1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes=loop-vectorize \
3; RUN: -force-ordered-reductions=true -hints-allow-reordering=false \
4; RUN: -force-tail-folding-style=data-with-evl \
5; RUN: -prefer-predicate-over-epilogue=predicate-dont-vectorize \
6; RUN: -mtriple=riscv64 -mattr=+v,+f -S < %s| FileCheck %s --check-prefix=IF-EVL
7
8; RUN: opt -passes=loop-vectorize \
9; RUN: -force-ordered-reductions=true -hints-allow-reordering=false \
10; RUN: -force-tail-folding-style=none \
11; RUN: -prefer-predicate-over-epilogue=predicate-dont-vectorize \
12; RUN: -mtriple=riscv64 -mattr=+v,+f -S < %s| FileCheck %s --check-prefix=NO-VP
13
14define float @fadd(ptr noalias nocapture readonly %a, i64 %n) {
15; IF-EVL-LABEL: @fadd(
16; IF-EVL-NEXT:  entry:
17; IF-EVL-NEXT:    [[TMP0:%.*]] = sub i64 -1, [[N:%.*]]
18; IF-EVL-NEXT:    [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
19; IF-EVL-NEXT:    [[TMP2:%.*]] = mul i64 [[TMP1]], 4
20; IF-EVL-NEXT:    [[TMP3:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]]
21; IF-EVL-NEXT:    br i1 [[TMP3]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
22; IF-EVL:       vector.ph:
23; IF-EVL-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
24; IF-EVL-NEXT:    [[TMP5:%.*]] = mul i64 [[TMP4]], 4
25; IF-EVL-NEXT:    [[TMP6:%.*]] = sub i64 [[TMP5]], 1
26; IF-EVL-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N]], [[TMP6]]
27; IF-EVL-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP5]]
28; IF-EVL-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
29; IF-EVL-NEXT:    [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
30; IF-EVL-NEXT:    [[TMP8:%.*]] = mul i64 [[TMP7]], 4
31; IF-EVL-NEXT:    br label [[VECTOR_BODY:%.*]]
32; IF-EVL:       vector.body:
33; IF-EVL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
34; IF-EVL-NEXT:    [[EVL_BASED_IV:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], [[VECTOR_BODY]] ]
35; IF-EVL-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]
36; IF-EVL-NEXT:    [[TMP9:%.*]] = sub i64 [[N]], [[EVL_BASED_IV]]
37; IF-EVL-NEXT:    [[TMP10:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[TMP9]], i32 4, i1 true)
38; IF-EVL-NEXT:    [[TMP11:%.*]] = add i64 [[EVL_BASED_IV]], 0
39; IF-EVL-NEXT:    [[TMP12:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP11]]
40; IF-EVL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds float, ptr [[TMP12]], i32 0
41; IF-EVL-NEXT:    [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[TMP13]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP10]])
42; IF-EVL-NEXT:    [[TMP14]] = call float @llvm.vp.reduce.fadd.nxv4f32(float [[VEC_PHI]], <vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP10]])
43; IF-EVL-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP10]] to i64
44; IF-EVL-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP15]], [[EVL_BASED_IV]]
45; IF-EVL-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP8]]
46; IF-EVL-NEXT:    [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
47; IF-EVL-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
48; IF-EVL:       middle.block:
49; IF-EVL-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
50; IF-EVL:       scalar.ph:
51; IF-EVL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
52; IF-EVL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP14]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
53; IF-EVL-NEXT:    br label [[FOR_BODY:%.*]]
54; IF-EVL:       for.body:
55; IF-EVL-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
56; IF-EVL-NEXT:    [[SUM_07:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
57; IF-EVL-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]]
58; IF-EVL-NEXT:    [[TMP17:%.*]] = load float, ptr [[ARRAYIDX]], align 4
59; IF-EVL-NEXT:    [[ADD]] = fadd float [[TMP17]], [[SUM_07]]
60; IF-EVL-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
61; IF-EVL-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
62; IF-EVL-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
63; IF-EVL:       for.end:
64; IF-EVL-NEXT:    [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]
65; IF-EVL-NEXT:    ret float [[ADD_LCSSA]]
66;
67; NO-VP-LABEL: @fadd(
68; NO-VP-NEXT:  entry:
69; NO-VP-NEXT:    br label [[FOR_BODY:%.*]]
70; NO-VP:       for.body:
71; NO-VP-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
72; NO-VP-NEXT:    [[SUM_07:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
73; NO-VP-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[IV]]
74; NO-VP-NEXT:    [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4
75; NO-VP-NEXT:    [[ADD]] = fadd float [[TMP0]], [[SUM_07]]
76; NO-VP-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
77; NO-VP-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]]
78; NO-VP-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
79; NO-VP:       for.end:
80; NO-VP-NEXT:    [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ]
81; NO-VP-NEXT:    ret float [[ADD_LCSSA]]
82;
83entry:
84  br label %for.body
85
86for.body:
87  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
88  %sum.07 = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ]
89  %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv
90  %0 = load float, ptr %arrayidx, align 4
91  %add = fadd float %0, %sum.07
92  %iv.next = add nuw nsw i64 %iv, 1
93  %exitcond.not = icmp eq i64 %iv.next, %n
94  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
95
96for.end:
97  ret float %add
98}
99
100!0 = distinct !{!0, !1}
101!1 = !{!"llvm.loop.vectorize.enable", i1 true}
102