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// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false 22// RUN: %{compile} | %{run} | FileCheck %s 23// 24// Do the same run, but now with vectorization. 25// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true 26// RUN: %{compile} | %{run} | FileCheck %s 27// 28// Do the same run, but now with VLA vectorization. 29// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %} 30 31#TensorCSR = #sparse_tensor.encoding<{ 32 map = (d0, d1, d2) -> (d0 : compressed, d1 : dense, d2 : compressed) 33}> 34 35#TensorRow = #sparse_tensor.encoding<{ 36 map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : dense) 37}> 38 39#CCoo = #sparse_tensor.encoding<{ 40 map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed(nonunique), d2 : singleton) 41}> 42 43#DCoo = #sparse_tensor.encoding<{ 44 map = (d0, d1, d2) -> (d0 : dense, d1 : compressed(nonunique), d2 : singleton) 45}> 46 47 48module { 49 // 50 // Main driver. 51 // 52 func.func @main() { 53 %c0 = arith.constant 0 : index 54 %c1 = arith.constant 1 : index 55 %c2 = arith.constant 2 : index 56 %c3 = arith.constant 3 : index 57 %c4 = arith.constant 4 : index 58 %f1 = arith.constant 1.1 : f64 59 %f2 = arith.constant 2.2 : f64 60 %f3 = arith.constant 3.3 : f64 61 %f4 = arith.constant 4.4 : f64 62 %f5 = arith.constant 5.5 : f64 63 64 // CHECK: ---- Sparse Tensor ---- 65 // CHECK-NEXT: nse = 5 66 // CHECK-NEXT: dim = ( 5, 4, 3 ) 67 // CHECK-NEXT: lvl = ( 5, 4, 3 ) 68 // CHECK-NEXT: pos[0] : ( 0, 2 ) 69 // CHECK-NEXT: crd[0] : ( 3, 4 ) 70 // CHECK-NEXT: pos[2] : ( 0, 2, 2, 2, 3, 3, 3, 4, 5 ) 71 // CHECK-NEXT: crd[2] : ( 1, 2, 1, 2, 2 ) 72 // CHECK-NEXT: values : ( 1.1, 2.2, 3.3, 4.4, 5.5 ) 73 // CHECK-NEXT: ---- 74 %tensora = tensor.empty() : tensor<5x4x3xf64, #TensorCSR> 75 %tensor1 = tensor.insert %f1 into %tensora[%c3, %c0, %c1] : tensor<5x4x3xf64, #TensorCSR> 76 %tensor2 = tensor.insert %f2 into %tensor1[%c3, %c0, %c2] : tensor<5x4x3xf64, #TensorCSR> 77 %tensor3 = tensor.insert %f3 into %tensor2[%c3, %c3, %c1] : tensor<5x4x3xf64, #TensorCSR> 78 %tensor4 = tensor.insert %f4 into %tensor3[%c4, %c2, %c2] : tensor<5x4x3xf64, #TensorCSR> 79 %tensor5 = tensor.insert %f5 into %tensor4[%c4, %c3, %c2] : tensor<5x4x3xf64, #TensorCSR> 80 %tensorm = sparse_tensor.load %tensor5 hasInserts : tensor<5x4x3xf64, #TensorCSR> 81 sparse_tensor.print %tensorm : tensor<5x4x3xf64, #TensorCSR> 82 83 // CHECK-NEXT: ---- Sparse Tensor ---- 84 // CHECK-NEXT: nse = 12 85 // CHECK-NEXT: dim = ( 5, 4, 3 ) 86 // CHECK-NEXT: lvl = ( 5, 4, 3 ) 87 // CHECK-NEXT: pos[0] : ( 0, 2 ) 88 // CHECK-NEXT: crd[0] : ( 3, 4 ) 89 // CHECK-NEXT: pos[1] : ( 0, 2, 4 ) 90 // CHECK-NEXT: crd[1] : ( 0, 3, 2, 3 ) 91 // CHECK-NEXT: values : ( 0, 1.1, 2.2, 0, 3.3, 0, 0, 0, 4.4, 0, 0, 5.5 ) 92 // CHECK-NEXT: ---- 93 %rowa = tensor.empty() : tensor<5x4x3xf64, #TensorRow> 94 %row1 = tensor.insert %f1 into %rowa[%c3, %c0, %c1] : tensor<5x4x3xf64, #TensorRow> 95 %row2 = tensor.insert %f2 into %row1[%c3, %c0, %c2] : tensor<5x4x3xf64, #TensorRow> 96 %row3 = tensor.insert %f3 into %row2[%c3, %c3, %c1] : tensor<5x4x3xf64, #TensorRow> 97 %row4 = tensor.insert %f4 into %row3[%c4, %c2, %c2] : tensor<5x4x3xf64, #TensorRow> 98 %row5 = tensor.insert %f5 into %row4[%c4, %c3, %c2] : tensor<5x4x3xf64, #TensorRow> 99 %rowm = sparse_tensor.load %row5 hasInserts : tensor<5x4x3xf64, #TensorRow> 100 sparse_tensor.print %rowm : tensor<5x4x3xf64, #TensorRow> 101 102 // CHECK-NEXT: ---- Sparse Tensor ---- 103 // CHECK-NEXT: nse = 5 104 // CHECK-NEXT: dim = ( 5, 4, 3 ) 105 // CHECK-NEXT: lvl = ( 5, 4, 3 ) 106 // CHECK-NEXT: pos[0] : ( 0, 2 ) 107 // CHECK-NEXT: crd[0] : ( 3, 4 ) 108 // CHECK-NEXT: pos[1] : ( 0, 3, 5 ) 109 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 2, 3, 1, 2, 2, 3, 2 ) 110 // CHECK-NEXT: values : ( 1.1, 2.2, 3.3, 4.4, 5.5 ) 111 // CHECK-NEXT: ---- 112 %ccoo = tensor.empty() : tensor<5x4x3xf64, #CCoo> 113 %ccoo1 = tensor.insert %f1 into %ccoo[%c3, %c0, %c1] : tensor<5x4x3xf64, #CCoo> 114 %ccoo2 = tensor.insert %f2 into %ccoo1[%c3, %c0, %c2] : tensor<5x4x3xf64, #CCoo> 115 %ccoo3 = tensor.insert %f3 into %ccoo2[%c3, %c3, %c1] : tensor<5x4x3xf64, #CCoo> 116 %ccoo4 = tensor.insert %f4 into %ccoo3[%c4, %c2, %c2] : tensor<5x4x3xf64, #CCoo> 117 %ccoo5 = tensor.insert %f5 into %ccoo4[%c4, %c3, %c2] : tensor<5x4x3xf64, #CCoo> 118 %ccoom = sparse_tensor.load %ccoo5 hasInserts : tensor<5x4x3xf64, #CCoo> 119 sparse_tensor.print %ccoom : tensor<5x4x3xf64, #CCoo> 120 121 // CHECK-NEXT: ---- Sparse Tensor ---- 122 // CHECK-NEXT: nse = 5 123 // CHECK-NEXT: dim = ( 5, 4, 3 ) 124 // CHECK-NEXT: lvl = ( 5, 4, 3 ) 125 // CHECK-NEXT: pos[1] : ( 0, 0, 0, 0, 3, 5 ) 126 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 2, 3, 1, 2, 2, 3, 2 ) 127 // CHECK-NEXT: values : ( 1.1, 2.2, 3.3, 4.4, 5.5 ) 128 // CHECK-NEXT: ---- 129 %dcoo = tensor.empty() : tensor<5x4x3xf64, #DCoo> 130 %dcoo1 = tensor.insert %f1 into %dcoo[%c3, %c0, %c1] : tensor<5x4x3xf64, #DCoo> 131 %dcoo2 = tensor.insert %f2 into %dcoo1[%c3, %c0, %c2] : tensor<5x4x3xf64, #DCoo> 132 %dcoo3 = tensor.insert %f3 into %dcoo2[%c3, %c3, %c1] : tensor<5x4x3xf64, #DCoo> 133 %dcoo4 = tensor.insert %f4 into %dcoo3[%c4, %c2, %c2] : tensor<5x4x3xf64, #DCoo> 134 %dcoo5 = tensor.insert %f5 into %dcoo4[%c4, %c3, %c2] : tensor<5x4x3xf64, #DCoo> 135 %dcoom = sparse_tensor.load %dcoo5 hasInserts : tensor<5x4x3xf64, #DCoo> 136 sparse_tensor.print %dcoom : tensor<5x4x3xf64, #DCoo> 137 138 // Release resources. 139 bufferization.dealloc_tensor %tensorm : tensor<5x4x3xf64, #TensorCSR> 140 bufferization.dealloc_tensor %rowm : tensor<5x4x3xf64, #TensorRow> 141 bufferization.dealloc_tensor %ccoom : tensor<5x4x3xf64, #CCoo> 142 bufferization.dealloc_tensor %dcoom : tensor<5x4x3xf64, #DCoo> 143 144 return 145 } 146} 147