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 vectorization. 28// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=4 enable-buffer-initialization=true 29// RUN: %{compile} | %{run} | FileCheck %s 30// 31// Do the same run, but now with VLA vectorization. 32// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %} 33 34#SparseVector = #sparse_tensor.encoding<{map = (d0) -> (d0 : compressed)}> 35#CSR = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : dense, d1 : compressed)}> 36#CSC = #sparse_tensor.encoding<{ 37 map = (d0, d1) -> (d1 : dense, d0 : compressed) 38}> 39 40// 41// Traits for tensor operations. 42// 43#trait_vec_select = { 44 indexing_maps = [ 45 affine_map<(i) -> (i)>, // A 46 affine_map<(i) -> (i)> // C (out) 47 ], 48 iterator_types = ["parallel"] 49} 50 51#trait_mat_select = { 52 indexing_maps = [ 53 affine_map<(i,j) -> (i,j)>, // A (in) 54 affine_map<(i,j) -> (i,j)> // X (out) 55 ], 56 iterator_types = ["parallel", "parallel"] 57} 58 59module { 60 func.func @vecSelect(%arga: tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector> { 61 %c0 = arith.constant 0 : index 62 %cf1 = arith.constant 1.0 : f64 63 %d0 = tensor.dim %arga, %c0 : tensor<?xf64, #SparseVector> 64 %xv = tensor.empty(%d0): tensor<?xf64, #SparseVector> 65 %0 = linalg.generic #trait_vec_select 66 ins(%arga: tensor<?xf64, #SparseVector>) 67 outs(%xv: tensor<?xf64, #SparseVector>) { 68 ^bb(%a: f64, %b: f64): 69 %1 = sparse_tensor.select %a : f64 { 70 ^bb0(%x: f64): 71 %keep = arith.cmpf "oge", %x, %cf1 : f64 72 sparse_tensor.yield %keep : i1 73 } 74 linalg.yield %1 : f64 75 } -> tensor<?xf64, #SparseVector> 76 return %0 : tensor<?xf64, #SparseVector> 77 } 78 79 func.func @matUpperTriangle(%arga: tensor<?x?xf64, #CSR>) -> tensor<?x?xf64, #CSR> { 80 %c0 = arith.constant 0 : index 81 %c1 = arith.constant 1 : index 82 %d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #CSR> 83 %d1 = tensor.dim %arga, %c1 : tensor<?x?xf64, #CSR> 84 %xv = tensor.empty(%d0, %d1): tensor<?x?xf64, #CSR> 85 %0 = linalg.generic #trait_mat_select 86 ins(%arga: tensor<?x?xf64, #CSR>) 87 outs(%xv: tensor<?x?xf64, #CSR>) { 88 ^bb(%a: f64, %b: f64): 89 %row = linalg.index 0 : index 90 %col = linalg.index 1 : index 91 %1 = sparse_tensor.select %a : f64 { 92 ^bb0(%x: f64): 93 %keep = arith.cmpi "ugt", %col, %row : index 94 sparse_tensor.yield %keep : i1 95 } 96 linalg.yield %1 : f64 97 } -> tensor<?x?xf64, #CSR> 98 return %0 : tensor<?x?xf64, #CSR> 99 } 100 101 // Driver method to call and verify vector kernels. 102 func.func @main() { 103 %c0 = arith.constant 0 : index 104 105 // Setup sparse matrices. 106 %v1 = arith.constant sparse< 107 [ [1], [3], [5], [7], [9] ], 108 [ 1.0, 2.0, -4.0, 0.0, 5.0 ] 109 > : tensor<10xf64> 110 %m1 = arith.constant sparse< 111 [ [0, 3], [1, 4], [2, 1], [2, 3], [3, 3], [3, 4], [4, 2] ], 112 [ 1., 2., 3., 4., 5., 6., 7.] 113 > : tensor<5x5xf64> 114 %sv1 = sparse_tensor.convert %v1 : tensor<10xf64> to tensor<?xf64, #SparseVector> 115 %sm1 = sparse_tensor.convert %m1 : tensor<5x5xf64> to tensor<?x?xf64, #CSR> 116 117 // Call sparse matrix kernels. 118 %1 = call @vecSelect(%sv1) : (tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector> 119 %2 = call @matUpperTriangle(%sm1) : (tensor<?x?xf64, #CSR>) -> tensor<?x?xf64, #CSR> 120 121 // 122 // Verify the results. 123 // 124 // CHECK: ---- Sparse Tensor ---- 125 // CHECK-NEXT: nse = 5 126 // CHECK-NEXT: dim = ( 10 ) 127 // CHECK-NEXT: lvl = ( 10 ) 128 // CHECK-NEXT: pos[0] : ( 0, 5 ) 129 // CHECK-NEXT: crd[0] : ( 1, 3, 5, 7, 9 ) 130 // CHECK-NEXT: values : ( 1, 2, -4, 0, 5 ) 131 // CHECK-NEXT: ---- 132 // CHECK: ---- Sparse Tensor ---- 133 // CHECK-NEXT: nse = 7 134 // CHECK-NEXT: dim = ( 5, 5 ) 135 // CHECK-NEXT: lvl = ( 5, 5 ) 136 // CHECK-NEXT: pos[1] : ( 0, 1, 2, 4, 6, 7 ) 137 // CHECK-NEXT: crd[1] : ( 3, 4, 1, 3, 3, 4, 2 ) 138 // CHECK-NEXT: values : ( 1, 2, 3, 4, 5, 6, 7 ) 139 // CHECK-NEXT: ---- 140 // CHECK: ---- Sparse Tensor ---- 141 // CHECK-NEXT: nse = 3 142 // CHECK-NEXT: dim = ( 10 ) 143 // CHECK-NEXT: lvl = ( 10 ) 144 // CHECK-NEXT: pos[0] : ( 0, 3 ) 145 // CHECK-NEXT: crd[0] : ( 1, 3, 9 ) 146 // CHECK-NEXT: values : ( 1, 2, 5 ) 147 // CHECK-NEXT: ---- 148 // CHECK: ---- Sparse Tensor ---- 149 // CHECK-NEXT: nse = 4 150 // CHECK-NEXT: dim = ( 5, 5 ) 151 // CHECK-NEXT: lvl = ( 5, 5 ) 152 // CHECK-NEXT: pos[1] : ( 0, 1, 2, 3, 4, 4 ) 153 // CHECK-NEXT: crd[1] : ( 3, 4, 3, 4 ) 154 // CHECK-NEXT: values : ( 1, 2, 4, 6 ) 155 // CHECK-NEXT: ---- 156 // 157 sparse_tensor.print %sv1 : tensor<?xf64, #SparseVector> 158 sparse_tensor.print %sm1 : tensor<?x?xf64, #CSR> 159 sparse_tensor.print %1 : tensor<?xf64, #SparseVector> 160 sparse_tensor.print %2 : tensor<?x?xf64, #CSR> 161 162 // Release the resources. 163 bufferization.dealloc_tensor %sv1 : tensor<?xf64, #SparseVector> 164 bufferization.dealloc_tensor %sm1 : tensor<?x?xf64, #CSR> 165 bufferization.dealloc_tensor %1 : tensor<?xf64, #SparseVector> 166 bufferization.dealloc_tensor %2 : tensor<?x?xf64, #CSR> 167 return 168 } 169} 170