; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=hexagon < %s | FileCheck %s

define void @f0(ptr %a0, ptr %a1, ptr %a2) #0 {
; CHECK-LABEL: f0:
; CHECK:       // %bb.0: // %b0
; CHECK-NEXT:    {
; CHECK-NEXT:     v0 = vmem(r0+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1 = vmem(r1+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v2.w = vmpye(v0.w,v1.uh)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v2.w += vmpyo(v0.w,v1.h):<<1:sat:shift
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     vmem(r2+#0) = v2
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }
b0:
  %v0 = load <32 x i32>, ptr %a0, align 128
  %v1 = load <32 x i32>, ptr %a1, align 128
  %v2 = sext <32 x i32> %v0 to <32 x i64>
  %v3 = sext <32 x i32> %v1 to <32 x i64>
  %v4 = mul nsw <32 x i64> %v2, %v3
  %v5 = lshr <32 x i64> %v4, <i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31>
  %v6 = trunc <32 x i64> %v5 to <32 x i32>
  store <32 x i32> %v6, ptr %a2, align 128
  ret void
}

define void @f1(ptr %a0, ptr %a1, ptr %a2) #0 {
; CHECK-LABEL: f1:
; CHECK:       // %bb.0: // %b0
; CHECK-NEXT:    {
; CHECK-NEXT:     v0 = vmem(r0+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1 = vmem(r1+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v2.w = vmpye(v0.w,v1.uh)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v2.w += vmpyo(v0.w,v1.h):<<1:rnd:sat:shift
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     vmem(r2+#0) = v2
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }
b0:
  %v0 = load <32 x i32>, ptr %a0, align 128
  %v1 = load <32 x i32>, ptr %a1, align 128
  %v2 = sext <32 x i32> %v0 to <32 x i64>
  %v3 = sext <32 x i32> %v1 to <32 x i64>
  %v4 = mul nsw <32 x i64> %v2, %v3
  %v5 = add nsw <32 x i64> %v4, <i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824, i64 1073741824>
  %v6 = lshr <32 x i64> %v5, <i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31, i64 31>
  %v7 = trunc <32 x i64> %v6 to <32 x i32>
  store <32 x i32> %v7, ptr %a2, align 128
  ret void
}

define void @f2(ptr %a0, ptr %a1, ptr %a2) #0 {
; CHECK-LABEL: f2:
; CHECK:       // %bb.0: // %b0
; CHECK-NEXT:    {
; CHECK-NEXT:     v0 = vmem(r1+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r7 = #124
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     r3 = #15
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1 = vmem(r0+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1:0.w = vmpy(v0.h,v1.h)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1:0 = vshuff(v1,v0,r7)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v0.uw = vlsr(v0.uw,r3)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1.uw = vlsr(v1.uw,r3)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v0.h = vpacke(v1.w,v0.w)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     vmem(r2+#0) = v0
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }
b0:
  %v0 = load <64 x i16>, ptr %a0, align 128
  %v1 = load <64 x i16>, ptr %a1, align 128
  %v2 = sext <64 x i16> %v0 to <64 x i32>
  %v3 = sext <64 x i16> %v1 to <64 x i32>
  %v4 = mul nsw <64 x i32> %v2, %v3
  %v5 = lshr <64 x i32> %v4, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
  %v6 = trunc <64 x i32> %v5 to <64 x i16>
  store <64 x i16> %v6, ptr %a2, align 128
  ret void
}

define void @f3(ptr %a0, ptr %a1, ptr %a2) #0 {
; CHECK-LABEL: f3:
; CHECK:       // %bb.0: // %b0
; CHECK-NEXT:    {
; CHECK-NEXT:     v0 = vmem(r0+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v1 = vmem(r1+#0)
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     v0.h = vmpy(v0.h,v1.h):<<1:rnd:sat
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     vmem(r2+#0) = v0
; CHECK-NEXT:    }
; CHECK-NEXT:    {
; CHECK-NEXT:     jumpr r31
; CHECK-NEXT:    }
b0:
  %v0 = load <64 x i16>, ptr %a0, align 128
  %v1 = load <64 x i16>, ptr %a1, align 128
  %v2 = sext <64 x i16> %v0 to <64 x i32>
  %v3 = sext <64 x i16> %v1 to <64 x i32>
  %v4 = mul nsw <64 x i32> %v2, %v3
  %v5 = add nsw <64 x i32> %v4, <i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384, i32 16384>
  %v6 = lshr <64 x i32> %v5, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
  %v7 = trunc <64 x i32> %v6 to <64 x i16>
  store <64 x i16> %v7, ptr %a2, align 128
  ret void
}

attributes #0 = { nounwind "target-features"="+v68,+hvxv68,+hvx-length128b,-packets" }