• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -relocation-model=pic -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr9 < %s | FileCheck -check-prefix=CHECK-P9 %s
3; RUN: llc -relocation-model=pic -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 < %s | FileCheck -check-prefix=CHECK-P8 %s
4
5; As the constant could be represented as float, a float is
6; loaded from constant pool.
7define double @doubleConstant1() {
8; CHECK-P9-LABEL: doubleConstant1:
9; CHECK-P9:       # %bb.0:
10; CHECK-P9-NEXT:    addis 3, 2, .LCPI0_0@toc@ha
11; CHECK-P9-NEXT:    lfs 1, .LCPI0_0@toc@l(3)
12; CHECK-P9-NEXT:    blr
13;
14; CHECK-P8-LABEL: doubleConstant1:
15; CHECK-P8:       # %bb.0:
16; CHECK-P8-NEXT:    addis 3, 2, .LCPI0_0@toc@ha
17; CHECK-P8-NEXT:    lfs 1, .LCPI0_0@toc@l(3)
18; CHECK-P8-NEXT:    blr
19  ret double 1.400000e+01
20}
21
22; As the constant couldn't be represented as float, a double is
23; loaded from constant pool.
24define double @doubleConstant2() {
25; CHECK-P9-LABEL: doubleConstant2:
26; CHECK-P9:       # %bb.0:
27; CHECK-P9-NEXT:    addis 3, 2, .LCPI1_0@toc@ha
28; CHECK-P9-NEXT:    lfd 1, .LCPI1_0@toc@l(3)
29; CHECK-P9-NEXT:    blr
30;
31; CHECK-P8-LABEL: doubleConstant2:
32; CHECK-P8:       # %bb.0:
33; CHECK-P8-NEXT:    addis 3, 2, .LCPI1_0@toc@ha
34; CHECK-P8-NEXT:    lfd 1, .LCPI1_0@toc@l(3)
35; CHECK-P8-NEXT:    blr
36  ret double 2.408904e+01
37}
38
39@FArr = hidden local_unnamed_addr global [10 x float] zeroinitializer, align 4
40
41define float @floatConstantArray() local_unnamed_addr  {
42; CHECK-P9-LABEL: floatConstantArray:
43; CHECK-P9:       # %bb.0:
44; CHECK-P9-NEXT:    addis 3, 2, FArr@toc@ha+12
45; CHECK-P9-NEXT:    lfs 0, FArr@toc@l+12(3)
46; CHECK-P9-NEXT:    addis 3, 2, .LCPI2_0@toc@ha
47; CHECK-P9-NEXT:    lfs 1, .LCPI2_0@toc@l(3)
48; CHECK-P9-NEXT:    xsaddsp 1, 0, 1
49; CHECK-P9-NEXT:    blr
50;
51; CHECK-P8-LABEL: floatConstantArray:
52; CHECK-P8:       # %bb.0:
53; CHECK-P8-NEXT:    addis 3, 2, FArr@toc@ha
54; CHECK-P8-NEXT:    addis 4, 2, .LCPI2_0@toc@ha
55; CHECK-P8-NEXT:    addi 3, 3, FArr@toc@l
56; CHECK-P8-NEXT:    lfs 1, .LCPI2_0@toc@l(4)
57; CHECK-P8-NEXT:    lfs 0, 12(3)
58; CHECK-P8-NEXT:    xsaddsp 1, 0, 1
59; CHECK-P8-NEXT:    blr
60  %1 = load float, float* getelementptr inbounds ([10 x float], [10 x float]* @FArr, i64 0, i64 3), align 4
61  %2 = fadd float %1, 0x400B333340000000
62  ret float %2
63}
64
65define float @floatConstant() {
66; CHECK-P9-LABEL: floatConstant:
67; CHECK-P9:       # %bb.0:
68; CHECK-P9-NEXT:    addis 3, 2, .LCPI3_0@toc@ha
69; CHECK-P9-NEXT:    lfs 1, .LCPI3_0@toc@l(3)
70; CHECK-P9-NEXT:    blr
71;
72; CHECK-P8-LABEL: floatConstant:
73; CHECK-P8:       # %bb.0:
74; CHECK-P8-NEXT:    addis 3, 2, .LCPI3_0@toc@ha
75; CHECK-P8-NEXT:    lfs 1, .LCPI3_0@toc@l(3)
76; CHECK-P8-NEXT:    blr
77  ret float 0x400470A3E0000000
78}
79
80; llvm put the hidden globals into the TOC table.
81; TODO - do some analysis and decide which globals could be put into TOC.
82@d = hidden local_unnamed_addr global [200 x double] zeroinitializer, align 8
83
84define double @doubleConstantArray()  {
85; CHECK-P9-LABEL: doubleConstantArray:
86; CHECK-P9:       # %bb.0:
87; CHECK-P9-NEXT:    addis 3, 2, d@toc@ha+24
88; CHECK-P9-NEXT:    lfd 0, d@toc@l+24(3)
89; CHECK-P9-NEXT:    addis 3, 2, .LCPI4_0@toc@ha
90; CHECK-P9-NEXT:    lfd 1, .LCPI4_0@toc@l(3)
91; CHECK-P9-NEXT:    xsadddp 1, 0, 1
92; CHECK-P9-NEXT:    blr
93;
94; CHECK-P8-LABEL: doubleConstantArray:
95; CHECK-P8:       # %bb.0:
96; CHECK-P8-NEXT:    addis 3, 2, d@toc@ha
97; CHECK-P8-NEXT:    addis 4, 2, .LCPI4_0@toc@ha
98; CHECK-P8-NEXT:    addi 3, 3, d@toc@l
99; CHECK-P8-NEXT:    lfd 1, .LCPI4_0@toc@l(4)
100; CHECK-P8-NEXT:    lfd 0, 24(3)
101; CHECK-P8-NEXT:    xsadddp 1, 0, 1
102; CHECK-P8-NEXT:    blr
103  %1 = load double, double* getelementptr inbounds ([200 x double], [200 x double]* @d, i64 0, i64 3), align 8
104  %2 = fadd double %1, 6.880000e+00
105  ret double %2
106}
107
108@arr = hidden local_unnamed_addr global [20000 x double] zeroinitializer, align 8
109
110; Access an element with an offset that doesn't fit in the displacement field of LFD.
111define double @doubleLargeConstantArray()  {
112; CHECK-P9-LABEL: doubleLargeConstantArray:
113; CHECK-P9:       # %bb.0:
114; CHECK-P9-NEXT:    addis 3, 2, arr@toc@ha
115; CHECK-P9-NEXT:    li 4, 0
116; CHECK-P9-NEXT:    addi 3, 3, arr@toc@l
117; CHECK-P9-NEXT:    ori 4, 4, 32768
118; CHECK-P9-NEXT:    lfdx 0, 3, 4
119; CHECK-P9-NEXT:    addis 3, 2, .LCPI5_0@toc@ha
120; CHECK-P9-NEXT:    lfd 1, .LCPI5_0@toc@l(3)
121; CHECK-P9-NEXT:    xsadddp 1, 0, 1
122; CHECK-P9-NEXT:    blr
123;
124; CHECK-P8-LABEL: doubleLargeConstantArray:
125; CHECK-P8:       # %bb.0:
126; CHECK-P8-NEXT:    addis 3, 2, arr@toc@ha
127; CHECK-P8-NEXT:    li 4, 0
128; CHECK-P8-NEXT:    addis 5, 2, .LCPI5_0@toc@ha
129; CHECK-P8-NEXT:    addi 3, 3, arr@toc@l
130; CHECK-P8-NEXT:    ori 4, 4, 32768
131; CHECK-P8-NEXT:    lfdx 0, 3, 4
132; CHECK-P8-NEXT:    lfd 1, .LCPI5_0@toc@l(5)
133; CHECK-P8-NEXT:    xsadddp 1, 0, 1
134; CHECK-P8-NEXT:    blr
135  %1 = load double, double* getelementptr inbounds ([20000 x double], [20000 x double]* @arr, i64 0, i64 4096), align 8
136  %2 = fadd double %1, 6.880000e+00
137  ret double %2
138}
139
140@vec_arr = global [10 x <4 x i32>] zeroinitializer, align 16
141
142define <4 x i32> @vectorArray() #0 {
143; CHECK-P9-LABEL: vectorArray:
144; CHECK-P9:       # %bb.0: # %entry
145; CHECK-P9-NEXT:    addis 3, 2, .LC0@toc@ha
146; CHECK-P9-NEXT:    ld 3, .LC0@toc@l(3)
147; CHECK-P9-NEXT:    lxv 34, 32(3)
148; CHECK-P9-NEXT:    blr
149;
150; CHECK-P8-LABEL: vectorArray:
151; CHECK-P8:       # %bb.0: # %entry
152; CHECK-P8-NEXT:    addis 3, 2, .LC0@toc@ha
153; CHECK-P8-NEXT:    ld 3, .LC0@toc@l(3)
154; CHECK-P8-NEXT:    addi 3, 3, 32
155; CHECK-P8-NEXT:    lvx 2, 0, 3
156; CHECK-P8-NEXT:    blr
157entry:
158  %0 = load <4 x i32>, <4 x i32>* getelementptr inbounds ([10 x <4 x i32>], [10 x <4 x i32>]* @vec_arr, i64 0, i64 2), align 16
159  ret <4 x i32> %0
160}
161