270 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=false enable-buffer-initialization=true25// 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: collapse.53//54module {55 56 func.func @collapse_dense(%arg0: tensor<3x4xf64>) -> tensor<12xf64> {57 %0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64> into tensor<12xf64>58 return %0 : tensor<12xf64>59 }60 61 func.func @collapse_from_sparse(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64> {62 %0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64, #SparseMatrix> into tensor<12xf64>63 return %0 : tensor<12xf64>64 }65 66 func.func @collapse_to_sparse(%arg0: tensor<3x4xf64>) -> tensor<12xf64, #SparseVector> {67 %0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64> into tensor<12xf64, #SparseVector>68 return %0 : tensor<12xf64, #SparseVector>69 }70 71 func.func @collapse_sparse2sparse(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64, #SparseVector> {72 %0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64, #SparseMatrix> into tensor<12xf64, #SparseVector>73 return %0 : tensor<12xf64, #SparseVector>74 }75 76 func.func @collapse_dense_6x10(%arg0: tensor<2x3x5x2xf64>) -> tensor<6x10xf64> {77 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64> into tensor<6x10xf64>78 return %0 : tensor<6x10xf64>79 }80 81 func.func @collapse_from_sparse_6x10(%arg0: tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64> {82 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64, #Sparse4dTensor> into tensor<6x10xf64>83 return %0 : tensor<6x10xf64>84 }85 86 func.func @collapse_to_sparse_6x10(%arg0: tensor<2x3x5x2xf64>) -> tensor<6x10xf64, #SparseMatrix> {87 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64> into tensor<6x10xf64, #SparseMatrix>88 return %0 : tensor<6x10xf64, #SparseMatrix>89 }90 91 func.func @collapse_sparse2sparse_6x10(%arg0: tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64, #SparseMatrix> {92 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64, #Sparse4dTensor> into tensor<6x10xf64, #SparseMatrix>93 return %0 : tensor<6x10xf64, #SparseMatrix>94 }95 96 func.func @collapse_dense_dyn(%arg0: tensor<?x?x?x?xf64>) -> tensor<?x?xf64> {97 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64> into tensor<?x?xf64>98 return %0 : tensor<?x?xf64>99 }100 101 func.func @collapse_from_sparse_dyn(%arg0: tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64> {102 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64, #Sparse4dTensor> into tensor<?x?xf64>103 return %0 : tensor<?x?xf64>104 }105 106 func.func @collapse_to_sparse_dyn(%arg0: tensor<?x?x?x?xf64>) -> tensor<?x?xf64, #SparseMatrix> {107 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64> into tensor<?x?xf64, #SparseMatrix>108 return %0 : tensor<?x?xf64, #SparseMatrix>109 }110 111 func.func @collapse_sparse2sparse_dyn(%arg0: tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64, #SparseMatrix> {112 %0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64, #Sparse4dTensor> into tensor<?x?xf64, #SparseMatrix>113 return %0 : tensor<?x?xf64, #SparseMatrix>114 }115 116 //117 // Main driver.118 //119 func.func @main() {120 %c0 = arith.constant 0 : index121 %df = arith.constant -1.0 : f64122 123 // Setup test vectors and matrices..124 %m = arith.constant dense <[ [ 1.1, 0.0, 1.3, 0.0 ],125 [ 2.1, 0.0, 2.3, 0.0 ],126 [ 3.1, 0.0, 3.3, 0.0 ]]> : tensor<3x4xf64>127 %n = arith.constant dense <[128 [ [[ 1.0, 0.0], [ 3.0, 0.0], [ 5.0, 0.0], [ 7.0, 0.0], [ 9.0, 0.0]],129 [[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]],130 [[21.0, 0.0], [23.0, 0.0], [25.0, 0.0], [27.0, 0.0], [29.0, 0.0]] ],131 [ [[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]],132 [[41.0, 0.0], [43.0, 0.0], [45.0, 0.0], [47.0, 0.0], [49.0, 0.0]],133 [[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]] ] ]> : tensor<2x3x5x2xf64>134 %sm = sparse_tensor.convert %m : tensor<3x4xf64> to tensor<3x4xf64, #SparseMatrix>135 %sn = sparse_tensor.convert %n : tensor<2x3x5x2xf64> to tensor<2x3x5x2xf64, #Sparse4dTensor>136 137 %dm = tensor.cast %m : tensor<3x4xf64> to tensor<?x?xf64>138 139 %dn = tensor.cast %n : tensor<2x3x5x2xf64> to tensor<?x?x?x?xf64>140 %sdn = sparse_tensor.convert %dn : tensor<?x?x?x?xf64> to tensor<?x?x?x?xf64, #Sparse4dTensor>141 142 // Call the kernels.143 %collapse0 = call @collapse_dense(%m) : (tensor<3x4xf64>) -> tensor<12xf64>144 %collapse1 = call @collapse_from_sparse(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64>145 %collapse2 = call @collapse_to_sparse(%m) : (tensor<3x4xf64>) -> tensor<12xf64, #SparseVector>146 %collapse3 = call @collapse_sparse2sparse(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64, #SparseVector>147 %collapse4 = call @collapse_dense_6x10(%n) : (tensor<2x3x5x2xf64>) -> tensor<6x10xf64>148 %collapse5 = call @collapse_from_sparse_6x10(%sn) : (tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64>149 %collapse6 = call @collapse_to_sparse_6x10(%n) : (tensor<2x3x5x2xf64>) -> tensor<6x10xf64, #SparseMatrix>150 %collapse7 = call @collapse_sparse2sparse_6x10(%sn) : (tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64, #SparseMatrix>151 %collapse8 = call @collapse_dense_dyn(%dn) : (tensor<?x?x?x?xf64>) -> tensor<?x?xf64>152 %collapse9 = call @collapse_from_sparse_dyn(%sdn) : (tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64>153 %collapse10 = call @collapse_to_sparse_dyn(%dn) : (tensor<?x?x?x?xf64>) -> tensor<?x?xf64, #SparseMatrix>154 %collapse11 = call @collapse_sparse2sparse_dyn(%sdn) : (tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64, #SparseMatrix>155 156 //157 // Verify results of collapse158 //159 // CHECK: ( 1.1, 0, 1.3, 0, 2.1, 0, 2.3, 0, 3.1, 0, 3.3, 0 )160 // CHECK-NEXT: ( 1.1, 0, 1.3, 0, 2.1, 0, 2.3, 0, 3.1, 0, 3.3, 0 )161 //162 // CHECK: ---- Sparse Tensor ----163 // CHECK-NEXT: nse = 6164 // CHECK-NEXT: dim = ( 12 )165 // CHECK-NEXT: lvl = ( 12 )166 // CHECK-NEXT: pos[0] : ( 0, 6 )167 // CHECK-NEXT: crd[0] : ( 0, 2, 4, 6, 8, 10 )168 // CHECK-NEXT: values : ( 1.1, 1.3, 2.1, 2.3, 3.1, 3.3 )169 // CHECK-NEXT: ----170 //171 // CHECK: ---- Sparse Tensor ----172 // CHECK-NEXT: nse = 6173 // CHECK-NEXT: dim = ( 12 )174 // CHECK-NEXT: lvl = ( 12 )175 // CHECK-NEXT: pos[0] : ( 0, 6 )176 // CHECK-NEXT: crd[0] : ( 0, 2, 4, 6, 8, 10 )177 // CHECK-NEXT: values : ( 1.1, 1.3, 2.1, 2.3, 3.1, 3.3 )178 // CHECK-NEXT: ----179 //180 // CHECK: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )181 // CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )182 //183 // CHECK: ---- Sparse Tensor ----184 // CHECK-NEXT: nse = 15185 // CHECK-NEXT: dim = ( 6, 10 )186 // CHECK-NEXT: lvl = ( 6, 10 )187 // CHECK-NEXT: pos[0] : ( 0, 3 )188 // CHECK-NEXT: crd[0] : ( 0, 2, 4 )189 // CHECK-NEXT: pos[1] : ( 0, 5, 10, 15 )190 // CHECK-NEXT: crd[1] : ( 0, 2, 4, 6, 8, 0, 2, 4, 6, 8, 0, 2, 4, 6, 8 )191 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49 )192 // CHECK-NEXT: ----193 //194 // CHECK: ---- Sparse Tensor ----195 // CHECK-NEXT: nse = 15196 // CHECK-NEXT: dim = ( 6, 10 )197 // CHECK-NEXT: lvl = ( 6, 10 )198 // CHECK-NEXT: pos[0] : ( 0, 3 )199 // CHECK-NEXT: crd[0] : ( 0, 2, 4 )200 // CHECK-NEXT: pos[1] : ( 0, 5, 10, 15 )201 // CHECK-NEXT: crd[1] : ( 0, 2, 4, 6, 8, 0, 2, 4, 6, 8, 0, 2, 4, 6, 8 )202 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49 )203 // CHECK-NEXT: ----204 //205 // CHECK: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )206 // CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ), ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ), ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )207 //208 // CHECK: ---- Sparse Tensor ----209 // CHECK-NEXT: nse = 15210 // CHECK-NEXT: dim = ( 6, 10 )211 // CHECK-NEXT: lvl = ( 6, 10 )212 // CHECK-NEXT: pos[0] : ( 0, 3 )213 // CHECK-NEXT: crd[0] : ( 0, 2, 4 )214 // CHECK-NEXT: pos[1] : ( 0, 5, 10, 15 )215 // CHECK-NEXT: crd[1] : ( 0, 2, 4, 6, 8, 0, 2, 4, 6, 8, 0, 2, 4, 6, 8 )216 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49 )217 // CHECK-NEXT: ----218 //219 // CHECK: ---- Sparse Tensor ----220 // CHECK-NEXT: nse = 15221 // CHECK-NEXT: dim = ( 6, 10 )222 // CHECK-NEXT: lvl = ( 6, 10 )223 // CHECK-NEXT: pos[0] : ( 0, 3 )224 // CHECK-NEXT: crd[0] : ( 0, 2, 4 )225 // CHECK-NEXT: pos[1] : ( 0, 5, 10, 15 )226 // CHECK-NEXT: crd[1] : ( 0, 2, 4, 6, 8, 0, 2, 4, 6, 8, 0, 2, 4, 6, 8 )227 // CHECK-NEXT: values : ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49 )228 // CHECK-NEXT: ----229 //230 %v0 = vector.transfer_read %collapse0[%c0], %df: tensor<12xf64>, vector<12xf64>231 vector.print %v0 : vector<12xf64>232 %v1 = vector.transfer_read %collapse1[%c0], %df: tensor<12xf64>, vector<12xf64>233 vector.print %v1 : vector<12xf64>234 sparse_tensor.print %collapse2 : tensor<12xf64, #SparseVector>235 sparse_tensor.print %collapse3 : tensor<12xf64, #SparseVector>236 237 %v4 = vector.transfer_read %collapse4[%c0, %c0], %df: tensor<6x10xf64>, vector<6x10xf64>238 vector.print %v4 : vector<6x10xf64>239 %v5 = vector.transfer_read %collapse5[%c0, %c0], %df: tensor<6x10xf64>, vector<6x10xf64>240 vector.print %v5 : vector<6x10xf64>241 sparse_tensor.print %collapse6 : tensor<6x10xf64, #SparseMatrix>242 sparse_tensor.print %collapse7 : tensor<6x10xf64, #SparseMatrix>243 244 %v8 = vector.transfer_read %collapse8[%c0, %c0], %df: tensor<?x?xf64>, vector<6x10xf64>245 vector.print %v8 : vector<6x10xf64>246 %v9 = vector.transfer_read %collapse9[%c0, %c0], %df: tensor<?x?xf64>, vector<6x10xf64>247 vector.print %v9 : vector<6x10xf64>248 sparse_tensor.print %collapse10 : tensor<?x?xf64, #SparseMatrix>249 sparse_tensor.print %collapse11 : tensor<?x?xf64, #SparseMatrix>250 251 // Release sparse resources.252 bufferization.dealloc_tensor %sm : tensor<3x4xf64, #SparseMatrix>253 bufferization.dealloc_tensor %sn : tensor<2x3x5x2xf64, #Sparse4dTensor>254 bufferization.dealloc_tensor %sdn : tensor<?x?x?x?xf64, #Sparse4dTensor>255 bufferization.dealloc_tensor %collapse2 : tensor<12xf64, #SparseVector>256 bufferization.dealloc_tensor %collapse3 : tensor<12xf64, #SparseVector>257 bufferization.dealloc_tensor %collapse6 : tensor<6x10xf64, #SparseMatrix>258 bufferization.dealloc_tensor %collapse7 : tensor<6x10xf64, #SparseMatrix>259 bufferization.dealloc_tensor %collapse10 : tensor<?x?xf64, #SparseMatrix>260 bufferization.dealloc_tensor %collapse11 : tensor<?x?xf64, #SparseMatrix>261 262 // Release dense resources.263 bufferization.dealloc_tensor %collapse1 : tensor<12xf64>264 bufferization.dealloc_tensor %collapse5 : tensor<6x10xf64>265 bufferization.dealloc_tensor %collapse9: tensor<?x?xf64>266 267 return268 }269}270