1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -loop-vectorize -mtriple aarch64-unknown-linux-gnu -mattr=+sve -force-vector-interleave=1 -force-vector-width=4 -prefer-inloop-reductions -S | FileCheck %s
3
4define float @cond_fadd(float* noalias nocapture readonly %a, float* noalias nocapture readonly %cond, i64 %N){
5; CHECK-LABEL: @cond_fadd(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
8; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 4
9; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]
10; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
11; CHECK:       vector.ph:
12; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
13; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 4
14; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
15; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
16; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
17; CHECK:       vector.body:
18; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
19; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi float [ 1.000000e+00, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]
20; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 0
21; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[COND:%.*]], i64 [[TMP4]]
22; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP5]], i32 0
23; CHECK-NEXT:    [[TMP7:%.*]] = bitcast float* [[TMP6]] to <vscale x 4 x float>*
24; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, <vscale x 4 x float>* [[TMP7]], align 4
25; CHECK-NEXT:    [[TMP8:%.*]] = fcmp une <vscale x 4 x float> [[WIDE_LOAD]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 2.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
26; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[TMP4]]
27; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr float, float* [[TMP9]], i32 0
28; CHECK-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <vscale x 4 x float>*
29; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP11]], i32 4, <vscale x 4 x i1> [[TMP8]], <vscale x 4 x float> poison)
30; CHECK-NEXT:    [[TMP12:%.*]] = select fast <vscale x 4 x i1> [[TMP8]], <vscale x 4 x float> [[WIDE_MASKED_LOAD]], <vscale x 4 x float> zeroinitializer
31; CHECK-NEXT:    [[TMP13:%.*]] = call fast float @llvm.vector.reduce.fadd.nxv4f32(float -0.000000e+00, <vscale x 4 x float> [[TMP12]])
32; CHECK-NEXT:    [[TMP14]] = fadd fast float [[TMP13]], [[VEC_PHI]]
33; CHECK-NEXT:    [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
34; CHECK-NEXT:    [[TMP16:%.*]] = mul i64 [[TMP15]], 4
35; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP16]]
36; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
37; CHECK-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
38; CHECK:       middle.block:
39; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
40; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
41; CHECK:       scalar.ph:
42; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
43; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ 1.000000e+00, [[ENTRY]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]
44; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
45; CHECK:       for.body:
46; CHECK-NEXT:    [[INDVARS:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_NEXT:%.*]], [[FOR_INC:%.*]] ]
47; CHECK-NEXT:    [[RDX:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[RES:%.*]], [[FOR_INC]] ]
48; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[COND]], i64 [[INDVARS]]
49; CHECK-NEXT:    [[TMP18:%.*]] = load float, float* [[ARRAYIDX]], align 4
50; CHECK-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP18]], 2.000000e+00
51; CHECK-NEXT:    br i1 [[TOBOOL]], label [[IF_THEN:%.*]], label [[FOR_INC]]
52; CHECK:       if.then:
53; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS]]
54; CHECK-NEXT:    [[TMP19:%.*]] = load float, float* [[ARRAYIDX2]], align 4
55; CHECK-NEXT:    [[FADD:%.*]] = fadd fast float [[RDX]], [[TMP19]]
56; CHECK-NEXT:    br label [[FOR_INC]]
57; CHECK:       for.inc:
58; CHECK-NEXT:    [[RES]] = phi float [ [[FADD]], [[IF_THEN]] ], [ [[RDX]], [[FOR_BODY]] ]
59; CHECK-NEXT:    [[INDVARS_NEXT]] = add nuw nsw i64 [[INDVARS]], 1
60; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_NEXT]], [[N]]
61; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
62; CHECK:       for.end:
63; CHECK-NEXT:    [[RES_LCSSA:%.*]] = phi float [ [[RES]], [[FOR_INC]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]
64; CHECK-NEXT:    ret float [[RES_LCSSA]]
65;
66entry:
67  br label %for.body
68
69for.body:
70  %indvars = phi i64 [ 0, %entry ], [ %indvars.next, %for.inc ]
71  %rdx = phi float [ 1.000000e+00, %entry ], [ %res, %for.inc ]
72  %arrayidx = getelementptr inbounds float, float* %cond, i64 %indvars
73  %0 = load float, float* %arrayidx
74  %tobool = fcmp une float %0, 2.000000e+00
75  br i1 %tobool, label %if.then, label %for.inc
76
77if.then:
78  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars
79  %1 = load float, float* %arrayidx2
80  %fadd = fadd fast float %rdx, %1
81  br label %for.inc
82
83for.inc:
84  %res = phi float [ %fadd, %if.then ], [ %rdx, %for.body ]
85  %indvars.next = add nuw nsw i64 %indvars, 1
86  %exitcond.not = icmp eq i64 %indvars.next, %N
87  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
88
89for.end:
90  ret float %res
91}
92
93define float @cond_cmp_sel(float* noalias %a, float* noalias %cond, i64 %N) {
94; CHECK-LABEL: @cond_cmp_sel(
95; CHECK-NEXT:  entry:
96; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
97; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 4
98; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]
99; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
100; CHECK:       vector.ph:
101; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
102; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 4
103; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
104; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
105; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
106; CHECK:       vector.body:
107; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
108; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi float [ 1.000000e+00, [[VECTOR_PH]] ], [ [[RDX_MINMAX_SELECT:%.*]], [[VECTOR_BODY]] ]
109; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 0
110; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[COND:%.*]], i64 [[TMP4]]
111; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP5]], i32 0
112; CHECK-NEXT:    [[TMP7:%.*]] = bitcast float* [[TMP6]] to <vscale x 4 x float>*
113; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, <vscale x 4 x float>* [[TMP7]], align 4
114; CHECK-NEXT:    [[TMP8:%.*]] = fcmp une <vscale x 4 x float> [[WIDE_LOAD]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
115; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[TMP4]]
116; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr float, float* [[TMP9]], i32 0
117; CHECK-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <vscale x 4 x float>*
118; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0nxv4f32(<vscale x 4 x float>* [[TMP11]], i32 4, <vscale x 4 x i1> [[TMP8]], <vscale x 4 x float> poison)
119; CHECK-NEXT:    [[TMP12:%.*]] = select fast <vscale x 4 x i1> [[TMP8]], <vscale x 4 x float> [[WIDE_MASKED_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 0xFFF0000000000000, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
120; CHECK-NEXT:    [[TMP13:%.*]] = call fast float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float> [[TMP12]])
121; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast olt float [[TMP13]], [[VEC_PHI]]
122; CHECK-NEXT:    [[RDX_MINMAX_SELECT]] = select fast i1 [[RDX_MINMAX_CMP]], float [[TMP13]], float [[VEC_PHI]]
123; CHECK-NEXT:    [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
124; CHECK-NEXT:    [[TMP15:%.*]] = mul i64 [[TMP14]], 4
125; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP15]]
126; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
127; CHECK-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
128; CHECK:       middle.block:
129; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
130; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
131; CHECK:       scalar.ph:
132; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
133; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ 1.000000e+00, [[ENTRY]] ], [ [[RDX_MINMAX_SELECT]], [[MIDDLE_BLOCK]] ]
134; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
135; CHECK:       for.body:
136; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_INC:%.*]] ]
137; CHECK-NEXT:    [[RDX:%.*]] = phi float [ [[RES:%.*]], [[FOR_INC]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
138; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[COND]], i64 [[IV]]
139; CHECK-NEXT:    [[TMP17:%.*]] = load float, float* [[ARRAYIDX]], align 4
140; CHECK-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP17]], 3.000000e+00
141; CHECK-NEXT:    br i1 [[TOBOOL]], label [[IF_THEN:%.*]], label [[FOR_INC]]
142; CHECK:       if.then:
143; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IV]]
144; CHECK-NEXT:    [[TMP18:%.*]] = load float, float* [[ARRAYIDX2]], align 4
145; CHECK-NEXT:    [[FCMP:%.*]] = fcmp fast olt float [[RDX]], [[TMP18]]
146; CHECK-NEXT:    [[FSEL:%.*]] = select fast i1 [[FCMP]], float [[RDX]], float [[TMP18]]
147; CHECK-NEXT:    br label [[FOR_INC]]
148; CHECK:       for.inc:
149; CHECK-NEXT:    [[RES]] = phi float [ [[RDX]], [[FOR_BODY]] ], [ [[FSEL]], [[IF_THEN]] ]
150; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
151; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
152; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
153; CHECK:       for.end:
154; CHECK-NEXT:    [[RES_LCSSA:%.*]] = phi float [ [[RES]], [[FOR_INC]] ], [ [[RDX_MINMAX_SELECT]], [[MIDDLE_BLOCK]] ]
155; CHECK-NEXT:    ret float [[RES_LCSSA]]
156;
157entry:
158  br label %for.body
159
160for.body:
161  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.inc ]
162  %rdx = phi float [ %res, %for.inc ], [ 1.000000e+00, %entry ]
163  %arrayidx = getelementptr inbounds float, float* %cond, i64 %iv
164  %0 = load float, float* %arrayidx
165  %tobool = fcmp une float %0, 3.000000e+00
166  br i1 %tobool, label %if.then, label %for.inc
167
168if.then:
169  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %iv
170  %1 = load float, float* %arrayidx2
171  %fcmp = fcmp fast olt float %rdx, %1
172  %fsel = select fast i1 %fcmp, float %rdx, float %1
173  br label %for.inc
174
175for.inc:
176  %res = phi float [ %rdx, %for.body ], [ %fsel, %if.then ]
177  %iv.next = add i64 %iv, 1
178  %exitcond.not = icmp eq i64 %iv.next, %N
179  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
180
181for.end:
182  ret float %res
183}
184
185!0 = distinct !{!0, !1}
186!1 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
187