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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2021-2025 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; RUN: %opt_typed_ptrs %use_old_pass_manager% -CMKernelArgOffset -march=genx64 -mcpu=XeHPG -S < %s | FileCheck %s --check-prefixes=CHECK,CHECK-TYPED-PTRS
; RUN: %opt_opaque_ptrs %use_old_pass_manager% -CMKernelArgOffset -march=genx64 -mcpu=XeHPG -S < %s | FileCheck %s --check-prefixes=CHECK,CHECK-OPAQUE-PTRS
; RUN: %opt_new_pm_typed -passes=CMKernelArgOffset -march=genx64 -mcpu=XeHPG -S < %s | FileCheck %s --check-prefixes=CHECK,CHECK-TYPED-PTRS
; RUN: %opt_new_pm_opaque -passes=CMKernelArgOffset -march=genx64 -mcpu=XeHPG -S < %s | FileCheck %s --check-prefixes=CHECK,CHECK-OPAQUE-PTRS
target datalayout = "e-p:64:64-i64:64-n8:16:32:64"
target triple = "spir64-unknown-unknown"
; CHECK: [[STRUCT:%struct.state]]
%struct.state = type { i8, [3 x i32], float }
; COM: Find the kernel definition
; CHECK: define dllexport [[FTYPE:void]] @[[FNAME:foo]]
; COM: Explicit arguments. The first argument must be byval svmptr_t
; CHECK-TYPED-PTRS-SAME: [[STRUCT]]* byval(%struct.state) %_arg_
; CHECK-TYPED-PTRS-SAME: [[EXPARG2:i32 addrspace\(1\)\*]] %_arg_1
; CHECK-OPAQUE-PTRS-SAME: ptr byval(%struct.state) %_arg_
; CHECK-OPAQUE-PTRS-SAME: ptr addrspace(1) %_arg_1
; COM: Private base
; CHECK-SAME: [[PRIVBASE:i64]]
; COM: Implicit linearization of %_arg_
; CHECK-SAME: [[IMPLIN1:i8]] [[LIN1NAME:%__arg_lin__arg_.0]]
; CHECK-SAME: [[IMPLIN2:i32]] %__arg_lin__arg_.4
; CHECK-SAME: [[IMPLIN3:i32]] %__arg_lin__arg_.8
; CHECK-SAME: [[IMPLIN4:i32]] %__arg_lin__arg_.12
; CHECK-SAME: [[IMPLIN5:float]] %__arg_lin__arg_.16
define dllexport void @foo(%struct.state* byval(%struct.state) %_arg_, i32 addrspace(1)* %_arg_1, i64 %privBase, i8 %__arg_lin__arg_.0, i32 %__arg_lin__arg_.4, i32 %__arg_lin__arg_.8, i32 %__arg_lin__arg_.12, float %__arg_lin__arg_.16) #0 {
entry:
; COM: Check linearization transformation.
; COM: [[STRUCT]] allocation.
; CHECK: [[ALLOCA:%_arg_.linearization]] = alloca [[STRUCT]]
; CHECK-TYPED-PTRS: [[I8P:%_arg_.linearization.i8]] = bitcast [[STRUCT]]* [[ALLOCA]] to i8*
;
; COM: The first linearization arg has i8 type, so it must be stored as-is in the right position
; CHECK-TYPED-PTRS: [[LIN1POS:%0]] = getelementptr i8, i8* [[I8P]], i32 [[LIN1OFF:0]]
; CHECK-TYPED-PTRS: store [[IMPLIN1]] [[LIN1NAME]], [[IMPLIN1]]* [[LIN1POS]]
; CHECK-OPAQUE-PTRS: [[LIN1POS:%0]] = getelementptr i8, ptr [[ALLOCA]], i32 [[LIN1OFF:0]]
; CHECK-OPAQUE-PTRS: store [[IMPLIN1]] [[LIN1NAME]], ptr [[LIN1POS]]
;
; COM: All the next linearization arguments
; COM: The check-count-4 itselt is confusing, but it checks the following sequence (additionally checking types correctness):
; %1 = getelementptr i8, i8* %_arg_.linearization.i8, i32 4
; %2 = bitcast i8* %1 to i32*
; store i32 %__arg_lin__arg_.4, i32* %2
; CHECK-TYPED-PTRS-COUNT-4: [[POS:%[0-9]+]] = getelementptr {{.+}}[[I8P]]{{.+[[:space:]]+}}[[CAST:%[0-9]+]] = bitcast i8* [[POS]] to [[LINTY:[[:alnum:]]+]]*{{[[:space:]]+}}store [[LINTY]]{{.+}} [[LINTY]]* [[CAST]]
; CHECK-OPAQUE-PTRS-COUNT-3: [[POS:%[0-9]+]] = getelementptr i8, ptr [[ALLOCA]]{{.+[[:space:]]+}}store i32 {{.+}}, ptr [[POS]]
; CHECK-OPAQUE-PTRS-NEXT: [[LIN5POS:%[0-9]+]] = getelementptr i8, ptr [[ALLOCA]]
; CHECK-OPAQUE-PTRS-NEXT: store float {{.+}}, ptr [[LIN5POS]]
;
; COM: Uses of the linearized %_arg_ must be replaced with ALLOCA
; CHECK-TYPED-PTRS: %.sroa.0.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* [[ALLOCA]], i64 0, i32 0
; CHECK-TYPED-PTRS: %.sroa.44.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* [[ALLOCA]], i64 0, i32 1, i64 1
; CHECK-TYPED-PTRS: %.sroa.57.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* [[ALLOCA]], i64 0, i32 2
; CHECK-OPAQUE-PTRS: %.sroa.0.0..sroa_idx = getelementptr inbounds %struct.state, ptr [[ALLOCA]], i64 0, i32 0
; CHECK-OPAQUE-PTRS: %.sroa.44.0..sroa_idx = getelementptr inbounds %struct.state, ptr [[ALLOCA]], i64 0, i32 1, i64 1
; CHECK-OPAQUE-PTRS: %.sroa.57.0..sroa_idx = getelementptr inbounds %struct.state, ptr [[ALLOCA]], i64 0, i32 2
%.sroa.0.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* %_arg_, i64 0, i32 0
%.sroa.0.0.copyload = load i8, i8* %.sroa.0.0..sroa_idx, align 4
%.sroa.44.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* %_arg_, i64 0, i32 1, i64 1
%.sroa.44.0.copyload = load i32, i32* %.sroa.44.0..sroa_idx, align 4
%.sroa.57.0..sroa_idx = getelementptr inbounds %struct.state, %struct.state* %_arg_, i64 0, i32 2
%.sroa.57.0.copyload = load float, float* %.sroa.57.0..sroa_idx, align 4
%conv.i = sitofp i32 %.sroa.44.0.copyload to float
%add.i = fadd float %.sroa.57.0.copyload, %conv.i
%conv3.i = sitofp i8 %.sroa.0.0.copyload to float
%add4.i = fadd float %add.i, %conv3.i
%conv5.i = fptosi float %add4.i to i32
store i32 %conv5.i, i32 addrspace(1)* %_arg_1, align 4
ret void
}
attributes #0 = { nounwind "CMGenxMain" "oclrt"="1" }
!spirv.Source = !{!0}
!opencl.spir.version = !{!1}
!opencl.ocl.version = !{!2}
!opencl.used.extensions = !{!3}
!opencl.used.optional.core.features = !{!3}
!spirv.Generator = !{!4}
!genx.kernels = !{!5}
!genx.kernel.internal = !{!9}
; CHECK: !genx.kernel.internal = !{[[INTERNAL:![0-9]+]]}
!0 = !{i32 0, i32 100000}
!1 = !{i32 1, i32 2}
!2 = !{i32 1, i32 0}
!3 = !{}
!4 = !{i16 6, i16 14}
!5 = !{void (%struct.state*, i32 addrspace(1)*, i64, i8, i32, i32, i32, float)* @foo, !"foo", !6, i32 0, i32 0, !7, !8, i32 0}
!6 = !{i32 112, i32 0, i32 96, i32 104, i32 104, i32 104, i32 104, i32 104}
!7 = !{i32 0, i32 0}
!8 = !{!"svmptr_t", !"svmptr_t"}
!9 = !{void (%struct.state*, i32 addrspace(1)*, i64, i8, i32, i32, i32, float)* @foo, null, null, !10, null}
; COM: Check OffsetInArgs and ArgIndexes
; CHECK-TYPED-PTRS: [[INTERNAL]] = !{[[FTYPE]] ([[STRUCT]]*, [[EXPARG2]], [[PRIVBASE]], [[IMPLIN1]], [[IMPLIN2]], [[IMPLIN3]], [[IMPLIN4]], [[IMPLIN5]])* @[[FNAME]], [[OFFSETINARGS:![0-9]+]], [[ARGINDEXES:![0-9]+]], [[LINMD:![0-9]+]], null, i32 0}
; CHECK-OPAQUE-PTRS: [[INTERNAL]] = !{ptr @[[FNAME]], [[OFFSETINARGS:![0-9]+]], [[ARGINDEXES:![0-9]+]], [[LINMD:![0-9]+]], null, i32 0}
; CHECK: [[OFFSETINARGS]] = !{i32 0, i32 0, i32 0, i32 0, i32 4, i32 8, i32 12, i32 16}
; CHECK: [[ARGINDEXES]] = !{i32 0, i32 1, i32 2, i32 0, i32 0, i32 0, i32 0, i32 0}
!10 = !{!11}
!11 = !{i32 0, !12}
!12 = !{!13, !14, !15, !16, !17}
!13 = !{i32 3, i32 0}
!14 = !{i32 4, i32 4}
!15 = !{i32 5, i32 8}
!16 = !{i32 6, i32 12}
!17 = !{i32 7, i32 16}
; COM: Also check that linearization MD is still here
; CHECK: [[LINMD]] = !{[[ARG1LINMD:![0-9]+]]}
; CHECK: [[ARG1LINMD]] = !{i32 0, [[IMPLIN:![0-9]+]]}
; CHECK: [[IMPLIN]] = !{[[IMPLINMD1:![0-9]+]], [[IMPLINMD2:![0-9]+]], [[IMPLINMD3:![0-9]+]], [[IMPLINMD4:![0-9]+]], [[IMPLINMD5:![0-9]+]]}
; CHECK: [[IMPLINMD1]] = !{i32 3, i32 [[LIN1OFF]]}
; CHECK: [[IMPLINMD2]] = !{i32 4, i32 4}
; CHECK: [[IMPLINMD3]] = !{i32 5, i32 8}
; CHECK: [[IMPLINMD4]] = !{i32 6, i32 12}
; CHECK: [[IMPLINMD5]] = !{i32 7, i32 16}
!18 = !{i32 1}
|
