1; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1032 %s
2; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1064 %s
3; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-early-ifcvt=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1032 %s
4; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-early-ifcvt=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1064 %s
5; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX1032,GFX10DEFWAVE %s
6
7; GCN-LABEL: {{^}}test_vopc_i32:
8; GFX1032: v_cmp_lt_i32_e32 vcc_lo, 0, v{{[0-9]+}}
9; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, vcc_lo
10; GFX1064: v_cmp_lt_i32_e32 vcc, 0, v{{[0-9]+}}
11; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, vcc{{$}}
12define amdgpu_kernel void @test_vopc_i32(i32 addrspace(1)* %arg) {
13  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
14  %gep = getelementptr inbounds i32, i32 addrspace(1)* %arg, i32 %lid
15  %load = load i32, i32 addrspace(1)* %gep, align 4
16  %cmp = icmp sgt i32 %load, 0
17  %sel = select i1 %cmp, i32 1, i32 2
18  store i32 %sel, i32 addrspace(1)* %gep, align 4
19  ret void
20}
21
22; GCN-LABEL: {{^}}test_vopc_f32:
23; GFX1032: v_cmp_nge_f32_e32 vcc_lo, 0, v{{[0-9]+}}
24; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, vcc_lo
25; GFX1064: v_cmp_nge_f32_e32 vcc, 0, v{{[0-9]+}}
26; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, vcc{{$}}
27define amdgpu_kernel void @test_vopc_f32(float addrspace(1)* %arg) {
28  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
29  %gep = getelementptr inbounds float, float addrspace(1)* %arg, i32 %lid
30  %load = load float, float addrspace(1)* %gep, align 4
31  %cmp = fcmp ugt float %load, 0.0
32  %sel = select i1 %cmp, float 1.0, float 2.0
33  store float %sel, float addrspace(1)* %gep, align 4
34  ret void
35}
36
37; GCN-LABEL: {{^}}test_vopc_vcmpx:
38; GFX1032: v_cmpx_le_f32_e32 0, v{{[0-9]+}}
39; GFX1064: v_cmpx_le_f32_e32 0, v{{[0-9]+}}
40define amdgpu_ps void @test_vopc_vcmpx(float %x) {
41  %cmp = fcmp oge float %x, 0.0
42  call void @llvm.amdgcn.kill(i1 %cmp)
43  ret void
44}
45
46; GCN-LABEL: {{^}}test_vopc_2xf16:
47; GFX1032: v_cmp_le_f16_sdwa [[SC:s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
48; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
49; GFX1064: v_cmp_le_f16_sdwa [[SC:s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
50; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
51define amdgpu_kernel void @test_vopc_2xf16(<2 x half> addrspace(1)* %arg) {
52  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
53  %gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %lid
54  %load = load <2 x half>, <2 x half> addrspace(1)* %gep, align 4
55  %elt = extractelement <2 x half> %load, i32 1
56  %cmp = fcmp ugt half %elt, 0.0
57  %sel = select i1 %cmp, <2 x half> <half 1.0, half 1.0>, <2 x half> %load
58  store <2 x half> %sel, <2 x half> addrspace(1)* %gep, align 4
59  ret void
60}
61
62; GCN-LABEL: {{^}}test_vopc_class:
63; GFX1032: v_cmp_class_f32_e64 [[C:vcc_lo|s[0-9:]+]], s{{[0-9]+}}, 0x204
64; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[C]]
65; GFX1064: v_cmp_class_f32_e64 [[C:vcc|s\[[0-9:]+\]]], s{{[0-9]+}}, 0x204
66; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[C]]{{$}}
67define amdgpu_kernel void @test_vopc_class(i32 addrspace(1)* %out, float %x) #0 {
68  %fabs = tail call float @llvm.fabs.f32(float %x)
69  %cmp = fcmp oeq float %fabs, 0x7FF0000000000000
70  %ext = zext i1 %cmp to i32
71  store i32 %ext, i32 addrspace(1)* %out, align 4
72  ret void
73}
74
75; GCN-LABEL: {{^}}test_vcmp_vcnd_f16:
76; GFX1032: v_cmp_neq_f16_e64 [[C:vcc_lo|s\[[0-9:]+\]]], 0x7c00, s{{[0-9]+}}
77; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c00, v{{[0-9]+}}, [[C]]
78
79; GFX1064: v_cmp_neq_f16_e64 [[C:vcc|s\[[0-9:]+\]]], 0x7c00, s{{[0-9]+}}
80; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c00, v{{[0-9]+}}, [[C]]{{$}}
81define amdgpu_kernel void @test_vcmp_vcnd_f16(half addrspace(1)* %out, half %x) #0 {
82  %cmp = fcmp oeq half %x, 0x7FF0000000000000
83  %sel = select i1 %cmp, half 1.0, half %x
84  store half %sel, half addrspace(1)* %out, align 2
85  ret void
86}
87
88; GCN-LABEL: {{^}}test_vop3_cmp_f32_sop_and:
89; GFX1032: v_cmp_nge_f32_e32 vcc_lo, 0, v{{[0-9]+}}
90; GFX1032: v_cmp_nle_f32_e64 [[C2:s[0-9]+]], 1.0, v{{[0-9]+}}
91; GFX1032: s_and_b32 [[AND:s[0-9]+]], vcc_lo, [[C2]]
92; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, [[AND]]
93; GFX1064: v_cmp_nge_f32_e32 vcc, 0, v{{[0-9]+}}
94; GFX1064: v_cmp_nle_f32_e64 [[C2:s\[[0-9:]+\]]], 1.0, v{{[0-9]+}}
95; GFX1064: s_and_b64 [[AND:s\[[0-9:]+\]]], vcc, [[C2]]
96; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, [[AND]]
97define amdgpu_kernel void @test_vop3_cmp_f32_sop_and(float addrspace(1)* %arg) {
98  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
99  %gep = getelementptr inbounds float, float addrspace(1)* %arg, i32 %lid
100  %load = load float, float addrspace(1)* %gep, align 4
101  %cmp = fcmp ugt float %load, 0.0
102  %cmp2 = fcmp ult float %load, 1.0
103  %and = and i1 %cmp, %cmp2
104  %sel = select i1 %and, float 1.0, float 2.0
105  store float %sel, float addrspace(1)* %gep, align 4
106  ret void
107}
108
109; GCN-LABEL: {{^}}test_vop3_cmp_i32_sop_xor:
110; GFX1032: v_cmp_lt_i32_e32 vcc_lo, 0, v{{[0-9]+}}
111; GFX1032: v_cmp_gt_i32_e64 [[C2:s[0-9]+]], 1, v{{[0-9]+}}
112; GFX1032: s_xor_b32 [[AND:s[0-9]+]], vcc_lo, [[C2]]
113; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, [[AND]]
114; GFX1064: v_cmp_lt_i32_e32 vcc, 0, v{{[0-9]+}}
115; GFX1064: v_cmp_gt_i32_e64 [[C2:s\[[0-9:]+\]]], 1, v{{[0-9]+}}
116; GFX1064: s_xor_b64 [[AND:s\[[0-9:]+\]]], vcc, [[C2]]
117; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, [[AND]]
118define amdgpu_kernel void @test_vop3_cmp_i32_sop_xor(i32 addrspace(1)* %arg) {
119  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
120  %gep = getelementptr inbounds i32, i32 addrspace(1)* %arg, i32 %lid
121  %load = load i32, i32 addrspace(1)* %gep, align 4
122  %cmp = icmp sgt i32 %load, 0
123  %cmp2 = icmp slt i32 %load, 1
124  %xor = xor i1 %cmp, %cmp2
125  %sel = select i1 %xor, i32 1, i32 2
126  store i32 %sel, i32 addrspace(1)* %gep, align 4
127  ret void
128}
129
130; GCN-LABEL: {{^}}test_vop3_cmp_u32_sop_or:
131; GFX1032: v_cmp_lt_u32_e32 vcc_lo, 3, v{{[0-9]+}}
132; GFX1032: v_cmp_gt_u32_e64 [[C2:s[0-9]+]], 2, v{{[0-9]+}}
133; GFX1032: s_or_b32 [[AND:s[0-9]+]], vcc_lo, [[C2]]
134; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, [[AND]]
135; GFX1064: v_cmp_lt_u32_e32 vcc, 3, v{{[0-9]+}}
136; GFX1064: v_cmp_gt_u32_e64 [[C2:s\[[0-9:]+\]]], 2, v{{[0-9]+}}
137; GFX1064: s_or_b64 [[AND:s\[[0-9:]+\]]], vcc, [[C2]]
138; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 1, [[AND]]
139define amdgpu_kernel void @test_vop3_cmp_u32_sop_or(i32 addrspace(1)* %arg) {
140  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
141  %gep = getelementptr inbounds i32, i32 addrspace(1)* %arg, i32 %lid
142  %load = load i32, i32 addrspace(1)* %gep, align 4
143  %cmp = icmp ugt i32 %load, 3
144  %cmp2 = icmp ult i32 %load, 2
145  %or = or i1 %cmp, %cmp2
146  %sel = select i1 %or, i32 1, i32 2
147  store i32 %sel, i32 addrspace(1)* %gep, align 4
148  ret void
149}
150
151; GCN-LABEL: {{^}}test_mask_if:
152; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
153; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
154; GCN: ; mask branch
155define amdgpu_kernel void @test_mask_if(i32 addrspace(1)* %arg) #0 {
156  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
157  %cmp = icmp ugt i32 %lid, 10
158  br i1 %cmp, label %if, label %endif
159
160if:
161  store i32 0, i32 addrspace(1)* %arg, align 4
162  br label %endif
163
164endif:
165  ret void
166}
167
168; GCN-LABEL: {{^}}test_loop_with_if:
169; GFX1032: s_or_b32 s{{[0-9]+}}, vcc_lo, s{{[0-9]+}}
170; GFX1032: s_andn2_b32 exec_lo, exec_lo, s{{[0-9]+}}
171; GFX1064: s_or_b64 s[{{[0-9:]+}}], vcc, s[{{[0-9:]+}}]
172; GFX1064: s_andn2_b64 exec, exec, s[{{[0-9:]+}}]
173; GCN:     s_cbranch_execz
174; GCN:   BB{{.*}}:
175; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
176; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
177; GCN:     s_cbranch_execz
178; GCN:   BB{{.*}}:
179; GCN:   BB{{.*}}:
180; GFX1032: s_xor_b32 s{{[0-9]+}}, exec_lo, s{{[0-9]+}}
181; GFX1064: s_xor_b64 s[{{[0-9:]+}}], exec, s[{{[0-9:]+}}]
182; GCN:     ; mask branch BB
183; GCN:   BB{{.*}}:
184; GCN:   BB{{.*}}:
185; GFX1032: s_or_b32 exec_lo, exec_lo, s{{[0-9]+}}
186; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, s{{[0-9]+}}
187; GFX1064: s_or_b64 exec, exec, s[{{[0-9:]+}}]
188; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}]{{$}}
189; GCN:     ; mask branch BB
190; GCN:   BB{{.*}}:
191; GCN:   BB{{.*}}:
192; GCN:     s_endpgm
193define amdgpu_kernel void @test_loop_with_if(i32 addrspace(1)* %arg) #0 {
194bb:
195  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
196  br label %bb2
197
198bb1:
199  ret void
200
201bb2:
202  %tmp3 = phi i32 [ 0, %bb ], [ %tmp15, %bb13 ]
203  %tmp4 = icmp slt i32 %tmp3, %tmp
204  br i1 %tmp4, label %bb5, label %bb11
205
206bb5:
207  %tmp6 = sext i32 %tmp3 to i64
208  %tmp7 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp6
209  %tmp8 = load i32, i32 addrspace(1)* %tmp7, align 4
210  %tmp9 = icmp sgt i32 %tmp8, 10
211  br i1 %tmp9, label %bb10, label %bb11
212
213bb10:
214  store i32 %tmp, i32 addrspace(1)* %tmp7, align 4
215  br label %bb13
216
217bb11:
218  %tmp12 = sdiv i32 %tmp3, 2
219  br label %bb13
220
221bb13:
222  %tmp14 = phi i32 [ %tmp3, %bb10 ], [ %tmp12, %bb11 ]
223  %tmp15 = add nsw i32 %tmp14, 1
224  %tmp16 = icmp slt i32 %tmp14, 255
225  br i1 %tmp16, label %bb2, label %bb1
226}
227
228; GCN-LABEL: {{^}}test_loop_with_if_else_break:
229; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
230; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
231; GCN:     ; mask branch
232; GCN:     s_cbranch_execz
233; GCN:   BB{{.*}}:
234; GCN:   BB{{.*}}:
235
236; GFX1032: s_or_b32 [[MASK0:s[0-9]+]], [[MASK0]], vcc_lo
237; GFX1064: s_or_b64 [[MASK0:s\[[0-9:]+\]]], [[MASK0]], vcc
238; GFX1032: s_andn2_b32 [[MASK1:s[0-9]+]], [[MASK1]], exec_lo
239; GFX1064: s_andn2_b64 [[MASK1:s\[[0-9:]+\]]], [[MASK1]], exec
240; GCN:     global_store_dword
241; GFX1032: s_and_b32 [[MASK0]], [[MASK0]], exec_lo
242; GFX1064: s_and_b64 [[MASK0]], [[MASK0]], exec
243; GFX1032: s_or_b32 [[MASK1]], [[MASK1]], [[MASK0]]
244; GFX1064: s_or_b64 [[MASK1]], [[MASK1]], [[MASK0]]
245; GCN:   BB{{.*}}: ; %Flow
246; GFX1032: s_and_b32 [[TMP0:s[0-9]+]], exec_lo, [[MASK1]]
247; GFX1064: s_and_b64 [[TMP0:s\[[0-9:]+\]]], exec, [[MASK1]]
248; GFX1032: s_or_b32  [[ACC:s[0-9]+]], [[TMP0]], [[ACC]]
249; GFX1064: s_or_b64  [[ACC:s\[[0-9:]+\]]], [[TMP0]], [[ACC]]
250; GFX1032: s_andn2_b32 exec_lo, exec_lo, [[ACC]]
251; GFX1064: s_andn2_b64 exec, exec, [[ACC]]
252; GCN:     s_cbranch_execz
253; GCN:   BB{{.*}}:
254; GCN: s_load_dword [[LOAD:s[0-9]+]]
255; GFX1032: s_or_b32 [[MASK1]], [[MASK1]], exec_lo
256; GFX1064: s_or_b64 [[MASK1]], [[MASK1]], exec
257; GCN: s_cmp_lt_i32 [[LOAD]], 11
258define amdgpu_kernel void @test_loop_with_if_else_break(i32 addrspace(1)* %arg) #0 {
259bb:
260  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
261  %tmp1 = icmp eq i32 %tmp, 0
262  br i1 %tmp1, label %.loopexit, label %.preheader
263
264.preheader:
265  br label %bb2
266
267bb2:
268  %tmp3 = phi i32 [ %tmp9, %bb8 ], [ 0, %.preheader ]
269  %tmp4 = zext i32 %tmp3 to i64
270  %tmp5 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp4
271  %tmp6 = load i32, i32 addrspace(1)* %tmp5, align 4
272  %tmp7 = icmp sgt i32 %tmp6, 10
273  br i1 %tmp7, label %bb8, label %.loopexit
274
275bb8:
276  store i32 %tmp, i32 addrspace(1)* %tmp5, align 4
277  %tmp9 = add nuw nsw i32 %tmp3, 1
278  %tmp10 = icmp ult i32 %tmp9, 256
279  %tmp11 = icmp ult i32 %tmp9, %tmp
280  %tmp12 = and i1 %tmp10, %tmp11
281  br i1 %tmp12, label %bb2, label %.loopexit
282
283.loopexit:
284  ret void
285}
286
287; GCN-LABEL: {{^}}test_addc_vop2b:
288; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, s{{[0-9]+}}
289; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}, vcc_lo
290; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, s{{[0-9]+}}
291; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}, vcc{{$}}
292define amdgpu_kernel void @test_addc_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
293bb:
294  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
295  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
296  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
297  %tmp5 = add nsw i64 %tmp4, %arg1
298  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
299  ret void
300}
301
302; GCN-LABEL: {{^}}test_subbrev_vop2b:
303; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s[0-9]+|vcc_lo]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
304; GFX1032: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, {{[vs][0-9]+}}, [[A0]]{{$}}
305; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
306; GFX1064: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, {{[vs][0-9]+}}, [[A0]]{{$}}
307define amdgpu_kernel void @test_subbrev_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
308bb:
309  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
310  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
311  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
312  %tmp5 = sub nsw i64 %tmp4, %arg1
313  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
314  ret void
315}
316
317; GCN-LABEL: {{^}}test_subb_vop2b:
318; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s[0-9]+|vcc_lo]], s{{[0-9]+}}, v{{[0-9]+}}{{$}}
319; GFX1032: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, v{{[0-9]+}}, [[A0]]{{$}}
320; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], s{{[0-9]+}}, v{{[0-9]+}}{{$}}
321; GFX1064: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, [[A0]]{{$}}
322define amdgpu_kernel void @test_subb_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
323bb:
324  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
325  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
326  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
327  %tmp5 = sub nsw i64 %arg1, %tmp4
328  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
329  ret void
330}
331
332; GCN-LABEL: {{^}}test_udiv64:
333; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, [[SDST:s[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}}
334; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
335; GFX1032: v_add_co_ci_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}, [[SDST]]
336; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
337; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
338; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
339; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
340; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
341; GFX1032: v_sub_co_ci_u32_e64 v{{[0-9]+}}, s{{[0-9]+}}, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
342; GFX1032: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
343; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, [[SDST:s\[[0-9:]+\]]], v{{[0-9]+}}, v{{[0-9]+}}
344; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
345; GFX1064: v_add_co_ci_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}, [[SDST]]
346; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
347; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
348; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
349; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
350; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
351; GFX1064: v_sub_co_ci_u32_e64 v{{[0-9]+}}, s[{{[0-9:]+}}], {{[vs][0-9]+}}, v{{[0-9]+}}, vcc{{$}}
352; GFX1064: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc{{$}}
353define amdgpu_kernel void @test_udiv64(i64 addrspace(1)* %arg) #0 {
354bb:
355  %tmp = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 1
356  %tmp1 = load i64, i64 addrspace(1)* %tmp, align 8
357  %tmp2 = load i64, i64 addrspace(1)* %arg, align 8
358  %tmp3 = udiv i64 %tmp1, %tmp2
359  %tmp4 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 2
360  store i64 %tmp3, i64 addrspace(1)* %tmp4, align 8
361  ret void
362}
363
364; GCN-LABEL: {{^}}test_div_scale_f32:
365; GFX1032: v_div_scale_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
366; GFX1064: v_div_scale_f32 v{{[0-9]+}}, s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
367define amdgpu_kernel void @test_div_scale_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
368  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
369  %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
370  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
371
372  %a = load volatile float, float addrspace(1)* %gep.0, align 4
373  %b = load volatile float, float addrspace(1)* %gep.1, align 4
374
375  %result = call { float, i1 } @llvm.amdgcn.div.scale.f32(float %a, float %b, i1 false) nounwind readnone
376  %result0 = extractvalue { float, i1 } %result, 0
377  store float %result0, float addrspace(1)* %out, align 4
378  ret void
379}
380
381; GCN-LABEL: {{^}}test_div_scale_f64:
382; GFX1032: v_div_scale_f64 v[{{[0-9:]+}}], s{{[0-9]+}}, v[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
383; GFX1064: v_div_scale_f64 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
384define amdgpu_kernel void @test_div_scale_f64(double addrspace(1)* %out, double addrspace(1)* %aptr, double addrspace(1)* %in) #0 {
385  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
386  %gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
387  %gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
388
389  %a = load volatile double, double addrspace(1)* %gep.0, align 8
390  %b = load volatile double, double addrspace(1)* %gep.1, align 8
391
392  %result = call { double, i1 } @llvm.amdgcn.div.scale.f64(double %a, double %b, i1 true) nounwind readnone
393  %result0 = extractvalue { double, i1 } %result, 0
394  store double %result0, double addrspace(1)* %out, align 8
395  ret void
396}
397
398; GCN-LABEL: {{^}}test_mad_i64_i32:
399; GFX1032: v_mad_i64_i32 v[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
400; GFX1064: v_mad_i64_i32 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
401define i64 @test_mad_i64_i32(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
402  %sext0 = sext i32 %arg0 to i64
403  %sext1 = sext i32 %arg1 to i64
404  %mul = mul i64 %sext0, %sext1
405  %mad = add i64 %mul, %arg2
406  ret i64 %mad
407}
408
409; GCN-LABEL: {{^}}test_mad_u64_u32:
410; GFX1032: v_mad_u64_u32 v[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
411; GFX1064: v_mad_u64_u32 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
412define i64 @test_mad_u64_u32(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
413  %sext0 = zext i32 %arg0 to i64
414  %sext1 = zext i32 %arg1 to i64
415  %mul = mul i64 %sext0, %sext1
416  %mad = add i64 %mul, %arg2
417  ret i64 %mad
418}
419
420; GCN-LABEL: {{^}}test_div_fmas_f32:
421; GFX1032: v_cmp_eq_u32_e64 vcc_lo,
422; GFX1064: v_cmp_eq_u32_e64 vcc,
423; GCN:     v_div_fmas_f32 v{{[0-9]+}}, {{[vs][0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
424define amdgpu_kernel void @test_div_fmas_f32(float addrspace(1)* %out, float %a, float %b, float %c, i1 %d) nounwind {
425  %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %d) nounwind readnone
426  store float %result, float addrspace(1)* %out, align 4
427  ret void
428}
429
430; GCN-LABEL: {{^}}test_div_fmas_f64:
431; GFX1032: v_cmp_eq_u32_e64 vcc_lo,
432; GFX1064: v_cmp_eq_u32_e64 vcc,
433; GCN-DAG: v_div_fmas_f64 v[{{[0-9:]+}}], {{[vs]}}[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
434define amdgpu_kernel void @test_div_fmas_f64(double addrspace(1)* %out, double %a, double %b, double %c, i1 %d) nounwind {
435  %result = call double @llvm.amdgcn.div.fmas.f64(double %a, double %b, double %c, i1 %d) nounwind readnone
436  store double %result, double addrspace(1)* %out, align 8
437  ret void
438}
439
440; GCN-LABEL: {{^}}test_div_fmas_f32_i1_phi_vcc:
441; GFX1032: s_mov_b32 [[VCC:vcc_lo]], 0{{$}}
442; GFX1064: s_mov_b64 [[VCC:vcc]], 0{{$}}
443; GFX1032: s_and_saveexec_b32 [[SAVE:s[0-9]+]], s{{[0-9]+}}{{$}}
444; GFX1064: s_and_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], s[{{[0-9:]+}}]{{$}}
445
446; GCN: load_dword [[LOAD:v[0-9]+]]
447; GCN: v_cmp_ne_u32_e32 [[VCC]], 0, [[LOAD]]
448
449; GCN: BB{{[0-9_]+}}:
450; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE]]
451; GFX1064: s_or_b64 exec, exec, [[SAVE]]
452; GCN: v_div_fmas_f32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
453define amdgpu_kernel void @test_div_fmas_f32_i1_phi_vcc(float addrspace(1)* %out, float addrspace(1)* %in, i32 addrspace(1)* %dummy) #0 {
454entry:
455  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
456  %gep.out = getelementptr float, float addrspace(1)* %out, i32 2
457  %gep.a = getelementptr float, float addrspace(1)* %in, i32 %tid
458  %gep.b = getelementptr float, float addrspace(1)* %gep.a, i32 1
459  %gep.c = getelementptr float, float addrspace(1)* %gep.a, i32 2
460
461  %a = load float, float addrspace(1)* %gep.a
462  %b = load float, float addrspace(1)* %gep.b
463  %c = load float, float addrspace(1)* %gep.c
464
465  %cmp0 = icmp eq i32 %tid, 0
466  br i1 %cmp0, label %bb, label %exit
467
468bb:
469  %val = load volatile i32, i32 addrspace(1)* %dummy
470  %cmp1 = icmp ne i32 %val, 0
471  br label %exit
472
473exit:
474  %cond = phi i1 [false, %entry], [%cmp1, %bb]
475  %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %cond) nounwind readnone
476  store float %result, float addrspace(1)* %gep.out, align 4
477  ret void
478}
479
480; GCN-LABEL: {{^}}fdiv_f32:
481; GFC1032: v_div_scale_f32 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}}
482; GFC1064: v_div_scale_f32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}}
483; GCN: v_rcp_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}
484; GCN-NOT: vcc
485; GCN: v_div_fmas_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
486define amdgpu_kernel void @fdiv_f32(float addrspace(1)* %out, float %a, float %b) #0 {
487entry:
488  %fdiv = fdiv float %a, %b
489  store float %fdiv, float addrspace(1)* %out
490  ret void
491}
492
493; GCN-LABEL: {{^}}test_br_cc_f16:
494; GFX1032:  v_cmp_nlt_f16_e32 vcc_lo,
495; GFX1032:  s_and_b32 vcc_lo, exec_lo, vcc_lo
496; GFX1064:  v_cmp_nlt_f16_e32 vcc,
497; GFX1064:  s_and_b64 vcc, exec, vcc{{$}}
498; GCN-NEXT: s_cbranch_vccnz
499define amdgpu_kernel void @test_br_cc_f16(
500    half addrspace(1)* %r,
501    half addrspace(1)* %a,
502    half addrspace(1)* %b) {
503entry:
504  %a.val = load half, half addrspace(1)* %a
505  %b.val = load half, half addrspace(1)* %b
506  %fcmp = fcmp olt half %a.val, %b.val
507  br i1 %fcmp, label %one, label %two
508
509one:
510  store half %a.val, half addrspace(1)* %r
511  ret void
512
513two:
514  store half %b.val, half addrspace(1)* %r
515  ret void
516}
517
518; GCN-LABEL: {{^}}test_brcc_i1:
519; GCN:      s_cmp_eq_u32 s{{[0-9]+}}, 0
520; GCN-NEXT: s_cbranch_scc1
521define amdgpu_kernel void @test_brcc_i1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in, i1 %val) #0 {
522  %cmp0 = icmp ne i1 %val, 0
523  br i1 %cmp0, label %store, label %end
524
525store:
526  store i32 222, i32 addrspace(1)* %out
527  ret void
528
529end:
530  ret void
531}
532
533; GCN-LABEL: {{^}}test_preserve_condition_undef_flag:
534; GFX1032: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
535; GFX1032: v_cmp_ngt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 0
536; GFX1032: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
537; GFX1032: s_or_b32 [[OR1:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
538; GFX1032: s_or_b32 [[OR2:s[0-9]+]], [[OR1]], s{{[0-9]+}}
539; GFX1032: s_and_b32 vcc_lo, exec_lo, [[OR2]]
540; GFX1064: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
541; GFX1064: v_cmp_ngt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 0
542; GFX1064: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
543; GFX1064: s_or_b64 [[OR1:s\[[0-9:]+\]]], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
544; GFX1064: s_or_b64 [[OR2:s\[[0-9:]+\]]], [[OR1]], s[{{[0-9:]+}}]
545; GFX1064: s_and_b64 vcc, exec, [[OR2]]
546; GCN:     s_cbranch_vccnz
547define amdgpu_kernel void @test_preserve_condition_undef_flag(float %arg, i32 %arg1, float %arg2) #0 {
548bb0:
549  %tmp = icmp sgt i32 %arg1, 4
550  %undef = call i1 @llvm.amdgcn.class.f32(float undef, i32 undef)
551  %tmp4 = select i1 %undef, float %arg, float 1.000000e+00
552  %tmp5 = fcmp ogt float %arg2, 0.000000e+00
553  %tmp6 = fcmp olt float %arg2, 1.000000e+00
554  %tmp7 = fcmp olt float %arg, %tmp4
555  %tmp8 = and i1 %tmp5, %tmp6
556  %tmp9 = and i1 %tmp8, %tmp7
557  br i1 %tmp9, label %bb1, label %bb2
558
559bb1:
560  store volatile i32 0, i32 addrspace(1)* undef
561  br label %bb2
562
563bb2:
564  ret void
565}
566
567; GCN-LABEL: {{^}}test_invert_true_phi_cond_break_loop:
568; GFX1032: s_xor_b32 s{{[0-9]+}}, s{{[0-9]+}}, -1
569; GFX1032: s_or_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
570; GFX1064: s_xor_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], -1
571; GFX1064: s_or_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
572define amdgpu_kernel void @test_invert_true_phi_cond_break_loop(i32 %arg) #0 {
573bb:
574  %id = call i32 @llvm.amdgcn.workitem.id.x()
575  %tmp = sub i32 %id, %arg
576  br label %bb1
577
578bb1:                                              ; preds = %Flow, %bb
579  %lsr.iv = phi i32 [ undef, %bb ], [ %tmp2, %Flow ]
580  %lsr.iv.next = add i32 %lsr.iv, 1
581  %cmp0 = icmp slt i32 %lsr.iv.next, 0
582  br i1 %cmp0, label %bb4, label %Flow
583
584bb4:                                              ; preds = %bb1
585  %load = load volatile i32, i32 addrspace(1)* undef, align 4
586  %cmp1 = icmp sge i32 %tmp, %load
587  br label %Flow
588
589Flow:                                             ; preds = %bb4, %bb1
590  %tmp2 = phi i32 [ %lsr.iv.next, %bb4 ], [ undef, %bb1 ]
591  %tmp3 = phi i1 [ %cmp1, %bb4 ], [ true, %bb1 ]
592  br i1 %tmp3, label %bb1, label %bb9
593
594bb9:                                              ; preds = %Flow
595  store volatile i32 7, i32 addrspace(3)* undef
596  ret void
597}
598
599; GCN-LABEL: {{^}}test_movrels_extract_neg_offset_vgpr:
600; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 1, v{{[0-9]+}}
601; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc_lo
602; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 2, v{{[0-9]+}}
603; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc_lo
604; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 3, v{{[0-9]+}}
605; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc_lo
606; GFX1064: v_cmp_eq_u32_e32 vcc, 1, v{{[0-9]+}}
607; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
608; GFX1064: v_cmp_ne_u32_e32 vcc, 2, v{{[0-9]+}}
609; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc
610; GFX1064: v_cmp_ne_u32_e32 vcc, 3, v{{[0-9]+}}
611; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc
612define amdgpu_kernel void @test_movrels_extract_neg_offset_vgpr(i32 addrspace(1)* %out) #0 {
613entry:
614  %id = call i32 @llvm.amdgcn.workitem.id.x() #1
615  %index = add i32 %id, -512
616  %value = extractelement <4 x i32> <i32 0, i32 1, i32 2, i32 3>, i32 %index
617  store i32 %value, i32 addrspace(1)* %out
618  ret void
619}
620
621; GCN-LABEL: {{^}}test_set_inactive:
622; GFX1032: s_not_b32 exec_lo, exec_lo
623; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 42
624; GFX1032: s_not_b32 exec_lo, exec_lo
625; GFX1064: s_not_b64 exec, exec{{$}}
626; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 42
627; GFX1064: s_not_b64 exec, exec{{$}}
628define amdgpu_kernel void @test_set_inactive(i32 addrspace(1)* %out, i32 %in) #0 {
629  %tmp = call i32 @llvm.amdgcn.set.inactive.i32(i32 %in, i32 42)
630  store i32 %tmp, i32 addrspace(1)* %out
631  ret void
632}
633
634; GCN-LABEL: {{^}}test_set_inactive_64:
635; GFX1032: s_not_b32 exec_lo, exec_lo
636; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
637; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
638; GFX1032: s_not_b32 exec_lo, exec_lo
639; GFX1064: s_not_b64 exec, exec{{$}}
640; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
641; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
642; GFX1064: s_not_b64 exec, exec{{$}}
643define amdgpu_kernel void @test_set_inactive_64(i64 addrspace(1)* %out, i64 %in) #0 {
644  %tmp = call i64 @llvm.amdgcn.set.inactive.i64(i64 %in, i64 0)
645  store i64 %tmp, i64 addrspace(1)* %out
646  ret void
647}
648
649; GCN-LABEL: {{^}}test_kill_i1_terminator_float:
650; GFX1032: s_mov_b32 exec_lo, 0
651; GFX1064: s_mov_b64 exec, 0
652define amdgpu_ps void @test_kill_i1_terminator_float() #0 {
653  call void @llvm.amdgcn.kill(i1 false)
654  ret void
655}
656
657; GCN-LABEL: {{^}}test_kill_i1_terminator_i1:
658; GFX1032: s_or_b32 [[OR:s[0-9]+]],
659; GFX1032: s_and_b32 exec_lo, exec_lo, [[OR]]
660; GFX1064: s_or_b64 [[OR:s\[[0-9:]+\]]],
661; GFX1064: s_and_b64 exec, exec, [[OR]]
662define amdgpu_gs void @test_kill_i1_terminator_i1(i32 %a, i32 %b, i32 %c, i32 %d) #0 {
663  %c1 = icmp slt i32 %a, %b
664  %c2 = icmp slt i32 %c, %d
665  %x = or i1 %c1, %c2
666  call void @llvm.amdgcn.kill(i1 %x)
667  ret void
668}
669
670; GCN-LABEL: {{^}}test_loop_vcc:
671; GFX1032: v_cmp_lt_f32_e32 vcc_lo,
672; GFX1064: v_cmp_lt_f32_e32 vcc,
673; GCN: s_cbranch_vccnz
674define amdgpu_ps <4 x float> @test_loop_vcc(<4 x float> %in) #0 {
675entry:
676  br label %loop
677
678loop:
679  %ctr.iv = phi float [ 0.0, %entry ], [ %ctr.next, %body ]
680  %c.iv = phi <4 x float> [ %in, %entry ], [ %c.next, %body ]
681  %cc = fcmp ogt float %ctr.iv, 7.0
682  br i1 %cc, label %break, label %body
683
684body:
685  %c.iv0 = extractelement <4 x float> %c.iv, i32 0
686  %c.next = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %c.iv0, <8 x i32> undef, <4 x i32> undef, i1 0, i32 0, i32 0)
687  %ctr.next = fadd float %ctr.iv, 2.0
688  br label %loop
689
690break:
691  ret <4 x float> %c.iv
692}
693
694; GCN-LABEL: {{^}}test_wwm1:
695; GFX1032: s_or_saveexec_b32 [[SAVE:s[0-9]+]], -1
696; GFX1032: s_mov_b32 exec_lo, [[SAVE]]
697; GFX1064: s_or_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], -1
698; GFX1064: s_mov_b64 exec, [[SAVE]]
699define amdgpu_ps float @test_wwm1(i32 inreg %idx0, i32 inreg %idx1, float %src0, float %src1) {
700main_body:
701  %out = fadd float %src0, %src1
702  %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
703  ret float %out.0
704}
705
706; GCN-LABEL: {{^}}test_wwm2:
707; GFX1032: v_cmp_gt_u32_e32 vcc_lo, 32, v{{[0-9]+}}
708; GFX1032: s_and_saveexec_b32 [[SAVE1:s[0-9]+]], vcc_lo
709; GFX1032: s_or_saveexec_b32 [[SAVE2:s[0-9]+]], -1
710; GFX1032: s_mov_b32 exec_lo, [[SAVE2]]
711; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE1]]
712; GFX1064: v_cmp_gt_u32_e32 vcc, 32, v{{[0-9]+}}
713; GFX1064: s_and_saveexec_b64 [[SAVE1:s\[[0-9:]+\]]], vcc{{$}}
714; GFX1064: s_or_saveexec_b64 [[SAVE2:s\[[0-9:]+\]]], -1
715; GFX1064: s_mov_b64 exec, [[SAVE2]]
716; GFX1064: s_or_b64 exec, exec, [[SAVE1]]
717define amdgpu_ps float @test_wwm2(i32 inreg %idx) {
718main_body:
719  ; use mbcnt to make sure the branch is divergent
720  %lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
721  %hi = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
722  %cc = icmp uge i32 %hi, 32
723  br i1 %cc, label %endif, label %if
724
725if:
726  %src = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> undef, i32 %idx, i32 0, i1 0, i1 0)
727  %out = fadd float %src, %src
728  %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
729  %out.1 = fadd float %src, %out.0
730  br label %endif
731
732endif:
733  %out.2 = phi float [ %out.1, %if ], [ 0.0, %main_body ]
734  ret float %out.2
735}
736
737; GCN-LABEL: {{^}}test_wqm1:
738; GFX1032: s_mov_b32 [[ORIG:s[0-9]+]], exec_lo
739; GFX1032: s_wqm_b32 exec_lo, exec_lo
740; GFX1032: s_and_b32 exec_lo, exec_lo, [[ORIG]]
741; GFX1064: s_mov_b64 [[ORIG:s\[[0-9]+:[0-9]+\]]], exec{{$}}
742; GFX1064: s_wqm_b64 exec, exec{{$}}
743; GFX1064: s_and_b64 exec, exec, [[ORIG]]
744define amdgpu_ps <4 x float> @test_wqm1(i32 inreg, i32 inreg, i32 inreg, i32 inreg %m0, <8 x i32> inreg %rsrc, <4 x i32> inreg %sampler, <2 x float> %pos) #0 {
745main_body:
746  %inst23 = extractelement <2 x float> %pos, i32 0
747  %inst24 = extractelement <2 x float> %pos, i32 1
748  %inst25 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 0, i32 0, i32 %m0)
749  %inst26 = tail call float @llvm.amdgcn.interp.p2(float %inst25, float %inst24, i32 0, i32 0, i32 %m0)
750  %inst28 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 1, i32 0, i32 %m0)
751  %inst29 = tail call float @llvm.amdgcn.interp.p2(float %inst28, float %inst24, i32 1, i32 0, i32 %m0)
752  %tex = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %inst26, float %inst29, <8 x i32> %rsrc, <4 x i32> %sampler, i1 0, i32 0, i32 0)
753  ret <4 x float> %tex
754}
755
756; GCN-LABEL: {{^}}test_wqm2:
757; GFX1032: s_wqm_b32 exec_lo, exec_lo
758; GFX1032: s_and_b32 exec_lo, exec_lo, s{{[0-9+]}}
759; GFX1064: s_wqm_b64 exec, exec{{$}}
760; GFX1064: s_and_b64 exec, exec, s[{{[0-9:]+}}]
761define amdgpu_ps float @test_wqm2(i32 inreg %idx0, i32 inreg %idx1) #0 {
762main_body:
763  %src0 = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> undef, i32 %idx0, i32 0, i1 0, i1 0)
764  %src1 = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> undef, i32 %idx1, i32 0, i1 0, i1 0)
765  %out = fadd float %src0, %src1
766  %out.0 = bitcast float %out to i32
767  %out.1 = call i32 @llvm.amdgcn.wqm.i32(i32 %out.0)
768  %out.2 = bitcast i32 %out.1 to float
769  ret float %out.2
770}
771
772; GCN-LABEL: {{^}}test_intr_fcmp_i64:
773; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
774; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
775; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
776; GFX1064:     v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
777; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
778; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
779; GCN:         store_dwordx2 v[{{[0-9:]+}}], v{{\[}}[[V_LO]]:[[V_HI]]],
780define amdgpu_kernel void @test_intr_fcmp_i64(i64 addrspace(1)* %out, float %src, float %a) {
781  %temp = call float @llvm.fabs.f32(float %a)
782  %result = call i64 @llvm.amdgcn.fcmp.i64.f32(float %src, float %temp, i32 1)
783  store i64 %result, i64 addrspace(1)* %out
784  ret void
785}
786
787; GCN-LABEL: {{^}}test_intr_icmp_i64:
788; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
789; GFX1032-DAG: v_cmp_eq_u32_e64 [[C_LO:vcc_lo|s[0-9]+]], 0x64, {{s[0-9]+}}
790; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], [[C_LO]]
791; GFX1064:     v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], 0x64, {{s[0-9]+}}
792; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
793; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
794; GCN:         store_dwordx2 v[{{[0-9:]+}}], v{{\[}}[[V_LO]]:[[V_HI]]],
795define amdgpu_kernel void @test_intr_icmp_i64(i64 addrspace(1)* %out, i32 %src) {
796  %result = call i64 @llvm.amdgcn.icmp.i64.i32(i32 %src, i32 100, i32 32)
797  store i64 %result, i64 addrspace(1)* %out
798  ret void
799}
800
801; GCN-LABEL: {{^}}test_intr_fcmp_i32:
802; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
803; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
804; GFX1064:     v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
805; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
806; GCN:         store_dword v[{{[0-9:]+}}], v[[V_LO]],
807define amdgpu_kernel void @test_intr_fcmp_i32(i32 addrspace(1)* %out, float %src, float %a) {
808  %temp = call float @llvm.fabs.f32(float %a)
809  %result = call i32 @llvm.amdgcn.fcmp.i32.f32(float %src, float %temp, i32 1)
810  store i32 %result, i32 addrspace(1)* %out
811  ret void
812}
813
814; GCN-LABEL: {{^}}test_intr_icmp_i32:
815; GFX1032-DAG: v_cmp_eq_u32_e64 s[[C_LO:[0-9]+]], 0x64, {{s[0-9]+}}
816; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
817; GFX1064:     v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:{{[0-9]+}}], 0x64, {{s[0-9]+}}
818; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
819; GCN:         store_dword v[{{[0-9:]+}}], v[[V_LO]],
820define amdgpu_kernel void @test_intr_icmp_i32(i32 addrspace(1)* %out, i32 %src) {
821  %result = call i32 @llvm.amdgcn.icmp.i32.i32(i32 %src, i32 100, i32 32)
822  store i32 %result, i32 addrspace(1)* %out
823  ret void
824}
825
826; GCN-LABEL: {{^}}test_wqm_vote:
827; GFX1032: v_cmp_neq_f32_e32 vcc_lo, 0
828; GFX1032: s_wqm_b32 [[WQM:s[0-9]+]], vcc_lo
829; GFX1032: s_and_b32 exec_lo, exec_lo, [[WQM]]
830; GFX1064: v_cmp_neq_f32_e32 vcc, 0
831; GFX1064: s_wqm_b64 [[WQM:s\[[0-9:]+\]]], vcc{{$}}
832; GFX1064: s_and_b64 exec, exec, [[WQM]]
833define amdgpu_ps void @test_wqm_vote(float %a) {
834  %c1 = fcmp une float %a, 0.0
835  %c2 = call i1 @llvm.amdgcn.wqm.vote(i1 %c1)
836  call void @llvm.amdgcn.kill(i1 %c2)
837  ret void
838}
839
840; GCN-LABEL: {{^}}test_branch_true:
841; GFX1032: s_and_b32 vcc_lo, exec_lo, -1
842; GFX1064: s_and_b64 vcc, exec, -1
843define amdgpu_kernel void @test_branch_true() #2 {
844entry:
845  br i1 true, label %for.end, label %for.body.lr.ph
846
847for.body.lr.ph:                                   ; preds = %entry
848  br label %for.body
849
850for.body:                                         ; preds = %for.body, %for.body.lr.ph
851  br i1 undef, label %for.end, label %for.body
852
853for.end:                                          ; preds = %for.body, %entry
854  ret void
855}
856
857; GCN-LABEL: {{^}}test_ps_live:
858; GFX1032: s_mov_b32 [[C:s[0-9]+]], exec_lo
859; GFX1064: s_mov_b64 [[C:s\[[0-9:]+\]]], exec{{$}}
860; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[C]]
861define amdgpu_ps float @test_ps_live() #0 {
862  %live = call i1 @llvm.amdgcn.ps.live()
863  %live.32 = zext i1 %live to i32
864  %r = bitcast i32 %live.32 to float
865  ret float %r
866}
867
868; GCN-LABEL: {{^}}test_vccnz_ifcvt_triangle64:
869; GFX1032: v_cmp_neq_f64_e64 [[C:s[0-9]+]], s[{{[0-9:]+}}], 1.0
870; GFX1032: s_and_b32 vcc_lo, exec_lo, [[C]]
871; GFX1064: v_cmp_neq_f64_e64 [[C:s\[[0-9:]+\]]], s[{{[0-9:]+}}], 1.0
872; GFX1064: s_and_b64 vcc, exec, [[C]]
873define amdgpu_kernel void @test_vccnz_ifcvt_triangle64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
874entry:
875  %v = load double, double addrspace(1)* %in
876  %cc = fcmp oeq double %v, 1.000000e+00
877  br i1 %cc, label %if, label %endif
878
879if:
880  %u = fadd double %v, %v
881  br label %endif
882
883endif:
884  %r = phi double [ %v, %entry ], [ %u, %if ]
885  store double %r, double addrspace(1)* %out
886  ret void
887}
888
889; GCN-LABEL: {{^}}test_vgprblocks_w32_attr:
890; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
891; GFX10DEFWAVE: ; VGPRBlocks: 1
892define amdgpu_gs float @test_vgprblocks_w32_attr(float %a, float %b, float %c, float %d, float %e,
893                                        float %f, float %g, float %h, float %i, float %j, float %k, float %l) #3 {
894main_body:
895  %s = fadd float %a, %b
896  %s.1 = fadd float %s, %c
897  %s.2 = fadd float %s.1, %d
898  %s.3 = fadd float %s.2, %e
899  %s.4 = fadd float %s.3, %f
900  %s.5 = fadd float %s.4, %g
901  %s.6 = fadd float %s.5, %h
902  %s.7 = fadd float %s.6, %i
903  %s.8 = fadd float %s.7, %j
904  %s.9 = fadd float %s.8, %k
905  %s.10 = fadd float %s.9, %l
906  ret float %s.10
907}
908
909; GCN-LABEL: {{^}}test_vgprblocks_w64_attr:
910; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
911; GFX10DEFWAVE: ; VGPRBlocks: 2
912define amdgpu_gs float @test_vgprblocks_w64_attr(float %a, float %b, float %c, float %d, float %e,
913                                        float %f, float %g, float %h, float %i, float %j, float %k, float %l) #4 {
914main_body:
915  %s = fadd float %a, %b
916  %s.1 = fadd float %s, %c
917  %s.2 = fadd float %s.1, %d
918  %s.3 = fadd float %s.2, %e
919  %s.4 = fadd float %s.3, %f
920  %s.5 = fadd float %s.4, %g
921  %s.6 = fadd float %s.5, %h
922  %s.7 = fadd float %s.6, %i
923  %s.8 = fadd float %s.7, %j
924  %s.9 = fadd float %s.8, %k
925  %s.10 = fadd float %s.9, %l
926  ret float %s.10
927}
928
929; GCN-LABEL: {{^}}icmp64:
930; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
931; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
932define amdgpu_kernel void @icmp64(i32 %n, i32 %s) {
933entry:
934  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
935  %mul4 = mul nsw i32 %s, %n
936  %cmp = icmp slt i32 0, %mul4
937  br label %if.end
938
939if.end:                                           ; preds = %entry
940  %rem = urem i32 %id, %s
941  %icmp = tail call i64 @llvm.amdgcn.icmp.i64.i32(i32 %rem, i32 0, i32 32)
942  %shr = lshr i64 %icmp, 1
943  %notmask = shl nsw i64 -1, 0
944  %and = and i64 %notmask, %shr
945  %or = or i64 %and, -9223372036854775808
946  %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
947  %cast = trunc i64 %cttz to i32
948  %cmp3 = icmp ugt i32 10, %cast
949  %cmp6 = icmp ne i32 %rem, 0
950  %brmerge = or i1 %cmp6, %cmp3
951  br i1 %brmerge, label %if.end2, label %if.then
952
953if.then:                                          ; preds = %if.end
954  unreachable
955
956if.end2:                                          ; preds = %if.end
957  ret void
958}
959
960; GCN-LABEL: {{^}}fcmp64:
961; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
962; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
963define amdgpu_kernel void @fcmp64(float %n, float %s) {
964entry:
965  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
966  %id.f = uitofp i32 %id to float
967  %mul4 = fmul float %s, %n
968  %cmp = fcmp ult float 0.0, %mul4
969  br label %if.end
970
971if.end:                                           ; preds = %entry
972  %rem.f = frem float %id.f, %s
973  %fcmp = tail call i64 @llvm.amdgcn.fcmp.i64.f32(float %rem.f, float 0.0, i32 1)
974  %shr = lshr i64 %fcmp, 1
975  %notmask = shl nsw i64 -1, 0
976  %and = and i64 %notmask, %shr
977  %or = or i64 %and, -9223372036854775808
978  %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
979  %cast = trunc i64 %cttz to i32
980  %cmp3 = icmp ugt i32 10, %cast
981  %cmp6 = fcmp one float %rem.f, 0.0
982  %brmerge = or i1 %cmp6, %cmp3
983  br i1 %brmerge, label %if.end2, label %if.then
984
985if.then:                                          ; preds = %if.end
986  unreachable
987
988if.end2:                                          ; preds = %if.end
989  ret void
990}
991
992; GCN-LABEL: {{^}}icmp32:
993; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
994; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
995define amdgpu_kernel void @icmp32(i32 %n, i32 %s) {
996entry:
997  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
998  %mul4 = mul nsw i32 %s, %n
999  %cmp = icmp slt i32 0, %mul4
1000  br label %if.end
1001
1002if.end:                                           ; preds = %entry
1003  %rem = urem i32 %id, %s
1004  %icmp = tail call i32 @llvm.amdgcn.icmp.i32.i32(i32 %rem, i32 0, i32 32)
1005  %shr = lshr i32 %icmp, 1
1006  %notmask = shl nsw i32 -1, 0
1007  %and = and i32 %notmask, %shr
1008  %or = or i32 %and, 2147483648
1009  %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1010  %cmp3 = icmp ugt i32 10, %cttz
1011  %cmp6 = icmp ne i32 %rem, 0
1012  %brmerge = or i1 %cmp6, %cmp3
1013  br i1 %brmerge, label %if.end2, label %if.then
1014
1015if.then:                                          ; preds = %if.end
1016  unreachable
1017
1018if.end2:                                          ; preds = %if.end
1019  ret void
1020}
1021
1022; GCN-LABEL: {{^}}fcmp32:
1023; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
1024; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
1025define amdgpu_kernel void @fcmp32(float %n, float %s) {
1026entry:
1027  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
1028  %id.f = uitofp i32 %id to float
1029  %mul4 = fmul float %s, %n
1030  %cmp = fcmp ult float 0.0, %mul4
1031  br label %if.end
1032
1033if.end:                                           ; preds = %entry
1034  %rem.f = frem float %id.f, %s
1035  %fcmp = tail call i32 @llvm.amdgcn.fcmp.i32.f32(float %rem.f, float 0.0, i32 1)
1036  %shr = lshr i32 %fcmp, 1
1037  %notmask = shl nsw i32 -1, 0
1038  %and = and i32 %notmask, %shr
1039  %or = or i32 %and, 2147483648
1040  %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1041  %cmp3 = icmp ugt i32 10, %cttz
1042  %cmp6 = fcmp one float %rem.f, 0.0
1043  %brmerge = or i1 %cmp6, %cmp3
1044  br i1 %brmerge, label %if.end2, label %if.then
1045
1046if.then:                                          ; preds = %if.end
1047  unreachable
1048
1049if.end2:                                          ; preds = %if.end
1050  ret void
1051}
1052
1053declare void @external_void_func_void() #1
1054
1055; Test save/restore of VGPR needed for SGPR spilling.
1056
1057; GCN-LABEL: {{^}}callee_no_stack_with_call:
1058; GCN: s_waitcnt
1059; GCN-NEXT: s_waitcnt_vscnt
1060
1061; GFX1064-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1062; GFX1032-NEXT: s_or_saveexec_b32 [[COPY_EXEC0:s[0-9]]], -1{{$}}
1063; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s32 ; 4-byte Folded Spill
1064; GCN-NEXT: v_nop
1065; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
1066; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC0]]
1067
1068; GCN-NEXT: v_writelane_b32 v32, s34, 2
1069; GCN: s_mov_b32 s34, s32
1070; GFX1064: s_add_u32 s32, s32, 0x400
1071; GFX1032: s_add_u32 s32, s32, 0x200
1072
1073
1074; GCN-DAG: v_writelane_b32 v32, s30, 0
1075; GCN-DAG: v_writelane_b32 v32, s31, 1
1076; GCN: s_swappc_b64
1077; GCN-DAG: v_readlane_b32 s4, v32, 0
1078; GCN-DAG: v_readlane_b32 s5, v32, 1
1079
1080
1081; GFX1064: s_sub_u32 s32, s32, 0x400
1082; GFX1032: s_sub_u32 s32, s32, 0x200
1083; GCN: v_readlane_b32 s34, v32, 2
1084; GFX1064: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1085; GFX1032: s_or_saveexec_b32 [[COPY_EXEC1:s[0-9]]], -1{{$}}
1086; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s32 ; 4-byte Folded Reload
1087; GCN-NEXT: v_nop
1088; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
1089; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC1]]
1090; GCN-NEXT: s_waitcnt vmcnt(0)
1091; GCN-NEXT: s_setpc_b64
1092define void @callee_no_stack_with_call() #1 {
1093  call void @external_void_func_void()
1094  ret void
1095}
1096
1097
1098declare i32 @llvm.amdgcn.workitem.id.x()
1099declare float @llvm.fabs.f32(float)
1100declare { float, i1 } @llvm.amdgcn.div.scale.f32(float, float, i1)
1101declare { double, i1 } @llvm.amdgcn.div.scale.f64(double, double, i1)
1102declare float @llvm.amdgcn.div.fmas.f32(float, float, float, i1)
1103declare double @llvm.amdgcn.div.fmas.f64(double, double, double, i1)
1104declare i1 @llvm.amdgcn.class.f32(float, i32)
1105declare i32 @llvm.amdgcn.set.inactive.i32(i32, i32)
1106declare i64 @llvm.amdgcn.set.inactive.i64(i64, i64)
1107declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1108declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1109declare float @llvm.amdgcn.wwm.f32(float)
1110declare i32 @llvm.amdgcn.wqm.i32(i32)
1111declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32)
1112declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32)
1113declare float @llvm.amdgcn.buffer.load.f32(<4 x i32>, i32, i32, i1, i1)
1114declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32)
1115declare i32 @llvm.amdgcn.mbcnt.hi(i32, i32)
1116declare i64 @llvm.amdgcn.fcmp.i64.f32(float, float, i32)
1117declare i64 @llvm.amdgcn.icmp.i64.i32(i32, i32, i32)
1118declare i32 @llvm.amdgcn.fcmp.i32.f32(float, float, i32)
1119declare i32 @llvm.amdgcn.icmp.i32.i32(i32, i32, i32)
1120declare void @llvm.amdgcn.kill(i1)
1121declare i1 @llvm.amdgcn.wqm.vote(i1)
1122declare i1 @llvm.amdgcn.ps.live()
1123declare i64 @llvm.cttz.i64(i64, i1)
1124declare i32 @llvm.cttz.i32(i32, i1)
1125
1126attributes #0 = { nounwind readnone speculatable }
1127attributes #1 = { nounwind }
1128attributes #2 = { nounwind readnone optnone noinline }
1129attributes #3 = { "target-features"="+wavefrontsize32" }
1130attributes #4 = { "target-features"="+wavefrontsize64" }
1131