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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2024 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; REQUIRES: regkeys, llvm-14-plus
;
; RUN: igc_opt --opaque-pointers %s -S -o - --igc-bfloat-funcs-resolution | FileCheck %s
; This test verifies if bfloat OCL validation functions are lowered correctly.
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-n8:16:32"
target triple = "spir64-unknown-unknown"
; Function Attrs: nounwind
declare spir_func zeroext i16 @_Z18__builtin_bf16_mintt(i16 zeroext, i16 zeroext) #0
; Function Attrs: nounwind
declare spir_func <2 x i16> @_Z18__builtin_bf16_minDv2_tS_(<2 x i16>, <2 x i16>) #0
; Function Attrs: nounwind
declare spir_func <4 x i16> @_Z18__builtin_bf16_minDv4_tS_(<4 x i16>, <4 x i16>) #0
; Function Attrs: nounwind
declare spir_func <8 x i16> @_Z18__builtin_bf16_minDv8_tS_(<8 x i16>, <8 x i16>) #0
; Function Attrs: nounwind
declare spir_func <16 x i16> @_Z18__builtin_bf16_minDv16_tS_(<16 x i16>, <16 x i16>) #0
; Function Attrs: nounwind
declare spir_func zeroext i16 @_Z18__builtin_bf16_maxtt(i16 zeroext, i16 zeroext) #0
; Function Attrs: nounwind
declare spir_func <2 x i16> @_Z18__builtin_bf16_maxDv2_tS_(<2 x i16>, <2 x i16>) #0
; Function Attrs: nounwind
declare spir_func <4 x i16> @_Z18__builtin_bf16_maxDv4_tS_(<4 x i16>, <4 x i16>) #0
; Function Attrs: nounwind
declare spir_func <8 x i16> @_Z18__builtin_bf16_maxDv8_tS_(<8 x i16>, <8 x i16>) #0
; Function Attrs: nounwind
declare spir_func <16 x i16> @_Z18__builtin_bf16_maxDv16_tS_(<16 x i16>, <16 x i16>) #0
; Function Attrs: convergent nounwind
; CHECK: define spir_kernel void @test_min
define spir_kernel void @test_min(i16 addrspace(1)* %out1, i16 zeroext %v1_1, i16 zeroext %v2_1, <2 x i16> addrspace(1)* %out2, <2 x i16> %v1_2, <2 x i16> %v2_2, <4 x i16> addrspace(1)* %out4, <4 x i16> %v1_4, <4 x i16> %v2_4, <8 x i16> addrspace(1)* %out8, <8 x i16> %v1_8, <8 x i16> %v2_8, <16 x i16> addrspace(1)* %out16, <16 x i16> %v1_16, <16 x i16> %v2_16) #1 {
entry:
; CHECK: %[[SRC0BF:.*]] = bitcast i16 %v1_1 to bfloat
; CHECK: %[[SRC1BF:.*]] = bitcast i16 %v2_1 to bfloat
; CHECK: %[[COND1:.*]] = fcmp olt bfloat %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select i1 %[[COND1]], bfloat %[[SRC0BF]], bfloat %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno bfloat %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select i1 %[[COND2]], bfloat %[[SRC1BF]], bfloat %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord bfloat %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select i1 %[[COND3]], bfloat %[[SELECT1RES]], bfloat %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast bfloat %[[SELECT3RES]] to i16
%call = call spir_func zeroext i16 @_Z18__builtin_bf16_mintt(i16 zeroext %v1_1, i16 zeroext %v2_1) #2
%arrayidx = getelementptr inbounds i16, i16 addrspace(1)* %out1, i64 0
store i16 %call, i16 addrspace(1)* %arrayidx, align 2
; CHECK: %[[SRC0BF:.*]] = bitcast <2 x i16> %v1_2 to <2 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <2 x i16> %v2_2 to <2 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp olt <2 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <2 x i1> %[[COND1]], <2 x bfloat> %[[SRC0BF]], <2 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <2 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <2 x i1> %[[COND2]], <2 x bfloat> %[[SRC1BF]], <2 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <2 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <2 x i1> %[[COND3]], <2 x bfloat> %[[SELECT1RES]], <2 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <2 x bfloat> %[[SELECT3RES]] to <2 x i16>
%call1 = call spir_func <2 x i16> @_Z18__builtin_bf16_minDv2_tS_(<2 x i16> %v1_2, <2 x i16> %v2_2) #2
%arrayidx2 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out2, i64 1
store <2 x i16> %call1, <2 x i16> addrspace(1)* %arrayidx2, align 4
; CHECK: %[[SRC0BF:.*]] = bitcast <4 x i16> %v1_4 to <4 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <4 x i16> %v2_4 to <4 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp olt <4 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <4 x i1> %[[COND1]], <4 x bfloat> %[[SRC0BF]], <4 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <4 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <4 x i1> %[[COND2]], <4 x bfloat> %[[SRC1BF]], <4 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <4 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <4 x i1> %[[COND3]], <4 x bfloat> %[[SELECT1RES]], <4 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <4 x bfloat> %[[SELECT3RES]] to <4 x i16>
%call3 = call spir_func <4 x i16> @_Z18__builtin_bf16_minDv4_tS_(<4 x i16> %v1_4, <4 x i16> %v2_4) #2
%arrayidx4 = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out4, i64 2
store <4 x i16> %call3, <4 x i16> addrspace(1)* %arrayidx4, align 8
; CHECK: %[[SRC0BF:.*]] = bitcast <8 x i16> %v1_8 to <8 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <8 x i16> %v2_8 to <8 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp olt <8 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <8 x i1> %[[COND1]], <8 x bfloat> %[[SRC0BF]], <8 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <8 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <8 x i1> %[[COND2]], <8 x bfloat> %[[SRC1BF]], <8 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <8 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <8 x i1> %[[COND3]], <8 x bfloat> %[[SELECT1RES]], <8 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <8 x bfloat> %[[SELECT3RES]] to <8 x i16>
%call5 = call spir_func <8 x i16> @_Z18__builtin_bf16_minDv8_tS_(<8 x i16> %v1_8, <8 x i16> %v2_8) #2
%arrayidx6 = getelementptr inbounds <8 x i16>, <8 x i16> addrspace(1)* %out8, i64 3
store <8 x i16> %call5, <8 x i16> addrspace(1)* %arrayidx6, align 16
; CHECK: %[[SRC0BF:.*]] = bitcast <16 x i16> %v1_16 to <16 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <16 x i16> %v2_16 to <16 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp olt <16 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <16 x i1> %[[COND1]], <16 x bfloat> %[[SRC0BF]], <16 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <16 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <16 x i1> %[[COND2]], <16 x bfloat> %[[SRC1BF]], <16 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <16 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <16 x i1> %[[COND3]], <16 x bfloat> %[[SELECT1RES]], <16 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <16 x bfloat> %[[SELECT3RES]] to <16 x i16>
%call7 = call spir_func <16 x i16> @_Z18__builtin_bf16_minDv16_tS_(<16 x i16> %v1_16, <16 x i16> %v2_16) #2
%arrayidx8 = getelementptr inbounds <16 x i16>, <16 x i16> addrspace(1)* %out16, i64 4
store <16 x i16> %call7, <16 x i16> addrspace(1)* %arrayidx8, align 32
ret void
}
; Function Attrs: convergent nounwind
; CHECK: define spir_kernel void @test_max
define spir_kernel void @test_max(i16 addrspace(1)* %out1, i16 zeroext %v1_1, i16 zeroext %v2_1, <2 x i16> addrspace(1)* %out2, <2 x i16> %v1_2, <2 x i16> %v2_2, <4 x i16> addrspace(1)* %out4, <4 x i16> %v1_4, <4 x i16> %v2_4, <8 x i16> addrspace(1)* %out8, <8 x i16> %v1_8, <8 x i16> %v2_8, <16 x i16> addrspace(1)* %out16, <16 x i16> %v1_16, <16 x i16> %v2_16) #1 {
entry:
; CHECK: %[[SRC0BF:.*]] = bitcast i16 %v1_1 to bfloat
; CHECK: %[[SRC1BF:.*]] = bitcast i16 %v2_1 to bfloat
; CHECK: %[[COND1:.*]] = fcmp ogt bfloat %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select i1 %[[COND1]], bfloat %[[SRC0BF]], bfloat %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno bfloat %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select i1 %[[COND2]], bfloat %[[SRC1BF]], bfloat %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord bfloat %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select i1 %[[COND3]], bfloat %[[SELECT1RES]], bfloat %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast bfloat %[[SELECT3RES]] to i16
%call = call spir_func zeroext i16 @_Z18__builtin_bf16_maxtt(i16 zeroext %v1_1, i16 zeroext %v2_1) #2
%arrayidx = getelementptr inbounds i16, i16 addrspace(1)* %out1, i64 0
store i16 %call, i16 addrspace(1)* %arrayidx, align 2
; CHECK: %[[SRC0BF:.*]] = bitcast <2 x i16> %v1_2 to <2 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <2 x i16> %v2_2 to <2 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp ogt <2 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <2 x i1> %[[COND1]], <2 x bfloat> %[[SRC0BF]], <2 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <2 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <2 x i1> %[[COND2]], <2 x bfloat> %[[SRC1BF]], <2 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <2 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <2 x i1> %[[COND3]], <2 x bfloat> %[[SELECT1RES]], <2 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <2 x bfloat> %[[SELECT3RES]] to <2 x i16>
%call1 = call spir_func <2 x i16> @_Z18__builtin_bf16_maxDv2_tS_(<2 x i16> %v1_2, <2 x i16> %v2_2) #2
%arrayidx2 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out2, i64 1
store <2 x i16> %call1, <2 x i16> addrspace(1)* %arrayidx2, align 4
; CHECK: %[[SRC0BF:.*]] = bitcast <4 x i16> %v1_4 to <4 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <4 x i16> %v2_4 to <4 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp ogt <4 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <4 x i1> %[[COND1]], <4 x bfloat> %[[SRC0BF]], <4 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <4 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <4 x i1> %[[COND2]], <4 x bfloat> %[[SRC1BF]], <4 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <4 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <4 x i1> %[[COND3]], <4 x bfloat> %[[SELECT1RES]], <4 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <4 x bfloat> %[[SELECT3RES]] to <4 x i16>
%call3 = call spir_func <4 x i16> @_Z18__builtin_bf16_maxDv4_tS_(<4 x i16> %v1_4, <4 x i16> %v2_4) #2
%arrayidx4 = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out4, i64 2
store <4 x i16> %call3, <4 x i16> addrspace(1)* %arrayidx4, align 8
; CHECK: %[[SRC0BF:.*]] = bitcast <8 x i16> %v1_8 to <8 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <8 x i16> %v2_8 to <8 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp ogt <8 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <8 x i1> %[[COND1]], <8 x bfloat> %[[SRC0BF]], <8 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <8 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <8 x i1> %[[COND2]], <8 x bfloat> %[[SRC1BF]], <8 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <8 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <8 x i1> %[[COND3]], <8 x bfloat> %[[SELECT1RES]], <8 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <8 x bfloat> %[[SELECT3RES]] to <8 x i16>
%call5 = call spir_func <8 x i16> @_Z18__builtin_bf16_maxDv8_tS_(<8 x i16> %v1_8, <8 x i16> %v2_8) #2
%arrayidx6 = getelementptr inbounds <8 x i16>, <8 x i16> addrspace(1)* %out8, i64 3
store <8 x i16> %call5, <8 x i16> addrspace(1)* %arrayidx6, align 16
; CHECK: %[[SRC0BF:.*]] = bitcast <16 x i16> %v1_16 to <16 x bfloat>
; CHECK: %[[SRC1BF:.*]] = bitcast <16 x i16> %v2_16 to <16 x bfloat>
; CHECK: %[[COND1:.*]] = fcmp ogt <16 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT1RES:.*]] = select <16 x i1> %[[COND1]], <16 x bfloat> %[[SRC0BF]], <16 x bfloat> %[[SRC1BF]]
; CHECK: %[[COND2:.*]] = fcmp uno <16 x bfloat> %[[SRC0BF]], %[[SRC0BF]]
; CHECK: %[[SELECT2RES:.*]] = select <16 x i1> %[[COND2]], <16 x bfloat> %[[SRC1BF]], <16 x bfloat> %[[SRC0BF]]
; CHECK: %[[COND3:.*]] = fcmp ord <16 x bfloat> %[[SRC0BF]], %[[SRC1BF]]
; CHECK: %[[SELECT3RES:.*]] = select <16 x i1> %[[COND3]], <16 x bfloat> %[[SELECT1RES]], <16 x bfloat> %[[SELECT2RES]]
; CHECK: %{{.*}} = bitcast <16 x bfloat> %[[SELECT3RES]] to <16 x i16>
%call7 = call spir_func <16 x i16> @_Z18__builtin_bf16_maxDv16_tS_(<16 x i16> %v1_16, <16 x i16> %v2_16) #2
%arrayidx8 = getelementptr inbounds <16 x i16>, <16 x i16> addrspace(1)* %out16, i64 4
store <16 x i16> %call7, <16 x i16> addrspace(1)* %arrayidx8, align 32
ret void
}
attributes #0 = { nounwind "referenced-indirectly" "visaStackCall" }
attributes #1 = { convergent nounwind }
attributes #2 = { nounwind }
|
