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:vcc_lo|s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
48; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]]
49; GFX1064: v_cmp_le_f16_sdwa [[SC:vcc|s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD
50; GFX1064: v_cndmask_b32_e32 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{{$}}
154; GCN: s_cbranch_execz
155define amdgpu_kernel void @test_mask_if(i32 addrspace(1)* %arg) #0 {
156  %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
157  %cmp = icmp ugt i32 %lid, 10
158  br i1 %cmp, label %if, label %endif
159
160if:
161  store i32 0, i32 addrspace(1)* %arg, align 4
162  br label %endif
163
164endif:
165  ret void
166}
167
168; GCN-LABEL: {{^}}test_loop_with_if:
169; GFX1032: s_or_b32 s{{[0-9]+}}, vcc_lo, s{{[0-9]+}}
170; GFX1032: s_andn2_b32 exec_lo, exec_lo, s{{[0-9]+}}
171; GFX1064: s_or_b64 s[{{[0-9:]+}}], vcc, s[{{[0-9:]+}}]
172; GFX1064: s_andn2_b64 exec, exec, s[{{[0-9:]+}}]
173; GCN:     s_cbranch_execz
174; GCN:   BB{{.*}}:
175; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
176; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
177; GCN:     s_cbranch_execz
178; GCN:   ; %bb.{{[0-9]+}}:
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:   ; %bb.{{[0-9]+}}:
183; GCN:   ; %bb.{{[0-9]+}}:
184; GFX1032: s_or_b32 exec_lo, exec_lo, s{{[0-9]+}}
185; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, s{{[0-9]+}}
186; GFX1064: s_or_b64 exec, exec, s[{{[0-9:]+}}]
187; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}]{{$}}
188; GCN:     s_cbranch_execz BB
189; GCN:   ; %bb.{{[0-9]+}}:
190; GCN:   BB{{.*}}:
191; GCN:     s_endpgm
192define amdgpu_kernel void @test_loop_with_if(i32 addrspace(1)* %arg) #0 {
193bb:
194  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
195  br label %bb2
196
197bb1:
198  ret void
199
200bb2:
201  %tmp3 = phi i32 [ 0, %bb ], [ %tmp15, %bb13 ]
202  %tmp4 = icmp slt i32 %tmp3, %tmp
203  br i1 %tmp4, label %bb5, label %bb11
204
205bb5:
206  %tmp6 = sext i32 %tmp3 to i64
207  %tmp7 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp6
208  %tmp8 = load i32, i32 addrspace(1)* %tmp7, align 4
209  %tmp9 = icmp sgt i32 %tmp8, 10
210  br i1 %tmp9, label %bb10, label %bb11
211
212bb10:
213  store i32 %tmp, i32 addrspace(1)* %tmp7, align 4
214  br label %bb13
215
216bb11:
217  %tmp12 = sdiv i32 %tmp3, 2
218  br label %bb13
219
220bb13:
221  %tmp14 = phi i32 [ %tmp3, %bb10 ], [ %tmp12, %bb11 ]
222  %tmp15 = add nsw i32 %tmp14, 1
223  %tmp16 = icmp slt i32 %tmp14, 255
224  br i1 %tmp16, label %bb2, label %bb1
225}
226
227; GCN-LABEL: {{^}}test_loop_with_if_else_break:
228; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
229; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
230; GCN:     s_cbranch_execz
231; GCN:   ; %bb.{{[0-9]+}}: ; %.preheader
232; GCN:   BB{{.*}}:
233
234; GFX1032: s_or_b32 [[MASK0:s[0-9]+]], [[MASK0]], vcc_lo
235; GFX1064: s_or_b64 [[MASK0:s\[[0-9:]+\]]], [[MASK0]], vcc
236; GCN:     global_store_dword
237; GFX1032: s_andn2_b32 [[MASK1:s[0-9]+]], [[MASK1]], exec_lo
238; GFX1064: s_andn2_b64 [[MASK1:s\[[0-9:]+\]]], [[MASK1]], exec
239; GFX1032: s_and_b32 [[MASK0]], [[MASK0]], exec_lo
240; GFX1064: s_and_b64 [[MASK0]], [[MASK0]], exec
241; GFX1032: s_or_b32 [[MASK1]], [[MASK1]], [[MASK0]]
242; GFX1064: s_or_b64 [[MASK1]], [[MASK1]], [[MASK0]]
243; GCN:   BB{{.*}}: ; %Flow
244; GFX1032: s_and_b32 [[TMP0:s[0-9]+]], exec_lo, [[MASK1]]
245; GFX1064: s_and_b64 [[TMP0:s\[[0-9:]+\]]], exec, [[MASK1]]
246; GFX1032: s_or_b32  [[ACC:s[0-9]+]], [[TMP0]], [[ACC]]
247; GFX1064: s_or_b64  [[ACC:s\[[0-9:]+\]]], [[TMP0]], [[ACC]]
248; GFX1032: s_andn2_b32 exec_lo, exec_lo, [[ACC]]
249; GFX1064: s_andn2_b64 exec, exec, [[ACC]]
250; GCN:     s_cbranch_execz
251; GCN:   BB{{.*}}:
252; GCN: s_load_dword [[LOAD:s[0-9]+]]
253; GFX1032: s_or_b32 [[MASK1]], [[MASK1]], exec_lo
254; GFX1064: s_or_b64 [[MASK1]], [[MASK1]], exec
255; GCN: s_cmp_lt_i32 [[LOAD]], 11
256define amdgpu_kernel void @test_loop_with_if_else_break(i32 addrspace(1)* %arg) #0 {
257bb:
258  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
259  %tmp1 = icmp eq i32 %tmp, 0
260  br i1 %tmp1, label %.loopexit, label %.preheader
261
262.preheader:
263  br label %bb2
264
265bb2:
266  %tmp3 = phi i32 [ %tmp9, %bb8 ], [ 0, %.preheader ]
267  %tmp4 = zext i32 %tmp3 to i64
268  %tmp5 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp4
269  %tmp6 = load i32, i32 addrspace(1)* %tmp5, align 4
270  %tmp7 = icmp sgt i32 %tmp6, 10
271  br i1 %tmp7, label %bb8, label %.loopexit
272
273bb8:
274  store i32 %tmp, i32 addrspace(1)* %tmp5, align 4
275  %tmp9 = add nuw nsw i32 %tmp3, 1
276  %tmp10 = icmp ult i32 %tmp9, 256
277  %tmp11 = icmp ult i32 %tmp9, %tmp
278  %tmp12 = and i1 %tmp10, %tmp11
279  br i1 %tmp12, label %bb2, label %.loopexit
280
281.loopexit:
282  ret void
283}
284
285; GCN-LABEL: {{^}}test_addc_vop2b:
286; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, s{{[0-9]+}}
287; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}, vcc_lo
288; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, s{{[0-9]+}}
289; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}, vcc{{$}}
290define amdgpu_kernel void @test_addc_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
291bb:
292  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
293  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
294  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
295  %tmp5 = add nsw i64 %tmp4, %arg1
296  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
297  ret void
298}
299
300; GCN-LABEL: {{^}}test_subbrev_vop2b:
301; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s[0-9]+|vcc_lo]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
302; GFX1032: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, {{[vs][0-9]+}}, [[A0]]{{$}}
303; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
304; GFX1064: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, {{[vs][0-9]+}}, [[A0]]{{$}}
305define amdgpu_kernel void @test_subbrev_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
306bb:
307  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
308  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
309  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
310  %tmp5 = sub nsw i64 %tmp4, %arg1
311  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
312  ret void
313}
314
315; GCN-LABEL: {{^}}test_subb_vop2b:
316; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s[0-9]+|vcc_lo]], s{{[0-9]+}}, v{{[0-9]+}}{{$}}
317; GFX1032: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, v{{[0-9]+}}, [[A0]]{{$}}
318; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], s{{[0-9]+}}, v{{[0-9]+}}{{$}}
319; GFX1064: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, [[A0]]{{$}}
320define amdgpu_kernel void @test_subb_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
321bb:
322  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
323  %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
324  %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
325  %tmp5 = sub nsw i64 %arg1, %tmp4
326  store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
327  ret void
328}
329
330; GCN-LABEL: {{^}}test_udiv64:
331; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, [[SDST:s[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}}
332; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
333; GFX1032: v_add_co_ci_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}, [[SDST]]
334; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
335; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
336; GFX1032: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
337; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
338; GFX1032: v_sub_co_u32_e64 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
339; GFX1032: v_subrev_co_ci_u32_e64 v{{[0-9]+}}, s{{[0-9]+}}, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
340; GFX1032: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
341; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, [[SDST:s\[[0-9:]+\]]], v{{[0-9]+}}, v{{[0-9]+}}
342; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
343; GFX1064: v_add_co_ci_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}, [[SDST]]
344; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
345; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
346; GFX1064: v_add_co_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
347; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
348; GFX1064: v_sub_co_u32_e64 v{{[0-9]+}}, s[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}
349; GFX1064: v_subrev_co_ci_u32_e64 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}]
350; GFX1064: v_sub_co_ci_u32_e64 v{{[0-9]+}}, s[{{[0-9:]+}}], {{[vs][0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}]
351define amdgpu_kernel void @test_udiv64(i64 addrspace(1)* %arg) #0 {
352bb:
353  %tmp = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 1
354  %tmp1 = load i64, i64 addrspace(1)* %tmp, align 8
355  %tmp2 = load i64, i64 addrspace(1)* %arg, align 8
356  %tmp3 = udiv i64 %tmp1, %tmp2
357  %tmp4 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 2
358  store i64 %tmp3, i64 addrspace(1)* %tmp4, align 8
359  ret void
360}
361
362; GCN-LABEL: {{^}}test_div_scale_f32:
363; GFX1032: v_div_scale_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
364; GFX1064: v_div_scale_f32 v{{[0-9]+}}, s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
365define amdgpu_kernel void @test_div_scale_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
366  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
367  %gep.0 = getelementptr float, float addrspace(1)* %in, i32 %tid
368  %gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
369
370  %a = load volatile float, float addrspace(1)* %gep.0, align 4
371  %b = load volatile float, float addrspace(1)* %gep.1, align 4
372
373  %result = call { float, i1 } @llvm.amdgcn.div.scale.f32(float %a, float %b, i1 false) nounwind readnone
374  %result0 = extractvalue { float, i1 } %result, 0
375  store float %result0, float addrspace(1)* %out, align 4
376  ret void
377}
378
379; GCN-LABEL: {{^}}test_div_scale_f64:
380; GFX1032: v_div_scale_f64 v[{{[0-9:]+}}], s{{[0-9]+}}, v[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
381; GFX1064: v_div_scale_f64 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
382define amdgpu_kernel void @test_div_scale_f64(double addrspace(1)* %out, double addrspace(1)* %aptr, double addrspace(1)* %in) #0 {
383  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
384  %gep.0 = getelementptr double, double addrspace(1)* %in, i32 %tid
385  %gep.1 = getelementptr double, double addrspace(1)* %gep.0, i32 1
386
387  %a = load volatile double, double addrspace(1)* %gep.0, align 8
388  %b = load volatile double, double addrspace(1)* %gep.1, align 8
389
390  %result = call { double, i1 } @llvm.amdgcn.div.scale.f64(double %a, double %b, i1 true) nounwind readnone
391  %result0 = extractvalue { double, i1 } %result, 0
392  store double %result0, double addrspace(1)* %out, align 8
393  ret void
394}
395
396; GCN-LABEL: {{^}}test_mad_i64_i32:
397; GFX1032: v_mad_i64_i32 v[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
398; GFX1064: v_mad_i64_i32 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
399define i64 @test_mad_i64_i32(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
400  %sext0 = sext i32 %arg0 to i64
401  %sext1 = sext i32 %arg1 to i64
402  %mul = mul i64 %sext0, %sext1
403  %mad = add i64 %mul, %arg2
404  ret i64 %mad
405}
406
407; GCN-LABEL: {{^}}test_mad_u64_u32:
408; GFX1032: v_mad_u64_u32 v[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
409; GFX1064: v_mad_u64_u32 v[{{[0-9:]+}}], s[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, v[{{[0-9:]+}}]
410define i64 @test_mad_u64_u32(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
411  %sext0 = zext i32 %arg0 to i64
412  %sext1 = zext i32 %arg1 to i64
413  %mul = mul i64 %sext0, %sext1
414  %mad = add i64 %mul, %arg2
415  ret i64 %mad
416}
417
418; GCN-LABEL: {{^}}test_div_fmas_f32:
419; GFX1032: v_cmp_eq_u32_e64 vcc_lo,
420; GFX1064: v_cmp_eq_u32_e64 vcc,
421; GCN:     v_div_fmas_f32 v{{[0-9]+}}, {{[vs][0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
422define amdgpu_kernel void @test_div_fmas_f32(float addrspace(1)* %out, float %a, float %b, float %c, i1 %d) nounwind {
423  %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %d) nounwind readnone
424  store float %result, float addrspace(1)* %out, align 4
425  ret void
426}
427
428; GCN-LABEL: {{^}}test_div_fmas_f64:
429; GFX1032: v_cmp_eq_u32_e64 vcc_lo,
430; GFX1064: v_cmp_eq_u32_e64 vcc,
431; GCN-DAG: v_div_fmas_f64 v[{{[0-9:]+}}], {{[vs]}}[{{[0-9:]+}}], v[{{[0-9:]+}}], v[{{[0-9:]+}}]
432define amdgpu_kernel void @test_div_fmas_f64(double addrspace(1)* %out, double %a, double %b, double %c, i1 %d) nounwind {
433  %result = call double @llvm.amdgcn.div.fmas.f64(double %a, double %b, double %c, i1 %d) nounwind readnone
434  store double %result, double addrspace(1)* %out, align 8
435  ret void
436}
437
438; GCN-LABEL: {{^}}test_div_fmas_f32_i1_phi_vcc:
439; GFX1032: s_mov_b32 [[VCC:vcc_lo]], 0{{$}}
440; GFX1064: s_mov_b64 [[VCC:vcc]], 0{{$}}
441; GFX1032: s_and_saveexec_b32 [[SAVE:s[0-9]+]], s{{[0-9]+}}{{$}}
442; GFX1064: s_and_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], s[{{[0-9:]+}}]{{$}}
443
444; GCN: load_dword [[LOAD:v[0-9]+]]
445; GCN: v_cmp_ne_u32_e32 [[VCC]], 0, [[LOAD]]
446
447; GCN: BB{{[0-9_]+}}:
448; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE]]
449; GFX1064: s_or_b64 exec, exec, [[SAVE]]
450; GCN: v_div_fmas_f32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
451define amdgpu_kernel void @test_div_fmas_f32_i1_phi_vcc(float addrspace(1)* %out, float addrspace(1)* %in, i32 addrspace(1)* %dummy) #0 {
452entry:
453  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
454  %gep.out = getelementptr float, float addrspace(1)* %out, i32 2
455  %gep.a = getelementptr float, float addrspace(1)* %in, i32 %tid
456  %gep.b = getelementptr float, float addrspace(1)* %gep.a, i32 1
457  %gep.c = getelementptr float, float addrspace(1)* %gep.a, i32 2
458
459  %a = load float, float addrspace(1)* %gep.a
460  %b = load float, float addrspace(1)* %gep.b
461  %c = load float, float addrspace(1)* %gep.c
462
463  %cmp0 = icmp eq i32 %tid, 0
464  br i1 %cmp0, label %bb, label %exit
465
466bb:
467  %val = load volatile i32, i32 addrspace(1)* %dummy
468  %cmp1 = icmp ne i32 %val, 0
469  br label %exit
470
471exit:
472  %cond = phi i1 [false, %entry], [%cmp1, %bb]
473  %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %cond) nounwind readnone
474  store float %result, float addrspace(1)* %gep.out, align 4
475  ret void
476}
477
478; GCN-LABEL: {{^}}fdiv_f32:
479; GFX1032: v_div_scale_f32 v{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
480; GFX1064: v_div_scale_f32 v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
481; GCN: v_rcp_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}
482; GFX1032: v_div_scale_f32 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
483; GFX1064: v_div_scale_f32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
484
485; GCN-NOT: vcc
486; GCN: v_div_fmas_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
487define amdgpu_kernel void @fdiv_f32(float addrspace(1)* %out, float %a, float %b) #0 {
488entry:
489  %fdiv = fdiv float %a, %b
490  store float %fdiv, float addrspace(1)* %out
491  ret void
492}
493
494; GCN-LABEL: {{^}}test_br_cc_f16:
495; GFX1032:  v_cmp_nlt_f16_e32 vcc_lo,
496; GFX1032:  s_and_b32 vcc_lo, exec_lo, vcc_lo
497; GFX1064:  v_cmp_nlt_f16_e32 vcc,
498; GFX1064:  s_and_b64 vcc, exec, vcc{{$}}
499; GCN-NEXT: s_cbranch_vccnz
500define amdgpu_kernel void @test_br_cc_f16(
501    half addrspace(1)* %r,
502    half addrspace(1)* %a,
503    half addrspace(1)* %b) {
504entry:
505  %a.val = load half, half addrspace(1)* %a
506  %b.val = load half, half addrspace(1)* %b
507  %fcmp = fcmp olt half %a.val, %b.val
508  br i1 %fcmp, label %one, label %two
509
510one:
511  store half %a.val, half addrspace(1)* %r
512  ret void
513
514two:
515  store half %b.val, half addrspace(1)* %r
516  ret void
517}
518
519; GCN-LABEL: {{^}}test_brcc_i1:
520; GCN:      s_cmp_eq_u32 s{{[0-9]+}}, 0
521; GCN-NEXT: s_cbranch_scc1
522define amdgpu_kernel void @test_brcc_i1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in, i1 %val) #0 {
523  %cmp0 = icmp ne i1 %val, 0
524  br i1 %cmp0, label %store, label %end
525
526store:
527  store i32 222, i32 addrspace(1)* %out
528  ret void
529
530end:
531  ret void
532}
533
534; GCN-LABEL: {{^}}test_preserve_condition_undef_flag:
535; GFX1032-DAG: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
536; GFX1032-DAG: v_cmp_ngt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 0
537; GFX1032: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
538; GFX1032: s_or_b32 [[OR1:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
539; GFX1032: s_or_b32 [[OR2:s[0-9]+]], [[OR1]], s{{[0-9]+}}
540; GFX1032: s_and_b32 vcc_lo, exec_lo, [[OR2]]
541; GFX1064-DAG: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
542; GFX1064-DAG: v_cmp_ngt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 0
543; GFX1064: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
544; GFX1064: s_or_b64 [[OR1:s\[[0-9:]+\]]], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
545; GFX1064: s_or_b64 [[OR2:s\[[0-9:]+\]]], [[OR1]], s[{{[0-9:]+}}]
546; GFX1064: s_and_b64 vcc, exec, [[OR2]]
547; GCN:     s_cbranch_vccnz
548define amdgpu_kernel void @test_preserve_condition_undef_flag(float %arg, i32 %arg1, float %arg2) #0 {
549bb0:
550  %tmp = icmp sgt i32 %arg1, 4
551  %undef = call i1 @llvm.amdgcn.class.f32(float undef, i32 undef)
552  %tmp4 = select i1 %undef, float %arg, float 1.000000e+00
553  %tmp5 = fcmp ogt float %arg2, 0.000000e+00
554  %tmp6 = fcmp olt float %arg2, 1.000000e+00
555  %tmp7 = fcmp olt float %arg, %tmp4
556  %tmp8 = and i1 %tmp5, %tmp6
557  %tmp9 = and i1 %tmp8, %tmp7
558  br i1 %tmp9, label %bb1, label %bb2
559
560bb1:
561  store volatile i32 0, i32 addrspace(1)* undef
562  br label %bb2
563
564bb2:
565  ret void
566}
567
568; GCN-LABEL: {{^}}test_invert_true_phi_cond_break_loop:
569; GFX1032: s_xor_b32 s{{[0-9]+}}, s{{[0-9]+}}, -1
570; GFX1032: s_or_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
571; GFX1064: s_xor_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], -1
572; GFX1064: s_or_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
573define amdgpu_kernel void @test_invert_true_phi_cond_break_loop(i32 %arg) #0 {
574bb:
575  %id = call i32 @llvm.amdgcn.workitem.id.x()
576  %tmp = sub i32 %id, %arg
577  br label %bb1
578
579bb1:                                              ; preds = %Flow, %bb
580  %lsr.iv = phi i32 [ undef, %bb ], [ %tmp2, %Flow ]
581  %lsr.iv.next = add i32 %lsr.iv, 1
582  %cmp0 = icmp slt i32 %lsr.iv.next, 0
583  br i1 %cmp0, label %bb4, label %Flow
584
585bb4:                                              ; preds = %bb1
586  %load = load volatile i32, i32 addrspace(1)* undef, align 4
587  %cmp1 = icmp sge i32 %tmp, %load
588  br label %Flow
589
590Flow:                                             ; preds = %bb4, %bb1
591  %tmp2 = phi i32 [ %lsr.iv.next, %bb4 ], [ undef, %bb1 ]
592  %tmp3 = phi i1 [ %cmp1, %bb4 ], [ true, %bb1 ]
593  br i1 %tmp3, label %bb1, label %bb9
594
595bb9:                                              ; preds = %Flow
596  store volatile i32 7, i32 addrspace(3)* undef
597  ret void
598}
599
600; GCN-LABEL: {{^}}test_movrels_extract_neg_offset_vgpr:
601; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 1, v{{[0-9]+}}
602; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc_lo
603; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 2, v{{[0-9]+}}
604; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc_lo
605; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 3, v{{[0-9]+}}
606; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc_lo
607; GFX1064: v_cmp_eq_u32_e32 vcc, 1, v{{[0-9]+}}
608; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
609; GFX1064: v_cmp_ne_u32_e32 vcc, 2, v{{[0-9]+}}
610; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc
611; GFX1064: v_cmp_ne_u32_e32 vcc, 3, v{{[0-9]+}}
612; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc
613define amdgpu_kernel void @test_movrels_extract_neg_offset_vgpr(i32 addrspace(1)* %out) #0 {
614entry:
615  %id = call i32 @llvm.amdgcn.workitem.id.x() #1
616  %index = add i32 %id, -512
617  %value = extractelement <4 x i32> <i32 0, i32 1, i32 2, i32 3>, i32 %index
618  store i32 %value, i32 addrspace(1)* %out
619  ret void
620}
621
622; GCN-LABEL: {{^}}test_set_inactive:
623; GFX1032: s_not_b32 exec_lo, exec_lo
624; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 42
625; GFX1032: s_not_b32 exec_lo, exec_lo
626; GFX1064: s_not_b64 exec, exec{{$}}
627; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 42
628; GFX1064: s_not_b64 exec, exec{{$}}
629define amdgpu_kernel void @test_set_inactive(i32 addrspace(1)* %out, i32 %in) #0 {
630  %tmp = call i32 @llvm.amdgcn.set.inactive.i32(i32 %in, i32 42)
631  store i32 %tmp, i32 addrspace(1)* %out
632  ret void
633}
634
635; GCN-LABEL: {{^}}test_set_inactive_64:
636; GFX1032: s_not_b32 exec_lo, exec_lo
637; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
638; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
639; GFX1032: s_not_b32 exec_lo, exec_lo
640; GFX1064: s_not_b64 exec, exec{{$}}
641; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
642; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
643; GFX1064: s_not_b64 exec, exec{{$}}
644define amdgpu_kernel void @test_set_inactive_64(i64 addrspace(1)* %out, i64 %in) #0 {
645  %tmp = call i64 @llvm.amdgcn.set.inactive.i64(i64 %in, i64 0)
646  store i64 %tmp, i64 addrspace(1)* %out
647  ret void
648}
649
650; GCN-LABEL: {{^}}test_kill_i1_terminator_float:
651; GFX1032: s_mov_b32 exec_lo, 0
652; GFX1064: s_mov_b64 exec, 0
653define amdgpu_ps void @test_kill_i1_terminator_float() #0 {
654  call void @llvm.amdgcn.kill(i1 false)
655  ret void
656}
657
658; GCN-LABEL: {{^}}test_kill_i1_terminator_i1:
659; GFX1032: s_or_b32 [[OR:s[0-9]+]],
660; GFX1032: s_and_b32 exec_lo, exec_lo, [[OR]]
661; GFX1064: s_or_b64 [[OR:s\[[0-9:]+\]]],
662; GFX1064: s_and_b64 exec, exec, [[OR]]
663define amdgpu_gs void @test_kill_i1_terminator_i1(i32 %a, i32 %b, i32 %c, i32 %d) #0 {
664  %c1 = icmp slt i32 %a, %b
665  %c2 = icmp slt i32 %c, %d
666  %x = or i1 %c1, %c2
667  call void @llvm.amdgcn.kill(i1 %x)
668  ret void
669}
670
671; GCN-LABEL: {{^}}test_loop_vcc:
672; GFX1032: v_cmp_lt_f32_e32 vcc_lo,
673; GFX1064: v_cmp_lt_f32_e32 vcc,
674; GCN: s_cbranch_vccz
675define amdgpu_ps <4 x float> @test_loop_vcc(<4 x float> %in) #0 {
676entry:
677  br label %loop
678
679loop:
680  %ctr.iv = phi float [ 0.0, %entry ], [ %ctr.next, %body ]
681  %c.iv = phi <4 x float> [ %in, %entry ], [ %c.next, %body ]
682  %cc = fcmp ogt float %ctr.iv, 7.0
683  br i1 %cc, label %break, label %body
684
685body:
686  %c.iv0 = extractelement <4 x float> %c.iv, i32 0
687  %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)
688  %ctr.next = fadd float %ctr.iv, 2.0
689  br label %loop
690
691break:
692  ret <4 x float> %c.iv
693}
694
695; GCN-LABEL: {{^}}test_wwm1:
696; GFX1032: s_or_saveexec_b32 [[SAVE:s[0-9]+]], -1
697; GFX1032: s_mov_b32 exec_lo, [[SAVE]]
698; GFX1064: s_or_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], -1
699; GFX1064: s_mov_b64 exec, [[SAVE]]
700define amdgpu_ps float @test_wwm1(i32 inreg %idx0, i32 inreg %idx1, float %src0, float %src1) {
701main_body:
702  %out = fadd float %src0, %src1
703  %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
704  ret float %out.0
705}
706
707; GCN-LABEL: {{^}}test_wwm2:
708; GFX1032: v_cmp_gt_u32_e32 vcc_lo, 32, v{{[0-9]+}}
709; GFX1032: s_and_saveexec_b32 [[SAVE1:s[0-9]+]], vcc_lo
710; GFX1032: s_or_saveexec_b32 [[SAVE2:s[0-9]+]], -1
711; GFX1032: s_mov_b32 exec_lo, [[SAVE2]]
712; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE1]]
713; GFX1064: v_cmp_gt_u32_e32 vcc, 32, v{{[0-9]+}}
714; GFX1064: s_and_saveexec_b64 [[SAVE1:s\[[0-9:]+\]]], vcc{{$}}
715; GFX1064: s_or_saveexec_b64 [[SAVE2:s\[[0-9:]+\]]], -1
716; GFX1064: s_mov_b64 exec, [[SAVE2]]
717; GFX1064: s_or_b64 exec, exec, [[SAVE1]]
718define amdgpu_ps float @test_wwm2(i32 inreg %idx) {
719main_body:
720  ; use mbcnt to make sure the branch is divergent
721  %lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
722  %hi = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
723  %cc = icmp uge i32 %hi, 32
724  br i1 %cc, label %endif, label %if
725
726if:
727  %src = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx, i32 0, i32 0, i32 0)
728  %out = fadd float %src, %src
729  %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
730  %out.1 = fadd float %src, %out.0
731  br label %endif
732
733endif:
734  %out.2 = phi float [ %out.1, %if ], [ 0.0, %main_body ]
735  ret float %out.2
736}
737
738; GCN-LABEL: {{^}}test_wqm1:
739; GFX1032: s_mov_b32 [[ORIG:s[0-9]+]], exec_lo
740; GFX1032: s_wqm_b32 exec_lo, exec_lo
741; GFX1032: s_and_b32 exec_lo, exec_lo, [[ORIG]]
742; GFX1064: s_mov_b64 [[ORIG:s\[[0-9]+:[0-9]+\]]], exec{{$}}
743; GFX1064: s_wqm_b64 exec, exec{{$}}
744; GFX1064: s_and_b64 exec, exec, [[ORIG]]
745define 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 {
746main_body:
747  %inst23 = extractelement <2 x float> %pos, i32 0
748  %inst24 = extractelement <2 x float> %pos, i32 1
749  %inst25 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 0, i32 0, i32 %m0)
750  %inst26 = tail call float @llvm.amdgcn.interp.p2(float %inst25, float %inst24, i32 0, i32 0, i32 %m0)
751  %inst28 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 1, i32 0, i32 %m0)
752  %inst29 = tail call float @llvm.amdgcn.interp.p2(float %inst28, float %inst24, i32 1, i32 0, i32 %m0)
753  %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)
754  ret <4 x float> %tex
755}
756
757; GCN-LABEL: {{^}}test_wqm2:
758; GFX1032: s_wqm_b32 exec_lo, exec_lo
759; GFX1032: s_and_b32 exec_lo, exec_lo, s{{[0-9+]}}
760; GFX1064: s_wqm_b64 exec, exec{{$}}
761; GFX1064: s_and_b64 exec, exec, s[{{[0-9:]+}}]
762define amdgpu_ps float @test_wqm2(i32 inreg %idx0, i32 inreg %idx1) #0 {
763main_body:
764  %src0 = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx0, i32 0, i32 0, i32 0)
765  %src1 = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx1, i32 0, i32 0, i32 0)
766  %out = fadd float %src0, %src1
767  %out.0 = bitcast float %out to i32
768  %out.1 = call i32 @llvm.amdgcn.wqm.i32(i32 %out.0)
769  %out.2 = bitcast i32 %out.1 to float
770  ret float %out.2
771}
772
773; GCN-LABEL: {{^}}test_intr_fcmp_i64:
774; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
775; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
776; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
777; GFX1064:     v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
778; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
779; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
780; GCN:         store_dwordx2 v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]], s
781define amdgpu_kernel void @test_intr_fcmp_i64(i64 addrspace(1)* %out, float %src, float %a) {
782  %temp = call float @llvm.fabs.f32(float %a)
783  %result = call i64 @llvm.amdgcn.fcmp.i64.f32(float %src, float %temp, i32 1)
784  store i64 %result, i64 addrspace(1)* %out
785  ret void
786}
787
788; GCN-LABEL: {{^}}test_intr_icmp_i64:
789; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
790; GFX1032-DAG: v_cmp_eq_u32_e64 [[C_LO:vcc_lo|s[0-9]+]], 0x64, {{s[0-9]+}}
791; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], [[C_LO]]
792; GFX1064:     v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], 0x64, {{s[0-9]+}}
793; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
794; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
795; GCN:         store_dwordx2 v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]], s
796define amdgpu_kernel void @test_intr_icmp_i64(i64 addrspace(1)* %out, i32 %src) {
797  %result = call i64 @llvm.amdgcn.icmp.i64.i32(i32 %src, i32 100, i32 32)
798  store i64 %result, i64 addrspace(1)* %out
799  ret void
800}
801
802; GCN-LABEL: {{^}}test_intr_fcmp_i32:
803; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
804; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
805; GFX1064:     v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
806; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
807; GCN:         store_dword v{{[0-9]+}}, v[[V_LO]], s
808define amdgpu_kernel void @test_intr_fcmp_i32(i32 addrspace(1)* %out, float %src, float %a) {
809  %temp = call float @llvm.fabs.f32(float %a)
810  %result = call i32 @llvm.amdgcn.fcmp.i32.f32(float %src, float %temp, i32 1)
811  store i32 %result, i32 addrspace(1)* %out
812  ret void
813}
814
815; GCN-LABEL: {{^}}test_intr_icmp_i32:
816; GFX1032-DAG: v_cmp_eq_u32_e64 s[[C_LO:[0-9]+]], 0x64, {{s[0-9]+}}
817; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
818; GFX1064:     v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:{{[0-9]+}}], 0x64, {{s[0-9]+}}
819; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
820; GCN:         store_dword v{{[0-9]+}}, v[[V_LO]], s
821define amdgpu_kernel void @test_intr_icmp_i32(i32 addrspace(1)* %out, i32 %src) {
822  %result = call i32 @llvm.amdgcn.icmp.i32.i32(i32 %src, i32 100, i32 32)
823  store i32 %result, i32 addrspace(1)* %out
824  ret void
825}
826
827; GCN-LABEL: {{^}}test_wqm_vote:
828; GFX1032: v_cmp_neq_f32_e32 vcc_lo, 0
829; GFX1032: s_wqm_b32 [[WQM:s[0-9]+]], vcc_lo
830; GFX1032: s_and_b32 exec_lo, exec_lo, [[WQM]]
831; GFX1064: v_cmp_neq_f32_e32 vcc, 0
832; GFX1064: s_wqm_b64 [[WQM:s\[[0-9:]+\]]], vcc{{$}}
833; GFX1064: s_and_b64 exec, exec, [[WQM]]
834define amdgpu_ps void @test_wqm_vote(float %a) {
835  %c1 = fcmp une float %a, 0.0
836  %c2 = call i1 @llvm.amdgcn.wqm.vote(i1 %c1)
837  call void @llvm.amdgcn.kill(i1 %c2)
838  ret void
839}
840
841; GCN-LABEL: {{^}}test_branch_true:
842; GFX1032: s_mov_b32 vcc_lo, exec_lo
843; GFX1064: s_mov_b64 vcc, exec
844define amdgpu_kernel void @test_branch_true() #2 {
845entry:
846  br i1 true, label %for.end, label %for.body.lr.ph
847
848for.body.lr.ph:                                   ; preds = %entry
849  br label %for.body
850
851for.body:                                         ; preds = %for.body, %for.body.lr.ph
852  br i1 undef, label %for.end, label %for.body
853
854for.end:                                          ; preds = %for.body, %entry
855  ret void
856}
857
858; GCN-LABEL: {{^}}test_ps_live:
859; GFX1032: s_mov_b32 [[C:s[0-9]+]], exec_lo
860; GFX1064: s_mov_b64 [[C:s\[[0-9:]+\]]], exec{{$}}
861; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[C]]
862define amdgpu_ps float @test_ps_live() #0 {
863  %live = call i1 @llvm.amdgcn.ps.live()
864  %live.32 = zext i1 %live to i32
865  %r = bitcast i32 %live.32 to float
866  ret float %r
867}
868
869; GCN-LABEL: {{^}}test_vccnz_ifcvt_triangle64:
870; GFX1032: v_cmp_neq_f64_e64 [[C:s[0-9]+]], s[{{[0-9:]+}}], 1.0
871; GFX1032: s_and_b32 vcc_lo, exec_lo, [[C]]
872; GFX1064: v_cmp_neq_f64_e64 [[C:s\[[0-9:]+\]]], s[{{[0-9:]+}}], 1.0
873; GFX1064: s_and_b64 vcc, exec, [[C]]
874define amdgpu_kernel void @test_vccnz_ifcvt_triangle64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
875entry:
876  %v = load double, double addrspace(1)* %in
877  %cc = fcmp oeq double %v, 1.000000e+00
878  br i1 %cc, label %if, label %endif
879
880if:
881  %u = fadd double %v, %v
882  br label %endif
883
884endif:
885  %r = phi double [ %v, %entry ], [ %u, %if ]
886  store double %r, double addrspace(1)* %out
887  ret void
888}
889
890; GCN-LABEL: {{^}}test_vgprblocks_w32_attr:
891; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
892; GFX10DEFWAVE: ; VGPRBlocks: 1
893define amdgpu_gs float @test_vgprblocks_w32_attr(float %a, float %b, float %c, float %d, float %e,
894                                        float %f, float %g, float %h, float %i, float %j, float %k, float %l) #3 {
895main_body:
896  %s = fadd float %a, %b
897  %s.1 = fadd float %s, %c
898  %s.2 = fadd float %s.1, %d
899  %s.3 = fadd float %s.2, %e
900  %s.4 = fadd float %s.3, %f
901  %s.5 = fadd float %s.4, %g
902  %s.6 = fadd float %s.5, %h
903  %s.7 = fadd float %s.6, %i
904  %s.8 = fadd float %s.7, %j
905  %s.9 = fadd float %s.8, %k
906  %s.10 = fadd float %s.9, %l
907  ret float %s.10
908}
909
910; GCN-LABEL: {{^}}test_vgprblocks_w64_attr:
911; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
912; GFX10DEFWAVE: ; VGPRBlocks: 2
913define amdgpu_gs float @test_vgprblocks_w64_attr(float %a, float %b, float %c, float %d, float %e,
914                                        float %f, float %g, float %h, float %i, float %j, float %k, float %l) #4 {
915main_body:
916  %s = fadd float %a, %b
917  %s.1 = fadd float %s, %c
918  %s.2 = fadd float %s.1, %d
919  %s.3 = fadd float %s.2, %e
920  %s.4 = fadd float %s.3, %f
921  %s.5 = fadd float %s.4, %g
922  %s.6 = fadd float %s.5, %h
923  %s.7 = fadd float %s.6, %i
924  %s.8 = fadd float %s.7, %j
925  %s.9 = fadd float %s.8, %k
926  %s.10 = fadd float %s.9, %l
927  ret float %s.10
928}
929
930; GCN-LABEL: {{^}}icmp64:
931; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
932; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
933define amdgpu_kernel void @icmp64(i32 %n, i32 %s) {
934entry:
935  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
936  %mul4 = mul nsw i32 %s, %n
937  %cmp = icmp slt i32 0, %mul4
938  br label %if.end
939
940if.end:                                           ; preds = %entry
941  %rem = urem i32 %id, %s
942  %icmp = tail call i64 @llvm.amdgcn.icmp.i64.i32(i32 %rem, i32 0, i32 32)
943  %shr = lshr i64 %icmp, 1
944  %notmask = shl nsw i64 -1, 0
945  %and = and i64 %notmask, %shr
946  %or = or i64 %and, -9223372036854775808
947  %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
948  %cast = trunc i64 %cttz to i32
949  %cmp3 = icmp ugt i32 10, %cast
950  %cmp6 = icmp ne i32 %rem, 0
951  %brmerge = or i1 %cmp6, %cmp3
952  br i1 %brmerge, label %if.end2, label %if.then
953
954if.then:                                          ; preds = %if.end
955  unreachable
956
957if.end2:                                          ; preds = %if.end
958  ret void
959}
960
961; GCN-LABEL: {{^}}fcmp64:
962; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
963; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
964define amdgpu_kernel void @fcmp64(float %n, float %s) {
965entry:
966  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
967  %id.f = uitofp i32 %id to float
968  %mul4 = fmul float %s, %n
969  %cmp = fcmp ult float 0.0, %mul4
970  br label %if.end
971
972if.end:                                           ; preds = %entry
973  %rem.f = frem float %id.f, %s
974  %fcmp = tail call i64 @llvm.amdgcn.fcmp.i64.f32(float %rem.f, float 0.0, i32 1)
975  %shr = lshr i64 %fcmp, 1
976  %notmask = shl nsw i64 -1, 0
977  %and = and i64 %notmask, %shr
978  %or = or i64 %and, -9223372036854775808
979  %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
980  %cast = trunc i64 %cttz to i32
981  %cmp3 = icmp ugt i32 10, %cast
982  %cmp6 = fcmp one float %rem.f, 0.0
983  %brmerge = or i1 %cmp6, %cmp3
984  br i1 %brmerge, label %if.end2, label %if.then
985
986if.then:                                          ; preds = %if.end
987  unreachable
988
989if.end2:                                          ; preds = %if.end
990  ret void
991}
992
993; GCN-LABEL: {{^}}icmp32:
994; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
995; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
996define amdgpu_kernel void @icmp32(i32 %n, i32 %s) {
997entry:
998  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
999  %mul4 = mul nsw i32 %s, %n
1000  %cmp = icmp slt i32 0, %mul4
1001  br label %if.end
1002
1003if.end:                                           ; preds = %entry
1004  %rem = urem i32 %id, %s
1005  %icmp = tail call i32 @llvm.amdgcn.icmp.i32.i32(i32 %rem, i32 0, i32 32)
1006  %shr = lshr i32 %icmp, 1
1007  %notmask = shl nsw i32 -1, 0
1008  %and = and i32 %notmask, %shr
1009  %or = or i32 %and, 2147483648
1010  %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1011  %cmp3 = icmp ugt i32 10, %cttz
1012  %cmp6 = icmp ne i32 %rem, 0
1013  %brmerge = or i1 %cmp6, %cmp3
1014  br i1 %brmerge, label %if.end2, label %if.then
1015
1016if.then:                                          ; preds = %if.end
1017  unreachable
1018
1019if.end2:                                          ; preds = %if.end
1020  ret void
1021}
1022
1023; GCN-LABEL: {{^}}fcmp32:
1024; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
1025; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
1026define amdgpu_kernel void @fcmp32(float %n, float %s) {
1027entry:
1028  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
1029  %id.f = uitofp i32 %id to float
1030  %mul4 = fmul float %s, %n
1031  %cmp = fcmp ult float 0.0, %mul4
1032  br label %if.end
1033
1034if.end:                                           ; preds = %entry
1035  %rem.f = frem float %id.f, %s
1036  %fcmp = tail call i32 @llvm.amdgcn.fcmp.i32.f32(float %rem.f, float 0.0, i32 1)
1037  %shr = lshr i32 %fcmp, 1
1038  %notmask = shl nsw i32 -1, 0
1039  %and = and i32 %notmask, %shr
1040  %or = or i32 %and, 2147483648
1041  %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1042  %cmp3 = icmp ugt i32 10, %cttz
1043  %cmp6 = fcmp one float %rem.f, 0.0
1044  %brmerge = or i1 %cmp6, %cmp3
1045  br i1 %brmerge, label %if.end2, label %if.then
1046
1047if.then:                                          ; preds = %if.end
1048  unreachable
1049
1050if.end2:                                          ; preds = %if.end
1051  ret void
1052}
1053
1054declare void @external_void_func_void() #1
1055
1056; Test save/restore of VGPR needed for SGPR spilling.
1057
1058; GCN-LABEL: {{^}}callee_no_stack_with_call:
1059; GCN: s_waitcnt
1060; GCN-NEXT: s_waitcnt_vscnt
1061
1062; GFX1064-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1063; GFX1032-NEXT: s_or_saveexec_b32 [[COPY_EXEC0:s[0-9]]], -1{{$}}
1064; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
1065; GCN-NEXT: s_waitcnt_depctr 0xffe3
1066; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
1067; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC0]]
1068
1069; GCN-NEXT: v_writelane_b32 v40, s33, 2
1070; GCN: s_mov_b32 s33, s32
1071; GFX1064: s_add_u32 s32, s32, 0x400
1072; GFX1032: s_add_u32 s32, s32, 0x200
1073
1074
1075; GCN-DAG: v_writelane_b32 v40, s30, 0
1076; GCN-DAG: v_writelane_b32 v40, s31, 1
1077; GCN: s_swappc_b64
1078; GCN-DAG: v_readlane_b32 s4, v40, 0
1079; GCN-DAG: v_readlane_b32 s5, v40, 1
1080
1081
1082; GFX1064: s_sub_u32 s32, s32, 0x400
1083; GFX1032: s_sub_u32 s32, s32, 0x200
1084; GCN: v_readlane_b32 s33, v40, 2
1085; GFX1064: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1086; GFX1032: s_or_saveexec_b32 [[COPY_EXEC1:s[0-9]]], -1{{$}}
1087; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
1088; GCN-NEXT: s_waitcnt_depctr 0xffe3
1089; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
1090; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC1]]
1091; GCN-NEXT: s_waitcnt vmcnt(0)
1092; GCN-NEXT: s_setpc_b64
1093define void @callee_no_stack_with_call() #1 {
1094  call void @external_void_func_void()
1095  ret void
1096}
1097
1098
1099declare i32 @llvm.amdgcn.workitem.id.x()
1100declare float @llvm.fabs.f32(float)
1101declare { float, i1 } @llvm.amdgcn.div.scale.f32(float, float, i1)
1102declare { double, i1 } @llvm.amdgcn.div.scale.f64(double, double, i1)
1103declare float @llvm.amdgcn.div.fmas.f32(float, float, float, i1)
1104declare double @llvm.amdgcn.div.fmas.f64(double, double, double, i1)
1105declare i1 @llvm.amdgcn.class.f32(float, i32)
1106declare i32 @llvm.amdgcn.set.inactive.i32(i32, i32)
1107declare i64 @llvm.amdgcn.set.inactive.i64(i64, i64)
1108declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1109declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1110declare float @llvm.amdgcn.wwm.f32(float)
1111declare i32 @llvm.amdgcn.wqm.i32(i32)
1112declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32)
1113declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32)
1114declare float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32>, i32, i32, i32, i32 immarg)
1115declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32)
1116declare i32 @llvm.amdgcn.mbcnt.hi(i32, i32)
1117declare i64 @llvm.amdgcn.fcmp.i64.f32(float, float, i32)
1118declare i64 @llvm.amdgcn.icmp.i64.i32(i32, i32, i32)
1119declare i32 @llvm.amdgcn.fcmp.i32.f32(float, float, i32)
1120declare i32 @llvm.amdgcn.icmp.i32.i32(i32, i32, i32)
1121declare void @llvm.amdgcn.kill(i1)
1122declare i1 @llvm.amdgcn.wqm.vote(i1)
1123declare i1 @llvm.amdgcn.ps.live()
1124declare i64 @llvm.cttz.i64(i64, i1)
1125declare i32 @llvm.cttz.i32(i32, i1)
1126
1127attributes #0 = { nounwind readnone speculatable }
1128attributes #1 = { nounwind }
1129attributes #2 = { nounwind readnone optnone noinline }
1130attributes #3 = { "target-features"="+wavefrontsize32" }
1131attributes #4 = { "target-features"="+wavefrontsize64" }
1132