; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mattr=+simd128 | FileCheck %s

; Test loads and stores with custom alignment values.

target triple = "wasm32-unknown-unknown"

; ==============================================================================
; 16 x i8
; ==============================================================================

define <16 x i8> @load_v16i8_a1(ptr %p) {
; CHECK-LABEL: load_v16i8_a1:
; CHECK:         .functype load_v16i8_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <16 x i8>, ptr %p, align 1
  ret <16 x i8> %v
}

define <16 x i8> @load_v16i8_a4(ptr %p) {
; CHECK-LABEL: load_v16i8_a4:
; CHECK:         .functype load_v16i8_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <16 x i8>, ptr %p, align 4
  ret <16 x i8> %v
}

; 16 is the default alignment for v128 so no attribute is needed.
define <16 x i8> @load_v16i8_a16(ptr %p) {
; CHECK-LABEL: load_v16i8_a16:
; CHECK:         .functype load_v16i8_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <16 x i8>, ptr %p, align 16
  ret <16 x i8> %v
}

; 32 is greater than the default alignment so it is ignored.
define <16 x i8> @load_v16i8_a32(ptr %p) {
; CHECK-LABEL: load_v16i8_a32:
; CHECK:         .functype load_v16i8_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <16 x i8>, ptr %p, align 32
  ret <16 x i8> %v
}

define void @store_v16i8_a1(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: store_v16i8_a1:
; CHECK:         .functype store_v16i8_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <16 x i8> %v, ptr %p, align 1
  ret void
}

define void @store_v16i8_a4(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: store_v16i8_a4:
; CHECK:         .functype store_v16i8_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <16 x i8> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v16i8_a16(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: store_v16i8_a16:
; CHECK:         .functype store_v16i8_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <16 x i8> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v16i8_a32(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: store_v16i8_a32:
; CHECK:         .functype store_v16i8_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <16 x i8> %v, ptr %p, align 32
  ret void
}

; 1 is the default alignment for v128.load8_splat so no attribute is needed.
define <16 x i8> @load_splat_v16i8_a1(ptr %p) {
; CHECK-LABEL: load_splat_v16i8_a1:
; CHECK:         .functype load_splat_v16i8_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load8_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i8, ptr %p, align 1
  %v1 = insertelement <16 x i8> undef, i8 %e, i32 0
  %v2 = shufflevector <16 x i8> %v1, <16 x i8> undef, <16 x i32> zeroinitializer
  ret <16 x i8> %v2
}

; 2 is greater than the default alignment so it is ignored.
define <16 x i8> @load_splat_v16i8_a2(ptr %p) {
; CHECK-LABEL: load_splat_v16i8_a2:
; CHECK:         .functype load_splat_v16i8_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load8_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i8, ptr %p, align 2
  %v1 = insertelement <16 x i8> undef, i8 %e, i32 0
  %v2 = shufflevector <16 x i8> %v1, <16 x i8> undef, <16 x i32> zeroinitializer
  ret <16 x i8> %v2
}

; 1 is the default alignment for v128.load8_lane so no attribute is needed.
define <16 x i8> @load_lane_i8_a1(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: load_lane_i8_a1:
; CHECK:         .functype load_lane_i8_a1 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load8_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i8, ptr %p, align 1
  %v1 = insertelement <16 x i8> %v, i8 %e, i32 0
  ret <16 x i8> %v1
}

; 2 is greater than the default alignment so it is ignored.
define <16 x i8> @load_lane_i8_a2(ptr %p, <16 x i8> %v) {
; CHECK-LABEL: load_lane_i8_a2:
; CHECK:         .functype load_lane_i8_a2 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load8_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i8, ptr %p, align 2
  %v1 = insertelement <16 x i8> %v, i8 %e, i32 0
  ret <16 x i8> %v1
}

; 1 is the default alignment for v128.store8_lane so no attribute is needed.
define void @store_lane_i8_a1(<16 x i8> %v, ptr %p) {
; CHECK-LABEL: store_lane_i8_a1:
; CHECK:         .functype store_lane_i8_a1 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store8_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <16 x i8> %v, i32 0
  store i8 %x, ptr %p, align 1
  ret void
}

; 2 is greater than the default alignment so it is ignored.
define void @store_lane_i8_a2(<16 x i8> %v, ptr %p) {
; CHECK-LABEL: store_lane_i8_a2:
; CHECK:         .functype store_lane_i8_a2 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store8_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <16 x i8> %v, i32 0
  store i8 %x, ptr %p, align 2
  ret void
}

; ==============================================================================
; 8 x i16
; ==============================================================================

define <8 x i16> @load_v8i16_a1(ptr %p) {
; CHECK-LABEL: load_v8i16_a1:
; CHECK:         .functype load_v8i16_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i16>, ptr %p, align 1
  ret <8 x i16> %v
}

define <8 x i16> @load_v8i16_a4(ptr %p) {
; CHECK-LABEL: load_v8i16_a4:
; CHECK:         .functype load_v8i16_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i16>, ptr %p, align 4
  ret <8 x i16> %v
}

; 8 is the default alignment for v128 so no attribute is needed.
define <8 x i16> @load_v8i16_a16(ptr %p) {
; CHECK-LABEL: load_v8i16_a16:
; CHECK:         .functype load_v8i16_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i16>, ptr %p, align 16
  ret <8 x i16> %v
}

; 32 is greater than the default alignment so it is ignored.
define <8 x i16> @load_v8i16_a32(ptr %p) {
; CHECK-LABEL: load_v8i16_a32:
; CHECK:         .functype load_v8i16_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i16>, ptr %p, align 32
  ret <8 x i16> %v
}

define void @store_v8i16_a1(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: store_v8i16_a1:
; CHECK:         .functype store_v8i16_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <8 x i16> %v, ptr %p, align 1
  ret void
}

define void @store_v8i16_a4(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: store_v8i16_a4:
; CHECK:         .functype store_v8i16_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <8 x i16> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v8i16_a16(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: store_v8i16_a16:
; CHECK:         .functype store_v8i16_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <8 x i16> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v8i16_a32(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: store_v8i16_a32:
; CHECK:         .functype store_v8i16_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <8 x i16> %v, ptr %p, align 32
  ret void
}

define <8 x i8> @load_ext_v8i16_a1(ptr %p) {
; CHECK-LABEL: load_ext_v8i16_a1:
; CHECK:         .functype load_ext_v8i16_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 1
  ret <8 x i8> %v
}

define <8 x i8> @load_ext_v8i16_a2(ptr %p) {
; CHECK-LABEL: load_ext_v8i16_a2:
; CHECK:         .functype load_ext_v8i16_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 2
  ret <8 x i8> %v
}

define <8 x i8> @load_ext_v8i16_a4(ptr %p) {
; CHECK-LABEL: load_ext_v8i16_a4:
; CHECK:         .functype load_ext_v8i16_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 4
  ret <8 x i8> %v
}

; 8 is the default alignment for v128 extending load so no attribute is needed.
define <8 x i8> @load_ext_v8i16_a8(ptr %p) {
; CHECK-LABEL: load_ext_v8i16_a8:
; CHECK:         .functype load_ext_v8i16_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 8
  ret <8 x i8> %v
}

; 16 is greater than the default alignment so it is ignored.
define <8 x i8> @load_ext_v8i16_a16(ptr %p) {
; CHECK-LABEL: load_ext_v8i16_a16:
; CHECK:         .functype load_ext_v8i16_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 16
  ret <8 x i8> %v
}

define <8 x i16> @load_sext_v8i16_a1(ptr %p) {
; CHECK-LABEL: load_sext_v8i16_a1:
; CHECK:         .functype load_sext_v8i16_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i16x8.load8x8_s 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 1
  %v2 = sext <8 x i8> %v to <8 x i16>
  ret <8 x i16> %v2
}

define <8 x i16> @load_sext_v8i16_a2(ptr %p) {
; CHECK-LABEL: load_sext_v8i16_a2:
; CHECK:         .functype load_sext_v8i16_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i16x8.load8x8_s 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 2
  %v2 = sext <8 x i8> %v to <8 x i16>
  ret <8 x i16> %v2
}

define <8 x i16> @load_sext_v8i16_a4(ptr %p) {
; CHECK-LABEL: load_sext_v8i16_a4:
; CHECK:         .functype load_sext_v8i16_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i16x8.load8x8_s 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 4
  %v2 = sext <8 x i8> %v to <8 x i16>
  ret <8 x i16> %v2
}

; 8 is the default alignment for v128 extending load so no attribute is needed.
define <8 x i16> @load_sext_v8i16_a8(ptr %p) {
; CHECK-LABEL: load_sext_v8i16_a8:
; CHECK:         .functype load_sext_v8i16_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i16x8.load8x8_s 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 8
  %v2 = sext <8 x i8> %v to <8 x i16>
  ret <8 x i16> %v2
}

; 16 is greater than the default alignment so it is ignored.
define <8 x i16> @load_sext_v8i16_a16(ptr %p) {
; CHECK-LABEL: load_sext_v8i16_a16:
; CHECK:         .functype load_sext_v8i16_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i16x8.load8x8_s 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <8 x i8>, ptr %p, align 16
  %v2 = sext <8 x i8> %v to <8 x i16>
  ret <8 x i16> %v2
}

define <8 x i16> @load_splat_v8i16_a1(ptr %p) {
; CHECK-LABEL: load_splat_v8i16_a1:
; CHECK:         .functype load_splat_v8i16_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load16_splat 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 1
  %v1 = insertelement <8 x i16> undef, i16 %e, i32 0
  %v2 = shufflevector <8 x i16> %v1, <8 x i16> undef, <8 x i32> zeroinitializer
  ret <8 x i16> %v2
}

; 2 is the default alignment for v128.load16_splat so no attribute is needed.
define <8 x i16> @load_splat_v8i16_a2(ptr %p) {
; CHECK-LABEL: load_splat_v8i16_a2:
; CHECK:         .functype load_splat_v8i16_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load16_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 2
  %v1 = insertelement <8 x i16> undef, i16 %e, i32 0
  %v2 = shufflevector <8 x i16> %v1, <8 x i16> undef, <8 x i32> zeroinitializer
  ret <8 x i16> %v2
}

; 4 is greater than the default alignment so it is ignored.
define <8 x i16> @load_splat_v8i16_a4(ptr %p) {
; CHECK-LABEL: load_splat_v8i16_a4:
; CHECK:         .functype load_splat_v8i16_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load16_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 4
  %v1 = insertelement <8 x i16> undef, i16 %e, i32 0
  %v2 = shufflevector <8 x i16> %v1, <8 x i16> undef, <8 x i32> zeroinitializer
  ret <8 x i16> %v2
}

define <8 x i16> @load_lane_i16_a1(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: load_lane_i16_a1:
; CHECK:         .functype load_lane_i16_a1 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load16_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 1
  %v1 = insertelement <8 x i16> %v, i16 %e, i32 0
  ret <8 x i16> %v1
}

; 2 is the default alignment for v128.load16_lane so no attribute is needed.
define <8 x i16> @load_lane_i16_a2(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: load_lane_i16_a2:
; CHECK:         .functype load_lane_i16_a2 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load16_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 2
  %v1 = insertelement <8 x i16> %v, i16 %e, i32 0
  ret <8 x i16> %v1
}

; 4 is greater than the default alignment so it is ignored.
define <8 x i16> @load_lane_i16_a4(ptr %p, <8 x i16> %v) {
; CHECK-LABEL: load_lane_i16_a4:
; CHECK:         .functype load_lane_i16_a4 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load16_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i16, ptr %p, align 4
  %v1 = insertelement <8 x i16> %v, i16 %e, i32 0
  ret <8 x i16> %v1
}

define void @store_lane_i16_a1(<8 x i16> %v, ptr %p) {
; CHECK-LABEL: store_lane_i16_a1:
; CHECK:         .functype store_lane_i16_a1 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store16_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <8 x i16> %v, i32 0
  store i16 %x, ptr %p, align 1
  ret void
}

; 2 is the default alignment for v128.store16_lane so no attribute is needed.
define void @store_lane_i16_a2(<8 x i16> %v, ptr %p) {
; CHECK-LABEL: store_lane_i16_a2:
; CHECK:         .functype store_lane_i16_a2 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store16_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <8 x i16> %v, i32 0
  store i16 %x, ptr %p, align 2
  ret void
}

; 4 is greater than the default alignment so it is ignored.
define void @store_lane_i16_a4(<8 x i16> %v, ptr %p) {
; CHECK-LABEL: store_lane_i16_a4:
; CHECK:         .functype store_lane_i16_a4 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store16_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <8 x i16> %v, i32 0
  store i16 %x, ptr %p, align 4
  ret void
}

; ==============================================================================
; 4 x i32
; ==============================================================================

define <4 x i32> @load_v4i32_a1(ptr %p) {
; CHECK-LABEL: load_v4i32_a1:
; CHECK:         .functype load_v4i32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i32>, ptr %p, align 1
  ret <4 x i32> %v
}

define <4 x i32> @load_v4i32_a4(ptr %p) {
; CHECK-LABEL: load_v4i32_a4:
; CHECK:         .functype load_v4i32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i32>, ptr %p, align 4
  ret <4 x i32> %v
}

; 4 is the default alignment for v128 so no attribute is needed.
define <4 x i32> @load_v4i32_a16(ptr %p) {
; CHECK-LABEL: load_v4i32_a16:
; CHECK:         .functype load_v4i32_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i32>, ptr %p, align 16
  ret <4 x i32> %v
}

; 32 is greater than the default alignment so it is ignored.
define <4 x i32> @load_v4i32_a32(ptr %p) {
; CHECK-LABEL: load_v4i32_a32:
; CHECK:         .functype load_v4i32_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i32>, ptr %p, align 32
  ret <4 x i32> %v
}

define void @store_v4i32_a1(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: store_v4i32_a1:
; CHECK:         .functype store_v4i32_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <4 x i32> %v, ptr %p, align 1
  ret void
}

define void @store_v4i32_a4(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: store_v4i32_a4:
; CHECK:         .functype store_v4i32_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <4 x i32> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v4i32_a16(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: store_v4i32_a16:
; CHECK:         .functype store_v4i32_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <4 x i32> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v4i32_a32(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: store_v4i32_a32:
; CHECK:         .functype store_v4i32_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <4 x i32> %v, ptr %p, align 32
  ret void
}

define <4 x i16> @load_ext_v4i32_a1(ptr %p) {
; CHECK-LABEL: load_ext_v4i32_a1:
; CHECK:         .functype load_ext_v4i32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 1
  ret <4 x i16> %v
}

define <4 x i16> @load_ext_v4i32_a2(ptr %p) {
; CHECK-LABEL: load_ext_v4i32_a2:
; CHECK:         .functype load_ext_v4i32_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 2
  ret <4 x i16> %v
}

define <4 x i16> @load_ext_v4i32_a4(ptr %p) {
; CHECK-LABEL: load_ext_v4i32_a4:
; CHECK:         .functype load_ext_v4i32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 4
  ret <4 x i16> %v
}

; 8 is the default alignment for v128 extending load so no attribute is needed.
define <4 x i16> @load_ext_v4i32_a8(ptr %p) {
; CHECK-LABEL: load_ext_v4i32_a8:
; CHECK:         .functype load_ext_v4i32_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 8
  ret <4 x i16> %v
}

; 16 is greater than the default alignment so it is ignored.
define <4 x i16> @load_ext_v4i32_a16(ptr %p) {
; CHECK-LABEL: load_ext_v4i32_a16:
; CHECK:         .functype load_ext_v4i32_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 16
  ret <4 x i16> %v
}

define <4 x i32> @load_sext_v4i32_a1(ptr %p) {
; CHECK-LABEL: load_sext_v4i32_a1:
; CHECK:         .functype load_sext_v4i32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32x4.load16x4_s 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 1
  %v2 = sext <4 x i16> %v to <4 x i32>
  ret <4 x i32> %v2
}

define <4 x i32> @load_sext_v4i32_a2(ptr %p) {
; CHECK-LABEL: load_sext_v4i32_a2:
; CHECK:         .functype load_sext_v4i32_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32x4.load16x4_s 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 2
  %v2 = sext <4 x i16> %v to <4 x i32>
  ret <4 x i32> %v2
}

define <4 x i32> @load_sext_v4i32_a4(ptr %p) {
; CHECK-LABEL: load_sext_v4i32_a4:
; CHECK:         .functype load_sext_v4i32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32x4.load16x4_s 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 4
  %v2 = sext <4 x i16> %v to <4 x i32>
  ret <4 x i32> %v2
}

; 8 is the default alignment for v128 extending load so no attribute is needed.
define <4 x i32> @load_sext_v4i32_a8(ptr %p) {
; CHECK-LABEL: load_sext_v4i32_a8:
; CHECK:         .functype load_sext_v4i32_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32x4.load16x4_s 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 8
  %v2 = sext <4 x i16> %v to <4 x i32>
  ret <4 x i32> %v2
}

; 16 is greater than the default alignment so it is ignored.
define <4 x i32> @load_sext_v4i32_a16(ptr %p) {
; CHECK-LABEL: load_sext_v4i32_a16:
; CHECK:         .functype load_sext_v4i32_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32x4.load16x4_s 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x i16>, ptr %p, align 16
  %v2 = sext <4 x i16> %v to <4 x i32>
  ret <4 x i32> %v2
}

define <4 x i32> @load_splat_v4i32_a1(ptr %addr) {
; CHECK-LABEL: load_splat_v4i32_a1:
; CHECK:         .functype load_splat_v4i32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_splat 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %addr, align 1
  %v1 = insertelement <4 x i32> undef, i32 %e, i32 0
  %v2 = shufflevector <4 x i32> %v1, <4 x i32> undef, <4 x i32> zeroinitializer
  ret <4 x i32> %v2
}

define <4 x i32> @load_splat_v4i32_a2(ptr %addr) {
; CHECK-LABEL: load_splat_v4i32_a2:
; CHECK:         .functype load_splat_v4i32_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_splat 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %addr, align 2
  %v1 = insertelement <4 x i32> undef, i32 %e, i32 0
  %v2 = shufflevector <4 x i32> %v1, <4 x i32> undef, <4 x i32> zeroinitializer
  ret <4 x i32> %v2
}

; 4 is the default alignment for v128.load32_splat so no attribute is needed.
define <4 x i32> @load_splat_v4i32_a4(ptr %addr) {
; CHECK-LABEL: load_splat_v4i32_a4:
; CHECK:         .functype load_splat_v4i32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %addr, align 4
  %v1 = insertelement <4 x i32> undef, i32 %e, i32 0
  %v2 = shufflevector <4 x i32> %v1, <4 x i32> undef, <4 x i32> zeroinitializer
  ret <4 x i32> %v2
}

; 8 is greater than the default alignment so it is ignored.
define <4 x i32> @load_splat_v4i32_a8(ptr %addr) {
; CHECK-LABEL: load_splat_v4i32_a8:
; CHECK:         .functype load_splat_v4i32_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %addr, align 8
  %v1 = insertelement <4 x i32> undef, i32 %e, i32 0
  %v2 = shufflevector <4 x i32> %v1, <4 x i32> undef, <4 x i32> zeroinitializer
  ret <4 x i32> %v2
}

define <4 x i32> @load_lane_i32_a1(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: load_lane_i32_a1:
; CHECK:         .functype load_lane_i32_a1 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load32_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %p, align 1
  %v1 = insertelement <4 x i32> %v, i32 %e, i32 0
  ret <4 x i32> %v1
}

define <4 x i32> @load_lane_i32_a2(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: load_lane_i32_a2:
; CHECK:         .functype load_lane_i32_a2 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load32_lane 0:p2align=1, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %p, align 2
  %v1 = insertelement <4 x i32> %v, i32 %e, i32 0
  ret <4 x i32> %v1
}

; 4 is the default alignment for v128.load32_lane so no attribute is needed.
define <4 x i32> @load_lane_i32_a4(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: load_lane_i32_a4:
; CHECK:         .functype load_lane_i32_a4 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load32_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %p, align 4
  %v1 = insertelement <4 x i32> %v, i32 %e, i32 0
  ret <4 x i32> %v1
}

; 8 is greater than the default alignment so it is ignored.
define <4 x i32> @load_lane_i32_a8(ptr %p, <4 x i32> %v) {
; CHECK-LABEL: load_lane_i32_a8:
; CHECK:         .functype load_lane_i32_a8 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load32_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i32, ptr %p, align 8
  %v1 = insertelement <4 x i32> %v, i32 %e, i32 0
  ret <4 x i32> %v1
}

define void @store_lane_i32_a1(<4 x i32> %v, ptr %p) {
; CHECK-LABEL: store_lane_i32_a1:
; CHECK:         .functype store_lane_i32_a1 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store32_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <4 x i32> %v, i32 0
  store i32 %x, ptr %p, align 1
  ret void
}

define void @store_lane_i32_a2(<4 x i32> %v, ptr %p) {
; CHECK-LABEL: store_lane_i32_a2:
; CHECK:         .functype store_lane_i32_a2 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store32_lane 0:p2align=1, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <4 x i32> %v, i32 0
  store i32 %x, ptr %p, align 2
  ret void
}

; 4 is the default alignment for v128.store32_lane so no attribute is needed.
define void @store_lane_i32_a4(<4 x i32> %v, ptr %p) {
; CHECK-LABEL: store_lane_i32_a4:
; CHECK:         .functype store_lane_i32_a4 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store32_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <4 x i32> %v, i32 0
  store i32 %x, ptr %p, align 4
  ret void
}

; 8 is greater than the default alignment so it is ignored.
define void @store_lane_i32_a8(<4 x i32> %v, ptr %p) {
; CHECK-LABEL: store_lane_i32_a8:
; CHECK:         .functype store_lane_i32_a8 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store32_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <4 x i32> %v, i32 0
  store i32 %x, ptr %p, align 8
  ret void
}

define <4 x i32> @load_zero_i32_a1(ptr %p) {
; CHECK-LABEL: load_zero_i32_a1:
; CHECK:         .functype load_zero_i32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_zero 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %x = load i32, ptr %p, align 1
  %v = insertelement <4 x i32> zeroinitializer, i32 %x, i32 0
  ret <4 x i32> %v
}

define <4 x i32> @load_zero_i32_a2(ptr %p) {
; CHECK-LABEL: load_zero_i32_a2:
; CHECK:         .functype load_zero_i32_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_zero 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %x = load i32, ptr %p, align 2
  %v = insertelement <4 x i32> zeroinitializer, i32 %x, i32 0
  ret <4 x i32> %v
}

; 4 is the default alignment for v128.load32_zero so no attribute is needed.
define <4 x i32> @load_zero_i32_a4(ptr %p) {
; CHECK-LABEL: load_zero_i32_a4:
; CHECK:         .functype load_zero_i32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_zero 0
; CHECK-NEXT:    # fallthrough-return
  %x = load i32, ptr %p, align 4
  %v = insertelement <4 x i32> zeroinitializer, i32 %x, i32 0
  ret <4 x i32> %v
}

; 8 is greater than the default alignment so it is ignored.
define <4 x i32> @load_zero_i32_a8(ptr %p) {
; CHECK-LABEL: load_zero_i32_a8:
; CHECK:         .functype load_zero_i32_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load32_zero 0
; CHECK-NEXT:    # fallthrough-return
  %x = load i32, ptr %p, align 8
  %v = insertelement <4 x i32> zeroinitializer, i32 %x, i32 0
  ret <4 x i32> %v
}

; ==============================================================================
; 2 x i64
; ==============================================================================

define <2 x i64> @load_v2i64_a1(ptr %p) {
; CHECK-LABEL: load_v2i64_a1:
; CHECK:         .functype load_v2i64_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x i64>, ptr %p, align 1
  ret <2 x i64> %v
}

define <2 x i64> @load_v2i64_a4(ptr %p) {
; CHECK-LABEL: load_v2i64_a4:
; CHECK:         .functype load_v2i64_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x i64>, ptr %p, align 4
  ret <2 x i64> %v
}

; 2 is the default alignment for v128 so no attribute is needed.
define <2 x i64> @load_v2i64_a16(ptr %p) {
; CHECK-LABEL: load_v2i64_a16:
; CHECK:         .functype load_v2i64_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x i64>, ptr %p, align 16
  ret <2 x i64> %v
}

; 32 is greater than the default alignment so it is ignored.
define <2 x i64> @load_v2i64_a32(ptr %p) {
; CHECK-LABEL: load_v2i64_a32:
; CHECK:         .functype load_v2i64_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x i64>, ptr %p, align 32
  ret <2 x i64> %v
}

define void @store_v2i64_a1(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: store_v2i64_a1:
; CHECK:         .functype store_v2i64_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <2 x i64> %v, ptr %p, align 1
  ret void
}

define void @store_v2i64_a4(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: store_v2i64_a4:
; CHECK:         .functype store_v2i64_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <2 x i64> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v2i64_a16(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: store_v2i64_a16:
; CHECK:         .functype store_v2i64_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <2 x i64> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v2i64_a32(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: store_v2i64_a32:
; CHECK:         .functype store_v2i64_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <2 x i64> %v, ptr %p, align 32
  ret void
}

define <2 x i64> @load_splat_v2i64_a1(ptr %p) {
; CHECK-LABEL: load_splat_v2i64_a1:
; CHECK:         .functype load_splat_v2i64_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_splat 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 1
  %v1 = insertelement <2 x i64> undef, i64 %e, i32 0
  %v2 = shufflevector <2 x i64> %v1, <2 x i64> undef, <2 x i32> zeroinitializer
  ret <2 x i64> %v2
}

define <2 x i64> @load_splat_v2i64_a2(ptr %p) {
; CHECK-LABEL: load_splat_v2i64_a2:
; CHECK:         .functype load_splat_v2i64_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_splat 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 2
  %v1 = insertelement <2 x i64> undef, i64 %e, i32 0
  %v2 = shufflevector <2 x i64> %v1, <2 x i64> undef, <2 x i32> zeroinitializer
  ret <2 x i64> %v2
}

define <2 x i64> @load_splat_v2i64_a4(ptr %p) {
; CHECK-LABEL: load_splat_v2i64_a4:
; CHECK:         .functype load_splat_v2i64_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_splat 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 4
  %v1 = insertelement <2 x i64> undef, i64 %e, i32 0
  %v2 = shufflevector <2 x i64> %v1, <2 x i64> undef, <2 x i32> zeroinitializer
  ret <2 x i64> %v2
}

; 8 is the default alignment for v128.load64_splat so no attribute is needed.
define <2 x i64> @load_splat_v2i64_a8(ptr %p) {
; CHECK-LABEL: load_splat_v2i64_a8:
; CHECK:         .functype load_splat_v2i64_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 8
  %v1 = insertelement <2 x i64> undef, i64 %e, i32 0
  %v2 = shufflevector <2 x i64> %v1, <2 x i64> undef, <2 x i32> zeroinitializer
  ret <2 x i64> %v2
}

; 16 is greater than the default alignment so it is ignored.
define <2 x i64> @load_splat_v2i64_a16(ptr %p) {
; CHECK-LABEL: load_splat_v2i64_a16:
; CHECK:         .functype load_splat_v2i64_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_splat 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 16
  %v1 = insertelement <2 x i64> undef, i64 %e, i32 0
  %v2 = shufflevector <2 x i64> %v1, <2 x i64> undef, <2 x i32> zeroinitializer
  ret <2 x i64> %v2
}

define <2 x i64> @load_lane_i64_a1(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: load_lane_i64_a1:
; CHECK:         .functype load_lane_i64_a1 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load64_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 1
  %v1 = insertelement <2 x i64> %v, i64 %e, i32 0
  ret <2 x i64> %v1
}

define <2 x i64> @load_lane_i64_a2(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: load_lane_i64_a2:
; CHECK:         .functype load_lane_i64_a2 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load64_lane 0:p2align=1, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 2
  %v1 = insertelement <2 x i64> %v, i64 %e, i32 0
  ret <2 x i64> %v1
}

define <2 x i64> @load_lane_i64_a4(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: load_lane_i64_a4:
; CHECK:         .functype load_lane_i64_a4 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load64_lane 0:p2align=2, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 4
  %v1 = insertelement <2 x i64> %v, i64 %e, i32 0
  ret <2 x i64> %v1
}

; 8 is the default alignment for v128.load64_lane so no attribute is needed.
define <2 x i64> @load_lane_i64_a8(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: load_lane_i64_a8:
; CHECK:         .functype load_lane_i64_a8 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load64_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 8
  %v1 = insertelement <2 x i64> %v, i64 %e, i32 0
  ret <2 x i64> %v1
}

; 16 is greater than the default alignment so it is ignored.
define <2 x i64> @load_lane_i64_a16(ptr %p, <2 x i64> %v) {
; CHECK-LABEL: load_lane_i64_a16:
; CHECK:         .functype load_lane_i64_a16 (i32, v128) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.load64_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %e = load i64, ptr %p, align 16
  %v1 = insertelement <2 x i64> %v, i64 %e, i32 0
  ret <2 x i64> %v1
}

define void @store_lane_i64_a1(<2 x i64> %v, ptr %p) {
; CHECK-LABEL: store_lane_i64_a1:
; CHECK:         .functype store_lane_i64_a1 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store64_lane 0:p2align=0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <2 x i64> %v, i32 0
  store i64 %x, ptr %p, align 1
  ret void
}

define void @store_lane_i64_a2(<2 x i64> %v, ptr %p) {
; CHECK-LABEL: store_lane_i64_a2:
; CHECK:         .functype store_lane_i64_a2 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store64_lane 0:p2align=1, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <2 x i64> %v, i32 0
  store i64 %x, ptr %p, align 2
  ret void
}

define void @store_lane_i64_a4(<2 x i64> %v, ptr %p) {
; CHECK-LABEL: store_lane_i64_a4:
; CHECK:         .functype store_lane_i64_a4 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store64_lane 0:p2align=2, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <2 x i64> %v, i32 0
  store i64 %x, ptr %p, align 4
  ret void
}

; 8 is the default alignment for v128.store64_lane so no attribute is needed.
define void @store_lane_i64_a8(<2 x i64> %v, ptr %p) {
; CHECK-LABEL: store_lane_i64_a8:
; CHECK:         .functype store_lane_i64_a8 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store64_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <2 x i64> %v, i32 0
  store i64 %x, ptr %p, align 8
  ret void
}

; 16 is greater than the default alignment so it is ignored.
define void @store_lane_i64_a16(<2 x i64> %v, ptr %p) {
; CHECK-LABEL: store_lane_i64_a16:
; CHECK:         .functype store_lane_i64_a16 (v128, i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.store64_lane 0, 0
; CHECK-NEXT:    # fallthrough-return
  %x = extractelement <2 x i64> %v, i32 0
  store i64 %x, ptr %p, align 16
  ret void
}

define <2 x i64> @load_zero_i64_a1(ptr %p) {
; CHECK-LABEL: load_zero_i64_a1:
; CHECK:         .functype load_zero_i64_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %x = load i64, ptr %p, align 1
  %v = insertelement <2 x i64> zeroinitializer, i64 %x, i32 0
  ret <2 x i64> %v
}

define <2 x i64> @load_zero_i64_a2(ptr %p) {
; CHECK-LABEL: load_zero_i64_a2:
; CHECK:         .functype load_zero_i64_a2 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=1
; CHECK-NEXT:    # fallthrough-return
  %x = load i64, ptr %p, align 2
  %v = insertelement <2 x i64> zeroinitializer, i64 %x, i32 0
  ret <2 x i64> %v
}

define <2 x i64> @load_zero_i64_a4(ptr %p) {
; CHECK-LABEL: load_zero_i64_a4:
; CHECK:         .functype load_zero_i64_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %x = load i64, ptr %p, align 4
  %v = insertelement <2 x i64> zeroinitializer, i64 %x, i32 0
  ret <2 x i64> %v
}

; 8 is the default alignment for v128.load64_zero so no attribute is needed.
define <2 x i64> @load_zero_i64_a8(ptr %p) {
; CHECK-LABEL: load_zero_i64_a8:
; CHECK:         .functype load_zero_i64_a8 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0
; CHECK-NEXT:    # fallthrough-return
  %x = load i64, ptr %p, align 8
  %v = insertelement <2 x i64> zeroinitializer, i64 %x, i32 0
  ret <2 x i64> %v
}

; 16 is greater than the default alignment so it is ignored.
define <2 x i64> @load_zero_i64_a16(ptr %p) {
; CHECK-LABEL: load_zero_i64_a16:
; CHECK:         .functype load_zero_i64_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load64_zero 0
; CHECK-NEXT:    # fallthrough-return
  %x = load i64, ptr %p, align 16
  %v = insertelement <2 x i64> zeroinitializer, i64 %x, i32 0
  ret <2 x i64> %v
}

; ==============================================================================
; 4 x float
; ==============================================================================

define <4 x float> @load_v4f32_a1(ptr %p) {
; CHECK-LABEL: load_v4f32_a1:
; CHECK:         .functype load_v4f32_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x float>, ptr %p, align 1
  ret <4 x float> %v
}

define <4 x float> @load_v4f32_a4(ptr %p) {
; CHECK-LABEL: load_v4f32_a4:
; CHECK:         .functype load_v4f32_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x float>, ptr %p, align 4
  ret <4 x float> %v
}

; 4 is the default alignment for v128 so no attribute is needed.
define <4 x float> @load_v4f32_a16(ptr %p) {
; CHECK-LABEL: load_v4f32_a16:
; CHECK:         .functype load_v4f32_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x float>, ptr %p, align 16
  ret <4 x float> %v
}

; 32 is greater than the default alignment so it is ignored.
define <4 x float> @load_v4f32_a32(ptr %p) {
; CHECK-LABEL: load_v4f32_a32:
; CHECK:         .functype load_v4f32_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <4 x float>, ptr %p, align 32
  ret <4 x float> %v
}

define void @store_v4f32_a1(ptr %p, <4 x float> %v) {
; CHECK-LABEL: store_v4f32_a1:
; CHECK:         .functype store_v4f32_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <4 x float> %v, ptr %p, align 1
  ret void
}

define void @store_v4f32_a4(ptr %p, <4 x float> %v) {
; CHECK-LABEL: store_v4f32_a4:
; CHECK:         .functype store_v4f32_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <4 x float> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v4f32_a16(ptr %p, <4 x float> %v) {
; CHECK-LABEL: store_v4f32_a16:
; CHECK:         .functype store_v4f32_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <4 x float> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v4f32_a32(ptr %p, <4 x float> %v) {
; CHECK-LABEL: store_v4f32_a32:
; CHECK:         .functype store_v4f32_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <4 x float> %v, ptr %p, align 32
  ret void
}

; ==============================================================================
; 2 x double
; ==============================================================================

define <2 x double> @load_v2f64_a1(ptr %p) {
; CHECK-LABEL: load_v2f64_a1:
; CHECK:         .functype load_v2f64_a1 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x double>, ptr %p, align 1
  ret <2 x double> %v
}

define <2 x double> @load_v2f64_a4(ptr %p) {
; CHECK-LABEL: load_v2f64_a4:
; CHECK:         .functype load_v2f64_a4 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x double>, ptr %p, align 4
  ret <2 x double> %v
}

; 2 is the default alignment for v128 so no attribute is needed.
define <2 x double> @load_v2f64_a16(ptr %p) {
; CHECK-LABEL: load_v2f64_a16:
; CHECK:         .functype load_v2f64_a16 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x double>, ptr %p, align 16
  ret <2 x double> %v
}

; 32 is greater than the default alignment so it is ignored.
define <2 x double> @load_v2f64_a32(ptr %p) {
; CHECK-LABEL: load_v2f64_a32:
; CHECK:         .functype load_v2f64_a32 (i32) -> (v128)
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    v128.load 0
; CHECK-NEXT:    # fallthrough-return
  %v = load <2 x double>, ptr %p, align 32
  ret <2 x double> %v
}

define void @store_v2f64_a1(ptr %p, <2 x double> %v) {
; CHECK-LABEL: store_v2f64_a1:
; CHECK:         .functype store_v2f64_a1 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=0
; CHECK-NEXT:    # fallthrough-return
  store <2 x double> %v, ptr %p, align 1
  ret void
}

define void @store_v2f64_a4(ptr %p, <2 x double> %v) {
; CHECK-LABEL: store_v2f64_a4:
; CHECK:         .functype store_v2f64_a4 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0:p2align=2
; CHECK-NEXT:    # fallthrough-return
  store <2 x double> %v, ptr %p, align 4
  ret void
}

; 16 is the default alignment for v128 so no attribute is needed.
define void @store_v2f64_a16(ptr %p, <2 x double> %v) {
; CHECK-LABEL: store_v2f64_a16:
; CHECK:         .functype store_v2f64_a16 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <2 x double> %v, ptr %p, align 16
  ret void
}

; 32 is greater than the default alignment so it is ignored.
define void @store_v2f64_a32(ptr %p, <2 x double> %v) {
; CHECK-LABEL: store_v2f64_a32:
; CHECK:         .functype store_v2f64_a32 (i32, v128) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    local.get 1
; CHECK-NEXT:    v128.store 0
; CHECK-NEXT:    # fallthrough-return
  store <2 x double> %v, ptr %p, align 32
  ret void
}