1; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds < %s | FileCheck %s
2; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s | FileCheck %s
3
4; Variables that are not lowered by this pass are left unchanged
5; CHECK-NOT: asm
6; CHECK-NOT: llvm.amdgcn.module.lds
7; CHECK-NOT: llvm.amdgcn.module.lds.t
8
9; var1, var2 would be transformed were they used from a non-kernel function
10; CHECK-NOT: @var1 =
11; CHECK: @var2 = addrspace(3) global float undef
12@var1 = addrspace(3) global i32 undef
13@var2 = addrspace(3) global float undef
14
15; constant variables are left to the optimizer / error diagnostics
16; CHECK: @const_undef = addrspace(3) constant i32 undef
17; CHECK: @const_with_init = addrspace(3) constant i64 8
18@const_undef = addrspace(3) constant i32 undef
19@const_with_init = addrspace(3) constant i64 8
20
21; External and constant are both left to the optimizer / error diagnostics
22; CHECK: @extern = external addrspace(3) global i32
23@extern = external addrspace(3) global i32
24
25; Use of an addrspace(3) variable with an initializer is skipped,
26; so as to preserve the unimplemented error from llc
27; CHECK: @with_init = addrspace(3) global i64 0
28@with_init = addrspace(3) global i64 0
29
30; Only local addrspace variables are transformed
31; CHECK: @addr4 = addrspace(4) global i64 undef
32@addr4 = addrspace(4) global i64 undef
33
34; Assign to self is treated as any other initializer, i.e. ignored by this pass
35; CHECK: @toself = addrspace(3) global float addrspace(3)* bitcast (float addrspace(3)* addrspace(3)* @toself to float addrspace(3)*), align 8
36@toself = addrspace(3) global float addrspace(3)* bitcast (float addrspace(3)* addrspace(3)* @toself to float addrspace(3)*), align 8
37
38; Use by .used lists doesn't trigger lowering
39; CHECK-NOT: @llvm.used =
40@llvm.used = appending global [1 x i8*] [i8* addrspacecast (i8 addrspace(3)* bitcast (i32 addrspace(3)* @var1 to i8 addrspace(3)*) to i8*)], section "llvm.metadata"
41
42; CHECK: @llvm.compiler.used = appending global [1 x i8*] [i8* addrspacecast (i8 addrspace(3)* bitcast (float addrspace(3)* @var2 to i8 addrspace(3)*) to i8*)], section "llvm.metadata"
43@llvm.compiler.used = appending global [1 x i8*] [i8* addrspacecast (i8 addrspace(3)* bitcast (float addrspace(3)* @var2 to i8 addrspace(3)*) to i8*)], section "llvm.metadata"
44
45; Access from a function would cause lowering for non-excluded cases
46; CHECK-LABEL: @use_variables()
47; CHECK: %c0 = load i32, i32 addrspace(3)* @const_undef, align 4
48; CHECK: %c1 = load i64, i64 addrspace(3)* @const_with_init, align 4
49; CHECK: %v0 = atomicrmw add i64 addrspace(3)* @with_init, i64 1 seq_cst
50; CHECK: %v1 = cmpxchg i32 addrspace(3)* @extern, i32 4, i32 %c0 acq_rel monotonic
51; CHECK: %v2 = atomicrmw add i64 addrspace(4)* @addr4, i64 %c1 monotonic
52define void @use_variables() {
53  %c0 = load i32, i32 addrspace(3)* @const_undef, align 4
54  %c1 = load i64, i64 addrspace(3)* @const_with_init, align 4
55  %v0 = atomicrmw add i64 addrspace(3)* @with_init, i64 1 seq_cst
56  %v1 = cmpxchg i32 addrspace(3)* @extern, i32 4, i32 %c0 acq_rel monotonic
57  %v2 = atomicrmw add i64 addrspace(4)* @addr4, i64 %c1 monotonic
58  ret void
59}
60
61; CHECK-LABEL: @kern_use()
62; CHECK: %inc = atomicrmw add i32 addrspace(3)* getelementptr inbounds (%llvm.amdgcn.kernel.kern_use.lds.t, %llvm.amdgcn.kernel.kern_use.lds.t addrspace(3)* @llvm.amdgcn.kernel.kern_use.lds, i32 0, i32 0), i32 1 monotonic, align 4
63define amdgpu_kernel void @kern_use() {
64  %inc = atomicrmw add i32 addrspace(3)* @var1, i32 1 monotonic
65  call void @use_variables()
66  ret void
67}
68