1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
2 ; XUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=EG -check-prefix=FUNC %s
5 declare i32 @llvm.r600.read.tidig.x() #0
7 ; FUNC-LABEL: {{^}}lshr_i32:
8 ; SI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
9 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
10 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
11 define amdgpu_kernel void @lshr_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
12 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
13 %a = load i32, i32 addrspace(1)* %in
14 %b = load i32, i32 addrspace(1)* %b_ptr
15 %result = lshr i32 %a, %b
16 store i32 %result, i32 addrspace(1)* %out
20 ; FUNC-LABEL: {{^}}lshr_v2i32:
21 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
22 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
24 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
25 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
27 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
28 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
29 define amdgpu_kernel void @lshr_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
30 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
31 %a = load <2 x i32>, <2 x i32> addrspace(1)* %in
32 %b = load <2 x i32>, <2 x i32> addrspace(1)* %b_ptr
33 %result = lshr <2 x i32> %a, %b
34 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
38 ; FUNC-LABEL: {{^}}lshr_v4i32:
39 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
40 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
41 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
42 ; SI: v_lshr_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
44 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
45 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
46 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
47 ; VI: v_lshrrev_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
49 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
50 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
51 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
52 ; EG: LSHR {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
53 define amdgpu_kernel void @lshr_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
54 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
55 %a = load <4 x i32>, <4 x i32> addrspace(1)* %in
56 %b = load <4 x i32>, <4 x i32> addrspace(1)* %b_ptr
57 %result = lshr <4 x i32> %a, %b
58 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
62 ; FUNC-LABEL: {{^}}lshr_i64:
63 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
64 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
66 ; EG: SUB_INT {{\*? *}}[[COMPSH:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHIFT:T[0-9]+\.[XYZW]]]
67 ; EG: LSHL {{\* *}}[[TEMP:T[0-9]+\.[XYZW]]], [[OPHI:T[0-9]+\.[XYZW]]], {{[[COMPSH]]|PV.[XYZW]}}
68 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSH:T[0-9]+\.[XYZW]]], [[SHIFT]], literal
69 ; EG-DAG: LSHL {{\*? *}}[[OVERF:T[0-9]+\.[XYZW]]], {{[[TEMP]]|PV.[XYZW]}}, 1
70 ; EG-DAG: LSHR {{\*? *}}[[LOSMTMP:T[0-9]+\.[XYZW]]], [[OPLO:T[0-9]+\.[XYZW]]], [[SHIFT]]
71 ; EG-DAG: OR_INT {{\*? *}}[[LOSM:T[0-9]+\.[XYZW]]], {{[[LOSMTMP]]|PV.[XYZW]|PS}}, {{[[OVERF]]|PV.[XYZW]}}
72 ; EG-DAG: LSHR {{\*? *}}[[HISM:T[0-9]+\.[XYZW]]], [[OPHI]], {{PS|[[SHIFT]]|PV\.[XYZW]}}
73 ; EG-DAG: SETGT_UINT {{\*? *}}[[RESC:T[0-9]+\.[XYZW]]], [[SHIFT]], literal
74 ; EG-DAG: CNDE_INT {{\*? *}}[[RESLO:T[0-9]+\.[XYZW]]], {{T[0-9]+\.[XYZW]|PS}}
75 ; EG-DAG: LSHR {{\*? *}}[[LOBIG:T[0-9]+\.[XYZW]]], [[OPHI]], [[SHIFT]]
76 ; EG-DAG: CNDE_INT {{\*? *}}[[RESHI:T[0-9]+\.[XYZW]]], {{T[0-9]+\.[XYZW], .*}}, 0.0
77 define amdgpu_kernel void @lshr_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
78 %b_ptr = getelementptr i64, i64 addrspace(1)* %in, i64 1
79 %a = load i64, i64 addrspace(1)* %in
80 %b = load i64, i64 addrspace(1)* %b_ptr
81 %result = lshr i64 %a, %b
82 store i64 %result, i64 addrspace(1)* %out
86 ; FUNC-LABEL: {{^}}lshr_v2i64:
87 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
88 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
90 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
91 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
93 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHA:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHA:T[0-9]+\.[XYZW]]]
94 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHB:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHB:T[0-9]+\.[XYZW]]]
95 ; EG-DAG: LSHL {{\*? *}}[[COMPSHA]]
96 ; EG-DAG: LSHL {{\*? *}}[[COMPSHB]]
97 ; EG-DAG: LSHL {{.*}}, 1
98 ; EG-DAG: LSHL {{.*}}, 1
99 ; EG-DAG: LSHR {{.*}}, [[SHA]]
100 ; EG-DAG: LSHR {{.*}}, [[SHB]]
101 ; EG-DAG: LSHR {{.*}}, [[SHA]]
102 ; EG-DAG: LSHR {{.*}}, [[SHB]]
105 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHA:T[0-9]+\.[XYZW]]]{{.*}}, literal
106 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHB:T[0-9]+\.[XYZW]]]{{.*}}, literal
109 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHA]], literal
110 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHB]], literal
111 ; EG-DAG: CNDE_INT {{.*}}, 0.0
112 ; EG-DAG: CNDE_INT {{.*}}, 0.0
115 define amdgpu_kernel void @lshr_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
116 %b_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %in, i64 1
117 %a = load <2 x i64>, <2 x i64> addrspace(1)* %in
118 %b = load <2 x i64>, <2 x i64> addrspace(1)* %b_ptr
119 %result = lshr <2 x i64> %a, %b
120 store <2 x i64> %result, <2 x i64> addrspace(1)* %out
124 ; FUNC-LABEL: {{^}}lshr_v4i64:
125 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
126 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
127 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
128 ; SI: v_lshr_b64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
130 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
131 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
132 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
133 ; VI: v_lshrrev_b64 {{v\[[0-9]+:[0-9]+\], v[0-9]+, v\[[0-9]+:[0-9]+\]}}
135 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHA:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHA:T[0-9]+\.[XYZW]]]
136 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHB:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHB:T[0-9]+\.[XYZW]]]
137 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHC:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHC:T[0-9]+\.[XYZW]]]
138 ; EG-DAG: SUB_INT {{\*? *}}[[COMPSHD:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHD:T[0-9]+\.[XYZW]]]
139 ; EG-DAG: LSHL {{\*? *}}[[COMPSHA]]
140 ; EG-DAG: LSHL {{\*? *}}[[COMPSHB]]
141 ; EG-DAG: LSHL {{\*? *}}[[COMPSHC]]
142 ; EG-DAG: LSHL {{\*? *}}[[COMPSHD]]
143 ; EG-DAG: LSHL {{.*}}, 1
144 ; EG-DAG: LSHL {{.*}}, 1
145 ; EG-DAG: LSHL {{.*}}, 1
146 ; EG-DAG: LSHL {{.*}}, 1
147 ; EG-DAG: LSHR {{.*}}, [[SHA]]
148 ; EG-DAG: LSHR {{.*}}, [[SHB]]
149 ; EG-DAG: LSHR {{.*}}, [[SHC]]
150 ; EG-DAG: LSHR {{.*}}, [[SHD]]
151 ; EG-DAG: LSHR {{.*}}, [[SHA]]
152 ; EG-DAG: LSHR {{.*}}, [[SHB]]
153 ; EG-DAG: LSHR {{.*}}, [[SHC]]
154 ; EG-DAG: LSHR {{.*}}, [[SHD]]
159 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHA:T[0-9]+\.[XYZW]]]{{.*}}, literal
160 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHB:T[0-9]+\.[XYZW]]]{{.*}}, literal
161 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHC:T[0-9]+\.[XYZW]]]{{.*}}, literal
162 ; EG-DAG: ADD_INT {{\*? *}}[[BIGSHD:T[0-9]+\.[XYZW]]]{{.*}}, literal
169 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHA]], literal
170 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHB]], literal
171 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHC]], literal
172 ; EG-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHD]], literal
173 ; EG-DAG: CNDE_INT {{.*}}, 0.0
174 ; EG-DAG: CNDE_INT {{.*}}, 0.0
175 ; EG-DAG: CNDE_INT {{.*}}, 0.0
176 ; EG-DAG: CNDE_INT {{.*}}, 0.0
181 define amdgpu_kernel void @lshr_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
182 %b_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i64 1
183 %a = load <4 x i64>, <4 x i64> addrspace(1)* %in
184 %b = load <4 x i64>, <4 x i64> addrspace(1)* %b_ptr
185 %result = lshr <4 x i64> %a, %b
186 store <4 x i64> %result, <4 x i64> addrspace(1)* %out
190 ; Make sure load width gets reduced to i32 load.
191 ; GCN-LABEL: {{^}}s_lshr_32_i64:
192 ; GCN-DAG: s_load_dword [[HI_A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x14{{$}}
193 ; GCN-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], 0{{$}}
194 ; GCN-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], [[HI_A]]
195 ; GCN: buffer_store_dwordx2 v{{\[}}[[VLO]]:[[VHI]]{{\]}}
196 define amdgpu_kernel void @s_lshr_32_i64(i64 addrspace(1)* %out, [8 x i32], i64 %a) {
197 %result = lshr i64 %a, 32
198 store i64 %result, i64 addrspace(1)* %out
202 ; GCN-LABEL: {{^}}v_lshr_32_i64:
203 ; GCN-DAG: buffer_load_dword v[[HI_A:[0-9]+]], v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
204 ; GCN-DAG: v_mov_b32_e32 v[[VHI1:[0-9]+]], 0{{$}}
205 ; GCN-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], v[[VHI1]]{{$}}
206 ; GCN: buffer_store_dwordx2 v{{\[}}[[HI_A]]:[[VHI]]{{\]}}
207 define amdgpu_kernel void @v_lshr_32_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
208 %tid = call i32 @llvm.r600.read.tidig.x() #0
209 %gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
210 %gep.out = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
211 %a = load i64, i64 addrspace(1)* %gep.in
212 %result = lshr i64 %a, 32
213 store i64 %result, i64 addrspace(1)* %gep.out
217 attributes #0 = { nounwind readnone }