1 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefixes=GCN,GFX1032 %s
2 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefixes=GCN,GFX1064 %s
3 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-early-ifcvt=1 -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefixes=GCN,GFX1032 %s
4 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-early-ifcvt=1 -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefixes=GCN,GFX1064 %s
5 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefixes=GCN,GFX1032,GFX10DEFWAVE %s
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{{$}}
12 define amdgpu_kernel void @test_vopc_i32(i32 addrspace(1)* %arg) {
13 %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
14 %gep = getelementptr inbounds i32, i32 addrspace(1)* %arg, i32 %lid
15 %load = load i32, i32 addrspace(1)* %gep, align 4
16 %cmp = icmp sgt i32 %load, 0
17 %sel = select i1 %cmp, i32 1, i32 2
18 store i32 %sel, i32 addrspace(1)* %gep, align 4
22 ; GCN-LABEL: {{^}}test_vopc_f32:
23 ; GFX1032: v_cmp_nge_f32_e32 vcc_lo, 0, v{{[0-9]+}}
24 ; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, vcc_lo
25 ; GFX1064: v_cmp_nge_f32_e32 vcc, 0, v{{[0-9]+}}
26 ; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0, vcc{{$}}
27 define 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
37 ; GCN-LABEL: {{^}}test_vopc_vcmp:
38 ; GFX1032: v_cmp_nle_f32_e32 vcc_lo, 0, v{{[0-9]+}}
39 ; GFX1064: v_cmp_nle_f32_e32 vcc, 0, v{{[0-9]+}}
40 define amdgpu_ps void @test_vopc_vcmp(float %x) {
41 %cmp = fcmp oge float %x, 0.0
42 call void @llvm.amdgcn.kill(i1 %cmp)
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]]
51 define 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
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]]{{$}}
67 define 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
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]]
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]]{{$}}
81 define 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
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]]
97 define 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
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]]
118 define 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
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]]
139 define 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
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{{$}}
154 ; GCN: s_cbranch_execz
155 define amdgpu_kernel void @test_mask_if(i32 addrspace(1)* %arg) #0 {
156 %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
157 %cmp = icmp ugt i32 %lid, 10
158 br i1 %cmp, label %if, label %endif
161 store i32 0, i32 addrspace(1)* %arg, align 4
168 ; GCN-LABEL: {{^}}test_loop_with_if:
169 ; GFX1032: s_or_b32 s{{[0-9]+}}, vcc_lo, s{{[0-9]+}}
170 ; GFX1032: s_andn2_b32 exec_lo, exec_lo, s{{[0-9]+}}
171 ; GFX1064: s_or_b64 s[{{[0-9:]+}}], vcc, s[{{[0-9:]+}}]
172 ; GFX1064: s_andn2_b64 exec, exec, s[{{[0-9:]+}}]
173 ; GCN: s_cbranch_execz
175 ; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
176 ; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
177 ; GCN: s_cbranch_execz
178 ; GCN: ; %bb.{{[0-9]+}}:
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:]+}}]{{$}}
188 ; GCN: s_cbranch_execz BB
189 ; GCN: ; %bb.{{[0-9]+}}:
192 define amdgpu_kernel void @test_loop_with_if(i32 addrspace(1)* %arg) #0 {
194 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
201 %tmp3 = phi i32 [ 0, %bb ], [ %tmp15, %bb13 ]
202 %tmp4 = icmp slt i32 %tmp3, %tmp
203 br i1 %tmp4, label %bb5, label %bb11
206 %tmp6 = sext i32 %tmp3 to i64
207 %tmp7 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp6
208 %tmp8 = load i32, i32 addrspace(1)* %tmp7, align 4
209 %tmp9 = icmp sgt i32 %tmp8, 10
210 br i1 %tmp9, label %bb10, label %bb11
213 store i32 %tmp, i32 addrspace(1)* %tmp7, align 4
217 %tmp12 = sdiv i32 %tmp3, 2
221 %tmp14 = phi i32 [ %tmp3, %bb10 ], [ %tmp12, %bb11 ]
222 %tmp15 = add nsw i32 %tmp14, 1
223 %tmp16 = icmp slt i32 %tmp14, 255
224 br i1 %tmp16, label %bb2, label %bb1
227 ; GCN-LABEL: {{^}}test_loop_with_if_else_break:
228 ; GFX1032: s_and_saveexec_b32 s{{[0-9]+}}, vcc_lo
229 ; GFX1064: s_and_saveexec_b64 s[{{[0-9:]+}}], vcc{{$}}
230 ; GCN: s_cbranch_execz
231 ; GCN: ; %bb.{{[0-9]+}}: ; %.preheader
234 ; GCN: global_store_dword
235 ; GFX1032: s_or_b32 [[MASK0:s[0-9]+]], [[MASK0]], vcc_lo
236 ; GFX1064: s_or_b64 [[MASK0:s\[[0-9:]+\]]], [[MASK0]], vcc
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
253 ; GFX1032: s_or_b32 [[MASK1]], [[MASK1]], exec_lo
254 ; GFX1064: s_or_b64 [[MASK1]], [[MASK1]], exec
255 ; GCN: global_load_dword [[LOAD:v[0-9]+]]
256 ; GFX1032: v_cmp_gt_i32_e32 vcc_lo, 11, [[LOAD]]
257 ; GFX1064: v_cmp_gt_i32_e32 vcc, 11, [[LOAD]]
258 define amdgpu_kernel void @test_loop_with_if_else_break(i32 addrspace(1)* %arg) #0 {
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
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
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
287 ; GCN-LABEL: {{^}}test_addc_vop2b:
288 ; GFX1032: v_add_co_u32 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 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{{$}}
292 define amdgpu_kernel void @test_addc_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
294 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
295 %tmp3 = getelementptr inbounds i64, i64 addrspace(1)* %arg, i32 %tmp
296 %tmp4 = load i64, i64 addrspace(1)* %tmp3, align 8
297 %tmp5 = add nsw i64 %tmp4, %arg1
298 store i64 %tmp5, i64 addrspace(1)* %tmp3, align 8
302 ; GCN-LABEL: {{^}}test_subbrev_vop2b:
303 ; GFX1032: v_sub_co_u32 v{{[0-9]+}}, [[A0:s[0-9]+|vcc_lo]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
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 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], v{{[0-9]+}}, s{{[0-9]+}}{{$}}
306 ; GFX1064: v_subrev_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, {{[vs][0-9]+}}, [[A0]]{{$}}
307 define amdgpu_kernel void @test_subbrev_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
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
317 ; GCN-LABEL: {{^}}test_subb_vop2b:
318 ; GFX1032: v_sub_co_u32 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 v{{[0-9]+}}, [[A0:s\[[0-9:]+\]|vcc]], s{{[0-9]+}}, v{{[0-9]+}}{{$}}
321 ; GFX1064: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, [[A0]]{{$}}
322 define amdgpu_kernel void @test_subb_vop2b(i64 addrspace(1)* %arg, i64 %arg1) #0 {
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
332 ; GCN-LABEL: {{^}}test_udiv64:
333 ; GFX1032: v_add_co_u32 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 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
337 ; GFX1032: v_add_co_u32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
338 ; GFX1032: v_add_co_u32 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 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
341 ; GFX1032: v_subrev_co_ci_u32_e64 v{{[0-9]+}}, s{{[0-9]+}}, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
342 ; GFX1032: v_sub_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, {{[vs][0-9]+}}, v{{[0-9]+}}, vcc_lo
343 ; GFX1064: v_add_co_u32 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{{$}}
345 ; GFX1064: v_add_co_ci_u32_e64 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}, [[SDST]]
346 ; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
347 ; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
348 ; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
349 ; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
350 ; GFX1064: v_sub_co_u32 v{{[0-9]+}}, s[{{[0-9:]+}}], s{{[0-9]+}}, v{{[0-9]+}}
351 ; GFX1064: v_subrev_co_ci_u32_e64 v{{[0-9]+}}, vcc, {{[vs][0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}]
352 ; GFX1064: v_sub_co_ci_u32_e64 v{{[0-9]+}}, s[{{[0-9:]+}}], {{[vs][0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}]
353 define amdgpu_kernel void @test_udiv64(i64 addrspace(1)* %arg) #0 {
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
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]+}}
367 define 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
372 %a = load volatile float, float addrspace(1)* %gep.0, align 4
373 %b = load volatile float, float addrspace(1)* %gep.1, align 4
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
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:]+}}]
384 define 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
389 %a = load volatile double, double addrspace(1)* %gep.0, align 8
390 %b = load volatile double, double addrspace(1)* %gep.1, align 8
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
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:]+}}]
401 define 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
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:]+}}]
412 define 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
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]+}}
424 define 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
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:]+}}]
434 define 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
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:]+}}]{{$}}
446 ; GCN: load_dword [[LOAD:v[0-9]+]]
447 ; GCN: v_cmp_ne_u32_e32 [[VCC]], 0, [[LOAD]]
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]+}}
453 define amdgpu_kernel void @test_div_fmas_f32_i1_phi_vcc(float addrspace(1)* %out, float addrspace(1)* %in, i32 addrspace(1)* %dummy) #0 {
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
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
465 %cmp0 = icmp eq i32 %tid, 0
466 br i1 %cmp0, label %bb, label %exit
469 %val = load volatile i32, i32 addrspace(1)* %dummy
470 %cmp1 = icmp ne i32 %val, 0
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
480 ; GCN-LABEL: {{^}}fdiv_f32:
481 ; GFX1032: v_div_scale_f32 v{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
482 ; GFX1064: v_div_scale_f32 v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
483 ; GCN: v_rcp_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}
484 ; GFX1032: v_div_scale_f32 v{{[0-9]+}}, vcc_lo, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
485 ; GFX1064: v_div_scale_f32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
488 ; GCN: v_div_fmas_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
489 define amdgpu_kernel void @fdiv_f32(float addrspace(1)* %out, float %a, float %b) #0 {
491 %fdiv = fdiv float %a, %b
492 store float %fdiv, float addrspace(1)* %out
496 ; GCN-LABEL: {{^}}test_br_cc_f16:
497 ; GFX1032: v_cmp_nlt_f16_e32 vcc_lo,
498 ; GFX1032: s_and_b32 vcc_lo, exec_lo, vcc_lo
499 ; GFX1064: v_cmp_nlt_f16_e32 vcc,
500 ; GFX1064: s_and_b64 vcc, exec, vcc{{$}}
501 ; GCN-NEXT: s_cbranch_vccnz
502 define amdgpu_kernel void @test_br_cc_f16(
503 half addrspace(1)* %r,
504 half addrspace(1)* %a,
505 half addrspace(1)* %b) {
507 %a.val = load half, half addrspace(1)* %a
508 %b.val = load half, half addrspace(1)* %b
509 %fcmp = fcmp olt half %a.val, %b.val
510 br i1 %fcmp, label %one, label %two
513 store half %a.val, half addrspace(1)* %r
517 store half %b.val, half addrspace(1)* %r
521 ; GCN-LABEL: {{^}}test_brcc_i1:
522 ; GCN: s_cmp_eq_u32 s{{[0-9]+}}, 0
523 ; GCN-NEXT: s_cbranch_scc1
524 define amdgpu_kernel void @test_brcc_i1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in, i1 %val) #0 {
525 %cmp0 = icmp ne i1 %val, 0
526 br i1 %cmp0, label %store, label %end
529 store i32 222, i32 addrspace(1)* %out
536 ; GCN-LABEL: {{^}}test_preserve_condition_undef_flag:
537 ; GFX1032-DAG: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
538 ; GFX1032-DAG: v_cmp_ngt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 0
539 ; GFX1032: v_cmp_nlt_f32_e64 s{{[0-9]+}}, s{{[0-9]+}}, 1.0
540 ; GFX1032: s_or_b32 [[OR1:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
541 ; GFX1032: s_or_b32 [[OR2:s[0-9]+]], [[OR1]], s{{[0-9]+}}
542 ; GFX1032: s_and_b32 vcc_lo, exec_lo, [[OR2]]
543 ; GFX1064-DAG: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
544 ; GFX1064-DAG: v_cmp_ngt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 0
545 ; GFX1064: v_cmp_nlt_f32_e64 s[{{[0-9:]+}}], s{{[0-9]+}}, 1.0
546 ; GFX1064: s_or_b64 [[OR1:s\[[0-9:]+\]]], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
547 ; GFX1064: s_or_b64 [[OR2:s\[[0-9:]+\]]], [[OR1]], s[{{[0-9:]+}}]
548 ; GFX1064: s_and_b64 vcc, exec, [[OR2]]
549 ; GCN: s_cbranch_vccnz
550 define amdgpu_kernel void @test_preserve_condition_undef_flag(float %arg, i32 %arg1, float %arg2) #0 {
552 %tmp = icmp sgt i32 %arg1, 4
553 %undef = call i1 @llvm.amdgcn.class.f32(float undef, i32 undef)
554 %tmp4 = select i1 %undef, float %arg, float 1.000000e+00
555 %tmp5 = fcmp ogt float %arg2, 0.000000e+00
556 %tmp6 = fcmp olt float %arg2, 1.000000e+00
557 %tmp7 = fcmp olt float %arg, %tmp4
558 %tmp8 = and i1 %tmp5, %tmp6
559 %tmp9 = and i1 %tmp8, %tmp7
560 br i1 %tmp9, label %bb1, label %bb2
563 store volatile i32 0, i32 addrspace(1)* undef
570 ; GCN-LABEL: {{^}}test_invert_true_phi_cond_break_loop:
571 ; GFX1032: s_xor_b32 s{{[0-9]+}}, s{{[0-9]+}}, -1
572 ; GFX1032: s_or_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
573 ; GFX1064: s_xor_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], -1
574 ; GFX1064: s_or_b64 s[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}]
575 define amdgpu_kernel void @test_invert_true_phi_cond_break_loop(i32 %arg) #0 {
577 %id = call i32 @llvm.amdgcn.workitem.id.x()
578 %tmp = sub i32 %id, %arg
581 bb1: ; preds = %Flow, %bb
582 %lsr.iv = phi i32 [ undef, %bb ], [ %tmp2, %Flow ]
583 %lsr.iv.next = add i32 %lsr.iv, 1
584 %cmp0 = icmp slt i32 %lsr.iv.next, 0
585 br i1 %cmp0, label %bb4, label %Flow
588 %load = load volatile i32, i32 addrspace(1)* undef, align 4
589 %cmp1 = icmp sge i32 %tmp, %load
592 Flow: ; preds = %bb4, %bb1
593 %tmp2 = phi i32 [ %lsr.iv.next, %bb4 ], [ undef, %bb1 ]
594 %tmp3 = phi i1 [ %cmp1, %bb4 ], [ true, %bb1 ]
595 br i1 %tmp3, label %bb1, label %bb9
598 store volatile i32 7, i32 addrspace(3)* undef
602 ; GCN-LABEL: {{^}}test_movrels_extract_neg_offset_vgpr:
603 ; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 1, v{{[0-9]+}}
604 ; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc_lo
605 ; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 2, v{{[0-9]+}}
606 ; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc_lo
607 ; GFX1032: v_cmp_ne_u32_e32 vcc_lo, 3, v{{[0-9]+}}
608 ; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc_lo
609 ; GFX1064: v_cmp_eq_u32_e32 vcc, 1, v{{[0-9]+}}
610 ; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
611 ; GFX1064: v_cmp_ne_u32_e32 vcc, 2, v{{[0-9]+}}
612 ; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 2, v{{[0-9]+}}, vcc
613 ; GFX1064: v_cmp_ne_u32_e32 vcc, 3, v{{[0-9]+}}
614 ; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 3, v{{[0-9]+}}, vcc
615 define amdgpu_kernel void @test_movrels_extract_neg_offset_vgpr(i32 addrspace(1)* %out) #0 {
617 %id = call i32 @llvm.amdgcn.workitem.id.x() #1
618 %index = add i32 %id, -512
619 %value = extractelement <4 x i32> <i32 0, i32 1, i32 2, i32 3>, i32 %index
620 store i32 %value, i32 addrspace(1)* %out
624 ; GCN-LABEL: {{^}}test_set_inactive:
625 ; GFX1032: s_not_b32 exec_lo, exec_lo
626 ; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 42
627 ; GFX1032: s_not_b32 exec_lo, exec_lo
628 ; GFX1064: s_not_b64 exec, exec{{$}}
629 ; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 42
630 ; GFX1064: s_not_b64 exec, exec{{$}}
631 define amdgpu_kernel void @test_set_inactive(i32 addrspace(1)* %out, i32 %in) #0 {
632 %tmp = call i32 @llvm.amdgcn.set.inactive.i32(i32 %in, i32 42)
633 store i32 %tmp, i32 addrspace(1)* %out
637 ; GCN-LABEL: {{^}}test_set_inactive_64:
638 ; GFX1032: s_not_b32 exec_lo, exec_lo
639 ; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
640 ; GFX1032: v_mov_b32_e32 {{v[0-9]+}}, 0
641 ; GFX1032: s_not_b32 exec_lo, exec_lo
642 ; GFX1064: s_not_b64 exec, exec{{$}}
643 ; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
644 ; GFX1064: v_mov_b32_e32 {{v[0-9]+}}, 0
645 ; GFX1064: s_not_b64 exec, exec{{$}}
646 define amdgpu_kernel void @test_set_inactive_64(i64 addrspace(1)* %out, i64 %in) #0 {
647 %tmp = call i64 @llvm.amdgcn.set.inactive.i64(i64 %in, i64 0)
648 store i64 %tmp, i64 addrspace(1)* %out
652 ; GCN-LABEL: {{^}}test_kill_i1_terminator_float:
653 ; GFX1032: s_mov_b32 exec_lo, 0
654 ; GFX1064: s_mov_b64 exec, 0
655 define amdgpu_ps void @test_kill_i1_terminator_float() #0 {
656 call void @llvm.amdgcn.kill(i1 false)
660 ; GCN-LABEL: {{^}}test_kill_i1_terminator_i1:
661 ; GFX1032: s_mov_b32 [[LIVE:s[0-9]+]], exec_lo
662 ; GFX1032: s_or_b32 [[OR:s[0-9]+]],
663 ; GFX1032: s_xor_b32 [[KILL:s[0-9]+]], [[OR]], exec_lo
664 ; GFX1032: s_andn2_b32 [[MASK:s[0-9]+]], [[LIVE]], [[KILL]]
665 ; GFX1032: s_and_b32 exec_lo, exec_lo, [[MASK]]
666 ; GFX1064: s_mov_b64 [[LIVE:s\[[0-9:]+\]]], exec
667 ; GFX1064: s_or_b64 [[OR:s\[[0-9:]+\]]],
668 ; GFX1064: s_xor_b64 [[KILL:s\[[0-9:]+\]]], [[OR]], exec
669 ; GFX1064: s_andn2_b64 [[MASK:s\[[0-9:]+\]]], [[LIVE]], [[KILL]]
670 ; GFX1064: s_and_b64 exec, exec, [[MASK]]
671 define amdgpu_gs void @test_kill_i1_terminator_i1(i32 %a, i32 %b, i32 %c, i32 %d) #0 {
672 %c1 = icmp slt i32 %a, %b
673 %c2 = icmp slt i32 %c, %d
675 call void @llvm.amdgcn.kill(i1 %x)
676 call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 false, i1 false)
680 ; GCN-LABEL: {{^}}test_loop_vcc:
681 ; GFX1032: v_cmp_lt_f32_e32 vcc_lo,
682 ; GFX1064: v_cmp_lt_f32_e32 vcc,
683 ; GCN: s_cbranch_vccz
684 define amdgpu_ps <4 x float> @test_loop_vcc(<4 x float> %in) #0 {
689 %ctr.iv = phi float [ 0.0, %entry ], [ %ctr.next, %body ]
690 %c.iv = phi <4 x float> [ %in, %entry ], [ %c.next, %body ]
691 %cc = fcmp ogt float %ctr.iv, 7.0
692 br i1 %cc, label %break, label %body
695 %c.iv0 = extractelement <4 x float> %c.iv, i32 0
696 %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)
697 %ctr.next = fadd float %ctr.iv, 2.0
701 ret <4 x float> %c.iv
704 ; NOTE: llvm.amdgcn.wwm is deprecated, use llvm.amdgcn.strict.wwm instead.
705 ; GCN-LABEL: {{^}}test_wwm1:
706 ; GFX1032: s_or_saveexec_b32 [[SAVE:s[0-9]+]], -1
707 ; GFX1032: s_mov_b32 exec_lo, [[SAVE]]
708 ; GFX1064: s_or_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], -1
709 ; GFX1064: s_mov_b64 exec, [[SAVE]]
710 define amdgpu_ps float @test_wwm1(i32 inreg %idx0, i32 inreg %idx1, float %src0, float %src1) {
712 %out = fadd float %src0, %src1
713 %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
717 ; GCN-LABEL: {{^}}test_wwm2:
718 ; GFX1032: v_cmp_gt_u32_e32 vcc_lo, 32, v{{[0-9]+}}
719 ; GFX1032: s_and_saveexec_b32 [[SAVE1:s[0-9]+]], vcc_lo
720 ; GFX1032: s_or_saveexec_b32 [[SAVE2:s[0-9]+]], -1
721 ; GFX1032: s_mov_b32 exec_lo, [[SAVE2]]
722 ; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE1]]
723 ; GFX1064: v_cmp_gt_u32_e32 vcc, 32, v{{[0-9]+}}
724 ; GFX1064: s_and_saveexec_b64 [[SAVE1:s\[[0-9:]+\]]], vcc{{$}}
725 ; GFX1064: s_or_saveexec_b64 [[SAVE2:s\[[0-9:]+\]]], -1
726 ; GFX1064: s_mov_b64 exec, [[SAVE2]]
727 ; GFX1064: s_or_b64 exec, exec, [[SAVE1]]
728 define amdgpu_ps float @test_wwm2(i32 inreg %idx) {
730 ; use mbcnt to make sure the branch is divergent
731 %lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
732 %hi = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
733 %cc = icmp uge i32 %hi, 32
734 br i1 %cc, label %endif, label %if
737 %src = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx, i32 0, i32 0, i32 0)
738 %out = fadd float %src, %src
739 %out.0 = call float @llvm.amdgcn.wwm.f32(float %out)
740 %out.1 = fadd float %src, %out.0
744 %out.2 = phi float [ %out.1, %if ], [ 0.0, %main_body ]
748 ; GCN-LABEL: {{^}}test_strict_wwm1:
749 ; GFX1032: s_or_saveexec_b32 [[SAVE:s[0-9]+]], -1
750 ; GFX1032: s_mov_b32 exec_lo, [[SAVE]]
751 ; GFX1064: s_or_saveexec_b64 [[SAVE:s\[[0-9]+:[0-9]+\]]], -1
752 ; GFX1064: s_mov_b64 exec, [[SAVE]]
753 define amdgpu_ps float @test_strict_wwm1(i32 inreg %idx0, i32 inreg %idx1, float %src0, float %src1) {
755 %out = fadd float %src0, %src1
756 %out.0 = call float @llvm.amdgcn.strict.wwm.f32(float %out)
760 ; GCN-LABEL: {{^}}test_strict_wwm2:
761 ; GFX1032: v_cmp_gt_u32_e32 vcc_lo, 32, v{{[0-9]+}}
762 ; GFX1032: s_and_saveexec_b32 [[SAVE1:s[0-9]+]], vcc_lo
763 ; GFX1032: s_or_saveexec_b32 [[SAVE2:s[0-9]+]], -1
764 ; GFX1032: s_mov_b32 exec_lo, [[SAVE2]]
765 ; GFX1032: s_or_b32 exec_lo, exec_lo, [[SAVE1]]
766 ; GFX1064: v_cmp_gt_u32_e32 vcc, 32, v{{[0-9]+}}
767 ; GFX1064: s_and_saveexec_b64 [[SAVE1:s\[[0-9:]+\]]], vcc{{$}}
768 ; GFX1064: s_or_saveexec_b64 [[SAVE2:s\[[0-9:]+\]]], -1
769 ; GFX1064: s_mov_b64 exec, [[SAVE2]]
770 ; GFX1064: s_or_b64 exec, exec, [[SAVE1]]
771 define amdgpu_ps float @test_strict_wwm2(i32 inreg %idx) {
773 ; use mbcnt to make sure the branch is divergent
774 %lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
775 %hi = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
776 %cc = icmp uge i32 %hi, 32
777 br i1 %cc, label %endif, label %if
780 %src = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx, i32 0, i32 0, i32 0)
781 %out = fadd float %src, %src
782 %out.0 = call float @llvm.amdgcn.strict.wwm.f32(float %out)
783 %out.1 = fadd float %src, %out.0
787 %out.2 = phi float [ %out.1, %if ], [ 0.0, %main_body ]
792 ; GCN-LABEL: {{^}}test_wqm1:
793 ; GFX1032: s_mov_b32 [[ORIG:s[0-9]+]], exec_lo
794 ; GFX1032: s_wqm_b32 exec_lo, exec_lo
795 ; GFX1032: s_and_b32 exec_lo, exec_lo, [[ORIG]]
796 ; GFX1064: s_mov_b64 [[ORIG:s\[[0-9]+:[0-9]+\]]], exec{{$}}
797 ; GFX1064: s_wqm_b64 exec, exec{{$}}
798 ; GFX1064: s_and_b64 exec, exec, [[ORIG]]
799 define 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 {
801 %inst23 = extractelement <2 x float> %pos, i32 0
802 %inst24 = extractelement <2 x float> %pos, i32 1
803 %inst25 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 0, i32 0, i32 %m0)
804 %inst26 = tail call float @llvm.amdgcn.interp.p2(float %inst25, float %inst24, i32 0, i32 0, i32 %m0)
805 %inst28 = tail call float @llvm.amdgcn.interp.p1(float %inst23, i32 1, i32 0, i32 %m0)
806 %inst29 = tail call float @llvm.amdgcn.interp.p2(float %inst28, float %inst24, i32 1, i32 0, i32 %m0)
807 %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)
811 ; GCN-LABEL: {{^}}test_wqm2:
812 ; GFX1032: s_wqm_b32 exec_lo, exec_lo
813 ; GFX1032: s_and_b32 exec_lo, exec_lo, s{{[0-9]+}}
814 ; GFX1064: s_wqm_b64 exec, exec{{$}}
815 ; GFX1064: s_and_b64 exec, exec, s[{{[0-9:]+}}]
816 define amdgpu_ps float @test_wqm2(i32 inreg %idx0, i32 inreg %idx1) #0 {
818 %src0 = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx0, i32 0, i32 0, i32 0)
819 %src1 = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> undef, i32 %idx1, i32 0, i32 0, i32 0)
820 %out = fadd float %src0, %src1
821 %out.0 = bitcast float %out to i32
822 %out.1 = call i32 @llvm.amdgcn.wqm.i32(i32 %out.0)
823 %out.2 = bitcast i32 %out.1 to float
827 ; GCN-LABEL: {{^}}test_intr_fcmp_i64:
828 ; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
829 ; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
830 ; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
831 ; GFX1064: v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
832 ; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
833 ; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
834 ; GCN: store_dwordx2 v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]], s
835 define amdgpu_kernel void @test_intr_fcmp_i64(i64 addrspace(1)* %out, float %src, float %a) {
836 %temp = call float @llvm.fabs.f32(float %a)
837 %result = call i64 @llvm.amdgcn.fcmp.i64.f32(float %src, float %temp, i32 1)
838 store i64 %result, i64 addrspace(1)* %out
842 ; GCN-LABEL: {{^}}test_intr_icmp_i64:
843 ; GFX1032-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], 0{{$}}
844 ; GFX1032-DAG: v_cmp_eq_u32_e64 [[C_LO:vcc_lo|s[0-9]+]], 0x64, {{s[0-9]+}}
845 ; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], [[C_LO]]
846 ; GFX1064: v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], 0x64, {{s[0-9]+}}
847 ; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
848 ; GFX1064-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], s[[C_HI]]
849 ; GCN: store_dwordx2 v{{[0-9]+}}, v{{\[}}[[V_LO]]:[[V_HI]]], s
850 define amdgpu_kernel void @test_intr_icmp_i64(i64 addrspace(1)* %out, i32 %src) {
851 %result = call i64 @llvm.amdgcn.icmp.i64.i32(i32 %src, i32 100, i32 32)
852 store i64 %result, i64 addrspace(1)* %out
856 ; GCN-LABEL: {{^}}test_intr_fcmp_i32:
857 ; GFX1032-DAG: v_cmp_eq_f32_e64 s[[C_LO:[0-9]+]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
858 ; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
859 ; GFX1064: v_cmp_eq_f32_e64 s{{\[}}[[C_LO:[0-9]+]]:[[C_HI:[0-9]+]]], {{s[0-9]+}}, |{{[vs][0-9]+}}|
860 ; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]
861 ; GCN: store_dword v{{[0-9]+}}, v[[V_LO]], s
862 define amdgpu_kernel void @test_intr_fcmp_i32(i32 addrspace(1)* %out, float %src, float %a) {
863 %temp = call float @llvm.fabs.f32(float %a)
864 %result = call i32 @llvm.amdgcn.fcmp.i32.f32(float %src, float %temp, i32 1)
865 store i32 %result, i32 addrspace(1)* %out
869 ; GCN-LABEL: {{^}}test_intr_icmp_i32:
870 ; GFX1032-DAG: v_cmp_eq_u32_e64 s[[C_LO:[0-9]+]], 0x64, {{s[0-9]+}}
871 ; GFX1032-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
872 ; GFX1064: v_cmp_eq_u32_e64 s{{\[}}[[C_LO:[0-9]+]]:{{[0-9]+}}], 0x64, {{s[0-9]+}}
873 ; GFX1064-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], s[[C_LO]]{{$}}
874 ; GCN: store_dword v{{[0-9]+}}, v[[V_LO]], s
875 define amdgpu_kernel void @test_intr_icmp_i32(i32 addrspace(1)* %out, i32 %src) {
876 %result = call i32 @llvm.amdgcn.icmp.i32.i32(i32 %src, i32 100, i32 32)
877 store i32 %result, i32 addrspace(1)* %out
881 ; GCN-LABEL: {{^}}test_wqm_vote:
882 ; GFX1032: v_cmp_neq_f32_e32 vcc_lo, 0
883 ; GFX1032: s_mov_b32 [[LIVE:s[0-9]+]], exec_lo
884 ; GFX1032: s_wqm_b32 [[WQM:s[0-9]+]], vcc_lo
885 ; GFX1032: s_xor_b32 [[KILL:s[0-9]+]], [[WQM]], exec_lo
886 ; GFX1032: s_andn2_b32 [[MASK:s[0-9]+]], [[LIVE]], [[KILL]]
887 ; GFX1032: s_and_b32 exec_lo, exec_lo, [[MASK]]
888 ; GFX1064: v_cmp_neq_f32_e32 vcc, 0
889 ; GFX1064: s_mov_b64 [[LIVE:s\[[0-9:]+\]]], exec
890 ; GFX1064: s_wqm_b64 [[WQM:s\[[0-9:]+\]]], vcc
891 ; GFX1064: s_xor_b64 [[KILL:s\[[0-9:]+\]]], [[WQM]], exec
892 ; GFX1064: s_andn2_b64 [[MASK:s\[[0-9:]+\]]], [[LIVE]], [[KILL]]
893 ; GFX1064: s_and_b64 exec, exec, [[MASK]]
894 define amdgpu_ps void @test_wqm_vote(float %a) {
895 %c1 = fcmp une float %a, 0.0
896 %c2 = call i1 @llvm.amdgcn.wqm.vote(i1 %c1)
897 call void @llvm.amdgcn.kill(i1 %c2)
898 call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 false, i1 false)
902 ; GCN-LABEL: {{^}}test_branch_true:
903 ; GFX1032: s_mov_b32 vcc_lo, exec_lo
904 ; GFX1064: s_mov_b64 vcc, exec
905 define amdgpu_kernel void @test_branch_true() #2 {
907 br i1 true, label %for.end, label %for.body.lr.ph
909 for.body.lr.ph: ; preds = %entry
912 for.body: ; preds = %for.body, %for.body.lr.ph
913 br i1 undef, label %for.end, label %for.body
915 for.end: ; preds = %for.body, %entry
919 ; GCN-LABEL: {{^}}test_ps_live:
920 ; GFX1032: s_mov_b32 [[C:s[0-9]+]], exec_lo
921 ; GFX1064: s_mov_b64 [[C:s\[[0-9:]+\]]], exec{{$}}
922 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[C]]
923 define amdgpu_ps float @test_ps_live() #0 {
924 %live = call i1 @llvm.amdgcn.ps.live()
925 %live.32 = zext i1 %live to i32
926 %r = bitcast i32 %live.32 to float
930 ; GCN-LABEL: {{^}}test_vccnz_ifcvt_triangle64:
931 ; GFX1032: v_cmp_neq_f64_e64 [[C:s[0-9]+]], s[{{[0-9:]+}}], 1.0
932 ; GFX1032: s_and_b32 vcc_lo, exec_lo, [[C]]
933 ; GFX1064: v_cmp_neq_f64_e64 [[C:s\[[0-9:]+\]]], s[{{[0-9:]+}}], 1.0
934 ; GFX1064: s_and_b64 vcc, exec, [[C]]
935 define amdgpu_kernel void @test_vccnz_ifcvt_triangle64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
937 %v = load double, double addrspace(1)* %in
938 %cc = fcmp oeq double %v, 1.000000e+00
939 br i1 %cc, label %if, label %endif
942 %u = fadd double %v, %v
946 %r = phi double [ %v, %entry ], [ %u, %if ]
947 store double %r, double addrspace(1)* %out
951 ; GCN-LABEL: {{^}}test_vgprblocks_w32_attr:
952 ; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
953 ; GFX10DEFWAVE: ; VGPRBlocks: 1
954 define amdgpu_gs float @test_vgprblocks_w32_attr(float %a, float %b, float %c, float %d, float %e,
955 float %f, float %g, float %h, float %i, float %j, float %k, float %l) #3 {
957 %s = fadd float %a, %b
958 %s.1 = fadd float %s, %c
959 %s.2 = fadd float %s.1, %d
960 %s.3 = fadd float %s.2, %e
961 %s.4 = fadd float %s.3, %f
962 %s.5 = fadd float %s.4, %g
963 %s.6 = fadd float %s.5, %h
964 %s.7 = fadd float %s.6, %i
965 %s.8 = fadd float %s.7, %j
966 %s.9 = fadd float %s.8, %k
967 %s.10 = fadd float %s.9, %l
971 ; GCN-LABEL: {{^}}test_vgprblocks_w64_attr:
972 ; Test that the wave size can be overridden in function attributes and that the block size is correct as a result
973 ; GFX10DEFWAVE: ; VGPRBlocks: 2
974 define amdgpu_gs float @test_vgprblocks_w64_attr(float %a, float %b, float %c, float %d, float %e,
975 float %f, float %g, float %h, float %i, float %j, float %k, float %l) #4 {
977 %s = fadd float %a, %b
978 %s.1 = fadd float %s, %c
979 %s.2 = fadd float %s.1, %d
980 %s.3 = fadd float %s.2, %e
981 %s.4 = fadd float %s.3, %f
982 %s.5 = fadd float %s.4, %g
983 %s.6 = fadd float %s.5, %h
984 %s.7 = fadd float %s.6, %i
985 %s.8 = fadd float %s.7, %j
986 %s.9 = fadd float %s.8, %k
987 %s.10 = fadd float %s.9, %l
991 ; GCN-LABEL: {{^}}icmp64:
992 ; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
993 ; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
994 define amdgpu_kernel void @icmp64(i32 %n, i32 %s) {
996 %id = tail call i32 @llvm.amdgcn.workitem.id.x()
997 %mul4 = mul nsw i32 %s, %n
998 %cmp = icmp slt i32 0, %mul4
1001 if.end: ; preds = %entry
1002 %rem = urem i32 %id, %s
1003 %icmp = tail call i64 @llvm.amdgcn.icmp.i64.i32(i32 %rem, i32 0, i32 32)
1004 %shr = lshr i64 %icmp, 1
1005 %notmask = shl nsw i64 -1, 0
1006 %and = and i64 %notmask, %shr
1007 %or = or i64 %and, -9223372036854775808
1008 %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
1009 %cast = trunc i64 %cttz to i32
1010 %cmp3 = icmp ugt i32 10, %cast
1011 %cmp6 = icmp ne i32 %rem, 0
1012 %brmerge = or i1 %cmp6, %cmp3
1013 br i1 %brmerge, label %if.end2, label %if.then
1015 if.then: ; preds = %if.end
1018 if.end2: ; preds = %if.end
1022 ; GCN-LABEL: {{^}}fcmp64:
1023 ; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
1024 ; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
1025 define amdgpu_kernel void @fcmp64(float %n, float %s) {
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
1033 if.end: ; preds = %entry
1034 %rem.f = frem float %id.f, %s
1035 %fcmp = tail call i64 @llvm.amdgcn.fcmp.i64.f32(float %rem.f, float 0.0, i32 1)
1036 %shr = lshr i64 %fcmp, 1
1037 %notmask = shl nsw i64 -1, 0
1038 %and = and i64 %notmask, %shr
1039 %or = or i64 %and, -9223372036854775808
1040 %cttz = tail call i64 @llvm.cttz.i64(i64 %or, i1 true)
1041 %cast = trunc i64 %cttz to i32
1042 %cmp3 = icmp ugt i32 10, %cast
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
1047 if.then: ; preds = %if.end
1050 if.end2: ; preds = %if.end
1054 ; GCN-LABEL: {{^}}icmp32:
1055 ; GFX1032: v_cmp_eq_u32_e32 vcc_lo, 0, v
1056 ; GFX1064: v_cmp_eq_u32_e32 vcc, 0, v
1057 define amdgpu_kernel void @icmp32(i32 %n, i32 %s) {
1059 %id = tail call i32 @llvm.amdgcn.workitem.id.x()
1060 %mul4 = mul nsw i32 %s, %n
1061 %cmp = icmp slt i32 0, %mul4
1064 if.end: ; preds = %entry
1065 %rem = urem i32 %id, %s
1066 %icmp = tail call i32 @llvm.amdgcn.icmp.i32.i32(i32 %rem, i32 0, i32 32)
1067 %shr = lshr i32 %icmp, 1
1068 %notmask = shl nsw i32 -1, 0
1069 %and = and i32 %notmask, %shr
1070 %or = or i32 %and, 2147483648
1071 %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1072 %cmp3 = icmp ugt i32 10, %cttz
1073 %cmp6 = icmp ne i32 %rem, 0
1074 %brmerge = or i1 %cmp6, %cmp3
1075 br i1 %brmerge, label %if.end2, label %if.then
1077 if.then: ; preds = %if.end
1080 if.end2: ; preds = %if.end
1084 ; GCN-LABEL: {{^}}fcmp32:
1085 ; GFX1032: v_cmp_eq_f32_e32 vcc_lo, 0, v
1086 ; GFX1064: v_cmp_eq_f32_e32 vcc, 0, v
1087 define amdgpu_kernel void @fcmp32(float %n, float %s) {
1089 %id = tail call i32 @llvm.amdgcn.workitem.id.x()
1090 %id.f = uitofp i32 %id to float
1091 %mul4 = fmul float %s, %n
1092 %cmp = fcmp ult float 0.0, %mul4
1095 if.end: ; preds = %entry
1096 %rem.f = frem float %id.f, %s
1097 %fcmp = tail call i32 @llvm.amdgcn.fcmp.i32.f32(float %rem.f, float 0.0, i32 1)
1098 %shr = lshr i32 %fcmp, 1
1099 %notmask = shl nsw i32 -1, 0
1100 %and = and i32 %notmask, %shr
1101 %or = or i32 %and, 2147483648
1102 %cttz = tail call i32 @llvm.cttz.i32(i32 %or, i1 true)
1103 %cmp3 = icmp ugt i32 10, %cttz
1104 %cmp6 = fcmp one float %rem.f, 0.0
1105 %brmerge = or i1 %cmp6, %cmp3
1106 br i1 %brmerge, label %if.end2, label %if.then
1108 if.then: ; preds = %if.end
1111 if.end2: ; preds = %if.end
1115 declare void @external_void_func_void() #1
1117 ; Test save/restore of VGPR needed for SGPR spilling.
1119 ; GCN-LABEL: {{^}}callee_no_stack_with_call:
1121 ; GCN-NEXT: s_waitcnt_vscnt
1123 ; GFX1064-NEXT: s_or_saveexec_b64 [[COPY_EXEC0:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1124 ; GFX1032-NEXT: s_or_saveexec_b32 [[COPY_EXEC0:s[0-9]+]], -1{{$}}
1125 ; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
1126 ; GCN-NEXT: s_waitcnt_depctr 0xffe3
1127 ; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC0]]
1128 ; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC0]]
1130 ; GCN-NEXT: v_writelane_b32 v40, s33, 2
1131 ; GCN: s_mov_b32 s33, s32
1132 ; GFX1064: s_addk_i32 s32, 0x400
1133 ; GFX1032: s_addk_i32 s32, 0x200
1136 ; GCN-DAG: v_writelane_b32 v40, s30, 0
1137 ; GCN-DAG: v_writelane_b32 v40, s31, 1
1139 ; GCN-DAG: v_readlane_b32 s4, v40, 0
1140 ; GCN-DAG: v_readlane_b32 s5, v40, 1
1143 ; GFX1064: s_addk_i32 s32, 0xfc00
1144 ; GFX1032: s_addk_i32 s32, 0xfe00
1145 ; GCN: v_readlane_b32 s33, v40, 2
1146 ; GFX1064: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
1147 ; GFX1032: s_or_saveexec_b32 [[COPY_EXEC1:s[0-9]]], -1{{$}}
1148 ; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
1149 ; GCN-NEXT: s_waitcnt_depctr 0xffe3
1150 ; GFX1064-NEXT: s_mov_b64 exec, [[COPY_EXEC1]]
1151 ; GFX1032-NEXT: s_mov_b32 exec_lo, [[COPY_EXEC1]]
1152 ; GCN-NEXT: s_waitcnt vmcnt(0)
1153 ; GCN-NEXT: s_setpc_b64
1154 define void @callee_no_stack_with_call() #1 {
1155 call void @external_void_func_void()
1160 declare i32 @llvm.amdgcn.workitem.id.x()
1161 declare float @llvm.fabs.f32(float)
1162 declare { float, i1 } @llvm.amdgcn.div.scale.f32(float, float, i1)
1163 declare { double, i1 } @llvm.amdgcn.div.scale.f64(double, double, i1)
1164 declare float @llvm.amdgcn.div.fmas.f32(float, float, float, i1)
1165 declare double @llvm.amdgcn.div.fmas.f64(double, double, double, i1)
1166 declare i1 @llvm.amdgcn.class.f32(float, i32)
1167 declare i32 @llvm.amdgcn.set.inactive.i32(i32, i32)
1168 declare i64 @llvm.amdgcn.set.inactive.i64(i64, i64)
1169 declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1170 declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)
1171 declare float @llvm.amdgcn.strict.wwm.f32(float)
1172 declare float @llvm.amdgcn.wwm.f32(float)
1173 declare i32 @llvm.amdgcn.wqm.i32(i32)
1174 declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32)
1175 declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32)
1176 declare float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32>, i32, i32, i32, i32 immarg)
1177 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32)
1178 declare i32 @llvm.amdgcn.mbcnt.hi(i32, i32)
1179 declare i64 @llvm.amdgcn.fcmp.i64.f32(float, float, i32)
1180 declare i64 @llvm.amdgcn.icmp.i64.i32(i32, i32, i32)
1181 declare i32 @llvm.amdgcn.fcmp.i32.f32(float, float, i32)
1182 declare i32 @llvm.amdgcn.icmp.i32.i32(i32, i32, i32)
1183 declare void @llvm.amdgcn.kill(i1)
1184 declare i1 @llvm.amdgcn.wqm.vote(i1)
1185 declare i1 @llvm.amdgcn.ps.live()
1186 declare i64 @llvm.cttz.i64(i64, i1)
1187 declare i32 @llvm.cttz.i32(i32, i1)
1188 declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #5
1190 attributes #0 = { nounwind readnone speculatable }
1191 attributes #1 = { nounwind }
1192 attributes #2 = { nounwind readnone optnone noinline }
1193 attributes #3 = { "target-features"="+wavefrontsize32" }
1194 attributes #4 = { "target-features"="+wavefrontsize64" }
1195 attributes #5 = { inaccessiblememonly nounwind }