1// RUN: mlir-opt %s --sparse-compiler | \ 2// RUN: mlir-cpu-runner -e entry -entry-point-result=void \ 3// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ 4// RUN: FileCheck %s 5// 6// Do the same run, but now with SIMDization as well. This should not change the outcome. 7// 8// RUN: mlir-opt %s --sparse-compiler="vectorization-strategy=2 vl=4" | \ 9// RUN: mlir-cpu-runner -e entry -entry-point-result=void \ 10// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ 11// RUN: FileCheck %s 12 13#SparseVector = #sparse_tensor.encoding<{ 14 dimLevelType = ["compressed"] 15}> 16 17#SparseMatrix = #sparse_tensor.encoding<{ 18 dimLevelType = ["compressed", "compressed"] 19}> 20 21#trait_1d = { 22 indexing_maps = [ 23 affine_map<(i) -> (i)>, // a 24 affine_map<(i) -> (i)> // x (out) 25 ], 26 iterator_types = ["parallel"], 27 doc = "X(i) = a(i) op i" 28} 29 30#trait_2d = { 31 indexing_maps = [ 32 affine_map<(i,j) -> (i,j)>, // A 33 affine_map<(i,j) -> (i,j)> // X (out) 34 ], 35 iterator_types = ["parallel", "parallel"], 36 doc = "X(i,j) = A(i,j) op i op j" 37} 38 39// 40// Test with indices and sparse inputs. All outputs are dense. 41// 42module { 43 44 // 45 // Kernel that uses index in the index notation (conjunction). 46 // 47 func.func @sparse_index_1d_conj(%arga: tensor<8xi64, #SparseVector>) -> tensor<8xi64> { 48 %init = linalg.init_tensor [8] : tensor<8xi64> 49 %r = linalg.generic #trait_1d 50 ins(%arga: tensor<8xi64, #SparseVector>) 51 outs(%init: tensor<8xi64>) { 52 ^bb(%a: i64, %x: i64): 53 %i = linalg.index 0 : index 54 %ii = arith.index_cast %i : index to i64 55 %m1 = arith.muli %a, %ii : i64 56 linalg.yield %m1 : i64 57 } -> tensor<8xi64> 58 return %r : tensor<8xi64> 59 } 60 61 // 62 // Kernel that uses index in the index notation (disjunction). 63 // 64 func.func @sparse_index_1d_disj(%arga: tensor<8xi64, #SparseVector>) -> tensor<8xi64> { 65 %init = linalg.init_tensor [8] : tensor<8xi64> 66 %r = linalg.generic #trait_1d 67 ins(%arga: tensor<8xi64, #SparseVector>) 68 outs(%init: tensor<8xi64>) { 69 ^bb(%a: i64, %x: i64): 70 %i = linalg.index 0 : index 71 %ii = arith.index_cast %i : index to i64 72 %m1 = arith.addi %a, %ii : i64 73 linalg.yield %m1 : i64 74 } -> tensor<8xi64> 75 return %r : tensor<8xi64> 76 } 77 78 // 79 // Kernel that uses indices in the index notation (conjunction). 80 // 81 func.func @sparse_index_2d_conj(%arga: tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> { 82 %init = linalg.init_tensor [3,4] : tensor<3x4xi64> 83 %r = linalg.generic #trait_2d 84 ins(%arga: tensor<3x4xi64, #SparseMatrix>) 85 outs(%init: tensor<3x4xi64>) { 86 ^bb(%a: i64, %x: i64): 87 %i = linalg.index 0 : index 88 %j = linalg.index 1 : index 89 %ii = arith.index_cast %i : index to i64 90 %jj = arith.index_cast %j : index to i64 91 %m1 = arith.muli %ii, %a : i64 92 %m2 = arith.muli %jj, %m1 : i64 93 linalg.yield %m2 : i64 94 } -> tensor<3x4xi64> 95 return %r : tensor<3x4xi64> 96 } 97 98 // 99 // Kernel that uses indices in the index notation (disjunction). 100 // 101 func.func @sparse_index_2d_disj(%arga: tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> { 102 %init = linalg.init_tensor [3,4] : tensor<3x4xi64> 103 %r = linalg.generic #trait_2d 104 ins(%arga: tensor<3x4xi64, #SparseMatrix>) 105 outs(%init: tensor<3x4xi64>) { 106 ^bb(%a: i64, %x: i64): 107 %i = linalg.index 0 : index 108 %j = linalg.index 1 : index 109 %ii = arith.index_cast %i : index to i64 110 %jj = arith.index_cast %j : index to i64 111 %m1 = arith.addi %ii, %a : i64 112 %m2 = arith.addi %jj, %m1 : i64 113 linalg.yield %m2 : i64 114 } -> tensor<3x4xi64> 115 return %r : tensor<3x4xi64> 116 } 117 118 // 119 // Main driver. 120 // 121 func.func @entry() { 122 %c0 = arith.constant 0 : index 123 %du = arith.constant -1 : i64 124 125 // Setup input sparse vector. 126 %v1 = arith.constant sparse<[[2], [4]], [ 10, 20]> : tensor<8xi64> 127 %sv = sparse_tensor.convert %v1 : tensor<8xi64> to tensor<8xi64, #SparseVector> 128 129 // Setup input "sparse" vector. 130 %v2 = arith.constant dense<[ 1, 2, 4, 8, 16, 32, 64, 128 ]> : tensor<8xi64> 131 %dv = sparse_tensor.convert %v2 : tensor<8xi64> to tensor<8xi64, #SparseVector> 132 133 // Setup input sparse matrix. 134 %m1 = arith.constant sparse<[[1,1], [2,3]], [10, 20]> : tensor<3x4xi64> 135 %sm = sparse_tensor.convert %m1 : tensor<3x4xi64> to tensor<3x4xi64, #SparseMatrix> 136 137 // Setup input "sparse" matrix. 138 %m2 = arith.constant dense <[ [ 1, 1, 1, 1 ], 139 [ 1, 2, 1, 1 ], 140 [ 1, 1, 3, 4 ] ]> : tensor<3x4xi64> 141 %dm = sparse_tensor.convert %m2 : tensor<3x4xi64> to tensor<3x4xi64, #SparseMatrix> 142 143 // Call the kernels. 144 %0 = call @sparse_index_1d_conj(%sv) : (tensor<8xi64, #SparseVector>) -> tensor<8xi64> 145 %1 = call @sparse_index_1d_disj(%sv) : (tensor<8xi64, #SparseVector>) -> tensor<8xi64> 146 %2 = call @sparse_index_1d_conj(%dv) : (tensor<8xi64, #SparseVector>) -> tensor<8xi64> 147 %3 = call @sparse_index_1d_disj(%dv) : (tensor<8xi64, #SparseVector>) -> tensor<8xi64> 148 %4 = call @sparse_index_2d_conj(%sm) : (tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> 149 %5 = call @sparse_index_2d_disj(%sm) : (tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> 150 %6 = call @sparse_index_2d_conj(%dm) : (tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> 151 %7 = call @sparse_index_2d_disj(%dm) : (tensor<3x4xi64, #SparseMatrix>) -> tensor<3x4xi64> 152 153 // Get the backing buffers. 154 %mem0 = bufferization.to_memref %0 : memref<8xi64> 155 %mem1 = bufferization.to_memref %1 : memref<8xi64> 156 %mem2 = bufferization.to_memref %2 : memref<8xi64> 157 %mem3 = bufferization.to_memref %3 : memref<8xi64> 158 %mem4 = bufferization.to_memref %4 : memref<3x4xi64> 159 %mem5 = bufferization.to_memref %5 : memref<3x4xi64> 160 %mem6 = bufferization.to_memref %6 : memref<3x4xi64> 161 %mem7 = bufferization.to_memref %7 : memref<3x4xi64> 162 163 // 164 // Verify result. 165 // 166 // CHECK: ( 0, 0, 20, 0, 80, 0, 0, 0 ) 167 // CHECK-NEXT: ( 0, 1, 12, 3, 24, 5, 6, 7 ) 168 // CHECK-NEXT: ( 0, 2, 8, 24, 64, 160, 384, 896 ) 169 // CHECK-NEXT: ( 1, 3, 6, 11, 20, 37, 70, 135 ) 170 // CHECK-NEXT: ( ( 0, 0, 0, 0 ), ( 0, 10, 0, 0 ), ( 0, 0, 0, 120 ) ) 171 // CHECK-NEXT: ( ( 0, 1, 2, 3 ), ( 1, 12, 3, 4 ), ( 2, 3, 4, 25 ) ) 172 // CHECK-NEXT: ( ( 0, 0, 0, 0 ), ( 0, 2, 2, 3 ), ( 0, 2, 12, 24 ) ) 173 // CHECK-NEXT: ( ( 1, 2, 3, 4 ), ( 2, 4, 4, 5 ), ( 3, 4, 7, 9 ) ) 174 // 175 %vv0 = vector.transfer_read %mem0[%c0], %du: memref<8xi64>, vector<8xi64> 176 %vv1 = vector.transfer_read %mem1[%c0], %du: memref<8xi64>, vector<8xi64> 177 %vv2 = vector.transfer_read %mem2[%c0], %du: memref<8xi64>, vector<8xi64> 178 %vv3 = vector.transfer_read %mem3[%c0], %du: memref<8xi64>, vector<8xi64> 179 %vv4 = vector.transfer_read %mem4[%c0,%c0], %du: memref<3x4xi64>, vector<3x4xi64> 180 %vv5 = vector.transfer_read %mem5[%c0,%c0], %du: memref<3x4xi64>, vector<3x4xi64> 181 %vv6 = vector.transfer_read %mem6[%c0,%c0], %du: memref<3x4xi64>, vector<3x4xi64> 182 %vv7 = vector.transfer_read %mem7[%c0,%c0], %du: memref<3x4xi64>, vector<3x4xi64> 183 vector.print %vv0 : vector<8xi64> 184 vector.print %vv1 : vector<8xi64> 185 vector.print %vv2 : vector<8xi64> 186 vector.print %vv3 : vector<8xi64> 187 vector.print %vv4 : vector<3x4xi64> 188 vector.print %vv5 : vector<3x4xi64> 189 vector.print %vv6 : vector<3x4xi64> 190 vector.print %vv7 : vector<3x4xi64> 191 192 // Release resources. 193 sparse_tensor.release %sv : tensor<8xi64, #SparseVector> 194 sparse_tensor.release %dv : tensor<8xi64, #SparseVector> 195 sparse_tensor.release %sm : tensor<3x4xi64, #SparseMatrix> 196 sparse_tensor.release %dm : tensor<3x4xi64, #SparseMatrix> 197 memref.dealloc %mem0 : memref<8xi64> 198 memref.dealloc %mem1 : memref<8xi64> 199 memref.dealloc %mem2 : memref<8xi64> 200 memref.dealloc %mem3 : memref<8xi64> 201 memref.dealloc %mem4 : memref<3x4xi64> 202 memref.dealloc %mem5 : memref<3x4xi64> 203 memref.dealloc %mem6 : memref<3x4xi64> 204 memref.dealloc %mem7 : memref<3x4xi64> 205 206 return 207 } 208} 209