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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck %s
; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca -disable-promote-alloca-to-lds< %s | FileCheck -check-prefix=NOLDS %s
; This normally would be fixed by instcombine to be compare to the GEP
; indices
define amdgpu_kernel void @lds_promoted_alloca_icmp_same_derived_pointer(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
; CHECK-LABEL: @lds_promoted_alloca_icmp_same_derived_pointer(
; CHECK-NEXT: [[TMP1:%.*]] = call noalias nonnull dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 addrspace(4)* [[TMP1]] to i32 addrspace(4)*
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 1
; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32 addrspace(4)* [[TMP3]], align 4, !invariant.load !0
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 2
; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[TMP5]], align 4, !range [[RNG1:![0-9]+]], !invariant.load !0
; CHECK-NEXT: [[TMP7:%.*]] = lshr i32 [[TMP4]], 16
; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.workitem.id.x(), !range [[RNG2:![0-9]+]]
; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.amdgcn.workitem.id.y(), !range [[RNG2]]
; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.workitem.id.z(), !range [[RNG2]]
; CHECK-NEXT: [[TMP11:%.*]] = mul nuw nsw i32 [[TMP7]], [[TMP6]]
; CHECK-NEXT: [[TMP12:%.*]] = mul i32 [[TMP11]], [[TMP8]]
; CHECK-NEXT: [[TMP13:%.*]] = mul nuw nsw i32 [[TMP9]], [[TMP6]]
; CHECK-NEXT: [[TMP14:%.*]] = add i32 [[TMP12]], [[TMP13]]
; CHECK-NEXT: [[TMP15:%.*]] = add i32 [[TMP14]], [[TMP10]]
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_same_derived_pointer.alloca, i32 0, i32 [[TMP15]]
; CHECK-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[TMP16]], i32 0, i32 [[A:%.*]]
; CHECK-NEXT: [[PTR1:%.*]] = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[TMP16]], i32 0, i32 [[B:%.*]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 addrspace(3)* [[PTR0]], [[PTR1]]
; CHECK-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; CHECK-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
; NOLDS-LABEL: @lds_promoted_alloca_icmp_same_derived_pointer(
; NOLDS-NEXT: [[ALLOCA:%.*]] = alloca [16 x i32], align 4
; NOLDS-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[A:%.*]]
; NOLDS-NEXT: [[PTR1:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[B:%.*]]
; NOLDS-NEXT: [[CMP:%.*]] = icmp eq i32* [[PTR0]], [[PTR1]]
; NOLDS-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; NOLDS-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; NOLDS-NEXT: ret void
;
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%ptr1 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %b
%cmp = icmp eq i32* %ptr0, %ptr1
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
define amdgpu_kernel void @lds_promoted_alloca_icmp_null_rhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
; CHECK-LABEL: @lds_promoted_alloca_icmp_null_rhs(
; CHECK-NEXT: [[TMP1:%.*]] = call noalias nonnull dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 addrspace(4)* [[TMP1]] to i32 addrspace(4)*
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 1
; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32 addrspace(4)* [[TMP3]], align 4, !invariant.load !0
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 2
; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[TMP5]], align 4, !range [[RNG1]], !invariant.load !0
; CHECK-NEXT: [[TMP7:%.*]] = lshr i32 [[TMP4]], 16
; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.workitem.id.x(), !range [[RNG2]]
; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.amdgcn.workitem.id.y(), !range [[RNG2]]
; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.workitem.id.z(), !range [[RNG2]]
; CHECK-NEXT: [[TMP11:%.*]] = mul nuw nsw i32 [[TMP7]], [[TMP6]]
; CHECK-NEXT: [[TMP12:%.*]] = mul i32 [[TMP11]], [[TMP8]]
; CHECK-NEXT: [[TMP13:%.*]] = mul nuw nsw i32 [[TMP9]], [[TMP6]]
; CHECK-NEXT: [[TMP14:%.*]] = add i32 [[TMP12]], [[TMP13]]
; CHECK-NEXT: [[TMP15:%.*]] = add i32 [[TMP14]], [[TMP10]]
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_rhs.alloca, i32 0, i32 [[TMP15]]
; CHECK-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[TMP16]], i32 0, i32 [[A:%.*]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 addrspace(3)* [[PTR0]], null
; CHECK-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; CHECK-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
; NOLDS-LABEL: @lds_promoted_alloca_icmp_null_rhs(
; NOLDS-NEXT: [[ALLOCA:%.*]] = alloca [16 x i32], align 4
; NOLDS-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[A:%.*]]
; NOLDS-NEXT: [[CMP:%.*]] = icmp eq i32* [[PTR0]], null
; NOLDS-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; NOLDS-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; NOLDS-NEXT: ret void
;
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%cmp = icmp eq i32* %ptr0, null
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
define amdgpu_kernel void @lds_promoted_alloca_icmp_null_lhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
; CHECK-LABEL: @lds_promoted_alloca_icmp_null_lhs(
; CHECK-NEXT: [[TMP1:%.*]] = call noalias nonnull dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 addrspace(4)* [[TMP1]] to i32 addrspace(4)*
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 1
; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32 addrspace(4)* [[TMP3]], align 4, !invariant.load !0
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32 addrspace(4)* [[TMP2]], i64 2
; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32 addrspace(4)* [[TMP5]], align 4, !range [[RNG1]], !invariant.load !0
; CHECK-NEXT: [[TMP7:%.*]] = lshr i32 [[TMP4]], 16
; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.amdgcn.workitem.id.x(), !range [[RNG2]]
; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.amdgcn.workitem.id.y(), !range [[RNG2]]
; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.workitem.id.z(), !range [[RNG2]]
; CHECK-NEXT: [[TMP11:%.*]] = mul nuw nsw i32 [[TMP7]], [[TMP6]]
; CHECK-NEXT: [[TMP12:%.*]] = mul i32 [[TMP11]], [[TMP8]]
; CHECK-NEXT: [[TMP13:%.*]] = mul nuw nsw i32 [[TMP9]], [[TMP6]]
; CHECK-NEXT: [[TMP14:%.*]] = add i32 [[TMP12]], [[TMP13]]
; CHECK-NEXT: [[TMP15:%.*]] = add i32 [[TMP14]], [[TMP10]]
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_lhs.alloca, i32 0, i32 [[TMP15]]
; CHECK-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[TMP16]], i32 0, i32 [[A:%.*]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 addrspace(3)* null, [[PTR0]]
; CHECK-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; CHECK-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
; NOLDS-LABEL: @lds_promoted_alloca_icmp_null_lhs(
; NOLDS-NEXT: [[ALLOCA:%.*]] = alloca [16 x i32], align 4
; NOLDS-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[A:%.*]]
; NOLDS-NEXT: [[CMP:%.*]] = icmp eq i32* null, [[PTR0]]
; NOLDS-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; NOLDS-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; NOLDS-NEXT: ret void
;
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%cmp = icmp eq i32* null, %ptr0
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
define amdgpu_kernel void @lds_promoted_alloca_icmp_unknown_ptr(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
; CHECK-LABEL: @lds_promoted_alloca_icmp_unknown_ptr(
; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [16 x i32], align 4
; CHECK-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[A:%.*]]
; CHECK-NEXT: [[PTR1:%.*]] = call i32* @get_unknown_pointer()
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32* [[PTR0]], [[PTR1]]
; CHECK-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; CHECK-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
; NOLDS-LABEL: @lds_promoted_alloca_icmp_unknown_ptr(
; NOLDS-NEXT: [[ALLOCA:%.*]] = alloca [16 x i32], align 4
; NOLDS-NEXT: [[PTR0:%.*]] = getelementptr inbounds [16 x i32], [16 x i32]* [[ALLOCA]], i32 0, i32 [[A:%.*]]
; NOLDS-NEXT: [[PTR1:%.*]] = call i32* @get_unknown_pointer()
; NOLDS-NEXT: [[CMP:%.*]] = icmp eq i32* [[PTR0]], [[PTR1]]
; NOLDS-NEXT: [[ZEXT:%.*]] = zext i1 [[CMP]] to i32
; NOLDS-NEXT: store volatile i32 [[ZEXT]], i32 addrspace(1)* [[OUT:%.*]], align 4
; NOLDS-NEXT: ret void
;
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%ptr1 = call i32* @get_unknown_pointer()
%cmp = icmp eq i32* %ptr0, %ptr1
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
declare i32* @get_unknown_pointer() #0
attributes #0 = { nounwind "amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="1,256" }
|
