1//-------------------------------------------------------------------------------------------------- 2// WHEN CREATING A NEW TEST, PLEASE JUST COPY & PASTE WITHOUT EDITS. 3// 4// Set-up that's shared across all tests in this directory. In principle, this 5// config could be moved to lit.local.cfg. However, there are downstream users that 6// do not use these LIT config files. Hence why this is kept inline. 7// 8// DEFINE: %{sparsifier_opts} = enable-runtime-library=true 9// DEFINE: %{sparsifier_opts_sve} = enable-arm-sve=true %{sparsifier_opts} 10// DEFINE: %{compile} = mlir-opt %s --sparsifier="%{sparsifier_opts}" 11// DEFINE: %{compile_sve} = mlir-opt %s --sparsifier="%{sparsifier_opts_sve}" 12// DEFINE: %{run_libs} = -shared-libs=%mlir_c_runner_utils,%mlir_runner_utils 13// DEFINE: %{run_libs_sve} = -shared-libs=%native_mlir_runner_utils,%native_mlir_c_runner_utils 14// DEFINE: %{run_opts} = -e main -entry-point-result=void 15// DEFINE: %{run} = mlir-runner %{run_opts} %{run_libs} 16// DEFINE: %{run_sve} = %mcr_aarch64_cmd --march=aarch64 --mattr="+sve" %{run_opts} %{run_libs_sve} 17// 18// DEFINE: %{env} = 19//-------------------------------------------------------------------------------------------------- 20 21// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false 22// RUN: %{compile} | %{run} | FileCheck %s 23// 24// Do the same run, but now with vectorization. 25// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true 26// RUN: %{compile} | %{run} | FileCheck %s 27// 28// Do the same run, but now VLA vectorization. 29// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %} 30 31#Dense = #sparse_tensor.encoding<{ 32 map = (d0, d1) -> (d0 : dense, d1 : dense) 33}> 34 35#SortedCOO = #sparse_tensor.encoding<{ 36 map = (d0, d1) -> (d0 : compressed(nonunique), d1 : singleton(soa)) 37}> 38 39#CSR = #sparse_tensor.encoding<{ 40 map = (d0, d1) -> (d0 : dense, d1 : compressed) 41}> 42 43#DCSR = #sparse_tensor.encoding<{ 44 map = (d0, d1) -> (d0 : compressed, d1 : compressed) 45}> 46 47#Row = #sparse_tensor.encoding<{ 48 map = (d0, d1) -> (d0 : compressed, d1 : dense) 49}> 50 51module { 52 // 53 // Main driver. We test the contents of various sparse tensor 54 // schemes when they are still empty and after a few insertions. 55 // 56 func.func @main() { 57 %c0 = arith.constant 0 : index 58 %c2 = arith.constant 2 : index 59 %c3 = arith.constant 3 : index 60 %f1 = arith.constant 1.0 : f64 61 %f2 = arith.constant 2.0 : f64 62 %f3 = arith.constant 3.0 : f64 63 %f4 = arith.constant 4.0 : f64 64 65 // 66 // Dense case. 67 // 68 // CHECK: ---- Sparse Tensor ---- 69 // CHECK-NEXT: nse = 12 70 // CHECK-NEXT: dim = ( 4, 3 ) 71 // CHECK-NEXT: lvl = ( 4, 3 ) 72 // CHECK-NEXT: values : ( 1, 0, 0, 0, 0, 0, 0, 0, 2, 3, 0, 4 ) 73 // CHECK-NEXT: ---- 74 // 75 %densea = tensor.empty() : tensor<4x3xf64, #Dense> 76 %dense1 = tensor.insert %f1 into %densea[%c0, %c0] : tensor<4x3xf64, #Dense> 77 %dense2 = tensor.insert %f2 into %dense1[%c2, %c2] : tensor<4x3xf64, #Dense> 78 %dense3 = tensor.insert %f3 into %dense2[%c3, %c0] : tensor<4x3xf64, #Dense> 79 %dense4 = tensor.insert %f4 into %dense3[%c3, %c2] : tensor<4x3xf64, #Dense> 80 %densem = sparse_tensor.load %dense4 hasInserts : tensor<4x3xf64, #Dense> 81 sparse_tensor.print %densem : tensor<4x3xf64, #Dense> 82 83 // 84 // COO case. 85 // 86 // CHECK-NEXT: ---- Sparse Tensor ---- 87 // CHECK-NEXT: nse = 4 88 // CHECK-NEXT: dim = ( 4, 3 ) 89 // CHECK-NEXT: lvl = ( 4, 3 ) 90 // CHECK-NEXT: pos[0] : ( 0, 4 ) 91 // CHECK-NEXT: crd[0] : ( 0, 2, 3, 3 ) 92 // CHECK-NEXT: crd[1] : ( 0, 2, 0, 2 ) 93 // CHECK-NEXT: values : ( 1, 2, 3, 4 ) 94 // CHECK-NEXT: ---- 95 // 96 %cooa = tensor.empty() : tensor<4x3xf64, #SortedCOO> 97 %coo1 = tensor.insert %f1 into %cooa[%c0, %c0] : tensor<4x3xf64, #SortedCOO> 98 %coo2 = tensor.insert %f2 into %coo1[%c2, %c2] : tensor<4x3xf64, #SortedCOO> 99 %coo3 = tensor.insert %f3 into %coo2[%c3, %c0] : tensor<4x3xf64, #SortedCOO> 100 %coo4 = tensor.insert %f4 into %coo3[%c3, %c2] : tensor<4x3xf64, #SortedCOO> 101 %coom = sparse_tensor.load %coo4 hasInserts : tensor<4x3xf64, #SortedCOO> 102 sparse_tensor.print %coom : tensor<4x3xf64, #SortedCOO> 103 104 // 105 // CSR case. 106 // 107 // CHECK-NEXT: ---- Sparse Tensor ---- 108 // CHECK-NEXT: nse = 4 109 // CHECK-NEXT: dim = ( 4, 3 ) 110 // CHECK-NEXT: lvl = ( 4, 3 ) 111 // CHECK-NEXT: pos[1] : ( 0, 1, 1, 2, 4 ) 112 // CHECK-NEXT: crd[1] : ( 0, 2, 0, 2 ) 113 // CHECK-NEXT: values : ( 1, 2, 3, 4 ) 114 // CHECK-NEXT: ---- 115 // 116 %csra = tensor.empty() : tensor<4x3xf64, #CSR> 117 %csr1 = tensor.insert %f1 into %csra[%c0, %c0] : tensor<4x3xf64, #CSR> 118 %csr2 = tensor.insert %f2 into %csr1[%c2, %c2] : tensor<4x3xf64, #CSR> 119 %csr3 = tensor.insert %f3 into %csr2[%c3, %c0] : tensor<4x3xf64, #CSR> 120 %csr4 = tensor.insert %f4 into %csr3[%c3, %c2] : tensor<4x3xf64, #CSR> 121 %csrm = sparse_tensor.load %csr4 hasInserts : tensor<4x3xf64, #CSR> 122 sparse_tensor.print %csrm : tensor<4x3xf64, #CSR> 123 124 // 125 // DCSR case. 126 // 127 // CHECK-NEXT: ---- Sparse Tensor ---- 128 // CHECK-NEXT: nse = 4 129 // CHECK-NEXT: dim = ( 4, 3 ) 130 // CHECK-NEXT: lvl = ( 4, 3 ) 131 // CHECK-NEXT: pos[0] : ( 0, 3 ) 132 // CHECK-NEXT: crd[0] : ( 0, 2, 3 ) 133 // CHECK-NEXT: pos[1] : ( 0, 1, 2, 4 ) 134 // CHECK-NEXT: crd[1] : ( 0, 2, 0, 2 ) 135 // CHECK-NEXT: values : ( 1, 2, 3, 4 ) 136 // CHECK-NEXT: ---- 137 // 138 %dcsra = tensor.empty() : tensor<4x3xf64, #DCSR> 139 %dcsr1 = tensor.insert %f1 into %dcsra[%c0, %c0] : tensor<4x3xf64, #DCSR> 140 %dcsr2 = tensor.insert %f2 into %dcsr1[%c2, %c2] : tensor<4x3xf64, #DCSR> 141 %dcsr3 = tensor.insert %f3 into %dcsr2[%c3, %c0] : tensor<4x3xf64, #DCSR> 142 %dcsr4 = tensor.insert %f4 into %dcsr3[%c3, %c2] : tensor<4x3xf64, #DCSR> 143 %dcsrm = sparse_tensor.load %dcsr4 hasInserts : tensor<4x3xf64, #DCSR> 144 sparse_tensor.print %dcsrm : tensor<4x3xf64, #DCSR> 145 146 // 147 // Row case. 148 // 149 // CHECK-NEXT: ---- Sparse Tensor ---- 150 // CHECK-NEXT: nse = 9 151 // CHECK-NEXT: dim = ( 4, 3 ) 152 // CHECK-NEXT: lvl = ( 4, 3 ) 153 // CHECK-NEXT: pos[0] : ( 0, 3 ) 154 // CHECK-NEXT: crd[0] : ( 0, 2, 3 ) 155 // CHECK-NEXT: values : ( 1, 0, 0, 0, 0, 2, 3, 0, 4 ) 156 // CHECK-NEXT: ---- 157 // 158 %rowa = tensor.empty() : tensor<4x3xf64, #Row> 159 %row1 = tensor.insert %f1 into %rowa[%c0, %c0] : tensor<4x3xf64, #Row> 160 %row2 = tensor.insert %f2 into %row1[%c2, %c2] : tensor<4x3xf64, #Row> 161 %row3 = tensor.insert %f3 into %row2[%c3, %c0] : tensor<4x3xf64, #Row> 162 %row4 = tensor.insert %f4 into %row3[%c3, %c2] : tensor<4x3xf64, #Row> 163 %rowm = sparse_tensor.load %row4 hasInserts : tensor<4x3xf64, #Row> 164 sparse_tensor.print %rowm : tensor<4x3xf64, #Row> 165 166 // Release resources. 167 bufferization.dealloc_tensor %densem : tensor<4x3xf64, #Dense> 168 bufferization.dealloc_tensor %coom : tensor<4x3xf64, #SortedCOO> 169 bufferization.dealloc_tensor %csrm : tensor<4x3xf64, #CSR> 170 bufferization.dealloc_tensor %dcsrm : tensor<4x3xf64, #DCSR> 171 bufferization.dealloc_tensor %rowm : tensor<4x3xf64, #Row> 172 173 return 174 } 175} 176