xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/AArch64/insertelement.ll (revision f61c9b75690e8c374a99332b9b837bf8f77bfc90)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=slp-vectorizer -S | FileCheck %s
3
4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
5target triple = "aarch64-unknown-linux-gnu"
6
7define <2 x float> @insertelement-fixed-vector() {
8; CHECK-LABEL: @insertelement-fixed-vector(
9; CHECK-NEXT:    [[TMP1:%.*]] = call fast <2 x float> @llvm.fabs.v2f32(<2 x float> undef)
10; CHECK-NEXT:    ret <2 x float> [[TMP1]]
11;
12  %f0 = tail call fast float @llvm.fabs.f32(float undef)
13  %f1 = tail call fast float @llvm.fabs.f32(float undef)
14  %i0 = insertelement <2 x float> undef, float %f0, i32 0
15  %i1 = insertelement <2 x float> %i0, float %f1, i32 1
16  ret <2 x float> %i1
17}
18
19; TODO: llvm.fabs could be optimized in vector form. It's legal to extract
20; elements from fixed-length vector and insert into scalable vector.
21define <vscale x 2 x float> @insertelement-scalable-vector() {
22; CHECK-LABEL: @insertelement-scalable-vector(
23; CHECK-NEXT:    [[F0:%.*]] = tail call fast float @llvm.fabs.f32(float undef)
24; CHECK-NEXT:    [[F1:%.*]] = tail call fast float @llvm.fabs.f32(float undef)
25; CHECK-NEXT:    [[I0:%.*]] = insertelement <vscale x 2 x float> undef, float [[F0]], i32 0
26; CHECK-NEXT:    [[I1:%.*]] = insertelement <vscale x 2 x float> [[I0]], float [[F1]], i32 1
27; CHECK-NEXT:    ret <vscale x 2 x float> [[I1]]
28;
29  %f0 = tail call fast float @llvm.fabs.f32(float undef)
30  %f1 = tail call fast float @llvm.fabs.f32(float undef)
31  %i0 = insertelement <vscale x 2 x float> undef, float %f0, i32 0
32  %i1 = insertelement <vscale x 2 x float> %i0, float %f1, i32 1
33  ret <vscale x 2 x float> %i1
34}
35
36; Function Attrs: nounwind readnone speculatable willreturn
37declare float @llvm.fabs.f32(float)
38
39
40define <4 x float> @insertelement_poison_lanes(ptr %0) {
41; CHECK-LABEL: @insertelement_poison_lanes(
42; CHECK-NEXT:    [[INS_1:%.*]] = insertelement <4 x float> zeroinitializer, float poison, i64 0
43; CHECK-NEXT:    [[INS_2:%.*]] = insertelement <4 x float> [[INS_1]], float 0.000000e+00, i64 0
44; CHECK-NEXT:    [[GEP_1:%.*]] = getelementptr double, ptr [[TMP0:%.*]], i64 1
45; CHECK-NEXT:    store <2 x double> <double 0.000000e+00, double 1.000000e+00>, ptr [[GEP_1]], align 8
46; CHECK-NEXT:    ret <4 x float> [[INS_2]]
47;
48  %trunc.1 = fptrunc double 0.000000e+00 to float
49  %trunc.2 = fptrunc double 1.000000e+00 to float
50  %ins.1 = insertelement <4 x float> zeroinitializer, float poison, i64 0
51  %ins.2 = insertelement <4 x float> %ins.1, float %trunc.1, i64 0
52  %ext.1 = fpext float %trunc.1 to double
53  %gep.1  = getelementptr double, ptr %0, i64 1
54  store double %ext.1, ptr %gep.1, align 8
55  %ext.2 = fpext float %trunc.2 to double
56  %gep.2 = getelementptr double, ptr %0, i64 2
57  store double %ext.2, ptr %gep.2, align 8
58  ret <4 x float> %ins.2
59}
60