1 ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn < %s | FileCheck --check-prefixes=GCN-NOHSA,FUNC,SI-NOHSA %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck --check-prefixes=GCN-HSA,FUNC,GCNX3-HSA %s
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck --check-prefixes=GCN-NOHSA,FUNC,GCNX3-NOHSA %s
5 ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=r600 -mcpu=redwood < %s | FileCheck --check-prefixes=R600,FUNC %s
6 ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=r600 -mcpu=cayman < %s | FileCheck --check-prefixes=R600,FUNC %s
8 ; FUNC-LABEL: {{^}}global_load_f32:
9 ; GCN-NOHSA: buffer_load_dword v{{[0-9]+}}
10 ; GCN-HSA: flat_load_dword
12 ; R600: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
13 define amdgpu_kernel void @global_load_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
15 %tmp0 = load float, ptr addrspace(1) %in
16 store float %tmp0, ptr addrspace(1) %out
20 ; FUNC-LABEL: {{^}}global_load_v2f32:
21 ; GCN-NOHSA: buffer_load_dwordx2
22 ; GCN-HSA: flat_load_dwordx2
25 define amdgpu_kernel void @global_load_v2f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
27 %tmp0 = load <2 x float>, ptr addrspace(1) %in
28 store <2 x float> %tmp0, ptr addrspace(1) %out
32 ; FUNC-LABEL: {{^}}global_load_v3f32:
33 ; SI-NOHSA: buffer_load_dwordx4
34 ; GCNX3-NOHSA: buffer_load_dwordx3
35 ; GCNX3-HSA: flat_load_dwordx3
38 define amdgpu_kernel void @global_load_v3f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
40 %tmp0 = load <3 x float>, ptr addrspace(1) %in
41 store <3 x float> %tmp0, ptr addrspace(1) %out
45 ; FUNC-LABEL: {{^}}global_load_v4f32:
46 ; GCN-NOHSA: buffer_load_dwordx4
47 ; GCN-HSA: flat_load_dwordx4
50 define amdgpu_kernel void @global_load_v4f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
52 %tmp0 = load <4 x float>, ptr addrspace(1) %in
53 store <4 x float> %tmp0, ptr addrspace(1) %out
57 ; FUNC-LABEL: {{^}}global_load_v8f32:
58 ; GCN-NOHSA: buffer_load_dwordx4
59 ; GCN-NOHSA: buffer_load_dwordx4
60 ; GCN-HSA: flat_load_dwordx4
61 ; GCN-HSA: flat_load_dwordx4
65 define amdgpu_kernel void @global_load_v8f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
67 %tmp0 = load <8 x float>, ptr addrspace(1) %in
68 store <8 x float> %tmp0, ptr addrspace(1) %out
72 ; FUNC-LABEL: {{^}}global_load_v9f32:
73 ; GCN-NOHSA: buffer_load_dword
74 ; GCN-NOHSA: buffer_load_dwordx4
75 ; GCN-NOHSA: buffer_load_dwordx4
76 ; GCN-HSA: flat_load_dwordx4
77 ; GCN-HSA: flat_load_dword
78 ; GCN-HSA: flat_load_dwordx4
83 define amdgpu_kernel void @global_load_v9f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
85 %tmp0 = load <9 x float>, ptr addrspace(1) %in
86 store <9 x float> %tmp0, ptr addrspace(1) %out
91 ; FUNC-LABEL: {{^}}global_load_v10f32:
92 ; GCN-NOHSA: buffer_load_dwordx4
93 ; GCN-NOHSA: buffer_load_dwordx4
94 ; GCN-NOHSA: buffer_load_dwordx2
95 ; GCN-HSA: flat_load_dwordx4
96 ; GCN-HSA: flat_load_dwordx4
97 ; GCN-HSA: flat_load_dwordx2
102 define amdgpu_kernel void @global_load_v10f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
104 %tmp0 = load <10 x float>, ptr addrspace(1) %in
105 store <10 x float> %tmp0, ptr addrspace(1) %out
109 ; FUNC-LABEL: {{^}}global_load_v11f32:
110 ; SI-NOHSA: buffer_load_dwordx4
111 ; SI-NOHSA: buffer_load_dwordx4
112 ; SI-NOHSA: buffer_load_dwordx4
113 ; GCNX3-NOHSA: buffer_load_dwordx4
114 ; GCNX3-NOHSA: buffer_load_dwordx4
115 ; GCNX3-NOHSA: buffer_load_dwordx3
116 ; GCN-HSA: flat_load_dwordx4
117 ; GCN-HSA: flat_load_dwordx4
118 ; GCN-HSA: flat_load_dwordx3
123 define amdgpu_kernel void @global_load_v11f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
125 %tmp0 = load <11 x float>, ptr addrspace(1) %in
126 store <11 x float> %tmp0, ptr addrspace(1) %out
130 ; FUNC-LABEL: {{^}}global_load_v12f32:
131 ; GCN-NOHSA: buffer_load_dwordx4
132 ; GCN-NOHSA: buffer_load_dwordx4
133 ; GCN-NOHSA: buffer_load_dwordx4
134 ; GCN-HSA: flat_load_dwordx4
135 ; GCN-HSA: flat_load_dwordx4
136 ; GCN-HSA: flat_load_dwordx4
141 define amdgpu_kernel void @global_load_v12f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
143 %tmp0 = load <12 x float>, ptr addrspace(1) %in
144 store <12 x float> %tmp0, ptr addrspace(1) %out
148 ; FUNC-LABEL: {{^}}global_load_v16f32:
149 ; GCN-NOHSA: buffer_load_dwordx4
150 ; GCN-NOHSA: buffer_load_dwordx4
151 ; GCN-NOHSA: buffer_load_dwordx4
152 ; GCN-NOHSA: buffer_load_dwordx4
154 ; GCN-HSA: flat_load_dwordx4
155 ; GCN-HSA: flat_load_dwordx4
156 ; GCN-HSA: flat_load_dwordx4
157 ; GCN-HSA: flat_load_dwordx4
163 define amdgpu_kernel void @global_load_v16f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
165 %tmp0 = load <16 x float>, ptr addrspace(1) %in
166 store <16 x float> %tmp0, ptr addrspace(1) %out
170 attributes #0 = { nounwind }