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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -instcombine -S | FileCheck %s
declare float @llvm.vector.reduce.fadd.f32.v4f32(float, <4 x float>)
declare float @llvm.vector.reduce.fadd.f32.v8f32(float, <8 x float>)
declare void @use_f32(float)
declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
declare i32 @llvm.vector.reduce.add.v8i32(<8 x i32>)
declare void @use_i32(i32)
define float @diff_of_sums_v4f32(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_v4f32(
; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc nsz <4 x float> [[V0:%.*]], [[V1:%.*]]
; CHECK-NEXT: [[TMP2:%.*]] = call reassoc nsz float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[TMP1]])
; CHECK-NEXT: [[R:%.*]] = fsub reassoc nsz float [[TMP2]], [[A1:%.*]]
; CHECK-NEXT: ret float [[R]]
;
%r0 = call float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
%r1 = call float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
%r = fsub reassoc nsz float %r0, %r1
ret float %r
}
; negative test - fsub must allow reassociation
define float @diff_of_sums_v4f32_fmf(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_v4f32_fmf(
; CHECK-NEXT: [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT: [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT: [[R:%.*]] = fsub nnan ninf nsz float [[R0]], [[R1]]
; CHECK-NEXT: ret float [[R]]
;
%r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
%r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
%r = fsub ninf nnan nsz float %r0, %r1
ret float %r
}
; negative test - extra uses could create extra instructions
define float @diff_of_sums_extra_use1(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_extra_use1(
; CHECK-NEXT: [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT: call void @use_f32(float [[R0]])
; CHECK-NEXT: [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT: [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT: ret float [[R]]
;
%r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
call void @use_f32(float %r0)
%r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
%r = fsub fast float %r0, %r1
ret float %r
}
; negative test - extra uses could create extra instructions
define float @diff_of_sums_extra_use2(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_extra_use2(
; CHECK-NEXT: [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT: [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
; CHECK-NEXT: call void @use_f32(float [[R1]])
; CHECK-NEXT: [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT: ret float [[R]]
;
%r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
%r1 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a1, <4 x float> %v1)
call void @use_f32(float %r1)
%r = fsub fast float %r0, %r1
ret float %r
}
; negative test - can't reassociate different vector types
define float @diff_of_sums_type_mismatch(float %a0, <4 x float> %v0, float %a1, <8 x float> %v1) {
; CHECK-LABEL: @diff_of_sums_type_mismatch(
; CHECK-NEXT: [[R0:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
; CHECK-NEXT: [[R1:%.*]] = call fast float @llvm.vector.reduce.fadd.v8f32(float [[A1:%.*]], <8 x float> [[V1:%.*]])
; CHECK-NEXT: [[R:%.*]] = fsub fast float [[R0]], [[R1]]
; CHECK-NEXT: ret float [[R]]
;
%r0 = call fast float @llvm.vector.reduce.fadd.f32.v4f32(float %a0, <4 x float> %v0)
%r1 = call fast float @llvm.vector.reduce.fadd.f32.v8f32(float %a1, <8 x float> %v1)
%r = fsub fast float %r0, %r1
ret float %r
}
define i32 @diff_of_sums_v4i32(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32(
; CHECK-NEXT: [[TMP1:%.*]] = sub <4 x i32> [[V0:%.*]], [[V1:%.*]]
; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP1]])
; CHECK-NEXT: ret i32 [[TMP2]]
;
%r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
%r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
%r = sub i32 %r0, %r1
ret i32 %r
}
; negative test - extra uses could create extra instructions
define i32 @diff_of_sums_v4i32_extra_use1(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32_extra_use1(
; CHECK-NEXT: [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT: call void @use_i32(i32 [[R0]])
; CHECK-NEXT: [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT: [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT: ret i32 [[R]]
;
%r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
call void @use_i32(i32 %r0)
%r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
%r = sub i32 %r0, %r1
ret i32 %r
}
; negative test - extra uses could create extra instructions
define i32 @diff_of_sums_v4i32_extra_use2(<4 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_v4i32_extra_use2(
; CHECK-NEXT: [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V0:%.*]])
; CHECK-NEXT: [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT: call void @use_i32(i32 [[R1]])
; CHECK-NEXT: [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT: ret i32 [[R]]
;
%r0 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v0)
%r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
call void @use_i32(i32 %r1)
%r = sub i32 %r0, %r1
ret i32 %r
}
; negative test - can't reassociate different vector types
define i32 @diff_of_sums_type_mismatch2(<8 x i32> %v0, <4 x i32> %v1) {
; CHECK-LABEL: @diff_of_sums_type_mismatch2(
; CHECK-NEXT: [[R0:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[V0:%.*]])
; CHECK-NEXT: [[R1:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[V1:%.*]])
; CHECK-NEXT: [[R:%.*]] = sub i32 [[R0]], [[R1]]
; CHECK-NEXT: ret i32 [[R]]
;
%r0 = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %v0)
%r1 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %v1)
%r = sub i32 %r0, %r1
ret i32 %r
}
|
