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
|
// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
// RUN: mlir-opt %s --linalg-generalize-named-ops \
// RUN: --sparsification --sparse-tensor-codegen \
// RUN: --canonicalize --cse | FileCheck %s
#CSR = #sparse_tensor.encoding<{
lvlTypes = [ "dense", "compressed" ],
dimToLvl = affine_map<(i,j) -> (i,j)>
}>
//
// Computes C = A x B with all matrices sparse (SpMSpM) in CSR.
//
// CHECK-LABEL: func.func private @_insert_dense_compressed_4_4_f64_0_0(
// CHECK-SAME: %[[VAL_0:.*0]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_1:.*1]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_2:.*2]]: memref<?xf64>,
// CHECK-SAME: %[[VAL_3:.*3]]: !sparse_tensor.storage_specifier
// CHECK-SAME: %[[VAL_4:.*4]]: index,
// CHECK-SAME: %[[VAL_5:.*5]]: index,
// CHECK-SAME: %[[VAL_6:.*6]]: f64) -> (memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: %[[VAL_7:.*]] = arith.constant false
// CHECK: %[[VAL_8:.*]] = arith.constant 1 : index
// CHECK: %[[VAL_9:.*]] = arith.addi %[[VAL_4]], %[[VAL_8]] : index
// CHECK: %[[VAL_10:.*]] = memref.load %[[VAL_0]]{{\[}}%[[VAL_4]]] : memref<?xindex>
// CHECK: %[[VAL_11:.*]] = memref.load %[[VAL_0]]{{\[}}%[[VAL_9]]] : memref<?xindex>
// CHECK: %[[VAL_13:.*]] = sparse_tensor.storage_specifier.get %[[VAL_3]] crd_mem_sz at 1 : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_14:.*]] = arith.subi %[[VAL_11]], %[[VAL_8]] : index
// CHECK: %[[VAL_15:.*]] = arith.cmpi ult, %[[VAL_10]], %[[VAL_11]] : index
// CHECK: %[[VAL_16:.*]] = scf.if %[[VAL_15]] -> (i1) {
// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_1]]{{\[}}%[[VAL_14]]] : memref<?xindex>
// CHECK: %[[VAL_18:.*]] = arith.cmpi eq, %[[VAL_17]], %[[VAL_5]] : index
// CHECK: scf.yield %[[VAL_18]] : i1
// CHECK: } else {
// CHECK: memref.store %[[VAL_13]], %[[VAL_0]]{{\[}}%[[VAL_4]]] : memref<?xindex>
// CHECK: scf.yield %[[VAL_7]] : i1
// CHECK: }
// CHECK: %[[VAL_19:.*]]:2 = scf.if %[[VAL_20:.*]] -> (memref<?xindex>, !sparse_tensor.storage_specifier
// CHECK: scf.yield %[[VAL_1]], %[[VAL_3]] : memref<?xindex>, !sparse_tensor.storage_specifier
// CHECK: } else {
// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_13]], %[[VAL_8]] : index
// CHECK: memref.store %[[VAL_21]], %[[VAL_0]]{{\[}}%[[VAL_9]]] : memref<?xindex>
// CHECK: %[[VAL_22:.*]], %[[VAL_24:.*]] = sparse_tensor.push_back %[[VAL_13]], %[[VAL_1]], %[[VAL_5]] : index, memref<?xindex>, index
// CHECK: %[[VAL_25:.*]] = sparse_tensor.storage_specifier.set %[[VAL_3]] crd_mem_sz at 1 with %[[VAL_24]] : !sparse_tensor.storage_specifier
// CHECK: scf.yield %[[VAL_22]], %[[VAL_25]] : memref<?xindex>, !sparse_tensor.storage_specifier
// CHECK: }
// CHECK: %[[VAL_28:.*]] = sparse_tensor.storage_specifier.get %[[VAL_27:.*]]#1 val_mem_sz : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_29:.*]], %[[VAL_30:.*]] = sparse_tensor.push_back %[[VAL_28]], %[[VAL_2]], %[[VAL_6]] : index, memref<?xf64>, f64
// CHECK: %[[VAL_32:.*]] = sparse_tensor.storage_specifier.set %[[VAL_27]]#1 val_mem_sz with %[[VAL_30]] : !sparse_tensor.storage_specifier
// CHECK: return %[[VAL_0]], %[[VAL_27]]#0, %[[VAL_29]], %[[VAL_32]] : memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: }
// CHECK-LABEL: func.func @matmul(
// CHECK-SAME: %[[VAL_0:.*0]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_1:.*1]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_2:.*2]]: memref<?xf64>,
// CHECK-SAME: %[[VAL_3:.*3]]: !sparse_tensor.storage_specifier
// CHECK-SAME: %[[VAL_4:.*4]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_5:.*5]]: memref<?xindex>,
// CHECK-SAME: %[[VAL_6:.*6]]: memref<?xf64>,
// CHECK-SAME: %[[VAL_7:.*7]]: !sparse_tensor.storage_specifier
// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 4 : index
// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 0.000000e+00 : f64
// CHECK-DAG: %[[VAL_10:.*]] = arith.constant 0 : index
// CHECK-DAG: %[[VAL_11:.*]] = arith.constant 1 : index
// CHECK-DAG: %[[VAL_12:.*]] = arith.constant false
// CHECK-DAG: %[[VAL_13:.*]] = arith.constant true
// CHECK: %[[VAL_14:.*]] = memref.alloc() : memref<16xindex>
// CHECK: %[[VAL_15:.*]] = memref.cast %[[VAL_14]] : memref<16xindex> to memref<?xindex>
// CHECK: %[[VAL_16:.*]] = memref.alloc() : memref<16xindex>
// CHECK: %[[VAL_17:.*]] = memref.cast %[[VAL_16]] : memref<16xindex> to memref<?xindex>
// CHECK: %[[VAL_18:.*]] = memref.alloc() : memref<16xf64>
// CHECK: %[[VAL_19:.*]] = memref.cast %[[VAL_18]] : memref<16xf64> to memref<?xf64>
// CHECK: %[[VAL_20:.*]] = sparse_tensor.storage_specifier.init : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_21:.*]] = sparse_tensor.storage_specifier.set %[[VAL_20]] lvl_sz at 0 with %[[VAL_8]] : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_22:.*]] = sparse_tensor.storage_specifier.set %[[VAL_21]] lvl_sz at 1 with %[[VAL_8]] : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_23:.*]] = sparse_tensor.storage_specifier.get %[[VAL_22]] pos_mem_sz at 1 : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_24:.*]], %[[VAL_25:.*]] = sparse_tensor.push_back %[[VAL_23]], %[[VAL_15]], %[[VAL_10]] : index, memref<?xindex>, index
// CHECK: %[[VAL_26:.*]] = sparse_tensor.storage_specifier.set %[[VAL_22]] pos_mem_sz at 1 with %[[VAL_25]] : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_27:.*]], %[[VAL_28:.*]] = sparse_tensor.push_back %[[VAL_25]], %[[VAL_24]], %[[VAL_10]], %[[VAL_8]] : index, memref<?xindex>, index, index
// CHECK: %[[VAL_29:.*]] = sparse_tensor.storage_specifier.set %[[VAL_26]] pos_mem_sz at 1 with %[[VAL_28]] : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_30:.*]] = memref.alloc() : memref<4xf64>
// CHECK: %[[VAL_31:.*]] = m
// CHECK: %[[VAL_32:.*]] = memref.alloc() : memref<4xindex>
// CHECK: %[[VAL_33:.*]] = memref.cast %[[VAL_32]] : memref<4xindex> to memref<?xindex>
// CHECK: linalg.fill ins(%[[VAL_9]] : f64) outs(%[[VAL_30]] : memref<4xf64>)
// CHECK: linalg.fill ins(%[[VAL_12]] : i1) outs(%[[VAL_31]] : memref<4xi1>)
// CHECK: %[[VAL_34:.*]]:4 = scf.for %[[VAL_35:.*]] = %[[VAL_10]] to %[[VAL_8]] step %[[VAL_11]] iter_args(%[[VAL_36:.*]] = %[[VAL_27]], %[[VAL_37:.*]] = %[[VAL_17]], %[[VAL_38:.*]] = %[[VAL_19]], %[[VAL_39:.*]] = %[[VAL_29]]) -> (memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: %[[VAL_40:.*]] = memref.load %[[VAL_0]]{{\[}}%[[VAL_35]]] : memref<?xindex>
// CHECK: %[[VAL_41:.*]] = arith.addi %[[VAL_35]], %[[VAL_11]] : index
// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_0]]{{\[}}%[[VAL_41]]] : memref<?xindex>
// CHECK: %[[VAL_43:.*]] = scf.for %[[VAL_44:.*]] = %[[VAL_40]] to %[[VAL_42]] step %[[VAL_11]] iter_args(%[[VAL_45:.*]] = %[[VAL_10]]) -> (index) {
// CHECK: %[[VAL_46:.*]] = memref.load %[[VAL_1]]{{\[}}%[[VAL_44]]] : memref<?xindex>
// CHECK: %[[VAL_47:.*]] = memref.load %[[VAL_2]]{{\[}}%[[VAL_44]]] : memref<?xf64>
// CHECK: %[[VAL_48:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_46]]] : memref<?xindex>
// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_46]], %[[VAL_11]] : index
// CHECK: %[[VAL_50:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_49]]] : memref<?xindex>
// CHECK: %[[VAL_51:.*]] = scf.for %[[VAL_52:.*]] = %[[VAL_48]] to %[[VAL_50]] step %[[VAL_11]] iter_args(%[[VAL_53:.*]] = %[[VAL_45]]) -> (index) {
// CHECK: %[[VAL_54:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_52]]] : memref<?xindex>
// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_30]]{{\[}}%[[VAL_54]]] : memref<4xf64>
// CHECK: %[[VAL_56:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_52]]] : memref<?xf64>
// CHECK: %[[VAL_57:.*]] = arith.mulf %[[VAL_47]], %[[VAL_56]] : f64
// CHECK: %[[VAL_58:.*]] = arith.addf %[[VAL_55]], %[[VAL_57]] : f64
// CHECK: %[[VAL_59:.*]] = memref.load %[[VAL_31]]{{\[}}%[[VAL_54]]] : memref<4xi1>
// CHECK: %[[VAL_60:.*]] = arith.cmpi eq, %[[VAL_59]], %[[VAL_12]] : i1
// CHECK: %[[VAL_61:.*]] = scf.if %[[VAL_60]] -> (index) {
// CHECK: memref.store %[[VAL_13]], %[[VAL_31]]{{\[}}%[[VAL_54]]] : memref<4xi1>
// CHECK: memref.store %[[VAL_54]], %[[VAL_32]]{{\[}}%[[VAL_53]]] : memref<4xindex>
// CHECK: %[[VAL_62:.*]] = arith.addi %[[VAL_53]], %[[VAL_11]] : index
// CHECK: scf.yield %[[VAL_62]] : index
// CHECK: } else {
// CHECK: scf.yield %[[VAL_53]] : index
// CHECK: }
// CHECK: memref.store %[[VAL_58]], %[[VAL_30]]{{\[}}%[[VAL_54]]] : memref<4xf64>
// CHECK: scf.yield %[[VAL_63:.*]] : index
// CHECK: } {"Emitted from" = "linalg.generic"}
// CHECK: scf.yield %[[VAL_64:.*]] : index
// CHECK: } {"Emitted from" = "linalg.generic"}
// CHECK: sparse_tensor.sort hybrid_quick_sort %[[VAL_65:.*]], %[[VAL_33]] : memref<?xindex>
// CHECK: %[[VAL_66:.*]]:4 = scf.for %[[VAL_67:.*]] = %[[VAL_10]] to %[[VAL_65]] step %[[VAL_11]] iter_args(%[[VAL_68:.*]] = %[[VAL_36]], %[[VAL_69:.*]] = %[[VAL_37]], %[[VAL_70:.*]] = %[[VAL_38]], %[[VAL_71:.*]] = %[[VAL_39]]) -> (memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: %[[VAL_72:.*]] = memref.load %[[VAL_32]]{{\[}}%[[VAL_67]]] : memref<4xindex>
// CHECK: %[[VAL_73:.*]] = memref.load %[[VAL_30]]{{\[}}%[[VAL_72]]] : memref<4xf64>
// CHECK: %[[VAL_74:.*]]:4 = func.call @_insert_dense_compressed_4_4_f64_0_0(%[[VAL_68]], %[[VAL_69]], %[[VAL_70]], %[[VAL_71]], %[[VAL_35]], %[[VAL_72]], %[[VAL_73]]) : (memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifie
// CHECK: memref.store %[[VAL_9]], %[[VAL_30]]{{\[}}%[[VAL_72]]] : memref<4xf64>
// CHECK: memref.store %[[VAL_12]], %[[VAL_31]]{{\[}}%[[VAL_72]]] : memref<4xi1>
// CHECK: scf.yield %[[VAL_74]]#0, %[[VAL_74]]#1, %[[VAL_74]]#2, %[[VAL_74]]#3 : memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: }
// CHECK: scf.yield %[[VAL_75:.*]]#0, %[[VAL_75]]#1, %[[VAL_75]]#2, %[[VAL_75]]#3 : memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
// CHECK: } {"Emitted from" = "linalg.generic"}
// CHECK: memref.dealloc %[[VAL_30]] : memref<4xf64>
// CHECK: memref.dealloc %[[VAL_31]] : memref<4xi1>
// CHECK: memref.dealloc %[[VAL_32]] : memref<4xindex>
// CHECK: %[[VAL_76:.*]] = sparse_tensor.storage_specifier.get %[[VAL_77:.*]]#3 pos_mem_sz at 1 : !sparse_tensor.storage_specifier
// CHECK: %[[VAL_78:.*]] = memref.load %[[VAL_77]]#0{{\[}}%[[VAL_10]]] : memref<?xindex>
// CHECK: %[[VAL_79:.*]] = scf.for %[[VAL_80:.*]] = %[[VAL_11]] to %[[VAL_76]] step %[[VAL_11]] iter_args(%[[VAL_81:.*]] = %[[VAL_78]]) -> (index) {
// CHECK: %[[VAL_82:.*]] = memref.load %[[VAL_77]]#0{{\[}}%[[VAL_80]]] : memref<?xindex>
// CHECK: %[[VAL_83:.*]] = arith.cmpi eq, %[[VAL_82]], %[[VAL_10]] : index
// CHECK: %[[VAL_84:.*]] = arith.select %[[VAL_83]], %[[VAL_81]], %[[VAL_82]] : index
// CHECK: scf.if %[[VAL_83]] {
// CHECK: memref.store %[[VAL_81]], %[[VAL_77]]#0{{\[}}%[[VAL_80]]] : memref<?xindex>
// CHECK: }
// CHECK: scf.yield %[[VAL_84]] : index
// CHECK: }
// CHECK: return %[[VAL_77]]#0, %[[VAL_77]]#1, %[[VAL_77]]#2, %[[VAL_77]]#3 : memref<?xindex>, memref<?xindex>, memref<?xf64>, !sparse_tensor.storage_specifier
func.func @matmul(%A: tensor<4x8xf64, #CSR>,
%B: tensor<8x4xf64, #CSR>) -> tensor<4x4xf64, #CSR> {
%C = bufferization.alloc_tensor() : tensor<4x4xf64, #CSR>
%D = linalg.matmul
ins(%A, %B: tensor<4x8xf64, #CSR>, tensor<8x4xf64, #CSR>)
outs(%C: tensor<4x4xf64, #CSR>) -> tensor<4x4xf64, #CSR>
return %D: tensor<4x4xf64, #CSR>
}
|
