xref: /llvm-project/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-peephole-vmerge-vops.ll (revision d8d131dfa99762ccdd2116661980b7d0493cd7b5) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+zvl256b | FileCheck %s
3
4declare <8 x i16> @llvm.vp.merge.v8i16(<8 x i1>, <8 x i16>, <8 x i16>, i32)
5declare <8 x i32> @llvm.vp.merge.v8i32(<8 x i1>, <8 x i32>, <8 x i32>, i32)
6declare <8 x float> @llvm.vp.merge.v8f32(<8 x i1>, <8 x float>, <8 x float>, i32)
7declare <8 x double> @llvm.vp.merge.v8f64(<8 x i1>, <8 x double>, <8 x double>, i32)
8
9; Test binary operator with vp.merge and vp.smax.
10declare <8 x i32> @llvm.vp.add.v8i32(<8 x i32>, <8 x i32>, <8 x i1>, i32)
11define <8 x i32> @vpmerge_vpadd(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, <8 x i1> %m, i32 zeroext %vl) {
12; CHECK-LABEL: vpmerge_vpadd:
13; CHECK:       # %bb.0:
14; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
15; CHECK-NEXT:    vadd.vv v8, v9, v10, v0.t
16; CHECK-NEXT:    ret
17  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
18  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
19  ret <8 x i32> %b
20}
21
22; Test glued node of merge should not be deleted.
23declare <8 x i1> @llvm.vp.icmp.v8i32(<8 x i32>, <8 x i32>, metadata, <8 x i1>, i32)
24define <8 x i32> @vpmerge_vpadd2(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
25; CHECK-LABEL: vpmerge_vpadd2:
26; CHECK:       # %bb.0:
27; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, ma
28; CHECK-NEXT:    vmseq.vv v0, v9, v10
29; CHECK-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
30; CHECK-NEXT:    vadd.vv v8, v9, v10, v0.t
31; CHECK-NEXT:    ret
32  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
33  %m = call <8 x i1> @llvm.vp.icmp.v8i32(<8 x i32> %x, <8 x i32> %y, metadata !"eq", <8 x i1> splat (i1 true), i32 %vl)
34  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
35  ret <8 x i32> %b
36}
37
38; Test vp.merge have all-ones mask.
39define <8 x i32> @vpmerge_vpadd3(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
40; CHECK-LABEL: vpmerge_vpadd3:
41; CHECK:       # %bb.0:
42; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, ma
43; CHECK-NEXT:    vadd.vv v8, v9, v10
44; CHECK-NEXT:    ret
45  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
46  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> splat (i1 true), <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
47  ret <8 x i32> %b
48}
49
50; Test float binary operator with vp.merge and vp.fadd.
51declare <8 x float> @llvm.vp.fadd.v8f32(<8 x float>, <8 x float>, <8 x i1>, i32)
52define <8 x float> @vpmerge_vpfadd(<8 x float> %passthru, <8 x float> %x, <8 x float> %y, <8 x i1> %m, i32 zeroext %vl) {
53; CHECK-LABEL: vpmerge_vpfadd:
54; CHECK:       # %bb.0:
55; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
56; CHECK-NEXT:    vfadd.vv v8, v9, v10, v0.t
57; CHECK-NEXT:    ret
58  %a = call <8 x float> @llvm.vp.fadd.v8f32(<8 x float> %x, <8 x float> %y, <8 x i1> splat (i1 true), i32 %vl)
59  %b = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
60  ret <8 x float> %b
61}
62
63; Test conversion by fptosi.
64declare <8 x i16> @llvm.vp.fptosi.v8i16.v8f32(<8 x float>, <8 x i1>, i32)
65define <8 x i16> @vpmerge_vpfptosi(<8 x i16> %passthru, <8 x float> %x, <8 x i1> %m, i32 zeroext %vl) {
66; CHECK-LABEL: vpmerge_vpfptosi:
67; CHECK:       # %bb.0:
68; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, mu
69; CHECK-NEXT:    vfncvt.rtz.x.f.w v8, v9, v0.t
70; CHECK-NEXT:    ret
71  %a = call <8 x i16> @llvm.vp.fptosi.v8i16.v8f32(<8 x float> %x, <8 x i1> splat (i1 true), i32 %vl)
72  %b = call <8 x i16> @llvm.vp.merge.v8i16(<8 x i1> %m, <8 x i16> %a, <8 x i16> %passthru, i32 %vl)
73  ret <8 x i16> %b
74}
75
76; Test conversion by sitofp.
77declare <8 x float> @llvm.vp.sitofp.v8f32.v8i64(<8 x i64>, <8 x i1>, i32)
78define <8 x float> @vpmerge_vpsitofp(<8 x float> %passthru, <8 x i64> %x, <8 x i1> %m, i32 zeroext %vl) {
79; CHECK-LABEL: vpmerge_vpsitofp:
80; CHECK:       # %bb.0:
81; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
82; CHECK-NEXT:    vfncvt.f.x.w v8, v10, v0.t
83; CHECK-NEXT:    ret
84  %a = call <8 x float> @llvm.vp.sitofp.v8f32.v8i64(<8 x i64> %x, <8 x i1> splat (i1 true), i32 %vl)
85  %b = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
86  ret <8 x float> %b
87}
88
89; Test integer extension by vp.zext.
90declare <8 x i32> @llvm.vp.zext.v8i32.v8i8(<8 x i8>, <8 x i1>, i32)
91define <8 x i32> @vpmerge_vpzext(<8 x i32> %passthru, <8 x i8> %x, <8 x i1> %m, i32 zeroext %vl) {
92; CHECK-LABEL: vpmerge_vpzext:
93; CHECK:       # %bb.0:
94; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
95; CHECK-NEXT:    vzext.vf4 v8, v9, v0.t
96; CHECK-NEXT:    ret
97  %a = call <8 x i32> @llvm.vp.zext.v8i32.v8i8(<8 x i8> %x, <8 x i1> splat (i1 true), i32 %vl)
98  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
99  ret <8 x i32> %b
100}
101
102; Test integer truncation by vp.trunc.
103declare <8 x i32> @llvm.vp.trunc.v8i32.v8i64(<8 x i64>, <8 x i1>, i32)
104define <8 x i32> @vpmerge_vptrunc(<8 x i32> %passthru, <8 x i64> %x, <8 x i1> %m, i32 zeroext %vl) {
105; CHECK-LABEL: vpmerge_vptrunc:
106; CHECK:       # %bb.0:
107; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
108; CHECK-NEXT:    vnsrl.wi v8, v10, 0, v0.t
109; CHECK-NEXT:    ret
110  %a = call <8 x i32> @llvm.vp.trunc.v8i32.v8i64(<8 x i64> %x, <8 x i1> splat (i1 true), i32 %vl)
111  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
112  ret <8 x i32> %b
113}
114
115; Test integer extension by vp.fpext.
116declare <8 x double> @llvm.vp.fpext.v8f64.v8f32(<8 x float>, <8 x i1>, i32)
117define <8 x double> @vpmerge_vpfpext(<8 x double> %passthru, <8 x float> %x, <8 x i1> %m, i32 zeroext %vl) {
118; CHECK-LABEL: vpmerge_vpfpext:
119; CHECK:       # %bb.0:
120; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
121; CHECK-NEXT:    vfwcvt.f.f.v v8, v10, v0.t
122; CHECK-NEXT:    ret
123  %a = call <8 x double> @llvm.vp.fpext.v8f64.v8f32(<8 x float> %x, <8 x i1> splat (i1 true), i32 %vl)
124  %b = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %m, <8 x double> %a, <8 x double> %passthru, i32 %vl)
125  ret <8 x double> %b
126}
127
128; Test integer truncation by vp.trunc.
129declare <8 x float> @llvm.vp.fptrunc.v8f32.v8f64(<8 x double>, <8 x i1>, i32)
130define <8 x float> @vpmerge_vpfptrunc(<8 x float> %passthru, <8 x double> %x, <8 x i1> %m, i32 zeroext %vl) {
131; CHECK-LABEL: vpmerge_vpfptrunc:
132; CHECK:       # %bb.0:
133; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
134; CHECK-NEXT:    vfncvt.f.f.w v8, v10, v0.t
135; CHECK-NEXT:    ret
136  %a = call <8 x float> @llvm.vp.fptrunc.v8f32.v8f64(<8 x double> %x, <8 x i1> splat (i1 true), i32 %vl)
137  %b = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
138  ret <8 x float> %b
139}
140
141; Test load operation by vp.load.
142declare <8 x i32> @llvm.vp.load.v8i32.p0(ptr, <8 x i1>, i32)
143
144define <8 x i32> @vpmerge_vpload(<8 x i32> %passthru, ptr %p, <8 x i1> %m, i32 zeroext %vl) {
145; CHECK-LABEL: vpmerge_vpload:
146; CHECK:       # %bb.0:
147; CHECK-NEXT:    vsetvli zero, a1, e32, m1, tu, mu
148; CHECK-NEXT:    vle32.v v8, (a0), v0.t
149; CHECK-NEXT:    ret
150  %a = call <8 x i32> @llvm.vp.load.v8i32.p0(ptr %p, <8 x i1> splat (i1 true), i32 %vl)
151  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
152  ret <8 x i32> %b
153}
154
155; Test result have chain and glued node.
156define <8 x i32> @vpmerge_vpload2(<8 x i32> %passthru, ptr %p, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
157; CHECK-LABEL: vpmerge_vpload2:
158; CHECK:       # %bb.0:
159; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, ma
160; CHECK-NEXT:    vmseq.vv v0, v9, v10
161; CHECK-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
162; CHECK-NEXT:    vle32.v v8, (a0), v0.t
163; CHECK-NEXT:    ret
164  %a = call <8 x i32> @llvm.vp.load.v8i32.p0(ptr %p, <8 x i1> splat (i1 true), i32 %vl)
165  %m = call <8 x i1> @llvm.vp.icmp.v8i32(<8 x i32> %x, <8 x i32> %y, metadata !"eq", <8 x i1> splat (i1 true), i32 %vl)
166  %b = call <8 x i32> @llvm.vp.merge.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
167  ret <8 x i32> %b
168}
169
170declare <8 x i16> @llvm.vp.select.v8i16(<8 x i1>, <8 x i16>, <8 x i16>, i32)
171declare <8 x i32> @llvm.vp.select.v8i32(<8 x i1>, <8 x i32>, <8 x i32>, i32)
172declare <8 x float> @llvm.vp.select.v8f32(<8 x i1>, <8 x float>, <8 x float>, i32)
173declare <8 x double> @llvm.vp.select.v8f64(<8 x i1>, <8 x double>, <8 x double>, i32)
174
175; Test binary operator with vp.select and vp.add.
176define <8 x i32> @vpselect_vpadd(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, <8 x i1> %m, i32 zeroext %vl) {
177; CHECK-LABEL: vpselect_vpadd:
178; CHECK:       # %bb.0:
179; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
180; CHECK-NEXT:    vadd.vv v8, v9, v10, v0.t
181; CHECK-NEXT:    ret
182  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
183  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
184  ret <8 x i32> %b
185}
186
187; Test glued node of select should not be deleted.
188define <8 x i32> @vpselect_vpadd2(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
189; CHECK-LABEL: vpselect_vpadd2:
190; CHECK:       # %bb.0:
191; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
192; CHECK-NEXT:    vmseq.vv v0, v9, v10
193; CHECK-NEXT:    vadd.vv v8, v9, v10, v0.t
194; CHECK-NEXT:    ret
195  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
196  %m = call <8 x i1> @llvm.vp.icmp.v8i32(<8 x i32> %x, <8 x i32> %y, metadata !"eq", <8 x i1> splat (i1 true), i32 %vl)
197  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
198  ret <8 x i32> %b
199}
200
201; Test vp.select have all-ones mask.
202define <8 x i32> @vpselect_vpadd3(<8 x i32> %passthru, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
203; CHECK-LABEL: vpselect_vpadd3:
204; CHECK:       # %bb.0:
205; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, ma
206; CHECK-NEXT:    vadd.vv v8, v9, v10
207; CHECK-NEXT:    ret
208  %a = call <8 x i32> @llvm.vp.add.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i1> splat (i1 true), i32 %vl)
209  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> splat (i1 true), <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
210  ret <8 x i32> %b
211}
212
213; Test float binary operator with vp.select and vp.fadd.
214define <8 x float> @vpselect_vpfadd(<8 x float> %passthru, <8 x float> %x, <8 x float> %y, <8 x i1> %m, i32 zeroext %vl) {
215; CHECK-LABEL: vpselect_vpfadd:
216; CHECK:       # %bb.0:
217; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
218; CHECK-NEXT:    vfadd.vv v8, v9, v10, v0.t
219; CHECK-NEXT:    ret
220  %a = call <8 x float> @llvm.vp.fadd.v8f32(<8 x float> %x, <8 x float> %y, <8 x i1> splat (i1 true), i32 %vl)
221  %b = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
222  ret <8 x float> %b
223}
224
225; Test conversion by fptosi.
226define <8 x i16> @vpselect_vpfptosi(<8 x i16> %passthru, <8 x float> %x, <8 x i1> %m, i32 zeroext %vl) {
227; CHECK-LABEL: vpselect_vpfptosi:
228; CHECK:       # %bb.0:
229; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
230; CHECK-NEXT:    vfncvt.rtz.x.f.w v8, v9, v0.t
231; CHECK-NEXT:    ret
232  %a = call <8 x i16> @llvm.vp.fptosi.v8i16.v8f32(<8 x float> %x, <8 x i1> splat (i1 true), i32 %vl)
233  %b = call <8 x i16> @llvm.vp.select.v8i16(<8 x i1> %m, <8 x i16> %a, <8 x i16> %passthru, i32 %vl)
234  ret <8 x i16> %b
235}
236
237; Test conversion by sitofp.
238define <8 x float> @vpselect_vpsitofp(<8 x float> %passthru, <8 x i64> %x, <8 x i1> %m, i32 zeroext %vl) {
239; CHECK-LABEL: vpselect_vpsitofp:
240; CHECK:       # %bb.0:
241; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
242; CHECK-NEXT:    vfncvt.f.x.w v8, v10, v0.t
243; CHECK-NEXT:    ret
244  %a = call <8 x float> @llvm.vp.sitofp.v8f32.v8i64(<8 x i64> %x, <8 x i1> splat (i1 true), i32 %vl)
245  %b = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
246  ret <8 x float> %b
247}
248
249; Test integer extension by vp.zext.
250define <8 x i32> @vpselect_vpzext(<8 x i32> %passthru, <8 x i8> %x, <8 x i1> %m, i32 zeroext %vl) {
251; CHECK-LABEL: vpselect_vpzext:
252; CHECK:       # %bb.0:
253; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
254; CHECK-NEXT:    vzext.vf4 v8, v9, v0.t
255; CHECK-NEXT:    ret
256  %a = call <8 x i32> @llvm.vp.zext.v8i32.v8i8(<8 x i8> %x, <8 x i1> splat (i1 true), i32 %vl)
257  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
258  ret <8 x i32> %b
259}
260
261; Test integer truncation by vp.trunc.
262define <8 x i32> @vpselect_vptrunc(<8 x i32> %passthru, <8 x i64> %x, <8 x i1> %m, i32 zeroext %vl) {
263; CHECK-LABEL: vpselect_vptrunc:
264; CHECK:       # %bb.0:
265; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
266; CHECK-NEXT:    vnsrl.wi v8, v10, 0, v0.t
267; CHECK-NEXT:    ret
268  %a = call <8 x i32> @llvm.vp.trunc.v8i32.v8i64(<8 x i64> %x, <8 x i1> splat (i1 true), i32 %vl)
269  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
270  ret <8 x i32> %b
271}
272
273; Test integer extension by vp.fpext.
274define <8 x double> @vpselect_vpfpext(<8 x double> %passthru, <8 x float> %x, <8 x i1> %m, i32 zeroext %vl) {
275; CHECK-LABEL: vpselect_vpfpext:
276; CHECK:       # %bb.0:
277; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
278; CHECK-NEXT:    vfwcvt.f.f.v v8, v10, v0.t
279; CHECK-NEXT:    ret
280  %a = call <8 x double> @llvm.vp.fpext.v8f64.v8f32(<8 x float> %x, <8 x i1> splat (i1 true), i32 %vl)
281  %b = call <8 x double> @llvm.vp.select.v8f64(<8 x i1> %m, <8 x double> %a, <8 x double> %passthru, i32 %vl)
282  ret <8 x double> %b
283}
284
285; Test integer truncation by vp.trunc.
286define <8 x float> @vpselect_vpfptrunc(<8 x float> %passthru, <8 x double> %x, <8 x i1> %m, i32 zeroext %vl) {
287; CHECK-LABEL: vpselect_vpfptrunc:
288; CHECK:       # %bb.0:
289; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
290; CHECK-NEXT:    vfncvt.f.f.w v8, v10, v0.t
291; CHECK-NEXT:    ret
292  %a = call <8 x float> @llvm.vp.fptrunc.v8f32.v8f64(<8 x double> %x, <8 x i1> splat (i1 true), i32 %vl)
293  %b = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %a, <8 x float> %passthru, i32 %vl)
294  ret <8 x float> %b
295}
296
297; Test load operation by vp.load.
298define <8 x i32> @vpselect_vpload(<8 x i32> %passthru, ptr %p, <8 x i1> %m, i32 zeroext %vl) {
299; CHECK-LABEL: vpselect_vpload:
300; CHECK:       # %bb.0:
301; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, mu
302; CHECK-NEXT:    vle32.v v8, (a0), v0.t
303; CHECK-NEXT:    ret
304  %a = call <8 x i32> @llvm.vp.load.v8i32.p0(ptr %p, <8 x i1> splat (i1 true), i32 %vl)
305  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
306  ret <8 x i32> %b
307}
308
309; Test result have chain and glued node.
310define <8 x i32> @vpselect_vpload2(<8 x i32> %passthru, ptr %p, <8 x i32> %x, <8 x i32> %y, i32 zeroext %vl) {
311; CHECK-LABEL: vpselect_vpload2:
312; CHECK:       # %bb.0:
313; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, mu
314; CHECK-NEXT:    vmseq.vv v0, v9, v10
315; CHECK-NEXT:    vle32.v v8, (a0), v0.t
316; CHECK-NEXT:    ret
317  %a = call <8 x i32> @llvm.vp.load.v8i32.p0(ptr %p, <8 x i1> splat (i1 true), i32 %vl)
318  %m = call <8 x i1> @llvm.vp.icmp.v8i32(<8 x i32> %x, <8 x i32> %y, metadata !"eq", <8 x i1> splat (i1 true), i32 %vl)
319  %b = call <8 x i32> @llvm.vp.select.v8i32(<8 x i1> %m, <8 x i32> %a, <8 x i32> %passthru, i32 %vl)
320  ret <8 x i32> %b
321}
322