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
|
// RUN: mlir-opt %s --test-vector-chained-reduction-folding-patterns | FileCheck %s
// CHECK-LABEL: func.func @reduce_1x_fp32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xf32>) -> f32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0.0
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ARG0]], %[[CST]] : vector<8xf32> into f32
// CHECK-NEXT: return %[[RES]] : f32
func.func @reduce_1x_fp32(%arg0: vector<8xf32>) -> f32 {
%cst0 = arith.constant 0.0 : f32
%0 = vector.reduction <add>, %arg0, %cst0 : vector<8xf32> into f32
return %0 : f32
}
// CHECK-LABEL: func.func @reduce_2x_fp32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xf32>, %[[ARG1:.+]]: vector<8xf32>) -> f32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0.0
// CHECK-DAG: %[[ADD:.+]] = arith.addf %[[ARG0]], %[[ARG1]] : vector<8xf32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]], %[[CST]] : vector<8xf32> into f32
// CHECK-NEXT: return %[[RES]] : f32
func.func @reduce_2x_fp32(%arg0: vector<8xf32>, %arg1: vector<8xf32>) -> f32 {
%cst0 = arith.constant 0.0 : f32
%0 = vector.reduction <add>, %arg0, %cst0 : vector<8xf32> into f32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xf32> into f32
return %1 : f32
}
// CHECK-LABEL: func.func @reduce_2x_no_acc_fp32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xf32>, %[[ARG1:.+]]: vector<8xf32>) -> f32 {
// CHECK: %[[ADD:.+]] = arith.addf %[[ARG0]], %[[ARG1]] : vector<8xf32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]] : vector<8xf32> into f32
// CHECK-NEXT: return %[[RES]] : f32
func.func @reduce_2x_no_acc_fp32(%arg0: vector<8xf32>, %arg1: vector<8xf32>) -> f32 {
%0 = vector.reduction <add>, %arg0 : vector<8xf32> into f32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xf32> into f32
return %1 : f32
}
// CHECK-LABEL: func.func @reduce_2x_zero_add_fp32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xf32>, %[[ARG1:.+]]: vector<8xf32>) -> f32 {
// CHECK: %[[ADD:.+]] = arith.addf %[[ARG0]], %[[ARG1]] : vector<8xf32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]] : vector<8xf32> into f32
// CHECK-NEXT: return %[[RES]] : f32
func.func @reduce_2x_zero_add_fp32(%arg0: vector<8xf32>, %arg1: vector<8xf32>) -> f32 {
%cst0 = arith.constant dense<0.0> : vector<8xf32>
%x = arith.addf %arg0, %cst0 : vector<8xf32>
%0 = vector.reduction <add>, %x : vector<8xf32> into f32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xf32> into f32
return %1 : f32
}
// CHECK-LABEL: func.func @reduce_3x_fp32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xf32>, %[[ARG1:.+]]: vector<8xf32>,
// CHECK-SAME: %[[ARG2:.+]]: vector<8xf32>) -> f32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0.0
// CHECK-DAG: %[[ADD0:.+]] = arith.addf %[[ARG1]], %[[ARG2]] : vector<8xf32>
// CHECK-DAG: %[[ADD1:.+]] = arith.addf %[[ARG0]], %[[ADD0]] : vector<8xf32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD1]], %[[CST]] : vector<8xf32> into f32
// CHECK-NEXT: return %[[RES]] : f32
func.func @reduce_3x_fp32(%arg0: vector<8xf32>, %arg1: vector<8xf32>,
%arg2: vector<8xf32>) -> f32 {
%cst0 = arith.constant 0.0 : f32
%0 = vector.reduction <add>, %arg0, %cst0 : vector<8xf32> into f32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xf32> into f32
%2 = vector.reduction <add>, %arg2, %1 : vector<8xf32> into f32
return %2 : f32
}
// CHECK-LABEL: func.func @reduce_1x_i32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xi32>) -> i32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ARG0]], %[[CST]] : vector<8xi32> into i32
// CHECK-NEXT: return %[[RES]] : i32
func.func @reduce_1x_i32(%arg0: vector<8xi32>) -> i32 {
%cst0 = arith.constant 0 : i32
%0 = vector.reduction <add>, %arg0, %cst0 : vector<8xi32> into i32
return %0 : i32
}
// CHECK-LABEL: func.func @reduce_2x_i32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xi32>, %[[ARG1:.+]]: vector<8xi32>) -> i32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0
// CHECK-DAG: %[[ADD:.+]] = arith.addi %[[ARG0]], %[[ARG1]] : vector<8xi32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]], %[[CST]] : vector<8xi32> into i32
// CHECK-NEXT: return %[[RES]] : i32
func.func @reduce_2x_i32(%arg0: vector<8xi32>, %arg1: vector<8xi32>) -> i32 {
%cst0 = arith.constant 0 : i32
%0 = vector.reduction <add>, %arg0, %cst0 : vector<8xi32> into i32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xi32> into i32
return %1 : i32
}
// CHECK-LABEL: func.func @reduce_2x_no_acc_i32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xi32>, %[[ARG1:.+]]: vector<8xi32>) -> i32 {
// CHECK: %[[ADD:.+]] = arith.addi %[[ARG0]], %[[ARG1]] : vector<8xi32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]] : vector<8xi32> into i32
// CHECK-NEXT: return %[[RES]] : i32
func.func @reduce_2x_no_acc_i32(%arg0: vector<8xi32>, %arg1: vector<8xi32>) -> i32 {
%0 = vector.reduction <add>, %arg0 : vector<8xi32> into i32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xi32> into i32
return %1 : i32
}
// CHECK-LABEL: func.func @reduce_2x_zero_add_i32(
// CHECK-SAME: %[[ARG0:.+]]: vector<8xi32>, %[[ARG1:.+]]: vector<8xi32>) -> i32 {
// CHECK-DAG: %[[CST:.+]] = arith.constant 0
// CHECK-DAG: %[[ADD:.+]] = arith.addi %[[ARG0]], %[[ARG1]] : vector<8xi32>
// CHECK-NEXT: %[[RES:.+]] = vector.reduction <add>, %[[ADD]], %[[CST]] : vector<8xi32> into i32
// CHECK-NEXT: return %[[RES]] : i32
func.func @reduce_2x_zero_add_i32(%arg0: vector<8xi32>, %arg1: vector<8xi32>) -> i32 {
%cst0 = arith.constant 0 : i32
%cstV = arith.constant dense<0> : vector<8xi32>
%x = arith.addi %arg0, %cstV : vector<8xi32>
%0 = vector.reduction <add>, %x, %cst0 : vector<8xi32> into i32
%1 = vector.reduction <add>, %arg1, %0 : vector<8xi32> into i32
return %1 : i32
}
|
