1// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
2// RUN: mlir-opt %s \
3// RUN: --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
4// RUN: --sparsification | FileCheck %s
5
6#DCSR = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed" ] }>
7
8// CHECK-LABEL:   func @matmul1(
9// CHECK-SAME:      %[[VAL_0:.*]]: tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>,
10// CHECK-SAME:      %[[VAL_1:.*]]: tensor<20x30xf32>,
11// CHECK-SAME:      %[[VAL_2:.*]]: tensor<10x30xf32>) -> tensor<10x30xf32> {
12// CHECK-DAG:       %[[VAL_3:.*]] = arith.constant 0 : index
13// CHECK-DAG:       %[[VAL_4:.*]] = arith.constant 1 : index
14// CHECK-DAG:       %[[VAL_5:.*]] = arith.constant 30 : index
15// CHECK:           %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
16// CHECK:           %[[VAL_7:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
17// CHECK:           %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
18// CHECK:           %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
19// CHECK:           %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
20// CHECK:           %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_1]] : memref<20x30xf32>
21// CHECK:           %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : memref<10x30xf32>
22// CHECK:           %[[VAL_13:.*]] = memref.alloc() : memref<10x30xf32>
23// CHECK:           memref.copy %[[VAL_12]], %[[VAL_13]] : memref<10x30xf32> to memref<10x30xf32>
24// CHECK:           %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_3]]] : memref<?xindex>
25// CHECK:           %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex>
26// CHECK:           scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_4]] {
27// CHECK:             %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xindex>
28// CHECK:             %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex>
29// CHECK:             %[[VAL_19:.*]] = arith.addi %[[VAL_16]], %[[VAL_4]] : index
30// CHECK:             %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex>
31// CHECK:             scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_4]] {
32// CHECK:               %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex>
33// CHECK:               %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xf32>
34// CHECK:               scf.for %[[VAL_24:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
35// CHECK:                 %[[VAL_25:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_24]]] : memref<10x30xf32>
36// CHECK:                 %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_22]], %[[VAL_24]]] : memref<20x30xf32>
37// CHECK:                 %[[VAL_27:.*]] = arith.mulf %[[VAL_23]], %[[VAL_26]] : f32
38// CHECK:                 %[[VAL_28:.*]] = arith.addf %[[VAL_25]], %[[VAL_27]] : f32
39// CHECK:                 memref.store %[[VAL_28]], %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_24]]] : memref<10x30xf32>
40// CHECK:               }
41// CHECK:             }
42// CHECK:           }
43// CHECK:           %[[VAL_29:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<10x30xf32>
44// CHECK:           return %[[VAL_29]] : tensor<10x30xf32>
45// CHECK:         }
46func @matmul1(%a: tensor<10x20xf32, #DCSR>,
47              %b: tensor<20x30xf32>,
48              %c: tensor<10x30xf32>) -> tensor<10x30xf32> {
49  %0 = linalg.matmul
50    ins(%a, %b: tensor<10x20xf32, #DCSR>, tensor<20x30xf32>)
51    outs(%c: tensor<10x30xf32>) -> tensor<10x30xf32>
52  return %0 : tensor<10x30xf32>
53}
54
55//
56// Computes C = A x B with all matrices sparse (SpMSpM) in DCSR.
57//
58// CHECK-LABEL:   func @matmul2(
59// CHECK-SAME:      %[[VAL_0:.*]]: tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>>,
60// CHECK-SAME:      %[[VAL_1:.*]]: tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> {
61// CHECK-DAG:       %[[VAL_2:.*]] = arith.constant 4 : index
62// CHECK-DAG:       %[[VAL_3:.*]] = arith.constant 0 : index
63// CHECK-DAG:       %[[VAL_4:.*]] = arith.constant 1 : index
64// CHECK-DAG:       %[[VAL_5:.*]] = arith.constant 2 : index
65// CHECK-DAG:       %[[VAL_6:.*]] = arith.constant false
66// CHECK-DAG:       %[[VAL_7:.*]] = arith.constant true
67// CHECK:           %[[VAL_8:.*]] = sparse_tensor.init{{\[}}%[[VAL_2]], %[[VAL_2]]] : tensor<4x4xf64, #sparse_tensor.encoding<{{{.*}}}>>
68// CHECK:           %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
69// CHECK:           %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
70// CHECK:           %[[VAL_11:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
71// CHECK:           %[[VAL_12:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
72// CHECK:           %[[VAL_13:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<4x8xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf64>
73// CHECK:           %[[VAL_14:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_3]] : tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
74// CHECK:           %[[VAL_15:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_3]] : tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
75// CHECK:           %[[VAL_16:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
76// CHECK:           %[[VAL_17:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex>
77// CHECK:           %[[VAL_18:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<8x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf64>
78// CHECK:           %[[VAL_19:.*]] = memref.alloca(%[[VAL_5]]) : memref<?xindex>
79// CHECK:           %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_3]]] : memref<?xindex>
80// CHECK:           %[[VAL_21:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_4]]] : memref<?xindex>
81// CHECK:           scf.for %[[VAL_22:.*]] = %[[VAL_20]] to %[[VAL_21]] step %[[VAL_4]] {
82// CHECK:             %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_22]]] : memref<?xindex>
83// CHECK:             memref.store %[[VAL_23]], %[[VAL_19]]{{\[}}%[[VAL_3]]] : memref<?xindex>
84// CHECK:             %[[VAL_24:.*]], %[[VAL_25:.*]], %[[VAL_26:.*]], %[[VAL_27:.*]] = sparse_tensor.expand %[[VAL_8]] : tensor<4x4xf64, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf64>, memref<?xi1>, memref<?xindex>, index
85// CHECK:             %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_22]]] : memref<?xindex>
86// CHECK:             %[[VAL_29:.*]] = arith.addi %[[VAL_22]], %[[VAL_4]] : index
87// CHECK:             %[[VAL_30:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_29]]] : memref<?xindex>
88// CHECK:             %[[VAL_31:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_3]]] : memref<?xindex>
89// CHECK:             %[[VAL_32:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_4]]] : memref<?xindex>
90// CHECK:             %[[VAL_33:.*]]:3 = scf.while (%[[VAL_34:.*]] = %[[VAL_28]], %[[VAL_35:.*]] = %[[VAL_31]], %[[VAL_36:.*]] = %[[VAL_27]]) : (index, index, index) -> (index, index, index) {
91// CHECK:               %[[VAL_37:.*]] = arith.cmpi ult, %[[VAL_34]], %[[VAL_30]] : index
92// CHECK:               %[[VAL_38:.*]] = arith.cmpi ult, %[[VAL_35]], %[[VAL_32]] : index
93// CHECK:               %[[VAL_39:.*]] = arith.andi %[[VAL_37]], %[[VAL_38]] : i1
94// CHECK:               scf.condition(%[[VAL_39]]) %[[VAL_34]], %[[VAL_35]], %[[VAL_36]] : index, index, index
95// CHECK:             } do {
96// CHECK:             ^bb0(%[[VAL_40:.*]]: index, %[[VAL_41:.*]]: index, %[[VAL_42:.*]]: index):
97// CHECK:               %[[VAL_43:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_40]]] : memref<?xindex>
98// CHECK:               %[[VAL_44:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_41]]] : memref<?xindex>
99// CHECK:               %[[VAL_45:.*]] = arith.cmpi ult, %[[VAL_44]], %[[VAL_43]] : index
100// CHECK:               %[[VAL_46:.*]] = select %[[VAL_45]], %[[VAL_44]], %[[VAL_43]] : index
101// CHECK:               %[[VAL_47:.*]] = arith.cmpi eq, %[[VAL_43]], %[[VAL_46]] : index
102// CHECK:               %[[VAL_48:.*]] = arith.cmpi eq, %[[VAL_44]], %[[VAL_46]] : index
103// CHECK:               %[[VAL_49:.*]] = arith.andi %[[VAL_47]], %[[VAL_48]] : i1
104// CHECK:               %[[VAL_50:.*]] = scf.if %[[VAL_49]] -> (index) {
105// CHECK:                 %[[VAL_51:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_40]]] : memref<?xf64>
106// CHECK:                 %[[VAL_52:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_41]]] : memref<?xindex>
107// CHECK:                 %[[VAL_53:.*]] = arith.addi %[[VAL_41]], %[[VAL_4]] : index
108// CHECK:                 %[[VAL_54:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_53]]] : memref<?xindex>
109// CHECK:                 %[[VAL_55:.*]] = scf.for %[[VAL_56:.*]] = %[[VAL_52]] to %[[VAL_54]] step %[[VAL_4]] iter_args(%[[VAL_57:.*]] = %[[VAL_42]]) -> (index) {
110// CHECK:                   %[[VAL_58:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_56]]] : memref<?xindex>
111// CHECK:                   %[[VAL_59:.*]] = memref.load %[[VAL_24]]{{\[}}%[[VAL_58]]] : memref<?xf64>
112// CHECK:                   %[[VAL_60:.*]] = memref.load %[[VAL_18]]{{\[}}%[[VAL_56]]] : memref<?xf64>
113// CHECK:                   %[[VAL_61:.*]] = arith.mulf %[[VAL_51]], %[[VAL_60]] : f64
114// CHECK:                   %[[VAL_62:.*]] = arith.addf %[[VAL_59]], %[[VAL_61]] : f64
115// CHECK:                   %[[VAL_63:.*]] = memref.load %[[VAL_25]]{{\[}}%[[VAL_58]]] : memref<?xi1>
116// CHECK:                   %[[VAL_64:.*]] = arith.cmpi eq, %[[VAL_63]], %[[VAL_6]] : i1
117// CHECK:                   %[[VAL_65:.*]] = scf.if %[[VAL_64]] -> (index) {
118// CHECK:                     memref.store %[[VAL_7]], %[[VAL_25]]{{\[}}%[[VAL_58]]] : memref<?xi1>
119// CHECK:                     memref.store %[[VAL_58]], %[[VAL_26]]{{\[}}%[[VAL_57]]] : memref<?xindex>
120// CHECK:                     %[[VAL_66:.*]] = arith.addi %[[VAL_57]], %[[VAL_4]] : index
121// CHECK:                     scf.yield %[[VAL_66]] : index
122// CHECK:                   } else {
123// CHECK:                     scf.yield %[[VAL_57]] : index
124// CHECK:                   }
125// CHECK:                   memref.store %[[VAL_62]], %[[VAL_24]]{{\[}}%[[VAL_58]]] : memref<?xf64>
126// CHECK:                   scf.yield %[[VAL_67:.*]] : index
127// CHECK:                 }
128// CHECK:                 scf.yield %[[VAL_68:.*]] : index
129// CHECK:               } else {
130// CHECK:                 scf.yield %[[VAL_42]] : index
131// CHECK:               }
132// CHECK:               %[[VAL_69:.*]] = arith.cmpi eq, %[[VAL_43]], %[[VAL_46]] : index
133// CHECK:               %[[VAL_70:.*]] = arith.addi %[[VAL_40]], %[[VAL_4]] : index
134// CHECK:               %[[VAL_71:.*]] = select %[[VAL_69]], %[[VAL_70]], %[[VAL_40]] : index
135// CHECK:               %[[VAL_72:.*]] = arith.cmpi eq, %[[VAL_44]], %[[VAL_46]] : index
136// CHECK:               %[[VAL_73:.*]] = arith.addi %[[VAL_41]], %[[VAL_4]] : index
137// CHECK:               %[[VAL_74:.*]] = select %[[VAL_72]], %[[VAL_73]], %[[VAL_41]] : index
138// CHECK:               scf.yield %[[VAL_71]], %[[VAL_74]], %[[VAL_75:.*]] : index, index, index
139// CHECK:             }
140// CHECK:             sparse_tensor.compress %[[VAL_8]], %[[VAL_19]], %[[VAL_24]], %[[VAL_25]], %[[VAL_26]], %[[VAL_76:.*]]#2 : tensor<4x4xf64, #sparse_tensor.encoding<{{{.*}}}>>, memref<?xindex>, memref<?xf64>, memref<?xi1>, memref<?xindex>, index
141// CHECK:           }
142// CHECK:           %[[VAL_77:.*]] = sparse_tensor.load %[[VAL_8]] hasInserts : tensor<4x4xf64, #sparse_tensor.encoding<{{{.*}}}>>
143// CHECK:           return %[[VAL_77]] : tensor<4x4xf64, #sparse_tensor.encoding<{{{.*}}}>>
144// CHECK:         }
145func @matmul2(%A: tensor<4x8xf64, #DCSR>,
146              %B: tensor<8x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
147  %c4 = arith.constant 4 : index
148  %C = sparse_tensor.init [%c4, %c4] : tensor<4x4xf64, #DCSR>
149  %D = linalg.matmul
150    ins(%A, %B: tensor<4x8xf64, #DCSR>, tensor<8x4xf64, #DCSR>)
151       outs(%C: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR>
152  return %D: tensor<4x4xf64, #DCSR>
153}
154
155// CHECK-LABEL:   func @conv2d(
156// CHECK-SAME:      %[[VAL_0:.*]]: tensor<8x8xi32>,
157// CHECK-SAME:      %[[VAL_1:.*]]: tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>,
158// CHECK-SAME:      %[[VAL_2:.*]]: tensor<6x6xi32>) -> tensor<6x6xi32> {
159// CHECK-DAG:       %[[VAL_3:.*]] = arith.constant 0 : index
160// CHECK-DAG:       %[[VAL_4:.*]] = arith.constant 1 : index
161// CHECK-DAG:       %[[VAL_5:.*]] = arith.constant 6 : index
162// CHECK:           %[[VAL_6:.*]] = bufferization.to_memref %[[VAL_0]] : memref<8x8xi32>
163// CHECK:           %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_3]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
164// CHECK:           %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_3]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
165// CHECK:           %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
166// CHECK:           %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
167// CHECK:           %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
168// CHECK:           %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : memref<6x6xi32>
169// CHECK:           %[[VAL_13:.*]] = memref.alloc() : memref<6x6xi32>
170// CHECK:           memref.copy %[[VAL_12]], %[[VAL_13]] : memref<6x6xi32> to memref<6x6xi32>
171// CHECK:           %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_3]]] : memref<?xindex>
172// CHECK:           %[[VAL_15:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_4]]] : memref<?xindex>
173// CHECK:           scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_4]] {
174// CHECK:             %[[VAL_17:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex>
175// CHECK:             %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_16]]] : memref<?xindex>
176// CHECK:             %[[VAL_19:.*]] = arith.addi %[[VAL_16]], %[[VAL_4]] : index
177// CHECK:             %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_19]]] : memref<?xindex>
178// CHECK:             scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_4]] {
179// CHECK:               %[[VAL_22:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex>
180// CHECK:               %[[VAL_23:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_21]]] : memref<?xi32>
181// CHECK:               scf.for %[[VAL_24:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
182// CHECK:                 scf.for %[[VAL_25:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
183// CHECK:                   %[[VAL_26:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_25]], %[[VAL_24]]] : memref<6x6xi32>
184// CHECK:                   %[[VAL_27:.*]] = arith.addi %[[VAL_25]], %[[VAL_17]] : index
185// CHECK:                   %[[VAL_28:.*]] = arith.addi %[[VAL_24]], %[[VAL_22]] : index
186// CHECK:                   %[[VAL_29:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_27]], %[[VAL_28]]] : memref<8x8xi32>
187// CHECK:                   %[[VAL_30:.*]] = arith.muli %[[VAL_29]], %[[VAL_23]] : i32
188// CHECK:                   %[[VAL_31:.*]] = arith.addi %[[VAL_26]], %[[VAL_30]] : i32
189// CHECK:                   memref.store %[[VAL_31]], %[[VAL_13]]{{\[}}%[[VAL_25]], %[[VAL_24]]] : memref<6x6xi32>
190// CHECK:                 }
191// CHECK:               }
192// CHECK:             }
193// CHECK:           }
194// CHECK:           %[[VAL_32:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<6x6xi32>
195// CHECK:           return %[[VAL_32]] : tensor<6x6xi32>
196// CHECK:         }
197func @conv2d(%input:  tensor<8x8xi32>,
198             %filter: tensor<3x3xi32, #DCSR>,
199             %output: tensor<6x6xi32>) -> tensor<6x6xi32> {
200  %0 = linalg.conv_2d
201    ins  (%input, %filter: tensor<8x8xi32>, tensor<3x3xi32, #DCSR>)
202    outs (%output: tensor<6x6xi32>) -> tensor<6x6xi32>
203  return %0 : tensor<6x6xi32>
204}
205
206// CHECK-LABEL:   func @quantized_matmul(
207// CHECK-SAME:      %[[VAL_0:.*]]: tensor<5x3xi8>,
208// CHECK-SAME:      %[[VAL_1:.*]]: tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>,
209// CHECK-SAME:      %[[VAL_2:.*]]: tensor<5x6xi64>) -> tensor<5x6xi64> {
210// CHECK-DAG:       %[[VAL_3:.*]] = arith.constant 2 : i64
211// CHECK-DAG:       %[[VAL_4:.*]] = arith.constant 0 : index
212// CHECK-DAG:       %[[VAL_5:.*]] = arith.constant 1 : index
213// CHECK-DAG:       %[[VAL_6:.*]] = arith.constant 5 : index
214// CHECK:           %[[VAL_7:.*]] = bufferization.to_memref %[[VAL_0]] : memref<5x3xi8>
215// CHECK:           %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
216// CHECK:           %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
217// CHECK:           %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_5]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
218// CHECK:           %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_5]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
219// CHECK:           %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
220// CHECK:           %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_2]] : memref<5x6xi64>
221// CHECK:           %[[VAL_14:.*]] = memref.alloc() : memref<5x6xi64>
222// CHECK:           memref.copy %[[VAL_13]], %[[VAL_14]] : memref<5x6xi64> to memref<5x6xi64>
223// CHECK:           %[[VAL_15:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_4]]] : memref<?xindex>
224// CHECK:           %[[VAL_16:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_5]]] : memref<?xindex>
225// CHECK:           scf.for %[[VAL_17:.*]] = %[[VAL_15]] to %[[VAL_16]] step %[[VAL_5]] {
226// CHECK:             %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_17]]] : memref<?xindex>
227// CHECK:             %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_17]]] : memref<?xindex>
228// CHECK:             %[[VAL_20:.*]] = arith.addi %[[VAL_17]], %[[VAL_5]] : index
229// CHECK:             %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_20]]] : memref<?xindex>
230// CHECK:             scf.for %[[VAL_22:.*]] = %[[VAL_19]] to %[[VAL_21]] step %[[VAL_5]] {
231// CHECK:               %[[VAL_23:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_22]]] : memref<?xindex>
232// CHECK:               %[[VAL_24:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]]] : memref<?xi8>
233// CHECK:               scf.for %[[VAL_25:.*]] = %[[VAL_4]] to %[[VAL_6]] step %[[VAL_5]] {
234// CHECK:                 %[[VAL_26:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_25]], %[[VAL_23]]] : memref<5x6xi64>
235// CHECK:                 %[[VAL_27:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_25]], %[[VAL_18]]] : memref<5x3xi8>
236// CHECK:                 %[[VAL_28:.*]] = arith.extsi %[[VAL_27]] : i8 to i64
237// CHECK:                 %[[VAL_29:.*]] = arith.subi %[[VAL_28]], %[[VAL_3]] : i64
238// CHECK:                 %[[VAL_30:.*]] = arith.extsi %[[VAL_24]] : i8 to i64
239// CHECK:                 %[[VAL_31:.*]] = arith.muli %[[VAL_29]], %[[VAL_30]] : i64
240// CHECK:                 %[[VAL_32:.*]] = arith.addi %[[VAL_26]], %[[VAL_31]] : i64
241// CHECK:                 memref.store %[[VAL_32]], %[[VAL_14]]{{\[}}%[[VAL_25]], %[[VAL_23]]] : memref<5x6xi64>
242// CHECK:               }
243// CHECK:             }
244// CHECK:           }
245// CHECK:           %[[VAL_33:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<5x6xi64>
246// CHECK:           return %[[VAL_33]] : tensor<5x6xi64>
247// CHECK:         }
248func @quantized_matmul(%input1: tensor<5x3xi8>,
249                       %input2: tensor<3x6xi8, #DCSR>,
250                       %output: tensor<5x6xi64>) -> tensor<5x6xi64> {
251  %c0 = arith.constant 0 : i32
252  %c2 = arith.constant 2 : i32
253  %0 = linalg.quantized_matmul
254    ins(%input1, %input2, %c2, %c0 : tensor<5x3xi8>, tensor<3x6xi8, #DCSR>, i32, i32)
255    outs(%output : tensor<5x6xi64>) -> tensor<5x6xi64>
256  return %0: tensor<5x6xi64>
257}
258