1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals --version 3
2 ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -O3 --amdgpu-lower-module-lds-strategy=module < %s | FileCheck -check-prefix=GCN %s
3 ; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s
4 ; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s
6 %vec_type = type { %vec_base }
7 %vec_base = type { %union.anon }
8 %union.anon = type { %"vec_base<char, 3>::n_vec_" }
9 %"vec_base<char, 3>::n_vec_" = type { [3 x i8] }
13 @_f1 = linkonce_odr hidden local_unnamed_addr addrspace(3) global %vec_type poison, comdat, align 1
14 @_f2 = linkonce_odr hidden local_unnamed_addr addrspace(3) global %vec_type poison, comdat, align 1
17 ; CHECK: @[[LLVM_AMDGCN_KERNEL_TEST_LDS:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [[LLVM_AMDGCN_KERNEL_TEST_LDS_T:%.*]] poison, align 4, !absolute_symbol !0
19 define protected amdgpu_kernel void @test(ptr addrspace(1) nocapture %ptr.coerce) local_unnamed_addr #0 {
21 ; GCN: ; %bb.0: ; %entry
22 ; GCN-NEXT: v_mov_b32_e32 v0, 0
23 ; GCN-NEXT: v_mov_b32_e32 v1, 2
24 ; GCN-NEXT: ds_write_b8 v0, v1
25 ; GCN-NEXT: ds_read_u8 v2, v0 offset:2
26 ; GCN-NEXT: ds_read_u16 v3, v0
27 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x24
28 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
29 ; GCN-NEXT: ds_write_b8 v0, v2 offset:6
30 ; GCN-NEXT: ds_write_b16 v0, v3 offset:4
31 ; GCN-NEXT: v_cmp_eq_u16_sdwa s[2:3], v3, v1 src0_sel:BYTE_0 src1_sel:DWORD
32 ; GCN-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[2:3]
33 ; GCN-NEXT: global_store_byte v0, v1, s[0:1]
35 ; CHECK-LABEL: define protected amdgpu_kernel void @test(
36 ; CHECK-SAME: ptr addrspace(1) nocapture [[PTR_COERCE:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
38 ; CHECK-NEXT: store i8 3, ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, align 4, !alias.scope !1, !noalias !4
39 ; CHECK-NEXT: tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T:%.*]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), ptr addrspace(3) noundef align 1 dereferenceable(3) @llvm.amdgcn.kernel.test.lds, i64 3, i1 false), !alias.scope !6, !noalias !7
40 ; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), align 4, !alias.scope !4, !noalias !1
41 ; CHECK-NEXT: [[CMP_I_I:%.*]] = icmp eq i8 [[TMP0]], 3
42 ; CHECK-NEXT: store i8 2, ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, align 4, !alias.scope !1, !noalias !4
43 ; CHECK-NEXT: tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), ptr addrspace(3) noundef align 1 dereferenceable(3) @llvm.amdgcn.kernel.test.lds, i64 3, i1 false), !alias.scope !6, !noalias !7
44 ; CHECK-NEXT: [[TMP1:%.*]] = load i8, ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), align 4, !alias.scope !4, !noalias !1
45 ; CHECK-NEXT: [[CMP_I_I19:%.*]] = icmp eq i8 [[TMP1]], 2
46 ; CHECK-NEXT: [[TMP2:%.*]] = and i1 [[CMP_I_I19]], [[CMP_I_I]]
47 ; CHECK-NEXT: [[FROMBOOL8:%.*]] = zext i1 [[TMP2]] to i8
48 ; CHECK-NEXT: store i8 [[FROMBOOL8]], ptr addrspace(1) [[PTR_COERCE]], align 1
49 ; CHECK-NEXT: ret void
52 store i8 3, ptr addrspace(3) @_f1, align 1
53 tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) @_f2, ptr addrspace(3) noundef align 1 dereferenceable(3) @_f1, i64 3, i1 false)
54 %0 = load i8, ptr addrspace(3) @_f2, align 1
55 %cmp.i.i = icmp eq i8 %0, 3
56 store i8 2, ptr addrspace(3) @_f1, align 1
57 tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) @_f2, ptr addrspace(3) noundef align 1 dereferenceable(3) @_f1, i64 3, i1 false)
58 %1 = load i8, ptr addrspace(3) @_f2, align 1
59 %cmp.i.i19 = icmp eq i8 %1, 2
60 %2 = and i1 %cmp.i.i19, %cmp.i.i
61 %frombool8 = zext i1 %2 to i8
62 store i8 %frombool8, ptr addrspace(1) %ptr.coerce, align 1
66 declare void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noalias nocapture writeonly, ptr addrspace(3) noalias nocapture readonly, i64, i1 immarg) #1
69 ; CHECK: attributes #[[ATTR0]] = { "amdgpu-lds-size"="7" }
70 ; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
72 ; CHECK: [[META0:![0-9]+]] = !{i32 0, i32 1}
73 ; CHECK: [[META1:![0-9]+]] = !{!2}
74 ; CHECK: [[META2:![0-9]+]] = distinct !{!2, !3}
75 ; CHECK: [[META3:![0-9]+]] = distinct !{!3}
76 ; CHECK: [[META4:![0-9]+]] = !{!5}
77 ; CHECK: [[META5:![0-9]+]] = distinct !{!5, !3}
78 ; CHECK: [[META6:![0-9]+]] = !{!5, !2}
79 ; CHECK: [[META7:![0-9]+]] = !{}