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
|
// RUN: mlir-opt %s -test-transform-dialect-interpreter --split-input-file | FileCheck %s
// CHECK-LABEL: func.func @matmul_tensors_1(
func.func @matmul_tensors_1(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>,
%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// This operation is marked for tiling only.
// CHECK-COUNT-3: scf.for
// CHECK-COUNT-3: tensor.extract_slice
// CHECK: linalg.matmul
// CHECK-SAME: -> tensor<4x4xf32>
%0 = linalg.matmul { test.attrA }
ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
func.return %0 : tensor<128x128xf32>
}
func.func @matmul_tensors_2(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>,
%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// This operation is marked f
// This operation is marked for tiling and vectorization.
// CHECK-COUNT-3: scf.for
// CHECK-COUNT-3: vector.transfer_read
// CHECK: vector.contract
// CHECK-NOT: linalg.matmul
// CHECK: vector.transfer_write
%0 = linalg.matmul { test.attrA, test.attrC }
ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
func.return %0 : tensor<128x128xf32>
}
func.func @matmul_tensors_3(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>,
%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// This operation is marked for vectorization only.
// CHECK-NOT: scf.for
// CHECK-COUNT-3: vector.transfer_read
// CHECK: vector.contract
// CHECK-SAME: into vector<128x128xf32>
// CHECK: vector.transfer_write
%0 = linalg.matmul { test.attrC }
ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
func.return %0 : tensor<128x128xf32>
}
transform.with_pdl_patterns {
^bb0(%arg0: !transform.any_op):
// Match matmul operations inside @matmul_tensors with test.attrA set.
pdl.pattern @pdl_target_attrA : benefit(1) {
%args = operands
%results = types
%attr = attribute
%0 = operation "linalg.matmul"(%args : !pdl.range<value>) {"test.attrA" = %attr}-> (%results : !pdl.range<type>)
// TODO: we don't want this, but it is the required terminator for pdl.pattern
rewrite %0 with "transform.dialect"
}
// Match matmul operations inside @matmul_tensors with test.attrC set.
pdl.pattern @pdl_target_attrC : benefit(1) {
%args = operands
%results = types
%attr = attribute
%0 = operation "linalg.matmul"(%args : !pdl.range<value>) {"test.attrC" = %attr}-> (%results : !pdl.range<type>)
// TODO: we don't want this, but it is the required terminator for pdl.pattern
rewrite %0 with "transform.dialect"
}
transform.sequence %arg0 : !transform.any_op failures(propagate) {
^bb1(%arg1: !transform.any_op):
%0 = pdl_match @pdl_target_attrA in %arg1 : (!transform.any_op) -> !transform.any_op
transform.structured.tile %0 [4, 4, 4] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)
%1 = pdl_match @pdl_target_attrC in %arg1 : (!transform.any_op) -> !transform.any_op
%2 = transform.get_closest_isolated_parent %1 : (!transform.any_op) -> !transform.any_op
transform.structured.vectorize %2 : (!transform.any_op) -> !transform.any_op
}
}
// -----
// CHECK-LABEL: @vectorize_one
func.func @vectorize_one(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>,
%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// CHECK: vector.contract
%0 = linalg.matmul {test.attrA}
ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
func.return %0 : tensor<128x128xf32>
}
func.func @vectorize_none(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>,
%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// CHECK: linalg.matmul
%0 = linalg.matmul ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
func.return %0 : tensor<128x128xf32>
}
transform.with_pdl_patterns {
^bb0(%arg0: !transform.any_op):
pdl.pattern @pdl_target : benefit(1) {
%args = operands
%results = types
%attr = attribute
%0 = operation "linalg.matmul"(%args : !pdl.range<value>) {"test.attrA" = %attr}-> (%results : !pdl.range<type>)
// TODO: we don't want this, but it is the required terminator for pdl.pattern
rewrite %0 with "transform.dialect"
}
transform.sequence %arg0 : !transform.any_op failures(propagate) {
^bb1(%arg1: !transform.any_op):
%0 = pdl_match @pdl_target in %arg1 : (!transform.any_op) -> !transform.any_op
%1 = get_closest_isolated_parent %0 : (!transform.any_op) -> !transform.any_op
transform.structured.vectorize %1 : (!transform.any_op) -> !transform.any_op
}
}
// -----
// CHECK-LABEL: @vectorize_all
func.func @vectorize_all(
%arg0: tensor<128x128xf32>, %arg1: tensor<128x128xf32>, %arg2: tensor<128x128xf32>,
%arg3: tensor<128x128xf32>)
-> tensor<128x128xf32> {
// CHECK: vector.contract
%0 = linalg.matmul {test.attrA}
ins(%arg0, %arg1: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg2: tensor<128x128xf32>)
-> tensor<128x128xf32>
// CHECK: vector.contract
%1 = linalg.matmul ins(%arg0, %0: tensor<128x128xf32>, tensor<128x128xf32>)
outs(%arg3: tensor<128x128xf32>)
-> tensor<128x128xf32>
return %1 : tensor<128x128xf32>
}
transform.sequence failures(propagate) {
^bb0(%arg0: !transform.any_op):
transform.structured.vectorize %arg0 : (!transform.any_op) -> !transform.any_op
}
|
