183 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 enable-buffer-initialization=true 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#MAT_C_C = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : compressed, d1 : compressed)}>35#MAT_D_C = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : dense, d1 : compressed)}>36#MAT_C_D = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : compressed, d1 : dense)}>37#MAT_D_D = #sparse_tensor.encoding<{38 map = (d0, d1) -> (d1 : dense, d0 : dense)39}>40 41#MAT_C_C_P = #sparse_tensor.encoding<{42 map = (d0, d1) -> (d1 : compressed, d0 : compressed)43}>44 45#MAT_C_D_P = #sparse_tensor.encoding<{46 map = (d0, d1) -> (d1 : compressed, d0 : dense),47}>48 49#MAT_D_C_P = #sparse_tensor.encoding<{50 map = (d0, d1) -> (d1 : dense, d0 : compressed)51}>52 53module {54 func.func private @printMemrefF64(%ptr : tensor<*xf64>)55 56 // Concats all sparse matrices (with different encodings) to a sparse matrix.57 func.func @concat_sparse_sparse(%arg0: tensor<2x4xf64, #MAT_C_C>, %arg1: tensor<3x4xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64, #MAT_C_C> {58 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 0 : index}59 : tensor<2x4xf64, #MAT_C_C>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C> to tensor<9x4xf64, #MAT_C_C>60 return %0 : tensor<9x4xf64, #MAT_C_C>61 }62 63 // Concats all sparse matrices (with different encodings) to a dense matrix.64 func.func @concat_sparse_dense(%arg0: tensor<2x4xf64, #MAT_C_C>, %arg1: tensor<3x4xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64> {65 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 0 : index}66 : tensor<2x4xf64, #MAT_C_C>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C> to tensor<9x4xf64>67 return %0 : tensor<9x4xf64>68 }69 70 // Concats mix sparse and dense matrices to a sparse matrix.71 func.func @concat_mix_sparse(%arg0: tensor<2x4xf64>, %arg1: tensor<3x4xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64, #MAT_C_C> {72 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 0 : index}73 : tensor<2x4xf64>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C> to tensor<9x4xf64, #MAT_C_C>74 return %0 : tensor<9x4xf64, #MAT_C_C>75 }76 77 // Concats mix sparse and dense matrices to a dense matrix.78 func.func @concat_mix_dense(%arg0: tensor<2x4xf64>, %arg1: tensor<3x4xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64> {79 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 0 : index}80 : tensor<2x4xf64>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C> to tensor<9x4xf64>81 return %0 : tensor<9x4xf64>82 }83 84 // Outputs dense matrix.85 func.func @dump_mat_dense_9x4(%A: tensor<9x4xf64>) {86 %u = tensor.cast %A : tensor<9x4xf64> to tensor<*xf64>87 call @printMemrefF64(%u) : (tensor<*xf64>) -> ()88 return89 }90 91 // Driver method to call and verify kernels.92 func.func @main() {93 %m24 = arith.constant dense<94 [ [ 1.0, 0.0, 3.0, 0.0],95 [ 0.0, 2.0, 0.0, 0.0] ]> : tensor<2x4xf64>96 %m34 = arith.constant dense<97 [ [ 1.0, 0.0, 1.0, 1.0],98 [ 0.0, 0.5, 0.0, 0.0],99 [ 1.0, 5.0, 2.0, 0.0] ]> : tensor<3x4xf64>100 %m44 = arith.constant dense<101 [ [ 0.0, 0.0, 1.5, 1.0],102 [ 0.0, 3.5, 0.0, 0.0],103 [ 1.0, 5.0, 2.0, 0.0],104 [ 1.0, 0.5, 0.0, 0.0] ]> : tensor<4x4xf64>105 106 %sm24cc = sparse_tensor.convert %m24 : tensor<2x4xf64> to tensor<2x4xf64, #MAT_C_C>107 %sm34cd = sparse_tensor.convert %m34 : tensor<3x4xf64> to tensor<3x4xf64, #MAT_C_D>108 %sm44dc = sparse_tensor.convert %m44 : tensor<4x4xf64> to tensor<4x4xf64, #MAT_D_C>109 110 //111 // CHECK: ---- Sparse Tensor ----112 // CHECK-NEXT: nse = 18113 // CHECK-NEXT: dim = ( 9, 4 )114 // CHECK-NEXT: lvl = ( 9, 4 )115 // CHECK-NEXT: pos[0] : ( 0, 9 )116 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3, 4, 5, 6, 7, 8 )117 // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 7, 10, 12, 13, 16, 18 )118 // CHECK-NEXT: crd[1] : ( 0, 2, 1, 0, 2, 3, 1, 0, 1, 2, 2, 3, 1, 0, 1, 2, 0, 1 )119 // CHECK-NEXT: values : ( 1, 3, 2, 1, 1, 1, 0.5, 1, 5, 2, 1.5, 1, 3.5, 1, 5, 2, 1, 0.5 )120 // CHECK-NEXT: ----121 //122 %0 = call @concat_sparse_sparse(%sm24cc, %sm34cd, %sm44dc)123 : (tensor<2x4xf64, #MAT_C_C>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64, #MAT_C_C>124 sparse_tensor.print %0 : tensor<9x4xf64, #MAT_C_C>125 126 //127 // CHECK: {{\[}}[1, 0, 3, 0],128 // CHECK-NEXT: [0, 2, 0, 0],129 // CHECK-NEXT: [1, 0, 1, 1],130 // CHECK-NEXT: [0, 0.5, 0, 0],131 // CHECK-NEXT: [1, 5, 2, 0],132 // CHECK-NEXT: [0, 0, 1.5, 1],133 // CHECK-NEXT: [0, 3.5, 0, 0],134 // CHECK-NEXT: [1, 5, 2, 0],135 // CHECK-NEXT: [1, 0.5, 0, 0]]136 //137 %1 = call @concat_sparse_dense(%sm24cc, %sm34cd, %sm44dc)138 : (tensor<2x4xf64, #MAT_C_C>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64>139 call @dump_mat_dense_9x4(%1) : (tensor<9x4xf64>) -> ()140 141 //142 // CHECK: ---- Sparse Tensor ----143 // CHECK-NEXT: nse = 18144 // CHECK-NEXT: dim = ( 9, 4 )145 // CHECK-NEXT: lvl = ( 9, 4 )146 // CHECK-NEXT: pos[0] : ( 0, 9 )147 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3, 4, 5, 6, 7, 8 )148 // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 7, 10, 12, 13, 16, 18 )149 // CHECK-NEXT: crd[1] : ( 0, 2, 1, 0, 2, 3, 1, 0, 1, 2, 2, 3, 1, 0, 1, 2, 0, 1 )150 // CHECK-NEXT: values : ( 1, 3, 2, 1, 1, 1, 0.5, 1, 5, 2, 1.5, 1, 3.5, 1, 5, 2, 1, 0.5 )151 // CHECK-NEXT: ----152 //153 %2 = call @concat_mix_sparse(%m24, %sm34cd, %sm44dc)154 : (tensor<2x4xf64>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64, #MAT_C_C>155 sparse_tensor.print %2 : tensor<9x4xf64, #MAT_C_C>156 157 //158 // CHECK: {{\[}}[1, 0, 3, 0],159 // CHECK-NEXT: [0, 2, 0, 0],160 // CHECK-NEXT: [1, 0, 1, 1],161 // CHECK-NEXT: [0, 0.5, 0, 0],162 // CHECK-NEXT: [1, 5, 2, 0],163 // CHECK-NEXT: [0, 0, 1.5, 1],164 // CHECK-NEXT: [0, 3.5, 0, 0],165 // CHECK-NEXT: [1, 5, 2, 0],166 // CHECK-NEXT: [1, 0.5, 0, 0]]167 //168 %3 = call @concat_mix_dense(%m24, %sm34cd, %sm44dc)169 : (tensor<2x4xf64>, tensor<3x4xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C>) -> tensor<9x4xf64>170 call @dump_mat_dense_9x4(%3) : (tensor<9x4xf64>) -> ()171 172 // Release resources.173 bufferization.dealloc_tensor %sm24cc : tensor<2x4xf64, #MAT_C_C>174 bufferization.dealloc_tensor %sm34cd : tensor<3x4xf64, #MAT_C_D>175 bufferization.dealloc_tensor %sm44dc : tensor<4x4xf64, #MAT_D_C>176 bufferization.dealloc_tensor %0 : tensor<9x4xf64, #MAT_C_C>177 bufferization.dealloc_tensor %1 : tensor<9x4xf64>178 bufferization.dealloc_tensor %2 : tensor<9x4xf64, #MAT_C_C>179 bufferization.dealloc_tensor %3 : tensor<9x4xf64>180 return181 }182}183