1// RUN: mlir-opt %s -sparsification | FileCheck %s --check-prefix=CHECK-HIR
2//
3// RUN: mlir-opt %s -sparsification --sparse-tensor-conversion                 \
4// RUN: --convert-linalg-to-loops | FileCheck %s --check-prefix=CHECK-MIR
5//
6// RUN: mlir-opt %s -sparsification --sparse-tensor-conversion                 \
7// RUN: --convert-linalg-to-loops --func-bufferize --tensor-constant-bufferize \
8// RUN: --tensor-bufferize --finalizing-bufferize |                            \
9// RUN: FileCheck %s --check-prefix=CHECK-LIR
10//
11// RUN: mlir-opt %s -sparsification="fast-output" --sparse-tensor-conversion   \
12// RUN: --convert-linalg-to-loops --func-bufferize --tensor-constant-bufferize \
13// RUN: --tensor-bufferize --finalizing-bufferize |                            \
14// RUN: FileCheck %s --check-prefix=CHECK-FAST
15
16#CSR = #sparse_tensor.encoding<{dimLevelType = [ "dense", "compressed" ]}>
17
18#trait_matvec = {
19  indexing_maps = [
20    affine_map<(i,j) -> (i,j)>,  // A
21    affine_map<(i,j) -> (j)>,    // b
22    affine_map<(i,j) -> (i)>     // x (out)
23  ],
24  iterator_types = ["parallel","reduction"],
25  doc = "x(i) += A(i,j) * b(j)"
26}
27
28// CHECK-HIR-LABEL:   func @matvec(
29// CHECK-HIR-SAME:                 %[[VAL_0:.*]]: tensor<64x64xf64, #sparse_tensor.encoding<{{.*}}>>,
30// CHECK-HIR-SAME:                 %[[VAL_1:.*]]: tensor<64xf64>,
31// CHECK-HIR-SAME:                 %[[VAL_2:.*]]: tensor<64xf64>) -> tensor<64xf64> {
32// CHECK-HIR:           %[[VAL_3:.*]] = constant 64 : index
33// CHECK-HIR:           %[[VAL_4:.*]] = constant 0 : index
34// CHECK-HIR:           %[[VAL_5:.*]] = constant 1 : index
35// CHECK-HIR:           %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_5]] : tensor<64x64xf64, #{{.*}}> to memref<?xindex>
36// CHECK-HIR:           %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_5]] : tensor<64x64xf64, #{{.*}}> to memref<?xindex>
37// CHECK-HIR:           %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<64x64xf64, #{{.*}}> to memref<?xf64>
38// CHECK-HIR:           %[[VAL_10:.*]] = memref.buffer_cast %[[VAL_1]] : memref<64xf64>
39// CHECK-HIR:           %[[VAL_11:.*]] = memref.buffer_cast %[[VAL_2]] : memref<64xf64>
40// CHECK-HIR:           %[[VAL_12:.*]] = memref.alloc() : memref<64xf64>
41// CHECK-HIR:           linalg.copy(%[[VAL_11]], %[[VAL_12]]) : memref<64xf64>, memref<64xf64>
42// CHECK-HIR:           scf.for %[[VAL_13:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
43// CHECK-HIR:             %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_13]]] : memref<?xindex>
44// CHECK-HIR:             %[[VAL_15:.*]] = addi %[[VAL_13]], %[[VAL_5]] : index
45// CHECK-HIR:             %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex>
46// CHECK-HIR:             %[[VAL_17:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_13]]] : memref<64xf64>
47// CHECK-HIR:             %[[VAL_18:.*]] = scf.for %[[VAL_19:.*]] = %[[VAL_14]] to %[[VAL_16]] step %[[VAL_5]] iter_args(%[[VAL_20:.*]] = %[[VAL_17]]) -> (f64) {
48// CHECK-HIR:               %[[VAL_21:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex>
49// CHECK-HIR:               %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_19]]] : memref<?xf64>
50// CHECK-HIR:               %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<64xf64>
51// CHECK-HIR:               %[[VAL_24:.*]] = mulf %[[VAL_22]], %[[VAL_23]] : f64
52// CHECK-HIR:               %[[VAL_25:.*]] = addf %[[VAL_20]], %[[VAL_24]] : f64
53// CHECK-HIR:               scf.yield %[[VAL_25]] : f64
54// CHECK-HIR:             }
55// CHECK-HIR:             store %[[VAL_26:.*]], %[[VAL_12]]{{\[}}%[[VAL_13]]] : memref<64xf64>
56// CHECK-HIR:           }
57// CHECK-HIR:           %[[VAL_27:.*]] = memref.tensor_load %[[VAL_12]] : memref<64xf64>
58// CHECK-HIR:           return %[[VAL_27]] : tensor<64xf64>
59// CHECK-HIR:         }
60
61// CHECK-MIR-LABEL:   func @matvec(
62// CHECK-MIR-SAME:                 %[[VAL_0:.*]]: !llvm.ptr<i8>,
63// CHECK-MIR-SAME:                 %[[VAL_1:.*]]: tensor<64xf64>,
64// CHECK-MIR-SAME:                 %[[VAL_2:.*]]: tensor<64xf64>) -> tensor<64xf64> {
65// CHECK-MIR:           %[[VAL_3:.*]] = constant 64 : index
66// CHECK-MIR:           %[[VAL_4:.*]] = constant 0 : index
67// CHECK-MIR:           %[[VAL_5:.*]] = constant 1 : index
68// CHECK-MIR:           %[[VAL_6:.*]] = call @sparsePointers(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
69// CHECK-MIR:           %[[VAL_7:.*]] = call @sparseIndices(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
70// CHECK-MIR:           %[[VAL_8:.*]] = call @sparseValuesF64(%[[VAL_0]]) : (!llvm.ptr<i8>) -> memref<?xf64>
71// CHECK-MIR:           %[[VAL_9:.*]] = memref.buffer_cast %[[VAL_1]] : memref<64xf64>
72// CHECK-MIR:           %[[VAL_10:.*]] = memref.buffer_cast %[[VAL_2]] : memref<64xf64>
73// CHECK-MIR:           %[[VAL_11:.*]] = memref.alloc() : memref<64xf64>
74// CHECK-MIR:           scf.for %[[VAL_12:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
75// CHECK-MIR:             %[[VAL_13:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_12]]] : memref<64xf64>
76// CHECK-MIR:             store %[[VAL_13]], %[[VAL_11]]{{\[}}%[[VAL_12]]] : memref<64xf64>
77// CHECK-MIR:           }
78// CHECK-MIR:           scf.for %[[VAL_14:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
79// CHECK-MIR:             %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_14]]] : memref<?xindex>
80// CHECK-MIR:             %[[VAL_16:.*]] = addi %[[VAL_14]], %[[VAL_5]] : index
81// CHECK-MIR:             %[[VAL_17:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_16]]] : memref<?xindex>
82// CHECK-MIR:             %[[VAL_18:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_14]]] : memref<64xf64>
83// CHECK-MIR:             %[[VAL_19:.*]] = scf.for %[[VAL_20:.*]] = %[[VAL_15]] to %[[VAL_17]] step %[[VAL_5]] iter_args(%[[VAL_21:.*]] = %[[VAL_18]]) -> (f64) {
84// CHECK-MIR:               %[[VAL_22:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_20]]] : memref<?xindex>
85// CHECK-MIR:               %[[VAL_23:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_20]]] : memref<?xf64>
86// CHECK-MIR:               %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_22]]] : memref<64xf64>
87// CHECK-MIR:               %[[VAL_25:.*]] = mulf %[[VAL_23]], %[[VAL_24]] : f64
88// CHECK-MIR:               %[[VAL_26:.*]] = addf %[[VAL_21]], %[[VAL_25]] : f64
89// CHECK-MIR:               scf.yield %[[VAL_26]] : f64
90// CHECK-MIR:             }
91// CHECK-MIR:             store %[[VAL_27:.*]], %[[VAL_11]]{{\[}}%[[VAL_14]]] : memref<64xf64>
92// CHECK-MIR:           }
93// CHECK-MIR:           %[[VAL_28:.*]] = memref.tensor_load %[[VAL_11]] : memref<64xf64>
94// CHECK-MIR:           return %[[VAL_28]] : tensor<64xf64>
95// CHECK-MIR:         }
96
97// CHECK-LIR-LABEL:   func @matvec(
98// CHECK-LIR-SAME:                 %[[VAL_0:.*]]: !llvm.ptr<i8>,
99// CHECK-LIR-SAME:                 %[[VAL_1:.*]]: memref<64xf64>,
100// CHECK-LIR-SAME:                 %[[VAL_2:.*]]: memref<64xf64>) -> memref<64xf64> {
101// CHECK-LIR:           %[[VAL_3:.*]] = constant 64 : index
102// CHECK-LIR:           %[[VAL_4:.*]] = constant 0 : index
103// CHECK-LIR:           %[[VAL_5:.*]] = constant 1 : index
104// CHECK-LIR:           %[[VAL_6:.*]] = call @sparsePointers(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
105// CHECK-LIR:           %[[VAL_7:.*]] = call @sparseIndices(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
106// CHECK-LIR:           %[[VAL_8:.*]] = call @sparseValuesF64(%[[VAL_0]]) : (!llvm.ptr<i8>) -> memref<?xf64>
107// CHECK-LIR:           %[[VAL_9:.*]] = memref.alloc() : memref<64xf64>
108// CHECK-LIR:           scf.for %[[VAL_10:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
109// CHECK-LIR:             %[[VAL_11:.*]] = memref.load %[[VAL_2]]{{\[}}%[[VAL_10]]] : memref<64xf64>
110// CHECK-LIR:             store %[[VAL_11]], %[[VAL_9]]{{\[}}%[[VAL_10]]] : memref<64xf64>
111// CHECK-LIR:           }
112// CHECK-LIR:           scf.for %[[VAL_12:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
113// CHECK-LIR:             %[[VAL_13:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_12]]] : memref<?xindex>
114// CHECK-LIR:             %[[VAL_14:.*]] = addi %[[VAL_12]], %[[VAL_5]] : index
115// CHECK-LIR:             %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_14]]] : memref<?xindex>
116// CHECK-LIR:             %[[VAL_16:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_12]]] : memref<64xf64>
117// CHECK-LIR:             %[[VAL_17:.*]] = scf.for %[[VAL_18:.*]] = %[[VAL_13]] to %[[VAL_15]] step %[[VAL_5]] iter_args(%[[VAL_19:.*]] = %[[VAL_16]]) -> (f64) {
118// CHECK-LIR:               %[[VAL_20:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_18]]] : memref<?xindex>
119// CHECK-LIR:               %[[VAL_21:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xf64>
120// CHECK-LIR:               %[[VAL_22:.*]] = memref.load %[[VAL_1]]{{\[}}%[[VAL_20]]] : memref<64xf64>
121// CHECK-LIR:               %[[VAL_23:.*]] = mulf %[[VAL_21]], %[[VAL_22]] : f64
122// CHECK-LIR:               %[[VAL_24:.*]] = addf %[[VAL_19]], %[[VAL_23]] : f64
123// CHECK-LIR:               scf.yield %[[VAL_24]] : f64
124// CHECK-LIR:             }
125// CHECK-LIR:             store %[[VAL_25:.*]], %[[VAL_9]]{{\[}}%[[VAL_12]]] : memref<64xf64>
126// CHECK-LIR:           }
127// CHECK-LIR:           return %[[VAL_9]] : memref<64xf64>
128// CHECK-LIR:         }
129
130// CHECK-FAST-LABEL:   func @matvec(
131// CHECK-FAST-SAME:                 %[[VAL_0:.*]]: !llvm.ptr<i8>,
132// CHECK-FAST-SAME:                 %[[VAL_1:.*]]: memref<64xf64>,
133// CHECK-FAST-SAME:                 %[[VAL_2:.*]]: memref<64xf64>) -> memref<64xf64> {
134// CHECK-FAST:           %[[VAL_3:.*]] = constant 64 : index
135// CHECK-FAST:           %[[VAL_4:.*]] = constant 0 : index
136// CHECK-FAST:           %[[VAL_5:.*]] = constant 1 : index
137// CHECK-FAST:           %[[VAL_6:.*]] = call @sparsePointers(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
138// CHECK-FAST:           %[[VAL_7:.*]] = call @sparseIndices(%[[VAL_0]], %[[VAL_5]]) : (!llvm.ptr<i8>, index) -> memref<?xindex>
139// CHECK-FAST:           %[[VAL_8:.*]] = call @sparseValuesF64(%[[VAL_0]]) : (!llvm.ptr<i8>) -> memref<?xf64>
140// CHECK-FAST:           scf.for %[[VAL_9:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] {
141// CHECK-FAST:             %[[VAL_10:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_9]]] : memref<?xindex>
142// CHECK-FAST:             %[[VAL_11:.*]] = addi %[[VAL_9]], %[[VAL_5]] : index
143// CHECK-FAST:             %[[VAL_12:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_11]]] : memref<?xindex>
144// CHECK-FAST:             %[[VAL_13:.*]] = memref.load %[[VAL_2]]{{\[}}%[[VAL_9]]] : memref<64xf64>
145// CHECK-FAST:             %[[VAL_14:.*]] = scf.for %[[VAL_15:.*]] = %[[VAL_10]] to %[[VAL_12]] step %[[VAL_5]] iter_args(%[[VAL_16:.*]] = %[[VAL_13]]) -> (f64) {
146// CHECK-FAST:               %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex>
147// CHECK-FAST:               %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]]] : memref<?xf64>
148// CHECK-FAST:               %[[VAL_19:.*]] = memref.load %[[VAL_1]]{{\[}}%[[VAL_17]]] : memref<64xf64>
149// CHECK-FAST:               %[[VAL_20:.*]] = mulf %[[VAL_18]], %[[VAL_19]] : f64
150// CHECK-FAST:               %[[VAL_21:.*]] = addf %[[VAL_16]], %[[VAL_20]] : f64
151// CHECK-FAST:               scf.yield %[[VAL_21]] : f64
152// CHECK-FAST:             }
153// CHECK-FAST:             store %[[VAL_22:.*]], %[[VAL_2]]{{\[}}%[[VAL_9]]] : memref<64xf64>
154// CHECK-FAST:           }
155// CHECK-FAST:           return %[[VAL_2]] : memref<64xf64>
156// CHECK-FAST:         }
157
158func @matvec(%arga: tensor<64x64xf64, #CSR>, %argb: tensor<64xf64>, %argx: tensor<64xf64>) -> tensor<64xf64> {
159  %0 = linalg.generic #trait_matvec
160      ins(%arga, %argb : tensor<64x64xf64, #CSR>, tensor<64xf64>)
161      outs(%argx: tensor<64xf64>) {
162    ^bb(%A: f64, %b: f64, %x: f64):
163      %0 = mulf %A, %b : f64
164      %1 = addf %x, %0 : f64
165      linalg.yield %1 : f64
166  } -> tensor<64xf64>
167  return %0 : tensor<64xf64>
168}
169