1 ; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S \
2 ; RUN: -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s --check-prefix=CHECK-VF4IC1
3 ; RUN: opt -passes=loop-vectorize -force-vector-interleave=4 -force-vector-width=4 -S \
4 ; RUN: -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s --check-prefix=CHECK-VF4IC4
6 target triple = "aarch64-linux-gnu"
8 define i32 @select_const_i32_from_icmp(ptr nocapture readonly %v, i64 %n) #0 {
9 ; CHECK-VF4IC1-LABEL: @select_const_i32_from_icmp
10 ; CHECK-VF4IC1: vector.body:
11 ; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
12 ; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x i32>
13 ; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp eq <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
14 ; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
15 ; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
16 ; CHECK-VF4IC1: middle.block:
17 ; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
18 ; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
19 ; CHECK-VF4IC1-NEXT: {{.*}} = select i1 [[FR]], i32 7, i32 3
20 ; CHECK-VF4IC1: %cmp.n = icmp eq i64 %n, %n.vec
22 ; CHECK-VF4IC4-LABEL: @select_const_i32_from_icmp
23 ; CHECK-VF4IC4: vector.body:
24 ; CHECK-VF4IC4: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL1:%.*]], %vector.body ]
25 ; CHECK-VF4IC4: [[VEC_PHI2:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL2:%.*]], %vector.body ]
26 ; CHECK-VF4IC4: [[VEC_PHI3:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL3:%.*]], %vector.body ]
27 ; CHECK-VF4IC4: [[VEC_PHI4:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL4:%.*]], %vector.body ]
28 ; CHECK-VF4IC4: [[VEC_ICMP1:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
29 ; CHECK-VF4IC4-NEXT: [[VEC_ICMP2:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
30 ; CHECK-VF4IC4-NEXT: [[VEC_ICMP3:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
31 ; CHECK-VF4IC4-NEXT: [[VEC_ICMP4:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
32 ; CHECK-VF4IC4-NEXT: [[NOT1:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP1]], splat (i1 true)
33 ; CHECK-VF4IC4-NEXT: [[NOT2:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP2]], splat (i1 true)
34 ; CHECK-VF4IC4-NEXT: [[NOT3:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP3]], splat (i1 true)
35 ; CHECK-VF4IC4-NEXT: [[NOT4:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP4]], splat (i1 true)
36 ; CHECK-VF4IC4-NEXT: [[VEC_SEL1:%.*]] = or <vscale x 4 x i1> [[VEC_PHI1]], [[NOT1]]
37 ; CHECK-VF4IC4-NEXT: [[VEC_SEL2:%.*]] = or <vscale x 4 x i1> [[VEC_PHI2]], [[NOT2]]
38 ; CHECK-VF4IC4-NEXT: [[VEC_SEL3:%.*]] = or <vscale x 4 x i1> [[VEC_PHI3]], [[NOT3]]
39 ; CHECK-VF4IC4-NEXT: [[VEC_SEL4:%.*]] = or <vscale x 4 x i1> [[VEC_PHI4]], [[NOT4]]
40 ; CHECK-VF4IC4: middle.block:
41 ; CHECK-VF4IC4-NEXT: [[OR1:%.*]] = or <vscale x 4 x i1> [[VEC_SEL2]], [[VEC_SEL1]]
42 ; CHECK-VF4IC4-NEXT: [[OR2:%.*]] = or <vscale x 4 x i1> [[VEC_SEL3]], [[OR1]]
43 ; CHECK-VF4IC4-NEXT: [[OR3:%.*]] = or <vscale x 4 x i1> [[VEC_SEL4]], [[OR2]]
44 ; CHECK-VF4IC4-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[OR3]])
45 ; CHECK-VF4IC4-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
46 ; CHECK-VF4IC4-NEXT: {{.*}} = select i1 [[FR]], i32 7, i32 3
47 ; CHECK-VF4IC4-NEXT: %cmp.n = icmp eq i64 %n, %n.vec
51 for.body: ; preds = %entry, %for.body
52 %0 = phi i64 [ 0, %entry ], [ %6, %for.body ]
53 %1 = phi i32 [ 3, %entry ], [ %5, %for.body ]
54 %2 = getelementptr inbounds i32, ptr %v, i64 %0
55 %3 = load i32, ptr %2, align 4
56 %4 = icmp eq i32 %3, 3
57 %5 = select i1 %4, i32 %1, i32 7
58 %6 = add nuw nsw i64 %0, 1
59 %7 = icmp eq i64 %6, %n
60 br i1 %7, label %exit, label %for.body, !llvm.loop !0
62 exit: ; preds = %for.body
66 define i32 @select_i32_from_icmp(ptr nocapture readonly %v, i32 %a, i32 %b, i64 %n) #0 {
67 ; CHECK-VF4IC1-LABEL: @select_i32_from_icmp
68 ; CHECK-VF4IC1: vector.ph:
69 ; CHECK-VF4IC1-NOT: shufflevector <vscale x 4 x i32>
70 ; CHECK-VF4IC1-NOT: shufflevector <vscale x 4 x i32>
71 ; CHECK-VF4IC1: vector.body:
72 ; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
73 ; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x i32>
74 ; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp eq <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
75 ; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
76 ; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
77 ; CHECK-VF4IC1: middle.block:
78 ; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
79 ; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
80 ; CHECK-VF4IC1-NEXT: {{.*}} = select i1 [[FR]], i32 %b, i32 %a
81 ; CHECK-VF4IC1: %cmp.n = icmp eq i64 %n, %n.vec
83 ; CHECK-VF4IC4-LABEL: @select_i32_from_icmp
84 ; CHECK-VF4IC4: vector.body:
88 for.body: ; preds = %entry, %for.body
89 %0 = phi i64 [ 0, %entry ], [ %6, %for.body ]
90 %1 = phi i32 [ %a, %entry ], [ %5, %for.body ]
91 %2 = getelementptr inbounds i32, ptr %v, i64 %0
92 %3 = load i32, ptr %2, align 4
93 %4 = icmp eq i32 %3, 3
94 %5 = select i1 %4, i32 %1, i32 %b
95 %6 = add nuw nsw i64 %0, 1
96 %7 = icmp eq i64 %6, %n
97 br i1 %7, label %exit, label %for.body, !llvm.loop !0
99 exit: ; preds = %for.body
103 define i32 @select_const_i32_from_fcmp(ptr nocapture readonly %v, i64 %n) #0 {
104 ; CHECK-VF4IC1-LABEL: @select_const_i32_from_fcmp
105 ; CHECK-VF4IC1: vector.body:
106 ; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
107 ; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x float>
108 ; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = fcmp fast ueq <vscale x 4 x float> [[VEC_LOAD]], splat (float 3.000000e+00)
109 ; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
110 ; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
111 ; CHECK-VF4IC1: middle.block:
112 ; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
113 ; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
114 ; CHECK-VF4IC1-NEXT: {{.*}} = select i1 [[FR]], i32 1, i32 2
115 ; CHECK-VF4IC1: %cmp.n = icmp eq i64 %n, %n.vec
117 ; CHECK-VF4IC4-LABEL: @select_const_i32_from_fcmp
118 ; CHECK-VF4IC4: vector.body:
122 for.body: ; preds = %entry, %for.body
123 %0 = phi i64 [ 0, %entry ], [ %6, %for.body ]
124 %1 = phi i32 [ 2, %entry ], [ %5, %for.body ]
125 %2 = getelementptr inbounds float, ptr %v, i64 %0
126 %3 = load float, ptr %2, align 4
127 %4 = fcmp fast ueq float %3, 3.0
128 %5 = select i1 %4, i32 %1, i32 1
129 %6 = add nuw nsw i64 %0, 1
130 %7 = icmp eq i64 %6, %n
131 br i1 %7, label %exit, label %for.body, !llvm.loop !0
133 exit: ; preds = %for.body
137 define float @select_const_f32_from_icmp(ptr nocapture readonly %v, i64 %n) #0 {
138 ; CHECK-VF4IC1-LABEL: @select_const_f32_from_icmp
139 ; CHECK-VF4IC1-NOT: vector.body
140 ; CHECK-VF4IC4-LABEL: @select_const_f32_from_icmp
141 ; CHECK-VF4IC4-NOT: vector.body
145 for.body: ; preds = %entry, %for.body
146 %0 = phi i64 [ 0, %entry ], [ %6, %for.body ]
147 %1 = phi fast float [ 3.0, %entry ], [ %5, %for.body ]
148 %2 = getelementptr inbounds i32, ptr %v, i64 %0
149 %3 = load i32, ptr %2, align 4
150 %4 = icmp eq i32 %3, 3
151 %5 = select fast i1 %4, float %1, float 7.0
152 %6 = add nuw nsw i64 %0, 1
153 %7 = icmp eq i64 %6, %n
154 br i1 %7, label %exit, label %for.body, !llvm.loop !0
156 exit: ; preds = %for.body
160 define i32 @pred_select_const_i32_from_icmp(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i64 %n) #0 {
161 ; CHECK-VF4IC1-LABEL: @pred_select_const_i32_from_icmp
162 ; CHECK-VF4IC1: vector.body:
163 ; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
164 ; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x i32>
165 ; CHECK-VF4IC1: [[MASK:%.*]] = icmp sgt <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 35)
166 ; CHECK-VF4IC1: [[MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr {{%.*}}, i32 4, <vscale x 4 x i1> [[MASK]], <vscale x 4 x i32> poison)
167 ; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp eq <vscale x 4 x i32> [[MASKED_LOAD]], splat (i32 2)
168 ; CHECK-VF4IC1-NEXT: [[VEC_SEL_TMP:%.*]] = or <vscale x 4 x i1> [[VEC_PHI]], [[VEC_ICMP]]
169 ; CHECK-VF4IC1: [[VEC_SEL:%.*]] = select <vscale x 4 x i1> [[MASK]], <vscale x 4 x i1> [[VEC_SEL_TMP]], <vscale x 4 x i1> [[VEC_PHI]]
170 ; CHECK-VF4IC1: middle.block:
171 ; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
172 ; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
173 ; CHECK-VF4IC1-NEXT: {{.*}} = select i1 [[FR]], i32 1, i32 0
174 ; CHECK-VF4IC1: %cmp.n = icmp eq i64 %n, %n.vec
176 ; CHECK-VF4IC4-LABEL: @pred_select_const_i32_from_icmp
177 ; CHECK-VF4IC4: vector.body:
181 for.body: ; preds = %entry, %for.inc
182 %i.013 = phi i64 [ %inc, %for.inc ], [ 0, %entry ]
183 %r.012 = phi i32 [ %r.1, %for.inc ], [ 0, %entry ]
184 %arrayidx = getelementptr inbounds i32, ptr %src1, i64 %i.013
185 %0 = load i32, ptr %arrayidx, align 4
186 %cmp1 = icmp sgt i32 %0, 35
187 br i1 %cmp1, label %if.then, label %for.inc
189 if.then: ; preds = %for.body
190 %arrayidx2 = getelementptr inbounds i32, ptr %src2, i64 %i.013
191 %1 = load i32, ptr %arrayidx2, align 4
192 %cmp3 = icmp eq i32 %1, 2
193 %spec.select = select i1 %cmp3, i32 1, i32 %r.012
196 for.inc: ; preds = %if.then, %for.body
197 %r.1 = phi i32 [ %r.012, %for.body ], [ %spec.select, %if.then ]
198 %inc = add nuw nsw i64 %i.013, 1
199 %exitcond.not = icmp eq i64 %inc, %n
200 br i1 %exitcond.not, label %for.end.loopexit, label %for.body, !llvm.loop !0
202 for.end.loopexit: ; preds = %for.inc
203 %r.1.lcssa = phi i32 [ %r.1, %for.inc ]
208 attributes #0 = { "target-features"="+sve" }
210 !0 = distinct !{!0, !1}
211 !1 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}