xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/pr19657.ll (revision 580210a0c938531ef9fd79f9ffedb93eeb2e66c2) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=slp-vectorizer -S -mcpu=corei7-avx | FileCheck %s --check-prefixes=ANY,AVX
3; RUN: opt < %s -passes=slp-vectorizer -slp-max-reg-size=128 -S -mcpu=corei7-avx | FileCheck %s --check-prefixes=ANY,MAX128
4
5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6target triple = "x86_64-unknown-linux-gnu"
7
8define void @store_chains(ptr %x) {
9; AVX-LABEL: @store_chains(
10; AVX-NEXT:    [[TMP2:%.*]] = load <4 x double>, ptr [[X:%.*]], align 8
11; AVX-NEXT:    [[TMP3:%.*]] = fadd <4 x double> [[TMP2]], [[TMP2]]
12; AVX-NEXT:    [[TMP4:%.*]] = fadd <4 x double> [[TMP3]], [[TMP2]]
13; AVX-NEXT:    store <4 x double> [[TMP4]], ptr [[X]], align 8
14; AVX-NEXT:    ret void
15;
16; MAX128-LABEL: @store_chains(
17; MAX128-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[X:%.*]], align 8
18; MAX128-NEXT:    [[TMP3:%.*]] = fadd <2 x double> [[TMP2]], [[TMP2]]
19; MAX128-NEXT:    [[TMP4:%.*]] = fadd <2 x double> [[TMP3]], [[TMP2]]
20; MAX128-NEXT:    store <2 x double> [[TMP4]], ptr [[X]], align 8
21; MAX128-NEXT:    [[TMP6:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
22; MAX128-NEXT:    [[TMP8:%.*]] = load <2 x double>, ptr [[TMP6]], align 8
23; MAX128-NEXT:    [[TMP9:%.*]] = fadd <2 x double> [[TMP8]], [[TMP8]]
24; MAX128-NEXT:    [[TMP10:%.*]] = fadd <2 x double> [[TMP9]], [[TMP8]]
25; MAX128-NEXT:    store <2 x double> [[TMP10]], ptr [[TMP6]], align 8
26; MAX128-NEXT:    ret void
27;
28  %1 = load double, ptr %x, align 8
29  %2 = fadd double %1, %1
30  %3 = fadd double %2, %1
31  store double %3, ptr %x, align 8
32  %4 = getelementptr inbounds double, ptr %x, i64 1
33  %5 = load double, ptr %4, align 8
34  %6 = fadd double %5, %5
35  %7 = fadd double %6, %5
36  store double %7, ptr %4, align 8
37  %8 = getelementptr inbounds double, ptr %x, i64 2
38  %9 = load double, ptr %8, align 8
39  %10 = fadd double %9, %9
40  %11 = fadd double %10, %9
41  store double %11, ptr %8, align 8
42  %12 = getelementptr inbounds double, ptr %x, i64 3
43  %13 = load double, ptr %12, align 8
44  %14 = fadd double %13, %13
45  %15 = fadd double %14, %13
46  store double %15, ptr %12, align 8
47  ret void
48}
49
50define void @store_chains_prefer_width_attr(ptr %x) #0 {
51; ANY-LABEL: @store_chains_prefer_width_attr(
52; ANY-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[X:%.*]], align 8
53; ANY-NEXT:    [[TMP3:%.*]] = fadd <2 x double> [[TMP2]], [[TMP2]]
54; ANY-NEXT:    [[TMP4:%.*]] = fadd <2 x double> [[TMP3]], [[TMP2]]
55; ANY-NEXT:    store <2 x double> [[TMP4]], ptr [[X]], align 8
56; ANY-NEXT:    [[TMP6:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
57; ANY-NEXT:    [[TMP8:%.*]] = load <2 x double>, ptr [[TMP6]], align 8
58; ANY-NEXT:    [[TMP9:%.*]] = fadd <2 x double> [[TMP8]], [[TMP8]]
59; ANY-NEXT:    [[TMP10:%.*]] = fadd <2 x double> [[TMP9]], [[TMP8]]
60; ANY-NEXT:    store <2 x double> [[TMP10]], ptr [[TMP6]], align 8
61; ANY-NEXT:    ret void
62;
63  %1 = load double, ptr %x, align 8
64  %2 = fadd double %1, %1
65  %3 = fadd double %2, %1
66  store double %3, ptr %x, align 8
67  %4 = getelementptr inbounds double, ptr %x, i64 1
68  %5 = load double, ptr %4, align 8
69  %6 = fadd double %5, %5
70  %7 = fadd double %6, %5
71  store double %7, ptr %4, align 8
72  %8 = getelementptr inbounds double, ptr %x, i64 2
73  %9 = load double, ptr %8, align 8
74  %10 = fadd double %9, %9
75  %11 = fadd double %10, %9
76  store double %11, ptr %8, align 8
77  %12 = getelementptr inbounds double, ptr %x, i64 3
78  %13 = load double, ptr %12, align 8
79  %14 = fadd double %13, %13
80  %15 = fadd double %14, %13
81  store double %15, ptr %12, align 8
82  ret void
83}
84
85attributes #0 = { "prefer-vector-width"="128" }
86