301 lines · plain
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, this5// config could be moved to lit.local.cfg. However, there are downstream users that6// do not use these LIT config files. Hence why this is kept inline.7//8// DEFINE: %{sparsifier_opts} = enable-runtime-library=true9// 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_utils13// DEFINE: %{run_libs_sve} = -shared-libs=%native_mlir_runner_utils,%native_mlir_c_runner_utils14// DEFINE: %{run_opts} = -e main -entry-point-result=void15// 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// RUN: %{compile} | %{run} | FileCheck %s22//23// Do the same run, but now with direct IR generation.24// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false25// RUN: %{compile} | %{run} | FileCheck %s26//27// Do the same run, but now with direct IR generation and vectorization.28// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true29// RUN: %{compile} | %{run} | FileCheck %s30//31// Do the same run, but now with direct IR generation and VLA vectorization.32// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}33 34#SparseVector = #sparse_tensor.encoding<{35 map = (d0) -> (d0 : compressed)36}>37 38#SparseMatrix = #sparse_tensor.encoding<{39 map = (d0, d1) -> (d0 : compressed, d1 : compressed)40}>41 42#Sparse3dTensor = #sparse_tensor.encoding<{43 map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : compressed)44}>45 46#Sparse4dTensor = #sparse_tensor.encoding<{47 map = (d0, d1, d2, d3) -> (d0 : compressed, d1 : compressed, d2 : compressed, d3 : compressed)48}>49 50//51// Test with various forms of the two most elementary reshape52// operations: expand53//54module {55 56 func.func @expand_dense(%arg0: tensor<12xf64>) -> tensor<3x4xf64> {57 %0 = tensor.expand_shape %arg0 [[0, 1]] output_shape [3, 4] : tensor<12xf64> into tensor<3x4xf64>58 return %0 : tensor<3x4xf64>59 }60 61 func.func @expand_from_sparse(%arg0: tensor<12xf64, #SparseVector>) -> tensor<3x4xf64> {62 %0 = tensor.expand_shape %arg0 [[0, 1]] output_shape [3, 4] : tensor<12xf64, #SparseVector> into tensor<3x4xf64>63 return %0 : tensor<3x4xf64>64 }65 66 func.func @expand_to_sparse(%arg0: tensor<12xf64>) -> tensor<3x4xf64, #SparseMatrix> {67 %0 = tensor.expand_shape %arg0 [[0, 1]] output_shape [3, 4] : tensor<12xf64> into tensor<3x4xf64, #SparseMatrix>68 return %0 : tensor<3x4xf64, #SparseMatrix>69 }70 71 func.func @expand_sparse2sparse(%arg0: tensor<12xf64, #SparseVector>) -> tensor<3x4xf64, #SparseMatrix> {72 %0 = tensor.expand_shape %arg0 [[0, 1]] output_shape [3, 4] : tensor<12xf64, #SparseVector> into tensor<3x4xf64, #SparseMatrix>73 return %0 : tensor<3x4xf64, #SparseMatrix>74 }75 76 func.func @expand_dense_3x2x2(%arg0: tensor<3x4xf64>) -> tensor<3x2x2xf64> {77 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [3, 2, 2] : tensor<3x4xf64> into tensor<3x2x2xf64>78 return %0 : tensor<3x2x2xf64>79 }80 81 func.func @expand_from_sparse_3x2x2(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64> {82 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [3, 2, 2] : tensor<3x4xf64, #SparseMatrix> into tensor<3x2x2xf64>83 return %0 : tensor<3x2x2xf64>84 }85 86 func.func @expand_to_sparse_3x2x2(%arg0: tensor<3x4xf64>) -> tensor<3x2x2xf64, #Sparse3dTensor> {87 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [3, 2, 2] : tensor<3x4xf64> into tensor<3x2x2xf64, #Sparse3dTensor>88 return %0 : tensor<3x2x2xf64, #Sparse3dTensor>89 }90 91 func.func @expand_sparse2sparse_3x2x2(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64, #Sparse3dTensor> {92 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [3, 2, 2] : tensor<3x4xf64, #SparseMatrix> into tensor<3x2x2xf64, #Sparse3dTensor>93 return %0 : tensor<3x2x2xf64, #Sparse3dTensor>94 }95 96 func.func @expand_dense_dyn(%arg0: tensor<?x?xf64>) -> tensor<?x2x?xf64> {97 %c0 = arith.constant 0 : index98 %c1 = arith.constant 1 : index99 %c2 = arith.constant 2 : index100 %d0 = tensor.dim %arg0, %c0 : tensor<?x?xf64>101 %d1 = tensor.dim %arg0, %c1 : tensor<?x?xf64>102 %d2 = arith.divui %d1, %c2 : index103 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [%d0, 2, %d2] : tensor<?x?xf64> into tensor<?x2x?xf64>104 return %0 : tensor<?x2x?xf64>105 }106 107 func.func @expand_from_sparse_dyn(%arg0: tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64> {108 %c0 = arith.constant 0 : index109 %c1 = arith.constant 1 : index110 %c2 = arith.constant 2 : index111 %d0 = tensor.dim %arg0, %c0 : tensor<?x?xf64, #SparseMatrix>112 %d1 = tensor.dim %arg0, %c1 : tensor<?x?xf64, #SparseMatrix>113 %d2 = arith.divui %d1, %c2 : index114 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [%d0, 2, %d2] : tensor<?x?xf64, #SparseMatrix> into tensor<?x2x?xf64>115 return %0 : tensor<?x2x?xf64>116 }117 118 func.func @expand_to_sparse_dyn(%arg0: tensor<?x?xf64>) -> tensor<?x2x?xf64, #Sparse3dTensor> {119 %c0 = arith.constant 0 : index120 %c1 = arith.constant 1 : index121 %c2 = arith.constant 2 : index122 %d0 = tensor.dim %arg0, %c0 : tensor<?x?xf64>123 %d1 = tensor.dim %arg0, %c1 : tensor<?x?xf64>124 %d2 = arith.divui %d1, %c2 : index125 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [%d0, 2, %d2] : tensor<?x?xf64> into tensor<?x2x?xf64, #Sparse3dTensor>126 return %0 : tensor<?x2x?xf64, #Sparse3dTensor>127 }128 129 func.func @expand_sparse2sparse_dyn(%arg0: tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64, #Sparse3dTensor> {130 %c0 = arith.constant 0 : index131 %c1 = arith.constant 1 : index132 %c2 = arith.constant 2 : index133 %d0 = tensor.dim %arg0, %c0 : tensor<?x?xf64, #SparseMatrix>134 %d1 = tensor.dim %arg0, %c1 : tensor<?x?xf64, #SparseMatrix>135 %d2 = arith.divui %d1, %c2 : index136 %0 = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [%d0, 2, %d2] : tensor<?x?xf64, #SparseMatrix> into tensor<?x2x?xf64, #Sparse3dTensor>137 return %0 : tensor<?x2x?xf64, #Sparse3dTensor>138 }139 140 //141 // Main driver.142 //143 func.func @main() {144 %c0 = arith.constant 0 : index145 %df = arith.constant -1.0 : f64146 147 // Setup test vectors and matrices..148 %v = arith.constant dense <[ 1.0, 0.0, 3.0, 0.0, 5.0, 0.0,149 7.0, 0.0, 9.0, 0.0, 11.0, 0.0]> : tensor<12xf64>150 %m = arith.constant dense <[ [ 1.1, 1.2, 1.3, 1.4 ],151 [ 2.1, 2.2, 2.3, 2.4 ],152 [ 3.1, 3.2, 3.3, 3.4 ]]> : tensor<3x4xf64>153 154 %sv = sparse_tensor.convert %v : tensor<12xf64> to tensor<12xf64, #SparseVector>155 %sm = sparse_tensor.convert %m : tensor<3x4xf64> to tensor<3x4xf64, #SparseMatrix>156 157 %dm = tensor.cast %m : tensor<3x4xf64> to tensor<?x?xf64>158 %sdm = sparse_tensor.convert %dm : tensor<?x?xf64> to tensor<?x?xf64, #SparseMatrix>159 160 // Call the kernels.161 %expand0 = call @expand_dense(%v) : (tensor<12xf64>) -> tensor<3x4xf64>162 %expand1 = call @expand_from_sparse(%sv) : (tensor<12xf64, #SparseVector>) -> tensor<3x4xf64>163 %expand2 = call @expand_to_sparse(%v) : (tensor<12xf64>) -> tensor<3x4xf64, #SparseMatrix>164 %expand3 = call @expand_sparse2sparse(%sv) : (tensor<12xf64, #SparseVector>) -> tensor<3x4xf64, #SparseMatrix>165 %expand4 = call @expand_dense_3x2x2(%m) : (tensor<3x4xf64>) -> tensor<3x2x2xf64>166 %expand5 = call @expand_from_sparse_3x2x2(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64>167 %expand6 = call @expand_to_sparse_3x2x2(%m) : (tensor<3x4xf64>) -> tensor<3x2x2xf64, #Sparse3dTensor>168 %expand7 = call @expand_sparse2sparse_3x2x2(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64, #Sparse3dTensor>169 %expand8 = call @expand_dense_dyn(%dm) : (tensor<?x?xf64>) -> tensor<?x2x?xf64>170 %expand9 = call @expand_from_sparse_dyn(%sdm) : (tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64>171 %expand10 = call @expand_to_sparse_dyn(%dm) : (tensor<?x?xf64>) -> tensor<?x2x?xf64, #Sparse3dTensor>172 %expand11 = call @expand_sparse2sparse_dyn(%sdm) : (tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64, #Sparse3dTensor>173 174 //175 // Verify results of expand with dense output.176 //177 // CHECK: ( ( 1, 0, 3, 0 ), ( 5, 0, 7, 0 ), ( 9, 0, 11, 0 ) )178 // CHECK-NEXT: ( ( 1, 0, 3, 0 ), ( 5, 0, 7, 0 ), ( 9, 0, 11, 0 ) )179 // CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )180 // CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )181 // CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )182 // CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )183 //184 %m0 = vector.transfer_read %expand0[%c0, %c0], %df: tensor<3x4xf64>, vector<3x4xf64>185 vector.print %m0 : vector<3x4xf64>186 %m1 = vector.transfer_read %expand1[%c0, %c0], %df: tensor<3x4xf64>, vector<3x4xf64>187 vector.print %m1 : vector<3x4xf64>188 %m4 = vector.transfer_read %expand4[%c0, %c0, %c0], %df: tensor<3x2x2xf64>, vector<3x2x2xf64>189 vector.print %m4 : vector<3x2x2xf64>190 %m5 = vector.transfer_read %expand5[%c0, %c0, %c0], %df: tensor<3x2x2xf64>, vector<3x2x2xf64>191 vector.print %m5 : vector<3x2x2xf64>192 %m8 = vector.transfer_read %expand8[%c0, %c0, %c0], %df: tensor<?x2x?xf64>, vector<3x2x2xf64>193 vector.print %m8 : vector<3x2x2xf64>194 %m9 = vector.transfer_read %expand9[%c0, %c0, %c0], %df: tensor<?x2x?xf64>, vector<3x2x2xf64>195 vector.print %m9 : vector<3x2x2xf64>196 197 //198 // Verify results of expand with sparse output.199 //200 // CHECK: ---- Sparse Tensor ----201 // CHECK-NEXT: nse = 6202 // CHECK-NEXT: dim = ( 3, 4 )203 // CHECK-NEXT: lvl = ( 3, 4 )204 // CHECK-NEXT: pos[0] : ( 0, 3 )205 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )206 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )207 // CHECK-NEXT: crd[1] : ( 0, 2, 0, 2, 0, 2 )208 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 11 )209 // CHECK-NEXT: ----210 //211 // CHECK: ---- Sparse Tensor ----212 // CHECK-NEXT: nse = 6213 // CHECK-NEXT: dim = ( 3, 4 )214 // CHECK-NEXT: lvl = ( 3, 4 )215 // CHECK-NEXT: pos[0] : ( 0, 3 )216 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )217 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )218 // CHECK-NEXT: crd[1] : ( 0, 2, 0, 2, 0, 2 )219 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 11 )220 // CHECK-NEXT: ----221 //222 // CHECK: ---- Sparse Tensor ----223 // CHECK-NEXT: nse = 12224 // CHECK-NEXT: dim = ( 3, 2, 2 )225 // CHECK-NEXT: lvl = ( 3, 2, 2 )226 // CHECK-NEXT: pos[0] : ( 0, 3 )227 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )228 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )229 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 1, 0, 1 )230 // CHECK-NEXT: pos[2] : ( 0, 2, 4, 6, 8, 10, 12 )231 // CHECK-NEXT: crd[2] : ( 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 )232 // CHECK-NEXT: values : ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )233 // CHECK-NEXT: ----234 //235 // CHECK: ---- Sparse Tensor ----236 // CHECK-NEXT: nse = 12237 // CHECK-NEXT: dim = ( 3, 2, 2 )238 // CHECK-NEXT: lvl = ( 3, 2, 2 )239 // CHECK-NEXT: pos[0] : ( 0, 3 )240 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )241 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )242 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 1, 0, 1 )243 // CHECK-NEXT: pos[2] : ( 0, 2, 4, 6, 8, 10, 12 )244 // CHECK-NEXT: crd[2] : ( 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 )245 // CHECK-NEXT: values : ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )246 // CHECK-NEXT: ----247 //248 // CHECK: ---- Sparse Tensor ----249 // CHECK-NEXT: nse = 12250 // CHECK-NEXT: dim = ( 3, 2, 2 )251 // CHECK-NEXT: lvl = ( 3, 2, 2 )252 // CHECK-NEXT: pos[0] : ( 0, 3 )253 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )254 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )255 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 1, 0, 1 )256 // CHECK-NEXT: pos[2] : ( 0, 2, 4, 6, 8, 10, 12 )257 // CHECK-NEXT: crd[2] : ( 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 )258 // CHECK-NEXT: values : ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )259 // CHECK-NEXT: ----260 //261 // CHECK: ---- Sparse Tensor ----262 // CHECK-NEXT: nse = 12263 // CHECK-NEXT: dim = ( 3, 2, 2 )264 // CHECK-NEXT: lvl = ( 3, 2, 2 )265 // CHECK-NEXT: pos[0] : ( 0, 3 )266 // CHECK-NEXT: crd[0] : ( 0, 1, 2 )267 // CHECK-NEXT: pos[1] : ( 0, 2, 4, 6 )268 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 1, 0, 1 )269 // CHECK-NEXT: pos[2] : ( 0, 2, 4, 6, 8, 10, 12 )270 // CHECK-NEXT: crd[2] : ( 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 )271 // CHECK-NEXT: values : ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )272 // CHECK-NEXT: ----273 //274 sparse_tensor.print %expand2 : tensor<3x4xf64, #SparseMatrix>275 sparse_tensor.print %expand3 : tensor<3x4xf64, #SparseMatrix>276 sparse_tensor.print %expand6 : tensor<3x2x2xf64, #Sparse3dTensor>277 sparse_tensor.print %expand7 : tensor<3x2x2xf64, #Sparse3dTensor>278 sparse_tensor.print %expand10 : tensor<?x2x?xf64, #Sparse3dTensor>279 sparse_tensor.print %expand11 : tensor<?x2x?xf64, #Sparse3dTensor>280 281 282 // Release sparse resources.283 bufferization.dealloc_tensor %sv : tensor<12xf64, #SparseVector>284 bufferization.dealloc_tensor %sm : tensor<3x4xf64, #SparseMatrix>285 bufferization.dealloc_tensor %sdm : tensor<?x?xf64, #SparseMatrix>286 bufferization.dealloc_tensor %expand2 : tensor<3x4xf64, #SparseMatrix>287 bufferization.dealloc_tensor %expand3 : tensor<3x4xf64, #SparseMatrix>288 bufferization.dealloc_tensor %expand6 : tensor<3x2x2xf64, #Sparse3dTensor>289 bufferization.dealloc_tensor %expand7 : tensor<3x2x2xf64, #Sparse3dTensor>290 bufferization.dealloc_tensor %expand10 : tensor<?x2x?xf64, #Sparse3dTensor>291 bufferization.dealloc_tensor %expand11 : tensor<?x2x?xf64, #Sparse3dTensor>292 293 // Release dense resources.294 bufferization.dealloc_tensor %expand1 : tensor<3x4xf64>295 bufferization.dealloc_tensor %expand5 : tensor<3x2x2xf64>296 bufferization.dealloc_tensor %expand9 : tensor<?x2x?xf64>297 298 return299 }300}301