253 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#Tensor1 = #sparse_tensor.encoding<{35 map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : compressed)36}>37 38#Tensor2 = #sparse_tensor.encoding<{39 map = (d0, d1, d2) -> (d1 : compressed, d2 : compressed, d0 : compressed),40}>41 42#Tensor3 = #sparse_tensor.encoding<{43 map = (d0, d1, d2) -> (d2 : compressed, d0 : compressed, d1 : compressed),44}>45 46#Tensor4 = #sparse_tensor.encoding<{47 map = (d0, d1, d2) -> (d0 : dense, d1 : compressed, d2 : compressed)48}>49 50#Tensor5 = #sparse_tensor.encoding<{51 map = (d0, d1, d2) -> (d1 : dense, d2 : compressed, d0 : compressed)52}>53 54#Tensor6 = #sparse_tensor.encoding<{55 map = (d0, d1, d2) -> (d2 : dense, d0 : compressed, d1 : compressed)56}>57 58//59// Integration test that tests conversions from sparse to dense tensors.60//61module {62 //63 // Utilities for output.64 //65 func.func @dump_234(%arg0: tensor<2x3x4xf64>) {66 %c0 = arith.constant 0 : index67 %d0 = arith.constant -1.0 : f6468 %0 = vector.transfer_read %arg0[%c0, %c0, %c0], %d0: tensor<2x3x4xf64>, vector<2x3x4xf64>69 vector.print %0 : vector<2x3x4xf64>70 return71 }72 func.func @dump_p34(%arg0: tensor<?x3x4xf64>) {73 %0 = tensor.cast %arg0 : tensor<?x3x4xf64> to tensor<2x3x4xf64>74 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()75 return76 }77 func.func @dump_2p4(%arg0: tensor<2x?x4xf64>) {78 %0 = tensor.cast %arg0 : tensor<2x?x4xf64> to tensor<2x3x4xf64>79 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()80 return81 }82 func.func @dump_23p(%arg0: tensor<2x3x?xf64>) {83 %0 = tensor.cast %arg0 : tensor<2x3x?xf64> to tensor<2x3x4xf64>84 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()85 return86 }87 func.func @dump_2pp(%arg0: tensor<2x?x?xf64>) {88 %0 = tensor.cast %arg0 : tensor<2x?x?xf64> to tensor<2x3x4xf64>89 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()90 return91 }92 func.func @dump_p3p(%arg0: tensor<?x3x?xf64>) {93 %0 = tensor.cast %arg0 : tensor<?x3x?xf64> to tensor<2x3x4xf64>94 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()95 return96 }97 func.func @dump_pp4(%arg0: tensor<?x?x4xf64>) {98 %0 = tensor.cast %arg0 : tensor<?x?x4xf64> to tensor<2x3x4xf64>99 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()100 return101 }102 func.func @dump_ppp(%arg0: tensor<?x?x?xf64>) {103 %0 = tensor.cast %arg0 : tensor<?x?x?xf64> to tensor<2x3x4xf64>104 call @dump_234(%0) : (tensor<2x3x4xf64>) -> ()105 return106 }107 108 //109 // Main driver.110 //111 func.func @main() {112 //113 // Initialize a 3-dim dense tensor.114 //115 %src = arith.constant dense<[116 [ [ 1.0, 2.0, 3.0, 4.0 ],117 [ 5.0, 6.0, 7.0, 8.0 ],118 [ 9.0, 10.0, 11.0, 12.0 ] ],119 [ [ 13.0, 14.0, 15.0, 16.0 ],120 [ 17.0, 18.0, 19.0, 20.0 ],121 [ 21.0, 22.0, 23.0, 24.0 ] ]122 ]> : tensor<2x3x4xf64>123 124 //125 // Convert dense tensor directly to various sparse tensors.126 //127 %s2341 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor1>128 %s2342 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor2>129 %s2343 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor3>130 %s2344 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor4>131 %s2345 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor5>132 %s2346 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x4xf64, #Tensor6>133 134 // NOTE: We commented out most cases with the dynamic-sized output tensor because finishing135 // all of them are currently taking too long, and they are not covering too many new code136 // paths.137 %sp344 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<?x3x4xf64, #Tensor4>138 // %sp345 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<?x3x4xf64, #Tensor5>139 // %sp346 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<?x3x4xf64, #Tensor6>140 // %s2p44 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x4xf64, #Tensor4>141 %s2p45 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x4xf64, #Tensor5>142 // %s2p46 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x4xf64, #Tensor6>143 // %s23p4 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x?xf64, #Tensor4>144 // %s23p5 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x?xf64, #Tensor5>145 %s23p6 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x3x?xf64, #Tensor6>146 // %s2pp4 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x?xf64, #Tensor4>147 // %s2pp5 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x?xf64, #Tensor5>148 // %s2pp6 = sparse_tensor.convert %src : tensor<2x3x4xf64> to tensor<2x?x?xf64, #Tensor6>149 150 //151 // Convert sparse tensor back to dense.152 //153 %d2341 = sparse_tensor.convert %s2341 : tensor<2x3x4xf64, #Tensor1> to tensor<2x3x4xf64>154 %d2342 = sparse_tensor.convert %s2342 : tensor<2x3x4xf64, #Tensor2> to tensor<2x3x4xf64>155 %d2343 = sparse_tensor.convert %s2343 : tensor<2x3x4xf64, #Tensor3> to tensor<2x3x4xf64>156 %d2344 = sparse_tensor.convert %s2344 : tensor<2x3x4xf64, #Tensor4> to tensor<2x3x4xf64>157 %d2345 = sparse_tensor.convert %s2345 : tensor<2x3x4xf64, #Tensor5> to tensor<2x3x4xf64>158 %d2346 = sparse_tensor.convert %s2346 : tensor<2x3x4xf64, #Tensor6> to tensor<2x3x4xf64>159 160 %dp344 = sparse_tensor.convert %sp344 : tensor<?x3x4xf64, #Tensor4> to tensor<?x3x4xf64>161 // %dp345 = sparse_tensor.convert %sp345 : tensor<?x3x4xf64, #Tensor5> to tensor<?x3x4xf64>162 // %dp346 = sparse_tensor.convert %sp346 : tensor<?x3x4xf64, #Tensor6> to tensor<?x3x4xf64>163 // %d2p44 = sparse_tensor.convert %s2p44 : tensor<2x?x4xf64, #Tensor4> to tensor<2x?x4xf64>164 %d2p45 = sparse_tensor.convert %s2p45 : tensor<2x?x4xf64, #Tensor5> to tensor<2x?x4xf64>165 // %d2p46 = sparse_tensor.convert %s2p46 : tensor<2x?x4xf64, #Tensor6> to tensor<2x?x4xf64>166 // %d23p4 = sparse_tensor.convert %s23p4 : tensor<2x3x?xf64, #Tensor4> to tensor<2x3x?xf64>167 // %d23p5 = sparse_tensor.convert %s23p5 : tensor<2x3x?xf64, #Tensor5> to tensor<2x3x?xf64>168 %d23p6 = sparse_tensor.convert %s23p6 : tensor<2x3x?xf64, #Tensor6> to tensor<2x3x?xf64>169 // %d2pp4 = sparse_tensor.convert %s2pp4 : tensor<2x?x?xf64, #Tensor4> to tensor<2x?x?xf64>170 // %d2pp5 = sparse_tensor.convert %s2pp5 : tensor<2x?x?xf64, #Tensor5> to tensor<2x?x?xf64>171 // %d2pp6 = sparse_tensor.convert %s2pp6 : tensor<2x?x?xf64, #Tensor6> to tensor<2x?x?xf64>172 173 %dp3p4 = sparse_tensor.convert %sp344 : tensor<?x3x4xf64, #Tensor4> to tensor<?x3x?xf64>174 // %dp3p5 = sparse_tensor.convert %sp345 : tensor<?x3x4xf64, #Tensor5> to tensor<?x3x?xf64>175 // %dp3p6 = sparse_tensor.convert %sp346 : tensor<?x3x4xf64, #Tensor6> to tensor<?x3x?xf64>176 // %dpp44 = sparse_tensor.convert %s2p44 : tensor<2x?x4xf64, #Tensor4> to tensor<?x?x4xf64>177 %dpp45 = sparse_tensor.convert %s2p45 : tensor<2x?x4xf64, #Tensor5> to tensor<?x?x4xf64>178 // %dpp46 = sparse_tensor.convert %s2p46 : tensor<2x?x4xf64, #Tensor6> to tensor<?x?x4xf64>179 // %dppp4 = sparse_tensor.convert %s2pp4 : tensor<2x?x?xf64, #Tensor4> to tensor<?x?x?xf64>180 // %dppp5 = sparse_tensor.convert %s2pp5 : tensor<2x?x?xf64, #Tensor5> to tensor<?x?x?xf64>181 // %dppp6 = sparse_tensor.convert %s2pp6 : tensor<2x?x?xf64, #Tensor6> to tensor<?x?x?xf64>182 183 //184 // Check round-trip equality. And release dense tensors.185 // CHECK-COUNT-12: ( ( ( 1, 2, 3, 4 ), ( 5, 6, 7, 8 ), ( 9, 10, 11, 12 ) ), ( ( 13, 14, 15, 16 ), ( 17, 18, 19, 20 ), ( 21, 22, 23, 24 ) ) )186 // was COUNT-28 before.187 call @dump_234(%src) : (tensor<2x3x4xf64>) -> ()188 call @dump_234(%d2341) : (tensor<2x3x4xf64>) -> ()189 call @dump_234(%d2342) : (tensor<2x3x4xf64>) -> ()190 call @dump_234(%d2343) : (tensor<2x3x4xf64>) -> ()191 call @dump_234(%d2344) : (tensor<2x3x4xf64>) -> ()192 call @dump_234(%d2345) : (tensor<2x3x4xf64>) -> ()193 call @dump_234(%d2346) : (tensor<2x3x4xf64>) -> ()194 call @dump_p34(%dp344) : (tensor<?x3x4xf64>) -> ()195 // call @dump_p34(%dp345) : (tensor<?x3x4xf64>) -> ()196 // call @dump_p34(%dp346) : (tensor<?x3x4xf64>) -> ()197 // call @dump_2p4(%d2p44) : (tensor<2x?x4xf64>) -> ()198 call @dump_2p4(%d2p45) : (tensor<2x?x4xf64>) -> ()199 // call @dump_2p4(%d2p46) : (tensor<2x?x4xf64>) -> ()200 // call @dump_23p(%d23p4) : (tensor<2x3x?xf64>) -> ()201 // call @dump_23p(%d23p5) : (tensor<2x3x?xf64>) -> ()202 call @dump_23p(%d23p6) : (tensor<2x3x?xf64>) -> ()203 // call @dump_2pp(%d2pp4) : (tensor<2x?x?xf64>) -> ()204 // call @dump_2pp(%d2pp5) : (tensor<2x?x?xf64>) -> ()205 // call @dump_2pp(%d2pp6) : (tensor<2x?x?xf64>) -> ()206 call @dump_p3p(%dp3p4) : (tensor<?x3x?xf64>) -> ()207 // call @dump_p3p(%dp3p5) : (tensor<?x3x?xf64>) -> ()208 // call @dump_p3p(%dp3p6) : (tensor<?x3x?xf64>) -> ()209 // call @dump_pp4(%dpp44) : (tensor<?x?x4xf64>) -> ()210 call @dump_pp4(%dpp45) : (tensor<?x?x4xf64>) -> ()211 // call @dump_pp4(%dpp46) : (tensor<?x?x4xf64>) -> ()212 // call @dump_ppp(%dppp4) : (tensor<?x?x?xf64>) -> ()213 // call @dump_ppp(%dppp5) : (tensor<?x?x?xf64>) -> ()214 // call @dump_ppp(%dppp6) : (tensor<?x?x?xf64>) -> ()215 216 //217 // Release sparse tensors.218 //219 bufferization.dealloc_tensor %s2341 : tensor<2x3x4xf64, #Tensor1>220 bufferization.dealloc_tensor %s2342 : tensor<2x3x4xf64, #Tensor2>221 bufferization.dealloc_tensor %s2343 : tensor<2x3x4xf64, #Tensor3>222 bufferization.dealloc_tensor %s2344 : tensor<2x3x4xf64, #Tensor4>223 bufferization.dealloc_tensor %s2345 : tensor<2x3x4xf64, #Tensor5>224 bufferization.dealloc_tensor %s2346 : tensor<2x3x4xf64, #Tensor6>225 bufferization.dealloc_tensor %sp344 : tensor<?x3x4xf64, #Tensor4>226 // bufferization.dealloc_tensor %sp345 : tensor<?x3x4xf64, #Tensor5>227 // bufferization.dealloc_tensor %sp346 : tensor<?x3x4xf64, #Tensor6>228 // bufferization.dealloc_tensor %s2p44 : tensor<2x?x4xf64, #Tensor4>229 bufferization.dealloc_tensor %s2p45 : tensor<2x?x4xf64, #Tensor5>230 // bufferization.dealloc_tensor %s2p46 : tensor<2x?x4xf64, #Tensor6>231 // bufferization.dealloc_tensor %s23p4 : tensor<2x3x?xf64, #Tensor4>232 // bufferization.dealloc_tensor %s23p5 : tensor<2x3x?xf64, #Tensor5>233 bufferization.dealloc_tensor %s23p6 : tensor<2x3x?xf64, #Tensor6>234 // bufferization.dealloc_tensor %s2pp4 : tensor<2x?x?xf64, #Tensor4>235 // bufferization.dealloc_tensor %s2pp5 : tensor<2x?x?xf64, #Tensor5>236 // bufferization.dealloc_tensor %s2pp6 : tensor<2x?x?xf64, #Tensor6>237 238 bufferization.dealloc_tensor %d2341 : tensor<2x3x4xf64>239 bufferization.dealloc_tensor %d2342 : tensor<2x3x4xf64>240 bufferization.dealloc_tensor %d2343 : tensor<2x3x4xf64>241 bufferization.dealloc_tensor %d2344 : tensor<2x3x4xf64>242 bufferization.dealloc_tensor %d2345 : tensor<2x3x4xf64>243 bufferization.dealloc_tensor %d2346 : tensor<2x3x4xf64>244 bufferization.dealloc_tensor %dp344 : tensor<?x3x4xf64>245 bufferization.dealloc_tensor %d2p45 : tensor<2x?x4xf64>246 bufferization.dealloc_tensor %d23p6 : tensor<2x3x?xf64>247 bufferization.dealloc_tensor %dp3p4 : tensor<?x3x?xf64>248 bufferization.dealloc_tensor %dpp45 : tensor<?x?x4xf64>249 250 return251 }252}253