arm-bf16-dotprod-intrinsics.ll
7.49 KB
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
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple armv8.6a-arm-none-eabi -mattr=+neon,+bf16 -float-abi=hard -verify-machineinstrs < %s -o - | FileCheck %s
define arm_aapcs_vfpcc <2 x float> @test_vbfdot_f32(<2 x float> %r, <4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vbfdot_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdot.bf16 d0, d1, d2
; CHECK-NEXT: bx lr
entry:
%vbfdot3.i = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> %r, <4 x bfloat> %a, <4 x bfloat> %b) #3
ret <2 x float> %vbfdot3.i
}
define <4 x float> @test_vbfdotq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfdotq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdot.bf16 q0, q1, q2
; CHECK-NEXT: bx lr
entry:
%vbfdot3.i = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) #3
ret <4 x float> %vbfdot3.i
}
define <2 x float> @test_vbfdot_lane_f32(<2 x float> %r, <4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vbfdot_lane_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdot.bf16 d0, d1, d2[0]
; CHECK-NEXT: bx lr
entry:
%.cast = bitcast <4 x bfloat> %b to <2 x float>
%lane = shufflevector <2 x float> %.cast, <2 x float> undef, <2 x i32> zeroinitializer
%.cast1 = bitcast <2 x float> %lane to <4 x bfloat>
%vbfdot3.i = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> %r, <4 x bfloat> %a, <4 x bfloat> %.cast1) #3
ret <2 x float> %vbfdot3.i
}
define <4 x float> @test_vbfdotq_laneq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfdotq_laneq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdup.32 q8, d5[1]
; CHECK-NEXT: vdot.bf16 q0, q1, q8
; CHECK-NEXT: bx lr
entry:
%.cast = bitcast <8 x bfloat> %b to <4 x float>
%lane = shufflevector <4 x float> %.cast, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
%.cast1 = bitcast <4 x float> %lane to <8 x bfloat>
%vbfdot3.i = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %.cast1) #3
ret <4 x float> %vbfdot3.i
}
define <2 x float> @test_vbfdot_laneq_f32(<2 x float> %r, <4 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfdot_laneq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdot.bf16 d0, d1, d3[1]
; CHECK-NEXT: bx lr
entry:
%.cast = bitcast <8 x bfloat> %b to <4 x float>
%lane = shufflevector <4 x float> %.cast, <4 x float> undef, <2 x i32> <i32 3, i32 3>
%.cast1 = bitcast <2 x float> %lane to <4 x bfloat>
%vbfdot3.i = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> %r, <4 x bfloat> %a, <4 x bfloat> %.cast1) #3
ret <2 x float> %vbfdot3.i
}
define <4 x float> @test_vbfdotq_lane_f32(<4 x float> %r, <8 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vbfdotq_lane_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: @ kill: def $d4 killed $d4 def $q2
; CHECK-NEXT: vdot.bf16 q0, q1, d4[0]
; CHECK-NEXT: bx lr
entry:
%.cast = bitcast <4 x bfloat> %b to <2 x float>
%lane = shufflevector <2 x float> %.cast, <2 x float> undef, <4 x i32> zeroinitializer
%.cast1 = bitcast <4 x float> %lane to <8 x bfloat>
%vbfdot3.i = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %.cast1) #3
ret <4 x float> %vbfdot3.i
}
define <4 x float> @test_vbfmmlaq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmmlaq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vmmla.bf16 q0, q1, q2
; CHECK-NEXT: bx lr
entry:
%vbfmmlaq_v3.i = call <4 x float> @llvm.arm.neon.bfmmla(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b)
ret <4 x float> %vbfmmlaq_v3.i
}
define <4 x float> @test_vbfmlalbq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlalbq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vfmab.bf16 q0, q1, q2
; CHECK-NEXT: bx lr
entry:
%vbfmlalbq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b)
ret <4 x float> %vbfmlalbq_v3.i
}
define <4 x float> @test_vbfmlaltq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlaltq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vfmat.bf16 q0, q1, q2
; CHECK-NEXT: bx lr
entry:
%vbfmlaltq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b)
ret <4 x float> %vbfmlaltq_v3.i
}
define <4 x float> @test_vbfmlalbq_lane_f32(<4 x float> %r, <8 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlalbq_lane_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: @ kill: def $d4 killed $d4 def $q2
; CHECK-NEXT: vfmab.bf16 q0, q1, d4[0]
; CHECK-NEXT: bx lr
entry:
%vecinit35 = shufflevector <4 x bfloat> %b, <4 x bfloat> undef, <8 x i32> zeroinitializer
%vbfmlalbq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %vecinit35)
ret <4 x float> %vbfmlalbq_v3.i
}
define <4 x float> @test_vbfmlalbq_laneq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlalbq_laneq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vfmab.bf16 q0, q1, d4[3]
; CHECK-NEXT: bx lr
entry:
%vecinit35 = shufflevector <8 x bfloat> %b, <8 x bfloat> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
%vbfmlalbq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %vecinit35)
ret <4 x float> %vbfmlalbq_v3.i
}
define <4 x float> @test_vbfmlaltq_lane_f32(<4 x float> %r, <8 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlaltq_lane_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: @ kill: def $d4 killed $d4 def $q2
; CHECK-NEXT: vfmat.bf16 q0, q1, d4[0]
; CHECK-NEXT: bx lr
entry:
%vecinit35 = shufflevector <4 x bfloat> %b, <4 x bfloat> undef, <8 x i32> zeroinitializer
%vbfmlaltq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %vecinit35)
ret <4 x float> %vbfmlaltq_v3.i
}
define <4 x float> @test_vbfmlaltq_laneq_f32(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlaltq_laneq_f32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vfmat.bf16 q0, q1, d4[3]
; CHECK-NEXT: bx lr
entry:
%vecinit35 = shufflevector <8 x bfloat> %b, <8 x bfloat> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
%vbfmlaltq_v3.i = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %vecinit35)
ret <4 x float> %vbfmlaltq_v3.i
}
define <4 x float> @test_vbfmlaltq_laneq_f32_v2(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vbfmlaltq_laneq_f32_v2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vdup.16 q8, d5[2]
; CHECK-NEXT: vfmat.bf16 q0, q1, q8
; CHECK-NEXT: bx lr
entry:
%vecinit35 = shufflevector <8 x bfloat> %b, <8 x bfloat> undef, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>
%vbfmlalt1.i = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> %r, <8 x bfloat> %a, <8 x bfloat> %vecinit35)
ret <4 x float> %vbfmlalt1.i
}
declare <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float>, <4 x bfloat>, <4 x bfloat>)
declare <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float>, <8 x bfloat>, <8 x bfloat>)
declare <4 x float> @llvm.arm.neon.bfmmla(<4 x float>, <8 x bfloat>, <8 x bfloat>)
declare <4 x float> @llvm.arm.neon.bfmlalb(<4 x float>, <8 x bfloat>, <8 x bfloat>)
declare <4 x float> @llvm.arm.neon.bfmlalt(<4 x float>, <8 x bfloat>, <8 x bfloat>)