half %8, i64 0
  %9 = shufflevector <vscale x 8 x half> %.splatinsert57.i.us, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
  %arrayidx3.i.us = getelementptr inbounds i8, ptr %add.ptr29.us, i64 8
  %10 = load float, ptr %arrayidx3.i.us, align 4
  %11 = fptrunc float %10 to half
  %.splatinsert58.i.us = insertelement <vscale x 8 x half> poison, half %11, i64 0
  %12 = shufflevector <vscale x 8 x half> %.splatinsert58.i.us, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
  %arrayidx4.i.us = getelementptr inbounds i8, ptr %add.ptr29.us, i64 12
  %13 = load float, ptr %arrayidx4.i.us, align 4
  %14 = fptrunc float %13 to half
  %.splatinsert59.i.us = insertelement <vscale x 8 x half> poison, half %14, i64 0
  %15 = shufflevector <vscale x 8 x half> %.splatinsert59.i.us, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
  br label %for.cond.i.preheader.us

for.cond.i.preheader.us:                          ; preds = %for.body.us, %for.cond.i.preheader.us
  %vdst.0.i138.us = phi ptr [ %call5.i.i.i119, %for.body.us ], [ %add.ptr15.i.us, %for.cond.i.preheader.us ]
  %s1.0.i137.us = phi ptr [ %add.ptr.us, %for.body.us ], [ %add.ptr16.i.us, %for.cond.i.preheader.us ]
  %s2.0.i136.us = phi ptr [ %add.ptr13.us, %for.body.us ], [ %add.ptr17.i.us, %for.cond.i.preheader.us ]
  %s3.0.i135.us = phi ptr [ %add.ptr20.us, %for.body.us ], [ %add.ptr18.i.us, %for.cond.i.preheader.us ]
  %s4.0.i134.us = phi ptr [ %add.ptr27.us, %for.body.us ], [ %add.ptr19.i.us, %for.cond.i.preheader.us ]
  %16 = load <vscale x 8 x half>, ptr %s1.0.i137.us, align 16
  %17 = load <vscale x 8 x half>, ptr %vdst.0.i138.us, align 16
  %18 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %17, <vscale x 8 x half> %16, <vscale x 8 x half> %6)
  %19 = load <vscale x 8 x half>, ptr %s2.0.i136.us, align 16
  %20 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %18, <vscale x 8 x half> %19, <vscale x 8 x half> %9)
  %21 = load <vscale x 8 x half>, ptr %s3.0.i135.us, align 16
  %22 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %20, <vscale x 8 x half> %21, <vscale x 8 x half> %12)
  %23 = load <vscale x 8 x half>, ptr %s4.0.i134.us, align 16
  %24 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %22, <vscale x 8 x half> %23, <vscale x 8 x half> %15)
  store <vscale x 8 x half> %24, ptr %vdst.0.i138.us, align 16
  %25 = getelementptr i8, ptr %s1.0.i137.us, i64 %vs2
  %26 = load <vscale x 8 x half>, ptr %25, align 16
  %27 = getelementptr i8, ptr %vdst.0.i138.us, i64 %vs2
  %28 = load <vscale x 8 x half>, ptr %27, align 16
  %29 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %28, <vscale x 8 x half> %26, <vscale x 8 x half> %6)
  %30 = getelementptr i8, ptr %s2.0.i136.us, i64 %vs2
  %31 = load <vscale x 8 x half>, ptr %30, align 16
  %32 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %29, <vscale x 8 x half> %31, <vscale x 8 x half> %9)
  %33 = getelementptr i8, ptr %s3.0.i135.us, i64 %vs2
  %34 = load <vscale x 8 x half>, ptr %33, align 16
  %35 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %32, <vscale x 8 x half> %34, <vscale x 8 x half> %12)
  %36 = getelementptr i8, ptr %s4.0.i134.us, i64 %vs2
  %37 = load <vscale x 8 x half>, ptr %36, align 16
  %38 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %35, <vscale x 8 x half> %37, <vscale x 8 x half> %15)
  store <vscale x 8 x half> %38, ptr %27, align 16
  %39 = getelementptr i8, ptr %s1.0.i137.us, i64 %.idx.i.us.2
  %40 = load <vscale x 8 x half>, ptr %39, align 16
  %41 = getelementptr i8, ptr %vdst.0.i138.us, i64 %.idx.i.us.2
  %42 = load <vscale x 8 x half>, ptr %41, align 16
  %43 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %42, <vscale x 8 x half> %40, <vscale x 8 x half> %6)
  %44 = getelementptr i8, ptr %s2.0.i136.us, i64 %.idx.i.us.2
  %45 = load <vscale x 8 x half>, ptr %44, align 16
  %46 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %43, <vscale x 8 x half> %45, <vscale x 8 x half> %9)
  %47 = getelementptr i8, ptr %s3.0.i135.us, i64 %.idx.i.us.2
  %48 = load <vscale x 8 x half>, ptr %47, align 16
  %49 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %46, <vscale x 8 x half> %48, <vscale x 8 x half> %12)
  %50 = getelementptr i8, ptr %s4.0.i134.us, i64 %.idx.i.us.2
  %51 = load <vscale x 8 x half>, ptr %50, align 16
  %52 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %49, <vscale x 8 x half> %51, <vscale x 8 x half> %15)
  store <vscale x 8 x half> %52, ptr %41, align 16
  %53 = getelementptr i8, ptr %s1.0.i137.us, i64 %.idx.i.us.3
  %54 = load <vscale x 8 x half>, ptr %53, align 16
  %55 = getelementptr i8, ptr %vdst.0.i138.us, i64 %.idx.i.us.3
  %56 = load <vscale x 8 x half>, ptr %55, align 16
  %57 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %56, <vscale x 8 x half> %54, <vscale x 8 x half> %6)
  %58 = getelementptr i8, ptr %s2.0.i136.us, i64 %.idx.i.us.3
  %59 = load <vscale x 8 x half>, ptr %58, align 16
  %60 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %57, <vscale x 8 x half> %59, <vscale x 8 x half> %9)
  %61 = getelementptr i8, ptr %s3.0.i135.us, i64 %.idx.i.us.3
  %62 = load <vscale x 8 x half>, ptr %61, align 16
  %63 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %60, <vscale x 8 x half> %62, <vscale x 8 x half> %12)
  %64 = getelementptr i8, ptr %s4.0.i134.us, i64 %.idx.i.us.3
  %65 = load <vscale x 8 x half>, ptr %64, align 16
  %66 = tail call fast <vscale x 8 x half> @llvm.aarch64.sve.fmla.u.nxv8f16(<vscale x 8 x i1> %0, <vscale x 8 x half> %63, <vscale x 8 x half> %65, <vscale x 8 x half> %15)
  store <vscale x 8 x half> %66, ptr %55, align 16
  %add.ptr15.i.us = getelementptr inbounds half, ptr %vdst.0.i138.us, i64 %mul5.i
  %add.ptr16.i.us = getelementptr inbounds half, ptr %s1.0.i137.us, i64 %mul5.i
  %add.ptr17.i.us = getelementptr inbounds half, ptr %s2.0.i136.us, i64 %mul5.i
  %add.ptr18.i.us = getelementptr inbounds half, ptr %s3.0.i135.us, i64 %mul5.i
  %add.ptr19.i.us = getelementptr inbounds half, ptr %s4.0.i134.us, i64 %mul5.i
  %cmp.i.us = icmp ult ptr %add.ptr15.i.us, %add.ptr.i
  br i1 %cmp.i.us, label %for.cond.i.preheader.us, label %while.cond.i..exit_crit_edge.us

while.cond.i..exit_crit_edge.us: ; preds = %for.cond.i.preheader.us
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 4
  %inc.us = add nuw nsw i32 %i4.0140.us, 1
  %exitcond.not = icmp eq i32 %inc.us, %kw
  br i1 %exitcond.not, label %exit78, label %for.body.us

exit78:                      ; preds = %while.cond.i..exit_crit_edge.us, %for.body.lr.ph, %entry
  ret void
}