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// RUN: %{compile} | %{run} | FileCheck %s 22// 23// Do the same run, but now with direct IR generation. 24// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false enable-buffer-initialization=true 25// RUN: %{compile} | %{run} | FileCheck %s 26// 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=true 29// RUN: %{compile} | %{run} | FileCheck %s 30// 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 func.func private @printMemref1dF64(%ptr : memref<?xf64>) attributes { llvm.emit_c_interface } 56 57 // 58 // Tests with permutation (concatenate on dimension 1) 59 // 60 61 // Concats all sparse matrices (with different encodings) to a sparse matrix. 62 func.func @concat_sparse_sparse_perm_dim1(%arg0: tensor<4x2xf64, #MAT_C_C_P>, %arg1: tensor<4x3xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64, #MAT_C_C_P> { 63 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 1 : index} 64 : tensor<4x2xf64, #MAT_C_C_P>, tensor<4x3xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C> to tensor<4x9xf64, #MAT_C_C_P> 65 return %0 : tensor<4x9xf64, #MAT_C_C_P> 66 } 67 68 // Concats all sparse matrices (with different encodings) to a dense matrix. 69 func.func @concat_sparse_dense_perm_dim1(%arg0: tensor<4x2xf64, #MAT_C_C_P>, %arg1: tensor<4x3xf64, #MAT_C_D_P>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64> { 70 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 1 : index} 71 : tensor<4x2xf64, #MAT_C_C_P>, tensor<4x3xf64, #MAT_C_D_P>, tensor<4x4xf64, #MAT_D_C> to tensor<4x9xf64> 72 return %0 : tensor<4x9xf64> 73 } 74 75 // Concats mix sparse and dense matrices to a sparse matrix. 76 func.func @concat_mix_sparse_perm_dim1(%arg0: tensor<4x2xf64>, %arg1: tensor<4x3xf64, #MAT_C_D_P>, %arg2: tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64, #MAT_C_C> { 77 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 1 : index} 78 : tensor<4x2xf64>, tensor<4x3xf64, #MAT_C_D_P>, tensor<4x4xf64, #MAT_D_C> to tensor<4x9xf64, #MAT_C_C> 79 return %0 : tensor<4x9xf64, #MAT_C_C> 80 } 81 82 // Concats mix sparse and dense matrices to a dense matrix. 83 func.func @concat_mix_dense_perm_dim1(%arg0: tensor<4x2xf64>, %arg1: tensor<4x3xf64, #MAT_C_D>, %arg2: tensor<4x4xf64, #MAT_D_C_P>) -> tensor<4x9xf64> { 84 %0 = sparse_tensor.concatenate %arg0, %arg1, %arg2 {dimension = 1 : index} 85 : tensor<4x2xf64>, tensor<4x3xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C_P> to tensor<4x9xf64> 86 return %0 : tensor<4x9xf64> 87 } 88 89 func.func @dump_mat_dense_4x9(%A: tensor<4x9xf64>) { 90 %1 = tensor.cast %A : tensor<4x9xf64> to tensor<*xf64> 91 call @printMemrefF64(%1) : (tensor<*xf64>) -> () 92 93 return 94 } 95 96 // Driver method to call and verify kernels. 97 func.func @main() { 98 %m42 = arith.constant dense< 99 [ [ 1.0, 0.0 ], 100 [ 3.1, 0.0 ], 101 [ 0.0, 2.0 ], 102 [ 0.0, 0.0 ] ]> : tensor<4x2xf64> 103 %m43 = arith.constant dense< 104 [ [ 1.0, 0.0, 1.0 ], 105 [ 1.0, 0.0, 0.5 ], 106 [ 0.0, 0.0, 1.0 ], 107 [ 5.0, 2.0, 0.0 ] ]> : tensor<4x3xf64> 108 %m24 = arith.constant dense< 109 [ [ 1.0, 0.0, 3.0, 0.0], 110 [ 0.0, 2.0, 0.0, 0.0] ]> : tensor<2x4xf64> 111 %m34 = arith.constant dense< 112 [ [ 1.0, 0.0, 1.0, 1.0], 113 [ 0.0, 0.5, 0.0, 0.0], 114 [ 1.0, 5.0, 2.0, 0.0] ]> : tensor<3x4xf64> 115 %m44 = arith.constant dense< 116 [ [ 0.0, 0.0, 1.5, 1.0], 117 [ 0.0, 3.5, 0.0, 0.0], 118 [ 1.0, 5.0, 2.0, 0.0], 119 [ 1.0, 0.5, 0.0, 0.0] ]> : tensor<4x4xf64> 120 121 %sm42cc = sparse_tensor.convert %m42 : tensor<4x2xf64> to tensor<4x2xf64, #MAT_C_C> 122 %sm43cd = sparse_tensor.convert %m43 : tensor<4x3xf64> to tensor<4x3xf64, #MAT_C_D> 123 %sm44dc = sparse_tensor.convert %m44 : tensor<4x4xf64> to tensor<4x4xf64, #MAT_D_C> 124 125 %sm42ccp = sparse_tensor.convert %m42 : tensor<4x2xf64> to tensor<4x2xf64, #MAT_C_C_P> 126 %sm43cdp = sparse_tensor.convert %m43 : tensor<4x3xf64> to tensor<4x3xf64, #MAT_C_D_P> 127 %sm44dcp = sparse_tensor.convert %m44 : tensor<4x4xf64> to tensor<4x4xf64, #MAT_D_C_P> 128 129 // 130 // CHECK: ---- Sparse Tensor ---- 131 // CHECK-NEXT: nse = 18 132 // CHECK-NEXT: dim = ( 4, 9 ) 133 // CHECK-NEXT: lvl = ( 9, 4 ) 134 // CHECK-NEXT: pos[0] : ( 0, 9 ) 135 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3, 4, 5, 6, 7, 8 ) 136 // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 7, 10, 12, 15, 17, 18 ) 137 // CHECK-NEXT: crd[1] : ( 0, 1, 2, 0, 1, 3, 3, 0, 1, 2, 2, 3, 1, 2, 3, 0, 2, 0 ) 138 // CHECK-NEXT: values : ( 1, 3.1, 2, 1, 1, 5, 2, 1, 0.5, 1, 1, 1, 3.5, 5, 0.5, 1.5, 2, 1 ) 139 // CHECK-NEXT: ---- 140 // 141 %12 = call @concat_sparse_sparse_perm_dim1(%sm42ccp, %sm43cd, %sm44dc) 142 : (tensor<4x2xf64, #MAT_C_C_P>, tensor<4x3xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64, #MAT_C_C_P> 143 sparse_tensor.print %12 : tensor<4x9xf64, #MAT_C_C_P> 144 145 // CHECK: {{\[}}[1, 0, 1, 0, 1, 0, 0, 1.5, 1], 146 // CHECK-NEXT: [3.1, 0, 1, 0, 0.5, 0, 3.5, 0, 0], 147 // CHECK-NEXT: [0, 2, 0, 0, 1, 1, 5, 2, 0], 148 // CHECK-NEXT: [0, 0, 5, 2, 0, 1, 0.5, 0, 0]] 149 %13 = call @concat_sparse_dense_perm_dim1(%sm42ccp, %sm43cdp, %sm44dc) 150 : (tensor<4x2xf64, #MAT_C_C_P>, tensor<4x3xf64, #MAT_C_D_P>, tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64> 151 call @dump_mat_dense_4x9(%13) : (tensor<4x9xf64>) -> () 152 153 // 154 // CHECK: ---- Sparse Tensor ---- 155 // CHECK-NEXT: nse = 18 156 // CHECK-NEXT: dim = ( 4, 9 ) 157 // CHECK-NEXT: lvl = ( 4, 9 ) 158 // CHECK-NEXT: pos[0] : ( 0, 4 ) 159 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3 ) 160 // CHECK-NEXT: pos[1] : ( 0, 5, 9, 14, 18 ) 161 // CHECK-NEXT: crd[1] : ( 0, 2, 4, 7, 8, 0, 2, 4, 6, 1, 4, 5, 6, 7, 2, 3, 5, 6 ) 162 // CHECK-NEXT: values : ( 1, 1, 1, 1.5, 1, 3.1, 1, 0.5, 3.5, 2, 1, 1, 5, 2, 5, 2, 1, 0.5 ) 163 // CHECK-NEXT: ---- 164 // 165 %14 = call @concat_mix_sparse_perm_dim1(%m42, %sm43cdp, %sm44dc) 166 : (tensor<4x2xf64>, tensor<4x3xf64, #MAT_C_D_P>, tensor<4x4xf64, #MAT_D_C>) -> tensor<4x9xf64, #MAT_C_C> 167 sparse_tensor.print %14 : tensor<4x9xf64, #MAT_C_C> 168 169 // CHECK: {{\[}}[1, 0, 1, 0, 1, 0, 0, 1.5, 1], 170 // CHECK-NEXT: [3.1, 0, 1, 0, 0.5, 0, 3.5, 0, 0], 171 // CHECK-NEXT: [0, 2, 0, 0, 1, 1, 5, 2, 0], 172 // CHECK-NEXT: [0, 0, 5, 2, 0, 1, 0.5, 0, 0]] 173 %15 = call @concat_mix_dense_perm_dim1(%m42, %sm43cd, %sm44dcp) 174 : (tensor<4x2xf64>, tensor<4x3xf64, #MAT_C_D>, tensor<4x4xf64, #MAT_D_C_P>) -> tensor<4x9xf64> 175 call @dump_mat_dense_4x9(%15) : (tensor<4x9xf64>) -> () 176 177 // Release resources. 178 bufferization.dealloc_tensor %sm42cc : tensor<4x2xf64, #MAT_C_C> 179 bufferization.dealloc_tensor %sm43cd : tensor<4x3xf64, #MAT_C_D> 180 bufferization.dealloc_tensor %sm44dc : tensor<4x4xf64, #MAT_D_C> 181 bufferization.dealloc_tensor %sm42ccp : tensor<4x2xf64, #MAT_C_C_P> 182 bufferization.dealloc_tensor %sm43cdp : tensor<4x3xf64, #MAT_C_D_P> 183 bufferization.dealloc_tensor %sm44dcp : tensor<4x4xf64, #MAT_D_C_P> 184 bufferization.dealloc_tensor %12 : tensor<4x9xf64, #MAT_C_C_P> 185 bufferization.dealloc_tensor %13 : tensor<4x9xf64> 186 bufferization.dealloc_tensor %14 : tensor<4x9xf64, #MAT_C_C> 187 bufferization.dealloc_tensor %15 : tensor<4x9xf64> 188 return 189 } 190} 191