1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
|
; RUN: llc -amdgpu-scalar-ir-passes=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
; GCN-LABEL: {{^}}select_undef_lhs:
; GCN: s_waitcnt
; GCN-NOT: v_cmp
; GCN-NOT: v_cndmask
; GCN-NEXT: s_setpc_b64
define float @select_undef_lhs(float %val, i1 %cond) {
%sel = select i1 %cond, float undef, float %val
ret float %sel
}
; GCN-LABEL: {{^}}select_undef_rhs:
; GCN: s_waitcnt
; GCN-NOT: v_cmp
; GCN-NOT: v_cndmask
; GCN-NEXT: s_setpc_b64
define float @select_undef_rhs(float %val, i1 %cond) {
%sel = select i1 %cond, float %val, float undef
ret float %sel
}
; GCN-LABEL: {{^}}select_undef_n1:
; GCN: v_mov_b32_e32 [[RES:v[0-9]+]], 1.0
; GCN: store_dword {{[^,]+}}, [[RES]]
define void @select_undef_n1(ptr addrspace(1) %a, i32 %c) {
%cc = icmp eq i32 %c, 0
%sel = select i1 %cc, float 1.000000e+00, float undef
store float %sel, ptr addrspace(1) %a
ret void
}
; GCN-LABEL: {{^}}select_undef_n2:
; GCN: v_mov_b32_e32 [[RES:v[0-9]+]], 1.0
; GCN: store_dword {{[^,]+}}, [[RES]]
define void @select_undef_n2(ptr addrspace(1) %a, i32 %c) {
%cc = icmp eq i32 %c, 0
%sel = select i1 %cc, float undef, float 1.000000e+00
store float %sel, ptr addrspace(1) %a
ret void
}
declare float @llvm.amdgcn.rcp.f32(float)
; Make sure the vector undef isn't lowered into 0s.
; GCN-LABEL: {{^}}undef_v6f32:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v6f32(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <6 x float> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <6 x float>, ptr addrspace(3) undef
%add = fadd <6 x float> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <6 x float> %add, ptr addrspace(3) undef
ret void
}
; GCN-LABEL: {{^}}undef_v6i32:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v6i32(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <6 x i32> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <6 x i32>, ptr addrspace(3) undef
%add = add <6 x i32> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <6 x i32> %add, ptr addrspace(3) undef
ret void
}
; Make sure the vector undef isn't lowered into 0s.
; GCN-LABEL: {{^}}undef_v5f32:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v5f32(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <5 x float> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <5 x float>, ptr addrspace(3) undef
%add = fadd <5 x float> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <5 x float> %add, ptr addrspace(3) undef
ret void
}
; GCN-LABEL: {{^}}undef_v5i32:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v5i32(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <5 x i32> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <5 x i32>, ptr addrspace(3) undef
%add = add <5 x i32> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <5 x i32> %add, ptr addrspace(3) undef
ret void
}
; Make sure the vector undef isn't lowered into 0s.
; GCN-LABEL: {{^}}undef_v3f64:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v3f64(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <3 x double> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <3 x double>, ptr addrspace(3) %ptr
%add = fadd <3 x double> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <3 x double> %add, ptr addrspace(3) %ptr
ret void
}
; GCN-LABEL: {{^}}undef_v3i64:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v3i64(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <3 x i64> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <3 x i64>, ptr addrspace(3) %ptr
%add = add <3 x i64> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <3 x i64> %add, ptr addrspace(3) %ptr
ret void
}
; Make sure the vector undef isn't lowered into 0s.
; GCN-LABEL: {{^}}undef_v4f16:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v4f16(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <4 x half> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <4 x half>, ptr addrspace(3) %ptr
%add = fadd <4 x half> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <4 x half> %add, ptr addrspace(3) %ptr
ret void
}
; GCN-LABEL: {{^}}undef_v4i16:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v4i16(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <4 x i16> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <4 x i16>, ptr addrspace(3) %ptr
%add = add <4 x i16> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <4 x i16> %add, ptr addrspace(3) %ptr
ret void
}
; Make sure the vector undef isn't lowered into 0s.
; GCN-LABEL: {{^}}undef_v2f16:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v2f16(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <2 x half> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <2 x half>, ptr addrspace(3) %ptr
%add = fadd <2 x half> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <2 x half> %add, ptr addrspace(3) %ptr
ret void
}
; GCN-LABEL: {{^}}undef_v2i16:
; GCN-NOT: v_mov_b32_e32 v{{[0-9]+}}, 0
; GCN-NOT: s_mov_b32 s{{[0-9]+}}, 0
; GCN: s_cbranch_vccnz
define amdgpu_kernel void @undef_v2i16(ptr addrspace(3) %ptr, i1 %cond) {
entry:
br label %loop
loop:
%phi = phi <2 x i16> [ undef, %entry ], [ %add, %loop ]
%load = load volatile <2 x i16>, ptr addrspace(3) %ptr
%add = add <2 x i16> %load, %phi
br i1 %cond, label %loop, label %ret
ret:
store volatile <2 x i16> %add, ptr addrspace(3) %ptr
ret void
}
; We were expanding undef vectors into zero vectors. Optimizations
; would then see we used no elements of the vector, and reform the
; undef vector resulting in a combiner loop.
; GCN-LABEL: {{^}}inf_loop_undef_vector:
; GCN: s_waitcnt
; GCN-NEXT: v_mad_u64_u32
; GCN-NEXT: v_mul_lo_u32
; GCN-NEXT: v_mul_lo_u32
; GCN-NEXT: v_add3_u32
; GCN-NEXT: global_store_dwordx2
define void @inf_loop_undef_vector(<6 x float> %arg, float %arg1, i64 %arg2) {
%i = insertelement <6 x float> %arg, float %arg1, i64 2
%i3 = bitcast <6 x float> %i to <3 x i64>
%i4 = extractelement <3 x i64> %i3, i64 0
%i5 = extractelement <3 x i64> %i3, i64 1
%i6 = mul i64 %i5, %arg2
%i7 = add i64 %i6, %i4
store volatile i64 %i7, ptr addrspace(1) undef, align 4
ret void
}
|
