xref: /llvm-project/llvm/test/CodeGen/PowerPC/unaligned-addressing-mode.ll (revision 5403c59c608c08c8ecd4303763f08eb046eb5e4d)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr9 -ppc-convert-rr-to-ri=false -ppc-asm-full-reg-names < %s | FileCheck %s
3
4; ISEL matches address mode xaddr.
5define i8 @test_xaddr(ptr %p) {
6; CHECK-LABEL: test_xaddr:
7; CHECK:       # %bb.0: # %entry
8; CHECK-NEXT:    li r4, 0
9; CHECK-NEXT:    std r3, -8(r1)
10; CHECK-NEXT:    ori r4, r4, 40000
11; CHECK-NEXT:    lbzx r3, r3, r4
12; CHECK-NEXT:    blr
13entry:
14  %p.addr = alloca ptr, align 8
15  store ptr %p, ptr %p.addr, align 8
16  %0 = load ptr, ptr %p.addr, align 8
17  %add.ptr = getelementptr inbounds i8, ptr %0, i64 40000
18  %1 = load i8, ptr %add.ptr, align 1
19  ret i8 %1
20}
21
22; ISEL matches address mode xaddrX4.
23define i64 @test_xaddrX4(ptr %p) {
24; CHECK-LABEL: test_xaddrX4:
25; CHECK:       # %bb.0: # %entry
26; CHECK-NEXT:    li r4, 3
27; CHECK-NEXT:    std r3, -8(r1)
28; CHECK-NEXT:    ldx r3, r3, r4
29; CHECK-NEXT:    blr
30entry:
31  %p.addr = alloca ptr, align 8
32  store ptr %p, ptr %p.addr, align 8
33  %0 = load ptr, ptr %p.addr, align 8
34  %add.ptr = getelementptr inbounds i8, ptr %0, i64 3
35  %1 = load i64, ptr %add.ptr, align 8
36  ret i64 %1
37}
38
39; ISEL matches address mode xaddrX16.
40define <2 x double> @test_xaddrX16(ptr %arr) {
41; CHECK-LABEL: test_xaddrX16:
42; CHECK:       # %bb.0: # %entry
43; CHECK-NEXT:    li r4, 40
44; CHECK-NEXT:    lxvx vs34, r3, r4
45; CHECK-NEXT:    blr
46entry:
47  %arrayidx1 = getelementptr inbounds double, ptr %arr, i64 5
48  %0 = load <2 x double>, ptr %arrayidx1, align 16
49  ret <2 x double> %0
50}
51
52; ISEL matches address mode xoaddr.
53define void @test_xoaddr(ptr %arr, ptr %arrTo) {
54; CHECK-LABEL: test_xoaddr:
55; CHECK:       # %bb.0: # %entry
56; CHECK-NEXT:    li r5, 8
57; CHECK-NEXT:    lxvx vs0, r3, r5
58; CHECK-NEXT:    li r3, 4
59; CHECK-NEXT:    stxvx vs0, r4, r3
60; CHECK-NEXT:    blr
61entry:
62  %arrayidx = getelementptr inbounds i32, ptr %arrTo, i64 1
63  %arrayidx1 = getelementptr inbounds i32, ptr %arr, i64 2
64  %0 = load <4 x i32>, ptr %arrayidx1, align 8
65  store <4 x i32> %0, ptr %arrayidx, align 8
66  ret void
67}
68
69; ISEL matches address mode xaddrX4 and generates LI which can be moved outside of
70; loop.
71define i64 @test_xaddrX4_loop(ptr %p) {
72; CHECK-LABEL: test_xaddrX4_loop:
73; CHECK:       # %bb.0: # %entry
74; CHECK-NEXT:    addi r4, r3, -8
75; CHECK-NEXT:    li r3, 8
76; CHECK-NEXT:    li r5, 3
77; CHECK-NEXT:    mtctr r3
78; CHECK-NEXT:    li r3, 0
79; CHECK-NEXT:    .p2align 4
80; CHECK-NEXT:  .LBB4_1: # %for.body
81; CHECK-NEXT:    #
82; CHECK-NEXT:    ldu r6, 8(r4)
83; CHECK-NEXT:    ldx r7, r4, r5
84; CHECK-NEXT:    maddld r3, r7, r6, r3
85; CHECK-NEXT:    bdnz .LBB4_1
86; CHECK-NEXT:  # %bb.2: # %for.end
87; CHECK-NEXT:    blr
88; loop instruction number is changed from 5 to 4, so its align is changed from 5 to 4.
89entry:
90  br label %for.body
91
92for.body:                                         ; preds = %for.body, %entry
93  %i.015 = phi i64 [ 0, %entry ], [ %inc, %for.body ]
94  %res.014 = phi i64 [ 0, %entry ], [ %add, %for.body ]
95  %mul = shl i64 %i.015, 3
96  %add.ptr = getelementptr inbounds i8, ptr %p, i64 %mul
97  %0 = load i64, ptr %add.ptr, align 8
98  %add.ptr3 = getelementptr inbounds i8, ptr %add.ptr, i64 3
99  %1 = load i64, ptr %add.ptr3, align 8
100  %mul4 = mul i64 %1, %0
101  %add = add i64 %mul4, %res.014
102  %inc = add nuw nsw i64 %i.015, 1
103  %exitcond = icmp eq i64 %inc, 8
104  br i1 %exitcond, label %for.end, label %for.body
105
106for.end:                                          ; preds = %for.body
107  ret i64 %add
108
109}
110