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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2022-2023 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; COM: This test verifies whether an SLM global variable uses are updated correctly
; COM: in the following cyclic graph:
; kernelA kernelB
; / \
; f0 f3
; / \\ /
; / \\ /
; f1 f2 /
; \ / /
; foo /
; \ /
; \ /
; \ /
; \ /
; \ /
; bar
; RUN: %opt %use_old_pass_manager% -GenXSLMResolution -march=genx64 -mcpu=Gen9 -S < %s | FileCheck %s
target datalayout = "e-p:64:64-i64:64-n8:16:32"
@SLM_GV = internal addrspace(3) global [4 x i32] undef, align 4 #0
; COM: @SLM_GV access in this node is not supported.
define internal spir_func i32 @bar(i32 addrspace(3)* %arg) #1 {
%arg.ld = load i32, i32 addrspace(3)* %arg, align 4
ret i32 %arg.ld
}
; CHECK: define internal spir_func i32 @foo
; CHECK: %bar.res = call spir_func i32 @bar(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 1))
define internal spir_func i32 @foo(i32 addrspace(3)* %arg) #1 {
%arg.ld = load i32, i32 addrspace(3)* %arg, align 4
%bar.res = call spir_func i32 @bar(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 1))
%res = add i32 %bar.res, %arg.ld
ret i32 %res
}
; CHECK: define internal spir_func i32 @f1
; CHECK: %gv.p3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 2
define internal spir_func i32 @f1() #1 {
%gv.p3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 2
%foo.res = call spir_func i32 @foo(i32 addrspace(3)* %gv.p3)
ret i32 %foo.res
}
; CHECK: define internal spir_func i32 @f2
define internal spir_func i32 @f2() #1 {
; CHECK: %gv.p3.0 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64
%gv.p3.0 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 0
%gv.ld.0 = load i32, i32 addrspace(3)* %gv.p3.0, align 4
; CHECK: %gv.p3.3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 3
%gv.p3.3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 3
%gv.ld.3 = load i32, i32 addrspace(3)* %gv.p3.3, align 4
%sum = add i32 %gv.ld.0, %gv.ld.3
store i32 %sum, i32 addrspace(3)* %gv.p3.0
%f0.res = call spir_func i32 @f0(i32 addrspace(3)* %gv.p3.0)
%foo.res = call spir_func i32 @foo(i32 addrspace(3)* %gv.p3.3)
%sum.res = add i32 %f0.res, %foo.res
ret i32 %sum.res
}
; CHECK: define internal spir_func i32 @f0
define internal spir_func i32 @f0(i32 addrspace(3)* %arg) #1 {
%arg.ld = load i32, i32 addrspace(3)* %arg, align 4
; CHECK: %gv.p3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 0
%gv.p3 = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 0
%gv.ld = load i32, i32 addrspace(3)* %gv.p3, align 4
%sum = add i32 %arg.ld, %gv.ld
%c = icmp uge i32 %sum, 25
br i1 %c, label %call_f1, label %call_f2
call_f1:
%f1.res = call spir_func i32 @f1()
br label %exit
call_f2:
%f2.res = call spir_func i32 @f2()
br label %exit
exit:
%f0.res = phi i32 [ %f1.res, %call_f1 ], [ %f2.res, %call_f2 ]
ret i32 %f0.res
}
; CHECK: define internal spir_func i32 @f3
; CHECK: %bar.res = call spir_func i32 @bar(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 0))
define internal spir_func i32 @f3(i32 addrspace(3)* %arg) #1 {
%bar.res = call spir_func i32 @bar(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 0))
%arg.ld = load i32, i32 addrspace(3)* %arg, align 4
%sum = add i32 %bar.res, %arg.ld
ret i32 %sum
}
; CHECK: define dllexport spir_kernel void @kernelA
; CHECK: %res = call spir_func i32 @f0(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 268435456 to [4 x i32] addrspace(3)*), i64 0, i64 0))
define dllexport spir_kernel void @kernelA() #2 {
%res = call spir_func i32 @f0(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 0))
ret void
}
; CHECK: define dllexport spir_kernel void @kernelB
; CHECK: %res = call spir_func i32 @f3(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* inttoptr (i32 64 to [4 x i32] addrspace(3)*), i64 0, i64 3))
define dllexport spir_kernel void @kernelB() #2 {
%res = call spir_func i32 @f3(i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @SLM_GV, i64 0, i64 3))
ret void
}
; CHECK: !{{[[:digit:]]}} = !{void ()* @kernelA, !"kernelA", !{{[[:digit:]]}}, i32 16, !{{[[:digit:]]}}, !{{[[:digit:]]}}, !{{[[:digit:]]}}, i32 0}
; CHECK: !{{[[:digit:]]}} = !{void ()* @kernelB, !"kernelB", !{{[[:digit:]]}}, i32 80, !{{[[:digit:]]}}, !{{[[:digit:]]}}, !{{[[:digit:]]}}, i32 0}
attributes #0 = { "VCGlobalVariable" }
attributes #1 = { noinline nounwind }
attributes #2 = { noinline nounwind "CMGenxMain" }
!genx.kernels = !{!0, !1}
!genx.kernel.internal = !{!4, !5}
; COM: Initial slm size for kernelA is 0
!0 = !{void ()* @kernelA, !"kernelA", !2, i32 0, !2, !2, !3, i32 0}
; COM: Initial slm size for kernelB is 64
!1 = !{void ()* @kernelB, !"kernelB", !2, i32 64, !2, !2, !3, i32 0}
!2 = !{}
!3 = !{!""}
!4 = !{void ()* @kernelA, !2, !2, !2, !2}
!5 = !{void ()* @kernelB, !2, !2, !2, !2}
|
