1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mattr=+simd128 | FileCheck %s
3
4;; Test that SIMD extending operations can be successfully selected
5
6target triple = "wasm32-unknown-unknown"
7
8define <8 x i16> @extend_low_i8x16_s(<16 x i8> %v) {
9; CHECK-LABEL: extend_low_i8x16_s:
10; CHECK:         .functype extend_low_i8x16_s (v128) -> (v128)
11; CHECK-NEXT:  # %bb.0:
12; CHECK-NEXT:    local.get 0
13; CHECK-NEXT:    i16x8.extend_low_i8x16_s
14; CHECK-NEXT:    # fallthrough-return
15  %low = shufflevector <16 x i8> %v, <16 x i8> undef,
16           <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
17  %extended = sext <8 x i8> %low to <8 x i16>
18  ret <8 x i16> %extended
19}
20
21define <8 x i16> @extend_low_i8x16_u(<16 x i8> %v) {
22; CHECK-LABEL: extend_low_i8x16_u:
23; CHECK:         .functype extend_low_i8x16_u (v128) -> (v128)
24; CHECK-NEXT:  # %bb.0:
25; CHECK-NEXT:    local.get 0
26; CHECK-NEXT:    i16x8.extend_low_i8x16_u
27; CHECK-NEXT:    # fallthrough-return
28  %low = shufflevector <16 x i8> %v, <16 x i8> undef,
29           <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
30  %extended = zext <8 x i8> %low to <8 x i16>
31  ret <8 x i16> %extended
32}
33
34define <8 x i16> @extend_high_i8x16_s(<16 x i8> %v) {
35; CHECK-LABEL: extend_high_i8x16_s:
36; CHECK:         .functype extend_high_i8x16_s (v128) -> (v128)
37; CHECK-NEXT:  # %bb.0:
38; CHECK-NEXT:    local.get 0
39; CHECK-NEXT:    i16x8.extend_high_i8x16_s
40; CHECK-NEXT:    # fallthrough-return
41  %low = shufflevector <16 x i8> %v, <16 x i8> undef,
42           <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
43  %extended = sext <8 x i8> %low to <8 x i16>
44  ret <8 x i16> %extended
45}
46
47define <8 x i16> @extend_high_i8x16_u(<16 x i8> %v) {
48; CHECK-LABEL: extend_high_i8x16_u:
49; CHECK:         .functype extend_high_i8x16_u (v128) -> (v128)
50; CHECK-NEXT:  # %bb.0:
51; CHECK-NEXT:    local.get 0
52; CHECK-NEXT:    i16x8.extend_high_i8x16_u
53; CHECK-NEXT:    # fallthrough-return
54  %low = shufflevector <16 x i8> %v, <16 x i8> undef,
55           <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
56  %extended = zext <8 x i8> %low to <8 x i16>
57  ret <8 x i16> %extended
58}
59
60define <4 x i32> @extend_low_i16x8_s(<8 x i16> %v) {
61; CHECK-LABEL: extend_low_i16x8_s:
62; CHECK:         .functype extend_low_i16x8_s (v128) -> (v128)
63; CHECK-NEXT:  # %bb.0:
64; CHECK-NEXT:    local.get 0
65; CHECK-NEXT:    i32x4.extend_low_i16x8_s
66; CHECK-NEXT:    # fallthrough-return
67  %low = shufflevector <8 x i16> %v, <8 x i16> undef,
68           <4 x i32> <i32 0, i32 1, i32 2, i32 3>
69  %extended = sext <4 x i16> %low to <4 x i32>
70  ret <4 x i32> %extended
71}
72
73define <4 x i32> @extend_low_i16x8_u(<8 x i16> %v) {
74; CHECK-LABEL: extend_low_i16x8_u:
75; CHECK:         .functype extend_low_i16x8_u (v128) -> (v128)
76; CHECK-NEXT:  # %bb.0:
77; CHECK-NEXT:    local.get 0
78; CHECK-NEXT:    i32x4.extend_low_i16x8_u
79; CHECK-NEXT:    # fallthrough-return
80  %low = shufflevector <8 x i16> %v, <8 x i16> undef,
81           <4 x i32> <i32 0, i32 1, i32 2, i32 3>
82  %extended = zext <4 x i16> %low to <4 x i32>
83  ret <4 x i32> %extended
84}
85
86define <4 x i32> @extend_high_i16x8_s(<8 x i16> %v) {
87; CHECK-LABEL: extend_high_i16x8_s:
88; CHECK:         .functype extend_high_i16x8_s (v128) -> (v128)
89; CHECK-NEXT:  # %bb.0:
90; CHECK-NEXT:    local.get 0
91; CHECK-NEXT:    i32x4.extend_high_i16x8_s
92; CHECK-NEXT:    # fallthrough-return
93  %low = shufflevector <8 x i16> %v, <8 x i16> undef,
94           <4 x i32> <i32 4, i32 5, i32 6, i32 7>
95  %extended = sext <4 x i16> %low to <4 x i32>
96  ret <4 x i32> %extended
97}
98
99define <4 x i32> @extend_high_i16x8_u(<8 x i16> %v) {
100; CHECK-LABEL: extend_high_i16x8_u:
101; CHECK:         .functype extend_high_i16x8_u (v128) -> (v128)
102; CHECK-NEXT:  # %bb.0:
103; CHECK-NEXT:    local.get 0
104; CHECK-NEXT:    i32x4.extend_high_i16x8_u
105; CHECK-NEXT:    # fallthrough-return
106  %low = shufflevector <8 x i16> %v, <8 x i16> undef,
107           <4 x i32> <i32 4, i32 5, i32 6, i32 7>
108  %extended = zext <4 x i16> %low to <4 x i32>
109  ret <4 x i32> %extended
110}
111
112define <2 x i64> @extend_low_i32x4_s(<4 x i32> %v) {
113; CHECK-LABEL: extend_low_i32x4_s:
114; CHECK:         .functype extend_low_i32x4_s (v128) -> (v128)
115; CHECK-NEXT:  # %bb.0:
116; CHECK-NEXT:    local.get 0
117; CHECK-NEXT:    i64x2.extend_low_i32x4_s
118; CHECK-NEXT:    # fallthrough-return
119  %low = shufflevector <4 x i32> %v, <4 x i32> undef,
120           <2 x i32> <i32 0, i32 1>
121  %extended = sext <2 x i32> %low to <2 x i64>
122  ret <2 x i64> %extended
123}
124
125define <2 x i64> @extend_low_i32x4_u(<4 x i32> %v) {
126; CHECK-LABEL: extend_low_i32x4_u:
127; CHECK:         .functype extend_low_i32x4_u (v128) -> (v128)
128; CHECK-NEXT:  # %bb.0:
129; CHECK-NEXT:    local.get 0
130; CHECK-NEXT:    i64x2.extend_low_i32x4_u
131; CHECK-NEXT:    # fallthrough-return
132  %low = shufflevector <4 x i32> %v, <4 x i32> undef,
133           <2 x i32> <i32 0, i32 1>
134  %extended = zext <2 x i32> %low to <2 x i64>
135  ret <2 x i64> %extended
136}
137
138define <2 x i64> @extend_high_i32x4_s(<4 x i32> %v) {
139; CHECK-LABEL: extend_high_i32x4_s:
140; CHECK:         .functype extend_high_i32x4_s (v128) -> (v128)
141; CHECK-NEXT:  # %bb.0:
142; CHECK-NEXT:    local.get 0
143; CHECK-NEXT:    i64x2.extend_high_i32x4_s
144; CHECK-NEXT:    # fallthrough-return
145  %low = shufflevector <4 x i32> %v, <4 x i32> undef,
146           <2 x i32> <i32 2, i32 3>
147  %extended = sext <2 x i32> %low to <2 x i64>
148  ret <2 x i64> %extended
149}
150
151define <2 x i64> @extend_high_i32x4_u(<4 x i32> %v) {
152; CHECK-LABEL: extend_high_i32x4_u:
153; CHECK:         .functype extend_high_i32x4_u (v128) -> (v128)
154; CHECK-NEXT:  # %bb.0:
155; CHECK-NEXT:    local.get 0
156; CHECK-NEXT:    i64x2.extend_high_i32x4_u
157; CHECK-NEXT:    # fallthrough-return
158  %low = shufflevector <4 x i32> %v, <4 x i32> undef,
159           <2 x i32> <i32 2, i32 3>
160  %extended = zext <2 x i32> %low to <2 x i64>
161  ret <2 x i64> %extended
162}
163
164;; Also test that similar patterns with offsets not corresponding to
165;; the low or high half are correctly expanded.
166
167define <8 x i16> @extend_lowish_i8x16_s(<16 x i8> %v) {
168; CHECK-LABEL: extend_lowish_i8x16_s:
169; CHECK:         .functype extend_lowish_i8x16_s (v128) -> (v128)
170; CHECK-NEXT:  # %bb.0:
171; CHECK-NEXT:    local.get 0
172; CHECK-NEXT:    local.get 0
173; CHECK-NEXT:    i8x16.shuffle 1, 0, 2, 0, 3, 0, 4, 0, 5, 0, 6, 0, 7, 0, 8, 0
174; CHECK-NEXT:    i32.const 8
175; CHECK-NEXT:    i16x8.shl
176; CHECK-NEXT:    i32.const 8
177; CHECK-NEXT:    i16x8.shr_s
178; CHECK-NEXT:    # fallthrough-return
179  %lowish = shufflevector <16 x i8> %v, <16 x i8> undef,
180           <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
181  %extended = sext <8 x i8> %lowish to <8 x i16>
182  ret <8 x i16> %extended
183}
184
185define <4 x i32> @extend_lowish_i16x8_s(<8 x i16> %v) {
186; CHECK-LABEL: extend_lowish_i16x8_s:
187; CHECK:         .functype extend_lowish_i16x8_s (v128) -> (v128)
188; CHECK-NEXT:  # %bb.0:
189; CHECK-NEXT:    local.get 0
190; CHECK-NEXT:    local.get 0
191; CHECK-NEXT:    i8x16.shuffle 2, 3, 0, 0, 4, 5, 0, 0, 6, 7, 0, 0, 8, 9, 0, 0
192; CHECK-NEXT:    i32.const 16
193; CHECK-NEXT:    i32x4.shl
194; CHECK-NEXT:    i32.const 16
195; CHECK-NEXT:    i32x4.shr_s
196; CHECK-NEXT:    # fallthrough-return
197  %lowish = shufflevector <8 x i16> %v, <8 x i16> undef,
198           <4 x i32> <i32 1, i32 2, i32 3, i32 4>
199  %extended = sext <4 x i16> %lowish to <4 x i32>
200  ret <4 x i32> %extended
201}
202