1; RUN: opt -S -mtriple=amdgcn--  -amdgpu-replace-lds-use-with-pointer -amdgpu-enable-lds-replace-with-pointer=true < %s | FileCheck %s
2
3; DESCRIPTION:
4;
5; There is one lds global defined here, and this lds is used within a single non-kernel
6; function, as an operand of nested constant expression, and this non-kernel function is
7; reachable from kernel. Hence nested constant expression should to be converted into a
8; series of instructons and pointer replacement should take place.
9;
10
11; Original LDS should exist.
12; CHECK: @used_only_within_func = addrspace(3) global [4 x i32] undef, align 4
13@used_only_within_func = addrspace(3) global [4 x i32] undef, align 4
14
15; Pointers should be created.
16; CHECK: @used_only_within_func.ptr = internal unnamed_addr addrspace(3) global i16 undef, align 2
17
18; Pointer replacement code should be added.
19define void @f0(i32 %x) {
20; CHECK-LABEL: entry:
21; CHECK:   %0 = load i16, i16 addrspace(3)* @used_only_within_func.ptr, align 2
22; CHECK:   %1 = getelementptr i8, i8 addrspace(3)* null, i16 %0
23; CHECK:   %2 = bitcast i8 addrspace(3)* %1 to [4 x i32] addrspace(3)*
24; CHECK:   %3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* %2, i32 0, i32 0
25; CHECK:   %4 = addrspacecast i32 addrspace(3)* %3 to i32*
26; CHECK:   %5 = ptrtoint i32* %4 to i64
27; CHECK:   %6 = add i64 %5, %5
28; CHECK:   %7 = inttoptr i64 %6 to i32*
29; CHECK:   store i32 %x, i32* %7, align 4
30; CHECK:   ret void
31entry:
32  store i32 %x, i32* inttoptr (i64 add (i64 ptrtoint (i32* addrspacecast (i32 addrspace(3)* bitcast ([4 x i32] addrspace(3)* @used_only_within_func to i32 addrspace(3)*) to i32*) to i64), i64 ptrtoint (i32* addrspacecast (i32 addrspace(3)* bitcast ([4 x i32] addrspace(3)* @used_only_within_func to i32 addrspace(3)*) to i32*) to i64)) to i32*), align 4
33  ret void
34}
35
36; Pointer initialization code shoud be added
37define amdgpu_kernel void @k0() {
38; CHECK-LABEL: entry:
39; CHECK:   %0 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
40; CHECK:   %1 = icmp eq i32 %0, 0
41; CHECK:   br i1 %1, label %2, label %3
42;
43; CHECK-LABEL: 2:
44; CHECK:   store i16 ptrtoint ([4 x i32] addrspace(3)* @used_only_within_func to i16), i16 addrspace(3)* @used_only_within_func.ptr, align 2
45; CHECK:   br label %3
46;
47; CHECK-LABEL: 3:
48; CHECK:   call void @llvm.amdgcn.wave.barrier()
49; CHECK:   call void @f0(i32 0)
50; CHECK:   ret void
51entry:
52  call void @f0(i32 0)
53  ret void
54}
55