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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=arm64_32-apple-ios7.0 -mcpu=cyclone %s -o - | FileCheck %s
define <2 x double> @test_insert_elt(<2 x double> %vec, double %val) {
; CHECK-LABEL: test_insert_elt:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $d1 killed $d1 def $q1
; CHECK-NEXT: mov.d v0[0], v1[0]
; CHECK-NEXT: ret
%res = insertelement <2 x double> %vec, double %val, i32 0
ret <2 x double> %res
}
define void @test_split_16B(<4 x float> %val, ptr %addr) {
; CHECK-LABEL: test_split_16B:
; CHECK: ; %bb.0:
; CHECK-NEXT: str q0, [x0]
; CHECK-NEXT: ret
store <4 x float> %val, ptr %addr, align 8
ret void
}
define void @test_split_16B_splat(<4 x i32>, ptr %addr) {
; CHECK-LABEL: test_split_16B_splat:
; CHECK: ; %bb.0:
; CHECK-NEXT: movi.4s v0, #42
; CHECK-NEXT: str q0, [x0]
; CHECK-NEXT: ret
%vec.tmp0 = insertelement <4 x i32> undef, i32 42, i32 0
%vec.tmp1 = insertelement <4 x i32> %vec.tmp0, i32 42, i32 1
%vec.tmp2 = insertelement <4 x i32> %vec.tmp1, i32 42, i32 2
%vec = insertelement <4 x i32> %vec.tmp2, i32 42, i32 3
store <4 x i32> %vec, ptr %addr, align 8
ret void
}
%vec = type <2 x double>
declare {%vec, %vec} @llvm.aarch64.neon.ld2r.v2f64.p0(ptr)
define {%vec, %vec} @test_neon_load(ptr %addr) {
; CHECK-LABEL: test_neon_load:
; CHECK: ; %bb.0:
; CHECK-NEXT: ld2r.2d { v0, v1 }, [x0]
; CHECK-NEXT: ret
%res = call {%vec, %vec} @llvm.aarch64.neon.ld2r.v2f64.p0(ptr %addr)
ret {%vec, %vec} %res
}
declare {%vec, %vec} @llvm.aarch64.neon.ld2lane.v2f64.p0(%vec, %vec, i64, ptr)
define {%vec, %vec} @test_neon_load_lane(ptr %addr, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_load_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ld2.d { v0, v1 }[0], [x0]
; CHECK-NEXT: ret
%res = call {%vec, %vec} @llvm.aarch64.neon.ld2lane.v2f64.p0(%vec %in1, %vec %in2, i64 0, ptr %addr)
ret {%vec, %vec} %res
}
declare void @llvm.aarch64.neon.st2.v2f64.p0(%vec, %vec, ptr)
define void @test_neon_store(ptr %addr, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_store:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: st2.2d { v0, v1 }, [x0]
; CHECK-NEXT: ret
call void @llvm.aarch64.neon.st2.v2f64.p0(%vec %in1, %vec %in2, ptr %addr)
ret void
}
declare void @llvm.aarch64.neon.st2lane.v2f64.p0(%vec, %vec, i64, ptr)
define void @test_neon_store_lane(ptr %addr, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_store_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: st2.d { v0, v1 }[1], [x0]
; CHECK-NEXT: ret
call void @llvm.aarch64.neon.st2lane.v2f64.p0(%vec %in1, %vec %in2, i64 1, ptr %addr)
ret void
}
declare {%vec, %vec} @llvm.aarch64.neon.ld2.v2f64.p0(ptr)
define {{%vec, %vec}, ptr} @test_neon_load_post(ptr %addr, i32 %offset) {
; CHECK-LABEL: test_neon_load_post:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: sxtw x8, w1
; CHECK-NEXT: ld2.2d { v0, v1 }, [x0], x8
; CHECK-NEXT: ret
%vecs = call {%vec, %vec} @llvm.aarch64.neon.ld2.v2f64.p0(ptr %addr)
%addr.new = getelementptr inbounds i8, ptr %addr, i32 %offset
%res.tmp = insertvalue {{%vec, %vec}, ptr} undef, {%vec, %vec} %vecs, 0
%res = insertvalue {{%vec, %vec}, ptr} %res.tmp, ptr %addr.new, 1
ret {{%vec, %vec}, ptr} %res
}
define {{%vec, %vec}, ptr} @test_neon_load_post_lane(ptr %addr, i32 %offset, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_load_post_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: sxtw x8, w1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ld2.d { v0, v1 }[1], [x0], x8
; CHECK-NEXT: ret
%vecs = call {%vec, %vec} @llvm.aarch64.neon.ld2lane.v2f64.p0(%vec %in1, %vec %in2, i64 1, ptr %addr)
%addr.new = getelementptr inbounds i8, ptr %addr, i32 %offset
%res.tmp = insertvalue {{%vec, %vec}, ptr} undef, {%vec, %vec} %vecs, 0
%res = insertvalue {{%vec, %vec}, ptr} %res.tmp, ptr %addr.new, 1
ret {{%vec, %vec}, ptr} %res
}
define ptr @test_neon_store_post(ptr %addr, i32 %offset, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_store_post:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: sxtw x8, w1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: st2.2d { v0, v1 }, [x0], x8
; CHECK-NEXT: ret
call void @llvm.aarch64.neon.st2.v2f64.p0(%vec %in1, %vec %in2, ptr %addr)
%addr.new = getelementptr inbounds i8, ptr %addr, i32 %offset
ret ptr %addr.new
}
define ptr @test_neon_store_post_lane(ptr %addr, i32 %offset, %vec %in1, %vec %in2) {
; CHECK-LABEL: test_neon_store_post_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
; CHECK-NEXT: ; kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: sxtw x8, w1
; CHECK-NEXT: ; kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
; CHECK-NEXT: st2.d { v0, v1 }[0], [x0], x8
; CHECK-NEXT: ret
call void @llvm.aarch64.neon.st2lane.v2f64.p0(%vec %in1, %vec %in2, i64 0, ptr %addr)
%addr.new = getelementptr inbounds i8, ptr %addr, i32 %offset
ret ptr %addr.new
}
; ld1 is slightly different because it goes via ISelLowering of normal IR ops
; rather than an intrinsic.
define {%vec, ptr} @test_neon_ld1_post_lane(ptr %addr, i32 %offset, %vec %in) {
; CHECK-LABEL: test_neon_ld1_post_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ; kill: def $w1 killed $w1 def $x1
; CHECK-NEXT: sbfiz x8, x1, #3, #32
; CHECK-NEXT: ld1.d { v0 }[0], [x0], x8
; CHECK-NEXT: ret
%loaded = load double, ptr %addr, align 8
%newvec = insertelement %vec %in, double %loaded, i32 0
%addr.new = getelementptr inbounds double, ptr %addr, i32 %offset
%res.tmp = insertvalue {%vec, ptr} undef, %vec %newvec, 0
%res = insertvalue {%vec, ptr} %res.tmp, ptr %addr.new, 1
ret {%vec, ptr} %res
}
define {{%vec, %vec}, ptr} @test_neon_load_post_exact(ptr %addr) {
; CHECK-LABEL: test_neon_load_post_exact:
; CHECK: ; %bb.0:
; CHECK-NEXT: ld2.2d { v0, v1 }, [x0], #32
; CHECK-NEXT: ret
%vecs = call {%vec, %vec} @llvm.aarch64.neon.ld2.v2f64.p0(ptr %addr)
%addr.new = getelementptr inbounds i8, ptr %addr, i32 32
%res.tmp = insertvalue {{%vec, %vec}, ptr} undef, {%vec, %vec} %vecs, 0
%res = insertvalue {{%vec, %vec}, ptr} %res.tmp, ptr %addr.new, 1
ret {{%vec, %vec}, ptr} %res
}
define {%vec, ptr} @test_neon_ld1_post_lane_exact(ptr %addr, %vec %in) {
; CHECK-LABEL: test_neon_ld1_post_lane_exact:
; CHECK: ; %bb.0:
; CHECK-NEXT: ld1.d { v0 }[0], [x0], #8
; CHECK-NEXT: ret
%loaded = load double, ptr %addr, align 8
%newvec = insertelement %vec %in, double %loaded, i32 0
%addr.new = getelementptr inbounds double, ptr %addr, i32 1
%res.tmp = insertvalue {%vec, ptr} undef, %vec %newvec, 0
%res = insertvalue {%vec, ptr} %res.tmp, ptr %addr.new, 1
ret {%vec, ptr} %res
}
; As in the general load/store case, this GEP has defined semantics when the
; address wraps. We cannot use post-indexed addressing.
define {%vec, ptr} @test_neon_ld1_notpost_lane_exact(ptr %addr, %vec %in) {
; CHECK-LABEL: test_neon_ld1_notpost_lane_exact:
; CHECK: ; %bb.0:
; CHECK-NEXT: ld1.d { v0 }[0], [x0]
; CHECK-NEXT: add w0, w0, #8
; CHECK-NEXT: ret
%loaded = load double, ptr %addr, align 8
%newvec = insertelement %vec %in, double %loaded, i32 0
%addr.new = getelementptr double, ptr %addr, i32 1
%res.tmp = insertvalue {%vec, ptr} undef, %vec %newvec, 0
%res = insertvalue {%vec, ptr} %res.tmp, ptr %addr.new, 1
ret {%vec, ptr} %res
}
define {%vec, ptr} @test_neon_ld1_notpost_lane(ptr %addr, i32 %offset, %vec %in) {
; CHECK-LABEL: test_neon_ld1_notpost_lane:
; CHECK: ; %bb.0:
; CHECK-NEXT: ld1.d { v0 }[0], [x0]
; CHECK-NEXT: add w0, w0, w1, lsl #3
; CHECK-NEXT: ret
%loaded = load double, ptr %addr, align 8
%newvec = insertelement %vec %in, double %loaded, i32 0
%addr.new = getelementptr double, ptr %addr, i32 %offset
%res.tmp = insertvalue {%vec, ptr} undef, %vec %newvec, 0
%res = insertvalue {%vec, ptr} %res.tmp, ptr %addr.new, 1
ret {%vec, ptr} %res
}
|
