xref: /llvm-project/llvm/test/Transforms/MemCpyOpt/align.ll (revision a11faeed446882a81e79d780125d93e7199df645)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -S -passes=memcpyopt -verify-memoryssa | FileCheck %s
3target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
4
5declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1) nounwind
6declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1) nounwind
7
8; The resulting memset is only 4-byte aligned, despite containing
9; a 16-byte aligned store in the middle.
10
11define void @foo(ptr %p) {
12; CHECK-LABEL: @foo(
13; CHECK-NEXT:    [[A1:%.*]] = getelementptr i32, ptr [[P:%.*]], i64 1
14; CHECK-NEXT:    [[A2:%.*]] = getelementptr i32, ptr [[P]], i64 2
15; CHECK-NEXT:    [[A3:%.*]] = getelementptr i32, ptr [[P]], i64 3
16; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 4 [[P]], i8 0, i64 16, i1 false)
17; CHECK-NEXT:    ret void
18;
19  store i32 0, ptr %p, align 4
20  %a1 = getelementptr i32, ptr %p, i64 1
21  store i32 0, ptr %a1, align 16
22  %a2 = getelementptr i32, ptr %p, i64 2
23  store i32 0, ptr %a2, align 4
24  %a3 = getelementptr i32, ptr %p, i64 3
25  store i32 0, ptr %a3, align 4
26  ret void
27}
28
29; Replacing %a8 with %a4 in the memset requires boosting the alignment of %a4.
30
31define void @bar() {
32; CHECK-LABEL: @bar(
33; CHECK-NEXT:    [[A4:%.*]] = alloca i32, align 8
34; CHECK-NEXT:    [[A8:%.*]] = alloca i32, align 8
35; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[A4]], i8 0, i64 4, i1 false)
36; CHECK-NEXT:    ret void
37;
38  %a4 = alloca i32, align 4
39  %a8 = alloca i32, align 8
40  call void @llvm.memset.p0.i64(ptr align 8 %a8, i8 0, i64 4, i1 false)
41  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %a4, ptr align 4 %a8, i64 4, i1 false)
42  ret void
43}
44