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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S %s -passes=scalarize-masked-mem-intrin -mtriple=amdgcn-amd-amdhsa | FileCheck %s
; COM: Test that, unlike on CPU targets, the mask doesn't get bitcast to a scalar,
; COM: since, on GPUs, each i1 takes up at least one register and so they should
; COM: be treated separately.
define void @scalarize_v2i32(ptr %p, <2 x i1> %mask, <2 x i32> %data) {
; CHECK-LABEL: define void @scalarize_v2i32(
; CHECK-SAME: ptr [[P:%.*]], <2 x i1> [[MASK:%.*]], <2 x i32> [[DATA:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i1> [[MASK]], i64 0
; CHECK-NEXT: br i1 [[TMP1]], label %[[COND_STORE:.*]], label %[[ELSE:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[DATA]], i64 0
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[P]], i32 0
; CHECK-NEXT: store i32 [[TMP3]], ptr [[TMP4]], align 4
; CHECK-NEXT: br label %[[ELSE]]
; CHECK: [[ELSE]]:
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP5]], label %[[COND_STORE1:.*]], label %[[ELSE2:.*]]
; CHECK: [[COND_STORE1]]:
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[DATA]], i64 1
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[P]], i32 1
; CHECK-NEXT: store i32 [[TMP7]], ptr [[TMP8]], align 4
; CHECK-NEXT: br label %[[ELSE2]]
; CHECK: [[ELSE2]]:
; CHECK-NEXT: ret void
;
call void @llvm.masked.store.v2i32.p0(<2 x i32> %data, ptr %p, i32 128, <2 x i1> %mask)
ret void
}
define void @scalarize_v2i32_splat_mask(ptr %p, <2 x i32> %data, i1 %mask) {
; CHECK-LABEL: define void @scalarize_v2i32_splat_mask(
; CHECK-SAME: ptr [[P:%.*]], <2 x i32> [[DATA:%.*]], i1 [[MASK:%.*]]) {
; CHECK-NEXT: [[MASK_VEC:%.*]] = insertelement <2 x i1> poison, i1 [[MASK]], i32 0
; CHECK-NEXT: [[MASK_SPLAT:%.*]] = shufflevector <2 x i1> [[MASK_VEC]], <2 x i1> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[MASK_SPLAT_FIRST:%.*]] = extractelement <2 x i1> [[MASK_SPLAT]], i64 0
; CHECK-NEXT: br i1 [[MASK_SPLAT_FIRST]], label %[[COND_STORE:.*]], label %[[BB1:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: store <2 x i32> [[DATA]], ptr [[P]], align 8
; CHECK-NEXT: br label %[[BB1]]
; CHECK: [[BB1]]:
; CHECK-NEXT: ret void
;
%mask.vec = insertelement <2 x i1> poison, i1 %mask, i32 0
%mask.splat = shufflevector <2 x i1> %mask.vec, <2 x i1> poison, <2 x i32> zeroinitializer
call void @llvm.masked.store.v2i32.p0(<2 x i32> %data, ptr %p, i32 8, <2 x i1> %mask.splat)
ret void
}
define void @scalarize_v2f16(ptr %p, <2 x i1> %mask, <2 x half> %data) {
; CHECK-LABEL: define void @scalarize_v2f16(
; CHECK-SAME: ptr [[P:%.*]], <2 x i1> [[MASK:%.*]], <2 x half> [[DATA:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i1> [[MASK]], i64 0
; CHECK-NEXT: br i1 [[TMP1]], label %[[COND_STORE:.*]], label %[[ELSE:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x half> [[DATA]], i64 0
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds half, ptr [[P]], i32 0
; CHECK-NEXT: store half [[TMP3]], ptr [[TMP4]], align 2
; CHECK-NEXT: br label %[[ELSE]]
; CHECK: [[ELSE]]:
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP5]], label %[[COND_STORE1:.*]], label %[[ELSE2:.*]]
; CHECK: [[COND_STORE1]]:
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x half> [[DATA]], i64 1
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds half, ptr [[P]], i32 1
; CHECK-NEXT: store half [[TMP7]], ptr [[TMP8]], align 2
; CHECK-NEXT: br label %[[ELSE2]]
; CHECK: [[ELSE2]]:
; CHECK-NEXT: ret void
;
call void @llvm.masked.store.v2f16.p0(<2 x half> %data, ptr %p, i32 128, <2 x i1> %mask)
ret void
}
define void @scalarize_v2i32_p3(ptr addrspace(3) %p, <2 x i1> %mask, <2 x i32> %data) {
; CHECK-LABEL: define void @scalarize_v2i32_p3(
; CHECK-SAME: ptr addrspace(3) [[P:%.*]], <2 x i1> [[MASK:%.*]], <2 x i32> [[DATA:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i1> [[MASK]], i64 0
; CHECK-NEXT: br i1 [[TMP1]], label %[[COND_STORE:.*]], label %[[ELSE:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[DATA]], i64 0
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr addrspace(3) [[P]], i32 0
; CHECK-NEXT: store i32 [[TMP3]], ptr addrspace(3) [[TMP4]], align 4
; CHECK-NEXT: br label %[[ELSE]]
; CHECK: [[ELSE]]:
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP5]], label %[[COND_STORE1:.*]], label %[[ELSE2:.*]]
; CHECK: [[COND_STORE1]]:
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[DATA]], i64 1
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr addrspace(3) [[P]], i32 1
; CHECK-NEXT: store i32 [[TMP7]], ptr addrspace(3) [[TMP8]], align 4
; CHECK-NEXT: br label %[[ELSE2]]
; CHECK: [[ELSE2]]:
; CHECK-NEXT: ret void
;
call void @llvm.masked.store.v2i32.p3(<2 x i32> %data, ptr addrspace(3) %p, i32 128, <2 x i1> %mask)
ret void
}
define void @scalarize_v2i32_lane_mask(ptr %p, <2 x i32> %data) {
; CHECK-LABEL: define void @scalarize_v2i32_lane_mask(
; CHECK-SAME: ptr [[P:%.*]], <2 x i32> [[DATA:%.*]]) {
; CHECK-NEXT: [[ITEM_ID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT: [[MASK:%.*]] = icmp ult i32 [[ITEM_ID]], 16
; CHECK-NEXT: [[MASK_VEC:%.*]] = insertelement <2 x i1> poison, i1 [[MASK]], i32 0
; CHECK-NEXT: [[MASK_SPLAT:%.*]] = shufflevector <2 x i1> [[MASK_VEC]], <2 x i1> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[MASK_SPLAT_FIRST:%.*]] = extractelement <2 x i1> [[MASK_SPLAT]], i64 0
; CHECK-NEXT: br i1 [[MASK_SPLAT_FIRST]], label %[[COND_STORE:.*]], label %[[BB1:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: store <2 x i32> [[DATA]], ptr [[P]], align 8
; CHECK-NEXT: br label %[[BB1]]
; CHECK: [[BB1]]:
; CHECK-NEXT: ret void
;
%item.id = call i32 @llvm.amdgcn.workitem.id.x()
%mask = icmp ult i32 %item.id, 16
%mask.vec = insertelement <2 x i1> poison, i1 %mask, i32 0
%mask.splat = shufflevector <2 x i1> %mask.vec, <2 x i1> poison, <2 x i32> zeroinitializer
call void @llvm.masked.store.v2i32.p0(<2 x i32> %data, ptr %p, i32 8, <2 x i1> %mask.splat)
ret void
}
define void @scalarize_v2i32_group_mask(ptr %p, <2 x i32> %data) {
; CHECK-LABEL: define void @scalarize_v2i32_group_mask(
; CHECK-SAME: ptr [[P:%.*]], <2 x i32> [[DATA:%.*]]) {
; CHECK-NEXT: [[ITEM_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
; CHECK-NEXT: [[MASK:%.*]] = icmp ult i32 [[ITEM_ID]], 4
; CHECK-NEXT: [[MASK_VEC:%.*]] = insertelement <2 x i1> poison, i1 [[MASK]], i32 0
; CHECK-NEXT: [[MASK_SPLAT:%.*]] = shufflevector <2 x i1> [[MASK_VEC]], <2 x i1> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[MASK_SPLAT_FIRST:%.*]] = extractelement <2 x i1> [[MASK_SPLAT]], i64 0
; CHECK-NEXT: br i1 [[MASK_SPLAT_FIRST]], label %[[COND_STORE:.*]], label %[[BB1:.*]]
; CHECK: [[COND_STORE]]:
; CHECK-NEXT: store <2 x i32> [[DATA]], ptr [[P]], align 8
; CHECK-NEXT: br label %[[BB1]]
; CHECK: [[BB1]]:
; CHECK-NEXT: ret void
;
%group.id = call i32 @llvm.amdgcn.workgroup.id.x()
%mask = icmp ult i32 %group.id, 4
%mask.vec = insertelement <2 x i1> poison, i1 %mask, i32 0
%mask.splat = shufflevector <2 x i1> %mask.vec, <2 x i1> poison, <2 x i32> zeroinitializer
call void @llvm.masked.store.v2i32.p0(<2 x i32> %data, ptr %p, i32 8, <2 x i1> %mask.splat)
ret void
}
declare void @llvm.masked.store.v2i32.p0(<2 x i32>, ptr, i32, <2 x i1>)
declare void @llvm.masked.store.v2f16.p0(<2 x half>, ptr, i32, <2 x i1>)
declare void @llvm.masked.store.v2i32.p3(<2 x i32>, ptr addrspace(3), i32, <2 x i1>)
declare noundef i32 @llvm.amdgcn.workitem.id.x()
declare noundef i32 @llvm.amdgcn.workgroup.id.x()
|
