1; RUN: llc -march=amdgcn -mcpu=SI -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-ALLOCA -check-prefix=SI %s
2; RUN: llc -march=amdgcn -mcpu=SI -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s
3; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-ALLOCA -check-prefix=SI %s
4; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s
5
6declare void @llvm.amdgcn.s.barrier() #1
7
8; SI-LABEL: {{^}}private_access_f64_alloca:
9
10; SI-ALLOCA: buffer_store_dwordx2
11; SI-ALLOCA: buffer_load_dwordx2
12
13; SI-PROMOTE: ds_write_b64
14; SI-PROMOTE: ds_read_b64
15define void @private_access_f64_alloca(double addrspace(1)* noalias %out, double addrspace(1)* noalias %in, i32 %b) #1 {
16  %val = load double, double addrspace(1)* %in, align 8
17  %array = alloca double, i32 16, align 8
18  %ptr = getelementptr inbounds double, double* %array, i32 %b
19  store double %val, double* %ptr, align 8
20  call void @llvm.amdgcn.s.barrier()
21  %result = load double, double* %ptr, align 8
22  store double %result, double addrspace(1)* %out, align 8
23  ret void
24}
25
26; SI-LABEL: {{^}}private_access_v2f64_alloca:
27
28; SI-ALLOCA: buffer_store_dwordx4
29; SI-ALLOCA: buffer_load_dwordx4
30
31; SI-PROMOTE: ds_write_b64
32; SI-PROMOTE: ds_write_b64
33; SI-PROMOTE: ds_read_b64
34; SI-PROMOTE: ds_read_b64
35define void @private_access_v2f64_alloca(<2 x double> addrspace(1)* noalias %out, <2 x double> addrspace(1)* noalias %in, i32 %b) #1 {
36  %val = load <2 x double>, <2 x double> addrspace(1)* %in, align 16
37  %array = alloca <2 x double>, i32 16, align 16
38  %ptr = getelementptr inbounds <2 x double>, <2 x double>* %array, i32 %b
39  store <2 x double> %val, <2 x double>* %ptr, align 16
40  call void @llvm.amdgcn.s.barrier()
41  %result = load <2 x double>, <2 x double>* %ptr, align 16
42  store <2 x double> %result, <2 x double> addrspace(1)* %out, align 16
43  ret void
44}
45
46; SI-LABEL: {{^}}private_access_i64_alloca:
47
48; SI-ALLOCA: buffer_store_dwordx2
49; SI-ALLOCA: buffer_load_dwordx2
50
51; SI-PROMOTE: ds_write_b64
52; SI-PROMOTE: ds_read_b64
53define void @private_access_i64_alloca(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in, i32 %b) #1 {
54  %val = load i64, i64 addrspace(1)* %in, align 8
55  %array = alloca i64, i32 16, align 8
56  %ptr = getelementptr inbounds i64, i64* %array, i32 %b
57  store i64 %val, i64* %ptr, align 8
58  call void @llvm.amdgcn.s.barrier()
59  %result = load i64, i64* %ptr, align 8
60  store i64 %result, i64 addrspace(1)* %out, align 8
61  ret void
62}
63
64; SI-LABEL: {{^}}private_access_v2i64_alloca:
65
66; SI-ALLOCA: buffer_store_dwordx4
67; SI-ALLOCA: buffer_load_dwordx4
68
69; SI-PROMOTE: ds_write_b64
70; SI-PROMOTE: ds_write_b64
71; SI-PROMOTE: ds_read_b64
72; SI-PROMOTE: ds_read_b64
73define void @private_access_v2i64_alloca(<2 x i64> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %in, i32 %b) #1 {
74  %val = load <2 x i64>, <2 x i64> addrspace(1)* %in, align 16
75  %array = alloca <2 x i64>, i32 16, align 16
76  %ptr = getelementptr inbounds <2 x i64>, <2 x i64>* %array, i32 %b
77  store <2 x i64> %val, <2 x i64>* %ptr, align 16
78  call void @llvm.amdgcn.s.barrier()
79  %result = load <2 x i64>, <2 x i64>* %ptr, align 16
80  store <2 x i64> %result, <2 x i64> addrspace(1)* %out, align 16
81  ret void
82}
83
84attributes #0 = { nounwind }
85attributes #1 = { nounwind convergent }
86