xref: /llvm-project/llvm/test/CodeGen/AArch64/aarch64-dup-ext-scalable.ll (revision cc82f1290a1e2157a6c0530d78d8cc84d2b8553d) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple aarch64-none-linux-gnu -mattr=+sve | FileCheck %s
3
4define <vscale x 2 x i16> @dupsext_v2i8_v2i16(i8 %src, <vscale x 2 x i16> %b) {
5; CHECK-LABEL: dupsext_v2i8_v2i16:
6; CHECK:       // %bb.0: // %entry
7; CHECK-NEXT:    sxtb w8, w0
8; CHECK-NEXT:    ptrue p0.d
9; CHECK-NEXT:    mov z1.d, x8
10; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
11; CHECK-NEXT:    ret
12entry:
13    %in = sext i8 %src to i16
14    %broadcast.splatinsert = insertelement <vscale x 2 x i16> undef, i16 %in, i16 0
15    %broadcast.splat = shufflevector <vscale x 2 x i16> %broadcast.splatinsert, <vscale x 2 x i16> undef, <vscale x 2 x i32> zeroinitializer
16    %out = mul nsw <vscale x 2 x i16> %broadcast.splat, %b
17    ret <vscale x 2 x i16> %out
18}
19
20define <vscale x 4 x i16> @dupsext_v4i8_v4i16(i8 %src, <vscale x 4 x i16> %b) {
21; CHECK-LABEL: dupsext_v4i8_v4i16:
22; CHECK:       // %bb.0: // %entry
23; CHECK-NEXT:    sxtb w8, w0
24; CHECK-NEXT:    ptrue p0.s
25; CHECK-NEXT:    mov z1.s, w8
26; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
27; CHECK-NEXT:    ret
28entry:
29    %in = sext i8 %src to i16
30    %broadcast.splatinsert = insertelement <vscale x 4 x i16> undef, i16 %in, i16 0
31    %broadcast.splat = shufflevector <vscale x 4 x i16> %broadcast.splatinsert, <vscale x 4 x i16> undef, <vscale x 4 x i32> zeroinitializer
32    %out = mul nsw <vscale x 4 x i16> %broadcast.splat, %b
33    ret <vscale x 4 x i16> %out
34}
35
36define <vscale x 8 x i16> @dupsext_v8i8_v8i16(i8 %src, <vscale x 8 x i16> %b) {
37; CHECK-LABEL: dupsext_v8i8_v8i16:
38; CHECK:       // %bb.0: // %entry
39; CHECK-NEXT:    sxtb w8, w0
40; CHECK-NEXT:    ptrue p0.h
41; CHECK-NEXT:    mov z1.h, w8
42; CHECK-NEXT:    mul z0.h, p0/m, z0.h, z1.h
43; CHECK-NEXT:    ret
44entry:
45    %in = sext i8 %src to i16
46    %broadcast.splatinsert = insertelement <vscale x 8 x i16> undef, i16 %in, i16 0
47    %broadcast.splat = shufflevector <vscale x 8 x i16> %broadcast.splatinsert, <vscale x 8 x i16> undef, <vscale x 8 x i32> zeroinitializer
48    %out = mul nsw <vscale x 8 x i16> %broadcast.splat, %b
49    ret <vscale x 8 x i16> %out
50}
51
52define <vscale x 2 x i32> @dupsext_v2i8_v2i32(i8 %src, <vscale x 2 x i32> %b) {
53; CHECK-LABEL: dupsext_v2i8_v2i32:
54; CHECK:       // %bb.0: // %entry
55; CHECK-NEXT:    sxtb w8, w0
56; CHECK-NEXT:    ptrue p0.d
57; CHECK-NEXT:    mov z1.d, x8
58; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
59; CHECK-NEXT:    ret
60entry:
61    %in = sext i8 %src to i32
62    %broadcast.splatinsert = insertelement <vscale x 2 x i32> undef, i32 %in, i32 0
63    %broadcast.splat = shufflevector <vscale x 2 x i32> %broadcast.splatinsert, <vscale x 2 x i32> undef, <vscale x 2 x i32> zeroinitializer
64    %out = mul nsw <vscale x 2 x i32> %broadcast.splat, %b
65    ret <vscale x 2 x i32> %out
66}
67
68define <vscale x 4 x i32> @dupsext_v4i8_v4i32(i8 %src, <vscale x 4 x i32> %b) {
69; CHECK-LABEL: dupsext_v4i8_v4i32:
70; CHECK:       // %bb.0: // %entry
71; CHECK-NEXT:    sxtb w8, w0
72; CHECK-NEXT:    ptrue p0.s
73; CHECK-NEXT:    mov z1.s, w8
74; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
75; CHECK-NEXT:    ret
76entry:
77    %in = sext i8 %src to i32
78    %broadcast.splatinsert = insertelement <vscale x 4 x i32> undef, i32 %in, i32 0
79    %broadcast.splat = shufflevector <vscale x 4 x i32> %broadcast.splatinsert, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
80    %out = mul nsw <vscale x 4 x i32> %broadcast.splat, %b
81    ret <vscale x 4 x i32> %out
82}
83
84define <vscale x 2 x i64> @dupsext_v2i8_v2i64(i8 %src, <vscale x 2 x i64> %b) {
85; CHECK-LABEL: dupsext_v2i8_v2i64:
86; CHECK:       // %bb.0: // %entry
87; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
88; CHECK-NEXT:    sxtb x8, w0
89; CHECK-NEXT:    ptrue p0.d
90; CHECK-NEXT:    mov z1.d, x8
91; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
92; CHECK-NEXT:    ret
93entry:
94    %in = sext i8 %src to i64
95    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
96    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
97    %out = mul nsw <vscale x 2 x i64> %broadcast.splat, %b
98    ret <vscale x 2 x i64> %out
99}
100
101define <vscale x 2 x i32> @dupsext_v2i16_v2i32(i16 %src, <vscale x 2 x i32> %b) {
102; CHECK-LABEL: dupsext_v2i16_v2i32:
103; CHECK:       // %bb.0: // %entry
104; CHECK-NEXT:    sxth w8, w0
105; CHECK-NEXT:    ptrue p0.d
106; CHECK-NEXT:    mov z1.d, x8
107; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
108; CHECK-NEXT:    ret
109entry:
110    %in = sext i16 %src to i32
111    %broadcast.splatinsert = insertelement <vscale x 2 x i32> undef, i32 %in, i32 0
112    %broadcast.splat = shufflevector <vscale x 2 x i32> %broadcast.splatinsert, <vscale x 2 x i32> undef, <vscale x 2 x i32> zeroinitializer
113    %out = mul nsw <vscale x 2 x i32> %broadcast.splat, %b
114    ret <vscale x 2 x i32> %out
115}
116
117define <vscale x 4 x i32> @dupsext_v4i16_v4i32(i16 %src, <vscale x 4 x i32> %b) {
118; CHECK-LABEL: dupsext_v4i16_v4i32:
119; CHECK:       // %bb.0: // %entry
120; CHECK-NEXT:    sxth w8, w0
121; CHECK-NEXT:    ptrue p0.s
122; CHECK-NEXT:    mov z1.s, w8
123; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
124; CHECK-NEXT:    ret
125entry:
126    %in = sext i16 %src to i32
127    %broadcast.splatinsert = insertelement <vscale x 4 x i32> undef, i32 %in, i32 0
128    %broadcast.splat = shufflevector <vscale x 4 x i32> %broadcast.splatinsert, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
129    %out = mul nsw <vscale x 4 x i32> %broadcast.splat, %b
130    ret <vscale x 4 x i32> %out
131}
132
133define <vscale x 2 x i64> @dupsext_v2i16_v2i64(i16 %src, <vscale x 2 x i64> %b) {
134; CHECK-LABEL: dupsext_v2i16_v2i64:
135; CHECK:       // %bb.0: // %entry
136; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
137; CHECK-NEXT:    sxth x8, w0
138; CHECK-NEXT:    ptrue p0.d
139; CHECK-NEXT:    mov z1.d, x8
140; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
141; CHECK-NEXT:    ret
142entry:
143    %in = sext i16 %src to i64
144    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
145    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
146    %out = mul nsw <vscale x 2 x i64> %broadcast.splat, %b
147    ret <vscale x 2 x i64> %out
148}
149
150define <vscale x 2 x i64> @dupsext_v2i32_v2i64(i32 %src, <vscale x 2 x i64> %b) {
151; CHECK-LABEL: dupsext_v2i32_v2i64:
152; CHECK:       // %bb.0: // %entry
153; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
154; CHECK-NEXT:    sxtw x8, w0
155; CHECK-NEXT:    ptrue p0.d
156; CHECK-NEXT:    mov z1.d, x8
157; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
158; CHECK-NEXT:    ret
159entry:
160    %in = sext i32 %src to i64
161    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
162    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
163    %out = mul nsw <vscale x 2 x i64> %broadcast.splat, %b
164    ret <vscale x 2 x i64> %out
165}
166
167define <vscale x 2 x i16> @dupzext_v2i8_v2i16(i8 %src, <vscale x 2 x i16> %b) {
168; CHECK-LABEL: dupzext_v2i8_v2i16:
169; CHECK:       // %bb.0: // %entry
170; CHECK-NEXT:    and w8, w0, #0xff
171; CHECK-NEXT:    ptrue p0.d
172; CHECK-NEXT:    mov z1.d, x8
173; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
174; CHECK-NEXT:    ret
175entry:
176    %in = zext i8 %src to i16
177    %broadcast.splatinsert = insertelement <vscale x 2 x i16> undef, i16 %in, i16 0
178    %broadcast.splat = shufflevector <vscale x 2 x i16> %broadcast.splatinsert, <vscale x 2 x i16> undef, <vscale x 2 x i32> zeroinitializer
179    %out = mul nuw <vscale x 2 x i16> %broadcast.splat, %b
180    ret <vscale x 2 x i16> %out
181}
182
183define <vscale x 4 x i16> @dupzext_v4i8_v4i16(i8 %src, <vscale x 4 x i16> %b) {
184; CHECK-LABEL: dupzext_v4i8_v4i16:
185; CHECK:       // %bb.0: // %entry
186; CHECK-NEXT:    and w8, w0, #0xff
187; CHECK-NEXT:    ptrue p0.s
188; CHECK-NEXT:    mov z1.s, w8
189; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
190; CHECK-NEXT:    ret
191entry:
192    %in = zext i8 %src to i16
193    %broadcast.splatinsert = insertelement <vscale x 4 x i16> undef, i16 %in, i16 0
194    %broadcast.splat = shufflevector <vscale x 4 x i16> %broadcast.splatinsert, <vscale x 4 x i16> undef, <vscale x 4 x i32> zeroinitializer
195    %out = mul nuw <vscale x 4 x i16> %broadcast.splat, %b
196    ret <vscale x 4 x i16> %out
197}
198
199define <vscale x 8 x i16> @dupzext_v8i8_v8i16(i8 %src, <vscale x 8 x i16> %b) {
200; CHECK-LABEL: dupzext_v8i8_v8i16:
201; CHECK:       // %bb.0: // %entry
202; CHECK-NEXT:    and w8, w0, #0xff
203; CHECK-NEXT:    ptrue p0.h
204; CHECK-NEXT:    mov z1.h, w8
205; CHECK-NEXT:    mul z0.h, p0/m, z0.h, z1.h
206; CHECK-NEXT:    ret
207entry:
208    %in = zext i8 %src to i16
209    %broadcast.splatinsert = insertelement <vscale x 8 x i16> undef, i16 %in, i16 0
210    %broadcast.splat = shufflevector <vscale x 8 x i16> %broadcast.splatinsert, <vscale x 8 x i16> undef, <vscale x 8 x i32> zeroinitializer
211    %out = mul nuw <vscale x 8 x i16> %broadcast.splat, %b
212    ret <vscale x 8 x i16> %out
213}
214
215define <vscale x 2 x i32> @dupzext_v2i8_v2i32(i8 %src, <vscale x 2 x i32> %b) {
216; CHECK-LABEL: dupzext_v2i8_v2i32:
217; CHECK:       // %bb.0: // %entry
218; CHECK-NEXT:    and w8, w0, #0xff
219; CHECK-NEXT:    ptrue p0.d
220; CHECK-NEXT:    mov z1.d, x8
221; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
222; CHECK-NEXT:    ret
223entry:
224    %in = zext i8 %src to i32
225    %broadcast.splatinsert = insertelement <vscale x 2 x i32> undef, i32 %in, i32 0
226    %broadcast.splat = shufflevector <vscale x 2 x i32> %broadcast.splatinsert, <vscale x 2 x i32> undef, <vscale x 2 x i32> zeroinitializer
227    %out = mul nuw <vscale x 2 x i32> %broadcast.splat, %b
228    ret <vscale x 2 x i32> %out
229}
230
231define <vscale x 4 x i32> @dupzext_v4i8_v4i32(i8 %src, <vscale x 4 x i32> %b) {
232; CHECK-LABEL: dupzext_v4i8_v4i32:
233; CHECK:       // %bb.0: // %entry
234; CHECK-NEXT:    and w8, w0, #0xff
235; CHECK-NEXT:    ptrue p0.s
236; CHECK-NEXT:    mov z1.s, w8
237; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
238; CHECK-NEXT:    ret
239entry:
240    %in = zext i8 %src to i32
241    %broadcast.splatinsert = insertelement <vscale x 4 x i32> undef, i32 %in, i32 0
242    %broadcast.splat = shufflevector <vscale x 4 x i32> %broadcast.splatinsert, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
243    %out = mul nuw <vscale x 4 x i32> %broadcast.splat, %b
244    ret <vscale x 4 x i32> %out
245}
246
247define <vscale x 2 x i64> @dupzext_v2i8_v2i64(i8 %src, <vscale x 2 x i64> %b) {
248; CHECK-LABEL: dupzext_v2i8_v2i64:
249; CHECK:       // %bb.0: // %entry
250; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
251; CHECK-NEXT:    and x8, x0, #0xff
252; CHECK-NEXT:    ptrue p0.d
253; CHECK-NEXT:    mov z1.d, x8
254; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
255; CHECK-NEXT:    ret
256entry:
257    %in = zext i8 %src to i64
258    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
259    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
260    %out = mul nuw <vscale x 2 x i64> %broadcast.splat, %b
261    ret <vscale x 2 x i64> %out
262}
263
264define <vscale x 2 x i32> @dupzext_v2i16_v2i32(i16 %src, <vscale x 2 x i32> %b) {
265; CHECK-LABEL: dupzext_v2i16_v2i32:
266; CHECK:       // %bb.0: // %entry
267; CHECK-NEXT:    and w8, w0, #0xffff
268; CHECK-NEXT:    ptrue p0.d
269; CHECK-NEXT:    mov z1.d, x8
270; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
271; CHECK-NEXT:    ret
272entry:
273    %in = zext i16 %src to i32
274    %broadcast.splatinsert = insertelement <vscale x 2 x i32> undef, i32 %in, i32 0
275    %broadcast.splat = shufflevector <vscale x 2 x i32> %broadcast.splatinsert, <vscale x 2 x i32> undef, <vscale x 2 x i32> zeroinitializer
276    %out = mul nuw <vscale x 2 x i32> %broadcast.splat, %b
277    ret <vscale x 2 x i32> %out
278}
279
280define <vscale x 4 x i32> @dupzext_v4i16_v4i32(i16 %src, <vscale x 4 x i32> %b) {
281; CHECK-LABEL: dupzext_v4i16_v4i32:
282; CHECK:       // %bb.0: // %entry
283; CHECK-NEXT:    and w8, w0, #0xffff
284; CHECK-NEXT:    ptrue p0.s
285; CHECK-NEXT:    mov z1.s, w8
286; CHECK-NEXT:    mul z0.s, p0/m, z0.s, z1.s
287; CHECK-NEXT:    ret
288entry:
289    %in = zext i16 %src to i32
290    %broadcast.splatinsert = insertelement <vscale x 4 x i32> undef, i32 %in, i32 0
291    %broadcast.splat = shufflevector <vscale x 4 x i32> %broadcast.splatinsert, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
292    %out = mul nuw <vscale x 4 x i32> %broadcast.splat, %b
293    ret <vscale x 4 x i32> %out
294}
295
296define <vscale x 2 x i64> @dupzext_v2i16_v2i64(i16 %src, <vscale x 2 x i64> %b) {
297; CHECK-LABEL: dupzext_v2i16_v2i64:
298; CHECK:       // %bb.0: // %entry
299; CHECK-NEXT:    // kill: def $w0 killed $w0 def $x0
300; CHECK-NEXT:    and x8, x0, #0xffff
301; CHECK-NEXT:    ptrue p0.d
302; CHECK-NEXT:    mov z1.d, x8
303; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
304; CHECK-NEXT:    ret
305entry:
306    %in = zext i16 %src to i64
307    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
308    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
309    %out = mul nuw <vscale x 2 x i64> %broadcast.splat, %b
310    ret <vscale x 2 x i64> %out
311}
312
313define <vscale x 2 x i64> @dupzext_v2i32_v2i64(i32 %src, <vscale x 2 x i64> %b) {
314; CHECK-LABEL: dupzext_v2i32_v2i64:
315; CHECK:       // %bb.0: // %entry
316; CHECK-NEXT:    mov w8, w0
317; CHECK-NEXT:    ptrue p0.d
318; CHECK-NEXT:    mov z1.d, x8
319; CHECK-NEXT:    mul z0.d, p0/m, z0.d, z1.d
320; CHECK-NEXT:    ret
321entry:
322    %in = zext i32 %src to i64
323    %broadcast.splatinsert = insertelement <vscale x 2 x i64> undef, i64 %in, i64 0
324    %broadcast.splat = shufflevector <vscale x 2 x i64> %broadcast.splatinsert, <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer
325    %out = mul nuw <vscale x 2 x i64> %broadcast.splat, %b
326    ret <vscale x 2 x i64> %out
327}
328